diff --git a/media/libaom/README_MCP b/media/libaom/README_MCP
index 5ae9f96c42..bab18e4b5a 100644
--- a/media/libaom/README_MCP
+++ b/media/libaom/README_MCP
@@ -10,4 +10,4 @@ The upstream aom git repository is:
 
     https://aomedia.googlesource.com/aom
 
-The git commit ID used was 1e227d41f0616de9548a673a83a21ef990b62591.
+The git commit ID used was 0ec86ac7ae1e32a7e70410fa4972a655ec3670a4.
diff --git a/media/libaom/config/generic/config/aom_config.asm b/media/libaom/config/generic/config/aom_config.asm
index 897eeb4fd1..a236ca9dd2 100644
--- a/media/libaom/config/generic/config/aom_config.asm
+++ b/media/libaom/config/generic/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/generic/config/aom_config.h b/media/libaom/config/generic/config/aom_config.h
index 54fb243e11..3b3798a451 100644
--- a/media/libaom/config/generic/config/aom_config.h
+++ b/media/libaom/config/generic/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 0
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/generic/config/aom_dsp_rtcd.h b/media/libaom/config/generic/config/aom_dsp_rtcd.h
index c3b61b8a8b..62222ee19e 100644
--- a/media/libaom/config/generic/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/generic/config/aom_dsp_rtcd.h
@@ -1360,9 +1360,6 @@ void aom_v_predictor_8x4_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 #define aom_v_predictor_8x8 aom_v_predictor_8x8_c
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-#define av1_round_shift_array av1_round_shift_array_c
-
 void aom_dsp_rtcd(void);
 
 #include "config/aom_config.h"
diff --git a/media/libaom/config/generic/config/aom_scale_rtcd.h b/media/libaom/config/generic/config/aom_scale_rtcd.h
index 1ee9bef3d7..8b92a2d57f 100644
--- a/media/libaom/config/generic/config/aom_scale_rtcd.h
+++ b/media/libaom/config/generic/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/generic/config/av1_rtcd.h b/media/libaom/config/generic/config/av1_rtcd.h
index e2fe11bbdc..ab42789f64 100644
--- a/media/libaom/config/generic/config/av1_rtcd.h
+++ b/media/libaom/config/generic/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -81,6 +83,18 @@ void av1_convolve_x_sr_c(const uint8_t *src, int src_stride, uint8_t *dst, int d
 void av1_convolve_y_sr_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 #define av1_convolve_y_sr av1_convolve_y_sr_c
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+#define av1_dist_wtd_convolve_2d av1_dist_wtd_convolve_2d_c
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+#define av1_dist_wtd_convolve_2d_copy av1_dist_wtd_convolve_2d_copy_c
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+#define av1_dist_wtd_convolve_x av1_dist_wtd_convolve_x_c
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+#define av1_dist_wtd_convolve_y av1_dist_wtd_convolve_y_c
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 #define av1_dr_prediction_z1 av1_dr_prediction_z1_c
 
@@ -132,6 +146,18 @@ void av1_highbd_convolve_x_sr_c(const uint16_t *src, int src_stride, uint16_t *d
 void av1_highbd_convolve_y_sr_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 #define av1_highbd_convolve_y_sr av1_highbd_convolve_y_sr_c
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_2d av1_highbd_dist_wtd_convolve_2d_c
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_2d_copy av1_highbd_dist_wtd_convolve_2d_copy_c
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_x av1_highbd_dist_wtd_convolve_x_c
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_y av1_highbd_dist_wtd_convolve_y_c
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
 
@@ -144,20 +170,20 @@ void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add av1_highbd_inv_txfm_add_c
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_16x16 av1_highbd_inv_txfm_add_16x16_c
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_16x4 av1_highbd_inv_txfm_add_16x4_c
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_16x8 av1_highbd_inv_txfm_add_16x8_c
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_32x32 av1_highbd_inv_txfm_add_32x32_c
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_4x16 av1_highbd_inv_txfm_add_4x16_c
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add_4x4 av1_highbd_inv_txfm_add_4x4_c
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_8x16 av1_highbd_inv_txfm_add_8x16_c
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_4x8 av1_highbd_inv_txfm_add_4x8_c
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_8x4 av1_highbd_inv_txfm_add_8x4_c
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add_8x8 av1_highbd_inv_txfm_add_8x8_c
@@ -168,18 +194,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_2d av1_highbd_jnt_convolve_2d_c
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_2d_copy av1_highbd_jnt_convolve_2d_copy_c
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_x av1_highbd_jnt_convolve_x_c
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_y av1_highbd_jnt_convolve_y_c
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 #define av1_highbd_warp_affine av1_highbd_warp_affine_c
 
@@ -246,17 +260,8 @@ void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output, int stride
 void av1_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_inv_txfm_add av1_inv_txfm_add_c
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-#define av1_jnt_convolve_2d av1_jnt_convolve_2d_c
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-#define av1_jnt_convolve_2d_copy av1_jnt_convolve_2d_copy_c
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-#define av1_jnt_convolve_x av1_jnt_convolve_x_c
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-#define av1_jnt_convolve_y av1_jnt_convolve_y_c
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+#define av1_round_shift_array av1_round_shift_array_c
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -275,7 +280,7 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width, int he
 void av1_wiener_convolve_add_src_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 #define av1_wiener_convolve_add_src av1_wiener_convolve_add_src_c
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 #define cdef_filter_block cdef_filter_block_c
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
diff --git a/media/libaom/config/linux/arm/config/aom_config.asm b/media/libaom/config/linux/arm/config/aom_config.asm
index f5db6f6ca6..c680ea1ba7 100644
--- a/media/libaom/config/linux/arm/config/aom_config.asm
+++ b/media/libaom/config/linux/arm/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@
 .equ CONFIG_BIG_ENDIAN, 0
 .equ CONFIG_BITSTREAM_DEBUG, 0
 .equ CONFIG_COEFFICIENT_RANGE_CHECKING, 0
-.equ CONFIG_COLLECT_INTER_MODE_RD_STATS, 0
+.equ CONFIG_COLLECT_PARTITION_STATS, 0
 .equ CONFIG_COLLECT_RD_STATS, 0
 .equ CONFIG_DEBUG, 0
 .equ CONFIG_DENOISE, 1
+.equ CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8, 1
 .equ CONFIG_DIST_8X8, 0
 .equ CONFIG_ENTROPY_STATS, 0
 .equ CONFIG_FILEOPTIONS, 1
-.equ CONFIG_FIX_GF_LENGTH, 1
 .equ CONFIG_FP_MB_STATS, 0
 .equ CONFIG_GCC, 1
 .equ CONFIG_GCOV, 0
-.equ CONFIG_GLOBAL_MOTION_SEARCH, 1
 .equ CONFIG_GPROF, 0
 .equ CONFIG_INSPECTION, 0
 .equ CONFIG_INTERNAL_STATS, 0
@@ -44,15 +43,16 @@
 .equ CONFIG_MISMATCH_DEBUG, 0
 .equ CONFIG_MULTITHREAD, 1
 .equ CONFIG_NORMAL_TILE_MODE, 0
+.equ CONFIG_ONE_PASS_SVM, 0
 .equ CONFIG_OS_SUPPORT, 1
 .equ CONFIG_PIC, 1
 .equ CONFIG_RD_DEBUG, 0
-.equ CONFIG_REDUCED_ENCODER_BORDER, 0
 .equ CONFIG_RUNTIME_CPU_DETECT, 1
 .equ CONFIG_SHARED, 0
 .equ CONFIG_SHARP_SETTINGS, 0
 .equ CONFIG_SIZE_LIMIT, 0
 .equ CONFIG_SPATIAL_RESAMPLING, 1
+.equ CONFIG_SPEED_STATS, 0
 .equ CONFIG_STATIC, 1
 .equ CONFIG_WEBM_IO, 0
 .equ DECODE_HEIGHT_LIMIT, 0
diff --git a/media/libaom/config/linux/arm/config/aom_config.h b/media/libaom/config/linux/arm/config/aom_config.h
index 850eebfbd4..ffc5188369 100644
--- a/media/libaom/config/linux/arm/config/aom_config.h
+++ b/media/libaom/config/linux/arm/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 1
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 0
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/linux/arm/config/aom_dsp_rtcd.h b/media/libaom/config/linux/arm/config/aom_dsp_rtcd.h
index d80f8972b1..22fa7a1bbb 100644
--- a/media/libaom/config/linux/arm/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/linux/arm/config/aom_dsp_rtcd.h
@@ -1400,10 +1400,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 RTCD_EXTERN void (*aom_v_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_neon(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #include "config/aom_config.h"
@@ -1496,8 +1492,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_NEON) aom_v_predictor_4x4 = aom_v_predictor_4x4_neon;
     aom_v_predictor_8x8 = aom_v_predictor_8x8_c;
     if (flags & HAS_NEON) aom_v_predictor_8x8 = aom_v_predictor_8x8_neon;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_NEON) av1_round_shift_array = av1_round_shift_array_neon;
 }
 #endif
 
diff --git a/media/libaom/config/linux/arm/config/aom_scale_rtcd.h b/media/libaom/config/linux/arm/config/aom_scale_rtcd.h
index 165d730936..7260bd372f 100644
--- a/media/libaom/config/linux/arm/config/aom_scale_rtcd.h
+++ b/media/libaom/config/linux/arm/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/linux/arm/config/av1_rtcd.h b/media/libaom/config/linux/arm/config/av1_rtcd.h
index b54b4514dd..d51b6816cf 100644
--- a/media/libaom/config/linux/arm/config/av1_rtcd.h
+++ b/media/libaom/config/linux/arm/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -87,6 +89,22 @@ void av1_convolve_y_sr_c(const uint8_t *src, int src_stride, uint8_t *dst, int d
 void av1_convolve_y_sr_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 #define av1_dr_prediction_z1 av1_dr_prediction_z1_c
 
@@ -138,6 +156,18 @@ void av1_highbd_convolve_x_sr_c(const uint16_t *src, int src_stride, uint16_t *d
 void av1_highbd_convolve_y_sr_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 #define av1_highbd_convolve_y_sr av1_highbd_convolve_y_sr_c
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_2d av1_highbd_dist_wtd_convolve_2d_c
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_2d_copy av1_highbd_dist_wtd_convolve_2d_copy_c
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_x av1_highbd_dist_wtd_convolve_x_c
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+#define av1_highbd_dist_wtd_convolve_y av1_highbd_dist_wtd_convolve_y_c
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
 
@@ -150,20 +180,20 @@ void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add av1_highbd_inv_txfm_add_c
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_16x16 av1_highbd_inv_txfm_add_16x16_c
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_16x4 av1_highbd_inv_txfm_add_16x4_c
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_16x8 av1_highbd_inv_txfm_add_16x8_c
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_32x32 av1_highbd_inv_txfm_add_32x32_c
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_4x16 av1_highbd_inv_txfm_add_4x16_c
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add_4x4 av1_highbd_inv_txfm_add_4x4_c
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-#define av1_highbd_inv_txfm_add_8x16 av1_highbd_inv_txfm_add_8x16_c
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_4x8 av1_highbd_inv_txfm_add_4x8_c
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+#define av1_highbd_inv_txfm_add_8x4 av1_highbd_inv_txfm_add_8x4_c
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 #define av1_highbd_inv_txfm_add_8x8 av1_highbd_inv_txfm_add_8x8_c
@@ -174,18 +204,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_2d av1_highbd_jnt_convolve_2d_c
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_2d_copy av1_highbd_jnt_convolve_2d_copy_c
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_x av1_highbd_jnt_convolve_x_c
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-#define av1_highbd_jnt_convolve_y av1_highbd_jnt_convolve_y_c
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 #define av1_highbd_warp_affine av1_highbd_warp_affine_c
 
@@ -253,21 +271,9 @@ void av1_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, con
 void av1_inv_txfm_add_neon(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_neon(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -293,9 +299,9 @@ void av1_wiener_convolve_add_src_c(const uint8_t *src, ptrdiff_t src_stride, uin
 void av1_wiener_convolve_add_src_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_neon(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_neon(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_neon(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -369,16 +375,18 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_NEON) av1_convolve_x_sr = av1_convolve_x_sr_neon;
     av1_convolve_y_sr = av1_convolve_y_sr_c;
     if (flags & HAS_NEON) av1_convolve_y_sr = av1_convolve_y_sr_neon;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_c;
+    if (flags & HAS_NEON) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_neon;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_NEON) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_neon;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_c;
+    if (flags & HAS_NEON) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_neon;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_c;
+    if (flags & HAS_NEON) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_neon;
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_NEON) av1_inv_txfm_add = av1_inv_txfm_add_neon;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_NEON) av1_jnt_convolve_2d = av1_jnt_convolve_2d_neon;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c;
-    if (flags & HAS_NEON) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_neon;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_c;
-    if (flags & HAS_NEON) av1_jnt_convolve_x = av1_jnt_convolve_x_neon;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_c;
-    if (flags & HAS_NEON) av1_jnt_convolve_y = av1_jnt_convolve_y_neon;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_NEON) av1_round_shift_array = av1_round_shift_array_neon;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_NEON) av1_selfguided_restoration = av1_selfguided_restoration_neon;
     av1_warp_affine = av1_warp_affine_c;
diff --git a/media/libaom/config/linux/ia32/config/aom_config.asm b/media/libaom/config/linux/ia32/config/aom_config.asm
index 128f85ec27..d037bc4231 100644
--- a/media/libaom/config/linux/ia32/config/aom_config.asm
+++ b/media/libaom/config/linux/ia32/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 1
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/linux/ia32/config/aom_config.h b/media/libaom/config/linux/ia32/config/aom_config.h
index 02f6f72434..67201081b2 100644
--- a/media/libaom/config/linux/ia32/config/aom_config.h
+++ b/media/libaom/config/linux/ia32/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 1
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/linux/ia32/config/aom_dsp_rtcd.h b/media/libaom/config/linux/ia32/config/aom_dsp_rtcd.h
index 5b3762f109..a60a7aea6e 100644
--- a/media/libaom/config/linux/ia32/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/linux/ia32/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 RTCD_EXTERN void (*aom_h_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -487,10 +489,12 @@ void aom_highbd_blend_a64_vmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const
 RTCD_EXTERN void (*aom_highbd_blend_a64_vmask)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 
 void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_horiz)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
 void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_vert)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
@@ -1709,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 RTCD_EXTERN void (*aom_v_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1955,6 +1955,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_h_predictor_8x4 = aom_h_predictor_8x4_sse2;
     aom_h_predictor_8x8 = aom_h_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_h_predictor_8x8 = aom_h_predictor_8x8_sse2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1962,8 +1965,10 @@ static void setup_rtcd_internal(void)
     aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_sse4_1;
     aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_avx2;
     aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_avx2;
     aom_highbd_convolve_copy = aom_highbd_convolve_copy_c;
     if (flags & HAS_SSE2) aom_highbd_convolve_copy = aom_highbd_convolve_copy_sse2;
@@ -2367,8 +2372,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_v_predictor_8x4 = aom_v_predictor_8x4_sse2;
     aom_v_predictor_8x8 = aom_v_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_v_predictor_8x8 = aom_v_predictor_8x8_sse2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/linux/ia32/config/aom_scale_rtcd.h b/media/libaom/config/linux/ia32/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/linux/ia32/config/aom_scale_rtcd.h
+++ b/media/libaom/config/linux/ia32/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/linux/ia32/config/av1_rtcd.h b/media/libaom/config/linux/ia32/config/av1_rtcd.h
index a2a6f14ff5..8e69b643ef 100644
--- a/media/libaom/config/linux/ia32/config/av1_rtcd.h
+++ b/media/libaom/config/linux/ia32/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -163,40 +189,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -208,26 +255,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -301,25 +328,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -351,12 +362,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -462,6 +473,25 @@ static void setup_rtcd_internal(void)
     av1_convolve_y_sr = av1_convolve_y_sr_c;
     if (flags & HAS_SSE2) av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -484,34 +514,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -524,18 +557,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/linux/x64/config/aom_config.asm b/media/libaom/config/linux/x64/config/aom_config.asm
index 4ab2dedb42..294d0294b6 100644
--- a/media/libaom/config/linux/x64/config/aom_config.asm
+++ b/media/libaom/config/linux/x64/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/linux/x64/config/aom_config.h b/media/libaom/config/linux/x64/config/aom_config.h
index 9a0be7cd11..ffff5abe6a 100644
--- a/media/libaom/config/linux/x64/config/aom_config.h
+++ b/media/libaom/config/linux/x64/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/linux/x64/config/aom_dsp_rtcd.h b/media/libaom/config/linux/x64/config/aom_dsp_rtcd.h
index 2856d4ede7..5826faff95 100644
--- a/media/libaom/config/linux/x64/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/linux/x64/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 #define aom_h_predictor_8x8 aom_h_predictor_8x8_sse2
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -1711,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 #define aom_v_predictor_8x8 aom_v_predictor_8x8_sse2
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1788,6 +1786,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_dc_top_predictor_64x64 = aom_dc_top_predictor_64x64_avx2;
     aom_h_predictor_32x32 = aom_h_predictor_32x32_sse2;
     if (flags & HAS_AVX2) aom_h_predictor_32x32 = aom_h_predictor_32x32_avx2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1989,8 +1990,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_v_predictor_64x32 = aom_v_predictor_64x32_avx2;
     aom_v_predictor_64x64 = aom_v_predictor_64x64_sse2;
     if (flags & HAS_AVX2) aom_v_predictor_64x64 = aom_v_predictor_64x64_avx2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/linux/x64/config/aom_scale_rtcd.h b/media/libaom/config/linux/x64/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/linux/x64/config/aom_scale_rtcd.h
+++ b/media/libaom/config/linux/x64/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/linux/x64/config/av1_rtcd.h b/media/libaom/config/linux/x64/config/av1_rtcd.h
index d273182086..80bacb8dd5 100644
--- a/media/libaom/config/linux/x64/config/av1_rtcd.h
+++ b/media/libaom/config/linux/x64/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -166,40 +192,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -211,26 +258,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -304,25 +331,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -354,12 +365,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -461,6 +472,21 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) av1_convolve_x_sr = av1_convolve_x_sr_avx2;
     av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -482,34 +508,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -522,15 +551,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/mac/x64/config/aom_config.asm b/media/libaom/config/mac/x64/config/aom_config.asm
index 4ab2dedb42..294d0294b6 100644
--- a/media/libaom/config/mac/x64/config/aom_config.asm
+++ b/media/libaom/config/mac/x64/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/mac/x64/config/aom_config.h b/media/libaom/config/mac/x64/config/aom_config.h
index 9a0be7cd11..ffff5abe6a 100644
--- a/media/libaom/config/mac/x64/config/aom_config.h
+++ b/media/libaom/config/mac/x64/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/mac/x64/config/aom_dsp_rtcd.h b/media/libaom/config/mac/x64/config/aom_dsp_rtcd.h
index 2856d4ede7..5826faff95 100644
--- a/media/libaom/config/mac/x64/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/mac/x64/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 #define aom_h_predictor_8x8 aom_h_predictor_8x8_sse2
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -1711,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 #define aom_v_predictor_8x8 aom_v_predictor_8x8_sse2
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1788,6 +1786,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_dc_top_predictor_64x64 = aom_dc_top_predictor_64x64_avx2;
     aom_h_predictor_32x32 = aom_h_predictor_32x32_sse2;
     if (flags & HAS_AVX2) aom_h_predictor_32x32 = aom_h_predictor_32x32_avx2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1989,8 +1990,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_v_predictor_64x32 = aom_v_predictor_64x32_avx2;
     aom_v_predictor_64x64 = aom_v_predictor_64x64_sse2;
     if (flags & HAS_AVX2) aom_v_predictor_64x64 = aom_v_predictor_64x64_avx2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/mac/x64/config/aom_scale_rtcd.h b/media/libaom/config/mac/x64/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/mac/x64/config/aom_scale_rtcd.h
+++ b/media/libaom/config/mac/x64/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/mac/x64/config/av1_rtcd.h b/media/libaom/config/mac/x64/config/av1_rtcd.h
index d273182086..80bacb8dd5 100644
--- a/media/libaom/config/mac/x64/config/av1_rtcd.h
+++ b/media/libaom/config/mac/x64/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -166,40 +192,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -211,26 +258,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -304,25 +331,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -354,12 +365,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -461,6 +472,21 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) av1_convolve_x_sr = av1_convolve_x_sr_avx2;
     av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -482,34 +508,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -522,15 +551,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/win/ia32/config/aom_config.asm b/media/libaom/config/win/ia32/config/aom_config.asm
index 6ae776c7c4..b3a8c2d596 100644
--- a/media/libaom/config/win/ia32/config/aom_config.asm
+++ b/media/libaom/config/win/ia32/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/win/ia32/config/aom_config.h b/media/libaom/config/win/ia32/config/aom_config.h
index 090a455778..394c9d574e 100644
--- a/media/libaom/config/win/ia32/config/aom_config.h
+++ b/media/libaom/config/win/ia32/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/win/ia32/config/aom_dsp_rtcd.h b/media/libaom/config/win/ia32/config/aom_dsp_rtcd.h
index 5b3762f109..a60a7aea6e 100644
--- a/media/libaom/config/win/ia32/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/win/ia32/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 RTCD_EXTERN void (*aom_h_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -487,10 +489,12 @@ void aom_highbd_blend_a64_vmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const
 RTCD_EXTERN void (*aom_highbd_blend_a64_vmask)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 
 void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_horiz)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
 void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_vert)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
@@ -1709,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 RTCD_EXTERN void (*aom_v_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1955,6 +1955,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_h_predictor_8x4 = aom_h_predictor_8x4_sse2;
     aom_h_predictor_8x8 = aom_h_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_h_predictor_8x8 = aom_h_predictor_8x8_sse2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1962,8 +1965,10 @@ static void setup_rtcd_internal(void)
     aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_sse4_1;
     aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_avx2;
     aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_avx2;
     aom_highbd_convolve_copy = aom_highbd_convolve_copy_c;
     if (flags & HAS_SSE2) aom_highbd_convolve_copy = aom_highbd_convolve_copy_sse2;
@@ -2367,8 +2372,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_v_predictor_8x4 = aom_v_predictor_8x4_sse2;
     aom_v_predictor_8x8 = aom_v_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_v_predictor_8x8 = aom_v_predictor_8x8_sse2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/win/ia32/config/aom_scale_rtcd.h b/media/libaom/config/win/ia32/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/win/ia32/config/aom_scale_rtcd.h
+++ b/media/libaom/config/win/ia32/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/win/ia32/config/av1_rtcd.h b/media/libaom/config/win/ia32/config/av1_rtcd.h
index a2a6f14ff5..8e69b643ef 100644
--- a/media/libaom/config/win/ia32/config/av1_rtcd.h
+++ b/media/libaom/config/win/ia32/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -163,40 +189,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -208,26 +255,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -301,25 +328,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -351,12 +362,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -462,6 +473,25 @@ static void setup_rtcd_internal(void)
     av1_convolve_y_sr = av1_convolve_y_sr_c;
     if (flags & HAS_SSE2) av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -484,34 +514,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -524,18 +557,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/win/mingw32/config/aom_config.asm b/media/libaom/config/win/mingw32/config/aom_config.asm
index 6ae776c7c4..b3a8c2d596 100644
--- a/media/libaom/config/win/mingw32/config/aom_config.asm
+++ b/media/libaom/config/win/mingw32/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/win/mingw32/config/aom_config.h b/media/libaom/config/win/mingw32/config/aom_config.h
index 090a455778..394c9d574e 100644
--- a/media/libaom/config/win/mingw32/config/aom_config.h
+++ b/media/libaom/config/win/mingw32/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/win/mingw32/config/aom_dsp_rtcd.h b/media/libaom/config/win/mingw32/config/aom_dsp_rtcd.h
index 5b3762f109..a60a7aea6e 100644
--- a/media/libaom/config/win/mingw32/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/win/mingw32/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 RTCD_EXTERN void (*aom_h_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -487,10 +489,12 @@ void aom_highbd_blend_a64_vmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const
 RTCD_EXTERN void (*aom_highbd_blend_a64_vmask)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 
 void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_horiz)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
 void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
+void aom_highbd_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 void aom_highbd_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 RTCD_EXTERN void (*aom_highbd_convolve8_vert)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps);
 
@@ -1709,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 RTCD_EXTERN void (*aom_v_predictor_8x8)(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1955,6 +1955,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_h_predictor_8x4 = aom_h_predictor_8x4_sse2;
     aom_h_predictor_8x8 = aom_h_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_h_predictor_8x8 = aom_h_predictor_8x8_sse2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1962,8 +1965,10 @@ static void setup_rtcd_internal(void)
     aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_vmask = aom_highbd_blend_a64_vmask_sse4_1;
     aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_horiz = aom_highbd_convolve8_horiz_avx2;
     aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_c;
+    if (flags & HAS_SSE2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_sse2;
     if (flags & HAS_AVX2) aom_highbd_convolve8_vert = aom_highbd_convolve8_vert_avx2;
     aom_highbd_convolve_copy = aom_highbd_convolve_copy_c;
     if (flags & HAS_SSE2) aom_highbd_convolve_copy = aom_highbd_convolve_copy_sse2;
@@ -2367,8 +2372,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_SSE2) aom_v_predictor_8x4 = aom_v_predictor_8x4_sse2;
     aom_v_predictor_8x8 = aom_v_predictor_8x8_c;
     if (flags & HAS_SSE2) aom_v_predictor_8x8 = aom_v_predictor_8x8_sse2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/win/mingw32/config/aom_scale_rtcd.h b/media/libaom/config/win/mingw32/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/win/mingw32/config/aom_scale_rtcd.h
+++ b/media/libaom/config/win/mingw32/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/win/mingw32/config/av1_rtcd.h b/media/libaom/config/win/mingw32/config/av1_rtcd.h
index a2a6f14ff5..8e69b643ef 100644
--- a/media/libaom/config/win/mingw32/config/av1_rtcd.h
+++ b/media/libaom/config/win/mingw32/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -163,40 +189,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -208,26 +255,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -301,25 +328,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -351,12 +362,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -462,6 +473,25 @@ static void setup_rtcd_internal(void)
     av1_convolve_y_sr = av1_convolve_y_sr_c;
     if (flags & HAS_SSE2) av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -484,34 +514,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -524,18 +557,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_c;
-    if (flags & HAS_SSE2) av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/win/mingw64/config/aom_config.asm b/media/libaom/config/win/mingw64/config/aom_config.asm
index 4ab2dedb42..294d0294b6 100644
--- a/media/libaom/config/win/mingw64/config/aom_config.asm
+++ b/media/libaom/config/win/mingw64/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/win/mingw64/config/aom_config.h b/media/libaom/config/win/mingw64/config/aom_config.h
index 9a0be7cd11..ffff5abe6a 100644
--- a/media/libaom/config/win/mingw64/config/aom_config.h
+++ b/media/libaom/config/win/mingw64/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/win/mingw64/config/aom_dsp_rtcd.h b/media/libaom/config/win/mingw64/config/aom_dsp_rtcd.h
index 2856d4ede7..5826faff95 100644
--- a/media/libaom/config/win/mingw64/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/win/mingw64/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 #define aom_h_predictor_8x8 aom_h_predictor_8x8_sse2
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -1711,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 #define aom_v_predictor_8x8 aom_v_predictor_8x8_sse2
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1788,6 +1786,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_dc_top_predictor_64x64 = aom_dc_top_predictor_64x64_avx2;
     aom_h_predictor_32x32 = aom_h_predictor_32x32_sse2;
     if (flags & HAS_AVX2) aom_h_predictor_32x32 = aom_h_predictor_32x32_avx2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1989,8 +1990,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_v_predictor_64x32 = aom_v_predictor_64x32_avx2;
     aom_v_predictor_64x64 = aom_v_predictor_64x64_sse2;
     if (flags & HAS_AVX2) aom_v_predictor_64x64 = aom_v_predictor_64x64_avx2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/win/mingw64/config/aom_scale_rtcd.h b/media/libaom/config/win/mingw64/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/win/mingw64/config/aom_scale_rtcd.h
+++ b/media/libaom/config/win/mingw64/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/win/mingw64/config/av1_rtcd.h b/media/libaom/config/win/mingw64/config/av1_rtcd.h
index d273182086..80bacb8dd5 100644
--- a/media/libaom/config/win/mingw64/config/av1_rtcd.h
+++ b/media/libaom/config/win/mingw64/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -166,40 +192,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -211,26 +258,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -304,25 +331,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -354,12 +365,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -461,6 +472,21 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) av1_convolve_x_sr = av1_convolve_x_sr_avx2;
     av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -482,34 +508,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -522,15 +551,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/config/win/x64/config/aom_config.asm b/media/libaom/config/win/x64/config/aom_config.asm
index 4ab2dedb42..294d0294b6 100644
--- a/media/libaom/config/win/x64/config/aom_config.asm
+++ b/media/libaom/config/win/x64/config/aom_config.asm
@@ -1,5 +1,5 @@
 ;
-; Copyright (c) 2018, Alliance for Open Media. All rights reserved
+; Copyright (c) 2019, Alliance for Open Media. All rights reserved
 ;
 ; This source code is subject to the terms of the BSD 2 Clause License and
 ; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -22,18 +22,17 @@ CONFIG_AV1_ENCODER equ 0
 CONFIG_BIG_ENDIAN equ 0
 CONFIG_BITSTREAM_DEBUG equ 0
 CONFIG_COEFFICIENT_RANGE_CHECKING equ 0
-CONFIG_COLLECT_INTER_MODE_RD_STATS equ 0
+CONFIG_COLLECT_PARTITION_STATS equ 0
 CONFIG_COLLECT_RD_STATS equ 0
 CONFIG_DEBUG equ 0
 CONFIG_DENOISE equ 1
+CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 equ 1
 CONFIG_DIST_8X8 equ 0
 CONFIG_ENTROPY_STATS equ 0
 CONFIG_FILEOPTIONS equ 1
-CONFIG_FIX_GF_LENGTH equ 1
 CONFIG_FP_MB_STATS equ 0
 CONFIG_GCC equ 1
 CONFIG_GCOV equ 0
-CONFIG_GLOBAL_MOTION_SEARCH equ 1
 CONFIG_GPROF equ 0
 CONFIG_INSPECTION equ 0
 CONFIG_INTERNAL_STATS equ 0
@@ -44,15 +43,16 @@ CONFIG_MAX_DECODE_PROFILE equ 2
 CONFIG_MISMATCH_DEBUG equ 0
 CONFIG_MULTITHREAD equ 1
 CONFIG_NORMAL_TILE_MODE equ 0
+CONFIG_ONE_PASS_SVM equ 0
 CONFIG_OS_SUPPORT equ 1
 CONFIG_PIC equ 0
 CONFIG_RD_DEBUG equ 0
-CONFIG_REDUCED_ENCODER_BORDER equ 0
 CONFIG_RUNTIME_CPU_DETECT equ 1
 CONFIG_SHARED equ 0
 CONFIG_SHARP_SETTINGS equ 0
 CONFIG_SIZE_LIMIT equ 0
 CONFIG_SPATIAL_RESAMPLING equ 1
+CONFIG_SPEED_STATS equ 0
 CONFIG_STATIC equ 1
 CONFIG_WEBM_IO equ 0
 DECODE_HEIGHT_LIMIT equ 0
diff --git a/media/libaom/config/win/x64/config/aom_config.h b/media/libaom/config/win/x64/config/aom_config.h
index 9a0be7cd11..ffff5abe6a 100644
--- a/media/libaom/config/win/x64/config/aom_config.h
+++ b/media/libaom/config/win/x64/config/aom_config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
@@ -24,18 +24,17 @@
 #define CONFIG_BIG_ENDIAN 0
 #define CONFIG_BITSTREAM_DEBUG 0
 #define CONFIG_COEFFICIENT_RANGE_CHECKING 0
-#define CONFIG_COLLECT_INTER_MODE_RD_STATS 0
+#define CONFIG_COLLECT_PARTITION_STATS 0
 #define CONFIG_COLLECT_RD_STATS 0
 #define CONFIG_DEBUG 0
 #define CONFIG_DENOISE 1
+#define CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1
 #define CONFIG_DIST_8X8 0
 #define CONFIG_ENTROPY_STATS 0
 #define CONFIG_FILEOPTIONS 1
-#define CONFIG_FIX_GF_LENGTH 1
 #define CONFIG_FP_MB_STATS 0
 #define CONFIG_GCC 1
 #define CONFIG_GCOV 0
-#define CONFIG_GLOBAL_MOTION_SEARCH 1
 #define CONFIG_GPROF 0
 #define CONFIG_INSPECTION 0
 #define CONFIG_INTERNAL_STATS 0
@@ -46,15 +45,16 @@
 #define CONFIG_MISMATCH_DEBUG 0
 #define CONFIG_MULTITHREAD 1
 #define CONFIG_NORMAL_TILE_MODE 0
+#define CONFIG_ONE_PASS_SVM 0
 #define CONFIG_OS_SUPPORT 1
 #define CONFIG_PIC 0
 #define CONFIG_RD_DEBUG 0
-#define CONFIG_REDUCED_ENCODER_BORDER 0
 #define CONFIG_RUNTIME_CPU_DETECT 1
 #define CONFIG_SHARED 0
 #define CONFIG_SHARP_SETTINGS 0
 #define CONFIG_SIZE_LIMIT 0
 #define CONFIG_SPATIAL_RESAMPLING 1
+#define CONFIG_SPEED_STATS 0
 #define CONFIG_STATIC 1
 #define CONFIG_WEBM_IO 0
 #define DECODE_HEIGHT_LIMIT 0
@@ -76,7 +76,5 @@
 #define HAVE_SSSE3 1
 #define HAVE_VSX 0
 #define HAVE_WXWIDGETS 0
-#define INCLUDE_INSTALL_DIR INSTALLDIR/include
 #define INLINE inline
-#define LIB_INSTALL_DIR INSTALLDIR/lib
 #endif /* AOM_CONFIG_H_ */
diff --git a/media/libaom/config/win/x64/config/aom_dsp_rtcd.h b/media/libaom/config/win/x64/config/aom_dsp_rtcd.h
index 2856d4ede7..5826faff95 100644
--- a/media/libaom/config/win/x64/config/aom_dsp_rtcd.h
+++ b/media/libaom/config/win/x64/config/aom_dsp_rtcd.h
@@ -472,7 +472,9 @@ void aom_h_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *a
 #define aom_h_predictor_8x8 aom_h_predictor_8x8_sse2
 
 void aom_highbd_blend_a64_d16_mask_c(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
-#define aom_highbd_blend_a64_d16_mask aom_highbd_blend_a64_d16_mask_c
+void aom_highbd_blend_a64_d16_mask_sse4_1(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+void aom_highbd_blend_a64_d16_mask_avx2(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
+RTCD_EXTERN void (*aom_highbd_blend_a64_d16_mask)(uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd);
 
 void aom_highbd_blend_a64_hmask_c(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
 void aom_highbd_blend_a64_hmask_sse4_1(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd);
@@ -1711,10 +1713,6 @@ void aom_v_predictor_8x8_c(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *abov
 void aom_v_predictor_8x8_sse2(uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left);
 #define aom_v_predictor_8x8 aom_v_predictor_8x8_sse2
 
-void av1_round_shift_array_c(int32_t *arr, int size, int bit);
-void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
-RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
-
 void aom_dsp_rtcd(void);
 
 #ifdef RTCD_C
@@ -1788,6 +1786,9 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_dc_top_predictor_64x64 = aom_dc_top_predictor_64x64_avx2;
     aom_h_predictor_32x32 = aom_h_predictor_32x32_sse2;
     if (flags & HAS_AVX2) aom_h_predictor_32x32 = aom_h_predictor_32x32_avx2;
+    aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_c;
+    if (flags & HAS_SSE4_1) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_sse4_1;
+    if (flags & HAS_AVX2) aom_highbd_blend_a64_d16_mask = aom_highbd_blend_a64_d16_mask_avx2;
     aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_c;
     if (flags & HAS_SSE4_1) aom_highbd_blend_a64_hmask = aom_highbd_blend_a64_hmask_sse4_1;
     aom_highbd_blend_a64_mask = aom_highbd_blend_a64_mask_c;
@@ -1989,8 +1990,6 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) aom_v_predictor_64x32 = aom_v_predictor_64x32_avx2;
     aom_v_predictor_64x64 = aom_v_predictor_64x64_sse2;
     if (flags & HAS_AVX2) aom_v_predictor_64x64 = aom_v_predictor_64x64_avx2;
-    av1_round_shift_array = av1_round_shift_array_c;
-    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
 }
 #endif
 
diff --git a/media/libaom/config/win/x64/config/aom_scale_rtcd.h b/media/libaom/config/win/x64/config/aom_scale_rtcd.h
index 6b56795cd4..b6e814905b 100644
--- a/media/libaom/config/win/x64/config/aom_scale_rtcd.h
+++ b/media/libaom/config/win/x64/config/aom_scale_rtcd.h
@@ -59,13 +59,22 @@ void aom_yv12_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buf
 void aom_yv12_extend_frame_borders_c(struct yv12_buffer_config *ybf, const int num_planes);
 #define aom_yv12_extend_frame_borders aom_yv12_extend_frame_borders_c
 
-void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_coloc_copy_u_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_u aom_yv12_partial_coloc_copy_u_c
+
+void aom_yv12_partial_coloc_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_v aom_yv12_partial_coloc_copy_v_c
+
+void aom_yv12_partial_coloc_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+#define aom_yv12_partial_coloc_copy_y aom_yv12_partial_coloc_copy_y_c
+
+void aom_yv12_partial_copy_u_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_u aom_yv12_partial_copy_u_c
 
-void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_v_c(const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_v aom_yv12_partial_copy_v_c
 
-void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend);
+void aom_yv12_partial_copy_y_c(const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2);
 #define aom_yv12_partial_copy_y aom_yv12_partial_copy_y_c
 
 void aom_scale_rtcd(void);
diff --git a/media/libaom/config/win/x64/config/av1_rtcd.h b/media/libaom/config/win/x64/config/av1_rtcd.h
index d273182086..80bacb8dd5 100644
--- a/media/libaom/config/win/x64/config/av1_rtcd.h
+++ b/media/libaom/config/win/x64/config/av1_rtcd.h
@@ -31,6 +31,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -99,14 +101,38 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst, in
 void av1_convolve_y_sr_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_convolve_y_sr)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
 
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+RTCD_EXTERN void (*av1_dist_wtd_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+
 void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
-#define av1_dr_prediction_z1 av1_dr_prediction_z1_c
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z1)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy);
 
 void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z2 av1_dr_prediction_z2_c
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z2)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy);
 
 void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
-#define av1_dr_prediction_z3 av1_dr_prediction_z3_c
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
+RTCD_EXTERN void (*av1_dr_prediction_z3)(uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy);
 
 void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength);
 void av1_filter_intra_edge_sse4_1(uint8_t *p, int sz, int strength);
@@ -166,40 +192,61 @@ void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride, uint16_
 void av1_highbd_convolve_y_sr_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 RTCD_EXTERN void (*av1_highbd_convolve_y_sr)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
 
+void av1_highbd_dist_wtd_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+void av1_highbd_dist_wtd_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+void av1_highbd_dist_wtd_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+RTCD_EXTERN void (*av1_highbd_dist_wtd_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
 void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z1 av1_highbd_dr_prediction_z1_c
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z1)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd);
 
 void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd);
 #define av1_highbd_dr_prediction_z2 av1_highbd_dr_prediction_z2_c
 
 void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
-#define av1_highbd_dr_prediction_z3 av1_highbd_dr_prediction_z3_c
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
+RTCD_EXTERN void (*av1_highbd_dr_prediction_z3)(uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd);
 
 void av1_highbd_inv_txfm_add_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_16x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_16x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-
-void av1_highbd_inv_txfm_add_32x32_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_32x32)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_highbd_inv_txfm_add_8x16_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
-RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x16)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_4x8)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
+RTCD_EXTERN void (*av1_highbd_inv_txfm_add_8x4)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
 void av1_highbd_inv_txfm_add_8x8_c(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
@@ -211,26 +258,6 @@ void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int des
 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int dest_stride, int bd);
 #define av1_highbd_iwht4x4_1_add av1_highbd_iwht4x4_1_add_c
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_2d_copy_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_2d_copy_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_2d_copy)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_x_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_x)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_sse4_1(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-void av1_highbd_jnt_convolve_y_avx2(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-RTCD_EXTERN void (*av1_highbd_jnt_convolve_y)(const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd);
-
 void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
 RTCD_EXTERN void (*av1_highbd_warp_affine)(const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta);
@@ -304,25 +331,9 @@ void av1_inv_txfm_add_ssse3(const tran_low_t *dqcoeff, uint8_t *dst, int stride,
 void av1_inv_txfm_add_avx2(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 RTCD_EXTERN void (*av1_inv_txfm_add)(const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param);
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_2d_copy)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_x)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
-RTCD_EXTERN void (*av1_jnt_convolve_y)(const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params);
+void av1_round_shift_array_c(int32_t *arr, int size, int bit);
+void av1_round_shift_array_sse4_1(int32_t *arr, int size, int bit);
+RTCD_EXTERN void (*av1_round_shift_array)(int32_t *arr, int size, int bit);
 
 int av1_selfguided_restoration_c(const uint8_t *dgd8, int width, int height,
                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
@@ -354,12 +365,12 @@ void av1_wiener_convolve_add_src_sse2(const uint8_t *src, ptrdiff_t src_stride,
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 RTCD_EXTERN void (*av1_wiener_convolve_add_src)(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params);
 
-void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
-RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift);
+void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_ssse3(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_sse4_1(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+void cdef_filter_block_avx2(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
+RTCD_EXTERN void (*cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift);
 
 int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
 int cdef_find_dir_sse2(const uint16_t *img, int stride, int32_t *var, int coeff_shift);
@@ -461,6 +472,21 @@ static void setup_rtcd_internal(void)
     if (flags & HAS_AVX2) av1_convolve_x_sr = av1_convolve_x_sr_avx2;
     av1_convolve_y_sr = av1_convolve_y_sr_sse2;
     if (flags & HAS_AVX2) av1_convolve_y_sr = av1_convolve_y_sr_avx2;
+    av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_sse2;
+    if (flags & HAS_SSSE3) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_ssse3;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d = av1_dist_wtd_convolve_2d_avx2;
+    av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_2d_copy = av1_dist_wtd_convolve_2d_copy_avx2;
+    av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_x = av1_dist_wtd_convolve_x_avx2;
+    av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_sse2;
+    if (flags & HAS_AVX2) av1_dist_wtd_convolve_y = av1_dist_wtd_convolve_y_avx2;
+    av1_dr_prediction_z1 = av1_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z1 = av1_dr_prediction_z1_avx2;
+    av1_dr_prediction_z2 = av1_dr_prediction_z2_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z2 = av1_dr_prediction_z2_avx2;
+    av1_dr_prediction_z3 = av1_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_dr_prediction_z3 = av1_dr_prediction_z3_avx2;
     av1_filter_intra_edge = av1_filter_intra_edge_c;
     if (flags & HAS_SSE4_1) av1_filter_intra_edge = av1_filter_intra_edge_sse4_1;
     av1_filter_intra_edge_high = av1_filter_intra_edge_high_c;
@@ -482,34 +508,37 @@ static void setup_rtcd_internal(void)
     av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_c;
     if (flags & HAS_SSSE3) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_ssse3;
     if (flags & HAS_AVX2) av1_highbd_convolve_y_sr = av1_highbd_convolve_y_sr_avx2;
+    av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d = av1_highbd_dist_wtd_convolve_2d_avx2;
+    av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_2d_copy = av1_highbd_dist_wtd_convolve_2d_copy_avx2;
+    av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_x = av1_highbd_dist_wtd_convolve_x_avx2;
+    av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_c;
+    if (flags & HAS_SSE4_1) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_sse4_1;
+    if (flags & HAS_AVX2) av1_highbd_dist_wtd_convolve_y = av1_highbd_dist_wtd_convolve_y_avx2;
+    av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z1 = av1_highbd_dr_prediction_z1_avx2;
+    av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_c;
+    if (flags & HAS_AVX2) av1_highbd_dr_prediction_z3 = av1_highbd_dr_prediction_z3_avx2;
     av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_sse4_1;
     if (flags & HAS_AVX2) av1_highbd_inv_txfm_add = av1_highbd_inv_txfm_add_avx2;
-    av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x16 = av1_highbd_inv_txfm_add_16x16_sse4_1;
-    av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x8 = av1_highbd_inv_txfm_add_16x8_sse4_1;
-    av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_inv_txfm_add_32x32 = av1_highbd_inv_txfm_add_32x32_avx2;
+    av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_16x4 = av1_highbd_inv_txfm_add_16x4_sse4_1;
+    av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x16 = av1_highbd_inv_txfm_add_4x16_sse4_1;
     av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x4 = av1_highbd_inv_txfm_add_4x4_sse4_1;
-    av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_c;
-    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x16 = av1_highbd_inv_txfm_add_8x16_sse4_1;
+    av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_4x8 = av1_highbd_inv_txfm_add_4x8_sse4_1;
+    av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_c;
+    if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x4 = av1_highbd_inv_txfm_add_8x4_sse4_1;
     av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_c;
     if (flags & HAS_SSE4_1) av1_highbd_inv_txfm_add_8x8 = av1_highbd_inv_txfm_add_8x8_sse4_1;
-    av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d = av1_highbd_jnt_convolve_2d_avx2;
-    av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_2d_copy = av1_highbd_jnt_convolve_2d_copy_avx2;
-    av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_x = av1_highbd_jnt_convolve_x_avx2;
-    av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_c;
-    if (flags & HAS_SSE4_1) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_sse4_1;
-    if (flags & HAS_AVX2) av1_highbd_jnt_convolve_y = av1_highbd_jnt_convolve_y_avx2;
     av1_highbd_warp_affine = av1_highbd_warp_affine_c;
     if (flags & HAS_SSE4_1) av1_highbd_warp_affine = av1_highbd_warp_affine_sse4_1;
     av1_highbd_wiener_convolve_add_src = av1_highbd_wiener_convolve_add_src_c;
@@ -522,15 +551,8 @@ static void setup_rtcd_internal(void)
     av1_inv_txfm_add = av1_inv_txfm_add_c;
     if (flags & HAS_SSSE3) av1_inv_txfm_add = av1_inv_txfm_add_ssse3;
     if (flags & HAS_AVX2) av1_inv_txfm_add = av1_inv_txfm_add_avx2;
-    av1_jnt_convolve_2d = av1_jnt_convolve_2d_c;
-    if (flags & HAS_SSSE3) av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
-    av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
-    av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
-    av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
-    if (flags & HAS_AVX2) av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
+    av1_round_shift_array = av1_round_shift_array_c;
+    if (flags & HAS_SSE4_1) av1_round_shift_array = av1_round_shift_array_sse4_1;
     av1_selfguided_restoration = av1_selfguided_restoration_c;
     if (flags & HAS_SSE4_1) av1_selfguided_restoration = av1_selfguided_restoration_sse4_1;
     if (flags & HAS_AVX2) av1_selfguided_restoration = av1_selfguided_restoration_avx2;
diff --git a/media/libaom/sources.mozbuild b/media/libaom/sources.mozbuild
index 9aff837c49..b939b54328 100644
--- a/media/libaom/sources.mozbuild
+++ b/media/libaom/sources.mozbuild
@@ -1,4 +1,4 @@
-# This file is automatically generated. Do not edit.
+# This file is generated. Do not edit.
 
 files = {
   'ARM_EXPORTS': [
@@ -34,6 +34,7 @@ files = {
     '../../third_party/aom/aom_dsp/arm/intrapred_neon.c',
     '../../third_party/aom/aom_dsp/arm/loopfilter_neon.c',
     '../../third_party/aom/aom_dsp/arm/subtract_neon.c',
+    '../../third_party/aom/aom_dsp/avg.c',
     '../../third_party/aom/aom_dsp/binary_codes_reader.c',
     '../../third_party/aom/aom_dsp/bitreader_buffer.c',
     '../../third_party/aom/aom_dsp/bitwriter_buffer.c',
@@ -109,7 +110,6 @@ files = {
     '../../third_party/aom/av1/decoder/decoder.c',
     '../../third_party/aom/av1/decoder/decodetxb.c',
     '../../third_party/aom/av1/decoder/detokenize.c',
-    '../../third_party/aom/av1/decoder/dthread.c',
     '../../third_party/aom/av1/decoder/obu.c',
     '../../third_party/aom/av1/encoder/arm/neon/quantize_neon.c',
   ],
@@ -140,6 +140,7 @@ files = {
     '../../third_party/aom/aom/src/aom_integer.c',
     '../../third_party/aom/aom_dsp/aom_convolve.c',
     '../../third_party/aom/aom_dsp/aom_dsp_rtcd.c',
+    '../../third_party/aom/aom_dsp/avg.c',
     '../../third_party/aom/aom_dsp/binary_codes_reader.c',
     '../../third_party/aom/aom_dsp/bitreader_buffer.c',
     '../../third_party/aom/aom_dsp/bitwriter_buffer.c',
@@ -202,7 +203,6 @@ files = {
     '../../third_party/aom/av1/decoder/decoder.c',
     '../../third_party/aom/av1/decoder/decodetxb.c',
     '../../third_party/aom/av1/decoder/detokenize.c',
-    '../../third_party/aom/av1/decoder/dthread.c',
     '../../third_party/aom/av1/decoder/obu.c',
   ],
   'IA32_EXPORTS': [
@@ -232,6 +232,7 @@ files = {
     '../../third_party/aom/aom/src/aom_integer.c',
     '../../third_party/aom/aom_dsp/aom_convolve.c',
     '../../third_party/aom/aom_dsp/aom_dsp_rtcd.c',
+    '../../third_party/aom/aom_dsp/avg.c',
     '../../third_party/aom/aom_dsp/binary_codes_reader.c',
     '../../third_party/aom/aom_dsp/bitreader_buffer.c',
     '../../third_party/aom/aom_dsp/bitwriter_buffer.c',
@@ -251,11 +252,14 @@ files = {
     '../../third_party/aom/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_ssse3.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm',
+    '../../third_party/aom/aom_dsp/x86/avg_intrin_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/avg_intrin_sse2.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_hmask_sse4.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c',
@@ -263,14 +267,15 @@ files = {
     '../../third_party/aom/aom_dsp/x86/fft_avx2.c',
     '../../third_party/aom/aom_dsp/x86/fft_sse2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/highbd_convolve_sse2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_convolve_ssse3.c',
+    '../../third_party/aom/aom_dsp/x86/highbd_intrapred_asm_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/highbd_intrapred_sse2.c',
-    '../../third_party/aom/aom_dsp/x86/highbd_intrapred_sse2_asm.asm',
     '../../third_party/aom/aom_dsp/x86/highbd_loopfilter_avx2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c',
+    '../../third_party/aom/aom_dsp/x86/intrapred_asm_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/intrapred_avx2.c',
     '../../third_party/aom/aom_dsp/x86/intrapred_sse2.c',
-    '../../third_party/aom/aom_dsp/x86/intrapred_sse2_asm.asm',
     '../../third_party/aom/aom_dsp/x86/intrapred_ssse3.c',
     '../../third_party/aom/aom_dsp/x86/inv_wht_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/loopfilter_sse2.c',
@@ -365,7 +370,6 @@ files = {
     '../../third_party/aom/av1/decoder/decoder.c',
     '../../third_party/aom/av1/decoder/decodetxb.c',
     '../../third_party/aom/av1/decoder/detokenize.c',
-    '../../third_party/aom/av1/decoder/dthread.c',
     '../../third_party/aom/av1/decoder/obu.c',
   ],
   'X64_EXPORTS': [
@@ -395,6 +399,7 @@ files = {
     '../../third_party/aom/aom/src/aom_integer.c',
     '../../third_party/aom/aom_dsp/aom_convolve.c',
     '../../third_party/aom/aom_dsp/aom_dsp_rtcd.c',
+    '../../third_party/aom/aom_dsp/avg.c',
     '../../third_party/aom/aom_dsp/binary_codes_reader.c',
     '../../third_party/aom/aom_dsp/bitreader_buffer.c',
     '../../third_party/aom/aom_dsp/bitwriter_buffer.c',
@@ -414,11 +419,14 @@ files = {
     '../../third_party/aom/aom_dsp/x86/aom_high_subpixel_8t_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_8t_ssse3.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm',
+    '../../third_party/aom/aom_dsp/x86/avg_intrin_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/avg_intrin_sse2.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_hmask_sse4.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c',
     '../../third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c',
@@ -426,14 +434,15 @@ files = {
     '../../third_party/aom/aom_dsp/x86/fft_avx2.c',
     '../../third_party/aom/aom_dsp/x86/fft_sse2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c',
+    '../../third_party/aom/aom_dsp/x86/highbd_convolve_sse2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_convolve_ssse3.c',
+    '../../third_party/aom/aom_dsp/x86/highbd_intrapred_asm_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/highbd_intrapred_sse2.c',
-    '../../third_party/aom/aom_dsp/x86/highbd_intrapred_sse2_asm.asm',
     '../../third_party/aom/aom_dsp/x86/highbd_loopfilter_avx2.c',
     '../../third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c',
+    '../../third_party/aom/aom_dsp/x86/intrapred_asm_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/intrapred_avx2.c',
     '../../third_party/aom/aom_dsp/x86/intrapred_sse2.c',
-    '../../third_party/aom/aom_dsp/x86/intrapred_sse2_asm.asm',
     '../../third_party/aom/aom_dsp/x86/intrapred_ssse3.c',
     '../../third_party/aom/aom_dsp/x86/inv_wht_sse2.asm',
     '../../third_party/aom/aom_dsp/x86/loopfilter_sse2.c',
@@ -527,7 +536,6 @@ files = {
     '../../third_party/aom/av1/decoder/decoder.c',
     '../../third_party/aom/av1/decoder/decodetxb.c',
     '../../third_party/aom/av1/decoder/detokenize.c',
-    '../../third_party/aom/av1/decoder/dthread.c',
     '../../third_party/aom/av1/decoder/obu.c',
   ],
 }
diff --git a/third_party/aom/CMakeLists.txt b/third_party/aom/CMakeLists.txt
index a58e54f40a..f409892d27 100644
--- a/third_party/aom/CMakeLists.txt
+++ b/third_party/aom/CMakeLists.txt
@@ -22,10 +22,6 @@ project(AOM C CXX)
 
 set(AOM_ROOT "${CMAKE_CURRENT_SOURCE_DIR}")
 set(AOM_CONFIG_DIR "${CMAKE_CURRENT_BINARY_DIR}")
-set(INCLUDE_INSTALL_DIR "${CMAKE_INSTALL_PREFIX}/include"
-    CACHE PATH "Installation path of includes")
-set(LIB_INSTALL_DIR "${CMAKE_INSTALL_PREFIX}/lib"
-    CACHE PATH "Installation path of libraries")
 
 if("${AOM_ROOT}" STREQUAL "${AOM_CONFIG_DIR}")
   message(FATAL_ERROR
@@ -42,9 +38,10 @@ include("${AOM_ROOT}/aom_ports/aom_ports.cmake")
 include("${AOM_ROOT}/aom_scale/aom_scale.cmake")
 include("${AOM_ROOT}/aom_util/aom_util.cmake")
 include("${AOM_ROOT}/av1/av1.cmake")
-include("${AOM_ROOT}/test/test.cmake")
+include("${AOM_ROOT}/build/cmake/aom_install.cmake")
 include("${AOM_ROOT}/build/cmake/sanitizers.cmake")
 include("${AOM_ROOT}/build/cmake/util.cmake")
+include("${AOM_ROOT}/test/test.cmake")
 
 list(APPEND AOM_RTCD_SOURCES
             "${AOM_CONFIG_DIR}/config/aom_dsp_rtcd.h"
@@ -158,8 +155,6 @@ list(APPEND AOM_ENCODER_STATS_SOURCES "${AOM_ROOT}/stats/aomstats.c"
             "${AOM_ROOT}/stats/aomstats.h" "${AOM_ROOT}/stats/rate_hist.c"
             "${AOM_ROOT}/stats/rate_hist.h")
 
-list(APPEND AOM_PKG_CONFIG_SOURCES "${AOM_CONFIG_DIR}/aom.pc")
-
 list(APPEND AOM_VERSION_SOURCES "${AOM_CONFIG_DIR}/config/aom_version.h")
 
 list(APPEND AOM_WEBM_DECODER_SOURCES "${AOM_ROOT}/common/webmdec.cc"
@@ -192,22 +187,6 @@ add_custom_target(aom_version_check
                   COMMENT "Updating version info if necessary." VERBATIM)
 add_dependencies(aom_version aom_version_check)
 
-if(NOT MSVC)
-  add_library(aom_pc ${AOM_PKG_CONFIG_SOURCES})
-  add_dummy_source_file_to_target(aom_pc c)
-  add_custom_command(OUTPUT "${AOM_CONFIG_DIR}/aom.pc"
-                     COMMAND ${CMAKE_COMMAND} ARGS
-                             -DAOM_CONFIG_DIR=${AOM_CONFIG_DIR}
-                             -DAOM_ROOT=${AOM_ROOT}
-                             -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}
-                             -DCMAKE_PROJECT_NAME=${CMAKE_PROJECT_NAME}
-                             -DCONFIG_MULTITHREAD=${CONFIG_MULTITHREAD}
-                             -DHAVE_PTHREAD_H=${HAVE_PTHREAD_H} -P
-                             "${AOM_ROOT}/build/cmake/pkg_config.cmake"
-                     COMMENT "Writing aom.pc" VERBATIM)
-  add_dependencies(aom_pc aom_version)
-endif()
-
 # TODO(tomfinegan): Move rtcd target setup where it belongs for each rtcd
 # source.
 add_rtcd_build_step("${AOM_ROOT}/aom_dsp/aom_dsp_rtcd_defs.pl"
@@ -234,8 +213,7 @@ if(NOT MSVC AND NOT APPLE)
 endif()
 
 # List of object and static library targets.
-set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_rtcd aom_encoder_stats aom_mem
-    aom_scale aom)
+set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_rtcd aom_mem aom_scale aom)
 
 # Setup dependencies.
 setup_aom_dsp_targets()
@@ -403,7 +381,6 @@ endif()
 
 if(ENABLE_TOOLS)
   if(CONFIG_AV1_DECODER)
-    require_cxx_flag_nomsvc("-std=c++11" NO)
     add_executable(dump_obu "${AOM_GEN_SRC_DIR}/usage_exit.cc"
                    "${AOM_ROOT}/tools/dump_obu.cc"
                    "${AOM_ROOT}/tools/obu_parser.cc"
@@ -414,10 +391,6 @@ if(ENABLE_TOOLS)
     list(APPEND AOM_TOOL_TARGETS dump_obu)
     list(APPEND AOM_APP_TARGETS dump_obu)
 
-    if(NOT MSVC)
-      target_compile_options(dump_obu PUBLIC -std=c++11)
-    endif()
-
     # Maintain a separate variable listing only the examples to facilitate
     # installation of example programs into an tools sub directory of
     # $AOM_DIST_DIR/bin when building the dist target.
@@ -489,8 +462,6 @@ if(ENABLE_EXAMPLES OR ENABLE_TESTS OR ENABLE_TOOLS)
   endif()
 
   if(CONFIG_WEBM_IO)
-    require_cxx_flag_nomsvc("-std=c++11" NO)
-
     add_library(webm OBJECT ${AOM_LIBWEBM_SOURCES})
     include_directories("${AOM_ROOT}/third_party/libwebm")
     target_compile_definitions(webm PRIVATE __STDC_CONSTANT_MACROS)
@@ -498,7 +469,6 @@ if(ENABLE_EXAMPLES OR ENABLE_TESTS OR ENABLE_TOOLS)
 
     if(NOT MSVC)
       target_compile_options(webm PRIVATE -Wno-shadow)
-      target_compile_options(webm PUBLIC -std=c++11)
     endif()
 
     # Add to existing targets.
@@ -593,46 +563,6 @@ if(ENABLE_DOCS)
   endif()
 endif()
 
-if(NOT (MSVC OR XCODE))
-
-  # Aomedia install rule.
-  list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aom.h"
-              "${AOM_ROOT}/aom/aom_codec.h"
-              "${AOM_ROOT}/aom/aom_frame_buffer.h"
-              "${AOM_ROOT}/aom/aom_image.h" "${AOM_ROOT}/aom/aom_integer.h"
-              "${AOM_ROOT}/aom/aom.h")
-
-  if(CONFIG_AV1_DECODER)
-    if(ENABLE_EXAMPLES)
-      list(APPEND AOM_INSTALL_BINS aomdec)
-    endif()
-
-    list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aom_decoder.h"
-                "${AOM_ROOT}/aom/aomdx.h")
-  endif()
-
-  if(CONFIG_AV1_ENCODER)
-    if(ENABLE_EXAMPLES)
-      list(APPEND AOM_INSTALL_BINS aomenc)
-    endif()
-
-    list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aomcx.h"
-                "${AOM_ROOT}/aom/aom_encoder.h")
-  endif()
-
-  set(AOM_INSTALL_LIBS aom)
-
-  install(FILES ${AOM_INSTALL_INCS} DESTINATION "${INCLUDE_INSTALL_DIR}/aom")
-  install(FILES "${AOM_CONFIG_DIR}/aom.pc" DESTINATION
-                "${LIB_INSTALL_DIR}/pkgconfig")
-  install(TARGETS ${AOM_INSTALL_LIBS} DESTINATION "${LIB_INSTALL_DIR}")
-
-  if(ENABLE_EXAMPLES)
-    install(TARGETS ${AOM_INSTALL_BINS} DESTINATION
-                    "${CMAKE_INSTALL_PREFIX}/bin")
-  endif()
-endif()
-
 # Aomedia dist rule.
 if(CONFIG_AV1_DECODER AND ENABLE_EXAMPLES)
   list(APPEND AOM_DIST_APPS $<TARGET_FILE:aomdec>)
@@ -756,3 +686,6 @@ foreach(aom_source_var ${aom_source_vars})
     file(APPEND "${libaom_srcs_gni_file}" "]\n")
   endif()
 endforeach()
+
+# Generate aom.pc and setup install rule.
+setup_aom_install_targets()
diff --git a/third_party/aom/PATENTS b/third_party/aom/PATENTS
index 97842e02f1..493f616370 100644
--- a/third_party/aom/PATENTS
+++ b/third_party/aom/PATENTS
@@ -57,10 +57,10 @@ Alliance for Open Media Patent License 1.0
 
 2. Definitions.
 
-2.1. Affiliate.  �Affiliate� means an entity that directly or indirectly
+2.1. Affiliate.  "Affiliate" means an entity that directly or indirectly
      Controls, is Controlled by, or is under common Control of that party.
 
-2.2. Control. �Control� means direct or indirect control of more than 50% of
+2.2. Control. "Control" means direct or indirect control of more than 50% of
      the voting power to elect directors of that corporation, or for any other
      entity, the power to direct management of such entity.
 
@@ -70,7 +70,7 @@ Alliance for Open Media Patent License 1.0
 2.4. Encoder.  "Encoder" means any encoder that produces a bitstream that can
      be decoded by a Decoder only to the extent it produces such a bitstream.
 
-2.5. Final Deliverable.  �Final Deliverable� means the final version of a
+2.5. Final Deliverable.  "Final Deliverable" means the final version of a
      deliverable approved by the Alliance for Open Media as a Final
      Deliverable.
 
@@ -79,9 +79,9 @@ Alliance for Open Media Patent License 1.0
      Implementation also includes components of an Implementation only to the
      extent they are used as part of an Implementation.
 
-2.7. License. �License� means this license.
+2.7. License. "License" means this license.
 
-2.8. Licensee. �Licensee� means any person or entity who exercises patent
+2.8. Licensee. "Licensee" means any person or entity who exercises patent
      rights granted under this License.
 
 2.9. Licensor.  "Licensor" means (i) any Licensee that makes, sells, offers
@@ -98,11 +98,11 @@ Alliance for Open Media Patent License 1.0
       as if the Specification was a W3C Recommendation; or (ii) are infringed
       by the Reference Implementation.
 
-2.11. Reference Implementation. �Reference Implementation� means an Encoder
+2.11. Reference Implementation. "Reference Implementation" means an Encoder
       and/or Decoder released by the Alliance for Open Media as a Final
       Deliverable.
 
-2.12. Specification. �Specification� means the specification designated by
+2.12. Specification. "Specification" means the specification designated by
       the Alliance for Open Media as a Final Deliverable for which this
       License was issued.
 
diff --git a/third_party/aom/README.md b/third_party/aom/README.md
index cab3f9993f..6b58d35408 100644
--- a/third_party/aom/README.md
+++ b/third_party/aom/README.md
@@ -210,15 +210,18 @@ compiler documentation to determine which, if any, are available.
 
 ### Microsoft Visual Studio builds
 
-Building the AV1 codec library in Microsoft Visual Studio is supported. The
-following example demonstrates generating projects and a solution for the
-Microsoft IDE:
+Building the AV1 codec library in Microsoft Visual Studio is supported. Visual
+Studio 2015 (14.0) or later is required. The following example demonstrates
+generating projects and a solution for the Microsoft IDE:
 
 ~~~
     # This does not require a bash shell; command.exe is fine.
     $ cmake path/to/aom -G "Visual Studio 15 2017"
 ~~~
 
+NOTE: The build system targets Windows 7 or later by compiling files with
+`-D_WIN32_WINNT=0x0601`.
+
 ### Xcode builds
 
 Building the AV1 codec library in Xcode is supported. The following example
diff --git a/third_party/aom/aom/aom_decoder.h b/third_party/aom/aom/aom_decoder.h
index 06c2dc5f7b..70420c39e7 100644
--- a/third_party/aom/aom/aom_decoder.h
+++ b/third_party/aom/aom/aom_decoder.h
@@ -55,8 +55,6 @@ extern "C" {
 #define AOM_CODEC_CAP_PUT_SLICE 0x10000 /**< Will issue put_slice callbacks */
 #define AOM_CODEC_CAP_PUT_FRAME 0x20000 /**< Will issue put_frame callbacks */
 #define AOM_CODEC_CAP_POSTPROC 0x40000  /**< Can postprocess decoded frame */
-/*!\brief Can receive encoded frames one fragment at a time */
-#define AOM_CODEC_CAP_INPUT_FRAGMENTS 0x100000
 
 /*! \brief Initialization-time Feature Enabling
  *
@@ -65,15 +63,10 @@ extern "C" {
  *
  *  The available flags are specified by AOM_CODEC_USE_* defines.
  */
-/*!\brief Can support frame-based multi-threading */
-#define AOM_CODEC_CAP_FRAME_THREADING 0x200000
 /*!brief Can support external frame buffers */
-#define AOM_CODEC_CAP_EXTERNAL_FRAME_BUFFER 0x400000
+#define AOM_CODEC_CAP_EXTERNAL_FRAME_BUFFER 0x200000
 
 #define AOM_CODEC_USE_POSTPROC 0x10000 /**< Postprocess decoded frame */
-/*!\brief The input frame should be passed to the decoder one fragment at a
- * time */
-#define AOM_CODEC_USE_INPUT_FRAGMENTS 0x40000
 
 /*!\brief Stream properties
  *
@@ -191,13 +184,6 @@ aom_codec_err_t aom_codec_get_stream_info(aom_codec_ctx_t *ctx,
  * generated, as appropriate. Encoded data \ref MUST be passed in DTS (decode
  * time stamp) order. Frames produced will always be in PTS (presentation
  * time stamp) order.
- * If the decoder is configured with AOM_CODEC_USE_INPUT_FRAGMENTS enabled,
- * data and data_sz can contain a fragment of the encoded frame. Fragment
- * \#n must contain at least partition \#n, but can also contain subsequent
- * partitions (\#n+1 - \#n+i), and if so, fragments \#n+1, .., \#n+i must
- * be empty. When no more data is available, this function should be called
- * with NULL as data and 0 as data_sz. The memory passed to this function
- * must be available until the frame has been decoded.
  *
  * \param[in] ctx          Pointer to this instance's context
  * \param[in] data         Pointer to this block of new coded data. If
diff --git a/third_party/aom/aom/aom_encoder.h b/third_party/aom/aom/aom_encoder.h
index 0894ca9e31..777236f8a2 100644
--- a/third_party/aom/aom/aom_encoder.h
+++ b/third_party/aom/aom/aom_encoder.h
@@ -97,10 +97,14 @@ typedef uint32_t aom_codec_frame_flags_t;
 /*!\brief frame can be dropped without affecting the stream (no future frame
  * depends on this one) */
 #define AOM_FRAME_IS_DROPPABLE 0x2
-/*!\brief frame should be decoded but will not be shown */
-#define AOM_FRAME_IS_INVISIBLE 0x4
-/*!\brief this is a fragment of the encoded frame */
-#define AOM_FRAME_IS_FRAGMENT 0x8
+/*!\brief this is an INTRA_ONLY frame */
+#define AOM_FRAME_IS_INTRAONLY 0x10
+/*!\brief this is an S-frame */
+#define AOM_FRAME_IS_SWITCH 0x20
+/*!\brief this is an error-resilient frame */
+#define AOM_FRAME_IS_ERROR_RESILIENT 0x40
+/*!\brief this is a key-frame dependent recovery-point frame */
+#define AOM_FRAME_IS_DELAYED_RANDOM_ACCESS_POINT 0x80
 
 /*!\brief Error Resilient flags
  *
diff --git a/third_party/aom/aom/aom_frame_buffer.h b/third_party/aom/aom/aom_frame_buffer.h
index fba4322f85..a715645a7a 100644
--- a/third_party/aom/aom/aom_frame_buffer.h
+++ b/third_party/aom/aom/aom_frame_buffer.h
@@ -53,9 +53,9 @@ typedef struct aom_codec_frame_buffer {
  * data. The callback is triggered when the decoder needs a frame buffer to
  * decode a compressed image into. This function may be called more than once
  * for every call to aom_codec_decode. The application may set fb->priv to
- * some data which will be passed back in the ximage and the release function
- * call. |fb| is guaranteed to not be NULL. On success the callback must
- * return 0. Any failure the callback must return a value less than 0.
+ * some data which will be passed back in the aom_image_t and the release
+ * function call. |fb| is guaranteed to not be NULL. On success the callback
+ * must return 0. Any failure the callback must return a value less than 0.
  *
  * \param[in] priv         Callback's private data
  * \param[in] new_size     Size in bytes needed by the buffer
diff --git a/third_party/aom/aom/aom_image.h b/third_party/aom/aom/aom_image.h
index a960127f15..245ef2c692 100644
--- a/third_party/aom/aom/aom_image.h
+++ b/third_party/aom/aom/aom_image.h
@@ -30,11 +30,11 @@ extern "C" {
  * types, removing or reassigning enums, adding/removing/rearranging
  * fields to structures
  */
-#define AOM_IMAGE_ABI_VERSION (4) /**<\hideinitializer*/
+#define AOM_IMAGE_ABI_VERSION (5) /**<\hideinitializer*/
 
-#define AOM_IMG_FMT_PLANAR 0x100       /**< Image is a planar format. */
-#define AOM_IMG_FMT_UV_FLIP 0x200      /**< V plane precedes U in memory. */
-#define AOM_IMG_FMT_HAS_ALPHA 0x400    /**< Image has an alpha channel. */
+#define AOM_IMG_FMT_PLANAR 0x100  /**< Image is a planar format. */
+#define AOM_IMG_FMT_UV_FLIP 0x200 /**< V plane precedes U in memory. */
+/** 0x400 used to signal alpha channel, skipping for backwards compatibility. */
 #define AOM_IMG_FMT_HIGHBITDEPTH 0x800 /**< Image uses 16bit framebuffer. */
 
 /*!\brief List of supported image formats */
@@ -48,8 +48,8 @@ typedef enum aom_img_fmt {
   AOM_IMG_FMT_AOMI420 = AOM_IMG_FMT_PLANAR | 4,
   AOM_IMG_FMT_I422 = AOM_IMG_FMT_PLANAR | 5,
   AOM_IMG_FMT_I444 = AOM_IMG_FMT_PLANAR | 6,
-  AOM_IMG_FMT_444A = AOM_IMG_FMT_PLANAR | AOM_IMG_FMT_HAS_ALPHA | 6,
   AOM_IMG_FMT_I42016 = AOM_IMG_FMT_I420 | AOM_IMG_FMT_HIGHBITDEPTH,
+  AOM_IMG_FMT_YV1216 = AOM_IMG_FMT_YV12 | AOM_IMG_FMT_HIGHBITDEPTH,
   AOM_IMG_FMT_I42216 = AOM_IMG_FMT_I422 | AOM_IMG_FMT_HIGHBITDEPTH,
   AOM_IMG_FMT_I44416 = AOM_IMG_FMT_I444 | AOM_IMG_FMT_HIGHBITDEPTH,
 } aom_img_fmt_t; /**< alias for enum aom_img_fmt */
@@ -169,9 +169,8 @@ typedef struct aom_image {
 #define AOM_PLANE_Y 0       /**< Y (Luminance) plane */
 #define AOM_PLANE_U 1       /**< U (Chroma) plane */
 #define AOM_PLANE_V 2       /**< V (Chroma) plane */
-#define AOM_PLANE_ALPHA 3   /**< A (Transparency) plane */
-  unsigned char *planes[4]; /**< pointer to the top left pixel for each plane */
-  int stride[4];            /**< stride between rows for each plane */
+  unsigned char *planes[3]; /**< pointer to the top left pixel for each plane */
+  int stride[3];            /**< stride between rows for each plane */
   size_t sz;                /**< data size */
 
   int bps; /**< bits per sample (for packed formats) */
diff --git a/third_party/aom/aom/aomcx.h b/third_party/aom/aom/aomcx.h
index 013ddf57e8..fc0f23315f 100644
--- a/third_party/aom/aom/aomcx.h
+++ b/third_party/aom/aom/aomcx.h
@@ -300,9 +300,6 @@ enum aome_enc_control_id {
    *             2 = 4 tile columns
    *             .....
    *             n = 2**n tile columns
-   * The requested tile columns will be capped by encoder based on image size
-   * limitation (The minimum width of a tile column is 256 pixel, the maximum
-   * is 4096).
    *
    * By default, the value is 0, i.e. one single column tile for entire image.
    */
@@ -324,6 +321,13 @@ enum aome_enc_control_id {
    */
   AV1E_SET_TILE_ROWS,
 
+  /*!\brief Codec control function to enable RDO modulated by frame temporal
+   * dependency.
+   *
+   * By default, this feature is off.
+   */
+  AV1E_SET_ENABLE_TPL_MODEL,
+
   /*!\brief Codec control function to enable frame parallel decoding feature.
    *
    * AV1 has a bitstream feature to reduce decoding dependency between frames
@@ -557,6 +561,16 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ENABLE_RESTORATION,
 
+  /*!\brief Codec control function to predict with OBMC mode.
+   *
+   *                          0 = do not allow OBMC mode
+   *                          1 = allow OBMC mode
+   *
+   *  By default, the encoder allows OBMC prediction mode.
+   *
+   */
+  AV1E_SET_ENABLE_OBMC,
+
   /*!\brief Codec control function to encode without trellis quantization.
    *
    *                          0 = apply trellis quantization
@@ -684,18 +698,6 @@ enum aome_enc_control_id {
    */
   AV1E_SET_MTU,
 
-  /*!\brief Codec control function to set dependent_horz_tiles.
-   *
-   * In encoding and decoding, AV1 allows enabling dependent horizontal tile
-   * The parameter for this control describes the value of this flag,
-   * which has a valid range [0, 1]:
-   *            0 = disable dependent horizontal tile
-   *            1 = enable dependent horizontal tile,
-   *
-   * By default, the value is 0, i.e. disable dependent horizontal tile.
-   */
-  AV1E_SET_TILE_DEPENDENT_ROWS,
-
   /*!\brief Codec control function to set the number of symbols in an ANS data
    * window.
    *
@@ -708,13 +710,22 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ANS_WINDOW_SIZE_LOG2,
 
-  /*!\brief Codec control function to turn on / off dual filter
-   * enabling/disabling.
+  /*!\brief Codec control function to enable/disable rectangular partitions.
    *
-   * This will enable or disable dual filter. The default value is 1
+   * This will enable or disable usage of rectangular partitions. The default
+   * value is 1.
    *
    */
-  AV1E_SET_ENABLE_DF,
+  AV1E_SET_ENABLE_RECT_PARTITIONS,
+
+  /*!\brief Codec control function to turn on / off intra edge filter
+   * at sequence level.
+   *
+   * This will enable or disable usage of intra-edge filtering. The default
+   * value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_INTRA_EDGE_FILTER,
 
   /*!\brief Codec control function to turn on / off frame order hint for a
    * few tools:
@@ -728,14 +739,23 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ENABLE_ORDER_HINT,
 
-  /*!\brief Codec control function to turn on / off joint compound mode
-   * at sequence level.
+  /*!\brief Codec control function to turn on / off 64-length transforms.
    *
-   * This will enable or disable joint compound mode. The default value is 1.
-   * If AV1E_SET_ENABLE_ORDER_HINT is 0, then this flag is forced to 0.
+   * This will enable or disable usage of length 64 transforms in any
+   * direction. The default value is 1.
    *
    */
-  AV1E_SET_ENABLE_JNT_COMP,
+  AV1E_SET_ENABLE_TX64,
+
+  /*!\brief Codec control function to turn on / off dist-wtd compound mode
+   * at sequence level.
+   *
+   * This will enable or disable distance-weighted compound mode. The default
+   * value is 1. If AV1E_SET_ENABLE_ORDER_HINT is 0, then this flag is forced
+   * to 0.
+   *
+   */
+  AV1E_SET_ENABLE_DIST_WTD_COMP,
 
   /*!\brief Codec control function to turn on / off ref frame mvs (mfmv) usage
    * at sequence level.
@@ -755,6 +775,77 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ALLOW_REF_FRAME_MVS,
 
+  /*!\brief Codec control function to turn on / off dual filter usage
+   * for a sequence.
+   *
+   * This will enable or disable use of dual interpolation filter.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_DUAL_FILTER,
+
+  /*!\brief Codec control function to turn on / off masked compound usage
+   * for a sequence.
+   *
+   * This will enable or disable usage of wedge and diff-wtd compound
+   * modes. The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_MASKED_COMP,
+
+  /*!\brief Codec control function to turn on / off interintra compound
+   * for a sequence.
+   *
+   * This will enable or disable usage of inter-intra compound modes.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_INTERINTRA_COMP,
+
+  /*!\brief Codec control function to turn on / off smooth inter-intra
+   * mode for a sequence.
+   *
+   * This will enable or disable usage of smooth inter-intra mode.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_SMOOTH_INTERINTRA,
+
+  /*!\brief Codec control function to turn on / off difference weighted
+   * compound.
+   *
+   * This will enable or disable usage of difference weighted compound.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_DIFF_WTD_COMP,
+
+  /*!\brief Codec control function to turn on / off interinter wedge
+   * compound.
+   *
+   * This will enable or disable usage of interinter wedge compound.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_INTERINTER_WEDGE,
+
+  /*!\brief Codec control function to turn on / off interintra wedge
+   * compound.
+   *
+   * This will enable or disable usage of interintra wedge compound.
+   * The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_INTERINTRA_WEDGE,
+
+  /*!\brief Codec control function to turn on / off global motion usage
+   * for a sequence.
+   *
+   * This will enable or disable usage of global motion. The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_GLOBAL_MOTION,
+
   /*!\brief Codec control function to turn on / off warped motion usage
    * at sequence level.
    *
@@ -772,6 +863,39 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ALLOW_WARPED_MOTION,
 
+  /*!\brief Codec control function to turn on / off filter intra usage at
+   * sequence level.
+   *
+   * This will enable or disable usage of filter intra. The default value is 1.
+   * If AV1E_SET_ENABLE_FILTER_INTRA is 0, then this flag is forced to 0.
+   *
+   */
+  AV1E_SET_ENABLE_FILTER_INTRA,
+
+  /*!\brief Codec control function to turn on / off smooth intra modes usage.
+   *
+   * This will enable or disable usage of smooth, smooth_h and smooth_v intra
+   * modes. The default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_SMOOTH_INTRA,
+
+  /*!\brief Codec control function to turn on / off Paeth intra mode usage.
+   *
+   * This will enable or disable usage of Paeth intra mode. The default value
+   * is 1.
+   *
+   */
+  AV1E_SET_ENABLE_PAETH_INTRA,
+
+  /*!\brief Codec control function to turn on / off CFL uv intra mode usage.
+   *
+   * This will enable or disable usage of chroma-from-luma intra mode. The
+   * default value is 1.
+   *
+   */
+  AV1E_SET_ENABLE_CFL_INTRA,
+
   /*!\brief Codec control function to turn on / off frame superresolution.
    *
    * This will enable or disable frame superresolution. The default value is 1
@@ -779,34 +903,14 @@ enum aome_enc_control_id {
    */
   AV1E_SET_ENABLE_SUPERRES,
 
-  /*!\brief Codec control function to set loop_filter_across_tiles_v_enabled
-   * and loop_filter_across_tiles_h_enabled.
-   * In encoding and decoding, AV1 allows disabling loop filter across tile
-   * boundary The parameter for this control describes the value of this flag,
-   * which has a valid range [0, 1]:
-   *            0 = disable loop filter across tile boundary
-   *            1 = enable loop filter across tile boundary
-   *
-   * By default, the value is 1, i.e. enable loop filter across tile boundary.
-   *
-   * Experiment: LOOPFILTERING_ACROSS_TILES_EXT
-   */
-  AV1E_SET_TILE_LOOPFILTER_V,
-  AV1E_SET_TILE_LOOPFILTER_H,
+  /*!\brief Codec control function to turn on/off palette mode */
+  AV1E_SET_ENABLE_PALETTE,
 
-  /*!\brief Codec control function to set loop_filter_across_tiles_enabled.
-   *
-   * In encoding and decoding, AV1 allows disabling loop filter across tile
-   * boundary The parameter for this control describes the value of this flag,
-   * which has a valid range [0, 1]:
-   *            0 = disable loop filter across tile boundary
-   *            1 = enable loop filter across tile boundary
-   *
-   * By default, the value is 1, i.e. enable loop filter across tile boundary.
-   *
-   * Experiment: LOOPFILTERING_ACROSS_TILES
-   */
-  AV1E_SET_TILE_LOOPFILTER,
+  /*!\brief Codec control function to turn on/off intra block copy mode */
+  AV1E_SET_ENABLE_INTRABC,
+
+  /*!\brief Codec control function to turn on/off intra angle delta */
+  AV1E_SET_ENABLE_ANGLE_DELTA,
 
   /*!\brief Codec control function to set the delta q mode
    *
@@ -865,6 +969,26 @@ enum aome_enc_control_id {
 
   /*!\brief Sets the chroma subsampling y value */
   AV1E_SET_CHROMA_SUBSAMPLING_Y,
+
+  /*!\brief Control to use a reduced tx type set */
+  AV1E_SET_REDUCED_TX_TYPE_SET,
+
+  /*!\brief Control to use dct only for intra modes */
+  AV1E_SET_INTRA_DCT_ONLY,
+
+  /*!\brief Control to use dct only for inter modes */
+  AV1E_SET_INTER_DCT_ONLY,
+
+  /*!\brief Control to use adaptive quantize_b */
+  AV1E_SET_QUANT_B_ADAPT,
+
+  /*!\brief Control to select maximum height for the GF group pyramid structure
+   * (valid values: 1 - 4) */
+  AV1E_SET_GF_MAX_PYRAMID_HEIGHT,
+
+  /*!\brief Control to select maximum reference frames allowed per frame
+   * (valid values: 3 - 7) */
+  AV1E_SET_MAX_REFERENCE_FRAMES,
 };
 
 /*!\brief aom 1-D scaling mode
@@ -1006,15 +1130,8 @@ AOM_CTRL_USE_TYPE(AV1E_SET_TILE_COLUMNS, int)
 AOM_CTRL_USE_TYPE(AV1E_SET_TILE_ROWS, int)
 #define AOM_CTRL_AV1E_SET_TILE_ROWS
 
-AOM_CTRL_USE_TYPE(AV1E_SET_TILE_DEPENDENT_ROWS, int)
-#define AOM_CTRL_AV1E_SET_TILE_DEPENDENT_ROWS
-
-AOM_CTRL_USE_TYPE(AV1E_SET_TILE_LOOPFILTER_V, int)
-#define AOM_CTRL_AV1E_SET_TILE_LOOPFILTER_V
-AOM_CTRL_USE_TYPE(AV1E_SET_TILE_LOOPFILTER_H, int)
-#define AOM_CTRL_AV1E_SET_TILE_LOOPFILTER_H
-AOM_CTRL_USE_TYPE(AV1E_SET_TILE_LOOPFILTER, int)
-#define AOM_CTRL_AV1E_SET_TILE_LOOPFILTER
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_TPL_MODEL, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_TPL_MODEL
 
 AOM_CTRL_USE_TYPE(AOME_GET_LAST_QUANTIZER, int *)
 #define AOM_CTRL_AOME_GET_LAST_QUANTIZER
@@ -1041,6 +1158,9 @@ AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_CDEF, unsigned int)
 AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_RESTORATION, unsigned int)
 #define AOM_CTRL_AV1E_SET_ENABLE_RESTORATION
 
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_OBMC, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_OBMC
+
 AOM_CTRL_USE_TYPE(AV1E_SET_DISABLE_TRELLIS_QUANT, unsigned int)
 #define AOM_CTRL_AV1E_SET_DISABLE_TRELLIS_QUANT
 
@@ -1070,17 +1190,23 @@ AOM_CTRL_USE_TYPE(AV1E_SET_NUM_TG, unsigned int)
 AOM_CTRL_USE_TYPE(AV1E_SET_MTU, unsigned int)
 #define AOM_CTRL_AV1E_SET_MTU
 
-AOM_CTRL_USE_TYPE(AV1E_SET_TIMING_INFO_TYPE, aom_timing_info_type_t)
+AOM_CTRL_USE_TYPE(AV1E_SET_TIMING_INFO_TYPE, int) /* aom_timing_info_type_t */
 #define AOM_CTRL_AV1E_SET_TIMING_INFO_TYPE
 
-AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_DF, unsigned int)
-#define AOM_CTRL_AV1E_SET_ENABLE_DF
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_RECT_PARTITIONS, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_RECT_PARTITIONS
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_INTRA_EDGE_FILTER, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_INTRA_EDGE_FILTER
 
 AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_ORDER_HINT, unsigned int)
 #define AOM_CTRL_AV1E_SET_ENABLE_ORDER_HINT
 
-AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_JNT_COMP, unsigned int)
-#define AOM_CTRL_AV1E_SET_ENABLE_JNT_COMP
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_TX64, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_TX64
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_DIST_WTD_COMP, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_DIST_WTD_COMP
 
 AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_REF_FRAME_MVS, unsigned int)
 #define AOM_CTRL_AV1E_SET_ENABLE_REF_FRAME_MVS
@@ -1088,15 +1214,60 @@ AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_REF_FRAME_MVS, unsigned int)
 AOM_CTRL_USE_TYPE(AV1E_SET_ALLOW_REF_FRAME_MVS, unsigned int)
 #define AOM_CTRL_AV1E_SET_ALLOW_REF_FRAME_MVS
 
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_DUAL_FILTER, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_DUAL_FILTER
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_MASKED_COMP, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_MASKED_COMP
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_INTERINTRA_COMP, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_INTERINTRA_COMP
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_SMOOTH_INTERINTRA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_SMOOTH_INTERINTRA
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_DIFF_WTD_COMP, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_DIFF_WTD_COMP
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_INTERINTER_WEDGE, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_INTERINTER_WEDGE
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_INTERINTRA_WEDGE, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_INTERINTRA_WEDGE
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_GLOBAL_MOTION, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_GLOBAL_MOTION
+
 AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_WARPED_MOTION, unsigned int)
 #define AOM_CTRL_AV1E_SET_ENABLE_WARPED_MOTION
 
 AOM_CTRL_USE_TYPE(AV1E_SET_ALLOW_WARPED_MOTION, unsigned int)
 #define AOM_CTRL_AV1E_SET_ALLOW_WARPED_MOTION
 
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_FILTER_INTRA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_FILTER_INTRA
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_SMOOTH_INTRA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_SMOOTH_INTRA
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_PAETH_INTRA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_PAETH_INTRA
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_CFL_INTRA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_CFL_INTRA
+
 AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_SUPERRES, unsigned int)
 #define AOM_CTRL_AV1E_SET_ENABLE_SUPERRES
 
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_PALETTE, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_PALETTE
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_INTRABC, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_INTRABC
+
+AOM_CTRL_USE_TYPE(AV1E_SET_ENABLE_ANGLE_DELTA, unsigned int)
+#define AOM_CTRL_AV1E_SET_ENABLE_ANGLE_DELTA
+
 AOM_CTRL_USE_TYPE(AV1E_SET_FRAME_PARALLEL_DECODING, unsigned int)
 #define AOM_CTRL_AV1E_SET_FRAME_PARALLEL_DECODING
 
@@ -1189,6 +1360,24 @@ AOM_CTRL_USE_TYPE(AV1E_SET_CHROMA_SUBSAMPLING_X, unsigned int)
 AOM_CTRL_USE_TYPE(AV1E_SET_CHROMA_SUBSAMPLING_Y, unsigned int)
 #define AOM_CTRL_AV1E_SET_CHROMA_SUBSAMPLING_Y
 
+AOM_CTRL_USE_TYPE(AV1E_SET_REDUCED_TX_TYPE_SET, unsigned int)
+#define AOM_CTRL_AV1E_SET_REDUCED_TX_TYPE_SET
+
+AOM_CTRL_USE_TYPE(AV1E_SET_INTRA_DCT_ONLY, unsigned int)
+#define AOM_CTRL_AV1E_SET_INTRA_DCT_ONLY
+
+AOM_CTRL_USE_TYPE(AV1E_SET_INTER_DCT_ONLY, unsigned int)
+#define AOM_CTRL_AV1E_SET_INTER_DCT_ONLY
+
+AOM_CTRL_USE_TYPE(AV1E_SET_QUANT_B_ADAPT, unsigned int)
+#define AOM_CTRL_AV1E_SET_QUANT_B_ADAPT
+
+AOM_CTRL_USE_TYPE(AV1E_SET_GF_MAX_PYRAMID_HEIGHT, unsigned int)
+#define AOM_CTRL_AV1E_SET_GF_MAX_PYRAMID_HEIGHT
+
+AOM_CTRL_USE_TYPE(AV1E_SET_MAX_REFERENCE_FRAMES, unsigned int)
+#define AOM_CTRL_AV1E_SET_MAX_REFERENCE_FRAMES
+
 /*!\endcond */
 /*! @} - end defgroup aom_encoder */
 #ifdef __cplusplus
diff --git a/third_party/aom/aom/aomdx.h b/third_party/aom/aom/aomdx.h
index 765856a1bb..c71eaf90d6 100644
--- a/third_party/aom/aom/aomdx.h
+++ b/third_party/aom/aom/aomdx.h
@@ -45,6 +45,7 @@ typedef struct Accounting Accounting;
 /** Callback that inspects decoder frame data.
  */
 typedef void (*aom_inspect_cb)(void *decoder, void *ctx);
+
 #endif
 
 /*!\brief Structure to hold inspection callback and context.
@@ -60,6 +61,21 @@ typedef struct aom_inspect_init {
   void *inspect_ctx;
 } aom_inspect_init;
 
+/*!\brief Structure to collect a buffer index when inspecting.
+ *
+ * Defines a structure to hold the buffer and return an index
+ * when calling decode from inspect. This enables us to decode
+ * non showable sub frames.
+ */
+typedef struct {
+  /*! Pointer for new position in compressed buffer after decoding 1 OBU. */
+  const unsigned char *buf;
+  /*! Index into reference buffer array to see result of decoding 1 OBU. */
+  int idx;
+  /*! Is a show existing frame. */
+  int show_existing;
+} Av1DecodeReturn;
+
 /*!\brief Structure to hold a tile's start address and size in the bitstream.
  *
  * Defines a structure to hold a tile's start address and size in the bitstream.
@@ -125,6 +141,9 @@ enum aom_dec_control_id {
   /** control function to get the size of the tile. */
   AV1D_GET_TILE_SIZE,
 
+  /** control function to get the tile count in a tile list. */
+  AV1D_GET_TILE_COUNT,
+
   /** control function to set the byte alignment of the planes in the reference
    * buffers. Valid values are power of 2, from 32 to 1024. A value of 0 sets
    * legacy alignment. I.e. Y plane is aligned to 32 bytes, U plane directly
@@ -260,6 +279,8 @@ AOM_CTRL_USE_TYPE(AV1D_GET_IMG_FORMAT, aom_img_fmt_t *)
 #define AOM_CTRL_AV1D_GET_IMG_FORMAT
 AOM_CTRL_USE_TYPE(AV1D_GET_TILE_SIZE, unsigned int *)
 #define AOM_CTRL_AV1D_GET_TILE_SIZE
+AOM_CTRL_USE_TYPE(AV1D_GET_TILE_COUNT, unsigned int *)
+#define AOM_CTRL_AV1D_GET_TILE_COUNT
 AOM_CTRL_USE_TYPE(AV1D_GET_FRAME_SIZE, int *)
 #define AOM_CTRL_AV1D_GET_FRAME_SIZE
 AOM_CTRL_USE_TYPE(AV1_INVERT_TILE_DECODE_ORDER, int)
diff --git a/third_party/aom/aom/exports_com b/third_party/aom/aom/exports_com
index 2798bd51ad..cf99bc5d95 100644
--- a/third_party/aom/aom/exports_com
+++ b/third_party/aom/aom/exports_com
@@ -9,6 +9,7 @@ text aom_codec_iface_name
 text aom_codec_version
 text aom_codec_version_extra_str
 text aom_codec_version_str
+text aom_free
 text aom_img_alloc
 text aom_img_alloc_with_border
 text aom_img_flip
diff --git a/third_party/aom/aom/src/aom_decoder.c b/third_party/aom/aom/src/aom_decoder.c
index 8c9111faf0..282ec8a756 100644
--- a/third_party/aom/aom/src/aom_decoder.c
+++ b/third_party/aom/aom/src/aom_decoder.c
@@ -37,9 +37,6 @@ aom_codec_err_t aom_codec_dec_init_ver(aom_codec_ctx_t *ctx,
   else if ((flags & AOM_CODEC_USE_POSTPROC) &&
            !(iface->caps & AOM_CODEC_CAP_POSTPROC))
     res = AOM_CODEC_INCAPABLE;
-  else if ((flags & AOM_CODEC_USE_INPUT_FRAGMENTS) &&
-           !(iface->caps & AOM_CODEC_CAP_INPUT_FRAGMENTS))
-    res = AOM_CODEC_INCAPABLE;
   else if (!(iface->caps & AOM_CODEC_CAP_DECODER))
     res = AOM_CODEC_INCAPABLE;
   else {
diff --git a/third_party/aom/aom/src/aom_image.c b/third_party/aom/aom/src/aom_image.c
index 437f0241eb..6504cdde01 100644
--- a/third_party/aom/aom/src/aom_image.c
+++ b/third_party/aom/aom/src/aom_image.c
@@ -59,6 +59,7 @@ static aom_image_t *img_alloc_helper(
     case AOM_IMG_FMT_AOMYV12: bps = 12; break;
     case AOM_IMG_FMT_I422:
     case AOM_IMG_FMT_I444: bps = 24; break;
+    case AOM_IMG_FMT_YV1216:
     case AOM_IMG_FMT_I42016: bps = 24; break;
     case AOM_IMG_FMT_I42216:
     case AOM_IMG_FMT_I44416: bps = 48; break;
@@ -73,6 +74,7 @@ static aom_image_t *img_alloc_helper(
     case AOM_IMG_FMT_AOMYV12:
     case AOM_IMG_FMT_I422:
     case AOM_IMG_FMT_I42016:
+    case AOM_IMG_FMT_YV1216:
     case AOM_IMG_FMT_I42216: xcs = 1; break;
     default: xcs = 0; break;
   }
@@ -82,6 +84,7 @@ static aom_image_t *img_alloc_helper(
     case AOM_IMG_FMT_YV12:
     case AOM_IMG_FMT_AOMI420:
     case AOM_IMG_FMT_AOMYV12:
+    case AOM_IMG_FMT_YV1216:
     case AOM_IMG_FMT_I42016: ycs = 1; break;
     default: ycs = 0; break;
   }
@@ -117,6 +120,7 @@ static aom_image_t *img_alloc_helper(
 
     img->img_data = (uint8_t *)aom_memalign(buf_align, (size_t)alloc_size);
     img->img_data_owner = 1;
+    img->sz = (size_t)alloc_size;
   }
 
   if (!img->img_data) goto fail;
@@ -131,7 +135,7 @@ static aom_image_t *img_alloc_helper(
   img->bps = bps;
 
   /* Calculate strides */
-  img->stride[AOM_PLANE_Y] = img->stride[AOM_PLANE_ALPHA] = stride_in_bytes;
+  img->stride[AOM_PLANE_Y] = stride_in_bytes;
   img->stride[AOM_PLANE_U] = img->stride[AOM_PLANE_V] = stride_in_bytes >> xcs;
 
   /* Default viewport to entire image */
@@ -185,12 +189,6 @@ int aom_img_set_rect(aom_image_t *img, unsigned int x, unsigned int y,
           (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1;
       data = img->img_data;
 
-      if (img->fmt & AOM_IMG_FMT_HAS_ALPHA) {
-        img->planes[AOM_PLANE_ALPHA] =
-            data + x * bytes_per_sample + y * img->stride[AOM_PLANE_ALPHA];
-        data += (img->h + 2 * border) * img->stride[AOM_PLANE_ALPHA];
-      }
-
       img->planes[AOM_PLANE_Y] =
           data + x * bytes_per_sample + y * img->stride[AOM_PLANE_Y];
       data += (img->h + 2 * border) * img->stride[AOM_PLANE_Y];
@@ -236,10 +234,6 @@ void aom_img_flip(aom_image_t *img) {
   img->planes[AOM_PLANE_V] += (signed)((img->d_h >> img->y_chroma_shift) - 1) *
                               img->stride[AOM_PLANE_V];
   img->stride[AOM_PLANE_V] = -img->stride[AOM_PLANE_V];
-
-  img->planes[AOM_PLANE_ALPHA] +=
-      (signed)(img->d_h - 1) * img->stride[AOM_PLANE_ALPHA];
-  img->stride[AOM_PLANE_ALPHA] = -img->stride[AOM_PLANE_ALPHA];
 }
 
 void aom_img_free(aom_image_t *img) {
diff --git a/third_party/aom/aom_dsp/add_noise.c b/third_party/aom/aom_dsp/add_noise.c
index bfb3e7e009..43587ca315 100644
--- a/third_party/aom/aom_dsp/add_noise.c
+++ b/third_party/aom/aom_dsp/add_noise.c
@@ -40,7 +40,7 @@ void aom_plane_add_noise_c(uint8_t *start, char *noise, char blackclamp[16],
 }
 
 static double gaussian(double sigma, double mu, double x) {
-  return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
+  return 1 / (sigma * sqrt(2.0 * PI)) *
          (exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
 }
 
diff --git a/third_party/aom/aom_dsp/aom_dsp.cmake b/third_party/aom/aom_dsp/aom_dsp.cmake
index 11ff73756b..c6a6e3df59 100644
--- a/third_party/aom/aom_dsp/aom_dsp.cmake
+++ b/third_party/aom/aom_dsp/aom_dsp.cmake
@@ -35,6 +35,7 @@ list(APPEND AOM_DSP_COMMON_SOURCES
             "${AOM_ROOT}/aom_dsp/intrapred_common.h"
             "${AOM_ROOT}/aom_dsp/loopfilter.c"
             "${AOM_ROOT}/aom_dsp/prob.h"
+            "${AOM_ROOT}/aom_dsp/recenter.h"
             "${AOM_ROOT}/aom_dsp/simd/v128_intrinsics.h"
             "${AOM_ROOT}/aom_dsp/simd/v128_intrinsics_c.h"
             "${AOM_ROOT}/aom_dsp/simd/v256_intrinsics.h"
@@ -43,7 +44,8 @@ list(APPEND AOM_DSP_COMMON_SOURCES
             "${AOM_ROOT}/aom_dsp/simd/v64_intrinsics_c.h"
             "${AOM_ROOT}/aom_dsp/subtract.c"
             "${AOM_ROOT}/aom_dsp/txfm_common.h"
-            "${AOM_ROOT}/aom_dsp/x86/convolve_common_intrin.h")
+            "${AOM_ROOT}/aom_dsp/x86/convolve_common_intrin.h"
+            "${AOM_ROOT}/aom_dsp/avg.c")
 
 list(APPEND AOM_DSP_COMMON_ASM_SSE2
             "${AOM_ROOT}/aom_dsp/x86/aom_convolve_copy_sse2.asm"
@@ -51,15 +53,17 @@ list(APPEND AOM_DSP_COMMON_ASM_SSE2
             "${AOM_ROOT}/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm"
             "${AOM_ROOT}/aom_dsp/x86/aom_subpixel_8t_sse2.asm"
             "${AOM_ROOT}/aom_dsp/x86/aom_subpixel_bilinear_sse2.asm"
-            "${AOM_ROOT}/aom_dsp/x86/highbd_intrapred_sse2.asm"
-            "${AOM_ROOT}/aom_dsp/x86/intrapred_sse2.asm"
+            "${AOM_ROOT}/aom_dsp/x86/highbd_intrapred_asm_sse2.asm"
+            "${AOM_ROOT}/aom_dsp/x86/intrapred_asm_sse2.asm"
             "${AOM_ROOT}/aom_dsp/x86/inv_wht_sse2.asm")
 
 list(APPEND AOM_DSP_COMMON_INTRIN_SSE2
+            "${AOM_ROOT}/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c"
             "${AOM_ROOT}/aom_dsp/x86/aom_asm_stubs.c"
             "${AOM_ROOT}/aom_dsp/x86/convolve.h"
             "${AOM_ROOT}/aom_dsp/x86/convolve_sse2.h"
             "${AOM_ROOT}/aom_dsp/x86/fft_sse2.c"
+            "${AOM_ROOT}/aom_dsp/x86/highbd_convolve_sse2.c"
             "${AOM_ROOT}/aom_dsp/x86/highbd_intrapred_sse2.c"
             "${AOM_ROOT}/aom_dsp/x86/highbd_loopfilter_sse2.c"
             "${AOM_ROOT}/aom_dsp/x86/intrapred_sse2.c"
@@ -68,7 +72,9 @@ list(APPEND AOM_DSP_COMMON_INTRIN_SSE2
             "${AOM_ROOT}/aom_dsp/x86/mem_sse2.h"
             "${AOM_ROOT}/aom_dsp/x86/transpose_sse2.h"
             "${AOM_ROOT}/aom_dsp/x86/txfm_common_sse2.h"
-            "${AOM_ROOT}/aom_dsp/x86/sum_squares_sse2.h")
+            "${AOM_ROOT}/aom_dsp/x86/sum_squares_sse2.h"
+            "${AOM_ROOT}/aom_dsp/x86/avg_intrin_sse2.c"
+            "${AOM_ROOT}/aom_dsp/x86/bitdepth_conversion_sse2.h")
 
 list(APPEND AOM_DSP_COMMON_ASM_SSSE3
             "${AOM_ROOT}/aom_dsp/x86/aom_subpixel_8t_ssse3.asm"
@@ -94,7 +100,9 @@ list(APPEND AOM_DSP_COMMON_INTRIN_AVX2
             "${AOM_ROOT}/aom_dsp/x86/highbd_convolve_avx2.c"
             "${AOM_ROOT}/aom_dsp/x86/highbd_loopfilter_avx2.c"
             "${AOM_ROOT}/aom_dsp/x86/intrapred_avx2.c"
-            "${AOM_ROOT}/aom_dsp/x86/blend_a64_mask_avx2.c")
+            "${AOM_ROOT}/aom_dsp/x86/blend_a64_mask_avx2.c"
+            "${AOM_ROOT}/aom_dsp/x86/avg_intrin_avx2.c"
+            "${AOM_ROOT}/aom_dsp/x86/bitdepth_conversion_avx2.h")
 
 list(APPEND AOM_DSP_COMMON_INTRIN_NEON
             "${AOM_ROOT}/aom_dsp/arm/fwd_txfm_neon.c"
@@ -175,6 +183,9 @@ if(CONFIG_AV1_ENCODER)
               "${AOM_ROOT}/aom_dsp/x86/subpel_variance_sse2.asm"
               "${AOM_ROOT}/aom_dsp/x86/subtract_sse2.asm")
 
+  list(APPEND AOM_DSP_ENCODER_ASM_SSE2_X86_64
+              "${AOM_ROOT}/aom_dsp/x86/ssim_sse2_x86_64.asm")
+
   list(APPEND AOM_DSP_ENCODER_INTRIN_SSE2
               "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_impl_sse2.h"
               "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_sse2.c"
@@ -189,7 +200,7 @@ if(CONFIG_AV1_ENCODER)
 
   list(APPEND AOM_DSP_ENCODER_ASM_SSSE3_X86_64
               "${AOM_ROOT}/aom_dsp/x86/fwd_txfm_ssse3_x86_64.asm"
-              "${AOM_ROOT}/aom_dsp/x86/ssim_opt_x86_64.asm")
+              "${AOM_ROOT}/aom_dsp/x86/quantize_ssse3_x86_64.asm")
 
   list(APPEND AOM_DSP_ENCODER_INTRIN_AVX2
               "${AOM_ROOT}/aom_dsp/x86/masked_sad_intrin_avx2.c"
@@ -207,9 +218,6 @@ if(CONFIG_AV1_ENCODER)
               "${AOM_ROOT}/aom_dsp/x86/obmc_variance_avx2.c"
               "${AOM_ROOT}/aom_dsp/x86/sum_squares_avx2.c")
 
-  list(APPEND AOM_DSP_ENCODER_ASM_SSSE3_X86_64
-              "${AOM_ROOT}/aom_dsp/x86/quantize_ssse3_x86_64.asm")
-
   list(APPEND AOM_DSP_ENCODER_AVX_ASM_X86_64
               "${AOM_ROOT}/aom_dsp/x86/quantize_avx_x86_64.asm")
 
@@ -218,6 +226,7 @@ if(CONFIG_AV1_ENCODER)
               "${AOM_ROOT}/aom_dsp/x86/masked_sad_intrin_ssse3.c"
               "${AOM_ROOT}/aom_dsp/x86/masked_variance_intrin_ssse3.h"
               "${AOM_ROOT}/aom_dsp/x86/masked_variance_intrin_ssse3.c"
+              "${AOM_ROOT}/aom_dsp/x86/quantize_ssse3.c"
               "${AOM_ROOT}/aom_dsp/x86/variance_impl_ssse3.c"
               "${AOM_ROOT}/aom_dsp/x86/jnt_variance_ssse3.c"
               "${AOM_ROOT}/aom_dsp/x86/jnt_sad_ssse3.c")
@@ -278,6 +287,9 @@ function(setup_aom_dsp_targets)
                                   "AOM_DSP_COMMON_INTRIN_SSE2" "aom")
 
     if(CONFIG_AV1_ENCODER)
+      if("${AOM_TARGET_CPU}" STREQUAL "x86_64")
+        list(APPEND AOM_DSP_ENCODER_ASM_SSE2 ${AOM_DSP_ENCODER_ASM_SSE2_X86_64})
+      endif()
       add_asm_library("aom_dsp_encoder_sse2" "AOM_DSP_ENCODER_ASM_SSE2" "aom")
       add_intrinsics_object_library("-msse2" "sse2" "aom_dsp_encoder"
                                     "AOM_DSP_ENCODER_INTRIN_SSE2" "aom")
diff --git a/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl b/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
index 8e8a480fec..ff0f65333c 100755
--- a/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
+++ b/third_party/aom/aom_dsp/aom_dsp_rtcd_defs.pl
@@ -362,10 +362,10 @@ add_proto qw/void aom_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_
 specialize qw/aom_highbd_convolve_copy sse2 avx2/;
 
 add_proto qw/void aom_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/aom_highbd_convolve8_horiz avx2/, "$sse2_x86_64";
+specialize qw/aom_highbd_convolve8_horiz sse2 avx2/;
 
 add_proto qw/void aom_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-specialize qw/aom_highbd_convolve8_vert avx2/, "$sse2_x86_64";
+specialize qw/aom_highbd_convolve8_vert sse2 avx2/;
 
 #
 # Loopfilter
@@ -466,10 +466,6 @@ specialize qw/aom_highbd_lpf_horizontal_4 sse2/;
 add_proto qw/void aom_highbd_lpf_horizontal_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
 specialize qw/aom_highbd_lpf_horizontal_4_dual sse2 avx2/;
 
-# Helper functions.
-add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit";
-specialize "av1_round_shift_array", qw/sse4_1 neon/;
-
 #
 # Encoder functions.
 #
@@ -526,6 +522,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   specialize qw/aom_quantize_b_32x32/, "$ssse3_x86_64", "$avx_x86_64";
 
   add_proto qw/void aom_quantize_b_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+  specialize qw/aom_quantize_b_64x64 ssse3/;
 }  # CONFIG_AV1_ENCODER
 
 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
@@ -544,7 +541,6 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 #
 add_proto qw/void aom_lowbd_blend_a64_d16_mask/, "uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params";
 specialize qw/aom_lowbd_blend_a64_d16_mask sse4_1 avx2 neon/;
-add_proto qw/void aom_highbd_blend_a64_d16_mask/, "uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd";
 add_proto qw/void aom_blend_a64_mask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby";
 add_proto qw/void aom_blend_a64_hmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h";
 add_proto qw/void aom_blend_a64_vmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h";
@@ -555,9 +551,11 @@ specialize "aom_blend_a64_vmask", qw/sse4_1 neon/;
 add_proto qw/void aom_highbd_blend_a64_mask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, int bd";
 add_proto qw/void aom_highbd_blend_a64_hmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd";
 add_proto qw/void aom_highbd_blend_a64_vmask/, "uint8_t *dst, uint32_t dst_stride, const uint8_t *src0, uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride, const uint8_t *mask, int w, int h, int bd";
+add_proto qw/void aom_highbd_blend_a64_d16_mask/, "uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0, uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride, const uint8_t *mask, uint32_t mask_stride, int w, int h, int subx, int suby, ConvolveParams *conv_params, const int bd";
 specialize "aom_highbd_blend_a64_mask", qw/sse4_1/;
 specialize "aom_highbd_blend_a64_hmask", qw/sse4_1/;
 specialize "aom_highbd_blend_a64_vmask", qw/sse4_1/;
+specialize "aom_highbd_blend_a64_d16_mask", qw/sse4_1 avx2/;
 
 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   #
@@ -595,7 +593,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
     ($w, $h) = @$_;
     add_proto qw/unsigned int/, "aom_sad${w}x${h}", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
     add_proto qw/unsigned int/, "aom_sad${w}x${h}_avg", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-    add_proto qw/unsigned int/, "aom_jnt_sad${w}x${h}_avg", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param";
+    add_proto qw/unsigned int/, "aom_dist_wtd_sad${w}x${h}_avg", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param";
   }
 
   specialize qw/aom_sad128x128    avx2          sse2/;
@@ -646,29 +644,29 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   specialize qw/aom_sad16x64_avg sse2/;
   specialize qw/aom_sad64x16_avg sse2/;
 
-  specialize qw/aom_jnt_sad128x128_avg ssse3/;
-  specialize qw/aom_jnt_sad128x64_avg  ssse3/;
-  specialize qw/aom_jnt_sad64x128_avg  ssse3/;
-  specialize qw/aom_jnt_sad64x64_avg   ssse3/;
-  specialize qw/aom_jnt_sad64x32_avg   ssse3/;
-  specialize qw/aom_jnt_sad32x64_avg   ssse3/;
-  specialize qw/aom_jnt_sad32x32_avg   ssse3/;
-  specialize qw/aom_jnt_sad32x16_avg   ssse3/;
-  specialize qw/aom_jnt_sad16x32_avg   ssse3/;
-  specialize qw/aom_jnt_sad16x16_avg   ssse3/;
-  specialize qw/aom_jnt_sad16x8_avg    ssse3/;
-  specialize qw/aom_jnt_sad8x16_avg    ssse3/;
-  specialize qw/aom_jnt_sad8x8_avg     ssse3/;
-  specialize qw/aom_jnt_sad8x4_avg     ssse3/;
-  specialize qw/aom_jnt_sad4x8_avg     ssse3/;
-  specialize qw/aom_jnt_sad4x4_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad128x128_avg ssse3/;
+  specialize qw/aom_dist_wtd_sad128x64_avg  ssse3/;
+  specialize qw/aom_dist_wtd_sad64x128_avg  ssse3/;
+  specialize qw/aom_dist_wtd_sad64x64_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad64x32_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad32x64_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad32x32_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad32x16_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad16x32_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad16x16_avg   ssse3/;
+  specialize qw/aom_dist_wtd_sad16x8_avg    ssse3/;
+  specialize qw/aom_dist_wtd_sad8x16_avg    ssse3/;
+  specialize qw/aom_dist_wtd_sad8x8_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad8x4_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad4x8_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad4x4_avg     ssse3/;
 
-  specialize qw/aom_jnt_sad4x16_avg     ssse3/;
-  specialize qw/aom_jnt_sad16x4_avg     ssse3/;
-  specialize qw/aom_jnt_sad8x32_avg     ssse3/;
-  specialize qw/aom_jnt_sad32x8_avg     ssse3/;
-  specialize qw/aom_jnt_sad16x64_avg     ssse3/;
-  specialize qw/aom_jnt_sad64x16_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad4x16_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad16x4_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad8x32_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad32x8_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad16x64_avg     ssse3/;
+  specialize qw/aom_dist_wtd_sad64x16_avg     ssse3/;
 
   add_proto qw/unsigned int/, "aom_sad4xh", "const uint8_t *a, int a_stride, const uint8_t *b, int b_stride, int width, int height";
   add_proto qw/unsigned int/, "aom_sad8xh", "const uint8_t *a, int a_stride, const uint8_t *b, int b_stride, int width, int height";
@@ -693,7 +691,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
         specialize "aom_highbd_sad${w}x${h}", qw/sse2/;
         specialize "aom_highbd_sad${w}x${h}_avg", qw/sse2/;
       }
-      add_proto qw/unsigned int/, "aom_highbd_jnt_sad${w}x${h}_avg", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred, const JNT_COMP_PARAMS* jcp_param";
+      add_proto qw/unsigned int/, "aom_highbd_dist_wtd_sad${w}x${h}_avg", "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS* jcp_param";
     }
     specialize qw/aom_highbd_sad128x128 avx2/;
     specialize qw/aom_highbd_sad128x64  avx2/;
@@ -837,6 +835,21 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   specialize qw/aom_highbd_sad16x64x4d sse2/;
   specialize qw/aom_highbd_sad64x16x4d sse2/;
 
+  #
+  # hamadard transform and satd for implmenting temporal dependency model
+  #
+  add_proto qw/void aom_hadamard_8x8/, "const int16_t *src_diff, ptrdiff_t src_stride, tran_low_t *coeff";
+  specialize qw/aom_hadamard_8x8 sse2/;
+
+  add_proto qw/void aom_hadamard_16x16/, "const int16_t *src_diff, ptrdiff_t src_stride, tran_low_t *coeff";
+  specialize qw/aom_hadamard_16x16 avx2 sse2/;
+
+  add_proto qw/void aom_hadamard_32x32/, "const int16_t *src_diff, ptrdiff_t src_stride, tran_low_t *coeff";
+  specialize qw/aom_hadamard_32x32 avx2 sse2/;
+
+  add_proto qw/int aom_satd/, "const tran_low_t *coeff, int length";
+  specialize qw/aom_satd avx2 sse2/;
+
   #
   # Structured Similarity (SSIM)
   #
@@ -903,11 +916,11 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
                                                    int ref_stride, int subpel_search";
   specialize qw/aom_comp_avg_upsampled_pred sse2/;
 
-  add_proto qw/void aom_jnt_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+  add_proto qw/void aom_dist_wtd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
                                                        const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
                                                        int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
-                                                       int ref_stride, const JNT_COMP_PARAMS *jcp_param, int subpel_search";
-  specialize qw/aom_jnt_comp_avg_upsampled_pred ssse3/;
+                                                       int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search";
+  specialize qw/aom_dist_wtd_comp_avg_upsampled_pred ssse3/;
 
   add_proto qw/void aom_comp_mask_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
                                                        const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
@@ -916,7 +929,6 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
                                                        int subpel_search";
   specialize qw/aom_comp_mask_upsampled_pred sse2/;
 
-
   add_proto qw/void aom_highbd_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
                                                  const MV *const mv, uint8_t *comp_pred8, int width, int height, int subpel_x_q3,
                                                  int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
@@ -927,11 +939,11 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
                                                           int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
   specialize qw/aom_highbd_comp_avg_upsampled_pred sse2/;
 
-  add_proto qw/void aom_highbd_jnt_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+  add_proto qw/void aom_highbd_dist_wtd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
                                                               const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
                                                               int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
-                                                              int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param, int subpel_search";
-  specialize qw/aom_highbd_jnt_comp_avg_upsampled_pred sse2/;
+                                                              int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search";
+  specialize qw/aom_highbd_dist_wtd_comp_avg_upsampled_pred sse2/;
 
 
   #
@@ -957,7 +969,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
     add_proto qw/unsigned int/, "aom_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
     add_proto qw/uint32_t/, "aom_sub_pixel_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
     add_proto qw/uint32_t/, "aom_sub_pixel_avg_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
-    add_proto qw/uint32_t/, "aom_jnt_sub_pixel_avg_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param";
+    add_proto qw/uint32_t/, "aom_dist_wtd_sub_pixel_avg_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param";
   }
   specialize qw/aom_variance128x128   sse2 avx2         /;
   specialize qw/aom_variance128x64    sse2 avx2         /;
@@ -1029,30 +1041,30 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   specialize qw/aom_sub_pixel_avg_variance16x64 sse2 ssse3/;
   specialize qw/aom_sub_pixel_avg_variance64x16 sse2 ssse3/;
 
-  specialize qw/aom_jnt_sub_pixel_avg_variance64x64 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance64x32 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance32x64 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance32x32 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance32x16 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance16x32 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance16x16 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance16x8  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance8x16  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance8x8   ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance8x4   ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance4x8   ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance4x4   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance64x64 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance64x32 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance32x64 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance32x32 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance32x16 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance16x32 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance16x16 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance16x8  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance8x16  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance8x8   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance8x4   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance4x8   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance4x4   ssse3/;
 
-  specialize qw/aom_jnt_sub_pixel_avg_variance4x16  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance16x4  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance8x32  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance32x8  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance16x64 ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance64x16 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance4x16  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance16x4  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance8x32  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance32x8  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance16x64 ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance64x16 ssse3/;
 
-  specialize qw/aom_jnt_sub_pixel_avg_variance128x128  ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance128x64   ssse3/;
-  specialize qw/aom_jnt_sub_pixel_avg_variance64x128   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance128x128  ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance128x64   ssse3/;
+  specialize qw/aom_dist_wtd_sub_pixel_avg_variance64x128   ssse3/;
 
 
   foreach $bd (8, 10, 12) {
@@ -1084,7 +1096,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
         specialize "aom_highbd_${bd}_sub_pixel_avg_variance${w}x${h}", "sse4_1";
       }
 
-      add_proto qw/uint32_t/, "aom_highbd_${bd}_jnt_sub_pixel_avg_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred, const JNT_COMP_PARAMS* jcp_param";
+      add_proto qw/uint32_t/, "aom_highbd_${bd}_dist_wtd_sub_pixel_avg_variance${w}x${h}", "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS* jcp_param";
     }
   }
 
@@ -1173,8 +1185,8 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   #
   add_proto qw/void aom_comp_avg_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride";
 
-  add_proto qw/void aom_jnt_comp_avg_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride, const JNT_COMP_PARAMS *jcp_param";
-  specialize qw/aom_jnt_comp_avg_pred ssse3/;
+  add_proto qw/void aom_dist_wtd_comp_avg_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param";
+  specialize qw/aom_dist_wtd_comp_avg_pred ssse3/;
 
     add_proto qw/unsigned int aom_highbd_12_variance128x128/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
     specialize qw/aom_highbd_12_variance128x128 sse2/;
@@ -1340,8 +1352,8 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 
     add_proto qw/void aom_highbd_comp_avg_pred/, "uint8_t *comp_pred8, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride";
 
-    add_proto qw/void aom_highbd_jnt_comp_avg_pred/, "uint8_t *comp_pred8, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride, const JNT_COMP_PARAMS *jcp_param";
-    specialize qw/aom_highbd_jnt_comp_avg_pred sse2/;
+    add_proto qw/void aom_highbd_dist_wtd_comp_avg_pred/, "uint8_t *comp_pred8, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param";
+    specialize qw/aom_highbd_dist_wtd_comp_avg_pred sse2/;
 
     #
     # Subpixel Variance
diff --git a/third_party/aom/aom_dsp/arm/loopfilter_neon.c b/third_party/aom/aom_dsp/arm/loopfilter_neon.c
index bdc67626d6..aafac8966d 100644
--- a/third_party/aom/aom_dsp/arm/loopfilter_neon.c
+++ b/third_party/aom/aom_dsp/arm/loopfilter_neon.c
@@ -785,16 +785,15 @@ void aom_lpf_vertical_4_neon(uint8_t *src, int stride, const uint8_t *blimit,
                              const uint8_t *limit, const uint8_t *thresh) {
   uint32x2x2_t p1q0_p0q1, p1q1_p0q0, p1p0_q1q0;
   uint32x2_t pq_rev;
-  uint8x8_t UNINITIALIZED_IS_SAFE(p1p0), q0q1, p0q0, p1q1;
+  uint8x8_t UNINITIALIZED_IS_SAFE(p1p0), UNINITIALIZED_IS_SAFE(q0q1);
+  uint8x8_t p0q0, p1q1;
 
   // row0: p1 p0 | q0 q1
   // row1: p1 p0 | q0 q1
   // row2: p1 p0 | q0 q1
   // row3: p1 p0 | q0 q1
-  load_u8_4x1(src - 2, &p1p0, 0);
-  load_u8_4x1((src - 2) + 1 * stride, &p1p0, 1);
-  load_u8_4x1((src - 2) + 2 * stride, &q0q1, 0);
-  load_u8_4x1((src - 2) + 3 * stride, &q0q1, 1);
+  load_unaligned_u8_4x4(src - 2, stride, (uint32x2_t *)&p1p0,
+                        (uint32x2_t *)&q0q1);
 
   transpose_u8_4x4(&p1p0, &q0q1);
 
@@ -815,10 +814,10 @@ void aom_lpf_vertical_4_neon(uint8_t *src, int stride, const uint8_t *blimit,
 
   transpose_u8_4x4(&p1p0, &q0q1);
 
-  store_u8_4x1(src - 2, p1p0, 0);
-  store_u8_4x1((src - 2) + 1 * stride, q0q1, 0);
-  store_u8_4x1((src - 2) + 2 * stride, p1p0, 1);
-  store_u8_4x1((src - 2) + 3 * stride, q0q1, 1);
+  store_unaligned_u8_4x1(src - 2, p1p0, 0);
+  store_unaligned_u8_4x1((src - 2) + 1 * stride, q0q1, 0);
+  store_unaligned_u8_4x1((src - 2) + 2 * stride, p1p0, 1);
+  store_unaligned_u8_4x1((src - 2) + 3 * stride, q0q1, 1);
 }
 
 void aom_lpf_horizontal_14_neon(uint8_t *src, int stride, const uint8_t *blimit,
diff --git a/third_party/aom/aom_dsp/avg.c b/third_party/aom/aom_dsp/avg.c
new file mode 100644
index 0000000000..4d78c9c569
--- /dev/null
+++ b/third_party/aom/aom_dsp/avg.c
@@ -0,0 +1,148 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdlib.h>
+
+#include "config/aom_dsp_rtcd.h"
+#include "aom_ports/mem.h"
+
+// src_diff: first pass, 9 bit, dynamic range [-255, 255]
+//           second pass, 12 bit, dynamic range [-2040, 2040]
+static void hadamard_col8(const int16_t *src_diff, ptrdiff_t src_stride,
+                          int16_t *coeff) {
+  int16_t b0 = src_diff[0 * src_stride] + src_diff[1 * src_stride];
+  int16_t b1 = src_diff[0 * src_stride] - src_diff[1 * src_stride];
+  int16_t b2 = src_diff[2 * src_stride] + src_diff[3 * src_stride];
+  int16_t b3 = src_diff[2 * src_stride] - src_diff[3 * src_stride];
+  int16_t b4 = src_diff[4 * src_stride] + src_diff[5 * src_stride];
+  int16_t b5 = src_diff[4 * src_stride] - src_diff[5 * src_stride];
+  int16_t b6 = src_diff[6 * src_stride] + src_diff[7 * src_stride];
+  int16_t b7 = src_diff[6 * src_stride] - src_diff[7 * src_stride];
+
+  int16_t c0 = b0 + b2;
+  int16_t c1 = b1 + b3;
+  int16_t c2 = b0 - b2;
+  int16_t c3 = b1 - b3;
+  int16_t c4 = b4 + b6;
+  int16_t c5 = b5 + b7;
+  int16_t c6 = b4 - b6;
+  int16_t c7 = b5 - b7;
+
+  coeff[0] = c0 + c4;
+  coeff[7] = c1 + c5;
+  coeff[3] = c2 + c6;
+  coeff[4] = c3 + c7;
+  coeff[2] = c0 - c4;
+  coeff[6] = c1 - c5;
+  coeff[1] = c2 - c6;
+  coeff[5] = c3 - c7;
+}
+
+// The order of the output coeff of the hadamard is not important. For
+// optimization purposes the final transpose may be skipped.
+void aom_hadamard_8x8_c(const int16_t *src_diff, ptrdiff_t src_stride,
+                        tran_low_t *coeff) {
+  int idx;
+  int16_t buffer[64];
+  int16_t buffer2[64];
+  int16_t *tmp_buf = &buffer[0];
+  for (idx = 0; idx < 8; ++idx) {
+    hadamard_col8(src_diff, src_stride, tmp_buf);  // src_diff: 9 bit
+                                                   // dynamic range [-255, 255]
+    tmp_buf += 8;
+    ++src_diff;
+  }
+
+  tmp_buf = &buffer[0];
+  for (idx = 0; idx < 8; ++idx) {
+    hadamard_col8(tmp_buf, 8, buffer2 + 8 * idx);  // tmp_buf: 12 bit
+    // dynamic range [-2040, 2040]
+    // buffer2: 15 bit
+    // dynamic range [-16320, 16320]
+    ++tmp_buf;
+  }
+
+  for (idx = 0; idx < 64; ++idx) coeff[idx] = (tran_low_t)buffer2[idx];
+}
+
+// In place 16x16 2D Hadamard transform
+void aom_hadamard_16x16_c(const int16_t *src_diff, ptrdiff_t src_stride,
+                          tran_low_t *coeff) {
+  int idx;
+  for (idx = 0; idx < 4; ++idx) {
+    // src_diff: 9 bit, dynamic range [-255, 255]
+    const int16_t *src_ptr =
+        src_diff + (idx >> 1) * 8 * src_stride + (idx & 0x01) * 8;
+    aom_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
+  }
+
+  // coeff: 15 bit, dynamic range [-16320, 16320]
+  for (idx = 0; idx < 64; ++idx) {
+    tran_low_t a0 = coeff[0];
+    tran_low_t a1 = coeff[64];
+    tran_low_t a2 = coeff[128];
+    tran_low_t a3 = coeff[192];
+
+    tran_low_t b0 = (a0 + a1) >> 1;  // (a0 + a1): 16 bit, [-32640, 32640]
+    tran_low_t b1 = (a0 - a1) >> 1;  // b0-b3: 15 bit, dynamic range
+    tran_low_t b2 = (a2 + a3) >> 1;  // [-16320, 16320]
+    tran_low_t b3 = (a2 - a3) >> 1;
+
+    coeff[0] = b0 + b2;  // 16 bit, [-32640, 32640]
+    coeff[64] = b1 + b3;
+    coeff[128] = b0 - b2;
+    coeff[192] = b1 - b3;
+
+    ++coeff;
+  }
+}
+
+void aom_hadamard_32x32_c(const int16_t *src_diff, ptrdiff_t src_stride,
+                          tran_low_t *coeff) {
+  int idx;
+  for (idx = 0; idx < 4; ++idx) {
+    // src_diff: 9 bit, dynamic range [-255, 255]
+    const int16_t *src_ptr =
+        src_diff + (idx >> 1) * 16 * src_stride + (idx & 0x01) * 16;
+    aom_hadamard_16x16_c(src_ptr, src_stride, coeff + idx * 256);
+  }
+
+  // coeff: 15 bit, dynamic range [-16320, 16320]
+  for (idx = 0; idx < 256; ++idx) {
+    tran_low_t a0 = coeff[0];
+    tran_low_t a1 = coeff[256];
+    tran_low_t a2 = coeff[512];
+    tran_low_t a3 = coeff[768];
+
+    tran_low_t b0 = (a0 + a1) >> 2;  // (a0 + a1): 16 bit, [-32640, 32640]
+    tran_low_t b1 = (a0 - a1) >> 2;  // b0-b3: 15 bit, dynamic range
+    tran_low_t b2 = (a2 + a3) >> 2;  // [-16320, 16320]
+    tran_low_t b3 = (a2 - a3) >> 2;
+
+    coeff[0] = b0 + b2;  // 16 bit, [-32640, 32640]
+    coeff[256] = b1 + b3;
+    coeff[512] = b0 - b2;
+    coeff[768] = b1 - b3;
+
+    ++coeff;
+  }
+}
+
+// coeff: 16 bits, dynamic range [-32640, 32640].
+// length: value range {16, 64, 256, 1024}.
+int aom_satd_c(const tran_low_t *coeff, int length) {
+  int i;
+  int satd = 0;
+  for (i = 0; i < length; ++i) satd += abs(coeff[i]);
+
+  // satd: 26 bits, dynamic range [-32640 * 1024, 32640 * 1024]
+  return satd;
+}
diff --git a/third_party/aom/aom_dsp/binary_codes_reader.c b/third_party/aom/aom_dsp/binary_codes_reader.c
index 01088010a1..7cd903d821 100644
--- a/third_party/aom/aom_dsp/binary_codes_reader.c
+++ b/third_party/aom/aom_dsp/binary_codes_reader.c
@@ -10,29 +10,9 @@
  */
 
 #include "aom_dsp/binary_codes_reader.h"
-
+#include "aom_dsp/recenter.h"
 #include "av1/common/common.h"
 
-// Inverse recenters a non-negative literal v around a reference r
-static uint16_t inv_recenter_nonneg(uint16_t r, uint16_t v) {
-  if (v > (r << 1))
-    return v;
-  else if ((v & 1) == 0)
-    return (v >> 1) + r;
-  else
-    return r - ((v + 1) >> 1);
-}
-
-// Inverse recenters a non-negative literal v in [0, n-1] around a
-// reference r also in [0, n-1]
-static uint16_t inv_recenter_finite_nonneg(uint16_t n, uint16_t r, uint16_t v) {
-  if ((r << 1) <= n) {
-    return inv_recenter_nonneg(r, v);
-  } else {
-    return n - 1 - inv_recenter_nonneg(n - 1 - r, v);
-  }
-}
-
 uint16_t aom_read_primitive_quniform_(aom_reader *r,
                                       uint16_t n ACCT_STR_PARAM) {
   if (n <= 1) return 0;
@@ -42,15 +22,6 @@ uint16_t aom_read_primitive_quniform_(aom_reader *r,
   return v < m ? v : (v << 1) - m + aom_read_bit(r, ACCT_STR_NAME);
 }
 
-static uint16_t aom_rb_read_primitive_quniform(struct aom_read_bit_buffer *rb,
-                                               uint16_t n) {
-  if (n <= 1) return 0;
-  const int l = get_msb(n) + 1;
-  const int m = (1 << l) - n;
-  const int v = aom_rb_read_literal(rb, l - 1);
-  return v < m ? v : (v << 1) - m + aom_rb_read_bit(rb);
-}
-
 // Decode finite subexponential code that for a symbol v in [0, n-1] with
 // parameter k
 uint16_t aom_read_primitive_subexpfin_(aom_reader *r, uint16_t n,
@@ -78,46 +49,8 @@ uint16_t aom_read_primitive_subexpfin_(aom_reader *r, uint16_t n,
   return 0;
 }
 
-static uint16_t aom_rb_read_primitive_subexpfin(struct aom_read_bit_buffer *rb,
-                                                uint16_t n, uint16_t k) {
-  int i = 0;
-  int mk = 0;
-
-  while (1) {
-    int b = (i ? k + i - 1 : k);
-    int a = (1 << b);
-
-    if (n <= mk + 3 * a) {
-      return aom_rb_read_primitive_quniform(rb, n - mk) + mk;
-    }
-
-    if (!aom_rb_read_bit(rb)) {
-      return aom_rb_read_literal(rb, b) + mk;
-    }
-
-    i = i + 1;
-    mk += a;
-  }
-
-  assert(0);
-  return 0;
-}
-
 uint16_t aom_read_primitive_refsubexpfin_(aom_reader *r, uint16_t n, uint16_t k,
                                           uint16_t ref ACCT_STR_PARAM) {
   return inv_recenter_finite_nonneg(
       n, ref, aom_read_primitive_subexpfin(r, n, k, ACCT_STR_NAME));
 }
-
-static uint16_t aom_rb_read_primitive_refsubexpfin(
-    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, uint16_t ref) {
-  return inv_recenter_finite_nonneg(n, ref,
-                                    aom_rb_read_primitive_subexpfin(rb, n, k));
-}
-
-int16_t aom_rb_read_signed_primitive_refsubexpfin(
-    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, int16_t ref) {
-  ref += n - 1;
-  const uint16_t scaled_n = (n << 1) - 1;
-  return aom_rb_read_primitive_refsubexpfin(rb, scaled_n, k, ref) - n + 1;
-}
diff --git a/third_party/aom/aom_dsp/binary_codes_reader.h b/third_party/aom/aom_dsp/binary_codes_reader.h
index 364a67469a..d218f0619f 100644
--- a/third_party/aom/aom_dsp/binary_codes_reader.h
+++ b/third_party/aom/aom_dsp/binary_codes_reader.h
@@ -37,9 +37,6 @@ uint16_t aom_read_primitive_subexpfin_(aom_reader *r, uint16_t n,
 uint16_t aom_read_primitive_refsubexpfin_(aom_reader *r, uint16_t n, uint16_t k,
                                           uint16_t ref ACCT_STR_PARAM);
 
-int16_t aom_rb_read_signed_primitive_refsubexpfin(
-    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, int16_t ref);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/aom_dsp/binary_codes_writer.c b/third_party/aom/aom_dsp/binary_codes_writer.c
index ee7a9f567e..adf1c1304c 100644
--- a/third_party/aom/aom_dsp/binary_codes_writer.c
+++ b/third_party/aom/aom_dsp/binary_codes_writer.c
@@ -11,29 +11,10 @@
 
 #include "aom_dsp/bitwriter.h"
 #include "aom_dsp/binary_codes_writer.h"
-
+#include "aom_dsp/recenter.h"
+#include "aom_ports/bitops.h"
 #include "av1/common/common.h"
 
-// Recenters a non-negative literal v around a reference r
-static uint16_t recenter_nonneg(uint16_t r, uint16_t v) {
-  if (v > (r << 1))
-    return v;
-  else if (v >= r)
-    return ((v - r) << 1);
-  else
-    return ((r - v) << 1) - 1;
-}
-
-// Recenters a non-negative literal v in [0, n-1] around a
-// reference r also in [0, n-1]
-static uint16_t recenter_finite_nonneg(uint16_t n, uint16_t r, uint16_t v) {
-  if ((r << 1) <= n) {
-    return recenter_nonneg(r, v);
-  } else {
-    return recenter_nonneg(n - 1 - r, n - 1 - v);
-  }
-}
-
 // Codes a symbol v in [-2^mag_bits, 2^mag_bits].
 // mag_bits is number of bits for magnitude. The alphabet is of size
 // 2 * 2^mag_bits + 1, symmetric around 0, where one bit is used to
@@ -69,19 +50,6 @@ void aom_write_primitive_quniform(aom_writer *w, uint16_t n, uint16_t v) {
   }
 }
 
-static void aom_wb_write_primitive_quniform(struct aom_write_bit_buffer *wb,
-                                            uint16_t n, uint16_t v) {
-  if (n <= 1) return;
-  const int l = get_msb(n) + 1;
-  const int m = (1 << l) - n;
-  if (v < m) {
-    aom_wb_write_literal(wb, v, l - 1);
-  } else {
-    aom_wb_write_literal(wb, m + ((v - m) >> 1), l - 1);
-    aom_wb_write_bit(wb, (v - m) & 1);
-  }
-}
-
 int aom_count_primitive_quniform(uint16_t n, uint16_t v) {
   if (n <= 1) return 0;
   const int l = get_msb(n) + 1;
@@ -114,31 +82,6 @@ void aom_write_primitive_subexpfin(aom_writer *w, uint16_t n, uint16_t k,
   }
 }
 
-static void aom_wb_write_primitive_subexpfin(struct aom_write_bit_buffer *wb,
-                                             uint16_t n, uint16_t k,
-                                             uint16_t v) {
-  int i = 0;
-  int mk = 0;
-  while (1) {
-    int b = (i ? k + i - 1 : k);
-    int a = (1 << b);
-    if (n <= mk + 3 * a) {
-      aom_wb_write_primitive_quniform(wb, n - mk, v - mk);
-      break;
-    } else {
-      int t = (v >= mk + a);
-      aom_wb_write_bit(wb, t);
-      if (t) {
-        i = i + 1;
-        mk += a;
-      } else {
-        aom_wb_write_literal(wb, v - mk, b);
-        break;
-      }
-    }
-  }
-}
-
 int aom_count_primitive_subexpfin(uint16_t n, uint16_t k, uint16_t v) {
   int count = 0;
   int i = 0;
@@ -172,12 +115,6 @@ void aom_write_primitive_refsubexpfin(aom_writer *w, uint16_t n, uint16_t k,
   aom_write_primitive_subexpfin(w, n, k, recenter_finite_nonneg(n, ref, v));
 }
 
-static void aom_wb_write_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
-                                                uint16_t n, uint16_t k,
-                                                uint16_t ref, uint16_t v) {
-  aom_wb_write_primitive_subexpfin(wb, n, k, recenter_finite_nonneg(n, ref, v));
-}
-
 void aom_write_signed_primitive_refsubexpfin(aom_writer *w, uint16_t n,
                                              uint16_t k, int16_t ref,
                                              int16_t v) {
@@ -187,15 +124,6 @@ void aom_write_signed_primitive_refsubexpfin(aom_writer *w, uint16_t n,
   aom_write_primitive_refsubexpfin(w, scaled_n, k, ref, v);
 }
 
-void aom_wb_write_signed_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
-                                                uint16_t n, uint16_t k,
-                                                int16_t ref, int16_t v) {
-  ref += n - 1;
-  v += n - 1;
-  const uint16_t scaled_n = (n << 1) - 1;
-  aom_wb_write_primitive_refsubexpfin(wb, scaled_n, k, ref, v);
-}
-
 int aom_count_primitive_refsubexpfin(uint16_t n, uint16_t k, uint16_t ref,
                                      uint16_t v) {
   return aom_count_primitive_subexpfin(n, k, recenter_finite_nonneg(n, ref, v));
diff --git a/third_party/aom/aom_dsp/binary_codes_writer.h b/third_party/aom/aom_dsp/binary_codes_writer.h
index c360e0e298..5ec8662139 100644
--- a/third_party/aom/aom_dsp/binary_codes_writer.h
+++ b/third_party/aom/aom_dsp/binary_codes_writer.h
@@ -49,10 +49,6 @@ void aom_write_signed_primitive_refsubexpfin(aom_writer *w, uint16_t n,
                                              uint16_t k, int16_t ref,
                                              int16_t v);
 
-void aom_wb_write_signed_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
-                                                uint16_t n, uint16_t k,
-                                                int16_t ref, int16_t v);
-
 // Functions that counts bits for the above primitives
 int aom_count_primitive_symmetric(int16_t v, unsigned int mag_bits);
 int aom_count_primitive_quniform(uint16_t n, uint16_t v);
@@ -61,6 +57,7 @@ int aom_count_primitive_refsubexpfin(uint16_t n, uint16_t k, uint16_t ref,
                                      uint16_t v);
 int aom_count_signed_primitive_refsubexpfin(uint16_t n, uint16_t k, int16_t ref,
                                             int16_t v);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/aom_dsp/bitreader.h b/third_party/aom/aom_dsp/bitreader.h
index 7c0efcc785..38b17eae5f 100644
--- a/third_party/aom/aom_dsp/bitreader.h
+++ b/third_party/aom/aom_dsp/bitreader.h
@@ -65,10 +65,6 @@ static INLINE const uint8_t *aom_reader_find_end(aom_reader *r) {
   return aom_daala_reader_find_end(r);
 }
 
-static INLINE int aom_reader_has_error(aom_reader *r) {
-  return aom_daala_reader_has_error(r);
-}
-
 // Returns true if the bit reader has tried to decode more data from the buffer
 // than was actually provided.
 static INLINE int aom_reader_has_overflowed(const aom_reader *r) {
diff --git a/third_party/aom/aom_dsp/bitreader_buffer.c b/third_party/aom/aom_dsp/bitreader_buffer.c
index b532117849..984b217738 100644
--- a/third_party/aom/aom_dsp/bitreader_buffer.c
+++ b/third_party/aom/aom_dsp/bitreader_buffer.c
@@ -14,6 +14,8 @@
 #include "config/aom_config.h"
 
 #include "aom_dsp/bitreader_buffer.h"
+#include "aom_dsp/recenter.h"
+#include "aom_ports/bitops.h"
 
 size_t aom_rb_bytes_read(const struct aom_read_bit_buffer *rb) {
   return (rb->bit_offset + 7) >> 3;
@@ -65,3 +67,50 @@ uint32_t aom_rb_read_uvlc(struct aom_read_bit_buffer *rb) {
   const uint32_t value = aom_rb_read_literal(rb, leading_zeros);
   return base + value;
 }
+
+static uint16_t aom_rb_read_primitive_quniform(struct aom_read_bit_buffer *rb,
+                                               uint16_t n) {
+  if (n <= 1) return 0;
+  const int l = get_msb(n) + 1;
+  const int m = (1 << l) - n;
+  const int v = aom_rb_read_literal(rb, l - 1);
+  return v < m ? v : (v << 1) - m + aom_rb_read_bit(rb);
+}
+
+static uint16_t aom_rb_read_primitive_subexpfin(struct aom_read_bit_buffer *rb,
+                                                uint16_t n, uint16_t k) {
+  int i = 0;
+  int mk = 0;
+
+  while (1) {
+    int b = (i ? k + i - 1 : k);
+    int a = (1 << b);
+
+    if (n <= mk + 3 * a) {
+      return aom_rb_read_primitive_quniform(rb, n - mk) + mk;
+    }
+
+    if (!aom_rb_read_bit(rb)) {
+      return aom_rb_read_literal(rb, b) + mk;
+    }
+
+    i = i + 1;
+    mk += a;
+  }
+
+  assert(0);
+  return 0;
+}
+
+static uint16_t aom_rb_read_primitive_refsubexpfin(
+    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, uint16_t ref) {
+  return inv_recenter_finite_nonneg(n, ref,
+                                    aom_rb_read_primitive_subexpfin(rb, n, k));
+}
+
+int16_t aom_rb_read_signed_primitive_refsubexpfin(
+    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, int16_t ref) {
+  ref += n - 1;
+  const uint16_t scaled_n = (n << 1) - 1;
+  return aom_rb_read_primitive_refsubexpfin(rb, scaled_n, k, ref) - n + 1;
+}
diff --git a/third_party/aom/aom_dsp/bitreader_buffer.h b/third_party/aom/aom_dsp/bitreader_buffer.h
index 725ca1ea2b..359fbe5194 100644
--- a/third_party/aom/aom_dsp/bitreader_buffer.h
+++ b/third_party/aom/aom_dsp/bitreader_buffer.h
@@ -43,6 +43,9 @@ int aom_rb_read_inv_signed_literal(struct aom_read_bit_buffer *rb, int bits);
 
 uint32_t aom_rb_read_uvlc(struct aom_read_bit_buffer *rb);
 
+int16_t aom_rb_read_signed_primitive_refsubexpfin(
+    struct aom_read_bit_buffer *rb, uint16_t n, uint16_t k, int16_t ref);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/aom_dsp/bitwriter_buffer.c b/third_party/aom/aom_dsp/bitwriter_buffer.c
index 596246deb2..c08cc9da0b 100644
--- a/third_party/aom/aom_dsp/bitwriter_buffer.c
+++ b/third_party/aom/aom_dsp/bitwriter_buffer.c
@@ -16,6 +16,8 @@
 #include "config/aom_config.h"
 
 #include "aom_dsp/bitwriter_buffer.h"
+#include "aom_dsp/recenter.h"
+#include "aom_ports/bitops.h"
 
 int aom_wb_is_byte_aligned(const struct aom_write_bit_buffer *wb) {
   return (wb->bit_offset % CHAR_BIT == 0);
@@ -85,3 +87,56 @@ void aom_wb_write_uvlc(struct aom_write_bit_buffer *wb, uint32_t v) {
   aom_wb_write_literal(wb, 0, leading_zeroes >> 1);
   aom_wb_write_unsigned_literal(wb, v, (leading_zeroes + 1) >> 1);
 }
+
+static void aom_wb_write_primitive_quniform(struct aom_write_bit_buffer *wb,
+                                            uint16_t n, uint16_t v) {
+  if (n <= 1) return;
+  const int l = get_msb(n) + 1;
+  const int m = (1 << l) - n;
+  if (v < m) {
+    aom_wb_write_literal(wb, v, l - 1);
+  } else {
+    aom_wb_write_literal(wb, m + ((v - m) >> 1), l - 1);
+    aom_wb_write_bit(wb, (v - m) & 1);
+  }
+}
+
+static void aom_wb_write_primitive_subexpfin(struct aom_write_bit_buffer *wb,
+                                             uint16_t n, uint16_t k,
+                                             uint16_t v) {
+  int i = 0;
+  int mk = 0;
+  while (1) {
+    int b = (i ? k + i - 1 : k);
+    int a = (1 << b);
+    if (n <= mk + 3 * a) {
+      aom_wb_write_primitive_quniform(wb, n - mk, v - mk);
+      break;
+    } else {
+      int t = (v >= mk + a);
+      aom_wb_write_bit(wb, t);
+      if (t) {
+        i = i + 1;
+        mk += a;
+      } else {
+        aom_wb_write_literal(wb, v - mk, b);
+        break;
+      }
+    }
+  }
+}
+
+static void aom_wb_write_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
+                                                uint16_t n, uint16_t k,
+                                                uint16_t ref, uint16_t v) {
+  aom_wb_write_primitive_subexpfin(wb, n, k, recenter_finite_nonneg(n, ref, v));
+}
+
+void aom_wb_write_signed_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
+                                                uint16_t n, uint16_t k,
+                                                int16_t ref, int16_t v) {
+  ref += n - 1;
+  v += n - 1;
+  const uint16_t scaled_n = (n << 1) - 1;
+  aom_wb_write_primitive_refsubexpfin(wb, scaled_n, k, ref, v);
+}
diff --git a/third_party/aom/aom_dsp/bitwriter_buffer.h b/third_party/aom/aom_dsp/bitwriter_buffer.h
index d0311284f2..fd10e01bb7 100644
--- a/third_party/aom/aom_dsp/bitwriter_buffer.h
+++ b/third_party/aom/aom_dsp/bitwriter_buffer.h
@@ -44,6 +44,10 @@ void aom_wb_write_inv_signed_literal(struct aom_write_bit_buffer *wb, int data,
 
 void aom_wb_write_uvlc(struct aom_write_bit_buffer *wb, uint32_t v);
 
+void aom_wb_write_signed_primitive_refsubexpfin(struct aom_write_bit_buffer *wb,
+                                                uint16_t n, uint16_t k,
+                                                int16_t ref, int16_t v);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/aom_dsp/blend_a64_mask.c b/third_party/aom/aom_dsp/blend_a64_mask.c
index 992cc5c0c4..79956c3a57 100644
--- a/third_party/aom/aom_dsp/blend_a64_mask.c
+++ b/third_party/aom/aom_dsp/blend_a64_mask.c
@@ -22,7 +22,7 @@
 // as described for AOM_BLEND_A64 in aom_dsp/blend.h. src0 or src1 can
 // be the same as dst, or dst can be different from both sources.
 
-// NOTE(david.barker): The input and output of aom_blend_a64_d32_mask_c() are
+// NOTE(david.barker): The input and output of aom_blend_a64_d16_mask_c() are
 // in a higher intermediate precision, and will later be rounded down to pixel
 // precision.
 // Thus, in order to avoid double-rounding, we want to use normal right shifts
@@ -30,7 +30,7 @@
 // This works because of the identity:
 // ROUND_POWER_OF_TWO(x >> y, z) == ROUND_POWER_OF_TWO(x, y+z)
 //
-// In contrast, the output of the non-d32 functions will not be further rounded,
+// In contrast, the output of the non-d16 functions will not be further rounded,
 // so we *should* use ROUND_POWER_OF_TWO there.
 
 void aom_lowbd_blend_a64_d16_mask_c(
diff --git a/third_party/aom/aom_dsp/daalaboolreader.h b/third_party/aom/aom_dsp/daalaboolreader.h
index ba78f916d7..b1810fcda0 100644
--- a/third_party/aom/aom_dsp/daalaboolreader.h
+++ b/third_party/aom/aom_dsp/daalaboolreader.h
@@ -101,10 +101,6 @@ static INLINE int aom_daala_read(daala_reader *r, int prob) {
   return bit;
 }
 
-static INLINE int aom_daala_reader_has_error(daala_reader *r) {
-  return r->ec.error;
-}
-
 static INLINE int daala_read_symbol(daala_reader *r, const aom_cdf_prob *cdf,
                                     int nsymbs) {
   int symb;
diff --git a/third_party/aom/aom_dsp/entcode.h b/third_party/aom/aom_dsp/entcode.h
index 7ba2b1c392..7518879217 100644
--- a/third_party/aom/aom_dsp/entcode.h
+++ b/third_party/aom/aom_dsp/entcode.h
@@ -24,6 +24,7 @@
    on a larger type, you can speed up the decoder by using it here.*/
 typedef uint32_t od_ec_window;
 
+/*The size in bits of od_ec_window.*/
 #define OD_EC_WINDOW_SIZE ((int)sizeof(od_ec_window) * CHAR_BIT)
 
 /*The resolution of fractional-precision bit usage measurements, i.e.,
diff --git a/third_party/aom/aom_dsp/entdec.c b/third_party/aom/aom_dsp/entdec.c
index d1764c47b1..da43e8a399 100644
--- a/third_party/aom/aom_dsp/entdec.c
+++ b/third_party/aom/aom_dsp/entdec.c
@@ -87,11 +87,25 @@ static void od_ec_dec_refill(od_ec_dec *dec) {
   end = dec->end;
   s = OD_EC_WINDOW_SIZE - 9 - (cnt + 15);
   for (; s >= 0 && bptr < end; s -= 8, bptr++) {
+    /*Each time a byte is inserted into the window (dif), bptr advances and cnt
+       is incremented by 8, so the total number of consumed bits (the return
+       value of od_ec_dec_tell) does not change.*/
     assert(s <= OD_EC_WINDOW_SIZE - 8);
     dif ^= (od_ec_window)bptr[0] << s;
     cnt += 8;
   }
   if (bptr >= end) {
+    /*We've reached the end of the buffer. It is perfectly valid for us to need
+       to fill the window with additional bits past the end of the buffer (and
+       this happens in normal operation). These bits should all just be taken
+       as zero. But we cannot increment bptr past 'end' (this is undefined
+       behavior), so we start to increment dec->tell_offs. We also don't want
+       to keep testing bptr against 'end', so we set cnt to OD_EC_LOTS_OF_BITS
+       and adjust dec->tell_offs so that the total number of unconsumed bits in
+       the window (dec->cnt - dec->tell_offs) does not change. This effectively
+       puts lots of zero bits into the window, and means we won't try to refill
+       it from the buffer for a very long time (at which point we'll put lots
+       of zero bits into the window again).*/
     dec->tell_offs += OD_EC_LOTS_OF_BITS - cnt;
     cnt = OD_EC_LOTS_OF_BITS;
   }
@@ -112,8 +126,9 @@ static int od_ec_dec_normalize(od_ec_dec *dec, od_ec_window dif, unsigned rng,
                                int ret) {
   int d;
   assert(rng <= 65535U);
-  // The number of leading zeros in the 16-bit binary representation of rng.
+  /*The number of leading zeros in the 16-bit binary representation of rng.*/
   d = 16 - OD_ILOG_NZ(rng);
+  /*d bits in dec->dif are consumed.*/
   dec->cnt -= d;
   /*This is equivalent to shifting in 1's instead of 0's.*/
   dec->dif = ((dif + 1) << d) - 1;
@@ -124,7 +139,7 @@ static int od_ec_dec_normalize(od_ec_dec *dec, od_ec_window dif, unsigned rng,
 
 /*Initializes the decoder.
   buf: The input buffer to use.
-  Return: 0 on success, or a negative value on error.*/
+  storage: The size in bytes of the input buffer.*/
 void od_ec_dec_init(od_ec_dec *dec, const unsigned char *buf,
                     uint32_t storage) {
   dec->buf = buf;
@@ -134,7 +149,6 @@ void od_ec_dec_init(od_ec_dec *dec, const unsigned char *buf,
   dec->dif = ((od_ec_window)1 << (OD_EC_WINDOW_SIZE - 1)) - 1;
   dec->rng = 0x8000;
   dec->cnt = -15;
-  dec->error = 0;
   od_ec_dec_refill(dec);
 }
 
@@ -215,6 +229,10 @@ int od_ec_decode_cdf_q15(od_ec_dec *dec, const uint16_t *icdf, int nsyms) {
           This will always be slightly larger than the exact value (e.g., all
            rounding error is in the positive direction).*/
 int od_ec_dec_tell(const od_ec_dec *dec) {
+  /*There is a window of bits stored in dec->dif. The difference
+     (dec->bptr - dec->buf) tells us how many bytes have been read into this
+     window. The difference (dec->cnt - dec->tell_offs) tells us how many of
+     the bits in that window remain unconsumed.*/
   return (int)((dec->bptr - dec->buf) * 8 - dec->cnt + dec->tell_offs);
 }
 
diff --git a/third_party/aom/aom_dsp/entdec.h b/third_party/aom/aom_dsp/entdec.h
index 283bf18315..c746167775 100644
--- a/third_party/aom/aom_dsp/entdec.h
+++ b/third_party/aom/aom_dsp/entdec.h
@@ -34,7 +34,7 @@ struct od_ec_dec {
   const unsigned char *buf;
   /*An offset used to keep track of tell after reaching the end of the stream.
     This is constant throughout most of the decoding process, but becomes
-     important once we hit the end of the buffer and stop incrementing pointers
+     important once we hit the end of the buffer and stop incrementing bptr
      (and instead pretend cnt has lots of bits).*/
   int32_t tell_offs;
   /*The end of the current input buffer.*/
@@ -53,8 +53,6 @@ struct od_ec_dec {
   uint16_t rng;
   /*The number of bits of data in the current value.*/
   int16_t cnt;
-  /*Nonzero if an error occurred.*/
-  int error;
 };
 
 /*See entdec.c for further documentation.*/
diff --git a/third_party/aom/aom_dsp/entenc.c b/third_party/aom/aom_dsp/entenc.c
index a61da263cb..2fd4493eab 100644
--- a/third_party/aom/aom_dsp/entenc.c
+++ b/third_party/aom/aom_dsp/entenc.c
@@ -60,7 +60,7 @@ static void od_ec_enc_normalize(od_ec_enc *enc, od_ec_window low,
   int s;
   c = enc->cnt;
   assert(rng <= 65535U);
-  // The number of leading zeros in the 16-bit binary representation of rng.
+  /*The number of leading zeros in the 16-bit binary representation of rng.*/
   d = 16 - OD_ILOG_NZ(rng);
   s = c + d;
   /*TODO: Right now we flush every time we have at least one byte available.
diff --git a/third_party/aom/aom_dsp/grain_synthesis.h b/third_party/aom/aom_dsp/grain_synthesis.h
index 7aee6f6f44..9155b39035 100644
--- a/third_party/aom/aom_dsp/grain_synthesis.h
+++ b/third_party/aom/aom_dsp/grain_synthesis.h
@@ -20,6 +20,8 @@
 extern "C" {
 #endif
 
+#include <string.h>
+
 #include "aom_dsp/aom_dsp_common.h"
 #include "aom/aom_image.h"
 
@@ -28,6 +30,9 @@ extern "C" {
  * This structure contains input parameters for film grain synthesis
  */
 typedef struct {
+  // This structure is compared element-by-element in the function
+  // av1_check_grain_params_equiv: this function must be updated if any changes
+  // are made to this structure.
   int apply_grain;
 
   int update_parameters;
@@ -79,8 +84,73 @@ typedef struct {
   int grain_scale_shift;
 
   uint16_t random_seed;
+  // This structure is compared element-by-element in the function
+  // av1_check_grain_params_equiv: this function must be updated if any changes
+  // are made to this structure.
 } aom_film_grain_t;
 
+/*!\brief Check if two film grain parameters structs are equivalent
+ *
+ * Check if two film grain parameters are equal, except for the
+ * update_parameters and random_seed elements which are ignored.
+ *
+ * \param[in]    pa               The first set of parameters to compare
+ * \param[in]    pb               The second set of parameters to compare
+ * \return       Returns 1 if the params are equivalent, 0 otherwise
+ */
+static INLINE int av1_check_grain_params_equiv(
+    const aom_film_grain_t *const pa, const aom_film_grain_t *const pb) {
+  if (pa->apply_grain != pb->apply_grain) return 0;
+  // Don't compare update_parameters
+
+  if (pa->num_y_points != pb->num_y_points) return 0;
+  if (memcmp(pa->scaling_points_y, pb->scaling_points_y,
+             pa->num_y_points * 2 * sizeof(*pa->scaling_points_y)) != 0)
+    return 0;
+
+  if (pa->num_cb_points != pb->num_cb_points) return 0;
+  if (memcmp(pa->scaling_points_cb, pb->scaling_points_cb,
+             pa->num_cb_points * 2 * sizeof(*pa->scaling_points_cb)) != 0)
+    return 0;
+
+  if (pa->num_cr_points != pb->num_cr_points) return 0;
+  if (memcmp(pa->scaling_points_cr, pb->scaling_points_cr,
+             pa->num_cr_points * 2 * sizeof(*pa->scaling_points_cr)) != 0)
+    return 0;
+
+  if (pa->scaling_shift != pb->scaling_shift) return 0;
+  if (pa->ar_coeff_lag != pb->ar_coeff_lag) return 0;
+
+  const int num_pos = 2 * pa->ar_coeff_lag * (pa->ar_coeff_lag + 1);
+  if (memcmp(pa->ar_coeffs_y, pb->ar_coeffs_y,
+             num_pos * sizeof(*pa->ar_coeffs_y)) != 0)
+    return 0;
+  if (memcmp(pa->ar_coeffs_cb, pb->ar_coeffs_cb,
+             num_pos * sizeof(*pa->ar_coeffs_cb)) != 0)
+    return 0;
+  if (memcmp(pa->ar_coeffs_cr, pb->ar_coeffs_cr,
+             num_pos * sizeof(*pa->ar_coeffs_cr)) != 0)
+    return 0;
+
+  if (pa->ar_coeff_shift != pb->ar_coeff_shift) return 0;
+
+  if (pa->cb_mult != pb->cb_mult) return 0;
+  if (pa->cb_luma_mult != pb->cb_luma_mult) return 0;
+  if (pa->cb_offset != pb->cb_offset) return 0;
+
+  if (pa->cr_mult != pb->cr_mult) return 0;
+  if (pa->cr_luma_mult != pb->cr_luma_mult) return 0;
+  if (pa->cr_offset != pb->cr_offset) return 0;
+
+  if (pa->overlap_flag != pb->overlap_flag) return 0;
+  if (pa->clip_to_restricted_range != pb->clip_to_restricted_range) return 0;
+  if (pa->bit_depth != pb->bit_depth) return 0;
+  if (pa->chroma_scaling_from_luma != pb->chroma_scaling_from_luma) return 0;
+  if (pa->grain_scale_shift != pb->grain_scale_shift) return 0;
+
+  return 1;
+}
+
 /*!\brief Add film grain
  *
  * Add film grain to an image
diff --git a/third_party/aom/aom_dsp/grain_table.c b/third_party/aom/aom_dsp/grain_table.c
index 0d6a73f558..5eb5b68b79 100644
--- a/third_party/aom/aom_dsp/grain_table.c
+++ b/third_party/aom/aom_dsp/grain_table.c
@@ -134,7 +134,8 @@ static void grain_table_entry_read(FILE *file,
   }
 }
 
-void grain_table_entry_write(FILE *file, aom_film_grain_table_entry_t *entry) {
+static void grain_table_entry_write(FILE *file,
+                                    aom_film_grain_table_entry_t *entry) {
   const aom_film_grain_t *pars = &entry->params;
   fprintf(file, "E %" PRId64 " %" PRId64 " %d %d %d\n", entry->start_time,
           entry->end_time, pars->apply_grain, pars->random_seed,
diff --git a/third_party/aom/aom_dsp/intrapred.c b/third_party/aom/aom_dsp/intrapred.c
index c6aa6b207b..72ccfd8358 100644
--- a/third_party/aom/aom_dsp/intrapred.c
+++ b/third_party/aom/aom_dsp/intrapred.c
@@ -239,23 +239,23 @@ static INLINE int divide_using_multiply_shift(int num, int shift1,
   return interm * multiplier >> shift2;
 }
 
-  // The constants (multiplier and shifts) for a given block size are obtained
-  // as follows:
-  // - Let sum_w_h =  block width + block height.
-  // - Shift 'sum_w_h' right until we reach an odd number. Let the number of
-  // shifts for that block size be called 'shift1' (see the parameter in
-  // dc_predictor_rect() function), and let the odd number be 'd'. [d has only 2
-  // possible values: d = 3 for a 1:2 rect block and d = 5 for a 1:4 rect
-  // block].
-  // - Find multipliers for (i) dividing by 3, and (ii) dividing by 5,
-  // using the "Algorithm 1" in:
-  // http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1467632
-  // by ensuring that m + n = 16 (in that algorithm). This ensures that our 2nd
-  // shift will be 16, regardless of the block size.
+// The constants (multiplier and shifts) for a given block size are obtained
+// as follows:
+// - Let sum_w_h =  block width + block height.
+// - Shift 'sum_w_h' right until we reach an odd number. Let the number of
+// shifts for that block size be called 'shift1' (see the parameter in
+// dc_predictor_rect() function), and let the odd number be 'd'. [d has only 2
+// possible values: d = 3 for a 1:2 rect block and d = 5 for a 1:4 rect
+// block].
+// - Find multipliers for (i) dividing by 3, and (ii) dividing by 5,
+// using the "Algorithm 1" in:
+// http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1467632
+// by ensuring that m + n = 16 (in that algorithm). This ensures that our 2nd
+// shift will be 16, regardless of the block size.
 
-  // Note: For low bitdepth, assembly code may be optimized by using smaller
-  // constants for smaller block sizes, where the range of the 'sum' is
-  // restricted to fewer bits.
+// Note: For low bitdepth, assembly code may be optimized by using smaller
+// constants for smaller block sizes, where the range of the 'sum' is
+// restricted to fewer bits.
 
 #define DC_MULTIPLIER_1X2 0x5556
 #define DC_MULTIPLIER_1X4 0x3334
diff --git a/third_party/aom/aom_dsp/noise_model.h b/third_party/aom/aom_dsp/noise_model.h
index 049d5be151..5e7de9bf2c 100644
--- a/third_party/aom/aom_dsp/noise_model.h
+++ b/third_party/aom/aom_dsp/noise_model.h
@@ -158,10 +158,10 @@ int aom_flat_block_finder_run(const aom_flat_block_finder_t *block_finder,
                               int stride, uint8_t *flat_blocks);
 
 // The noise shape indicates the allowed coefficients in the AR model.
-typedef enum {
+enum {
   AOM_NOISE_SHAPE_DIAMOND = 0,
   AOM_NOISE_SHAPE_SQUARE = 1
-} aom_noise_shape;
+} UENUM1BYTE(aom_noise_shape);
 
 // The parameters of the noise model include the shape type, lag, the
 // bit depth of the input images provided, and whether the input images
@@ -202,13 +202,13 @@ typedef struct {
 } aom_noise_model_t;
 
 /*!\brief Result of a noise model update. */
-typedef enum {
+enum {
   AOM_NOISE_STATUS_OK = 0,
   AOM_NOISE_STATUS_INVALID_ARGUMENT,
   AOM_NOISE_STATUS_INSUFFICIENT_FLAT_BLOCKS,
   AOM_NOISE_STATUS_DIFFERENT_NOISE_TYPE,
   AOM_NOISE_STATUS_INTERNAL_ERROR,
-} aom_noise_status_t;
+} UENUM1BYTE(aom_noise_status_t);
 
 /*!\brief Initializes a noise model with the given parameters.
  *
diff --git a/third_party/aom/aom_dsp/prob.h b/third_party/aom/aom_dsp/prob.h
index d003a986e6..20ffdeaeaf 100644
--- a/third_party/aom/aom_dsp/prob.h
+++ b/third_party/aom/aom_dsp/prob.h
@@ -641,7 +641,7 @@ static INLINE uint8_t get_prob(unsigned int num, unsigned int den) {
   }
 }
 
-static INLINE void update_cdf(aom_cdf_prob *cdf, int val, int nsymbs) {
+static INLINE void update_cdf(aom_cdf_prob *cdf, int8_t val, int nsymbs) {
   int rate;
   int i, tmp;
 
diff --git a/third_party/aom/aom_dsp/quantize.c b/third_party/aom/aom_dsp/quantize.c
index 62dbd86a92..4acbf20939 100644
--- a/third_party/aom/aom_dsp/quantize.c
+++ b/third_party/aom/aom_dsp/quantize.c
@@ -12,6 +12,68 @@
 #include "aom_dsp/quantize.h"
 #include "aom_mem/aom_mem.h"
 
+void quantize_b_adaptive_helper_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
+    const qm_val_t *iqm_ptr, const int log_scale) {
+  const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0], log_scale),
+                         ROUND_POWER_OF_TWO(zbin_ptr[1], log_scale) };
+  const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
+  int i, non_zero_count = (int)n_coeffs, eob = -1;
+  (void)iscan;
+
+  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+  // Pre-scan pass
+  for (i = (int)n_coeffs - 1; i >= 0; i--) {
+    const int rc = scan[i];
+    const qm_val_t wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
+    const int coeff = coeff_ptr[rc] * wt;
+
+    int prescan_add = ROUND_POWER_OF_TWO(dequant_ptr[rc != 0] * 325, 7);
+    if (coeff < (zbins[rc != 0] * (1 << AOM_QM_BITS) + prescan_add) &&
+        coeff > (nzbins[rc != 0] * (1 << AOM_QM_BITS) - prescan_add))
+      non_zero_count--;
+    else
+      break;
+  }
+
+  // Quantization pass: All coefficients with index >= zero_flag are
+  // skippable. Note: zero_flag can be zero.
+  for (i = 0; i < non_zero_count; i++) {
+    const int rc = scan[i];
+    const int coeff = coeff_ptr[rc];
+    const int coeff_sign = (coeff >> 31);
+    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+    int tmp32;
+
+    const qm_val_t wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
+    if (abs_coeff * wt >= (zbins[rc != 0] << AOM_QM_BITS)) {
+      int64_t tmp =
+          clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale),
+                INT16_MIN, INT16_MAX);
+      tmp *= wt;
+      tmp32 = (int)(((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+                     quant_shift_ptr[rc != 0]) >>
+                    (16 - log_scale + AOM_QM_BITS));  // quantization
+      qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+      const int iwt = iqm_ptr != NULL ? iqm_ptr[rc] : (1 << AOM_QM_BITS);
+      const int dequant =
+          (dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
+          AOM_QM_BITS;
+      const tran_low_t abs_dqcoeff = (tmp32 * dequant) >> log_scale;
+      dqcoeff_ptr[rc] = (tran_low_t)((abs_dqcoeff ^ coeff_sign) - coeff_sign);
+
+      if (tmp32) eob = i;
+    }
+  }
+  *eob_ptr = eob + 1;
+}
+
 void quantize_b_helper_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          const int16_t *zbin_ptr, const int16_t *round_ptr,
                          const int16_t *quant_ptr,
@@ -74,6 +136,64 @@ void quantize_b_helper_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
   *eob_ptr = eob + 1;
 }
 
+void highbd_quantize_b_adaptive_helper_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
+    const qm_val_t *iqm_ptr, const int log_scale) {
+  int i, eob = -1;
+  const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0], log_scale),
+                         ROUND_POWER_OF_TWO(zbin_ptr[1], log_scale) };
+  const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
+  int dequant;
+  int idx_arr[4096];
+  (void)iscan;
+  int idx = 0;
+
+  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+  // Pre-scan pass
+  for (i = 0; i < n_coeffs; i++) {
+    const int rc = scan[i];
+    const qm_val_t wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
+    const int coeff = coeff_ptr[rc] * wt;
+
+    // If the coefficient is out of the base ZBIN range, keep it for
+    // quantization.
+    int prescan_add = ROUND_POWER_OF_TWO(dequant_ptr[rc != 0] * 325, 7);
+    if (coeff >= (zbins[rc != 0] * (1 << AOM_QM_BITS) + prescan_add) &&
+        coeff <= (nzbins[rc != 0] * (1 << AOM_QM_BITS) - prescan_add))
+      idx_arr[idx++] = i;
+  }
+
+  // Quantization pass: only process the coefficients selected in
+  // pre-scan pass. Note: idx can be zero.
+  for (i = 0; i < idx; i++) {
+    const int rc = scan[idx_arr[i]];
+    const int coeff = coeff_ptr[rc];
+    const int coeff_sign = (coeff >> 31);
+    const qm_val_t wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
+    const qm_val_t iwt = iqm_ptr != NULL ? iqm_ptr[rc] : (1 << AOM_QM_BITS);
+    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+    const int64_t tmp1 =
+        abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale);
+    const int64_t tmpw = tmp1 * wt;
+    const int64_t tmp2 = ((tmpw * quant_ptr[rc != 0]) >> 16) + tmpw;
+    const int abs_qcoeff = (int)((tmp2 * quant_shift_ptr[rc != 0]) >>
+                                 (16 - log_scale + AOM_QM_BITS));
+    qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+    dequant =
+        (dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+    const tran_low_t abs_dqcoeff = (abs_qcoeff * dequant) >> log_scale;
+    dqcoeff_ptr[rc] = (tran_low_t)((abs_dqcoeff ^ coeff_sign) - coeff_sign);
+    if (abs_qcoeff) eob = idx_arr[i];
+  }
+  *eob_ptr = eob + 1;
+}
+
 void highbd_quantize_b_helper_c(
     const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
     const int16_t *round_ptr, const int16_t *quant_ptr,
@@ -133,6 +253,80 @@ void highbd_quantize_b_helper_c(
 
 /* These functions should only be called when quantisation matrices
    are not used. */
+void aom_quantize_b_adaptive_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                               const int16_t *zbin_ptr,
+                               const int16_t *round_ptr,
+                               const int16_t *quant_ptr,
+                               const int16_t *quant_shift_ptr,
+                               tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                               const int16_t *dequant_ptr, uint16_t *eob_ptr,
+                               const int16_t *scan, const int16_t *iscan) {
+  quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                               quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                               dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+                               NULL, NULL, 0);
+}
+
+void aom_quantize_b_32x32_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan) {
+  quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                               quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                               dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+                               NULL, NULL, 1);
+}
+
+void aom_quantize_b_64x64_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan) {
+  quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                               quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                               dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+                               NULL, NULL, 2);
+}
+
+void aom_highbd_quantize_b_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan) {
+  highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                                      quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                                      dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+                                      iscan, NULL, NULL, 0);
+}
+
+void aom_highbd_quantize_b_32x32_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan) {
+  highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                                      quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                                      dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+                                      iscan, NULL, NULL, 1);
+}
+
+void aom_highbd_quantize_b_64x64_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan) {
+  highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+                                      quant_ptr, quant_shift_ptr, qcoeff_ptr,
+                                      dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+                                      iscan, NULL, NULL, 2);
+}
+
 void aom_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                       const int16_t *zbin_ptr, const int16_t *round_ptr,
                       const int16_t *quant_ptr, const int16_t *quant_shift_ptr,
diff --git a/third_party/aom/aom_dsp/quantize.h b/third_party/aom/aom_dsp/quantize.h
index c55ab234e7..43c30ee090 100644
--- a/third_party/aom/aom_dsp/quantize.h
+++ b/third_party/aom/aom_dsp/quantize.h
@@ -20,6 +20,66 @@
 extern "C" {
 #endif
 
+void quantize_b_adaptive_helper_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
+    const qm_val_t *iqm_ptr, const int log_scale);
+
+void aom_quantize_b_adaptive_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                               const int16_t *zbin_ptr,
+                               const int16_t *round_ptr,
+                               const int16_t *quant_ptr,
+                               const int16_t *quant_shift_ptr,
+                               tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                               const int16_t *dequant_ptr, uint16_t *eob_ptr,
+                               const int16_t *scan, const int16_t *iscan);
+
+void aom_quantize_b_32x32_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan);
+
+void aom_quantize_b_64x64_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan);
+
+void highbd_quantize_b_adaptive_helper_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
+    const qm_val_t *iqm_ptr, const int log_scale);
+
+void aom_highbd_quantize_b_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan);
+
+void aom_highbd_quantize_b_32x32_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan);
+
+void aom_highbd_quantize_b_64x64_adaptive_c(
+    const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan);
+
 void quantize_b_helper_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          const int16_t *zbin_ptr, const int16_t *round_ptr,
                          const int16_t *quant_ptr,
diff --git a/third_party/aom/aom_dsp/recenter.h b/third_party/aom/aom_dsp/recenter.h
new file mode 100644
index 0000000000..b3fd412907
--- /dev/null
+++ b/third_party/aom/aom_dsp/recenter.h
@@ -0,0 +1,61 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_AOM_DSP_RECENTER_H_
+#define AOM_AOM_DSP_RECENTER_H_
+
+#include "config/aom_config.h"
+
+#include "aom/aom_integer.h"
+
+// Inverse recenters a non-negative literal v around a reference r
+static INLINE uint16_t inv_recenter_nonneg(uint16_t r, uint16_t v) {
+  if (v > (r << 1))
+    return v;
+  else if ((v & 1) == 0)
+    return (v >> 1) + r;
+  else
+    return r - ((v + 1) >> 1);
+}
+
+// Inverse recenters a non-negative literal v in [0, n-1] around a
+// reference r also in [0, n-1]
+static INLINE uint16_t inv_recenter_finite_nonneg(uint16_t n, uint16_t r,
+                                                  uint16_t v) {
+  if ((r << 1) <= n) {
+    return inv_recenter_nonneg(r, v);
+  } else {
+    return n - 1 - inv_recenter_nonneg(n - 1 - r, v);
+  }
+}
+
+// Recenters a non-negative literal v around a reference r
+static INLINE uint16_t recenter_nonneg(uint16_t r, uint16_t v) {
+  if (v > (r << 1))
+    return v;
+  else if (v >= r)
+    return ((v - r) << 1);
+  else
+    return ((r - v) << 1) - 1;
+}
+
+// Recenters a non-negative literal v in [0, n-1] around a
+// reference r also in [0, n-1]
+static INLINE uint16_t recenter_finite_nonneg(uint16_t n, uint16_t r,
+                                              uint16_t v) {
+  if ((r << 1) <= n) {
+    return recenter_nonneg(r, v);
+  } else {
+    return recenter_nonneg(n - 1 - r, n - 1 - v);
+  }
+}
+
+#endif  // AOM_AOM_DSP_RECENTER_H_
diff --git a/third_party/aom/aom_dsp/sad.c b/third_party/aom/aom_dsp/sad.c
index 1e24df4a56..9169e78cd1 100644
--- a/third_party/aom/aom_dsp/sad.c
+++ b/third_party/aom/aom_dsp/sad.c
@@ -25,7 +25,9 @@ static INLINE unsigned int sad(const uint8_t *a, int a_stride, const uint8_t *b,
   unsigned int sad = 0;
 
   for (y = 0; y < height; y++) {
-    for (x = 0; x < width; x++) sad += abs(a[x] - b[x]);
+    for (x = 0; x < width; x++) {
+      sad += abs(a[x] - b[x]);
+    }
 
     a += a_stride;
     b += b_stride;
@@ -52,12 +54,12 @@ static INLINE unsigned int sad(const uint8_t *a, int a_stride, const uint8_t *b,
     aom_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride);         \
     return sad(src, src_stride, comp_pred, m, m, n);                          \
   }                                                                           \
-  unsigned int aom_jnt_sad##m##x##n##_avg_c(                                  \
+  unsigned int aom_dist_wtd_sad##m##x##n##_avg_c(                             \
       const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {         \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
     uint8_t comp_pred[m * n];                                                 \
-    aom_jnt_comp_avg_pred_c(comp_pred, second_pred, m, n, ref, ref_stride,    \
-                            jcp_param);                                       \
+    aom_dist_wtd_comp_avg_pred_c(comp_pred, second_pred, m, n, ref,           \
+                                 ref_stride, jcp_param);                      \
     return sad(src, src_stride, comp_pred, m, m, n);                          \
   }
 
@@ -67,75 +69,75 @@ static INLINE unsigned int sad(const uint8_t *a, int a_stride, const uint8_t *b,
                                const uint8_t *const ref_array[],           \
                                int ref_stride, uint32_t *sad_array) {      \
     int i;                                                                 \
-    for (i = 0; i < 4; ++i)                                                \
+    for (i = 0; i < 4; ++i) {                                              \
       sad_array[i] =                                                       \
           aom_sad##m##x##n##_c(src, src_stride, ref_array[i], ref_stride); \
+    }                                                                      \
   }
 
-/* clang-format off */
 // 128x128
-sadMxN(128, 128)
-sadMxNx4D(128, 128)
+sadMxN(128, 128);
+sadMxNx4D(128, 128);
 
 // 128x64
-sadMxN(128, 64)
-sadMxNx4D(128, 64)
+sadMxN(128, 64);
+sadMxNx4D(128, 64);
 
 // 64x128
-sadMxN(64, 128)
-sadMxNx4D(64, 128)
+sadMxN(64, 128);
+sadMxNx4D(64, 128);
 
 // 64x64
-sadMxN(64, 64)
-sadMxNx4D(64, 64)
+sadMxN(64, 64);
+sadMxNx4D(64, 64);
 
 // 64x32
-sadMxN(64, 32)
-sadMxNx4D(64, 32)
+sadMxN(64, 32);
+sadMxNx4D(64, 32);
 
 // 32x64
-sadMxN(32, 64)
-sadMxNx4D(32, 64)
+sadMxN(32, 64);
+sadMxNx4D(32, 64);
 
 // 32x32
-sadMxN(32, 32)
-sadMxNx4D(32, 32)
+sadMxN(32, 32);
+sadMxNx4D(32, 32);
 
 // 32x16
-sadMxN(32, 16)
-sadMxNx4D(32, 16)
+sadMxN(32, 16);
+sadMxNx4D(32, 16);
 
 // 16x32
-sadMxN(16, 32)
-sadMxNx4D(16, 32)
+sadMxN(16, 32);
+sadMxNx4D(16, 32);
 
 // 16x16
-sadMxN(16, 16)
-sadMxNx4D(16, 16)
+sadMxN(16, 16);
+sadMxNx4D(16, 16);
 
 // 16x8
-sadMxN(16, 8)
-sadMxNx4D(16, 8)
+sadMxN(16, 8);
+sadMxNx4D(16, 8);
 
 // 8x16
-sadMxN(8, 16)
-sadMxNx4D(8, 16)
+sadMxN(8, 16);
+sadMxNx4D(8, 16);
 
 // 8x8
-sadMxN(8, 8)
-sadMxNx4D(8, 8)
+sadMxN(8, 8);
+sadMxNx4D(8, 8);
 
 // 8x4
-sadMxN(8, 4)
-sadMxNx4D(8, 4)
+sadMxN(8, 4);
+sadMxNx4D(8, 4);
 
 // 4x8
-sadMxN(4, 8)
-sadMxNx4D(4, 8)
+sadMxN(4, 8);
+sadMxNx4D(4, 8);
 
 // 4x4
-sadMxN(4, 4)
-sadMxNx4D(4, 4)
+sadMxN(4, 4);
+sadMxNx4D(4, 4);
 
 sadMxh(128);
 sadMxh(64);
@@ -144,30 +146,30 @@ sadMxh(16);
 sadMxh(8);
 sadMxh(4);
 
-sadMxN(4, 16)
-sadMxNx4D(4, 16)
-sadMxN(16, 4)
-sadMxNx4D(16, 4)
-sadMxN(8, 32)
-sadMxNx4D(8, 32)
-sadMxN(32, 8)
-sadMxNx4D(32, 8)
-sadMxN(16, 64)
-sadMxNx4D(16, 64)
-sadMxN(64, 16)
-sadMxNx4D(64, 16)
+sadMxN(4, 16);
+sadMxNx4D(4, 16);
+sadMxN(16, 4);
+sadMxNx4D(16, 4);
+sadMxN(8, 32);
+sadMxNx4D(8, 32);
+sadMxN(32, 8);
+sadMxNx4D(32, 8);
+sadMxN(16, 64);
+sadMxNx4D(16, 64);
+sadMxN(64, 16);
+sadMxNx4D(64, 16);
 
-    /* clang-format on */
-
-    static INLINE
-    unsigned int highbd_sad(const uint8_t *a8, int a_stride, const uint8_t *b8,
-                            int b_stride, int width, int height) {
+static INLINE unsigned int highbd_sad(const uint8_t *a8, int a_stride,
+                                      const uint8_t *b8, int b_stride,
+                                      int width, int height) {
   int y, x;
   unsigned int sad = 0;
   const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
   const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
   for (y = 0; y < height; y++) {
-    for (x = 0; x < width; x++) sad += abs(a[x] - b[x]);
+    for (x = 0; x < width; x++) {
+      sad += abs(a[x] - b[x]);
+    }
 
     a += a_stride;
     b += b_stride;
@@ -182,7 +184,9 @@ static INLINE unsigned int highbd_sadb(const uint8_t *a8, int a_stride,
   unsigned int sad = 0;
   const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
   for (y = 0; y < height; y++) {
-    for (x = 0; x < width; x++) sad += abs(a[x] - b[x]);
+    for (x = 0; x < width; x++) {
+      sad += abs(a[x] - b[x]);
+    }
 
     a += a_stride;
     b += b_stride;
@@ -204,12 +208,13 @@ static INLINE unsigned int highbd_sadb(const uint8_t *a8, int a_stride,
                              ref, ref_stride);                                 \
     return highbd_sadb(src, src_stride, comp_pred, m, m, n);                   \
   }                                                                            \
-  unsigned int aom_highbd_jnt_sad##m##x##n##_avg_c(                            \
+  unsigned int aom_highbd_dist_wtd_sad##m##x##n##_avg_c(                       \
       const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride,  \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {          \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {     \
     uint16_t comp_pred[m * n];                                                 \
-    aom_highbd_jnt_comp_avg_pred(CONVERT_TO_BYTEPTR(comp_pred), second_pred,   \
-                                 m, n, ref, ref_stride, jcp_param);            \
+    aom_highbd_dist_wtd_comp_avg_pred(CONVERT_TO_BYTEPTR(comp_pred),           \
+                                      second_pred, m, n, ref, ref_stride,      \
+                                      jcp_param);                              \
     return highbd_sadb(src, src_stride, comp_pred, m, m, n);                   \
   }
 
@@ -224,81 +229,79 @@ static INLINE unsigned int highbd_sadb(const uint8_t *a8, int a_stride,
     }                                                                        \
   }
 
-/* clang-format off */
 // 128x128
-highbd_sadMxN(128, 128)
-highbd_sadMxNx4D(128, 128)
+highbd_sadMxN(128, 128);
+highbd_sadMxNx4D(128, 128);
 
 // 128x64
-highbd_sadMxN(128, 64)
-highbd_sadMxNx4D(128, 64)
+highbd_sadMxN(128, 64);
+highbd_sadMxNx4D(128, 64);
 
 // 64x128
-highbd_sadMxN(64, 128)
-highbd_sadMxNx4D(64, 128)
+highbd_sadMxN(64, 128);
+highbd_sadMxNx4D(64, 128);
 
 // 64x64
-highbd_sadMxN(64, 64)
-highbd_sadMxNx4D(64, 64)
+highbd_sadMxN(64, 64);
+highbd_sadMxNx4D(64, 64);
 
 // 64x32
-highbd_sadMxN(64, 32)
-highbd_sadMxNx4D(64, 32)
+highbd_sadMxN(64, 32);
+highbd_sadMxNx4D(64, 32);
 
 // 32x64
-highbd_sadMxN(32, 64)
-highbd_sadMxNx4D(32, 64)
+highbd_sadMxN(32, 64);
+highbd_sadMxNx4D(32, 64);
 
 // 32x32
-highbd_sadMxN(32, 32)
-highbd_sadMxNx4D(32, 32)
+highbd_sadMxN(32, 32);
+highbd_sadMxNx4D(32, 32);
 
 // 32x16
-highbd_sadMxN(32, 16)
-highbd_sadMxNx4D(32, 16)
+highbd_sadMxN(32, 16);
+highbd_sadMxNx4D(32, 16);
 
 // 16x32
-highbd_sadMxN(16, 32)
-highbd_sadMxNx4D(16, 32)
+highbd_sadMxN(16, 32);
+highbd_sadMxNx4D(16, 32);
 
 // 16x16
-highbd_sadMxN(16, 16)
-highbd_sadMxNx4D(16, 16)
+highbd_sadMxN(16, 16);
+highbd_sadMxNx4D(16, 16);
 
 // 16x8
-highbd_sadMxN(16, 8)
-highbd_sadMxNx4D(16, 8)
+highbd_sadMxN(16, 8);
+highbd_sadMxNx4D(16, 8);
 
 // 8x16
-highbd_sadMxN(8, 16)
-highbd_sadMxNx4D(8, 16)
+highbd_sadMxN(8, 16);
+highbd_sadMxNx4D(8, 16);
 
 // 8x8
-highbd_sadMxN(8, 8)
-highbd_sadMxNx4D(8, 8)
+highbd_sadMxN(8, 8);
+highbd_sadMxNx4D(8, 8);
 
 // 8x4
-highbd_sadMxN(8, 4)
-highbd_sadMxNx4D(8, 4)
+highbd_sadMxN(8, 4);
+highbd_sadMxNx4D(8, 4);
 
 // 4x8
-highbd_sadMxN(4, 8)
-highbd_sadMxNx4D(4, 8)
+highbd_sadMxN(4, 8);
+highbd_sadMxNx4D(4, 8);
 
 // 4x4
-highbd_sadMxN(4, 4)
-highbd_sadMxNx4D(4, 4)
+highbd_sadMxN(4, 4);
+highbd_sadMxNx4D(4, 4);
 
-highbd_sadMxN(4, 16)
-highbd_sadMxNx4D(4, 16)
-highbd_sadMxN(16, 4)
-highbd_sadMxNx4D(16, 4)
-highbd_sadMxN(8, 32)
-highbd_sadMxNx4D(8, 32)
-highbd_sadMxN(32, 8)
-highbd_sadMxNx4D(32, 8)
-highbd_sadMxN(16, 64)
-highbd_sadMxNx4D(16, 64)
-highbd_sadMxN(64, 16)
-highbd_sadMxNx4D(64, 16)
-    /* clang-format on */
+highbd_sadMxN(4, 16);
+highbd_sadMxNx4D(4, 16);
+highbd_sadMxN(16, 4);
+highbd_sadMxNx4D(16, 4);
+highbd_sadMxN(8, 32);
+highbd_sadMxNx4D(8, 32);
+highbd_sadMxN(32, 8);
+highbd_sadMxNx4D(32, 8);
+highbd_sadMxN(16, 64);
+highbd_sadMxNx4D(16, 64);
+highbd_sadMxN(64, 16);
+highbd_sadMxNx4D(64, 16);
diff --git a/third_party/aom/aom_dsp/txfm_common.h b/third_party/aom/aom_dsp/txfm_common.h
index f98242840c..f13d69092d 100644
--- a/third_party/aom/aom_dsp/txfm_common.h
+++ b/third_party/aom/aom_dsp/txfm_common.h
@@ -39,7 +39,7 @@ typedef struct txfm_param {
 // Constants:
 //  for (int i = 1; i< 32; ++i)
 //    printf("static const int cospi_%d_64 = %.0f;\n", i,
-//           round(16384 * cos(i*M_PI/64)));
+//           round(16384 * cos(i*PI/64)));
 // Note: sin(k*Pi/64) = cos((32-k)*Pi/64)
 static const tran_high_t cospi_1_64 = 16364;
 static const tran_high_t cospi_2_64 = 16305;
diff --git a/third_party/aom/aom_dsp/variance.c b/third_party/aom/aom_dsp/variance.c
index 23b7153098..18a33c5467 100644
--- a/third_party/aom/aom_dsp/variance.c
+++ b/third_party/aom/aom_dsp/variance.c
@@ -164,40 +164,40 @@ void aom_var_filter_block2d_bil_second_pass_c(const uint16_t *a, uint8_t *b,
     return aom_variance##W##x##H##_c(temp2, W, b, b_stride, sse);             \
   }
 
-#define SUBPIX_AVG_VAR(W, H)                                                  \
-  uint32_t aom_sub_pixel_avg_variance##W##x##H##_c(                           \
-      const uint8_t *a, int a_stride, int xoffset, int yoffset,               \
-      const uint8_t *b, int b_stride, uint32_t *sse,                          \
-      const uint8_t *second_pred) {                                           \
-    uint16_t fdata3[(H + 1) * W];                                             \
-    uint8_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                               \
-                                                                              \
-    aom_var_filter_block2d_bil_first_pass_c(a, fdata3, a_stride, 1, H + 1, W, \
-                                            bilinear_filters_2t[xoffset]);    \
-    aom_var_filter_block2d_bil_second_pass_c(fdata3, temp2, W, W, H, W,       \
-                                             bilinear_filters_2t[yoffset]);   \
-                                                                              \
-    aom_comp_avg_pred(temp3, second_pred, W, H, temp2, W);                    \
-                                                                              \
-    return aom_variance##W##x##H##_c(temp3, W, b, b_stride, sse);             \
-  }                                                                           \
-  uint32_t aom_jnt_sub_pixel_avg_variance##W##x##H##_c(                       \
-      const uint8_t *a, int a_stride, int xoffset, int yoffset,               \
-      const uint8_t *b, int b_stride, uint32_t *sse,                          \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {         \
-    uint16_t fdata3[(H + 1) * W];                                             \
-    uint8_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                               \
-                                                                              \
-    aom_var_filter_block2d_bil_first_pass_c(a, fdata3, a_stride, 1, H + 1, W, \
-                                            bilinear_filters_2t[xoffset]);    \
-    aom_var_filter_block2d_bil_second_pass_c(fdata3, temp2, W, W, H, W,       \
-                                             bilinear_filters_2t[yoffset]);   \
-                                                                              \
-    aom_jnt_comp_avg_pred(temp3, second_pred, W, H, temp2, W, jcp_param);     \
-                                                                              \
-    return aom_variance##W##x##H(temp3, W, b, b_stride, sse);                 \
+#define SUBPIX_AVG_VAR(W, H)                                                   \
+  uint32_t aom_sub_pixel_avg_variance##W##x##H##_c(                            \
+      const uint8_t *a, int a_stride, int xoffset, int yoffset,                \
+      const uint8_t *b, int b_stride, uint32_t *sse,                           \
+      const uint8_t *second_pred) {                                            \
+    uint16_t fdata3[(H + 1) * W];                                              \
+    uint8_t temp2[H * W];                                                      \
+    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                                \
+                                                                               \
+    aom_var_filter_block2d_bil_first_pass_c(a, fdata3, a_stride, 1, H + 1, W,  \
+                                            bilinear_filters_2t[xoffset]);     \
+    aom_var_filter_block2d_bil_second_pass_c(fdata3, temp2, W, W, H, W,        \
+                                             bilinear_filters_2t[yoffset]);    \
+                                                                               \
+    aom_comp_avg_pred(temp3, second_pred, W, H, temp2, W);                     \
+                                                                               \
+    return aom_variance##W##x##H##_c(temp3, W, b, b_stride, sse);              \
+  }                                                                            \
+  uint32_t aom_dist_wtd_sub_pixel_avg_variance##W##x##H##_c(                   \
+      const uint8_t *a, int a_stride, int xoffset, int yoffset,                \
+      const uint8_t *b, int b_stride, uint32_t *sse,                           \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {     \
+    uint16_t fdata3[(H + 1) * W];                                              \
+    uint8_t temp2[H * W];                                                      \
+    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                                \
+                                                                               \
+    aom_var_filter_block2d_bil_first_pass_c(a, fdata3, a_stride, 1, H + 1, W,  \
+                                            bilinear_filters_2t[xoffset]);     \
+    aom_var_filter_block2d_bil_second_pass_c(fdata3, temp2, W, W, H, W,        \
+                                             bilinear_filters_2t[yoffset]);    \
+                                                                               \
+    aom_dist_wtd_comp_avg_pred(temp3, second_pred, W, H, temp2, W, jcp_param); \
+                                                                               \
+    return aom_variance##W##x##H(temp3, W, b, b_stride, sse);                  \
   }
 
 /* Identical to the variance call except it takes an additional parameter, sum,
@@ -291,7 +291,7 @@ void aom_upsampled_pred_c(MACROBLOCKD *xd, const AV1_COMMON *const cm,
     const int ref_num = 0;
     const int is_intrabc = is_intrabc_block(mi);
     const struct scale_factors *const sf =
-        is_intrabc ? &cm->sf_identity : &xd->block_refs[ref_num]->sf;
+        is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref_num];
     const int is_scaled = av1_is_scaled(sf);
 
     if (is_scaled) {
@@ -368,10 +368,7 @@ void aom_upsampled_pred_c(MACROBLOCKD *xd, const AV1_COMMON *const cm,
     }
   }
 
-  const InterpFilterParams *filter =
-      (subpel_search == 1)
-          ? av1_get_4tap_interp_filter_params(EIGHTTAP_REGULAR)
-          : av1_get_interp_filter_params_with_block_size(EIGHTTAP_REGULAR, 8);
+  const InterpFilterParams *filter = av1_get_filter(subpel_search);
 
   if (!subpel_x_q3 && !subpel_y_q3) {
     for (int i = 0; i < height; i++) {
@@ -427,9 +424,10 @@ void aom_comp_avg_upsampled_pred_c(MACROBLOCKD *xd, const AV1_COMMON *const cm,
   }
 }
 
-void aom_jnt_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
-                             int height, const uint8_t *ref, int ref_stride,
-                             const JNT_COMP_PARAMS *jcp_param) {
+void aom_dist_wtd_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred,
+                                  int width, int height, const uint8_t *ref,
+                                  int ref_stride,
+                                  const DIST_WTD_COMP_PARAMS *jcp_param) {
   int i, j;
   const int fwd_offset = jcp_param->fwd_offset;
   const int bck_offset = jcp_param->bck_offset;
@@ -446,11 +444,11 @@ void aom_jnt_comp_avg_pred_c(uint8_t *comp_pred, const uint8_t *pred, int width,
   }
 }
 
-void aom_jnt_comp_avg_upsampled_pred_c(
+void aom_dist_wtd_comp_avg_upsampled_pred_c(
     MACROBLOCKD *xd, const AV1_COMMON *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
-    int ref_stride, const JNT_COMP_PARAMS *jcp_param, int subpel_search) {
+    int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search) {
   int i, j;
   const int fwd_offset = jcp_param->fwd_offset;
   const int bck_offset = jcp_param->bck_offset;
@@ -691,125 +689,128 @@ void aom_highbd_var_filter_block2d_bil_second_pass(
                                                dst, dst_stride, sse);        \
   }
 
-#define HIGHBD_SUBPIX_AVG_VAR(W, H)                                            \
-  uint32_t aom_highbd_8_sub_pixel_avg_variance##W##x##H##_c(                   \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred) {                                            \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,   \
-                               CONVERT_TO_BYTEPTR(temp2), W);                  \
-                                                                               \
-    return aom_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,    \
-                                              dst, dst_stride, sse);           \
-  }                                                                            \
-                                                                               \
-  uint32_t aom_highbd_10_sub_pixel_avg_variance##W##x##H##_c(                  \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred) {                                            \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,   \
-                               CONVERT_TO_BYTEPTR(temp2), W);                  \
-                                                                               \
-    return aom_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,   \
-                                               dst, dst_stride, sse);          \
-  }                                                                            \
-                                                                               \
-  uint32_t aom_highbd_12_sub_pixel_avg_variance##W##x##H##_c(                  \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred) {                                            \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,   \
-                               CONVERT_TO_BYTEPTR(temp2), W);                  \
-                                                                               \
-    return aom_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,   \
-                                               dst, dst_stride, sse);          \
-  }                                                                            \
-                                                                               \
-  uint32_t aom_highbd_8_jnt_sub_pixel_avg_variance##W##x##H##_c(               \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {          \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_jnt_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H, \
-                                 CONVERT_TO_BYTEPTR(temp2), W, jcp_param);     \
-                                                                               \
-    return aom_highbd_8_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst,   \
-                                          dst_stride, sse);                    \
-  }                                                                            \
-                                                                               \
-  uint32_t aom_highbd_10_jnt_sub_pixel_avg_variance##W##x##H##_c(              \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {          \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_jnt_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H, \
-                                 CONVERT_TO_BYTEPTR(temp2), W, jcp_param);     \
-                                                                               \
-    return aom_highbd_10_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst,  \
-                                           dst_stride, sse);                   \
-  }                                                                            \
-                                                                               \
-  uint32_t aom_highbd_12_jnt_sub_pixel_avg_variance##W##x##H##_c(              \
-      const uint8_t *src, int src_stride, int xoffset, int yoffset,            \
-      const uint8_t *dst, int dst_stride, uint32_t *sse,                       \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {          \
-    uint16_t fdata3[(H + 1) * W];                                              \
-    uint16_t temp2[H * W];                                                     \
-    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                               \
-                                                                               \
-    aom_highbd_var_filter_block2d_bil_first_pass(                              \
-        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
-    aom_highbd_var_filter_block2d_bil_second_pass(                             \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);              \
-                                                                               \
-    aom_highbd_jnt_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H, \
-                                 CONVERT_TO_BYTEPTR(temp2), W, jcp_param);     \
-                                                                               \
-    return aom_highbd_12_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst,  \
-                                           dst_stride, sse);                   \
+#define HIGHBD_SUBPIX_AVG_VAR(W, H)                                           \
+  uint32_t aom_highbd_8_sub_pixel_avg_variance##W##x##H##_c(                  \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred) {                                           \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,  \
+                               CONVERT_TO_BYTEPTR(temp2), W);                 \
+                                                                              \
+    return aom_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,   \
+                                              dst, dst_stride, sse);          \
+  }                                                                           \
+                                                                              \
+  uint32_t aom_highbd_10_sub_pixel_avg_variance##W##x##H##_c(                 \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred) {                                           \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,  \
+                               CONVERT_TO_BYTEPTR(temp2), W);                 \
+                                                                              \
+    return aom_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,  \
+                                               dst, dst_stride, sse);         \
+  }                                                                           \
+                                                                              \
+  uint32_t aom_highbd_12_sub_pixel_avg_variance##W##x##H##_c(                 \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred) {                                           \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_comp_avg_pred_c(CONVERT_TO_BYTEPTR(temp3), second_pred, W, H,  \
+                               CONVERT_TO_BYTEPTR(temp2), W);                 \
+                                                                              \
+    return aom_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W,  \
+                                               dst, dst_stride, sse);         \
+  }                                                                           \
+                                                                              \
+  uint32_t aom_highbd_8_dist_wtd_sub_pixel_avg_variance##W##x##H##_c(         \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_dist_wtd_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, \
+                                      W, H, CONVERT_TO_BYTEPTR(temp2), W,     \
+                                      jcp_param);                             \
+                                                                              \
+    return aom_highbd_8_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst,  \
+                                          dst_stride, sse);                   \
+  }                                                                           \
+                                                                              \
+  uint32_t aom_highbd_10_dist_wtd_sub_pixel_avg_variance##W##x##H##_c(        \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_dist_wtd_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, \
+                                      W, H, CONVERT_TO_BYTEPTR(temp2), W,     \
+                                      jcp_param);                             \
+                                                                              \
+    return aom_highbd_10_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst, \
+                                           dst_stride, sse);                  \
+  }                                                                           \
+                                                                              \
+  uint32_t aom_highbd_12_dist_wtd_sub_pixel_avg_variance##W##x##H##_c(        \
+      const uint8_t *src, int src_stride, int xoffset, int yoffset,           \
+      const uint8_t *dst, int dst_stride, uint32_t *sse,                      \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
+    uint16_t fdata3[(H + 1) * W];                                             \
+    uint16_t temp2[H * W];                                                    \
+    DECLARE_ALIGNED(16, uint16_t, temp3[H * W]);                              \
+                                                                              \
+    aom_highbd_var_filter_block2d_bil_first_pass(                             \
+        src, fdata3, src_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);  \
+    aom_highbd_var_filter_block2d_bil_second_pass(                            \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);             \
+                                                                              \
+    aom_highbd_dist_wtd_comp_avg_pred(CONVERT_TO_BYTEPTR(temp3), second_pred, \
+                                      W, H, CONVERT_TO_BYTEPTR(temp2), W,     \
+                                      jcp_param);                             \
+                                                                              \
+    return aom_highbd_12_variance##W##x##H(CONVERT_TO_BYTEPTR(temp3), W, dst, \
+                                           dst_stride, sse);                  \
   }
 
 /* All three forms of the variance are available in the same sizes. */
@@ -883,7 +884,7 @@ void aom_highbd_upsampled_pred_c(MACROBLOCKD *xd,
     const int ref_num = 0;
     const int is_intrabc = is_intrabc_block(mi);
     const struct scale_factors *const sf =
-        is_intrabc ? &cm->sf_identity : &xd->block_refs[ref_num]->sf;
+        is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref_num];
     const int is_scaled = av1_is_scaled(sf);
 
     if (is_scaled) {
@@ -959,10 +960,7 @@ void aom_highbd_upsampled_pred_c(MACROBLOCKD *xd,
     }
   }
 
-  const InterpFilterParams *filter =
-      (subpel_search == 1)
-          ? av1_get_4tap_interp_filter_params(EIGHTTAP_REGULAR)
-          : av1_get_interp_filter_params_with_block_size(EIGHTTAP_REGULAR, 8);
+  const InterpFilterParams *filter = av1_get_filter(subpel_search);
 
   if (!subpel_x_q3 && !subpel_y_q3) {
     const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
@@ -975,13 +973,13 @@ void aom_highbd_upsampled_pred_c(MACROBLOCKD *xd,
   } else if (!subpel_y_q3) {
     const int16_t *const kernel =
         av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
-    aom_highbd_convolve8_horiz(ref8, ref_stride, comp_pred8, width, kernel, 16,
-                               NULL, -1, width, height, bd);
+    aom_highbd_convolve8_horiz_c(ref8, ref_stride, comp_pred8, width, kernel,
+                                 16, NULL, -1, width, height, bd);
   } else if (!subpel_x_q3) {
     const int16_t *const kernel =
         av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
-    aom_highbd_convolve8_vert(ref8, ref_stride, comp_pred8, width, NULL, -1,
-                              kernel, 16, width, height, bd);
+    aom_highbd_convolve8_vert_c(ref8, ref_stride, comp_pred8, width, NULL, -1,
+                                kernel, 16, width, height, bd);
   } else {
     DECLARE_ALIGNED(16, uint16_t,
                     temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
@@ -992,11 +990,11 @@ void aom_highbd_upsampled_pred_c(MACROBLOCKD *xd,
     const int intermediate_height =
         (((height - 1) * 8 + subpel_y_q3) >> 3) + filter->taps;
     assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
-    aom_highbd_convolve8_horiz(ref8 - ref_stride * ((filter->taps >> 1) - 1),
-                               ref_stride, CONVERT_TO_BYTEPTR(temp),
-                               MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
-                               intermediate_height, bd);
-    aom_highbd_convolve8_vert(
+    aom_highbd_convolve8_horiz_c(ref8 - ref_stride * ((filter->taps >> 1) - 1),
+                                 ref_stride, CONVERT_TO_BYTEPTR(temp),
+                                 MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
+                                 intermediate_height, bd);
+    aom_highbd_convolve8_vert_c(
         CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter->taps >> 1) - 1)),
         MAX_SB_SIZE, comp_pred8, width, NULL, -1, kernel_y, 16, width, height,
         bd);
@@ -1024,10 +1022,10 @@ void aom_highbd_comp_avg_upsampled_pred_c(
   }
 }
 
-void aom_highbd_jnt_comp_avg_pred_c(uint8_t *comp_pred8, const uint8_t *pred8,
-                                    int width, int height, const uint8_t *ref8,
-                                    int ref_stride,
-                                    const JNT_COMP_PARAMS *jcp_param) {
+void aom_highbd_dist_wtd_comp_avg_pred_c(
+    uint8_t *comp_pred8, const uint8_t *pred8, int width, int height,
+    const uint8_t *ref8, int ref_stride,
+    const DIST_WTD_COMP_PARAMS *jcp_param) {
   int i, j;
   const int fwd_offset = jcp_param->fwd_offset;
   const int bck_offset = jcp_param->bck_offset;
@@ -1047,20 +1045,20 @@ void aom_highbd_jnt_comp_avg_pred_c(uint8_t *comp_pred8, const uint8_t *pred8,
   }
 }
 
-void aom_highbd_jnt_comp_avg_upsampled_pred_c(
+void aom_highbd_dist_wtd_comp_avg_upsampled_pred_c(
     MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
-    int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param,
+    int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param,
     int subpel_search) {
   int i, j;
   const int fwd_offset = jcp_param->fwd_offset;
   const int bck_offset = jcp_param->bck_offset;
   const uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
   uint16_t *comp_pred = CONVERT_TO_SHORTPTR(comp_pred8);
-  aom_highbd_upsampled_pred(xd, cm, mi_row, mi_col, mv, comp_pred8, width,
-                            height, subpel_x_q3, subpel_y_q3, ref8, ref_stride,
-                            bd, subpel_search);
+  aom_highbd_upsampled_pred_c(xd, cm, mi_row, mi_col, mv, comp_pred8, width,
+                              height, subpel_x_q3, subpel_y_q3, ref8,
+                              ref_stride, bd, subpel_search);
 
   for (i = 0; i < height; i++) {
     for (j = 0; j < width; j++) {
diff --git a/third_party/aom/aom_dsp/variance.h b/third_party/aom/aom_dsp/variance.h
index 362da29d33..4550c17b34 100644
--- a/third_party/aom/aom_dsp/variance.h
+++ b/third_party/aom/aom_dsp/variance.h
@@ -50,15 +50,14 @@ typedef unsigned int (*aom_subp_avg_variance_fn_t)(
     const uint8_t *a, int a_stride, int xoffset, int yoffset, const uint8_t *b,
     int b_stride, unsigned int *sse, const uint8_t *second_pred);
 
-typedef unsigned int (*aom_jnt_sad_avg_fn_t)(const uint8_t *a, int a_stride,
-                                             const uint8_t *b, int b_stride,
-                                             const uint8_t *second_pred,
-                                             const JNT_COMP_PARAMS *jcp_param);
+typedef unsigned int (*aom_dist_wtd_sad_avg_fn_t)(
+    const uint8_t *a, int a_stride, const uint8_t *b, int b_stride,
+    const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param);
 
-typedef unsigned int (*aom_jnt_subp_avg_variance_fn_t)(
+typedef unsigned int (*aom_dist_wtd_subp_avg_variance_fn_t)(
     const uint8_t *a, int a_stride, int xoffset, int yoffset, const uint8_t *b,
     int b_stride, unsigned int *sse, const uint8_t *second_pred,
-    const JNT_COMP_PARAMS *jcp_param);
+    const DIST_WTD_COMP_PARAMS *jcp_param);
 
 typedef unsigned int (*aom_masked_sad_fn_t)(const uint8_t *src, int src_stride,
                                             const uint8_t *ref, int ref_stride,
@@ -101,8 +100,8 @@ typedef struct aom_variance_vtable {
   aom_obmc_sad_fn_t osdf;
   aom_obmc_variance_fn_t ovf;
   aom_obmc_subpixvariance_fn_t osvf;
-  aom_jnt_sad_avg_fn_t jsdaf;
-  aom_jnt_subp_avg_variance_fn_t jsvaf;
+  aom_dist_wtd_sad_avg_fn_t jsdaf;
+  aom_dist_wtd_subp_avg_variance_fn_t jsvaf;
 } aom_variance_fn_ptr_t;
 
 void aom_highbd_var_filter_block2d_bil_first_pass(
diff --git a/third_party/aom/aom_dsp/x86/aom_asm_stubs.c b/third_party/aom/aom_dsp/x86/aom_asm_stubs.c
index 5f5bf5f14e..3bf7b5597a 100644
--- a/third_party/aom/aom_dsp/x86/aom_asm_stubs.c
+++ b/third_party/aom/aom_dsp/x86/aom_asm_stubs.c
@@ -21,13 +21,13 @@ filter8_1dfunction aom_filter_block1d8_v8_sse2;
 filter8_1dfunction aom_filter_block1d8_h8_sse2;
 filter8_1dfunction aom_filter_block1d4_v8_sse2;
 filter8_1dfunction aom_filter_block1d4_h8_sse2;
+filter8_1dfunction aom_filter_block1d16_v4_sse2;
+filter8_1dfunction aom_filter_block1d16_h4_sse2;
 
-#define aom_filter_block1d16_h4_sse2 aom_filter_block1d16_h8_sse2
-#define aom_filter_block1d16_v4_sse2 aom_filter_block1d16_v8_sse2
-#define aom_filter_block1d8_h4_sse2 aom_filter_block1d8_h8_sse2
-#define aom_filter_block1d8_v4_sse2 aom_filter_block1d8_v8_sse2
-#define aom_filter_block1d4_h4_sse2 aom_filter_block1d4_h8_sse2
-#define aom_filter_block1d4_v4_sse2 aom_filter_block1d4_v8_sse2
+filter8_1dfunction aom_filter_block1d8_h4_sse2;
+filter8_1dfunction aom_filter_block1d8_v4_sse2;
+filter8_1dfunction aom_filter_block1d4_h4_sse2;
+filter8_1dfunction aom_filter_block1d4_v4_sse2;
 
 filter8_1dfunction aom_filter_block1d16_v2_sse2;
 filter8_1dfunction aom_filter_block1d16_h2_sse2;
@@ -49,7 +49,6 @@ filter8_1dfunction aom_filter_block1d4_h2_sse2;
 FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
 FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
 
-#if ARCH_X86_64
 highbd_filter8_1dfunction aom_highbd_filter_block1d16_v8_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d16_h8_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d8_v8_sse2;
@@ -57,6 +56,13 @@ highbd_filter8_1dfunction aom_highbd_filter_block1d8_h8_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d4_v8_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d4_h8_sse2;
 
+highbd_filter8_1dfunction aom_highbd_filter_block1d16_v4_sse2;
+highbd_filter8_1dfunction aom_highbd_filter_block1d16_h4_sse2;
+highbd_filter8_1dfunction aom_highbd_filter_block1d8_v4_sse2;
+highbd_filter8_1dfunction aom_highbd_filter_block1d8_h4_sse2;
+highbd_filter8_1dfunction aom_highbd_filter_block1d4_v4_sse2;
+highbd_filter8_1dfunction aom_highbd_filter_block1d4_h4_sse2;
+
 highbd_filter8_1dfunction aom_highbd_filter_block1d16_v2_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d16_h2_sse2;
 highbd_filter8_1dfunction aom_highbd_filter_block1d8_v2_sse2;
@@ -85,5 +91,4 @@ highbd_filter8_1dfunction aom_highbd_filter_block1d4_h2_sse2;
 HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
 HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
 
-#endif  // ARCH_X86_64
 #endif  // HAVE_SSE2
diff --git a/third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm b/third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
index 7b3fe6419a..a7152be57c 100644
--- a/third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/aom_high_subpixel_bilinear_sse2.asm
@@ -67,7 +67,6 @@
     dec         rcx
 %endm
 
-%if ARCH_X86_64
 %macro HIGH_GET_PARAM 0
     mov         rdx, arg(5)                 ;filter ptr
     mov         rsi, arg(0)                 ;src_ptr
@@ -86,14 +85,17 @@
 
     mov         rdx, 0x00010001
     movsxd      rcx, DWORD PTR arg(6)       ;bps
-    movq        xmm8, rdx
+    movq        xmm3, rdx
     movq        xmm5, rcx
-    pshufd      xmm8, xmm8, 0b
-    movdqa      xmm1, xmm8
-    psllw       xmm8, xmm5
-    psubw       xmm8, xmm1                  ;max value (for clamping)
+    pshufd      xmm3, xmm3, 0b
+    movdqa      xmm1, xmm3
+    psllw       xmm3, xmm5
+    psubw       xmm3, xmm1                  ;max value (for clamping)
     pxor        xmm5, xmm5                  ;min value (for clamping)
 
+    movdqa      max, xmm3
+    movdqa      min, xmm5
+
     movsxd      rax, DWORD PTR arg(1)       ;pixels_per_line
     movsxd      rdx, DWORD PTR arg(3)       ;out_pitch
     movsxd      rcx, DWORD PTR arg(4)       ;output_height
@@ -113,8 +115,8 @@
     packssdw    xmm0, xmm6                  ;pack back to word
 
     ;clamp the values
-    pminsw      xmm0, xmm8
-    pmaxsw      xmm0, xmm5
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
 
 %if %1
     movdqu      xmm1, [rdi]
@@ -128,36 +130,36 @@
 %endm
 
 %macro HIGH_APPLY_FILTER_16 1
-    movdqa      xmm9, xmm0
+    movdqa      xmm5, xmm0
     movdqa      xmm6, xmm2
-    punpckhwd   xmm9, xmm1
+    punpckhwd   xmm5, xmm1
     punpckhwd   xmm6, xmm3
     punpcklwd   xmm0, xmm1
     punpcklwd   xmm2, xmm3
 
-    pmaddwd     xmm9, xmm7
+    pmaddwd     xmm5, xmm7
     pmaddwd     xmm6, xmm7
     pmaddwd     xmm0, xmm7
     pmaddwd     xmm2, xmm7
 
-    paddd       xmm9, xmm4                  ;rounding
+    paddd       xmm5, xmm4                  ;rounding
     paddd       xmm6, xmm4
     paddd       xmm0, xmm4
     paddd       xmm2, xmm4
 
-    psrad       xmm9, 7                     ;shift
+    psrad       xmm5, 7                     ;shift
     psrad       xmm6, 7
     psrad       xmm0, 7
     psrad       xmm2, 7
 
-    packssdw    xmm0, xmm9                  ;pack back to word
+    packssdw    xmm0, xmm5                  ;pack back to word
     packssdw    xmm2, xmm6                  ;pack back to word
 
     ;clamp the values
-    pminsw      xmm0, xmm8
-    pmaxsw      xmm0, xmm5
-    pminsw      xmm2, xmm8
-    pmaxsw      xmm2, xmm5
+    pminsw      xmm0, max
+    pmaxsw      xmm0, min
+    pminsw      xmm2, max
+    pmaxsw      xmm2, min
 
 %if %1
     movdqu      xmm1, [rdi]
@@ -172,7 +174,6 @@
     lea         rdi, [rdi + 2*rdx]
     dec         rcx
 %endm
-%endif
 
 SECTION .text
 
@@ -200,7 +201,6 @@ sym(aom_highbd_filter_block1d4_v2_sse2):
     pop         rbp
     ret
 
-%if ARCH_X86_64
 global sym(aom_highbd_filter_block1d8_v2_sse2) PRIVATE
 sym(aom_highbd_filter_block1d8_v2_sse2):
     push        rbp
@@ -211,6 +211,11 @@ sym(aom_highbd_filter_block1d8_v2_sse2):
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 2
+    %define max [rsp + 16 * 0]
+    %define min [rsp + 16 * 1]
+
     HIGH_GET_PARAM
 .loop:
     movdqu      xmm0, [rsi]                 ;0
@@ -219,6 +224,9 @@ sym(aom_highbd_filter_block1d8_v2_sse2):
     HIGH_APPLY_FILTER_8 0
     jnz         .loop
 
+    add rsp, 16 * 2
+    pop rsp
+
     ; begin epilog
     pop         rdi
     pop         rsi
@@ -237,6 +245,11 @@ sym(aom_highbd_filter_block1d16_v2_sse2):
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 2
+    %define max [rsp + 16 * 0]
+    %define min [rsp + 16 * 1]
+
     HIGH_GET_PARAM
 .loop:
     movdqu        xmm0, [rsi]               ;0
@@ -247,6 +260,9 @@ sym(aom_highbd_filter_block1d16_v2_sse2):
     HIGH_APPLY_FILTER_16 0
     jnz         .loop
 
+    add rsp, 16 * 2
+    pop rsp
+
     ; begin epilog
     pop         rdi
     pop         rsi
@@ -254,7 +270,6 @@ sym(aom_highbd_filter_block1d16_v2_sse2):
     UNSHADOW_ARGS
     pop         rbp
     ret
-%endif
 
 global sym(aom_highbd_filter_block1d4_h2_sse2) PRIVATE
 sym(aom_highbd_filter_block1d4_h2_sse2):
@@ -281,7 +296,6 @@ sym(aom_highbd_filter_block1d4_h2_sse2):
     pop         rbp
     ret
 
-%if ARCH_X86_64
 global sym(aom_highbd_filter_block1d8_h2_sse2) PRIVATE
 sym(aom_highbd_filter_block1d8_h2_sse2):
     push        rbp
@@ -292,6 +306,11 @@ sym(aom_highbd_filter_block1d8_h2_sse2):
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 2
+    %define max [rsp + 16 * 0]
+    %define min [rsp + 16 * 1]
+
     HIGH_GET_PARAM
 .loop:
     movdqu      xmm0, [rsi]                 ;load src
@@ -300,6 +319,9 @@ sym(aom_highbd_filter_block1d8_h2_sse2):
     HIGH_APPLY_FILTER_8 0
     jnz         .loop
 
+    add rsp, 16 * 2
+    pop rsp
+
     ; begin epilog
     pop         rdi
     pop         rsi
@@ -318,6 +340,11 @@ sym(aom_highbd_filter_block1d16_h2_sse2):
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 2
+    %define max [rsp + 16 * 0]
+    %define min [rsp + 16 * 1]
+
     HIGH_GET_PARAM
 .loop:
     movdqu      xmm0,   [rsi]               ;load src
@@ -328,6 +355,9 @@ sym(aom_highbd_filter_block1d16_h2_sse2):
     HIGH_APPLY_FILTER_16 0
     jnz         .loop
 
+    add rsp, 16 * 2
+    pop rsp
+
     ; begin epilog
     pop         rdi
     pop         rsi
@@ -335,4 +365,3 @@ sym(aom_highbd_filter_block1d16_h2_sse2):
     UNSHADOW_ARGS
     pop         rbp
     ret
-%endif
diff --git a/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c b/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c
new file mode 100644
index 0000000000..cff7f43eee
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_sse2.c
@@ -0,0 +1,569 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <emmintrin.h>  // SSE2
+
+#include "config/aom_dsp_rtcd.h"
+#include "aom_dsp/x86/convolve.h"
+#include "aom_ports/mem.h"
+
+void aom_filter_block1d16_h4_sse2(const uint8_t *src_ptr,
+                                  ptrdiff_t src_pixels_per_line,
+                                  uint8_t *output_ptr, ptrdiff_t output_pitch,
+                                  uint32_t output_height,
+                                  const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1, srcRegFilt32b1_2, srcRegFilt32b2_1,
+      srcRegFilt32b2_2;
+  __m128i srcReg32b1, srcReg32b2;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp_0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp_1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp_0, tmp_0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp_1, tmp_1);   // coeffs 4 5 4 5 4 5 4 5
+
+  for (i = output_height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    __m128i ss_2 = _mm_srli_si128(srcReg32b1, 2);
+    __m128i ss_4 = _mm_srli_si128(srcReg32b1, 4);
+    __m128i ss_1_1 = _mm_unpacklo_epi8(ss_2, _mm_setzero_si128());
+    __m128i ss_2_1 = _mm_unpacklo_epi8(ss_4, _mm_setzero_si128());
+    __m128i d1 = _mm_madd_epi16(ss_1_1, secondFilters);
+    __m128i d2 = _mm_madd_epi16(ss_2_1, thirdFilters);
+    srcRegFilt32b1_1 = _mm_add_epi32(d1, d2);
+
+    __m128i ss_1 = _mm_srli_si128(srcReg32b1, 3);
+    __m128i ss_3 = _mm_srli_si128(srcReg32b1, 5);
+    __m128i ss_1_2 = _mm_unpacklo_epi8(ss_1, _mm_setzero_si128());
+    __m128i ss_2_2 = _mm_unpacklo_epi8(ss_3, _mm_setzero_si128());
+    d1 = _mm_madd_epi16(ss_1_2, secondFilters);
+    d2 = _mm_madd_epi16(ss_2_2, thirdFilters);
+    srcRegFilt32b1_2 = _mm_add_epi32(d1, d2);
+
+    __m128i res_lo = _mm_unpacklo_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+    __m128i res_hi = _mm_unpackhi_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+    srcRegFilt32b1_1 = _mm_packs_epi32(res_lo, res_hi);
+
+    // reading stride of the next 16 bytes
+    // (part of it was being read by earlier read)
+    srcReg32b2 = _mm_loadu_si128((const __m128i *)(src_ptr + 8));
+
+    ss_2 = _mm_srli_si128(srcReg32b2, 2);
+    ss_4 = _mm_srli_si128(srcReg32b2, 4);
+    ss_1_1 = _mm_unpacklo_epi8(ss_2, _mm_setzero_si128());
+    ss_2_1 = _mm_unpacklo_epi8(ss_4, _mm_setzero_si128());
+    d1 = _mm_madd_epi16(ss_1_1, secondFilters);
+    d2 = _mm_madd_epi16(ss_2_1, thirdFilters);
+    srcRegFilt32b2_1 = _mm_add_epi32(d1, d2);
+
+    ss_1 = _mm_srli_si128(srcReg32b2, 3);
+    ss_3 = _mm_srli_si128(srcReg32b2, 5);
+    ss_1_2 = _mm_unpacklo_epi8(ss_1, _mm_setzero_si128());
+    ss_2_2 = _mm_unpacklo_epi8(ss_3, _mm_setzero_si128());
+    d1 = _mm_madd_epi16(ss_1_2, secondFilters);
+    d2 = _mm_madd_epi16(ss_2_2, thirdFilters);
+    srcRegFilt32b2_2 = _mm_add_epi32(d1, d2);
+
+    res_lo = _mm_unpacklo_epi32(srcRegFilt32b2_1, srcRegFilt32b2_2);
+    res_hi = _mm_unpackhi_epi32(srcRegFilt32b2_1, srcRegFilt32b2_2);
+    srcRegFilt32b2_1 = _mm_packs_epi32(res_lo, res_hi);
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b2_1 = _mm_adds_epi16(srcRegFilt32b2_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+    srcRegFilt32b2_1 = _mm_srai_epi16(srcRegFilt32b2_1, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve result
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, srcRegFilt32b2_1);
+
+    src_ptr += src_pixels_per_line;
+
+    _mm_store_si128((__m128i *)output_ptr, srcRegFilt32b1_1);
+
+    output_ptr += output_pitch;
+  }
+}
+
+void aom_filter_block1d16_v4_sse2(const uint8_t *src_ptr, ptrdiff_t src_pitch,
+                                  uint8_t *output_ptr, ptrdiff_t out_pitch,
+                                  uint32_t output_height,
+                                  const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23_lo, srcReg23_hi, srcReg34_lo, srcReg34_hi;
+  __m128i srcReg45_lo, srcReg45_hi, srcReg56_lo, srcReg56_hi;
+  __m128i resReg23_lo, resReg34_lo, resReg45_lo, resReg56_lo;
+  __m128i resReg23_hi, resReg34_hi, resReg45_hi, resReg56_hi;
+  __m128i resReg23_45_lo, resReg34_56_lo, resReg23_45_hi, resReg34_56_hi;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg32, secondFilters, thirdFilters;
+  __m128i tmp_0, tmp_1;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp0, tmp0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp1, tmp1);   // coeffs 4 5 4 5 4 5 4 5
+
+  // multiply the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23_lo = _mm_unpacklo_epi8(srcReg2, srcReg3);
+  srcReg23_hi = _mm_unpackhi_epi8(srcReg2, srcReg3);
+  __m128i resReg23_lo_1 = _mm_unpacklo_epi8(srcReg23_lo, _mm_setzero_si128());
+  __m128i resReg23_lo_2 = _mm_unpackhi_epi8(srcReg23_lo, _mm_setzero_si128());
+  __m128i resReg23_hi_1 = _mm_unpacklo_epi8(srcReg23_hi, _mm_setzero_si128());
+  __m128i resReg23_hi_2 = _mm_unpackhi_epi8(srcReg23_hi, _mm_setzero_si128());
+
+  srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+  srcReg34_lo = _mm_unpacklo_epi8(srcReg3, srcReg4);
+  srcReg34_hi = _mm_unpackhi_epi8(srcReg3, srcReg4);
+  __m128i resReg34_lo_1 = _mm_unpacklo_epi8(srcReg34_lo, _mm_setzero_si128());
+  __m128i resReg34_lo_2 = _mm_unpackhi_epi8(srcReg34_lo, _mm_setzero_si128());
+  __m128i resReg34_hi_1 = _mm_unpacklo_epi8(srcReg34_hi, _mm_setzero_si128());
+  __m128i resReg34_hi_2 = _mm_unpackhi_epi8(srcReg34_hi, _mm_setzero_si128());
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+
+    srcReg45_lo = _mm_unpacklo_epi8(srcReg4, srcReg5);
+    srcReg45_hi = _mm_unpackhi_epi8(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+
+    srcReg56_lo = _mm_unpacklo_epi8(srcReg5, srcReg6);
+    srcReg56_hi = _mm_unpackhi_epi8(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    tmp_0 = _mm_madd_epi16(resReg23_lo_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg23_lo_2, secondFilters);
+    resReg23_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    tmp_0 = _mm_madd_epi16(resReg34_lo_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg34_lo_2, secondFilters);
+    resReg34_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg45_lo_1 = _mm_unpacklo_epi8(srcReg45_lo, _mm_setzero_si128());
+    __m128i resReg45_lo_2 = _mm_unpackhi_epi8(srcReg45_lo, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg45_lo_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg45_lo_2, thirdFilters);
+    resReg45_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg56_lo_1 = _mm_unpacklo_epi8(srcReg56_lo, _mm_setzero_si128());
+    __m128i resReg56_lo_2 = _mm_unpackhi_epi8(srcReg56_lo, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg56_lo_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg56_lo_2, thirdFilters);
+    resReg56_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    // add and saturate the results together
+    resReg23_45_lo = _mm_adds_epi16(resReg23_lo, resReg45_lo);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_lo, resReg56_lo);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    tmp_0 = _mm_madd_epi16(resReg23_hi_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg23_hi_2, secondFilters);
+    resReg23_hi = _mm_packs_epi32(tmp_0, tmp_1);
+
+    tmp_0 = _mm_madd_epi16(resReg34_hi_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg34_hi_2, secondFilters);
+    resReg34_hi = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg45_hi_1 = _mm_unpacklo_epi8(srcReg45_hi, _mm_setzero_si128());
+    __m128i resReg45_hi_2 = _mm_unpackhi_epi8(srcReg45_hi, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg45_hi_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg45_hi_2, thirdFilters);
+    resReg45_hi = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg56_hi_1 = _mm_unpacklo_epi8(srcReg56_hi, _mm_setzero_si128());
+    __m128i resReg56_hi_2 = _mm_unpackhi_epi8(srcReg56_hi, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg56_hi_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg56_hi_2, thirdFilters);
+    resReg56_hi = _mm_packs_epi32(tmp_0, tmp_1);
+
+    // add and saturate the results together
+    resReg23_45_hi = _mm_adds_epi16(resReg23_hi, resReg45_hi);
+    resReg34_56_hi = _mm_adds_epi16(resReg34_hi, resReg56_hi);
+
+    // shift by 6 bit each 16 bit
+    resReg23_45_lo = _mm_adds_epi16(resReg23_45_lo, addFilterReg32);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_56_lo, addFilterReg32);
+    resReg23_45_hi = _mm_adds_epi16(resReg23_45_hi, addFilterReg32);
+    resReg34_56_hi = _mm_adds_epi16(resReg34_56_hi, addFilterReg32);
+    resReg23_45_lo = _mm_srai_epi16(resReg23_45_lo, 6);
+    resReg34_56_lo = _mm_srai_epi16(resReg34_56_lo, 6);
+    resReg23_45_hi = _mm_srai_epi16(resReg23_45_hi, 6);
+    resReg34_56_hi = _mm_srai_epi16(resReg34_56_hi, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packus_epi16(resReg23_45_lo, resReg23_45_hi);
+    resReg34_56 = _mm_packus_epi16(resReg34_56_lo, resReg34_56_hi);
+
+    src_ptr += src_stride;
+
+    _mm_store_si128((__m128i *)output_ptr, (resReg23_45));
+    _mm_store_si128((__m128i *)(output_ptr + out_pitch), (resReg34_56));
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    resReg23_lo_1 = resReg45_lo_1;
+    resReg23_lo_2 = resReg45_lo_2;
+    resReg23_hi_1 = resReg45_hi_1;
+    resReg23_hi_2 = resReg45_hi_2;
+    resReg34_lo_1 = resReg56_lo_1;
+    resReg34_lo_2 = resReg56_lo_2;
+    resReg34_hi_1 = resReg56_hi_1;
+    resReg34_hi_2 = resReg56_hi_2;
+    srcReg4 = srcReg6;
+  }
+}
+
+void aom_filter_block1d8_h4_sse2(const uint8_t *src_ptr,
+                                 ptrdiff_t src_pixels_per_line,
+                                 uint8_t *output_ptr, ptrdiff_t output_pitch,
+                                 uint32_t output_height,
+                                 const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1, srcRegFilt32b1_2;
+  __m128i srcReg32b1;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp_0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp_1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp_0, tmp_0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp_1, tmp_1);   // coeffs 4 5 4 5 4 5 4 5
+
+  for (i = output_height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    __m128i ss_2 = _mm_srli_si128(srcReg32b1, 2);
+    __m128i ss_4 = _mm_srli_si128(srcReg32b1, 4);
+    ss_2 = _mm_unpacklo_epi8(ss_2, _mm_setzero_si128());
+    ss_4 = _mm_unpacklo_epi8(ss_4, _mm_setzero_si128());
+    __m128i d1 = _mm_madd_epi16(ss_2, secondFilters);
+    __m128i d2 = _mm_madd_epi16(ss_4, thirdFilters);
+    srcRegFilt32b1_1 = _mm_add_epi32(d1, d2);
+
+    __m128i ss_3 = _mm_srli_si128(srcReg32b1, 3);
+    __m128i ss_5 = _mm_srli_si128(srcReg32b1, 5);
+    ss_3 = _mm_unpacklo_epi8(ss_3, _mm_setzero_si128());
+    ss_5 = _mm_unpacklo_epi8(ss_5, _mm_setzero_si128());
+    d1 = _mm_madd_epi16(ss_3, secondFilters);
+    d2 = _mm_madd_epi16(ss_5, thirdFilters);
+    srcRegFilt32b1_2 = _mm_add_epi32(d1, d2);
+
+    __m128i res_lo = _mm_unpacklo_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+    __m128i res_hi = _mm_unpackhi_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+    srcRegFilt32b1_1 = _mm_packs_epi32(res_lo, res_hi);
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve result
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    src_ptr += src_pixels_per_line;
+
+    _mm_storel_epi64((__m128i *)output_ptr, srcRegFilt32b1_1);
+
+    output_ptr += output_pitch;
+  }
+}
+
+void aom_filter_block1d8_v4_sse2(const uint8_t *src_ptr, ptrdiff_t src_pitch,
+                                 uint8_t *output_ptr, ptrdiff_t out_pitch,
+                                 uint32_t output_height,
+                                 const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23_lo, srcReg34_lo;
+  __m128i srcReg45_lo, srcReg56_lo;
+  __m128i resReg23_lo, resReg34_lo, resReg45_lo, resReg56_lo;
+  __m128i resReg23_45_lo, resReg34_56_lo;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg32, secondFilters, thirdFilters;
+  __m128i tmp_0, tmp_1;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp0, tmp0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp1, tmp1);   // coeffs 4 5 4 5 4 5 4 5
+
+  // multiply the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23_lo = _mm_unpacklo_epi8(srcReg2, srcReg3);
+  __m128i resReg23_lo_1 = _mm_unpacklo_epi8(srcReg23_lo, _mm_setzero_si128());
+  __m128i resReg23_lo_2 = _mm_unpackhi_epi8(srcReg23_lo, _mm_setzero_si128());
+
+  srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+  srcReg34_lo = _mm_unpacklo_epi8(srcReg3, srcReg4);
+  __m128i resReg34_lo_1 = _mm_unpacklo_epi8(srcReg34_lo, _mm_setzero_si128());
+  __m128i resReg34_lo_2 = _mm_unpackhi_epi8(srcReg34_lo, _mm_setzero_si128());
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+    srcReg45_lo = _mm_unpacklo_epi8(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+    srcReg56_lo = _mm_unpacklo_epi8(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    tmp_0 = _mm_madd_epi16(resReg23_lo_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg23_lo_2, secondFilters);
+    resReg23_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    tmp_0 = _mm_madd_epi16(resReg34_lo_1, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg34_lo_2, secondFilters);
+    resReg34_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg45_lo_1 = _mm_unpacklo_epi8(srcReg45_lo, _mm_setzero_si128());
+    __m128i resReg45_lo_2 = _mm_unpackhi_epi8(srcReg45_lo, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg45_lo_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg45_lo_2, thirdFilters);
+    resReg45_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg56_lo_1 = _mm_unpacklo_epi8(srcReg56_lo, _mm_setzero_si128());
+    __m128i resReg56_lo_2 = _mm_unpackhi_epi8(srcReg56_lo, _mm_setzero_si128());
+    tmp_0 = _mm_madd_epi16(resReg56_lo_1, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg56_lo_2, thirdFilters);
+    resReg56_lo = _mm_packs_epi32(tmp_0, tmp_1);
+
+    // add and saturate the results together
+    resReg23_45_lo = _mm_adds_epi16(resReg23_lo, resReg45_lo);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_lo, resReg56_lo);
+
+    // shift by 6 bit each 16 bit
+    resReg23_45_lo = _mm_adds_epi16(resReg23_45_lo, addFilterReg32);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_56_lo, addFilterReg32);
+    resReg23_45_lo = _mm_srai_epi16(resReg23_45_lo, 6);
+    resReg34_56_lo = _mm_srai_epi16(resReg34_56_lo, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packus_epi16(resReg23_45_lo, _mm_setzero_si128());
+    resReg34_56 = _mm_packus_epi16(resReg34_56_lo, _mm_setzero_si128());
+
+    src_ptr += src_stride;
+
+    _mm_storel_epi64((__m128i *)output_ptr, (resReg23_45));
+    _mm_storel_epi64((__m128i *)(output_ptr + out_pitch), (resReg34_56));
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    resReg23_lo_1 = resReg45_lo_1;
+    resReg23_lo_2 = resReg45_lo_2;
+    resReg34_lo_1 = resReg56_lo_1;
+    resReg34_lo_2 = resReg56_lo_2;
+    srcReg4 = srcReg6;
+  }
+}
+
+void aom_filter_block1d4_h4_sse2(const uint8_t *src_ptr,
+                                 ptrdiff_t src_pixels_per_line,
+                                 uint8_t *output_ptr, ptrdiff_t output_pitch,
+                                 uint32_t output_height,
+                                 const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1;
+  __m128i srcReg32b1;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp_0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp_1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp_0, tmp_0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp_1, tmp_1);   // coeffs 4 5 4 5 4 5 4 5
+
+  for (i = output_height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    __m128i ss_2 = _mm_srli_si128(srcReg32b1, 2);
+    __m128i ss_3 = _mm_srli_si128(srcReg32b1, 3);
+    __m128i ss_4 = _mm_srli_si128(srcReg32b1, 4);
+    __m128i ss_5 = _mm_srli_si128(srcReg32b1, 5);
+
+    ss_2 = _mm_unpacklo_epi8(ss_2, _mm_setzero_si128());
+    ss_3 = _mm_unpacklo_epi8(ss_3, _mm_setzero_si128());
+    ss_4 = _mm_unpacklo_epi8(ss_4, _mm_setzero_si128());
+    ss_5 = _mm_unpacklo_epi8(ss_5, _mm_setzero_si128());
+
+    __m128i ss_1_1 = _mm_unpacklo_epi32(ss_2, ss_3);
+    __m128i ss_1_2 = _mm_unpacklo_epi32(ss_4, ss_5);
+
+    __m128i d1 = _mm_madd_epi16(ss_1_1, secondFilters);
+    __m128i d2 = _mm_madd_epi16(ss_1_2, thirdFilters);
+    srcRegFilt32b1_1 = _mm_add_epi32(d1, d2);
+
+    srcRegFilt32b1_1 = _mm_packs_epi32(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve result
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    src_ptr += src_pixels_per_line;
+
+    *((uint32_t *)(output_ptr)) = _mm_cvtsi128_si32(srcRegFilt32b1_1);
+
+    output_ptr += output_pitch;
+  }
+}
+
+void aom_filter_block1d4_v4_sse2(const uint8_t *src_ptr, ptrdiff_t src_pitch,
+                                 uint8_t *output_ptr, ptrdiff_t out_pitch,
+                                 uint32_t output_height,
+                                 const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23, srcReg34, srcReg45, srcReg56;
+  __m128i resReg23_34, resReg45_56;
+  __m128i resReg23_34_45_56;
+  __m128i addFilterReg32, secondFilters, thirdFilters;
+  __m128i tmp_0, tmp_1;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp0, tmp0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp1, tmp1);   // coeffs 4 5 4 5 4 5 4 5
+
+  // multiply the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23 = _mm_unpacklo_epi8(srcReg2, srcReg3);
+  __m128i resReg23 = _mm_unpacklo_epi8(srcReg23, _mm_setzero_si128());
+
+  srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
+  srcReg34 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+  __m128i resReg34 = _mm_unpacklo_epi8(srcReg34, _mm_setzero_si128());
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
+    srcReg45 = _mm_unpacklo_epi8(srcReg4, srcReg5);
+    srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+    srcReg56 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    tmp_0 = _mm_madd_epi16(resReg23, secondFilters);
+    tmp_1 = _mm_madd_epi16(resReg34, secondFilters);
+    resReg23_34 = _mm_packs_epi32(tmp_0, tmp_1);
+
+    __m128i resReg45 = _mm_unpacklo_epi8(srcReg45, _mm_setzero_si128());
+    __m128i resReg56 = _mm_unpacklo_epi8(srcReg56, _mm_setzero_si128());
+
+    tmp_0 = _mm_madd_epi16(resReg45, thirdFilters);
+    tmp_1 = _mm_madd_epi16(resReg56, thirdFilters);
+    resReg45_56 = _mm_packs_epi32(tmp_0, tmp_1);
+
+    // add and saturate the results together
+    resReg23_34_45_56 = _mm_adds_epi16(resReg23_34, resReg45_56);
+
+    // shift by 6 bit each 16 bit
+    resReg23_34_45_56 = _mm_adds_epi16(resReg23_34_45_56, addFilterReg32);
+    resReg23_34_45_56 = _mm_srai_epi16(resReg23_34_45_56, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_34_45_56 =
+        _mm_packus_epi16(resReg23_34_45_56, _mm_setzero_si128());
+
+    src_ptr += src_stride;
+
+    *((uint32_t *)(output_ptr)) = _mm_cvtsi128_si32(resReg23_34_45_56);
+    *((uint32_t *)(output_ptr + out_pitch)) =
+        _mm_cvtsi128_si32(_mm_srli_si128(resReg23_34_45_56, 4));
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    resReg23 = resReg45;
+    resReg34 = resReg56;
+    srcReg4 = srcReg6;
+  }
+}
diff --git a/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c b/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
index 325a21b761..79324f5457 100644
--- a/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/aom_subpixel_8t_intrin_ssse3.c
@@ -45,11 +45,153 @@ DECLARE_ALIGNED(16, static const uint8_t, filt4_global[16]) = {
   6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
 };
 
+DECLARE_ALIGNED(32, static const uint8_t, filt_h4[]) = {
+  0,  1,  1,  2,  2, 3,  3,  4,  4,  5,  5,  6,  6,  7,  7,  8,  0,  1,  1,
+  2,  2,  3,  3,  4, 4,  5,  5,  6,  6,  7,  7,  8,  2,  3,  3,  4,  4,  5,
+  5,  6,  6,  7,  7, 8,  8,  9,  9,  10, 2,  3,  3,  4,  4,  5,  5,  6,  6,
+  7,  7,  8,  8,  9, 9,  10, 4,  5,  5,  6,  6,  7,  7,  8,  8,  9,  9,  10,
+  10, 11, 11, 12, 4, 5,  5,  6,  6,  7,  7,  8,  8,  9,  9,  10, 10, 11, 11,
+  12, 6,  7,  7,  8, 8,  9,  9,  10, 10, 11, 11, 12, 12, 13, 13, 14, 6,  7,
+  7,  8,  8,  9,  9, 10, 10, 11, 11, 12, 12, 13, 13, 14
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filtd4[]) = {
+  2, 3, 4, 5, 3, 4, 5, 6, 4, 5, 6, 7, 5, 6, 7, 8,
+  2, 3, 4, 5, 3, 4, 5, 6, 4, 5, 6, 7, 5, 6, 7, 8,
+};
+
 // These are reused by the avx2 intrinsics.
 filter8_1dfunction aom_filter_block1d8_v8_intrin_ssse3;
 filter8_1dfunction aom_filter_block1d8_h8_intrin_ssse3;
 filter8_1dfunction aom_filter_block1d4_h8_intrin_ssse3;
 
+static void aom_filter_block1d4_h4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
+    ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32, filt1Reg, firstFilters, srcReg32b1, srcRegFilt32b1_1;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  // converting the 16 bit (short) to 8 bit (byte) and have the same data
+  // in both lanes of 128 bit register.
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi32(0x5040302u));
+  filt1Reg = _mm_load_si128((__m128i const *)(filtd4));
+
+  for (i = output_height; i > 0; i -= 1) {
+    // load the 2 strides of source
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    // filter the source buffer
+    srcRegFilt32b1_1 = _mm_shuffle_epi8(srcReg32b1, filt1Reg);
+
+    // multiply 4 adjacent elements with the filter and add the result
+    srcRegFilt32b1_1 = _mm_maddubs_epi16(srcRegFilt32b1_1, firstFilters);
+
+    srcRegFilt32b1_1 = _mm_hadds_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve result
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    src_ptr += src_pixels_per_line;
+
+    *((uint32_t *)(output_ptr)) = _mm_cvtsi128_si32(srcRegFilt32b1_1);
+    output_ptr += output_pitch;
+  }
+}
+
+static void aom_filter_block1d4_v4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
+    ptrdiff_t out_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32;
+  __m128i srcReg2, srcReg3, srcReg23, srcReg4, srcReg34, srcReg5, srcReg45,
+      srcReg6, srcReg56;
+  __m128i srcReg23_34_lo, srcReg45_56_lo;
+  __m128i srcReg2345_3456_lo, srcReg2345_3456_hi;
+  __m128i resReglo, resReghi;
+  __m128i firstFilters;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  // converting the 16 bit (short) to  8 bit (byte) and have the
+  // same data in both lanes of 128 bit register.
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi32(0x5040302u));
+
+  // multiple the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23 = _mm_unpacklo_epi32(srcReg2, srcReg3);
+
+  srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
+
+  // have consecutive loads on the same 256 register
+  srcReg34 = _mm_unpacklo_epi32(srcReg3, srcReg4);
+
+  srcReg23_34_lo = _mm_unpacklo_epi8(srcReg23, srcReg34);
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
+    srcReg45 = _mm_unpacklo_epi32(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+    srcReg56 = _mm_unpacklo_epi32(srcReg5, srcReg6);
+
+    // merge every two consecutive registers
+    srcReg45_56_lo = _mm_unpacklo_epi8(srcReg45, srcReg56);
+
+    srcReg2345_3456_lo = _mm_unpacklo_epi16(srcReg23_34_lo, srcReg45_56_lo);
+    srcReg2345_3456_hi = _mm_unpackhi_epi16(srcReg23_34_lo, srcReg45_56_lo);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    resReglo = _mm_maddubs_epi16(srcReg2345_3456_lo, firstFilters);
+    resReghi = _mm_maddubs_epi16(srcReg2345_3456_hi, firstFilters);
+
+    resReglo = _mm_hadds_epi16(resReglo, _mm_setzero_si128());
+    resReghi = _mm_hadds_epi16(resReghi, _mm_setzero_si128());
+
+    // shift by 6 bit each 16 bit
+    resReglo = _mm_adds_epi16(resReglo, addFilterReg32);
+    resReghi = _mm_adds_epi16(resReghi, addFilterReg32);
+    resReglo = _mm_srai_epi16(resReglo, 6);
+    resReghi = _mm_srai_epi16(resReghi, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReglo = _mm_packus_epi16(resReglo, resReglo);
+    resReghi = _mm_packus_epi16(resReghi, resReghi);
+
+    src_ptr += src_stride;
+
+    *((uint32_t *)(output_ptr)) = _mm_cvtsi128_si32(resReglo);
+    *((uint32_t *)(output_ptr + out_pitch)) = _mm_cvtsi128_si32(resReghi);
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    srcReg23_34_lo = srcReg45_56_lo;
+    srcReg4 = srcReg6;
+  }
+}
+
 void aom_filter_block1d4_h8_intrin_ssse3(
     const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
     ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
@@ -118,6 +260,145 @@ void aom_filter_block1d4_h8_intrin_ssse3(
   }
 }
 
+static void aom_filter_block1d8_h4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
+    ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32, filt2Reg, filt3Reg;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1, srcRegFilt32b2, srcRegFilt32b3;
+  __m128i srcReg32b1;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  // converting the 16 bit (short) to 8 bit (byte) and have the same data
+  // in both lanes of 128 bit register.
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  // duplicate only the second 16 bits (third and forth byte)
+  // across 256 bit register
+  secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+  // duplicate only the third 16 bits (fifth and sixth byte)
+  // across 256 bit register
+  thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+
+  filt2Reg = _mm_load_si128((__m128i const *)(filt_h4 + 32));
+  filt3Reg = _mm_load_si128((__m128i const *)(filt_h4 + 32 * 2));
+
+  for (i = output_height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    // filter the source buffer
+    srcRegFilt32b3 = _mm_shuffle_epi8(srcReg32b1, filt2Reg);
+    srcRegFilt32b2 = _mm_shuffle_epi8(srcReg32b1, filt3Reg);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    srcRegFilt32b3 = _mm_maddubs_epi16(srcRegFilt32b3, secondFilters);
+    srcRegFilt32b2 = _mm_maddubs_epi16(srcRegFilt32b2, thirdFilters);
+
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b3, srcRegFilt32b2);
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+
+    // shrink to 8 bit each 16 bits
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+
+    src_ptr += src_pixels_per_line;
+
+    _mm_storel_epi64((__m128i *)output_ptr, srcRegFilt32b1_1);
+
+    output_ptr += output_pitch;
+  }
+}
+
+static void aom_filter_block1d8_v4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
+    ptrdiff_t out_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23, srcReg34, srcReg45, srcReg56;
+  __m128i resReg23, resReg34, resReg45, resReg56;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg32, secondFilters, thirdFilters;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  // converting the 16 bit (short) to  8 bit (byte) and have the
+  // same data in both lanes of 128 bit register.
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  // duplicate only the second 16 bits (third and forth byte)
+  // across 128 bit register
+  secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+  // duplicate only the third 16 bits (fifth and sixth byte)
+  // across 128 bit register
+  thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+
+  // multiple the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23 = _mm_unpacklo_epi8(srcReg2, srcReg3);
+
+  srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
+
+  // have consecutive loads on the same 256 register
+  srcReg34 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
+
+    srcReg45 = _mm_unpacklo_epi8(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+
+    srcReg56 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    resReg23 = _mm_maddubs_epi16(srcReg23, secondFilters);
+    resReg34 = _mm_maddubs_epi16(srcReg34, secondFilters);
+    resReg45 = _mm_maddubs_epi16(srcReg45, thirdFilters);
+    resReg56 = _mm_maddubs_epi16(srcReg56, thirdFilters);
+
+    // add and saturate the results together
+    resReg23_45 = _mm_adds_epi16(resReg23, resReg45);
+    resReg34_56 = _mm_adds_epi16(resReg34, resReg56);
+
+    // shift by 6 bit each 16 bit
+    resReg23_45 = _mm_adds_epi16(resReg23_45, addFilterReg32);
+    resReg34_56 = _mm_adds_epi16(resReg34_56, addFilterReg32);
+    resReg23_45 = _mm_srai_epi16(resReg23_45, 6);
+    resReg34_56 = _mm_srai_epi16(resReg34_56, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packus_epi16(resReg23_45, _mm_setzero_si128());
+    resReg34_56 = _mm_packus_epi16(resReg34_56, _mm_setzero_si128());
+
+    src_ptr += src_stride;
+
+    _mm_storel_epi64((__m128i *)output_ptr, (resReg23_45));
+    _mm_storel_epi64((__m128i *)(output_ptr + out_pitch), (resReg34_56));
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    srcReg23 = srcReg45;
+    srcReg34 = srcReg56;
+    srcReg4 = srcReg6;
+  }
+}
+
 void aom_filter_block1d8_h8_intrin_ssse3(
     const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
     ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
@@ -280,6 +561,187 @@ void aom_filter_block1d8_v8_intrin_ssse3(
   }
 }
 
+static void aom_filter_block1d16_h4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
+    ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i addFilterReg32, filt2Reg, filt3Reg;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1, srcRegFilt32b2_1, srcRegFilt32b2, srcRegFilt32b3;
+  __m128i srcReg32b1, srcReg32b2;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  // converting the 16 bit (short) to 8 bit (byte) and have the same data
+  // in both lanes of 128 bit register.
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  // duplicate only the second 16 bits (third and forth byte)
+  // across 256 bit register
+  secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+  // duplicate only the third 16 bits (fifth and sixth byte)
+  // across 256 bit register
+  thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+
+  filt2Reg = _mm_load_si128((__m128i const *)(filt_h4 + 32));
+  filt3Reg = _mm_load_si128((__m128i const *)(filt_h4 + 32 * 2));
+
+  for (i = output_height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)src_ptr);
+
+    // filter the source buffer
+    srcRegFilt32b3 = _mm_shuffle_epi8(srcReg32b1, filt2Reg);
+    srcRegFilt32b2 = _mm_shuffle_epi8(srcReg32b1, filt3Reg);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    srcRegFilt32b3 = _mm_maddubs_epi16(srcRegFilt32b3, secondFilters);
+    srcRegFilt32b2 = _mm_maddubs_epi16(srcRegFilt32b2, thirdFilters);
+
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b3, srcRegFilt32b2);
+
+    // reading stride of the next 16 bytes
+    // (part of it was being read by earlier read)
+    srcReg32b2 = _mm_loadu_si128((const __m128i *)(src_ptr + 8));
+
+    // filter the source buffer
+    srcRegFilt32b3 = _mm_shuffle_epi8(srcReg32b2, filt2Reg);
+    srcRegFilt32b2 = _mm_shuffle_epi8(srcReg32b2, filt3Reg);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    srcRegFilt32b3 = _mm_maddubs_epi16(srcRegFilt32b3, secondFilters);
+    srcRegFilt32b2 = _mm_maddubs_epi16(srcRegFilt32b2, thirdFilters);
+
+    // add and saturate the results together
+    srcRegFilt32b2_1 = _mm_adds_epi16(srcRegFilt32b3, srcRegFilt32b2);
+
+    // shift by 6 bit each 16 bit
+    srcRegFilt32b1_1 = _mm_adds_epi16(srcRegFilt32b1_1, addFilterReg32);
+    srcRegFilt32b2_1 = _mm_adds_epi16(srcRegFilt32b2_1, addFilterReg32);
+    srcRegFilt32b1_1 = _mm_srai_epi16(srcRegFilt32b1_1, 6);
+    srcRegFilt32b2_1 = _mm_srai_epi16(srcRegFilt32b2_1, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve result
+    srcRegFilt32b1_1 = _mm_packus_epi16(srcRegFilt32b1_1, srcRegFilt32b2_1);
+
+    src_ptr += src_pixels_per_line;
+
+    _mm_store_si128((__m128i *)output_ptr, srcRegFilt32b1_1);
+
+    output_ptr += output_pitch;
+  }
+}
+
+static void aom_filter_block1d16_v4_ssse3(
+    const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
+    ptrdiff_t out_pitch, uint32_t output_height, const int16_t *filter) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23_lo, srcReg23_hi, srcReg34_lo, srcReg34_hi;
+  __m128i srcReg45_lo, srcReg45_hi, srcReg56_lo, srcReg56_hi;
+  __m128i resReg23_lo, resReg34_lo, resReg45_lo, resReg56_lo;
+  __m128i resReg23_hi, resReg34_hi, resReg45_hi, resReg56_hi;
+  __m128i resReg23_45_lo, resReg34_56_lo, resReg23_45_hi, resReg34_56_hi;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg32, secondFilters, thirdFilters;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  addFilterReg32 = _mm_set1_epi16(32);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  // converting the 16 bit (short) to  8 bit (byte) and have the
+  // same data in both lanes of 128 bit register.
+  filtersReg = _mm_srai_epi16(filtersReg, 1);
+  filtersReg = _mm_packs_epi16(filtersReg, filtersReg);
+
+  // duplicate only the second 16 bits (third and forth byte)
+  // across 128 bit register
+  secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+  // duplicate only the third 16 bits (fifth and sixth byte)
+  // across 128 bit register
+  thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+
+  // multiple the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = out_pitch << 1;
+
+  srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23_lo = _mm_unpacklo_epi8(srcReg2, srcReg3);
+  srcReg23_hi = _mm_unpackhi_epi8(srcReg2, srcReg3);
+
+  srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+
+  // have consecutive loads on the same 256 register
+  srcReg34_lo = _mm_unpacklo_epi8(srcReg3, srcReg4);
+  srcReg34_hi = _mm_unpackhi_epi8(srcReg3, srcReg4);
+
+  for (i = output_height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+
+    srcReg45_lo = _mm_unpacklo_epi8(srcReg4, srcReg5);
+    srcReg45_hi = _mm_unpackhi_epi8(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+
+    srcReg56_lo = _mm_unpacklo_epi8(srcReg5, srcReg6);
+    srcReg56_hi = _mm_unpackhi_epi8(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+    resReg23_lo = _mm_maddubs_epi16(srcReg23_lo, secondFilters);
+    resReg34_lo = _mm_maddubs_epi16(srcReg34_lo, secondFilters);
+    resReg45_lo = _mm_maddubs_epi16(srcReg45_lo, thirdFilters);
+    resReg56_lo = _mm_maddubs_epi16(srcReg56_lo, thirdFilters);
+
+    // add and saturate the results together
+    resReg23_45_lo = _mm_adds_epi16(resReg23_lo, resReg45_lo);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_lo, resReg56_lo);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    resReg23_hi = _mm_maddubs_epi16(srcReg23_hi, secondFilters);
+    resReg34_hi = _mm_maddubs_epi16(srcReg34_hi, secondFilters);
+    resReg45_hi = _mm_maddubs_epi16(srcReg45_hi, thirdFilters);
+    resReg56_hi = _mm_maddubs_epi16(srcReg56_hi, thirdFilters);
+
+    // add and saturate the results together
+    resReg23_45_hi = _mm_adds_epi16(resReg23_hi, resReg45_hi);
+    resReg34_56_hi = _mm_adds_epi16(resReg34_hi, resReg56_hi);
+
+    // shift by 6 bit each 16 bit
+    resReg23_45_lo = _mm_adds_epi16(resReg23_45_lo, addFilterReg32);
+    resReg34_56_lo = _mm_adds_epi16(resReg34_56_lo, addFilterReg32);
+    resReg23_45_hi = _mm_adds_epi16(resReg23_45_hi, addFilterReg32);
+    resReg34_56_hi = _mm_adds_epi16(resReg34_56_hi, addFilterReg32);
+    resReg23_45_lo = _mm_srai_epi16(resReg23_45_lo, 6);
+    resReg34_56_lo = _mm_srai_epi16(resReg34_56_lo, 6);
+    resReg23_45_hi = _mm_srai_epi16(resReg23_45_hi, 6);
+    resReg34_56_hi = _mm_srai_epi16(resReg34_56_hi, 6);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packus_epi16(resReg23_45_lo, resReg23_45_hi);
+    resReg34_56 = _mm_packus_epi16(resReg34_56_lo, resReg34_56_hi);
+
+    src_ptr += src_stride;
+
+    _mm_store_si128((__m128i *)output_ptr, (resReg23_45));
+    _mm_store_si128((__m128i *)(output_ptr + out_pitch), (resReg34_56));
+
+    output_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    srcReg23_lo = srcReg45_lo;
+    srcReg34_lo = srcReg56_lo;
+    srcReg23_hi = srcReg45_hi;
+    srcReg34_hi = srcReg56_hi;
+    srcReg4 = srcReg6;
+  }
+}
+
 filter8_1dfunction aom_filter_block1d16_v8_ssse3;
 filter8_1dfunction aom_filter_block1d16_h8_ssse3;
 filter8_1dfunction aom_filter_block1d8_v8_ssse3;
@@ -287,13 +749,6 @@ filter8_1dfunction aom_filter_block1d8_h8_ssse3;
 filter8_1dfunction aom_filter_block1d4_v8_ssse3;
 filter8_1dfunction aom_filter_block1d4_h8_ssse3;
 
-#define aom_filter_block1d16_h4_ssse3 aom_filter_block1d16_h8_ssse3
-#define aom_filter_block1d16_v4_ssse3 aom_filter_block1d16_v8_ssse3
-#define aom_filter_block1d8_h4_ssse3 aom_filter_block1d8_h8_ssse3
-#define aom_filter_block1d8_v4_ssse3 aom_filter_block1d8_v8_ssse3
-#define aom_filter_block1d4_h4_ssse3 aom_filter_block1d4_h8_ssse3
-#define aom_filter_block1d4_v4_ssse3 aom_filter_block1d4_v8_ssse3
-
 filter8_1dfunction aom_filter_block1d16_v2_ssse3;
 filter8_1dfunction aom_filter_block1d16_h2_ssse3;
 filter8_1dfunction aom_filter_block1d8_v2_ssse3;
diff --git a/third_party/aom/aom_dsp/x86/avg_intrin_avx2.c b/third_party/aom/aom_dsp/x86/avg_intrin_avx2.c
new file mode 100644
index 0000000000..e0ba8d5de6
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/avg_intrin_avx2.c
@@ -0,0 +1,249 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <immintrin.h>
+
+#include "config/aom_dsp_rtcd.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/x86/bitdepth_conversion_avx2.h"
+#include "aom_ports/mem.h"
+
+static void hadamard_col8x2_avx2(__m256i *in, int iter) {
+  __m256i a0 = in[0];
+  __m256i a1 = in[1];
+  __m256i a2 = in[2];
+  __m256i a3 = in[3];
+  __m256i a4 = in[4];
+  __m256i a5 = in[5];
+  __m256i a6 = in[6];
+  __m256i a7 = in[7];
+
+  __m256i b0 = _mm256_add_epi16(a0, a1);
+  __m256i b1 = _mm256_sub_epi16(a0, a1);
+  __m256i b2 = _mm256_add_epi16(a2, a3);
+  __m256i b3 = _mm256_sub_epi16(a2, a3);
+  __m256i b4 = _mm256_add_epi16(a4, a5);
+  __m256i b5 = _mm256_sub_epi16(a4, a5);
+  __m256i b6 = _mm256_add_epi16(a6, a7);
+  __m256i b7 = _mm256_sub_epi16(a6, a7);
+
+  a0 = _mm256_add_epi16(b0, b2);
+  a1 = _mm256_add_epi16(b1, b3);
+  a2 = _mm256_sub_epi16(b0, b2);
+  a3 = _mm256_sub_epi16(b1, b3);
+  a4 = _mm256_add_epi16(b4, b6);
+  a5 = _mm256_add_epi16(b5, b7);
+  a6 = _mm256_sub_epi16(b4, b6);
+  a7 = _mm256_sub_epi16(b5, b7);
+
+  if (iter == 0) {
+    b0 = _mm256_add_epi16(a0, a4);
+    b7 = _mm256_add_epi16(a1, a5);
+    b3 = _mm256_add_epi16(a2, a6);
+    b4 = _mm256_add_epi16(a3, a7);
+    b2 = _mm256_sub_epi16(a0, a4);
+    b6 = _mm256_sub_epi16(a1, a5);
+    b1 = _mm256_sub_epi16(a2, a6);
+    b5 = _mm256_sub_epi16(a3, a7);
+
+    a0 = _mm256_unpacklo_epi16(b0, b1);
+    a1 = _mm256_unpacklo_epi16(b2, b3);
+    a2 = _mm256_unpackhi_epi16(b0, b1);
+    a3 = _mm256_unpackhi_epi16(b2, b3);
+    a4 = _mm256_unpacklo_epi16(b4, b5);
+    a5 = _mm256_unpacklo_epi16(b6, b7);
+    a6 = _mm256_unpackhi_epi16(b4, b5);
+    a7 = _mm256_unpackhi_epi16(b6, b7);
+
+    b0 = _mm256_unpacklo_epi32(a0, a1);
+    b1 = _mm256_unpacklo_epi32(a4, a5);
+    b2 = _mm256_unpackhi_epi32(a0, a1);
+    b3 = _mm256_unpackhi_epi32(a4, a5);
+    b4 = _mm256_unpacklo_epi32(a2, a3);
+    b5 = _mm256_unpacklo_epi32(a6, a7);
+    b6 = _mm256_unpackhi_epi32(a2, a3);
+    b7 = _mm256_unpackhi_epi32(a6, a7);
+
+    in[0] = _mm256_unpacklo_epi64(b0, b1);
+    in[1] = _mm256_unpackhi_epi64(b0, b1);
+    in[2] = _mm256_unpacklo_epi64(b2, b3);
+    in[3] = _mm256_unpackhi_epi64(b2, b3);
+    in[4] = _mm256_unpacklo_epi64(b4, b5);
+    in[5] = _mm256_unpackhi_epi64(b4, b5);
+    in[6] = _mm256_unpacklo_epi64(b6, b7);
+    in[7] = _mm256_unpackhi_epi64(b6, b7);
+  } else {
+    in[0] = _mm256_add_epi16(a0, a4);
+    in[7] = _mm256_add_epi16(a1, a5);
+    in[3] = _mm256_add_epi16(a2, a6);
+    in[4] = _mm256_add_epi16(a3, a7);
+    in[2] = _mm256_sub_epi16(a0, a4);
+    in[6] = _mm256_sub_epi16(a1, a5);
+    in[1] = _mm256_sub_epi16(a2, a6);
+    in[5] = _mm256_sub_epi16(a3, a7);
+  }
+}
+
+static void hadamard_8x8x2_avx2(const int16_t *src_diff, ptrdiff_t src_stride,
+                                int16_t *coeff) {
+  __m256i src[8];
+  src[0] = _mm256_loadu_si256((const __m256i *)src_diff);
+  src[1] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[2] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[3] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[4] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[5] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[6] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+  src[7] = _mm256_loadu_si256((const __m256i *)(src_diff += src_stride));
+
+  hadamard_col8x2_avx2(src, 0);
+  hadamard_col8x2_avx2(src, 1);
+
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[0], src[1], 0x20));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[2], src[3], 0x20));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[4], src[5], 0x20));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[6], src[7], 0x20));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[0], src[1], 0x31));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[2], src[3], 0x31));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[4], src[5], 0x31));
+  coeff += 16;
+  _mm256_storeu_si256((__m256i *)coeff,
+                      _mm256_permute2x128_si256(src[6], src[7], 0x31));
+}
+
+static INLINE void hadamard_16x16_avx2(const int16_t *src_diff,
+                                       ptrdiff_t src_stride, tran_low_t *coeff,
+                                       int is_final) {
+  DECLARE_ALIGNED(32, int16_t, temp_coeff[16 * 16]);
+  int16_t *t_coeff = temp_coeff;
+  int16_t *coeff16 = (int16_t *)coeff;
+  int idx;
+  for (idx = 0; idx < 2; ++idx) {
+    const int16_t *src_ptr = src_diff + idx * 8 * src_stride;
+    hadamard_8x8x2_avx2(src_ptr, src_stride, t_coeff + (idx * 64 * 2));
+  }
+
+  for (idx = 0; idx < 64; idx += 16) {
+    const __m256i coeff0 = _mm256_loadu_si256((const __m256i *)t_coeff);
+    const __m256i coeff1 = _mm256_loadu_si256((const __m256i *)(t_coeff + 64));
+    const __m256i coeff2 = _mm256_loadu_si256((const __m256i *)(t_coeff + 128));
+    const __m256i coeff3 = _mm256_loadu_si256((const __m256i *)(t_coeff + 192));
+
+    __m256i b0 = _mm256_add_epi16(coeff0, coeff1);
+    __m256i b1 = _mm256_sub_epi16(coeff0, coeff1);
+    __m256i b2 = _mm256_add_epi16(coeff2, coeff3);
+    __m256i b3 = _mm256_sub_epi16(coeff2, coeff3);
+
+    b0 = _mm256_srai_epi16(b0, 1);
+    b1 = _mm256_srai_epi16(b1, 1);
+    b2 = _mm256_srai_epi16(b2, 1);
+    b3 = _mm256_srai_epi16(b3, 1);
+    if (is_final) {
+      store_tran_low(_mm256_add_epi16(b0, b2), coeff);
+      store_tran_low(_mm256_add_epi16(b1, b3), coeff + 64);
+      store_tran_low(_mm256_sub_epi16(b0, b2), coeff + 128);
+      store_tran_low(_mm256_sub_epi16(b1, b3), coeff + 192);
+      coeff += 16;
+    } else {
+      _mm256_storeu_si256((__m256i *)coeff16, _mm256_add_epi16(b0, b2));
+      _mm256_storeu_si256((__m256i *)(coeff16 + 64), _mm256_add_epi16(b1, b3));
+      _mm256_storeu_si256((__m256i *)(coeff16 + 128), _mm256_sub_epi16(b0, b2));
+      _mm256_storeu_si256((__m256i *)(coeff16 + 192), _mm256_sub_epi16(b1, b3));
+      coeff16 += 16;
+    }
+    t_coeff += 16;
+  }
+}
+
+void aom_hadamard_16x16_avx2(const int16_t *src_diff, ptrdiff_t src_stride,
+                             tran_low_t *coeff) {
+  hadamard_16x16_avx2(src_diff, src_stride, coeff, 1);
+}
+
+void aom_hadamard_32x32_avx2(const int16_t *src_diff, ptrdiff_t src_stride,
+                             tran_low_t *coeff) {
+  // For high bitdepths, it is unnecessary to store_tran_low
+  // (mult/unpack/store), then load_tran_low (load/pack) the same memory in the
+  // next stage.  Output to an intermediate buffer first, then store_tran_low()
+  // in the final stage.
+  DECLARE_ALIGNED(32, int16_t, temp_coeff[32 * 32]);
+  int16_t *t_coeff = temp_coeff;
+  int idx;
+  for (idx = 0; idx < 4; ++idx) {
+    // src_diff: 9 bit, dynamic range [-255, 255]
+    const int16_t *src_ptr =
+        src_diff + (idx >> 1) * 16 * src_stride + (idx & 0x01) * 16;
+    hadamard_16x16_avx2(src_ptr, src_stride,
+                        (tran_low_t *)(t_coeff + idx * 256), 0);
+  }
+
+  for (idx = 0; idx < 256; idx += 16) {
+    const __m256i coeff0 = _mm256_loadu_si256((const __m256i *)t_coeff);
+    const __m256i coeff1 = _mm256_loadu_si256((const __m256i *)(t_coeff + 256));
+    const __m256i coeff2 = _mm256_loadu_si256((const __m256i *)(t_coeff + 512));
+    const __m256i coeff3 = _mm256_loadu_si256((const __m256i *)(t_coeff + 768));
+
+    __m256i b0 = _mm256_add_epi16(coeff0, coeff1);
+    __m256i b1 = _mm256_sub_epi16(coeff0, coeff1);
+    __m256i b2 = _mm256_add_epi16(coeff2, coeff3);
+    __m256i b3 = _mm256_sub_epi16(coeff2, coeff3);
+
+    b0 = _mm256_srai_epi16(b0, 2);
+    b1 = _mm256_srai_epi16(b1, 2);
+    b2 = _mm256_srai_epi16(b2, 2);
+    b3 = _mm256_srai_epi16(b3, 2);
+
+    store_tran_low(_mm256_add_epi16(b0, b2), coeff);
+    store_tran_low(_mm256_add_epi16(b1, b3), coeff + 256);
+    store_tran_low(_mm256_sub_epi16(b0, b2), coeff + 512);
+    store_tran_low(_mm256_sub_epi16(b1, b3), coeff + 768);
+
+    coeff += 16;
+    t_coeff += 16;
+  }
+}
+
+int aom_satd_avx2(const tran_low_t *coeff, int length) {
+  const __m256i one = _mm256_set1_epi16(1);
+  __m256i accum = _mm256_setzero_si256();
+  int i;
+
+  for (i = 0; i < length; i += 16) {
+    const __m256i src_line = load_tran_low(coeff);
+    const __m256i abs = _mm256_abs_epi16(src_line);
+    const __m256i sum = _mm256_madd_epi16(abs, one);
+    accum = _mm256_add_epi32(accum, sum);
+    coeff += 16;
+  }
+
+  {  // 32 bit horizontal add
+    const __m256i a = _mm256_srli_si256(accum, 8);
+    const __m256i b = _mm256_add_epi32(accum, a);
+    const __m256i c = _mm256_srli_epi64(b, 32);
+    const __m256i d = _mm256_add_epi32(b, c);
+    const __m128i accum_128 = _mm_add_epi32(_mm256_castsi256_si128(d),
+                                            _mm256_extractf128_si256(d, 1));
+    return _mm_cvtsi128_si32(accum_128);
+  }
+}
diff --git a/third_party/aom/aom_dsp/x86/avg_intrin_sse2.c b/third_party/aom/aom_dsp/x86/avg_intrin_sse2.c
new file mode 100644
index 0000000000..969e4e195d
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/avg_intrin_sse2.c
@@ -0,0 +1,285 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <immintrin.h>
+
+#include "config/aom_dsp_rtcd.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/x86/bitdepth_conversion_sse2.h"
+#include "aom_ports/mem.h"
+
+static void hadamard_col8_sse2(__m128i *in, int iter) {
+  __m128i a0 = in[0];
+  __m128i a1 = in[1];
+  __m128i a2 = in[2];
+  __m128i a3 = in[3];
+  __m128i a4 = in[4];
+  __m128i a5 = in[5];
+  __m128i a6 = in[6];
+  __m128i a7 = in[7];
+
+  __m128i b0 = _mm_add_epi16(a0, a1);
+  __m128i b1 = _mm_sub_epi16(a0, a1);
+  __m128i b2 = _mm_add_epi16(a2, a3);
+  __m128i b3 = _mm_sub_epi16(a2, a3);
+  __m128i b4 = _mm_add_epi16(a4, a5);
+  __m128i b5 = _mm_sub_epi16(a4, a5);
+  __m128i b6 = _mm_add_epi16(a6, a7);
+  __m128i b7 = _mm_sub_epi16(a6, a7);
+
+  a0 = _mm_add_epi16(b0, b2);
+  a1 = _mm_add_epi16(b1, b3);
+  a2 = _mm_sub_epi16(b0, b2);
+  a3 = _mm_sub_epi16(b1, b3);
+  a4 = _mm_add_epi16(b4, b6);
+  a5 = _mm_add_epi16(b5, b7);
+  a6 = _mm_sub_epi16(b4, b6);
+  a7 = _mm_sub_epi16(b5, b7);
+
+  if (iter == 0) {
+    b0 = _mm_add_epi16(a0, a4);
+    b7 = _mm_add_epi16(a1, a5);
+    b3 = _mm_add_epi16(a2, a6);
+    b4 = _mm_add_epi16(a3, a7);
+    b2 = _mm_sub_epi16(a0, a4);
+    b6 = _mm_sub_epi16(a1, a5);
+    b1 = _mm_sub_epi16(a2, a6);
+    b5 = _mm_sub_epi16(a3, a7);
+
+    a0 = _mm_unpacklo_epi16(b0, b1);
+    a1 = _mm_unpacklo_epi16(b2, b3);
+    a2 = _mm_unpackhi_epi16(b0, b1);
+    a3 = _mm_unpackhi_epi16(b2, b3);
+    a4 = _mm_unpacklo_epi16(b4, b5);
+    a5 = _mm_unpacklo_epi16(b6, b7);
+    a6 = _mm_unpackhi_epi16(b4, b5);
+    a7 = _mm_unpackhi_epi16(b6, b7);
+
+    b0 = _mm_unpacklo_epi32(a0, a1);
+    b1 = _mm_unpacklo_epi32(a4, a5);
+    b2 = _mm_unpackhi_epi32(a0, a1);
+    b3 = _mm_unpackhi_epi32(a4, a5);
+    b4 = _mm_unpacklo_epi32(a2, a3);
+    b5 = _mm_unpacklo_epi32(a6, a7);
+    b6 = _mm_unpackhi_epi32(a2, a3);
+    b7 = _mm_unpackhi_epi32(a6, a7);
+
+    in[0] = _mm_unpacklo_epi64(b0, b1);
+    in[1] = _mm_unpackhi_epi64(b0, b1);
+    in[2] = _mm_unpacklo_epi64(b2, b3);
+    in[3] = _mm_unpackhi_epi64(b2, b3);
+    in[4] = _mm_unpacklo_epi64(b4, b5);
+    in[5] = _mm_unpackhi_epi64(b4, b5);
+    in[6] = _mm_unpacklo_epi64(b6, b7);
+    in[7] = _mm_unpackhi_epi64(b6, b7);
+  } else {
+    in[0] = _mm_add_epi16(a0, a4);
+    in[7] = _mm_add_epi16(a1, a5);
+    in[3] = _mm_add_epi16(a2, a6);
+    in[4] = _mm_add_epi16(a3, a7);
+    in[2] = _mm_sub_epi16(a0, a4);
+    in[6] = _mm_sub_epi16(a1, a5);
+    in[1] = _mm_sub_epi16(a2, a6);
+    in[5] = _mm_sub_epi16(a3, a7);
+  }
+}
+
+static INLINE void hadamard_8x8_sse2(const int16_t *src_diff,
+                                     ptrdiff_t src_stride, tran_low_t *coeff,
+                                     int is_final) {
+  __m128i src[8];
+  src[0] = _mm_load_si128((const __m128i *)src_diff);
+  src[1] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[2] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[3] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[4] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[5] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[6] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+  src[7] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
+
+  hadamard_col8_sse2(src, 0);
+  hadamard_col8_sse2(src, 1);
+
+  if (is_final) {
+    store_tran_low(src[0], coeff);
+    coeff += 8;
+    store_tran_low(src[1], coeff);
+    coeff += 8;
+    store_tran_low(src[2], coeff);
+    coeff += 8;
+    store_tran_low(src[3], coeff);
+    coeff += 8;
+    store_tran_low(src[4], coeff);
+    coeff += 8;
+    store_tran_low(src[5], coeff);
+    coeff += 8;
+    store_tran_low(src[6], coeff);
+    coeff += 8;
+    store_tran_low(src[7], coeff);
+  } else {
+    int16_t *coeff16 = (int16_t *)coeff;
+    _mm_store_si128((__m128i *)coeff16, src[0]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[1]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[2]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[3]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[4]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[5]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[6]);
+    coeff16 += 8;
+    _mm_store_si128((__m128i *)coeff16, src[7]);
+  }
+}
+
+void aom_hadamard_8x8_sse2(const int16_t *src_diff, ptrdiff_t src_stride,
+                           tran_low_t *coeff) {
+  hadamard_8x8_sse2(src_diff, src_stride, coeff, 1);
+}
+
+static INLINE void hadamard_16x16_sse2(const int16_t *src_diff,
+                                       ptrdiff_t src_stride, tran_low_t *coeff,
+                                       int is_final) {
+  // For high bitdepths, it is unnecessary to store_tran_low
+  // (mult/unpack/store), then load_tran_low (load/pack) the same memory in the
+  // next stage.  Output to an intermediate buffer first, then store_tran_low()
+  // in the final stage.
+  DECLARE_ALIGNED(32, int16_t, temp_coeff[16 * 16]);
+  int16_t *t_coeff = temp_coeff;
+  int16_t *coeff16 = (int16_t *)coeff;
+  int idx;
+  for (idx = 0; idx < 4; ++idx) {
+    const int16_t *src_ptr =
+        src_diff + (idx >> 1) * 8 * src_stride + (idx & 0x01) * 8;
+    hadamard_8x8_sse2(src_ptr, src_stride, (tran_low_t *)(t_coeff + idx * 64),
+                      0);
+  }
+
+  for (idx = 0; idx < 64; idx += 8) {
+    __m128i coeff0 = _mm_load_si128((const __m128i *)t_coeff);
+    __m128i coeff1 = _mm_load_si128((const __m128i *)(t_coeff + 64));
+    __m128i coeff2 = _mm_load_si128((const __m128i *)(t_coeff + 128));
+    __m128i coeff3 = _mm_load_si128((const __m128i *)(t_coeff + 192));
+
+    __m128i b0 = _mm_add_epi16(coeff0, coeff1);
+    __m128i b1 = _mm_sub_epi16(coeff0, coeff1);
+    __m128i b2 = _mm_add_epi16(coeff2, coeff3);
+    __m128i b3 = _mm_sub_epi16(coeff2, coeff3);
+
+    b0 = _mm_srai_epi16(b0, 1);
+    b1 = _mm_srai_epi16(b1, 1);
+    b2 = _mm_srai_epi16(b2, 1);
+    b3 = _mm_srai_epi16(b3, 1);
+
+    coeff0 = _mm_add_epi16(b0, b2);
+    coeff1 = _mm_add_epi16(b1, b3);
+    coeff2 = _mm_sub_epi16(b0, b2);
+    coeff3 = _mm_sub_epi16(b1, b3);
+
+    if (is_final) {
+      store_tran_low(coeff0, coeff);
+      store_tran_low(coeff1, coeff + 64);
+      store_tran_low(coeff2, coeff + 128);
+      store_tran_low(coeff3, coeff + 192);
+      coeff += 8;
+    } else {
+      _mm_store_si128((__m128i *)coeff16, coeff0);
+      _mm_store_si128((__m128i *)(coeff16 + 64), coeff1);
+      _mm_store_si128((__m128i *)(coeff16 + 128), coeff2);
+      _mm_store_si128((__m128i *)(coeff16 + 192), coeff3);
+      coeff16 += 8;
+    }
+
+    t_coeff += 8;
+  }
+}
+
+void aom_hadamard_16x16_sse2(const int16_t *src_diff, ptrdiff_t src_stride,
+                             tran_low_t *coeff) {
+  hadamard_16x16_sse2(src_diff, src_stride, coeff, 1);
+}
+
+void aom_hadamard_32x32_sse2(const int16_t *src_diff, ptrdiff_t src_stride,
+                             tran_low_t *coeff) {
+  // For high bitdepths, it is unnecessary to store_tran_low
+  // (mult/unpack/store), then load_tran_low (load/pack) the same memory in the
+  // next stage.  Output to an intermediate buffer first, then store_tran_low()
+  // in the final stage.
+  DECLARE_ALIGNED(32, int16_t, temp_coeff[32 * 32]);
+  int16_t *t_coeff = temp_coeff;
+  int idx;
+  for (idx = 0; idx < 4; ++idx) {
+    const int16_t *src_ptr =
+        src_diff + (idx >> 1) * 16 * src_stride + (idx & 0x01) * 16;
+    hadamard_16x16_sse2(src_ptr, src_stride,
+                        (tran_low_t *)(t_coeff + idx * 256), 0);
+  }
+
+  for (idx = 0; idx < 256; idx += 8) {
+    __m128i coeff0 = _mm_load_si128((const __m128i *)t_coeff);
+    __m128i coeff1 = _mm_load_si128((const __m128i *)(t_coeff + 256));
+    __m128i coeff2 = _mm_load_si128((const __m128i *)(t_coeff + 512));
+    __m128i coeff3 = _mm_load_si128((const __m128i *)(t_coeff + 768));
+
+    __m128i b0 = _mm_add_epi16(coeff0, coeff1);
+    __m128i b1 = _mm_sub_epi16(coeff0, coeff1);
+    __m128i b2 = _mm_add_epi16(coeff2, coeff3);
+    __m128i b3 = _mm_sub_epi16(coeff2, coeff3);
+
+    b0 = _mm_srai_epi16(b0, 2);
+    b1 = _mm_srai_epi16(b1, 2);
+    b2 = _mm_srai_epi16(b2, 2);
+    b3 = _mm_srai_epi16(b3, 2);
+
+    coeff0 = _mm_add_epi16(b0, b2);
+    coeff1 = _mm_add_epi16(b1, b3);
+    store_tran_low(coeff0, coeff);
+    store_tran_low(coeff1, coeff + 256);
+
+    coeff2 = _mm_sub_epi16(b0, b2);
+    coeff3 = _mm_sub_epi16(b1, b3);
+    store_tran_low(coeff2, coeff + 512);
+    store_tran_low(coeff3, coeff + 768);
+
+    coeff += 8;
+    t_coeff += 8;
+  }
+}
+
+int aom_satd_sse2(const tran_low_t *coeff, int length) {
+  int i;
+  const __m128i zero = _mm_setzero_si128();
+  __m128i accum = zero;
+
+  for (i = 0; i < length; i += 8) {
+    const __m128i src_line = load_tran_low(coeff);
+    const __m128i inv = _mm_sub_epi16(zero, src_line);
+    const __m128i abs = _mm_max_epi16(src_line, inv);  // abs(src_line)
+    const __m128i abs_lo = _mm_unpacklo_epi16(abs, zero);
+    const __m128i abs_hi = _mm_unpackhi_epi16(abs, zero);
+    const __m128i sum = _mm_add_epi32(abs_lo, abs_hi);
+    accum = _mm_add_epi32(accum, sum);
+    coeff += 8;
+  }
+
+  {  // cascading summation of accum
+    __m128i hi = _mm_srli_si128(accum, 8);
+    accum = _mm_add_epi32(accum, hi);
+    hi = _mm_srli_epi64(accum, 32);
+    accum = _mm_add_epi32(accum, hi);
+  }
+
+  return _mm_cvtsi128_si32(accum);
+}
diff --git a/third_party/aom/aom_dsp/x86/bitdepth_conversion_avx2.h b/third_party/aom/aom_dsp/x86/bitdepth_conversion_avx2.h
new file mode 100644
index 0000000000..85896e2768
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/bitdepth_conversion_avx2.h
@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <immintrin.h>
+
+#include "config/aom_config.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/aom_dsp_common.h"
+
+static INLINE __m256i load_tran_low(const tran_low_t *a) {
+  const __m256i a_low = _mm256_loadu_si256((const __m256i *)a);
+  const __m256i a_high = _mm256_loadu_si256((const __m256i *)(a + 8));
+  return _mm256_packs_epi32(a_low, a_high);
+}
+
+static INLINE void store_tran_low(__m256i a, tran_low_t *b) {
+  const __m256i one = _mm256_set1_epi16(1);
+  const __m256i a_hi = _mm256_mulhi_epi16(a, one);
+  const __m256i a_lo = _mm256_mullo_epi16(a, one);
+  const __m256i a_1 = _mm256_unpacklo_epi16(a_lo, a_hi);
+  const __m256i a_2 = _mm256_unpackhi_epi16(a_lo, a_hi);
+  _mm256_storeu_si256((__m256i *)b, a_1);
+  _mm256_storeu_si256((__m256i *)(b + 8), a_2);
+}
diff --git a/third_party/aom/aom_dsp/x86/bitdepth_conversion_sse2.h b/third_party/aom/aom_dsp/x86/bitdepth_conversion_sse2.h
new file mode 100644
index 0000000000..42bb2d1d32
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/bitdepth_conversion_sse2.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <xmmintrin.h>
+
+#include "config/aom_config.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/aom_dsp_common.h"
+
+// Load 8 16 bit values. If the source is 32 bits then pack down with
+// saturation.
+static INLINE __m128i load_tran_low(const tran_low_t *a) {
+  const __m128i a_low = _mm_load_si128((const __m128i *)a);
+  return _mm_packs_epi32(a_low, *(const __m128i *)(a + 4));
+}
+
+// Store 8 16 bit values. If the destination is 32 bits then sign extend the
+// values by multiplying by 1.
+static INLINE void store_tran_low(__m128i a, tran_low_t *b) {
+  const __m128i one = _mm_set1_epi16(1);
+  const __m128i a_hi = _mm_mulhi_epi16(a, one);
+  const __m128i a_lo = _mm_mullo_epi16(a, one);
+  const __m128i a_1 = _mm_unpacklo_epi16(a_lo, a_hi);
+  const __m128i a_2 = _mm_unpackhi_epi16(a_lo, a_hi);
+  _mm_store_si128((__m128i *)(b), a_1);
+  _mm_store_si128((__m128i *)(b + 4), a_2);
+}
diff --git a/third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c b/third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c
index 67fb4d32bd..057f615c20 100644
--- a/third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c
+++ b/third_party/aom/aom_dsp/x86/blend_a64_mask_avx2.c
@@ -898,3 +898,475 @@ void aom_blend_a64_mask_avx2(uint8_t *dst, uint32_t dst_stride,
     }
   }
 }
+
+//////////////////////////////////////////////////////////////////////////////
+// aom_highbd_blend_a64_d16_mask_avx2()
+//////////////////////////////////////////////////////////////////////////////
+
+static INLINE void highbd_blend_a64_d16_mask_w4_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0,
+    const __m256i *round_offset, int shift, const __m256i *clip_low,
+    const __m256i *clip_high, const __m256i *mask_max) {
+  // Load 4x u16 pixels from each of 4 rows from each source
+  const __m256i s0 = _mm256_set_epi64x(*(uint64_t *)(src0 + 3 * src0_stride),
+                                       *(uint64_t *)(src0 + 2 * src0_stride),
+                                       *(uint64_t *)(src0 + 1 * src0_stride),
+                                       *(uint64_t *)(src0 + 0 * src0_stride));
+  const __m256i s1 = _mm256_set_epi64x(*(uint64_t *)(src1 + 3 * src1_stride),
+                                       *(uint64_t *)(src1 + 2 * src1_stride),
+                                       *(uint64_t *)(src1 + 1 * src1_stride),
+                                       *(uint64_t *)(src1 + 0 * src1_stride));
+  // Generate the inverse mask
+  const __m256i mask1 = _mm256_sub_epi16(*mask_max, *mask0);
+
+  // Multiply each mask by the respective source
+  const __m256i mul0_highs = _mm256_mulhi_epu16(*mask0, s0);
+  const __m256i mul0_lows = _mm256_mullo_epi16(*mask0, s0);
+  const __m256i mul0h = _mm256_unpackhi_epi16(mul0_lows, mul0_highs);
+  const __m256i mul0l = _mm256_unpacklo_epi16(mul0_lows, mul0_highs);
+  // Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
+  // lanes Later, packs does the same again which cancels this out with no need
+  // for a permute.  The intermediate values being reordered makes no difference
+
+  const __m256i mul1_highs = _mm256_mulhi_epu16(mask1, s1);
+  const __m256i mul1_lows = _mm256_mullo_epi16(mask1, s1);
+  const __m256i mul1h = _mm256_unpackhi_epi16(mul1_lows, mul1_highs);
+  const __m256i mul1l = _mm256_unpacklo_epi16(mul1_lows, mul1_highs);
+
+  const __m256i sumh = _mm256_add_epi32(mul0h, mul1h);
+  const __m256i suml = _mm256_add_epi32(mul0l, mul1l);
+
+  const __m256i roundh =
+      _mm256_srai_epi32(_mm256_sub_epi32(sumh, *round_offset), shift);
+  const __m256i roundl =
+      _mm256_srai_epi32(_mm256_sub_epi32(suml, *round_offset), shift);
+
+  const __m256i pack = _mm256_packs_epi32(roundl, roundh);
+  const __m256i clip =
+      _mm256_min_epi16(_mm256_max_epi16(pack, *clip_low), *clip_high);
+
+  // _mm256_extract_epi64 doesn't exist on x86, so do it the old-fashioned way:
+  const __m128i cliph = _mm256_extracti128_si256(clip, 1);
+  xx_storel_64(dst + 3 * dst_stride, _mm_srli_si128(cliph, 8));
+  xx_storel_64(dst + 2 * dst_stride, cliph);
+  const __m128i clipl = _mm256_castsi256_si128(clip);
+  xx_storel_64(dst + 1 * dst_stride, _mm_srli_si128(clipl, 8));
+  xx_storel_64(dst + 0 * dst_stride, clipl);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w4_avx2(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m256i *round_offset, int shift, const __m256i *clip_low,
+    const __m256i *clip_high, const __m256i *mask_max) {
+  do {
+    // Load 8x u8 pixels from each of 4 rows of the mask, pad each to u16
+    const __m128i mask08 = _mm_set_epi32(*(uint32_t *)(mask + 3 * mask_stride),
+                                         *(uint32_t *)(mask + 2 * mask_stride),
+                                         *(uint32_t *)(mask + 1 * mask_stride),
+                                         *(uint32_t *)(mask + 0 * mask_stride));
+    const __m256i mask0 = _mm256_cvtepu8_epi16(mask08);
+
+    highbd_blend_a64_d16_mask_w4_avx2(dst, dst_stride, src0, src0_stride, src1,
+                                      src1_stride, &mask0, round_offset, shift,
+                                      clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 4;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w4_avx2(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m256i *round_offset, int shift, const __m256i *clip_low,
+    const __m256i *clip_high, const __m256i *mask_max) {
+  const __m256i one_b = _mm256_set1_epi8(1);
+  const __m256i two_w = _mm256_set1_epi16(2);
+  do {
+    // Load 8 pixels from each of 8 rows of mask,
+    // (saturating) add together rows then use madd to add adjacent pixels
+    // Finally, divide each value by 4 (with rounding)
+    const __m256i m0246 =
+        _mm256_set_epi64x(*(uint64_t *)(mask + 6 * mask_stride),
+                          *(uint64_t *)(mask + 4 * mask_stride),
+                          *(uint64_t *)(mask + 2 * mask_stride),
+                          *(uint64_t *)(mask + 0 * mask_stride));
+    const __m256i m1357 =
+        _mm256_set_epi64x(*(uint64_t *)(mask + 7 * mask_stride),
+                          *(uint64_t *)(mask + 5 * mask_stride),
+                          *(uint64_t *)(mask + 3 * mask_stride),
+                          *(uint64_t *)(mask + 1 * mask_stride));
+    const __m256i addrows = _mm256_adds_epu8(m0246, m1357);
+    const __m256i adjacent = _mm256_maddubs_epi16(addrows, one_b);
+    const __m256i mask0 =
+        _mm256_srli_epi16(_mm256_add_epi16(adjacent, two_w), 2);
+
+    highbd_blend_a64_d16_mask_w4_avx2(dst, dst_stride, src0, src0_stride, src1,
+                                      src1_stride, &mask0, round_offset, shift,
+                                      clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 8;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_w8_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0a,
+    const __m256i *mask0b, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  // Load 8x u16 pixels from each of 4 rows from each source
+  const __m256i s0a =
+      yy_loadu2_128(src0 + 0 * src0_stride, src0 + 1 * src0_stride);
+  const __m256i s0b =
+      yy_loadu2_128(src0 + 2 * src0_stride, src0 + 3 * src0_stride);
+  const __m256i s1a =
+      yy_loadu2_128(src1 + 0 * src1_stride, src1 + 1 * src1_stride);
+  const __m256i s1b =
+      yy_loadu2_128(src1 + 2 * src1_stride, src1 + 3 * src1_stride);
+
+  // Generate inverse masks
+  const __m256i mask1a = _mm256_sub_epi16(*mask_max, *mask0a);
+  const __m256i mask1b = _mm256_sub_epi16(*mask_max, *mask0b);
+
+  // Multiply sources by respective masks
+  const __m256i mul0a_highs = _mm256_mulhi_epu16(*mask0a, s0a);
+  const __m256i mul0a_lows = _mm256_mullo_epi16(*mask0a, s0a);
+  const __m256i mul0ah = _mm256_unpackhi_epi16(mul0a_lows, mul0a_highs);
+  const __m256i mul0al = _mm256_unpacklo_epi16(mul0a_lows, mul0a_highs);
+  // Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
+  // lanes Later, packs does the same again which cancels this out with no need
+  // for a permute.  The intermediate values being reordered makes no difference
+
+  const __m256i mul1a_highs = _mm256_mulhi_epu16(mask1a, s1a);
+  const __m256i mul1a_lows = _mm256_mullo_epi16(mask1a, s1a);
+  const __m256i mul1ah = _mm256_unpackhi_epi16(mul1a_lows, mul1a_highs);
+  const __m256i mul1al = _mm256_unpacklo_epi16(mul1a_lows, mul1a_highs);
+
+  const __m256i sumah = _mm256_add_epi32(mul0ah, mul1ah);
+  const __m256i sumal = _mm256_add_epi32(mul0al, mul1al);
+
+  const __m256i mul0b_highs = _mm256_mulhi_epu16(*mask0b, s0b);
+  const __m256i mul0b_lows = _mm256_mullo_epi16(*mask0b, s0b);
+  const __m256i mul0bh = _mm256_unpackhi_epi16(mul0b_lows, mul0b_highs);
+  const __m256i mul0bl = _mm256_unpacklo_epi16(mul0b_lows, mul0b_highs);
+
+  const __m256i mul1b_highs = _mm256_mulhi_epu16(mask1b, s1b);
+  const __m256i mul1b_lows = _mm256_mullo_epi16(mask1b, s1b);
+  const __m256i mul1bh = _mm256_unpackhi_epi16(mul1b_lows, mul1b_highs);
+  const __m256i mul1bl = _mm256_unpacklo_epi16(mul1b_lows, mul1b_highs);
+
+  const __m256i sumbh = _mm256_add_epi32(mul0bh, mul1bh);
+  const __m256i sumbl = _mm256_add_epi32(mul0bl, mul1bl);
+
+  // Divide down each result, with rounding
+  const __m256i roundah =
+      _mm256_srai_epi32(_mm256_sub_epi32(sumah, *round_offset), shift);
+  const __m256i roundal =
+      _mm256_srai_epi32(_mm256_sub_epi32(sumal, *round_offset), shift);
+  const __m256i roundbh =
+      _mm256_srai_epi32(_mm256_sub_epi32(sumbh, *round_offset), shift);
+  const __m256i roundbl =
+      _mm256_srai_epi32(_mm256_sub_epi32(sumbl, *round_offset), shift);
+
+  // Pack each i32 down to an i16 with saturation, then clip to valid range
+  const __m256i packa = _mm256_packs_epi32(roundal, roundah);
+  const __m256i clipa =
+      _mm256_min_epi16(_mm256_max_epi16(packa, *clip_low), *clip_high);
+  const __m256i packb = _mm256_packs_epi32(roundbl, roundbh);
+  const __m256i clipb =
+      _mm256_min_epi16(_mm256_max_epi16(packb, *clip_low), *clip_high);
+
+  // Store 8x u16 pixels to each of 4 rows in the destination
+  yy_storeu2_128(dst + 0 * dst_stride, dst + 1 * dst_stride, clipa);
+  yy_storeu2_128(dst + 2 * dst_stride, dst + 3 * dst_stride, clipb);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w8_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
+    int mask_stride, int h, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  do {
+    // Load 8x u8 pixels from each of 4 rows in the mask
+    const __m128i mask0a8 =
+        _mm_set_epi64x(*(uint64_t *)mask, *(uint64_t *)(mask + mask_stride));
+    const __m128i mask0b8 =
+        _mm_set_epi64x(*(uint64_t *)(mask + 2 * mask_stride),
+                       *(uint64_t *)(mask + 3 * mask_stride));
+    const __m256i mask0a = _mm256_cvtepu8_epi16(mask0a8);
+    const __m256i mask0b = _mm256_cvtepu8_epi16(mask0b8);
+
+    highbd_blend_a64_d16_mask_w8_avx2(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
+        round_offset, shift, clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 4;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w8_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
+    int mask_stride, int h, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  const __m256i one_b = _mm256_set1_epi8(1);
+  const __m256i two_w = _mm256_set1_epi16(2);
+  do {
+    // Load 16x u8 pixels from each of 8 rows in the mask,
+    // (saturating) add together rows then use madd to add adjacent pixels
+    // Finally, divide each value by 4 (with rounding)
+    const __m256i m02 =
+        yy_loadu2_128(mask + 0 * mask_stride, mask + 2 * mask_stride);
+    const __m256i m13 =
+        yy_loadu2_128(mask + 1 * mask_stride, mask + 3 * mask_stride);
+    const __m256i m0123 =
+        _mm256_maddubs_epi16(_mm256_adds_epu8(m02, m13), one_b);
+    const __m256i mask_0a =
+        _mm256_srli_epi16(_mm256_add_epi16(m0123, two_w), 2);
+    const __m256i m46 =
+        yy_loadu2_128(mask + 4 * mask_stride, mask + 6 * mask_stride);
+    const __m256i m57 =
+        yy_loadu2_128(mask + 5 * mask_stride, mask + 7 * mask_stride);
+    const __m256i m4567 =
+        _mm256_maddubs_epi16(_mm256_adds_epu8(m46, m57), one_b);
+    const __m256i mask_0b =
+        _mm256_srli_epi16(_mm256_add_epi16(m4567, two_w), 2);
+
+    highbd_blend_a64_d16_mask_w8_avx2(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask_0a,
+        &mask_0b, round_offset, shift, clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 8;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_w16_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const __m256i *mask0a,
+    const __m256i *mask0b, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  // Load 16x pixels from each of 2 rows from each source
+  const __m256i s0a = yy_loadu_256(src0);
+  const __m256i s0b = yy_loadu_256(src0 + src0_stride);
+  const __m256i s1a = yy_loadu_256(src1);
+  const __m256i s1b = yy_loadu_256(src1 + src1_stride);
+
+  // Calculate inverse masks
+  const __m256i mask1a = _mm256_sub_epi16(*mask_max, *mask0a);
+  const __m256i mask1b = _mm256_sub_epi16(*mask_max, *mask0b);
+
+  // Multiply each source by appropriate mask
+  const __m256i mul0a_highs = _mm256_mulhi_epu16(*mask0a, s0a);
+  const __m256i mul0a_lows = _mm256_mullo_epi16(*mask0a, s0a);
+  const __m256i mul0ah = _mm256_unpackhi_epi16(mul0a_lows, mul0a_highs);
+  const __m256i mul0al = _mm256_unpacklo_epi16(mul0a_lows, mul0a_highs);
+  // Note that AVX2 unpack orders 64-bit words as [3 1] [2 0] to keep within
+  // lanes Later, packs does the same again which cancels this out with no need
+  // for a permute.  The intermediate values being reordered makes no difference
+
+  const __m256i mul1a_highs = _mm256_mulhi_epu16(mask1a, s1a);
+  const __m256i mul1a_lows = _mm256_mullo_epi16(mask1a, s1a);
+  const __m256i mul1ah = _mm256_unpackhi_epi16(mul1a_lows, mul1a_highs);
+  const __m256i mul1al = _mm256_unpacklo_epi16(mul1a_lows, mul1a_highs);
+
+  const __m256i mulah = _mm256_add_epi32(mul0ah, mul1ah);
+  const __m256i mulal = _mm256_add_epi32(mul0al, mul1al);
+
+  const __m256i mul0b_highs = _mm256_mulhi_epu16(*mask0b, s0b);
+  const __m256i mul0b_lows = _mm256_mullo_epi16(*mask0b, s0b);
+  const __m256i mul0bh = _mm256_unpackhi_epi16(mul0b_lows, mul0b_highs);
+  const __m256i mul0bl = _mm256_unpacklo_epi16(mul0b_lows, mul0b_highs);
+
+  const __m256i mul1b_highs = _mm256_mulhi_epu16(mask1b, s1b);
+  const __m256i mul1b_lows = _mm256_mullo_epi16(mask1b, s1b);
+  const __m256i mul1bh = _mm256_unpackhi_epi16(mul1b_lows, mul1b_highs);
+  const __m256i mul1bl = _mm256_unpacklo_epi16(mul1b_lows, mul1b_highs);
+
+  const __m256i mulbh = _mm256_add_epi32(mul0bh, mul1bh);
+  const __m256i mulbl = _mm256_add_epi32(mul0bl, mul1bl);
+
+  const __m256i resah =
+      _mm256_srai_epi32(_mm256_sub_epi32(mulah, *round_offset), shift);
+  const __m256i resal =
+      _mm256_srai_epi32(_mm256_sub_epi32(mulal, *round_offset), shift);
+  const __m256i resbh =
+      _mm256_srai_epi32(_mm256_sub_epi32(mulbh, *round_offset), shift);
+  const __m256i resbl =
+      _mm256_srai_epi32(_mm256_sub_epi32(mulbl, *round_offset), shift);
+
+  // Signed saturating pack from i32 to i16:
+  const __m256i packa = _mm256_packs_epi32(resal, resah);
+  const __m256i packb = _mm256_packs_epi32(resbl, resbh);
+
+  // Clip the values to the valid range
+  const __m256i clipa =
+      _mm256_min_epi16(_mm256_max_epi16(packa, *clip_low), *clip_high);
+  const __m256i clipb =
+      _mm256_min_epi16(_mm256_max_epi16(packb, *clip_low), *clip_high);
+
+  // Store 16 pixels
+  yy_storeu_256(dst, clipa);
+  yy_storeu_256(dst + dst_stride, clipb);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
+    int mask_stride, int h, int w, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  for (int i = 0; i < h; i += 2) {
+    for (int j = 0; j < w; j += 16) {
+      // Load 16x u8 alpha-mask values from each of two rows and pad to u16
+      const __m128i masks_a8 = xx_loadu_128(mask + j);
+      const __m128i masks_b8 = xx_loadu_128(mask + mask_stride + j);
+      const __m256i mask0a = _mm256_cvtepu8_epi16(masks_a8);
+      const __m256i mask0b = _mm256_cvtepu8_epi16(masks_b8);
+
+      highbd_blend_a64_d16_mask_w16_avx2(
+          dst + j, dst_stride, src0 + j, src0_stride, src1 + j, src1_stride,
+          &mask0a, &mask0b, round_offset, shift, clip_low, clip_high, mask_max);
+    }
+    dst += dst_stride * 2;
+    src0 += src0_stride * 2;
+    src1 += src1_stride * 2;
+    mask += mask_stride * 2;
+  }
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const uint8_t *mask,
+    int mask_stride, int h, int w, const __m256i *round_offset, int shift,
+    const __m256i *clip_low, const __m256i *clip_high,
+    const __m256i *mask_max) {
+  const __m256i one_b = _mm256_set1_epi8(1);
+  const __m256i two_w = _mm256_set1_epi16(2);
+  for (int i = 0; i < h; i += 2) {
+    for (int j = 0; j < w; j += 16) {
+      // Load 32x u8 alpha-mask values from each of four rows
+      // (saturating) add pairs of rows, then use madd to add adjacent values
+      // Finally, divide down each result with rounding
+      const __m256i m0 = yy_loadu_256(mask + 0 * mask_stride + 2 * j);
+      const __m256i m1 = yy_loadu_256(mask + 1 * mask_stride + 2 * j);
+      const __m256i m2 = yy_loadu_256(mask + 2 * mask_stride + 2 * j);
+      const __m256i m3 = yy_loadu_256(mask + 3 * mask_stride + 2 * j);
+
+      const __m256i m01_8 = _mm256_adds_epu8(m0, m1);
+      const __m256i m23_8 = _mm256_adds_epu8(m2, m3);
+
+      const __m256i m01 = _mm256_maddubs_epi16(m01_8, one_b);
+      const __m256i m23 = _mm256_maddubs_epi16(m23_8, one_b);
+
+      const __m256i mask0a = _mm256_srli_epi16(_mm256_add_epi16(m01, two_w), 2);
+      const __m256i mask0b = _mm256_srli_epi16(_mm256_add_epi16(m23, two_w), 2);
+
+      highbd_blend_a64_d16_mask_w16_avx2(
+          dst + j, dst_stride, src0 + j, src0_stride, src1 + j, src1_stride,
+          &mask0a, &mask0b, round_offset, shift, clip_low, clip_high, mask_max);
+    }
+    dst += dst_stride * 2;
+    src0 += src0_stride * 2;
+    src1 += src1_stride * 2;
+    mask += mask_stride * 4;
+  }
+}
+
+void aom_highbd_blend_a64_d16_mask_avx2(
+    uint8_t *dst8, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
+    ConvolveParams *conv_params, const int bd) {
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  const int round_bits =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const int32_t round_offset =
+      ((1 << (round_bits + bd)) + (1 << (round_bits + bd - 1)) -
+       (1 << (round_bits - 1)))
+      << AOM_BLEND_A64_ROUND_BITS;
+  const __m256i v_round_offset = _mm256_set1_epi32(round_offset);
+  const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;
+
+  const __m256i clip_low = _mm256_set1_epi16(0);
+  const __m256i clip_high = _mm256_set1_epi16((1 << bd) - 1);
+  const __m256i mask_max = _mm256_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
+
+  assert(IMPLIES((void *)src0 == dst, src0_stride == dst_stride));
+  assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));
+
+  assert(h >= 4);
+  assert(w >= 4);
+  assert(IS_POWER_OF_TWO(h));
+  assert(IS_POWER_OF_TWO(w));
+
+  if (subw == 0 && subh == 0) {
+    switch (w) {
+      case 4:
+        highbd_blend_a64_d16_mask_subw0_subh0_w4_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      case 8:
+        highbd_blend_a64_d16_mask_subw0_subh0_w8_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      default:  // >= 16
+        highbd_blend_a64_d16_mask_subw0_subh0_w16_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+    }
+
+  } else if (subw == 1 && subh == 1) {
+    switch (w) {
+      case 4:
+        highbd_blend_a64_d16_mask_subw1_subh1_w4_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      case 8:
+        highbd_blend_a64_d16_mask_subw1_subh1_w8_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      default:  // >= 16
+        highbd_blend_a64_d16_mask_subw1_subh1_w16_avx2(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+    }
+  } else {
+    // Sub-sampling in only one axis doesn't seem to happen very much, so fall
+    // back to the vanilla C implementation instead of having all the optimised
+    // code for these.
+    aom_highbd_blend_a64_d16_mask_c(dst8, dst_stride, src0, src0_stride, src1,
+                                    src1_stride, mask, mask_stride, w, h, subw,
+                                    subh, conv_params, bd);
+  }
+}
diff --git a/third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c b/third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c
index 9d6b4c2f74..b7a2468045 100644
--- a/third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c
+++ b/third_party/aom/aom_dsp/x86/blend_a64_mask_sse4.c
@@ -1107,3 +1107,452 @@ void aom_lowbd_blend_a64_d16_mask_sse4_1(
     }
   }
 }
+
+//////////////////////////////////////////////////////////////////////////////
+// aom_highbd_blend_a64_d16_mask_sse4_1()
+//////////////////////////////////////////////////////////////////////////////
+
+static INLINE void highbd_blend_a64_d16_mask_w4_sse4_1(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const __m128i *mask0a,
+    const __m128i *mask0b, const __m128i *round_offset, int shift,
+    const __m128i *clip_low, const __m128i *clip_high,
+    const __m128i *mask_max) {
+  // Load 4 pixels from each of 4 rows from each source
+  const __m128i s0a =
+      _mm_set_epi64x(*(uint64_t *)src0, *(uint64_t *)(src0 + src0_stride));
+  const __m128i s0b = _mm_set_epi64x(*(uint64_t *)(src0 + 2 * src0_stride),
+                                     *(uint64_t *)(src0 + 3 * src0_stride));
+  const __m128i s1a =
+      _mm_set_epi64x(*(uint64_t *)(src1), *(uint64_t *)(src1 + src1_stride));
+  const __m128i s1b = _mm_set_epi64x(*(uint64_t *)(src1 + 2 * src1_stride),
+                                     *(uint64_t *)(src1 + 3 * src1_stride));
+
+  // Generate the inverse masks
+  const __m128i mask1a = _mm_sub_epi16(*mask_max, *mask0a);
+  const __m128i mask1b = _mm_sub_epi16(*mask_max, *mask0b);
+
+  // Multiply each mask by the respective source
+  const __m128i mul0a_highs = _mm_mulhi_epu16(*mask0a, s0a);
+  const __m128i mul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
+  const __m128i mul0ah = _mm_unpackhi_epi16(mul0a_lows, mul0a_highs);
+  const __m128i mul0al = _mm_unpacklo_epi16(mul0a_lows, mul0a_highs);
+  const __m128i mul1a_highs = _mm_mulhi_epu16(mask1a, s1a);
+  const __m128i mul1a_lows = _mm_mullo_epi16(mask1a, s1a);
+  const __m128i mul1ah = _mm_unpackhi_epi16(mul1a_lows, mul1a_highs);
+  const __m128i mul1al = _mm_unpacklo_epi16(mul1a_lows, mul1a_highs);
+
+  const __m128i mul0b_highs = _mm_mulhi_epu16(*mask0b, s0b);
+  const __m128i mul0b_lows = _mm_mullo_epi16(*mask0b, s0b);
+  const __m128i mul0bh = _mm_unpackhi_epi16(mul0b_lows, mul0b_highs);
+  const __m128i mul0bl = _mm_unpacklo_epi16(mul0b_lows, mul0b_highs);
+  const __m128i mul1b_highs = _mm_mulhi_epu16(mask1b, s1b);
+  const __m128i mul1b_lows = _mm_mullo_epi16(mask1b, s1b);
+  const __m128i mul1bh = _mm_unpackhi_epi16(mul1b_lows, mul1b_highs);
+  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);
+
+  const __m128i sumah = _mm_add_epi32(mul0ah, mul1ah);
+  const __m128i sumal = _mm_add_epi32(mul0al, mul1al);
+  const __m128i sumbh = _mm_add_epi32(mul0bh, mul1bh);
+  const __m128i sumbl = _mm_add_epi32(mul0bl, mul1bl);
+
+  const __m128i roundah =
+      _mm_srai_epi32(_mm_sub_epi32(sumah, *round_offset), shift);
+  const __m128i roundbh =
+      _mm_srai_epi32(_mm_sub_epi32(sumbh, *round_offset), shift);
+  const __m128i roundal =
+      _mm_srai_epi32(_mm_sub_epi32(sumal, *round_offset), shift);
+  const __m128i roundbl =
+      _mm_srai_epi32(_mm_sub_epi32(sumbl, *round_offset), shift);
+
+  const __m128i packa = _mm_packs_epi32(roundal, roundah);
+  const __m128i packb = _mm_packs_epi32(roundbl, roundbh);
+
+  const __m128i clipa =
+      _mm_min_epi16(_mm_max_epi16(packa, *clip_low), *clip_high);
+  const __m128i clipb =
+      _mm_min_epi16(_mm_max_epi16(packb, *clip_low), *clip_high);
+
+  xx_storel_64(dst, _mm_srli_si128(clipa, 8));
+  xx_storel_64(dst + dst_stride, clipa);
+  xx_storel_64(dst + 2 * dst_stride, _mm_srli_si128(clipb, 8));
+  xx_storel_64(dst + 3 * dst_stride, clipb);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *mask_max) {
+  do {
+    const __m128i mask0a8 = _mm_set_epi32(0, 0, *(uint32_t *)mask,
+                                          *(uint32_t *)(mask + mask_stride));
+    const __m128i mask0b8 =
+        _mm_set_epi32(0, 0, *(uint32_t *)(mask + 2 * mask_stride),
+                      *(uint32_t *)(mask + 3 * mask_stride));
+    const __m128i mask0a = _mm_cvtepu8_epi16(mask0a8);
+    const __m128i mask0b = _mm_cvtepu8_epi16(mask0b8);
+
+    highbd_blend_a64_d16_mask_w4_sse4_1(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
+        round_offset, shift, clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 4;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *mask_max) {
+  const __m128i one_b = _mm_set1_epi8(1);
+  const __m128i two_w = _mm_set1_epi16(2);
+  do {
+    // Load 8 pixels from each of 8 rows of mask,
+    // (saturating) add together rows then use madd to add adjacent pixels
+    // Finally, divide each value by 4 (with rounding)
+    const __m128i m02 = _mm_set_epi64x(*(uint64_t *)(mask),
+                                       *(uint64_t *)(mask + 2 * mask_stride));
+    const __m128i m13 = _mm_set_epi64x(*(uint64_t *)(mask + mask_stride),
+                                       *(uint64_t *)(mask + 3 * mask_stride));
+    const __m128i m0123 = _mm_maddubs_epi16(_mm_adds_epu8(m02, m13), one_b);
+    const __m128i mask_0a = _mm_srli_epi16(_mm_add_epi16(m0123, two_w), 2);
+    const __m128i m46 = _mm_set_epi64x(*(uint64_t *)(mask + 4 * mask_stride),
+                                       *(uint64_t *)(mask + 6 * mask_stride));
+    const __m128i m57 = _mm_set_epi64x(*(uint64_t *)(mask + 5 * mask_stride),
+                                       *(uint64_t *)(mask + 7 * mask_stride));
+    const __m128i m4567 = _mm_maddubs_epi16(_mm_adds_epu8(m46, m57), one_b);
+    const __m128i mask_0b = _mm_srli_epi16(_mm_add_epi16(m4567, two_w), 2);
+
+    highbd_blend_a64_d16_mask_w4_sse4_1(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask_0a,
+        &mask_0b, round_offset, shift, clip_low, clip_high, mask_max);
+
+    dst += dst_stride * 4;
+    src0 += src0_stride * 4;
+    src1 += src1_stride * 4;
+    mask += mask_stride * 8;
+  } while (h -= 4);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_w8_sse4_1(
+    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
+    const CONV_BUF_TYPE *src1, int src1_stride, const __m128i *mask0a,
+    const __m128i *mask0b, const __m128i *round_offset, int shift,
+    const __m128i *clip_low, const __m128i *clip_high,
+    const __m128i *max_mask) {
+  // Load 8x pixels from each of 2 rows from each source
+  const __m128i s0a = xx_loadu_128(src0);
+  const __m128i s0b = xx_loadu_128(src0 + src0_stride);
+  const __m128i s1a = xx_loadu_128(src1);
+  const __m128i s1b = xx_loadu_128(src1 + src1_stride);
+
+  // Generate inverse masks
+  const __m128i mask1a = _mm_sub_epi16(*max_mask, *mask0a);
+  const __m128i mask1b = _mm_sub_epi16(*max_mask, *mask0b);
+
+  // Multiply sources by respective masks
+  const __m128i mul0a_highs = _mm_mulhi_epu16(*mask0a, s0a);
+  const __m128i mul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
+  const __m128i mul0ah = _mm_unpackhi_epi16(mul0a_lows, mul0a_highs);
+  const __m128i mul0al = _mm_unpacklo_epi16(mul0a_lows, mul0a_highs);
+
+  const __m128i mul1a_highs = _mm_mulhi_epu16(mask1a, s1a);
+  const __m128i mul1a_lows = _mm_mullo_epi16(mask1a, s1a);
+  const __m128i mul1ah = _mm_unpackhi_epi16(mul1a_lows, mul1a_highs);
+  const __m128i mul1al = _mm_unpacklo_epi16(mul1a_lows, mul1a_highs);
+
+  const __m128i sumah = _mm_add_epi32(mul0ah, mul1ah);
+  const __m128i sumal = _mm_add_epi32(mul0al, mul1al);
+
+  const __m128i mul0b_highs = _mm_mulhi_epu16(*mask0b, s0b);
+  const __m128i mul0b_lows = _mm_mullo_epi16(*mask0b, s0b);
+  const __m128i mul0bh = _mm_unpackhi_epi16(mul0b_lows, mul0b_highs);
+  const __m128i mul0bl = _mm_unpacklo_epi16(mul0b_lows, mul0b_highs);
+
+  const __m128i mul1b_highs = _mm_mulhi_epu16(mask1b, s1b);
+  const __m128i mul1b_lows = _mm_mullo_epi16(mask1b, s1b);
+  const __m128i mul1bh = _mm_unpackhi_epi16(mul1b_lows, mul1b_highs);
+  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);
+
+  const __m128i sumbh = _mm_add_epi32(mul0bh, mul1bh);
+  const __m128i sumbl = _mm_add_epi32(mul0bl, mul1bl);
+
+  const __m128i roundah =
+      _mm_srai_epi32(_mm_sub_epi32(sumah, *round_offset), shift);
+  const __m128i roundal =
+      _mm_srai_epi32(_mm_sub_epi32(sumal, *round_offset), shift);
+  const __m128i roundbh =
+      _mm_srai_epi32(_mm_sub_epi32(sumbh, *round_offset), shift);
+  const __m128i roundbl =
+      _mm_srai_epi32(_mm_sub_epi32(sumbl, *round_offset), shift);
+
+  const __m128i packa = _mm_packs_epi32(roundal, roundah);
+  const __m128i clipa =
+      _mm_min_epi16(_mm_max_epi16(packa, *clip_low), *clip_high);
+  const __m128i packb = _mm_packs_epi32(roundbl, roundbh);
+  const __m128i clipb =
+      _mm_min_epi16(_mm_max_epi16(packb, *clip_low), *clip_high);
+
+  xx_storeu_128(dst, clipa);
+  xx_storeu_128(dst + dst_stride, clipb);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *max_mask) {
+  do {
+    const __m128i mask0a = _mm_cvtepu8_epi16(xx_loadl_64(mask));
+    const __m128i mask0b = _mm_cvtepu8_epi16(xx_loadl_64(mask + mask_stride));
+    highbd_blend_a64_d16_mask_w8_sse4_1(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
+        round_offset, shift, clip_low, clip_high, max_mask);
+
+    dst += dst_stride * 2;
+    src0 += src0_stride * 2;
+    src1 += src1_stride * 2;
+    mask += mask_stride * 2;
+  } while (h -= 2);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *max_mask) {
+  const __m128i one_b = _mm_set1_epi8(1);
+  const __m128i two_w = _mm_set1_epi16(2);
+  do {
+    const __m128i mask_thisrowa = xx_loadu_128(mask);
+    const __m128i mask_nextrowa = xx_loadu_128(mask + mask_stride);
+    const __m128i mask_thisrowb = xx_loadu_128(mask + 2 * mask_stride);
+    const __m128i mask_nextrowb = xx_loadu_128(mask + 3 * mask_stride);
+    const __m128i mask_bothrowsa = _mm_adds_epu8(mask_thisrowa, mask_nextrowa);
+    const __m128i mask_bothrowsb = _mm_adds_epu8(mask_thisrowb, mask_nextrowb);
+    const __m128i mask_16a = _mm_maddubs_epi16(mask_bothrowsa, one_b);
+    const __m128i mask_16b = _mm_maddubs_epi16(mask_bothrowsb, one_b);
+    const __m128i mask_sa = _mm_srli_epi16(_mm_add_epi16(mask_16a, two_w), 2);
+    const __m128i mask_sb = _mm_srli_epi16(_mm_add_epi16(mask_16b, two_w), 2);
+
+    highbd_blend_a64_d16_mask_w8_sse4_1(
+        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask_sa,
+        &mask_sb, round_offset, shift, clip_low, clip_high, max_mask);
+
+    dst += dst_stride * 2;
+    src0 += src0_stride * 2;
+    src1 += src1_stride * 2;
+    mask += mask_stride * 4;
+  } while (h -= 2);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_w16_sse4_1(
+    uint16_t *dst, const CONV_BUF_TYPE *src0, const CONV_BUF_TYPE *src1,
+    const __m128i *round_offset, int shift, const __m128i *mask0l,
+    const __m128i *mask0h, const __m128i *clip_low, const __m128i *clip_high,
+    const __m128i *mask_max) {
+  // Load 16x u16 pixels for this row from each src
+  const __m128i s0l = xx_loadu_128(src0);
+  const __m128i s0h = xx_loadu_128(src0 + 8);
+  const __m128i s1l = xx_loadu_128(src1);
+  const __m128i s1h = xx_loadu_128(src1 + 8);
+
+  // Calculate inverse masks
+  const __m128i mask1h = _mm_sub_epi16(*mask_max, *mask0h);
+  const __m128i mask1l = _mm_sub_epi16(*mask_max, *mask0l);
+
+  const __m128i mul0_highs = _mm_mulhi_epu16(*mask0h, s0h);
+  const __m128i mul0_lows = _mm_mullo_epi16(*mask0h, s0h);
+  const __m128i mul0h = _mm_unpackhi_epi16(mul0_lows, mul0_highs);
+  const __m128i mul0l = _mm_unpacklo_epi16(mul0_lows, mul0_highs);
+
+  const __m128i mul1_highs = _mm_mulhi_epu16(mask1h, s1h);
+  const __m128i mul1_lows = _mm_mullo_epi16(mask1h, s1h);
+  const __m128i mul1h = _mm_unpackhi_epi16(mul1_lows, mul1_highs);
+  const __m128i mul1l = _mm_unpacklo_epi16(mul1_lows, mul1_highs);
+
+  const __m128i mulhh = _mm_add_epi32(mul0h, mul1h);
+  const __m128i mulhl = _mm_add_epi32(mul0l, mul1l);
+
+  const __m128i mul2_highs = _mm_mulhi_epu16(*mask0l, s0l);
+  const __m128i mul2_lows = _mm_mullo_epi16(*mask0l, s0l);
+  const __m128i mul2h = _mm_unpackhi_epi16(mul2_lows, mul2_highs);
+  const __m128i mul2l = _mm_unpacklo_epi16(mul2_lows, mul2_highs);
+
+  const __m128i mul3_highs = _mm_mulhi_epu16(mask1l, s1l);
+  const __m128i mul3_lows = _mm_mullo_epi16(mask1l, s1l);
+  const __m128i mul3h = _mm_unpackhi_epi16(mul3_lows, mul3_highs);
+  const __m128i mul3l = _mm_unpacklo_epi16(mul3_lows, mul3_highs);
+
+  const __m128i mullh = _mm_add_epi32(mul2h, mul3h);
+  const __m128i mulll = _mm_add_epi32(mul2l, mul3l);
+
+  const __m128i reshh =
+      _mm_srai_epi32(_mm_sub_epi32(mulhh, *round_offset), shift);
+  const __m128i reshl =
+      _mm_srai_epi32(_mm_sub_epi32(mulhl, *round_offset), shift);
+  const __m128i reslh =
+      _mm_srai_epi32(_mm_sub_epi32(mullh, *round_offset), shift);
+  const __m128i resll =
+      _mm_srai_epi32(_mm_sub_epi32(mulll, *round_offset), shift);
+
+  // Signed saturating pack from i32 to i16:
+  const __m128i packh = _mm_packs_epi32(reshl, reshh);
+  const __m128i packl = _mm_packs_epi32(resll, reslh);
+
+  // Clip the values to the valid range
+  const __m128i cliph =
+      _mm_min_epi16(_mm_max_epi16(packh, *clip_low), *clip_high);
+  const __m128i clipl =
+      _mm_min_epi16(_mm_max_epi16(packl, *clip_low), *clip_high);
+
+  // Store 16 pixels
+  xx_storeu_128(dst, clipl);
+  xx_storeu_128(dst + 8, cliph);
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h, int w,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *mask_max) {
+  for (int i = 0; i < h; i++) {
+    for (int j = 0; j < w; j += 16) {
+      // Load 16x u8 alpha-mask values and pad to u16
+      const __m128i masks_u8 = xx_loadu_128(mask + j);
+      const __m128i mask0l = _mm_cvtepu8_epi16(masks_u8);
+      const __m128i mask0h = _mm_cvtepu8_epi16(_mm_srli_si128(masks_u8, 8));
+
+      highbd_blend_a64_d16_mask_w16_sse4_1(
+          dst + j, src0 + j, src1 + j, round_offset, shift, &mask0l, &mask0h,
+          clip_low, clip_high, mask_max);
+    }
+    dst += dst_stride;
+    src0 += src0_stride;
+    src1 += src1_stride;
+    mask += mask_stride;
+  }
+}
+
+static INLINE void highbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1(
+    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int h, int w,
+    const __m128i *round_offset, int shift, const __m128i *clip_low,
+    const __m128i *clip_high, const __m128i *mask_max) {
+  const __m128i one_b = _mm_set1_epi8(1);
+  const __m128i two_w = _mm_set1_epi16(2);
+  for (int i = 0; i < h; i++) {
+    for (int j = 0; j < w; j += 16) {
+      const __m128i m_i00 = xx_loadu_128(mask + 2 * j);
+      const __m128i m_i01 = xx_loadu_128(mask + 2 * j + 16);
+      const __m128i m_i10 = xx_loadu_128(mask + mask_stride + 2 * j);
+      const __m128i m_i11 = xx_loadu_128(mask + mask_stride + 2 * j + 16);
+
+      const __m128i m0_ac = _mm_adds_epu8(m_i00, m_i10);
+      const __m128i m1_ac = _mm_adds_epu8(m_i01, m_i11);
+      const __m128i m0_acbd = _mm_maddubs_epi16(m0_ac, one_b);
+      const __m128i m1_acbd = _mm_maddubs_epi16(m1_ac, one_b);
+      const __m128i mask_l = _mm_srli_epi16(_mm_add_epi16(m0_acbd, two_w), 2);
+      const __m128i mask_h = _mm_srli_epi16(_mm_add_epi16(m1_acbd, two_w), 2);
+
+      highbd_blend_a64_d16_mask_w16_sse4_1(
+          dst + j, src0 + j, src1 + j, round_offset, shift, &mask_l, &mask_h,
+          clip_low, clip_high, mask_max);
+    }
+    dst += dst_stride;
+    src0 += src0_stride;
+    src1 += src1_stride;
+    mask += mask_stride * 2;
+  }
+}
+
+void aom_highbd_blend_a64_d16_mask_sse4_1(
+    uint8_t *dst8, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
+    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+    const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
+    ConvolveParams *conv_params, const int bd) {
+  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  const int round_bits =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const int32_t round_offset =
+      ((1 << (round_bits + bd)) + (1 << (round_bits + bd - 1)) -
+       (1 << (round_bits - 1)))
+      << AOM_BLEND_A64_ROUND_BITS;
+  const __m128i v_round_offset = _mm_set1_epi32(round_offset);
+  const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;
+
+  const __m128i clip_low = _mm_set1_epi16(0);
+  const __m128i clip_high = _mm_set1_epi16((1 << bd) - 1);
+  const __m128i mask_max = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
+
+  assert(IMPLIES((void *)src0 == dst, src0_stride == dst_stride));
+  assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));
+
+  assert(h >= 4);
+  assert(w >= 4);
+  assert(IS_POWER_OF_TWO(h));
+  assert(IS_POWER_OF_TWO(w));
+
+  if (subw == 0 && subh == 0) {
+    switch (w) {
+      case 4:
+        highbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      case 8:
+        highbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      default:  // >=16
+        highbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+    }
+
+  } else if (subw == 1 && subh == 1) {
+    switch (w) {
+      case 4:
+        highbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      case 8:
+        highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+      default:  // >=16
+        highbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1(
+            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
+            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
+            &mask_max);
+        break;
+    }
+  } else {
+    // Sub-sampling in only one axis doesn't seem to happen very much, so fall
+    // back to the vanilla C implementation instead of having all the optimised
+    // code for these.
+    aom_highbd_blend_a64_d16_mask_c(dst8, dst_stride, src0, src0_stride, src1,
+                                    src1_stride, mask, mask_stride, w, h, subw,
+                                    subh, conv_params, bd);
+  }
+}
diff --git a/third_party/aom/aom_dsp/x86/convolve.h b/third_party/aom/aom_dsp/x86/convolve.h
index 3e19682cd9..ff3224e5ad 100644
--- a/third_party/aom/aom_dsp/x86/convolve.h
+++ b/third_party/aom/aom_dsp/x86/convolve.h
@@ -122,7 +122,30 @@ typedef void highbd_filter8_1dfunction(const uint16_t *src_ptr,
     uint16_t *src = CONVERT_TO_SHORTPTR(src8);                             \
     uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);                             \
     if (step_q4 == 16 && filter[3] != 128) {                               \
-      if (filter[0] | filter[1] | filter[2]) {                             \
+      if (((filter[0] | filter[1] | filter[6] | filter[7]) == 0) &&        \
+          (filter[2] | filter[5])) {                                       \
+        while (w >= 16) {                                                  \
+          aom_highbd_filter_block1d16_##dir##4_##avg##opt(                 \
+              src_start, src_stride, dst, dst_stride, h, filter, bd);      \
+          src += 16;                                                       \
+          dst += 16;                                                       \
+          w -= 16;                                                         \
+        }                                                                  \
+        while (w >= 8) {                                                   \
+          aom_highbd_filter_block1d8_##dir##4_##avg##opt(                  \
+              src_start, src_stride, dst, dst_stride, h, filter, bd);      \
+          src += 8;                                                        \
+          dst += 8;                                                        \
+          w -= 8;                                                          \
+        }                                                                  \
+        while (w >= 4) {                                                   \
+          aom_highbd_filter_block1d4_##dir##4_##avg##opt(                  \
+              src_start, src_stride, dst, dst_stride, h, filter, bd);      \
+          src += 4;                                                        \
+          dst += 4;                                                        \
+          w -= 4;                                                          \
+        }                                                                  \
+      } else if (filter[0] | filter[1] | filter[2]) {                      \
         while (w >= 16) {                                                  \
           aom_highbd_filter_block1d16_##dir##8_##avg##opt(                 \
               src_start, src_stride, dst, dst_stride, h, filter, bd);      \
diff --git a/third_party/aom/aom_dsp/x86/convolve_avx2.h b/third_party/aom/aom_dsp/x86/convolve_avx2.h
index 30253f65c2..e041a6230e 100644
--- a/third_party/aom/aom_dsp/x86/convolve_avx2.h
+++ b/third_party/aom/aom_dsp/x86/convolve_avx2.h
@@ -34,6 +34,31 @@ DECLARE_ALIGNED(32, static const uint8_t, filt4_d4_global_avx2[]) = {
   2, 3, 4, 5, 3, 4, 5, 6, 4, 5, 6, 7, 5, 6, 7, 8,
 };
 
+DECLARE_ALIGNED(32, static const uint8_t, filt_center_global_avx2[32]) = {
+  3, 255, 4, 255, 5, 255, 6, 255, 7, 255, 8, 255, 9, 255, 10, 255,
+  3, 255, 4, 255, 5, 255, 6, 255, 7, 255, 8, 255, 9, 255, 10, 255
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt1_global_avx2[32]) = {
+  0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
+  0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt2_global_avx2[32]) = {
+  2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10,
+  2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt3_global_avx2[32]) = {
+  4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12,
+  4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt4_global_avx2[32]) = {
+  6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14,
+  6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
+};
+
 static INLINE void prepare_coeffs_lowbd(
     const InterpFilterParams *const filter_params, const int subpel_q4,
     __m256i *const coeffs /* [4] */) {
@@ -108,6 +133,15 @@ static INLINE __m256i convolve(const __m256i *const s,
   return res;
 }
 
+static INLINE __m256i convolve_4tap(const __m256i *const s,
+                                    const __m256i *const coeffs) {
+  const __m256i res_1 = _mm256_madd_epi16(s[0], coeffs[0]);
+  const __m256i res_2 = _mm256_madd_epi16(s[1], coeffs[1]);
+
+  const __m256i res = _mm256_add_epi32(res_1, res_2);
+  return res;
+}
+
 static INLINE __m256i convolve_lowbd_x(const __m256i data,
                                        const __m256i *const coeffs,
                                        const __m256i *const filt) {
@@ -138,9 +172,9 @@ static INLINE void add_store_aligned_256(CONV_BUF_TYPE *const dst,
 static INLINE __m256i comp_avg(const __m256i *const data_ref_0,
                                const __m256i *const res_unsigned,
                                const __m256i *const wt,
-                               const int use_jnt_comp_avg) {
+                               const int use_dist_wtd_comp_avg) {
   __m256i res;
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const __m256i data_lo = _mm256_unpacklo_epi16(*data_ref_0, *res_unsigned);
     const __m256i data_hi = _mm256_unpackhi_epi16(*data_ref_0, *res_unsigned);
 
@@ -172,9 +206,9 @@ static INLINE __m256i highbd_comp_avg(const __m256i *const data_ref_0,
                                       const __m256i *const res_unsigned,
                                       const __m256i *const wt0,
                                       const __m256i *const wt1,
-                                      const int use_jnt_comp_avg) {
+                                      const int use_dist_wtd_comp_avg) {
   __m256i res;
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const __m256i wt0_res = _mm256_mullo_epi32(*data_ref_0, *wt0);
     const __m256i wt1_res = _mm256_mullo_epi32(*res_unsigned, *wt1);
     const __m256i wt_res = _mm256_add_epi32(wt0_res, wt1_res);
diff --git a/third_party/aom/aom_dsp/x86/convolve_sse2.h b/third_party/aom/aom_dsp/x86/convolve_sse2.h
index 445d04b103..385c7c7e18 100644
--- a/third_party/aom/aom_dsp/x86/convolve_sse2.h
+++ b/third_party/aom/aom_dsp/x86/convolve_sse2.h
@@ -78,9 +78,9 @@ static INLINE __m128i convolve_hi_y(const __m128i *const s,
 static INLINE __m128i comp_avg(const __m128i *const data_ref_0,
                                const __m128i *const res_unsigned,
                                const __m128i *const wt,
-                               const int use_jnt_comp_avg) {
+                               const int use_dist_wtd_avg) {
   __m128i res;
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_avg) {
     const __m128i data_lo = _mm_unpacklo_epi16(*data_ref_0, *res_unsigned);
     const __m128i data_hi = _mm_unpackhi_epi16(*data_ref_0, *res_unsigned);
 
diff --git a/third_party/aom/aom_dsp/x86/convolve_sse4_1.h b/third_party/aom/aom_dsp/x86/convolve_sse4_1.h
index 6b8388d84a..b1a3bb4664 100644
--- a/third_party/aom/aom_dsp/x86/convolve_sse4_1.h
+++ b/third_party/aom/aom_dsp/x86/convolve_sse4_1.h
@@ -35,9 +35,9 @@ static INLINE __m128i highbd_comp_avg_sse4_1(const __m128i *const data_ref_0,
                                              const __m128i *const res_unsigned,
                                              const __m128i *const wt0,
                                              const __m128i *const wt1,
-                                             const int use_jnt_comp_avg) {
+                                             const int use_dist_wtd_avg) {
   __m128i res;
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_avg) {
     const __m128i wt0_res = _mm_mullo_epi32(*data_ref_0, *wt0);
     const __m128i wt1_res = _mm_mullo_epi32(*res_unsigned, *wt1);
 
diff --git a/third_party/aom/aom_dsp/x86/fft_avx2.c b/third_party/aom/aom_dsp/x86/fft_avx2.c
index 54da022538..4cccc5f00f 100644
--- a/third_party/aom/aom_dsp/x86/fft_avx2.c
+++ b/third_party/aom/aom_dsp/x86/fft_avx2.c
@@ -11,6 +11,7 @@
 
 #include <immintrin.h>
 
+#include "config/aom_dsp_rtcd.h"
 #include "aom_dsp/aom_dsp_common.h"
 #include "aom_dsp/fft_common.h"
 
diff --git a/third_party/aom/aom_dsp/x86/fft_sse2.c b/third_party/aom/aom_dsp/x86/fft_sse2.c
index 12bdc3e185..6f20a3cc01 100644
--- a/third_party/aom/aom_dsp/x86/fft_sse2.c
+++ b/third_party/aom/aom_dsp/x86/fft_sse2.c
@@ -11,6 +11,7 @@ s * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 
 #include <xmmintrin.h>
 
+#include "config/aom_dsp_rtcd.h"
 #include "aom_dsp/aom_dsp_common.h"
 #include "aom_dsp/fft_common.h"
 
diff --git a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
index 2d8f8f71e4..6b7c11b18d 100644
--- a/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
+++ b/third_party/aom/aom_dsp/x86/fwd_txfm_sse2.c
@@ -17,46 +17,6 @@
 #include "aom_dsp/aom_dsp_common.h"
 #include "aom_dsp/x86/fwd_txfm_sse2.h"
 
-void aom_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
-  __m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
-  __m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
-  __m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
-  __m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
-  __m128i u0, u1, sum;
-
-  u0 = _mm_add_epi16(in0, in1);
-  u1 = _mm_add_epi16(in2, in3);
-
-  in0 = _mm_load_si128((const __m128i *)(input + 4 * stride));
-  in1 = _mm_load_si128((const __m128i *)(input + 5 * stride));
-  in2 = _mm_load_si128((const __m128i *)(input + 6 * stride));
-  in3 = _mm_load_si128((const __m128i *)(input + 7 * stride));
-
-  sum = _mm_add_epi16(u0, u1);
-
-  in0 = _mm_add_epi16(in0, in1);
-  in2 = _mm_add_epi16(in2, in3);
-  sum = _mm_add_epi16(sum, in0);
-
-  u0 = _mm_setzero_si128();
-  sum = _mm_add_epi16(sum, in2);
-
-  in0 = _mm_unpacklo_epi16(u0, sum);
-  in1 = _mm_unpackhi_epi16(u0, sum);
-  in0 = _mm_srai_epi32(in0, 16);
-  in1 = _mm_srai_epi32(in1, 16);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_unpacklo_epi32(sum, u0);
-  in1 = _mm_unpackhi_epi32(sum, u0);
-
-  sum = _mm_add_epi32(in0, in1);
-  in0 = _mm_srli_si128(sum, 8);
-
-  in1 = _mm_add_epi32(sum, in0);
-  output[0] = (tran_low_t)_mm_cvtsi128_si32(in1);
-}
-
 #define DCT_HIGH_BIT_DEPTH 0
 #define FDCT8x8_2D aom_fdct8x8_sse2
 #include "aom_dsp/x86/fwd_txfm_impl_sse2.h"
diff --git a/third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c b/third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c
index 099fcf7fc6..ebcb5ac2e6 100644
--- a/third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_convolve_avx2.c
@@ -20,6 +20,14 @@
 // -----------------------------------------------------------------------------
 // Copy and average
 
+static const uint8_t ip_shuffle_f2f3[32] = { 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6,
+                                             7, 6, 7, 8, 9, 0, 1, 2, 3, 2, 3,
+                                             4, 5, 4, 5, 6, 7, 6, 7, 8, 9 };
+static const uint8_t ip_shuffle_f4f5[32] = { 4, 5, 6,  7,  6,  7,  8,  9,
+                                             8, 9, 10, 11, 10, 11, 12, 13,
+                                             4, 5, 6,  7,  6,  7,  8,  9,
+                                             8, 9, 10, 11, 10, 11, 12, 13 };
+
 void aom_highbd_convolve_copy_avx2(const uint8_t *src8, ptrdiff_t src_stride,
                                    uint8_t *dst8, ptrdiff_t dst_stride,
                                    const int16_t *filter_x, int filter_x_stride,
@@ -444,6 +452,17 @@ static INLINE void pack_filters(const int16_t *filter, __m256i *f /*f[4]*/) {
   f[3] = _mm256_shuffle_epi8(hh, p3);
 }
 
+static INLINE void pack_filters_4tap(const int16_t *filter,
+                                     __m256i *f /*f[4]*/) {
+  const __m128i h = _mm_loadu_si128((const __m128i *)filter);
+  const __m256i coeff = _mm256_broadcastsi128_si256(h);
+
+  // coeffs 2 3 2 3 2 3 2 3
+  f[0] = _mm256_shuffle_epi32(coeff, 0x55);
+  // coeffs 4 5 4 5 4 5 4 5
+  f[1] = _mm256_shuffle_epi32(coeff, 0xaa);
+}
+
 static INLINE void filter_8x1_pixels(const __m256i *sig /*sig[4]*/,
                                      const __m256i *fil /*fil[4]*/,
                                      __m256i *y) {
@@ -544,6 +563,176 @@ static void aom_highbd_filter_block1d16_h8_avx2(
   } while (height > 0);
 }
 
+static void aom_highbd_filter_block1d4_h4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  const __m256i rounding = _mm256_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
+  __m256i ff[2], s[2];
+  uint32_t i;
+  const __m256i clip_pixel =
+      _mm256_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+  const __m256i zero = _mm256_setzero_si256();
+
+  static const uint8_t shuffle_mask[32] = { 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6,
+                                            7, 6, 7, 8, 9, 0, 1, 2, 3, 2, 3,
+                                            4, 5, 4, 5, 6, 7, 6, 7, 8, 9 };
+
+  __m256i mask = _mm256_loadu_si256((__m256i *)shuffle_mask);
+  __m256i ip_mask_f2f3 = _mm256_loadu_si256((__m256i *)ip_shuffle_f2f3);
+  __m256i ip_mask_f4f5 = _mm256_loadu_si256((__m256i *)ip_shuffle_f4f5);
+
+  pack_filters_4tap(filter, ff);
+  src_ptr -= 3;
+  for (i = 0; i <= (height - 2); i += 2) {
+    __m256i row0 = _mm256_castsi128_si256(
+        _mm_loadu_si128((__m128i *)&src_ptr[i * src_pitch + 2]));
+    __m256i row1 = _mm256_castsi128_si256(
+        _mm_loadu_si128((__m128i *)&src_ptr[(i + 1) * src_pitch + 2]));
+
+    s[0] = _mm256_inserti128_si256(row0, _mm256_castsi256_si128(row1), 1);
+    s[1] = _mm256_alignr_epi8(s[0], s[0], 4);
+
+    s[0] = _mm256_shuffle_epi8(s[0], mask);
+    s[1] = _mm256_shuffle_epi8(s[1], mask);
+
+    __m256i res = convolve_4tap(s, ff);
+    res =
+        _mm256_srai_epi32(_mm256_add_epi32(res, rounding), CONV8_ROUNDING_BITS);
+
+    res = _mm256_packs_epi32(res, res);
+    res = _mm256_min_epi16(res, clip_pixel);
+    res = _mm256_max_epi16(res, zero);
+
+    _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch],
+                     _mm256_castsi256_si128(res));
+    _mm_storel_epi64((__m128i *)&dst_ptr[(i + 1) * dst_pitch],
+                     _mm256_extracti128_si256(res, 1));
+  }
+  if (height % 2 != 0) {
+    i = height - 1;
+    const __m256i row0_0 = _mm256_castsi128_si256(
+        _mm_loadu_si128((__m128i *)&src_ptr[i * src_pitch + 2]));
+    const __m256i row0_1 = _mm256_castsi128_si256(
+        _mm_loadu_si128((__m128i *)&src_ptr[i * src_pitch + 6]));
+
+    const __m256i r0 =
+        _mm256_inserti128_si256(row0_0, _mm256_castsi256_si128(row0_1), 1);
+
+    s[0] = _mm256_shuffle_epi8(r0, ip_mask_f2f3);
+    s[1] = _mm256_shuffle_epi8(r0, ip_mask_f4f5);
+
+    __m256i res = convolve_4tap(s, ff);
+    res =
+        _mm256_srai_epi32(_mm256_add_epi32(res, rounding), CONV8_ROUNDING_BITS);
+
+    res = _mm256_packs_epi32(res, res);
+    res = _mm256_min_epi16(res, clip_pixel);
+    res = _mm256_max_epi16(res, zero);
+
+    _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch],
+                     _mm256_castsi256_si128(res));
+  }
+}
+
+static void aom_highbd_filter_block1d8_h4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  const __m256i rounding = _mm256_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
+  __m256i ff[2], s[2];
+  uint32_t i = 0;
+  const __m256i clip_pixel =
+      _mm256_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+  const __m256i zero = _mm256_setzero_si256();
+
+  static const uint8_t shuffle_mask[32] = { 0, 1, 8,  9,  2, 3, 10, 11,
+                                            4, 5, 12, 13, 6, 7, 14, 15,
+                                            0, 1, 8,  9,  2, 3, 10, 11,
+                                            4, 5, 12, 13, 6, 7, 14, 15 };
+
+  __m256i mask = _mm256_loadu_si256((__m256i *)shuffle_mask);
+  __m256i ip_mask_f2f3 = _mm256_loadu_si256((__m256i *)ip_shuffle_f2f3);
+  __m256i ip_mask_f4f5 = _mm256_loadu_si256((__m256i *)ip_shuffle_f4f5);
+
+  pack_filters_4tap(filter, ff);
+  src_ptr -= 3;
+
+  /* Horizontal filter */
+
+  for (i = 0; i <= (height - 2); i += 2) {
+    const __m256i row0 =
+        _mm256_loadu_si256((__m256i *)&src_ptr[i * src_pitch + 2]);
+    __m256i row1 =
+        _mm256_loadu_si256((__m256i *)&src_ptr[(i + 1) * src_pitch + 2]);
+
+    const __m256i r0 =
+        _mm256_inserti128_si256(row0, _mm256_castsi256_si128(row1), 1);
+    const __m256i r1 = _mm256_permute2x128_si256(row0, row1, 0x31);
+
+    // even pixels
+    s[0] = r0;
+    s[1] = _mm256_alignr_epi8(r1, r0, 4);
+
+    __m256i res_even = convolve_4tap(s, ff);
+    res_even = _mm256_srai_epi32(_mm256_add_epi32(res_even, rounding),
+                                 CONV8_ROUNDING_BITS);
+
+    // odd pixels
+    s[0] = _mm256_alignr_epi8(r1, r0, 2);
+    s[1] = _mm256_alignr_epi8(r1, r0, 6);
+
+    __m256i res_odd = convolve_4tap(s, ff);
+    res_odd = _mm256_srai_epi32(_mm256_add_epi32(res_odd, rounding),
+                                CONV8_ROUNDING_BITS);
+
+    __m256i res = _mm256_packs_epi32(res_even, res_odd);
+    res = _mm256_shuffle_epi8(res, mask);
+
+    res = _mm256_min_epi16(res, clip_pixel);
+    res = _mm256_max_epi16(res, zero);
+
+    _mm_storeu_si128((__m128i *)&dst_ptr[i * dst_pitch],
+                     _mm256_castsi256_si128(res));
+    _mm_storeu_si128((__m128i *)&dst_ptr[i * dst_pitch + dst_pitch],
+                     _mm256_extracti128_si256(res, 1));
+  }
+
+  if (height % 2 != 0) {
+    i = height - 1;
+    const __m256i row0_0 =
+        _mm256_loadu_si256((__m256i *)&src_ptr[i * src_pitch + 2]);
+    const __m256i row0_1 =
+        _mm256_loadu_si256((__m256i *)&src_ptr[i * src_pitch + 6]);
+
+    const __m256i r0 =
+        _mm256_inserti128_si256(row0_0, _mm256_castsi256_si128(row0_1), 1);
+
+    s[0] = _mm256_shuffle_epi8(r0, ip_mask_f2f3);
+    s[1] = _mm256_shuffle_epi8(r0, ip_mask_f4f5);
+
+    __m256i res = convolve_4tap(s, ff);
+    res =
+        _mm256_srai_epi32(_mm256_add_epi32(res, rounding), CONV8_ROUNDING_BITS);
+
+    res = _mm256_packs_epi32(res, res);
+    res = _mm256_min_epi16(res, clip_pixel);
+    res = _mm256_max_epi16(res, zero);
+
+    _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch],
+                     _mm256_castsi256_si128(res));
+    _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch + 4],
+                     _mm256_extracti128_si256(res, 1));
+  }
+}
+
+static void aom_highbd_filter_block1d16_h4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  aom_highbd_filter_block1d8_h4_avx2(src_ptr, src_pitch, dst_ptr, dst_pitch,
+                                     height, filter, bd);
+  aom_highbd_filter_block1d8_h4_avx2(src_ptr + 8, src_pitch, dst_ptr + 8,
+                                     dst_pitch, height, filter, bd);
+}
+
 // -----------------------------------------------------------------------------
 // 2-tap horizontal filtering
 
@@ -875,6 +1064,142 @@ static void aom_highbd_filter_block1d16_v8_avx2(
   } while (height > 0);
 }
 
+static void aom_highbd_filter_block1d4_v4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  const int bits = FILTER_BITS;
+
+  const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
+  const __m256i round_const_bits = _mm256_set1_epi32((1 << bits) >> 1);
+  const __m256i clip_pixel =
+      _mm256_set1_epi32(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+  const __m256i zero = _mm256_setzero_si256();
+  uint32_t i;
+  __m256i s[2], ff[2];
+
+  pack_filters_4tap(filter, ff);
+
+  const uint16_t *data = src_ptr;
+  /* Vertical filter */
+  {
+    __m128i s2 = _mm_loadl_epi64((__m128i *)(data + 2 * src_pitch));
+    __m128i s3 = _mm_loadl_epi64((__m128i *)(data + 3 * src_pitch));
+
+    __m256i s23 = _mm256_inserti128_si256(_mm256_castsi128_si256(s2), s3, 1);
+
+    __m128i s4 = _mm_loadl_epi64((__m128i *)(data + 4 * src_pitch));
+
+    __m256i s34 = _mm256_inserti128_si256(_mm256_castsi128_si256(s3), s4, 1);
+
+    s[0] = _mm256_unpacklo_epi16(s23, s34);
+
+    for (i = 0; i < height; i += 2) {
+      data = &src_ptr[i * src_pitch];
+
+      __m128i s5 = _mm_loadl_epi64((__m128i *)(data + 5 * src_pitch));
+      __m128i s6 = _mm_loadl_epi64((__m128i *)(data + 6 * src_pitch));
+
+      __m256i s45 = _mm256_inserti128_si256(_mm256_castsi128_si256(s4), s5, 1);
+      __m256i s56 = _mm256_inserti128_si256(_mm256_castsi128_si256(s5), s6, 1);
+
+      s[1] = _mm256_unpacklo_epi16(s45, s56);
+
+      const __m256i res_a = convolve_4tap(s, ff);
+
+      __m256i res_a_round = _mm256_sra_epi32(
+          _mm256_add_epi32(res_a, round_const_bits), round_shift_bits);
+
+      __m256i res_16bit = _mm256_min_epi32(res_a_round, clip_pixel);
+      res_16bit = _mm256_max_epi32(res_16bit, zero);
+      res_16bit = _mm256_packs_epi32(res_16bit, res_16bit);
+
+      _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch],
+                       _mm256_castsi256_si128(res_16bit));
+      _mm_storel_epi64((__m128i *)&dst_ptr[i * dst_pitch + dst_pitch],
+                       _mm256_extracti128_si256(res_16bit, 1));
+
+      s[0] = s[1];
+      s4 = s6;
+    }
+  }
+}
+
+static void aom_highbd_filter_block1d8_v4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  const int bits = FILTER_BITS;
+
+  const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
+  const __m256i round_const_bits = _mm256_set1_epi32((1 << bits) >> 1);
+  const __m256i clip_pixel =
+      _mm256_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
+  const __m256i zero = _mm256_setzero_si256();
+  __m256i s[4], ff[2];
+  uint32_t i;
+  pack_filters_4tap(filter, ff);
+
+  const uint16_t *data = src_ptr;
+  /* Vertical filter */
+  {
+    __m128i s2 = _mm_loadu_si128((__m128i *)(data + 2 * src_pitch));
+    __m128i s3 = _mm_loadu_si128((__m128i *)(data + 3 * src_pitch));
+
+    __m256i s23 = _mm256_inserti128_si256(_mm256_castsi128_si256(s2), s3, 1);
+
+    __m128i s4 = _mm_loadu_si128((__m128i *)(data + 4 * src_pitch));
+
+    __m256i s34 = _mm256_inserti128_si256(_mm256_castsi128_si256(s3), s4, 1);
+
+    s[0] = _mm256_unpacklo_epi16(s23, s34);
+    s[2] = _mm256_unpackhi_epi16(s23, s34);
+
+    for (i = 0; i < height; i += 2) {
+      data = &src_ptr[i * src_pitch];
+
+      __m128i s5 = _mm_loadu_si128((__m128i *)(data + 5 * src_pitch));
+      __m128i s6 = _mm_loadu_si128((__m128i *)(data + 6 * src_pitch));
+
+      __m256i s45 = _mm256_inserti128_si256(_mm256_castsi128_si256(s4), s5, 1);
+      __m256i s56 = _mm256_inserti128_si256(_mm256_castsi128_si256(s5), s6, 1);
+
+      s[1] = _mm256_unpacklo_epi16(s45, s56);
+      s[3] = _mm256_unpackhi_epi16(s45, s56);
+
+      const __m256i res_a = convolve_4tap(s, ff);
+
+      __m256i res_a_round = _mm256_sra_epi32(
+          _mm256_add_epi32(res_a, round_const_bits), round_shift_bits);
+
+      const __m256i res_b = convolve_4tap(s + 2, ff);
+      __m256i res_b_round = _mm256_sra_epi32(
+          _mm256_add_epi32(res_b, round_const_bits), round_shift_bits);
+
+      __m256i res_16bit = _mm256_packs_epi32(res_a_round, res_b_round);
+      res_16bit = _mm256_min_epi16(res_16bit, clip_pixel);
+      res_16bit = _mm256_max_epi16(res_16bit, zero);
+
+      _mm_storeu_si128((__m128i *)&dst_ptr[i * dst_pitch],
+                       _mm256_castsi256_si128(res_16bit));
+      _mm_storeu_si128((__m128i *)&dst_ptr[i * dst_pitch + dst_pitch],
+                       _mm256_extracti128_si256(res_16bit, 1));
+
+      s[0] = s[1];
+      s[2] = s[3];
+      s4 = s6;
+    }
+  }
+}
+
+static void aom_highbd_filter_block1d16_v4_avx2(
+    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
+    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
+  aom_highbd_filter_block1d8_v4_avx2(src_ptr, src_pitch, dst_ptr, dst_pitch,
+                                     height, filter, bd);
+
+  aom_highbd_filter_block1d8_v4_avx2(src_ptr + 8, src_pitch, dst_ptr + 8,
+                                     dst_pitch, height, filter, bd);
+}
+
 // -----------------------------------------------------------------------------
 // 2-tap vertical filtering
 
diff --git a/third_party/aom/aom_dsp/x86/highbd_convolve_sse2.c b/third_party/aom/aom_dsp/x86/highbd_convolve_sse2.c
new file mode 100644
index 0000000000..a2bb283222
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/highbd_convolve_sse2.c
@@ -0,0 +1,351 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include <emmintrin.h>
+
+#include "config/aom_dsp_rtcd.h"
+#include "aom_dsp/x86/convolve.h"
+
+// -----------------------------------------------------------------------------
+
+void aom_highbd_filter_block1d4_v4_sse2(const uint16_t *src_ptr,
+                                        ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                        ptrdiff_t dst_pitch, uint32_t height,
+                                        const int16_t *filter, int bd) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23_lo, srcReg34_lo;
+  __m128i srcReg45_lo, srcReg56_lo;
+  __m128i resReg23_lo, resReg34_lo, resReg45_lo, resReg56_lo;
+  __m128i resReg23_45_lo, resReg34_56_lo;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg64, secondFilters, thirdFilters;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
+  addFilterReg64 = _mm_set1_epi32(64);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp0, tmp0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp1, tmp1);   // coeffs 4 5 4 5 4 5 4 5
+
+  // multiply the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = dst_pitch << 1;
+
+  srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23_lo = _mm_unpacklo_epi16(srcReg2, srcReg3);
+
+  srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
+  srcReg34_lo = _mm_unpacklo_epi16(srcReg3, srcReg4);
+
+  for (i = height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
+    srcReg45_lo = _mm_unpacklo_epi16(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+    srcReg56_lo = _mm_unpacklo_epi16(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    resReg23_lo = _mm_madd_epi16(srcReg23_lo, secondFilters);
+    resReg34_lo = _mm_madd_epi16(srcReg34_lo, secondFilters);
+    resReg45_lo = _mm_madd_epi16(srcReg45_lo, thirdFilters);
+    resReg56_lo = _mm_madd_epi16(srcReg56_lo, thirdFilters);
+
+    resReg23_45_lo = _mm_add_epi32(resReg23_lo, resReg45_lo);
+    resReg34_56_lo = _mm_add_epi32(resReg34_lo, resReg56_lo);
+
+    // shift by 7 bit each 32 bit
+    resReg23_45_lo = _mm_add_epi32(resReg23_45_lo, addFilterReg64);
+    resReg34_56_lo = _mm_add_epi32(resReg34_56_lo, addFilterReg64);
+    resReg23_45_lo = _mm_srai_epi32(resReg23_45_lo, 7);
+    resReg34_56_lo = _mm_srai_epi32(resReg34_56_lo, 7);
+
+    // shrink to 16 bit each 32 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packs_epi32(resReg23_45_lo, _mm_setzero_si128());
+    resReg34_56 = _mm_packs_epi32(resReg34_56_lo, _mm_setzero_si128());
+
+    resReg23_45 = _mm_max_epi16(resReg23_45, _mm_setzero_si128());
+    resReg23_45 = _mm_min_epi16(resReg23_45, max);
+    resReg34_56 = _mm_max_epi16(resReg34_56, _mm_setzero_si128());
+    resReg34_56 = _mm_min_epi16(resReg34_56, max);
+
+    src_ptr += src_stride;
+
+    _mm_storel_epi64((__m128i *)dst_ptr, (resReg23_45));
+    _mm_storel_epi64((__m128i *)(dst_ptr + dst_pitch), (resReg34_56));
+
+    dst_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    srcReg23_lo = srcReg45_lo;
+    srcReg34_lo = srcReg56_lo;
+    srcReg4 = srcReg6;
+  }
+}
+
+void aom_highbd_filter_block1d4_h4_sse2(const uint16_t *src_ptr,
+                                        ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                        ptrdiff_t dst_pitch, uint32_t height,
+                                        const int16_t *filter, int bd) {
+  __m128i filtersReg;
+  __m128i addFilterReg64;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1;
+  __m128i srcReg32b1;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg64 = _mm_set1_epi32(64);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp_0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp_1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp_0, tmp_0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp_1, tmp_1);   // coeffs 4 5 4 5 4 5 4 5
+
+  for (i = height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)(src_ptr + 2));
+
+    __m128i ss_3_1 = _mm_srli_si128(srcReg32b1, 2);
+    __m128i ss_4_1 = _mm_srli_si128(srcReg32b1, 4);
+    __m128i ss_5_1 = _mm_srli_si128(srcReg32b1, 6);
+    __m128i ss_23 = _mm_unpacklo_epi32(srcReg32b1, ss_3_1);
+    __m128i ss_45 = _mm_unpacklo_epi32(ss_4_1, ss_5_1);
+
+    ss_23 = _mm_madd_epi16(ss_23, secondFilters);
+    ss_45 = _mm_madd_epi16(ss_45, thirdFilters);
+    srcRegFilt32b1_1 = _mm_add_epi32(ss_23, ss_45);
+
+    // shift by 7 bit each 32 bit
+    srcRegFilt32b1_1 = _mm_add_epi32(srcRegFilt32b1_1, addFilterReg64);
+    srcRegFilt32b1_1 = _mm_srai_epi32(srcRegFilt32b1_1, 7);
+
+    srcRegFilt32b1_1 = _mm_packs_epi32(srcRegFilt32b1_1, _mm_setzero_si128());
+    srcRegFilt32b1_1 = _mm_max_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+    srcRegFilt32b1_1 = _mm_min_epi16(srcRegFilt32b1_1, max);
+
+    src_ptr += src_pitch;
+
+    _mm_storel_epi64((__m128i *)dst_ptr, srcRegFilt32b1_1);
+
+    dst_ptr += dst_pitch;
+  }
+}
+
+void aom_highbd_filter_block1d8_v4_sse2(const uint16_t *src_ptr,
+                                        ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                        ptrdiff_t dst_pitch, uint32_t height,
+                                        const int16_t *filter, int bd) {
+  __m128i filtersReg;
+  __m128i srcReg2, srcReg3, srcReg4, srcReg5, srcReg6;
+  __m128i srcReg23_lo, srcReg23_hi, srcReg34_lo, srcReg34_hi;
+  __m128i srcReg45_lo, srcReg45_hi, srcReg56_lo, srcReg56_hi;
+  __m128i resReg23_lo, resReg34_lo, resReg45_lo, resReg56_lo;
+  __m128i resReg23_hi, resReg34_hi, resReg45_hi, resReg56_hi;
+  __m128i resReg23_45_lo, resReg34_56_lo, resReg23_45_hi, resReg34_56_hi;
+  __m128i resReg23_45, resReg34_56;
+  __m128i addFilterReg64, secondFilters, thirdFilters;
+  unsigned int i;
+  ptrdiff_t src_stride, dst_stride;
+
+  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
+  addFilterReg64 = _mm_set1_epi32(64);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp0, tmp0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp1, tmp1);   // coeffs 4 5 4 5 4 5 4 5
+
+  // multiple the size of the source and destination stride by two
+  src_stride = src_pitch << 1;
+  dst_stride = dst_pitch << 1;
+
+  srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+  srcReg23_lo = _mm_unpacklo_epi16(srcReg2, srcReg3);
+  srcReg23_hi = _mm_unpackhi_epi16(srcReg2, srcReg3);
+
+  srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+  srcReg34_lo = _mm_unpacklo_epi16(srcReg3, srcReg4);
+  srcReg34_hi = _mm_unpackhi_epi16(srcReg3, srcReg4);
+
+  for (i = height; i > 1; i -= 2) {
+    srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+
+    srcReg45_lo = _mm_unpacklo_epi16(srcReg4, srcReg5);
+    srcReg45_hi = _mm_unpackhi_epi16(srcReg4, srcReg5);
+
+    srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+
+    srcReg56_lo = _mm_unpacklo_epi16(srcReg5, srcReg6);
+    srcReg56_hi = _mm_unpackhi_epi16(srcReg5, srcReg6);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    resReg23_lo = _mm_madd_epi16(srcReg23_lo, secondFilters);
+    resReg34_lo = _mm_madd_epi16(srcReg34_lo, secondFilters);
+    resReg45_lo = _mm_madd_epi16(srcReg45_lo, thirdFilters);
+    resReg56_lo = _mm_madd_epi16(srcReg56_lo, thirdFilters);
+
+    resReg23_45_lo = _mm_add_epi32(resReg23_lo, resReg45_lo);
+    resReg34_56_lo = _mm_add_epi32(resReg34_lo, resReg56_lo);
+
+    // multiply 2 adjacent elements with the filter and add the result
+
+    resReg23_hi = _mm_madd_epi16(srcReg23_hi, secondFilters);
+    resReg34_hi = _mm_madd_epi16(srcReg34_hi, secondFilters);
+    resReg45_hi = _mm_madd_epi16(srcReg45_hi, thirdFilters);
+    resReg56_hi = _mm_madd_epi16(srcReg56_hi, thirdFilters);
+
+    resReg23_45_hi = _mm_add_epi32(resReg23_hi, resReg45_hi);
+    resReg34_56_hi = _mm_add_epi32(resReg34_hi, resReg56_hi);
+
+    // shift by 7 bit each 32 bit
+    resReg23_45_lo = _mm_add_epi32(resReg23_45_lo, addFilterReg64);
+    resReg34_56_lo = _mm_add_epi32(resReg34_56_lo, addFilterReg64);
+    resReg23_45_hi = _mm_add_epi32(resReg23_45_hi, addFilterReg64);
+    resReg34_56_hi = _mm_add_epi32(resReg34_56_hi, addFilterReg64);
+    resReg23_45_lo = _mm_srai_epi32(resReg23_45_lo, 7);
+    resReg34_56_lo = _mm_srai_epi32(resReg34_56_lo, 7);
+    resReg23_45_hi = _mm_srai_epi32(resReg23_45_hi, 7);
+    resReg34_56_hi = _mm_srai_epi32(resReg34_56_hi, 7);
+
+    // shrink to 16 bit each 32 bits, the first lane contain the first
+    // convolve result and the second lane contain the second convolve
+    // result
+    resReg23_45 = _mm_packs_epi32(resReg23_45_lo, resReg23_45_hi);
+    resReg34_56 = _mm_packs_epi32(resReg34_56_lo, resReg34_56_hi);
+
+    resReg23_45 = _mm_max_epi16(resReg23_45, _mm_setzero_si128());
+    resReg23_45 = _mm_min_epi16(resReg23_45, max);
+    resReg34_56 = _mm_max_epi16(resReg34_56, _mm_setzero_si128());
+    resReg34_56 = _mm_min_epi16(resReg34_56, max);
+
+    src_ptr += src_stride;
+
+    _mm_store_si128((__m128i *)dst_ptr, (resReg23_45));
+    _mm_store_si128((__m128i *)(dst_ptr + dst_pitch), (resReg34_56));
+
+    dst_ptr += dst_stride;
+
+    // save part of the registers for next strides
+    srcReg23_lo = srcReg45_lo;
+    srcReg23_hi = srcReg45_hi;
+    srcReg34_lo = srcReg56_lo;
+    srcReg34_hi = srcReg56_hi;
+    srcReg4 = srcReg6;
+  }
+}
+
+void aom_highbd_filter_block1d8_h4_sse2(const uint16_t *src_ptr,
+                                        ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                        ptrdiff_t dst_pitch, uint32_t height,
+                                        const int16_t *filter, int bd) {
+  __m128i filtersReg;
+  __m128i addFilterReg64;
+  __m128i secondFilters, thirdFilters;
+  __m128i srcRegFilt32b1_1, srcRegFilt32b1_2;
+  __m128i srcReg32b1, srcReg32b2;
+  unsigned int i;
+  src_ptr -= 3;
+  addFilterReg64 = _mm_set1_epi32(64);
+  filtersReg = _mm_loadu_si128((const __m128i *)filter);
+  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
+
+  // coeffs 0 1 0 1 2 3 2 3
+  const __m128i tmp_0 = _mm_unpacklo_epi32(filtersReg, filtersReg);
+  // coeffs 4 5 4 5 6 7 6 7
+  const __m128i tmp_1 = _mm_unpackhi_epi32(filtersReg, filtersReg);
+
+  secondFilters = _mm_unpackhi_epi64(tmp_0, tmp_0);  // coeffs 2 3 2 3 2 3 2 3
+  thirdFilters = _mm_unpacklo_epi64(tmp_1, tmp_1);   // coeffs 4 5 4 5 4 5 4 5
+
+  for (i = height; i > 0; i -= 1) {
+    srcReg32b1 = _mm_loadu_si128((const __m128i *)(src_ptr + 2));
+    srcReg32b2 = _mm_loadu_si128((const __m128i *)(src_ptr + 6));
+
+    __m128i ss_4_1 = _mm_srli_si128(srcReg32b1, 4);
+    __m128i ss_4_2 = _mm_srli_si128(srcReg32b2, 4);
+    __m128i ss_4 = _mm_unpacklo_epi64(ss_4_1, ss_4_2);
+
+    __m128i d1 = _mm_madd_epi16(srcReg32b1, secondFilters);
+    __m128i d2 = _mm_madd_epi16(ss_4, thirdFilters);
+    srcRegFilt32b1_1 = _mm_add_epi32(d1, d2);
+
+    __m128i ss_3_1 = _mm_srli_si128(srcReg32b1, 2);
+    __m128i ss_5_1 = _mm_srli_si128(srcReg32b1, 6);
+    __m128i ss_3_2 = _mm_srli_si128(srcReg32b2, 2);
+    __m128i ss_5_2 = _mm_srli_si128(srcReg32b2, 6);
+    __m128i ss_3 = _mm_unpacklo_epi64(ss_3_1, ss_3_2);
+    __m128i ss_5 = _mm_unpacklo_epi64(ss_5_1, ss_5_2);
+
+    d1 = _mm_madd_epi16(ss_3, secondFilters);
+    d2 = _mm_madd_epi16(ss_5, thirdFilters);
+    srcRegFilt32b1_2 = _mm_add_epi32(d1, d2);
+
+    __m128i res_lo_1 = _mm_unpacklo_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+    __m128i res_hi_1 = _mm_unpackhi_epi32(srcRegFilt32b1_1, srcRegFilt32b1_2);
+
+    // shift by 7 bit each 32 bit
+    res_lo_1 = _mm_add_epi32(res_lo_1, addFilterReg64);
+    res_hi_1 = _mm_add_epi32(res_hi_1, addFilterReg64);
+    res_lo_1 = _mm_srai_epi32(res_lo_1, 7);
+    res_hi_1 = _mm_srai_epi32(res_hi_1, 7);
+
+    srcRegFilt32b1_1 = _mm_packs_epi32(res_lo_1, res_hi_1);
+
+    srcRegFilt32b1_1 = _mm_max_epi16(srcRegFilt32b1_1, _mm_setzero_si128());
+    srcRegFilt32b1_1 = _mm_min_epi16(srcRegFilt32b1_1, max);
+
+    src_ptr += src_pitch;
+
+    _mm_store_si128((__m128i *)dst_ptr, srcRegFilt32b1_1);
+
+    dst_ptr += dst_pitch;
+  }
+}
+
+void aom_highbd_filter_block1d16_v4_sse2(const uint16_t *src_ptr,
+                                         ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                         ptrdiff_t dst_pitch, uint32_t height,
+                                         const int16_t *filter, int bd) {
+  aom_highbd_filter_block1d8_v4_sse2(src_ptr, src_pitch, dst_ptr, dst_pitch,
+                                     height, filter, bd);
+  aom_highbd_filter_block1d8_v4_sse2((src_ptr + 8), src_pitch, (dst_ptr + 8),
+                                     dst_pitch, height, filter, bd);
+}
+
+void aom_highbd_filter_block1d16_h4_sse2(const uint16_t *src_ptr,
+                                         ptrdiff_t src_pitch, uint16_t *dst_ptr,
+                                         ptrdiff_t dst_pitch, uint32_t height,
+                                         const int16_t *filter, int bd) {
+  aom_highbd_filter_block1d8_h4_sse2(src_ptr, src_pitch, dst_ptr, dst_pitch,
+                                     height, filter, bd);
+  aom_highbd_filter_block1d8_h4_sse2((src_ptr + 8), src_pitch, (dst_ptr + 8),
+                                     dst_pitch, height, filter, bd);
+}
diff --git a/third_party/aom/aom_dsp/x86/highbd_intrapred_asm_sse2.asm b/third_party/aom/aom_dsp/x86/highbd_intrapred_asm_sse2.asm
new file mode 100644
index 0000000000..91b3d126ca
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/highbd_intrapred_asm_sse2.asm
@@ -0,0 +1,259 @@
+;
+; Copyright (c) 2016, Alliance for Open Media. All rights reserved
+;
+; This source code is subject to the terms of the BSD 2 Clause License and
+; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+; was not distributed with this source code in the LICENSE file, you can
+; obtain it at www.aomedia.org/license/software. If the Alliance for Open
+; Media Patent License 1.0 was not distributed with this source code in the
+; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+;
+
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pw_4:  times 8 dw 4
+pw_8:  times 8 dw 8
+pw_16: times 4 dd 16
+pw_32: times 4 dd 32
+
+SECTION .text
+INIT_XMM sse2
+cglobal highbd_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  movq                  m0, [aboveq]
+  movq                  m2, [leftq]
+  paddw                 m0, m2
+  pshuflw               m1, m0, 0xe
+  paddw                 m0, m1
+  pshuflw               m1, m0, 0x1
+  paddw                 m0, m1
+  paddw                 m0, [GLOBAL(pw_4)]
+  psraw                 m0, 3
+  pshuflw               m0, m0, 0x0
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal highbd_dc_predictor_8x8, 4, 5, 4, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [leftq]
+  DEFINE_ARGS dst, stride, stride3, one
+  mov                 oned, 0x00010001
+  lea             stride3q, [strideq*3]
+  movd                  m3, oned
+  pshufd                m3, m3, 0x0
+  paddw                 m0, m2
+  pmaddwd               m0, m3
+  packssdw              m0, m1
+  pmaddwd               m0, m3
+  packssdw              m0, m1
+  pmaddwd               m0, m3
+  paddw                 m0, [GLOBAL(pw_8)]
+  psrlw                 m0, 4
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  lea                 dstq, [dstq+strideq*8]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal highbd_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m3, [aboveq+16]
+  mova                  m2, [leftq]
+  mova                  m4, [leftq+16]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  paddw                 m0, m2
+  paddw                 m0, m3
+  paddw                 m0, m4
+  movhlps               m2, m0
+  paddw                 m0, m2
+  punpcklwd             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  punpckldq             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  paddd                 m0, [GLOBAL(pw_16)]
+  psrad                 m0, 5
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+.loop:
+  mova   [dstq              ], m0
+  mova   [dstq           +16], m0
+  mova   [dstq+strideq*2    ], m0
+  mova   [dstq+strideq*2 +16], m0
+  mova   [dstq+strideq*4    ], m0
+  mova   [dstq+strideq*4 +16], m0
+  mova   [dstq+stride3q*2   ], m0
+  mova   [dstq+stride3q*2+16], m0
+  lea                 dstq, [dstq+strideq*8]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal highbd_dc_predictor_32x32, 4, 5, 7, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  mova                  m0, [aboveq]
+  mova                  m2, [aboveq+16]
+  mova                  m3, [aboveq+32]
+  mova                  m4, [aboveq+48]
+  paddw                 m0, m2
+  paddw                 m3, m4
+  mova                  m2, [leftq]
+  mova                  m4, [leftq+16]
+  mova                  m5, [leftq+32]
+  mova                  m6, [leftq+48]
+  paddw                 m2, m4
+  paddw                 m5, m6
+  paddw                 m0, m3
+  paddw                 m2, m5
+  pxor                  m1, m1
+  paddw                 m0, m2
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  movhlps               m2, m0
+  paddw                 m0, m2
+  punpcklwd             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  punpckldq             m0, m1
+  movhlps               m2, m0
+  paddd                 m0, m2
+  paddd                 m0, [GLOBAL(pw_32)]
+  psrad                 m0, 6
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+.loop:
+  mova [dstq               ], m0
+  mova [dstq          +16  ], m0
+  mova [dstq          +32  ], m0
+  mova [dstq          +48  ], m0
+  mova [dstq+strideq*2     ], m0
+  mova [dstq+strideq*2+16  ], m0
+  mova [dstq+strideq*2+32  ], m0
+  mova [dstq+strideq*2+48  ], m0
+  mova [dstq+strideq*4     ], m0
+  mova [dstq+strideq*4+16  ], m0
+  mova [dstq+strideq*4+32  ], m0
+  mova [dstq+strideq*4+48  ], m0
+  mova [dstq+stride3q*2    ], m0
+  mova [dstq+stride3q*2 +16], m0
+  mova [dstq+stride3q*2 +32], m0
+  mova [dstq+stride3q*2 +48], m0
+  lea                 dstq, [dstq+strideq*8]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above
+  movq                  m0, [aboveq]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq*2], m0
+  RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_8x8, 3, 3, 1, dst, stride, above
+  mova                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  lea                 dstq, [dstq+strideq*8]
+  mova   [dstq           ], m0
+  mova   [dstq+strideq*2 ], m0
+  mova   [dstq+strideq*4 ], m0
+  mova   [dstq+stride3q*2], m0
+  RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_16x16, 3, 4, 2, dst, stride, above
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 4
+.loop:
+  mova    [dstq              ], m0
+  mova    [dstq           +16], m1
+  mova    [dstq+strideq*2    ], m0
+  mova    [dstq+strideq*2 +16], m1
+  mova    [dstq+strideq*4    ], m0
+  mova    [dstq+strideq*4 +16], m1
+  mova    [dstq+stride3q*2   ], m0
+  mova    [dstq+stride3q*2+16], m1
+  lea                 dstq, [dstq+strideq*8]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_32x32, 3, 4, 4, dst, stride, above
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  mova                  m2, [aboveq+32]
+  mova                  m3, [aboveq+48]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 8
+.loop:
+  mova [dstq               ], m0
+  mova [dstq            +16], m1
+  mova [dstq            +32], m2
+  mova [dstq            +48], m3
+  mova [dstq+strideq*2     ], m0
+  mova [dstq+strideq*2  +16], m1
+  mova [dstq+strideq*2  +32], m2
+  mova [dstq+strideq*2  +48], m3
+  mova [dstq+strideq*4     ], m0
+  mova [dstq+strideq*4  +16], m1
+  mova [dstq+strideq*4  +32], m2
+  mova [dstq+strideq*4  +48], m3
+  mova [dstq+stride3q*2    ], m0
+  mova [dstq+stride3q*2 +16], m1
+  mova [dstq+stride3q*2 +32], m2
+  mova [dstq+stride3q*2 +48], m3
+  lea                 dstq, [dstq+strideq*8]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
diff --git a/third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c b/third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c
index 097e0778ff..70b91c651d 100644
--- a/third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_loopfilter_sse2.c
@@ -727,8 +727,8 @@ void aom_highbd_lpf_horizontal_14_dual_sse2(
                                    _limit1, _thresh1, bd);
 
   for (i = 0; i < 6; i++) {
-    _mm_store_si128((__m128i *)(s - (i + 1) * pitch), p[i]);
-    _mm_store_si128((__m128i *)(s + i * pitch), q[i]);
+    _mm_storeu_si128((__m128i *)(s - (i + 1) * pitch), p[i]);
+    _mm_storeu_si128((__m128i *)(s + i * pitch), q[i]);
   }
 }
 
diff --git a/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm b/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
index 61f5b8e865..5c78933df5 100644
--- a/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/highbd_subpel_variance_impl_sse2.asm
@@ -114,45 +114,33 @@ SECTION .text
       cglobal highbd_sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
                                         x_offset, y_offset, \
                                         dst, dst_stride, \
-                                        sec, sec_stride, height, sse, \
-                                        g_bilin_filter, g_pw_8
+                                        sec, sec_stride, height, sse
       %define block_height dword heightm
       %define sec_str sec_stridemp
-
-      ; Store bilin_filter and pw_8 location in stack
-      %if GET_GOT_DEFINED == 1
-        GET_GOT eax
-        add esp, 4                ; restore esp
-      %endif
-
-      lea ecx, [GLOBAL(bilin_filter_m)]
-      mov g_bilin_filterm, ecx
-
-      lea ecx, [GLOBAL(pw_8)]
-      mov g_pw_8m, ecx
-
-      LOAD_IF_USED 0, 1         ; load eax, ecx back
     %else
       cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
                                     x_offset, y_offset, \
-                                    dst, dst_stride, height, sse, \
-                                    g_bilin_filter, g_pw_8
+                                    dst, dst_stride, height, sse
       %define block_height heightd
-
-      ; Store bilin_filter and pw_8 location in stack
-      %if GET_GOT_DEFINED == 1
-        GET_GOT eax
-        add esp, 4                ; restore esp
-      %endif
-
-      lea ecx, [GLOBAL(bilin_filter_m)]
-      mov g_bilin_filterm, ecx
-
-      lea ecx, [GLOBAL(pw_8)]
-      mov g_pw_8m, ecx
-
-      LOAD_IF_USED 0, 1         ; load eax, ecx back
     %endif
+
+    ; reuse argument stack space
+    %define g_bilin_filterm x_offsetm
+    %define g_pw_8m y_offsetm
+
+    ; Store bilin_filter and pw_8 location in stack
+    %if GET_GOT_DEFINED == 1
+      GET_GOT eax
+      add esp, 4                ; restore esp
+    %endif
+
+    lea ecx, [GLOBAL(bilin_filter_m)]
+    mov g_bilin_filterm, ecx
+
+    lea ecx, [GLOBAL(pw_8)]
+    mov g_pw_8m, ecx
+
+    LOAD_IF_USED 0, 1         ; load eax, ecx back
   %else
     %if %2 == 1 ; avg
       cglobal highbd_sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
diff --git a/third_party/aom/aom_dsp/x86/highbd_subtract_sse2.c b/third_party/aom/aom_dsp/x86/highbd_subtract_sse2.c
index 18eb03d12c..b72d1cf8ba 100644
--- a/third_party/aom/aom_dsp/x86/highbd_subtract_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_subtract_sse2.c
@@ -29,15 +29,15 @@ static void subtract_4x4(int16_t *diff, ptrdiff_t diff_stride,
   __m128i x0, x1, x2, x3;
   int64_t *store_diff = (int64_t *)(diff + 0 * diff_stride);
 
-  u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
-  u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
-  u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
-  u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
+  u0 = _mm_loadl_epi64((__m128i const *)(src + 0 * src_stride));
+  u1 = _mm_loadl_epi64((__m128i const *)(src + 1 * src_stride));
+  u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
 
-  v0 = _mm_loadu_si128((__m128i const *)(pred + 0 * pred_stride));
-  v1 = _mm_loadu_si128((__m128i const *)(pred + 1 * pred_stride));
-  v2 = _mm_loadu_si128((__m128i const *)(pred + 2 * pred_stride));
-  v3 = _mm_loadu_si128((__m128i const *)(pred + 3 * pred_stride));
+  v0 = _mm_loadl_epi64((__m128i const *)(pred + 0 * pred_stride));
+  v1 = _mm_loadl_epi64((__m128i const *)(pred + 1 * pred_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(pred + 2 * pred_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(pred + 3 * pred_stride));
 
   x0 = _mm_sub_epi16(u0, v0);
   x1 = _mm_sub_epi16(u1, v1);
@@ -61,23 +61,23 @@ static void subtract_4x8(int16_t *diff, ptrdiff_t diff_stride,
   __m128i x0, x1, x2, x3, x4, x5, x6, x7;
   int64_t *store_diff = (int64_t *)(diff + 0 * diff_stride);
 
-  u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
-  u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
-  u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
-  u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
-  u4 = _mm_loadu_si128((__m128i const *)(src + 4 * src_stride));
-  u5 = _mm_loadu_si128((__m128i const *)(src + 5 * src_stride));
-  u6 = _mm_loadu_si128((__m128i const *)(src + 6 * src_stride));
-  u7 = _mm_loadu_si128((__m128i const *)(src + 7 * src_stride));
+  u0 = _mm_loadl_epi64((__m128i const *)(src + 0 * src_stride));
+  u1 = _mm_loadl_epi64((__m128i const *)(src + 1 * src_stride));
+  u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
+  u4 = _mm_loadl_epi64((__m128i const *)(src + 4 * src_stride));
+  u5 = _mm_loadl_epi64((__m128i const *)(src + 5 * src_stride));
+  u6 = _mm_loadl_epi64((__m128i const *)(src + 6 * src_stride));
+  u7 = _mm_loadl_epi64((__m128i const *)(src + 7 * src_stride));
 
-  v0 = _mm_loadu_si128((__m128i const *)(pred + 0 * pred_stride));
-  v1 = _mm_loadu_si128((__m128i const *)(pred + 1 * pred_stride));
-  v2 = _mm_loadu_si128((__m128i const *)(pred + 2 * pred_stride));
-  v3 = _mm_loadu_si128((__m128i const *)(pred + 3 * pred_stride));
-  v4 = _mm_loadu_si128((__m128i const *)(pred + 4 * pred_stride));
-  v5 = _mm_loadu_si128((__m128i const *)(pred + 5 * pred_stride));
-  v6 = _mm_loadu_si128((__m128i const *)(pred + 6 * pred_stride));
-  v7 = _mm_loadu_si128((__m128i const *)(pred + 7 * pred_stride));
+  v0 = _mm_loadl_epi64((__m128i const *)(pred + 0 * pred_stride));
+  v1 = _mm_loadl_epi64((__m128i const *)(pred + 1 * pred_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(pred + 2 * pred_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(pred + 3 * pred_stride));
+  v4 = _mm_loadl_epi64((__m128i const *)(pred + 4 * pred_stride));
+  v5 = _mm_loadl_epi64((__m128i const *)(pred + 5 * pred_stride));
+  v6 = _mm_loadl_epi64((__m128i const *)(pred + 6 * pred_stride));
+  v7 = _mm_loadl_epi64((__m128i const *)(pred + 7 * pred_stride));
 
   x0 = _mm_sub_epi16(u0, v0);
   x1 = _mm_sub_epi16(u1, v1);
diff --git a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
index 47b052abc9..2373d3aecb 100644
--- a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
@@ -603,7 +603,7 @@ void aom_highbd_upsampled_pred_sse2(MACROBLOCKD *xd,
     const int ref_num = 0;
     const int is_intrabc = is_intrabc_block(mi);
     const struct scale_factors *const sf =
-        is_intrabc ? &cm->sf_identity : &xd->block_refs[ref_num]->sf;
+        is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref_num];
     const int is_scaled = av1_is_scaled(sf);
 
     if (is_scaled) {
@@ -677,11 +677,8 @@ void aom_highbd_upsampled_pred_sse2(MACROBLOCKD *xd,
     }
   }
 
-  const InterpFilterParams *filter =
-      (subpel_search == 1)
-          ? av1_get_4tap_interp_filter_params(EIGHTTAP_REGULAR)
-          : av1_get_interp_filter_params_with_block_size(EIGHTTAP_REGULAR, 8);
-
+  const InterpFilterParams *filter = av1_get_filter(subpel_search);
+  int filter_taps = (subpel_search <= USE_4_TAPS) ? 4 : SUBPEL_TAPS;
   if (!subpel_x_q3 && !subpel_y_q3) {
     uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
     uint16_t *comp_pred = CONVERT_TO_SHORTPTR(comp_pred8);
@@ -729,17 +726,20 @@ void aom_highbd_upsampled_pred_sse2(MACROBLOCKD *xd,
         av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
     const int16_t *const kernel_y =
         av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+    const uint8_t *ref_start = ref8 - ref_stride * ((filter_taps >> 1) - 1);
+    uint16_t *temp_start_horiz = (subpel_search <= USE_4_TAPS)
+                                     ? temp + (filter_taps >> 1) * MAX_SB_SIZE
+                                     : temp;
+    uint16_t *temp_start_vert = temp + MAX_SB_SIZE * ((filter->taps >> 1) - 1);
     const int intermediate_height =
-        (((height - 1) * 8 + subpel_y_q3) >> 3) + filter->taps;
+        (((height - 1) * 8 + subpel_y_q3) >> 3) + filter_taps;
     assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
-    aom_highbd_convolve8_horiz(ref8 - ref_stride * ((filter->taps >> 1) - 1),
-                               ref_stride, CONVERT_TO_BYTEPTR(temp),
-                               MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
-                               intermediate_height, bd);
-    aom_highbd_convolve8_vert(
-        CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter->taps >> 1) - 1)),
-        MAX_SB_SIZE, comp_pred8, width, NULL, -1, kernel_y, 16, width, height,
-        bd);
+    aom_highbd_convolve8_horiz(
+        ref_start, ref_stride, CONVERT_TO_BYTEPTR(temp_start_horiz),
+        MAX_SB_SIZE, kernel_x, 16, NULL, -1, width, intermediate_height, bd);
+    aom_highbd_convolve8_vert(CONVERT_TO_BYTEPTR(temp_start_vert), MAX_SB_SIZE,
+                              comp_pred8, width, NULL, -1, kernel_y, 16, width,
+                              height, bd);
   }
 }
 
@@ -765,11 +765,11 @@ void aom_highbd_comp_avg_upsampled_pred_sse2(
   }
 }
 
-static INLINE void highbd_compute_jnt_comp_avg(__m128i *p0, __m128i *p1,
-                                               const __m128i *w0,
-                                               const __m128i *w1,
-                                               const __m128i *r,
-                                               void *const result) {
+static INLINE void highbd_compute_dist_wtd_comp_avg(__m128i *p0, __m128i *p1,
+                                                    const __m128i *w0,
+                                                    const __m128i *w1,
+                                                    const __m128i *r,
+                                                    void *const result) {
   assert(DIST_PRECISION_BITS <= 4);
   __m128i mult0 = _mm_mullo_epi16(*p0, *w0);
   __m128i mult1 = _mm_mullo_epi16(*p1, *w1);
@@ -780,11 +780,10 @@ static INLINE void highbd_compute_jnt_comp_avg(__m128i *p0, __m128i *p1,
   xx_storeu_128(result, shift);
 }
 
-void aom_highbd_jnt_comp_avg_pred_sse2(uint8_t *comp_pred8,
-                                       const uint8_t *pred8, int width,
-                                       int height, const uint8_t *ref8,
-                                       int ref_stride,
-                                       const JNT_COMP_PARAMS *jcp_param) {
+void aom_highbd_dist_wtd_comp_avg_pred_sse2(
+    uint8_t *comp_pred8, const uint8_t *pred8, int width, int height,
+    const uint8_t *ref8, int ref_stride,
+    const DIST_WTD_COMP_PARAMS *jcp_param) {
   int i;
   const uint16_t wt0 = (uint16_t)jcp_param->fwd_offset;
   const uint16_t wt1 = (uint16_t)jcp_param->bck_offset;
@@ -806,7 +805,7 @@ void aom_highbd_jnt_comp_avg_pred_sse2(uint8_t *comp_pred8,
         __m128i p0 = xx_loadu_128(ref);
         __m128i p1 = xx_loadu_128(pred);
 
-        highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
+        highbd_compute_dist_wtd_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
 
         comp_pred += 8;
         pred += 8;
@@ -823,7 +822,7 @@ void aom_highbd_jnt_comp_avg_pred_sse2(uint8_t *comp_pred8,
       __m128i p0 = _mm_unpacklo_epi64(p0_0, p0_1);
       __m128i p1 = xx_loadu_128(pred);
 
-      highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
+      highbd_compute_dist_wtd_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
 
       comp_pred += 8;
       pred += 8;
@@ -832,11 +831,11 @@ void aom_highbd_jnt_comp_avg_pred_sse2(uint8_t *comp_pred8,
   }
 }
 
-void aom_highbd_jnt_comp_avg_upsampled_pred_sse2(
+void aom_highbd_dist_wtd_comp_avg_upsampled_pred_sse2(
     MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
-    int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param,
+    int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param,
     int subpel_search) {
   uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
   int n;
@@ -860,7 +859,7 @@ void aom_highbd_jnt_comp_avg_upsampled_pred_sse2(
     __m128i p0 = xx_loadu_128(comp_pred16);
     __m128i p1 = xx_loadu_128(pred);
 
-    highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred16);
+    highbd_compute_dist_wtd_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred16);
 
     comp_pred16 += 8;
     pred += 8;
diff --git a/third_party/aom/aom_dsp/x86/intrapred_asm_sse2.asm b/third_party/aom/aom_dsp/x86/intrapred_asm_sse2.asm
new file mode 100644
index 0000000000..0eb632326b
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/intrapred_asm_sse2.asm
@@ -0,0 +1,608 @@
+;
+; Copyright (c) 2016, Alliance for Open Media. All rights reserved
+;
+; This source code is subject to the terms of the BSD 2 Clause License and
+; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+; was not distributed with this source code in the LICENSE file, you can
+; obtain it at www.aomedia.org/license/software. If the Alliance for Open
+; Media Patent License 1.0 was not distributed with this source code in the
+; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+;
+
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pb_1: times 16 db 1
+pw_4:  times 8 dw 4
+pw_8:  times 8 dw 8
+pw_16: times 8 dw 16
+pw_32: times 8 dw 32
+dc_128: times 16 db 128
+pw2_4:  times 8 dw 2
+pw2_8:  times 8 dw 4
+pw2_16:  times 8 dw 8
+pw2_32:  times 8 dw 16
+
+SECTION .text
+
+INIT_XMM sse2
+cglobal dc_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  movd                  m2, [leftq]
+  movd                  m0, [aboveq]
+  pxor                  m1, m1
+  punpckldq             m0, m2
+  psadbw                m0, m1
+  paddw                 m0, [GLOBAL(pw_4)]
+  psraw                 m0, 3
+  pshuflw               m0, m0, 0x0
+  packuswb              m0, m0
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+  lea                 dstq, [dstq+strideq*2]
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_4x4, 2, 5, 2, dst, stride, above, left, goffset
+  movifnidn          leftq, leftmp
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  movd                  m0, [leftq]
+  psadbw                m0, m1
+  paddw                 m0, [GLOBAL(pw2_4)]
+  psraw                 m0, 2
+  pshuflw               m0, m0, 0x0
+  packuswb              m0, m0
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+  lea                 dstq, [dstq+strideq*2]
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_top_predictor_4x4, 3, 5, 2, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  movd                  m0, [aboveq]
+  psadbw                m0, m1
+  paddw                 m0, [GLOBAL(pw2_4)]
+  psraw                 m0, 2
+  pshuflw               m0, m0, 0x0
+  packuswb              m0, m0
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+  lea                 dstq, [dstq+strideq*2]
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  movq                  m0, [aboveq]
+  movq                  m2, [leftq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  psadbw                m0, m1
+  psadbw                m2, m1
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw_8)]
+  psraw                 m0, 4
+  punpcklbw             m0, m0
+  pshuflw               m0, m0, 0x0
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_top_predictor_8x8, 3, 5, 2, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  movq                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  psadbw                m0, m1
+  paddw                 m0, [GLOBAL(pw2_8)]
+  psraw                 m0, 3
+  punpcklbw             m0, m0
+  pshuflw               m0, m0, 0x0
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_8x8, 2, 5, 2, dst, stride, above, left, goffset
+  movifnidn          leftq, leftmp
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  movq                  m0, [leftq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  psadbw                m0, m1
+  paddw                 m0, [GLOBAL(pw2_8)]
+  psraw                 m0, 3
+  punpcklbw             m0, m0
+  pshuflw               m0, m0, 0x0
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_4x4, 2, 5, 1, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  movd     m0,        [GLOBAL(dc_128)]
+  movd    [dstq          ], m0
+  movd    [dstq+strideq  ], m0
+  movd    [dstq+strideq*2], m0
+  movd    [dstq+stride3q ], m0
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_8x8, 2, 5, 1, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  movq    m0,        [GLOBAL(dc_128)]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal dc_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [leftq]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  psadbw                m0, m1
+  psadbw                m2, m1
+  paddw                 m0, m2
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw_16)]
+  psraw                 m0, 5
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova    [dstq          ], m0
+  mova    [dstq+strideq  ], m0
+  mova    [dstq+strideq*2], m0
+  mova    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+
+INIT_XMM sse2
+cglobal dc_top_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  psadbw                m0, m1
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw2_16)]
+  psraw                 m0, 4
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova    [dstq          ], m0
+  mova    [dstq+strideq  ], m0
+  mova    [dstq+strideq*2], m0
+  mova    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [leftq]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  psadbw                m0, m1
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw2_16)]
+  psraw                 m0, 4
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova    [dstq          ], m0
+  mova    [dstq+strideq  ], m0
+  mova    [dstq+strideq*2], m0
+  mova    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 4
+  mova    m0,        [GLOBAL(dc_128)]
+.loop:
+  mova    [dstq          ], m0
+  mova    [dstq+strideq  ], m0
+  mova    [dstq+strideq*2], m0
+  mova    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+  RESTORE_GOT
+  RET
+
+
+INIT_XMM sse2
+cglobal dc_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [aboveq+16]
+  mova                  m3, [leftq]
+  mova                  m4, [leftq+16]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  psadbw                m0, m1
+  psadbw                m2, m1
+  psadbw                m3, m1
+  psadbw                m4, m1
+  paddw                 m0, m2
+  paddw                 m0, m3
+  paddw                 m0, m4
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw_32)]
+  psraw                 m0, 6
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova [dstq             ], m0
+  mova [dstq          +16], m0
+  mova [dstq+strideq     ], m0
+  mova [dstq+strideq  +16], m0
+  mova [dstq+strideq*2   ], m0
+  mova [dstq+strideq*2+16], m0
+  mova [dstq+stride3q    ], m0
+  mova [dstq+stride3q +16], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal dc_top_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [aboveq]
+  mova                  m2, [aboveq+16]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  psadbw                m0, m1
+  psadbw                m2, m1
+  paddw                 m0, m2
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw2_32)]
+  psraw                 m0, 5
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova [dstq             ], m0
+  mova [dstq          +16], m0
+  mova [dstq+strideq     ], m0
+  mova [dstq+strideq  +16], m0
+  mova [dstq+strideq*2   ], m0
+  mova [dstq+strideq*2+16], m0
+  mova [dstq+stride3q    ], m0
+  mova [dstq+stride3q +16], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  pxor                  m1, m1
+  mova                  m0, [leftq]
+  mova                  m2, [leftq+16]
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  psadbw                m0, m1
+  psadbw                m2, m1
+  paddw                 m0, m2
+  movhlps               m2, m0
+  paddw                 m0, m2
+  paddw                 m0, [GLOBAL(pw2_32)]
+  psraw                 m0, 5
+  pshuflw               m0, m0, 0x0
+  punpcklqdq            m0, m0
+  packuswb              m0, m0
+.loop:
+  mova [dstq             ], m0
+  mova [dstq          +16], m0
+  mova [dstq+strideq     ], m0
+  mova [dstq+strideq  +16], m0
+  mova [dstq+strideq*2   ], m0
+  mova [dstq+strideq*2+16], m0
+  mova [dstq+stride3q    ], m0
+  mova [dstq+stride3q +16], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+
+  RESTORE_GOT
+  REP_RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_32x32, 4, 5, 3, dst, stride, above, left, goffset
+  GET_GOT     goffsetq
+
+  DEFINE_ARGS dst, stride, stride3, lines4
+  lea             stride3q, [strideq*3]
+  mov              lines4d, 8
+  mova    m0,        [GLOBAL(dc_128)]
+.loop:
+  mova [dstq             ], m0
+  mova [dstq          +16], m0
+  mova [dstq+strideq     ], m0
+  mova [dstq+strideq  +16], m0
+  mova [dstq+strideq*2   ], m0
+  mova [dstq+strideq*2+16], m0
+  mova [dstq+stride3q    ], m0
+  mova [dstq+stride3q +16], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec              lines4d
+  jnz .loop
+  RESTORE_GOT
+  RET
+
+INIT_XMM sse2
+cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above
+  movd                  m0, [aboveq]
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+  lea                 dstq, [dstq+strideq*2]
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m0
+  RET
+
+INIT_XMM sse2
+cglobal v_predictor_8x8, 3, 3, 1, dst, stride, above
+  movq                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3
+  lea             stride3q, [strideq*3]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  movq    [dstq          ], m0
+  movq    [dstq+strideq  ], m0
+  movq    [dstq+strideq*2], m0
+  movq    [dstq+stride3q ], m0
+  RET
+
+INIT_XMM sse2
+cglobal v_predictor_16x16, 3, 4, 1, dst, stride, above
+  mova                  m0, [aboveq]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 4
+.loop:
+  mova    [dstq          ], m0
+  mova    [dstq+strideq  ], m0
+  mova    [dstq+strideq*2], m0
+  mova    [dstq+stride3q ], m0
+  lea                 dstq, [dstq+strideq*4]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal v_predictor_32x32, 3, 4, 2, dst, stride, above
+  mova                  m0, [aboveq]
+  mova                  m1, [aboveq+16]
+  DEFINE_ARGS dst, stride, stride3, nlines4
+  lea             stride3q, [strideq*3]
+  mov              nlines4d, 8
+.loop:
+  mova [dstq             ], m0
+  mova [dstq          +16], m1
+  mova [dstq+strideq     ], m0
+  mova [dstq+strideq  +16], m1
+  mova [dstq+strideq*2   ], m0
+  mova [dstq+strideq*2+16], m1
+  mova [dstq+stride3q    ], m0
+  mova [dstq+stride3q +16], m1
+  lea                 dstq, [dstq+strideq*4]
+  dec             nlines4d
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal h_predictor_4x4, 2, 4, 4, dst, stride, line, left
+  movifnidn          leftq, leftmp
+  movd                  m0, [leftq]
+  punpcklbw             m0, m0
+  punpcklbw             m0, m0
+  pshufd                m1, m0, 0x1
+  movd      [dstq        ], m0
+  movd      [dstq+strideq], m1
+  pshufd                m2, m0, 0x2
+  lea                 dstq, [dstq+strideq*2]
+  pshufd                m3, m0, 0x3
+  movd      [dstq        ], m2
+  movd      [dstq+strideq], m3
+  RET
+
+INIT_XMM sse2
+cglobal h_predictor_8x8, 2, 5, 3, dst, stride, line, left
+  movifnidn          leftq, leftmp
+  mov                lineq, -2
+  DEFINE_ARGS  dst, stride, line, left, stride3
+  lea             stride3q, [strideq*3]
+  movq                  m0, [leftq    ]
+  punpcklbw             m0, m0              ; l1 l1 l2 l2 ... l8 l8
+.loop:
+  pshuflw               m1, m0, 0x0         ; l1 l1 l1 l1 l1 l1 l1 l1
+  pshuflw               m2, m0, 0x55        ; l2 l2 l2 l2 l2 l2 l2 l2
+  movq      [dstq        ], m1
+  movq      [dstq+strideq], m2
+  pshuflw               m1, m0, 0xaa
+  pshuflw               m2, m0, 0xff
+  movq    [dstq+strideq*2], m1
+  movq    [dstq+stride3q ], m2
+  pshufd                m0, m0, 0xe         ; [63:0] l5 l5 l6 l6 l7 l7 l8 l8
+  inc                lineq
+  lea                 dstq, [dstq+strideq*4]
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal h_predictor_16x16, 2, 5, 3, dst, stride, line, left
+  movifnidn          leftq, leftmp
+  mov                lineq, -4
+  DEFINE_ARGS dst, stride, line, left, stride3
+  lea             stride3q, [strideq*3]
+.loop:
+  movd                  m0, [leftq]
+  punpcklbw             m0, m0
+  punpcklbw             m0, m0              ; l1 to l4 each repeated 4 times
+  pshufd            m1, m0, 0x0             ; l1 repeated 16 times
+  pshufd            m2, m0, 0x55            ; l2 repeated 16 times
+  mova    [dstq          ], m1
+  mova    [dstq+strideq  ], m2
+  pshufd            m1, m0, 0xaa
+  pshufd            m2, m0, 0xff
+  mova    [dstq+strideq*2], m1
+  mova    [dstq+stride3q ], m2
+  inc                lineq
+  lea                leftq, [leftq+4       ]
+  lea                 dstq, [dstq+strideq*4]
+  jnz .loop
+  REP_RET
+
+INIT_XMM sse2
+cglobal h_predictor_32x32, 2, 5, 3, dst, stride, line, left
+  movifnidn              leftq, leftmp
+  mov                    lineq, -8
+  DEFINE_ARGS dst, stride, line, left, stride3
+  lea                 stride3q, [strideq*3]
+.loop:
+  movd                      m0, [leftq]
+  punpcklbw                 m0, m0
+  punpcklbw                 m0, m0              ; l1 to l4 each repeated 4 times
+  pshufd                m1, m0, 0x0             ; l1 repeated 16 times
+  pshufd                m2, m0, 0x55            ; l2 repeated 16 times
+  mova     [dstq             ], m1
+  mova     [dstq+16          ], m1
+  mova     [dstq+strideq     ], m2
+  mova     [dstq+strideq+16  ], m2
+  pshufd                m1, m0, 0xaa
+  pshufd                m2, m0, 0xff
+  mova     [dstq+strideq*2   ], m1
+  mova     [dstq+strideq*2+16], m1
+  mova     [dstq+stride3q    ], m2
+  mova     [dstq+stride3q+16 ], m2
+  inc                    lineq
+  lea                    leftq, [leftq+4       ]
+  lea                     dstq, [dstq+strideq*4]
+  jnz .loop
+  REP_RET
diff --git a/third_party/aom/aom_dsp/x86/intrapred_avx2.c b/third_party/aom/aom_dsp/x86/intrapred_avx2.c
index 1e67d392e8..97569b3ed1 100644
--- a/third_party/aom/aom_dsp/x86/intrapred_avx2.c
+++ b/third_party/aom/aom_dsp/x86/intrapred_avx2.c
@@ -12,6 +12,7 @@
 #include <immintrin.h>
 
 #include "config/aom_dsp_rtcd.h"
+#include "aom_dsp/x86/lpf_common_sse2.h"
 
 static INLINE __m256i dc_sum_64(const uint8_t *ref) {
   const __m256i x0 = _mm256_loadu_si256((const __m256i *)ref);
@@ -63,6 +64,187 @@ static INLINE void row_store_64xh(const __m256i *r, int height, uint8_t *dst,
   }
 }
 
+static INLINE void highbd_transpose16x4_8x8_sse2(__m128i *x, __m128i *d) {
+  __m128i r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
+
+  r0 = _mm_unpacklo_epi16(x[0], x[1]);
+  r1 = _mm_unpacklo_epi16(x[2], x[3]);
+  r2 = _mm_unpacklo_epi16(x[4], x[5]);
+  r3 = _mm_unpacklo_epi16(x[6], x[7]);
+
+  r4 = _mm_unpacklo_epi16(x[8], x[9]);
+  r5 = _mm_unpacklo_epi16(x[10], x[11]);
+  r6 = _mm_unpacklo_epi16(x[12], x[13]);
+  r7 = _mm_unpacklo_epi16(x[14], x[15]);
+
+  r8 = _mm_unpacklo_epi32(r0, r1);
+  r9 = _mm_unpackhi_epi32(r0, r1);
+  r10 = _mm_unpacklo_epi32(r2, r3);
+  r11 = _mm_unpackhi_epi32(r2, r3);
+
+  r12 = _mm_unpacklo_epi32(r4, r5);
+  r13 = _mm_unpackhi_epi32(r4, r5);
+  r14 = _mm_unpacklo_epi32(r6, r7);
+  r15 = _mm_unpackhi_epi32(r6, r7);
+
+  r0 = _mm_unpacklo_epi64(r8, r9);
+  r1 = _mm_unpackhi_epi64(r8, r9);
+  r2 = _mm_unpacklo_epi64(r10, r11);
+  r3 = _mm_unpackhi_epi64(r10, r11);
+
+  r4 = _mm_unpacklo_epi64(r12, r13);
+  r5 = _mm_unpackhi_epi64(r12, r13);
+  r6 = _mm_unpacklo_epi64(r14, r15);
+  r7 = _mm_unpackhi_epi64(r14, r15);
+
+  d[0] = _mm_unpacklo_epi64(r0, r2);
+  d[1] = _mm_unpacklo_epi64(r4, r6);
+  d[2] = _mm_unpacklo_epi64(r1, r3);
+  d[3] = _mm_unpacklo_epi64(r5, r7);
+
+  d[4] = _mm_unpackhi_epi64(r0, r2);
+  d[5] = _mm_unpackhi_epi64(r4, r6);
+  d[6] = _mm_unpackhi_epi64(r1, r3);
+  d[7] = _mm_unpackhi_epi64(r5, r7);
+}
+
+static INLINE void highbd_transpose4x16_avx2(__m256i *x, __m256i *d) {
+  __m256i w0, w1, w2, w3, ww0, ww1;
+
+  w0 = _mm256_unpacklo_epi16(x[0], x[1]);  // 00 10 01 11 02 12 03 13
+  w1 = _mm256_unpacklo_epi16(x[2], x[3]);  // 20 30 21 31 22 32 23 33
+  w2 = _mm256_unpackhi_epi16(x[0], x[1]);  // 40 50 41 51 42 52 43 53
+  w3 = _mm256_unpackhi_epi16(x[2], x[3]);  // 60 70 61 71 62 72 63 73
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);  // 00 10 20 30 01 11 21 31
+  ww1 = _mm256_unpacklo_epi32(w2, w3);  // 40 50 60 70 41 51 61 71
+
+  d[0] = _mm256_unpacklo_epi64(ww0, ww1);  // 00 10 20 30 40 50 60 70
+  d[1] = _mm256_unpackhi_epi64(ww0, ww1);  // 01 11 21 31 41 51 61 71
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);  // 02 12 22 32 03 13 23 33
+  ww1 = _mm256_unpackhi_epi32(w2, w3);  // 42 52 62 72 43 53 63 73
+
+  d[2] = _mm256_unpacklo_epi64(ww0, ww1);  // 02 12 22 32 42 52 62 72
+  d[3] = _mm256_unpackhi_epi64(ww0, ww1);  // 03 13 23 33 43 53 63 73
+}
+
+static INLINE void highbd_transpose8x16_16x8_avx2(__m256i *x, __m256i *d) {
+  __m256i w0, w1, w2, w3, ww0, ww1;
+
+  w0 = _mm256_unpacklo_epi16(x[0], x[1]);  // 00 10 01 11 02 12 03 13
+  w1 = _mm256_unpacklo_epi16(x[2], x[3]);  // 20 30 21 31 22 32 23 33
+  w2 = _mm256_unpacklo_epi16(x[4], x[5]);  // 40 50 41 51 42 52 43 53
+  w3 = _mm256_unpacklo_epi16(x[6], x[7]);  // 60 70 61 71 62 72 63 73
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);  // 00 10 20 30 01 11 21 31
+  ww1 = _mm256_unpacklo_epi32(w2, w3);  // 40 50 60 70 41 51 61 71
+
+  d[0] = _mm256_unpacklo_epi64(ww0, ww1);  // 00 10 20 30 40 50 60 70
+  d[1] = _mm256_unpackhi_epi64(ww0, ww1);  // 01 11 21 31 41 51 61 71
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);  // 02 12 22 32 03 13 23 33
+  ww1 = _mm256_unpackhi_epi32(w2, w3);  // 42 52 62 72 43 53 63 73
+
+  d[2] = _mm256_unpacklo_epi64(ww0, ww1);  // 02 12 22 32 42 52 62 72
+  d[3] = _mm256_unpackhi_epi64(ww0, ww1);  // 03 13 23 33 43 53 63 73
+
+  w0 = _mm256_unpackhi_epi16(x[0], x[1]);  // 04 14 05 15 06 16 07 17
+  w1 = _mm256_unpackhi_epi16(x[2], x[3]);  // 24 34 25 35 26 36 27 37
+  w2 = _mm256_unpackhi_epi16(x[4], x[5]);  // 44 54 45 55 46 56 47 57
+  w3 = _mm256_unpackhi_epi16(x[6], x[7]);  // 64 74 65 75 66 76 67 77
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);  // 04 14 24 34 05 15 25 35
+  ww1 = _mm256_unpacklo_epi32(w2, w3);  // 44 54 64 74 45 55 65 75
+
+  d[4] = _mm256_unpacklo_epi64(ww0, ww1);  // 04 14 24 34 44 54 64 74
+  d[5] = _mm256_unpackhi_epi64(ww0, ww1);  // 05 15 25 35 45 55 65 75
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);  // 06 16 26 36 07 17 27 37
+  ww1 = _mm256_unpackhi_epi32(w2, w3);  // 46 56 66 76 47 57 67 77
+
+  d[6] = _mm256_unpacklo_epi64(ww0, ww1);  // 06 16 26 36 46 56 66 76
+  d[7] = _mm256_unpackhi_epi64(ww0, ww1);  // 07 17 27 37 47 57 67 77
+}
+
+static INLINE void highbd_transpose16x16_avx2(__m256i *x, __m256i *d) {
+  __m256i w0, w1, w2, w3, ww0, ww1;
+  __m256i dd[16];
+  w0 = _mm256_unpacklo_epi16(x[0], x[1]);
+  w1 = _mm256_unpacklo_epi16(x[2], x[3]);
+  w2 = _mm256_unpacklo_epi16(x[4], x[5]);
+  w3 = _mm256_unpacklo_epi16(x[6], x[7]);
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);  //
+  ww1 = _mm256_unpacklo_epi32(w2, w3);  //
+
+  dd[0] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[1] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);  //
+  ww1 = _mm256_unpackhi_epi32(w2, w3);  //
+
+  dd[2] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[3] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  w0 = _mm256_unpackhi_epi16(x[0], x[1]);
+  w1 = _mm256_unpackhi_epi16(x[2], x[3]);
+  w2 = _mm256_unpackhi_epi16(x[4], x[5]);
+  w3 = _mm256_unpackhi_epi16(x[6], x[7]);
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);  //
+  ww1 = _mm256_unpacklo_epi32(w2, w3);  //
+
+  dd[4] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[5] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);  //
+  ww1 = _mm256_unpackhi_epi32(w2, w3);  //
+
+  dd[6] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[7] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  w0 = _mm256_unpacklo_epi16(x[8], x[9]);
+  w1 = _mm256_unpacklo_epi16(x[10], x[11]);
+  w2 = _mm256_unpacklo_epi16(x[12], x[13]);
+  w3 = _mm256_unpacklo_epi16(x[14], x[15]);
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);
+  ww1 = _mm256_unpacklo_epi32(w2, w3);
+
+  dd[8] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[9] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);
+  ww1 = _mm256_unpackhi_epi32(w2, w3);
+
+  dd[10] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[11] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  w0 = _mm256_unpackhi_epi16(x[8], x[9]);
+  w1 = _mm256_unpackhi_epi16(x[10], x[11]);
+  w2 = _mm256_unpackhi_epi16(x[12], x[13]);
+  w3 = _mm256_unpackhi_epi16(x[14], x[15]);
+
+  ww0 = _mm256_unpacklo_epi32(w0, w1);
+  ww1 = _mm256_unpacklo_epi32(w2, w3);
+
+  dd[12] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[13] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  ww0 = _mm256_unpackhi_epi32(w0, w1);
+  ww1 = _mm256_unpackhi_epi32(w2, w3);
+
+  dd[14] = _mm256_unpacklo_epi64(ww0, ww1);
+  dd[15] = _mm256_unpackhi_epi64(ww0, ww1);
+
+  for (int i = 0; i < 8; i++) {
+    d[i] = _mm256_insertf128_si256(dd[i], _mm256_castsi256_si128(dd[i + 8]), 1);
+    d[i + 8] = _mm256_insertf128_si256(dd[i + 8],
+                                       _mm256_extracti128_si256(dd[i], 1), 0);
+  }
+}
+
 void aom_dc_predictor_32x32_avx2(uint8_t *dst, ptrdiff_t stride,
                                  const uint8_t *above, const uint8_t *left) {
   const __m256i sum_above = dc_sum_32(above);
@@ -196,7 +378,7 @@ void aom_dc_predictor_32x16_avx2(uint8_t *dst, ptrdiff_t stride,
   const __m128i top_sum = dc_sum_32_sse2(above);
   __m128i left_sum = dc_sum_16_sse2(left);
   left_sum = _mm_add_epi16(top_sum, left_sum);
-  uint32_t sum = _mm_cvtsi128_si32(left_sum);
+  uint16_t sum = _mm_cvtsi128_si32(left_sum);
   sum += 24;
   sum /= 48;
   const __m256i row = _mm256_set1_epi8((uint8_t)sum);
@@ -208,7 +390,7 @@ void aom_dc_predictor_32x64_avx2(uint8_t *dst, ptrdiff_t stride,
   const __m256i sum_above = dc_sum_32(above);
   __m256i sum_left = dc_sum_64(left);
   sum_left = _mm256_add_epi16(sum_left, sum_above);
-  uint32_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
+  uint16_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
   sum += 48;
   sum /= 96;
   const __m256i row = _mm256_set1_epi8((uint8_t)sum);
@@ -220,7 +402,7 @@ void aom_dc_predictor_64x64_avx2(uint8_t *dst, ptrdiff_t stride,
   const __m256i sum_above = dc_sum_64(above);
   __m256i sum_left = dc_sum_64(left);
   sum_left = _mm256_add_epi16(sum_left, sum_above);
-  uint32_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
+  uint16_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
   sum += 64;
   sum /= 128;
   const __m256i row = _mm256_set1_epi8((uint8_t)sum);
@@ -232,7 +414,7 @@ void aom_dc_predictor_64x32_avx2(uint8_t *dst, ptrdiff_t stride,
   const __m256i sum_above = dc_sum_64(above);
   __m256i sum_left = dc_sum_32(left);
   sum_left = _mm256_add_epi16(sum_left, sum_above);
-  uint32_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
+  uint16_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
   sum += 48;
   sum /= 96;
   const __m256i row = _mm256_set1_epi8((uint8_t)sum);
@@ -244,7 +426,7 @@ void aom_dc_predictor_64x16_avx2(uint8_t *dst, ptrdiff_t stride,
   const __m256i sum_above = dc_sum_64(above);
   __m256i sum_left = _mm256_castsi128_si256(dc_sum_16_sse2(left));
   sum_left = _mm256_add_epi16(sum_left, sum_above);
-  uint32_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
+  uint16_t sum = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_left));
   sum += 40;
   sum /= 80;
   const __m256i row = _mm256_set1_epi8((uint8_t)sum);
@@ -809,3 +991,3384 @@ void aom_paeth_predictor_64x16_avx2(uint8_t *dst, ptrdiff_t stride,
     rep = _mm256_add_epi16(rep, one);
   }
 }
+
+#define PERM4x64(c0, c1, c2, c3) c0 + (c1 << 2) + (c2 << 4) + (c3 << 6)
+#define PERM2x128(c0, c1) c0 + (c1 << 4)
+
+static AOM_FORCE_INLINE void highbd_dr_prediction_z1_4xN_internal_avx2(
+    int N, __m128i *dst, const uint16_t *above, int upsample_above, int dx) {
+  const int frac_bits = 6 - upsample_above;
+  const int max_base_x = ((N + 4) - 1) << upsample_above;
+  int x;
+  // a assert(dx > 0);
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a1, a32, a16;
+  __m256i diff;
+  __m128i a_mbase_x, max_base_x128, base_inc128, mask128;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm_set1_epi16(above[max_base_x]);
+  max_base_x128 = _mm_set1_epi32(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i res1;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        dst[i] = a_mbase_x;  // save 4 values
+      }
+      return;
+    }
+
+    a0 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base)));
+    a1 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 1)));
+
+    if (upsample_above) {
+      a0 = _mm256_permutevar8x32_epi32(
+          a0, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0, 1));
+      base_inc128 = _mm_setr_epi32(base, base + 2, base + 4, base + 6);
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(
+              _mm256_slli_epi32(_mm256_set1_epi32(x), upsample_above),
+              _mm256_set1_epi32(0x3f)),
+          1);
+    } else {
+      base_inc128 = _mm_setr_epi32(base, base + 1, base + 2, base + 3);
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+    }
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+
+    b = _mm256_mullo_epi32(diff, shift);
+    res = _mm256_add_epi32(a32, b);
+    res = _mm256_srli_epi32(res, 5);
+
+    res1 = _mm256_castsi256_si128(res);
+    res1 = _mm_packus_epi32(res1, res1);
+
+    mask128 = _mm_cmpgt_epi32(max_base_x128, base_inc128);
+    mask128 = _mm_packs_epi32(mask128, mask128);  // goto 16 bit
+    dst[r] = _mm_blendv_epi8(a_mbase_x, res1, mask128);
+    x += dx;
+  }
+}
+
+static void highbd_dr_prediction_z1_4xN_avx2(int N, uint16_t *dst,
+                                             ptrdiff_t stride,
+                                             const uint16_t *above,
+                                             int upsample_above, int dx) {
+  __m128i dstvec[16];
+
+  highbd_dr_prediction_z1_4xN_internal_avx2(N, dstvec, above, upsample_above,
+                                            dx);
+  for (int i = 0; i < N; i++) {
+    _mm_storel_epi64((__m128i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void highbd_dr_prediction_z1_8xN_internal_avx2(
+    int N, __m128i *dst, const uint16_t *above, int upsample_above, int dx) {
+  const int frac_bits = 6 - upsample_above;
+  const int max_base_x = ((8 + N) - 1) << upsample_above;
+
+  int x;
+  // a assert(dx > 0);
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a1, a0_1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff, max_base_x256, base_inc256, mask256;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi16(above[max_base_x]);
+  max_base_x256 = _mm256_set1_epi32(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, res1, shift;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        dst[i] = _mm256_castsi256_si128(a_mbase_x);  // save 8 values
+      }
+      return;
+    }
+
+    a0 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base)));
+    a1 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 1)));
+
+    if (upsample_above) {
+      a0 = _mm256_permutevar8x32_epi32(
+          a0, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0, 1));
+
+      a0_1 =
+          _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 8)));
+      a0_1 = _mm256_permutevar8x32_epi32(
+          a0_1, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1_1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0_1, 1));
+
+      a0 = _mm256_inserti128_si256(a0, _mm256_castsi256_si128(a0_1), 1);
+      a1 = _mm256_inserti128_si256(a1, _mm256_castsi256_si128(a1_1), 1);
+      base_inc256 =
+          _mm256_setr_epi32(base, base + 2, base + 4, base + 6, base + 8,
+                            base + 10, base + 12, base + 14);
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(
+              _mm256_slli_epi32(_mm256_set1_epi32(x), upsample_above),
+              _mm256_set1_epi32(0x3f)),
+          1);
+    } else {
+      base_inc256 = _mm256_setr_epi32(base, base + 1, base + 2, base + 3,
+                                      base + 4, base + 5, base + 6, base + 7);
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+    }
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+
+    b = _mm256_mullo_epi32(diff, shift);
+    res = _mm256_add_epi32(a32, b);
+    res = _mm256_srli_epi32(res, 5);
+
+    res1 = _mm256_packus_epi32(
+        res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+
+    mask256 = _mm256_cmpgt_epi32(max_base_x256, base_inc256);
+    mask256 = _mm256_packs_epi32(
+        mask256, _mm256_castsi128_si256(
+                     _mm256_extracti128_si256(mask256, 1)));  // goto 16 bit
+    res1 = _mm256_blendv_epi8(a_mbase_x, res1, mask256);
+    dst[r] = _mm256_castsi256_si128(res1);
+    x += dx;
+  }
+}
+
+static void highbd_dr_prediction_z1_8xN_avx2(int N, uint16_t *dst,
+                                             ptrdiff_t stride,
+                                             const uint16_t *above,
+                                             int upsample_above, int dx) {
+  __m128i dstvec[32];
+
+  highbd_dr_prediction_z1_8xN_internal_avx2(N, dstvec, above, upsample_above,
+                                            dx);
+  for (int i = 0; i < N; i++) {
+    _mm_storeu_si128((__m128i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void highbd_dr_prediction_z1_16xN_internal_avx2(
+    int N, __m256i *dstvec, const uint16_t *above, int upsample_above, int dx) {
+  int x;
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((16 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff, max_base_x256, base_inc256, mask256;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi16(above[max_base_x]);
+  max_base_x256 = _mm256_set1_epi16(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res[2], res1;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        dstvec[i] = a_mbase_x;  // save 16 values
+      }
+      return;
+    }
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    a0 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base)));
+    a1 = _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 1)));
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+    b = _mm256_mullo_epi32(diff, shift);
+
+    res[0] = _mm256_add_epi32(a32, b);
+    res[0] = _mm256_srli_epi32(res[0], 5);
+    res[0] = _mm256_packus_epi32(
+        res[0], _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+
+    int mdif = max_base_x - base;
+    if (mdif > 8) {
+      a0_1 =
+          _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 8)));
+      a1_1 =
+          _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i *)(above + base + 9)));
+
+      diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+      b = _mm256_mullo_epi32(diff, shift);
+
+      res[1] = _mm256_add_epi32(a32, b);
+      res[1] = _mm256_srli_epi32(res[1], 5);
+      res[1] = _mm256_packus_epi32(
+          res[1], _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+    } else {
+      res[1] = a_mbase_x;
+    }
+    res1 = _mm256_inserti128_si256(res[0], _mm256_castsi256_si128(res[1]),
+                                   1);  // 16 16bit values
+
+    base_inc256 = _mm256_setr_epi16(base, base + 1, base + 2, base + 3,
+                                    base + 4, base + 5, base + 6, base + 7,
+                                    base + 8, base + 9, base + 10, base + 11,
+                                    base + 12, base + 13, base + 14, base + 15);
+    mask256 = _mm256_cmpgt_epi16(max_base_x256, base_inc256);
+    dstvec[r] = _mm256_blendv_epi8(a_mbase_x, res1, mask256);
+    x += dx;
+  }
+}
+
+static void highbd_dr_prediction_z1_16xN_avx2(int N, uint16_t *dst,
+                                              ptrdiff_t stride,
+                                              const uint16_t *above,
+                                              int upsample_above, int dx) {
+  __m256i dstvec[64];
+  highbd_dr_prediction_z1_16xN_internal_avx2(N, dstvec, above, upsample_above,
+                                             dx);
+  for (int i = 0; i < N; i++) {
+    _mm256_storeu_si256((__m256i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void highbd_dr_prediction_z1_32xN_internal_avx2(
+    int N, __m256i *dstvec, const uint16_t *above, int upsample_above, int dx) {
+  int x;
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((32 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff, max_base_x256, base_inc256, mask256;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi16(above[max_base_x]);
+  max_base_x256 = _mm256_set1_epi16(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res[2], res1;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        dstvec[i] = a_mbase_x;  // save 32 values
+        dstvec[i + N] = a_mbase_x;
+      }
+      return;
+    }
+
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    for (int j = 0; j < 32; j += 16) {
+      int mdif = max_base_x - (base + j);
+      if (mdif <= 0) {
+        res1 = a_mbase_x;
+      } else {
+        a0 = _mm256_cvtepu16_epi32(
+            _mm_loadu_si128((__m128i *)(above + base + j)));
+        a1 = _mm256_cvtepu16_epi32(
+            _mm_loadu_si128((__m128i *)(above + base + 1 + j)));
+
+        diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        res[0] = _mm256_add_epi32(a32, b);
+        res[0] = _mm256_srli_epi32(res[0], 5);
+        res[0] = _mm256_packus_epi32(
+            res[0],
+            _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+        if (mdif > 8) {
+          a0_1 = _mm256_cvtepu16_epi32(
+              _mm_loadu_si128((__m128i *)(above + base + 8 + j)));
+          a1_1 = _mm256_cvtepu16_epi32(
+              _mm_loadu_si128((__m128i *)(above + base + 9 + j)));
+
+          diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+          a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+          a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+          b = _mm256_mullo_epi32(diff, shift);
+
+          res[1] = _mm256_add_epi32(a32, b);
+          res[1] = _mm256_srli_epi32(res[1], 5);
+          res[1] = _mm256_packus_epi32(
+              res[1],
+              _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+        } else {
+          res[1] = a_mbase_x;
+        }
+        res1 = _mm256_inserti128_si256(res[0], _mm256_castsi256_si128(res[1]),
+                                       1);  // 16 16bit values
+        base_inc256 = _mm256_setr_epi16(
+            base + j, base + j + 1, base + j + 2, base + j + 3, base + j + 4,
+            base + j + 5, base + j + 6, base + j + 7, base + j + 8,
+            base + j + 9, base + j + 10, base + j + 11, base + j + 12,
+            base + j + 13, base + j + 14, base + j + 15);
+
+        mask256 = _mm256_cmpgt_epi16(max_base_x256, base_inc256);
+        res1 = _mm256_blendv_epi8(a_mbase_x, res1, mask256);
+      }
+      if (!j)
+        dstvec[r] = res1;
+      else
+        dstvec[r + N] = res1;
+    }
+    x += dx;
+  }
+}
+
+static void highbd_dr_prediction_z1_32xN_avx2(int N, uint16_t *dst,
+                                              ptrdiff_t stride,
+                                              const uint16_t *above,
+                                              int upsample_above, int dx) {
+  __m256i dstvec[128];
+
+  highbd_dr_prediction_z1_32xN_internal_avx2(N, dstvec, above, upsample_above,
+                                             dx);
+  for (int i = 0; i < N; i++) {
+    _mm256_storeu_si256((__m256i *)(dst + stride * i), dstvec[i]);
+    _mm256_storeu_si256((__m256i *)(dst + stride * i + 16), dstvec[i + N]);
+  }
+}
+
+static void highbd_dr_prediction_z1_64xN_avx2(int N, uint16_t *dst,
+                                              ptrdiff_t stride,
+                                              const uint16_t *above,
+                                              int upsample_above, int dx) {
+  int x;
+
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((64 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff, max_base_x256, base_inc256, mask256;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi16(above[max_base_x]);
+  max_base_x256 = _mm256_set1_epi16(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++, dst += stride) {
+    __m256i b, res[2], res1;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        _mm256_storeu_si256((__m256i *)dst, a_mbase_x);  // save 32 values
+        _mm256_storeu_si256((__m256i *)(dst + 16), a_mbase_x);
+        _mm256_storeu_si256((__m256i *)(dst + 32), a_mbase_x);
+        _mm256_storeu_si256((__m256i *)(dst + 48), a_mbase_x);
+        dst += stride;
+      }
+      return;
+    }
+
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    __m128i a0_128, a0_1_128, a1_128, a1_1_128;
+    for (int j = 0; j < 64; j += 16) {
+      int mdif = max_base_x - (base + j);
+      if (mdif <= 0) {
+        _mm256_storeu_si256((__m256i *)(dst + j), a_mbase_x);
+      } else {
+        a0_128 = _mm_loadu_si128((__m128i *)(above + base + j));
+        a1_128 = _mm_loadu_si128((__m128i *)(above + base + 1 + j));
+        a0 = _mm256_cvtepu16_epi32(a0_128);
+        a1 = _mm256_cvtepu16_epi32(a1_128);
+
+        diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        res[0] = _mm256_add_epi32(a32, b);
+        res[0] = _mm256_srli_epi32(res[0], 5);
+        res[0] = _mm256_packus_epi32(
+            res[0],
+            _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+        if (mdif > 8) {
+          a0_1_128 = _mm_loadu_si128((__m128i *)(above + base + 8 + j));
+          a1_1_128 = _mm_loadu_si128((__m128i *)(above + base + 9 + j));
+          a0_1 = _mm256_cvtepu16_epi32(a0_1_128);
+          a1_1 = _mm256_cvtepu16_epi32(a1_1_128);
+
+          diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+          a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+          a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+          b = _mm256_mullo_epi32(diff, shift);
+
+          res[1] = _mm256_add_epi32(a32, b);
+          res[1] = _mm256_srli_epi32(res[1], 5);
+          res[1] = _mm256_packus_epi32(
+              res[1],
+              _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+        } else {
+          res[1] = a_mbase_x;
+        }
+        res1 = _mm256_inserti128_si256(res[0], _mm256_castsi256_si128(res[1]),
+                                       1);  // 16 16bit values
+        base_inc256 = _mm256_setr_epi16(
+            base + j, base + j + 1, base + j + 2, base + j + 3, base + j + 4,
+            base + j + 5, base + j + 6, base + j + 7, base + j + 8,
+            base + j + 9, base + j + 10, base + j + 11, base + j + 12,
+            base + j + 13, base + j + 14, base + j + 15);
+
+        mask256 = _mm256_cmpgt_epi16(max_base_x256, base_inc256);
+        res1 = _mm256_blendv_epi8(a_mbase_x, res1, mask256);
+        _mm256_storeu_si256((__m256i *)(dst + j), res1);
+      }
+    }
+    x += dx;
+  }
+}
+
+// Directional prediction, zone 1: 0 < angle < 90
+void av1_highbd_dr_prediction_z1_avx2(uint16_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint16_t *above,
+                                      const uint16_t *left, int upsample_above,
+                                      int dx, int dy, int bd) {
+  (void)left;
+  (void)dy;
+  (void)bd;
+
+  switch (bw) {
+    case 4:
+      highbd_dr_prediction_z1_4xN_avx2(bh, dst, stride, above, upsample_above,
+                                       dx);
+      break;
+    case 8:
+      highbd_dr_prediction_z1_8xN_avx2(bh, dst, stride, above, upsample_above,
+                                       dx);
+      break;
+    case 16:
+      highbd_dr_prediction_z1_16xN_avx2(bh, dst, stride, above, upsample_above,
+                                        dx);
+      break;
+    case 32:
+      highbd_dr_prediction_z1_32xN_avx2(bh, dst, stride, above, upsample_above,
+                                        dx);
+      break;
+    case 64:
+      highbd_dr_prediction_z1_64xN_avx2(bh, dst, stride, above, upsample_above,
+                                        dx);
+      break;
+    default: break;
+  }
+  return;
+}
+
+static void highbd_transpose_TX_8X8(const uint16_t *src, ptrdiff_t pitchSrc,
+                                    uint16_t *dst, ptrdiff_t pitchDst) {
+  __m128i r0, r1, r2, r3, r4, r5, r6, r7, r0_Lo, r1_Lo, r2_Lo, r3_Lo, r4_Lo,
+      r5_Lo, r6_Lo;
+  r0 = _mm_load_si128(
+      (__m128i *)(src + 0 * pitchSrc));  // 07,06,05,04,03,02,01,00
+  r1 = _mm_load_si128(
+      (__m128i *)(src + 1 * pitchSrc));  // 17,16,15,14,13,12,11,10
+  r2 = _mm_load_si128(
+      (__m128i *)(src + 2 * pitchSrc));  // 27,26,25,24,23,22,21,20
+  r3 = _mm_load_si128(
+      (__m128i *)(src + 3 * pitchSrc));  // 37,36,35,34,33,32,31,30
+  r4 = _mm_load_si128(
+      (__m128i *)(src + 4 * pitchSrc));  // 47,46,45,44,43,42,41,40
+  r5 = _mm_load_si128(
+      (__m128i *)(src + 5 * pitchSrc));  // 57,56,55,54,53,52,51,50
+  r6 = _mm_load_si128(
+      (__m128i *)(src + 6 * pitchSrc));  // 67,66,65,64,63,62,61,60
+  r7 = _mm_load_si128(
+      (__m128i *)(src + 7 * pitchSrc));  // 77,76,75,74,73,72,71,70
+
+  r0_Lo = _mm_unpacklo_epi16(r0, r1);
+  r2_Lo = _mm_unpacklo_epi16(r2, r3);
+  r4_Lo = _mm_unpacklo_epi16(r4, r5);
+  r6_Lo = _mm_unpacklo_epi16(r6, r7);
+
+  r1_Lo = r0_Lo;
+  r0_Lo = _mm_unpacklo_epi32(r0_Lo, r2_Lo);
+  r1_Lo = _mm_unpackhi_epi32(r1_Lo, r2_Lo);
+  r5_Lo = r4_Lo;
+  r4_Lo = _mm_unpacklo_epi32(r4_Lo, r6_Lo);
+  r5_Lo = _mm_unpackhi_epi32(r5_Lo, r6_Lo);
+  r2_Lo = r0_Lo;
+  r0_Lo = _mm_unpacklo_epi64(r0_Lo, r4_Lo);  // 64
+  r2_Lo = _mm_unpackhi_epi64(r2_Lo, r4_Lo);
+  r3_Lo = r1_Lo;
+  r1_Lo = _mm_unpacklo_epi64(r1_Lo, r5_Lo);
+  r3_Lo = _mm_unpackhi_epi64(r3_Lo, r5_Lo);
+
+  _mm_storeu_si128((__m128i *)(dst + 0 * pitchDst), r0_Lo);
+  _mm_storeu_si128((__m128i *)(dst + 1 * pitchDst), r2_Lo);
+  _mm_storeu_si128((__m128i *)(dst + 2 * pitchDst), r1_Lo);
+  _mm_storeu_si128((__m128i *)(dst + 3 * pitchDst), r3_Lo);
+
+  r0 = _mm_unpackhi_epi16(r0, r1);
+  r2 = _mm_unpackhi_epi16(r2, r3);
+  r4 = _mm_unpackhi_epi16(r4, r5);
+  r6 = _mm_unpackhi_epi16(r6, r7);
+
+  r1 = r0;
+  r0 = _mm_unpacklo_epi32(r0, r2);
+  r1 = _mm_unpackhi_epi32(r1, r2);
+  r5 = r4;
+  r4 = _mm_unpacklo_epi32(r4, r6);
+  r5 = _mm_unpackhi_epi32(r5, r6);
+  r2 = r0;
+  r0 = _mm_unpacklo_epi64(r0, r4);
+  r2 = _mm_unpackhi_epi64(r2, r4);
+  r3 = r1;
+  r1 = _mm_unpacklo_epi64(r1, r5);
+  r3 = _mm_unpackhi_epi64(r3, r5);
+
+  _mm_storeu_si128((__m128i *)(dst + 4 * pitchDst), r0);
+  _mm_storeu_si128((__m128i *)(dst + 5 * pitchDst), r2);
+  _mm_storeu_si128((__m128i *)(dst + 6 * pitchDst), r1);
+  _mm_storeu_si128((__m128i *)(dst + 7 * pitchDst), r3);
+}
+
+static uint8_t HighbdLoadMaskx[8][16] = {
+  { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+  { 0, 1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 },
+  { 0, 1, 0, 1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2, 3, 4, 5, 6, 7 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2, 3, 4, 5 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2, 3 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 },
+};
+
+static uint8_t HighbdEvenOddMaskx4[8][16] = {
+  { 0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14,
+    15 },  // 0=0,1, 1=2,3, 2=4,5, 3=6,7, 4=8,9, 5=10,11, 6=12,13, 7=14,15,
+           // >7=0,1
+  { 0, 1, 2, 3, 6, 7, 10, 11, 14, 15, 4, 5, 8, 9, 12, 13 },
+  { 0, 1, 0, 1, 4, 5, 8, 9, 12, 13, 0, 1, 6, 7, 10, 11 },
+  { 0, 1, 0, 1, 0, 1, 6, 7, 10, 11, 14, 15, 0, 1, 8, 9 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 8, 9, 12, 13, 0, 1, 0, 1 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 10, 11, 14, 15, 0, 1 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 12, 13, 0, 1 },
+  { 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 14, 15 }
+};
+
+static uint16_t HighbdEvenOddMaskx8_2[8][16] = {
+  { 0, 2, 4, 6, 8, 10, 12, 14 },      { 2, 2, 4, 6, 8, 10, 12, 14 },
+  { 4, 4, 4, 6, 8, 10, 12, 14 },      { 6, 6, 6, 6, 8, 10, 12, 14 },
+  { 8, 8, 8, 8, 8, 10, 12, 14 },      { 10, 10, 10, 10, 10, 10, 12, 14 },
+  { 12, 12, 12, 12, 12, 12, 12, 14 }, { 14, 14, 14, 14, 14, 14, 14, 14 },
+};
+
+static uint16_t HighbdBaseMask[17][16] = {
+  {
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+      0,
+  },
+  { 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0, 0, 0, 0, 0,
+    0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0, 0,
+    0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0,
+    0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0xffff, 0, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0xffff, 0xffff, 0, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0 },
+  { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff,
+    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff }
+};
+
+static void highbd_dr_prediction_z2_Nx4_avx2(
+    int N, uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+    const uint16_t *left, int upsample_above, int upsample_left, int dx,
+    int dy) {
+  const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  const int frac_bits_x = 6 - upsample_above;
+  const int frac_bits_y = 6 - upsample_left;
+
+  // a assert(dx > 0);
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a16;
+  __m256i diff;
+  __m128i c3f, min_base_y128;
+
+  a16 = _mm256_set1_epi32(16);
+  c3f = _mm_set1_epi32(0x3f);
+  min_base_y128 = _mm_set1_epi32(min_base_y);
+
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i resx, resy, resxy;
+    __m128i a0_x128, a1_x128;
+    int y = r + 1;
+    int base_x = (-y * dx) >> frac_bits_x;
+    int base_shift = 0;
+    if (base_x < (min_base_x - 1)) {
+      base_shift = (min_base_x - base_x - 1) >> upsample_above;
+    }
+    int base_min_diff =
+        (min_base_x - base_x + upsample_above) >> upsample_above;
+    if (base_min_diff > 4) {
+      base_min_diff = 4;
+    } else {
+      if (base_min_diff < 0) base_min_diff = 0;
+    }
+
+    if (base_shift > 3) {
+      resx = _mm_setzero_si128();
+    } else {
+      a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
+      a1_x128 = _mm_loadu_si128((__m128i *)(above + base_x + 1 + base_shift));
+
+      if (upsample_above) {
+        a0_x128 = _mm_shuffle_epi8(a0_x128,
+                                   *(__m128i *)HighbdEvenOddMaskx4[base_shift]);
+        a1_x128 = _mm_shuffle_epi8(a1_x128,
+                                   *(__m128i *)HighbdEvenOddMaskx4[base_shift]);
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(
+                _mm_slli_epi32(
+                    _mm_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                   (2 << 6) - y * dx, (3 << 6) - y * dx),
+                    upsample_above),
+                c3f),
+            1));
+      } else {
+        a0_x128 =
+            _mm_shuffle_epi8(a0_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+        a1_x128 =
+            _mm_shuffle_epi8(a1_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(_mm_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                         (2 << 6) - y * dx, (3 << 6) - y * dx),
+                          c3f),
+            1));
+      }
+      a0_x = _mm256_cvtepu16_epi32(a0_x128);
+      a1_x = _mm256_cvtepu16_epi32(a1_x128);
+
+      diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resx = _mm256_castsi256_si128(res);
+      resx = _mm_packus_epi32(resx, resx);
+    }
+    // y calc
+    if (base_x < min_base_x) {
+      DECLARE_ALIGNED(32, int, base_y_c[4]);
+      __m128i r6, c1234, dy128, y_c128, base_y_c128, mask128;
+      r6 = _mm_set1_epi32(r << 6);
+      dy128 = _mm_set1_epi32(dy);
+      c1234 = _mm_setr_epi32(1, 2, 3, 4);
+      y_c128 = _mm_sub_epi32(r6, _mm_mullo_epi32(c1234, dy128));
+      base_y_c128 = _mm_srai_epi32(y_c128, frac_bits_y);
+      mask128 = _mm_cmpgt_epi32(min_base_y128, base_y_c128);
+      base_y_c128 = _mm_andnot_si128(mask128, base_y_c128);
+      _mm_store_si128((__m128i *)base_y_c, base_y_c128);
+
+      a0_y = _mm256_castsi128_si256(
+          _mm_setr_epi32(left[base_y_c[0]], left[base_y_c[1]],
+                         left[base_y_c[2]], left[base_y_c[3]]));
+      a1_y = _mm256_castsi128_si256(
+          _mm_setr_epi32(left[base_y_c[0] + 1], left[base_y_c[1] + 1],
+                         left[base_y_c[2] + 1], left[base_y_c[3] + 1]));
+
+      if (upsample_left) {
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(_mm_slli_epi32(y_c128, upsample_left), c3f), 1));
+      } else {
+        shift = _mm256_castsi128_si256(
+            _mm_srli_epi32(_mm_and_si128(y_c128, c3f), 1));
+      }
+      diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resy = _mm256_castsi256_si128(res);
+      resy = _mm_packus_epi32(resy, resy);
+    } else {
+      resy = resx;
+    }
+    resxy =
+        _mm_blendv_epi8(resx, resy, *(__m128i *)HighbdBaseMask[base_min_diff]);
+    _mm_storel_epi64((__m128i *)(dst), resxy);
+    dst += stride;
+  }
+}
+
+static void highbd_dr_prediction_z2_Nx8_avx2(
+    int N, uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+    const uint16_t *left, int upsample_above, int upsample_left, int dx,
+    int dy) {
+  const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  const int frac_bits_x = 6 - upsample_above;
+  const int frac_bits_y = 6 - upsample_left;
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a16, c3f, min_base_y256;
+  __m256i diff;
+  __m128i a0_x128, a1_x128;
+
+  a16 = _mm256_set1_epi32(16);
+  c3f = _mm256_set1_epi32(0x3f);
+  min_base_y256 = _mm256_set1_epi32(min_base_y);
+
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i resx, resy, resxy;
+    int y = r + 1;
+    int base_x = (-y * dx) >> frac_bits_x;
+    int base_shift = 0;
+    if (base_x < (min_base_x - 1)) {
+      base_shift = (min_base_x - base_x - 1) >> upsample_above;
+    }
+    int base_min_diff =
+        (min_base_x - base_x + upsample_above) >> upsample_above;
+    if (base_min_diff > 8) {
+      base_min_diff = 8;
+    } else {
+      if (base_min_diff < 0) base_min_diff = 0;
+    }
+
+    if (base_shift > 7) {
+      resx = _mm_setzero_si128();
+    } else {
+      if (upsample_above) {
+        a0_x128 = _mm_setr_epi16(
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][0]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][1]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][2]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][3]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][4]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][5]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][6]],
+            above[base_x + HighbdEvenOddMaskx8_2[base_shift][7]]);
+        a1_x128 = _mm_setr_epi16(
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][0]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][1]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][2]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][3]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][4]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][5]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][6]],
+            above[base_x + 1 + HighbdEvenOddMaskx8_2[base_shift][7]]);
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_slli_epi32(
+                    _mm256_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                      (2 << 6) - y * dx, (3 << 6) - y * dx,
+                                      (4 << 6) - y * dx, (5 << 6) - y * dx,
+                                      (6 << 6) - y * dx, (7 << 6) - y * dx),
+                    upsample_above),
+                c3f),
+            1);
+      } else {
+        a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
+        a1_x128 = _mm_loadu_si128((__m128i *)(above + base_x + 1 + base_shift));
+        a0_x128 =
+            _mm_shuffle_epi8(a0_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+        a1_x128 =
+            _mm_shuffle_epi8(a1_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(-y * dx, (1 << 6) - y * dx, (2 << 6) - y * dx,
+                                  (3 << 6) - y * dx, (4 << 6) - y * dx,
+                                  (5 << 6) - y * dx, (6 << 6) - y * dx,
+                                  (7 << 6) - y * dx),
+                c3f),
+            1);
+      }
+
+      a0_x = _mm256_cvtepu16_epi32(a0_x128);
+      a1_x = _mm256_cvtepu16_epi32(a1_x128);
+
+      diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resx = _mm256_castsi256_si128(_mm256_packus_epi32(
+          res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1))));
+    }
+    // y calc
+    if (base_x < min_base_x) {
+      DECLARE_ALIGNED(32, int, base_y_c[8]);
+      __m256i r6, c256, dy256, y_c256, base_y_c256, mask256;
+      r6 = _mm256_set1_epi32(r << 6);
+      dy256 = _mm256_set1_epi32(dy);
+      c256 = _mm256_setr_epi32(1, 2, 3, 4, 5, 6, 7, 8);
+      y_c256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+      base_y_c256 = _mm256_srai_epi32(y_c256, frac_bits_y);
+      mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+      base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+      _mm256_store_si256((__m256i *)base_y_c, base_y_c256);
+
+      a0_y = _mm256_cvtepu16_epi32(_mm_setr_epi16(
+          left[base_y_c[0]], left[base_y_c[1]], left[base_y_c[2]],
+          left[base_y_c[3]], left[base_y_c[4]], left[base_y_c[5]],
+          left[base_y_c[6]], left[base_y_c[7]]));
+      a1_y = _mm256_cvtepu16_epi32(_mm_setr_epi16(
+          left[base_y_c[0] + 1], left[base_y_c[1] + 1], left[base_y_c[2] + 1],
+          left[base_y_c[3] + 1], left[base_y_c[4] + 1], left[base_y_c[5] + 1],
+          left[base_y_c[6] + 1], left[base_y_c[7] + 1]));
+
+      if (upsample_left) {
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(_mm256_slli_epi32((y_c256), upsample_left), c3f),
+            1);
+      } else {
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c256, c3f), 1);
+      }
+      diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resy = _mm256_castsi256_si128(_mm256_packus_epi32(
+          res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1))));
+    } else {
+      resy = resx;
+    }
+    resxy =
+        _mm_blendv_epi8(resx, resy, *(__m128i *)HighbdBaseMask[base_min_diff]);
+    _mm_storeu_si128((__m128i *)(dst), resxy);
+    dst += stride;
+  }
+}
+
+static void highbd_dr_prediction_z2_HxW_avx2(
+    int H, int W, uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+    const uint16_t *left, int upsample_above, int upsample_left, int dx,
+    int dy) {
+  // here upsample_above and upsample_left are 0 by design of
+  // av1_use_intra_edge_upsample
+  const int min_base_x = -1;
+  const int min_base_y = -1;
+  (void)upsample_above;
+  (void)upsample_left;
+  const int frac_bits_x = 6;
+  const int frac_bits_y = 6;
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a0_1_x, a1_1_x, a16;
+  __m256i diff, min_base_y256, c3f;
+  __m128i a0_x128, a1_x128, a0_1_x128, a1_1_x128;
+
+  a16 = _mm256_set1_epi32(16);
+  min_base_y256 = _mm256_set1_epi16(min_base_y);
+  c3f = _mm256_set1_epi32(0x3f);
+  for (int r = 0; r < H; r++) {
+    __m256i b, res, shift;
+    __m256i resx[2], resy[2];
+    __m256i resxy;
+    for (int j = 0; j < W; j += 16) {
+      int y = r + 1;
+      int base_x = (-y * dx) >> frac_bits_x;
+      int base_shift = 0;
+      if ((base_x + j) < (min_base_x - 1)) {
+        base_shift = (min_base_x - (base_x + j) - 1);
+      }
+      int base_min_diff = (min_base_x - base_x - j);
+      if (base_min_diff > 16) {
+        base_min_diff = 16;
+      } else {
+        if (base_min_diff < 0) base_min_diff = 0;
+      }
+
+      if (base_shift > 7) {
+        resx[0] = _mm256_setzero_si256();
+      } else {
+        a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift + j));
+        a1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 1 + j));
+        a0_x128 =
+            _mm_shuffle_epi8(a0_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+        a1_x128 =
+            _mm_shuffle_epi8(a1_x128, *(__m128i *)HighbdLoadMaskx[base_shift]);
+
+        a0_x = _mm256_cvtepu16_epi32(a0_x128);
+        a1_x = _mm256_cvtepu16_epi32(a1_x128);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(
+                    ((0 + j) << 6) - y * dx, ((1 + j) << 6) - y * dx,
+                    ((2 + j) << 6) - y * dx, ((3 + j) << 6) - y * dx,
+                    ((4 + j) << 6) - y * dx, ((5 + j) << 6) - y * dx,
+                    ((6 + j) << 6) - y * dx, ((7 + j) << 6) - y * dx),
+                _mm256_set1_epi32(0x3f)),
+            1);
+
+        diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        resx[0] = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+      }
+      base_shift = 0;
+      if ((base_x + j + 8) < (min_base_x - 1)) {
+        base_shift = (min_base_x - (base_x + j + 8) - 1);
+      }
+      if (base_shift > 7) {
+        resx[1] = _mm256_setzero_si256();
+      } else {
+        a0_1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 8 + j));
+        a1_1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 9 + j));
+        a0_1_x128 = _mm_shuffle_epi8(a0_1_x128,
+                                     *(__m128i *)HighbdLoadMaskx[base_shift]);
+        a1_1_x128 = _mm_shuffle_epi8(a1_1_x128,
+                                     *(__m128i *)HighbdLoadMaskx[base_shift]);
+
+        a0_1_x = _mm256_cvtepu16_epi32(a0_1_x128);
+        a1_1_x = _mm256_cvtepu16_epi32(a1_1_x128);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(
+                    ((8 + j) << 6) - y * dx, ((9 + j) << 6) - y * dx,
+                    ((10 + j) << 6) - y * dx, ((11 + j) << 6) - y * dx,
+                    ((12 + j) << 6) - y * dx, ((13 + j) << 6) - y * dx,
+                    ((14 + j) << 6) - y * dx, ((15 + j) << 6) - y * dx),
+                _mm256_set1_epi32(0x3f)),
+            1);
+
+        diff = _mm256_sub_epi32(a1_1_x, a0_1_x);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_1_x, 5);       // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);         // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        resx[1] = _mm256_add_epi32(a32, b);
+        resx[1] = _mm256_srli_epi32(resx[1], 5);
+        resx[1] = _mm256_packus_epi32(
+            resx[1],
+            _mm256_castsi128_si256(_mm256_extracti128_si256(resx[1], 1)));
+      }
+      resx[0] =
+          _mm256_inserti128_si256(resx[0], _mm256_castsi256_si128(resx[1]),
+                                  1);  // 16 16bit values
+
+      // y calc
+      if ((base_x < min_base_x)) {
+        DECLARE_ALIGNED(32, int, base_y_c[16]);
+        __m256i r6, c256, dy256, y_c256, y_c_1_256, base_y_c256, mask256;
+        r6 = _mm256_set1_epi32(r << 6);
+        dy256 = _mm256_set1_epi32(dy);
+        c256 = _mm256_setr_epi32(1 + j, 2 + j, 3 + j, 4 + j, 5 + j, 6 + j,
+                                 7 + j, 8 + j);
+        y_c256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+        base_y_c256 = _mm256_srai_epi32(y_c256, frac_bits_y);
+        mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+        base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+        _mm256_store_si256((__m256i *)base_y_c, base_y_c256);
+        c256 = _mm256_setr_epi32(9 + j, 10 + j, 11 + j, 12 + j, 13 + j, 14 + j,
+                                 15 + j, 16 + j);
+        y_c_1_256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+        base_y_c256 = _mm256_srai_epi32(y_c_1_256, frac_bits_y);
+        mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+        base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+        _mm256_store_si256((__m256i *)(base_y_c + 8), base_y_c256);
+
+        a0_y = _mm256_cvtepu16_epi32(_mm_setr_epi16(
+            left[base_y_c[0]], left[base_y_c[1]], left[base_y_c[2]],
+            left[base_y_c[3]], left[base_y_c[4]], left[base_y_c[5]],
+            left[base_y_c[6]], left[base_y_c[7]]));
+        a1_y = _mm256_cvtepu16_epi32(_mm_setr_epi16(
+            left[base_y_c[0] + 1], left[base_y_c[1] + 1], left[base_y_c[2] + 1],
+            left[base_y_c[3] + 1], left[base_y_c[4] + 1], left[base_y_c[5] + 1],
+            left[base_y_c[6] + 1], left[base_y_c[7] + 1]));
+
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c256, c3f), 1);
+
+        diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        resy[0] = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+
+        a0_y = _mm256_cvtepu16_epi32(_mm_setr_epi16(
+            left[base_y_c[8]], left[base_y_c[9]], left[base_y_c[10]],
+            left[base_y_c[11]], left[base_y_c[12]], left[base_y_c[13]],
+            left[base_y_c[14]], left[base_y_c[15]]));
+        a1_y = _mm256_cvtepu16_epi32(
+            _mm_setr_epi16(left[base_y_c[8] + 1], left[base_y_c[9] + 1],
+                           left[base_y_c[10] + 1], left[base_y_c[11] + 1],
+                           left[base_y_c[12] + 1], left[base_y_c[13] + 1],
+                           left[base_y_c[14] + 1], left[base_y_c[15] + 1]));
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c_1_256, c3f), 1);
+
+        diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        resy[1] = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+
+        resy[0] =
+            _mm256_inserti128_si256(resy[0], _mm256_castsi256_si128(resy[1]),
+                                    1);  // 16 16bit values
+      } else {
+        resy[0] = resx[0];
+      }
+      resxy = _mm256_blendv_epi8(resx[0], resy[0],
+                                 *(__m256i *)HighbdBaseMask[base_min_diff]);
+      _mm256_storeu_si256((__m256i *)(dst + j), resxy);
+    }  // for j
+    dst += stride;
+  }
+}
+
+// Directional prediction, zone 2: 90 < angle < 180
+void av1_highbd_dr_prediction_z2_avx2(uint16_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint16_t *above,
+                                      const uint16_t *left, int upsample_above,
+                                      int upsample_left, int dx, int dy,
+                                      int bd) {
+  (void)bd;
+  assert(dx > 0);
+  assert(dy > 0);
+  switch (bw) {
+    case 4:
+      highbd_dr_prediction_z2_Nx4_avx2(bh, dst, stride, above, left,
+                                       upsample_above, upsample_left, dx, dy);
+      break;
+    case 8:
+      highbd_dr_prediction_z2_Nx8_avx2(bh, dst, stride, above, left,
+                                       upsample_above, upsample_left, dx, dy);
+      break;
+    default:
+      highbd_dr_prediction_z2_HxW_avx2(bh, bw, dst, stride, above, left,
+                                       upsample_above, upsample_left, dx, dy);
+      break;
+  }
+  return;
+}
+
+static void highbd_transpose(const uint16_t *src, ptrdiff_t pitchSrc,
+                             uint16_t *dst, ptrdiff_t pitchDst, int width,
+                             int height) {
+  for (int j = 0; j < height; j += 8)
+    for (int i = 0; i < width; i += 8)
+      highbd_transpose_TX_8X8(src + i * pitchSrc + j, pitchSrc,
+                              dst + j * pitchDst + i, pitchDst);
+}
+
+static void highbd_dr_prediction_z3_4x4_avx2(uint16_t *dst, ptrdiff_t stride,
+                                             const uint16_t *left,
+                                             int upsample_left, int dy) {
+  __m128i dstvec[4], d[4];
+
+  highbd_dr_prediction_z1_4xN_internal_avx2(4, dstvec, left, upsample_left, dy);
+  highbd_transpose4x8_8x4_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2],
+                                   &dstvec[3], &d[0], &d[1], &d[2], &d[3]);
+  _mm_storel_epi64((__m128i *)(dst + 0 * stride), d[0]);
+  _mm_storel_epi64((__m128i *)(dst + 1 * stride), d[1]);
+  _mm_storel_epi64((__m128i *)(dst + 2 * stride), d[2]);
+  _mm_storel_epi64((__m128i *)(dst + 3 * stride), d[3]);
+  return;
+}
+
+static void highbd_dr_prediction_z3_8x8_avx2(uint16_t *dst, ptrdiff_t stride,
+                                             const uint16_t *left,
+                                             int upsample_left, int dy) {
+  __m128i dstvec[8], d[8];
+
+  highbd_dr_prediction_z1_8xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  highbd_transpose8x8_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
+                           &dstvec[4], &dstvec[5], &dstvec[6], &dstvec[7],
+                           &d[0], &d[1], &d[2], &d[3], &d[4], &d[5], &d[6],
+                           &d[7]);
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void highbd_dr_prediction_z3_4x8_avx2(uint16_t *dst, ptrdiff_t stride,
+                                             const uint16_t *left,
+                                             int upsample_left, int dy) {
+  __m128i dstvec[4], d[8];
+
+  highbd_dr_prediction_z1_8xN_internal_avx2(4, dstvec, left, upsample_left, dy);
+  highbd_transpose4x8_8x4_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
+                               &d[0], &d[1], &d[2], &d[3], &d[4], &d[5], &d[6],
+                               &d[7]);
+  for (int i = 0; i < 8; i++) {
+    _mm_storel_epi64((__m128i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void highbd_dr_prediction_z3_8x4_avx2(uint16_t *dst, ptrdiff_t stride,
+                                             const uint16_t *left,
+                                             int upsample_left, int dy) {
+  __m128i dstvec[8], d[4];
+
+  highbd_dr_prediction_z1_4xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  highbd_transpose8x8_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
+                               &dstvec[4], &dstvec[5], &dstvec[6], &dstvec[7],
+                               &d[0], &d[1], &d[2], &d[3]);
+  _mm_storeu_si128((__m128i *)(dst + 0 * stride), d[0]);
+  _mm_storeu_si128((__m128i *)(dst + 1 * stride), d[1]);
+  _mm_storeu_si128((__m128i *)(dst + 2 * stride), d[2]);
+  _mm_storeu_si128((__m128i *)(dst + 3 * stride), d[3]);
+}
+
+static void highbd_dr_prediction_z3_8x16_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m256i dstvec[8], d[16];
+
+  highbd_dr_prediction_z1_16xN_internal_avx2(8, dstvec, left, upsample_left,
+                                             dy);
+  highbd_transpose8x16_16x8_avx2(dstvec, d);
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride),
+                     _mm256_castsi256_si128(d[i]));
+  }
+  for (int i = 8; i < 16; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride),
+                     _mm256_extracti128_si256(d[i - 8], 1));
+  }
+}
+
+static void highbd_dr_prediction_z3_16x8_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m128i dstvec[16], d[16];
+
+  highbd_dr_prediction_z1_8xN_internal_avx2(16, dstvec, left, upsample_left,
+                                            dy);
+  for (int i = 0; i < 16; i += 8) {
+    highbd_transpose8x8_sse2(&dstvec[0 + i], &dstvec[1 + i], &dstvec[2 + i],
+                             &dstvec[3 + i], &dstvec[4 + i], &dstvec[5 + i],
+                             &dstvec[6 + i], &dstvec[7 + i], &d[0 + i],
+                             &d[1 + i], &d[2 + i], &d[3 + i], &d[4 + i],
+                             &d[5 + i], &d[6 + i], &d[7 + i]);
+  }
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+    _mm_storeu_si128((__m128i *)(dst + i * stride + 8), d[i + 8]);
+  }
+}
+
+static void highbd_dr_prediction_z3_4x16_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m256i dstvec[4], d[4], d1;
+
+  highbd_dr_prediction_z1_16xN_internal_avx2(4, dstvec, left, upsample_left,
+                                             dy);
+  highbd_transpose4x16_avx2(dstvec, d);
+  for (int i = 0; i < 4; i++) {
+    _mm_storel_epi64((__m128i *)(dst + i * stride),
+                     _mm256_castsi256_si128(d[i]));
+    d1 = _mm256_bsrli_epi128(d[i], 8);
+    _mm_storel_epi64((__m128i *)(dst + (i + 4) * stride),
+                     _mm256_castsi256_si128(d1));
+    _mm_storel_epi64((__m128i *)(dst + (i + 8) * stride),
+                     _mm256_extracti128_si256(d[i], 1));
+    _mm_storel_epi64((__m128i *)(dst + (i + 12) * stride),
+                     _mm256_extracti128_si256(d1, 1));
+  }
+}
+
+static void highbd_dr_prediction_z3_16x4_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m128i dstvec[16], d[8];
+
+  highbd_dr_prediction_z1_4xN_internal_avx2(16, dstvec, left, upsample_left,
+                                            dy);
+  highbd_transpose16x4_8x8_sse2(dstvec, d);
+
+  _mm_storeu_si128((__m128i *)(dst + 0 * stride), d[0]);
+  _mm_storeu_si128((__m128i *)(dst + 0 * stride + 8), d[1]);
+  _mm_storeu_si128((__m128i *)(dst + 1 * stride), d[2]);
+  _mm_storeu_si128((__m128i *)(dst + 1 * stride + 8), d[3]);
+  _mm_storeu_si128((__m128i *)(dst + 2 * stride), d[4]);
+  _mm_storeu_si128((__m128i *)(dst + 2 * stride + 8), d[5]);
+  _mm_storeu_si128((__m128i *)(dst + 3 * stride), d[6]);
+  _mm_storeu_si128((__m128i *)(dst + 3 * stride + 8), d[7]);
+}
+
+static void highbd_dr_prediction_z3_8x32_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m256i dstvec[16], d[16];
+
+  highbd_dr_prediction_z1_32xN_internal_avx2(8, dstvec, left, upsample_left,
+                                             dy);
+  for (int i = 0; i < 16; i += 8) {
+    highbd_transpose8x16_16x8_avx2(dstvec + i, d + i);
+  }
+
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride),
+                     _mm256_castsi256_si128(d[i]));
+  }
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + (i + 8) * stride),
+                     _mm256_extracti128_si256(d[i], 1));
+  }
+  for (int i = 8; i < 16; i++) {
+    _mm_storeu_si128((__m128i *)(dst + (i + 8) * stride),
+                     _mm256_castsi256_si128(d[i]));
+  }
+  for (int i = 8; i < 16; i++) {
+    _mm_storeu_si128((__m128i *)(dst + (i + 16) * stride),
+                     _mm256_extracti128_si256(d[i], 1));
+  }
+}
+
+static void highbd_dr_prediction_z3_32x8_avx2(uint16_t *dst, ptrdiff_t stride,
+                                              const uint16_t *left,
+                                              int upsample_left, int dy) {
+  __m128i dstvec[32], d[32];
+
+  highbd_dr_prediction_z1_8xN_internal_avx2(32, dstvec, left, upsample_left,
+                                            dy);
+  for (int i = 0; i < 32; i += 8) {
+    highbd_transpose8x8_sse2(&dstvec[0 + i], &dstvec[1 + i], &dstvec[2 + i],
+                             &dstvec[3 + i], &dstvec[4 + i], &dstvec[5 + i],
+                             &dstvec[6 + i], &dstvec[7 + i], &d[0 + i],
+                             &d[1 + i], &d[2 + i], &d[3 + i], &d[4 + i],
+                             &d[5 + i], &d[6 + i], &d[7 + i]);
+  }
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+    _mm_storeu_si128((__m128i *)(dst + i * stride + 8), d[i + 8]);
+    _mm_storeu_si128((__m128i *)(dst + i * stride + 16), d[i + 16]);
+    _mm_storeu_si128((__m128i *)(dst + i * stride + 24), d[i + 24]);
+  }
+}
+
+static void highbd_dr_prediction_z3_16x16_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  __m256i dstvec[16], d[16];
+
+  highbd_dr_prediction_z1_16xN_internal_avx2(16, dstvec, left, upsample_left,
+                                             dy);
+  highbd_transpose16x16_avx2(dstvec, d);
+
+  for (int i = 0; i < 16; i++) {
+    _mm256_storeu_si256((__m256i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void highbd_dr_prediction_z3_32x32_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  __m256i dstvec[64], d[16];
+
+  highbd_dr_prediction_z1_32xN_internal_avx2(32, dstvec, left, upsample_left,
+                                             dy);
+
+  highbd_transpose16x16_avx2(dstvec, d);
+  for (int j = 0; j < 16; j++) {
+    _mm256_storeu_si256((__m256i *)(dst + j * stride), d[j]);
+  }
+  highbd_transpose16x16_avx2(dstvec + 16, d);
+  for (int j = 0; j < 16; j++) {
+    _mm256_storeu_si256((__m256i *)(dst + j * stride + 16), d[j]);
+  }
+  highbd_transpose16x16_avx2(dstvec + 32, d);
+  for (int j = 0; j < 16; j++) {
+    _mm256_storeu_si256((__m256i *)(dst + (j + 16) * stride), d[j]);
+  }
+  highbd_transpose16x16_avx2(dstvec + 48, d);
+  for (int j = 0; j < 16; j++) {
+    _mm256_storeu_si256((__m256i *)(dst + (j + 16) * stride + 16), d[j]);
+  }
+}
+
+static void highbd_dr_prediction_z3_64x64_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  DECLARE_ALIGNED(16, uint16_t, dstT[64 * 64]);
+  highbd_dr_prediction_z1_64xN_avx2(64, dstT, 64, left, upsample_left, dy);
+  highbd_transpose(dstT, 64, dst, stride, 64, 64);
+}
+
+static void highbd_dr_prediction_z3_16x32_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  __m256i dstvec[32], d[32];
+
+  highbd_dr_prediction_z1_32xN_internal_avx2(16, dstvec, left, upsample_left,
+                                             dy);
+  for (int i = 0; i < 32; i += 8) {
+    highbd_transpose8x16_16x8_avx2(dstvec + i, d + i);
+  }
+  // store
+  for (int j = 0; j < 32; j += 16) {
+    for (int i = 0; i < 8; i++) {
+      _mm_storeu_si128((__m128i *)(dst + (i + j) * stride),
+                       _mm256_castsi256_si128(d[(i + j)]));
+    }
+    for (int i = 0; i < 8; i++) {
+      _mm_storeu_si128((__m128i *)(dst + (i + j) * stride + 8),
+                       _mm256_castsi256_si128(d[(i + j) + 8]));
+    }
+    for (int i = 8; i < 16; i++) {
+      _mm256_storeu_si256(
+          (__m256i *)(dst + (i + j) * stride),
+          _mm256_inserti128_si256(
+              d[(i + j)], _mm256_extracti128_si256(d[(i + j) - 8], 1), 0));
+    }
+  }
+}
+
+static void highbd_dr_prediction_z3_32x16_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  __m256i dstvec[32], d[16];
+
+  highbd_dr_prediction_z1_16xN_internal_avx2(32, dstvec, left, upsample_left,
+                                             dy);
+  for (int i = 0; i < 32; i += 16) {
+    highbd_transpose16x16_avx2((dstvec + i), d);
+    for (int j = 0; j < 16; j++) {
+      _mm256_storeu_si256((__m256i *)(dst + j * stride + i), d[j]);
+    }
+  }
+}
+
+static void highbd_dr_prediction_z3_32x64_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  uint16_t dstT[64 * 32];
+  highbd_dr_prediction_z1_64xN_avx2(32, dstT, 64, left, upsample_left, dy);
+  highbd_transpose(dstT, 64, dst, stride, 32, 64);
+}
+
+static void highbd_dr_prediction_z3_64x32_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  DECLARE_ALIGNED(16, uint16_t, dstT[32 * 64]);
+  highbd_dr_prediction_z1_32xN_avx2(64, dstT, 32, left, upsample_left, dy);
+  highbd_transpose(dstT, 32, dst, stride, 64, 32);
+  return;
+}
+
+static void highbd_dr_prediction_z3_16x64_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  DECLARE_ALIGNED(16, uint16_t, dstT[64 * 16]);
+  highbd_dr_prediction_z1_64xN_avx2(16, dstT, 64, left, upsample_left, dy);
+  highbd_transpose(dstT, 64, dst, stride, 16, 64);
+}
+
+static void highbd_dr_prediction_z3_64x16_avx2(uint16_t *dst, ptrdiff_t stride,
+                                               const uint16_t *left,
+                                               int upsample_left, int dy) {
+  __m256i dstvec[64], d[16];
+
+  highbd_dr_prediction_z1_16xN_internal_avx2(64, dstvec, left, upsample_left,
+                                             dy);
+  for (int i = 0; i < 64; i += 16) {
+    highbd_transpose16x16_avx2((dstvec + i), d);
+    for (int j = 0; j < 16; j++) {
+      _mm256_storeu_si256((__m256i *)(dst + j * stride + i), d[j]);
+    }
+  }
+}
+
+void av1_highbd_dr_prediction_z3_avx2(uint16_t *dst, ptrdiff_t stride, int bw,
+                                      int bh, const uint16_t *above,
+                                      const uint16_t *left, int upsample_left,
+                                      int dx, int dy, int bd) {
+  (void)above;
+  (void)dx;
+  (void)bd;
+  assert(dx == 1);
+  assert(dy > 0);
+  if (bw == bh) {
+    switch (bw) {
+      case 4:
+        highbd_dr_prediction_z3_4x4_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 8:
+        highbd_dr_prediction_z3_8x8_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 16:
+        highbd_dr_prediction_z3_16x16_avx2(dst, stride, left, upsample_left,
+                                           dy);
+        break;
+      case 32:
+        highbd_dr_prediction_z3_32x32_avx2(dst, stride, left, upsample_left,
+                                           dy);
+        break;
+      case 64:
+        highbd_dr_prediction_z3_64x64_avx2(dst, stride, left, upsample_left,
+                                           dy);
+        break;
+    }
+  } else {
+    if (bw < bh) {
+      if (bw + bw == bh) {
+        switch (bw) {
+          case 4:
+            highbd_dr_prediction_z3_4x8_avx2(dst, stride, left, upsample_left,
+                                             dy);
+            break;
+          case 8:
+            highbd_dr_prediction_z3_8x16_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 16:
+            highbd_dr_prediction_z3_16x32_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+          case 32:
+            highbd_dr_prediction_z3_32x64_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+        }
+      } else {
+        switch (bw) {
+          case 4:
+            highbd_dr_prediction_z3_4x16_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 8:
+            highbd_dr_prediction_z3_8x32_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 16:
+            highbd_dr_prediction_z3_16x64_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+        }
+      }
+    } else {
+      if (bh + bh == bw) {
+        switch (bh) {
+          case 4:
+            highbd_dr_prediction_z3_8x4_avx2(dst, stride, left, upsample_left,
+                                             dy);
+            break;
+          case 8:
+            highbd_dr_prediction_z3_16x8_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 16:
+            highbd_dr_prediction_z3_32x16_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+          case 32:
+            highbd_dr_prediction_z3_64x32_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+        }
+      } else {
+        switch (bh) {
+          case 4:
+            highbd_dr_prediction_z3_16x4_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 8:
+            highbd_dr_prediction_z3_32x8_avx2(dst, stride, left, upsample_left,
+                                              dy);
+            break;
+          case 16:
+            highbd_dr_prediction_z3_64x16_avx2(dst, stride, left, upsample_left,
+                                               dy);
+            break;
+        }
+      }
+    }
+  }
+  return;
+}
+
+// Low bit depth functions
+static uint8_t BaseMask[33][32] = {
+  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0, 0, 0, 0, 0, 0, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0,    0,    0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0,    0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0,    0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0,    0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0,    0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0,    0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,    0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0 },
+  { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+    0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
+};
+
+static AOM_FORCE_INLINE void dr_prediction_z1_4xN_internal_avx2(
+    int N, __m128i *dst, const uint8_t *above, int upsample_above, int dx) {
+  const int frac_bits = 6 - upsample_above;
+  const int max_base_x = ((N + 4) - 1) << upsample_above;
+  int x;
+  // a assert(dx > 0);
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a1, a32, a16;
+  __m256i diff;
+  __m128i a_mbase_x;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm_set1_epi8(above[max_base_x]);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i res1, a0_128, a1_128;
+
+    int base = x >> frac_bits;
+    int base_max_diff = (max_base_x - base) >> upsample_above;
+    if (base_max_diff <= 0) {
+      for (int i = r; i < N; ++i) {
+        dst[i] = a_mbase_x;  // save 4 values
+      }
+      return;
+    }
+    if (base_max_diff > 4) base_max_diff = 4;
+    a0_128 = _mm_loadu_si128((__m128i *)(above + base));
+    a1_128 = _mm_srli_si128(a0_128, 1);
+    a0 = _mm256_cvtepu8_epi32(a0_128);
+    a1 = _mm256_cvtepu8_epi32(a1_128);
+
+    if (upsample_above) {
+      a0 = _mm256_permutevar8x32_epi32(
+          a0, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0, 1));
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(
+              _mm256_slli_epi32(_mm256_set1_epi32(x), upsample_above),
+              _mm256_set1_epi32(0x3f)),
+          1);
+    } else {
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+    }
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+
+    b = _mm256_mullo_epi32(diff, shift);
+    res = _mm256_add_epi32(a32, b);
+    res = _mm256_srli_epi32(res, 5);
+
+    res1 = _mm256_castsi256_si128(res);
+    res1 = _mm_packus_epi32(res1, res1);
+    res1 = _mm_packus_epi16(res1, res1);
+
+    dst[r] =
+        _mm_blendv_epi8(a_mbase_x, res1, *(__m128i *)BaseMask[base_max_diff]);
+    x += dx;
+  }
+}
+
+static void dr_prediction_z1_4xN_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *above, int upsample_above,
+                                      int dx) {
+  __m128i dstvec[16];
+
+  dr_prediction_z1_4xN_internal_avx2(N, dstvec, above, upsample_above, dx);
+  for (int i = 0; i < N; i++) {
+    *(uint32_t *)(dst + stride * i) = _mm_cvtsi128_si32(dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void dr_prediction_z1_8xN_internal_avx2(
+    int N, __m128i *dst, const uint8_t *above, int upsample_above, int dx) {
+  const int frac_bits = 6 - upsample_above;
+  const int max_base_x = ((8 + N) - 1) << upsample_above;
+
+  int x;
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a1, a0_1, a1_1, a32, a16, diff;
+  __m128i a_mbase_x;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm_set1_epi8(above[max_base_x]);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, res1, shift;
+    __m128i res128;
+
+    int base = x >> frac_bits;
+    int base_max_diff = (max_base_x - base) >> upsample_above;
+    if (base_max_diff <= 0) {
+      for (int i = r; i < N; ++i) {
+        dst[i] = a_mbase_x;  // save 16 values, 8 to be used furter
+      }
+      return;
+    }
+    if (base_max_diff > 8) base_max_diff = 8;
+
+    a0 = _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base)));
+    a1 = _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base + 1)));
+
+    if (upsample_above) {
+      a0 = _mm256_permutevar8x32_epi32(
+          a0, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0, 1));
+
+      a0_1 =
+          _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base + 8)));
+      a0_1 = _mm256_permutevar8x32_epi32(
+          a0_1, _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0));
+      a1_1 = _mm256_castsi128_si256(_mm256_extracti128_si256(a0_1, 1));
+
+      a0 = _mm256_inserti128_si256(a0, _mm256_castsi256_si128(a0_1), 1);
+      a1 = _mm256_inserti128_si256(a1, _mm256_castsi256_si128(a1_1), 1);
+
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(
+              _mm256_slli_epi32(_mm256_set1_epi32(x), upsample_above),
+              _mm256_set1_epi32(0x3f)),
+          1);
+    } else {
+      shift = _mm256_srli_epi32(
+          _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+    }
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+
+    b = _mm256_mullo_epi32(diff, shift);
+    res = _mm256_add_epi32(a32, b);
+    res = _mm256_srli_epi32(res, 5);
+
+    res1 = _mm256_packus_epi32(
+        res, _mm256_castsi128_si256(
+                 _mm256_extracti128_si256(res, 1)));  // goto 16 bit
+
+    res128 = _mm_packus_epi16(_mm256_castsi256_si128(res1),
+                              _mm256_castsi256_si128(res1));  // goto 8 bit
+
+    res128 =
+        _mm_blendv_epi8(a_mbase_x, res128, *(__m128i *)BaseMask[base_max_diff]);
+    dst[r] = res128;
+    x += dx;
+  }
+}
+
+static void dr_prediction_z1_8xN_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *above, int upsample_above,
+                                      int dx) {
+  __m128i dstvec[32];
+
+  dr_prediction_z1_8xN_internal_avx2(N, dstvec, above, upsample_above, dx);
+  for (int i = 0; i < N; i++) {
+    _mm_storel_epi64((__m128i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void dr_prediction_z1_16xN_internal_avx2(
+    int N, __m128i *dstvec, const uint8_t *above, int upsample_above, int dx) {
+  int x;
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((16 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, diff, a32, a16;
+  __m128i a_mbase_x;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm_set1_epi8((uint8_t)above[max_base_x]);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res[2];
+    __m128i res128[2];
+    int base = x >> frac_bits;
+    int base_max_diff = (max_base_x - base);
+    if (base_max_diff <= 0) {
+      for (int i = r; i < N; ++i) {
+        dstvec[i] = a_mbase_x;  // save 16 values
+      }
+      return;
+    }
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    a0 = _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base)));
+    a1 = _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base + 1)));
+
+    diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+    a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+    a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+    b = _mm256_mullo_epi32(diff, shift);
+
+    res[0] = _mm256_add_epi32(a32, b);
+    res[0] = _mm256_srli_epi32(res[0], 5);
+    res[0] = _mm256_packus_epi32(
+        res[0], _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+    res128[0] = _mm_packus_epi16(_mm256_castsi256_si128(res[0]),
+                                 _mm256_castsi256_si128(res[0]));  // goto 8 bit
+
+    if (base_max_diff > 8) {
+      if (base_max_diff > 16) base_max_diff = 16;
+      a0_1 =
+          _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base + 8)));
+      a1_1 =
+          _mm256_cvtepu8_epi32(_mm_loadu_si128((__m128i *)(above + base + 9)));
+
+      diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+      b = _mm256_mullo_epi32(diff, shift);
+
+      res[1] = _mm256_add_epi32(a32, b);
+      res[1] = _mm256_srli_epi32(res[1], 5);
+      res[1] = _mm256_packus_epi32(
+          res[1], _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+      res128[1] =
+          _mm_packus_epi16(_mm256_castsi256_si128(res[1]),
+                           _mm256_castsi256_si128(res[1]));  // goto 8 bit
+
+    } else {
+      res128[1] = a_mbase_x;
+    }
+    res128[0] = _mm_unpacklo_epi64(res128[0], res128[1]);  // 16 8bit values
+
+    dstvec[r] = _mm_blendv_epi8(a_mbase_x, res128[0],
+                                *(__m128i *)BaseMask[base_max_diff]);
+    x += dx;
+  }
+}
+static void dr_prediction_z1_16xN_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *above, int upsample_above,
+                                       int dx) {
+  __m128i dstvec[64];
+
+  dr_prediction_z1_16xN_internal_avx2(N, dstvec, above, upsample_above, dx);
+  for (int i = 0; i < N; i++) {
+    _mm_storeu_si128((__m128i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static AOM_FORCE_INLINE void dr_prediction_z1_32xN_internal_avx2(
+    int N, __m256i *dstvec, const uint8_t *above, int upsample_above, int dx) {
+  int x;
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((32 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi8(above[max_base_x]);
+
+  x = dx;
+  for (int r = 0; r < N; r++) {
+    __m256i b, res[2], res16[2];
+
+    int base = x >> frac_bits;
+    int base_max_diff = (max_base_x - base);
+    if (base_max_diff <= 0) {
+      for (int i = r; i < N; ++i) {
+        dstvec[i] = a_mbase_x;  // save 32 values
+      }
+      return;
+    }
+    if (base_max_diff > 32) base_max_diff = 32;
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    for (int j = 0, jj = 0; j < 32; j += 16, jj++) {
+      int mdiff = base_max_diff - j;
+      if (mdiff <= 0) {
+        res16[jj] = a_mbase_x;
+      } else {
+        a0 = _mm256_cvtepu8_epi32(
+            _mm_loadu_si128((__m128i *)(above + base + j)));
+        a1 = _mm256_cvtepu8_epi32(
+            _mm_loadu_si128((__m128i *)(above + base + 1 + j)));
+
+        diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        res[0] = _mm256_add_epi32(a32, b);
+        res[0] = _mm256_srli_epi32(res[0], 5);
+        res[0] = _mm256_packus_epi32(
+            res[0],
+            _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+
+        // goto 8 bit
+        res[0] = _mm256_packus_epi16(res[0], res[0]);
+
+        if (mdiff > 8) {
+          a0_1 = _mm256_cvtepu8_epi32(
+              _mm_loadu_si128((__m128i *)(above + base + 8 + j)));
+          a1_1 = _mm256_cvtepu8_epi32(
+              _mm_loadu_si128((__m128i *)(above + base + 9 + j)));
+
+          diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+          a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+          a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+          b = _mm256_mullo_epi32(diff, shift);
+
+          res[1] = _mm256_add_epi32(a32, b);
+          res[1] = _mm256_srli_epi32(res[1], 5);
+          res[1] = _mm256_packus_epi32(
+              res[1],
+              _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+          res[1] = _mm256_packus_epi16(res[1], res[1]);
+          // goto 8 bit
+        } else {
+          res[1] = a_mbase_x;
+        }
+        res16[jj] = _mm256_unpacklo_epi64(res[0], res[1]);  // 16 8bit values
+      }
+    }
+    res16[1] =
+        _mm256_inserti128_si256(res16[0], _mm256_castsi256_si128(res16[1]),
+                                1);  // 32 8bit values
+
+    dstvec[r] = _mm256_blendv_epi8(
+        a_mbase_x, res16[1],
+        *(__m256i *)BaseMask[base_max_diff]);  // 32 8bit values
+    x += dx;
+  }
+}
+
+static void dr_prediction_z1_32xN_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *above, int upsample_above,
+                                       int dx) {
+  __m256i dstvec[64];
+  dr_prediction_z1_32xN_internal_avx2(N, dstvec, above, upsample_above, dx);
+  for (int i = 0; i < N; i++) {
+    _mm256_storeu_si256((__m256i *)(dst + stride * i), dstvec[i]);
+  }
+}
+
+static void dr_prediction_z1_64xN_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *above, int upsample_above,
+                                       int dx) {
+  int x;
+
+  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
+  (void)upsample_above;
+  const int frac_bits = 6;
+  const int max_base_x = ((64 + N) - 1);
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0, a0_1, a1, a1_1, a32, a16;
+  __m256i a_mbase_x, diff;
+  __m128i max_base_x128, base_inc128, mask128;
+
+  a16 = _mm256_set1_epi32(16);
+  a_mbase_x = _mm256_set1_epi8(above[max_base_x]);
+  max_base_x128 = _mm_set1_epi8(max_base_x);
+
+  x = dx;
+  for (int r = 0; r < N; r++, dst += stride) {
+    __m256i b, res[2];
+    __m128i res1;
+
+    int base = x >> frac_bits;
+    if (base >= max_base_x) {
+      for (int i = r; i < N; ++i) {
+        _mm256_storeu_si256((__m256i *)dst, a_mbase_x);  // save 32 values
+        _mm256_storeu_si256((__m256i *)(dst + 32), a_mbase_x);
+        dst += stride;
+      }
+      return;
+    }
+
+    __m256i shift = _mm256_srli_epi32(
+        _mm256_and_si256(_mm256_set1_epi32(x), _mm256_set1_epi32(0x3f)), 1);
+
+    __m128i a0_128, a0_1_128, a1_128, a1_1_128;
+    for (int j = 0; j < 64; j += 16) {
+      int mdif = max_base_x - (base + j);
+      if (mdif <= 0) {
+        _mm_storeu_si128((__m128i *)(dst + j),
+                         _mm256_castsi256_si128(a_mbase_x));
+      } else {
+        a0_128 = _mm_loadu_si128((__m128i *)(above + base + j));
+        a1_128 = _mm_loadu_si128((__m128i *)(above + base + 1 + j));
+        a0 = _mm256_cvtepu8_epi32(a0_128);
+        a1 = _mm256_cvtepu8_epi32(a1_128);
+
+        diff = _mm256_sub_epi32(a1, a0);   // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0, 5);    // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);  // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        res[0] = _mm256_add_epi32(a32, b);
+        res[0] = _mm256_srli_epi32(res[0], 5);
+        res[0] = _mm256_packus_epi32(
+            res[0],
+            _mm256_castsi128_si256(_mm256_extracti128_si256(res[0], 1)));
+        // goto 8 bit
+        res[0] = _mm256_packus_epi16(res[0], res[0]);
+
+        if (mdif > 8) {
+          a0_1_128 = _mm_loadu_si128((__m128i *)(above + base + 8 + j));
+          a1_1_128 = _mm_loadu_si128((__m128i *)(above + base + 9 + j));
+          a0_1 = _mm256_cvtepu8_epi32(a0_1_128);
+          a1_1 = _mm256_cvtepu8_epi32(a1_1_128);
+
+          diff = _mm256_sub_epi32(a1_1, a0_1);  // a[x+1] - a[x]
+          a32 = _mm256_slli_epi32(a0_1, 5);     // a[x] * 32
+          a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+          b = _mm256_mullo_epi32(diff, shift);
+
+          res[1] = _mm256_add_epi32(a32, b);
+          res[1] = _mm256_srli_epi32(res[1], 5);
+          res[1] = _mm256_packus_epi32(
+              res[1],
+              _mm256_castsi128_si256(_mm256_extracti128_si256(res[1], 1)));
+          res[1] = _mm256_packus_epi16(res[1], res[1]);
+
+        } else {
+          res[1] = a_mbase_x;
+        }
+        res1 = _mm_unpacklo_epi64(
+            _mm256_castsi256_si128(res[0]),
+            _mm256_castsi256_si128(res[1]));  // 16 8bit values
+
+        base_inc128 = _mm_setr_epi8(
+            base + j, base + j + 1, base + j + 2, base + j + 3, base + j + 4,
+            base + j + 5, base + j + 6, base + j + 7, base + j + 8,
+            base + j + 9, base + j + 10, base + j + 11, base + j + 12,
+            base + j + 13, base + j + 14, base + j + 15);
+
+        mask128 = _mm_cmpgt_epi8(_mm_subs_epu8(max_base_x128, base_inc128),
+                                 _mm_setzero_si128());
+        res1 =
+            _mm_blendv_epi8(_mm256_castsi256_si128(a_mbase_x), res1, mask128);
+        _mm_storeu_si128((__m128i *)(dst + j), res1);
+      }
+    }
+    x += dx;
+  }
+}
+
+// Directional prediction, zone 1: 0 < angle < 90
+void av1_dr_prediction_z1_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                               const uint8_t *above, const uint8_t *left,
+                               int upsample_above, int dx, int dy) {
+  (void)left;
+  (void)dy;
+  switch (bw) {
+    case 4:
+      dr_prediction_z1_4xN_avx2(bh, dst, stride, above, upsample_above, dx);
+      break;
+    case 8:
+      dr_prediction_z1_8xN_avx2(bh, dst, stride, above, upsample_above, dx);
+      break;
+    case 16:
+      dr_prediction_z1_16xN_avx2(bh, dst, stride, above, upsample_above, dx);
+      break;
+    case 32:
+      dr_prediction_z1_32xN_avx2(bh, dst, stride, above, upsample_above, dx);
+      break;
+    case 64:
+      dr_prediction_z1_64xN_avx2(bh, dst, stride, above, upsample_above, dx);
+      break;
+    default: break;
+  }
+  return;
+}
+
+static uint8_t LoadMaskx[8][16] = {
+  { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+  { 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 },
+  { 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 },
+  { 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 },
+  { 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 },
+  { 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 },
+  { 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 },
+  { 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8 },
+};
+
+static uint8_t EvenOddMaskx4[8][16] = {
+  { 0, 2, 4, 6, 1, 3, 5, 7, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 0, 1, 3, 5, 7, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0 },
+  { 0, 0, 2, 4, 6, 8, 3, 5, 7, 9, 0, 0, 0, 0, 0, 0 },
+  { 0, 0, 0, 3, 5, 7, 9, 4, 6, 8, 10, 0, 0, 0, 0, 0 },
+  { 0, 0, 0, 0, 4, 6, 8, 10, 5, 7, 9, 11, 0, 0, 0, 0 },
+  { 0, 0, 0, 0, 0, 5, 7, 9, 11, 6, 8, 10, 12, 0, 0, 0 },
+  { 0, 0, 0, 0, 0, 0, 6, 8, 10, 12, 7, 9, 11, 13, 0, 0 },
+  { 0, 0, 0, 0, 0, 0, 0, 7, 9, 11, 13, 8, 10, 12, 14, 0 }
+};
+
+static uint8_t EvenOddMaskx[8][16] = {
+  { 0, 2, 4, 6, 8, 10, 12, 14, 1, 3, 5, 7, 0, 0, 0, 0 },
+  { 0, 1, 3, 5, 7, 9, 11, 13, 15, 2, 4, 6, 8, 0, 0, 0 },
+  { 0, 0, 2, 4, 6, 8, 10, 12, 14, 3, 5, 7, 9, 0, 0, 0 },
+  { 0, 0, 0, 3, 5, 7, 9, 11, 13, 15, 4, 6, 8, 10, 0 },
+  { 0, 0, 0, 0, 4, 6, 8, 10, 12, 14, 5, 7, 9, 11, 0, 0 },
+  { 0, 0, 0, 0, 0, 5, 7, 9, 11, 13, 15, 6, 8, 10, 12, 0 },
+  { 0, 0, 0, 0, 0, 0, 6, 8, 10, 12, 14, 7, 9, 11, 13, 0 },
+  { 0, 0, 0, 0, 0, 0, 0, 7, 9, 11, 13, 15, 8, 10, 12, 14 }
+};
+
+static void dr_prediction_z2_Nx4_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *above, const uint8_t *left,
+                                      int upsample_above, int upsample_left,
+                                      int dx, int dy) {
+  const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  const int frac_bits_x = 6 - upsample_above;
+  const int frac_bits_y = 6 - upsample_left;
+
+  // a assert(dx > 0);
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a16, diff;
+  __m128i c3f, min_base_y128;
+
+  a16 = _mm256_set1_epi32(16);
+  c3f = _mm_set1_epi32(0x3f);
+  min_base_y128 = _mm_set1_epi32(min_base_y);
+
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i resx, resy, resxy;
+    __m128i a0_x128, a1_x128;
+    int y = r + 1;
+    int base_x = (-y * dx) >> frac_bits_x;
+    int base_shift = 0;
+    if (base_x < (min_base_x - 1)) {
+      base_shift = (min_base_x - base_x - 1) >> upsample_above;
+    }
+    int base_min_diff =
+        (min_base_x - base_x + upsample_above) >> upsample_above;
+    if (base_min_diff > 4) {
+      base_min_diff = 4;
+    } else {
+      if (base_min_diff < 0) base_min_diff = 0;
+    }
+
+    if (base_shift > 3) {
+      resx = _mm_setzero_si128();
+    } else {
+      a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
+      if (upsample_above) {
+        a0_x128 =
+            _mm_shuffle_epi8(a0_x128, *(__m128i *)EvenOddMaskx4[base_shift]);
+        a1_x128 = _mm_srli_si128(a0_x128, 4);
+
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(
+                _mm_slli_epi32(
+                    _mm_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                   (2 << 6) - y * dx, (3 << 6) - y * dx),
+                    upsample_above),
+                c3f),
+            1));
+      } else {
+        a0_x128 = _mm_shuffle_epi8(a0_x128, *(__m128i *)LoadMaskx[base_shift]);
+        a1_x128 = _mm_srli_si128(a0_x128, 1);
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(_mm_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                         (2 << 6) - y * dx, (3 << 6) - y * dx),
+                          c3f),
+            1));
+      }
+      a0_x = _mm256_cvtepu8_epi32(a0_x128);
+      a1_x = _mm256_cvtepu8_epi32(a1_x128);
+
+      diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resx = _mm256_castsi256_si128(res);
+      resx = _mm_packus_epi32(resx, resx);
+      resx = _mm_packus_epi16(resx, resx);
+    }
+    // y calc
+    if (base_x < min_base_x) {
+      DECLARE_ALIGNED(32, int, base_y_c[4]);
+      __m128i r6, c1234, dy128, y_c128, base_y_c128, mask128;
+      r6 = _mm_set1_epi32(r << 6);
+      dy128 = _mm_set1_epi32(dy);
+      c1234 = _mm_setr_epi32(1, 2, 3, 4);
+      y_c128 = _mm_sub_epi32(r6, _mm_mullo_epi32(c1234, dy128));
+      base_y_c128 = _mm_srai_epi32(y_c128, frac_bits_y);
+      mask128 = _mm_cmpgt_epi32(min_base_y128, base_y_c128);
+      base_y_c128 = _mm_andnot_si128(mask128, base_y_c128);
+      _mm_store_si128((__m128i *)base_y_c, base_y_c128);
+
+      a0_y = _mm256_castsi128_si256(
+          _mm_setr_epi32(left[base_y_c[0]], left[base_y_c[1]],
+                         left[base_y_c[2]], left[base_y_c[3]]));
+      a1_y = _mm256_castsi128_si256(
+          _mm_setr_epi32(left[base_y_c[0] + 1], left[base_y_c[1] + 1],
+                         left[base_y_c[2] + 1], left[base_y_c[3] + 1]));
+
+      if (upsample_left) {
+        shift = _mm256_castsi128_si256(_mm_srli_epi32(
+            _mm_and_si128(_mm_slli_epi32(y_c128, upsample_left), c3f), 1));
+      } else {
+        shift = _mm256_castsi128_si256(
+            _mm_srli_epi32(_mm_and_si128(y_c128, c3f), 1));
+      }
+      diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      resy = _mm256_castsi256_si128(res);
+      resy = _mm_packus_epi32(resy, resy);
+      resy = _mm_packus_epi16(resy, resy);
+    } else {
+      resy = resx;
+    }
+    resxy = _mm_blendv_epi8(resx, resy, *(__m128i *)BaseMask[base_min_diff]);
+    *(uint32_t *)(dst) = _mm_cvtsi128_si32(resxy);
+    dst += stride;
+  }
+}
+
+static void dr_prediction_z2_Nx8_avx2(int N, uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *above, const uint8_t *left,
+                                      int upsample_above, int upsample_left,
+                                      int dx, int dy) {
+  const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  const int frac_bits_x = 6 - upsample_above;
+  const int frac_bits_y = 6 - upsample_left;
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a16, c3f;
+  __m256i diff, min_base_y256;
+  __m128i a0_x128, a1_x128;
+
+  a16 = _mm256_set1_epi32(16);
+  c3f = _mm256_set1_epi32(0x3f);
+  min_base_y256 = _mm256_set1_epi32(min_base_y);
+
+  for (int r = 0; r < N; r++) {
+    __m256i b, res, shift;
+    __m128i resx, resy, resxy;
+
+    int y = r + 1;
+    int base_x = (-y * dx) >> frac_bits_x;
+    int base_shift = 0;
+    if (base_x < (min_base_x - 1)) {
+      base_shift = (min_base_x - base_x - 1) >> upsample_above;
+    }
+    int base_min_diff =
+        (min_base_x - base_x + upsample_above) >> upsample_above;
+    if (base_min_diff > 8) {
+      base_min_diff = 8;
+    } else {
+      if (base_min_diff < 0) base_min_diff = 0;
+    }
+
+    if (base_shift > 7) {
+      resx = _mm_setzero_si128();
+    } else {
+      a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
+      a1_x128 = _mm_loadu_si128((__m128i *)(above + base_x + 1 + base_shift));
+      if (upsample_above) {
+        a0_x128 =
+            _mm_shuffle_epi8(a0_x128, *(__m128i *)EvenOddMaskx[base_shift]);
+        a1_x128 =
+            _mm_shuffle_epi8(a1_x128, *(__m128i *)EvenOddMaskx[base_shift]);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_slli_epi32(
+                    _mm256_setr_epi32(-y * dx, (1 << 6) - y * dx,
+                                      (2 << 6) - y * dx, (3 << 6) - y * dx,
+                                      (4 << 6) - y * dx, (5 << 6) - y * dx,
+                                      (6 << 6) - y * dx, (7 << 6) - y * dx),
+                    upsample_above),
+                c3f),
+            1);
+      } else {
+        a0_x128 = _mm_shuffle_epi8(a0_x128, *(__m128i *)LoadMaskx[base_shift]);
+        a1_x128 = _mm_shuffle_epi8(a1_x128, *(__m128i *)LoadMaskx[base_shift]);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(-y * dx, (1 << 6) - y * dx, (2 << 6) - y * dx,
+                                  (3 << 6) - y * dx, (4 << 6) - y * dx,
+                                  (5 << 6) - y * dx, (6 << 6) - y * dx,
+                                  (7 << 6) - y * dx),
+                c3f),
+            1);
+      }
+      a0_x = _mm256_cvtepu8_epi32(a0_x128);
+      a1_x = _mm256_cvtepu8_epi32(a1_x128);
+
+      diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      res = _mm256_packus_epi32(
+          res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+      resx = _mm_packus_epi16(_mm256_castsi256_si128(res),
+                              _mm256_castsi256_si128(res));
+    }
+    // y calc
+    if (base_x < min_base_x) {
+      DECLARE_ALIGNED(32, int, base_y_c[8]);
+      __m256i r6, c256, dy256, y_c256, base_y_c256, mask256;
+      r6 = _mm256_set1_epi32(r << 6);
+      dy256 = _mm256_set1_epi32(dy);
+      c256 = _mm256_setr_epi32(1, 2, 3, 4, 5, 6, 7, 8);
+      y_c256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+      base_y_c256 = _mm256_srai_epi32(y_c256, frac_bits_y);
+      mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+      base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+      _mm256_store_si256((__m256i *)base_y_c, base_y_c256);
+
+      a0_y = _mm256_setr_epi32(left[base_y_c[0]], left[base_y_c[1]],
+                               left[base_y_c[2]], left[base_y_c[3]],
+                               left[base_y_c[4]], left[base_y_c[5]],
+                               left[base_y_c[6]], left[base_y_c[7]]);
+      a1_y = _mm256_setr_epi32(left[base_y_c[0] + 1], left[base_y_c[1] + 1],
+                               left[base_y_c[2] + 1], left[base_y_c[3] + 1],
+                               left[base_y_c[4] + 1], left[base_y_c[5] + 1],
+                               left[base_y_c[6] + 1], left[base_y_c[7] + 1]);
+
+      if (upsample_left) {
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(_mm256_slli_epi32(y_c256, upsample_left), c3f), 1);
+      } else {
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c256, c3f), 1);
+      }
+      diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+      a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+      a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+      b = _mm256_mullo_epi32(diff, shift);
+      res = _mm256_add_epi32(a32, b);
+      res = _mm256_srli_epi32(res, 5);
+
+      res = _mm256_packus_epi32(
+          res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+      resy = _mm_packus_epi16(_mm256_castsi256_si128(res),
+                              _mm256_castsi256_si128(res));
+    } else {
+      resy = resx;
+    }
+    resxy = _mm_blendv_epi8(resx, resy, *(__m128i *)BaseMask[base_min_diff]);
+    _mm_storel_epi64((__m128i *)(dst), resxy);
+    dst += stride;
+  }
+}
+
+static void dr_prediction_z2_HxW_avx2(int H, int W, uint8_t *dst,
+                                      ptrdiff_t stride, const uint8_t *above,
+                                      const uint8_t *left, int upsample_above,
+                                      int upsample_left, int dx, int dy) {
+  // here upsample_above and upsample_left are 0 by design of
+  // av1_use_intra_edge_upsample
+  const int min_base_x = -1;
+  const int min_base_y = -1;
+  (void)upsample_above;
+  (void)upsample_left;
+  const int frac_bits_x = 6;
+  const int frac_bits_y = 6;
+
+  // pre-filter above pixels
+  // store in temp buffers:
+  //   above[x] * 32 + 16
+  //   above[x+1] - above[x]
+  // final pixels will be caluculated as:
+  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
+  __m256i a0_x, a1_x, a0_y, a1_y, a32, a0_1_x, a1_1_x, a16;
+  __m256i diff, min_base_y256, c3f;
+  __m128i a0_x128, a1_x128, a0_1_x128, a1_1_x128;
+
+  a16 = _mm256_set1_epi32(16);
+  min_base_y256 = _mm256_set1_epi32(min_base_y);
+  c3f = _mm256_set1_epi32(0x3f);
+
+  for (int r = 0; r < H; r++) {
+    __m256i b, res, shift;
+    __m128i resx[2], resy[2];
+    __m128i resxy;
+    for (int j = 0; j < W; j += 16) {
+      int y = r + 1;
+      int base_x = (-y * dx) >> frac_bits_x;
+
+      int base_shift = 0;
+      if ((base_x + j) < (min_base_x - 1)) {
+        base_shift = (min_base_x - (base_x + j) - 1);
+      }
+      int base_min_diff = (min_base_x - base_x - j);
+      if (base_min_diff > 16) {
+        base_min_diff = 16;
+      } else {
+        if (base_min_diff < 0) base_min_diff = 0;
+      }
+      if (base_shift > 7) {
+        resx[0] = _mm_setzero_si128();
+      } else {
+        a0_x128 = _mm_loadu_si128((__m128i *)(above + base_x + base_shift + j));
+        a1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 1 + j));
+        a0_x128 = _mm_shuffle_epi8(a0_x128, *(__m128i *)LoadMaskx[base_shift]);
+        a1_x128 = _mm_shuffle_epi8(a1_x128, *(__m128i *)LoadMaskx[base_shift]);
+
+        a0_x = _mm256_cvtepu8_epi32(a0_x128);
+        a1_x = _mm256_cvtepu8_epi32(a1_x128);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(
+                    ((0 + j) << 6) - y * dx, ((1 + j) << 6) - y * dx,
+                    ((2 + j) << 6) - y * dx, ((3 + j) << 6) - y * dx,
+                    ((4 + j) << 6) - y * dx, ((5 + j) << 6) - y * dx,
+                    ((6 + j) << 6) - y * dx, ((7 + j) << 6) - y * dx),
+                c3f),
+            1);
+
+        diff = _mm256_sub_epi32(a1_x, a0_x);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_x, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        res = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+        resx[0] = _mm_packus_epi16(_mm256_castsi256_si128(res),
+                                   _mm256_castsi256_si128(res));
+      }
+      base_shift = 0;
+      if ((base_x + j + 8) < (min_base_x - 1)) {
+        base_shift = (min_base_x - (base_x + j + 8) - 1);
+      }
+      if (base_shift > 7) {
+        resx[1] = _mm_setzero_si128();
+      } else {
+        a0_1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 8 + j));
+        a1_1_x128 =
+            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 9 + j));
+        a0_1_x128 =
+            _mm_shuffle_epi8(a0_1_x128, *(__m128i *)LoadMaskx[base_shift]);
+        a1_1_x128 =
+            _mm_shuffle_epi8(a1_1_x128, *(__m128i *)LoadMaskx[base_shift]);
+
+        a0_1_x = _mm256_cvtepu8_epi32(a0_1_x128);
+        a1_1_x = _mm256_cvtepu8_epi32(a1_1_x128);
+
+        shift = _mm256_srli_epi32(
+            _mm256_and_si256(
+                _mm256_setr_epi32(
+                    ((8 + j) << 6) - y * dx, ((9 + j) << 6) - y * dx,
+                    ((10 + j) << 6) - y * dx, ((11 + j) << 6) - y * dx,
+                    ((12 + j) << 6) - y * dx, ((13 + j) << 6) - y * dx,
+                    ((14 + j) << 6) - y * dx, ((15 + j) << 6) - y * dx),
+                _mm256_set1_epi32(0x3f)),
+            1);
+
+        diff = _mm256_sub_epi32(a1_1_x, a0_1_x);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_1_x, 5);       // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);         // a[x] * 32 + 16
+        b = _mm256_mullo_epi32(diff, shift);
+
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+        res = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+        resx[1] = _mm_packus_epi16(_mm256_castsi256_si128(res),
+                                   _mm256_castsi256_si128(res));
+      }
+      resx[0] = _mm_unpacklo_epi64(resx[0], resx[1]);
+
+      // y calc
+      if ((base_x < min_base_x)) {
+        DECLARE_ALIGNED(32, int, base_y_c[16]);
+        __m256i r6, c256, dy256, y_c256, y_c_1_256, base_y_c256, mask256;
+        r6 = _mm256_set1_epi32(r << 6);
+        dy256 = _mm256_set1_epi32(dy);
+        c256 = _mm256_setr_epi32(1 + j, 2 + j, 3 + j, 4 + j, 5 + j, 6 + j,
+                                 7 + j, 8 + j);
+        y_c256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+        base_y_c256 = _mm256_srai_epi32(y_c256, frac_bits_y);
+        mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+        base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+        _mm256_store_si256((__m256i *)base_y_c, base_y_c256);
+        c256 = _mm256_setr_epi32(9 + j, 10 + j, 11 + j, 12 + j, 13 + j, 14 + j,
+                                 15 + j, 16 + j);
+        y_c_1_256 = _mm256_sub_epi32(r6, _mm256_mullo_epi32(c256, dy256));
+        base_y_c256 = _mm256_srai_epi32(y_c_1_256, frac_bits_y);
+        mask256 = _mm256_cmpgt_epi32(min_base_y256, base_y_c256);
+        base_y_c256 = _mm256_andnot_si256(mask256, base_y_c256);
+        _mm256_store_si256((__m256i *)(base_y_c + 8), base_y_c256);
+
+        a0_y = _mm256_cvtepu8_epi32(_mm_setr_epi8(
+            left[base_y_c[0]], left[base_y_c[1]], left[base_y_c[2]],
+            left[base_y_c[3]], left[base_y_c[4]], left[base_y_c[5]],
+            left[base_y_c[6]], left[base_y_c[7]], 0, 0, 0, 0, 0, 0, 0, 0));
+        a1_y = _mm256_cvtepu8_epi32(_mm_setr_epi8(
+            left[base_y_c[0] + 1], left[base_y_c[1] + 1], left[base_y_c[2] + 1],
+            left[base_y_c[3] + 1], left[base_y_c[4] + 1], left[base_y_c[5] + 1],
+            left[base_y_c[6] + 1], left[base_y_c[7] + 1], 0, 0, 0, 0, 0, 0, 0,
+            0));
+
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c256, c3f), 1);
+
+        diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        res = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+        resy[0] = _mm_packus_epi16(_mm256_castsi256_si128(res),
+                                   _mm256_castsi256_si128(res));
+
+        a0_y = _mm256_cvtepu8_epi32(_mm_setr_epi8(
+            left[base_y_c[8]], left[base_y_c[9]], left[base_y_c[10]],
+            left[base_y_c[11]], left[base_y_c[12]], left[base_y_c[13]],
+            left[base_y_c[14]], left[base_y_c[15]], 0, 0, 0, 0, 0, 0, 0, 0));
+        a1_y = _mm256_cvtepu8_epi32(
+            _mm_setr_epi8(left[base_y_c[8] + 1], left[base_y_c[9] + 1],
+                          left[base_y_c[10] + 1], left[base_y_c[11] + 1],
+                          left[base_y_c[12] + 1], left[base_y_c[13] + 1],
+                          left[base_y_c[14] + 1], left[base_y_c[15] + 1], 0, 0,
+                          0, 0, 0, 0, 0, 0));
+        shift = _mm256_srli_epi32(_mm256_and_si256(y_c_1_256, c3f), 1);
+
+        diff = _mm256_sub_epi32(a1_y, a0_y);  // a[x+1] - a[x]
+        a32 = _mm256_slli_epi32(a0_y, 5);     // a[x] * 32
+        a32 = _mm256_add_epi32(a32, a16);     // a[x] * 32 + 16
+
+        b = _mm256_mullo_epi32(diff, shift);
+        res = _mm256_add_epi32(a32, b);
+        res = _mm256_srli_epi32(res, 5);
+
+        res = _mm256_packus_epi32(
+            res, _mm256_castsi128_si256(_mm256_extracti128_si256(res, 1)));
+        resy[1] =
+            _mm_packus_epi16(_mm256_castsi256_si128(res),
+                             _mm256_castsi256_si128(res));  // 8 16bit values
+        resy[0] = _mm_unpacklo_epi64(resy[0], resy[1]);
+      } else {
+        resy[0] = resx[0];
+      }
+      resxy = _mm_blendv_epi8(resx[0], resy[0],
+                              *(__m128i *)BaseMask[base_min_diff]);
+      _mm_storeu_si128((__m128i *)(dst + j), resxy);
+    }  // for j
+    dst += stride;
+  }
+}
+
+// Directional prediction, zone 2: 90 < angle < 180
+void av1_dr_prediction_z2_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                               const uint8_t *above, const uint8_t *left,
+                               int upsample_above, int upsample_left, int dx,
+                               int dy) {
+  assert(dx > 0);
+  assert(dy > 0);
+  switch (bw) {
+    case 4:
+      dr_prediction_z2_Nx4_avx2(bh, dst, stride, above, left, upsample_above,
+                                upsample_left, dx, dy);
+      break;
+    case 8:
+      dr_prediction_z2_Nx8_avx2(bh, dst, stride, above, left, upsample_above,
+                                upsample_left, dx, dy);
+
+      break;
+    default:
+      dr_prediction_z2_HxW_avx2(bh, bw, dst, stride, above, left,
+                                upsample_above, upsample_left, dx, dy);
+      break;
+  }
+  return;
+}
+
+// z3 functions
+static INLINE void transpose4x16_sse2(__m128i *x, __m128i *d) {
+  __m128i w0, w1, w2, w3, ww0, ww1, ww2, ww3;
+  w0 = _mm_unpacklo_epi8(x[0], x[1]);
+  w1 = _mm_unpacklo_epi8(x[2], x[3]);
+  w2 = _mm_unpackhi_epi8(x[0], x[1]);
+  w3 = _mm_unpackhi_epi8(x[2], x[3]);
+
+  ww0 = _mm_unpacklo_epi16(w0, w1);
+  ww1 = _mm_unpacklo_epi16(w2, w3);
+  ww2 = _mm_unpackhi_epi16(w0, w1);
+  ww3 = _mm_unpackhi_epi16(w2, w3);
+
+  w0 = _mm_unpacklo_epi32(ww0, ww1);
+  w2 = _mm_unpacklo_epi32(ww2, ww3);
+  w1 = _mm_unpackhi_epi32(ww0, ww1);
+  w3 = _mm_unpackhi_epi32(ww2, ww3);
+
+  d[0] = _mm_unpacklo_epi64(w0, w2);
+  d[1] = _mm_unpackhi_epi64(w0, w2);
+  d[2] = _mm_unpacklo_epi64(w1, w3);
+  d[3] = _mm_unpackhi_epi64(w1, w3);
+
+  d[4] = _mm_srli_si128(d[0], 8);
+  d[5] = _mm_srli_si128(d[1], 8);
+  d[6] = _mm_srli_si128(d[2], 8);
+  d[7] = _mm_srli_si128(d[3], 8);
+
+  d[8] = _mm_srli_si128(d[0], 4);
+  d[9] = _mm_srli_si128(d[1], 4);
+  d[10] = _mm_srli_si128(d[2], 4);
+  d[11] = _mm_srli_si128(d[3], 4);
+
+  d[12] = _mm_srli_si128(d[0], 12);
+  d[13] = _mm_srli_si128(d[1], 12);
+  d[14] = _mm_srli_si128(d[2], 12);
+  d[15] = _mm_srli_si128(d[3], 12);
+}
+
+static INLINE void transpose16x32_avx2(__m256i *x, __m256i *d) {
+  __m256i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
+  __m256i w10, w11, w12, w13, w14, w15;
+
+  w0 = _mm256_unpacklo_epi8(x[0], x[1]);
+  w1 = _mm256_unpacklo_epi8(x[2], x[3]);
+  w2 = _mm256_unpacklo_epi8(x[4], x[5]);
+  w3 = _mm256_unpacklo_epi8(x[6], x[7]);
+
+  w8 = _mm256_unpacklo_epi8(x[8], x[9]);
+  w9 = _mm256_unpacklo_epi8(x[10], x[11]);
+  w10 = _mm256_unpacklo_epi8(x[12], x[13]);
+  w11 = _mm256_unpacklo_epi8(x[14], x[15]);
+
+  w4 = _mm256_unpacklo_epi16(w0, w1);
+  w5 = _mm256_unpacklo_epi16(w2, w3);
+  w12 = _mm256_unpacklo_epi16(w8, w9);
+  w13 = _mm256_unpacklo_epi16(w10, w11);
+
+  w6 = _mm256_unpacklo_epi32(w4, w5);
+  w7 = _mm256_unpackhi_epi32(w4, w5);
+  w14 = _mm256_unpacklo_epi32(w12, w13);
+  w15 = _mm256_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  d[0] = _mm256_unpacklo_epi64(w6, w14);
+  d[1] = _mm256_unpackhi_epi64(w6, w14);
+  d[2] = _mm256_unpacklo_epi64(w7, w15);
+  d[3] = _mm256_unpackhi_epi64(w7, w15);
+
+  w4 = _mm256_unpackhi_epi16(w0, w1);
+  w5 = _mm256_unpackhi_epi16(w2, w3);
+  w12 = _mm256_unpackhi_epi16(w8, w9);
+  w13 = _mm256_unpackhi_epi16(w10, w11);
+
+  w6 = _mm256_unpacklo_epi32(w4, w5);
+  w7 = _mm256_unpackhi_epi32(w4, w5);
+  w14 = _mm256_unpacklo_epi32(w12, w13);
+  w15 = _mm256_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  d[4] = _mm256_unpacklo_epi64(w6, w14);
+  d[5] = _mm256_unpackhi_epi64(w6, w14);
+  d[6] = _mm256_unpacklo_epi64(w7, w15);
+  d[7] = _mm256_unpackhi_epi64(w7, w15);
+
+  // upper half
+  w0 = _mm256_unpackhi_epi8(x[0], x[1]);
+  w1 = _mm256_unpackhi_epi8(x[2], x[3]);
+  w2 = _mm256_unpackhi_epi8(x[4], x[5]);
+  w3 = _mm256_unpackhi_epi8(x[6], x[7]);
+
+  w8 = _mm256_unpackhi_epi8(x[8], x[9]);
+  w9 = _mm256_unpackhi_epi8(x[10], x[11]);
+  w10 = _mm256_unpackhi_epi8(x[12], x[13]);
+  w11 = _mm256_unpackhi_epi8(x[14], x[15]);
+
+  w4 = _mm256_unpacklo_epi16(w0, w1);
+  w5 = _mm256_unpacklo_epi16(w2, w3);
+  w12 = _mm256_unpacklo_epi16(w8, w9);
+  w13 = _mm256_unpacklo_epi16(w10, w11);
+
+  w6 = _mm256_unpacklo_epi32(w4, w5);
+  w7 = _mm256_unpackhi_epi32(w4, w5);
+  w14 = _mm256_unpacklo_epi32(w12, w13);
+  w15 = _mm256_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  d[8] = _mm256_unpacklo_epi64(w6, w14);
+  d[9] = _mm256_unpackhi_epi64(w6, w14);
+  d[10] = _mm256_unpacklo_epi64(w7, w15);
+  d[11] = _mm256_unpackhi_epi64(w7, w15);
+
+  w4 = _mm256_unpackhi_epi16(w0, w1);
+  w5 = _mm256_unpackhi_epi16(w2, w3);
+  w12 = _mm256_unpackhi_epi16(w8, w9);
+  w13 = _mm256_unpackhi_epi16(w10, w11);
+
+  w6 = _mm256_unpacklo_epi32(w4, w5);
+  w7 = _mm256_unpackhi_epi32(w4, w5);
+  w14 = _mm256_unpacklo_epi32(w12, w13);
+  w15 = _mm256_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  d[12] = _mm256_unpacklo_epi64(w6, w14);
+  d[13] = _mm256_unpackhi_epi64(w6, w14);
+  d[14] = _mm256_unpacklo_epi64(w7, w15);
+  d[15] = _mm256_unpackhi_epi64(w7, w15);
+}
+
+static INLINE void transpose16x16_sse2(__m128i *x, __m128i *d) {
+  __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
+  __m128i w10, w11, w12, w13, w14, w15;
+
+  w0 = _mm_unpacklo_epi8(x[0], x[1]);
+  w1 = _mm_unpacklo_epi8(x[2], x[3]);
+  w2 = _mm_unpacklo_epi8(x[4], x[5]);
+  w3 = _mm_unpacklo_epi8(x[6], x[7]);
+
+  w8 = _mm_unpacklo_epi8(x[8], x[9]);
+  w9 = _mm_unpacklo_epi8(x[10], x[11]);
+  w10 = _mm_unpacklo_epi8(x[12], x[13]);
+  w11 = _mm_unpacklo_epi8(x[14], x[15]);
+
+  w4 = _mm_unpacklo_epi16(w0, w1);
+  w5 = _mm_unpacklo_epi16(w2, w3);
+  w12 = _mm_unpacklo_epi16(w8, w9);
+  w13 = _mm_unpacklo_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  d[0] = _mm_unpacklo_epi64(w6, w14);
+  d[1] = _mm_unpackhi_epi64(w6, w14);
+  d[2] = _mm_unpacklo_epi64(w7, w15);
+  d[3] = _mm_unpackhi_epi64(w7, w15);
+
+  w4 = _mm_unpackhi_epi16(w0, w1);
+  w5 = _mm_unpackhi_epi16(w2, w3);
+  w12 = _mm_unpackhi_epi16(w8, w9);
+  w13 = _mm_unpackhi_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  d[4] = _mm_unpacklo_epi64(w6, w14);
+  d[5] = _mm_unpackhi_epi64(w6, w14);
+  d[6] = _mm_unpacklo_epi64(w7, w15);
+  d[7] = _mm_unpackhi_epi64(w7, w15);
+
+  // upper half
+  w0 = _mm_unpackhi_epi8(x[0], x[1]);
+  w1 = _mm_unpackhi_epi8(x[2], x[3]);
+  w2 = _mm_unpackhi_epi8(x[4], x[5]);
+  w3 = _mm_unpackhi_epi8(x[6], x[7]);
+
+  w8 = _mm_unpackhi_epi8(x[8], x[9]);
+  w9 = _mm_unpackhi_epi8(x[10], x[11]);
+  w10 = _mm_unpackhi_epi8(x[12], x[13]);
+  w11 = _mm_unpackhi_epi8(x[14], x[15]);
+
+  w4 = _mm_unpacklo_epi16(w0, w1);
+  w5 = _mm_unpacklo_epi16(w2, w3);
+  w12 = _mm_unpacklo_epi16(w8, w9);
+  w13 = _mm_unpacklo_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  d[8] = _mm_unpacklo_epi64(w6, w14);
+  d[9] = _mm_unpackhi_epi64(w6, w14);
+  d[10] = _mm_unpacklo_epi64(w7, w15);
+  d[11] = _mm_unpackhi_epi64(w7, w15);
+
+  w4 = _mm_unpackhi_epi16(w0, w1);
+  w5 = _mm_unpackhi_epi16(w2, w3);
+  w12 = _mm_unpackhi_epi16(w8, w9);
+  w13 = _mm_unpackhi_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  d[12] = _mm_unpacklo_epi64(w6, w14);
+  d[13] = _mm_unpackhi_epi64(w6, w14);
+  d[14] = _mm_unpacklo_epi64(w7, w15);
+  d[15] = _mm_unpackhi_epi64(w7, w15);
+}
+
+static void transpose_TX_8X8(const uint8_t *src, ptrdiff_t pitchSrc,
+                             uint8_t *dst, ptrdiff_t pitchDst) {
+  __m128i r0, r1, r2, r3, r4, r5, r6, r7;
+  __m128i d0d1, d2d3, d4d5, d6d7;
+  r0 = _mm_loadl_epi64((__m128i *)(src + 0 * pitchSrc));
+  r1 = _mm_loadl_epi64((__m128i *)(src + 1 * pitchSrc));
+  r2 = _mm_loadl_epi64((__m128i *)(src + 2 * pitchSrc));
+  r3 = _mm_loadl_epi64((__m128i *)(src + 3 * pitchSrc));
+  r4 = _mm_loadl_epi64((__m128i *)(src + 4 * pitchSrc));
+  r5 = _mm_loadl_epi64((__m128i *)(src + 5 * pitchSrc));
+  r6 = _mm_loadl_epi64((__m128i *)(src + 6 * pitchSrc));
+  r7 = _mm_loadl_epi64((__m128i *)(src + 7 * pitchSrc));
+
+  transpose8x8_sse2(&r0, &r1, &r2, &r3, &r4, &r5, &r6, &r7, &d0d1, &d2d3, &d4d5,
+                    &d6d7);
+
+  _mm_storel_epi64((__m128i *)(dst + 0 * pitchDst), d0d1);
+  _mm_storel_epi64((__m128i *)(dst + 1 * pitchDst), _mm_srli_si128(d0d1, 8));
+  _mm_storel_epi64((__m128i *)(dst + 2 * pitchDst), d2d3);
+  _mm_storel_epi64((__m128i *)(dst + 3 * pitchDst), _mm_srli_si128(d2d3, 8));
+  _mm_storel_epi64((__m128i *)(dst + 4 * pitchDst), d4d5);
+  _mm_storel_epi64((__m128i *)(dst + 5 * pitchDst), _mm_srli_si128(d4d5, 8));
+  _mm_storel_epi64((__m128i *)(dst + 6 * pitchDst), d6d7);
+  _mm_storel_epi64((__m128i *)(dst + 7 * pitchDst), _mm_srli_si128(d6d7, 8));
+}
+
+static void transpose(const uint8_t *src, ptrdiff_t pitchSrc, uint8_t *dst,
+                      ptrdiff_t pitchDst, int width, int height) {
+  for (int j = 0; j < height; j += 8)
+    for (int i = 0; i < width; i += 8)
+      transpose_TX_8X8(src + i * pitchSrc + j, pitchSrc, dst + j * pitchDst + i,
+                       pitchDst);
+}
+
+static void dr_prediction_z3_4x4_avx2(uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *left, int upsample_left,
+                                      int dy) {
+  __m128i dstvec[4], d[4];
+
+  dr_prediction_z1_4xN_internal_avx2(4, dstvec, left, upsample_left, dy);
+  transpose4x8_8x4_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
+                            &d[0], &d[1], &d[2], &d[3]);
+
+  *(uint32_t *)(dst + stride * 0) = _mm_cvtsi128_si32(d[0]);
+  *(uint32_t *)(dst + stride * 1) = _mm_cvtsi128_si32(d[1]);
+  *(uint32_t *)(dst + stride * 2) = _mm_cvtsi128_si32(d[2]);
+  *(uint32_t *)(dst + stride * 3) = _mm_cvtsi128_si32(d[3]);
+  return;
+}
+
+static void dr_prediction_z3_8x8_avx2(uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *left, int upsample_left,
+                                      int dy) {
+  __m128i dstvec[8], d[8];
+
+  dr_prediction_z1_8xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  transpose8x8_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4],
+                    &dstvec[5], &dstvec[6], &dstvec[7], &d[0], &d[1], &d[2],
+                    &d[3]);
+
+  _mm_storel_epi64((__m128i *)(dst + 0 * stride), d[0]);
+  _mm_storel_epi64((__m128i *)(dst + 1 * stride), _mm_srli_si128(d[0], 8));
+  _mm_storel_epi64((__m128i *)(dst + 2 * stride), d[1]);
+  _mm_storel_epi64((__m128i *)(dst + 3 * stride), _mm_srli_si128(d[1], 8));
+  _mm_storel_epi64((__m128i *)(dst + 4 * stride), d[2]);
+  _mm_storel_epi64((__m128i *)(dst + 5 * stride), _mm_srli_si128(d[2], 8));
+  _mm_storel_epi64((__m128i *)(dst + 6 * stride), d[3]);
+  _mm_storel_epi64((__m128i *)(dst + 7 * stride), _mm_srli_si128(d[3], 8));
+}
+
+static void dr_prediction_z3_4x8_avx2(uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *left, int upsample_left,
+                                      int dy) {
+  __m128i dstvec[4], d[8];
+
+  dr_prediction_z1_8xN_internal_avx2(4, dstvec, left, upsample_left, dy);
+  transpose4x8_8x4_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &d[0],
+                        &d[1], &d[2], &d[3], &d[4], &d[5], &d[6], &d[7]);
+  for (int i = 0; i < 8; i++) {
+    *(uint32_t *)(dst + stride * i) = _mm_cvtsi128_si32(d[i]);
+  }
+}
+
+static void dr_prediction_z3_8x4_avx2(uint8_t *dst, ptrdiff_t stride,
+                                      const uint8_t *left, int upsample_left,
+                                      int dy) {
+  __m128i dstvec[8], d[4];
+
+  dr_prediction_z1_4xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  transpose8x8_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
+                        &dstvec[4], &dstvec[5], &dstvec[6], &dstvec[7], &d[0],
+                        &d[1], &d[2], &d[3]);
+  _mm_storel_epi64((__m128i *)(dst + 0 * stride), d[0]);
+  _mm_storel_epi64((__m128i *)(dst + 1 * stride), d[1]);
+  _mm_storel_epi64((__m128i *)(dst + 2 * stride), d[2]);
+  _mm_storel_epi64((__m128i *)(dst + 3 * stride), d[3]);
+}
+
+static void dr_prediction_z3_8x16_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m128i dstvec[8], d[8];
+
+  dr_prediction_z1_16xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  transpose8x16_16x8_sse2(dstvec, dstvec + 1, dstvec + 2, dstvec + 3,
+                          dstvec + 4, dstvec + 5, dstvec + 6, dstvec + 7, d,
+                          d + 1, d + 2, d + 3, d + 4, d + 5, d + 6, d + 7);
+  for (int i = 0; i < 8; i++) {
+    _mm_storel_epi64((__m128i *)(dst + i * stride), d[i]);
+    _mm_storel_epi64((__m128i *)(dst + (i + 8) * stride),
+                     _mm_srli_si128(d[i], 8));
+  }
+}
+
+static void dr_prediction_z3_16x8_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m128i dstvec[16], d[16];
+
+  dr_prediction_z1_8xN_internal_avx2(16, dstvec, left, upsample_left, dy);
+  transpose16x8_8x16_sse2(
+      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
+      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
+      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
+      &d[3], &d[4], &d[5], &d[6], &d[7]);
+
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void dr_prediction_z3_4x16_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m128i dstvec[4], d[16];
+
+  dr_prediction_z1_16xN_internal_avx2(4, dstvec, left, upsample_left, dy);
+  transpose4x16_sse2(dstvec, d);
+  for (int i = 0; i < 16; i++) {
+    *(uint32_t *)(dst + stride * i) = _mm_cvtsi128_si32(d[i]);
+  }
+}
+
+static void dr_prediction_z3_16x4_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m128i dstvec[16], d[8];
+
+  dr_prediction_z1_4xN_internal_avx2(16, dstvec, left, upsample_left, dy);
+  for (int i = 4; i < 8; i++) {
+    d[i] = _mm_setzero_si128();
+  }
+  transpose16x8_8x16_sse2(
+      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
+      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
+      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
+      &d[3], &d[4], &d[5], &d[6], &d[7]);
+
+  for (int i = 0; i < 4; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void dr_prediction_z3_8x32_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m256i dstvec[16], d[16];
+
+  dr_prediction_z1_32xN_internal_avx2(8, dstvec, left, upsample_left, dy);
+  for (int i = 8; i < 16; i++) {
+    dstvec[i] = _mm256_setzero_si256();
+  }
+  transpose16x32_avx2(dstvec, d);
+
+  for (int i = 0; i < 16; i++) {
+    _mm_storel_epi64((__m128i *)(dst + i * stride),
+                     _mm256_castsi256_si128(d[i]));
+  }
+  for (int i = 0; i < 16; i++) {
+    _mm_storel_epi64((__m128i *)(dst + (i + 16) * stride),
+                     _mm256_extracti128_si256(d[i], 1));
+  }
+}
+
+static void dr_prediction_z3_32x8_avx2(uint8_t *dst, ptrdiff_t stride,
+                                       const uint8_t *left, int upsample_left,
+                                       int dy) {
+  __m128i dstvec[32], d[16];
+
+  dr_prediction_z1_8xN_internal_avx2(32, dstvec, left, upsample_left, dy);
+
+  transpose16x8_8x16_sse2(
+      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
+      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
+      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
+      &d[3], &d[4], &d[5], &d[6], &d[7]);
+  transpose16x8_8x16_sse2(
+      &dstvec[0 + 16], &dstvec[1 + 16], &dstvec[2 + 16], &dstvec[3 + 16],
+      &dstvec[4 + 16], &dstvec[5 + 16], &dstvec[6 + 16], &dstvec[7 + 16],
+      &dstvec[8 + 16], &dstvec[9 + 16], &dstvec[10 + 16], &dstvec[11 + 16],
+      &dstvec[12 + 16], &dstvec[13 + 16], &dstvec[14 + 16], &dstvec[15 + 16],
+      &d[0 + 8], &d[1 + 8], &d[2 + 8], &d[3 + 8], &d[4 + 8], &d[5 + 8],
+      &d[6 + 8], &d[7 + 8]);
+
+  for (int i = 0; i < 8; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+    _mm_storeu_si128((__m128i *)(dst + i * stride + 16), d[i + 8]);
+  }
+}
+
+static void dr_prediction_z3_16x16_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  __m128i dstvec[16], d[16];
+
+  dr_prediction_z1_16xN_internal_avx2(16, dstvec, left, upsample_left, dy);
+  transpose16x16_sse2(dstvec, d);
+
+  for (int i = 0; i < 16; i++) {
+    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
+  }
+}
+
+static void dr_prediction_z3_32x32_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  __m256i dstvec[32], d[32];
+
+  dr_prediction_z1_32xN_internal_avx2(32, dstvec, left, upsample_left, dy);
+  transpose16x32_avx2(dstvec, d);
+  transpose16x32_avx2(dstvec + 16, d + 16);
+  for (int j = 0; j < 16; j++) {
+    _mm_storeu_si128((__m128i *)(dst + j * stride),
+                     _mm256_castsi256_si128(d[j]));
+    _mm_storeu_si128((__m128i *)(dst + j * stride + 16),
+                     _mm256_castsi256_si128(d[j + 16]));
+  }
+  for (int j = 0; j < 16; j++) {
+    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride),
+                     _mm256_extracti128_si256(d[j], 1));
+    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride + 16),
+                     _mm256_extracti128_si256(d[j + 16], 1));
+  }
+}
+
+static void dr_prediction_z3_64x64_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  DECLARE_ALIGNED(16, uint8_t, dstT[64 * 64]);
+  dr_prediction_z1_64xN_avx2(64, dstT, 64, left, upsample_left, dy);
+  transpose(dstT, 64, dst, stride, 64, 64);
+}
+
+static void dr_prediction_z3_16x32_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  __m256i dstvec[16], d[16];
+
+  dr_prediction_z1_32xN_internal_avx2(16, dstvec, left, upsample_left, dy);
+  transpose16x32_avx2(dstvec, d);
+  // store
+  for (int j = 0; j < 16; j++) {
+    _mm_storeu_si128((__m128i *)(dst + j * stride),
+                     _mm256_castsi256_si128(d[j]));
+    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride),
+                     _mm256_extracti128_si256(d[j], 1));
+  }
+}
+
+static void dr_prediction_z3_32x16_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  __m128i dstvec[32], d[16];
+
+  dr_prediction_z1_16xN_internal_avx2(32, dstvec, left, upsample_left, dy);
+  for (int i = 0; i < 32; i += 16) {
+    transpose16x16_sse2((dstvec + i), d);
+    for (int j = 0; j < 16; j++) {
+      _mm_storeu_si128((__m128i *)(dst + j * stride + i), d[j]);
+    }
+  }
+}
+
+static void dr_prediction_z3_32x64_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  uint8_t dstT[64 * 32];
+  dr_prediction_z1_64xN_avx2(32, dstT, 64, left, upsample_left, dy);
+  transpose(dstT, 64, dst, stride, 32, 64);
+}
+
+static void dr_prediction_z3_64x32_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  uint8_t dstT[32 * 64];
+  dr_prediction_z1_32xN_avx2(64, dstT, 32, left, upsample_left, dy);
+  transpose(dstT, 32, dst, stride, 64, 32);
+  return;
+}
+
+static void dr_prediction_z3_16x64_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  uint8_t dstT[64 * 16];
+  dr_prediction_z1_64xN_avx2(16, dstT, 64, left, upsample_left, dy);
+  transpose(dstT, 64, dst, stride, 16, 64);
+}
+
+static void dr_prediction_z3_64x16_avx2(uint8_t *dst, ptrdiff_t stride,
+                                        const uint8_t *left, int upsample_left,
+                                        int dy) {
+  __m128i dstvec[64], d[16];
+
+  dr_prediction_z1_16xN_internal_avx2(64, dstvec, left, upsample_left, dy);
+  for (int i = 0; i < 64; i += 16) {
+    transpose16x16_sse2((dstvec + i), d);
+    for (int j = 0; j < 16; j++) {
+      _mm_storeu_si128((__m128i *)(dst + j * stride + i), d[j]);
+    }
+  }
+}
+
+void av1_dr_prediction_z3_avx2(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                               const uint8_t *above, const uint8_t *left,
+                               int upsample_left, int dx, int dy) {
+  (void)above;
+  (void)dx;
+  assert(dx == 1);
+  assert(dy > 0);
+
+  if (bw == bh) {
+    switch (bw) {
+      case 4:
+        dr_prediction_z3_4x4_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 8:
+        dr_prediction_z3_8x8_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 16:
+        dr_prediction_z3_16x16_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 32:
+        dr_prediction_z3_32x32_avx2(dst, stride, left, upsample_left, dy);
+        break;
+      case 64:
+        dr_prediction_z3_64x64_avx2(dst, stride, left, upsample_left, dy);
+        break;
+    }
+  } else {
+    if (bw < bh) {
+      if (bw + bw == bh) {
+        switch (bw) {
+          case 4:
+            dr_prediction_z3_4x8_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 8:
+            dr_prediction_z3_8x16_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 16:
+            dr_prediction_z3_16x32_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 32:
+            dr_prediction_z3_32x64_avx2(dst, stride, left, upsample_left, dy);
+            break;
+        }
+      } else {
+        switch (bw) {
+          case 4:
+            dr_prediction_z3_4x16_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 8:
+            dr_prediction_z3_8x32_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 16:
+            dr_prediction_z3_16x64_avx2(dst, stride, left, upsample_left, dy);
+            break;
+        }
+      }
+    } else {
+      if (bh + bh == bw) {
+        switch (bh) {
+          case 4:
+            dr_prediction_z3_8x4_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 8:
+            dr_prediction_z3_16x8_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 16:
+            dr_prediction_z3_32x16_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 32:
+            dr_prediction_z3_64x32_avx2(dst, stride, left, upsample_left, dy);
+            break;
+        }
+      } else {
+        switch (bh) {
+          case 4:
+            dr_prediction_z3_16x4_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 8:
+            dr_prediction_z3_32x8_avx2(dst, stride, left, upsample_left, dy);
+            break;
+          case 16:
+            dr_prediction_z3_64x16_avx2(dst, stride, left, upsample_left, dy);
+            break;
+        }
+      }
+    }
+  }
+  return;
+}
diff --git a/third_party/aom/aom_dsp/x86/jnt_sad_ssse3.c b/third_party/aom/aom_dsp/x86/jnt_sad_ssse3.c
index c3c88245a4..2e3e2be105 100644
--- a/third_party/aom/aom_dsp/x86/jnt_sad_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/jnt_sad_ssse3.c
@@ -192,47 +192,47 @@ unsigned int aom_sad128xh_sse2(const uint8_t *a, int a_stride, const uint8_t *b,
   return res;
 }
 
-#define jnt_sadMxN_sse2(m, n)                                                 \
-  unsigned int aom_jnt_sad##m##x##n##_avg_ssse3(                              \
+#define dist_wtd_sadMxN_sse2(m, n)                                            \
+  unsigned int aom_dist_wtd_sad##m##x##n##_avg_ssse3(                         \
       const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {         \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
     uint8_t comp_pred[m * n];                                                 \
-    aom_jnt_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride,      \
-                          jcp_param);                                         \
+    aom_dist_wtd_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride, \
+                               jcp_param);                                    \
     return aom_sad##m##xh_sse2(src, src_stride, comp_pred, m, m, n);          \
   }
 
-#define jnt_sadMxN_avx2(m, n)                                                 \
-  unsigned int aom_jnt_sad##m##x##n##_avg_avx2(                               \
+#define dist_wtd_sadMxN_avx2(m, n)                                            \
+  unsigned int aom_dist_wtd_sad##m##x##n##_avg_avx2(                          \
       const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {         \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) {    \
     uint8_t comp_pred[m * n];                                                 \
-    aom_jnt_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride,      \
-                          jcp_param);                                         \
+    aom_dist_wtd_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride, \
+                               jcp_param);                                    \
     return aom_sad##m##xh_avx2(src, src_stride, comp_pred, m, m, n);          \
   }
 
 /* clang-format off */
-jnt_sadMxN_sse2(128, 128)
-jnt_sadMxN_sse2(128, 64)
-jnt_sadMxN_sse2(64, 128)
-jnt_sadMxN_sse2(64, 64)
-jnt_sadMxN_sse2(64, 32)
-jnt_sadMxN_sse2(32, 64)
-jnt_sadMxN_sse2(32, 32)
-jnt_sadMxN_sse2(32, 16)
-jnt_sadMxN_sse2(16, 32)
-jnt_sadMxN_sse2(16, 16)
-jnt_sadMxN_sse2(16, 8)
-jnt_sadMxN_sse2(8, 16)
-jnt_sadMxN_sse2(8, 8)
-jnt_sadMxN_sse2(8, 4)
-jnt_sadMxN_sse2(4, 8)
-jnt_sadMxN_sse2(4, 4)
-jnt_sadMxN_sse2(4, 16)
-jnt_sadMxN_sse2(16, 4)
-jnt_sadMxN_sse2(8, 32)
-jnt_sadMxN_sse2(32, 8)
-jnt_sadMxN_sse2(16, 64)
-jnt_sadMxN_sse2(64, 16)
+dist_wtd_sadMxN_sse2(128, 128)
+dist_wtd_sadMxN_sse2(128, 64)
+dist_wtd_sadMxN_sse2(64, 128)
+dist_wtd_sadMxN_sse2(64, 64)
+dist_wtd_sadMxN_sse2(64, 32)
+dist_wtd_sadMxN_sse2(32, 64)
+dist_wtd_sadMxN_sse2(32, 32)
+dist_wtd_sadMxN_sse2(32, 16)
+dist_wtd_sadMxN_sse2(16, 32)
+dist_wtd_sadMxN_sse2(16, 16)
+dist_wtd_sadMxN_sse2(16, 8)
+dist_wtd_sadMxN_sse2(8, 16)
+dist_wtd_sadMxN_sse2(8, 8)
+dist_wtd_sadMxN_sse2(8, 4)
+dist_wtd_sadMxN_sse2(4, 8)
+dist_wtd_sadMxN_sse2(4, 4)
+dist_wtd_sadMxN_sse2(4, 16)
+dist_wtd_sadMxN_sse2(16, 4)
+dist_wtd_sadMxN_sse2(8, 32)
+dist_wtd_sadMxN_sse2(32, 8)
+dist_wtd_sadMxN_sse2(16, 64)
+dist_wtd_sadMxN_sse2(64, 16)
     /* clang-format on */
diff --git a/third_party/aom/aom_dsp/x86/jnt_variance_ssse3.c b/third_party/aom/aom_dsp/x86/jnt_variance_ssse3.c
index f9a41a210b..c8b02f5560 100644
--- a/third_party/aom/aom_dsp/x86/jnt_variance_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/jnt_variance_ssse3.c
@@ -29,7 +29,7 @@ void aom_var_filter_block2d_bil_second_pass_ssse3(
     unsigned int pixel_step, unsigned int output_height,
     unsigned int output_width, const uint8_t *filter);
 
-static INLINE void compute_jnt_comp_avg(__m128i *p0, __m128i *p1,
+static INLINE void compute_dist_wtd_avg(__m128i *p0, __m128i *p1,
                                         const __m128i *w, const __m128i *r,
                                         void *const result) {
   __m128i p_lo = _mm_unpacklo_epi8(*p0, *p1);
@@ -45,10 +45,10 @@ static INLINE void compute_jnt_comp_avg(__m128i *p0, __m128i *p1,
   xx_storeu_128(result, _mm_packus_epi16(shift_lo, shift_hi));
 }
 
-void aom_jnt_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
-                                 int width, int height, const uint8_t *ref,
-                                 int ref_stride,
-                                 const JNT_COMP_PARAMS *jcp_param) {
+void aom_dist_wtd_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
+                                      int width, int height, const uint8_t *ref,
+                                      int ref_stride,
+                                      const DIST_WTD_COMP_PARAMS *jcp_param) {
   int i;
   const uint8_t w0 = (uint8_t)jcp_param->fwd_offset;
   const uint8_t w1 = (uint8_t)jcp_param->bck_offset;
@@ -67,7 +67,7 @@ void aom_jnt_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
         __m128i p0 = xx_loadu_128(ref);
         __m128i p1 = xx_loadu_128(pred);
 
-        compute_jnt_comp_avg(&p0, &p1, &w, &r, comp_pred);
+        compute_dist_wtd_avg(&p0, &p1, &w, &r, comp_pred);
 
         comp_pred += 16;
         pred += 16;
@@ -85,7 +85,7 @@ void aom_jnt_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
       __m128i p0 = _mm_unpacklo_epi64(p0_0, p0_1);
       __m128i p1 = xx_loadu_128(pred);
 
-      compute_jnt_comp_avg(&p0, &p1, &w, &r, comp_pred);
+      compute_dist_wtd_avg(&p0, &p1, &w, &r, comp_pred);
 
       comp_pred += 16;
       pred += 16;
@@ -107,7 +107,7 @@ void aom_jnt_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
                         row3[0], row3[1], row3[2], row3[3]);
       __m128i p1 = xx_loadu_128(pred);
 
-      compute_jnt_comp_avg(&p0, &p1, &w, &r, comp_pred);
+      compute_dist_wtd_avg(&p0, &p1, &w, &r, comp_pred);
 
       comp_pred += 16;
       pred += 16;
@@ -116,11 +116,11 @@ void aom_jnt_comp_avg_pred_ssse3(uint8_t *comp_pred, const uint8_t *pred,
   }
 }
 
-void aom_jnt_comp_avg_upsampled_pred_ssse3(
+void aom_dist_wtd_comp_avg_upsampled_pred_ssse3(
     MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
-    int ref_stride, const JNT_COMP_PARAMS *jcp_param, int subpel_search) {
+    int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search) {
   int n;
   int i;
   aom_upsampled_pred(xd, cm, mi_row, mi_col, mv, comp_pred, width, height,
@@ -141,52 +141,52 @@ void aom_jnt_comp_avg_upsampled_pred_ssse3(
     __m128i p0 = xx_loadu_128(comp_pred);
     __m128i p1 = xx_loadu_128(pred);
 
-    compute_jnt_comp_avg(&p0, &p1, &w, &r, comp_pred);
+    compute_dist_wtd_avg(&p0, &p1, &w, &r, comp_pred);
 
     comp_pred += 16;
     pred += 16;
   }
 }
 
-#define JNT_SUBPIX_AVG_VAR(W, H)                                         \
-  uint32_t aom_jnt_sub_pixel_avg_variance##W##x##H##_ssse3(              \
-      const uint8_t *a, int a_stride, int xoffset, int yoffset,          \
-      const uint8_t *b, int b_stride, uint32_t *sse,                     \
-      const uint8_t *second_pred, const JNT_COMP_PARAMS *jcp_param) {    \
-    uint16_t fdata3[(H + 1) * W];                                        \
-    uint8_t temp2[H * W];                                                \
-    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                          \
-                                                                         \
-    aom_var_filter_block2d_bil_first_pass_ssse3(                         \
-        a, fdata3, a_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]); \
-    aom_var_filter_block2d_bil_second_pass_ssse3(                        \
-        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);        \
-                                                                         \
-    aom_jnt_comp_avg_pred_ssse3(temp3, second_pred, W, H, temp2, W,      \
-                                jcp_param);                              \
-                                                                         \
-    return aom_variance##W##x##H(temp3, W, b, b_stride, sse);            \
+#define DIST_WTD_SUBPIX_AVG_VAR(W, H)                                      \
+  uint32_t aom_dist_wtd_sub_pixel_avg_variance##W##x##H##_ssse3(           \
+      const uint8_t *a, int a_stride, int xoffset, int yoffset,            \
+      const uint8_t *b, int b_stride, uint32_t *sse,                       \
+      const uint8_t *second_pred, const DIST_WTD_COMP_PARAMS *jcp_param) { \
+    uint16_t fdata3[(H + 1) * W];                                          \
+    uint8_t temp2[H * W];                                                  \
+    DECLARE_ALIGNED(16, uint8_t, temp3[H * W]);                            \
+                                                                           \
+    aom_var_filter_block2d_bil_first_pass_ssse3(                           \
+        a, fdata3, a_stride, 1, H + 1, W, bilinear_filters_2t[xoffset]);   \
+    aom_var_filter_block2d_bil_second_pass_ssse3(                          \
+        fdata3, temp2, W, W, H, W, bilinear_filters_2t[yoffset]);          \
+                                                                           \
+    aom_dist_wtd_comp_avg_pred_ssse3(temp3, second_pred, W, H, temp2, W,   \
+                                     jcp_param);                           \
+                                                                           \
+    return aom_variance##W##x##H(temp3, W, b, b_stride, sse);              \
   }
 
-JNT_SUBPIX_AVG_VAR(128, 128)
-JNT_SUBPIX_AVG_VAR(128, 64)
-JNT_SUBPIX_AVG_VAR(64, 128)
-JNT_SUBPIX_AVG_VAR(64, 64)
-JNT_SUBPIX_AVG_VAR(64, 32)
-JNT_SUBPIX_AVG_VAR(32, 64)
-JNT_SUBPIX_AVG_VAR(32, 32)
-JNT_SUBPIX_AVG_VAR(32, 16)
-JNT_SUBPIX_AVG_VAR(16, 32)
-JNT_SUBPIX_AVG_VAR(16, 16)
-JNT_SUBPIX_AVG_VAR(16, 8)
-JNT_SUBPIX_AVG_VAR(8, 16)
-JNT_SUBPIX_AVG_VAR(8, 8)
-JNT_SUBPIX_AVG_VAR(8, 4)
-JNT_SUBPIX_AVG_VAR(4, 8)
-JNT_SUBPIX_AVG_VAR(4, 4)
-JNT_SUBPIX_AVG_VAR(4, 16)
-JNT_SUBPIX_AVG_VAR(16, 4)
-JNT_SUBPIX_AVG_VAR(8, 32)
-JNT_SUBPIX_AVG_VAR(32, 8)
-JNT_SUBPIX_AVG_VAR(16, 64)
-JNT_SUBPIX_AVG_VAR(64, 16)
+DIST_WTD_SUBPIX_AVG_VAR(128, 128)
+DIST_WTD_SUBPIX_AVG_VAR(128, 64)
+DIST_WTD_SUBPIX_AVG_VAR(64, 128)
+DIST_WTD_SUBPIX_AVG_VAR(64, 64)
+DIST_WTD_SUBPIX_AVG_VAR(64, 32)
+DIST_WTD_SUBPIX_AVG_VAR(32, 64)
+DIST_WTD_SUBPIX_AVG_VAR(32, 32)
+DIST_WTD_SUBPIX_AVG_VAR(32, 16)
+DIST_WTD_SUBPIX_AVG_VAR(16, 32)
+DIST_WTD_SUBPIX_AVG_VAR(16, 16)
+DIST_WTD_SUBPIX_AVG_VAR(16, 8)
+DIST_WTD_SUBPIX_AVG_VAR(8, 16)
+DIST_WTD_SUBPIX_AVG_VAR(8, 8)
+DIST_WTD_SUBPIX_AVG_VAR(8, 4)
+DIST_WTD_SUBPIX_AVG_VAR(4, 8)
+DIST_WTD_SUBPIX_AVG_VAR(4, 4)
+DIST_WTD_SUBPIX_AVG_VAR(4, 16)
+DIST_WTD_SUBPIX_AVG_VAR(16, 4)
+DIST_WTD_SUBPIX_AVG_VAR(8, 32)
+DIST_WTD_SUBPIX_AVG_VAR(32, 8)
+DIST_WTD_SUBPIX_AVG_VAR(16, 64)
+DIST_WTD_SUBPIX_AVG_VAR(64, 16)
diff --git a/third_party/aom/aom_dsp/x86/loopfilter_sse2.c b/third_party/aom/aom_dsp/x86/loopfilter_sse2.c
index 9d88b5e493..e1981e555a 100644
--- a/third_party/aom/aom_dsp/x86/loopfilter_sse2.c
+++ b/third_party/aom/aom_dsp/x86/loopfilter_sse2.c
@@ -16,347 +16,12 @@
 #include "aom_dsp/x86/synonyms.h"
 #include "aom_ports/mem.h"
 #include "aom_ports/emmintrin_compat.h"
+#include "aom_dsp/x86/lpf_common_sse2.h"
 
 static INLINE __m128i abs_diff(__m128i a, __m128i b) {
   return _mm_or_si128(_mm_subs_epu8(a, b), _mm_subs_epu8(b, a));
 }
 
-static INLINE void transpose4x8_8x4_low_sse2(__m128i *x0, __m128i *x1,
-                                             __m128i *x2, __m128i *x3,
-                                             __m128i *d0, __m128i *d1,
-                                             __m128i *d2, __m128i *d3) {
-  // input
-  // x0   00 01 02 03 04 05 06 07 xx xx xx xx xx xx xx xx
-  // x1   10 11 12 13 14 15 16 17 xx xx xx xx xx xx xx xx
-  // x2   20 21 22 23 24 25 26 27 xx xx xx xx xx xx xx xx
-  // x3   30 31 32 33 34 35 36 37 xx xx xx xx xx xx xx xx
-  // output
-  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
-
-  __m128i w0, w1;
-
-  w0 = _mm_unpacklo_epi8(
-      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-  w1 = _mm_unpacklo_epi8(
-      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-
-  *d0 = _mm_unpacklo_epi16(
-      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-
-  *d1 = _mm_srli_si128(*d0,
-                       4);  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d2 = _mm_srli_si128(*d0,
-                       8);  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d3 = _mm_srli_si128(*d0,
-                       12);  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
-}
-
-static INLINE void transpose4x8_8x4_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
-                                         __m128i *x3, __m128i *d0, __m128i *d1,
-                                         __m128i *d2, __m128i *d3, __m128i *d4,
-                                         __m128i *d5, __m128i *d6,
-                                         __m128i *d7) {
-  // input
-  // x0   00 01 02 03 04 05 06 07 xx xx xx xx xx xx xx xx
-  // x1   10 11 12 13 14 15 16 17 xx xx xx xx xx xx xx xx
-  // x2   20 21 22 23 24 25 26 27 xx xx xx xx xx xx xx xx
-  // x3   30 31 32 33 34 35 36 37 xx xx xx xx xx xx xx xx
-  // output
-  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 04 14 24 34 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 05 15 25 35 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 06 16 26 36 xx xx xx xx xx xx xx xx xx xx xx xx
-  // 07 17 27 37 xx xx xx xx xx xx xx xx xx xx xx xx
-
-  __m128i w0, w1, ww0, ww1;
-
-  w0 = _mm_unpacklo_epi8(
-      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-  w1 = _mm_unpacklo_epi8(
-      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-
-  ww0 = _mm_unpacklo_epi16(
-      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  ww1 = _mm_unpackhi_epi16(
-      w0, w1);  // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
-
-  *d0 = ww0;  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d1 = _mm_srli_si128(ww0,
-                       4);  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d2 = _mm_srli_si128(ww0,
-                       8);  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d3 = _mm_srli_si128(ww0,
-                       12);  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
-
-  *d4 = ww1;  // 04 14 24 34 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d5 = _mm_srli_si128(ww1,
-                       4);  // 05 15 25 35 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d6 = _mm_srli_si128(ww1,
-                       8);  // 06 16 26 36 xx xx xx xx xx xx xx xx xx xx xx xx
-  *d7 = _mm_srli_si128(ww1,
-                       12);  // 07 17 27 37 xx xx xx xx xx xx xx xx xx xx xx xx
-}
-
-static INLINE void transpose8x8_low_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
-                                         __m128i *x3, __m128i *x4, __m128i *x5,
-                                         __m128i *x6, __m128i *x7, __m128i *d0,
-                                         __m128i *d1, __m128i *d2,
-                                         __m128i *d3) {
-  // input
-  // x0 00 01 02 03 04 05 06 07
-  // x1 10 11 12 13 14 15 16 17
-  // x2 20 21 22 23 24 25 26 27
-  // x3 30 31 32 33 34 35 36 37
-  // x4 40 41 42 43 44 45 46 47
-  // x5  50 51 52 53 54 55 56 57
-  // x6  60 61 62 63 64 65 66 67
-  // x7 70 71 72 73 74 75 76 77
-  // output
-  // d0 00 10 20 30 40 50 60 70 xx xx xx xx xx xx xx
-  // d1 01 11 21 31 41 51 61 71 xx xx xx xx xx xx xx xx
-  // d2 02 12 22 32 42 52 62 72 xx xx xx xx xx xx xx xx
-  // d3 03 13 23 33 43 53 63 73 xx xx xx xx xx xx xx xx
-
-  __m128i w0, w1, w2, w3, w4, w5;
-
-  w0 = _mm_unpacklo_epi8(
-      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-
-  w1 = _mm_unpacklo_epi8(
-      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-
-  w2 = _mm_unpacklo_epi8(
-      *x4, *x5);  // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
-
-  w3 = _mm_unpacklo_epi8(
-      *x6, *x7);  // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
-
-  w4 = _mm_unpacklo_epi16(
-      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  w5 = _mm_unpacklo_epi16(
-      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
-
-  *d0 = _mm_unpacklo_epi32(
-      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
-  *d1 = _mm_srli_si128(*d0, 8);
-  *d2 = _mm_unpackhi_epi32(
-      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
-  *d3 = _mm_srli_si128(*d2, 8);
-}
-
-static INLINE void transpose8x8_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
-                                     __m128i *x3, __m128i *x4, __m128i *x5,
-                                     __m128i *x6, __m128i *x7, __m128i *d0d1,
-                                     __m128i *d2d3, __m128i *d4d5,
-                                     __m128i *d6d7) {
-  __m128i w0, w1, w2, w3, w4, w5, w6, w7;
-  // x0 00 01 02 03 04 05 06 07
-  // x1 10 11 12 13 14 15 16 17
-  w0 = _mm_unpacklo_epi8(
-      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-
-  // x2 20 21 22 23 24 25 26 27
-  // x3 30 31 32 33 34 35 36 37
-  w1 = _mm_unpacklo_epi8(
-      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-
-  // x4 40 41 42 43 44 45 46 47
-  // x5  50 51 52 53 54 55 56 57
-  w2 = _mm_unpacklo_epi8(
-      *x4, *x5);  // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
-
-  // x6  60 61 62 63 64 65 66 67
-  // x7 70 71 72 73 74 75 76 77
-  w3 = _mm_unpacklo_epi8(
-      *x6, *x7);  // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
-
-  w4 = _mm_unpacklo_epi16(
-      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  w5 = _mm_unpacklo_epi16(
-      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
-
-  *d0d1 = _mm_unpacklo_epi32(
-      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
-  *d2d3 = _mm_unpackhi_epi32(
-      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
-
-  w6 = _mm_unpackhi_epi16(
-      w0, w1);  // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
-  w7 = _mm_unpackhi_epi16(
-      w2, w3);  // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77
-
-  *d4d5 = _mm_unpacklo_epi32(
-      w6, w7);  // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75
-  *d6d7 = _mm_unpackhi_epi32(
-      w6, w7);  // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77
-}
-
-static INLINE void transpose16x8_8x16_sse2(
-    __m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3, __m128i *x4,
-    __m128i *x5, __m128i *x6, __m128i *x7, __m128i *x8, __m128i *x9,
-    __m128i *x10, __m128i *x11, __m128i *x12, __m128i *x13, __m128i *x14,
-    __m128i *x15, __m128i *d0, __m128i *d1, __m128i *d2, __m128i *d3,
-    __m128i *d4, __m128i *d5, __m128i *d6, __m128i *d7) {
-  __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
-  __m128i w10, w11, w12, w13, w14, w15;
-
-  w0 = _mm_unpacklo_epi8(*x0, *x1);
-  w1 = _mm_unpacklo_epi8(*x2, *x3);
-  w2 = _mm_unpacklo_epi8(*x4, *x5);
-  w3 = _mm_unpacklo_epi8(*x6, *x7);
-
-  w8 = _mm_unpacklo_epi8(*x8, *x9);
-  w9 = _mm_unpacklo_epi8(*x10, *x11);
-  w10 = _mm_unpacklo_epi8(*x12, *x13);
-  w11 = _mm_unpacklo_epi8(*x14, *x15);
-
-  w4 = _mm_unpacklo_epi16(w0, w1);
-  w5 = _mm_unpacklo_epi16(w2, w3);
-  w12 = _mm_unpacklo_epi16(w8, w9);
-  w13 = _mm_unpacklo_epi16(w10, w11);
-
-  w6 = _mm_unpacklo_epi32(w4, w5);
-  w7 = _mm_unpackhi_epi32(w4, w5);
-  w14 = _mm_unpacklo_epi32(w12, w13);
-  w15 = _mm_unpackhi_epi32(w12, w13);
-
-  // Store first 4-line result
-  *d0 = _mm_unpacklo_epi64(w6, w14);
-  *d1 = _mm_unpackhi_epi64(w6, w14);
-  *d2 = _mm_unpacklo_epi64(w7, w15);
-  *d3 = _mm_unpackhi_epi64(w7, w15);
-
-  w4 = _mm_unpackhi_epi16(w0, w1);
-  w5 = _mm_unpackhi_epi16(w2, w3);
-  w12 = _mm_unpackhi_epi16(w8, w9);
-  w13 = _mm_unpackhi_epi16(w10, w11);
-
-  w6 = _mm_unpacklo_epi32(w4, w5);
-  w7 = _mm_unpackhi_epi32(w4, w5);
-  w14 = _mm_unpacklo_epi32(w12, w13);
-  w15 = _mm_unpackhi_epi32(w12, w13);
-
-  // Store second 4-line result
-  *d4 = _mm_unpacklo_epi64(w6, w14);
-  *d5 = _mm_unpackhi_epi64(w6, w14);
-  *d6 = _mm_unpacklo_epi64(w7, w15);
-  *d7 = _mm_unpackhi_epi64(w7, w15);
-}
-
-// this function treats its input as 2 parallel 8x4 matrices, transposes each of
-// them  independently while flipping the second matrix horizontaly  Used for 14
-// taps filter pq pairs inverse
-static INLINE void transpose_pq_14_inv_sse2(__m128i *x0, __m128i *x1,
-                                            __m128i *x2, __m128i *x3,
-                                            __m128i *x4, __m128i *x5,
-                                            __m128i *x6, __m128i *x7,
-                                            __m128i *pq0, __m128i *pq1,
-                                            __m128i *pq2, __m128i *pq3) {
-  __m128i w10, w11, w12, w13;
-  __m128i w0, w1, w2, w3, w4, w5;
-  __m128i d0, d1, d2, d3;
-
-  w0 = _mm_unpacklo_epi8(
-      *x0, *x1);  // p 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-  w1 = _mm_unpacklo_epi8(
-      *x2, *x3);  // p 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-  w2 = _mm_unpacklo_epi8(
-      *x4, *x5);  // p 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
-  w3 = _mm_unpacklo_epi8(
-      *x6, *x7);  // p 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
-
-  w4 = _mm_unpacklo_epi16(
-      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  w5 = _mm_unpacklo_epi16(
-      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
-
-  d0 = _mm_unpacklo_epi32(
-      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
-  d2 = _mm_unpackhi_epi32(
-      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
-
-  w10 = _mm_unpacklo_epi8(
-      *x7, *x6);  // q xx xx xx xx xx xx xx xx 00 10 01 11 02 12 03 13
-  w11 = _mm_unpacklo_epi8(
-      *x5, *x4);  // q  xx xx xx xx xx xx xx xx 20 30 21 31 22 32 23 33
-  w12 = _mm_unpacklo_epi8(
-      *x3, *x2);  // q  xx xx xx xx xx xx xx xx 40 50 41 51 42 52 43 53
-  w13 = _mm_unpacklo_epi8(
-      *x1, *x0);  // q  xx xx xx xx xx xx xx xx 60 70 61 71 62 72 63 73
-
-  w4 = _mm_unpackhi_epi16(
-      w10, w11);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  w5 = _mm_unpackhi_epi16(
-      w12, w13);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
-
-  d1 = _mm_unpacklo_epi32(
-      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
-  d3 = _mm_unpackhi_epi32(
-      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
-
-  *pq0 = _mm_unpacklo_epi64(d0, d1);  // pq
-  *pq1 = _mm_unpackhi_epi64(d0, d1);  // pq
-  *pq2 = _mm_unpacklo_epi64(d2, d3);  // pq
-  *pq3 = _mm_unpackhi_epi64(d2, d3);  // pq
-}
-
-static INLINE void transpose8x16_16x8_sse2(
-    __m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3, __m128i *x4,
-    __m128i *x5, __m128i *x6, __m128i *x7, __m128i *d0d1, __m128i *d2d3,
-    __m128i *d4d5, __m128i *d6d7, __m128i *d8d9, __m128i *d10d11,
-    __m128i *d12d13, __m128i *d14d15) {
-  __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
-  __m128i w10, w11, w12, w13, w14, w15;
-
-  w0 = _mm_unpacklo_epi8(*x0, *x1);
-  w1 = _mm_unpacklo_epi8(*x2, *x3);
-  w2 = _mm_unpacklo_epi8(*x4, *x5);
-  w3 = _mm_unpacklo_epi8(*x6, *x7);
-
-  w8 = _mm_unpackhi_epi8(*x0, *x1);
-  w9 = _mm_unpackhi_epi8(*x2, *x3);
-  w10 = _mm_unpackhi_epi8(*x4, *x5);
-  w11 = _mm_unpackhi_epi8(*x6, *x7);
-
-  w4 = _mm_unpacklo_epi16(w0, w1);
-  w5 = _mm_unpacklo_epi16(w2, w3);
-  w12 = _mm_unpacklo_epi16(w8, w9);
-  w13 = _mm_unpacklo_epi16(w10, w11);
-
-  w6 = _mm_unpacklo_epi32(w4, w5);
-  w7 = _mm_unpackhi_epi32(w4, w5);
-  w14 = _mm_unpacklo_epi32(w12, w13);
-  w15 = _mm_unpackhi_epi32(w12, w13);
-
-  // Store first 4-line result
-  *d0d1 = _mm_unpacklo_epi64(w6, w14);
-  *d2d3 = _mm_unpackhi_epi64(w6, w14);
-  *d4d5 = _mm_unpacklo_epi64(w7, w15);
-  *d6d7 = _mm_unpackhi_epi64(w7, w15);
-
-  w4 = _mm_unpackhi_epi16(w0, w1);
-  w5 = _mm_unpackhi_epi16(w2, w3);
-  w12 = _mm_unpackhi_epi16(w8, w9);
-  w13 = _mm_unpackhi_epi16(w10, w11);
-
-  w6 = _mm_unpacklo_epi32(w4, w5);
-  w7 = _mm_unpackhi_epi32(w4, w5);
-  w14 = _mm_unpacklo_epi32(w12, w13);
-  w15 = _mm_unpackhi_epi32(w12, w13);
-
-  // Store second 4-line result
-  *d8d9 = _mm_unpacklo_epi64(w6, w14);
-  *d10d11 = _mm_unpackhi_epi64(w6, w14);
-  *d12d13 = _mm_unpacklo_epi64(w7, w15);
-  *d14d15 = _mm_unpackhi_epi64(w7, w15);
-}
-
 // this function treats its input as 2 parallel 8x4 matrices, transposes each of
 // them to 4x8  independently while flipping the second matrix horizontaly. Used
 // for 14 taps pq pairs creation
@@ -416,6 +81,63 @@ static INLINE void transpose_pq_14_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
       ww2);  // 07 17 27 37  08 18 28 38     xx xx xx xx xx xx xx xx
 }
 
+// this function treats its input as 2 parallel 8x4 matrices, transposes each of
+// them  independently while flipping the second matrix horizontaly  Used for 14
+// taps filter pq pairs inverse
+static INLINE void transpose_pq_14_inv_sse2(__m128i *x0, __m128i *x1,
+                                            __m128i *x2, __m128i *x3,
+                                            __m128i *x4, __m128i *x5,
+                                            __m128i *x6, __m128i *x7,
+                                            __m128i *pq0, __m128i *pq1,
+                                            __m128i *pq2, __m128i *pq3) {
+  __m128i w10, w11, w12, w13;
+  __m128i w0, w1, w2, w3, w4, w5;
+  __m128i d0, d1, d2, d3;
+
+  w0 = _mm_unpacklo_epi8(
+      *x0, *x1);  // p 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+  w1 = _mm_unpacklo_epi8(
+      *x2, *x3);  // p 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+  w2 = _mm_unpacklo_epi8(
+      *x4, *x5);  // p 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
+  w3 = _mm_unpacklo_epi8(
+      *x6, *x7);  // p 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
+
+  w4 = _mm_unpacklo_epi16(
+      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+  w5 = _mm_unpacklo_epi16(
+      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+
+  d0 = _mm_unpacklo_epi32(
+      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+  d2 = _mm_unpackhi_epi32(
+      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+
+  w10 = _mm_unpacklo_epi8(
+      *x7, *x6);  // q xx xx xx xx xx xx xx xx 00 10 01 11 02 12 03 13
+  w11 = _mm_unpacklo_epi8(
+      *x5, *x4);  // q  xx xx xx xx xx xx xx xx 20 30 21 31 22 32 23 33
+  w12 = _mm_unpacklo_epi8(
+      *x3, *x2);  // q  xx xx xx xx xx xx xx xx 40 50 41 51 42 52 43 53
+  w13 = _mm_unpacklo_epi8(
+      *x1, *x0);  // q  xx xx xx xx xx xx xx xx 60 70 61 71 62 72 63 73
+
+  w4 = _mm_unpackhi_epi16(
+      w10, w11);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+  w5 = _mm_unpackhi_epi16(
+      w12, w13);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+
+  d1 = _mm_unpacklo_epi32(
+      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+  d3 = _mm_unpackhi_epi32(
+      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+
+  *pq0 = _mm_unpacklo_epi64(d0, d1);  // pq
+  *pq1 = _mm_unpackhi_epi64(d0, d1);  // pq
+  *pq2 = _mm_unpacklo_epi64(d2, d3);  // pq
+  *pq3 = _mm_unpackhi_epi64(d2, d3);  // pq
+}
+
 static AOM_FORCE_INLINE void filter4_sse2(__m128i *p1p0, __m128i *q1q0,
                                           __m128i *hev, __m128i *mask,
                                           __m128i *qs1qs0, __m128i *ps1ps0) {
@@ -616,10 +338,10 @@ void aom_lpf_horizontal_4_sse2(uint8_t *s, int p /* pitch */,
   __m128i qs1qs0, ps1ps0;
   __m128i p1, p0, q0, q1;
 
-  p1 = _mm_cvtsi32_si128(*(int *)(s - 2 * p));
-  p0 = _mm_cvtsi32_si128(*(int *)(s - 1 * p));
-  q0 = _mm_cvtsi32_si128(*(int *)(s + 0 * p));
-  q1 = _mm_cvtsi32_si128(*(int *)(s + 1 * p));
+  p1 = xx_loadl_32(s - 2 * p);
+  p0 = xx_loadl_32(s - 1 * p);
+  q0 = xx_loadl_32(s - 0 * p);
+  q1 = xx_loadl_32(s + 1 * p);
 
   lpf_internal_4_sse2(&p1, &p0, &q0, &q1, &limit, &thresh, &qs1qs0, &ps1ps0);
 
@@ -1241,23 +963,16 @@ void aom_lpf_horizontal_14_sse2(unsigned char *s, int p,
   __m128i limit = _mm_load_si128((const __m128i *)_limit);
   __m128i thresh = _mm_load_si128((const __m128i *)_thresh);
 
-  q4p4 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 5 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 4 * p)));
-  q3p3 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 4 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 3 * p)));
-  q2p2 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 3 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 2 * p)));
-  q1p1 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 2 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 1 * p)));
+  q4p4 = _mm_unpacklo_epi32(xx_loadl_32(s - 5 * p), xx_loadl_32(s + 4 * p));
+  q3p3 = _mm_unpacklo_epi32(xx_loadl_32(s - 4 * p), xx_loadl_32(s + 3 * p));
+  q2p2 = _mm_unpacklo_epi32(xx_loadl_32(s - 3 * p), xx_loadl_32(s + 2 * p));
+  q1p1 = _mm_unpacklo_epi32(xx_loadl_32(s - 2 * p), xx_loadl_32(s + 1 * p));
 
-  q0p0 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 1 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s - 0 * p)));
+  q0p0 = _mm_unpacklo_epi32(xx_loadl_32(s - 1 * p), xx_loadl_32(s - 0 * p));
 
-  q5p5 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 6 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 5 * p)));
+  q5p5 = _mm_unpacklo_epi32(xx_loadl_32(s - 6 * p), xx_loadl_32(s + 5 * p));
 
-  q6p6 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(*(int *)(s - 7 * p)),
-                            _mm_cvtsi32_si128(*(int *)(s + 6 * p)));
+  q6p6 = _mm_unpacklo_epi32(xx_loadl_32(s - 7 * p), xx_loadl_32(s + 6 * p));
 
   lpf_internal_14_sse2(&q6p6, &q5p5, &q4p4, &q3p3, &q2p2, &q1p1, &q0p0, &blimit,
                        &limit, &thresh);
@@ -1543,12 +1258,12 @@ void aom_lpf_horizontal_6_sse2(unsigned char *s, int p,
   __m128i limit = _mm_load_si128((__m128i *)_limit);
   __m128i thresh = _mm_load_si128((__m128i *)_thresh);
 
-  p2 = _mm_cvtsi32_si128(*(int *)(s - 3 * p));
-  p1 = _mm_cvtsi32_si128(*(int *)(s - 2 * p));
-  p0 = _mm_cvtsi32_si128(*(int *)(s - 1 * p));
-  q0 = _mm_cvtsi32_si128(*(int *)(s - 0 * p));
-  q1 = _mm_cvtsi32_si128(*(int *)(s + 1 * p));
-  q2 = _mm_cvtsi32_si128(*(int *)(s + 2 * p));
+  p2 = xx_loadl_32(s - 3 * p);
+  p1 = xx_loadl_32(s - 2 * p);
+  p0 = xx_loadl_32(s - 1 * p);
+  q0 = xx_loadl_32(s - 0 * p);
+  q1 = xx_loadl_32(s + 1 * p);
+  q2 = xx_loadl_32(s + 2 * p);
 
   lpf_internal_6_sse2(&p2, &q2, &p1, &q1, &p0, &q0, &q1q0, &p1p0, &blimit,
                       &limit, &thresh);
@@ -1895,20 +1610,20 @@ void aom_lpf_horizontal_8_sse2(unsigned char *s, int p,
                                const unsigned char *_blimit,
                                const unsigned char *_limit,
                                const unsigned char *_thresh) {
-  __m128i p2, p1, p0, q0, q1, q2, p3, q3;
+  __m128i p3, p2, p1, p0, q0, q1, q2, q3;
   __m128i q1q0, p1p0;
   __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
   __m128i limit = _mm_load_si128((const __m128i *)_limit);
   __m128i thresh = _mm_load_si128((const __m128i *)_thresh);
 
-  p3 = _mm_cvtsi32_si128(*(int *)(s - 4 * p));
-  p2 = _mm_cvtsi32_si128(*(int *)(s - 3 * p));
-  p1 = _mm_cvtsi32_si128(*(int *)(s - 2 * p));
-  p0 = _mm_cvtsi32_si128(*(int *)(s - 1 * p));
-  q0 = _mm_cvtsi32_si128(*(int *)(s - 0 * p));
-  q1 = _mm_cvtsi32_si128(*(int *)(s + 1 * p));
-  q2 = _mm_cvtsi32_si128(*(int *)(s + 2 * p));
-  q3 = _mm_cvtsi32_si128(*(int *)(s + 3 * p));
+  p3 = xx_loadl_32(s - 4 * p);
+  p2 = xx_loadl_32(s - 3 * p);
+  p1 = xx_loadl_32(s - 2 * p);
+  p0 = xx_loadl_32(s - 1 * p);
+  q0 = xx_loadl_32(s - 0 * p);
+  q1 = xx_loadl_32(s + 1 * p);
+  q2 = xx_loadl_32(s + 2 * p);
+  q3 = xx_loadl_32(s + 3 * p);
 
   lpf_internal_8_sse2(&p3, &q3, &p2, &q2, &p1, &q1, &p0, &q0, &q1q0, &p1p0,
                       &blimit, &limit, &thresh);
diff --git a/third_party/aom/aom_dsp/x86/lpf_common_sse2.h b/third_party/aom/aom_dsp/x86/lpf_common_sse2.h
index 8970fe7dd6..6ed2cbfdf4 100644
--- a/third_party/aom/aom_dsp/x86/lpf_common_sse2.h
+++ b/third_party/aom/aom_dsp/x86/lpf_common_sse2.h
@@ -212,4 +212,284 @@ static INLINE void highbd_transpose8x16_sse2(
                            d4 + 1, d5 + 1, d6 + 1, d7 + 1);
 }
 
+// Low bit depth functions
+static INLINE void transpose4x8_8x4_low_sse2(__m128i *x0, __m128i *x1,
+                                             __m128i *x2, __m128i *x3,
+                                             __m128i *d0, __m128i *d1,
+                                             __m128i *d2, __m128i *d3) {
+  // input
+  // x0   00 01 02 03 04 05 06 07 xx xx xx xx xx xx xx xx
+  // x1   10 11 12 13 14 15 16 17 xx xx xx xx xx xx xx xx
+  // x2   20 21 22 23 24 25 26 27 xx xx xx xx xx xx xx xx
+  // x3   30 31 32 33 34 35 36 37 xx xx xx xx xx xx xx xx
+  // output
+  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
+
+  __m128i w0, w1;
+
+  w0 = _mm_unpacklo_epi8(
+      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+  w1 = _mm_unpacklo_epi8(
+      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+
+  *d0 = _mm_unpacklo_epi16(
+      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+
+  *d1 = _mm_srli_si128(*d0,
+                       4);  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d2 = _mm_srli_si128(*d0,
+                       8);  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d3 = _mm_srli_si128(*d0,
+                       12);  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
+}
+
+static INLINE void transpose4x8_8x4_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
+                                         __m128i *x3, __m128i *d0, __m128i *d1,
+                                         __m128i *d2, __m128i *d3, __m128i *d4,
+                                         __m128i *d5, __m128i *d6,
+                                         __m128i *d7) {
+  // input
+  // x0   00 01 02 03 04 05 06 07 xx xx xx xx xx xx xx xx
+  // x1   10 11 12 13 14 15 16 17 xx xx xx xx xx xx xx xx
+  // x2   20 21 22 23 24 25 26 27 xx xx xx xx xx xx xx xx
+  // x3   30 31 32 33 34 35 36 37 xx xx xx xx xx xx xx xx
+  // output
+  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 04 14 24 34 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 05 15 25 35 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 06 16 26 36 xx xx xx xx xx xx xx xx xx xx xx xx
+  // 07 17 27 37 xx xx xx xx xx xx xx xx xx xx xx xx
+
+  __m128i w0, w1, ww0, ww1;
+
+  w0 = _mm_unpacklo_epi8(
+      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+  w1 = _mm_unpacklo_epi8(
+      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+
+  ww0 = _mm_unpacklo_epi16(
+      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+  ww1 = _mm_unpackhi_epi16(
+      w0, w1);  // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+
+  *d0 = ww0;  // 00 10 20 30 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d1 = _mm_srli_si128(ww0,
+                       4);  // 01 11 21 31 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d2 = _mm_srli_si128(ww0,
+                       8);  // 02 12 22 32 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d3 = _mm_srli_si128(ww0,
+                       12);  // 03 13 23 33 xx xx xx xx xx xx xx xx xx xx xx xx
+
+  *d4 = ww1;  // 04 14 24 34 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d5 = _mm_srli_si128(ww1,
+                       4);  // 05 15 25 35 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d6 = _mm_srli_si128(ww1,
+                       8);  // 06 16 26 36 xx xx xx xx xx xx xx xx xx xx xx xx
+  *d7 = _mm_srli_si128(ww1,
+                       12);  // 07 17 27 37 xx xx xx xx xx xx xx xx xx xx xx xx
+}
+
+static INLINE void transpose8x8_low_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
+                                         __m128i *x3, __m128i *x4, __m128i *x5,
+                                         __m128i *x6, __m128i *x7, __m128i *d0,
+                                         __m128i *d1, __m128i *d2,
+                                         __m128i *d3) {
+  // input
+  // x0 00 01 02 03 04 05 06 07
+  // x1 10 11 12 13 14 15 16 17
+  // x2 20 21 22 23 24 25 26 27
+  // x3 30 31 32 33 34 35 36 37
+  // x4 40 41 42 43 44 45 46 47
+  // x5  50 51 52 53 54 55 56 57
+  // x6  60 61 62 63 64 65 66 67
+  // x7 70 71 72 73 74 75 76 77
+  // output
+  // d0 00 10 20 30 40 50 60 70 xx xx xx xx xx xx xx
+  // d1 01 11 21 31 41 51 61 71 xx xx xx xx xx xx xx xx
+  // d2 02 12 22 32 42 52 62 72 xx xx xx xx xx xx xx xx
+  // d3 03 13 23 33 43 53 63 73 xx xx xx xx xx xx xx xx
+
+  __m128i w0, w1, w2, w3, w4, w5;
+
+  w0 = _mm_unpacklo_epi8(
+      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+
+  w1 = _mm_unpacklo_epi8(
+      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+
+  w2 = _mm_unpacklo_epi8(
+      *x4, *x5);  // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
+
+  w3 = _mm_unpacklo_epi8(
+      *x6, *x7);  // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
+
+  w4 = _mm_unpacklo_epi16(
+      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+  w5 = _mm_unpacklo_epi16(
+      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+
+  *d0 = _mm_unpacklo_epi32(
+      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+  *d1 = _mm_srli_si128(*d0, 8);
+  *d2 = _mm_unpackhi_epi32(
+      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+  *d3 = _mm_srli_si128(*d2, 8);
+}
+
+static INLINE void transpose8x8_sse2(__m128i *x0, __m128i *x1, __m128i *x2,
+                                     __m128i *x3, __m128i *x4, __m128i *x5,
+                                     __m128i *x6, __m128i *x7, __m128i *d0d1,
+                                     __m128i *d2d3, __m128i *d4d5,
+                                     __m128i *d6d7) {
+  __m128i w0, w1, w2, w3, w4, w5, w6, w7;
+  // x0 00 01 02 03 04 05 06 07
+  // x1 10 11 12 13 14 15 16 17
+  w0 = _mm_unpacklo_epi8(
+      *x0, *x1);  // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
+
+  // x2 20 21 22 23 24 25 26 27
+  // x3 30 31 32 33 34 35 36 37
+  w1 = _mm_unpacklo_epi8(
+      *x2, *x3);  // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
+
+  // x4 40 41 42 43 44 45 46 47
+  // x5  50 51 52 53 54 55 56 57
+  w2 = _mm_unpacklo_epi8(
+      *x4, *x5);  // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
+
+  // x6  60 61 62 63 64 65 66 67
+  // x7 70 71 72 73 74 75 76 77
+  w3 = _mm_unpacklo_epi8(
+      *x6, *x7);  // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
+
+  w4 = _mm_unpacklo_epi16(
+      w0, w1);  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
+  w5 = _mm_unpacklo_epi16(
+      w2, w3);  // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
+
+  *d0d1 = _mm_unpacklo_epi32(
+      w4, w5);  // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
+  *d2d3 = _mm_unpackhi_epi32(
+      w4, w5);  // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
+
+  w6 = _mm_unpackhi_epi16(
+      w0, w1);  // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
+  w7 = _mm_unpackhi_epi16(
+      w2, w3);  // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77
+
+  *d4d5 = _mm_unpacklo_epi32(
+      w6, w7);  // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75
+  *d6d7 = _mm_unpackhi_epi32(
+      w6, w7);  // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77
+}
+
+static INLINE void transpose16x8_8x16_sse2(
+    __m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3, __m128i *x4,
+    __m128i *x5, __m128i *x6, __m128i *x7, __m128i *x8, __m128i *x9,
+    __m128i *x10, __m128i *x11, __m128i *x12, __m128i *x13, __m128i *x14,
+    __m128i *x15, __m128i *d0, __m128i *d1, __m128i *d2, __m128i *d3,
+    __m128i *d4, __m128i *d5, __m128i *d6, __m128i *d7) {
+  __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
+  __m128i w10, w11, w12, w13, w14, w15;
+
+  w0 = _mm_unpacklo_epi8(*x0, *x1);
+  w1 = _mm_unpacklo_epi8(*x2, *x3);
+  w2 = _mm_unpacklo_epi8(*x4, *x5);
+  w3 = _mm_unpacklo_epi8(*x6, *x7);
+
+  w8 = _mm_unpacklo_epi8(*x8, *x9);
+  w9 = _mm_unpacklo_epi8(*x10, *x11);
+  w10 = _mm_unpacklo_epi8(*x12, *x13);
+  w11 = _mm_unpacklo_epi8(*x14, *x15);
+
+  w4 = _mm_unpacklo_epi16(w0, w1);
+  w5 = _mm_unpacklo_epi16(w2, w3);
+  w12 = _mm_unpacklo_epi16(w8, w9);
+  w13 = _mm_unpacklo_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  *d0 = _mm_unpacklo_epi64(w6, w14);
+  *d1 = _mm_unpackhi_epi64(w6, w14);
+  *d2 = _mm_unpacklo_epi64(w7, w15);
+  *d3 = _mm_unpackhi_epi64(w7, w15);
+
+  w4 = _mm_unpackhi_epi16(w0, w1);
+  w5 = _mm_unpackhi_epi16(w2, w3);
+  w12 = _mm_unpackhi_epi16(w8, w9);
+  w13 = _mm_unpackhi_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  *d4 = _mm_unpacklo_epi64(w6, w14);
+  *d5 = _mm_unpackhi_epi64(w6, w14);
+  *d6 = _mm_unpacklo_epi64(w7, w15);
+  *d7 = _mm_unpackhi_epi64(w7, w15);
+}
+
+static INLINE void transpose8x16_16x8_sse2(
+    __m128i *x0, __m128i *x1, __m128i *x2, __m128i *x3, __m128i *x4,
+    __m128i *x5, __m128i *x6, __m128i *x7, __m128i *d0d1, __m128i *d2d3,
+    __m128i *d4d5, __m128i *d6d7, __m128i *d8d9, __m128i *d10d11,
+    __m128i *d12d13, __m128i *d14d15) {
+  __m128i w0, w1, w2, w3, w4, w5, w6, w7, w8, w9;
+  __m128i w10, w11, w12, w13, w14, w15;
+
+  w0 = _mm_unpacklo_epi8(*x0, *x1);
+  w1 = _mm_unpacklo_epi8(*x2, *x3);
+  w2 = _mm_unpacklo_epi8(*x4, *x5);
+  w3 = _mm_unpacklo_epi8(*x6, *x7);
+
+  w8 = _mm_unpackhi_epi8(*x0, *x1);
+  w9 = _mm_unpackhi_epi8(*x2, *x3);
+  w10 = _mm_unpackhi_epi8(*x4, *x5);
+  w11 = _mm_unpackhi_epi8(*x6, *x7);
+
+  w4 = _mm_unpacklo_epi16(w0, w1);
+  w5 = _mm_unpacklo_epi16(w2, w3);
+  w12 = _mm_unpacklo_epi16(w8, w9);
+  w13 = _mm_unpacklo_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store first 4-line result
+  *d0d1 = _mm_unpacklo_epi64(w6, w14);
+  *d2d3 = _mm_unpackhi_epi64(w6, w14);
+  *d4d5 = _mm_unpacklo_epi64(w7, w15);
+  *d6d7 = _mm_unpackhi_epi64(w7, w15);
+
+  w4 = _mm_unpackhi_epi16(w0, w1);
+  w5 = _mm_unpackhi_epi16(w2, w3);
+  w12 = _mm_unpackhi_epi16(w8, w9);
+  w13 = _mm_unpackhi_epi16(w10, w11);
+
+  w6 = _mm_unpacklo_epi32(w4, w5);
+  w7 = _mm_unpackhi_epi32(w4, w5);
+  w14 = _mm_unpacklo_epi32(w12, w13);
+  w15 = _mm_unpackhi_epi32(w12, w13);
+
+  // Store second 4-line result
+  *d8d9 = _mm_unpacklo_epi64(w6, w14);
+  *d10d11 = _mm_unpackhi_epi64(w6, w14);
+  *d12d13 = _mm_unpacklo_epi64(w7, w15);
+  *d14d15 = _mm_unpackhi_epi64(w7, w15);
+}
+
 #endif  // AOM_AOM_DSP_X86_LPF_COMMON_SSE2_H_
diff --git a/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c b/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
index d7dbefd7d9..ebf4631ba7 100644
--- a/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
+++ b/third_party/aom/aom_dsp/x86/masked_variance_intrin_ssse3.c
@@ -218,15 +218,15 @@ static void bilinear_filter(const uint8_t *src, int src_stride, int xoffset,
   }
 }
 
-static INLINE __m128i filter_block_2rows(const __m128i a0, const __m128i b0,
-                                         const __m128i a1, const __m128i b1,
-                                         const __m128i filter) {
-  __m128i v0 = _mm_unpacklo_epi8(a0, b0);
-  v0 = _mm_maddubs_epi16(v0, filter);
+static INLINE __m128i filter_block_2rows(const __m128i *a0, const __m128i *b0,
+                                         const __m128i *a1, const __m128i *b1,
+                                         const __m128i *filter) {
+  __m128i v0 = _mm_unpacklo_epi8(*a0, *b0);
+  v0 = _mm_maddubs_epi16(v0, *filter);
   v0 = xx_roundn_epu16(v0, FILTER_BITS);
 
-  __m128i v1 = _mm_unpacklo_epi8(a1, b1);
-  v1 = _mm_maddubs_epi16(v1, filter);
+  __m128i v1 = _mm_unpacklo_epi8(*a1, *b1);
+  v1 = _mm_maddubs_epi16(v1, *filter);
   v1 = xx_roundn_epu16(v1, FILTER_BITS);
 
   return _mm_packus_epi16(v0, v1);
@@ -262,7 +262,7 @@ static void bilinear_filter8xh(const uint8_t *src, int src_stride, int xoffset,
       const __m128i z0 = _mm_srli_si128(x0, 1);
       const __m128i x1 = _mm_loadu_si128((__m128i *)&src[src_stride]);
       const __m128i z1 = _mm_srli_si128(x1, 1);
-      const __m128i res = filter_block_2rows(x0, z0, x1, z1, hfilter_vec);
+      const __m128i res = filter_block_2rows(&x0, &z0, &x1, &z1, &hfilter_vec);
       _mm_storeu_si128((__m128i *)b, res);
 
       src += src_stride * 2;
@@ -296,7 +296,7 @@ static void bilinear_filter8xh(const uint8_t *src, int src_stride, int xoffset,
       const __m128i x = _mm_loadl_epi64((__m128i *)dst);
       const __m128i y = _mm_loadl_epi64((__m128i *)&dst[8]);
       const __m128i z = _mm_loadl_epi64((__m128i *)&dst[16]);
-      const __m128i res = filter_block_2rows(x, y, y, z, vfilter_vec);
+      const __m128i res = filter_block_2rows(&x, &y, &y, &z, &vfilter_vec);
       _mm_storeu_si128((__m128i *)dst, res);
 
       dst += 16;
@@ -343,7 +343,7 @@ static void bilinear_filter4xh(const uint8_t *src, int src_stride, int xoffset,
       const __m128i b0 = _mm_unpacklo_epi32(z0, z1);
       const __m128i a1 = _mm_unpacklo_epi32(x2, x3);
       const __m128i b1 = _mm_unpacklo_epi32(z2, z3);
-      const __m128i res = filter_block_2rows(a0, b0, a1, b1, hfilter_vec);
+      const __m128i res = filter_block_2rows(&a0, &b0, &a1, &b1, &hfilter_vec);
       _mm_storeu_si128((__m128i *)b, res);
 
       src += src_stride * 4;
@@ -384,7 +384,7 @@ static void bilinear_filter4xh(const uint8_t *src, int src_stride, int xoffset,
       const __m128i b0 = _mm_unpacklo_epi32(b, c);
       const __m128i a1 = _mm_unpacklo_epi32(c, d);
       const __m128i b1 = _mm_unpacklo_epi32(d, e);
-      const __m128i res = filter_block_2rows(a0, b0, a1, b1, vfilter_vec);
+      const __m128i res = filter_block_2rows(&a0, &b0, &a1, &b1, &vfilter_vec);
       _mm_storeu_si128((__m128i *)dst, res);
 
       dst += 16;
@@ -392,29 +392,29 @@ static void bilinear_filter4xh(const uint8_t *src, int src_stride, int xoffset,
   }
 }
 
-static INLINE void accumulate_block(const __m128i src, const __m128i a,
-                                    const __m128i b, const __m128i m,
+static INLINE void accumulate_block(const __m128i *src, const __m128i *a,
+                                    const __m128i *b, const __m128i *m,
                                     __m128i *sum, __m128i *sum_sq) {
   const __m128i zero = _mm_setzero_si128();
   const __m128i one = _mm_set1_epi16(1);
   const __m128i mask_max = _mm_set1_epi8((1 << AOM_BLEND_A64_ROUND_BITS));
-  const __m128i m_inv = _mm_sub_epi8(mask_max, m);
+  const __m128i m_inv = _mm_sub_epi8(mask_max, *m);
 
   // Calculate 16 predicted pixels.
   // Note that the maximum value of any entry of 'pred_l' or 'pred_r'
   // is 64 * 255, so we have plenty of space to add rounding constants.
-  const __m128i data_l = _mm_unpacklo_epi8(a, b);
-  const __m128i mask_l = _mm_unpacklo_epi8(m, m_inv);
+  const __m128i data_l = _mm_unpacklo_epi8(*a, *b);
+  const __m128i mask_l = _mm_unpacklo_epi8(*m, m_inv);
   __m128i pred_l = _mm_maddubs_epi16(data_l, mask_l);
   pred_l = xx_roundn_epu16(pred_l, AOM_BLEND_A64_ROUND_BITS);
 
-  const __m128i data_r = _mm_unpackhi_epi8(a, b);
-  const __m128i mask_r = _mm_unpackhi_epi8(m, m_inv);
+  const __m128i data_r = _mm_unpackhi_epi8(*a, *b);
+  const __m128i mask_r = _mm_unpackhi_epi8(*m, m_inv);
   __m128i pred_r = _mm_maddubs_epi16(data_r, mask_r);
   pred_r = xx_roundn_epu16(pred_r, AOM_BLEND_A64_ROUND_BITS);
 
-  const __m128i src_l = _mm_unpacklo_epi8(src, zero);
-  const __m128i src_r = _mm_unpackhi_epi8(src, zero);
+  const __m128i src_l = _mm_unpacklo_epi8(*src, zero);
+  const __m128i src_r = _mm_unpackhi_epi8(*src, zero);
   const __m128i diff_l = _mm_sub_epi16(pred_l, src_l);
   const __m128i diff_r = _mm_sub_epi16(pred_r, src_r);
 
@@ -440,7 +440,7 @@ static void masked_variance(const uint8_t *src_ptr, int src_stride,
       const __m128i a = _mm_loadu_si128((const __m128i *)&a_ptr[x]);
       const __m128i b = _mm_loadu_si128((const __m128i *)&b_ptr[x]);
       const __m128i m = _mm_loadu_si128((const __m128i *)&m_ptr[x]);
-      accumulate_block(src, a, b, m, &sum, &sum_sq);
+      accumulate_block(&src, &a, &b, &m, &sum, &sum_sq);
     }
 
     src_ptr += src_stride;
@@ -471,7 +471,7 @@ static void masked_variance8xh(const uint8_t *src_ptr, int src_stride,
     const __m128i m =
         _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)m_ptr),
                            _mm_loadl_epi64((const __m128i *)&m_ptr[m_stride]));
-    accumulate_block(src, a, b, m, &sum, &sum_sq);
+    accumulate_block(&src, &a, &b, &m, &sum, &sum_sq);
 
     src_ptr += src_stride * 2;
     a_ptr += 16;
@@ -503,7 +503,7 @@ static void masked_variance4xh(const uint8_t *src_ptr, int src_stride,
     const __m128i m = _mm_setr_epi32(
         *(uint32_t *)m_ptr, *(uint32_t *)&m_ptr[m_stride],
         *(uint32_t *)&m_ptr[m_stride * 2], *(uint32_t *)&m_ptr[m_stride * 3]);
-    accumulate_block(src, a, b, m, &sum, &sum_sq);
+    accumulate_block(&src, &a, &b, &m, &sum, &sum_sq);
 
     src_ptr += src_stride * 4;
     a_ptr += 16;
@@ -797,17 +797,17 @@ static void highbd_bilinear_filter(const uint16_t *src, int src_stride,
   }
 }
 
-static INLINE __m128i highbd_filter_block_2rows(const __m128i a0,
-                                                const __m128i b0,
-                                                const __m128i a1,
-                                                const __m128i b1,
-                                                const __m128i filter) {
-  __m128i v0 = _mm_unpacklo_epi16(a0, b0);
-  v0 = _mm_madd_epi16(v0, filter);
+static INLINE __m128i highbd_filter_block_2rows(const __m128i *a0,
+                                                const __m128i *b0,
+                                                const __m128i *a1,
+                                                const __m128i *b1,
+                                                const __m128i *filter) {
+  __m128i v0 = _mm_unpacklo_epi16(*a0, *b0);
+  v0 = _mm_madd_epi16(v0, *filter);
   v0 = xx_roundn_epu32(v0, FILTER_BITS);
 
-  __m128i v1 = _mm_unpacklo_epi16(a1, b1);
-  v1 = _mm_madd_epi16(v1, filter);
+  __m128i v1 = _mm_unpacklo_epi16(*a1, *b1);
+  v1 = _mm_madd_epi16(v1, *filter);
   v1 = xx_roundn_epu32(v1, FILTER_BITS);
 
   return _mm_packs_epi32(v0, v1);
@@ -845,7 +845,7 @@ static void highbd_bilinear_filter4xh(const uint16_t *src, int src_stride,
       const __m128i x1 = _mm_loadu_si128((__m128i *)&src[src_stride]);
       const __m128i z1 = _mm_srli_si128(x1, 2);
       const __m128i res =
-          highbd_filter_block_2rows(x0, z0, x1, z1, hfilter_vec);
+          highbd_filter_block_2rows(&x0, &z0, &x1, &z1, &hfilter_vec);
       _mm_storeu_si128((__m128i *)b, res);
 
       src += src_stride * 2;
@@ -879,7 +879,8 @@ static void highbd_bilinear_filter4xh(const uint16_t *src, int src_stride,
       const __m128i x = _mm_loadl_epi64((__m128i *)dst);
       const __m128i y = _mm_loadl_epi64((__m128i *)&dst[4]);
       const __m128i z = _mm_loadl_epi64((__m128i *)&dst[8]);
-      const __m128i res = highbd_filter_block_2rows(x, y, y, z, vfilter_vec);
+      const __m128i res =
+          highbd_filter_block_2rows(&x, &y, &y, &z, &vfilter_vec);
       _mm_storeu_si128((__m128i *)dst, res);
 
       dst += 8;
diff --git a/third_party/aom/aom_dsp/x86/quantize_avx_x86_64.asm b/third_party/aom/aom_dsp/x86/quantize_avx_x86_64.asm
index 216a0bd8f9..d6e15c4be5 100644
--- a/third_party/aom/aom_dsp/x86/quantize_avx_x86_64.asm
+++ b/third_party/aom/aom_dsp/x86/quantize_avx_x86_64.asm
@@ -126,7 +126,7 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   punpckhqdq                      m3, m3
   pmullw                         m13, m3                   ; dqc[i] = qc[i] * q
 
-  ; Store 16bit numbers as 32bit numbers in array pointed to by qcoeff
+  ; Store 16bit numbers as 32bit numbers in array pointed to by dqcoeff
   pcmpgtw                         m6, m5, m8
   punpckhwd                       m6, m8, m6
   pmovsxwd                       m11, m8
@@ -198,10 +198,7 @@ DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
   mova                            m4, [r2]            ; m4 = shift
   mov                             r4, dqcoeffmp
   mov                             r5, iscanmp
-%ifidn %1, b_32x32
-  psllw                           m4, 1
-%endif
-  pxor                            m5, m5                   ; m5 = dedicated zero
+  pxor                            m5, m5              ; m5 = dedicated zero
 
   DEFINE_ARGS coeff, ncoeff, d1, qcoeff, dqcoeff, iscan, d2, d3, d4, eob
 
@@ -255,9 +252,26 @@ DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   paddw                           m8, m6                   ; m8 += m6
   paddw                          m13, m11                  ; m13 += m11
+  %ifidn %1, b_32x32
+  pmullw                          m5, m8, m4               ; store the lower 16 bits of m8*qsh
+  %endif
   pmulhw                          m8, m4                   ; m8 = m8*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                           m8, 1
+  psrlw                           m5, 15
+  por                             m8, m5
+  %endif
   punpckhqdq                      m4, m4
+  %ifidn %1, b_32x32
+  pmullw                          m5, m13, m4              ; store the lower 16 bits of m13*qsh
+  %endif
   pmulhw                         m13, m4                   ; m13 = m13*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m13, 1
+  psrlw                           m5, 15
+  por                            m13, m5
+  pxor                            m5, m5                   ; reset m5 to zero register
+  %endif
   psignw                          m8, m9                   ; m8 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   pand                            m8, m7
@@ -289,7 +303,7 @@ DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
   psignw                         m13, m10
 %endif
 
-  ; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
+  ; store 16bit numbers as 32bit numbers in array pointed to by dqcoeff
   pcmpgtw                         m6, m5, m8
   punpckhwd                       m6, m8, m6
   pmovsxwd                       m11, m8
@@ -359,8 +373,23 @@ DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   paddw                          m14, m6                   ; m14 += m6
   paddw                          m13, m11                  ; m13 += m11
+  %ifidn %1, b_32x32
+  pmullw                          m5, m14, m4              ; store the lower 16 bits of m14*qsh
+  %endif
   pmulhw                         m14, m4                   ; m14 = m14*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m14, 1
+  psrlw                           m5, 15
+  por                            m14, m5
+  pmullw                          m5, m13, m4              ; store the lower 16 bits of m13*qsh
+  %endif
   pmulhw                         m13, m4                   ; m13 = m13*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m13, 1
+  psrlw                           m5, 15
+  por                            m13, m5
+  pxor                            m5, m5                   ; reset m5 to zero register
+  %endif
   psignw                         m14, m9                   ; m14 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   pand                           m14, m7
@@ -391,7 +420,7 @@ DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
   psignw                         m13, m10
 %endif
 
-  ; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
+  ; store 16bit numbers as 32bit numbers in array pointed to by dqcoeff
   pcmpgtw                         m6, m5, m14
   punpckhwd                       m6, m14, m6
   pmovsxwd                       m11, m14
diff --git a/third_party/aom/aom_dsp/x86/quantize_sse2.c b/third_party/aom/aom_dsp/x86/quantize_sse2.c
index d3de6e24db..ebef1fbac2 100644
--- a/third_party/aom/aom_dsp/x86/quantize_sse2.c
+++ b/third_party/aom/aom_dsp/x86/quantize_sse2.c
@@ -18,28 +18,6 @@
 #include "aom/aom_integer.h"
 #include "aom_dsp/x86/quantize_x86.h"
 
-static INLINE __m128i load_coefficients(const tran_low_t *coeff_ptr) {
-  assert(sizeof(tran_low_t) == 4);
-
-  return _mm_setr_epi16((int16_t)coeff_ptr[0], (int16_t)coeff_ptr[1],
-                        (int16_t)coeff_ptr[2], (int16_t)coeff_ptr[3],
-                        (int16_t)coeff_ptr[4], (int16_t)coeff_ptr[5],
-                        (int16_t)coeff_ptr[6], (int16_t)coeff_ptr[7]);
-}
-
-static INLINE void store_coefficients(__m128i coeff_vals,
-                                      tran_low_t *coeff_ptr) {
-  assert(sizeof(tran_low_t) == 4);
-
-  __m128i one = _mm_set1_epi16(1);
-  __m128i coeff_vals_hi = _mm_mulhi_epi16(coeff_vals, one);
-  __m128i coeff_vals_lo = _mm_mullo_epi16(coeff_vals, one);
-  __m128i coeff_vals_1 = _mm_unpacklo_epi16(coeff_vals_lo, coeff_vals_hi);
-  __m128i coeff_vals_2 = _mm_unpackhi_epi16(coeff_vals_lo, coeff_vals_hi);
-  _mm_store_si128((__m128i *)(coeff_ptr), coeff_vals_1);
-  _mm_store_si128((__m128i *)(coeff_ptr + 4), coeff_vals_2);
-}
-
 void aom_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          const int16_t *zbin_ptr, const int16_t *round_ptr,
                          const int16_t *quant_ptr,
diff --git a/third_party/aom/aom_dsp/x86/quantize_ssse3.c b/third_party/aom/aom_dsp/x86/quantize_ssse3.c
new file mode 100644
index 0000000000..25980a055a
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/quantize_ssse3.c
@@ -0,0 +1,192 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+#include <tmmintrin.h>
+#include <emmintrin.h>
+#include <xmmintrin.h>
+
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "aom_dsp/x86/quantize_x86.h"
+
+static INLINE void calculate_qcoeff_64x64(__m128i *coeff, const __m128i round,
+                                          const __m128i quant,
+                                          const __m128i *shift) {
+  __m128i tmp, qcoeff, tmp1;
+  qcoeff = _mm_adds_epi16(*coeff, round);
+  tmp = _mm_mulhi_epi16(qcoeff, quant);
+  qcoeff = _mm_add_epi16(tmp, qcoeff);
+  tmp = _mm_mullo_epi16(qcoeff, *shift);
+  tmp = _mm_srli_epi16(tmp, 14);
+  tmp1 = _mm_mulhi_epi16(qcoeff, *shift);
+  tmp1 = _mm_slli_epi16(tmp1, 2);
+  *coeff = _mm_or_si128(tmp, tmp1);
+}
+
+static INLINE void calculate_dqcoeff_and_store_64x64(const __m128i qcoeff,
+                                                     const __m128i dequant,
+                                                     const __m128i zero,
+                                                     tran_low_t *dqcoeff) {
+  // Un-sign to bias rounding like C.
+  const __m128i coeff = _mm_abs_epi16(qcoeff);
+
+  const __m128i sign_0 = _mm_unpacklo_epi16(zero, qcoeff);
+  const __m128i sign_1 = _mm_unpackhi_epi16(zero, qcoeff);
+
+  const __m128i low = _mm_mullo_epi16(coeff, dequant);
+  const __m128i high = _mm_mulhi_epi16(coeff, dequant);
+  __m128i dqcoeff32_0 = _mm_unpacklo_epi16(low, high);
+  __m128i dqcoeff32_1 = _mm_unpackhi_epi16(low, high);
+
+  // "Divide" by 4.
+  dqcoeff32_0 = _mm_srli_epi32(dqcoeff32_0, 2);
+  dqcoeff32_1 = _mm_srli_epi32(dqcoeff32_1, 2);
+
+  dqcoeff32_0 = _mm_sign_epi32(dqcoeff32_0, sign_0);
+  dqcoeff32_1 = _mm_sign_epi32(dqcoeff32_1, sign_1);
+
+  _mm_store_si128((__m128i *)(dqcoeff), dqcoeff32_0);
+  _mm_store_si128((__m128i *)(dqcoeff + 4), dqcoeff32_1);
+}
+
+void aom_quantize_b_64x64_ssse3(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
+                                const int16_t *zbin_ptr,
+                                const int16_t *round_ptr,
+                                const int16_t *quant_ptr,
+                                const int16_t *quant_shift_ptr,
+                                tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                                const int16_t *dequant_ptr, uint16_t *eob_ptr,
+                                const int16_t *scan, const int16_t *iscan) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i one = _mm_set1_epi16(1);
+  const __m128i two = _mm_set1_epi16(2);
+  int index;
+
+  __m128i zbin, round, quant, dequant, shift;
+  __m128i coeff0, coeff1, qcoeff0, qcoeff1;
+  __m128i cmp_mask0, cmp_mask1, all_zero;
+  __m128i eob = zero, eob0;
+
+  (void)scan;
+  (void)n_coeffs;
+
+  // Setup global values.
+  zbin = _mm_load_si128((const __m128i *)zbin_ptr);
+  round = _mm_load_si128((const __m128i *)round_ptr);
+  quant = _mm_load_si128((const __m128i *)quant_ptr);
+  dequant = _mm_load_si128((const __m128i *)dequant_ptr);
+  shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
+
+  // Shift with rounding.
+  zbin = _mm_add_epi16(zbin, two);
+  round = _mm_add_epi16(round, two);
+  zbin = _mm_srli_epi16(zbin, 2);
+  round = _mm_srli_epi16(round, 2);
+  zbin = _mm_sub_epi16(zbin, one);
+  // Do DC and first 15 AC.
+  coeff0 = load_coefficients(coeff_ptr);
+  coeff1 = load_coefficients(coeff_ptr + 8);
+
+  qcoeff0 = _mm_abs_epi16(coeff0);
+  qcoeff1 = _mm_abs_epi16(coeff1);
+
+  cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
+  zbin = _mm_unpackhi_epi64(zbin, zbin);
+  cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
+  all_zero = _mm_or_si128(cmp_mask0, cmp_mask1);
+  if (_mm_movemask_epi8(all_zero) == 0) {
+    _mm_store_si128((__m128i *)(qcoeff_ptr), zero);
+    _mm_store_si128((__m128i *)(qcoeff_ptr + 4), zero);
+    _mm_store_si128((__m128i *)(qcoeff_ptr + 8), zero);
+    _mm_store_si128((__m128i *)(qcoeff_ptr + 12), zero);
+    _mm_store_si128((__m128i *)(dqcoeff_ptr), zero);
+    _mm_store_si128((__m128i *)(dqcoeff_ptr + 4), zero);
+    _mm_store_si128((__m128i *)(dqcoeff_ptr + 8), zero);
+    _mm_store_si128((__m128i *)(dqcoeff_ptr + 12), zero);
+    round = _mm_unpackhi_epi64(round, round);
+    quant = _mm_unpackhi_epi64(quant, quant);
+    shift = _mm_unpackhi_epi64(shift, shift);
+    dequant = _mm_unpackhi_epi64(dequant, dequant);
+  } else {
+    calculate_qcoeff_64x64(&qcoeff0, round, quant, &shift);
+    round = _mm_unpackhi_epi64(round, round);
+    quant = _mm_unpackhi_epi64(quant, quant);
+    shift = _mm_unpackhi_epi64(shift, shift);
+    calculate_qcoeff_64x64(&qcoeff1, round, quant, &shift);
+
+    // Reinsert signs.
+    qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
+    qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
+
+    // Mask out zbin threshold coeffs.
+    qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
+    qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
+
+    store_coefficients(qcoeff0, qcoeff_ptr);
+    store_coefficients(qcoeff1, qcoeff_ptr + 8);
+
+    calculate_dqcoeff_and_store_64x64(qcoeff0, dequant, zero, dqcoeff_ptr);
+    dequant = _mm_unpackhi_epi64(dequant, dequant);
+    calculate_dqcoeff_and_store_64x64(qcoeff1, dequant, zero, dqcoeff_ptr + 8);
+
+    eob =
+        scan_for_eob(&qcoeff0, &qcoeff1, cmp_mask0, cmp_mask1, iscan, 0, zero);
+  }
+
+  // AC only loop.
+  for (index = 16; index < 1024; index += 16) {
+    coeff0 = load_coefficients(coeff_ptr + index);
+    coeff1 = load_coefficients(coeff_ptr + index + 8);
+
+    qcoeff0 = _mm_abs_epi16(coeff0);
+    qcoeff1 = _mm_abs_epi16(coeff1);
+
+    cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
+    cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
+
+    all_zero = _mm_or_si128(cmp_mask0, cmp_mask1);
+    if (_mm_movemask_epi8(all_zero) == 0) {
+      _mm_store_si128((__m128i *)(qcoeff_ptr + index), zero);
+      _mm_store_si128((__m128i *)(qcoeff_ptr + index + 4), zero);
+      _mm_store_si128((__m128i *)(qcoeff_ptr + index + 8), zero);
+      _mm_store_si128((__m128i *)(qcoeff_ptr + index + 12), zero);
+      _mm_store_si128((__m128i *)(dqcoeff_ptr + index), zero);
+      _mm_store_si128((__m128i *)(dqcoeff_ptr + index + 4), zero);
+      _mm_store_si128((__m128i *)(dqcoeff_ptr + index + 8), zero);
+      _mm_store_si128((__m128i *)(dqcoeff_ptr + index + 12), zero);
+      continue;
+    }
+    calculate_qcoeff_64x64(&qcoeff0, round, quant, &shift);
+    calculate_qcoeff_64x64(&qcoeff1, round, quant, &shift);
+
+    qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
+    qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
+
+    qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
+    qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
+
+    store_coefficients(qcoeff0, qcoeff_ptr + index);
+    store_coefficients(qcoeff1, qcoeff_ptr + index + 8);
+
+    calculate_dqcoeff_and_store_64x64(qcoeff0, dequant, zero,
+                                      dqcoeff_ptr + index);
+    calculate_dqcoeff_and_store_64x64(qcoeff1, dequant, zero,
+                                      dqcoeff_ptr + 8 + index);
+
+    eob0 = scan_for_eob(&qcoeff0, &qcoeff1, cmp_mask0, cmp_mask1, iscan, index,
+                        zero);
+    eob = _mm_max_epi16(eob, eob0);
+  }
+
+  *eob_ptr = accumulate_eob(eob);
+}
diff --git a/third_party/aom/aom_dsp/x86/quantize_ssse3_x86_64.asm b/third_party/aom/aom_dsp/x86/quantize_ssse3_x86_64.asm
index 39d4ca674c..fa616a6f1a 100644
--- a/third_party/aom/aom_dsp/x86/quantize_ssse3_x86_64.asm
+++ b/third_party/aom/aom_dsp/x86/quantize_ssse3_x86_64.asm
@@ -48,9 +48,6 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   mov                             r3, qcoeffmp
   mov                             r4, dqcoeffmp
   mov                             r5, iscanmp
-%ifidn %1, b_32x32
-  psllw                           m4, 1
-%endif
   pxor                            m5, m5                   ; m5 = dedicated zero
   DEFINE_ARGS coeff, ncoeff, d1, qcoeff, dqcoeff, iscan, d2, d3, d4, eob
   lea                         coeffq, [  coeffq+ncoeffq*4]
@@ -78,9 +75,26 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   paddw                           m8, m6                   ; m8 += m6
   paddw                          m13, m11                  ; m13 += m11
+  %ifidn %1, b_32x32
+  pmullw                          m5, m8, m4               ; store the lower 16 bits of m8*qsh
+  %endif
   pmulhw                          m8, m4                   ; m8 = m8*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                           m8, 1
+  psrlw                           m5, 15
+  por                             m8, m5
+  %endif
   punpckhqdq                      m4, m4
+  %ifidn %1, b_32x32
+  pmullw                          m5, m13, m4              ; store the lower 16 bits of m13*qsh
+  %endif
   pmulhw                         m13, m4                   ; m13 = m13*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m13, 1
+  psrlw                           m5, 15
+  por                            m13, m5
+  pxor                            m5, m5                   ; reset m5 to zero register
+  %endif
   psignw                          m8, m9                   ; m8 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   pand                            m8, m7
@@ -117,7 +131,7 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   psignw                          m8, m9
   psignw                         m13, m10
 %endif
-  ; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
+  ; store 16bit numbers as 32bit numbers in array pointed to by dqcoeff
   mova                            m11, m8
   mova                            m6, m8
   pcmpgtw                         m5, m8
@@ -169,12 +183,28 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   pmulhw                         m13, m11, m2              ; m13 = m11*q>>16
   paddw                          m14, m6                   ; m14 += m6
   paddw                          m13, m11                  ; m13 += m11
+  %ifidn %1, b_32x32
+  pmullw                          m5, m14, m4              ; store the lower 16 bits of m14*qsh
+  %endif
   pmulhw                         m14, m4                   ; m14 = m14*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m14, 1
+  psrlw                           m5, 15
+  por                            m14, m5
+  pmullw                          m5, m13, m4              ; store the lower 16 bits of m13*qsh
+  %endif
   pmulhw                         m13, m4                   ; m13 = m13*qsh>>16
+  %ifidn %1, b_32x32
+  psllw                          m13, 1
+  psrlw                           m5, 15
+  por                            m13, m5
+  pxor                            m5, m5                   ; reset m5 to zero register
+  %endif
   psignw                         m14, m9                   ; m14 = reinsert sign
   psignw                         m13, m10                  ; m13 = reinsert sign
   pand                           m14, m7
   pand                           m13, m12
+
   ; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
   pxor                           m11, m11
   mova                           m11, m14
@@ -207,7 +237,7 @@ cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
   psignw                         m13, m10
 %endif
 
-  ; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
+  ; store 16bit numbers as 32bit numbers in array pointed to by dqcoeff
   mova                           m11, m14
   mova                            m6, m14
   pcmpgtw                         m5, m14
diff --git a/third_party/aom/aom_dsp/x86/quantize_x86.h b/third_party/aom/aom_dsp/x86/quantize_x86.h
index 4eed7dd29a..3b94bb05e3 100644
--- a/third_party/aom/aom_dsp/x86/quantize_x86.h
+++ b/third_party/aom/aom_dsp/x86/quantize_x86.h
@@ -75,3 +75,23 @@ static INLINE int16_t accumulate_eob(__m128i eob) {
   eob = _mm_max_epi16(eob, eob_shuffled);
   return _mm_extract_epi16(eob, 1);
 }
+
+static INLINE __m128i load_coefficients(const tran_low_t *coeff_ptr) {
+  assert(sizeof(tran_low_t) == 4);
+  const __m128i coeff1 = _mm_load_si128((__m128i *)(coeff_ptr));
+  const __m128i coeff2 = _mm_load_si128((__m128i *)(coeff_ptr + 4));
+  return _mm_packs_epi32(coeff1, coeff2);
+}
+
+static INLINE void store_coefficients(__m128i coeff_vals,
+                                      tran_low_t *coeff_ptr) {
+  assert(sizeof(tran_low_t) == 4);
+
+  __m128i one = _mm_set1_epi16(1);
+  __m128i coeff_vals_hi = _mm_mulhi_epi16(coeff_vals, one);
+  __m128i coeff_vals_lo = _mm_mullo_epi16(coeff_vals, one);
+  __m128i coeff_vals_1 = _mm_unpacklo_epi16(coeff_vals_lo, coeff_vals_hi);
+  __m128i coeff_vals_2 = _mm_unpackhi_epi16(coeff_vals_lo, coeff_vals_hi);
+  _mm_store_si128((__m128i *)(coeff_ptr), coeff_vals_1);
+  _mm_store_si128((__m128i *)(coeff_ptr + 4), coeff_vals_2);
+}
diff --git a/third_party/aom/aom_dsp/x86/sse_avx2.c b/third_party/aom/aom_dsp/x86/sse_avx2.c
index 305dde5c08..42df9815cf 100644
--- a/third_party/aom/aom_dsp/x86/sse_avx2.c
+++ b/third_party/aom/aom_dsp/x86/sse_avx2.c
@@ -21,12 +21,11 @@ static INLINE void sse_w32_avx2(__m256i *sum, const uint8_t *a,
                                 const uint8_t *b) {
   const __m256i v_a0 = yy_loadu_256(a);
   const __m256i v_b0 = yy_loadu_256(b);
-  const __m256i v_a00_w = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(v_a0));
-  const __m256i v_a01_w =
-      _mm256_cvtepu8_epi16(_mm256_extracti128_si256(v_a0, 1));
-  const __m256i v_b00_w = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(v_b0));
-  const __m256i v_b01_w =
-      _mm256_cvtepu8_epi16(_mm256_extracti128_si256(v_b0, 1));
+  const __m256i zero = _mm256_setzero_si256();
+  const __m256i v_a00_w = _mm256_unpacklo_epi8(v_a0, zero);
+  const __m256i v_a01_w = _mm256_unpackhi_epi8(v_a0, zero);
+  const __m256i v_b00_w = _mm256_unpacklo_epi8(v_b0, zero);
+  const __m256i v_b01_w = _mm256_unpackhi_epi8(v_b0, zero);
   const __m256i v_d00_w = _mm256_sub_epi16(v_a00_w, v_b00_w);
   const __m256i v_d01_w = _mm256_sub_epi16(v_a01_w, v_b01_w);
   *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d00_w, v_d00_w));
@@ -35,43 +34,76 @@ static INLINE void sse_w32_avx2(__m256i *sum, const uint8_t *a,
 
 static INLINE int64_t summary_all_avx2(const __m256i *sum_all) {
   int64_t sum;
-  const __m256i sum0_4x64 =
-      _mm256_cvtepu32_epi64(_mm256_castsi256_si128(*sum_all));
-  const __m256i sum1_4x64 =
-      _mm256_cvtepu32_epi64(_mm256_extracti128_si256(*sum_all, 1));
+  __m256i zero = _mm256_setzero_si256();
+  const __m256i sum0_4x64 = _mm256_unpacklo_epi32(*sum_all, zero);
+  const __m256i sum1_4x64 = _mm256_unpackhi_epi32(*sum_all, zero);
   const __m256i sum_4x64 = _mm256_add_epi64(sum0_4x64, sum1_4x64);
   const __m128i sum_2x64 = _mm_add_epi64(_mm256_castsi256_si128(sum_4x64),
                                          _mm256_extracti128_si256(sum_4x64, 1));
   const __m128i sum_1x64 = _mm_add_epi64(sum_2x64, _mm_srli_si128(sum_2x64, 8));
+  xx_storel_64(&sum, sum_1x64);
+  return sum;
+}
+
+static INLINE void summary_32_avx2(const __m256i *sum32, __m256i *sum) {
+  const __m256i sum0_4x64 =
+      _mm256_cvtepu32_epi64(_mm256_castsi256_si128(*sum32));
+  const __m256i sum1_4x64 =
+      _mm256_cvtepu32_epi64(_mm256_extracti128_si256(*sum32, 1));
+  const __m256i sum_4x64 = _mm256_add_epi64(sum0_4x64, sum1_4x64);
+  *sum = _mm256_add_epi64(*sum, sum_4x64);
+}
+
+static INLINE int64_t summary_4x64_avx2(const __m256i sum_4x64) {
+  int64_t sum;
+  const __m128i sum_2x64 = _mm_add_epi64(_mm256_castsi256_si128(sum_4x64),
+                                         _mm256_extracti128_si256(sum_4x64, 1));
+  const __m128i sum_1x64 = _mm_add_epi64(sum_2x64, _mm_srli_si128(sum_2x64, 8));
 
   xx_storel_64(&sum, sum_1x64);
   return sum;
 }
 
+static INLINE void sse_w4x4_avx2(const uint8_t *a, int a_stride,
+                                 const uint8_t *b, int b_stride, __m256i *sum) {
+  const __m128i v_a0 = xx_loadl_32(a);
+  const __m128i v_a1 = xx_loadl_32(a + a_stride);
+  const __m128i v_a2 = xx_loadl_32(a + a_stride * 2);
+  const __m128i v_a3 = xx_loadl_32(a + a_stride * 3);
+  const __m128i v_b0 = xx_loadl_32(b);
+  const __m128i v_b1 = xx_loadl_32(b + b_stride);
+  const __m128i v_b2 = xx_loadl_32(b + b_stride * 2);
+  const __m128i v_b3 = xx_loadl_32(b + b_stride * 3);
+  const __m128i v_a0123 = _mm_unpacklo_epi64(_mm_unpacklo_epi32(v_a0, v_a1),
+                                             _mm_unpacklo_epi32(v_a2, v_a3));
+  const __m128i v_b0123 = _mm_unpacklo_epi64(_mm_unpacklo_epi32(v_b0, v_b1),
+                                             _mm_unpacklo_epi32(v_b2, v_b3));
+  const __m256i v_a_w = _mm256_cvtepu8_epi16(v_a0123);
+  const __m256i v_b_w = _mm256_cvtepu8_epi16(v_b0123);
+  const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d_w, v_d_w));
+}
+static INLINE void sse_w8x2_avx2(const uint8_t *a, int a_stride,
+                                 const uint8_t *b, int b_stride, __m256i *sum) {
+  const __m128i v_a0 = xx_loadl_64(a);
+  const __m128i v_a1 = xx_loadl_64(a + a_stride);
+  const __m128i v_b0 = xx_loadl_64(b);
+  const __m128i v_b1 = xx_loadl_64(b + b_stride);
+  const __m256i v_a_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(v_a0, v_a1));
+  const __m256i v_b_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(v_b0, v_b1));
+  const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d_w, v_d_w));
+}
 int64_t aom_sse_avx2(const uint8_t *a, int a_stride, const uint8_t *b,
                      int b_stride, int width, int height) {
   int32_t y = 0;
   int64_t sse = 0;
   __m256i sum = _mm256_setzero_si256();
+  __m256i zero = _mm256_setzero_si256();
   switch (width) {
     case 4:
       do {
-        const __m128i v_a0 = xx_loadl_32(a);
-        const __m128i v_a1 = xx_loadl_32(a + a_stride);
-        const __m128i v_a2 = xx_loadl_32(a + a_stride * 2);
-        const __m128i v_a3 = xx_loadl_32(a + a_stride * 3);
-        const __m128i v_b0 = xx_loadl_32(b);
-        const __m128i v_b1 = xx_loadl_32(b + b_stride);
-        const __m128i v_b2 = xx_loadl_32(b + b_stride * 2);
-        const __m128i v_b3 = xx_loadl_32(b + b_stride * 3);
-        const __m128i v_a0123 = _mm_unpacklo_epi64(
-            _mm_unpacklo_epi32(v_a0, v_a1), _mm_unpacklo_epi32(v_a2, v_a3));
-        const __m128i v_b0123 = _mm_unpacklo_epi64(
-            _mm_unpacklo_epi32(v_b0, v_b1), _mm_unpacklo_epi32(v_b2, v_b3));
-        const __m256i v_a_w = _mm256_cvtepu8_epi16(v_a0123);
-        const __m256i v_b_w = _mm256_cvtepu8_epi16(v_b0123);
-        const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
-        sum = _mm256_add_epi32(sum, _mm256_madd_epi16(v_d_w, v_d_w));
+        sse_w4x4_avx2(a, a_stride, b, b_stride, &sum);
         a += a_stride << 2;
         b += b_stride << 2;
         y += 4;
@@ -80,16 +112,7 @@ int64_t aom_sse_avx2(const uint8_t *a, int a_stride, const uint8_t *b,
       break;
     case 8:
       do {
-        const __m128i v_a0 = xx_loadl_64(a);
-        const __m128i v_a1 = xx_loadl_64(a + a_stride);
-        const __m128i v_b0 = xx_loadl_64(b);
-        const __m128i v_b1 = xx_loadl_64(b + b_stride);
-        const __m256i v_a_w =
-            _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(v_a0, v_a1));
-        const __m256i v_b_w =
-            _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(v_b0, v_b1));
-        const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
-        sum = _mm256_add_epi32(sum, _mm256_madd_epi16(v_d_w, v_d_w));
+        sse_w8x2_avx2(a, a_stride, b, b_stride, &sum);
         a += a_stride << 1;
         b += b_stride << 1;
         y += 2;
@@ -99,14 +122,26 @@ int64_t aom_sse_avx2(const uint8_t *a, int a_stride, const uint8_t *b,
     case 16:
       do {
         const __m128i v_a0 = xx_loadu_128(a);
+        const __m128i v_a1 = xx_loadu_128(a + a_stride);
         const __m128i v_b0 = xx_loadu_128(b);
-        const __m256i v_a_w = _mm256_cvtepu8_epi16(v_a0);
-        const __m256i v_b_w = _mm256_cvtepu8_epi16(v_b0);
-        const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
-        sum = _mm256_add_epi32(sum, _mm256_madd_epi16(v_d_w, v_d_w));
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        const __m128i v_b1 = xx_loadu_128(b + b_stride);
+        const __m256i v_a =
+            _mm256_insertf128_si256(_mm256_castsi128_si256(v_a0), v_a1, 0x01);
+        const __m256i v_b =
+            _mm256_insertf128_si256(_mm256_castsi128_si256(v_b0), v_b1, 0x01);
+        const __m256i v_al = _mm256_unpacklo_epi8(v_a, zero);
+        const __m256i v_au = _mm256_unpackhi_epi8(v_a, zero);
+        const __m256i v_bl = _mm256_unpacklo_epi8(v_b, zero);
+        const __m256i v_bu = _mm256_unpackhi_epi8(v_b, zero);
+        const __m256i v_asub = _mm256_sub_epi16(v_al, v_bl);
+        const __m256i v_bsub = _mm256_sub_epi16(v_au, v_bu);
+        const __m256i temp =
+            _mm256_add_epi32(_mm256_madd_epi16(v_asub, v_asub),
+                             _mm256_madd_epi16(v_bsub, v_bsub));
+        sum = _mm256_add_epi32(sum, temp);
+        a += a_stride << 1;
+        b += b_stride << 1;
+        y += 2;
       } while (y < height);
       sse = summary_all_avx2(&sum);
       break;
@@ -141,7 +176,36 @@ int64_t aom_sse_avx2(const uint8_t *a, int a_stride, const uint8_t *b,
       } while (y < height);
       sse = summary_all_avx2(&sum);
       break;
-    default: break;
+    default:
+      if ((width & 0x07) == 0) {
+        do {
+          int i = 0;
+          do {
+            sse_w8x2_avx2(a + i, a_stride, b + i, b_stride, &sum);
+            i += 8;
+          } while (i < width);
+          a += a_stride << 1;
+          b += b_stride << 1;
+          y += 2;
+        } while (y < height);
+      } else {
+        do {
+          int i = 0;
+          do {
+            sse_w8x2_avx2(a + i, a_stride, b + i, b_stride, &sum);
+            const uint8_t *a2 = a + i + (a_stride << 1);
+            const uint8_t *b2 = b + i + (b_stride << 1);
+            sse_w8x2_avx2(a2, a_stride, b2, b_stride, &sum);
+            i += 8;
+          } while (i + 4 < width);
+          sse_w4x4_avx2(a + i, a_stride, b + i, b_stride, &sum);
+          a += a_stride << 2;
+          b += b_stride << 2;
+          y += 4;
+        } while (y < height);
+      }
+      sse = summary_all_avx2(&sum);
+      break;
   }
 
   return sse;
@@ -155,6 +219,33 @@ static INLINE void highbd_sse_w16_avx2(__m256i *sum, const uint16_t *a,
   *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d_w, v_d_w));
 }
 
+static INLINE void highbd_sse_w4x4_avx2(__m256i *sum, const uint16_t *a,
+                                        int a_stride, const uint16_t *b,
+                                        int b_stride) {
+  const __m128i v_a0 = xx_loadl_64(a);
+  const __m128i v_a1 = xx_loadl_64(a + a_stride);
+  const __m128i v_a2 = xx_loadl_64(a + a_stride * 2);
+  const __m128i v_a3 = xx_loadl_64(a + a_stride * 3);
+  const __m128i v_b0 = xx_loadl_64(b);
+  const __m128i v_b1 = xx_loadl_64(b + b_stride);
+  const __m128i v_b2 = xx_loadl_64(b + b_stride * 2);
+  const __m128i v_b3 = xx_loadl_64(b + b_stride * 3);
+  const __m256i v_a_w = yy_set_m128i(_mm_unpacklo_epi64(v_a0, v_a1),
+                                     _mm_unpacklo_epi64(v_a2, v_a3));
+  const __m256i v_b_w = yy_set_m128i(_mm_unpacklo_epi64(v_b0, v_b1),
+                                     _mm_unpacklo_epi64(v_b2, v_b3));
+  const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d_w, v_d_w));
+}
+
+static INLINE void highbd_sse_w8x2_avx2(__m256i *sum, const uint16_t *a,
+                                        int a_stride, const uint16_t *b,
+                                        int b_stride) {
+  const __m256i v_a_w = yy_loadu2_128(a + a_stride, a);
+  const __m256i v_b_w = yy_loadu2_128(b + b_stride, b);
+  const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm256_add_epi32(*sum, _mm256_madd_epi16(v_d_w, v_d_w));
+}
 int64_t aom_highbd_sse_avx2(const uint8_t *a8, int a_stride, const uint8_t *b8,
                             int b_stride, int width, int height) {
   int32_t y = 0;
@@ -165,20 +256,7 @@ int64_t aom_highbd_sse_avx2(const uint8_t *a8, int a_stride, const uint8_t *b8,
   switch (width) {
     case 4:
       do {
-        const __m128i v_a0 = xx_loadl_64(a);
-        const __m128i v_a1 = xx_loadl_64(a + a_stride);
-        const __m128i v_a2 = xx_loadl_64(a + a_stride * 2);
-        const __m128i v_a3 = xx_loadl_64(a + a_stride * 3);
-        const __m128i v_b0 = xx_loadl_64(b);
-        const __m128i v_b1 = xx_loadl_64(b + b_stride);
-        const __m128i v_b2 = xx_loadl_64(b + b_stride * 2);
-        const __m128i v_b3 = xx_loadl_64(b + b_stride * 3);
-        const __m256i v_a_w = yy_set_m128i(_mm_unpacklo_epi64(v_a0, v_a1),
-                                           _mm_unpacklo_epi64(v_a2, v_a3));
-        const __m256i v_b_w = yy_set_m128i(_mm_unpacklo_epi64(v_b0, v_b1),
-                                           _mm_unpacklo_epi64(v_b2, v_b3));
-        const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
-        sum = _mm256_add_epi32(sum, _mm256_madd_epi16(v_d_w, v_d_w));
+        highbd_sse_w4x4_avx2(&sum, a, a_stride, b, b_stride);
         a += a_stride << 2;
         b += b_stride << 2;
         y += 4;
@@ -187,10 +265,7 @@ int64_t aom_highbd_sse_avx2(const uint8_t *a8, int a_stride, const uint8_t *b8,
       break;
     case 8:
       do {
-        const __m256i v_a_w = yy_loadu2_128(a + a_stride, a);
-        const __m256i v_b_w = yy_loadu2_128(b + b_stride, b);
-        const __m256i v_d_w = _mm256_sub_epi16(v_a_w, v_b_w);
-        sum = _mm256_add_epi32(sum, _mm256_madd_epi16(v_d_w, v_d_w));
+        highbd_sse_w8x2_avx2(&sum, a, a_stride, b, b_stride);
         a += a_stride << 1;
         b += b_stride << 1;
         y += 2;
@@ -208,43 +283,98 @@ int64_t aom_highbd_sse_avx2(const uint8_t *a8, int a_stride, const uint8_t *b8,
       break;
     case 32:
       do {
-        highbd_sse_w16_avx2(&sum, a, b);
-        highbd_sse_w16_avx2(&sum, a + 16, b + 16);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m256i sum32 = _mm256_setzero_si256();
+        do {
+          highbd_sse_w16_avx2(&sum32, a, b);
+          highbd_sse_w16_avx2(&sum32, a + 16, b + 16);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 64 && l < (height - y));
+        summary_32_avx2(&sum32, &sum);
+        y += 64;
       } while (y < height);
-      sse = summary_all_avx2(&sum);
+      sse = summary_4x64_avx2(sum);
       break;
     case 64:
       do {
-        highbd_sse_w16_avx2(&sum, a, b);
-        highbd_sse_w16_avx2(&sum, a + 16 * 1, b + 16 * 1);
-        highbd_sse_w16_avx2(&sum, a + 16 * 2, b + 16 * 2);
-        highbd_sse_w16_avx2(&sum, a + 16 * 3, b + 16 * 3);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m256i sum32 = _mm256_setzero_si256();
+        do {
+          highbd_sse_w16_avx2(&sum32, a, b);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 1, b + 16 * 1);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 2, b + 16 * 2);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 3, b + 16 * 3);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 32 && l < (height - y));
+        summary_32_avx2(&sum32, &sum);
+        y += 32;
       } while (y < height);
-      sse = summary_all_avx2(&sum);
+      sse = summary_4x64_avx2(sum);
       break;
     case 128:
       do {
-        highbd_sse_w16_avx2(&sum, a, b);
-        highbd_sse_w16_avx2(&sum, a + 16 * 1, b + 16 * 1);
-        highbd_sse_w16_avx2(&sum, a + 16 * 2, b + 16 * 2);
-        highbd_sse_w16_avx2(&sum, a + 16 * 3, b + 16 * 3);
-        highbd_sse_w16_avx2(&sum, a + 16 * 4, b + 16 * 4);
-        highbd_sse_w16_avx2(&sum, a + 16 * 5, b + 16 * 5);
-        highbd_sse_w16_avx2(&sum, a + 16 * 6, b + 16 * 6);
-        highbd_sse_w16_avx2(&sum, a + 16 * 7, b + 16 * 7);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m256i sum32 = _mm256_setzero_si256();
+        do {
+          highbd_sse_w16_avx2(&sum32, a, b);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 1, b + 16 * 1);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 2, b + 16 * 2);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 3, b + 16 * 3);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 4, b + 16 * 4);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 5, b + 16 * 5);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 6, b + 16 * 6);
+          highbd_sse_w16_avx2(&sum32, a + 16 * 7, b + 16 * 7);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 16 && l < (height - y));
+        summary_32_avx2(&sum32, &sum);
+        y += 16;
       } while (y < height);
-      sse = summary_all_avx2(&sum);
+      sse = summary_4x64_avx2(sum);
+      break;
+    default:
+      if (width & 0x7) {
+        do {
+          int i = 0;
+          __m256i sum32 = _mm256_setzero_si256();
+          do {
+            highbd_sse_w8x2_avx2(&sum32, a + i, a_stride, b + i, b_stride);
+            const uint16_t *a2 = a + i + (a_stride << 1);
+            const uint16_t *b2 = b + i + (b_stride << 1);
+            highbd_sse_w8x2_avx2(&sum32, a2, a_stride, b2, b_stride);
+            i += 8;
+          } while (i + 4 < width);
+          highbd_sse_w4x4_avx2(&sum32, a + i, a_stride, b + i, b_stride);
+          summary_32_avx2(&sum32, &sum);
+          a += a_stride << 2;
+          b += b_stride << 2;
+          y += 4;
+        } while (y < height);
+      } else {
+        do {
+          int l = 0;
+          __m256i sum32 = _mm256_setzero_si256();
+          do {
+            int i = 0;
+            do {
+              highbd_sse_w8x2_avx2(&sum32, a + i, a_stride, b + i, b_stride);
+              i += 8;
+            } while (i < width);
+            a += a_stride << 1;
+            b += b_stride << 1;
+            l += 2;
+          } while (l < 8 && l < (height - y));
+          summary_32_avx2(&sum32, &sum);
+          y += 8;
+        } while (y < height);
+      }
+      sse = summary_4x64_avx2(sum);
       break;
-    default: break;
   }
   return sse;
 }
diff --git a/third_party/aom/aom_dsp/x86/sse_sse4.c b/third_party/aom/aom_dsp/x86/sse_sse4.c
index 8b5af84691..0d45003e92 100644
--- a/third_party/aom/aom_dsp/x86/sse_sse4.c
+++ b/third_party/aom/aom_dsp/x86/sse_sse4.c
@@ -28,6 +28,13 @@ static INLINE int64_t summary_all_sse4(const __m128i *sum_all) {
   return sum;
 }
 
+static INLINE void summary_32_sse4(const __m128i *sum32, __m128i *sum64) {
+  const __m128i sum0 = _mm_cvtepu32_epi64(*sum32);
+  const __m128i sum1 = _mm_cvtepu32_epi64(_mm_srli_si128(*sum32, 8));
+  *sum64 = _mm_add_epi64(sum0, *sum64);
+  *sum64 = _mm_add_epi64(sum1, *sum64);
+}
+
 static INLINE void sse_w16_sse4_1(__m128i *sum, const uint8_t *a,
                                   const uint8_t *b) {
   const __m128i v_a0 = xx_loadu_128(a);
@@ -42,6 +49,28 @@ static INLINE void sse_w16_sse4_1(__m128i *sum, const uint8_t *a,
   *sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d01_w, v_d01_w));
 }
 
+static INLINE void aom_sse4x2_sse4_1(const uint8_t *a, int a_stride,
+                                     const uint8_t *b, int b_stride,
+                                     __m128i *sum) {
+  const __m128i v_a0 = xx_loadl_32(a);
+  const __m128i v_a1 = xx_loadl_32(a + a_stride);
+  const __m128i v_b0 = xx_loadl_32(b);
+  const __m128i v_b1 = xx_loadl_32(b + b_stride);
+  const __m128i v_a_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_a0, v_a1));
+  const __m128i v_b_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_b0, v_b1));
+  const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d_w, v_d_w));
+}
+static INLINE void aom_sse8_sse4_1(const uint8_t *a, const uint8_t *b,
+                                   __m128i *sum) {
+  const __m128i v_a0 = xx_loadl_64(a);
+  const __m128i v_b0 = xx_loadl_64(b);
+  const __m128i v_a_w = _mm_cvtepu8_epi16(v_a0);
+  const __m128i v_b_w = _mm_cvtepu8_epi16(v_b0);
+  const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d_w, v_d_w));
+}
+
 int64_t aom_sse_sse4_1(const uint8_t *a, int a_stride, const uint8_t *b,
                        int b_stride, int width, int height) {
   int y = 0;
@@ -50,14 +79,7 @@ int64_t aom_sse_sse4_1(const uint8_t *a, int a_stride, const uint8_t *b,
   switch (width) {
     case 4:
       do {
-        const __m128i v_a0 = xx_loadl_32(a);
-        const __m128i v_a1 = xx_loadl_32(a + a_stride);
-        const __m128i v_b0 = xx_loadl_32(b);
-        const __m128i v_b1 = xx_loadl_32(b + b_stride);
-        const __m128i v_a_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_a0, v_a1));
-        const __m128i v_b_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_b0, v_b1));
-        const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
-        sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
+        aom_sse4x2_sse4_1(a, a_stride, b, b_stride, &sum);
         a += a_stride << 1;
         b += b_stride << 1;
         y += 2;
@@ -66,12 +88,7 @@ int64_t aom_sse_sse4_1(const uint8_t *a, int a_stride, const uint8_t *b,
       break;
     case 8:
       do {
-        const __m128i v_a0 = xx_loadl_64(a);
-        const __m128i v_b0 = xx_loadl_64(b);
-        const __m128i v_a_w = _mm_cvtepu8_epi16(v_a0);
-        const __m128i v_b_w = _mm_cvtepu8_epi16(v_b0);
-        const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
-        sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
+        aom_sse8_sse4_1(a, b, &sum);
         a += a_stride;
         b += b_stride;
         y += 1;
@@ -125,12 +142,52 @@ int64_t aom_sse_sse4_1(const uint8_t *a, int a_stride, const uint8_t *b,
       } while (y < height);
       sse = summary_all_sse4(&sum);
       break;
-    default: break;
+    default:
+      if (width & 0x07) {
+        do {
+          int i = 0;
+          do {
+            aom_sse8_sse4_1(a + i, b + i, &sum);
+            aom_sse8_sse4_1(a + i + a_stride, b + i + b_stride, &sum);
+            i += 8;
+          } while (i + 4 < width);
+          aom_sse4x2_sse4_1(a + i, a_stride, b + i, b_stride, &sum);
+          a += (a_stride << 1);
+          b += (b_stride << 1);
+          y += 2;
+        } while (y < height);
+      } else {
+        do {
+          int i = 0;
+          do {
+            aom_sse8_sse4_1(a + i, b + i, &sum);
+            i += 8;
+          } while (i < width);
+          a += a_stride;
+          b += b_stride;
+          y += 1;
+        } while (y < height);
+      }
+      sse = summary_all_sse4(&sum);
+      break;
   }
 
   return sse;
 }
 
+static INLINE void highbd_sse_w4x2_sse4_1(__m128i *sum, const uint16_t *a,
+                                          int a_stride, const uint16_t *b,
+                                          int b_stride) {
+  const __m128i v_a0 = xx_loadl_64(a);
+  const __m128i v_a1 = xx_loadl_64(a + a_stride);
+  const __m128i v_b0 = xx_loadl_64(b);
+  const __m128i v_b1 = xx_loadl_64(b + b_stride);
+  const __m128i v_a_w = _mm_unpacklo_epi64(v_a0, v_a1);
+  const __m128i v_b_w = _mm_unpacklo_epi64(v_b0, v_b1);
+  const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
+  *sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d_w, v_d_w));
+}
+
 static INLINE void highbd_sse_w8_sse4_1(__m128i *sum, const uint16_t *a,
                                         const uint16_t *b) {
   const __m128i v_a_w = xx_loadu_128(a);
@@ -150,14 +207,7 @@ int64_t aom_highbd_sse_sse4_1(const uint8_t *a8, int a_stride,
   switch (width) {
     case 4:
       do {
-        const __m128i v_a0 = xx_loadl_64(a);
-        const __m128i v_a1 = xx_loadl_64(a + a_stride);
-        const __m128i v_b0 = xx_loadl_64(b);
-        const __m128i v_b1 = xx_loadl_64(b + b_stride);
-        const __m128i v_a_w = _mm_unpacklo_epi64(v_a0, v_a1);
-        const __m128i v_b_w = _mm_unpacklo_epi64(v_b0, v_b1);
-        const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
-        sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
+        highbd_sse_w4x2_sse4_1(&sum, a, a_stride, b, b_stride);
         a += a_stride << 1;
         b += b_stride << 1;
         y += 2;
@@ -175,67 +225,126 @@ int64_t aom_highbd_sse_sse4_1(const uint8_t *a8, int a_stride,
       break;
     case 16:
       do {
-        highbd_sse_w8_sse4_1(&sum, a, b);
-        highbd_sse_w8_sse4_1(&sum, a + 8, b + 8);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m128i sum32 = _mm_setzero_si128();
+        do {
+          highbd_sse_w8_sse4_1(&sum32, a, b);
+          highbd_sse_w8_sse4_1(&sum32, a + 8, b + 8);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 64 && l < (height - y));
+        summary_32_sse4(&sum32, &sum);
+        y += 64;
       } while (y < height);
-      sse = summary_all_sse4(&sum);
+      xx_storel_64(&sse, _mm_add_epi64(sum, _mm_srli_si128(sum, 8)));
       break;
     case 32:
       do {
-        highbd_sse_w8_sse4_1(&sum, a, b);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m128i sum32 = _mm_setzero_si128();
+        do {
+          highbd_sse_w8_sse4_1(&sum32, a, b);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 1, b + 8 * 1);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 2, b + 8 * 2);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 3, b + 8 * 3);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 32 && l < (height - y));
+        summary_32_sse4(&sum32, &sum);
+        y += 32;
       } while (y < height);
-      sse = summary_all_sse4(&sum);
+      xx_storel_64(&sse, _mm_add_epi64(sum, _mm_srli_si128(sum, 8)));
       break;
     case 64:
       do {
-        highbd_sse_w8_sse4_1(&sum, a, b);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 4, b + 8 * 4);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 5, b + 8 * 5);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 6, b + 8 * 6);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 7, b + 8 * 7);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m128i sum32 = _mm_setzero_si128();
+        do {
+          highbd_sse_w8_sse4_1(&sum32, a, b);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 1, b + 8 * 1);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 2, b + 8 * 2);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 3, b + 8 * 3);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 4, b + 8 * 4);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 5, b + 8 * 5);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 6, b + 8 * 6);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 7, b + 8 * 7);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 16 && l < (height - y));
+        summary_32_sse4(&sum32, &sum);
+        y += 16;
       } while (y < height);
-      sse = summary_all_sse4(&sum);
+      xx_storel_64(&sse, _mm_add_epi64(sum, _mm_srli_si128(sum, 8)));
       break;
     case 128:
       do {
-        highbd_sse_w8_sse4_1(&sum, a, b);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 4, b + 8 * 4);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 5, b + 8 * 5);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 6, b + 8 * 6);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 7, b + 8 * 7);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 8, b + 8 * 8);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 9, b + 8 * 9);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 10, b + 8 * 10);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 11, b + 8 * 11);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 12, b + 8 * 12);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 13, b + 8 * 13);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 14, b + 8 * 14);
-        highbd_sse_w8_sse4_1(&sum, a + 8 * 15, b + 8 * 15);
-        a += a_stride;
-        b += b_stride;
-        y += 1;
+        int l = 0;
+        __m128i sum32 = _mm_setzero_si128();
+        do {
+          highbd_sse_w8_sse4_1(&sum32, a, b);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 1, b + 8 * 1);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 2, b + 8 * 2);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 3, b + 8 * 3);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 4, b + 8 * 4);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 5, b + 8 * 5);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 6, b + 8 * 6);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 7, b + 8 * 7);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 8, b + 8 * 8);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 9, b + 8 * 9);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 10, b + 8 * 10);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 11, b + 8 * 11);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 12, b + 8 * 12);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 13, b + 8 * 13);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 14, b + 8 * 14);
+          highbd_sse_w8_sse4_1(&sum32, a + 8 * 15, b + 8 * 15);
+          a += a_stride;
+          b += b_stride;
+          l += 1;
+        } while (l < 8 && l < (height - y));
+        summary_32_sse4(&sum32, &sum);
+        y += 8;
       } while (y < height);
-      sse = summary_all_sse4(&sum);
+      xx_storel_64(&sse, _mm_add_epi64(sum, _mm_srli_si128(sum, 8)));
+      break;
+    default:
+      if (width & 0x7) {
+        do {
+          __m128i sum32 = _mm_setzero_si128();
+          int i = 0;
+          do {
+            highbd_sse_w8_sse4_1(&sum32, a + i, b + i);
+            highbd_sse_w8_sse4_1(&sum32, a + i + a_stride, b + i + b_stride);
+            i += 8;
+          } while (i + 4 < width);
+          highbd_sse_w4x2_sse4_1(&sum32, a + i, a_stride, b + i, b_stride);
+          a += (a_stride << 1);
+          b += (b_stride << 1);
+          y += 2;
+          summary_32_sse4(&sum32, &sum);
+        } while (y < height);
+      } else {
+        do {
+          int l = 0;
+          __m128i sum32 = _mm_setzero_si128();
+          do {
+            int i = 0;
+            do {
+              highbd_sse_w8_sse4_1(&sum32, a + i, b + i);
+              i += 8;
+            } while (i < width);
+            a += a_stride;
+            b += b_stride;
+            l += 1;
+          } while (l < 8 && l < (height - y));
+          summary_32_sse4(&sum32, &sum);
+          y += 8;
+        } while (y < height);
+      }
+      xx_storel_64(&sse, _mm_add_epi64(sum, _mm_srli_si128(sum, 8)));
       break;
-    default: break;
   }
   return sse;
 }
diff --git a/third_party/aom/aom_dsp/x86/ssim_sse2_x86_64.asm b/third_party/aom/aom_dsp/x86/ssim_sse2_x86_64.asm
new file mode 100644
index 0000000000..6d9b5a12f1
--- /dev/null
+++ b/third_party/aom/aom_dsp/x86/ssim_sse2_x86_64.asm
@@ -0,0 +1,222 @@
+;
+; Copyright (c) 2016, Alliance for Open Media. All rights reserved
+;
+; This source code is subject to the terms of the BSD 2 Clause License and
+; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+; was not distributed with this source code in the LICENSE file, you can
+; obtain it at www.aomedia.org/license/software. If the Alliance for Open
+; Media Patent License 1.0 was not distributed with this source code in the
+; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+;
+
+;
+
+%include "aom_ports/x86_abi_support.asm"
+
+; tabulate_ssim - sums sum_s,sum_r,sum_sq_s,sum_sq_r, sum_sxr
+%macro TABULATE_SSIM 0
+        paddusw         xmm15, xmm3  ; sum_s
+        paddusw         xmm14, xmm4  ; sum_r
+        movdqa          xmm1, xmm3
+        pmaddwd         xmm1, xmm1
+        paddd           xmm13, xmm1 ; sum_sq_s
+        movdqa          xmm2, xmm4
+        pmaddwd         xmm2, xmm2
+        paddd           xmm12, xmm2 ; sum_sq_r
+        pmaddwd         xmm3, xmm4
+        paddd           xmm11, xmm3  ; sum_sxr
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_Q 1
+        movdqa          xmm2,%1
+        punpckldq       %1,xmm0
+        punpckhdq       xmm2,xmm0
+        paddq           %1,xmm2
+        movdqa          xmm2,%1
+        punpcklqdq      %1,xmm0
+        punpckhqdq      xmm2,xmm0
+        paddq           %1,xmm2
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_W 1
+        movdqa          xmm1, %1
+        punpcklwd       %1,xmm0
+        punpckhwd       xmm1,xmm0
+        paddd           %1, xmm1
+        SUM_ACROSS_Q    %1
+%endmacro
+
+SECTION .text
+
+;void ssim_parms_sse2(
+;    unsigned char *s,
+;    int sp,
+;    unsigned char *r,
+;    int rp
+;    uint32_t *sum_s,
+;    uint32_t *sum_r,
+;    uint32_t *sum_sq_s,
+;    uint32_t *sum_sq_r,
+;    uint32_t *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(aom_ssim_parms_16x16_sse2) PRIVATE
+sym(aom_ssim_parms_16x16_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 15
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0) ;s
+    mov             rcx,        arg(1) ;sp
+    mov             rdi,        arg(2) ;r
+    mov             rax,        arg(3) ;rp
+
+    pxor            xmm0, xmm0
+    pxor            xmm15,xmm15  ;sum_s
+    pxor            xmm14,xmm14  ;sum_r
+    pxor            xmm13,xmm13  ;sum_sq_s
+    pxor            xmm12,xmm12  ;sum_sq_r
+    pxor            xmm11,xmm11  ;sum_sxr
+
+    mov             rdx, 16      ;row counter
+.NextRow:
+
+    ;grab source and reference pixels
+    movdqu          xmm5, [rsi]
+    movdqu          xmm6, [rdi]
+    movdqa          xmm3, xmm5
+    movdqa          xmm4, xmm6
+    punpckhbw       xmm3, xmm0 ; high_s
+    punpckhbw       xmm4, xmm0 ; high_r
+
+    TABULATE_SSIM
+
+    movdqa          xmm3, xmm5
+    movdqa          xmm4, xmm6
+    punpcklbw       xmm3, xmm0 ; low_s
+    punpcklbw       xmm4, xmm0 ; low_r
+
+    TABULATE_SSIM
+
+    add             rsi, rcx   ; next s row
+    add             rdi, rax   ; next r row
+
+    dec             rdx        ; counter
+    jnz .NextRow
+
+    SUM_ACROSS_W    xmm15
+    SUM_ACROSS_W    xmm14
+    SUM_ACROSS_Q    xmm13
+    SUM_ACROSS_Q    xmm12
+    SUM_ACROSS_Q    xmm11
+
+    mov             rdi,arg(4)
+    movd            [rdi], xmm15;
+    mov             rdi,arg(5)
+    movd            [rdi], xmm14;
+    mov             rdi,arg(6)
+    movd            [rdi], xmm13;
+    mov             rdi,arg(7)
+    movd            [rdi], xmm12;
+    mov             rdi,arg(8)
+    movd            [rdi], xmm11;
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
+
+;void ssim_parms_sse2(
+;    unsigned char *s,
+;    int sp,
+;    unsigned char *r,
+;    int rp
+;    uint32_t *sum_s,
+;    uint32_t *sum_r,
+;    uint32_t *sum_sq_s,
+;    uint32_t *sum_sq_r,
+;    uint32_t *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(aom_ssim_parms_8x8_sse2) PRIVATE
+sym(aom_ssim_parms_8x8_sse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 9
+    SAVE_XMM 15
+    push        rsi
+    push        rdi
+    ; end prolog
+
+    mov             rsi,        arg(0) ;s
+    mov             rcx,        arg(1) ;sp
+    mov             rdi,        arg(2) ;r
+    mov             rax,        arg(3) ;rp
+
+    pxor            xmm0, xmm0
+    pxor            xmm15,xmm15  ;sum_s
+    pxor            xmm14,xmm14  ;sum_r
+    pxor            xmm13,xmm13  ;sum_sq_s
+    pxor            xmm12,xmm12  ;sum_sq_r
+    pxor            xmm11,xmm11  ;sum_sxr
+
+    mov             rdx, 8      ;row counter
+.NextRow:
+
+    ;grab source and reference pixels
+    movq            xmm3, [rsi]
+    movq            xmm4, [rdi]
+    punpcklbw       xmm3, xmm0 ; low_s
+    punpcklbw       xmm4, xmm0 ; low_r
+
+    TABULATE_SSIM
+
+    add             rsi, rcx   ; next s row
+    add             rdi, rax   ; next r row
+
+    dec             rdx        ; counter
+    jnz .NextRow
+
+    SUM_ACROSS_W    xmm15
+    SUM_ACROSS_W    xmm14
+    SUM_ACROSS_Q    xmm13
+    SUM_ACROSS_Q    xmm12
+    SUM_ACROSS_Q    xmm11
+
+    mov             rdi,arg(4)
+    movd            [rdi], xmm15;
+    mov             rdi,arg(5)
+    movd            [rdi], xmm14;
+    mov             rdi,arg(6)
+    movd            [rdi], xmm13;
+    mov             rdi,arg(7)
+    movd            [rdi], xmm12;
+    mov             rdi,arg(8)
+    movd            [rdi], xmm11;
+
+    ; begin epilog
+    pop         rdi
+    pop         rsi
+    RESTORE_XMM
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/third_party/aom/aom_dsp/x86/subpel_variance_sse2.asm b/third_party/aom/aom_dsp/x86/subpel_variance_sse2.asm
index 45bf6ec3c5..cbf28901be 100644
--- a/third_party/aom/aom_dsp/x86/subpel_variance_sse2.asm
+++ b/third_party/aom/aom_dsp/x86/subpel_variance_sse2.asm
@@ -135,44 +135,33 @@ SECTION .text
     %if %2 == 1 ; avg
       cglobal sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
                                           x_offset, y_offset, dst, dst_stride, \
-                                          sec, sec_stride, height, sse, \
-                                          g_bilin_filter, g_pw_8
+                                          sec, sec_stride, height, sse
       %define block_height dword heightm
       %define sec_str sec_stridemp
-
-      ;Store bilin_filter and pw_8 location in stack
-      %if GET_GOT_DEFINED == 1
-        GET_GOT eax
-        add esp, 4                ; restore esp
-      %endif
-
-      lea ecx, [GLOBAL(bilin_filter_m)]
-      mov g_bilin_filterm, ecx
-
-      lea ecx, [GLOBAL(pw_8)]
-      mov g_pw_8m, ecx
-
-      LOAD_IF_USED 0, 1         ; load eax, ecx back
     %else
       cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
                                       x_offset, y_offset, dst, dst_stride, \
-                                      height, sse, g_bilin_filter, g_pw_8
+                                      height, sse
       %define block_height heightd
-
-      ;Store bilin_filter and pw_8 location in stack
-      %if GET_GOT_DEFINED == 1
-        GET_GOT eax
-        add esp, 4                ; restore esp
-      %endif
-
-      lea ecx, [GLOBAL(bilin_filter_m)]
-      mov g_bilin_filterm, ecx
-
-      lea ecx, [GLOBAL(pw_8)]
-      mov g_pw_8m, ecx
-
-      LOAD_IF_USED 0, 1         ; load eax, ecx back
     %endif
+
+    ; reuse argument stack space
+    %define g_bilin_filterm x_offsetm
+    %define g_pw_8m y_offsetm
+
+    ;Store bilin_filter and pw_8 location in stack
+    %if GET_GOT_DEFINED == 1
+      GET_GOT eax
+      add esp, 4                ; restore esp
+    %endif
+
+    lea ecx, [GLOBAL(bilin_filter_m)]
+    mov g_bilin_filterm, ecx
+
+    lea ecx, [GLOBAL(pw_8)]
+    mov g_pw_8m, ecx
+
+    LOAD_IF_USED 0, 1         ; load eax, ecx back
   %else
     %if %2 == 1 ; avg
       cglobal sub_pixel_avg_variance%1xh, 7, 7, 13, src, src_stride, \
diff --git a/third_party/aom/aom_dsp/x86/synonyms.h b/third_party/aom/aom_dsp/x86/synonyms.h
index 1e9f1e27b8..2e99bee3e9 100644
--- a/third_party/aom/aom_dsp/x86/synonyms.h
+++ b/third_party/aom/aom_dsp/x86/synonyms.h
@@ -13,6 +13,7 @@
 #define AOM_AOM_DSP_X86_SYNONYMS_H_
 
 #include <immintrin.h>
+#include <string.h>
 
 #include "config/aom_config.h"
 
@@ -28,7 +29,9 @@
 // Loads and stores to do away with the tedium of casting the address
 // to the right type.
 static INLINE __m128i xx_loadl_32(const void *a) {
-  return _mm_cvtsi32_si128(*(const uint32_t *)a);
+  int val;
+  memcpy(&val, a, sizeof(val));
+  return _mm_cvtsi32_si128(val);
 }
 
 static INLINE __m128i xx_loadl_64(const void *a) {
@@ -44,7 +47,8 @@ static INLINE __m128i xx_loadu_128(const void *a) {
 }
 
 static INLINE void xx_storel_32(void *const a, const __m128i v) {
-  *(uint32_t *)a = _mm_cvtsi128_si32(v);
+  const int val = _mm_cvtsi128_si32(v);
+  memcpy(a, &val, sizeof(val));
 }
 
 static INLINE void xx_storel_64(void *const a, const __m128i v) {
diff --git a/third_party/aom/aom_dsp/x86/synonyms_avx2.h b/third_party/aom/aom_dsp/x86/synonyms_avx2.h
index 3f69b120ea..4d6ee6ad64 100644
--- a/third_party/aom/aom_dsp/x86/synonyms_avx2.h
+++ b/third_party/aom/aom_dsp/x86/synonyms_avx2.h
@@ -67,6 +67,11 @@ static INLINE __m256i yy_loadu2_128(const void *hi, const void *lo) {
   return yy_set_m128i(mhi, mlo);
 }
 
+static INLINE void yy_storeu2_128(void *hi, void *lo, const __m256i a) {
+  _mm_storeu_si128((__m128i *)hi, _mm256_extracti128_si256(a, 1));
+  _mm_storeu_si128((__m128i *)lo, _mm256_castsi256_si128(a));
+}
+
 static INLINE __m256i yy_roundn_epu16(__m256i v_val_w, int bits) {
   const __m256i v_s_w = _mm256_srli_epi16(v_val_w, bits - 1);
   return _mm256_avg_epu16(v_s_w, _mm256_setzero_si256());
diff --git a/third_party/aom/aom_dsp/x86/txfm_common_avx2.h b/third_party/aom/aom_dsp/x86/txfm_common_avx2.h
index b1611ba870..8a40508986 100644
--- a/third_party/aom/aom_dsp/x86/txfm_common_avx2.h
+++ b/third_party/aom/aom_dsp/x86/txfm_common_avx2.h
@@ -20,9 +20,6 @@
 extern "C" {
 #endif
 
-typedef void (*transform_1d_avx2)(const __m256i *input, __m256i *output,
-                                  int8_t cos_bit);
-
 static INLINE __m256i pair_set_w16_epi16(int16_t a, int16_t b) {
   return _mm256_set1_epi32(
       (int32_t)(((uint16_t)(a)) | (((uint32_t)(b)) << 16)));
@@ -192,6 +189,53 @@ static INLINE void round_shift_16bit_w16_avx2(__m256i *in, int size, int bit) {
   }
 }
 
+static INLINE __m256i av1_round_shift_32_avx2(__m256i vec, int bit) {
+  __m256i tmp, round;
+  round = _mm256_set1_epi32(1 << (bit - 1));
+  tmp = _mm256_add_epi32(vec, round);
+  return _mm256_srai_epi32(tmp, bit);
+}
+
+static INLINE void av1_round_shift_array_32_avx2(__m256i *input,
+                                                 __m256i *output,
+                                                 const int size,
+                                                 const int bit) {
+  if (bit > 0) {
+    int i;
+    for (i = 0; i < size; i++) {
+      output[i] = av1_round_shift_32_avx2(input[i], bit);
+    }
+  } else {
+    int i;
+    for (i = 0; i < size; i++) {
+      output[i] = _mm256_slli_epi32(input[i], -bit);
+    }
+  }
+}
+
+static INLINE void av1_round_shift_rect_array_32_avx2(__m256i *input,
+                                                      __m256i *output,
+                                                      const int size,
+                                                      const int bit,
+                                                      const int val) {
+  const __m256i sqrt2 = _mm256_set1_epi32(val);
+  if (bit > 0) {
+    int i;
+    for (i = 0; i < size; i++) {
+      const __m256i r0 = av1_round_shift_32_avx2(input[i], bit);
+      const __m256i r1 = _mm256_mullo_epi32(sqrt2, r0);
+      output[i] = av1_round_shift_32_avx2(r1, NewSqrt2Bits);
+    }
+  } else {
+    int i;
+    for (i = 0; i < size; i++) {
+      const __m256i r0 = _mm256_slli_epi32(input[i], -bit);
+      const __m256i r1 = _mm256_mullo_epi32(sqrt2, r0);
+      output[i] = av1_round_shift_32_avx2(r1, NewSqrt2Bits);
+    }
+  }
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/third_party/aom/aom_dsp/x86/variance_sse2.c b/third_party/aom/aom_dsp/x86/variance_sse2.c
index 3c37e77c06..f3efc157f9 100644
--- a/third_party/aom/aom_dsp/x86/variance_sse2.c
+++ b/third_party/aom/aom_dsp/x86/variance_sse2.c
@@ -494,7 +494,7 @@ void aom_upsampled_pred_sse2(MACROBLOCKD *xd, const struct AV1Common *const cm,
     const int ref_num = 0;
     const int is_intrabc = is_intrabc_block(mi);
     const struct scale_factors *const sf =
-        is_intrabc ? &cm->sf_identity : &xd->block_refs[ref_num]->sf;
+        is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref_num];
     const int is_scaled = av1_is_scaled(sf);
 
     if (is_scaled) {
@@ -571,11 +571,10 @@ void aom_upsampled_pred_sse2(MACROBLOCKD *xd, const struct AV1Common *const cm,
     }
   }
 
-  const InterpFilterParams *filter =
-      (subpel_search == 1)
-          ? av1_get_4tap_interp_filter_params(EIGHTTAP_REGULAR)
-          : av1_get_interp_filter_params_with_block_size(EIGHTTAP_REGULAR, 8);
-  int filter_taps = (subpel_search == 1) ? 4 : SUBPEL_TAPS;
+  const InterpFilterParams *filter = av1_get_filter(subpel_search);
+  // (TODO:yunqing) 2-tap case uses 4-tap functions since there is no SIMD for
+  // 2-tap yet.
+  int filter_taps = (subpel_search <= USE_4_TAPS) ? 4 : SUBPEL_TAPS;
 
   if (!subpel_x_q3 && !subpel_y_q3) {
     if (width >= 16) {
@@ -638,20 +637,13 @@ void aom_upsampled_pred_sse2(MACROBLOCKD *xd, const struct AV1Common *const cm,
     const int16_t *const kernel_y =
         av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
     const uint8_t *ref_start = ref - ref_stride * ((filter_taps >> 1) - 1);
-    uint8_t *temp_start_horiz =
-        (subpel_search == 1) ? temp + (filter_taps >> 1) * MAX_SB_SIZE : temp;
+    uint8_t *temp_start_horiz = (subpel_search <= USE_4_TAPS)
+                                    ? temp + (filter_taps >> 1) * MAX_SB_SIZE
+                                    : temp;
     uint8_t *temp_start_vert = temp + MAX_SB_SIZE * ((filter->taps >> 1) - 1);
     int intermediate_height =
         (((height - 1) * 8 + subpel_y_q3) >> 3) + filter_taps;
     assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
-    // TODO(Deepa): Remove the memset below when we have
-    // 4 tap simd for sse2 and ssse3.
-    if (subpel_search == 1) {
-      memset(temp_start_vert - 3 * MAX_SB_SIZE, 0, width);
-      memset(temp_start_vert - 2 * MAX_SB_SIZE, 0, width);
-      memset(temp_start_vert + (height + 2) * MAX_SB_SIZE, 0, width);
-      memset(temp_start_vert + (height + 3) * MAX_SB_SIZE, 0, width);
-    }
     aom_convolve8_horiz(ref_start, ref_stride, temp_start_horiz, MAX_SB_SIZE,
                         kernel_x, 16, NULL, -1, width, intermediate_height);
     aom_convolve8_vert(temp_start_vert, MAX_SB_SIZE, comp_pred, width, NULL, -1,
diff --git a/third_party/aom/aom_ports/mem.h b/third_party/aom/aom_ports/mem.h
index 3ffea3cd6d..9e3d424037 100644
--- a/third_party/aom/aom_ports/mem.h
+++ b/third_party/aom/aom_ports/mem.h
@@ -66,4 +66,34 @@
 #define CONVERT_TO_SHORTPTR(x) ((uint16_t *)(((uintptr_t)(x)) << 1))
 #define CONVERT_TO_BYTEPTR(x) ((uint8_t *)(((uintptr_t)(x)) >> 1))
 
+/*!\brief force enum to be unsigned 1 byte*/
+#define UENUM1BYTE(enumvar) \
+  ;                         \
+  typedef uint8_t enumvar
+
+/*!\brief force enum to be signed 1 byte*/
+#define SENUM1BYTE(enumvar) \
+  ;                         \
+  typedef int8_t enumvar
+
+/*!\brief force enum to be unsigned 2 byte*/
+#define UENUM2BYTE(enumvar) \
+  ;                         \
+  typedef uint16_t enumvar
+
+/*!\brief force enum to be signed 2 byte*/
+#define SENUM2BYTE(enumvar) \
+  ;                         \
+  typedef int16_t enumvar
+
+/*!\brief force enum to be unsigned 4 byte*/
+#define UENUM4BYTE(enumvar) \
+  ;                         \
+  typedef uint32_t enumvar
+
+/*!\brief force enum to be unsigned 4 byte*/
+#define SENUM4BYTE(enumvar) \
+  ;                         \
+  typedef int32_t enumvar
+
 #endif  // AOM_AOM_PORTS_MEM_H_
diff --git a/third_party/aom/aom_ports/x86.h b/third_party/aom/aom_ports/x86.h
index 52ee49cb3c..8c18448714 100644
--- a/third_party/aom/aom_ports/x86.h
+++ b/third_party/aom/aom_ports/x86.h
@@ -222,11 +222,26 @@ static INLINE int x86_simd_caps(void) {
   return flags & mask;
 }
 
-// Note:
-//  32-bit CPU cycle counter is light-weighted for most function performance
-//  measurement. For large function (CPU time > a couple of seconds), 64-bit
-//  counter should be used.
-// 32-bit CPU cycle counter
+// Fine-Grain Measurement Functions
+//
+// If you are a timing a small region of code, access the timestamp counter
+// (TSC) via:
+//
+// unsigned int start = x86_tsc_start();
+//   ...
+// unsigned int end = x86_tsc_end();
+// unsigned int diff = end - start;
+//
+// The start/end functions introduce a few more instructions than using
+// x86_readtsc directly, but prevent the CPU's out-of-order execution from
+// affecting the measurement (by having earlier/later instructions be evaluated
+// in the time interval). See the white paper, "How to Benchmark Code
+// Execution Times on Intel® IA-32 and IA-64 Instruction Set Architectures" by
+// Gabriele Paoloni for more information.
+//
+// If you are timing a large function (CPU time > a couple of seconds), use
+// x86_readtsc64 to read the timestamp counter in a 64-bit integer. The
+// out-of-order leakage that can occur is minimal compared to total runtime.
 static INLINE unsigned int x86_readtsc(void) {
 #if defined(__GNUC__) && __GNUC__
   unsigned int tsc;
@@ -263,6 +278,41 @@ static INLINE uint64_t x86_readtsc64(void) {
 #endif
 }
 
+// 32-bit CPU cycle counter with a partial fence against out-of-order execution.
+static INLINE unsigned int x86_readtscp(void) {
+#if defined(__GNUC__) && __GNUC__
+  unsigned int tscp;
+  __asm__ __volatile__("rdtscp\n\t" : "=a"(tscp) :);
+  return tscp;
+#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
+  unsigned int tscp;
+  asm volatile("rdtscp\n\t" : "=a"(tscp) :);
+  return tscp;
+#elif defined(_MSC_VER)
+  unsigned int ui;
+  return (unsigned int)__rdtscp(&ui);
+#else
+#if ARCH_X86_64
+  return (unsigned int)__rdtscp();
+#else
+  __asm rdtscp;
+#endif
+#endif
+}
+
+static INLINE unsigned int x86_tsc_start(void) {
+  unsigned int reg_eax, reg_ebx, reg_ecx, reg_edx;
+  cpuid(0, 0, reg_eax, reg_ebx, reg_ecx, reg_edx);
+  return x86_readtsc();
+}
+
+static INLINE unsigned int x86_tsc_end(void) {
+  uint32_t v = x86_readtscp();
+  unsigned int reg_eax, reg_ebx, reg_ecx, reg_edx;
+  cpuid(0, 0, reg_eax, reg_ebx, reg_ecx, reg_edx);
+  return v;
+}
+
 #if defined(__GNUC__) && __GNUC__
 #define x86_pause_hint() __asm__ __volatile__("pause \n\t")
 #elif defined(__SUNPRO_C) || defined(__SUNPRO_CC)
diff --git a/third_party/aom/aom_scale/aom_scale_rtcd.pl b/third_party/aom/aom_scale/aom_scale_rtcd.pl
index c5990b1bb3..27378c7ef4 100644
--- a/third_party/aom/aom_scale/aom_scale_rtcd.pl
+++ b/third_party/aom/aom_scale/aom_scale_rtcd.pl
@@ -36,11 +36,12 @@ add_proto qw/void aom_yv12_copy_u/, "const struct yv12_buffer_config *src_bc, st
 
 add_proto qw/void aom_yv12_copy_v/, "const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc";
 
-add_proto qw/void aom_yv12_partial_copy_y/, "const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend";
-
-add_proto qw/void aom_yv12_partial_copy_u/, "const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend";
-
-add_proto qw/void aom_yv12_partial_copy_v/, "const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend";
+add_proto qw/void aom_yv12_partial_copy_y/, "const struct yv12_buffer_config *src_ybc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_ybc, int hstart2, int vstart2";
+add_proto qw/void aom_yv12_partial_coloc_copy_y/, "const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc, int hstart, int hend, int vstart, int vend";
+add_proto qw/void aom_yv12_partial_copy_u/, "const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2";
+add_proto qw/void aom_yv12_partial_coloc_copy_u/, "const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend";
+add_proto qw/void aom_yv12_partial_copy_v/, "const struct yv12_buffer_config *src_bc, int hstart1, int hend1, int vstart1, int vend1, struct yv12_buffer_config *dst_bc, int hstart2, int vstart2";
+add_proto qw/void aom_yv12_partial_coloc_copy_v/, "const struct yv12_buffer_config *src_bc, struct yv12_buffer_config *dst_bc, int hstart, int hend, int vstart, int vend";
 
 add_proto qw/void aom_extend_frame_borders/, "struct yv12_buffer_config *ybf, const int num_planes";
 specialize qw/aom_extend_frame_borders dspr2/;
diff --git a/third_party/aom/aom_scale/generic/yv12config.c b/third_party/aom/aom_scale/generic/yv12config.c
index 84705e2d8b..dafe02b171 100644
--- a/third_party/aom/aom_scale/generic/yv12config.c
+++ b/third_party/aom/aom_scale/generic/yv12config.c
@@ -55,6 +55,13 @@ int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
   if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) return -1;
 #endif
 
+  /* Only support allocating buffers that have a border that's a multiple
+   * of 32. The border restriction is required to get 16-byte alignment of
+   * the start of the chroma rows without introducing an arbitrary gap
+   * between planes, which would break the semantics of things like
+   * aom_img_set_rect(). */
+  if (border & 0x1f) return -3;
+
   if (ybf) {
     const int aom_byte_align = (byte_alignment == 0) ? 1 : byte_alignment;
     const int aligned_width = (width + 7) & ~7;
@@ -109,10 +116,11 @@ int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
       memset(ybf->buffer_alloc, 0, (size_t)frame_size);
 #endif
 #endif
-    } else if (frame_size > (size_t)ybf->buffer_alloc_sz) {
+    } else if (frame_size > ybf->buffer_alloc_sz) {
       // Allocation to hold larger frame, or first allocation.
       aom_free(ybf->buffer_alloc);
       ybf->buffer_alloc = NULL;
+      ybf->buffer_alloc_sz = 0;
 
       if (frame_size != (size_t)frame_size) return -1;
 
@@ -127,13 +135,6 @@ int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
       memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz);
     }
 
-    /* Only support allocating buffers that have a border that's a multiple
-     * of 32. The border restriction is required to get 16-byte alignment of
-     * the start of the chroma rows without introducing an arbitrary gap
-     * between planes, which would break the semantics of things like
-     * aom_img_set_rect(). */
-    if (border & 0x1f) return -3;
-
     ybf->y_crop_width = width;
     ybf->y_crop_height = height;
     ybf->y_width = aligned_width;
diff --git a/third_party/aom/aom_scale/generic/yv12extend.c b/third_party/aom/aom_scale/generic/yv12extend.c
index ba183520a0..127ca23116 100644
--- a/third_party/aom/aom_scale/generic/yv12extend.c
+++ b/third_party/aom/aom_scale/generic/yv12extend.c
@@ -321,91 +321,116 @@ void aom_yv12_copy_v_c(const YV12_BUFFER_CONFIG *src_bc,
   }
 }
 
-void aom_yv12_partial_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc,
-                               YV12_BUFFER_CONFIG *dst_ybc, int hstart,
-                               int hend, int vstart, int vend) {
+void aom_yv12_partial_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc, int hstart1,
+                               int hend1, int vstart1, int vend1,
+                               YV12_BUFFER_CONFIG *dst_ybc, int hstart2,
+                               int vstart2) {
   int row;
   const uint8_t *src = src_ybc->y_buffer;
   uint8_t *dst = dst_ybc->y_buffer;
 
   if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) {
     const uint16_t *src16 =
-        CONVERT_TO_SHORTPTR(src + vstart * src_ybc->y_stride + hstart);
+        CONVERT_TO_SHORTPTR(src + vstart1 * src_ybc->y_stride + hstart1);
     uint16_t *dst16 =
-        CONVERT_TO_SHORTPTR(dst + vstart * dst_ybc->y_stride + hstart);
-    for (row = vstart; row < vend; ++row) {
-      memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t));
+        CONVERT_TO_SHORTPTR(dst + vstart2 * dst_ybc->y_stride + hstart2);
+
+    for (row = vstart1; row < vend1; ++row) {
+      memcpy(dst16, src16, (hend1 - hstart1) * sizeof(uint16_t));
       src16 += src_ybc->y_stride;
       dst16 += dst_ybc->y_stride;
     }
     return;
   }
-  src = (src + vstart * src_ybc->y_stride + hstart);
-  dst = (dst + vstart * dst_ybc->y_stride + hstart);
+  src = (src + vstart1 * src_ybc->y_stride + hstart1);
+  dst = (dst + vstart2 * dst_ybc->y_stride + hstart2);
 
-  for (row = vstart; row < vend; ++row) {
-    memcpy(dst, src, (hend - hstart));
+  for (row = vstart1; row < vend1; ++row) {
+    memcpy(dst, src, (hend1 - hstart1));
     src += src_ybc->y_stride;
     dst += dst_ybc->y_stride;
   }
 }
 
-void aom_yv12_partial_copy_u_c(const YV12_BUFFER_CONFIG *src_bc,
-                               YV12_BUFFER_CONFIG *dst_bc, int hstart, int hend,
-                               int vstart, int vend) {
+void aom_yv12_partial_coloc_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc,
+                                     YV12_BUFFER_CONFIG *dst_ybc, int hstart,
+                                     int hend, int vstart, int vend) {
+  aom_yv12_partial_copy_y_c(src_ybc, hstart, hend, vstart, vend, dst_ybc,
+                            hstart, vstart);
+}
+
+void aom_yv12_partial_copy_u_c(const YV12_BUFFER_CONFIG *src_bc, int hstart1,
+                               int hend1, int vstart1, int vend1,
+                               YV12_BUFFER_CONFIG *dst_bc, int hstart2,
+                               int vstart2) {
   int row;
   const uint8_t *src = src_bc->u_buffer;
   uint8_t *dst = dst_bc->u_buffer;
 
   if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) {
     const uint16_t *src16 =
-        CONVERT_TO_SHORTPTR(src + vstart * src_bc->uv_stride + hstart);
+        CONVERT_TO_SHORTPTR(src + vstart1 * src_bc->uv_stride + hstart1);
     uint16_t *dst16 =
-        CONVERT_TO_SHORTPTR(dst + vstart * dst_bc->uv_stride + hstart);
-    for (row = vstart; row < vend; ++row) {
-      memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t));
+        CONVERT_TO_SHORTPTR(dst + vstart2 * dst_bc->uv_stride + hstart2);
+    for (row = vstart1; row < vend1; ++row) {
+      memcpy(dst16, src16, (hend1 - hstart1) * sizeof(uint16_t));
       src16 += src_bc->uv_stride;
       dst16 += dst_bc->uv_stride;
     }
     return;
   }
 
-  src = (src + vstart * src_bc->uv_stride + hstart);
-  dst = (dst + vstart * dst_bc->uv_stride + hstart);
+  src = (src + vstart1 * src_bc->uv_stride + hstart1);
+  dst = (dst + vstart2 * dst_bc->uv_stride + hstart2);
 
-  for (row = vstart; row < vend; ++row) {
-    memcpy(dst, src, (hend - hstart));
+  for (row = vstart1; row < vend1; ++row) {
+    memcpy(dst, src, (hend1 - hstart1));
     src += src_bc->uv_stride;
     dst += dst_bc->uv_stride;
   }
 }
 
-void aom_yv12_partial_copy_v_c(const YV12_BUFFER_CONFIG *src_bc,
-                               YV12_BUFFER_CONFIG *dst_bc, int hstart, int hend,
-                               int vstart, int vend) {
+void aom_yv12_partial_coloc_copy_u_c(const YV12_BUFFER_CONFIG *src_bc,
+                                     YV12_BUFFER_CONFIG *dst_bc, int hstart,
+                                     int hend, int vstart, int vend) {
+  aom_yv12_partial_copy_u_c(src_bc, hstart, hend, vstart, vend, dst_bc, hstart,
+                            vstart);
+}
+
+void aom_yv12_partial_copy_v_c(const YV12_BUFFER_CONFIG *src_bc, int hstart1,
+                               int hend1, int vstart1, int vend1,
+                               YV12_BUFFER_CONFIG *dst_bc, int hstart2,
+                               int vstart2) {
   int row;
   const uint8_t *src = src_bc->v_buffer;
   uint8_t *dst = dst_bc->v_buffer;
 
   if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) {
     const uint16_t *src16 =
-        CONVERT_TO_SHORTPTR(src + vstart * src_bc->uv_stride + hstart);
+        CONVERT_TO_SHORTPTR(src + vstart1 * src_bc->uv_stride + hstart1);
     uint16_t *dst16 =
-        CONVERT_TO_SHORTPTR(dst + vstart * dst_bc->uv_stride + hstart);
-    for (row = vstart; row < vend; ++row) {
-      memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t));
+        CONVERT_TO_SHORTPTR(dst + vstart2 * dst_bc->uv_stride + hstart2);
+    for (row = vstart1; row < vend1; ++row) {
+      memcpy(dst16, src16, (hend1 - hstart1) * sizeof(uint16_t));
       src16 += src_bc->uv_stride;
       dst16 += dst_bc->uv_stride;
     }
     return;
   }
 
-  src = (src + vstart * src_bc->uv_stride + hstart);
-  dst = (dst + vstart * dst_bc->uv_stride + hstart);
+  src = (src + vstart1 * src_bc->uv_stride + hstart1);
+  dst = (dst + vstart2 * dst_bc->uv_stride + hstart2);
 
-  for (row = vstart; row < vend; ++row) {
-    memcpy(dst, src, (hend - hstart));
+  for (row = vstart1; row < vend1; ++row) {
+    memcpy(dst, src, (hend1 - hstart1));
     src += src_bc->uv_stride;
     dst += dst_bc->uv_stride;
   }
 }
+
+void aom_yv12_partial_coloc_copy_v_c(const YV12_BUFFER_CONFIG *src_bc,
+                                     YV12_BUFFER_CONFIG *dst_bc, int hstart,
+                                     int hend, int vstart, int vend) {
+  aom_yv12_partial_copy_v_c(src_bc, hstart, hend, vstart, vend, dst_bc, hstart,
+                            vstart);
+}
diff --git a/third_party/aom/aom_scale/yv12config.h b/third_party/aom/aom_scale/yv12config.h
index 2fb81acd7c..bb0d56430b 100644
--- a/third_party/aom/aom_scale/yv12config.h
+++ b/third_party/aom/aom_scale/yv12config.h
@@ -24,32 +24,24 @@ extern "C" {
 
 #define AOMINNERBORDERINPIXELS 160
 #define AOM_INTERP_EXTEND 4
-
-// TODO(jingning): Use unified inter predictor for encoder and
-// decoder during the development process. Revisit the frame border
-// to improve the decoder performance.
-#if CONFIG_REDUCED_ENCODER_BORDER
-#define AOM_BORDER_IN_PIXELS 160
-#else
 #define AOM_BORDER_IN_PIXELS 288
-#endif  // CONFIG_REDUCED_ENCODER_BORDER
+#define AOM_ENC_NO_SCALE_BORDER 160
+#define AOM_DEC_BORDER_IN_PIXELS 64
 
 typedef struct yv12_buffer_config {
   union {
     struct {
       int y_width;
       int uv_width;
-      int alpha_width;
     };
-    int widths[3];
+    int widths[2];
   };
   union {
     struct {
       int y_height;
       int uv_height;
-      int alpha_height;
     };
-    int heights[3];
+    int heights[2];
   };
   union {
     struct {
@@ -69,18 +61,16 @@ typedef struct yv12_buffer_config {
     struct {
       int y_stride;
       int uv_stride;
-      int alpha_stride;
     };
-    int strides[3];
+    int strides[2];
   };
   union {
     struct {
       uint8_t *y_buffer;
       uint8_t *u_buffer;
       uint8_t *v_buffer;
-      uint8_t *alpha_buffer;
     };
-    uint8_t *buffers[4];
+    uint8_t *buffers[3];
   };
 
   // Indicate whether y_buffer, u_buffer, and v_buffer points to the internally
@@ -106,7 +96,7 @@ typedef struct yv12_buffer_config {
   aom_color_primaries_t color_primaries;
   aom_transfer_characteristics_t transfer_characteristics;
   aom_matrix_coefficients_t matrix_coefficients;
-  int monochrome;
+  uint8_t monochrome;
   aom_chroma_sample_position_t chroma_sample_position;
   aom_color_range_t color_range;
   int render_width;
diff --git a/third_party/aom/aom_util/aom_thread.c b/third_party/aom/aom_util/aom_thread.c
index cae9f5e250..244ac3bc29 100644
--- a/third_party/aom/aom_util/aom_thread.c
+++ b/third_party/aom/aom_util/aom_thread.c
@@ -14,6 +14,12 @@
 // Original source:
 //  https://chromium.googlesource.com/webm/libwebp
 
+// Enable GNU extensions in glibc so that we can call pthread_setname_np().
+// This must be before any #include statements.
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+
 #include <assert.h>
 #include <string.h>  // for memset()
 
@@ -34,6 +40,28 @@ static void execute(AVxWorker *const worker);  // Forward declaration.
 
 static THREADFN thread_loop(void *ptr) {
   AVxWorker *const worker = (AVxWorker *)ptr;
+#ifdef __APPLE__
+  if (worker->thread_name != NULL) {
+    // Apple's version of pthread_setname_np takes one argument and operates on
+    // the current thread only. The maximum size of the thread_name buffer was
+    // noted in the Chromium source code and was confirmed by experiments. If
+    // thread_name is too long, pthread_setname_np returns -1 with errno
+    // ENAMETOOLONG (63).
+    char thread_name[64];
+    strncpy(thread_name, worker->thread_name, sizeof(thread_name));
+    thread_name[sizeof(thread_name) - 1] = '\0';
+    pthread_setname_np(thread_name);
+  }
+#elif defined(__GLIBC__) || defined(__BIONIC__)
+  if (worker->thread_name != NULL) {
+    // Linux and Android require names (with nul) fit in 16 chars, otherwise
+    // pthread_setname_np() returns ERANGE (34).
+    char thread_name[16];
+    strncpy(thread_name, worker->thread_name, sizeof(thread_name));
+    thread_name[sizeof(thread_name) - 1] = '\0';
+    pthread_setname_np(pthread_self(), thread_name);
+  }
+#endif
   int done = 0;
   while (!done) {
     pthread_mutex_lock(&worker->impl_->mutex_);
diff --git a/third_party/aom/aom_util/aom_thread.h b/third_party/aom/aom_util/aom_thread.h
index f14c1ac18c..bda8b756ec 100644
--- a/third_party/aom/aom_util/aom_thread.h
+++ b/third_party/aom/aom_util/aom_thread.h
@@ -26,6 +26,7 @@ extern "C" {
 // Set maximum decode threads to be 8 due to the limit of frame buffers
 // and not enough semaphores in the emulation layer on windows.
 #define MAX_DECODE_THREADS 8
+#define MAX_NUM_THREADS 64
 
 #if CONFIG_MULTITHREAD
 
@@ -380,6 +381,10 @@ typedef struct AVxWorkerImpl AVxWorkerImpl;
 typedef struct {
   AVxWorkerImpl *impl_;
   AVxWorkerStatus status_;
+  // Thread name for the debugger. If not NULL, must point to a string that
+  // outlives the worker thread. For portability, use a name <= 15 characters
+  // long (not including the terminating NUL character).
+  const char *thread_name;
   AVxWorkerHook hook;  // hook to call
   void *data1;         // first argument passed to 'hook'
   void *data2;         // second argument passed to 'hook'
diff --git a/third_party/aom/apps/aomdec.c b/third_party/aom/apps/aomdec.c
index ff13b6f500..c054425777 100644
--- a/third_party/aom/apps/aomdec.c
+++ b/third_party/aom/apps/aomdec.c
@@ -143,7 +143,7 @@ static INLINE int libyuv_scale(aom_image_t *src, aom_image_t *dst,
 }
 #endif
 
-void show_help(FILE *fout, int shorthelp) {
+static void show_help(FILE *fout, int shorthelp) {
   fprintf(fout, "Usage: %s <options> filename\n\n", exec_name);
 
   if (shorthelp) {
@@ -427,13 +427,6 @@ static FILE *open_outfile(const char *name) {
   }
 }
 
-static int img_shifted_realloc_required(const aom_image_t *img,
-                                        const aom_image_t *shifted,
-                                        aom_img_fmt_t required_fmt) {
-  return img->d_w != shifted->d_w || img->d_h != shifted->d_h ||
-         required_fmt != shifted->fmt;
-}
-
 static int main_loop(int argc, const char **argv_) {
   aom_codec_ctx_t decoder;
   char *fn = NULL;
@@ -459,7 +452,7 @@ static int main_loop(int argc, const char **argv_) {
   int opt_i420 = 0;
   int opt_raw = 0;
   aom_codec_dec_cfg_t cfg = { 0, 0, 0, CONFIG_LOWBITDEPTH, { 1 } };
-  unsigned int output_bit_depth = 0;
+  unsigned int fixed_output_bit_depth = 0;
   unsigned int is_annexb = 0;
   int frames_corrupted = 0;
   int dec_flags = 0;
@@ -491,6 +484,7 @@ static int main_loop(int argc, const char **argv_) {
   input.webm_ctx = &webm_ctx;
 #endif
   struct ObuDecInputContext obu_ctx = { NULL, NULL, 0, 0, 0 };
+  int is_ivf = 0;
 
   obu_ctx.avx_ctx = &aom_input_ctx;
   input.obu_ctx = &obu_ctx;
@@ -572,7 +566,7 @@ static int main_loop(int argc, const char **argv_) {
     } else if (arg_match(&arg, &continuearg, argi)) {
       keep_going = 1;
     } else if (arg_match(&arg, &outbitdeptharg, argi)) {
-      output_bit_depth = arg_parse_uint(&arg);
+      fixed_output_bit_depth = arg_parse_uint(&arg);
     } else if (arg_match(&arg, &isannexb, argi)) {
       is_annexb = 1;
       input.obu_ctx->is_annexb = 1;
@@ -617,8 +611,10 @@ static int main_loop(int argc, const char **argv_) {
 #endif
   input.aom_input_ctx->filename = fn;
   input.aom_input_ctx->file = infile;
-  if (file_is_ivf(input.aom_input_ctx))
+  if (file_is_ivf(input.aom_input_ctx)) {
     input.aom_input_ctx->file_type = FILE_TYPE_IVF;
+    is_ivf = 1;
+  }
 #if CONFIG_WEBM_IO
   else if (file_is_webm(input.webm_ctx, input.aom_input_ctx))
     input.aom_input_ctx->file_type = FILE_TYPE_WEBM;
@@ -668,6 +664,10 @@ static int main_loop(int argc, const char **argv_) {
   }
 
   fourcc_interface = get_aom_decoder_by_fourcc(aom_input_ctx.fourcc);
+
+  if (is_ivf && !fourcc_interface)
+    fatal("Unsupported fourcc: %x\n", aom_input_ctx.fourcc);
+
   if (interface && fourcc_interface && interface != fourcc_interface)
     warn("Header indicates codec: %s\n", fourcc_interface->name);
   else
@@ -779,7 +779,7 @@ static int main_loop(int argc, const char **argv_) {
     aom_usec_timer_start(&timer);
 
     if (flush_decoder) {
-      // Flush the decoder in frame parallel decode.
+      // Flush the decoder.
       if (aom_codec_decode(&decoder, NULL, 0, NULL)) {
         warn("Failed to flush decoder: %s", aom_codec_error(&decoder));
       }
@@ -850,41 +850,15 @@ static int main_loop(int argc, const char **argv_) {
           }
         }
         // Default to codec bit depth if output bit depth not set
-        if (!output_bit_depth && single_file && !do_md5) {
+        unsigned int output_bit_depth;
+        if (!fixed_output_bit_depth && single_file && !do_md5) {
           output_bit_depth = img->bit_depth;
+        } else {
+          output_bit_depth = fixed_output_bit_depth;
         }
         // Shift up or down if necessary
-        if (output_bit_depth != 0) {
-          const aom_img_fmt_t shifted_fmt =
-              output_bit_depth == 8 ? img->fmt & ~AOM_IMG_FMT_HIGHBITDEPTH
-                                    : img->fmt | AOM_IMG_FMT_HIGHBITDEPTH;
-
-          if (shifted_fmt != img->fmt || output_bit_depth != img->bit_depth) {
-            if (img_shifted &&
-                img_shifted_realloc_required(img, img_shifted, shifted_fmt)) {
-              aom_img_free(img_shifted);
-              img_shifted = NULL;
-            }
-            if (img_shifted) {
-              img_shifted->monochrome = img->monochrome;
-            }
-            if (!img_shifted) {
-              img_shifted =
-                  aom_img_alloc(NULL, shifted_fmt, img->d_w, img->d_h, 16);
-              img_shifted->bit_depth = output_bit_depth;
-              img_shifted->monochrome = img->monochrome;
-              img_shifted->csp = img->csp;
-            }
-            if (output_bit_depth > img->bit_depth) {
-              aom_img_upshift(img_shifted, img,
-                              output_bit_depth - img->bit_depth);
-            } else {
-              aom_img_downshift(img_shifted, img,
-                                img->bit_depth - output_bit_depth);
-            }
-            img = img_shifted;
-          }
-        }
+        if (output_bit_depth != 0)
+          aom_shift_img(output_bit_depth, &img, &img_shifted);
 
         aom_input_ctx.width = img->d_w;
         aom_input_ctx.height = img->d_h;
@@ -900,6 +874,11 @@ static int main_loop(int argc, const char **argv_) {
                   y4m_buf, sizeof(y4m_buf), aom_input_ctx.width,
                   aom_input_ctx.height, &aom_input_ctx.framerate,
                   img->monochrome, img->csp, img->fmt, img->bit_depth);
+              if (img->csp == AOM_CSP_COLOCATED) {
+                fprintf(stderr,
+                        "Warning: Y4M lacks a colorspace for colocated "
+                        "chroma. Using a placeholder.\n");
+              }
               if (do_md5) {
                 MD5Update(&md5_ctx, (md5byte *)y4m_buf, (unsigned int)len);
               } else {
diff --git a/third_party/aom/apps/aomenc.c b/third_party/aom/apps/aomenc.c
index 2e5d35cfef..83c1353de5 100644
--- a/third_party/aom/apps/aomenc.c
+++ b/third_party/aom/apps/aomenc.c
@@ -263,9 +263,9 @@ static const arg_def_t global_error_resilient =
             "Enable global error resiliency features");
 static const arg_def_t lag_in_frames =
     ARG_DEF(NULL, "lag-in-frames", 1, "Max number of frames to lag");
-static const arg_def_t large_scale_tile =
-    ARG_DEF(NULL, "large-scale-tile", 1,
-            "Large scale tile coding (0: off (default), 1: on)");
+static const arg_def_t large_scale_tile = ARG_DEF(
+    NULL, "large-scale-tile", 1,
+    "Large scale tile coding (0: off (default), 1: on (ivf output only))");
 static const arg_def_t monochrome =
     ARG_DEF(NULL, "monochrome", 0, "Monochrome video (no chroma planes)");
 static const arg_def_t full_still_picture_hdr = ARG_DEF(
@@ -415,11 +415,14 @@ static const arg_def_t cpu_used_av1 =
     ARG_DEF(NULL, "cpu-used", 1, "CPU Used (0..8)");
 static const arg_def_t rowmtarg =
     ARG_DEF(NULL, "row-mt", 1,
-            "Enable row based multi-threading (0: off (default), 1: on)");
+            "Enable row based multi-threading (0: off, 1: on (default))");
 static const arg_def_t tile_cols =
     ARG_DEF(NULL, "tile-columns", 1, "Number of tile columns to use, log2");
 static const arg_def_t tile_rows =
     ARG_DEF(NULL, "tile-rows", 1, "Number of tile rows to use, log2");
+static const arg_def_t enable_tpl_model =
+    ARG_DEF(NULL, "enable-tpl-model", 1,
+            "RDO modulation based on frame temporal dependency");
 static const arg_def_t tile_width =
     ARG_DEF(NULL, "tile-width", 1, "Tile widths (comma separated)");
 static const arg_def_t tile_height =
@@ -434,6 +437,88 @@ static const arg_def_t enable_restoration =
     ARG_DEF(NULL, "enable-restoration", 1,
             "Enable the loop restoration filter (0: false, "
             "1: true (default))");
+static const arg_def_t enable_rect_partitions =
+    ARG_DEF(NULL, "enable-rect-partitions", 1,
+            "Enable rectangular partitions "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_dual_filter =
+    ARG_DEF(NULL, "enable-dual-filter", 1,
+            "Enable dual filter "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_intra_edge_filter =
+    ARG_DEF(NULL, "enable-intra-edge-filter", 1,
+            "Enable intra edge filtering "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_order_hint =
+    ARG_DEF(NULL, "enable-order-hint", 1,
+            "Enable order hint "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_tx64 =
+    ARG_DEF(NULL, "enable-tx64", 1,
+            "Enable 64-pt transform (0: false, 1: true (default))");
+static const arg_def_t enable_dist_wtd_comp =
+    ARG_DEF(NULL, "enable-dist-wtd-comp", 1,
+            "Enable distance-weighted compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_masked_comp =
+    ARG_DEF(NULL, "enable-masked-comp", 1,
+            "Enable masked (wedge/diff-wtd) compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_interintra_comp =
+    ARG_DEF(NULL, "enable-interintra-comp", 1,
+            "Enable interintra compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_smooth_interintra =
+    ARG_DEF(NULL, "enable-smooth-interintra", 1,
+            "Enable smooth interintra mode "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_diff_wtd_comp =
+    ARG_DEF(NULL, "enable-diff-wtd-comp", 1,
+            "Enable difference-weighted compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_interinter_wedge =
+    ARG_DEF(NULL, "enable-interinter-wedge", 1,
+            "Enable interinter wedge compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_interintra_wedge =
+    ARG_DEF(NULL, "enable-interintra-wedge", 1,
+            "Enable interintra wedge compound "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_global_motion =
+    ARG_DEF(NULL, "enable-global-motion", 1,
+            "Enable global motion "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_warped_motion =
+    ARG_DEF(NULL, "enable-warped-motion", 1,
+            "Enable local warped motion "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_filter_intra =
+    ARG_DEF(NULL, "enable-filter-intra", 1,
+            "Enable filter intra prediction mode "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_smooth_intra =
+    ARG_DEF(NULL, "enable-smooth-intra", 1,
+            "Enable smooth intra prediction modes "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_paeth_intra =
+    ARG_DEF(NULL, "enable-paeth-intra", 1,
+            "Enable Paeth intra prediction mode (0: false, 1: true (default))");
+static const arg_def_t enable_cfl_intra =
+    ARG_DEF(NULL, "enable-cfl-intra", 1,
+            "Enable chroma from luma intra prediction mode "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_obmc = ARG_DEF(
+    NULL, "enable-obmc", 1, "Enable OBMC (0: false, 1: true (default))");
+static const arg_def_t enable_palette =
+    ARG_DEF(NULL, "enable-palette", 1,
+            "Enable palette prediction mode (0: false, 1: true (default))");
+static const arg_def_t enable_intrabc =
+    ARG_DEF(NULL, "enable-intrabc", 1,
+            "Enable intra block copy prediction mode "
+            "(0: false, 1: true (default))");
+static const arg_def_t enable_angle_delta =
+    ARG_DEF(NULL, "enable-angle-delta", 1,
+            "Enable intra angle delta (0: false, 1: true (default))");
 static const arg_def_t disable_trellis_quant =
     ARG_DEF(NULL, "disable-trellis-quant", 1,
             "Disable trellis optimization of quantized coefficients (0: false ("
@@ -445,6 +530,14 @@ static const arg_def_t qm_min = ARG_DEF(
     NULL, "qm-min", 1, "Min quant matrix flatness (0..15), default is 8");
 static const arg_def_t qm_max = ARG_DEF(
     NULL, "qm-max", 1, "Max quant matrix flatness (0..15), default is 15");
+static const arg_def_t reduced_tx_type_set = ARG_DEF(
+    NULL, "reduced-tx-type-set", 1, "Use reduced set of transform types");
+static const arg_def_t use_intra_dct_only =
+    ARG_DEF(NULL, "use-intra-dct-only", 1, "Use DCT only for INTRA modes");
+static const arg_def_t use_inter_dct_only =
+    ARG_DEF(NULL, "use-inter-dct-only", 1, "Use DCT only for INTER modes");
+static const arg_def_t quant_b_adapt =
+    ARG_DEF(NULL, "quant-b-adapt", 1, "Use adaptive quantize_b");
 #if CONFIG_DIST_8X8
 static const arg_def_t enable_dist_8x8 =
     ARG_DEF(NULL, "enable-dist-8x8", 1,
@@ -512,6 +605,12 @@ static const arg_def_t min_gf_interval = ARG_DEF(
 static const arg_def_t max_gf_interval = ARG_DEF(
     NULL, "max-gf-interval", 1,
     "max gf/arf frame interval (default 0, indicating in-built behavior)");
+static const arg_def_t gf_max_pyr_height =
+    ARG_DEF(NULL, "gf-max-pyr-height", 1,
+            "maximum height for GF group pyramid structure (1 to 4 (default))");
+static const arg_def_t max_reference_frames = ARG_DEF(
+    NULL, "max-reference-frames", 1,
+    "maximum number of reference frames allowed per frame (3 to 7 (default))");
 
 static const struct arg_enum_list color_primaries_enum[] = {
   { "bt709", AOM_CICP_CP_BT_709 },
@@ -624,6 +723,7 @@ static const arg_def_t *av1_args[] = { &cpu_used_av1,
                                        &rowmtarg,
                                        &tile_cols,
                                        &tile_rows,
+                                       &enable_tpl_model,
                                        &arnr_maxframes,
                                        &arnr_strength,
                                        &tune_metric,
@@ -634,10 +734,36 @@ static const arg_def_t *av1_args[] = { &cpu_used_av1,
                                        &lossless,
                                        &enable_cdef,
                                        &enable_restoration,
+                                       &enable_rect_partitions,
+                                       &enable_dual_filter,
+                                       &enable_intra_edge_filter,
+                                       &enable_order_hint,
+                                       &enable_tx64,
+                                       &enable_dist_wtd_comp,
+                                       &enable_masked_comp,
+                                       &enable_interintra_comp,
+                                       &enable_smooth_interintra,
+                                       &enable_diff_wtd_comp,
+                                       &enable_interinter_wedge,
+                                       &enable_interintra_wedge,
+                                       &enable_global_motion,
+                                       &enable_warped_motion,
+                                       &enable_filter_intra,
+                                       &enable_smooth_intra,
+                                       &enable_paeth_intra,
+                                       &enable_cfl_intra,
+                                       &enable_obmc,
+                                       &enable_palette,
+                                       &enable_intrabc,
+                                       &enable_angle_delta,
                                        &disable_trellis_quant,
                                        &enable_qm,
                                        &qm_min,
                                        &qm_max,
+                                       &reduced_tx_type_set,
+                                       &use_intra_dct_only,
+                                       &use_inter_dct_only,
+                                       &quant_b_adapt,
 #if CONFIG_DIST_8X8
                                        &enable_dist_8x8,
 #endif
@@ -655,6 +781,7 @@ static const arg_def_t *av1_args[] = { &cpu_used_av1,
                                        &input_chroma_sample_position,
                                        &min_gf_interval,
                                        &max_gf_interval,
+                                       &gf_max_pyr_height,
                                        &superblock_size,
                                        &num_tg,
                                        &mtu_size,
@@ -664,7 +791,8 @@ static const arg_def_t *av1_args[] = { &cpu_used_av1,
 #if CONFIG_DENOISE
                                        &denoise_noise_level,
                                        &denoise_block_size,
-#endif
+#endif  // CONFIG_DENOISE
+                                       &max_reference_frames,
                                        &enable_ref_frame_mvs,
                                        &bitdeptharg,
                                        &inbitdeptharg,
@@ -681,6 +809,7 @@ static const int av1_arg_ctrl_map[] = { AOME_SET_CPUUSED,
                                         AV1E_SET_ROW_MT,
                                         AV1E_SET_TILE_COLUMNS,
                                         AV1E_SET_TILE_ROWS,
+                                        AV1E_SET_ENABLE_TPL_MODEL,
                                         AOME_SET_ARNR_MAXFRAMES,
                                         AOME_SET_ARNR_STRENGTH,
                                         AOME_SET_TUNING,
@@ -691,10 +820,36 @@ static const int av1_arg_ctrl_map[] = { AOME_SET_CPUUSED,
                                         AV1E_SET_LOSSLESS,
                                         AV1E_SET_ENABLE_CDEF,
                                         AV1E_SET_ENABLE_RESTORATION,
+                                        AV1E_SET_ENABLE_RECT_PARTITIONS,
+                                        AV1E_SET_ENABLE_DUAL_FILTER,
+                                        AV1E_SET_ENABLE_INTRA_EDGE_FILTER,
+                                        AV1E_SET_ENABLE_ORDER_HINT,
+                                        AV1E_SET_ENABLE_TX64,
+                                        AV1E_SET_ENABLE_DIST_WTD_COMP,
+                                        AV1E_SET_ENABLE_MASKED_COMP,
+                                        AV1E_SET_ENABLE_INTERINTRA_COMP,
+                                        AV1E_SET_ENABLE_SMOOTH_INTERINTRA,
+                                        AV1E_SET_ENABLE_DIFF_WTD_COMP,
+                                        AV1E_SET_ENABLE_INTERINTER_WEDGE,
+                                        AV1E_SET_ENABLE_INTERINTRA_WEDGE,
+                                        AV1E_SET_ENABLE_GLOBAL_MOTION,
+                                        AV1E_SET_ENABLE_WARPED_MOTION,
+                                        AV1E_SET_ENABLE_FILTER_INTRA,
+                                        AV1E_SET_ENABLE_SMOOTH_INTRA,
+                                        AV1E_SET_ENABLE_PAETH_INTRA,
+                                        AV1E_SET_ENABLE_CFL_INTRA,
+                                        AV1E_SET_ENABLE_OBMC,
+                                        AV1E_SET_ENABLE_PALETTE,
+                                        AV1E_SET_ENABLE_INTRABC,
+                                        AV1E_SET_ENABLE_ANGLE_DELTA,
                                         AV1E_SET_DISABLE_TRELLIS_QUANT,
                                         AV1E_SET_ENABLE_QM,
                                         AV1E_SET_QM_MIN,
                                         AV1E_SET_QM_MAX,
+                                        AV1E_SET_REDUCED_TX_TYPE_SET,
+                                        AV1E_SET_INTRA_DCT_ONLY,
+                                        AV1E_SET_INTER_DCT_ONLY,
+                                        AV1E_SET_QUANT_B_ADAPT,
 #if CONFIG_DIST_8X8
                                         AV1E_SET_ENABLE_DIST_8X8,
 #endif
@@ -712,6 +867,7 @@ static const int av1_arg_ctrl_map[] = { AOME_SET_CPUUSED,
                                         AV1E_SET_CHROMA_SAMPLE_POSITION,
                                         AV1E_SET_MIN_GF_INTERVAL,
                                         AV1E_SET_MAX_GF_INTERVAL,
+                                        AV1E_SET_GF_MAX_PYRAMID_HEIGHT,
                                         AV1E_SET_SUPERBLOCK_SIZE,
                                         AV1E_SET_NUM_TG,
                                         AV1E_SET_MTU,
@@ -721,18 +877,15 @@ static const int av1_arg_ctrl_map[] = { AOME_SET_CPUUSED,
 #if CONFIG_DENOISE
                                         AV1E_SET_DENOISE_NOISE_LEVEL,
                                         AV1E_SET_DENOISE_BLOCK_SIZE,
-#endif
+#endif  // CONFIG_DENOISE
+                                        AV1E_SET_MAX_REFERENCE_FRAMES,
                                         AV1E_SET_ENABLE_REF_FRAME_MVS,
-                                        AV1E_SET_ENABLE_DF,
-                                        AV1E_SET_ENABLE_ORDER_HINT,
-                                        AV1E_SET_ENABLE_JNT_COMP,
-                                        AV1E_SET_ENABLE_SUPERRES,
                                         0 };
 #endif  // CONFIG_AV1_ENCODER
 
 static const arg_def_t *no_args[] = { NULL };
 
-void show_help(FILE *fout, int shorthelp) {
+static void show_help(FILE *fout, int shorthelp) {
   fprintf(fout, "Usage: %s <options> -o dst_filename src_filename \n",
           exec_name);
 
@@ -845,14 +998,17 @@ static void validate_positive_rational(const char *msg,
   if (!rat->den) die("Error: %s has zero denominator\n", msg);
 }
 
-static void parse_global_config(struct AvxEncoderConfig *global, int *argc,
+/* Parses global config arguments into the AvxEncoderConfig. Note that
+ * argv is modified and overwrites all parsed arguments.
+ */
+static void parse_global_config(struct AvxEncoderConfig *global, int argc,
                                 char ***argv) {
   char **argi, **argj;
   struct arg arg;
   const int num_encoder = get_aom_encoder_count();
   char **argv_local = (char **)*argv;
 #if CONFIG_FILEOPTIONS
-  int argc_local = *argc;
+  int argc_local = argc;
 #endif
   if (num_encoder < 1) die("Error: no valid encoder available\n");
 
@@ -861,6 +1017,7 @@ static void parse_global_config(struct AvxEncoderConfig *global, int *argc,
   global->codec = get_aom_encoder_by_index(num_encoder - 1);
   global->passes = 0;
   global->color_type = I420;
+  global->csp = AOM_CSP_UNKNOWN;
 
 #if CONFIG_FILEOPTIONS
   const char *cfg = NULL;
@@ -900,6 +1057,10 @@ static void parse_global_config(struct AvxEncoderConfig *global, int *argc,
 
       if (global->pass < 1 || global->pass > 2)
         die("Error: Invalid pass selected (%d)\n", global->pass);
+    } else if (arg_match(&arg, &input_chroma_sample_position, argi)) {
+      global->csp = arg_parse_enum(&arg);
+      /* Flag is used by later code as well, preserve it. */
+      argj++;
     } else if (arg_match(&arg, &usage, argi))
       global->usage = arg_parse_uint(&arg);
     else if (arg_match(&arg, &good_dl, argi))
@@ -963,7 +1124,8 @@ static void parse_global_config(struct AvxEncoderConfig *global, int *argc,
   }
 }
 
-static void open_input_file(struct AvxInputContext *input) {
+static void open_input_file(struct AvxInputContext *input,
+                            aom_chroma_sample_position_t csp) {
   /* Parse certain options from the input file, if possible */
   input->file = strcmp(input->filename, "-") ? fopen(input->filename, "rb")
                                              : set_binary_mode(stdin);
@@ -989,7 +1151,7 @@ static void open_input_file(struct AvxInputContext *input) {
   input->detect.position = 0;
 
   if (input->detect.buf_read == 4 && file_is_y4m(input->detect.buf)) {
-    if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4,
+    if (y4m_input_open(&input->y4m, input->file, input->detect.buf, 4, csp,
                        input->only_i420) >= 0) {
       input->file_type = FILE_TYPE_Y4M;
       input->width = input->y4m.pic_w;
@@ -1195,6 +1357,7 @@ static int parse_stream_params(struct AvxEncoderConfig *global,
       config->cfg.g_lag_in_frames = arg_parse_uint(&arg);
     } else if (arg_match(&arg, &large_scale_tile, argi)) {
       config->cfg.large_scale_tile = arg_parse_uint(&arg);
+      if (config->cfg.large_scale_tile) global->codec = get_aom_lst_encoder();
     } else if (arg_match(&arg, &monochrome, argi)) {
       config->cfg.monochrome = 1;
     } else if (arg_match(&arg, &full_still_picture_hdr, argi)) {
@@ -1306,13 +1469,11 @@ static void validate_stream_config(const struct stream_state *stream,
         " and --height (-h)",
         stream->index);
 
-  // Check that the codec bit depth is greater than the input bit depth.
-  if (stream->config.cfg.g_input_bit_depth >
-      (unsigned int)stream->config.cfg.g_bit_depth) {
-    fatal("Stream %d: codec bit depth (%d) less than input bit depth (%d)",
-          stream->index, (int)stream->config.cfg.g_bit_depth,
-          stream->config.cfg.g_input_bit_depth);
-  }
+  /* Even if bit depth is set on the command line flag to be lower,
+   * it is upgraded to at least match the input bit depth.
+   */
+  assert(stream->config.cfg.g_input_bit_depth <=
+         (unsigned int)stream->config.cfg.g_bit_depth);
 
   for (streami = stream; streami; streami = streami->next) {
     /* All streams require output files */
@@ -1378,6 +1539,7 @@ static const char *image_format_to_string(aom_img_fmt_t f) {
     case AOM_IMG_FMT_I422: return "I422";
     case AOM_IMG_FMT_I444: return "I444";
     case AOM_IMG_FMT_YV12: return "YV12";
+    case AOM_IMG_FMT_YV1216: return "YV1216";
     case AOM_IMG_FMT_I42016: return "I42016";
     case AOM_IMG_FMT_I42216: return "I42216";
     case AOM_IMG_FMT_I44416: return "I44416";
@@ -1463,8 +1625,11 @@ static void open_output_file(struct stream_state *stream,
 #if CONFIG_WEBM_IO
   if (stream->config.write_webm) {
     stream->webm_ctx.stream = stream->file;
-    write_webm_file_header(&stream->webm_ctx, cfg, stream->config.stereo_fmt,
-                           global->codec->fourcc, pixel_aspect_ratio);
+    if (write_webm_file_header(&stream->webm_ctx, &stream->encoder, cfg,
+                               stream->config.stereo_fmt, global->codec->fourcc,
+                               pixel_aspect_ratio) != 0) {
+      fatal("WebM writer initialization failed.");
+    }
   }
 #else
   (void)pixel_aspect_ratio;
@@ -1483,7 +1648,9 @@ static void close_output_file(struct stream_state *stream,
 
 #if CONFIG_WEBM_IO
   if (stream->config.write_webm) {
-    write_webm_file_footer(&stream->webm_ctx);
+    if (write_webm_file_footer(&stream->webm_ctx) != 0) {
+      fatal("WebM writer finalization failed.");
+    }
   }
 #endif
 
@@ -1699,16 +1866,16 @@ static void get_cx_data(struct stream_state *stream,
 
     switch (pkt->kind) {
       case AOM_CODEC_CX_FRAME_PKT:
-        if (!(pkt->data.frame.flags & AOM_FRAME_IS_FRAGMENT)) {
-          stream->frames_out++;
-        }
+        ++stream->frames_out;
         if (!global->quiet)
           fprintf(stderr, " %6luF", (unsigned long)pkt->data.frame.sz);
 
         update_rate_histogram(stream->rate_hist, cfg, pkt);
 #if CONFIG_WEBM_IO
         if (stream->config.write_webm) {
-          write_webm_block(&stream->webm_ctx, cfg, pkt);
+          if (write_webm_block(&stream->webm_ctx, cfg, pkt) != 0) {
+            fatal("WebM writer failed.");
+          }
         }
 #endif
         if (!stream->config.write_webm) {
@@ -1721,12 +1888,10 @@ static void get_cx_data(struct stream_state *stream,
             } else {
               fsize += pkt->data.frame.sz;
 
-              if (!(pkt->data.frame.flags & AOM_FRAME_IS_FRAGMENT)) {
-                const FileOffset currpos = ftello(stream->file);
-                fseeko(stream->file, ivf_header_pos, SEEK_SET);
-                ivf_write_frame_size(stream->file, fsize);
-                fseeko(stream->file, currpos, SEEK_SET);
-              }
+              const FileOffset currpos = ftello(stream->file);
+              fseeko(stream->file, ivf_header_pos, SEEK_SET);
+              ivf_write_frame_size(stream->file, fsize);
+              fseeko(stream->file, currpos, SEEK_SET);
             }
           }
 
@@ -1913,7 +2078,7 @@ int main(int argc, const char **argv_) {
    * codec.
    */
   argv = argv_dup(argc - 1, argv_ + 1);
-  parse_global_config(&global, &argc, &argv);
+  parse_global_config(&global, argc, &argv);
 
 #if CONFIG_FILEOPTIONS
   if (argc < 2) usage_exit();
@@ -1950,6 +2115,10 @@ int main(int argc, const char **argv_) {
   FOREACH_STREAM(stream, streams) {
     check_encoder_config(global.disable_warning_prompt, &global,
                          &stream->config.cfg);
+
+    // If large_scale_tile = 1, only support to output to ivf format.
+    if (stream->config.cfg.large_scale_tile && !stream->config.write_ivf)
+      die("only support ivf output format while large-scale-tile=1\n");
   }
 
   /* Handle non-option arguments */
@@ -1969,7 +2138,7 @@ int main(int argc, const char **argv_) {
     int64_t average_rate = -1;
     int64_t lagged_count = 0;
 
-    open_input_file(&input);
+    open_input_file(&input, global.csp);
 
     /* If the input file doesn't specify its w/h (raw files), try to get
      * the data from the first stream's configuration.
@@ -2069,6 +2238,18 @@ int main(int argc, const char **argv_) {
           default: break;
         }
       }
+      /* Automatically set the codec bit depth to match the input bit depth.
+       * Upgrade the profile if required. */
+      if (stream->config.cfg.g_input_bit_depth >
+          (unsigned int)stream->config.cfg.g_bit_depth) {
+        stream->config.cfg.g_bit_depth = stream->config.cfg.g_input_bit_depth;
+        if (!global.quiet) {
+          fprintf(stderr,
+                  "Warning: automatically updating bit depth to %d to "
+                  "match input format.\n",
+                  stream->config.cfg.g_input_bit_depth);
+        }
+      }
       if (stream->config.cfg.g_bit_depth > 10) {
         switch (stream->config.cfg.g_profile) {
           case 0:
@@ -2153,10 +2334,11 @@ int main(int argc, const char **argv_) {
     }
 
     FOREACH_STREAM(stream, streams) { setup_pass(stream, &global, pass); }
+    FOREACH_STREAM(stream, streams) { initialize_encoder(stream, &global); }
     FOREACH_STREAM(stream, streams) {
       open_output_file(stream, &global, &input.pixel_aspect_ratio);
     }
-    FOREACH_STREAM(stream, streams) { initialize_encoder(stream, &global); }
+
     if (strcmp(global.codec->name, "av1") == 0 ||
         strcmp(global.codec->name, "av1") == 0) {
       // Check to see if at least one stream uses 16 bit internal.
diff --git a/third_party/aom/apps/aomenc.h b/third_party/aom/apps/aomenc.h
index 7c23df0068..5e59c1ac5d 100644
--- a/third_party/aom/apps/aomenc.h
+++ b/third_party/aom/apps/aomenc.h
@@ -53,6 +53,7 @@ struct AvxEncoderConfig {
   int disable_warnings;
   int disable_warning_prompt;
   int experimental_bitstream;
+  aom_chroma_sample_position_t csp;
 };
 
 #ifdef __cplusplus
diff --git a/third_party/aom/av1/av1.cmake b/third_party/aom/av1/av1.cmake
index 3a7cd7ee13..0b4901a619 100644
--- a/third_party/aom/av1/av1.cmake
+++ b/third_party/aom/av1/av1.cmake
@@ -100,7 +100,6 @@ list(APPEND AOM_AV1_DECODER_SOURCES
             "${AOM_ROOT}/av1/decoder/decodetxb.h"
             "${AOM_ROOT}/av1/decoder/detokenize.c"
             "${AOM_ROOT}/av1/decoder/detokenize.h"
-            "${AOM_ROOT}/av1/decoder/dthread.c"
             "${AOM_ROOT}/av1/decoder/dthread.h"
             "${AOM_ROOT}/av1/decoder/obu.h"
             "${AOM_ROOT}/av1/decoder/obu.c")
@@ -117,6 +116,8 @@ list(APPEND AOM_AV1_ENCODER_SOURCES
             "${AOM_ROOT}/av1/encoder/av1_fwd_txfm1d.h"
             "${AOM_ROOT}/av1/encoder/av1_fwd_txfm1d_cfg.h"
             "${AOM_ROOT}/av1/encoder/av1_fwd_txfm2d.c"
+            "${AOM_ROOT}/av1/encoder/av1_multi_thread.c"
+            "${AOM_ROOT}/av1/encoder/av1_multi_thread.h"
             "${AOM_ROOT}/av1/encoder/av1_quantize.c"
             "${AOM_ROOT}/av1/encoder/av1_quantize.h"
             "${AOM_ROOT}/av1/encoder/bitstream.c"
@@ -136,6 +137,8 @@ list(APPEND AOM_AV1_ENCODER_SOURCES
             "${AOM_ROOT}/av1/encoder/encodemb.h"
             "${AOM_ROOT}/av1/encoder/encodemv.c"
             "${AOM_ROOT}/av1/encoder/encodemv.h"
+            "${AOM_ROOT}/av1/encoder/encode_strategy.c"
+            "${AOM_ROOT}/av1/encoder/encode_strategy.h"
             "${AOM_ROOT}/av1/encoder/encoder.c"
             "${AOM_ROOT}/av1/encoder/encoder.h"
             "${AOM_ROOT}/av1/encoder/encodetxb.c"
@@ -188,6 +191,8 @@ list(APPEND AOM_AV1_ENCODER_SOURCES
             "${AOM_ROOT}/av1/encoder/temporal_filter.h"
             "${AOM_ROOT}/av1/encoder/tokenize.c"
             "${AOM_ROOT}/av1/encoder/tokenize.h"
+            "${AOM_ROOT}/av1/encoder/tpl_model.c"
+            "${AOM_ROOT}/av1/encoder/tpl_model.h"
             "${AOM_ROOT}/av1/encoder/wedge_utils.c"
             "${AOM_ROOT}/third_party/fastfeat/fast.c"
             "${AOM_ROOT}/third_party/fastfeat/fast.h"
@@ -263,6 +268,8 @@ list(APPEND AOM_AV1_ENCODER_INTRIN_SSE2
             "${AOM_ROOT}/av1/encoder/x86/highbd_block_error_intrin_sse2.c"
             "${AOM_ROOT}/av1/encoder/x86/wedge_utils_sse2.c")
 
+list(APPEND AOM_AV1_ENCODER_INTRIN_SSE3 "${AOM_ROOT}/av1/encoder/x86/ml_sse3.c")
+
 list(APPEND AOM_AV1_ENCODER_ASM_SSSE3_X86_64
             "${AOM_ROOT}/av1/encoder/x86/av1_quantize_ssse3_x86_64.asm")
 
@@ -273,16 +280,20 @@ list(APPEND AOM_AV1_ENCODER_INTRIN_SSE4_1
             "${AOM_ROOT}/av1/encoder/x86/corner_match_sse4.c"
             "${AOM_ROOT}/av1/encoder/x86/encodetxb_sse4.c"
             "${AOM_ROOT}/av1/encoder/x86/highbd_fwd_txfm_sse4.c"
+            "${AOM_ROOT}/av1/encoder/x86/rdopt_sse4.c"
             "${AOM_ROOT}/av1/encoder/x86/pickrst_sse4.c")
 
 list(APPEND AOM_AV1_ENCODER_INTRIN_AVX2
             "${AOM_ROOT}/av1/encoder/x86/av1_quantize_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/av1_highbd_quantize_avx2.c"
+            "${AOM_ROOT}/av1/encoder/x86/corner_match_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/error_intrin_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/av1_fwd_txfm_avx2.h"
             "${AOM_ROOT}/av1/encoder/x86/av1_fwd_txfm2d_avx2.c"
+            "${AOM_ROOT}/av1/encoder/x86/highbd_fwd_txfm_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/wedge_utils_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/encodetxb_avx2.c"
+            "${AOM_ROOT}/av1/encoder/x86/rdopt_avx2.c"
             "${AOM_ROOT}/av1/encoder/x86/pickrst_avx2.c")
 
 list(APPEND AOM_AV1_ENCODER_INTRIN_NEON
@@ -379,6 +390,14 @@ function(setup_av1_targets)
     endif()
   endif()
 
+  if(HAVE_SSE3)
+    require_compiler_flag_nomsvc("-msse3" NO)
+    if(CONFIG_AV1_ENCODER)
+      add_intrinsics_object_library("-msse3" "sse3" "aom_av1_encoder"
+                                    "AOM_AV1_ENCODER_INTRIN_SSE3" "aom")
+    endif()
+  endif()
+
   if(HAVE_SSSE3)
     require_compiler_flag_nomsvc("-mssse3" NO)
     add_intrinsics_object_library("-mssse3" "ssse3" "aom_av1_common"
diff --git a/third_party/aom/av1/av1_cx_iface.c b/third_party/aom/av1/av1_cx_iface.c
index 3295f618aa..285e6a5086 100644
--- a/third_party/aom/av1/av1_cx_iface.c
+++ b/third_party/aom/av1/av1_cx_iface.c
@@ -39,10 +39,12 @@ struct av1_extracfg {
   unsigned int row_mt;
   unsigned int tile_columns;  // log2 number of tile columns
   unsigned int tile_rows;     // log2 number of tile rows
+  unsigned int enable_tpl_model;
   unsigned int arnr_max_frames;
   unsigned int arnr_strength;
   unsigned int min_gf_interval;
   unsigned int max_gf_interval;
+  unsigned int gf_max_pyr_height;
   aom_tune_metric tuning;
   unsigned int cq_level;  // constrained quality level
   unsigned int rc_max_intra_bitrate_pct;
@@ -51,6 +53,7 @@ struct av1_extracfg {
   unsigned int lossless;
   unsigned int enable_cdef;
   unsigned int enable_restoration;
+  unsigned int enable_obmc;
   unsigned int disable_trellis_quant;
   unsigned int enable_qm;
   unsigned int qm_y;
@@ -66,7 +69,7 @@ struct av1_extracfg {
 
   aom_timing_info_type_t timing_info_type;
   unsigned int frame_parallel_decoding_mode;
-  int use_dual_filter;
+  int enable_dual_filter;
   AQ_MODE aq_mode;
   DELTAQ_MODE deltaq_mode;
   unsigned int frame_periodic_boost;
@@ -88,13 +91,31 @@ struct av1_extracfg {
   const char *film_grain_table_filename;
   unsigned int motion_vector_unit_test;
   unsigned int cdf_update_mode;
-  int enable_order_hint;
-  int enable_jnt_comp;
-  int enable_ref_frame_mvs;  // sequence level
-  int allow_ref_frame_mvs;   // frame level
-  int enable_warped_motion;  // sequence level
-  int allow_warped_motion;   // frame level
+  int enable_rect_partitions;    // enable rectangular partitions for sequence
+  int enable_intra_edge_filter;  // enable intra-edge filter for sequence
+  int enable_order_hint;         // enable order hint for sequence
+  int enable_tx64;               // enable 64-pt transform usage for sequence
+  int enable_dist_wtd_comp;      // enable dist wtd compound for sequence
+  int max_reference_frames;      // maximum number of references per frame
+  int enable_ref_frame_mvs;      // sequence level
+  int allow_ref_frame_mvs;       // frame level
+  int enable_masked_comp;        // enable masked compound for sequence
+  int enable_interintra_comp;    // enable interintra compound for sequence
+  int enable_smooth_interintra;  // enable smooth interintra mode usage
+  int enable_diff_wtd_comp;      // enable diff-wtd compound usage
+  int enable_interinter_wedge;   // enable interinter-wedge compound usage
+  int enable_interintra_wedge;   // enable interintra-wedge compound usage
+  int enable_global_motion;      // enable global motion usage for sequence
+  int enable_warped_motion;      // sequence level
+  int allow_warped_motion;       // frame level
+  int enable_filter_intra;       // enable filter intra for sequence
+  int enable_smooth_intra;       // enable smooth intra modes for sequence
+  int enable_paeth_intra;        // enable Peeth intra mode for sequence
+  int enable_cfl_intra;          // enable CFL uv intra mode for sequence
   int enable_superres;
+  int enable_palette;
+  int enable_intrabc;
+  int enable_angle_delta;
 #if CONFIG_DENOISE
   float noise_level;
   int noise_block_size;
@@ -102,6 +123,10 @@ struct av1_extracfg {
 
   unsigned int chroma_subsampling_x;
   unsigned int chroma_subsampling_y;
+  int reduced_tx_type_set;
+  int use_intra_dct_only;
+  int use_inter_dct_only;
+  int quant_b_adapt;
 };
 
 static struct av1_extracfg default_extra_cfg = {
@@ -111,13 +136,15 @@ static struct av1_extracfg default_extra_cfg = {
   0,                       // noise_sensitivity
   CONFIG_SHARP_SETTINGS,   // sharpness
   0,                       // static_thresh
-  0,                       // row_mt
+  1,                       // row_mt
   0,                       // tile_columns
   0,                       // tile_rows
+  0,                       // enable_tpl_model
   7,                       // arnr_max_frames
   5,                       // arnr_strength
   0,                       // min_gf_interval; 0 -> default decision
   0,                       // max_gf_interval; 0 -> default decision
+  4,                       // gf_max_pyr_height
   AOM_TUNE_PSNR,           // tuning
   10,                      // cq_level
   0,                       // rc_max_intra_bitrate_pct
@@ -126,6 +153,7 @@ static struct av1_extracfg default_extra_cfg = {
   0,                       // lossless
   !CONFIG_SHARP_SETTINGS,  // enable_cdef
   1,                       // enable_restoration
+  1,                       // enable_obmc
   0,                       // disable_trellis_quant
   0,                       // enable_qm
   DEFAULT_QM_Y,            // qm_y
@@ -161,19 +189,41 @@ static struct av1_extracfg default_extra_cfg = {
   0,                            // film_grain_table_filename
   0,                            // motion_vector_unit_test
   1,                            // CDF update mode
+  1,                            // enable rectangular partitions
+  1,                            // enable intra edge filter
   1,                            // frame order hint
-  1,                            // jnt_comp
+  1,                            // enable 64-pt transform usage
+  1,                            // dist-wtd compound
+  7,                            // max_reference_frames
   1,                            // enable_ref_frame_mvs sequence level
   1,                            // allow ref_frame_mvs frame level
+  1,                            // enable masked compound at sequence level
+  1,                            // enable interintra compound at sequence level
+  1,                            // enable smooth interintra mode
+  1,                            // enable difference-weighted compound
+  1,                            // enable interinter wedge compound
+  1,                            // enable interintra wedge compound
+  1,                            // enable_global_motion usage
   1,                            // enable_warped_motion at sequence level
   1,                            // allow_warped_motion at frame level
+  1,                            // enable filter intra at sequence level
+  1,                            // enable smooth intra modes usage for sequence
+  1,                            // enable Paeth intra mode usage for sequence
+  1,                            // enable CFL uv intra mode usage for sequence
   1,                            // superres
+  1,                            // enable palette
+  1,                            // enable intrabc
+  1,                            // enable angle delta
 #if CONFIG_DENOISE
   0,   // noise_level
   32,  // noise_block_size
 #endif
   0,  // chroma_subsampling_x
   0,  // chroma_subsampling_y
+  0,  // reduced_tx_type_set
+  0,  // use_intra_dct_only
+  0,  // use_inter_dct_only
+  0,  // quant_b_adapt
 };
 
 struct aom_codec_alg_priv {
@@ -245,7 +295,7 @@ static aom_codec_err_t validate_config(aom_codec_alg_priv_t *ctx,
   RANGE_CHECK_HI(extra_cfg, aq_mode, AQ_MODE_COUNT - 1);
   RANGE_CHECK_HI(extra_cfg, deltaq_mode, DELTAQ_MODE_COUNT - 1);
   RANGE_CHECK_HI(extra_cfg, frame_periodic_boost, 1);
-  RANGE_CHECK_HI(cfg, g_threads, 64);
+  RANGE_CHECK_HI(cfg, g_threads, MAX_NUM_THREADS);
   RANGE_CHECK_HI(cfg, g_lag_in_frames, MAX_LAG_BUFFERS);
   RANGE_CHECK(cfg, rc_end_usage, AOM_VBR, AOM_Q);
   RANGE_CHECK_HI(cfg, rc_undershoot_pct, 100);
@@ -260,6 +310,7 @@ static aom_codec_err_t validate_config(aom_codec_alg_priv_t *ctx,
     RANGE_CHECK(extra_cfg, max_gf_interval, MAX(2, extra_cfg->min_gf_interval),
                 (MAX_LAG_BUFFERS - 1));
   }
+  RANGE_CHECK(extra_cfg, gf_max_pyr_height, 1, 4);
 
   RANGE_CHECK_HI(cfg, rc_resize_mode, RESIZE_MODES - 1);
   RANGE_CHECK(cfg, rc_resize_denominator, SCALE_NUMERATOR,
@@ -284,7 +335,7 @@ static aom_codec_err_t validate_config(aom_codec_alg_priv_t *ctx,
         "or kf_max_dist instead.");
 
   RANGE_CHECK_HI(extra_cfg, motion_vector_unit_test, 2);
-  RANGE_CHECK_HI(extra_cfg, enable_auto_alt_ref, 2);
+  RANGE_CHECK_HI(extra_cfg, enable_auto_alt_ref, 1);
   RANGE_CHECK_HI(extra_cfg, enable_auto_bwd_ref, 2);
   RANGE_CHECK(extra_cfg, cpu_used, 0, 8);
   RANGE_CHECK_HI(extra_cfg, noise_sensitivity, 6);
@@ -376,6 +427,7 @@ static aom_codec_err_t validate_config(aom_codec_alg_priv_t *ctx,
 #endif
   }
 
+  RANGE_CHECK(extra_cfg, max_reference_frames, 3, 7);
   RANGE_CHECK_HI(extra_cfg, chroma_subsampling_x, 1);
   RANGE_CHECK_HI(extra_cfg, chroma_subsampling_y, 1);
 
@@ -387,6 +439,7 @@ static aom_codec_err_t validate_img(aom_codec_alg_priv_t *ctx,
   switch (img->fmt) {
     case AOM_IMG_FMT_YV12:
     case AOM_IMG_FMT_I420:
+    case AOM_IMG_FMT_YV1216:
     case AOM_IMG_FMT_I42016: break;
     case AOM_IMG_FMT_I444:
     case AOM_IMG_FMT_I44416:
@@ -420,6 +473,7 @@ static int get_image_bps(const aom_image_t *img) {
     case AOM_IMG_FMT_I420: return 12;
     case AOM_IMG_FMT_I422: return 16;
     case AOM_IMG_FMT_I444: return 24;
+    case AOM_IMG_FMT_YV1216:
     case AOM_IMG_FMT_I42016: return 24;
     case AOM_IMG_FMT_I42216: return 32;
     case AOM_IMG_FMT_I44416: return 48;
@@ -514,6 +568,10 @@ static aom_codec_err_t set_encoder_config(
 
   oxcf->enable_cdef = extra_cfg->enable_cdef;
   oxcf->enable_restoration = extra_cfg->enable_restoration;
+  oxcf->enable_obmc = extra_cfg->enable_obmc;
+  oxcf->enable_palette = extra_cfg->enable_palette;
+  oxcf->enable_intrabc = extra_cfg->enable_intrabc;
+  oxcf->enable_angle_delta = extra_cfg->enable_angle_delta;
   oxcf->disable_trellis_quant = extra_cfg->disable_trellis_quant;
   oxcf->using_qm = extra_cfg->enable_qm;
   oxcf->qm_y = extra_cfg->qm_y;
@@ -521,6 +579,10 @@ static aom_codec_err_t set_encoder_config(
   oxcf->qm_v = extra_cfg->qm_v;
   oxcf->qm_minlevel = extra_cfg->qm_min;
   oxcf->qm_maxlevel = extra_cfg->qm_max;
+  oxcf->reduced_tx_type_set = extra_cfg->reduced_tx_type_set;
+  oxcf->use_intra_dct_only = extra_cfg->use_intra_dct_only;
+  oxcf->use_inter_dct_only = extra_cfg->use_inter_dct_only;
+  oxcf->quant_b_adapt = extra_cfg->quant_b_adapt;
 #if CONFIG_DIST_8X8
   oxcf->using_dist_8x8 = extra_cfg->enable_dist_8x8;
   if (extra_cfg->tuning == AOM_TUNE_CDEF_DIST ||
@@ -570,6 +632,9 @@ static aom_codec_err_t set_encoder_config(
     }
   }
 
+  oxcf->enable_tpl_model =
+      extra_cfg->enable_tpl_model && (oxcf->superres_mode == SUPERRES_NONE);
+
   oxcf->maximum_buffer_size_ms = is_vbr ? 240000 : cfg->rc_buf_sz;
   oxcf->starting_buffer_level_ms = is_vbr ? 60000 : cfg->rc_buf_initial_sz;
   oxcf->optimal_buffer_level_ms = is_vbr ? 60000 : cfg->rc_buf_optimal_sz;
@@ -614,6 +679,7 @@ static aom_codec_err_t set_encoder_config(
   oxcf->arnr_strength = extra_cfg->arnr_strength;
   oxcf->min_gf_interval = extra_cfg->min_gf_interval;
   oxcf->max_gf_interval = extra_cfg->max_gf_interval;
+  oxcf->gf_max_pyr_height = extra_cfg->gf_max_pyr_height;
 
   oxcf->tuning = extra_cfg->tuning;
   oxcf->content = extra_cfg->content;
@@ -650,16 +716,39 @@ static aom_codec_err_t set_encoder_config(
 
   oxcf->monochrome = cfg->monochrome;
   oxcf->full_still_picture_hdr = cfg->full_still_picture_hdr;
-  oxcf->enable_dual_filter = extra_cfg->use_dual_filter;
+  oxcf->enable_dual_filter = extra_cfg->enable_dual_filter;
+  oxcf->enable_rect_partitions = extra_cfg->enable_rect_partitions;
+  oxcf->enable_intra_edge_filter = extra_cfg->enable_intra_edge_filter;
+  oxcf->enable_tx64 = extra_cfg->enable_tx64;
   oxcf->enable_order_hint = extra_cfg->enable_order_hint;
-  oxcf->enable_jnt_comp =
-      extra_cfg->enable_jnt_comp & extra_cfg->enable_order_hint;
+  oxcf->enable_dist_wtd_comp =
+      extra_cfg->enable_dist_wtd_comp & extra_cfg->enable_order_hint;
+  oxcf->max_reference_frames = extra_cfg->max_reference_frames;
+  if (oxcf->max_reference_frames > 3 && oxcf->max_reference_frames < 7) {
+    // TODO(urvang): Enable all possible values, after they work properly.
+    oxcf->max_reference_frames = 3;
+  }
+  oxcf->enable_masked_comp = extra_cfg->enable_masked_comp;
+  oxcf->enable_diff_wtd_comp =
+      extra_cfg->enable_masked_comp & extra_cfg->enable_diff_wtd_comp;
+  oxcf->enable_interinter_wedge =
+      extra_cfg->enable_masked_comp & extra_cfg->enable_interinter_wedge;
+  oxcf->enable_interintra_comp = extra_cfg->enable_interintra_comp;
+  oxcf->enable_smooth_interintra =
+      extra_cfg->enable_interintra_comp && extra_cfg->enable_smooth_interintra;
+  oxcf->enable_interintra_wedge =
+      extra_cfg->enable_interintra_comp & extra_cfg->enable_interintra_wedge;
   oxcf->enable_ref_frame_mvs =
       extra_cfg->enable_ref_frame_mvs & extra_cfg->enable_order_hint;
 
+  oxcf->enable_global_motion = extra_cfg->enable_global_motion;
   oxcf->enable_warped_motion = extra_cfg->enable_warped_motion;
   oxcf->allow_warped_motion =
       extra_cfg->allow_warped_motion & extra_cfg->enable_warped_motion;
+  oxcf->enable_filter_intra = extra_cfg->enable_filter_intra;
+  oxcf->enable_smooth_intra = extra_cfg->enable_smooth_intra;
+  oxcf->enable_paeth_intra = extra_cfg->enable_paeth_intra;
+  oxcf->enable_cfl_intra = extra_cfg->enable_cfl_intra;
 
   oxcf->enable_superres =
       (oxcf->superres_mode != SUPERRES_NONE) && extra_cfg->enable_superres;
@@ -701,23 +790,11 @@ static aom_codec_err_t set_encoder_config(
   oxcf->frame_periodic_boost = extra_cfg->frame_periodic_boost;
   oxcf->motion_vector_unit_test = extra_cfg->motion_vector_unit_test;
 
-#if CONFIG_REDUCED_ENCODER_BORDER
-  if (oxcf->superres_mode != SUPERRES_NONE ||
-      oxcf->resize_mode != RESIZE_NONE) {
-    warn(
-        "Superres / resize cannot be used with CONFIG_REDUCED_ENCODER_BORDER. "
-        "Disabling superres/resize.\n");
-    // return AOM_CODEC_INVALID_PARAM;
-    disable_superres(oxcf);
-    oxcf->resize_mode = RESIZE_NONE;
-    oxcf->resize_scale_denominator = SCALE_NUMERATOR;
-    oxcf->resize_kf_scale_denominator = SCALE_NUMERATOR;
-  }
-#endif  // CONFIG_REDUCED_ENCODER_BORDER
-
   oxcf->chroma_subsampling_x = extra_cfg->chroma_subsampling_x;
   oxcf->chroma_subsampling_y = extra_cfg->chroma_subsampling_y;
-
+  oxcf->border_in_pixels = (oxcf->resize_mode || oxcf->superres_mode)
+                               ? AOM_BORDER_IN_PIXELS
+                               : AOM_ENC_NO_SCALE_BORDER;
   return AOM_CODEC_OK;
 }
 
@@ -851,6 +928,13 @@ static aom_codec_err_t ctrl_set_tile_rows(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_enable_tpl_model(aom_codec_alg_priv_t *ctx,
+                                                 va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_tpl_model = CAST(AV1E_SET_ENABLE_TPL_MODEL, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_arnr_max_frames(aom_codec_alg_priv_t *ctx,
                                                 va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -923,6 +1007,13 @@ static aom_codec_err_t ctrl_set_enable_restoration(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_enable_obmc(aom_codec_alg_priv_t *ctx,
+                                            va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_obmc = CAST(AV1E_SET_ENABLE_OBMC, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_disable_trellis_quant(aom_codec_alg_priv_t *ctx,
                                                       va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -991,10 +1082,26 @@ static aom_codec_err_t ctrl_set_timing_info_type(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
-static aom_codec_err_t ctrl_set_enable_df(aom_codec_alg_priv_t *ctx,
-                                          va_list args) {
+static aom_codec_err_t ctrl_set_enable_dual_filter(aom_codec_alg_priv_t *ctx,
+                                                   va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
-  extra_cfg.use_dual_filter = CAST(AV1E_SET_ENABLE_DF, args);
+  extra_cfg.enable_dual_filter = CAST(AV1E_SET_ENABLE_DUAL_FILTER, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_rect_partitions(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_rect_partitions =
+      CAST(AV1E_SET_ENABLE_RECT_PARTITIONS, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_intra_edge_filter(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_intra_edge_filter =
+      CAST(AV1E_SET_ENABLE_INTRA_EDGE_FILTER, args);
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
@@ -1005,10 +1112,24 @@ static aom_codec_err_t ctrl_set_enable_order_hint(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
-static aom_codec_err_t ctrl_set_enable_jnt_comp(aom_codec_alg_priv_t *ctx,
-                                                va_list args) {
+static aom_codec_err_t ctrl_set_enable_tx64(aom_codec_alg_priv_t *ctx,
+                                            va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
-  extra_cfg.enable_jnt_comp = CAST(AV1E_SET_ENABLE_JNT_COMP, args);
+  extra_cfg.enable_tx64 = CAST(AV1E_SET_ENABLE_TX64, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_dist_wtd_comp(aom_codec_alg_priv_t *ctx,
+                                                     va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_dist_wtd_comp = CAST(AV1E_SET_ENABLE_DIST_WTD_COMP, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_max_reference_frames(aom_codec_alg_priv_t *ctx,
+                                                     va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.max_reference_frames = CAST(AV1E_SET_MAX_REFERENCE_FRAMES, args);
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
@@ -1026,6 +1147,59 @@ static aom_codec_err_t ctrl_set_allow_ref_frame_mvs(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_enable_masked_comp(aom_codec_alg_priv_t *ctx,
+                                                   va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_masked_comp = CAST(AV1E_SET_ENABLE_MASKED_COMP, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_interintra_comp(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_interintra_comp =
+      CAST(AV1E_SET_ENABLE_INTERINTRA_COMP, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_smooth_interintra(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_smooth_interintra =
+      CAST(AV1E_SET_ENABLE_SMOOTH_INTERINTRA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_diff_wtd_comp(aom_codec_alg_priv_t *ctx,
+                                                     va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_diff_wtd_comp = CAST(AV1E_SET_ENABLE_DIFF_WTD_COMP, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_interinter_wedge(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_interinter_wedge =
+      CAST(AV1E_SET_ENABLE_INTERINTER_WEDGE, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_interintra_wedge(
+    aom_codec_alg_priv_t *ctx, va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_interintra_wedge =
+      CAST(AV1E_SET_ENABLE_INTERINTRA_WEDGE, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_global_motion(aom_codec_alg_priv_t *ctx,
+                                                     va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_global_motion = CAST(AV1E_SET_ENABLE_GLOBAL_MOTION, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_enable_warped_motion(aom_codec_alg_priv_t *ctx,
                                                      va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -1040,6 +1214,34 @@ static aom_codec_err_t ctrl_set_allow_warped_motion(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_enable_filter_intra(aom_codec_alg_priv_t *ctx,
+                                                    va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_filter_intra = CAST(AV1E_SET_ENABLE_FILTER_INTRA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_smooth_intra(aom_codec_alg_priv_t *ctx,
+                                                    va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_smooth_intra = CAST(AV1E_SET_ENABLE_SMOOTH_INTRA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_paeth_intra(aom_codec_alg_priv_t *ctx,
+                                                   va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_paeth_intra = CAST(AV1E_SET_ENABLE_PAETH_INTRA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_cfl_intra(aom_codec_alg_priv_t *ctx,
+                                                 va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_cfl_intra = CAST(AV1E_SET_ENABLE_CFL_INTRA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_enable_superres(aom_codec_alg_priv_t *ctx,
                                                 va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -1047,6 +1249,27 @@ static aom_codec_err_t ctrl_set_enable_superres(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_enable_palette(aom_codec_alg_priv_t *ctx,
+                                               va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_palette = CAST(AV1E_SET_ENABLE_PALETTE, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_intrabc(aom_codec_alg_priv_t *ctx,
+                                               va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_intrabc = CAST(AV1E_SET_ENABLE_INTRABC, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_enable_angle_delta(aom_codec_alg_priv_t *ctx,
+                                                   va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.enable_angle_delta = CAST(AV1E_SET_ENABLE_ANGLE_DELTA, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_error_resilient_mode(aom_codec_alg_priv_t *ctx,
                                                      va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -1083,6 +1306,34 @@ static aom_codec_err_t ctrl_set_aq_mode(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_reduced_tx_type_set(aom_codec_alg_priv_t *ctx,
+                                                    va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.reduced_tx_type_set = CAST(AV1E_SET_REDUCED_TX_TYPE_SET, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_intra_dct_only(aom_codec_alg_priv_t *ctx,
+                                               va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.use_intra_dct_only = CAST(AV1E_SET_INTRA_DCT_ONLY, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_inter_dct_only(aom_codec_alg_priv_t *ctx,
+                                               va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.use_inter_dct_only = CAST(AV1E_SET_INTER_DCT_ONLY, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
+static aom_codec_err_t ctrl_set_quant_b_adapt(aom_codec_alg_priv_t *ctx,
+                                              va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.quant_b_adapt = CAST(AV1E_SET_QUANT_B_ADAPT, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_film_grain_test_vector(
     aom_codec_alg_priv_t *ctx, va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -1136,6 +1387,13 @@ static aom_codec_err_t ctrl_set_max_gf_interval(aom_codec_alg_priv_t *ctx,
   return update_extra_cfg(ctx, &extra_cfg);
 }
 
+static aom_codec_err_t ctrl_set_gf_max_pyr_height(aom_codec_alg_priv_t *ctx,
+                                                  va_list args) {
+  struct av1_extracfg extra_cfg = ctx->extra_cfg;
+  extra_cfg.gf_max_pyr_height = CAST(AV1E_SET_GF_MAX_PYRAMID_HEIGHT, args);
+  return update_extra_cfg(ctx, &extra_cfg);
+}
+
 static aom_codec_err_t ctrl_set_frame_periodic_boost(aom_codec_alg_priv_t *ctx,
                                                      va_list args) {
   struct av1_extracfg extra_cfg = ctx->extra_cfg;
@@ -1214,7 +1472,10 @@ static aom_codec_frame_flags_t get_frame_pkt_flags(const AV1_COMP *cpi,
   aom_codec_frame_flags_t flags = lib_flags << 16;
 
   if (lib_flags & FRAMEFLAGS_KEY) flags |= AOM_FRAME_IS_KEY;
-
+  if (lib_flags & FRAMEFLAGS_INTRAONLY) flags |= AOM_FRAME_IS_INTRAONLY;
+  if (lib_flags & FRAMEFLAGS_SWITCH) flags |= AOM_FRAME_IS_SWITCH;
+  if (lib_flags & FRAMEFLAGS_ERROR_RESILIENT)
+    flags |= AOM_FRAME_IS_ERROR_RESILIENT;
   if (cpi->droppable) flags |= AOM_FRAME_IS_DROPPABLE;
 
   return flags;
@@ -1329,18 +1590,13 @@ static aom_codec_err_t encoder_encode(aom_codec_alg_priv_t *ctx,
     unsigned int lib_flags = 0;
     int is_frame_visible = 0;
     int index_size = 0;
+    int has_fwd_keyframe = 0;
     // invisible frames get packed with the next visible frame
     while (cx_data_sz - index_size >= ctx->cx_data_sz / 2 &&
            !is_frame_visible &&
            -1 != av1_get_compressed_data(cpi, &lib_flags, &frame_size, cx_data,
                                          &dst_time_stamp, &dst_end_time_stamp,
                                          !img, timebase)) {
-      if (cpi->common.seq_params.frame_id_numbers_present_flag) {
-        if (cpi->common.invalid_delta_frame_id_minus_1) {
-          aom_internal_error(&cpi->common.error, AOM_CODEC_ERROR,
-                             "Invalid delta_frame_id_minus_1");
-        }
-      }
       cpi->seq_params_locked = 1;
       if (frame_size) {
         if (ctx->pending_cx_data == 0) ctx->pending_cx_data = cx_data;
@@ -1403,6 +1659,9 @@ static aom_codec_err_t encoder_encode(aom_codec_alg_priv_t *ctx,
         index_size = MAG_SIZE * (ctx->pending_frame_count - 1) + 2;
 
         is_frame_visible = cpi->common.show_frame;
+
+        has_fwd_keyframe |= (!is_frame_visible &&
+                             cpi->common.current_frame.frame_type == KEY_FRAME);
       }
     }
     if (is_frame_visible) {
@@ -1434,6 +1693,11 @@ static aom_codec_err_t encoder_encode(aom_codec_alg_priv_t *ctx,
 
       pkt.data.frame.pts = ticks_to_timebase_units(timebase, dst_time_stamp);
       pkt.data.frame.flags = get_frame_pkt_flags(cpi, lib_flags);
+      if (has_fwd_keyframe) {
+        // If one of the invisible frames in the packet is a keyframe, set
+        // the delayed random access point flag.
+        pkt.data.frame.flags |= AOM_FRAME_IS_DELAYED_RANDOM_ACCESS_POINT;
+      }
       pkt.data.frame.duration = (uint32_t)ticks_to_timebase_units(
           timebase, dst_end_time_stamp - dst_time_stamp);
 
@@ -1733,6 +1997,7 @@ static aom_codec_ctrl_fn_map_t encoder_ctrl_maps[] = {
   { AV1E_SET_ROW_MT, ctrl_set_row_mt },
   { AV1E_SET_TILE_COLUMNS, ctrl_set_tile_columns },
   { AV1E_SET_TILE_ROWS, ctrl_set_tile_rows },
+  { AV1E_SET_ENABLE_TPL_MODEL, ctrl_set_enable_tpl_model },
   { AOME_SET_ARNR_MAXFRAMES, ctrl_set_arnr_max_frames },
   { AOME_SET_ARNR_STRENGTH, ctrl_set_arnr_strength },
   { AOME_SET_TUNING, ctrl_set_tuning },
@@ -1744,6 +2009,7 @@ static aom_codec_ctrl_fn_map_t encoder_ctrl_maps[] = {
   { AV1E_SET_LOSSLESS, ctrl_set_lossless },
   { AV1E_SET_ENABLE_CDEF, ctrl_set_enable_cdef },
   { AV1E_SET_ENABLE_RESTORATION, ctrl_set_enable_restoration },
+  { AV1E_SET_ENABLE_OBMC, ctrl_set_enable_obmc },
   { AV1E_SET_DISABLE_TRELLIS_QUANT, ctrl_set_disable_trellis_quant },
   { AV1E_SET_ENABLE_QM, ctrl_set_enable_qm },
   { AV1E_SET_QM_Y, ctrl_set_qm_y },
@@ -1760,15 +2026,37 @@ static aom_codec_ctrl_fn_map_t encoder_ctrl_maps[] = {
   { AV1E_SET_FRAME_PARALLEL_DECODING, ctrl_set_frame_parallel_decoding_mode },
   { AV1E_SET_ERROR_RESILIENT_MODE, ctrl_set_error_resilient_mode },
   { AV1E_SET_S_FRAME_MODE, ctrl_set_s_frame_mode },
-  { AV1E_SET_ENABLE_DF, ctrl_set_enable_df },
+  { AV1E_SET_ENABLE_RECT_PARTITIONS, ctrl_set_enable_rect_partitions },
+  { AV1E_SET_ENABLE_DUAL_FILTER, ctrl_set_enable_dual_filter },
+  { AV1E_SET_ENABLE_INTRA_EDGE_FILTER, ctrl_set_enable_intra_edge_filter },
   { AV1E_SET_ENABLE_ORDER_HINT, ctrl_set_enable_order_hint },
-  { AV1E_SET_ENABLE_JNT_COMP, ctrl_set_enable_jnt_comp },
+  { AV1E_SET_ENABLE_TX64, ctrl_set_enable_tx64 },
+  { AV1E_SET_ENABLE_DIST_WTD_COMP, ctrl_set_enable_dist_wtd_comp },
+  { AV1E_SET_MAX_REFERENCE_FRAMES, ctrl_set_max_reference_frames },
   { AV1E_SET_ENABLE_REF_FRAME_MVS, ctrl_set_enable_ref_frame_mvs },
   { AV1E_SET_ALLOW_REF_FRAME_MVS, ctrl_set_allow_ref_frame_mvs },
+  { AV1E_SET_ENABLE_MASKED_COMP, ctrl_set_enable_masked_comp },
+  { AV1E_SET_ENABLE_INTERINTRA_COMP, ctrl_set_enable_interintra_comp },
+  { AV1E_SET_ENABLE_SMOOTH_INTERINTRA, ctrl_set_enable_smooth_interintra },
+  { AV1E_SET_ENABLE_DIFF_WTD_COMP, ctrl_set_enable_diff_wtd_comp },
+  { AV1E_SET_ENABLE_INTERINTER_WEDGE, ctrl_set_enable_interinter_wedge },
+  { AV1E_SET_ENABLE_INTERINTRA_WEDGE, ctrl_set_enable_interintra_wedge },
+  { AV1E_SET_ENABLE_GLOBAL_MOTION, ctrl_set_enable_global_motion },
   { AV1E_SET_ENABLE_WARPED_MOTION, ctrl_set_enable_warped_motion },
   { AV1E_SET_ALLOW_WARPED_MOTION, ctrl_set_allow_warped_motion },
+  { AV1E_SET_ENABLE_FILTER_INTRA, ctrl_set_enable_filter_intra },
+  { AV1E_SET_ENABLE_SMOOTH_INTRA, ctrl_set_enable_smooth_intra },
+  { AV1E_SET_ENABLE_PAETH_INTRA, ctrl_set_enable_paeth_intra },
+  { AV1E_SET_ENABLE_CFL_INTRA, ctrl_set_enable_cfl_intra },
   { AV1E_SET_ENABLE_SUPERRES, ctrl_set_enable_superres },
+  { AV1E_SET_ENABLE_PALETTE, ctrl_set_enable_palette },
+  { AV1E_SET_ENABLE_INTRABC, ctrl_set_enable_intrabc },
+  { AV1E_SET_ENABLE_ANGLE_DELTA, ctrl_set_enable_angle_delta },
   { AV1E_SET_AQ_MODE, ctrl_set_aq_mode },
+  { AV1E_SET_REDUCED_TX_TYPE_SET, ctrl_set_reduced_tx_type_set },
+  { AV1E_SET_INTRA_DCT_ONLY, ctrl_set_intra_dct_only },
+  { AV1E_SET_INTER_DCT_ONLY, ctrl_set_inter_dct_only },
+  { AV1E_SET_QUANT_B_ADAPT, ctrl_set_quant_b_adapt },
   { AV1E_SET_DELTAQ_MODE, ctrl_set_deltaq_mode },
   { AV1E_SET_FRAME_PERIODIC_BOOST, ctrl_set_frame_periodic_boost },
   { AV1E_SET_TUNE_CONTENT, ctrl_set_tune_content },
@@ -1781,6 +2069,7 @@ static aom_codec_ctrl_fn_map_t encoder_ctrl_maps[] = {
   { AV1E_SET_NOISE_SENSITIVITY, ctrl_set_noise_sensitivity },
   { AV1E_SET_MIN_GF_INTERVAL, ctrl_set_min_gf_interval },
   { AV1E_SET_MAX_GF_INTERVAL, ctrl_set_max_gf_interval },
+  { AV1E_SET_GF_MAX_PYRAMID_HEIGHT, ctrl_set_gf_max_pyr_height },
   { AV1E_SET_RENDER_SIZE, ctrl_set_render_size },
   { AV1E_SET_SUPERBLOCK_SIZE, ctrl_set_superblock_size },
   { AV1E_SET_SINGLE_TILE_DECODING, ctrl_set_single_tile_decoding },
@@ -1837,7 +2126,7 @@ static aom_codec_enc_cfg_map_t encoder_usage_cfg_map[] = {
         SCALE_NUMERATOR,  // rc_superres_denominator
         SCALE_NUMERATOR,  // rc_superres_kf_denominator
         63,               // rc_superres_qthresh
-        63,               // rc_superres_kf_qthresh
+        32,               // rc_superres_kf_qthresh
 
         AOM_VBR,      // rc_end_usage
         { NULL, 0 },  // rc_twopass_stats_in
diff --git a/third_party/aom/av1/av1_dx_iface.c b/third_party/aom/av1/av1_dx_iface.c
index 4a66310471..87a6e3dca0 100644
--- a/third_party/aom/av1/av1_dx_iface.c
+++ b/third_party/aom/av1/av1_dx_iface.c
@@ -44,7 +44,7 @@ struct aom_codec_alg_priv {
   int img_avail;
   int flushed;
   int invert_tile_order;
-  int last_show_frame;  // Index of last output frame.
+  RefCntBuffer *last_show_frame;  // Last output frame buffer
   int byte_alignment;
   int skip_loop_filter;
   int skip_film_grain;
@@ -58,12 +58,13 @@ struct aom_codec_alg_priv {
   int operating_point;
   int output_all_layers;
 
+  // TODO(wtc): This can be simplified. num_frame_workers is always 1, and
+  // next_output_worker_id is always 0. The frame_workers array of size 1 can
+  // be replaced by a single AVxWorker.
   AVxWorker *frame_workers;
   int num_frame_workers;
-  int next_submit_worker_id;
-  int last_submit_worker_id;
   int next_output_worker_id;
-  int available_threads;
+
   aom_image_t *image_with_grain[MAX_NUM_SPATIAL_LAYERS];
   int need_resync;  // wait for key/intra-only frame
   // BufferPool that holds all reference frames. Shared by all the FrameWorkers.
@@ -132,11 +133,6 @@ static aom_codec_err_t decoder_destroy(aom_codec_alg_priv_t *ctx) {
       av1_remove_common(&frame_worker_data->pbi->common);
       av1_free_restoration_buffers(&frame_worker_data->pbi->common);
       av1_decoder_remove(frame_worker_data->pbi);
-      aom_free(frame_worker_data->scratch_buffer);
-#if CONFIG_MULTITHREAD
-      pthread_mutex_destroy(&frame_worker_data->stats_mutex);
-      pthread_cond_destroy(&frame_worker_data->stats_cond);
-#endif
       aom_free(frame_worker_data);
     }
 #if CONFIG_MULTITHREAD
@@ -207,7 +203,7 @@ static aom_codec_err_t decoder_peek_si_internal(const uint8_t *data,
   memset(&obu_header, 0, sizeof(obu_header));
   size_t payload_size = 0;
   size_t bytes_read = 0;
-  int reduced_still_picture_hdr = 0;
+  uint8_t reduced_still_picture_hdr = 0;
   aom_codec_err_t status = aom_read_obu_header_and_size(
       data, data_sz, si->is_annexb, &obu_header, &payload_size, &bytes_read);
   if (status != AOM_CODEC_OK) return status;
@@ -236,7 +232,7 @@ static aom_codec_err_t decoder_peek_si_internal(const uint8_t *data,
       struct aom_read_bit_buffer rb = { data, data + data_sz, 0, NULL, NULL };
 
       av1_read_profile(&rb);  // profile
-      const int still_picture = aom_rb_read_bit(&rb);
+      const uint8_t still_picture = aom_rb_read_bit(&rb);
       reduced_still_picture_hdr = aom_rb_read_bit(&rb);
 
       if (!still_picture && reduced_still_picture_hdr) {
@@ -321,7 +317,7 @@ static void init_buffer_callbacks(aom_codec_alg_priv_t *ctx) {
     AV1_COMMON *const cm = &frame_worker_data->pbi->common;
     BufferPool *const pool = cm->buffer_pool;
 
-    cm->new_fb_idx = INVALID_IDX;
+    cm->cur_frame = NULL;
     cm->byte_alignment = ctx->byte_alignment;
     cm->skip_loop_filter = ctx->skip_loop_filter;
     cm->skip_film_grain = ctx->skip_film_grain;
@@ -360,7 +356,6 @@ static int frame_worker_hook(void *arg1, void *arg2) {
 
   if (result != 0) {
     // Check decode result in serial decode.
-    frame_worker_data->pbi->cur_buf->buf.corrupted = 1;
     frame_worker_data->pbi->need_resync = 1;
   }
   return !result;
@@ -370,15 +365,12 @@ static aom_codec_err_t init_decoder(aom_codec_alg_priv_t *ctx) {
   int i;
   const AVxWorkerInterface *const winterface = aom_get_worker_interface();
 
-  ctx->last_show_frame = -1;
-  ctx->next_submit_worker_id = 0;
-  ctx->last_submit_worker_id = 0;
+  ctx->last_show_frame = NULL;
   ctx->next_output_worker_id = 0;
   ctx->need_resync = 1;
   ctx->num_frame_workers = 1;
   if (ctx->num_frame_workers > MAX_DECODE_THREADS)
     ctx->num_frame_workers = MAX_DECODE_THREADS;
-  ctx->available_threads = ctx->num_frame_workers;
   ctx->flushed = 0;
 
   ctx->buffer_pool = (BufferPool *)aom_calloc(1, sizeof(BufferPool));
@@ -402,6 +394,7 @@ static aom_codec_err_t init_decoder(aom_codec_alg_priv_t *ctx) {
     AVxWorker *const worker = &ctx->frame_workers[i];
     FrameWorkerData *frame_worker_data = NULL;
     winterface->init(worker);
+    worker->thread_name = "aom frameworker";
     worker->data1 = aom_memalign(32, sizeof(FrameWorkerData));
     if (worker->data1 == NULL) {
       set_error_detail(ctx, "Failed to allocate frame_worker_data");
@@ -414,23 +407,9 @@ static aom_codec_err_t init_decoder(aom_codec_alg_priv_t *ctx) {
       return AOM_CODEC_MEM_ERROR;
     }
     frame_worker_data->pbi->common.options = &ctx->cfg.cfg;
-    frame_worker_data->pbi->frame_worker_owner = worker;
     frame_worker_data->worker_id = i;
-    frame_worker_data->scratch_buffer = NULL;
-    frame_worker_data->scratch_buffer_size = 0;
     frame_worker_data->frame_context_ready = 0;
     frame_worker_data->received_frame = 0;
-#if CONFIG_MULTITHREAD
-    if (pthread_mutex_init(&frame_worker_data->stats_mutex, NULL)) {
-      set_error_detail(ctx, "Failed to allocate frame_worker_data mutex");
-      return AOM_CODEC_MEM_ERROR;
-    }
-
-    if (pthread_cond_init(&frame_worker_data->stats_cond, NULL)) {
-      set_error_detail(ctx, "Failed to allocate frame_worker_data cond");
-      return AOM_CODEC_MEM_ERROR;
-    }
-#endif
     frame_worker_data->pbi->allow_lowbitdepth = ctx->cfg.allow_lowbitdepth;
 
     // If decoding in serial mode, FrameWorker thread could create tile worker
@@ -447,7 +426,8 @@ static aom_codec_err_t init_decoder(aom_codec_alg_priv_t *ctx) {
     frame_worker_data->pbi->row_mt = ctx->row_mt;
 
     worker->hook = frame_worker_hook;
-    if (!winterface->reset(worker)) {
+    // The main thread acts as Frame Worker 0.
+    if (i != 0 && !winterface->reset(worker)) {
       set_error_detail(ctx, "Frame Worker thread creation failed");
       return AOM_CODEC_MEM_ERROR;
     }
@@ -467,7 +447,7 @@ static INLINE void check_resync(aom_codec_alg_priv_t *const ctx,
                                 const AV1Decoder *const pbi) {
   // Clear resync flag if worker got a key frame or intra only frame.
   if (ctx->need_resync == 1 && pbi->need_resync == 0 &&
-      (pbi->common.intra_only || pbi->common.frame_type == KEY_FRAME))
+      frame_is_intra_only(&pbi->common))
     ctx->need_resync = 0;
 }
 
@@ -496,11 +476,6 @@ static aom_codec_err_t decode_one(aom_codec_alg_priv_t *ctx,
   frame_worker_data->user_priv = user_priv;
   frame_worker_data->received_frame = 1;
 
-#if CONFIG_INSPECTION
-  frame_worker_data->pbi->inspect_cb = ctx->inspect_cb;
-  frame_worker_data->pbi->inspect_ctx = ctx->inspect_ctx;
-#endif
-
   frame_worker_data->pbi->common.large_scale_tile = ctx->tile_mode;
   frame_worker_data->pbi->dec_tile_row = ctx->decode_tile_row;
   frame_worker_data->pbi->dec_tile_col = ctx->decode_tile_col;
@@ -524,17 +499,55 @@ static aom_codec_err_t decode_one(aom_codec_alg_priv_t *ctx,
   return AOM_CODEC_OK;
 }
 
+#if CONFIG_INSPECTION
+// This function enables the inspector to inspect non visible frames.
+static aom_codec_err_t decoder_inspect(aom_codec_alg_priv_t *ctx,
+                                       const uint8_t *data, size_t data_sz,
+                                       void *user_priv) {
+  aom_codec_err_t res = AOM_CODEC_OK;
+
+  Av1DecodeReturn *data2 = (Av1DecodeReturn *)user_priv;
+
+  if (ctx->frame_workers == NULL) {
+    res = init_decoder(ctx);
+    if (res != AOM_CODEC_OK) return res;
+  }
+  FrameWorkerData *const frame_worker_data =
+      (FrameWorkerData *)ctx->frame_workers[0].data1;
+  AV1Decoder *const pbi = frame_worker_data->pbi;
+  AV1_COMMON *const cm = &pbi->common;
+  frame_worker_data->pbi->inspect_cb = ctx->inspect_cb;
+  frame_worker_data->pbi->inspect_ctx = ctx->inspect_ctx;
+  res = av1_receive_compressed_data(frame_worker_data->pbi, data_sz, &data);
+  check_resync(ctx, frame_worker_data->pbi);
+
+  if (ctx->frame_workers->had_error)
+    return update_error_state(ctx, &frame_worker_data->pbi->common.error);
+
+  data2->idx = -1;
+  for (int i = 0; i < REF_FRAMES; ++i)
+    if (cm->ref_frame_map[i] == cm->cur_frame) data2->idx = i;
+  data2->buf = data;
+  data2->show_existing = cm->show_existing_frame;
+  return res;
+}
+#endif
+
 static aom_codec_err_t decoder_decode(aom_codec_alg_priv_t *ctx,
                                       const uint8_t *data, size_t data_sz,
                                       void *user_priv) {
   aom_codec_err_t res = AOM_CODEC_OK;
 
+#if CONFIG_INSPECTION
+  if (user_priv != 0) {
+    return decoder_inspect(ctx, data, data_sz, user_priv);
+  }
+#endif
   // Release any pending output frames from the previous decoder_decode call.
   // We need to do this even if the decoder is being flushed or the input
   // arguments are invalid.
   if (ctx->frame_workers) {
     BufferPool *const pool = ctx->buffer_pool;
-    RefCntBuffer *const frame_bufs = pool->frame_bufs;
     lock_buffer_pool(pool);
     for (int i = 0; i < ctx->num_frame_workers; ++i) {
       AVxWorker *const worker = &ctx->frame_workers[i];
@@ -542,7 +555,7 @@ static aom_codec_err_t decoder_decode(aom_codec_alg_priv_t *ctx,
           (FrameWorkerData *)worker->data1;
       struct AV1Decoder *pbi = frame_worker_data->pbi;
       for (size_t j = 0; j < pbi->num_output_frames; j++) {
-        decrease_ref_count((int)pbi->output_frame_index[j], frame_bufs, pool);
+        decrease_ref_count(pbi->output_frames[j], pool);
       }
       pbi->num_output_frames = 0;
     }
@@ -644,6 +657,7 @@ static aom_image_t *add_grain_if_needed(aom_image_t *img,
 
   if (grain_img_buf) {
     grain_img_buf->user_priv = img->user_priv;
+    grain_img_buf->fb_priv = img->fb_priv;
     if (av1_add_film_grain(grain_params, img, grain_img_buf)) {
       aom_img_free(grain_img_buf);
       grain_img_buf = NULL;
@@ -676,39 +690,40 @@ static aom_image_t *decoder_get_frame(aom_codec_alg_priv_t *ctx,
       AVxWorker *const worker = &ctx->frame_workers[ctx->next_output_worker_id];
       FrameWorkerData *const frame_worker_data =
           (FrameWorkerData *)worker->data1;
+      AV1Decoder *const pbi = frame_worker_data->pbi;
+      AV1_COMMON *const cm = &pbi->common;
       ctx->next_output_worker_id =
           (ctx->next_output_worker_id + 1) % ctx->num_frame_workers;
       // Wait for the frame from worker thread.
       if (winterface->sync(worker)) {
         // Check if worker has received any frames.
         if (frame_worker_data->received_frame == 1) {
-          ++ctx->available_threads;
           frame_worker_data->received_frame = 0;
           check_resync(ctx, frame_worker_data->pbi);
         }
         aom_film_grain_t *grain_params;
         if (av1_get_raw_frame(frame_worker_data->pbi, *index, &sd,
                               &grain_params) == 0) {
-          AV1Decoder *const pbi = frame_worker_data->pbi;
-          AV1_COMMON *const cm = &pbi->common;
-          RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
-          ctx->last_show_frame = cm->new_fb_idx;
+          RefCntBuffer *const output_frame_buf = pbi->output_frames[*index];
+          ctx->last_show_frame = output_frame_buf;
           if (ctx->need_resync) return NULL;
           yuvconfig2image(&ctx->img, sd, frame_worker_data->user_priv);
 
           if (!pbi->ext_tile_debug && cm->large_scale_tile) {
             *index += 1;  // Advance the iterator to point to the next image
+
+            yuvconfig2image(&ctx->img, &pbi->tile_list_outbuf, NULL);
             img = &ctx->img;
-            img->img_data = pbi->tile_list_output;
-            img->sz = pbi->tile_list_size;
             return img;
           }
 
           const int num_planes = av1_num_planes(cm);
           if (pbi->ext_tile_debug && cm->single_tile_decoding &&
               pbi->dec_tile_row >= 0) {
+            int tile_width, tile_height;
+            av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
             const int tile_row = AOMMIN(pbi->dec_tile_row, cm->tile_rows - 1);
-            const int mi_row = tile_row * cm->tile_height;
+            const int mi_row = tile_row * tile_height;
             const int ssy = ctx->img.y_chroma_shift;
             int plane;
             ctx->img.planes[0] += mi_row * MI_SIZE * ctx->img.stride[0];
@@ -718,14 +733,15 @@ static aom_image_t *decoder_get_frame(aom_codec_alg_priv_t *ctx,
                     mi_row * (MI_SIZE >> ssy) * ctx->img.stride[plane];
               }
             }
-            ctx->img.d_h =
-                AOMMIN(cm->tile_height, cm->mi_rows - mi_row) * MI_SIZE;
+            ctx->img.d_h = AOMMIN(tile_height, cm->mi_rows - mi_row) * MI_SIZE;
           }
 
           if (pbi->ext_tile_debug && cm->single_tile_decoding &&
               pbi->dec_tile_col >= 0) {
+            int tile_width, tile_height;
+            av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
             const int tile_col = AOMMIN(pbi->dec_tile_col, cm->tile_cols - 1);
-            const int mi_col = tile_col * cm->tile_width;
+            const int mi_col = tile_col * tile_width;
             const int ssx = ctx->img.x_chroma_shift;
             const int is_hbd =
                 (ctx->img.fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 1 : 0;
@@ -737,11 +753,10 @@ static aom_image_t *decoder_get_frame(aom_codec_alg_priv_t *ctx,
                     mi_col * (MI_SIZE >> ssx) * (1 + is_hbd);
               }
             }
-            ctx->img.d_w =
-                AOMMIN(cm->tile_width, cm->mi_cols - mi_col) * MI_SIZE;
+            ctx->img.d_w = AOMMIN(tile_width, cm->mi_cols - mi_col) * MI_SIZE;
           }
 
-          ctx->img.fb_priv = frame_bufs[cm->new_fb_idx].raw_frame_buffer.priv;
+          ctx->img.fb_priv = output_frame_buf->raw_frame_buffer.priv;
           img = &ctx->img;
           img->temporal_id = cm->temporal_layer_id;
           img->spatial_id = cm->spatial_layer_id;
@@ -758,11 +773,10 @@ static aom_image_t *decoder_get_frame(aom_codec_alg_priv_t *ctx,
       } else {
         // Decoding failed. Release the worker thread.
         frame_worker_data->received_frame = 0;
-        ++ctx->available_threads;
         ctx->need_resync = 1;
         if (ctx->flushed != 1) return NULL;
       }
-    } while (ctx->next_output_worker_id != ctx->next_submit_worker_id);
+    } while (ctx->next_output_worker_id != 0);
   }
   return NULL;
 }
@@ -894,7 +908,8 @@ static aom_codec_err_t ctrl_get_last_ref_updates(aom_codec_alg_priv_t *ctx,
       AVxWorker *const worker = ctx->frame_workers;
       FrameWorkerData *const frame_worker_data =
           (FrameWorkerData *)worker->data1;
-      *update_info = frame_worker_data->pbi->refresh_frame_flags;
+      *update_info =
+          frame_worker_data->pbi->common.current_frame.refresh_frame_flags;
       return AOM_CODEC_OK;
     } else {
       return AOM_CODEC_ERROR;
@@ -923,11 +938,10 @@ static aom_codec_err_t ctrl_get_frame_corrupted(aom_codec_alg_priv_t *ctx,
       FrameWorkerData *const frame_worker_data =
           (FrameWorkerData *)worker->data1;
       AV1Decoder *const pbi = frame_worker_data->pbi;
-      RefCntBuffer *const frame_bufs = pbi->common.buffer_pool->frame_bufs;
       if (pbi->seen_frame_header && pbi->num_output_frames == 0)
         return AOM_CODEC_ERROR;
-      if (ctx->last_show_frame >= 0)
-        *corrupted = frame_bufs[ctx->last_show_frame].buf.corrupted;
+      if (ctx->last_show_frame != NULL)
+        *corrupted = ctx->last_show_frame->buf.corrupted;
       return AOM_CODEC_OK;
     } else {
       return AOM_CODEC_ERROR;
@@ -1107,8 +1121,27 @@ static aom_codec_err_t ctrl_get_tile_size(aom_codec_alg_priv_t *ctx,
       FrameWorkerData *const frame_worker_data =
           (FrameWorkerData *)worker->data1;
       const AV1_COMMON *const cm = &frame_worker_data->pbi->common;
-      *tile_size =
-          ((cm->tile_width * MI_SIZE) << 16) + cm->tile_height * MI_SIZE;
+      int tile_width, tile_height;
+      av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
+      *tile_size = ((tile_width * MI_SIZE) << 16) + tile_height * MI_SIZE;
+      return AOM_CODEC_OK;
+    } else {
+      return AOM_CODEC_ERROR;
+    }
+  }
+  return AOM_CODEC_INVALID_PARAM;
+}
+
+static aom_codec_err_t ctrl_get_tile_count(aom_codec_alg_priv_t *ctx,
+                                           va_list args) {
+  unsigned int *const tile_count = va_arg(args, unsigned int *);
+
+  if (tile_count) {
+    AVxWorker *const worker = &ctx->frame_workers[ctx->next_output_worker_id];
+    if (worker) {
+      FrameWorkerData *const frame_worker_data =
+          (FrameWorkerData *)worker->data1;
+      *tile_count = frame_worker_data->pbi->tile_count_minus_1 + 1;
       return AOM_CODEC_OK;
     } else {
       return AOM_CODEC_ERROR;
@@ -1283,6 +1316,7 @@ static aom_codec_ctrl_fn_map_t decoder_ctrl_maps[] = {
   { AV1D_GET_BIT_DEPTH, ctrl_get_bit_depth },
   { AV1D_GET_IMG_FORMAT, ctrl_get_img_format },
   { AV1D_GET_TILE_SIZE, ctrl_get_tile_size },
+  { AV1D_GET_TILE_COUNT, ctrl_get_tile_count },
   { AV1D_GET_DISPLAY_SIZE, ctrl_get_render_size },
   { AV1D_GET_FRAME_SIZE, ctrl_get_frame_size },
   { AV1_GET_ACCOUNTING, ctrl_get_accounting },
diff --git a/third_party/aom/av1/av1_iface_common.h b/third_party/aom/av1/av1_iface_common.h
index 4a7af580b8..5568c89611 100644
--- a/third_party/aom/av1/av1_iface_common.h
+++ b/third_party/aom/av1/av1_iface_common.h
@@ -52,11 +52,9 @@ static void yuvconfig2image(aom_image_t *img, const YV12_BUFFER_CONFIG *yv12,
   img->planes[AOM_PLANE_Y] = yv12->y_buffer;
   img->planes[AOM_PLANE_U] = yv12->u_buffer;
   img->planes[AOM_PLANE_V] = yv12->v_buffer;
-  img->planes[AOM_PLANE_ALPHA] = NULL;
   img->stride[AOM_PLANE_Y] = yv12->y_stride;
   img->stride[AOM_PLANE_U] = yv12->uv_stride;
   img->stride[AOM_PLANE_V] = yv12->uv_stride;
-  img->stride[AOM_PLANE_ALPHA] = yv12->y_stride;
   if (yv12->flags & YV12_FLAG_HIGHBITDEPTH) {
     // aom_image_t uses byte strides and a pointer to the first byte
     // of the image.
@@ -65,17 +63,16 @@ static void yuvconfig2image(aom_image_t *img, const YV12_BUFFER_CONFIG *yv12,
     img->planes[AOM_PLANE_Y] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->y_buffer);
     img->planes[AOM_PLANE_U] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->u_buffer);
     img->planes[AOM_PLANE_V] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->v_buffer);
-    img->planes[AOM_PLANE_ALPHA] = NULL;
     img->stride[AOM_PLANE_Y] = 2 * yv12->y_stride;
     img->stride[AOM_PLANE_U] = 2 * yv12->uv_stride;
     img->stride[AOM_PLANE_V] = 2 * yv12->uv_stride;
-    img->stride[AOM_PLANE_ALPHA] = 2 * yv12->y_stride;
   }
   img->bps = bps;
   img->user_priv = user_priv;
   img->img_data = yv12->buffer_alloc;
   img->img_data_owner = 0;
   img->self_allocd = 0;
+  img->sz = yv12->frame_size;
 }
 
 static aom_codec_err_t image2yuvconfig(const aom_image_t *img,
@@ -127,7 +124,12 @@ static aom_codec_err_t image2yuvconfig(const aom_image_t *img,
   } else {
     yv12->flags = 0;
   }
-  yv12->border = (yv12->y_stride - img->w) / 2;
+
+  // Note(yunqing): if img is allocated the same as the frame buffer, y_stride
+  // is 32-byte aligned. Also, handle the cases while allocating img without a
+  // border or stride_align is less than 32.
+  int border = (yv12->y_stride - (int)((img->w + 31) & ~31)) / 2;
+  yv12->border = (border < 0) ? 0 : border;
   yv12->subsampling_x = img->x_chroma_shift;
   yv12->subsampling_y = img->y_chroma_shift;
   return AOM_CODEC_OK;
diff --git a/third_party/aom/av1/common/alloccommon.c b/third_party/aom/av1/common/alloccommon.c
index 1bf81c91d4..1c8528a468 100644
--- a/third_party/aom/av1/common/alloccommon.c
+++ b/third_party/aom/av1/common/alloccommon.c
@@ -92,6 +92,9 @@ void av1_free_ref_frame_buffers(BufferPool *pool) {
     if (pool->frame_bufs[i].ref_count > 0 &&
         pool->frame_bufs[i].raw_frame_buffer.data != NULL) {
       pool->release_fb_cb(pool->cb_priv, &pool->frame_bufs[i].raw_frame_buffer);
+      pool->frame_bufs[i].raw_frame_buffer.data = NULL;
+      pool->frame_bufs[i].raw_frame_buffer.size = 0;
+      pool->frame_bufs[i].raw_frame_buffer.priv = NULL;
       pool->frame_bufs[i].ref_count = 0;
     }
     aom_free(pool->frame_bufs[i].mvs);
@@ -131,13 +134,12 @@ void av1_alloc_restoration_buffers(AV1_COMMON *cm) {
     const int ext_h = RESTORATION_UNIT_OFFSET + (mi_h << MI_SIZE_LOG2);
     const int tile_stripes = (ext_h + 63) / 64;
     num_stripes += tile_stripes;
-    cm->rst_end_stripe[i] = num_stripes;
   }
 
   // Now we need to allocate enough space to store the line buffers for the
   // stripes
   const int frame_w = cm->superres_upscaled_width;
-  const int use_highbd = cm->seq_params.use_highbitdepth ? 1 : 0;
+  const int use_highbd = cm->seq_params.use_highbitdepth;
 
   for (int p = 0; p < num_planes; ++p) {
     const int is_uv = p > 0;
@@ -293,8 +295,8 @@ void av1_remove_common(AV1_COMMON *cm) {
 
   aom_free(cm->fc);
   cm->fc = NULL;
-  aom_free(cm->frame_contexts);
-  cm->frame_contexts = NULL;
+  aom_free(cm->default_frame_context);
+  cm->default_frame_context = NULL;
 }
 
 void av1_init_context_buffers(AV1_COMMON *cm) { cm->setup_mi(cm); }
diff --git a/third_party/aom/av1/common/arm/av1_inv_txfm_neon.c b/third_party/aom/av1/common/arm/av1_inv_txfm_neon.c
index bad411743d..7a23174be2 100644
--- a/third_party/aom/av1/common/arm/av1_inv_txfm_neon.c
+++ b/third_party/aom/av1/common/arm/av1_inv_txfm_neon.c
@@ -2086,33 +2086,1570 @@ static INLINE void idct32_low16_new_neon(int16x8_t *in, int16x8_t *out,
   out[30] = vqsubq_s16(step2[1], step2[30]);
   out[31] = vqsubq_s16(step2[0], step2[31]);
 }
+static INLINE void idct64_stage9_neon(int16x8_t *step2, int16x8_t *step1,
+                                      int8_t cos_bit) {
+  const int32_t *cospi = cospi_arr(cos_bit);
+  const int16x4_t c3 =
+      create_s16x4_neon((int16_t *)(cospi + 32), (int16_t *)(cospi + 32),
+                        (int16_t *)(cospi + 16), (int16_t *)(cospi + 48));
+
+  btf_16_lane_0_1_neon(step2[27], step2[20], c3, &step1[27], &step1[20]);
+  btf_16_lane_0_1_neon(step2[26], step2[21], c3, &step1[26], &step1[21]);
+  btf_16_lane_0_1_neon(step2[25], step2[22], c3, &step1[25], &step1[22]);
+  btf_16_lane_0_1_neon(step2[24], step2[23], c3, &step1[24], &step1[23]);
+
+  step1[0] = vqaddq_s16(step2[0], step2[15]);
+  step1[1] = vqaddq_s16(step2[1], step2[14]);
+  step1[2] = vqaddq_s16(step2[2], step2[13]);
+  step1[3] = vqaddq_s16(step2[3], step2[12]);
+  step1[4] = vqaddq_s16(step2[4], step2[11]);
+  step1[5] = vqaddq_s16(step2[5], step2[10]);
+  step1[6] = vqaddq_s16(step2[6], step2[9]);
+  step1[7] = vqaddq_s16(step2[7], step2[8]);
+  step1[8] = vqsubq_s16(step2[7], step2[8]);
+  step1[9] = vqsubq_s16(step2[6], step2[9]);
+  step1[10] = vqsubq_s16(step2[5], step2[10]);
+  step1[11] = vqsubq_s16(step2[4], step2[11]);
+  step1[12] = vqsubq_s16(step2[3], step2[12]);
+  step1[13] = vqsubq_s16(step2[2], step2[13]);
+  step1[14] = vqsubq_s16(step2[1], step2[14]);
+  step1[15] = vqsubq_s16(step2[0], step2[15]);
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[18] = step2[18];
+  step1[19] = step2[19];
+  step1[28] = step2[28];
+  step1[29] = step2[29];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[47]);
+  step1[33] = vqaddq_s16(step2[33], step2[46]);
+  step1[34] = vqaddq_s16(step2[34], step2[45]);
+  step1[35] = vqaddq_s16(step2[35], step2[44]);
+  step1[36] = vqaddq_s16(step2[36], step2[43]);
+  step1[37] = vqaddq_s16(step2[37], step2[42]);
+  step1[38] = vqaddq_s16(step2[38], step2[41]);
+  step1[39] = vqaddq_s16(step2[39], step2[40]);
+  step1[40] = vqsubq_s16(step2[39], step2[40]);
+  step1[41] = vqsubq_s16(step2[38], step2[41]);
+  step1[42] = vqsubq_s16(step2[37], step2[42]);
+  step1[43] = vqsubq_s16(step2[36], step2[43]);
+  step1[44] = vqsubq_s16(step2[35], step2[44]);
+  step1[45] = vqsubq_s16(step2[34], step2[45]);
+  step1[46] = vqsubq_s16(step2[33], step2[46]);
+  step1[47] = vqsubq_s16(step2[32], step2[47]);
+  step1[48] = vqsubq_s16(step2[63], step2[48]);
+  step1[49] = vqsubq_s16(step2[62], step2[49]);
+  step1[50] = vqsubq_s16(step2[61], step2[50]);
+  step1[51] = vqsubq_s16(step2[60], step2[51]);
+  step1[52] = vqsubq_s16(step2[59], step2[52]);
+  step1[53] = vqsubq_s16(step2[58], step2[53]);
+  step1[54] = vqsubq_s16(step2[57], step2[54]);
+  step1[55] = vqsubq_s16(step2[56], step2[55]);
+  step1[56] = vqaddq_s16(step2[56], step2[55]);
+  step1[57] = vqaddq_s16(step2[57], step2[54]);
+  step1[58] = vqaddq_s16(step2[58], step2[53]);
+  step1[59] = vqaddq_s16(step2[59], step2[52]);
+  step1[60] = vqaddq_s16(step2[60], step2[51]);
+  step1[61] = vqaddq_s16(step2[61], step2[50]);
+  step1[62] = vqaddq_s16(step2[62], step2[49]);
+  step1[63] = vqaddq_s16(step2[63], step2[48]);
+}
+
+static INLINE void idct64_stage10_neon(int16x8_t *step1, int16x8_t *step2,
+                                       int8_t cos_bit) {
+  const int32_t *cospi = cospi_arr(cos_bit);
+  const int16x4_t c3 =
+      create_s16x4_neon((int16_t *)(cospi + 32), (int16_t *)(cospi + 32),
+                        (int16_t *)(cospi + 16), (int16_t *)(cospi + 48));
+
+  btf_16_lane_0_1_neon(step1[55], step1[40], c3, &step2[55], &step2[40]);
+  btf_16_lane_0_1_neon(step1[54], step1[41], c3, &step2[54], &step2[41]);
+  btf_16_lane_0_1_neon(step1[53], step1[42], c3, &step2[53], &step2[42]);
+  btf_16_lane_0_1_neon(step1[52], step1[43], c3, &step2[52], &step2[43]);
+  btf_16_lane_0_1_neon(step1[51], step1[44], c3, &step2[51], &step2[44]);
+  btf_16_lane_0_1_neon(step1[50], step1[45], c3, &step2[50], &step2[45]);
+  btf_16_lane_0_1_neon(step1[49], step1[46], c3, &step2[49], &step2[46]);
+  btf_16_lane_0_1_neon(step1[48], step1[47], c3, &step2[48], &step2[47]);
+
+  step2[0] = vqaddq_s16(step1[0], step1[31]);
+  step2[1] = vqaddq_s16(step1[1], step1[30]);
+  step2[2] = vqaddq_s16(step1[2], step1[29]);
+  step2[3] = vqaddq_s16(step1[3], step1[28]);
+  step2[4] = vqaddq_s16(step1[4], step1[27]);
+  step2[5] = vqaddq_s16(step1[5], step1[26]);
+  step2[6] = vqaddq_s16(step1[6], step1[25]);
+  step2[7] = vqaddq_s16(step1[7], step1[24]);
+  step2[8] = vqaddq_s16(step1[8], step1[23]);
+  step2[9] = vqaddq_s16(step1[9], step1[22]);
+  step2[10] = vqaddq_s16(step1[10], step1[21]);
+  step2[11] = vqaddq_s16(step1[11], step1[20]);
+  step2[12] = vqaddq_s16(step1[12], step1[19]);
+  step2[13] = vqaddq_s16(step1[13], step1[18]);
+  step2[14] = vqaddq_s16(step1[14], step1[17]);
+  step2[15] = vqaddq_s16(step1[15], step1[16]);
+  step2[16] = vqsubq_s16(step1[15], step1[16]);
+  step2[17] = vqsubq_s16(step1[14], step1[17]);
+  step2[18] = vqsubq_s16(step1[13], step1[18]);
+  step2[19] = vqsubq_s16(step1[12], step1[19]);
+  step2[20] = vqsubq_s16(step1[11], step1[20]);
+  step2[21] = vqsubq_s16(step1[10], step1[21]);
+  step2[22] = vqsubq_s16(step1[9], step1[22]);
+  step2[23] = vqsubq_s16(step1[8], step1[23]);
+  step2[24] = vqsubq_s16(step1[7], step1[24]);
+  step2[25] = vqsubq_s16(step1[6], step1[25]);
+  step2[26] = vqsubq_s16(step1[5], step1[26]);
+  step2[27] = vqsubq_s16(step1[4], step1[27]);
+  step2[28] = vqsubq_s16(step1[3], step1[28]);
+  step2[29] = vqsubq_s16(step1[2], step1[29]);
+  step2[30] = vqsubq_s16(step1[1], step1[30]);
+  step2[31] = vqsubq_s16(step1[0], step1[31]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[34] = step1[34];
+  step2[35] = step1[35];
+  step2[36] = step1[36];
+  step2[37] = step1[37];
+  step2[38] = step1[38];
+  step2[39] = step1[39];
+  step2[56] = step1[56];
+  step2[57] = step1[57];
+  step2[58] = step1[58];
+  step2[59] = step1[59];
+  step2[60] = step1[60];
+  step2[61] = step1[61];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+}
+
+static INLINE void idct64_low32_new_neon(int16x8_t *in, int16x8_t *out,
+                                         int8_t cos_bit, int bit) {
+  (void)bit;
+  const int32_t *cospi = cospi_arr(cos_bit);
+  int16x8_t step2[64], step1[64];
+  const int16x4_t c0 =
+      create_s16x4_neon((int16_t *)(cospi + 4), (int16_t *)(cospi + 60),
+                        (int16_t *)(cospi + 36), (int16_t *)(cospi + 28));
+  const int16x4_t c1 =
+      create_s16x4_neon((int16_t *)(cospi + 20), (int16_t *)(cospi + 44),
+                        (int16_t *)(cospi + 52), (int16_t *)(cospi + 12));
+  const int16x4_t c2 =
+      create_s16x4_neon((int16_t *)(cospi + 8), (int16_t *)(cospi + 56),
+                        (int16_t *)(cospi + 40), (int16_t *)(cospi + 24));
+  const int16x4_t c3 =
+      create_s16x4_neon((int16_t *)(cospi + 32), (int16_t *)(cospi + 32),
+                        (int16_t *)(cospi + 16), (int16_t *)(cospi + 48));
+
+  // stage 1
+  // stage 2
+
+  step2[0] = in[0];
+  step2[2] = in[16];
+  step2[4] = in[8];
+  step2[6] = in[24];
+  step2[8] = in[4];
+  step2[10] = in[20];
+  step2[12] = in[12];
+  step2[14] = in[28];
+  step2[16] = in[2];
+  step2[18] = in[18];
+  step2[20] = in[10];
+  step2[22] = in[26];
+  step2[24] = in[6];
+  step2[26] = in[22];
+  step2[28] = in[14];
+  step2[30] = in[30];
+
+  btf_16_neon(in[1], cospi[63], cospi[1], &step2[32], &step2[63]);
+  btf_16_neon(in[31], -cospi[33], cospi[31], &step2[33], &step2[62]);
+  btf_16_neon(in[17], cospi[47], cospi[17], &step2[34], &step2[61]);
+  btf_16_neon(in[15], -cospi[49], cospi[15], &step2[35], &step2[60]);
+  btf_16_neon(in[9], cospi[55], cospi[9], &step2[36], &step2[59]);
+  btf_16_neon(in[23], -cospi[41], cospi[23], &step2[37], &step2[58]);
+  btf_16_neon(in[25], cospi[39], cospi[25], &step2[38], &step2[57]);
+  btf_16_neon(in[7], -cospi[57], cospi[7], &step2[39], &step2[56]);
+  btf_16_neon(in[5], cospi[59], cospi[5], &step2[40], &step2[55]);
+  btf_16_neon(in[27], -cospi[37], cospi[27], &step2[41], &step2[54]);
+  btf_16_neon(in[21], cospi[43], cospi[21], &step2[42], &step2[53]);
+  btf_16_neon(in[11], -cospi[53], cospi[11], &step2[43], &step2[52]);
+  btf_16_neon(in[13], cospi[51], cospi[13], &step2[44], &step2[51]);
+  btf_16_neon(in[19], -cospi[45], cospi[19], &step2[45], &step2[50]);
+  btf_16_neon(in[29], cospi[35], cospi[29], &step2[46], &step2[49]);
+  btf_16_neon(in[3], -cospi[61], cospi[3], &step2[47], &step2[48]);
+
+  // stage 3
+
+  step1[0] = step2[0];
+  step1[2] = step2[2];
+  step1[4] = step2[4];
+  step1[6] = step2[6];
+  step1[8] = step2[8];
+  step1[10] = step2[10];
+  step1[12] = step2[12];
+  step1[14] = step2[14];
+
+  btf_16_neon(step2[16], cospi[62], cospi[2], &step1[16], &step1[31]);
+  btf_16_neon(step2[30], -cospi[34], cospi[30], &step1[17], &step1[30]);
+  btf_16_neon(step2[18], cospi[46], cospi[18], &step1[18], &step1[29]);
+  btf_16_neon(step2[28], -cospi[50], cospi[14], &step1[19], &step1[28]);
+  btf_16_neon(step2[20], cospi[54], cospi[10], &step1[20], &step1[27]);
+  btf_16_neon(step2[26], -cospi[42], cospi[22], &step1[21], &step1[26]);
+  btf_16_neon(step2[22], cospi[38], cospi[26], &step1[22], &step1[25]);
+  btf_16_neon(step2[24], -cospi[58], cospi[6], &step1[23], &step1[24]);
+
+  step1[32] = vqaddq_s16(step2[32], step2[33]);
+  step1[33] = vqsubq_s16(step2[32], step2[33]);
+  step1[34] = vqsubq_s16(step2[35], step2[34]);
+  step1[35] = vqaddq_s16(step2[35], step2[34]);
+  step1[36] = vqaddq_s16(step2[36], step2[37]);
+  step1[37] = vqsubq_s16(step2[36], step2[37]);
+  step1[38] = vqsubq_s16(step2[39], step2[38]);
+  step1[39] = vqaddq_s16(step2[39], step2[38]);
+  step1[40] = vqaddq_s16(step2[40], step2[41]);
+  step1[41] = vqsubq_s16(step2[40], step2[41]);
+  step1[42] = vqsubq_s16(step2[43], step2[42]);
+  step1[43] = vqaddq_s16(step2[43], step2[42]);
+  step1[44] = vqaddq_s16(step2[44], step2[45]);
+  step1[45] = vqsubq_s16(step2[44], step2[45]);
+  step1[46] = vqsubq_s16(step2[47], step2[46]);
+  step1[47] = vqaddq_s16(step2[47], step2[46]);
+  step1[48] = vqaddq_s16(step2[48], step2[49]);
+  step1[49] = vqsubq_s16(step2[48], step2[49]);
+  step1[50] = vqsubq_s16(step2[51], step2[50]);
+  step1[51] = vqaddq_s16(step2[51], step2[50]);
+  step1[52] = vqaddq_s16(step2[52], step2[53]);
+  step1[53] = vqsubq_s16(step2[52], step2[53]);
+  step1[54] = vqsubq_s16(step2[55], step2[54]);
+  step1[55] = vqaddq_s16(step2[55], step2[54]);
+  step1[56] = vqaddq_s16(step2[56], step2[57]);
+  step1[57] = vqsubq_s16(step2[56], step2[57]);
+  step1[58] = vqsubq_s16(step2[59], step2[58]);
+  step1[59] = vqaddq_s16(step2[59], step2[58]);
+  step1[60] = vqaddq_s16(step2[60], step2[61]);
+  step1[61] = vqsubq_s16(step2[60], step2[61]);
+  step1[62] = vqsubq_s16(step2[63], step2[62]);
+  step1[63] = vqaddq_s16(step2[63], step2[62]);
+
+  // stage 4
+
+  step2[0] = step1[0];
+  step2[2] = step1[2];
+  step2[4] = step1[4];
+  step2[6] = step1[6];
+
+  btf_16_neon(step1[8], cospi[60], cospi[4], &step2[8], &step2[15]);
+  btf_16_neon(step1[14], -cospi[36], cospi[28], &step2[9], &step2[14]);
+  btf_16_neon(step1[10], cospi[44], cospi[20], &step2[10], &step2[13]);
+  btf_16_neon(step1[12], -cospi[52], cospi[12], &step2[11], &step2[12]);
+  btf_16_lane_0_1_neon(step1[62], step1[33], c0, &step2[62], &step2[33]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[34]), vnegq_s16(step1[61]), c0,
+                       &step2[34], &step2[61]);
+  btf_16_lane_2_3_neon(step1[58], step1[37], c0, &step2[58], &step2[37]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[38]), vnegq_s16(step1[57]), c0,
+                       &step2[38], &step2[57]);
+  btf_16_lane_0_1_neon(step1[54], step1[41], c1, &step2[54], &step2[41]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[42]), vnegq_s16(step1[53]), c1,
+                       &step2[42], &step2[53]);
+  btf_16_lane_2_3_neon(step1[50], step1[45], c1, &step2[50], &step2[45]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[46]), vnegq_s16(step1[49]), c1,
+                       &step2[46], &step2[49]);
+
+  step2[16] = vqaddq_s16(step1[16], step1[17]);
+  step2[17] = vqsubq_s16(step1[16], step1[17]);
+  step2[18] = vqsubq_s16(step1[19], step1[18]);
+  step2[19] = vqaddq_s16(step1[19], step1[18]);
+  step2[20] = vqaddq_s16(step1[20], step1[21]);
+  step2[21] = vqsubq_s16(step1[20], step1[21]);
+  step2[22] = vqsubq_s16(step1[23], step1[22]);
+  step2[23] = vqaddq_s16(step1[23], step1[22]);
+  step2[24] = vqaddq_s16(step1[24], step1[25]);
+  step2[25] = vqsubq_s16(step1[24], step1[25]);
+  step2[26] = vqsubq_s16(step1[27], step1[26]);
+  step2[27] = vqaddq_s16(step1[27], step1[26]);
+  step2[28] = vqaddq_s16(step1[28], step1[29]);
+  step2[29] = vqsubq_s16(step1[28], step1[29]);
+  step2[30] = vqsubq_s16(step1[31], step1[30]);
+  step2[31] = vqaddq_s16(step1[31], step1[30]);
+  step2[32] = step1[32];
+  step2[35] = step1[35];
+  step2[36] = step1[36];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[43] = step1[43];
+  step2[44] = step1[44];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[51] = step1[51];
+  step2[52] = step1[52];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[59] = step1[59];
+  step2[60] = step1[60];
+  step2[63] = step1[63];
+
+  // stage 5
+
+  step1[0] = step2[0];
+  step1[2] = step2[2];
+
+  btf_16_neon(step2[4], cospi[56], cospi[8], &step1[4], &step1[7]);
+  btf_16_neon(step2[6], -cospi[40], cospi[24], &step1[5], &step1[6]);
+  btf_16_lane_0_1_neon(step2[30], step2[17], c2, &step1[30], &step1[17]);
+  btf_16_lane_1_0_neon(vnegq_s16(step2[18]), vnegq_s16(step2[29]), c2,
+                       &step1[18], &step1[29]);
+  btf_16_lane_2_3_neon(step2[26], step2[21], c2, &step1[26], &step1[21]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[22]), vnegq_s16(step2[25]), c2,
+                       &step1[22], &step1[25]);
+
+  step1[8] = vqaddq_s16(step2[8], step2[9]);
+  step1[9] = vqsubq_s16(step2[8], step2[9]);
+  step1[10] = vqsubq_s16(step2[11], step2[10]);
+  step1[11] = vqaddq_s16(step2[11], step2[10]);
+  step1[12] = vqaddq_s16(step2[12], step2[13]);
+  step1[13] = vqsubq_s16(step2[12], step2[13]);
+  step1[14] = vqsubq_s16(step2[15], step2[14]);
+  step1[15] = vqaddq_s16(step2[15], step2[14]);
+  step1[16] = step2[16];
+  step1[19] = step2[19];
+  step1[20] = step2[20];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[27] = step2[27];
+  step1[28] = step2[28];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[35]);
+  step1[33] = vqaddq_s16(step2[33], step2[34]);
+  step1[34] = vqsubq_s16(step2[33], step2[34]);
+  step1[35] = vqsubq_s16(step2[32], step2[35]);
+  step1[36] = vqsubq_s16(step2[39], step2[36]);
+  step1[37] = vqsubq_s16(step2[38], step2[37]);
+  step1[38] = vqaddq_s16(step2[38], step2[37]);
+  step1[39] = vqaddq_s16(step2[39], step2[36]);
+  step1[40] = vqaddq_s16(step2[40], step2[43]);
+  step1[41] = vqaddq_s16(step2[41], step2[42]);
+  step1[42] = vqsubq_s16(step2[41], step2[42]);
+  step1[43] = vqsubq_s16(step2[40], step2[43]);
+  step1[44] = vqsubq_s16(step2[47], step2[44]);
+  step1[45] = vqsubq_s16(step2[46], step2[45]);
+  step1[46] = vqaddq_s16(step2[46], step2[45]);
+  step1[47] = vqaddq_s16(step2[47], step2[44]);
+  step1[48] = vqaddq_s16(step2[48], step2[51]);
+  step1[49] = vqaddq_s16(step2[49], step2[50]);
+  step1[50] = vqsubq_s16(step2[49], step2[50]);
+  step1[51] = vqsubq_s16(step2[48], step2[51]);
+  step1[52] = vqsubq_s16(step2[55], step2[52]);
+  step1[53] = vqsubq_s16(step2[54], step2[53]);
+  step1[54] = vqaddq_s16(step2[54], step2[53]);
+  step1[55] = vqaddq_s16(step2[55], step2[52]);
+  step1[56] = vqaddq_s16(step2[56], step2[59]);
+  step1[57] = vqaddq_s16(step2[57], step2[58]);
+  step1[58] = vqsubq_s16(step2[57], step2[58]);
+  step1[59] = vqsubq_s16(step2[56], step2[59]);
+  step1[60] = vqsubq_s16(step2[63], step2[60]);
+  step1[61] = vqsubq_s16(step2[62], step2[61]);
+  step1[62] = vqaddq_s16(step2[62], step2[61]);
+  step1[63] = vqaddq_s16(step2[63], step2[60]);
+
+  // stage 6
+
+  btf_16_neon(step1[0], cospi[32], cospi[32], &step2[0], &step2[1]);
+  btf_16_neon(step1[2], cospi[48], cospi[16], &step2[2], &step2[3]);
+  btf_16_lane_2_3_neon(step1[14], step1[9], c3, &step2[14], &step2[9]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[10]), vnegq_s16(step1[13]), c3,
+                       &step2[10], &step2[13]);
+  btf_16_lane_0_1_neon(step1[61], step1[34], c2, &step2[61], &step2[34]);
+  btf_16_lane_0_1_neon(step1[60], step1[35], c2, &step2[60], &step2[35]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[36]), vnegq_s16(step1[59]), c2,
+                       &step2[36], &step2[59]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[37]), vnegq_s16(step1[58]), c2,
+                       &step2[37], &step2[58]);
+  btf_16_lane_2_3_neon(step1[53], step1[42], c2, &step2[53], &step2[42]);
+  btf_16_lane_2_3_neon(step1[52], step1[43], c2, &step2[52], &step2[43]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[44]), vnegq_s16(step1[51]), c2,
+                       &step2[44], &step2[51]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[45]), vnegq_s16(step1[50]), c2,
+                       &step2[45], &step2[50]);
+
+  step2[4] = vqaddq_s16(step1[4], step1[5]);
+  step2[5] = vqsubq_s16(step1[4], step1[5]);
+  step2[6] = vqsubq_s16(step1[7], step1[6]);
+  step2[7] = vqaddq_s16(step1[7], step1[6]);
+  step2[8] = step1[8];
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+  step2[15] = step1[15];
+  step2[16] = vqaddq_s16(step1[16], step1[19]);
+  step2[17] = vqaddq_s16(step1[17], step1[18]);
+  step2[18] = vqsubq_s16(step1[17], step1[18]);
+  step2[19] = vqsubq_s16(step1[16], step1[19]);
+  step2[20] = vqsubq_s16(step1[23], step1[20]);
+  step2[21] = vqsubq_s16(step1[22], step1[21]);
+  step2[22] = vqaddq_s16(step1[22], step1[21]);
+  step2[23] = vqaddq_s16(step1[23], step1[20]);
+  step2[24] = vqaddq_s16(step1[24], step1[27]);
+  step2[25] = vqaddq_s16(step1[25], step1[26]);
+  step2[26] = vqsubq_s16(step1[25], step1[26]);
+  step2[27] = vqsubq_s16(step1[24], step1[27]);
+  step2[28] = vqsubq_s16(step1[31], step1[28]);
+  step2[29] = vqsubq_s16(step1[30], step1[29]);
+  step2[30] = vqaddq_s16(step1[30], step1[29]);
+  step2[31] = vqaddq_s16(step1[31], step1[28]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[38] = step1[38];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[41] = step1[41];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[54] = step1[54];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[57] = step1[57];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 7
+
+  btf_16_lane_0_1_neon(step2[6], step2[5], c3, &step1[6], &step1[5]);
+  btf_16_lane_2_3_neon(step2[29], step2[18], c3, &step1[29], &step1[18]);
+  btf_16_lane_2_3_neon(step2[28], step2[19], c3, &step1[28], &step1[19]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[20]), vnegq_s16(step2[27]), c3,
+                       &step1[20], &step1[27]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[21]), vnegq_s16(step2[26]), c3,
+                       &step1[21], &step1[26]);
+
+  step1[0] = vqaddq_s16(step2[0], step2[3]);
+  step1[1] = vqaddq_s16(step2[1], step2[2]);
+  step1[2] = vqsubq_s16(step2[1], step2[2]);
+  step1[3] = vqsubq_s16(step2[0], step2[3]);
+  step1[4] = step2[4];
+  step1[7] = step2[7];
+  step1[8] = vqaddq_s16(step2[8], step2[11]);
+  step1[9] = vqaddq_s16(step2[9], step2[10]);
+  step1[10] = vqsubq_s16(step2[9], step2[10]);
+  step1[11] = vqsubq_s16(step2[8], step2[11]);
+  step1[12] = vqsubq_s16(step2[15], step2[12]);
+  step1[13] = vqsubq_s16(step2[14], step2[13]);
+  step1[14] = vqaddq_s16(step2[14], step2[13]);
+  step1[15] = vqaddq_s16(step2[15], step2[12]);
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[22] = step2[22];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[25] = step2[25];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[39]);
+  step1[33] = vqaddq_s16(step2[33], step2[38]);
+  step1[34] = vqaddq_s16(step2[34], step2[37]);
+  step1[35] = vqaddq_s16(step2[35], step2[36]);
+  step1[36] = vqsubq_s16(step2[35], step2[36]);
+  step1[37] = vqsubq_s16(step2[34], step2[37]);
+  step1[38] = vqsubq_s16(step2[33], step2[38]);
+  step1[39] = vqsubq_s16(step2[32], step2[39]);
+  step1[40] = vqsubq_s16(step2[47], step2[40]);
+  step1[41] = vqsubq_s16(step2[46], step2[41]);
+  step1[42] = vqsubq_s16(step2[45], step2[42]);
+  step1[43] = vqsubq_s16(step2[44], step2[43]);
+  step1[44] = vqaddq_s16(step2[43], step2[44]);
+  step1[45] = vqaddq_s16(step2[42], step2[45]);
+  step1[46] = vqaddq_s16(step2[41], step2[46]);
+  step1[47] = vqaddq_s16(step2[40], step2[47]);
+  step1[48] = vqaddq_s16(step2[48], step2[55]);
+  step1[49] = vqaddq_s16(step2[49], step2[54]);
+  step1[50] = vqaddq_s16(step2[50], step2[53]);
+  step1[51] = vqaddq_s16(step2[51], step2[52]);
+  step1[52] = vqsubq_s16(step2[51], step2[52]);
+  step1[53] = vqsubq_s16(step2[50], step2[53]);
+  step1[54] = vqsubq_s16(step2[49], step2[54]);
+  step1[55] = vqsubq_s16(step2[48], step2[55]);
+  step1[56] = vqsubq_s16(step2[63], step2[56]);
+  step1[57] = vqsubq_s16(step2[62], step2[57]);
+  step1[58] = vqsubq_s16(step2[61], step2[58]);
+  step1[59] = vqsubq_s16(step2[60], step2[59]);
+  step1[60] = vqaddq_s16(step2[59], step2[60]);
+  step1[61] = vqaddq_s16(step2[58], step2[61]);
+  step1[62] = vqaddq_s16(step2[57], step2[62]);
+  step1[63] = vqaddq_s16(step2[56], step2[63]);
+
+  // stage 8
+
+  btf_16_lane_0_1_neon(step1[13], step1[10], c3, &step2[13], &step2[10]);
+  btf_16_lane_0_1_neon(step1[12], step1[11], c3, &step2[12], &step2[11]);
+  btf_16_lane_2_3_neon(step1[59], step1[36], c3, &step2[59], &step2[36]);
+  btf_16_lane_2_3_neon(step1[58], step1[37], c3, &step2[58], &step2[37]);
+  btf_16_lane_2_3_neon(step1[57], step1[38], c3, &step2[57], &step2[38]);
+  btf_16_lane_2_3_neon(step1[56], step1[39], c3, &step2[56], &step2[39]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[40]), vnegq_s16(step1[55]), c3,
+                       &step2[40], &step2[55]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[41]), vnegq_s16(step1[54]), c3,
+                       &step2[41], &step2[54]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[42]), vnegq_s16(step1[53]), c3,
+                       &step2[42], &step2[53]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[43]), vnegq_s16(step1[52]), c3,
+                       &step2[43], &step2[52]);
+
+  step2[0] = vqaddq_s16(step1[0], step1[7]);
+  step2[1] = vqaddq_s16(step1[1], step1[6]);
+  step2[2] = vqaddq_s16(step1[2], step1[5]);
+  step2[3] = vqaddq_s16(step1[3], step1[4]);
+  step2[4] = vqsubq_s16(step1[3], step1[4]);
+  step2[5] = vqsubq_s16(step1[2], step1[5]);
+  step2[6] = vqsubq_s16(step1[1], step1[6]);
+  step2[7] = vqsubq_s16(step1[0], step1[7]);
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+  step2[16] = vqaddq_s16(step1[16], step1[23]);
+  step2[17] = vqaddq_s16(step1[17], step1[22]);
+  step2[18] = vqaddq_s16(step1[18], step1[21]);
+  step2[19] = vqaddq_s16(step1[19], step1[20]);
+  step2[20] = vqsubq_s16(step1[19], step1[20]);
+  step2[21] = vqsubq_s16(step1[18], step1[21]);
+  step2[22] = vqsubq_s16(step1[17], step1[22]);
+  step2[23] = vqsubq_s16(step1[16], step1[23]);
+  step2[24] = vqsubq_s16(step1[31], step1[24]);
+  step2[25] = vqsubq_s16(step1[30], step1[25]);
+  step2[26] = vqsubq_s16(step1[29], step1[26]);
+  step2[27] = vqsubq_s16(step1[28], step1[27]);
+  step2[28] = vqaddq_s16(step1[28], step1[27]);
+  step2[29] = vqaddq_s16(step1[29], step1[26]);
+  step2[30] = vqaddq_s16(step1[30], step1[25]);
+  step2[31] = vqaddq_s16(step1[31], step1[24]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[34] = step1[34];
+  step2[35] = step1[35];
+  step2[44] = step1[44];
+  step2[45] = step1[45];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[50] = step1[50];
+  step2[51] = step1[51];
+  step2[60] = step1[60];
+  step2[61] = step1[61];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 9
+  idct64_stage9_neon(step2, step1, cos_bit);
+
+  // stage 10
+  idct64_stage10_neon(step1, step2, cos_bit);
+
+  // stage 11
+
+  out[0] = vqaddq_s16(step2[0], step2[63]);
+  out[1] = vqaddq_s16(step2[1], step2[62]);
+  out[2] = vqaddq_s16(step2[2], step2[61]);
+  out[3] = vqaddq_s16(step2[3], step2[60]);
+  out[4] = vqaddq_s16(step2[4], step2[59]);
+  out[5] = vqaddq_s16(step2[5], step2[58]);
+  out[6] = vqaddq_s16(step2[6], step2[57]);
+  out[7] = vqaddq_s16(step2[7], step2[56]);
+  out[8] = vqaddq_s16(step2[8], step2[55]);
+  out[9] = vqaddq_s16(step2[9], step2[54]);
+  out[10] = vqaddq_s16(step2[10], step2[53]);
+  out[11] = vqaddq_s16(step2[11], step2[52]);
+  out[12] = vqaddq_s16(step2[12], step2[51]);
+  out[13] = vqaddq_s16(step2[13], step2[50]);
+  out[14] = vqaddq_s16(step2[14], step2[49]);
+  out[15] = vqaddq_s16(step2[15], step2[48]);
+  out[16] = vqaddq_s16(step2[16], step2[47]);
+  out[17] = vqaddq_s16(step2[17], step2[46]);
+  out[18] = vqaddq_s16(step2[18], step2[45]);
+  out[19] = vqaddq_s16(step2[19], step2[44]);
+  out[20] = vqaddq_s16(step2[20], step2[43]);
+  out[21] = vqaddq_s16(step2[21], step2[42]);
+  out[22] = vqaddq_s16(step2[22], step2[41]);
+  out[23] = vqaddq_s16(step2[23], step2[40]);
+  out[24] = vqaddq_s16(step2[24], step2[39]);
+  out[25] = vqaddq_s16(step2[25], step2[38]);
+  out[26] = vqaddq_s16(step2[26], step2[37]);
+  out[27] = vqaddq_s16(step2[27], step2[36]);
+  out[28] = vqaddq_s16(step2[28], step2[35]);
+  out[29] = vqaddq_s16(step2[29], step2[34]);
+  out[30] = vqaddq_s16(step2[30], step2[33]);
+  out[31] = vqaddq_s16(step2[31], step2[32]);
+  out[32] = vqsubq_s16(step2[31], step2[32]);
+  out[33] = vqsubq_s16(step2[30], step2[33]);
+  out[34] = vqsubq_s16(step2[29], step2[34]);
+  out[35] = vqsubq_s16(step2[28], step2[35]);
+  out[36] = vqsubq_s16(step2[27], step2[36]);
+  out[37] = vqsubq_s16(step2[26], step2[37]);
+  out[38] = vqsubq_s16(step2[25], step2[38]);
+  out[39] = vqsubq_s16(step2[24], step2[39]);
+  out[40] = vqsubq_s16(step2[23], step2[40]);
+  out[41] = vqsubq_s16(step2[22], step2[41]);
+  out[42] = vqsubq_s16(step2[21], step2[42]);
+  out[43] = vqsubq_s16(step2[20], step2[43]);
+  out[44] = vqsubq_s16(step2[19], step2[44]);
+  out[45] = vqsubq_s16(step2[18], step2[45]);
+  out[46] = vqsubq_s16(step2[17], step2[46]);
+  out[47] = vqsubq_s16(step2[16], step2[47]);
+  out[48] = vqsubq_s16(step2[15], step2[48]);
+  out[49] = vqsubq_s16(step2[14], step2[49]);
+  out[50] = vqsubq_s16(step2[13], step2[50]);
+  out[51] = vqsubq_s16(step2[12], step2[51]);
+  out[52] = vqsubq_s16(step2[11], step2[52]);
+  out[53] = vqsubq_s16(step2[10], step2[53]);
+  out[54] = vqsubq_s16(step2[9], step2[54]);
+  out[55] = vqsubq_s16(step2[8], step2[55]);
+  out[56] = vqsubq_s16(step2[7], step2[56]);
+  out[57] = vqsubq_s16(step2[6], step2[57]);
+  out[58] = vqsubq_s16(step2[5], step2[58]);
+  out[59] = vqsubq_s16(step2[4], step2[59]);
+  out[60] = vqsubq_s16(step2[3], step2[60]);
+  out[61] = vqsubq_s16(step2[2], step2[61]);
+  out[62] = vqsubq_s16(step2[1], step2[62]);
+  out[63] = vqsubq_s16(step2[0], step2[63]);
+}
+
+static INLINE void idct64_low1_new_neon(int16x8_t *input, int16x8_t *out,
+                                        int8_t cos_bit, int bit) {
+  (void)bit;
+  const int32_t *cospi = cospi_arr(cos_bit);
+  int16x8_t step1;
+  int32x4_t t32[2];
+
+  // stage 1
+  // stage 2
+  // stage 3
+  // stage 4
+  // stage 5
+  // stage 6
+
+  t32[0] = vmull_n_s16(vget_low_s16(input[0]), cospi[32]);
+  t32[1] = vmull_n_s16(vget_high_s16(input[0]), cospi[32]);
+
+  step1 = vcombine_s16(vrshrn_n_s32(t32[0], INV_COS_BIT),
+                       vrshrn_n_s32(t32[1], INV_COS_BIT));
+  // stage 7
+  // stage 8
+  // stage 9
+  // stage 10
+  // stage 11
+  out[0] = step1;
+  out[1] = step1;
+  out[2] = step1;
+  out[3] = step1;
+  out[4] = step1;
+  out[5] = step1;
+  out[6] = step1;
+  out[7] = step1;
+  out[8] = step1;
+  out[9] = step1;
+  out[10] = step1;
+  out[11] = step1;
+  out[12] = step1;
+  out[13] = step1;
+  out[14] = step1;
+  out[15] = step1;
+  out[16] = step1;
+  out[17] = step1;
+  out[18] = step1;
+  out[19] = step1;
+  out[20] = step1;
+  out[21] = step1;
+  out[22] = step1;
+  out[23] = step1;
+  out[24] = step1;
+  out[25] = step1;
+  out[26] = step1;
+  out[27] = step1;
+  out[28] = step1;
+  out[29] = step1;
+  out[30] = step1;
+  out[31] = step1;
+  out[32] = step1;
+  out[33] = step1;
+  out[34] = step1;
+  out[35] = step1;
+  out[36] = step1;
+  out[37] = step1;
+  out[38] = step1;
+  out[39] = step1;
+  out[40] = step1;
+  out[41] = step1;
+  out[42] = step1;
+  out[43] = step1;
+  out[44] = step1;
+  out[45] = step1;
+  out[46] = step1;
+  out[47] = step1;
+  out[48] = step1;
+  out[49] = step1;
+  out[50] = step1;
+  out[51] = step1;
+  out[52] = step1;
+  out[53] = step1;
+  out[54] = step1;
+  out[55] = step1;
+  out[56] = step1;
+  out[57] = step1;
+  out[58] = step1;
+  out[59] = step1;
+  out[60] = step1;
+  out[61] = step1;
+  out[62] = step1;
+  out[63] = step1;
+}
+
+static INLINE void idct64_low8_new_neon(int16x8_t *in, int16x8_t *out,
+                                        int8_t cos_bit, int bit) {
+  (void)bit;
+  const int32_t *cospi = cospi_arr(cos_bit);
+  int16x8_t step2[64], step1[64];
+
+  const int16x4_t c0 =
+      create_s16x4_neon((int16_t *)(cospi + 4), (int16_t *)(cospi + 60),
+                        (int16_t *)(cospi + 36), (int16_t *)(cospi + 28));
+  const int16x4_t c1 =
+      create_s16x4_neon((int16_t *)(cospi + 20), (int16_t *)(cospi + 44),
+                        (int16_t *)(cospi + 52), (int16_t *)(cospi + 12));
+  const int16x4_t c2 =
+      create_s16x4_neon((int16_t *)(cospi + 8), (int16_t *)(cospi + 56),
+                        (int16_t *)(cospi + 40), (int16_t *)(cospi + 24));
+  const int16x4_t c3 =
+      create_s16x4_neon((int16_t *)(cospi + 32), (int16_t *)(cospi + 32),
+                        (int16_t *)(cospi + 16), (int16_t *)(cospi + 48));
+
+  // stage 1
+  // stage 2
+
+  step2[0] = in[0];
+  step2[8] = in[4];
+  step2[16] = in[2];
+  step2[24] = in[6];
+
+  btf_16_neon(in[1], cospi[63], cospi[1], &step2[32], &step2[63]);
+  btf_16_neon(in[7], -cospi[57], cospi[7], &step2[39], &step2[56]);
+  btf_16_neon(in[5], cospi[59], cospi[5], &step2[40], &step2[55]);
+  btf_16_neon(in[3], -cospi[61], cospi[3], &step2[47], &step2[48]);
+
+  // stage 3
+
+  step1[0] = step2[0];
+  step1[8] = step2[8];
+
+  btf_16_neon(step2[16], cospi[62], cospi[2], &step1[16], &step1[31]);
+  btf_16_neon(step2[24], -cospi[58], cospi[6], &step1[23], &step1[24]);
+
+  step1[32] = step2[32];
+  step1[33] = step2[32];
+  step1[38] = step2[39];
+  step1[39] = step2[39];
+  step1[40] = step2[40];
+  step1[41] = step2[40];
+  step1[46] = step2[47];
+  step1[47] = step2[47];
+  step1[48] = step2[48];
+  step1[49] = step2[48];
+  step1[54] = step2[55];
+  step1[55] = step2[55];
+  step1[56] = step2[56];
+  step1[57] = step2[56];
+  step1[62] = step2[63];
+  step1[63] = step2[63];
+
+  // stage 4
+
+  step2[0] = step1[0];
+
+  btf_16_neon(step1[8], cospi[60], cospi[4], &step2[8], &step2[15]);
+  btf_16_lane_0_1_neon(step1[62], step1[33], c0, &step2[62], &step2[33]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[38]), vnegq_s16(step1[57]), c0,
+                       &step2[38], &step2[57]);
+  btf_16_lane_0_1_neon(step1[54], step1[41], c1, &step2[54], &step2[41]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[46]), vnegq_s16(step1[49]), c1,
+                       &step2[46], &step2[49]);
+
+  step2[16] = step1[16];
+  step2[17] = step1[16];
+  step2[22] = step1[23];
+  step2[23] = step1[23];
+  step2[24] = step1[24];
+  step2[25] = step1[24];
+  step2[30] = step1[31];
+  step2[31] = step1[31];
+  step2[32] = step1[32];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[63] = step1[63];
+
+  // stage 5
+
+  step1[0] = step2[0];
+
+  btf_16_lane_0_1_neon(step2[30], step2[17], c2, &step1[30], &step1[17]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[22]), vnegq_s16(step2[25]), c2,
+                       &step1[22], &step1[25]);
+
+  step1[8] = step2[8];
+  step1[9] = step2[8];
+  step1[14] = step2[15];
+  step1[15] = step2[15];
+
+  step1[16] = step2[16];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[31] = step2[31];
+  step1[32] = step2[32];
+  step1[33] = step2[33];
+  step1[34] = step2[33];
+  step1[35] = step2[32];
+  step1[36] = step2[39];
+  step1[37] = step2[38];
+  step1[38] = step2[38];
+  step1[39] = step2[39];
+  step1[40] = step2[40];
+  step1[41] = step2[41];
+  step1[42] = step2[41];
+  step1[43] = step2[40];
+  step1[44] = step2[47];
+  step1[45] = step2[46];
+  step1[46] = step2[46];
+  step1[47] = step2[47];
+  step1[48] = step2[48];
+  step1[49] = step2[49];
+  step1[50] = step2[49];
+  step1[51] = step2[48];
+  step1[52] = step2[55];
+  step1[53] = step2[54];
+  step1[54] = step2[54];
+  step1[55] = step2[55];
+  step1[56] = step2[56];
+  step1[57] = step2[57];
+  step1[58] = step2[57];
+  step1[59] = step2[56];
+  step1[60] = step2[63];
+  step1[61] = step2[62];
+  step1[62] = step2[62];
+  step1[63] = step2[63];
+
+  // stage 6
+
+  btf_16_neon(step1[0], cospi[32], cospi[32], &step2[0], &step2[1]);
+  btf_16_lane_2_3_neon(step1[14], step1[9], c3, &step2[14], &step2[9]);
+  btf_16_lane_0_1_neon(step1[61], step1[34], c2, &step2[61], &step2[34]);
+  btf_16_lane_0_1_neon(step1[60], step1[35], c2, &step2[60], &step2[35]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[36]), vnegq_s16(step1[59]), c2,
+                       &step2[36], &step2[59]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[37]), vnegq_s16(step1[58]), c2,
+                       &step2[37], &step2[58]);
+  btf_16_lane_2_3_neon(step1[53], step1[42], c2, &step2[53], &step2[42]);
+  btf_16_lane_2_3_neon(step1[52], step1[43], c2, &step2[52], &step2[43]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[44]), vnegq_s16(step1[51]), c2,
+                       &step2[44], &step2[51]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[45]), vnegq_s16(step1[50]), c2,
+                       &step2[45], &step2[50]);
+
+  step2[8] = step1[8];
+  step2[15] = step1[15];
+  step2[16] = step1[16];
+  step2[17] = step1[17];
+  step2[18] = step1[17];
+  step2[19] = step1[16];
+  step2[20] = step1[23];
+  step2[21] = step1[22];
+  step2[22] = step1[22];
+  step2[23] = step1[23];
+  step2[24] = step1[24];
+  step2[25] = step1[25];
+  step2[26] = step1[25];
+  step2[27] = step1[24];
+  step2[28] = step1[31];
+  step2[29] = step1[30];
+  step2[30] = step1[30];
+  step2[31] = step1[31];
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[38] = step1[38];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[41] = step1[41];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[54] = step1[54];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[57] = step1[57];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 7
+
+  btf_16_lane_2_3_neon(step2[29], step2[18], c3, &step1[29], &step1[18]);
+  btf_16_lane_2_3_neon(step2[28], step2[19], c3, &step1[28], &step1[19]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[20]), vnegq_s16(step2[27]), c3,
+                       &step1[20], &step1[27]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[21]), vnegq_s16(step2[26]), c3,
+                       &step1[21], &step1[26]);
+
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[1];
+  step1[3] = step2[0];
+  step1[8] = step2[8];
+  step1[9] = step2[9];
+  step1[10] = step2[9];
+  step1[11] = step2[8];
+  step1[12] = step2[15];
+  step1[13] = step2[14];
+  step1[14] = step2[14];
+  step1[15] = step2[15];
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[22] = step2[22];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[25] = step2[25];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[39]);
+  step1[33] = vqaddq_s16(step2[33], step2[38]);
+  step1[34] = vqaddq_s16(step2[34], step2[37]);
+  step1[35] = vqaddq_s16(step2[35], step2[36]);
+  step1[36] = vqsubq_s16(step2[35], step2[36]);
+  step1[37] = vqsubq_s16(step2[34], step2[37]);
+  step1[38] = vqsubq_s16(step2[33], step2[38]);
+  step1[39] = vqsubq_s16(step2[32], step2[39]);
+  step1[40] = vqsubq_s16(step2[47], step2[40]);
+  step1[41] = vqsubq_s16(step2[46], step2[41]);
+  step1[42] = vqsubq_s16(step2[45], step2[42]);
+  step1[43] = vqsubq_s16(step2[44], step2[43]);
+  step1[44] = vqaddq_s16(step2[43], step2[44]);
+  step1[45] = vqaddq_s16(step2[42], step2[45]);
+  step1[46] = vqaddq_s16(step2[41], step2[46]);
+  step1[47] = vqaddq_s16(step2[40], step2[47]);
+  step1[48] = vqaddq_s16(step2[48], step2[55]);
+  step1[49] = vqaddq_s16(step2[49], step2[54]);
+  step1[50] = vqaddq_s16(step2[50], step2[53]);
+  step1[51] = vqaddq_s16(step2[51], step2[52]);
+  step1[52] = vqsubq_s16(step2[51], step2[52]);
+  step1[53] = vqsubq_s16(step2[50], step2[53]);
+  step1[54] = vqsubq_s16(step2[49], step2[54]);
+  step1[55] = vqsubq_s16(step2[48], step2[55]);
+  step1[56] = vqsubq_s16(step2[63], step2[56]);
+  step1[57] = vqsubq_s16(step2[62], step2[57]);
+  step1[58] = vqsubq_s16(step2[61], step2[58]);
+  step1[59] = vqsubq_s16(step2[60], step2[59]);
+  step1[60] = vqaddq_s16(step2[59], step2[60]);
+  step1[61] = vqaddq_s16(step2[58], step2[61]);
+  step1[62] = vqaddq_s16(step2[57], step2[62]);
+  step1[63] = vqaddq_s16(step2[56], step2[63]);
+
+  // stage 8
+
+  btf_16_lane_0_1_neon(step1[13], step1[10], c3, &step2[13], &step2[10]);
+  btf_16_lane_0_1_neon(step1[12], step1[11], c3, &step2[12], &step2[11]);
+  btf_16_lane_2_3_neon(step1[59], step1[36], c3, &step2[59], &step2[36]);
+  btf_16_lane_2_3_neon(step1[58], step1[37], c3, &step2[58], &step2[37]);
+  btf_16_lane_2_3_neon(step1[57], step1[38], c3, &step2[57], &step2[38]);
+  btf_16_lane_2_3_neon(step1[56], step1[39], c3, &step2[56], &step2[39]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[40]), vnegq_s16(step1[55]), c3,
+                       &step2[40], &step2[55]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[41]), vnegq_s16(step1[54]), c3,
+                       &step2[41], &step2[54]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[42]), vnegq_s16(step1[53]), c3,
+                       &step2[42], &step2[53]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[43]), vnegq_s16(step1[52]), c3,
+                       &step2[43], &step2[52]);
+
+  step2[0] = step1[0];
+  step2[1] = step1[1];
+  step2[2] = step1[2];
+  step2[3] = step1[3];
+  step2[4] = step1[3];
+  step2[5] = step1[2];
+  step2[6] = step1[1];
+  step2[7] = step1[0];
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+  step2[16] = vqaddq_s16(step1[16], step1[23]);
+  step2[17] = vqaddq_s16(step1[17], step1[22]);
+  step2[18] = vqaddq_s16(step1[18], step1[21]);
+  step2[19] = vqaddq_s16(step1[19], step1[20]);
+  step2[20] = vqsubq_s16(step1[19], step1[20]);
+  step2[21] = vqsubq_s16(step1[18], step1[21]);
+  step2[22] = vqsubq_s16(step1[17], step1[22]);
+  step2[23] = vqsubq_s16(step1[16], step1[23]);
+  step2[24] = vqsubq_s16(step1[31], step1[24]);
+  step2[25] = vqsubq_s16(step1[30], step1[25]);
+  step2[26] = vqsubq_s16(step1[29], step1[26]);
+  step2[27] = vqsubq_s16(step1[28], step1[27]);
+  step2[28] = vqaddq_s16(step1[28], step1[27]);
+  step2[29] = vqaddq_s16(step1[29], step1[26]);
+  step2[30] = vqaddq_s16(step1[30], step1[25]);
+  step2[31] = vqaddq_s16(step1[31], step1[24]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[34] = step1[34];
+  step2[35] = step1[35];
+  step2[44] = step1[44];
+  step2[45] = step1[45];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[50] = step1[50];
+  step2[51] = step1[51];
+  step2[60] = step1[60];
+  step2[61] = step1[61];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 9
+  idct64_stage9_neon(step2, step1, cos_bit);
+
+  // stage 10
+  idct64_stage10_neon(step1, step2, cos_bit);
+
+  // stage 11
+
+  out[0] = vqaddq_s16(step2[0], step2[63]);
+  out[1] = vqaddq_s16(step2[1], step2[62]);
+  out[2] = vqaddq_s16(step2[2], step2[61]);
+  out[3] = vqaddq_s16(step2[3], step2[60]);
+  out[4] = vqaddq_s16(step2[4], step2[59]);
+  out[5] = vqaddq_s16(step2[5], step2[58]);
+  out[6] = vqaddq_s16(step2[6], step2[57]);
+  out[7] = vqaddq_s16(step2[7], step2[56]);
+  out[8] = vqaddq_s16(step2[8], step2[55]);
+  out[9] = vqaddq_s16(step2[9], step2[54]);
+  out[10] = vqaddq_s16(step2[10], step2[53]);
+  out[11] = vqaddq_s16(step2[11], step2[52]);
+  out[12] = vqaddq_s16(step2[12], step2[51]);
+  out[13] = vqaddq_s16(step2[13], step2[50]);
+  out[14] = vqaddq_s16(step2[14], step2[49]);
+  out[15] = vqaddq_s16(step2[15], step2[48]);
+  out[16] = vqaddq_s16(step2[16], step2[47]);
+  out[17] = vqaddq_s16(step2[17], step2[46]);
+  out[18] = vqaddq_s16(step2[18], step2[45]);
+  out[19] = vqaddq_s16(step2[19], step2[44]);
+  out[20] = vqaddq_s16(step2[20], step2[43]);
+  out[21] = vqaddq_s16(step2[21], step2[42]);
+  out[22] = vqaddq_s16(step2[22], step2[41]);
+  out[23] = vqaddq_s16(step2[23], step2[40]);
+  out[24] = vqaddq_s16(step2[24], step2[39]);
+  out[25] = vqaddq_s16(step2[25], step2[38]);
+  out[26] = vqaddq_s16(step2[26], step2[37]);
+  out[27] = vqaddq_s16(step2[27], step2[36]);
+  out[28] = vqaddq_s16(step2[28], step2[35]);
+  out[29] = vqaddq_s16(step2[29], step2[34]);
+  out[30] = vqaddq_s16(step2[30], step2[33]);
+  out[31] = vqaddq_s16(step2[31], step2[32]);
+  out[32] = vqsubq_s16(step2[31], step2[32]);
+  out[33] = vqsubq_s16(step2[30], step2[33]);
+  out[34] = vqsubq_s16(step2[29], step2[34]);
+  out[35] = vqsubq_s16(step2[28], step2[35]);
+  out[36] = vqsubq_s16(step2[27], step2[36]);
+  out[37] = vqsubq_s16(step2[26], step2[37]);
+  out[38] = vqsubq_s16(step2[25], step2[38]);
+  out[39] = vqsubq_s16(step2[24], step2[39]);
+  out[40] = vqsubq_s16(step2[23], step2[40]);
+  out[41] = vqsubq_s16(step2[22], step2[41]);
+  out[42] = vqsubq_s16(step2[21], step2[42]);
+  out[43] = vqsubq_s16(step2[20], step2[43]);
+  out[44] = vqsubq_s16(step2[19], step2[44]);
+  out[45] = vqsubq_s16(step2[18], step2[45]);
+  out[46] = vqsubq_s16(step2[17], step2[46]);
+  out[47] = vqsubq_s16(step2[16], step2[47]);
+  out[48] = vqsubq_s16(step2[15], step2[48]);
+  out[49] = vqsubq_s16(step2[14], step2[49]);
+  out[50] = vqsubq_s16(step2[13], step2[50]);
+  out[51] = vqsubq_s16(step2[12], step2[51]);
+  out[52] = vqsubq_s16(step2[11], step2[52]);
+  out[53] = vqsubq_s16(step2[10], step2[53]);
+  out[54] = vqsubq_s16(step2[9], step2[54]);
+  out[55] = vqsubq_s16(step2[8], step2[55]);
+  out[56] = vqsubq_s16(step2[7], step2[56]);
+  out[57] = vqsubq_s16(step2[6], step2[57]);
+  out[58] = vqsubq_s16(step2[5], step2[58]);
+  out[59] = vqsubq_s16(step2[4], step2[59]);
+  out[60] = vqsubq_s16(step2[3], step2[60]);
+  out[61] = vqsubq_s16(step2[2], step2[61]);
+  out[62] = vqsubq_s16(step2[1], step2[62]);
+  out[63] = vqsubq_s16(step2[0], step2[63]);
+}
+
+static INLINE void idct64_low16_new_neon(int16x8_t *in, int16x8_t *out,
+                                         int8_t cos_bit, int bit) {
+  (void)bit;
+  const int32_t *cospi = cospi_arr(cos_bit);
+  int16x8_t step2[64], step1[64];
+
+  const int16x4_t c0 =
+      create_s16x4_neon((int16_t *)(cospi + 4), (int16_t *)(cospi + 60),
+                        (int16_t *)(cospi + 36), (int16_t *)(cospi + 28));
+  const int16x4_t c1 =
+      create_s16x4_neon((int16_t *)(cospi + 20), (int16_t *)(cospi + 44),
+                        (int16_t *)(cospi + 52), (int16_t *)(cospi + 12));
+  const int16x4_t c2 =
+      create_s16x4_neon((int16_t *)(cospi + 8), (int16_t *)(cospi + 56),
+                        (int16_t *)(cospi + 40), (int16_t *)(cospi + 24));
+  const int16x4_t c3 =
+      create_s16x4_neon((int16_t *)(cospi + 32), (int16_t *)(cospi + 32),
+                        (int16_t *)(cospi + 16), (int16_t *)(cospi + 48));
+
+  // stage 1
+  // stage 2
+
+  step2[0] = in[0];
+  step2[4] = in[8];
+  step2[8] = in[4];
+  step2[12] = in[12];
+  step2[16] = in[2];
+  step2[20] = in[10];
+  step2[24] = in[6];
+  step2[28] = in[14];
+
+  btf_16_neon(in[1], cospi[63], cospi[1], &step2[32], &step2[63]);
+  btf_16_neon(in[15], -cospi[49], cospi[15], &step2[35], &step2[60]);
+  btf_16_neon(in[9], cospi[55], cospi[9], &step2[36], &step2[59]);
+  btf_16_neon(in[7], -cospi[57], cospi[7], &step2[39], &step2[56]);
+  btf_16_neon(in[5], cospi[59], cospi[5], &step2[40], &step2[55]);
+  btf_16_neon(in[11], -cospi[53], cospi[11], &step2[43], &step2[52]);
+  btf_16_neon(in[13], cospi[51], cospi[13], &step2[44], &step2[51]);
+  btf_16_neon(in[3], -cospi[61], cospi[3], &step2[47], &step2[48]);
+
+  // stage 3
+
+  step1[0] = step2[0];
+  step1[4] = step2[4];
+  step1[8] = step2[8];
+  step1[12] = step2[12];
+
+  btf_16_neon(step2[16], cospi[62], cospi[2], &step1[16], &step1[31]);
+  btf_16_neon(step2[20], cospi[54], cospi[10], &step1[20], &step1[27]);
+  btf_16_neon(step2[24], -cospi[58], cospi[6], &step1[23], &step1[24]);
+  btf_16_neon(step2[28], -cospi[50], cospi[14], &step1[19], &step1[28]);
+
+  step1[32] = step2[32];
+  step1[33] = step2[32];
+  step1[34] = step2[35];
+  step1[35] = step2[35];
+  step1[36] = step2[36];
+  step1[37] = step2[36];
+  step1[38] = step2[39];
+  step1[39] = step2[39];
+  step1[40] = step2[40];
+  step1[41] = step2[40];
+  step1[42] = step2[43];
+  step1[43] = step2[43];
+  step1[44] = step2[44];
+  step1[45] = step2[44];
+  step1[46] = step2[47];
+  step1[47] = step2[47];
+  step1[48] = step2[48];
+  step1[49] = step2[48];
+  step1[50] = step2[51];
+  step1[51] = step2[51];
+  step1[52] = step2[52];
+  step1[53] = step2[52];
+  step1[54] = step2[55];
+  step1[55] = step2[55];
+  step1[56] = step2[56];
+  step1[57] = step2[56];
+  step1[58] = step2[59];
+  step1[59] = step2[59];
+  step1[60] = step2[60];
+  step1[61] = step2[60];
+  step1[62] = step2[63];
+  step1[63] = step2[63];
+
+  // stage 4
+
+  step2[0] = step1[0];
+  step2[4] = step1[4];
+
+  btf_16_neon(step1[8], cospi[60], cospi[4], &step2[8], &step2[15]);
+  btf_16_neon(step1[12], -cospi[52], cospi[12], &step2[11], &step2[12]);
+  btf_16_lane_0_1_neon(step1[62], step1[33], c0, &step2[62], &step2[33]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[34]), vnegq_s16(step1[61]), c0,
+                       &step2[34], &step2[61]);
+  btf_16_lane_2_3_neon(step1[58], step1[37], c0, &step2[58], &step2[37]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[38]), vnegq_s16(step1[57]), c0,
+                       &step2[38], &step2[57]);
+  btf_16_lane_0_1_neon(step1[54], step1[41], c1, &step2[54], &step2[41]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[42]), vnegq_s16(step1[53]), c1,
+                       &step2[42], &step2[53]);
+  btf_16_lane_2_3_neon(step1[50], step1[45], c1, &step2[50], &step2[45]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[46]), vnegq_s16(step1[49]), c1,
+                       &step2[46], &step2[49]);
+
+  step2[16] = step1[16];
+  step2[17] = step1[16];
+  step2[18] = step1[19];
+  step2[19] = step1[19];
+  step2[20] = step1[20];
+  step2[21] = step1[20];
+  step2[22] = step1[23];
+  step2[23] = step1[23];
+  step2[24] = step1[24];
+  step2[25] = step1[24];
+  step2[26] = step1[27];
+  step2[27] = step1[27];
+  step2[28] = step1[28];
+  step2[29] = step1[28];
+  step2[30] = step1[31];
+  step2[31] = step1[31];
+  step2[32] = step1[32];
+  step2[35] = step1[35];
+  step2[36] = step1[36];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[43] = step1[43];
+  step2[44] = step1[44];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[51] = step1[51];
+  step2[52] = step1[52];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[59] = step1[59];
+  step2[60] = step1[60];
+  step2[63] = step1[63];
+
+  // stage 5
+
+  step1[0] = step2[0];
+
+  btf_16_neon(step2[4], cospi[56], cospi[8], &step1[4], &step1[7]);
+  btf_16_lane_0_1_neon(step2[30], step2[17], c2, &step1[30], &step1[17]);
+  btf_16_lane_1_0_neon(vnegq_s16(step2[18]), vnegq_s16(step2[29]), c2,
+                       &step1[18], &step1[29]);
+  btf_16_lane_2_3_neon(step2[26], step2[21], c2, &step1[26], &step1[21]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[22]), vnegq_s16(step2[25]), c2,
+                       &step1[22], &step1[25]);
+
+  step1[8] = step2[8];
+  step1[9] = step2[8];
+  step1[10] = step2[11];
+  step1[11] = step2[11];
+  step1[12] = step2[12];
+  step1[13] = step2[12];
+  step1[14] = step2[15];
+  step1[15] = step2[15];
+  step1[16] = step2[16];
+  step1[19] = step2[19];
+  step1[20] = step2[20];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[27] = step2[27];
+  step1[28] = step2[28];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[35]);
+  step1[33] = vqaddq_s16(step2[33], step2[34]);
+  step1[34] = vqsubq_s16(step2[33], step2[34]);
+  step1[35] = vqsubq_s16(step2[32], step2[35]);
+  step1[36] = vqsubq_s16(step2[39], step2[36]);
+  step1[37] = vqsubq_s16(step2[38], step2[37]);
+  step1[38] = vqaddq_s16(step2[38], step2[37]);
+  step1[39] = vqaddq_s16(step2[39], step2[36]);
+  step1[40] = vqaddq_s16(step2[40], step2[43]);
+  step1[41] = vqaddq_s16(step2[41], step2[42]);
+  step1[42] = vqsubq_s16(step2[41], step2[42]);
+  step1[43] = vqsubq_s16(step2[40], step2[43]);
+  step1[44] = vqsubq_s16(step2[47], step2[44]);
+  step1[45] = vqsubq_s16(step2[46], step2[45]);
+  step1[46] = vqaddq_s16(step2[46], step2[45]);
+  step1[47] = vqaddq_s16(step2[47], step2[44]);
+  step1[48] = vqaddq_s16(step2[48], step2[51]);
+  step1[49] = vqaddq_s16(step2[49], step2[50]);
+  step1[50] = vqsubq_s16(step2[49], step2[50]);
+  step1[51] = vqsubq_s16(step2[48], step2[51]);
+  step1[52] = vqsubq_s16(step2[55], step2[52]);
+  step1[53] = vqsubq_s16(step2[54], step2[53]);
+  step1[54] = vqaddq_s16(step2[54], step2[53]);
+  step1[55] = vqaddq_s16(step2[55], step2[52]);
+  step1[56] = vqaddq_s16(step2[56], step2[59]);
+  step1[57] = vqaddq_s16(step2[57], step2[58]);
+  step1[58] = vqsubq_s16(step2[57], step2[58]);
+  step1[59] = vqsubq_s16(step2[56], step2[59]);
+  step1[60] = vqsubq_s16(step2[63], step2[60]);
+  step1[61] = vqsubq_s16(step2[62], step2[61]);
+  step1[62] = vqaddq_s16(step2[62], step2[61]);
+  step1[63] = vqaddq_s16(step2[63], step2[60]);
+
+  // stage 6
+
+  btf_16_neon(step1[0], cospi[32], cospi[32], &step2[0], &step2[1]);
+  btf_16_lane_2_3_neon(step1[14], step1[9], c3, &step2[14], &step2[9]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[10]), vnegq_s16(step1[13]), c3,
+                       &step2[10], &step2[13]);
+  btf_16_lane_0_1_neon(step1[61], step1[34], c2, &step2[61], &step2[34]);
+  btf_16_lane_0_1_neon(step1[60], step1[35], c2, &step2[60], &step2[35]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[36]), vnegq_s16(step1[59]), c2,
+                       &step2[36], &step2[59]);
+  btf_16_lane_1_0_neon(vnegq_s16(step1[37]), vnegq_s16(step1[58]), c2,
+                       &step2[37], &step2[58]);
+  btf_16_lane_2_3_neon(step1[53], step1[42], c2, &step2[53], &step2[42]);
+  btf_16_lane_2_3_neon(step1[52], step1[43], c2, &step2[52], &step2[43]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[44]), vnegq_s16(step1[51]), c2,
+                       &step2[44], &step2[51]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[45]), vnegq_s16(step1[50]), c2,
+                       &step2[45], &step2[50]);
+
+  step2[4] = step1[4];
+  step2[5] = step1[4];
+  step2[6] = step1[7];
+  step2[7] = step1[7];
+  step2[8] = step1[8];
+  step2[11] = step1[11];
+  step2[12] = step1[12];
+  step2[15] = step1[15];
+  step2[16] = vqaddq_s16(step1[16], step1[19]);
+  step2[17] = vqaddq_s16(step1[17], step1[18]);
+  step2[18] = vqsubq_s16(step1[17], step1[18]);
+  step2[19] = vqsubq_s16(step1[16], step1[19]);
+  step2[20] = vqsubq_s16(step1[23], step1[20]);
+  step2[21] = vqsubq_s16(step1[22], step1[21]);
+  step2[22] = vqaddq_s16(step1[22], step1[21]);
+  step2[23] = vqaddq_s16(step1[23], step1[20]);
+  step2[24] = vqaddq_s16(step1[24], step1[27]);
+  step2[25] = vqaddq_s16(step1[25], step1[26]);
+  step2[26] = vqsubq_s16(step1[25], step1[26]);
+  step2[27] = vqsubq_s16(step1[24], step1[27]);
+  step2[28] = vqsubq_s16(step1[31], step1[28]);
+  step2[29] = vqsubq_s16(step1[30], step1[29]);
+  step2[30] = vqaddq_s16(step1[30], step1[29]);
+  step2[31] = vqaddq_s16(step1[31], step1[28]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[38] = step1[38];
+  step2[39] = step1[39];
+  step2[40] = step1[40];
+  step2[41] = step1[41];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[54] = step1[54];
+  step2[55] = step1[55];
+  step2[56] = step1[56];
+  step2[57] = step1[57];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 7
+
+  btf_16_lane_0_1_neon(step2[6], step2[5], c3, &step1[6], &step1[5]);
+  btf_16_lane_2_3_neon(step2[29], step2[18], c3, &step1[29], &step1[18]);
+  btf_16_lane_2_3_neon(step2[28], step2[19], c3, &step1[28], &step1[19]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[20]), vnegq_s16(step2[27]), c3,
+                       &step1[20], &step1[27]);
+  btf_16_lane_3_2_neon(vnegq_s16(step2[21]), vnegq_s16(step2[26]), c3,
+                       &step1[21], &step1[26]);
+
+  step1[0] = step2[0];
+  step1[1] = step2[1];
+  step1[2] = step2[1];
+  step1[3] = step2[0];
+  step1[4] = step2[4];
+  step1[7] = step2[7];
+  step1[8] = vqaddq_s16(step2[8], step2[11]);
+  step1[9] = vqaddq_s16(step2[9], step2[10]);
+  step1[10] = vqsubq_s16(step2[9], step2[10]);
+  step1[11] = vqsubq_s16(step2[8], step2[11]);
+  step1[12] = vqsubq_s16(step2[15], step2[12]);
+  step1[13] = vqsubq_s16(step2[14], step2[13]);
+  step1[14] = vqaddq_s16(step2[14], step2[13]);
+  step1[15] = vqaddq_s16(step2[15], step2[12]);
+  step1[16] = step2[16];
+  step1[17] = step2[17];
+  step1[22] = step2[22];
+  step1[23] = step2[23];
+  step1[24] = step2[24];
+  step1[25] = step2[25];
+  step1[30] = step2[30];
+  step1[31] = step2[31];
+  step1[32] = vqaddq_s16(step2[32], step2[39]);
+  step1[33] = vqaddq_s16(step2[33], step2[38]);
+  step1[34] = vqaddq_s16(step2[34], step2[37]);
+  step1[35] = vqaddq_s16(step2[35], step2[36]);
+  step1[36] = vqsubq_s16(step2[35], step2[36]);
+  step1[37] = vqsubq_s16(step2[34], step2[37]);
+  step1[38] = vqsubq_s16(step2[33], step2[38]);
+  step1[39] = vqsubq_s16(step2[32], step2[39]);
+  step1[40] = vqsubq_s16(step2[47], step2[40]);
+  step1[41] = vqsubq_s16(step2[46], step2[41]);
+  step1[42] = vqsubq_s16(step2[45], step2[42]);
+  step1[43] = vqsubq_s16(step2[44], step2[43]);
+  step1[44] = vqaddq_s16(step2[43], step2[44]);
+  step1[45] = vqaddq_s16(step2[42], step2[45]);
+  step1[46] = vqaddq_s16(step2[41], step2[46]);
+  step1[47] = vqaddq_s16(step2[40], step2[47]);
+  step1[48] = vqaddq_s16(step2[48], step2[55]);
+  step1[49] = vqaddq_s16(step2[49], step2[54]);
+  step1[50] = vqaddq_s16(step2[50], step2[53]);
+  step1[51] = vqaddq_s16(step2[51], step2[52]);
+  step1[52] = vqsubq_s16(step2[51], step2[52]);
+  step1[53] = vqsubq_s16(step2[50], step2[53]);
+  step1[54] = vqsubq_s16(step2[49], step2[54]);
+  step1[55] = vqsubq_s16(step2[48], step2[55]);
+  step1[56] = vqsubq_s16(step2[63], step2[56]);
+  step1[57] = vqsubq_s16(step2[62], step2[57]);
+  step1[58] = vqsubq_s16(step2[61], step2[58]);
+  step1[59] = vqsubq_s16(step2[60], step2[59]);
+  step1[60] = vqaddq_s16(step2[59], step2[60]);
+  step1[61] = vqaddq_s16(step2[58], step2[61]);
+  step1[62] = vqaddq_s16(step2[57], step2[62]);
+  step1[63] = vqaddq_s16(step2[56], step2[63]);
+
+  // stage 8
+
+  btf_16_lane_0_1_neon(step1[13], step1[10], c3, &step2[13], &step2[10]);
+  btf_16_lane_0_1_neon(step1[12], step1[11], c3, &step2[12], &step2[11]);
+  btf_16_lane_2_3_neon(step1[59], step1[36], c3, &step2[59], &step2[36]);
+  btf_16_lane_2_3_neon(step1[58], step1[37], c3, &step2[58], &step2[37]);
+  btf_16_lane_2_3_neon(step1[57], step1[38], c3, &step2[57], &step2[38]);
+  btf_16_lane_2_3_neon(step1[56], step1[39], c3, &step2[56], &step2[39]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[40]), vnegq_s16(step1[55]), c3,
+                       &step2[40], &step2[55]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[41]), vnegq_s16(step1[54]), c3,
+                       &step2[41], &step2[54]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[42]), vnegq_s16(step1[53]), c3,
+                       &step2[42], &step2[53]);
+  btf_16_lane_3_2_neon(vnegq_s16(step1[43]), vnegq_s16(step1[52]), c3,
+                       &step2[43], &step2[52]);
+
+  step2[0] = vqaddq_s16(step1[0], step1[7]);
+  step2[1] = vqaddq_s16(step1[1], step1[6]);
+  step2[2] = vqaddq_s16(step1[2], step1[5]);
+  step2[3] = vqaddq_s16(step1[3], step1[4]);
+  step2[4] = vqsubq_s16(step1[3], step1[4]);
+  step2[5] = vqsubq_s16(step1[2], step1[5]);
+  step2[6] = vqsubq_s16(step1[1], step1[6]);
+  step2[7] = vqsubq_s16(step1[0], step1[7]);
+  step2[8] = step1[8];
+  step2[9] = step1[9];
+  step2[14] = step1[14];
+  step2[15] = step1[15];
+  step2[16] = vqaddq_s16(step1[16], step1[23]);
+  step2[17] = vqaddq_s16(step1[17], step1[22]);
+  step2[18] = vqaddq_s16(step1[18], step1[21]);
+  step2[19] = vqaddq_s16(step1[19], step1[20]);
+  step2[20] = vqsubq_s16(step1[19], step1[20]);
+  step2[21] = vqsubq_s16(step1[18], step1[21]);
+  step2[22] = vqsubq_s16(step1[17], step1[22]);
+  step2[23] = vqsubq_s16(step1[16], step1[23]);
+  step2[24] = vqsubq_s16(step1[31], step1[24]);
+  step2[25] = vqsubq_s16(step1[30], step1[25]);
+  step2[26] = vqsubq_s16(step1[29], step1[26]);
+  step2[27] = vqsubq_s16(step1[28], step1[27]);
+  step2[28] = vqaddq_s16(step1[28], step1[27]);
+  step2[29] = vqaddq_s16(step1[29], step1[26]);
+  step2[30] = vqaddq_s16(step1[30], step1[25]);
+  step2[31] = vqaddq_s16(step1[31], step1[24]);
+  step2[32] = step1[32];
+  step2[33] = step1[33];
+  step2[34] = step1[34];
+  step2[35] = step1[35];
+  step2[44] = step1[44];
+  step2[45] = step1[45];
+  step2[46] = step1[46];
+  step2[47] = step1[47];
+  step2[48] = step1[48];
+  step2[49] = step1[49];
+  step2[50] = step1[50];
+  step2[51] = step1[51];
+  step2[60] = step1[60];
+  step2[61] = step1[61];
+  step2[62] = step1[62];
+  step2[63] = step1[63];
+
+  // stage 9
+  idct64_stage9_neon(step2, step1, cos_bit);
+
+  // stage 10
+  idct64_stage10_neon(step1, step2, cos_bit);
+
+  // stage 11
+
+  out[0] = vqaddq_s16(step2[0], step2[63]);
+  out[1] = vqaddq_s16(step2[1], step2[62]);
+  out[2] = vqaddq_s16(step2[2], step2[61]);
+  out[3] = vqaddq_s16(step2[3], step2[60]);
+  out[4] = vqaddq_s16(step2[4], step2[59]);
+  out[5] = vqaddq_s16(step2[5], step2[58]);
+  out[6] = vqaddq_s16(step2[6], step2[57]);
+  out[7] = vqaddq_s16(step2[7], step2[56]);
+  out[8] = vqaddq_s16(step2[8], step2[55]);
+  out[9] = vqaddq_s16(step2[9], step2[54]);
+  out[10] = vqaddq_s16(step2[10], step2[53]);
+  out[11] = vqaddq_s16(step2[11], step2[52]);
+  out[12] = vqaddq_s16(step2[12], step2[51]);
+  out[13] = vqaddq_s16(step2[13], step2[50]);
+  out[14] = vqaddq_s16(step2[14], step2[49]);
+  out[15] = vqaddq_s16(step2[15], step2[48]);
+  out[16] = vqaddq_s16(step2[16], step2[47]);
+  out[17] = vqaddq_s16(step2[17], step2[46]);
+  out[18] = vqaddq_s16(step2[18], step2[45]);
+  out[19] = vqaddq_s16(step2[19], step2[44]);
+  out[20] = vqaddq_s16(step2[20], step2[43]);
+  out[21] = vqaddq_s16(step2[21], step2[42]);
+  out[22] = vqaddq_s16(step2[22], step2[41]);
+  out[23] = vqaddq_s16(step2[23], step2[40]);
+  out[24] = vqaddq_s16(step2[24], step2[39]);
+  out[25] = vqaddq_s16(step2[25], step2[38]);
+  out[26] = vqaddq_s16(step2[26], step2[37]);
+  out[27] = vqaddq_s16(step2[27], step2[36]);
+  out[28] = vqaddq_s16(step2[28], step2[35]);
+  out[29] = vqaddq_s16(step2[29], step2[34]);
+  out[30] = vqaddq_s16(step2[30], step2[33]);
+  out[31] = vqaddq_s16(step2[31], step2[32]);
+  out[32] = vqsubq_s16(step2[31], step2[32]);
+  out[33] = vqsubq_s16(step2[30], step2[33]);
+  out[34] = vqsubq_s16(step2[29], step2[34]);
+  out[35] = vqsubq_s16(step2[28], step2[35]);
+  out[36] = vqsubq_s16(step2[27], step2[36]);
+  out[37] = vqsubq_s16(step2[26], step2[37]);
+  out[38] = vqsubq_s16(step2[25], step2[38]);
+  out[39] = vqsubq_s16(step2[24], step2[39]);
+  out[40] = vqsubq_s16(step2[23], step2[40]);
+  out[41] = vqsubq_s16(step2[22], step2[41]);
+  out[42] = vqsubq_s16(step2[21], step2[42]);
+  out[43] = vqsubq_s16(step2[20], step2[43]);
+  out[44] = vqsubq_s16(step2[19], step2[44]);
+  out[45] = vqsubq_s16(step2[18], step2[45]);
+  out[46] = vqsubq_s16(step2[17], step2[46]);
+  out[47] = vqsubq_s16(step2[16], step2[47]);
+  out[48] = vqsubq_s16(step2[15], step2[48]);
+  out[49] = vqsubq_s16(step2[14], step2[49]);
+  out[50] = vqsubq_s16(step2[13], step2[50]);
+  out[51] = vqsubq_s16(step2[12], step2[51]);
+  out[52] = vqsubq_s16(step2[11], step2[52]);
+  out[53] = vqsubq_s16(step2[10], step2[53]);
+  out[54] = vqsubq_s16(step2[9], step2[54]);
+  out[55] = vqsubq_s16(step2[8], step2[55]);
+  out[56] = vqsubq_s16(step2[7], step2[56]);
+  out[57] = vqsubq_s16(step2[6], step2[57]);
+  out[58] = vqsubq_s16(step2[5], step2[58]);
+  out[59] = vqsubq_s16(step2[4], step2[59]);
+  out[60] = vqsubq_s16(step2[3], step2[60]);
+  out[61] = vqsubq_s16(step2[2], step2[61]);
+  out[62] = vqsubq_s16(step2[1], step2[62]);
+  out[63] = vqsubq_s16(step2[0], step2[63]);
+}
 
 // Functions for blocks with eob at DC and within
 // topleft 8x8, 16x16, 32x32 corner
-static const transform_1d_neon
-    lowbd_txfm_all_1d_zeros_w8_arr[TX_SIZES][ITX_TYPES_1D][4] = {
-      {
-          { av1_idct4_new, av1_idct4_new, NULL, NULL },
-          { av1_iadst4_new, av1_iadst4_new, NULL, NULL },
-          { av1_iidentity4_c, av1_iidentity4_c, NULL, NULL },
-      },
-      { { av1_idct8_new, av1_idct8_new, NULL, NULL },
-        { av1_iadst8_new, av1_iadst8_new, NULL, NULL },
-        { av1_iidentity8_c, av1_iidentity8_c, NULL, NULL } },
-      {
-          { av1_idct16_new, av1_idct16_new, av1_idct16_new, NULL },
-          { av1_iadst16_new, av1_iadst16_new, av1_iadst16_new, NULL },
-          { av1_iidentity16_c, av1_iidentity16_c, av1_iidentity16_c, NULL },
-      },
-      { { av1_idct32_new, av1_idct32_new, av1_idct32_new, av1_idct32_new },
-        { NULL, NULL, NULL, NULL },
-        { av1_iidentity32_c, av1_iidentity32_c, av1_iidentity32_c,
-          av1_iidentity32_c } },
-      { { av1_idct64_new, av1_idct64_new, av1_idct64_new, av1_idct64_new },
-        { NULL, NULL, NULL, NULL },
-        { NULL, NULL, NULL, NULL } }
-    };
-
 static const transform_neon
     lowbd_txfm_all_1d_zeros_w_arr[TX_SIZES][ITX_TYPES_1D][4] = {
       {
@@ -2135,80 +3672,12 @@ static const transform_neon
         { NULL, NULL, NULL, NULL },
         { identity32_new_neon, identity32_new_neon, identity32_new_neon,
           identity32_new_neon } },
-      { { NULL, NULL, NULL, NULL },
+      { { idct64_low1_new_neon, idct64_low8_new_neon, idct64_low16_new_neon,
+          idct64_low32_new_neon },
         { NULL, NULL, NULL, NULL },
         { NULL, NULL, NULL, NULL } }
     };
 
-static INLINE void lowbd_inv_txfm2d_add_wxh_idtx_neon(
-    const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
-    TX_SIZE tx_size, int eob) {
-  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
-  int32_t *temp_in = txfm_buf;
-
-  int eobx, eoby;
-  get_eobx_eoby_scan_default(&eobx, &eoby, tx_size, eob);
-  const int8_t *shift = inv_txfm_shift_ls[tx_size];
-  const int txw_idx = get_txw_idx(tx_size);
-  const int txh_idx = get_txh_idx(tx_size);
-  const int cos_bit_col = inv_cos_bit_col[txw_idx][txh_idx];
-  const int cos_bit_row = inv_cos_bit_row[txw_idx][txh_idx];
-  const int txfm_size_col = tx_size_wide[tx_size];
-  const int txfm_size_row = tx_size_high[tx_size];
-  const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
-
-  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
-  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
-
-  int32_t *temp_out = temp_in + buf_offset;
-  int32_t *buf = temp_out + buf_offset;
-  int32_t *buf_ptr = buf;
-  const int8_t stage_range[MAX_TXFM_STAGE_NUM] = { 16 };
-  int r, bd = 8;
-
-  const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
-  const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
-  const transform_1d_neon row_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][fun_idx_x];
-  const transform_1d_neon col_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][fun_idx_y];
-
-  assert(col_txfm != NULL);
-  assert(row_txfm != NULL);
-
-  // row tx
-  int row_start = (buf_size_nonzero_h_div8 * 8);
-  for (int i = 0; i < row_start; i++) {
-    if (abs(rect_type) == 1) {
-      for (int j = 0; j < txfm_size_col; j++)
-        temp_in[j] = round_shift((int64_t)input[j] * NewInvSqrt2, NewSqrt2Bits);
-      row_txfm(temp_in, buf_ptr, cos_bit_row, stage_range);
-    } else {
-      row_txfm(input, buf_ptr, cos_bit_row, stage_range);
-    }
-    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
-    input += txfm_size_col;
-    buf_ptr += txfm_size_col;
-  }
-
-  // Doing memset for the rows which are not processed in row transform.
-  memset(buf_ptr, 0,
-         sizeof(int32_t) * txfm_size_col * (txfm_size_row - row_start));
-
-  // col tx
-  for (int c = 0; c < txfm_size_col; c++) {
-    for (r = 0; r < txfm_size_row; ++r) temp_in[r] = buf[r * txfm_size_col + c];
-
-    col_txfm(temp_in, temp_out, cos_bit_col, stage_range);
-    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
-
-    for (r = 0; r < txfm_size_row; ++r) {
-      output[r * stride + c] =
-          highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
-    }
-  }
-}
-
 static INLINE void lowbd_inv_txfm2d_add_idtx_neon(const int32_t *input,
                                                   uint8_t *output, int stride,
                                                   TX_TYPE tx_type,
@@ -2281,90 +3750,6 @@ static INLINE void lowbd_inv_txfm2d_add_idtx_neon(const int32_t *input,
   }
 }
 
-static INLINE void lowbd_inv_txfm2d_add_v_wxh_identity_neon(
-    const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
-    TX_SIZE tx_size, int eob) {
-  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
-  int32_t *temp_in = txfm_buf;
-
-  int eobx, eoby;
-  get_eobx_eoby_scan_v_identity(&eobx, &eoby, tx_size, eob);
-  const int8_t *shift = inv_txfm_shift_ls[tx_size];
-  const int txw_idx = get_txw_idx(tx_size);
-  const int txh_idx = get_txh_idx(tx_size);
-  const int cos_bit_col = inv_cos_bit_col[txw_idx][txh_idx];
-  const int cos_bit_row = inv_cos_bit_row[txw_idx][txh_idx];
-  const int txfm_size_col = tx_size_wide[tx_size];
-  const int txfm_size_row = tx_size_high[tx_size];
-  const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
-
-  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
-  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
-
-  int32_t *temp_out = temp_in + buf_offset;
-  int32_t *buf = temp_out + buf_offset;
-  int32_t *buf_ptr = buf;
-  const int8_t stage_range[MAX_TXFM_STAGE_NUM] = { 16 };
-  int r, bd = 8;
-
-  const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
-  const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
-  const transform_1d_neon row_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][fun_idx_x];
-  const transform_1d_neon col_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][fun_idx_y];
-
-  assert(col_txfm != NULL);
-  assert(row_txfm != NULL);
-  int ud_flip, lr_flip;
-  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
-
-  // row tx
-  int row_start = (buf_size_nonzero_h_div8 * 8);
-  for (int i = 0; i < row_start; i++) {
-    if (abs(rect_type) == 1) {
-      for (int j = 0; j < txfm_size_col; j++)
-        temp_in[j] = round_shift((int64_t)input[j] * NewInvSqrt2, NewSqrt2Bits);
-      row_txfm(temp_in, buf_ptr, cos_bit_row, stage_range);
-    } else {
-      row_txfm(input, buf_ptr, cos_bit_row, stage_range);
-    }
-    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
-    input += txfm_size_col;
-    buf_ptr += txfm_size_col;
-  }
-  // Doing memset for the rows which are not processed in row transform.
-  memset(buf_ptr, 0,
-         sizeof(int32_t) * txfm_size_col * (txfm_size_row - row_start));
-
-  // col tx
-  for (int c = 0; c < txfm_size_col; c++) {
-    if (lr_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + c];
-    } else {
-      // flip left right
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
-    }
-    col_txfm(temp_in, temp_out, cos_bit_col, stage_range);
-    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
-
-    if (ud_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] =
-            highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
-      }
-    } else {
-      // flip upside down
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] = highbd_clip_pixel_add(
-            output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
-      }
-    }
-  }
-}
-
 static INLINE void lowbd_inv_txfm2d_add_v_identity_neon(
     const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
     TX_SIZE tx_size, int eob) {
@@ -2446,90 +3831,6 @@ static INLINE void lowbd_inv_txfm2d_add_v_identity_neon(
   }
 }
 
-static INLINE void lowbd_inv_txfm2d_add_h_wxh_identity_neon(
-    const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
-    TX_SIZE tx_size, int eob) {
-  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
-  int32_t *temp_in = txfm_buf;
-
-  int eobx, eoby;
-  get_eobx_eoby_scan_h_identity(&eobx, &eoby, tx_size, eob);
-  const int8_t *shift = inv_txfm_shift_ls[tx_size];
-  const int txw_idx = get_txw_idx(tx_size);
-  const int txh_idx = get_txh_idx(tx_size);
-  const int cos_bit_col = inv_cos_bit_col[txw_idx][txh_idx];
-  const int cos_bit_row = inv_cos_bit_row[txw_idx][txh_idx];
-  const int txfm_size_col = tx_size_wide[tx_size];
-  const int txfm_size_row = tx_size_high[tx_size];
-  const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
-
-  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
-  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
-
-  int32_t *temp_out = temp_in + buf_offset;
-  int32_t *buf = temp_out + buf_offset;
-  int32_t *buf_ptr = buf;
-  const int8_t stage_range[MAX_TXFM_STAGE_NUM] = { 16 };
-  int r, bd = 8;
-
-  const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
-  const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
-  const transform_1d_neon row_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][fun_idx_x];
-  const transform_1d_neon col_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][fun_idx_y];
-
-  assert(col_txfm != NULL);
-  assert(row_txfm != NULL);
-  int ud_flip, lr_flip;
-  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
-
-  // row tx
-  int row_start = (buf_size_nonzero_h_div8 * 8);
-  for (int i = 0; i < row_start; i++) {
-    if (abs(rect_type) == 1) {
-      for (int j = 0; j < txfm_size_col; j++)
-        temp_in[j] = round_shift((int64_t)input[j] * NewInvSqrt2, NewSqrt2Bits);
-      row_txfm(temp_in, buf_ptr, cos_bit_row, stage_range);
-    } else {
-      row_txfm(input, buf_ptr, cos_bit_row, stage_range);
-    }
-    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
-    input += txfm_size_col;
-    buf_ptr += txfm_size_col;
-  }
-  // Doing memset for the rows which are not processed in row transform.
-  memset(buf_ptr, 0,
-         sizeof(int32_t) * txfm_size_col * (txfm_size_row - row_start));
-
-  // col tx
-  for (int c = 0; c < txfm_size_col; c++) {
-    if (lr_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + c];
-    } else {
-      // flip left right
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
-    }
-    col_txfm(temp_in, temp_out, cos_bit_col, stage_range);
-    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
-
-    if (ud_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] =
-            highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
-      }
-    } else {
-      // flip upside down
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] = highbd_clip_pixel_add(
-            output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
-      }
-    }
-  }
-}
-
 static INLINE void lowbd_inv_txfm2d_add_h_identity_neon(
     const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
     TX_SIZE tx_size, int eob) {
@@ -2604,9 +3905,9 @@ static INLINE void lowbd_inv_txfm2d_add_h_identity_neon(
 
 static INLINE void lowbd_inv_txfm2d_add_4x4_neon(const int32_t *input,
                                                  uint8_t *output, int stride,
-                                                 TX_TYPE tx_type,
-                                                 TX_SIZE tx_size, int eob) {
+                                                 TX_TYPE tx_type, int eob) {
   (void)eob;
+  TX_SIZE tx_size = TX_4X4;
   DECLARE_ALIGNED(32, int, txfm_buf[4 * 4 + 8 + 8]);
   int32_t *temp_in = txfm_buf;
 
@@ -2666,9 +3967,9 @@ static INLINE void lowbd_inv_txfm2d_add_4x4_neon(const int32_t *input,
 }
 
 void lowbd_inv_txfm2d_add_4x8_neon(const int32_t *input, uint8_t *output,
-                                   int stride, TX_TYPE tx_type, TX_SIZE tx_size,
-                                   int eob) {
+                                   int stride, TX_TYPE tx_type, int eob) {
   (void)eob;
+  TX_SIZE tx_size = TX_4X8;
   DECLARE_ALIGNED(32, int, txfm_buf[4 * 8 + 8 + 8]);
   int32_t *temp_in = txfm_buf;
 
@@ -2730,9 +4031,9 @@ void lowbd_inv_txfm2d_add_4x8_neon(const int32_t *input, uint8_t *output,
 }
 
 void lowbd_inv_txfm2d_add_8x4_neon(const int32_t *input, uint8_t *output,
-                                   int stride, TX_TYPE tx_type, TX_SIZE tx_size,
-                                   int eob) {
+                                   int stride, TX_TYPE tx_type, int eob) {
   (void)eob;
+  TX_SIZE tx_size = TX_8X4;
   DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
   int32_t *temp_in = txfm_buf;
 
@@ -2794,9 +4095,9 @@ void lowbd_inv_txfm2d_add_8x4_neon(const int32_t *input, uint8_t *output,
 }
 
 void lowbd_inv_txfm2d_add_4x16_neon(const int32_t *input, uint8_t *output,
-                                    int stride, TX_TYPE tx_type,
-                                    TX_SIZE tx_size, int eob) {
+                                    int stride, TX_TYPE tx_type, int eob) {
   (void)eob;
+  TX_SIZE tx_size = TX_4X16;
   DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
   int32_t *temp_in = txfm_buf;
 
@@ -2856,10 +4157,9 @@ void lowbd_inv_txfm2d_add_4x16_neon(const int32_t *input, uint8_t *output,
 }
 
 void lowbd_inv_txfm2d_add_16x4_neon(const int32_t *input, uint8_t *output,
-                                    int stride, TX_TYPE tx_type,
-                                    TX_SIZE tx_size, int eob) {
+                                    int stride, TX_TYPE tx_type, int eob) {
   (void)eob;
-
+  TX_SIZE tx_size = TX_16X4;
   DECLARE_ALIGNED(32, int, txfm_buf[16 * 4 + 16 + 16]);
   int32_t *temp_in = txfm_buf;
 
@@ -2918,89 +4218,6 @@ void lowbd_inv_txfm2d_add_16x4_neon(const int32_t *input, uint8_t *output,
   }
 }
 
-static INLINE void lowbd_inv_txfm2d_add_wxh_no_identity_neon(
-    const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
-    TX_SIZE tx_size, int eob) {
-  DECLARE_ALIGNED(32, int, txfm_buf[64 * 64 + 64 + 64]);
-  int32_t *temp_in = txfm_buf;
-
-  int eobx, eoby, ud_flip, lr_flip, row_start;
-  get_eobx_eoby_scan_default(&eobx, &eoby, tx_size, eob);
-  const int8_t *shift = inv_txfm_shift_ls[tx_size];
-  const int txw_idx = get_txw_idx(tx_size);
-  const int txh_idx = get_txh_idx(tx_size);
-  const int cos_bit_col = inv_cos_bit_col[txw_idx][txh_idx];
-  const int cos_bit_row = inv_cos_bit_row[txw_idx][txh_idx];
-  const int txfm_size_col = tx_size_wide[tx_size];
-  const int txfm_size_row = tx_size_high[tx_size];
-  const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
-  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
-  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
-
-  int32_t *temp_out = temp_in + buf_offset;
-  int32_t *buf = temp_out + buf_offset;
-  int32_t *buf_ptr = buf;
-  const int8_t stage_range[MAX_TXFM_STAGE_NUM] = { 16 };
-  const int bd = 8;
-  int r;
-
-  const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
-  const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
-  const transform_1d_neon row_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][fun_idx_x];
-  const transform_1d_neon col_txfm =
-      lowbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][fun_idx_y];
-
-  assert(col_txfm != NULL);
-  assert(row_txfm != NULL);
-
-  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
-  row_start = (buf_size_nonzero_h_div8 << 3);
-
-  for (int i = 0; i < row_start; i++) {
-    if (abs(rect_type) == 1) {
-      for (int j = 0; j < txfm_size_col; j++)
-        temp_in[j] = round_shift((int64_t)input[j] * NewInvSqrt2, NewSqrt2Bits);
-      row_txfm(temp_in, buf_ptr, cos_bit_row, stage_range);
-    } else {
-      row_txfm(input, buf_ptr, cos_bit_row, stage_range);
-    }
-    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
-    input += txfm_size_col;
-    buf_ptr += txfm_size_col;
-  }
-
-  // Doing memset for the rows which are not processed in row transform.
-  memset(buf_ptr, 0,
-         sizeof(int32_t) * txfm_size_col * (txfm_size_row - row_start));
-
-  for (int c = 0; c < txfm_size_col; c++) {
-    if (lr_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + c];
-    } else {
-      // flip left right
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
-    }
-    col_txfm(temp_in, temp_out, cos_bit_col, stage_range);
-    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
-
-    if (ud_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] =
-            highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
-      }
-    } else {
-      // flip upside down
-      for (r = 0; r < txfm_size_row; ++r) {
-        output[r * stride + c] = highbd_clip_pixel_add(
-            output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
-      }
-    }
-  }
-}
-
 static INLINE void lowbd_inv_txfm2d_add_no_identity_neon(
     const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
     TX_SIZE tx_size, int eob) {
@@ -3019,6 +4236,7 @@ static INLINE void lowbd_inv_txfm2d_add_no_identity_neon(
   const int buf_size_w_div8 = txfm_size_col >> 3;
   const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
   const int buf_size_nonzero_w_div8 = (eobx + 8) >> 3;
+  const int input_stride = AOMMIN(32, txfm_size_col);
   const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
   const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
   const int32_t *input_1;
@@ -3038,14 +4256,14 @@ static INLINE void lowbd_inv_txfm2d_add_no_identity_neon(
     input_1 = input;
     for (int j = 0; j < buf_size_nonzero_w_div8; ++j) {
       int k = j * 8 + i * txfm_size_col;
-      load_buffer_32bit_to_16bit_neon(input_1, &a[k], txfm_size_col);
+      load_buffer_32bit_to_16bit_neon(input_1, &a[k], input_stride);
       transpose_s16_8x8q(&a[k], &a[k]);
       input_1 += 8;
     }
-    input += (txfm_size_col * 8);
+    input += (input_stride * 8);
     if (abs(rect_type) == 1) {
       int y = i * txfm_size_col;
-      round_shift_for_rect(&a[y], &a[y], txfm_size_col);
+      round_shift_for_rect(&a[y], &a[y], input_stride);
     }
     row_txfm(&a[i * txfm_size_col], &a[i * txfm_size_col], cos_bit_row, 0);
     av1_round_shift_array_16_neon(&a[i * txfm_size_col], txfm_size_col,
@@ -3083,36 +4301,6 @@ static INLINE void lowbd_inv_txfm2d_add_no_identity_neon(
   }
 }
 
-static INLINE void lowbd_inv_txfm2d_add_wxh_universe_neon(
-    const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
-    TX_SIZE tx_size, int eob) {
-  switch (tx_type) {
-    case IDTX:
-      lowbd_inv_txfm2d_add_wxh_idtx_neon(input, output, stride, tx_type,
-                                         tx_size, eob);
-      break;
-
-    case H_DCT:
-    case H_ADST:
-    case H_FLIPADST:
-      lowbd_inv_txfm2d_add_v_wxh_identity_neon(input, output, stride, tx_type,
-                                               tx_size, eob);
-      break;
-
-    case V_DCT:
-    case V_ADST:
-    case V_FLIPADST:
-      lowbd_inv_txfm2d_add_h_wxh_identity_neon(input, output, stride, tx_type,
-                                               tx_size, eob);
-      break;
-
-    default:
-      lowbd_inv_txfm2d_add_wxh_no_identity_neon(input, output, stride, tx_type,
-                                                tx_size, eob);
-      break;
-  }
-}
-
 static INLINE void lowbd_inv_txfm2d_add_universe_neon(
     const int32_t *input, uint8_t *output, int stride, TX_TYPE tx_type,
     TX_SIZE tx_size, int eob) {
@@ -3146,73 +4334,27 @@ static INLINE void lowbd_inv_txfm2d_add_universe_neon(
 void av1_lowbd_inv_txfm2d_add_neon(const int32_t *input, uint8_t *output,
                                    int stride, TX_TYPE tx_type, TX_SIZE tx_size,
                                    int eob) {
-  int row;
   switch (tx_size) {
     case TX_4X4:
-      lowbd_inv_txfm2d_add_4x4_neon(input, output, stride, tx_type, tx_size,
-                                    eob);
+      lowbd_inv_txfm2d_add_4x4_neon(input, output, stride, tx_type, eob);
       break;
 
     case TX_4X8:
-      lowbd_inv_txfm2d_add_4x8_neon(input, output, stride, tx_type, tx_size,
-                                    eob);
+      lowbd_inv_txfm2d_add_4x8_neon(input, output, stride, tx_type, eob);
       break;
 
     case TX_8X4:
-      lowbd_inv_txfm2d_add_8x4_neon(input, output, stride, tx_type, tx_size,
-                                    eob);
+      lowbd_inv_txfm2d_add_8x4_neon(input, output, stride, tx_type, eob);
       break;
 
     case TX_4X16:
-      lowbd_inv_txfm2d_add_4x16_neon(input, output, stride, tx_type, tx_size,
-                                     eob);
+      lowbd_inv_txfm2d_add_4x16_neon(input, output, stride, tx_type, eob);
       break;
 
     case TX_16X4:
-      lowbd_inv_txfm2d_add_16x4_neon(input, output, stride, tx_type, tx_size,
-                                     eob);
+      lowbd_inv_txfm2d_add_16x4_neon(input, output, stride, tx_type, eob);
       break;
 
-    case TX_16X64: {
-      lowbd_inv_txfm2d_add_wxh_universe_neon(input, output, stride, tx_type,
-                                             tx_size, eob);
-    } break;
-
-    case TX_64X16: {
-      int32_t mod_input[64 * 16];
-      for (row = 0; row < 16; ++row) {
-        memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
-        memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
-      }
-      lowbd_inv_txfm2d_add_wxh_universe_neon(mod_input, output, stride, tx_type,
-                                             tx_size, eob);
-    } break;
-
-    case TX_32X64: {
-      lowbd_inv_txfm2d_add_wxh_universe_neon(input, output, stride, tx_type,
-                                             tx_size, eob);
-    } break;
-
-    case TX_64X32: {
-      int32_t mod_input[64 * 32];
-      for (row = 0; row < 32; ++row) {
-        memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
-        memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
-      }
-      lowbd_inv_txfm2d_add_wxh_universe_neon(mod_input, output, stride, tx_type,
-                                             tx_size, eob);
-    } break;
-
-    case TX_64X64: {
-      int32_t mod_input[64 * 64];
-      for (row = 0; row < 32; ++row) {
-        memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
-        memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
-      }
-      lowbd_inv_txfm2d_add_wxh_universe_neon(mod_input, output, stride, tx_type,
-                                             tx_size, eob);
-    } break;
-
     default:
       lowbd_inv_txfm2d_add_universe_neon(input, output, stride, tx_type,
                                          tx_size, eob);
diff --git a/third_party/aom/av1/common/arm/av1_txfm_neon.c b/third_party/aom/av1/common/arm/av1_txfm_neon.c
index de3c547248..7e3a05ab7e 100644
--- a/third_party/aom/av1/common/arm/av1_txfm_neon.c
+++ b/third_party/aom/av1/common/arm/av1_txfm_neon.c
@@ -12,6 +12,8 @@
 #include <arm_neon.h>
 #include <assert.h>
 
+#include "config/av1_rtcd.h"
+
 #include "aom_ports/mem.h"
 #include "av1/common/arm/mem_neon.h"
 
diff --git a/third_party/aom/av1/common/arm/jnt_convolve_neon.c b/third_party/aom/av1/common/arm/jnt_convolve_neon.c
index e5674ef7c2..379ff9899d 100644
--- a/third_party/aom/av1/common/arm/jnt_convolve_neon.c
+++ b/third_party/aom/av1/common/arm/jnt_convolve_neon.c
@@ -23,19 +23,17 @@
 #include "av1/common/arm/transpose_neon.h"
 
 #if !defined(__aarch64__)
-static INLINE void compute_avg_4x1(uint16x4_t res0, uint16x4_t d0,
-                                   const uint16_t fwd_offset,
-                                   const uint16_t bck_offset,
-                                   const int16x4_t sub_const_vec,
-                                   const int16_t round_bits,
-                                   const int use_jnt_comp_avg, uint8x8_t *t0) {
+static INLINE void compute_avg_4x1(
+    uint16x4_t res0, uint16x4_t d0, const uint16_t fwd_offset,
+    const uint16_t bck_offset, const int16x4_t sub_const_vec,
+    const int16_t round_bits, const int use_dist_wtd_comp_avg, uint8x8_t *t0) {
   int16x4_t tmp0;
   uint16x4_t tmp_u0;
   uint32x4_t sum0;
   int32x4_t dst0;
   int16x8_t tmp4;
 
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const int32x4_t round_bits_vec = vdupq_n_s32((int32_t)(-round_bits));
 
     sum0 = vmull_n_u16(res0, fwd_offset);
@@ -65,12 +63,10 @@ static INLINE void compute_avg_4x1(uint16x4_t res0, uint16x4_t d0,
   }
 }
 
-static INLINE void compute_avg_8x1(uint16x8_t res0, uint16x8_t d0,
-                                   const uint16_t fwd_offset,
-                                   const uint16_t bck_offset,
-                                   const int16x4_t sub_const,
-                                   const int16_t round_bits,
-                                   const int use_jnt_comp_avg, uint8x8_t *t0) {
+static INLINE void compute_avg_8x1(
+    uint16x8_t res0, uint16x8_t d0, const uint16_t fwd_offset,
+    const uint16_t bck_offset, const int16x4_t sub_const,
+    const int16_t round_bits, const int use_dist_wtd_comp_avg, uint8x8_t *t0) {
   int16x4_t tmp0, tmp2;
   int16x8_t f0;
   uint32x4_t sum0, sum2;
@@ -78,7 +74,7 @@ static INLINE void compute_avg_8x1(uint16x8_t res0, uint16x8_t d0,
 
   uint16x8_t tmp_u0;
 
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const int32x4_t sub_const_vec = vmovl_s16(sub_const);
     const int32x4_t round_bits_vec = vdupq_n_s32(-(int32_t)round_bits);
 
@@ -123,7 +119,7 @@ static INLINE void compute_avg_4x4(
     uint16x4_t d0, uint16x4_t d1, uint16x4_t d2, uint16x4_t d3,
     const uint16_t fwd_offset, const uint16_t bck_offset,
     const int16x4_t sub_const_vec, const int16_t round_bits,
-    const int use_jnt_comp_avg, uint8x8_t *t0, uint8x8_t *t1) {
+    const int use_dist_wtd_comp_avg, uint8x8_t *t0, uint8x8_t *t1) {
   int16x4_t tmp0, tmp1, tmp2, tmp3;
   uint16x4_t tmp_u0, tmp_u1, tmp_u2, tmp_u3;
   uint32x4_t sum0, sum1, sum2, sum3;
@@ -132,7 +128,7 @@ static INLINE void compute_avg_4x4(
   int16x8_t tmp4, tmp5;
   const int16x8_t zero = vdupq_n_s16(0);
 
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const int32x4_t round_bits_vec = vdupq_n_s32((int32_t)(-round_bits));
     const int32x4_t const_vec = vmovl_s16(sub_const_vec);
 
@@ -203,8 +199,8 @@ static INLINE void compute_avg_8x4(
     uint16x8_t d0, uint16x8_t d1, uint16x8_t d2, uint16x8_t d3,
     const uint16_t fwd_offset, const uint16_t bck_offset,
     const int16x4_t sub_const, const int16_t round_bits,
-    const int use_jnt_comp_avg, uint8x8_t *t0, uint8x8_t *t1, uint8x8_t *t2,
-    uint8x8_t *t3) {
+    const int use_dist_wtd_comp_avg, uint8x8_t *t0, uint8x8_t *t1,
+    uint8x8_t *t2, uint8x8_t *t3) {
   int16x4_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   int16x8_t f0, f1, f2, f3;
   uint32x4_t sum0, sum1, sum2, sum3;
@@ -214,7 +210,7 @@ static INLINE void compute_avg_8x4(
   uint16x8_t tmp_u0, tmp_u1, tmp_u2, tmp_u3;
   const int16x8_t zero = vdupq_n_s16(0);
 
-  if (use_jnt_comp_avg) {
+  if (use_dist_wtd_comp_avg) {
     const int32x4_t sub_const_vec = vmovl_s16(sub_const);
     const int32x4_t round_bits_vec = vdupq_n_s32(-(int32_t)round_bits);
 
@@ -319,7 +315,7 @@ static INLINE void compute_avg_8x4(
   }
 }
 
-static INLINE void jnt_convolve_2d_horiz_neon(
+static INLINE void dist_wtd_convolve_2d_horiz_neon(
     const uint8_t *src, int src_stride, int16_t *im_block, const int im_stride,
     int16_t *x_filter_tmp, const int im_h, int w, const int round_0) {
   const int bd = 8;
@@ -563,7 +559,7 @@ static INLINE void jnt_convolve_2d_horiz_neon(
   }
 }
 
-static INLINE void jnt_convolve_2d_vert_neon(
+static INLINE void dist_wtd_convolve_2d_vert_neon(
     int16_t *im_block, const int im_stride, uint8_t *dst8, int dst8_stride,
     ConvolveParams *conv_params, const int16_t *y_filter, int h, int w) {
   uint8_t *dst_u8_ptr, *d_u8;
@@ -587,7 +583,7 @@ static INLINE void jnt_convolve_2d_vert_neon(
   const uint16_t fwd_offset = conv_params->fwd_offset;
   const uint16_t bck_offset = conv_params->bck_offset;
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
 
   int16x4_t s0, s1, s2, s3, s4, s5, s6, s7;
   uint16x4_t res4, d0;
@@ -652,8 +648,8 @@ static INLINE void jnt_convolve_2d_vert_neon(
         d += (dst_stride << 2);
 
         compute_avg_4x4(res4, res5, res6, res7, d0, d1, d2, d3, fwd_offset,
-                        bck_offset, sub_const_vec, round_bits, use_jnt_comp_avg,
-                        &t0, &t1);
+                        bck_offset, sub_const_vec, round_bits,
+                        use_dist_wtd_comp_avg, &t0, &t1);
 
         vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0), 0);
         d_u8 += dst8_stride;
@@ -691,7 +687,7 @@ static INLINE void jnt_convolve_2d_vert_neon(
         d += (dst_stride);
 
         compute_avg_4x1(res4, d0, fwd_offset, bck_offset, sub_const_vec,
-                        round_bits, use_jnt_comp_avg, &t0);
+                        round_bits, use_dist_wtd_comp_avg, &t0);
 
         vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0), 0);
         d_u8 += dst8_stride;
@@ -717,12 +713,12 @@ static INLINE void jnt_convolve_2d_vert_neon(
   } while (w > 0);
 }
 
-void av1_jnt_convolve_2d_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
-                              int dst8_stride, int w, int h,
-                              const InterpFilterParams *filter_params_x,
-                              const InterpFilterParams *filter_params_y,
-                              const int subpel_x_q4, const int subpel_y_q4,
-                              ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_neon(const uint8_t *src, int src_stride,
+                                   uint8_t *dst8, int dst8_stride, int w, int h,
+                                   const InterpFilterParams *filter_params_x,
+                                   const InterpFilterParams *filter_params_y,
+                                   const int subpel_x_q4, const int subpel_y_q4,
+                                   ConvolveParams *conv_params) {
   assert(!(w % 4));
   assert(!(h % 4));
 
@@ -748,19 +744,18 @@ void av1_jnt_convolve_2d_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
   filter_x_coef = vshrq_n_s16(filter_x_coef, 1);
   vst1q_s16(&x_filter_tmp[0], filter_x_coef);
 
-  jnt_convolve_2d_horiz_neon(src_ptr, src_stride, im_block, im_stride,
-                             x_filter_tmp, im_h, w, round_0);
+  dist_wtd_convolve_2d_horiz_neon(src_ptr, src_stride, im_block, im_stride,
+                                  x_filter_tmp, im_h, w, round_0);
 
-  jnt_convolve_2d_vert_neon(im_block, im_stride, dst8, dst8_stride, conv_params,
-                            y_filter, h, w);
+  dist_wtd_convolve_2d_vert_neon(im_block, im_stride, dst8, dst8_stride,
+                                 conv_params, y_filter, h, w);
 }
 
-void av1_jnt_convolve_2d_copy_neon(const uint8_t *src, int src_stride,
-                                   uint8_t *dst8, int dst8_stride, int w, int h,
-                                   const InterpFilterParams *filter_params_x,
-                                   const InterpFilterParams *filter_params_y,
-                                   const int subpel_x_q4, const int subpel_y_q4,
-                                   ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_copy_neon(
+    const uint8_t *src, int src_stride, uint8_t *dst8, int dst8_stride, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params) {
   uint8x8_t res0_8, res1_8, res2_8, res3_8, tmp_shift0, tmp_shift1, tmp_shift2,
       tmp_shift3;
   uint16x8_t res_q0, res_q1, res_q2, res_q3, tmp_q0, tmp_q1, tmp_q2, tmp_q3;
@@ -811,7 +806,7 @@ void av1_jnt_convolve_2d_copy_neon(const uint8_t *src, int src_stride,
           compute_avg_8x4(tmp_q0, tmp_q1, tmp_q2, tmp_q3, res_q0, res_q1,
                           res_q2, res_q3, conv_params->fwd_offset,
                           conv_params->bck_offset, sub_const_vec, bits,
-                          conv_params->use_jnt_comp_avg, &tmp_shift0,
+                          conv_params->use_dist_wtd_comp_avg, &tmp_shift0,
                           &tmp_shift1, &tmp_shift2, &tmp_shift3);
 
           vst1_u8(dst8_1 + (0 * dst8_stride), tmp_shift0);
@@ -854,7 +849,7 @@ void av1_jnt_convolve_2d_copy_neon(const uint8_t *src, int src_stride,
 
         compute_avg_4x4(tmp4, tmp5, tmp6, tmp7, res4, res5, res6, res7,
                         conv_params->fwd_offset, conv_params->bck_offset,
-                        sub_const_vec, bits, conv_params->use_jnt_comp_avg,
+                        sub_const_vec, bits, conv_params->use_dist_wtd_comp_avg,
                         &tmp_shift0, &tmp_shift1);
 
         vst1_lane_u32((uint32_t *)(dst8_1), vreinterpret_u32_u8(tmp_shift0), 0);
@@ -881,12 +876,12 @@ void av1_jnt_convolve_2d_copy_neon(const uint8_t *src, int src_stride,
   }
 }
 
-void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
-                             int dst8_stride, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_x_neon(const uint8_t *src, int src_stride,
+                                  uint8_t *dst8, int dst8_stride, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   assert(!(w % 4));
   assert(!(h % 4));
 
@@ -902,7 +897,7 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
       2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const uint16_t fwd_offset = conv_params->fwd_offset;
   const uint16_t bck_offset = conv_params->bck_offset;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
 
   (void)filter_params_y;
   (void)subpel_y_q4;
@@ -1031,8 +1026,8 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           compute_avg_4x4(res4, res5, res6, res7, vreinterpret_u16_s16(d0),
                           vreinterpret_u16_s16(d1), vreinterpret_u16_s16(d2),
                           vreinterpret_u16_s16(d3), fwd_offset, bck_offset,
-                          round_offset_vec, round_bits, use_jnt_comp_avg, &t0,
-                          &t1);
+                          round_offset_vec, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1);
 
           vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0),
                         0);  // 00 01 02 03
@@ -1103,7 +1098,7 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
 
           compute_avg_4x1(res4, vreinterpret_u16_s16(d0), fwd_offset,
                           bck_offset, round_offset_vec, round_bits,
-                          use_jnt_comp_avg, &t0);
+                          use_dist_wtd_comp_avg, &t0);
 
           vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0),
                         0);  // 00 01 02 03
@@ -1231,11 +1226,12 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           load_u16_8x4(d_tmp, dst_stride, &res8, &res9, &res10, &res11);
           d_tmp += (dst_stride << 2);
 
-          compute_avg_8x4(
-              res8, res9, res10, res11, vreinterpretq_u16_s16(res0),
-              vreinterpretq_u16_s16(res1), vreinterpretq_u16_s16(res2),
-              vreinterpretq_u16_s16(res3), fwd_offset, bck_offset,
-              round_offset64, round_bits, use_jnt_comp_avg, &t0, &t1, &t2, &t3);
+          compute_avg_8x4(res8, res9, res10, res11, vreinterpretq_u16_s16(res0),
+                          vreinterpretq_u16_s16(res1),
+                          vreinterpretq_u16_s16(res2),
+                          vreinterpretq_u16_s16(res3), fwd_offset, bck_offset,
+                          round_offset64, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1, &t2, &t3);
 
           store_u8_8x4(d_u8, dst8_stride, t0, t1, t2, t3);
           d_u8 += (dst8_stride << 2);
@@ -1243,11 +1239,12 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           load_u16_8x4(d_tmp, dst_stride, &res8, &res9, &res10, &res11);
           d_tmp += (dst_stride << 2);
 
-          compute_avg_8x4(
-              res8, res9, res10, res11, vreinterpretq_u16_s16(res4),
-              vreinterpretq_u16_s16(res5), vreinterpretq_u16_s16(res6),
-              vreinterpretq_u16_s16(res7), fwd_offset, bck_offset,
-              round_offset64, round_bits, use_jnt_comp_avg, &t0, &t1, &t2, &t3);
+          compute_avg_8x4(res8, res9, res10, res11, vreinterpretq_u16_s16(res4),
+                          vreinterpretq_u16_s16(res5),
+                          vreinterpretq_u16_s16(res6),
+                          vreinterpretq_u16_s16(res7), fwd_offset, bck_offset,
+                          round_offset64, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1, &t2, &t3);
 
           store_u8_8x4(d_u8, dst8_stride, t0, t1, t2, t3);
           d_u8 += (dst8_stride << 2);
@@ -1319,7 +1316,7 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
 
           compute_avg_8x1(res8, vreinterpretq_u16_s16(res0), fwd_offset,
                           bck_offset, round_offset64, round_bits,
-                          use_jnt_comp_avg, &t0);
+                          use_dist_wtd_comp_avg, &t0);
 
           vst1_u8(d_u8, t0);
           d_u8 += (dst8_stride);
@@ -1342,12 +1339,12 @@ void av1_jnt_convolve_x_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
   }
 }
 
-void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
-                             int dst8_stride, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_y_neon(const uint8_t *src, int src_stride,
+                                  uint8_t *dst8, int dst8_stride, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   assert(!(w % 4));
   assert(!(h % 4));
 
@@ -1363,7 +1360,7 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
       2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const uint16_t fwd_offset = conv_params->fwd_offset;
   const uint16_t bck_offset = conv_params->bck_offset;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int shift_value = (conv_params->round_1 - 1 - bits);
 
   (void)filter_params_x;
@@ -1489,8 +1486,8 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           compute_avg_4x4(res4, res5, res6, res7, vreinterpret_u16_s16(d0),
                           vreinterpret_u16_s16(d1), vreinterpret_u16_s16(d2),
                           vreinterpret_u16_s16(d3), fwd_offset, bck_offset,
-                          round_offset64, round_bits, use_jnt_comp_avg, &t0,
-                          &t1);
+                          round_offset64, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1);
 
           vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0), 0);
           d_u8 += dst8_stride;
@@ -1535,7 +1532,7 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
 
           compute_avg_4x1(res4, vreinterpret_u16_s16(d0), fwd_offset,
                           bck_offset, round_offset64, round_bits,
-                          use_jnt_comp_avg, &t0);
+                          use_dist_wtd_comp_avg, &t0);
 
           vst1_lane_u32((uint32_t *)d_u8, vreinterpret_u32_u8(t0), 0);
           d_u8 += dst8_stride;
@@ -1654,11 +1651,12 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           load_u16_8x4(d_tmp, dst_stride, &res8, &res9, &res10, &res11);
           d_tmp += (dst_stride << 2);
 
-          compute_avg_8x4(
-              res8, res9, res10, res11, vreinterpretq_u16_s16(res0),
-              vreinterpretq_u16_s16(res1), vreinterpretq_u16_s16(res2),
-              vreinterpretq_u16_s16(res3), fwd_offset, bck_offset,
-              round_offset64, round_bits, use_jnt_comp_avg, &t0, &t1, &t2, &t3);
+          compute_avg_8x4(res8, res9, res10, res11, vreinterpretq_u16_s16(res0),
+                          vreinterpretq_u16_s16(res1),
+                          vreinterpretq_u16_s16(res2),
+                          vreinterpretq_u16_s16(res3), fwd_offset, bck_offset,
+                          round_offset64, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1, &t2, &t3);
 
           store_u8_8x4(d_u8, dst8_stride, t0, t1, t2, t3);
           d_u8 += (dst8_stride << 2);
@@ -1666,11 +1664,12 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
           load_u16_8x4(d_tmp, dst_stride, &res8, &res9, &res10, &res11);
           d_tmp += (dst_stride << 2);
 
-          compute_avg_8x4(
-              res8, res9, res10, res11, vreinterpretq_u16_s16(res4),
-              vreinterpretq_u16_s16(res5), vreinterpretq_u16_s16(res6),
-              vreinterpretq_u16_s16(res7), fwd_offset, bck_offset,
-              round_offset64, round_bits, use_jnt_comp_avg, &t0, &t1, &t2, &t3);
+          compute_avg_8x4(res8, res9, res10, res11, vreinterpretq_u16_s16(res4),
+                          vreinterpretq_u16_s16(res5),
+                          vreinterpretq_u16_s16(res6),
+                          vreinterpretq_u16_s16(res7), fwd_offset, bck_offset,
+                          round_offset64, round_bits, use_dist_wtd_comp_avg,
+                          &t0, &t1, &t2, &t3);
 
           store_u8_8x4(d_u8, dst8_stride, t0, t1, t2, t3);
           d_u8 += (dst8_stride << 2);
@@ -1718,7 +1717,7 @@ void av1_jnt_convolve_y_neon(const uint8_t *src, int src_stride, uint8_t *dst8,
 
           compute_avg_8x1(res8, vreinterpretq_u16_s16(res0), fwd_offset,
                           bck_offset, round_offset64, round_bits,
-                          use_jnt_comp_avg, &t0);
+                          use_dist_wtd_comp_avg, &t0);
 
           vst1_u8(d_u8, t0);
           d_u8 += (dst8_stride);
diff --git a/third_party/aom/av1/common/arm/mem_neon.h b/third_party/aom/av1/common/arm/mem_neon.h
index c4ae2e7849..beae4ed0db 100644
--- a/third_party/aom/av1/common/arm/mem_neon.h
+++ b/third_party/aom/av1/common/arm/mem_neon.h
@@ -383,6 +383,15 @@ static INLINE void load_unaligned_u8_4x2(const uint8_t *buf, int stride,
   *tu0 = vset_lane_u32(a, *tu0, 1);
 }
 
+/* These intrinsics require immediate values, so we must use #defines
+   to enforce that. */
+#define store_unaligned_u8_4x1(dst, src, lane)         \
+  do {                                                 \
+    uint32_t a;                                        \
+    a = vget_lane_u32(vreinterpret_u32_u8(src), lane); \
+    memcpy(dst, &a, 4);                                \
+  } while (0)
+
 static INLINE void load_unaligned_u8_2x2(const uint8_t *buf, int stride,
                                          uint16x4_t *tu0) {
   uint16_t a;
diff --git a/third_party/aom/av1/common/arm/warp_plane_neon.c b/third_party/aom/av1/common/arm/warp_plane_neon.c
index 7f02d42a73..1062cc3f9d 100644
--- a/third_party/aom/av1/common/arm/warp_plane_neon.c
+++ b/third_party/aom/av1/common/arm/warp_plane_neon.c
@@ -640,7 +640,7 @@ void av1_warp_affine_neon(const int32_t *mat, const uint8_t *ref, int width,
             uint16x4_t tmp16_lo = vld1_u16(p);
             int32x4_t tmp32_lo = vreinterpretq_s32_u32(vmovl_u16(tmp16_lo));
             int16x4_t tmp16_low;
-            if (conv_params->use_jnt_comp_avg) {
+            if (conv_params->use_dist_wtd_comp_avg) {
               res_lo = vmulq_s32(res_lo, bwd);
               tmp32_lo = vmulq_s32(tmp32_lo, fwd);
               tmp32_lo = vaddq_s32(tmp32_lo, res_lo);
@@ -671,7 +671,7 @@ void av1_warp_affine_neon(const int32_t *mat, const uint8_t *ref, int width,
               uint16x4_t tmp16_hi = vld1_u16(p4);
               int32x4_t tmp32_hi = vreinterpretq_s32_u32(vmovl_u16(tmp16_hi));
               int16x4_t tmp16_high;
-              if (conv_params->use_jnt_comp_avg) {
+              if (conv_params->use_dist_wtd_comp_avg) {
                 res_hi = vmulq_s32(res_hi, bwd);
                 tmp32_hi = vmulq_s32(tmp32_hi, fwd);
                 tmp32_hi = vaddq_s32(tmp32_hi, res_hi);
diff --git a/third_party/aom/av1/common/av1_inv_txfm2d.c b/third_party/aom/av1/common/av1_inv_txfm2d.c
index 4e69443145..4f2d57b802 100644
--- a/third_party/aom/av1/common/av1_inv_txfm2d.c
+++ b/third_party/aom/av1/common/av1_inv_txfm2d.c
@@ -183,7 +183,6 @@ void av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
                           TXFM_2D_FLIP_CFG *cfg) {
   assert(cfg != NULL);
   cfg->tx_size = tx_size;
-  set_flip_cfg(tx_type, cfg);
   av1_zero(cfg->stage_range_col);
   av1_zero(cfg->stage_range_row);
   set_flip_cfg(tx_type, cfg);
diff --git a/third_party/aom/av1/common/av1_loopfilter.c b/third_party/aom/av1/common/av1_loopfilter.c
index 537d8dfe92..0aa1f9b907 100644
--- a/third_party/aom/av1/common/av1_loopfilter.c
+++ b/third_party/aom/av1/common/av1_loopfilter.c
@@ -32,7 +32,7 @@ static const int delta_lf_id_lut[MAX_MB_PLANE][2] = {
   { 0, 1 }, { 2, 2 }, { 3, 3 }
 };
 
-typedef enum EDGE_DIR { VERT_EDGE = 0, HORZ_EDGE = 1, NUM_EDGE_DIRS } EDGE_DIR;
+enum { VERT_EDGE = 0, HORZ_EDGE = 1, NUM_EDGE_DIRS } UENUM1BYTE(EDGE_DIR);
 
 static const int mode_lf_lut[] = {
   0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  // INTRA_MODES
@@ -40,7 +40,6 @@ static const int mode_lf_lut[] = {
   1, 1, 1, 1, 1, 1, 0, 1  // INTER_COMPOUND_MODES (GLOBAL_GLOBALMV == 0)
 };
 
-#if LOOP_FILTER_BITMASK
 // 256 bit masks (64x64 / 4x4) for left transform size for Y plane.
 // We use 4 uint64_t to represent the 256 bit.
 // Each 1 represents a position where we should apply a loop filter
@@ -113,6 +112,10 @@ const int mask_id_table_tx_32x32[BLOCK_SIZES_ALL] = { -1, -1, -1, -1, -1, -1,
                                                       -1, -1, -1, 0,  1,  2,
                                                       3,  -1, -1, -1, -1, -1,
                                                       -1, -1, -1, -1 };
+const int mask_id_table_vert_border[BLOCK_SIZES_ALL] = { 0,  47, 49, 19, 51, 53,
+                                                         33, 55, 57, 42, 59, 60,
+                                                         46, -1, -1, -1, 61, 62,
+                                                         63, 64, 65, 66 };
 
 const FilterMask left_mask_univariant_reordered[67] = {
   // TX_4X4
@@ -402,6 +405,7 @@ const FilterMask above_mask_univariant_reordered[67] = {
       0x0000000000000000ULL } },  // block size 64X16, TX_64X16
 };
 
+#if LOOP_FILTER_BITMASK
 LoopFilterMask *get_loop_filter_mask(const AV1_COMMON *const cm, int mi_row,
                                      int mi_col) {
   assert(cm->lf.lfm != NULL);
@@ -452,9 +456,9 @@ uint8_t get_filter_level(const AV1_COMMON *cm, const loop_filter_info_n *lfi_n,
                          const int dir_idx, int plane,
                          const MB_MODE_INFO *mbmi) {
   const int segment_id = mbmi->segment_id;
-  if (cm->delta_lf_present_flag) {
+  if (cm->delta_q_info.delta_lf_present_flag) {
     int delta_lf;
-    if (cm->delta_lf_multi) {
+    if (cm->delta_q_info.delta_lf_multi) {
       const int delta_lf_idx = delta_lf_id_lut[plane][dir_idx];
       delta_lf = mbmi->delta_lf[delta_lf_idx];
     } else {
@@ -531,6 +535,9 @@ void av1_loop_filter_frame_init(AV1_COMMON *cm, int plane_start,
   filt_lvl_r[1] = cm->lf.filter_level_u;
   filt_lvl_r[2] = cm->lf.filter_level_v;
 
+  assert(plane_start >= AOM_PLANE_Y);
+  assert(plane_end <= MAX_MB_PLANE);
+
   for (plane = plane_start; plane < plane_end; plane++) {
     if (plane == 0 && !filt_lvl[0] && !filt_lvl_r[0])
       break;
@@ -542,7 +549,6 @@ void av1_loop_filter_frame_init(AV1_COMMON *cm, int plane_start,
     for (seg_id = 0; seg_id < MAX_SEGMENTS; seg_id++) {
       for (int dir = 0; dir < 2; ++dir) {
         int lvl_seg = (dir == 0) ? filt_lvl[plane] : filt_lvl_r[plane];
-        assert(plane >= 0 && plane <= 2);
         const int seg_lf_feature_id = seg_lvl_lf_lut[plane][dir];
         if (segfeature_active(seg, seg_id, seg_lf_feature_id)) {
           const int data = get_segdata(&cm->seg, seg_id, seg_lf_feature_id);
@@ -763,9 +769,11 @@ static void setup_masks(AV1_COMMON *const cm, int mi_row, int mi_col, int plane,
           else
             lfm->lfl_y_hor[row][col] = level;
         } else if (plane == 1) {
-          lfm->lfl_u[row][col] = level;
+          lfm->lfl_u_ver[row][col] = level;
+          lfm->lfl_u_hor[row][col] = level;
         } else {
-          lfm->lfl_v[row][col] = level;
+          lfm->lfl_v_ver[row][col] = level;
+          lfm->lfl_v_hor[row][col] = level;
         }
       }
     }
@@ -1053,11 +1061,13 @@ void av1_setup_bitmask(AV1_COMMON *const cm, int mi_row, int mi_col, int plane,
       } else if (plane == 1) {
         av1_zero(lfm->left_u);
         av1_zero(lfm->above_u);
-        av1_zero(lfm->lfl_u);
+        av1_zero(lfm->lfl_u_ver);
+        av1_zero(lfm->lfl_u_hor);
       } else {
         av1_zero(lfm->left_v);
         av1_zero(lfm->above_v);
-        av1_zero(lfm->lfl_v);
+        av1_zero(lfm->lfl_v_ver);
+        av1_zero(lfm->lfl_v_hor);
       }
     }
   }
@@ -1311,11 +1321,16 @@ static void filter_selectively_horiz(uint8_t *s, int pitch, int plane,
   int count;
   const int step = 1 << subsampling;
   const unsigned int two_block_mask = subsampling ? 5 : 3;
+  int offset = 0;
 
   for (mask = mask_16x16 | mask_8x8 | mask_4x4; mask; mask >>= step * count) {
     const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
-    // Next block's thresholds.
-    const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + step);
+    // Next block's thresholds, when it is within current 64x64 block.
+    // If it is out of bound, its mask is zero, and it points to current edge's
+    // filter parameters, instead of next edge's.
+    int next_edge = step;
+    if (offset + next_edge >= MI_SIZE_64X64) next_edge = 0;
+    const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + next_edge);
 
     count = 1;
     if (mask & 1) {
@@ -1376,6 +1391,7 @@ static void filter_selectively_horiz(uint8_t *s, int pitch, int plane,
     mask_16x16 >>= step * count;
     mask_8x8 >>= step * count;
     mask_4x4 >>= step * count;
+    offset += step * count;
   }
 }
 
@@ -1387,11 +1403,16 @@ static void highbd_filter_selectively_horiz(
   int count;
   const int step = 1 << subsampling;
   const unsigned int two_block_mask = subsampling ? 5 : 3;
+  int offset = 0;
 
   for (mask = mask_16x16 | mask_8x8 | mask_4x4; mask; mask >>= step * count) {
     const loop_filter_thresh *lfi = lfi_n->lfthr + *lfl;
-    // Next block's thresholds.
-    const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + step);
+    // Next block's thresholds, when it is within current 64x64 block.
+    // If it is out of bound, its mask is zero, and it points to current edge's
+    // filter parameters, instead of next edge's.
+    int next_edge = step;
+    if (offset + next_edge >= MI_SIZE_64X64) next_edge = 0;
+    const loop_filter_thresh *lfin = lfi_n->lfthr + *(lfl + next_edge);
 
     count = 1;
     if (mask & 1) {
@@ -1401,9 +1422,9 @@ static void highbd_filter_selectively_horiz(
 
         if ((mask_16x16 & two_block_mask) == two_block_mask) {
           if (plane) {
-            aom_highbd_lpf_horizontal_6_dual(s, pitch, lfi->mblim, lfi->lim,
-                                             lfi->hev_thr, lfin->mblim,
-                                             lfin->lim, lfin->hev_thr, bd);
+            aom_highbd_lpf_horizontal_6_dual_c(s, pitch, lfi->mblim, lfi->lim,
+                                               lfi->hev_thr, lfin->mblim,
+                                               lfin->lim, lfin->hev_thr, bd);
           } else {
             aom_highbd_lpf_horizontal_14_dual(s, pitch, lfi->mblim, lfi->lim,
                                               lfi->hev_thr, lfin->mblim,
@@ -1420,13 +1441,13 @@ static void highbd_filter_selectively_horiz(
 
         if ((mask_8x8 & two_block_mask) == two_block_mask) {
           if (plane) {
-            aom_highbd_lpf_horizontal_6_dual(s, pitch, lfi->mblim, lfi->lim,
-                                             lfi->hev_thr, lfin->mblim,
-                                             lfin->lim, lfin->hev_thr, bd);
+            aom_highbd_lpf_horizontal_6_dual_c(s, pitch, lfi->mblim, lfi->lim,
+                                               lfi->hev_thr, lfin->mblim,
+                                               lfin->lim, lfin->hev_thr, bd);
           } else {
-            aom_highbd_lpf_horizontal_8_dual(s, pitch, lfi->mblim, lfi->lim,
-                                             lfi->hev_thr, lfin->mblim,
-                                             lfin->lim, lfin->hev_thr, bd);
+            aom_highbd_lpf_horizontal_8_dual_c(s, pitch, lfi->mblim, lfi->lim,
+                                               lfi->hev_thr, lfin->mblim,
+                                               lfin->lim, lfin->hev_thr, bd);
           }
           count = 2;
         } else {
@@ -1435,9 +1456,9 @@ static void highbd_filter_selectively_horiz(
         }
       } else if (mask_4x4 & 1) {
         if ((mask_4x4 & two_block_mask) == two_block_mask) {
-          aom_highbd_lpf_horizontal_4_dual(s, pitch, lfi->mblim, lfi->lim,
-                                           lfi->hev_thr, lfin->mblim, lfin->lim,
-                                           lfin->hev_thr, bd);
+          aom_highbd_lpf_horizontal_4_dual_c(s, pitch, lfi->mblim, lfi->lim,
+                                             lfi->hev_thr, lfin->mblim,
+                                             lfin->lim, lfin->hev_thr, bd);
           count = 2;
         } else {
           aom_highbd_lpf_horizontal_4(s, pitch, lfi->mblim, lfi->lim,
@@ -1451,6 +1472,7 @@ static void highbd_filter_selectively_horiz(
     mask_16x16 >>= step * count;
     mask_8x8 >>= step * count;
     mask_4x4 >>= step * count;
+    offset += step * count;
   }
 }
 
@@ -1463,12 +1485,13 @@ void av1_build_bitmask_vert_info(
   const int is_uv = plane > 0;
   TX_SIZE tx_size = TX_16X16, prev_tx_size = TX_16X16;
   uint8_t level, prev_level = 1;
-  int skip, prev_skip = 0;
-  int is_coding_block_border;
+  uint64_t skip, prev_skip = 0;
+  uint64_t is_coding_block_border;
 
   for (int r = 0; (r << MI_SIZE_LOG2) < plane_ptr->dst.height; r += row_step) {
     const int mi_row = r << subsampling_y;
     const int row = mi_row % MI_SIZE_64X64;
+    const int row_uv = row | subsampling_y;
     int index = 0;
     const int shift = get_index_shift(0, row, &index);
 
@@ -1482,13 +1505,14 @@ void av1_build_bitmask_vert_info(
         const int x = (c + col_in_unit) << MI_SIZE_LOG2;
         if (x >= plane_ptr->dst.width) break;
         const int col = col_in_unit << subsampling_x;
+        const int col_uv = col | subsampling_x;
         const uint64_t mask = ((uint64_t)1 << (shift | col));
         skip = lfm->skip.bits[index] & mask;
         is_coding_block_border = lfm->is_vert_border.bits[index] & mask;
         switch (plane) {
-          case 0: level = lfm->lfl_y_ver[row][col]; break;
-          case 1: level = lfm->lfl_u[row][col]; break;
-          case 2: level = lfm->lfl_v[row][col]; break;
+          case 0: level = lfm->lfl_y_ver[row_uv][col_uv]; break;
+          case 1: level = lfm->lfl_u_ver[row_uv][col_uv]; break;
+          case 2: level = lfm->lfl_v_ver[row_uv][col_uv]; break;
           default: assert(plane >= 0 && plane <= 2); return;
         }
         for (TX_SIZE ts = TX_4X4; ts <= TX_64X64; ++ts) {
@@ -1502,9 +1526,7 @@ void av1_build_bitmask_vert_info(
             (!prev_skip || !skip || is_coding_block_border)) {
           const TX_SIZE min_tx_size =
               AOMMIN(TX_16X16, AOMMIN(tx_size, prev_tx_size));
-          const int tmp_row = (mi_row | subsampling_y) % MI_SIZE_64X64;
-          const int tmp_col = (col | subsampling_x) % MI_SIZE_64X64;
-          const int shift_1 = get_index_shift(tmp_col, tmp_row, &index);
+          const int shift_1 = get_index_shift(col_uv, row_uv, &index);
           const uint64_t mask_1 = ((uint64_t)1 << shift_1);
           switch (plane) {
             case 0: lfm->left_y[min_tx_size].bits[index] |= mask_1; break;
@@ -1512,6 +1534,14 @@ void av1_build_bitmask_vert_info(
             case 2: lfm->left_v[min_tx_size].bits[index] |= mask_1; break;
             default: assert(plane >= 0 && plane <= 2); return;
           }
+          if (level == 0 && prev_level != 0) {
+            switch (plane) {
+              case 0: lfm->lfl_y_ver[row_uv][col_uv] = prev_level; break;
+              case 1: lfm->lfl_u_ver[row_uv][col_uv] = prev_level; break;
+              case 2: lfm->lfl_v_ver[row_uv][col_uv] = prev_level; break;
+              default: assert(plane >= 0 && plane <= 2); return;
+            }
+          }
         }
 
         // update prev info
@@ -1534,12 +1564,13 @@ void av1_build_bitmask_horz_info(
   const int is_uv = plane > 0;
   TX_SIZE tx_size = TX_16X16, prev_tx_size = TX_16X16;
   uint8_t level, prev_level = 1;
-  int skip, prev_skip = 0;
-  int is_coding_block_border;
+  uint64_t skip, prev_skip = 0;
+  uint64_t is_coding_block_border;
 
   for (int c = 0; (c << MI_SIZE_LOG2) < plane_ptr->dst.width; c += col_step) {
     const int mi_col = c << subsampling_x;
     const int col = mi_col % MI_SIZE_64X64;
+    const int col_uv = col | subsampling_x;
 
     for (int r = 0; (r << MI_SIZE_LOG2) < plane_ptr->dst.height;
          r += (tx_size_high_unit[TX_64X64] >> subsampling_y)) {
@@ -1551,15 +1582,16 @@ void av1_build_bitmask_horz_info(
         const int y = (r + r_in_unit) << MI_SIZE_LOG2;
         if (y >= plane_ptr->dst.height) break;
         const int row = r_in_unit << subsampling_y;
+        const int row_uv = row | subsampling_y;
         int index = 0;
         const int shift = get_index_shift(col, row, &index);
         const uint64_t mask = ((uint64_t)1 << shift);
         skip = lfm->skip.bits[index] & mask;
         is_coding_block_border = lfm->is_horz_border.bits[index] & mask;
         switch (plane) {
-          case 0: level = lfm->lfl_y_hor[row][col]; break;
-          case 1: level = lfm->lfl_u[row][col]; break;
-          case 2: level = lfm->lfl_v[row][col]; break;
+          case 0: level = lfm->lfl_y_hor[row_uv][col_uv]; break;
+          case 1: level = lfm->lfl_u_hor[row_uv][col_uv]; break;
+          case 2: level = lfm->lfl_v_hor[row_uv][col_uv]; break;
           default: assert(plane >= 0 && plane <= 2); return;
         }
         for (TX_SIZE ts = TX_4X4; ts <= TX_64X64; ++ts) {
@@ -1573,9 +1605,7 @@ void av1_build_bitmask_horz_info(
             (!prev_skip || !skip || is_coding_block_border)) {
           const TX_SIZE min_tx_size =
               AOMMIN(TX_16X16, AOMMIN(tx_size, prev_tx_size));
-          const int tmp_row = (row | subsampling_y) % MI_SIZE_64X64;
-          const int tmp_col = (mi_col | subsampling_x) % MI_SIZE_64X64;
-          const int shift_1 = get_index_shift(tmp_col, tmp_row, &index);
+          const int shift_1 = get_index_shift(col_uv, row_uv, &index);
           const uint64_t mask_1 = ((uint64_t)1 << shift_1);
 
           switch (plane) {
@@ -1584,6 +1614,14 @@ void av1_build_bitmask_horz_info(
             case 2: lfm->above_v[min_tx_size].bits[index] |= mask_1; break;
             default: assert(plane >= 0 && plane <= 2); return;
           }
+          if (level == 0 && prev_level != 0) {
+            switch (plane) {
+              case 0: lfm->lfl_y_hor[row_uv][col_uv] = prev_level; break;
+              case 1: lfm->lfl_u_hor[row_uv][col_uv] = prev_level; break;
+              case 2: lfm->lfl_v_hor[row_uv][col_uv] = prev_level; break;
+              default: assert(plane >= 0 && plane <= 2); return;
+            }
+          }
         }
 
         // update prev info
@@ -1628,6 +1666,7 @@ void av1_filter_block_plane_bitmask_vert(
     const int shift = get_index_shift(col, row, &index);
     int index_next = 0;
     const int shift_next = get_index_shift(col, row_next, &index_next);
+    const int has_next_row = row_next < cm->mi_rows;
     switch (pl) {
       case 0:
         mask_16x16 = lfm->left_y[TX_16X16].bits[index];
@@ -1640,15 +1679,15 @@ void av1_filter_block_plane_bitmask_vert(
         mask_16x16 = lfm->left_u[TX_16X16].bits[index];
         mask_8x8 = lfm->left_u[TX_8X8].bits[index];
         mask_4x4 = lfm->left_u[TX_4X4].bits[index];
-        lfl = &lfm->lfl_u[row][col];
-        lfl2 = &lfm->lfl_u[row_next][col];
+        lfl = &lfm->lfl_u_ver[row][col];
+        lfl2 = &lfm->lfl_u_ver[row_next][col];
         break;
       case 2:
         mask_16x16 = lfm->left_v[TX_16X16].bits[index];
         mask_8x8 = lfm->left_v[TX_8X8].bits[index];
         mask_4x4 = lfm->left_v[TX_4X4].bits[index];
-        lfl = &lfm->lfl_v[row][col];
-        lfl2 = &lfm->lfl_v[row_next][col];
+        lfl = &lfm->lfl_v_ver[row][col];
+        lfl2 = &lfm->lfl_v_ver[row_next][col];
         break;
       default: assert(pl >= 0 && pl <= 2); return;
     }
@@ -1658,6 +1697,11 @@ void av1_filter_block_plane_bitmask_vert(
     uint64_t mask_16x16_1 = (mask_16x16 >> shift_next) & mask_cutoff;
     uint64_t mask_8x8_1 = (mask_8x8 >> shift_next) & mask_cutoff;
     uint64_t mask_4x4_1 = (mask_4x4 >> shift_next) & mask_cutoff;
+    if (!has_next_row) {
+      mask_16x16_1 = 0;
+      mask_8x8_1 = 0;
+      mask_4x4_1 = 0;
+    }
 
     if (cm->seq_params.use_highbitdepth)
       highbd_filter_selectively_vert_row2(
@@ -1712,13 +1756,13 @@ void av1_filter_block_plane_bitmask_horz(
         mask_16x16 = lfm->above_u[TX_16X16].bits[index];
         mask_8x8 = lfm->above_u[TX_8X8].bits[index];
         mask_4x4 = lfm->above_u[TX_4X4].bits[index];
-        lfl = &lfm->lfl_u[row][col];
+        lfl = &lfm->lfl_u_hor[row][col];
         break;
       case 2:
         mask_16x16 = lfm->above_v[TX_16X16].bits[index];
         mask_8x8 = lfm->above_v[TX_8X8].bits[index];
         mask_4x4 = lfm->above_v[TX_4X4].bits[index];
-        lfl = &lfm->lfl_v[row][col];
+        lfl = &lfm->lfl_v_hor[row][col];
         break;
       default: assert(pl >= 0 && pl <= 2); return;
     }
@@ -1786,15 +1830,15 @@ void av1_filter_block_plane_ver(AV1_COMMON *const cm,
           mask_16x16 = lfm->left_u[TX_16X16].bits[index];
           mask_8x8 = lfm->left_u[TX_8X8].bits[index];
           mask_4x4 = lfm->left_u[TX_4X4].bits[index];
-          lfl = &lfm->lfl_u[row][col];
-          lfl2 = &lfm->lfl_u[row_next][col];
+          lfl = &lfm->lfl_u_ver[row][col];
+          lfl2 = &lfm->lfl_u_ver[row_next][col];
           break;
         case 2:
           mask_16x16 = lfm->left_v[TX_16X16].bits[index];
           mask_8x8 = lfm->left_v[TX_8X8].bits[index];
           mask_4x4 = lfm->left_v[TX_4X4].bits[index];
-          lfl = &lfm->lfl_v[row][col];
-          lfl2 = &lfm->lfl_v[row_next][col];
+          lfl = &lfm->lfl_v_ver[row][col];
+          lfl2 = &lfm->lfl_v_ver[row_next][col];
           break;
         default: assert(pl >= 0 && pl <= 2); return;
       }
@@ -1861,13 +1905,13 @@ void av1_filter_block_plane_hor(AV1_COMMON *const cm,
           mask_16x16 = lfm->above_u[TX_16X16].bits[index];
           mask_8x8 = lfm->above_u[TX_8X8].bits[index];
           mask_4x4 = lfm->above_u[TX_4X4].bits[index];
-          lfl = &lfm->lfl_u[row][col];
+          lfl = &lfm->lfl_u_hor[row][col];
           break;
         case 2:
           mask_16x16 = lfm->above_v[TX_16X16].bits[index];
           mask_8x8 = lfm->above_v[TX_8X8].bits[index];
           mask_4x4 = lfm->above_v[TX_4X4].bits[index];
-          lfl = &lfm->lfl_v[row][col];
+          lfl = &lfm->lfl_v_hor[row][col];
           break;
         default: assert(pl >= 0 && pl <= 2); return;
       }
@@ -2219,6 +2263,92 @@ void av1_filter_block_plane_horz(const AV1_COMMON *const cm,
   }
 }
 
+void av1_filter_block_plane_vert_test(const AV1_COMMON *const cm,
+                                      const MACROBLOCKD *const xd,
+                                      const int plane,
+                                      const MACROBLOCKD_PLANE *const plane_ptr,
+                                      const uint32_t mi_row,
+                                      const uint32_t mi_col) {
+  const int row_step = MI_SIZE >> MI_SIZE_LOG2;
+  const uint32_t scale_horz = plane_ptr->subsampling_x;
+  const uint32_t scale_vert = plane_ptr->subsampling_y;
+  uint8_t *const dst_ptr = plane_ptr->dst.buf;
+  const int dst_stride = plane_ptr->dst.stride;
+  const int y_range = cm->mi_rows >> scale_vert;
+  const int x_range = cm->mi_cols >> scale_horz;
+  for (int y = 0; y < y_range; y += row_step) {
+    uint8_t *p = dst_ptr + y * MI_SIZE * dst_stride;
+    for (int x = 0; x < x_range;) {
+      // inner loop always filter vertical edges in a MI block. If MI size
+      // is 8x8, it will filter the vertical edge aligned with a 8x8 block.
+      // If 4x4 trasnform is used, it will then filter the internal edge
+      //  aligned with a 4x4 block
+      const uint32_t curr_x = ((mi_col * MI_SIZE) >> scale_horz) + x * MI_SIZE;
+      const uint32_t curr_y = ((mi_row * MI_SIZE) >> scale_vert) + y * MI_SIZE;
+      uint32_t advance_units;
+      TX_SIZE tx_size;
+      AV1_DEBLOCKING_PARAMETERS params;
+      memset(&params, 0, sizeof(params));
+
+      tx_size =
+          set_lpf_parameters(&params, ((ptrdiff_t)1 << scale_horz), cm, xd,
+                             VERT_EDGE, curr_x, curr_y, plane, plane_ptr);
+      if (tx_size == TX_INVALID) {
+        params.filter_length = 0;
+        tx_size = TX_4X4;
+      }
+
+      // advance the destination pointer
+      advance_units = tx_size_wide_unit[tx_size];
+      x += advance_units;
+      p += advance_units * MI_SIZE;
+    }
+  }
+}
+
+void av1_filter_block_plane_horz_test(const AV1_COMMON *const cm,
+                                      const MACROBLOCKD *const xd,
+                                      const int plane,
+                                      const MACROBLOCKD_PLANE *const plane_ptr,
+                                      const uint32_t mi_row,
+                                      const uint32_t mi_col) {
+  const int col_step = MI_SIZE >> MI_SIZE_LOG2;
+  const uint32_t scale_horz = plane_ptr->subsampling_x;
+  const uint32_t scale_vert = plane_ptr->subsampling_y;
+  uint8_t *const dst_ptr = plane_ptr->dst.buf;
+  const int dst_stride = plane_ptr->dst.stride;
+  const int y_range = cm->mi_rows >> scale_vert;
+  const int x_range = cm->mi_cols >> scale_horz;
+  for (int x = 0; x < x_range; x += col_step) {
+    uint8_t *p = dst_ptr + x * MI_SIZE;
+    for (int y = 0; y < y_range;) {
+      // inner loop always filter vertical edges in a MI block. If MI size
+      // is 8x8, it will first filter the vertical edge aligned with a 8x8
+      // block. If 4x4 trasnform is used, it will then filter the internal
+      // edge aligned with a 4x4 block
+      const uint32_t curr_x = ((mi_col * MI_SIZE) >> scale_horz) + x * MI_SIZE;
+      const uint32_t curr_y = ((mi_row * MI_SIZE) >> scale_vert) + y * MI_SIZE;
+      uint32_t advance_units;
+      TX_SIZE tx_size;
+      AV1_DEBLOCKING_PARAMETERS params;
+      memset(&params, 0, sizeof(params));
+
+      tx_size =
+          set_lpf_parameters(&params, (cm->mi_stride << scale_vert), cm, xd,
+                             HORZ_EDGE, curr_x, curr_y, plane, plane_ptr);
+      if (tx_size == TX_INVALID) {
+        params.filter_length = 0;
+        tx_size = TX_4X4;
+      }
+
+      // advance the destination pointer
+      advance_units = tx_size_high_unit[tx_size];
+      y += advance_units;
+      p += advance_units * dst_stride * MI_SIZE;
+    }
+  }
+}
+
 static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
                              MACROBLOCKD *xd, int start, int stop,
 #if LOOP_FILTER_BITMASK
@@ -2233,6 +2363,7 @@ static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
 
 #if LOOP_FILTER_BITMASK
   if (is_decoding) {
+    cm->is_decoding = is_decoding;
     for (plane = plane_start; plane < plane_end; plane++) {
       if (plane == 0 && !(cm->lf.filter_level[0]) && !(cm->lf.filter_level[1]))
         break;
@@ -2243,24 +2374,25 @@ static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
 
       av1_setup_dst_planes(pd, cm->seq_params.sb_size, frame_buffer, 0, 0,
                            plane, plane + 1);
+
       av1_build_bitmask_vert_info(cm, &pd[plane], plane);
       av1_build_bitmask_horz_info(cm, &pd[plane], plane);
 
       // apply loop filtering which only goes through buffer once
       for (mi_row = start; mi_row < stop; mi_row += MI_SIZE_64X64) {
         for (mi_col = col_start; mi_col < col_end; mi_col += MI_SIZE_64X64) {
-          av1_setup_dst_planes(pd, MI_SIZE_64X64, frame_buffer, mi_row, mi_col,
+          av1_setup_dst_planes(pd, BLOCK_64X64, frame_buffer, mi_row, mi_col,
                                plane, plane + 1);
           av1_filter_block_plane_bitmask_vert(cm, &pd[plane], plane, mi_row,
                                               mi_col);
           if (mi_col - MI_SIZE_64X64 >= 0) {
-            av1_setup_dst_planes(pd, MI_SIZE_64X64, frame_buffer, mi_row,
+            av1_setup_dst_planes(pd, BLOCK_64X64, frame_buffer, mi_row,
                                  mi_col - MI_SIZE_64X64, plane, plane + 1);
             av1_filter_block_plane_bitmask_horz(cm, &pd[plane], plane, mi_row,
                                                 mi_col - MI_SIZE_64X64);
           }
         }
-        av1_setup_dst_planes(pd, MI_SIZE_64X64, frame_buffer, mi_row,
+        av1_setup_dst_planes(pd, BLOCK_64X64, frame_buffer, mi_row,
                              mi_col - MI_SIZE_64X64, plane, plane + 1);
         av1_filter_block_plane_bitmask_horz(cm, &pd[plane], plane, mi_row,
                                             mi_col - MI_SIZE_64X64);
@@ -2278,31 +2410,6 @@ static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
     else if (plane == 2 && !(cm->lf.filter_level_v))
       continue;
 
-#if LOOP_FILTER_BITMASK
-    // filter all vertical edges every superblock (could be 128x128 or 64x64)
-    for (mi_row = start; mi_row < stop; mi_row += cm->seq_params.mib_size) {
-      for (mi_col = col_start; mi_col < col_end;
-           mi_col += cm->seq_params.mib_size) {
-        av1_setup_dst_planes(pd, cm->seq_params.sb_size, frame_buffer, mi_row,
-                             mi_col, plane, plane + 1);
-
-        av1_setup_bitmask(cm, mi_row, mi_col, plane, pd[plane].subsampling_x,
-                          pd[plane].subsampling_y, stop, col_end);
-        av1_filter_block_plane_ver(cm, &pd[plane], plane, mi_row, mi_col);
-      }
-    }
-
-    // filter all horizontal edges every superblock
-    for (mi_row = start; mi_row < stop; mi_row += cm->seq_params.mib_size) {
-      for (mi_col = col_start; mi_col < col_end;
-           mi_col += cm->seq_params.mib_size) {
-        av1_setup_dst_planes(pd, cm->seq_params.sb_size, frame_buffer, mi_row,
-                             mi_col, plane, plane + 1);
-
-        av1_filter_block_plane_hor(cm, &pd[plane], plane, mi_row, mi_col);
-      }
-    }
-#else
     if (cm->lf.combine_vert_horz_lf) {
       // filter all vertical and horizontal edges in every 128x128 super block
       for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
@@ -2348,7 +2455,6 @@ static void loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
         }
       }
     }
-#endif  // LOOP_FILTER_BITMASK
   }
 }
 
diff --git a/third_party/aom/av1/common/av1_loopfilter.h b/third_party/aom/av1/common/av1_loopfilter.h
index 80ac611781..ae4d372fad 100644
--- a/third_party/aom/av1/common/av1_loopfilter.h
+++ b/third_party/aom/av1/common/av1_loopfilter.h
@@ -33,11 +33,11 @@ enum lf_path {
   LF_PATH_SLOW,
 };
 
-#if LOOP_FILTER_BITMASK
 typedef struct {
   uint64_t bits[4];
 } FilterMask;
 
+#if LOOP_FILTER_BITMASK
 // This structure holds bit masks for all 4x4 blocks in a 64x64 region.
 // Each 1 bit represents a position in which we want to apply the loop filter.
 // For Y plane, 4x4 in 64x64 requires 16x16 = 256 bit, therefore we use 4
@@ -61,10 +61,12 @@ typedef struct {
   uint8_t lfl_y_ver[MI_SIZE_64X64][MI_SIZE_64X64];
 
   // U plane filter level
-  uint8_t lfl_u[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_u_ver[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_u_hor[MI_SIZE_64X64][MI_SIZE_64X64];
 
   // V plane filter level
-  uint8_t lfl_v[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_v_ver[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_v_hor[MI_SIZE_64X64][MI_SIZE_64X64];
 
   // other info
   FilterMask skip;
@@ -173,39 +175,23 @@ LoopFilterMask *get_loop_filter_mask(const struct AV1Common *const cm,
                                      int mi_row, int mi_col);
 int get_index_shift(int mi_col, int mi_row, int *index);
 
-static const FilterMask left_txform_mask[TX_SIZES] = {
-  { { 0x0000000000000001ULL,  // TX_4X4,
-      0x0000000000000000ULL, 0x0000000000000000ULL, 0x0000000000000000ULL } },
+void av1_build_bitmask_vert_info(
+    struct AV1Common *const cm, const struct macroblockd_plane *const plane_ptr,
+    int plane);
 
-  { { 0x0000000000010001ULL,  // TX_8X8,
-      0x0000000000000000ULL, 0x0000000000000000ULL, 0x0000000000000000ULL } },
+void av1_build_bitmask_horz_info(
+    struct AV1Common *const cm, const struct macroblockd_plane *const plane_ptr,
+    int plane);
 
-  { { 0x0001000100010001ULL,  // TX_16X16,
-      0x0000000000000000ULL, 0x0000000000000000ULL, 0x0000000000000000ULL } },
+void av1_filter_block_plane_bitmask_vert(
+    struct AV1Common *const cm, struct macroblockd_plane *const plane_ptr,
+    int pl, int mi_row, int mi_col);
 
-  { { 0x0001000100010001ULL,  // TX_32X32,
-      0x0001000100010001ULL, 0x0000000000000000ULL, 0x0000000000000000ULL } },
+void av1_filter_block_plane_bitmask_horz(
+    struct AV1Common *const cm, struct macroblockd_plane *const plane_ptr,
+    int pl, int mi_row, int mi_col);
 
-  { { 0x0001000100010001ULL,  // TX_64X64,
-      0x0001000100010001ULL, 0x0001000100010001ULL, 0x0001000100010001ULL } },
-};
-
-static const uint64_t above_txform_mask[2][TX_SIZES] = {
-  {
-      0x0000000000000001ULL,  // TX_4X4
-      0x0000000000000003ULL,  // TX_8X8
-      0x000000000000000fULL,  // TX_16X16
-      0x00000000000000ffULL,  // TX_32X32
-      0x000000000000ffffULL,  // TX_64X64
-  },
-  {
-      0x0000000000000001ULL,  // TX_4X4
-      0x0000000000000005ULL,  // TX_8X8
-      0x0000000000000055ULL,  // TX_16X16
-      0x0000000000005555ULL,  // TX_32X32
-      0x0000000055555555ULL,  // TX_64X64
-  },
-};
+#endif  // LOOP_FILTER_BITMASK
 
 extern const int mask_id_table_tx_4x4[BLOCK_SIZES_ALL];
 
@@ -215,10 +201,13 @@ extern const int mask_id_table_tx_16x16[BLOCK_SIZES_ALL];
 
 extern const int mask_id_table_tx_32x32[BLOCK_SIZES_ALL];
 
+// corresponds to entry id in table left_mask_univariant_reordered,
+// of block size mxn and TX_mxn.
+extern const int mask_id_table_vert_border[BLOCK_SIZES_ALL];
+
 extern const FilterMask left_mask_univariant_reordered[67];
 
 extern const FilterMask above_mask_univariant_reordered[67];
-#endif
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/common/av1_rtcd_defs.pl b/third_party/aom/av1/common/av1_rtcd_defs.pl
index dee1f1c79b..7d67d37c2b 100755
--- a/third_party/aom/av1/common/av1_rtcd_defs.pl
+++ b/third_party/aom/av1/common/av1_rtcd_defs.pl
@@ -33,6 +33,8 @@ struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
 struct yv12_buffer_config;
+struct NN_CONFIG;
+typedef struct NN_CONFIG NN_CONFIG;
 
 /* Function pointers return by CfL functions */
 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
@@ -79,8 +81,11 @@ specialize qw/av1_highbd_wiener_convolve_add_src avx2/;
 
 # directional intra predictor functions
 add_proto qw/void av1_dr_prediction_z1/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy";
+specialize qw/av1_dr_prediction_z1 avx2/;
 add_proto qw/void av1_dr_prediction_z2/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy";
+specialize qw/av1_dr_prediction_z2 avx2/;
 add_proto qw/void av1_dr_prediction_z3/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy";
+specialize qw/av1_dr_prediction_z3 avx2/;
 
 # FILTER_INTRA predictor functions
 add_proto qw/void av1_filter_intra_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left, int mode";
@@ -115,14 +120,14 @@ add_proto qw/void av1_highbd_inv_txfm_add_4x4/,  "const tran_low_t *dqcoeff, uin
 specialize qw/av1_highbd_inv_txfm_add_4x4 sse4_1/;
 add_proto qw/void av1_highbd_inv_txfm_add_8x8/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
 specialize qw/av1_highbd_inv_txfm_add_8x8 sse4_1/;
-add_proto qw/void av1_highbd_inv_txfm_add_16x8/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
-specialize qw/av1_highbd_inv_txfm_add_16x8 sse4_1/;
-add_proto qw/void av1_highbd_inv_txfm_add_8x16/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
-specialize qw/av1_highbd_inv_txfm_add_8x16 sse4_1/;
-add_proto qw/void av1_highbd_inv_txfm_add_16x16/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
-specialize qw/av1_highbd_inv_txfm_add_16x16 sse4_1/;
-add_proto qw/void av1_highbd_inv_txfm_add_32x32/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
-specialize qw/av1_highbd_inv_txfm_add_32x32 sse4_1 avx2/;
+add_proto qw/void av1_highbd_inv_txfm_add_4x8/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_4x8 sse4_1/;
+add_proto qw/void av1_highbd_inv_txfm_add_8x4/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_8x4 sse4_1/;
+add_proto qw/void av1_highbd_inv_txfm_add_4x16/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_4x16 sse4_1/;
+add_proto qw/void av1_highbd_inv_txfm_add_16x4/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_16x4 sse4_1/;
 
 add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
 add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
@@ -153,8 +158,12 @@ add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *out
 
 # directional intra predictor functions
 add_proto qw/void av1_highbd_dr_prediction_z1/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd";
+specialize qw/av1_highbd_dr_prediction_z1 avx2/;
 add_proto qw/void av1_highbd_dr_prediction_z2/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd";
+# TODO(https://crbug.com/aomedia/2288)
+#specialize qw/av1_highbd_dr_prediction_z2 avx2/;
 add_proto qw/void av1_highbd_dr_prediction_z3/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd";
+specialize qw/av1_highbd_dr_prediction_z3 avx2/;
 
 # build compound seg mask functions
 add_proto qw/void av1_build_compound_diffwtd_mask/, "uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const uint8_t *src0, int src0_stride, const uint8_t *src1, int src1_stride, int h, int w";
@@ -166,6 +175,10 @@ specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2/;
 add_proto qw/void av1_build_compound_diffwtd_mask_d16/, "uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const CONV_BUF_TYPE *src0, int src0_stride, const CONV_BUF_TYPE *src1, int src1_stride, int h, int w, ConvolveParams *conv_params, int bd";
 specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/;
 
+# Helper functions.
+add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit";
+specialize "av1_round_shift_array", qw/sse4_1 neon/;
+
 #
 # Encoder functions below this point.
 #
@@ -196,32 +209,44 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   specialize qw/av1_lowbd_fwd_txfm sse2 sse4_1 avx2/;
 
   add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_4x8 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_8x4 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_8x16 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2/;
   add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_16x8 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2/;
   add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_16x32 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_32x16 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_4x16 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_16x4 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_8x32 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_32x8 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   specialize qw/av1_fwd_txfm2d_4x4 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_8x8 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2/;
   add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_16x16 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2/;
   add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_32x32 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2/;
 
   add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  specialize qw/av1_fwd_txfm2d_64x64 sse4_1/;
+  specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2/;
   add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_32x64 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_64x32 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_16x64 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  specialize qw/av1_fwd_txfm2d_64x16 sse4_1/;
 
   #
   # Motion search
@@ -230,8 +255,9 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 
   add_proto qw/int av1_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, MV *ref_mv, MV *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const MV *center_mv";
 
-  add_proto qw/void av1_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
-  specialize qw/av1_temporal_filter_apply sse2 msa/;
+  # TODO(yunqing): Add back the optimizations.
+  # add_proto qw/void av1_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
+  # specialize qw/av1_temporal_filter_apply sse2 msa/;
 
   add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
 
@@ -240,7 +266,7 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   add_proto qw/int64_t av1_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
   specialize qw/av1_highbd_block_error sse2/;
 
-  add_proto qw/void av1_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
+  # add_proto qw/void av1_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
 
   add_proto qw/void av1_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
   specialize qw/av1_highbd_quantize_fp sse4_1 avx2/;
@@ -266,18 +292,30 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, uint8_t *p, int length";
   specialize qw/av1_get_crc32c_value sse4_2/;
 
-  add_proto qw/void av1_compute_stats/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride,  double *M, double *H";
+  add_proto qw/void av1_compute_stats/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H";
   specialize qw/av1_compute_stats sse4_1 avx2/;
 
+  add_proto qw/void av1_compute_stats_highbd/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H, aom_bit_depth_t bit_depth";
+  specialize qw/av1_compute_stats_highbd sse4_1 avx2/;
+
   add_proto qw/int64_t av1_lowbd_pixel_proj_error/, " const uint8_t *src8, int width, int height, int src_stride, const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride, int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params";
   specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2/;
+
+  add_proto qw/int64_t av1_highbd_pixel_proj_error/, " const uint8_t *src8, int width, int height, int src_stride, const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride, int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params";
+  specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2/;
+
+  add_proto qw/void av1_get_horver_correlation_full/, " const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr";
+  specialize qw/av1_get_horver_correlation_full sse4_1 avx2/;
+
+  add_proto qw/void av1_nn_predict/, " const float *input_nodes, const NN_CONFIG *const nn_config, float *const output";
+  specialize qw/av1_nn_predict sse3/;
 }
 # end encoder functions
 
 # Deringing Functions
 
 add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift";
-add_proto qw/void cdef_filter_block/, "uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int max, int coeff_shift";
+add_proto qw/void cdef_filter_block/, "uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift";
 
 add_proto qw/void copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int v, int h";
 add_proto qw/void copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int v, int h";
@@ -302,7 +340,7 @@ specialize qw/av1_highbd_warp_affine sse4_1/;
 
 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   add_proto qw/double compute_cross_correlation/, "unsigned char *im1, int stride1, int x1, int y1, unsigned char *im2, int stride2, int x2, int y2";
-  specialize qw/compute_cross_correlation sse4_1/;
+  specialize qw/compute_cross_correlation sse4_1 avx2/;
 }
 
 # LOOP_RESTORATION functions
@@ -321,18 +359,18 @@ add_proto qw/void av1_convolve_2d_sr/, "const uint8_t *src, int src_stride, uint
 add_proto qw/void av1_convolve_2d_copy_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
 add_proto qw/void av1_convolve_x_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
 add_proto qw/void av1_convolve_y_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
-add_proto qw/void av1_jnt_convolve_2d/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
-add_proto qw/void av1_jnt_convolve_2d_copy/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
-add_proto qw/void av1_jnt_convolve_x/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
-add_proto qw/void av1_jnt_convolve_y/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+add_proto qw/void av1_dist_wtd_convolve_2d/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+add_proto qw/void av1_dist_wtd_convolve_2d_copy/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+add_proto qw/void av1_dist_wtd_convolve_x/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+add_proto qw/void av1_dist_wtd_convolve_y/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
 add_proto qw/void av1_highbd_convolve_2d_copy_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
 add_proto qw/void av1_highbd_convolve_2d_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
 add_proto qw/void av1_highbd_convolve_x_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
 add_proto qw/void av1_highbd_convolve_y_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
-add_proto qw/void av1_highbd_jnt_convolve_2d/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
-add_proto qw/void av1_highbd_jnt_convolve_x/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
-add_proto qw/void av1_highbd_jnt_convolve_y/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
-add_proto qw/void av1_highbd_jnt_convolve_2d_copy/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
+add_proto qw/void av1_highbd_dist_wtd_convolve_2d/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
+add_proto qw/void av1_highbd_dist_wtd_convolve_x/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
+add_proto qw/void av1_highbd_dist_wtd_convolve_y/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
+add_proto qw/void av1_highbd_dist_wtd_convolve_2d_copy/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
 
   add_proto qw/void av1_convolve_2d_scale/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int x_step_qn, const int subpel_y_q4, const int y_step_qn, ConvolveParams *conv_params";
   add_proto qw/void av1_highbd_convolve_2d_scale/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_q4, const int x_step_qn, const int subpel_y_q4, const int y_step_qn, ConvolveParams *conv_params, int bd";
@@ -342,19 +380,19 @@ add_proto qw/void av1_highbd_jnt_convolve_2d_copy/, "const uint16_t *src, int sr
   specialize qw/av1_convolve_x_sr sse2 avx2 neon/;
   specialize qw/av1_convolve_y_sr sse2 avx2 neon/;
   specialize qw/av1_convolve_2d_scale sse4_1/;
-  specialize qw/av1_jnt_convolve_2d ssse3 avx2 neon/;
-  specialize qw/av1_jnt_convolve_2d_copy sse2 avx2 neon/;
-  specialize qw/av1_jnt_convolve_x sse2 avx2 neon/;
-  specialize qw/av1_jnt_convolve_y sse2 avx2 neon/;
+  specialize qw/av1_dist_wtd_convolve_2d sse2 ssse3 avx2 neon/;
+  specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/;
+  specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon/;
+  specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/;
   specialize qw/av1_highbd_convolve_2d_copy_sr sse2 avx2/;
   specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2/;
   specialize qw/av1_highbd_convolve_x_sr ssse3 avx2/;
   specialize qw/av1_highbd_convolve_y_sr ssse3 avx2/;
   specialize qw/av1_highbd_convolve_2d_scale sse4_1/;
-  specialize qw/av1_highbd_jnt_convolve_2d sse4_1 avx2/;
-  specialize qw/av1_highbd_jnt_convolve_x sse4_1 avx2/;
-  specialize qw/av1_highbd_jnt_convolve_y sse4_1 avx2/;
-  specialize qw/av1_highbd_jnt_convolve_2d_copy sse4_1 avx2/;
+  specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2/;
+  specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2/;
+  specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2/;
+  specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2/;
 
 # INTRA_EDGE functions
 add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
diff --git a/third_party/aom/av1/common/av1_txfm.c b/third_party/aom/av1/common/av1_txfm.c
index bb70eab703..ac43402f49 100644
--- a/third_party/aom/av1/common/av1_txfm.c
+++ b/third_party/aom/av1/common/av1_txfm.c
@@ -10,10 +10,11 @@
  */
 
 #include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
 
 #include "av1/common/av1_txfm.h"
 
-// av1_cospi_arr[i][j] = (int)round(cos(M_PI*j/128) * (1<<(cos_bit_min+i)));
+// av1_cospi_arr[i][j] = (int)round(cos(PI*j/128) * (1<<(cos_bit_min+i)));
 const int32_t av1_cospi_arr_data[7][64] = {
   { 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980,
     972,  964,  955,  946,  936,  926,  915,  903,  891,  878, 865, 851, 837,
diff --git a/third_party/aom/av1/common/av1_txfm.h b/third_party/aom/av1/common/av1_txfm.h
index 59d64ca4af..14e2c0e9c2 100644
--- a/third_party/aom/av1/common/av1_txfm.h
+++ b/third_party/aom/av1/common/av1_txfm.h
@@ -110,7 +110,7 @@ typedef void (*TxfmFunc)(const int32_t *input, int32_t *output, int8_t cos_bit,
 typedef void (*FwdTxfm2dFunc)(const int16_t *input, int32_t *output, int stride,
                               TX_TYPE tx_type, int bd);
 
-typedef enum TXFM_TYPE {
+enum {
   TXFM_TYPE_DCT4,
   TXFM_TYPE_DCT8,
   TXFM_TYPE_DCT16,
@@ -125,7 +125,7 @@ typedef enum TXFM_TYPE {
   TXFM_TYPE_IDENTITY32,
   TXFM_TYPES,
   TXFM_TYPE_INVALID,
-} TXFM_TYPE;
+} UENUM1BYTE(TXFM_TYPE);
 
 typedef struct TXFM_2D_FLIP_CFG {
   TX_SIZE tx_size;
diff --git a/third_party/aom/av1/common/blockd.h b/third_party/aom/av1/common/blockd.h
index a2311c1b00..1cb669e29b 100644
--- a/third_party/aom/av1/common/blockd.h
+++ b/third_party/aom/av1/common/blockd.h
@@ -38,19 +38,19 @@ extern "C" {
 #define MAX_DIFFWTD_MASK_BITS 1
 
 // DIFFWTD_MASK_TYPES should not surpass 1 << MAX_DIFFWTD_MASK_BITS
-typedef enum ATTRIBUTE_PACKED {
+enum {
   DIFFWTD_38 = 0,
   DIFFWTD_38_INV,
   DIFFWTD_MASK_TYPES,
-} DIFFWTD_MASK_TYPE;
+} UENUM1BYTE(DIFFWTD_MASK_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   KEY_FRAME = 0,
   INTER_FRAME = 1,
   INTRA_ONLY_FRAME = 2,  // replaces intra-only
   S_FRAME = 3,
   FRAME_TYPES,
-} FRAME_TYPE;
+} UENUM1BYTE(FRAME_TYPE);
 
 static INLINE int is_comp_ref_allowed(BLOCK_SIZE bsize) {
   return AOMMIN(block_size_wide[bsize], block_size_high[bsize]) >= 8;
@@ -156,18 +156,16 @@ static INLINE int is_masked_compound_type(COMPOUND_TYPE type) {
    modes for the Y blocks to the left and above us; for interframes, there
    is a single probability table. */
 
-typedef int8_t MV_REFERENCE_FRAME;
-
 typedef struct {
-  // Number of base colors for Y (0) and UV (1)
-  uint8_t palette_size[2];
   // Value of base colors for Y, U, and V
   uint16_t palette_colors[3 * PALETTE_MAX_SIZE];
+  // Number of base colors for Y (0) and UV (1)
+  uint8_t palette_size[2];
 } PALETTE_MODE_INFO;
 
 typedef struct {
-  uint8_t use_filter_intra;
   FILTER_INTRA_MODE filter_intra_mode;
+  uint8_t use_filter_intra;
 } FILTER_INTRA_MODE_INFO;
 
 static const PREDICTION_MODE fimode_to_intradir[FILTER_INTRA_MODES] = {
@@ -192,6 +190,11 @@ typedef struct RD_STATS {
   int64_t ref_rdcost;
   int zero_rate;
   uint8_t invalid_rate;
+#if CONFIG_ONE_PASS_SVM
+  int eob, eob_0, eob_1, eob_2, eob_3;
+  int64_t rd, rd_0, rd_1, rd_2, rd_3;
+  int64_t y_sse, sse_0, sse_1, sse_2, sse_3;
+#endif
 #if CONFIG_RD_DEBUG
   int txb_coeff_cost[MAX_MB_PLANE];
   int txb_coeff_cost_map[MAX_MB_PLANE][TXB_COEFF_COST_MAP_SIZE]
@@ -202,10 +205,10 @@ typedef struct RD_STATS {
 // This struct is used to group function args that are commonly
 // sent together in functions related to interinter compound modes
 typedef struct {
+  uint8_t *seg_mask;
   int wedge_index;
   int wedge_sign;
   DIFFWTD_MASK_TYPE mask_type;
-  uint8_t *seg_mask;
   COMPOUND_TYPE type;
 } INTERINTER_COMPOUND_DATA;
 
@@ -213,48 +216,18 @@ typedef struct {
 #define TXK_TYPE_BUF_LEN 64
 // This structure now relates to 4x4 block regions.
 typedef struct MB_MODE_INFO {
-  // Common for both INTER and INTRA blocks
-  BLOCK_SIZE sb_type;
-  PREDICTION_MODE mode;
-  TX_SIZE tx_size;
-  uint8_t inter_tx_size[INTER_TX_SIZE_BUF_LEN];
-  int8_t skip;
-  int8_t skip_mode;
-  int8_t segment_id;
-  int8_t seg_id_predicted;  // valid only when temporal_update is enabled
-
-  // Only for INTRA blocks
-  UV_PREDICTION_MODE uv_mode;
-
   PALETTE_MODE_INFO palette_mode_info;
-  uint8_t use_intrabc;
-
-  // Only for INTER blocks
-  InterpFilters interp_filters;
-  MV_REFERENCE_FRAME ref_frame[2];
-
-  TX_TYPE txk_type[TXK_TYPE_BUF_LEN];
-
-  FILTER_INTRA_MODE_INFO filter_intra_mode_info;
-
-  // The actual prediction angle is the base angle + (angle_delta * step).
-  int8_t angle_delta[PLANE_TYPES];
-
-  // interintra members
-  INTERINTRA_MODE interintra_mode;
-  // TODO(debargha): Consolidate these flags
-  int use_wedge_interintra;
-  int interintra_wedge_index;
-  int interintra_wedge_sign;
+  WarpedMotionParams wm_params;
   // interinter members
   INTERINTER_COMPOUND_DATA interinter_comp;
-  MOTION_MODE motion_mode;
-  int overlappable_neighbors[2];
+  FILTER_INTRA_MODE_INFO filter_intra_mode_info;
   int_mv mv[2];
-  uint8_t ref_mv_idx;
-  PARTITION_TYPE partition;
-  /* deringing gain *per-superblock* */
-  int8_t cdef_strength;
+  // Only for INTER blocks
+  InterpFilters interp_filters;
+  // TODO(debargha): Consolidate these flags
+  int interintra_wedge_index;
+  int interintra_wedge_sign;
+  int overlappable_neighbors[2];
   int current_qindex;
   int delta_lf_from_base;
   int delta_lf[FRAME_LF_COUNT];
@@ -264,15 +237,43 @@ typedef struct MB_MODE_INFO {
   int mi_col;
 #endif
   int num_proj_ref;
-  WarpedMotionParams wm_params;
 
   // Index of the alpha Cb and alpha Cr combination
   int cfl_alpha_idx;
   // Joint sign of alpha Cb and alpha Cr
   int cfl_alpha_signs;
 
-  int compound_idx;
+  // Indicate if masked compound is used(1) or not(0).
   int comp_group_idx;
+  // If comp_group_idx=0, indicate if dist_wtd_comp(0) or avg_comp(1) is used.
+  int compound_idx;
+#if CONFIG_INSPECTION
+  int16_t tx_skip[TXK_TYPE_BUF_LEN];
+#endif
+  // Common for both INTER and INTRA blocks
+  BLOCK_SIZE sb_type;
+  PREDICTION_MODE mode;
+  // Only for INTRA blocks
+  UV_PREDICTION_MODE uv_mode;
+  // interintra members
+  INTERINTRA_MODE interintra_mode;
+  MOTION_MODE motion_mode;
+  PARTITION_TYPE partition;
+  TX_TYPE txk_type[TXK_TYPE_BUF_LEN];
+  MV_REFERENCE_FRAME ref_frame[2];
+  int8_t use_wedge_interintra;
+  int8_t skip;
+  int8_t skip_mode;
+  uint8_t inter_tx_size[INTER_TX_SIZE_BUF_LEN];
+  TX_SIZE tx_size;
+  int8_t segment_id;
+  int8_t seg_id_predicted;  // valid only when temporal_update is enabled
+  uint8_t use_intrabc;
+  // The actual prediction angle is the base angle + (angle_delta * step).
+  int8_t angle_delta[PLANE_TYPES];
+  /* deringing gain *per-superblock* */
+  int8_t cdef_strength;
+  uint8_t ref_mv_idx;
 } MB_MODE_INFO;
 
 static INLINE int is_intrabc_block(const MB_MODE_INFO *mbmi) {
@@ -372,7 +373,7 @@ static INLINE void mi_to_pixel_loc(int *pixel_c, int *pixel_r, int mi_col,
 }
 #endif
 
-enum ATTRIBUTE_PACKED mv_precision { MV_PRECISION_Q3, MV_PRECISION_Q4 };
+enum { MV_PRECISION_Q3, MV_PRECISION_Q4 } UENUM1BYTE(mv_precision);
 
 struct buf_2d {
   uint8_t *buf;
@@ -406,7 +407,7 @@ typedef struct macroblockd_plane {
   ENTROPY_CONTEXT *above_context;
   ENTROPY_CONTEXT *left_context;
 
-  // The dequantizers below are true dequntizers used only in the
+  // The dequantizers below are true dequantizers used only in the
   // dequantization process.  They have the same coefficient
   // shift/scale as TX.
   int16_t seg_dequant_QTX[MAX_SEGMENTS][2];
@@ -428,13 +429,6 @@ typedef struct macroblockd_plane {
 #define BLOCK_OFFSET(x, i) \
   ((x) + (i) * (1 << (tx_size_wide_log2[0] + tx_size_high_log2[0])))
 
-typedef struct RefBuffer {
-  int idx;      // frame buf idx
-  int map_idx;  // frame map idx
-  YV12_BUFFER_CONFIG *buf;
-  struct scale_factors sf;
-} RefBuffer;
-
 typedef struct {
   DECLARE_ALIGNED(16, InterpKernel, vfilter);
   DECLARE_ALIGNED(16, InterpKernel, hfilter);
@@ -490,11 +484,13 @@ typedef struct cfl_ctx {
   int is_chroma_reference;
 } CFL_CTX;
 
-typedef struct jnt_comp_params {
-  int use_jnt_comp_avg;
+typedef struct dist_wtd_comp_params {
+  int use_dist_wtd_comp_avg;
   int fwd_offset;
   int bck_offset;
-} JNT_COMP_PARAMS;
+} DIST_WTD_COMP_PARAMS;
+
+struct scale_factors;
 
 // Most/all of the pointers are mere pointers to actual arrays are allocated
 // elsewhere. This is mostly for coding convenience.
@@ -522,8 +518,8 @@ typedef struct macroblockd {
   int mb_to_top_edge;
   int mb_to_bottom_edge;
 
-  /* pointers to reference frames */
-  const RefBuffer *block_refs[2];
+  /* pointers to reference frame scale factors */
+  const struct scale_factors *block_ref_scale_factors[2];
 
   /* pointer to current frame */
   const YV12_BUFFER_CONFIG *cur_buf;
@@ -592,7 +588,7 @@ typedef struct macroblockd {
   uint8_t *mc_buf[2];
   CFL_CTX cfl;
 
-  JNT_COMP_PARAMS jcp_param;
+  DIST_WTD_COMP_PARAMS jcp_param;
 
   uint16_t cb_offset[MAX_MB_PLANE];
   uint16_t txb_offset[MAX_MB_PLANE];
@@ -602,7 +598,7 @@ typedef struct macroblockd {
   uint8_t *tmp_obmc_bufs[2];
 } MACROBLOCKD;
 
-static INLINE int get_bitdepth_data_path_index(const MACROBLOCKD *xd) {
+static INLINE int is_cur_buf_hbd(const MACROBLOCKD *xd) {
   return xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH ? 1 : 0;
 }
 
@@ -646,19 +642,19 @@ static INLINE BLOCK_SIZE get_partition_subsize(BLOCK_SIZE bsize,
 static TX_TYPE intra_mode_to_tx_type(const MB_MODE_INFO *mbmi,
                                      PLANE_TYPE plane_type) {
   static const TX_TYPE _intra_mode_to_tx_type[INTRA_MODES] = {
-    DCT_DCT,    // DC
-    ADST_DCT,   // V
-    DCT_ADST,   // H
-    DCT_DCT,    // D45
-    ADST_ADST,  // D135
-    ADST_DCT,   // D117
-    DCT_ADST,   // D153
-    DCT_ADST,   // D207
-    ADST_DCT,   // D63
-    ADST_ADST,  // SMOOTH
-    ADST_DCT,   // SMOOTH_V
-    DCT_ADST,   // SMOOTH_H
-    ADST_ADST,  // PAETH
+    DCT_DCT,    // DC_PRED
+    ADST_DCT,   // V_PRED
+    DCT_ADST,   // H_PRED
+    DCT_DCT,    // D45_PRED
+    ADST_ADST,  // D135_PRED
+    ADST_DCT,   // D113_PRED
+    DCT_ADST,   // D157_PRED
+    DCT_ADST,   // D203_PRED
+    ADST_DCT,   // D67_PRED
+    ADST_ADST,  // SMOOTH_PRED
+    ADST_DCT,   // SMOOTH_V_PRED
+    DCT_ADST,   // SMOOTH_H_PRED
+    ADST_ADST,  // PAETH_PRED
   };
   const PREDICTION_MODE mode =
       (plane_type == PLANE_TYPE_Y) ? mbmi->mode : get_uv_mode(mbmi->uv_mode);
@@ -777,11 +773,13 @@ static INLINE int av1_raster_order_to_block_index(TX_SIZE tx_size,
 
 static INLINE TX_TYPE get_default_tx_type(PLANE_TYPE plane_type,
                                           const MACROBLOCKD *xd,
-                                          TX_SIZE tx_size) {
+                                          TX_SIZE tx_size,
+                                          int is_screen_content_type) {
   const MB_MODE_INFO *const mbmi = xd->mi[0];
 
   if (is_inter_block(mbmi) || plane_type != PLANE_TYPE_Y ||
-      xd->lossless[mbmi->segment_id] || tx_size >= TX_32X32)
+      xd->lossless[mbmi->segment_id] || tx_size >= TX_32X32 ||
+      is_screen_content_type)
     return DCT_DCT;
 
   return intra_mode_to_tx_type(mbmi, plane_type);
@@ -1045,7 +1043,8 @@ motion_mode_allowed(const WarpedMotionParams *gm_params, const MACROBLOCKD *xd,
     if (!check_num_overlappable_neighbors(mbmi)) return SIMPLE_TRANSLATION;
     assert(!has_second_ref(mbmi));
     if (mbmi->num_proj_ref >= 1 &&
-        (allow_warped_motion && !av1_is_scaled(&(xd->block_refs[0]->sf)))) {
+        (allow_warped_motion &&
+         !av1_is_scaled(xd->block_ref_scale_factors[0]))) {
       if (xd->cur_frame_force_integer_mv) {
         return OBMC_CAUSAL;
       }
diff --git a/third_party/aom/av1/common/cdef.c b/third_party/aom/av1/common/cdef.c
index e9e2b0e42c..556deded96 100644
--- a/third_party/aom/av1/common/cdef.c
+++ b/third_party/aom/av1/common/cdef.c
@@ -107,7 +107,7 @@ void copy_rect8_16bit_to_16bit_c(uint16_t *dst, int dstride,
   }
 }
 
-static void copy_sb8_16(AOM_UNUSED AV1_COMMON *cm, uint16_t *dst, int dstride,
+static void copy_sb8_16(AV1_COMMON *cm, uint16_t *dst, int dstride,
                         const uint8_t *src, int src_voffset, int src_hoffset,
                         int sstride, int vsize, int hsize) {
   if (cm->seq_params.use_highbitdepth) {
@@ -140,6 +140,7 @@ static INLINE void copy_rect(uint16_t *dst, int dstride, const uint16_t *src,
 
 void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
                     MACROBLOCKD *xd) {
+  const CdefInfo *const cdef_info = &cm->cdef_info;
   const int num_planes = av1_num_planes(cm);
   DECLARE_ALIGNED(16, uint16_t, src[CDEF_INBUF_SIZE]);
   uint16_t *linebuf[3];
@@ -231,13 +232,15 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
           cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
                               MI_SIZE_64X64 * fbc]
               ->cdef_strength;
-      level = cm->cdef_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
+      level =
+          cdef_info->cdef_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
       sec_strength =
-          cm->cdef_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
+          cdef_info->cdef_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
       sec_strength += sec_strength == 3;
-      uv_level = cm->cdef_uv_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
+      uv_level =
+          cdef_info->cdef_uv_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
       uv_sec_strength =
-          cm->cdef_uv_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
+          cdef_info->cdef_uv_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
       uv_sec_strength += uv_sec_strength == 3;
       if ((level == 0 && sec_strength == 0 && uv_level == 0 &&
            uv_sec_strength == 0) ||
@@ -252,8 +255,8 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
       for (int pli = 0; pli < num_planes; pli++) {
         int coffset;
         int rend, cend;
-        int pri_damping = cm->cdef_pri_damping;
-        int sec_damping = cm->cdef_sec_damping;
+        int pri_damping = cdef_info->cdef_pri_damping;
+        int sec_damping = cdef_info->cdef_sec_damping;
         int hsize = nhb << mi_wide_l2[pli];
         int vsize = nvb << mi_high_l2[pli];
 
diff --git a/third_party/aom/av1/common/cdef_block.c b/third_party/aom/av1/common/cdef_block.c
index df1de89be3..845df377b7 100644
--- a/third_party/aom/av1/common/cdef_block.c
+++ b/third_party/aom/av1/common/cdef_block.c
@@ -114,7 +114,7 @@ const int cdef_sec_taps[2][2] = { { 2, 1 }, { 2, 1 } };
 void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride,
                          const uint16_t *in, int pri_strength, int sec_strength,
                          int dir, int pri_damping, int sec_damping, int bsize,
-                         AOM_UNUSED int max_unused, int coeff_shift) {
+                         int coeff_shift) {
   int i, j, k;
   const int s = CDEF_BSTRIDE;
   const int *pri_taps = cdef_pri_taps[(pri_strength >> coeff_shift) & 1];
@@ -237,12 +237,11 @@ void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
     by = dlist[bi].by;
     bx = dlist[bi].bx;
     if (dst8)
-      cdef_filter_block(&dst8[(by << bsizey) * dstride + (bx << bsizex)], NULL,
-                        dstride,
-                        &in[(by * CDEF_BSTRIDE << bsizey) + (bx << bsizex)],
-                        (pli ? t : adjust_strength(t, var[by][bx])), s,
-                        t ? dir[by][bx] : 0, pri_damping, sec_damping, bsize,
-                        (256 << coeff_shift) - 1, coeff_shift);
+      cdef_filter_block(
+          &dst8[(by << bsizey) * dstride + (bx << bsizex)], NULL, dstride,
+          &in[(by * CDEF_BSTRIDE << bsizey) + (bx << bsizex)],
+          (pli ? t : adjust_strength(t, var[by][bx])), s, t ? dir[by][bx] : 0,
+          pri_damping, sec_damping, bsize, coeff_shift);
     else
       cdef_filter_block(
           NULL,
@@ -251,7 +250,6 @@ void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
           dirinit ? 1 << bsizex : dstride,
           &in[(by * CDEF_BSTRIDE << bsizey) + (bx << bsizex)],
           (pli ? t : adjust_strength(t, var[by][bx])), s, t ? dir[by][bx] : 0,
-          pri_damping, sec_damping, bsize, (256 << coeff_shift) - 1,
-          coeff_shift);
+          pri_damping, sec_damping, bsize, coeff_shift);
   }
 }
diff --git a/third_party/aom/av1/common/cdef_block.h b/third_party/aom/av1/common/cdef_block.h
index 6b4452cd64..0e921e0005 100644
--- a/third_party/aom/av1/common/cdef_block.h
+++ b/third_party/aom/av1/common/cdef_block.h
@@ -45,7 +45,7 @@ typedef void (*cdef_filter_block_func)(uint8_t *dst8, uint16_t *dst16,
                                        int dstride, const uint16_t *in,
                                        int pri_strength, int sec_strength,
                                        int dir, int pri_damping,
-                                       int sec_damping, int bsize, int max,
+                                       int sec_damping, int bsize,
                                        int coeff_shift);
 void copy_cdef_16bit_to_16bit(uint16_t *dst, int dstride, uint16_t *src,
                               cdef_list *dlist, int cdef_count, int bsize);
diff --git a/third_party/aom/av1/common/cdef_block_simd.h b/third_party/aom/av1/common/cdef_block_simd.h
index 14587a023a..a3368ec40b 100644
--- a/third_party/aom/av1/common/cdef_block_simd.h
+++ b/third_party/aom/av1/common/cdef_block_simd.h
@@ -226,7 +226,6 @@ void SIMD_FUNC(cdef_filter_block_4x4_8)(uint8_t *dst, int dstride,
                                         const uint16_t *in, int pri_strength,
                                         int sec_strength, int dir,
                                         int pri_damping, int sec_damping,
-                                        AOM_UNUSED int max_unused,
                                         int coeff_shift) {
   v128 p0, p1, p2, p3;
   v256 sum, row, tap, res;
@@ -393,7 +392,6 @@ void SIMD_FUNC(cdef_filter_block_8x8_8)(uint8_t *dst, int dstride,
                                         const uint16_t *in, int pri_strength,
                                         int sec_strength, int dir,
                                         int pri_damping, int sec_damping,
-                                        AOM_UNUSED int max_unused,
                                         int coeff_shift) {
   int i;
   v128 p0, p1, p2, p3;
@@ -541,7 +539,6 @@ void SIMD_FUNC(cdef_filter_block_4x4_16)(uint16_t *dst, int dstride,
                                          const uint16_t *in, int pri_strength,
                                          int sec_strength, int dir,
                                          int pri_damping, int sec_damping,
-                                         AOM_UNUSED int max_unused,
                                          int coeff_shift) {
   int i;
   v256 p0, p1, p2, p3, sum, row, res;
@@ -699,7 +696,6 @@ void SIMD_FUNC(cdef_filter_block_8x8_16)(uint16_t *dst, int dstride,
                                          const uint16_t *in, int pri_strength,
                                          int sec_strength, int dir,
                                          int pri_damping, int sec_damping,
-                                         AOM_UNUSED int max_unused,
                                          int coeff_shift) {
   int i;
   v256 sum, p0, p1, p2, p3, row, res;
@@ -833,56 +829,55 @@ void SIMD_FUNC(cdef_filter_block_8x8_16)(uint16_t *dst, int dstride,
 void SIMD_FUNC(cdef_filter_block)(uint8_t *dst8, uint16_t *dst16, int dstride,
                                   const uint16_t *in, int pri_strength,
                                   int sec_strength, int dir, int pri_damping,
-                                  int sec_damping, int bsize, int max,
-                                  int coeff_shift) {
+                                  int sec_damping, int bsize, int coeff_shift) {
   if (dst8) {
     if (bsize == BLOCK_8X8) {
       SIMD_FUNC(cdef_filter_block_8x8_8)
       (dst8, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     } else if (bsize == BLOCK_4X8) {
       SIMD_FUNC(cdef_filter_block_4x4_8)
       (dst8, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
       SIMD_FUNC(cdef_filter_block_4x4_8)
       (dst8 + 4 * dstride, dstride, in + 4 * CDEF_BSTRIDE, pri_strength,
-       sec_strength, dir, pri_damping, sec_damping, max, coeff_shift);
+       sec_strength, dir, pri_damping, sec_damping, coeff_shift);
     } else if (bsize == BLOCK_8X4) {
       SIMD_FUNC(cdef_filter_block_4x4_8)
       (dst8, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
       SIMD_FUNC(cdef_filter_block_4x4_8)
       (dst8 + 4, dstride, in + 4, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     } else {
       SIMD_FUNC(cdef_filter_block_4x4_8)
       (dst8, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     }
   } else {
     if (bsize == BLOCK_8X8) {
       SIMD_FUNC(cdef_filter_block_8x8_16)
       (dst16, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     } else if (bsize == BLOCK_4X8) {
       SIMD_FUNC(cdef_filter_block_4x4_16)
       (dst16, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
       SIMD_FUNC(cdef_filter_block_4x4_16)
       (dst16 + 4 * dstride, dstride, in + 4 * CDEF_BSTRIDE, pri_strength,
-       sec_strength, dir, pri_damping, sec_damping, max, coeff_shift);
+       sec_strength, dir, pri_damping, sec_damping, coeff_shift);
     } else if (bsize == BLOCK_8X4) {
       SIMD_FUNC(cdef_filter_block_4x4_16)
       (dst16, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
       SIMD_FUNC(cdef_filter_block_4x4_16)
       (dst16 + 4, dstride, in + 4, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     } else {
       assert(bsize == BLOCK_4X4);
       SIMD_FUNC(cdef_filter_block_4x4_16)
       (dst16, dstride, in, pri_strength, sec_strength, dir, pri_damping,
-       sec_damping, max, coeff_shift);
+       sec_damping, coeff_shift);
     }
   }
 }
diff --git a/third_party/aom/av1/common/cfl.c b/third_party/aom/av1/common/cfl.c
index ccc59b4eb7..65e18e89da 100644
--- a/third_party/aom/av1/common/cfl.c
+++ b/third_party/aom/av1/common/cfl.c
@@ -37,7 +37,7 @@ void cfl_store_dc_pred(MACROBLOCKD *const xd, const uint8_t *input,
   assert(pred_plane < CFL_PRED_PLANES);
   assert(width <= CFL_BUF_LINE);
 
-  if (get_bitdepth_data_path_index(xd)) {
+  if (is_cur_buf_hbd(xd)) {
     uint16_t *const input_16 = CONVERT_TO_SHORTPTR(input);
     memcpy(xd->cfl.dc_pred_cache[pred_plane], input_16, width << 1);
     return;
@@ -69,7 +69,7 @@ void cfl_load_dc_pred(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
   assert(pred_plane < CFL_PRED_PLANES);
   assert(width <= CFL_BUF_LINE);
   assert(height <= CFL_BUF_LINE);
-  if (get_bitdepth_data_path_index(xd)) {
+  if (is_cur_buf_hbd(xd)) {
     uint16_t *dst_16 = CONVERT_TO_SHORTPTR(dst);
     cfl_load_dc_pred_hbd(xd->cfl.dc_pred_cache[pred_plane], dst_16, dst_stride,
                          width, height);
@@ -158,16 +158,6 @@ static INLINE void cfl_predict_lbd_c(const int16_t *ac_buf_q3, uint8_t *dst,
   }
 }
 
-// Null function used for invalid tx_sizes
-void cfl_predict_lbd_null(const int16_t *ac_buf_q3, uint8_t *dst,
-                          int dst_stride, int alpha_q3) {
-  (void)ac_buf_q3;
-  (void)dst;
-  (void)dst_stride;
-  (void)alpha_q3;
-  assert(0);
-}
-
 CFL_PREDICT_FN(c, lbd)
 
 void cfl_predict_hbd_c(const int16_t *ac_buf_q3, uint16_t *dst, int dst_stride,
@@ -182,17 +172,6 @@ void cfl_predict_hbd_c(const int16_t *ac_buf_q3, uint16_t *dst, int dst_stride,
   }
 }
 
-// Null function used for invalid tx_sizes
-void cfl_predict_hbd_null(const int16_t *ac_buf_q3, uint16_t *dst,
-                          int dst_stride, int alpha_q3, int bd) {
-  (void)ac_buf_q3;
-  (void)dst;
-  (void)dst_stride;
-  (void)alpha_q3;
-  (void)bd;
-  assert(0);
-}
-
 CFL_PREDICT_FN(c, hbd)
 
 static void cfl_compute_parameters(MACROBLOCKD *const xd, TX_SIZE tx_size) {
@@ -217,7 +196,7 @@ void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
       cfl_idx_to_alpha(mbmi->cfl_alpha_idx, mbmi->cfl_alpha_signs, plane - 1);
   assert((tx_size_high[tx_size] - 1) * CFL_BUF_LINE + tx_size_wide[tx_size] <=
          CFL_BUF_SQUARE);
-  if (get_bitdepth_data_path_index(xd)) {
+  if (is_cur_buf_hbd(xd)) {
     uint16_t *dst_16 = CONVERT_TO_SHORTPTR(dst);
     get_predict_hbd_fn(tx_size)(cfl->ac_buf_q3, dst_16, dst_stride, alpha_q3,
                                 xd->bd);
@@ -226,24 +205,6 @@ void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
   get_predict_lbd_fn(tx_size)(cfl->ac_buf_q3, dst, dst_stride, alpha_q3);
 }
 
-// Null function used for invalid tx_sizes
-void cfl_subsample_lbd_null(const uint8_t *input, int input_stride,
-                            uint16_t *output_q3) {
-  (void)input;
-  (void)input_stride;
-  (void)output_q3;
-  assert(0);
-}
-
-// Null function used for invalid tx_sizes
-void cfl_subsample_hbd_null(const uint16_t *input, int input_stride,
-                            uint16_t *output_q3) {
-  (void)input;
-  (void)input_stride;
-  (void)output_q3;
-  assert(0);
-}
-
 static void cfl_luma_subsampling_420_lbd_c(const uint8_t *input,
                                            int input_stride,
                                            uint16_t *output_q3, int width,
@@ -427,8 +388,7 @@ void cfl_store_tx(MACROBLOCKD *const xd, int row, int col, TX_SIZE tx_size,
     assert(!((row & 1) && tx_size_high[tx_size] != 4));
     sub8x8_adjust_offset(cfl, &row, &col);
   }
-  cfl_store(cfl, dst, pd->dst.stride, row, col, tx_size,
-            get_bitdepth_data_path_index(xd));
+  cfl_store(cfl, dst, pd->dst.stride, row, col, tx_size, is_cur_buf_hbd(xd));
 }
 
 void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size) {
@@ -444,5 +404,5 @@ void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size) {
   const int height = max_intra_block_height(xd, bsize, AOM_PLANE_Y, tx_size);
   tx_size = get_tx_size(width, height);
   cfl_store(cfl, pd->dst.buf, pd->dst.stride, row, col, tx_size,
-            get_bitdepth_data_path_index(xd));
+            is_cur_buf_hbd(xd));
 }
diff --git a/third_party/aom/av1/common/cfl.h b/third_party/aom/av1/common/cfl.h
index d627891bf5..3b91d85d1d 100644
--- a/third_party/aom/av1/common/cfl.h
+++ b/third_party/aom/av1/common/cfl.h
@@ -80,14 +80,6 @@ void cfl_store_dc_pred(MACROBLOCKD *const xd, const uint8_t *input,
 void cfl_load_dc_pred(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
                       TX_SIZE tx_size, CFL_PRED_TYPE pred_plane);
 
-// Null function used for invalid tx_sizes
-void cfl_subsample_lbd_null(const uint8_t *input, int input_stride,
-                            uint16_t *output_q3);
-
-// Null function used for invalid tx_sizes
-void cfl_subsample_hbd_null(const uint16_t *input, int input_stride,
-                            uint16_t *output_q3);
-
 // Allows the CFL_SUBSAMPLE function to switch types depending on the bitdepth.
 #define CFL_lbd_TYPE uint8_t *cfl_type
 #define CFL_hbd_TYPE uint16_t *cfl_type
@@ -133,21 +125,21 @@ void cfl_subsample_hbd_null(const uint16_t *input, int input_stride,
     subsample_##bd##_##sub##_8x8_##arch,   /* 8x8 */                      \
     subsample_##bd##_##sub##_16x16_##arch, /* 16x16 */                    \
     subsample_##bd##_##sub##_32x32_##arch, /* 32x32 */                    \
-    cfl_subsample_##bd##_null,             /* 64x64 (invalid CFL size) */ \
+    NULL,                                  /* 64x64 (invalid CFL size) */ \
     subsample_##bd##_##sub##_4x8_##arch,   /* 4x8 */                      \
     subsample_##bd##_##sub##_8x4_##arch,   /* 8x4 */                      \
     subsample_##bd##_##sub##_8x16_##arch,  /* 8x16 */                     \
     subsample_##bd##_##sub##_16x8_##arch,  /* 16x8 */                     \
     subsample_##bd##_##sub##_16x32_##arch, /* 16x32 */                    \
     subsample_##bd##_##sub##_32x16_##arch, /* 32x16 */                    \
-    cfl_subsample_##bd##_null,             /* 32x64 (invalid CFL size) */ \
-    cfl_subsample_##bd##_null,             /* 64x32 (invalid CFL size) */ \
+    NULL,                                  /* 32x64 (invalid CFL size) */ \
+    NULL,                                  /* 64x32 (invalid CFL size) */ \
     subsample_##bd##_##sub##_4x16_##arch,  /* 4x16  */                    \
     subsample_##bd##_##sub##_16x4_##arch,  /* 16x4  */                    \
     subsample_##bd##_##sub##_8x32_##arch,  /* 8x32  */                    \
     subsample_##bd##_##sub##_32x8_##arch,  /* 32x8  */                    \
-    cfl_subsample_##bd##_null,             /* 16x64 (invalid CFL size) */ \
-    cfl_subsample_##bd##_null,             /* 64x16 (invalid CFL size) */ \
+    NULL,                                  /* 16x64 (invalid CFL size) */ \
+    NULL,                                  /* 64x16 (invalid CFL size) */ \
   };
 
 // The RTCD script does not support passing in an array, so we wrap it in this
@@ -160,14 +152,6 @@ void cfl_subsample_hbd_null(const uint16_t *input, int input_stride,
   CFL_SUBSAMPLE_FUNCTIONS(arch, 422, hbd) \
   CFL_SUBSAMPLE_FUNCTIONS(arch, 444, hbd)
 
-// Null function used for invalid tx_sizes
-static INLINE void cfl_subtract_average_null(const uint16_t *src,
-                                             int16_t *dst) {
-  (void)dst;
-  (void)src;
-  assert(0);
-}
-
 // Declare a size-specific wrapper for the size-generic function. The compiler
 // will inline the size generic function in here, the advantage is that the size
 // will be constant allowing for loop unrolling and other constant propagated
@@ -201,21 +185,21 @@ static INLINE void cfl_subtract_average_null(const uint16_t *src,
       subtract_average_8x8_##arch,   /* 8x8 */                              \
       subtract_average_16x16_##arch, /* 16x16 */                            \
       subtract_average_32x32_##arch, /* 32x32 */                            \
-      cfl_subtract_average_null,     /* 64x64 (invalid CFL size) */         \
+      NULL,                          /* 64x64 (invalid CFL size) */         \
       subtract_average_4x8_##arch,   /* 4x8 */                              \
       subtract_average_8x4_##arch,   /* 8x4 */                              \
       subtract_average_8x16_##arch,  /* 8x16 */                             \
       subtract_average_16x8_##arch,  /* 16x8 */                             \
       subtract_average_16x32_##arch, /* 16x32 */                            \
       subtract_average_32x16_##arch, /* 32x16 */                            \
-      cfl_subtract_average_null,     /* 32x64 (invalid CFL size) */         \
-      cfl_subtract_average_null,     /* 64x32 (invalid CFL size) */         \
+      NULL,                          /* 32x64 (invalid CFL size) */         \
+      NULL,                          /* 64x32 (invalid CFL size) */         \
       subtract_average_4x16_##arch,  /* 4x16 (invalid CFL size) */          \
       subtract_average_16x4_##arch,  /* 16x4 (invalid CFL size) */          \
       subtract_average_8x32_##arch,  /* 8x32 (invalid CFL size) */          \
       subtract_average_32x8_##arch,  /* 32x8 (invalid CFL size) */          \
-      cfl_subtract_average_null,     /* 16x64 (invalid CFL size) */         \
-      cfl_subtract_average_null,     /* 64x16 (invalid CFL size) */         \
+      NULL,                          /* 16x64 (invalid CFL size) */         \
+      NULL,                          /* 64x16 (invalid CFL size) */         \
     };                                                                      \
     /* Modulo TX_SIZES_ALL to ensure that an attacker won't be able to */   \
     /* index the function pointer array out of bounds. */                   \
@@ -249,14 +233,6 @@ void subtract_average_4x16_c(const uint16_t *src, int16_t *dst);
 #define CFL_PREDICT_X(arch, width, height, bd) \
   CFL_PREDICT_##bd(arch, width, height)
 
-// Null function used for invalid tx_sizes
-void cfl_predict_lbd_null(const int16_t *pred_buf_q3, uint8_t *dst,
-                          int dst_stride, int alpha_q3);
-
-// Null function used for invalid tx_sizes
-void cfl_predict_hbd_null(const int16_t *pred_buf_q3, uint16_t *dst,
-                          int dst_stride, int alpha_q3, int bd);
-
 #define CFL_PREDICT_FN(arch, bd)                                          \
   CFL_PREDICT_X(arch, 4, 4, bd)                                           \
   CFL_PREDICT_X(arch, 4, 8, bd)                                           \
@@ -278,21 +254,21 @@ void cfl_predict_hbd_null(const int16_t *pred_buf_q3, uint16_t *dst,
       predict_##bd##_8x8_##arch,   /* 8x8 */                              \
       predict_##bd##_16x16_##arch, /* 16x16 */                            \
       predict_##bd##_32x32_##arch, /* 32x32 */                            \
-      cfl_predict_##bd##_null,     /* 64x64 (invalid CFL size) */         \
+      NULL,                        /* 64x64 (invalid CFL size) */         \
       predict_##bd##_4x8_##arch,   /* 4x8 */                              \
       predict_##bd##_8x4_##arch,   /* 8x4 */                              \
       predict_##bd##_8x16_##arch,  /* 8x16 */                             \
       predict_##bd##_16x8_##arch,  /* 16x8 */                             \
       predict_##bd##_16x32_##arch, /* 16x32 */                            \
       predict_##bd##_32x16_##arch, /* 32x16 */                            \
-      cfl_predict_##bd##_null,     /* 32x64 (invalid CFL size) */         \
-      cfl_predict_##bd##_null,     /* 64x32 (invalid CFL size) */         \
+      NULL,                        /* 32x64 (invalid CFL size) */         \
+      NULL,                        /* 64x32 (invalid CFL size) */         \
       predict_##bd##_4x16_##arch,  /* 4x16  */                            \
       predict_##bd##_16x4_##arch,  /* 16x4  */                            \
       predict_##bd##_8x32_##arch,  /* 8x32  */                            \
       predict_##bd##_32x8_##arch,  /* 32x8  */                            \
-      cfl_predict_##bd##_null,     /* 16x64 (invalid CFL size) */         \
-      cfl_predict_##bd##_null,     /* 64x16 (invalid CFL size) */         \
+      NULL,                        /* 16x64 (invalid CFL size) */         \
+      NULL,                        /* 64x16 (invalid CFL size) */         \
     };                                                                    \
     /* Modulo TX_SIZES_ALL to ensure that an attacker won't be able to */ \
     /* index the function pointer array out of bounds. */                 \
diff --git a/third_party/aom/av1/common/convolve.c b/third_party/aom/av1/common/convolve.c
index 1f11126fc3..5a55ece671 100644
--- a/third_party/aom/av1/common/convolve.c
+++ b/third_party/aom/av1/common/convolve.c
@@ -73,6 +73,45 @@ void av1_highbd_convolve_horiz_rs_c(const uint16_t *src, int src_stride,
   }
 }
 
+void av1_convolve_2d_sobel_y_c(const uint8_t *src, int src_stride, double *dst,
+                               int dst_stride, int w, int h, int dir,
+                               double norm) {
+  int16_t im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
+  DECLARE_ALIGNED(256, static const int16_t, sobel_a[3]) = { 1, 0, -1 };
+  DECLARE_ALIGNED(256, static const int16_t, sobel_b[3]) = { 1, 2, 1 };
+  const int taps = 3;
+  int im_h = h + taps - 1;
+  int im_stride = w;
+  const int fo_vert = 1;
+  const int fo_horiz = 1;
+
+  // horizontal filter
+  const uint8_t *src_horiz = src - fo_vert * src_stride;
+  const int16_t *x_filter = dir ? sobel_a : sobel_b;
+  for (int y = 0; y < im_h; ++y) {
+    for (int x = 0; x < w; ++x) {
+      int16_t sum = 0;
+      for (int k = 0; k < taps; ++k) {
+        sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
+      }
+      im_block[y * im_stride + x] = sum;
+    }
+  }
+
+  // vertical filter
+  int16_t *src_vert = im_block + fo_vert * im_stride;
+  const int16_t *y_filter = dir ? sobel_b : sobel_a;
+  for (int y = 0; y < h; ++y) {
+    for (int x = 0; x < w; ++x) {
+      int16_t sum = 0;
+      for (int k = 0; k < taps; ++k) {
+        sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
+      }
+      dst[y * dst_stride + x] = sum * norm;
+    }
+  }
+}
+
 void av1_convolve_2d_sr_c(const uint8_t *src, int src_stride, uint8_t *dst,
                           int dst_stride, int w, int h,
                           const InterpFilterParams *filter_params_x,
@@ -199,16 +238,16 @@ void av1_convolve_2d_copy_sr_c(const uint8_t *src, int src_stride, uint8_t *dst,
   (void)conv_params;
 
   for (int y = 0; y < h; ++y) {
-    memcpy(dst + y * dst_stride, src + y * src_stride, w * sizeof(src[0]));
+    memmove(dst + y * dst_stride, src + y * src_stride, w * sizeof(src[0]));
   }
 }
 
-void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst8,
-                           int dst8_stride, int w, int h,
-                           const InterpFilterParams *filter_params_x,
-                           const InterpFilterParams *filter_params_y,
-                           const int subpel_x_q4, const int subpel_y_q4,
-                           ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_c(const uint8_t *src, int src_stride,
+                                uint8_t *dst8, int dst8_stride, int w, int h,
+                                const InterpFilterParams *filter_params_x,
+                                const InterpFilterParams *filter_params_y,
+                                const int subpel_x_q4, const int subpel_y_q4,
+                                ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   int16_t im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
@@ -251,7 +290,7 @@ void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst8,
       CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -269,12 +308,12 @@ void av1_jnt_convolve_2d_c(const uint8_t *src, int src_stride, uint8_t *dst8,
   }
 }
 
-void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst8,
-                          int dst8_stride, int w, int h,
-                          const InterpFilterParams *filter_params_x,
-                          const InterpFilterParams *filter_params_y,
-                          const int subpel_x_q4, const int subpel_y_q4,
-                          ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_y_c(const uint8_t *src, int src_stride,
+                               uint8_t *dst8, int dst8_stride, int w, int h,
+                               const InterpFilterParams *filter_params_x,
+                               const InterpFilterParams *filter_params_y,
+                               const int subpel_x_q4, const int subpel_y_q4,
+                               ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int fo_vert = filter_params_y->taps / 2 - 1;
@@ -302,7 +341,7 @@ void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst8,
 
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -319,12 +358,12 @@ void av1_jnt_convolve_y_c(const uint8_t *src, int src_stride, uint8_t *dst8,
   }
 }
 
-void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst8,
-                          int dst8_stride, int w, int h,
-                          const InterpFilterParams *filter_params_x,
-                          const InterpFilterParams *filter_params_y,
-                          const int subpel_x_q4, const int subpel_y_q4,
-                          ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_x_c(const uint8_t *src, int src_stride,
+                               uint8_t *dst8, int dst8_stride, int w, int h,
+                               const InterpFilterParams *filter_params_x,
+                               const InterpFilterParams *filter_params_y,
+                               const int subpel_x_q4, const int subpel_y_q4,
+                               ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int fo_horiz = filter_params_x->taps / 2 - 1;
@@ -352,7 +391,7 @@ void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst8,
 
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -369,12 +408,11 @@ void av1_jnt_convolve_x_c(const uint8_t *src, int src_stride, uint8_t *dst8,
   }
 }
 
-void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride,
-                                uint8_t *dst8, int dst8_stride, int w, int h,
-                                const InterpFilterParams *filter_params_x,
-                                const InterpFilterParams *filter_params_y,
-                                const int subpel_x_q4, const int subpel_y_q4,
-                                ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_copy_c(
+    const uint8_t *src, int src_stride, uint8_t *dst8, int dst8_stride, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int bits =
@@ -395,7 +433,7 @@ void av1_jnt_convolve_2d_copy_c(const uint8_t *src, int src_stride,
 
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -472,7 +510,7 @@ void av1_convolve_2d_scale_c(const uint8_t *src, int src_stride, uint8_t *dst8,
       if (conv_params->is_compound) {
         if (conv_params->do_average) {
           int32_t tmp = dst16[y * dst16_stride + x];
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
             tmp = tmp >> DIST_PRECISION_BITS;
           } else {
@@ -593,7 +631,7 @@ void av1_highbd_convolve_2d_copy_sr_c(
   (void)bd;
 
   for (int y = 0; y < h; ++y) {
-    memcpy(dst + y * dst_stride, src + y * src_stride, w * sizeof(src[0]));
+    memmove(dst + y * dst_stride, src + y * src_stride, w * sizeof(src[0]));
   }
 }
 
@@ -709,13 +747,11 @@ void av1_highbd_convolve_2d_sr_c(const uint16_t *src, int src_stride,
   }
 }
 
-void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride,
-                                  uint16_t *dst16, int dst16_stride, int w,
-                                  int h,
-                                  const InterpFilterParams *filter_params_x,
-                                  const InterpFilterParams *filter_params_y,
-                                  const int subpel_x_q4, const int subpel_y_q4,
-                                  ConvolveParams *conv_params, int bd) {
+void av1_highbd_dist_wtd_convolve_2d_c(
+    const uint16_t *src, int src_stride, uint16_t *dst16, int dst16_stride,
+    int w, int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
   int x, y, k;
   int16_t im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
   CONV_BUF_TYPE *dst = conv_params->dst;
@@ -760,7 +796,7 @@ void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride,
       CONV_BUF_TYPE res = ROUND_POWER_OF_TWO(sum, conv_params->round_1);
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -778,13 +814,11 @@ void av1_highbd_jnt_convolve_2d_c(const uint16_t *src, int src_stride,
   }
 }
 
-void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride,
-                                 uint16_t *dst16, int dst16_stride, int w,
-                                 int h,
-                                 const InterpFilterParams *filter_params_x,
-                                 const InterpFilterParams *filter_params_y,
-                                 const int subpel_x_q4, const int subpel_y_q4,
-                                 ConvolveParams *conv_params, int bd) {
+void av1_highbd_dist_wtd_convolve_x_c(
+    const uint16_t *src, int src_stride, uint16_t *dst16, int dst16_stride,
+    int w, int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int fo_horiz = filter_params_x->taps / 2 - 1;
@@ -812,7 +846,7 @@ void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride,
 
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -829,13 +863,11 @@ void av1_highbd_jnt_convolve_x_c(const uint16_t *src, int src_stride,
   }
 }
 
-void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride,
-                                 uint16_t *dst16, int dst16_stride, int w,
-                                 int h,
-                                 const InterpFilterParams *filter_params_x,
-                                 const InterpFilterParams *filter_params_y,
-                                 const int subpel_x_q4, const int subpel_y_q4,
-                                 ConvolveParams *conv_params, int bd) {
+void av1_highbd_dist_wtd_convolve_y_c(
+    const uint16_t *src, int src_stride, uint16_t *dst16, int dst16_stride,
+    int w, int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int fo_vert = filter_params_y->taps / 2 - 1;
@@ -863,7 +895,7 @@ void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride,
 
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -880,7 +912,7 @@ void av1_highbd_jnt_convolve_y_c(const uint16_t *src, int src_stride,
   }
 }
 
-void av1_highbd_jnt_convolve_2d_copy_c(
+void av1_highbd_dist_wtd_convolve_2d_copy_c(
     const uint16_t *src, int src_stride, uint16_t *dst16, int dst16_stride,
     int w, int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -904,7 +936,7 @@ void av1_highbd_jnt_convolve_2d_copy_c(
       res += round_offset;
       if (conv_params->do_average) {
         int32_t tmp = dst[y * dst_stride + x];
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
           tmp = tmp >> DIST_PRECISION_BITS;
         } else {
@@ -980,7 +1012,7 @@ void av1_highbd_convolve_2d_scale_c(const uint16_t *src, int src_stride,
       if (conv_params->is_compound) {
         if (conv_params->do_average) {
           int32_t tmp = dst16[y * dst16_stride + x];
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
             tmp = tmp >> DIST_PRECISION_BITS;
           } else {
diff --git a/third_party/aom/av1/common/convolve.h b/third_party/aom/av1/common/convolve.h
index 4109dd8433..e5479e6664 100644
--- a/third_party/aom/av1/common/convolve.h
+++ b/third_party/aom/av1/common/convolve.h
@@ -26,7 +26,7 @@ typedef struct ConvolveParams {
   int round_1;
   int plane;
   int is_compound;
-  int use_jnt_comp_avg;
+  int use_dist_wtd_comp_avg;
   int fwd_offset;
   int bck_offset;
 } ConvolveParams;
@@ -118,6 +118,11 @@ void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
                                    const struct scale_factors *sf,
                                    int is_intrabc, int bd);
 
+// TODO(sarahparker) This will need to be integerized and optimized
+void av1_convolve_2d_sobel_y_c(const uint8_t *src, int src_stride, double *dst,
+                               int dst_stride, int w, int h, int dir,
+                               double norm);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/common/debugmodes.c b/third_party/aom/av1/common/debugmodes.c
index 868f341b5b..b26c7ddaa2 100644
--- a/third_party/aom/av1/common/debugmodes.c
+++ b/third_party/aom/av1/common/debugmodes.c
@@ -17,7 +17,7 @@
 
 static void log_frame_info(AV1_COMMON *cm, const char *str, FILE *f) {
   fprintf(f, "%s", str);
-  fprintf(f, "(Frame %d, Show:%d, Q:%d): \n", cm->current_video_frame,
+  fprintf(f, "(Frame %d, Show:%d, Q:%d): \n", cm->current_frame.frame_number,
           cm->show_frame, cm->base_qindex);
 }
 /* This function dereferences a pointer to the mbmi structure
@@ -40,7 +40,7 @@ static void print_mi_data(AV1_COMMON *cm, FILE *file, const char *descriptor,
       mi++;
     }
     fprintf(file, "\n");
-    mi += MAX_MIB_SIZE;
+    mi += cm->mi_stride - cols;
   }
   fprintf(file, "\n");
 }
@@ -68,7 +68,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
       mi++;
     }
     fprintf(mvs, "\n");
-    mi += MAX_MIB_SIZE;
+    mi += cm->mi_stride - cols;
   }
   fprintf(mvs, "\n");
 
@@ -82,7 +82,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
       mi++;
     }
     fprintf(mvs, "\n");
-    mi += MAX_MIB_SIZE;
+    mi += cm->mi_stride - cols;
   }
   fprintf(mvs, "\n");
 
diff --git a/third_party/aom/av1/common/entropy.h b/third_party/aom/av1/common/entropy.h
index 991692c2f0..41218d30e7 100644
--- a/third_party/aom/av1/common/entropy.h
+++ b/third_party/aom/av1/common/entropy.h
@@ -54,12 +54,12 @@ extern "C" {
 
 #define BASE_CONTEXT_POSITION_NUM 12
 
-typedef enum TX_CLASS {
+enum {
   TX_CLASS_2D = 0,
   TX_CLASS_HORIZ = 1,
   TX_CLASS_VERT = 2,
   TX_CLASSES = 3,
-} TX_CLASS;
+} UENUM1BYTE(TX_CLASS);
 
 #define DCT_MAX_VALUE 16384
 #define DCT_MAX_VALUE_HIGH10 65536
diff --git a/third_party/aom/av1/common/entropymode.c b/third_party/aom/av1/common/entropymode.c
index 41dc30ddb8..8e7e952e2c 100644
--- a/third_party/aom/av1/common/entropymode.c
+++ b/third_party/aom/av1/common/entropymode.c
@@ -1068,9 +1068,16 @@ void av1_setup_frame_contexts(AV1_COMMON *cm) {
   // This function must ONLY be called when cm->fc has been initialized with
   // default probs, either by av1_setup_past_independence or after manually
   // initializing them
-  cm->frame_contexts[FRAME_CONTEXT_DEFAULTS] = *cm->fc;
+  *cm->default_frame_context = *cm->fc;
+  // TODO(jack.haughton@argondesign.com): don't think this should be necessary,
+  // but could do with fuller testing
   if (cm->large_scale_tile) {
-    for (int i = 0; i < FRAME_CONTEXTS; ++i) cm->frame_contexts[i] = *cm->fc;
+    for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+      RefCntBuffer *const buf = get_ref_frame_buf(cm, i);
+      if (buf != NULL) buf->frame_context = *cm->fc;
+    }
+    for (int i = 0; i < FRAME_BUFFERS; ++i)
+      cm->buffer_pool->frame_bufs[i].frame_context = *cm->fc;
   }
 }
 
@@ -1079,10 +1086,8 @@ void av1_setup_past_independence(AV1_COMMON *cm) {
   // Features disabled, 0, with delta coding (Default state).
   av1_clearall_segfeatures(&cm->seg);
 
-  cm->current_frame_seg_map = cm->cur_frame->seg_map;
-
-  if (cm->current_frame_seg_map)
-    memset(cm->current_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
+  if (cm->cur_frame->seg_map)
+    memset(cm->cur_frame->seg_map, 0, (cm->mi_rows * cm->mi_cols));
 
   // reset mode ref deltas
   av1_set_default_ref_deltas(cm->cur_frame->ref_deltas);
@@ -1092,7 +1097,6 @@ void av1_setup_past_independence(AV1_COMMON *cm) {
   av1_default_coef_probs(cm);
   init_mode_probs(cm->fc);
   av1_init_mv_probs(cm);
-  av1_init_lv_map(cm);
   cm->fc->initialized = 1;
   av1_setup_frame_contexts(cm);
 
diff --git a/third_party/aom/av1/common/entropymv.h b/third_party/aom/av1/common/entropymv.h
index fa818a2c16..cddc80768c 100644
--- a/third_party/aom/av1/common/entropymv.h
+++ b/third_party/aom/av1/common/entropymv.h
@@ -30,12 +30,12 @@ void av1_init_mv_probs(struct AV1Common *cm);
 
 /* Symbols for coding which components are zero jointly */
 #define MV_JOINTS 4
-typedef enum {
+enum {
   MV_JOINT_ZERO = 0,   /* Zero vector */
   MV_JOINT_HNZVZ = 1,  /* Vert zero, hor nonzero */
   MV_JOINT_HZVNZ = 2,  /* Hor zero, vert nonzero */
   MV_JOINT_HNZVNZ = 3, /* Both components nonzero */
-} MV_JOINT_TYPE;
+} UENUM1BYTE(MV_JOINT_TYPE);
 
 static INLINE int mv_joint_vertical(MV_JOINT_TYPE type) {
   return type == MV_JOINT_HZVNZ || type == MV_JOINT_HNZVNZ;
@@ -47,7 +47,7 @@ static INLINE int mv_joint_horizontal(MV_JOINT_TYPE type) {
 
 /* Symbols for coding magnitude class of nonzero components */
 #define MV_CLASSES 11
-typedef enum {
+enum {
   MV_CLASS_0 = 0,   /* (0, 2]     integer pel */
   MV_CLASS_1 = 1,   /* (2, 4]     integer pel */
   MV_CLASS_2 = 2,   /* (4, 8]     integer pel */
@@ -59,7 +59,7 @@ typedef enum {
   MV_CLASS_8 = 8,   /* (256, 512] integer pel */
   MV_CLASS_9 = 9,   /* (512, 1024] integer pel */
   MV_CLASS_10 = 10, /* (1024,2048] integer pel */
-} MV_CLASS_TYPE;
+} UENUM1BYTE(MV_CLASS_TYPE);
 
 #define CLASS0_BITS 1 /* bits at integer precision for class 0 */
 #define CLASS0_SIZE (1 << CLASS0_BITS)
@@ -91,11 +91,11 @@ typedef struct {
   nmv_component comps[2];
 } nmv_context;
 
-typedef enum {
+enum {
   MV_SUBPEL_NONE = -1,
   MV_SUBPEL_LOW_PRECISION = 0,
   MV_SUBPEL_HIGH_PRECISION,
-} MvSubpelPrecision;
+} SENUM1BYTE(MvSubpelPrecision);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/common/enums.h b/third_party/aom/av1/common/enums.h
index 869c06ef2f..9af7b0f366 100644
--- a/third_party/aom/av1/common/enums.h
+++ b/third_party/aom/av1/common/enums.h
@@ -16,6 +16,7 @@
 
 #include "aom/aom_codec.h"
 #include "aom/aom_integer.h"
+#include "aom_ports/mem.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -63,17 +64,6 @@ extern "C" {
 #define FRAME_OFFSET_BITS 5
 #define MAX_FRAME_DISTANCE ((1 << FRAME_OFFSET_BITS) - 1)
 
-#define REF_FRAMES_LOG2 3
-#define REF_FRAMES (1 << REF_FRAMES_LOG2)
-
-// 4 scratch frames for the new frames to support a maximum of 4 cores decoding
-// in parallel, 3 for scaled references on the encoder.
-// TODO(hkuang): Add ondemand frame buffers instead of hardcoding the number
-// of framebuffers.
-// TODO(jkoleszar): These 3 extra references could probably come from the
-// normal reference pool.
-#define FRAME_BUFFERS (REF_FRAMES + 7)
-
 // 4 frame filter levels: y plane vertical, y plane horizontal,
 // u plane, and v plane
 #define FRAME_LF_COUNT 4
@@ -95,12 +85,12 @@ extern "C" {
 // Profile 2.  8-bit and 10-bit 4:2:2
 //            12-bit  4:0:0, 4:2:2 and 4:4:4
 // Since we have three bits for the profiles, it can be extended later.
-typedef enum BITSTREAM_PROFILE {
+enum {
   PROFILE_0,
   PROFILE_1,
   PROFILE_2,
   MAX_PROFILES,
-} BITSTREAM_PROFILE;
+} SENUM1BYTE(BITSTREAM_PROFILE);
 
 #define LEVEL_MAJOR_BITS 3
 #define LEVEL_MINOR_BITS 2
@@ -149,7 +139,7 @@ typedef enum ATTRIBUTE_PACKED {
 // 4X4, 8X8, 16X16, 32X32, 64X64, 128X128
 #define SQR_BLOCK_SIZES 6
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   PARTITION_NONE,
   PARTITION_HORZ,
   PARTITION_VERT,
@@ -163,7 +153,7 @@ typedef enum ATTRIBUTE_PACKED {
   EXT_PARTITION_TYPES,
   PARTITION_TYPES = PARTITION_SPLIT + 1,
   PARTITION_INVALID = 255
-} PARTITION_TYPE;
+} UENUM1BYTE(PARTITION_TYPE);
 
 typedef char PARTITION_CONTEXT;
 #define PARTITION_PLOFFSET 4  // number of probability models per block size
@@ -171,12 +161,7 @@ typedef char PARTITION_CONTEXT;
 #define PARTITION_CONTEXTS (PARTITION_BLOCK_SIZES * PARTITION_PLOFFSET)
 
 // block transform size
-#if defined(_MSC_VER)
-typedef uint8_t TX_SIZE;
-enum ATTRIBUTE_PACKED {
-#else
-typedef enum ATTRIBUTE_PACKED {
-#endif
+enum {
   TX_4X4,             // 4x4 transform
   TX_8X8,             // 8x8 transform
   TX_16X16,           // 16x16 transform
@@ -200,11 +185,7 @@ typedef enum ATTRIBUTE_PACKED {
   TX_SIZES = TX_4X8,  // Does NOT include rectangular transforms
   TX_SIZES_LARGEST = TX_64X64,
   TX_INVALID = 255  // Invalid transform size
-#if defined(_MSC_VER)
-};
-#else
-} TX_SIZE;
-#endif
+} UENUM1BYTE(TX_SIZE);
 
 #define TX_SIZE_LUMA_MIN (TX_4X4)
 /* We don't need to code a transform size unless the allowed size is at least
@@ -226,7 +207,7 @@ typedef enum ATTRIBUTE_PACKED {
 #define TX_PAD_HOR 4
 // Pad 6 extra rows (2 on top and 4 on bottom) to remove vertical availability
 // check.
-#define TX_PAD_TOP 2
+#define TX_PAD_TOP 0
 #define TX_PAD_BOTTOM 4
 #define TX_PAD_VER (TX_PAD_TOP + TX_PAD_BOTTOM)
 // Pad 16 extra bytes to avoid reading overflow in SIMD optimization.
@@ -238,43 +219,43 @@ typedef enum ATTRIBUTE_PACKED {
 #define MAX_TX_BLOCKS_IN_MAX_SB (1 << MAX_TX_BLOCKS_IN_MAX_SB_LOG2)
 
 // frame transform mode
-typedef enum ATTRIBUTE_PACKED {
+enum {
   ONLY_4X4,         // use only 4x4 transform
   TX_MODE_LARGEST,  // transform size is the largest possible for pu size
   TX_MODE_SELECT,   // transform specified for each block
   TX_MODES,
-} TX_MODE;
+} UENUM1BYTE(TX_MODE);
 
 // 1D tx types
-typedef enum ATTRIBUTE_PACKED {
+enum {
   DCT_1D,
   ADST_1D,
   FLIPADST_1D,
   IDTX_1D,
   TX_TYPES_1D,
-} TX_TYPE_1D;
+} UENUM1BYTE(TX_TYPE_1D);
 
-typedef enum ATTRIBUTE_PACKED {
-  DCT_DCT,    // DCT  in both horizontal and vertical
-  ADST_DCT,   // ADST in vertical, DCT in horizontal
-  DCT_ADST,   // DCT  in vertical, ADST in horizontal
-  ADST_ADST,  // ADST in both directions
-  FLIPADST_DCT,
-  DCT_FLIPADST,
-  FLIPADST_FLIPADST,
-  ADST_FLIPADST,
-  FLIPADST_ADST,
-  IDTX,
-  V_DCT,
-  H_DCT,
-  V_ADST,
-  H_ADST,
-  V_FLIPADST,
-  H_FLIPADST,
+enum {
+  DCT_DCT,            // DCT in both horizontal and vertical
+  ADST_DCT,           // ADST in vertical, DCT in horizontal
+  DCT_ADST,           // DCT in vertical, ADST in horizontal
+  ADST_ADST,          // ADST in both directions
+  FLIPADST_DCT,       // FLIPADST in vertical, DCT in horizontal
+  DCT_FLIPADST,       // DCT in vertical, FLIPADST in horizontal
+  FLIPADST_FLIPADST,  // FLIPADST in both directions
+  ADST_FLIPADST,      // ADST in vertical, FLIPADST in horizontal
+  FLIPADST_ADST,      // FLIPADST in vertical, ADST in horizontal
+  IDTX,               // Identity in both directions
+  V_DCT,              // DCT in vertical, identity in horizontal
+  H_DCT,              // Identity in vertical, DCT in horizontal
+  V_ADST,             // ADST in vertical, identity in horizontal
+  H_ADST,             // Identity in vertical, ADST in horizontal
+  V_FLIPADST,         // FLIPADST in vertical, identity in horizontal
+  H_FLIPADST,         // Identity in vertical, FLIPADST in horizontal
   TX_TYPES,
-} TX_TYPE;
+} UENUM1BYTE(TX_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   REG_REG,
   REG_SMOOTH,
   REG_SHARP,
@@ -284,9 +265,9 @@ typedef enum ATTRIBUTE_PACKED {
   SHARP_REG,
   SHARP_SMOOTH,
   SHARP_SHARP,
-} DUAL_FILTER_TYPE;
+} UENUM1BYTE(DUAL_FILTER_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   // DCT only
   EXT_TX_SET_DCTONLY,
   // DCT + Identity only
@@ -300,7 +281,7 @@ typedef enum ATTRIBUTE_PACKED {
   // Discrete Trig transforms w/ flip (9) + Identity (1) + 1D Hor/Ver (6)
   EXT_TX_SET_ALL16,
   EXT_TX_SET_TYPES
-} TxSetType;
+} UENUM1BYTE(TxSetType);
 
 #define IS_2D_TRANSFORM(tx_type) (tx_type < IDTX)
 
@@ -308,7 +289,7 @@ typedef enum ATTRIBUTE_PACKED {
 #define EXT_TX_SETS_INTER 4  // Sets of transform selections for INTER
 #define EXT_TX_SETS_INTRA 3  // Sets of transform selections for INTRA
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   AOM_LAST_FLAG = 1 << 0,
   AOM_LAST2_FLAG = 1 << 1,
   AOM_LAST3_FLAG = 1 << 2,
@@ -317,19 +298,15 @@ typedef enum ATTRIBUTE_PACKED {
   AOM_ALT2_FLAG = 1 << 5,
   AOM_ALT_FLAG = 1 << 6,
   AOM_REFFRAME_ALL = (1 << 7) - 1
-} AOM_REFFRAME;
+} UENUM1BYTE(AOM_REFFRAME);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   UNIDIR_COMP_REFERENCE,
   BIDIR_COMP_REFERENCE,
   COMP_REFERENCE_TYPES,
-} COMP_REFERENCE_TYPE;
+} UENUM1BYTE(COMP_REFERENCE_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
-  PLANE_TYPE_Y,
-  PLANE_TYPE_UV,
-  PLANE_TYPES
-} PLANE_TYPE;
+enum { PLANE_TYPE_Y, PLANE_TYPE_UV, PLANE_TYPES } UENUM1BYTE(PLANE_TYPE);
 
 #define CFL_ALPHABET_SIZE_LOG2 4
 #define CFL_ALPHABET_SIZE (1 << CFL_ALPHABET_SIZE_LOG2)
@@ -337,24 +314,20 @@ typedef enum ATTRIBUTE_PACKED {
 #define CFL_IDX_U(idx) (idx >> CFL_ALPHABET_SIZE_LOG2)
 #define CFL_IDX_V(idx) (idx & (CFL_ALPHABET_SIZE - 1))
 
-typedef enum ATTRIBUTE_PACKED {
-  CFL_PRED_U,
-  CFL_PRED_V,
-  CFL_PRED_PLANES
-} CFL_PRED_TYPE;
+enum { CFL_PRED_U, CFL_PRED_V, CFL_PRED_PLANES } UENUM1BYTE(CFL_PRED_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   CFL_SIGN_ZERO,
   CFL_SIGN_NEG,
   CFL_SIGN_POS,
   CFL_SIGNS
-} CFL_SIGN_TYPE;
+} UENUM1BYTE(CFL_SIGN_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   CFL_DISALLOWED,
   CFL_ALLOWED,
   CFL_ALLOWED_TYPES
-} CFL_ALLOWED_TYPE;
+} UENUM1BYTE(CFL_ALLOWED_TYPE);
 
 // CFL_SIGN_ZERO,CFL_SIGN_ZERO is invalid
 #define CFL_JOINT_SIGNS (CFL_SIGNS * CFL_SIGNS - 1)
@@ -371,12 +344,12 @@ typedef enum ATTRIBUTE_PACKED {
 #define CFL_CONTEXT_V(js) \
   (CFL_SIGN_V(js) * CFL_SIGNS + CFL_SIGN_U(js) - CFL_SIGNS)
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   PALETTE_MAP,
   COLOR_MAP_TYPES,
-} COLOR_MAP_TYPE;
+} UENUM1BYTE(COLOR_MAP_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   TWO_COLORS,
   THREE_COLORS,
   FOUR_COLORS,
@@ -385,9 +358,9 @@ typedef enum ATTRIBUTE_PACKED {
   SEVEN_COLORS,
   EIGHT_COLORS,
   PALETTE_SIZES
-} PALETTE_SIZE;
+} UENUM1BYTE(PALETTE_SIZE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   PALETTE_COLOR_ONE,
   PALETTE_COLOR_TWO,
   PALETTE_COLOR_THREE,
@@ -397,11 +370,11 @@ typedef enum ATTRIBUTE_PACKED {
   PALETTE_COLOR_SEVEN,
   PALETTE_COLOR_EIGHT,
   PALETTE_COLORS
-} PALETTE_COLOR;
+} UENUM1BYTE(PALETTE_COLOR);
 
 // Note: All directional predictors must be between V_PRED and D67_PRED (both
 // inclusive).
-typedef enum ATTRIBUTE_PACKED {
+enum {
   DC_PRED,        // Average of above and left pixels
   V_PRED,         // Vertical
   H_PRED,         // Horizontal
@@ -442,11 +415,11 @@ typedef enum ATTRIBUTE_PACKED {
   INTER_MODE_END = MB_MODE_COUNT,
   INTRA_MODES = PAETH_PRED + 1,  // PAETH_PRED has to be the last intra mode.
   INTRA_INVALID = MB_MODE_COUNT  // For uv_mode in inter blocks
-} PREDICTION_MODE;
+} UENUM1BYTE(PREDICTION_MODE);
 
 // TODO(ltrudeau) Do we really want to pack this?
 // TODO(ltrudeau) Do we match with PREDICTION_MODE?
-typedef enum ATTRIBUTE_PACKED {
+enum {
   UV_DC_PRED,        // Average of above and left pixels
   UV_V_PRED,         // Vertical
   UV_H_PRED,         // Horizontal
@@ -463,38 +436,38 @@ typedef enum ATTRIBUTE_PACKED {
   UV_CFL_PRED,       // Chroma-from-Luma
   UV_INTRA_MODES,
   UV_MODE_INVALID,  // For uv_mode in inter blocks
-} UV_PREDICTION_MODE;
+} UENUM1BYTE(UV_PREDICTION_MODE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   SIMPLE_TRANSLATION,
   OBMC_CAUSAL,    // 2-sided OBMC
   WARPED_CAUSAL,  // 2-sided WARPED
   MOTION_MODES
-} MOTION_MODE;
+} UENUM1BYTE(MOTION_MODE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   II_DC_PRED,
   II_V_PRED,
   II_H_PRED,
   II_SMOOTH_PRED,
   INTERINTRA_MODES
-} INTERINTRA_MODE;
+} UENUM1BYTE(INTERINTRA_MODE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   COMPOUND_AVERAGE,
   COMPOUND_WEDGE,
   COMPOUND_DIFFWTD,
   COMPOUND_TYPES,
-} COMPOUND_TYPE;
+} UENUM1BYTE(COMPOUND_TYPE);
 
-typedef enum ATTRIBUTE_PACKED {
+enum {
   FILTER_DC_PRED,
   FILTER_V_PRED,
   FILTER_H_PRED,
   FILTER_D157_PRED,
   FILTER_PAETH_PRED,
   FILTER_INTRA_MODES,
-} FILTER_INTRA_MODE;
+} UENUM1BYTE(FILTER_INTRA_MODE);
 
 #define DIRECTIONAL_MODES 8
 #define MAX_ANGLE_DELTA 3
@@ -550,28 +523,47 @@ typedef enum ATTRIBUTE_PACKED {
 #define TXFM_PARTITION_CONTEXTS ((TX_SIZES - TX_8X8) * 6 - 3)
 typedef uint8_t TXFM_CONTEXT;
 
-#define NONE_FRAME -1
-#define INTRA_FRAME 0
-#define LAST_FRAME 1
-#define LAST2_FRAME 2
-#define LAST3_FRAME 3
-#define GOLDEN_FRAME 4
-#define BWDREF_FRAME 5
-#define ALTREF2_FRAME 6
-#define ALTREF_FRAME 7
-#define EXTREF_FRAME REF_FRAMES
-#define LAST_REF_FRAMES (LAST3_FRAME - LAST_FRAME + 1)
+// An enum for single reference types (and some derived values).
+enum {
+  NONE_FRAME = -1,
+  INTRA_FRAME,
+  LAST_FRAME,
+  LAST2_FRAME,
+  LAST3_FRAME,
+  GOLDEN_FRAME,
+  BWDREF_FRAME,
+  ALTREF2_FRAME,
+  ALTREF_FRAME,
+  REF_FRAMES,
 
-#define INTER_REFS_PER_FRAME (ALTREF_FRAME - LAST_FRAME + 1)
+  // Extra/scratch reference frame. It may be:
+  // - used to update the ALTREF2_FRAME ref (see lshift_bwd_ref_frames()), or
+  // - updated from ALTREF2_FRAME ref (see rshift_bwd_ref_frames()).
+  EXTREF_FRAME = REF_FRAMES,
+
+  // Number of inter (non-intra) reference types.
+  INTER_REFS_PER_FRAME = ALTREF_FRAME - LAST_FRAME + 1,
+
+  // Number of forward (aka past) reference types.
+  FWD_REFS = GOLDEN_FRAME - LAST_FRAME + 1,
+
+  // Number of backward (aka future) reference types.
+  BWD_REFS = ALTREF_FRAME - BWDREF_FRAME + 1,
+
+  SINGLE_REFS = FWD_REFS + BWD_REFS,
+};
+
+#define REF_FRAMES_LOG2 3
+
+// REF_FRAMES for the cm->ref_frame_map array, 1 scratch frame for the new
+// frame in cm->cur_frame, INTER_REFS_PER_FRAME for scaled references on the
+// encoder in the cpi->scaled_ref_buf array.
+#define FRAME_BUFFERS (REF_FRAMES + 1 + INTER_REFS_PER_FRAME)
 
-#define FWD_REFS (GOLDEN_FRAME - LAST_FRAME + 1)
 #define FWD_RF_OFFSET(ref) (ref - LAST_FRAME)
-#define BWD_REFS (ALTREF_FRAME - BWDREF_FRAME + 1)
 #define BWD_RF_OFFSET(ref) (ref - BWDREF_FRAME)
 
-#define SINGLE_REFS (FWD_REFS + BWD_REFS)
-
-typedef enum ATTRIBUTE_PACKED {
+enum {
   LAST_LAST2_FRAMES,      // { LAST_FRAME, LAST2_FRAME }
   LAST_LAST3_FRAMES,      // { LAST_FRAME, LAST3_FRAME }
   LAST_GOLDEN_FRAMES,     // { LAST_FRAME, GOLDEN_FRAME }
@@ -585,7 +577,7 @@ typedef enum ATTRIBUTE_PACKED {
   // NOTE: UNIDIR_COMP_REFS is the number of uni-directional reference pairs
   //       that are explicitly signaled.
   UNIDIR_COMP_REFS = BWDREF_ALTREF_FRAMES + 1,
-} UNIDIR_COMP_REF;
+} UENUM1BYTE(UNIDIR_COMP_REF);
 
 #define TOTAL_COMP_REFS (FWD_REFS * BWD_REFS + TOTAL_UNIDIR_COMP_REFS)
 
@@ -596,14 +588,18 @@ typedef enum ATTRIBUTE_PACKED {
 //       possible to have a reference pair not listed for explicit signaling.
 #define MODE_CTX_REF_FRAMES (REF_FRAMES + TOTAL_COMP_REFS)
 
-typedef enum ATTRIBUTE_PACKED {
+// Note: It includes single and compound references. So, it can take values from
+// NONE_FRAME to (MODE_CTX_REF_FRAMES - 1). Hence, it is not defined as an enum.
+typedef int8_t MV_REFERENCE_FRAME;
+
+enum {
   RESTORE_NONE,
   RESTORE_WIENER,
   RESTORE_SGRPROJ,
   RESTORE_SWITCHABLE,
   RESTORE_SWITCHABLE_TYPES = RESTORE_SWITCHABLE,
   RESTORE_TYPES = 4,
-} RestorationType;
+} UENUM1BYTE(RestorationType);
 
 #define SUPERRES_SCALE_BITS 3
 #define SUPERRES_SCALE_DENOMINATOR_MIN (SCALE_NUMERATOR + 1)
diff --git a/third_party/aom/av1/common/filter.h b/third_party/aom/av1/common/filter.h
index 571422d114..184f5b2d0a 100644
--- a/third_party/aom/av1/common/filter.h
+++ b/third_party/aom/av1/common/filter.h
@@ -37,12 +37,16 @@ typedef enum ATTRIBUTE_PACKED {
   EXTRA_FILTERS = INTERP_FILTERS_ALL - SWITCHABLE_FILTERS,
 } InterpFilter;
 
-// With CONFIG_DUAL_FILTER, pack two InterpFilter's into a uint32_t: since
-// there are at most 10 filters, we can use 16 bits for each and have more than
-// enough space. This reduces argument passing and unifies the operation of
-// setting a (pair of) filters.
-//
-// Without CONFIG_DUAL_FILTER,
+enum {
+  USE_2_TAPS_ORIG = 0,  // This is used in temporal filtering.
+  USE_2_TAPS,
+  USE_4_TAPS,
+  USE_8_TAPS,
+} UENUM1BYTE(SUBPEL_SEARCH_TYPE);
+
+// Pack two InterpFilter's into a uint32_t: since there are at most 10 filters,
+// we can use 16 bits for each and have more than enough space. This reduces
+// argument passing and unifies the operation of setting a (pair of) filters.
 typedef uint32_t InterpFilters;
 static INLINE InterpFilter av1_extract_interp_filter(InterpFilters filters,
                                                      int x_filter) {
@@ -192,14 +196,19 @@ av1_get_interp_filter_params_with_block_size(const InterpFilter interp_filter,
   return &av1_interp_filter_params_list[interp_filter];
 }
 
-static INLINE const InterpFilterParams *av1_get_4tap_interp_filter_params(
+static INLINE const InterpFilterParams *get_4tap_interp_filter_params(
     const InterpFilter interp_filter) {
   return &av1_interp_4tap[interp_filter];
 }
 
 static INLINE const int16_t *av1_get_interp_filter_kernel(
-    const InterpFilter interp_filter) {
-  return av1_interp_filter_params_list[interp_filter].filter_ptr;
+    const InterpFilter interp_filter, int subpel_search) {
+  assert(subpel_search >= USE_2_TAPS);
+  return (subpel_search == USE_2_TAPS)
+             ? av1_interp_4tap[BILINEAR].filter_ptr
+             : ((subpel_search == USE_4_TAPS)
+                    ? av1_interp_4tap[interp_filter].filter_ptr
+                    : av1_interp_filter_params_list[interp_filter].filter_ptr);
 }
 
 static INLINE const int16_t *av1_get_interp_filter_subpel_kernel(
@@ -207,6 +216,17 @@ static INLINE const int16_t *av1_get_interp_filter_subpel_kernel(
   return filter_params->filter_ptr + filter_params->taps * subpel;
 }
 
+static INLINE const InterpFilterParams *av1_get_filter(int subpel_search) {
+  assert(subpel_search >= USE_2_TAPS);
+
+  switch (subpel_search) {
+    case USE_2_TAPS: return get_4tap_interp_filter_params(BILINEAR);
+    case USE_4_TAPS: return get_4tap_interp_filter_params(EIGHTTAP_REGULAR);
+    case USE_8_TAPS: return &av1_interp_filter_params_list[EIGHTTAP_REGULAR];
+    default: assert(0); return NULL;
+  }
+}
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/common/frame_buffers.c b/third_party/aom/av1/common/frame_buffers.c
index fd6c4bc799..f10ccd5942 100644
--- a/third_party/aom/av1/common/frame_buffers.c
+++ b/third_party/aom/av1/common/frame_buffers.c
@@ -22,7 +22,11 @@ int av1_alloc_internal_frame_buffers(InternalFrameBufferList *list) {
       AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
   list->int_fb = (InternalFrameBuffer *)aom_calloc(
       list->num_internal_frame_buffers, sizeof(*list->int_fb));
-  return (list->int_fb == NULL);
+  if (list->int_fb == NULL) {
+    list->num_internal_frame_buffers = 0;
+    return 1;
+  }
+  return 0;
 }
 
 void av1_free_internal_frame_buffers(InternalFrameBufferList *list) {
@@ -36,6 +40,7 @@ void av1_free_internal_frame_buffers(InternalFrameBufferList *list) {
   }
   aom_free(list->int_fb);
   list->int_fb = NULL;
+  list->num_internal_frame_buffers = 0;
 }
 
 void av1_zero_unused_internal_frame_buffers(InternalFrameBufferList *list) {
@@ -69,7 +74,10 @@ int av1_get_frame_buffer(void *cb_priv, size_t min_size,
     // due to access uninitialized memory in frame border. It could be
     // skipped if border were totally removed.
     int_fb_list->int_fb[i].data = (uint8_t *)aom_calloc(1, min_size);
-    if (!int_fb_list->int_fb[i].data) return -1;
+    if (!int_fb_list->int_fb[i].data) {
+      int_fb_list->int_fb[i].size = 0;
+      return -1;
+    }
     int_fb_list->int_fb[i].size = min_size;
   }
 
@@ -86,6 +94,5 @@ int av1_release_frame_buffer(void *cb_priv, aom_codec_frame_buffer_t *fb) {
   InternalFrameBuffer *const int_fb = (InternalFrameBuffer *)fb->priv;
   (void)cb_priv;
   if (int_fb) int_fb->in_use = 0;
-  fb->priv = NULL;
   return 0;
 }
diff --git a/third_party/aom/av1/common/idct.c b/third_party/aom/av1/common/idct.c
index 2c1cb98271..bff438f3c6 100644
--- a/third_party/aom/av1/common/idct.c
+++ b/third_party/aom/av1/common/idct.c
@@ -56,87 +56,87 @@ void av1_highbd_inv_txfm_add_4x4_c(const tran_low_t *input, uint8_t *dest,
   av1_inv_txfm2d_add_4x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type, bd);
 }
 
-void av1_highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest,
-                                 int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_4x8_c(const tran_low_t *input, uint8_t *dest,
+                                   int stride, const TxfmParam *txfm_param) {
   assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]);
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_4x8_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                            txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
-                                 int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_8x4_c(const tran_low_t *input, uint8_t *dest,
+                                   int stride, const TxfmParam *txfm_param) {
   assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]);
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_8x4_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                            txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_16x32(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_16x32_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_16x32_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_32x16(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_32x16_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_32x16_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_16x4(const tran_low_t *input, uint8_t *dest,
-                                  int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_16x4_c(const tran_low_t *input, uint8_t *dest,
+                                    int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_16x4_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                             txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_4x16(const tran_low_t *input, uint8_t *dest,
-                                  int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_4x16_c(const tran_low_t *input, uint8_t *dest,
+                                    int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_4x16_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                             txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_32x8(const tran_low_t *input, uint8_t *dest,
-                                  int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_32x8_c(const tran_low_t *input, uint8_t *dest,
+                                    int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_32x8_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                             txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_8x32(const tran_low_t *input, uint8_t *dest,
-                                  int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_8x32_c(const tran_low_t *input, uint8_t *dest,
+                                    int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_8x32_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                             txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_32x64(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_32x64_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_32x64_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_64x32(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_64x32_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_64x32_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_16x64(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_16x64_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_16x64_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
 }
 
-void av1_highbd_inv_txfm_add_64x16(const tran_low_t *input, uint8_t *dest,
-                                   int stride, const TxfmParam *txfm_param) {
+void av1_highbd_inv_txfm_add_64x16_c(const tran_low_t *input, uint8_t *dest,
+                                     int stride, const TxfmParam *txfm_param) {
   const int32_t *src = cast_to_int32(input);
   av1_inv_txfm2d_add_64x16_c(src, CONVERT_TO_SHORTPTR(dest), stride,
                              txfm_param->tx_type, txfm_param->bd);
@@ -204,7 +204,7 @@ static void init_txfm_param(const MACROBLOCKD *xd, int plane, TX_SIZE tx_size,
   txfm_param->eob = eob;
   txfm_param->lossless = xd->lossless[xd->mi[0]->segment_id];
   txfm_param->bd = xd->bd;
-  txfm_param->is_hbd = get_bitdepth_data_path_index(xd);
+  txfm_param->is_hbd = is_cur_buf_hbd(xd);
   txfm_param->tx_set_type = av1_get_ext_tx_set_type(
       txfm_param->tx_size, is_inter_block(xd->mi[0]), reduced_tx_set);
 }
@@ -224,10 +224,10 @@ void av1_highbd_inv_txfm_add_c(const tran_low_t *input, uint8_t *dest,
       av1_highbd_inv_txfm_add_8x8_c(input, dest, stride, txfm_param);
       break;
     case TX_4X8:
-      av1_highbd_inv_txfm_add_4x8(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x8_c(input, dest, stride, txfm_param);
       break;
     case TX_8X4:
-      av1_highbd_inv_txfm_add_8x4(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_8x4_c(input, dest, stride, txfm_param);
       break;
     case TX_8X16:
       av1_highbd_inv_txfm_add_8x16_c(input, dest, stride, txfm_param);
@@ -236,25 +236,25 @@ void av1_highbd_inv_txfm_add_c(const tran_low_t *input, uint8_t *dest,
       av1_highbd_inv_txfm_add_16x8_c(input, dest, stride, txfm_param);
       break;
     case TX_16X32:
-      av1_highbd_inv_txfm_add_16x32(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_16x32_c(input, dest, stride, txfm_param);
       break;
     case TX_32X16:
-      av1_highbd_inv_txfm_add_32x16(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_32x16_c(input, dest, stride, txfm_param);
       break;
     case TX_64X64:
       av1_highbd_inv_txfm_add_64x64_c(input, dest, stride, txfm_param);
       break;
     case TX_32X64:
-      av1_highbd_inv_txfm_add_32x64(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_32x64_c(input, dest, stride, txfm_param);
       break;
     case TX_64X32:
-      av1_highbd_inv_txfm_add_64x32(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_64x32_c(input, dest, stride, txfm_param);
       break;
     case TX_16X64:
-      av1_highbd_inv_txfm_add_16x64(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_16x64_c(input, dest, stride, txfm_param);
       break;
     case TX_64X16:
-      av1_highbd_inv_txfm_add_64x16(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_64x16_c(input, dest, stride, txfm_param);
       break;
     case TX_4X4:
       // this is like av1_short_idct4x4 but has a special case around eob<=1
@@ -263,16 +263,16 @@ void av1_highbd_inv_txfm_add_c(const tran_low_t *input, uint8_t *dest,
       av1_highbd_inv_txfm_add_4x4_c(input, dest, stride, txfm_param);
       break;
     case TX_16X4:
-      av1_highbd_inv_txfm_add_16x4(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_16x4_c(input, dest, stride, txfm_param);
       break;
     case TX_4X16:
-      av1_highbd_inv_txfm_add_4x16(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x16_c(input, dest, stride, txfm_param);
       break;
     case TX_8X32:
-      av1_highbd_inv_txfm_add_8x32(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_8x32_c(input, dest, stride, txfm_param);
       break;
     case TX_32X8:
-      av1_highbd_inv_txfm_add_32x8(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_32x8_c(input, dest, stride, txfm_param);
       break;
     default: assert(0 && "Invalid transform size"); break;
   }
diff --git a/third_party/aom/av1/common/idct.h b/third_party/aom/av1/common/idct.h
index d9454e73fe..004d25d49a 100644
--- a/third_party/aom/av1/common/idct.h
+++ b/third_party/aom/av1/common/idct.h
@@ -44,22 +44,6 @@ static INLINE const int32_t *cast_to_int32(const tran_low_t *input) {
   return (const int32_t *)input;
 }
 
-typedef void(highbd_inv_txfm_add)(const tran_low_t *input, uint8_t *dest,
-                                  int stride, const TxfmParam *param);
-
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_4x8;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_8x4;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_16x32;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_32x16;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_32x64;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_64x32;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_16x64;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_64x16;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_16x4;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_4x16;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_8x32;
-highbd_inv_txfm_add av1_highbd_inv_txfm_add_32x8;
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/common/mv.h b/third_party/aom/av1/common/mv.h
index 5b0225192a..d097f9e5f1 100644
--- a/third_party/aom/av1/common/mv.h
+++ b/third_party/aom/av1/common/mv.h
@@ -56,13 +56,13 @@ typedef struct mv32 {
 #define WARPEDDIFF_PREC_BITS (WARPEDMODEL_PREC_BITS - WARPEDPIXEL_PREC_BITS)
 
 /* clang-format off */
-typedef enum ATTRIBUTE_PACKED {
+enum {
   IDENTITY = 0,      // identity transformation, 0-parameter
   TRANSLATION = 1,   // translational motion 2-parameter
   ROTZOOM = 2,       // simplified affine with rotation + zoom only, 4-parameter
   AFFINE = 3,        // affine, 6-parameter
   TRANS_TYPES,
-} TransformationType;
+} UENUM1BYTE(TransformationType);
 /* clang-format on */
 
 // Number of types used for global motion (must be >= 3 and <= TRANS_TYPES)
@@ -87,18 +87,18 @@ static const int trans_model_params[TRANS_TYPES] = { 0, 2, 4, 6 };
 //  z .  y'  =   m4 m5 m1 *  y
 //       1]      m6 m7 1)    1]
 typedef struct {
-  TransformationType wmtype;
   int32_t wmmat[8];
   int16_t alpha, beta, gamma, delta;
+  TransformationType wmtype;
   int8_t invalid;
 } WarpedMotionParams;
 
 /* clang-format off */
 static const WarpedMotionParams default_warp_params = {
-  IDENTITY,
   { 0, 0, (1 << WARPEDMODEL_PREC_BITS), 0, 0, (1 << WARPEDMODEL_PREC_BITS), 0,
     0 },
   0, 0, 0, 0,
+  IDENTITY,
   0,
 };
 /* clang-format on */
@@ -263,7 +263,7 @@ static INLINE int_mv gm_get_motion_vector(const WarpedMotionParams *gm,
   return res;
 }
 
-static INLINE TransformationType get_gmtype(const WarpedMotionParams *gm) {
+static INLINE TransformationType get_wmtype(const WarpedMotionParams *gm) {
   if (gm->wmmat[5] == (1 << WARPEDMODEL_PREC_BITS) && !gm->wmmat[4] &&
       gm->wmmat[2] == (1 << WARPEDMODEL_PREC_BITS) && !gm->wmmat[3]) {
     return ((!gm->wmmat[1] && !gm->wmmat[0]) ? IDENTITY : TRANSLATION);
diff --git a/third_party/aom/av1/common/mvref_common.c b/third_party/aom/av1/common/mvref_common.c
index 7f24ab4e6d..3bdffe4ad4 100644
--- a/third_party/aom/av1/common/mvref_common.c
+++ b/third_party/aom/av1/common/mvref_common.c
@@ -346,10 +346,11 @@ static int add_tpl_ref_mv(const AV1_COMMON *cm, const MACROBLOCKD *xd,
   av1_set_ref_frame(rf, ref_frame);
 
   if (rf[1] == NONE_FRAME) {
-    int cur_frame_index = cm->cur_frame->cur_frame_offset;
-    int buf_idx_0 = cm->frame_refs[FWD_RF_OFFSET(rf[0])].idx;
-    int frame0_index = cm->buffer_pool->frame_bufs[buf_idx_0].cur_frame_offset;
-    int cur_offset_0 = get_relative_dist(cm, cur_frame_index, frame0_index);
+    int cur_frame_index = cm->cur_frame->order_hint;
+    const RefCntBuffer *const buf_0 = get_ref_frame_buf(cm, rf[0]);
+    int frame0_index = buf_0->order_hint;
+    int cur_offset_0 = get_relative_dist(&cm->seq_params.order_hint_info,
+                                         cur_frame_index, frame0_index);
     CANDIDATE_MV *ref_mv_stack = ref_mv_stacks[rf[0]];
 
     if (prev_frame_mvs->mfmv0.as_int != INVALID_MV) {
@@ -380,14 +381,16 @@ static int add_tpl_ref_mv(const AV1_COMMON *cm, const MACROBLOCKD *xd,
     }
   } else {
     // Process compound inter mode
-    int cur_frame_index = cm->cur_frame->cur_frame_offset;
-    int buf_idx_0 = cm->frame_refs[FWD_RF_OFFSET(rf[0])].idx;
-    int frame0_index = cm->buffer_pool->frame_bufs[buf_idx_0].cur_frame_offset;
+    int cur_frame_index = cm->cur_frame->order_hint;
+    const RefCntBuffer *const buf_0 = get_ref_frame_buf(cm, rf[0]);
+    int frame0_index = buf_0->order_hint;
 
-    int cur_offset_0 = get_relative_dist(cm, cur_frame_index, frame0_index);
-    int buf_idx_1 = cm->frame_refs[FWD_RF_OFFSET(rf[1])].idx;
-    int frame1_index = cm->buffer_pool->frame_bufs[buf_idx_1].cur_frame_offset;
-    int cur_offset_1 = get_relative_dist(cm, cur_frame_index, frame1_index);
+    int cur_offset_0 = get_relative_dist(&cm->seq_params.order_hint_info,
+                                         cur_frame_index, frame0_index);
+    const RefCntBuffer *const buf_1 = get_ref_frame_buf(cm, rf[1]);
+    int frame1_index = buf_1->order_hint;
+    int cur_offset_1 = get_relative_dist(&cm->seq_params.order_hint_info,
+                                         cur_frame_index, frame1_index);
     CANDIDATE_MV *ref_mv_stack = ref_mv_stacks[ref_frame];
 
     if (prev_frame_mvs->mfmv0.as_int != INVALID_MV) {
@@ -861,26 +864,25 @@ void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
 }
 
 void av1_setup_frame_buf_refs(AV1_COMMON *cm) {
-  cm->cur_frame->cur_frame_offset = cm->frame_offset;
+  cm->cur_frame->order_hint = cm->current_frame.order_hint;
 
   MV_REFERENCE_FRAME ref_frame;
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    const int buf_idx = cm->frame_refs[ref_frame - LAST_FRAME].idx;
-    if (buf_idx >= 0)
-      cm->cur_frame->ref_frame_offset[ref_frame - LAST_FRAME] =
-          cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset;
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    if (buf != NULL)
+      cm->cur_frame->ref_order_hints[ref_frame - LAST_FRAME] = buf->order_hint;
   }
 }
 
 void av1_setup_frame_sign_bias(AV1_COMMON *cm) {
   MV_REFERENCE_FRAME ref_frame;
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    const int buf_idx = cm->frame_refs[ref_frame - LAST_FRAME].idx;
-    if (cm->seq_params.enable_order_hint && buf_idx != INVALID_IDX) {
-      const int ref_frame_offset =
-          cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset;
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    if (cm->seq_params.order_hint_info.enable_order_hint && buf != NULL) {
+      const int ref_order_hint = buf->order_hint;
       cm->ref_frame_sign_bias[ref_frame] =
-          (get_relative_dist(cm, ref_frame_offset, (int)cm->frame_offset) <= 0)
+          (get_relative_dist(&cm->seq_params.order_hint_info, ref_order_hint,
+                             (int)cm->current_frame.order_hint) <= 0)
               ? 0
               : 1;
     } else {
@@ -935,33 +937,34 @@ static int motion_field_projection(AV1_COMMON *cm,
   TPL_MV_REF *tpl_mvs_base = cm->tpl_mvs;
   int ref_offset[REF_FRAMES] = { 0 };
 
-  (void)dir;
+  const RefCntBuffer *const start_frame_buf =
+      get_ref_frame_buf(cm, start_frame);
+  if (start_frame_buf == NULL) return 0;
 
-  const int start_frame_idx = cm->frame_refs[FWD_RF_OFFSET(start_frame)].idx;
-  if (start_frame_idx < 0) return 0;
-
-  if (cm->buffer_pool->frame_bufs[start_frame_idx].intra_only) return 0;
-
-  if (cm->buffer_pool->frame_bufs[start_frame_idx].mi_rows != cm->mi_rows ||
-      cm->buffer_pool->frame_bufs[start_frame_idx].mi_cols != cm->mi_cols)
+  if (start_frame_buf->frame_type == KEY_FRAME ||
+      start_frame_buf->frame_type == INTRA_ONLY_FRAME)
     return 0;
 
-  const int start_frame_offset =
-      cm->buffer_pool->frame_bufs[start_frame_idx].cur_frame_offset;
-  const unsigned int *const ref_frame_offsets =
-      &cm->buffer_pool->frame_bufs[start_frame_idx].ref_frame_offset[0];
-  const int cur_frame_offset = cm->cur_frame->cur_frame_offset;
-  int start_to_current_frame_offset =
-      get_relative_dist(cm, start_frame_offset, cur_frame_offset);
+  if (start_frame_buf->mi_rows != cm->mi_rows ||
+      start_frame_buf->mi_cols != cm->mi_cols)
+    return 0;
+
+  const int start_frame_order_hint = start_frame_buf->order_hint;
+  const unsigned int *const ref_order_hints =
+      &start_frame_buf->ref_order_hints[0];
+  const int cur_order_hint = cm->cur_frame->order_hint;
+  int start_to_current_frame_offset = get_relative_dist(
+      &cm->seq_params.order_hint_info, start_frame_order_hint, cur_order_hint);
 
   for (MV_REFERENCE_FRAME rf = LAST_FRAME; rf <= INTER_REFS_PER_FRAME; ++rf) {
-    ref_offset[rf] = get_relative_dist(cm, start_frame_offset,
-                                       ref_frame_offsets[rf - LAST_FRAME]);
+    ref_offset[rf] = get_relative_dist(&cm->seq_params.order_hint_info,
+                                       start_frame_order_hint,
+                                       ref_order_hints[rf - LAST_FRAME]);
   }
 
   if (dir == 2) start_to_current_frame_offset = -start_to_current_frame_offset;
 
-  MV_REF *mv_ref_base = cm->buffer_pool->frame_bufs[start_frame_idx].mvs;
+  MV_REF *mv_ref_base = start_frame_buf->mvs;
   const int mvs_rows = (cm->mi_rows + 1) >> 1;
   const int mvs_cols = (cm->mi_cols + 1) >> 1;
 
@@ -1002,8 +1005,10 @@ static int motion_field_projection(AV1_COMMON *cm,
 }
 
 void av1_setup_motion_field(AV1_COMMON *cm) {
+  const OrderHintInfo *const order_hint_info = &cm->seq_params.order_hint_info;
+
   memset(cm->ref_frame_side, 0, sizeof(cm->ref_frame_side));
-  if (!cm->seq_params.enable_order_hint) return;
+  if (!order_hint_info->enable_order_hint) return;
 
   TPL_MV_REF *tpl_mvs_base = cm->tpl_mvs;
   int size = ((cm->mi_rows + MAX_MIB_SIZE) >> 1) * (cm->mi_stride >> 1);
@@ -1012,23 +1017,22 @@ void av1_setup_motion_field(AV1_COMMON *cm) {
     tpl_mvs_base[idx].ref_frame_offset = 0;
   }
 
-  const int cur_order_hint = cm->cur_frame->cur_frame_offset;
-  RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
+  const int cur_order_hint = cm->cur_frame->order_hint;
 
-  int ref_buf_idx[INTER_REFS_PER_FRAME];
+  const RefCntBuffer *ref_buf[INTER_REFS_PER_FRAME];
   int ref_order_hint[INTER_REFS_PER_FRAME];
 
   for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
     const int ref_idx = ref_frame - LAST_FRAME;
-    const int buf_idx = cm->frame_refs[ref_idx].idx;
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
     int order_hint = 0;
 
-    if (buf_idx >= 0) order_hint = frame_bufs[buf_idx].cur_frame_offset;
+    if (buf != NULL) order_hint = buf->order_hint;
 
-    ref_buf_idx[ref_idx] = buf_idx;
+    ref_buf[ref_idx] = buf;
     ref_order_hint[ref_idx] = order_hint;
 
-    if (get_relative_dist(cm, order_hint, cur_order_hint) > 0)
+    if (get_relative_dist(order_hint_info, order_hint, cur_order_hint) > 0)
       cm->ref_frame_side[ref_frame] = 1;
     else if (order_hint == cur_order_hint)
       cm->ref_frame_side[ref_frame] = -1;
@@ -1036,10 +1040,10 @@ void av1_setup_motion_field(AV1_COMMON *cm) {
 
   int ref_stamp = MFMV_STACK_SIZE - 1;
 
-  if (ref_buf_idx[LAST_FRAME - LAST_FRAME] >= 0) {
+  if (ref_buf[LAST_FRAME - LAST_FRAME] != NULL) {
     const int alt_of_lst_order_hint =
-        frame_bufs[ref_buf_idx[LAST_FRAME - LAST_FRAME]]
-            .ref_frame_offset[ALTREF_FRAME - LAST_FRAME];
+        ref_buf[LAST_FRAME - LAST_FRAME]
+            ->ref_order_hints[ALTREF_FRAME - LAST_FRAME];
 
     const int is_lst_overlay =
         (alt_of_lst_order_hint == ref_order_hint[GOLDEN_FRAME - LAST_FRAME]);
@@ -1047,23 +1051,25 @@ void av1_setup_motion_field(AV1_COMMON *cm) {
     --ref_stamp;
   }
 
-  if (get_relative_dist(cm, ref_order_hint[BWDREF_FRAME - LAST_FRAME],
+  if (get_relative_dist(order_hint_info,
+                        ref_order_hint[BWDREF_FRAME - LAST_FRAME],
                         cur_order_hint) > 0) {
     if (motion_field_projection(cm, BWDREF_FRAME, 0)) --ref_stamp;
   }
 
-  if (get_relative_dist(cm, ref_order_hint[ALTREF2_FRAME - LAST_FRAME],
+  if (get_relative_dist(order_hint_info,
+                        ref_order_hint[ALTREF2_FRAME - LAST_FRAME],
                         cur_order_hint) > 0) {
     if (motion_field_projection(cm, ALTREF2_FRAME, 0)) --ref_stamp;
   }
 
-  if (get_relative_dist(cm, ref_order_hint[ALTREF_FRAME - LAST_FRAME],
+  if (get_relative_dist(order_hint_info,
+                        ref_order_hint[ALTREF_FRAME - LAST_FRAME],
                         cur_order_hint) > 0 &&
       ref_stamp >= 0)
     if (motion_field_projection(cm, ALTREF_FRAME, 0)) --ref_stamp;
 
-  if (ref_stamp >= 0 && ref_buf_idx[LAST2_FRAME - LAST_FRAME] >= 0)
-    if (motion_field_projection(cm, LAST2_FRAME, 2)) --ref_stamp;
+  if (ref_stamp >= 0) motion_field_projection(cm, LAST2_FRAME, 2);
 }
 
 static INLINE void record_samples(MB_MODE_INFO *mbmi, int *pts, int *pts_inref,
@@ -1264,36 +1270,43 @@ int findSamples(const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
 }
 
 void av1_setup_skip_mode_allowed(AV1_COMMON *cm) {
-  cm->is_skip_mode_allowed = 0;
-  cm->ref_frame_idx_0 = cm->ref_frame_idx_1 = INVALID_IDX;
+  const OrderHintInfo *const order_hint_info = &cm->seq_params.order_hint_info;
+  SkipModeInfo *const skip_mode_info = &cm->current_frame.skip_mode_info;
 
-  if (!cm->seq_params.enable_order_hint || frame_is_intra_only(cm) ||
-      cm->reference_mode == SINGLE_REFERENCE)
+  skip_mode_info->skip_mode_allowed = 0;
+  skip_mode_info->ref_frame_idx_0 = INVALID_IDX;
+  skip_mode_info->ref_frame_idx_1 = INVALID_IDX;
+
+  if (!order_hint_info->enable_order_hint || frame_is_intra_only(cm) ||
+      cm->current_frame.reference_mode == SINGLE_REFERENCE)
     return;
 
-  RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
-  const int cur_frame_offset = cm->frame_offset;
-  int ref_frame_offset[2] = { -1, INT_MAX };
+  const int cur_order_hint = cm->current_frame.order_hint;
+  int ref_order_hints[2] = { -1, INT_MAX };
   int ref_idx[2] = { INVALID_IDX, INVALID_IDX };
 
   // Identify the nearest forward and backward references.
   for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-    const int buf_idx = cm->frame_refs[i].idx;
-    if (buf_idx == INVALID_IDX) continue;
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, LAST_FRAME + i);
+    if (buf == NULL) continue;
 
-    const int ref_offset = frame_bufs[buf_idx].cur_frame_offset;
-    if (get_relative_dist(cm, ref_offset, cur_frame_offset) < 0) {
+    const int ref_order_hint = buf->order_hint;
+    if (get_relative_dist(order_hint_info, ref_order_hint, cur_order_hint) <
+        0) {
       // Forward reference
-      if (ref_frame_offset[0] == -1 ||
-          get_relative_dist(cm, ref_offset, ref_frame_offset[0]) > 0) {
-        ref_frame_offset[0] = ref_offset;
+      if (ref_order_hints[0] == -1 ||
+          get_relative_dist(order_hint_info, ref_order_hint,
+                            ref_order_hints[0]) > 0) {
+        ref_order_hints[0] = ref_order_hint;
         ref_idx[0] = i;
       }
-    } else if (get_relative_dist(cm, ref_offset, cur_frame_offset) > 0) {
+    } else if (get_relative_dist(order_hint_info, ref_order_hint,
+                                 cur_order_hint) > 0) {
       // Backward reference
-      if (ref_frame_offset[1] == INT_MAX ||
-          get_relative_dist(cm, ref_offset, ref_frame_offset[1]) < 0) {
-        ref_frame_offset[1] = ref_offset;
+      if (ref_order_hints[1] == INT_MAX ||
+          get_relative_dist(order_hint_info, ref_order_hint,
+                            ref_order_hints[1]) < 0) {
+        ref_order_hints[1] = ref_order_hint;
         ref_idx[1] = i;
       }
     }
@@ -1301,75 +1314,71 @@ void av1_setup_skip_mode_allowed(AV1_COMMON *cm) {
 
   if (ref_idx[0] != INVALID_IDX && ref_idx[1] != INVALID_IDX) {
     // == Bi-directional prediction ==
-    cm->is_skip_mode_allowed = 1;
-    cm->ref_frame_idx_0 = AOMMIN(ref_idx[0], ref_idx[1]);
-    cm->ref_frame_idx_1 = AOMMAX(ref_idx[0], ref_idx[1]);
+    skip_mode_info->skip_mode_allowed = 1;
+    skip_mode_info->ref_frame_idx_0 = AOMMIN(ref_idx[0], ref_idx[1]);
+    skip_mode_info->ref_frame_idx_1 = AOMMAX(ref_idx[0], ref_idx[1]);
   } else if (ref_idx[0] != INVALID_IDX && ref_idx[1] == INVALID_IDX) {
     // == Forward prediction only ==
     // Identify the second nearest forward reference.
-    ref_frame_offset[1] = -1;
+    ref_order_hints[1] = -1;
     for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-      const int buf_idx = cm->frame_refs[i].idx;
-      if (buf_idx == INVALID_IDX) continue;
+      const RefCntBuffer *const buf = get_ref_frame_buf(cm, LAST_FRAME + i);
+      if (buf == NULL) continue;
 
-      const int ref_offset = frame_bufs[buf_idx].cur_frame_offset;
-      if ((ref_frame_offset[0] != -1 &&
-           get_relative_dist(cm, ref_offset, ref_frame_offset[0]) < 0) &&
-          (ref_frame_offset[1] == -1 ||
-           get_relative_dist(cm, ref_offset, ref_frame_offset[1]) > 0)) {
+      const int ref_order_hint = buf->order_hint;
+      if ((ref_order_hints[0] != -1 &&
+           get_relative_dist(order_hint_info, ref_order_hint,
+                             ref_order_hints[0]) < 0) &&
+          (ref_order_hints[1] == -1 ||
+           get_relative_dist(order_hint_info, ref_order_hint,
+                             ref_order_hints[1]) > 0)) {
         // Second closest forward reference
-        ref_frame_offset[1] = ref_offset;
+        ref_order_hints[1] = ref_order_hint;
         ref_idx[1] = i;
       }
     }
-    if (ref_frame_offset[1] != -1) {
-      cm->is_skip_mode_allowed = 1;
-      cm->ref_frame_idx_0 = AOMMIN(ref_idx[0], ref_idx[1]);
-      cm->ref_frame_idx_1 = AOMMAX(ref_idx[0], ref_idx[1]);
+    if (ref_order_hints[1] != -1) {
+      skip_mode_info->skip_mode_allowed = 1;
+      skip_mode_info->ref_frame_idx_0 = AOMMIN(ref_idx[0], ref_idx[1]);
+      skip_mode_info->ref_frame_idx_1 = AOMMAX(ref_idx[0], ref_idx[1]);
     }
   }
 }
 
 typedef struct {
-  int map_idx;   // frame map index
-  int buf_idx;   // frame buffer index
-  int sort_idx;  // index based on the offset to be used for sorting
+  int map_idx;        // frame map index
+  RefCntBuffer *buf;  // frame buffer
+  int sort_idx;       // index based on the offset to be used for sorting
 } REF_FRAME_INFO;
 
+// Compares the sort_idx fields. If they are equal, then compares the map_idx
+// fields to break the tie. This ensures a stable sort.
 static int compare_ref_frame_info(const void *arg_a, const void *arg_b) {
   const REF_FRAME_INFO *info_a = (REF_FRAME_INFO *)arg_a;
   const REF_FRAME_INFO *info_b = (REF_FRAME_INFO *)arg_b;
 
-  if (info_a->sort_idx < info_b->sort_idx) return -1;
-  if (info_a->sort_idx > info_b->sort_idx) return 1;
-  return (info_a->map_idx < info_b->map_idx)
-             ? -1
-             : ((info_a->map_idx > info_b->map_idx) ? 1 : 0);
+  const int sort_idx_diff = info_a->sort_idx - info_b->sort_idx;
+  if (sort_idx_diff != 0) return sort_idx_diff;
+  return info_a->map_idx - info_b->map_idx;
 }
 
 static void set_ref_frame_info(AV1_COMMON *const cm, int frame_idx,
                                REF_FRAME_INFO *ref_info) {
   assert(frame_idx >= 0 && frame_idx < INTER_REFS_PER_FRAME);
 
-  const int buf_idx = ref_info->buf_idx;
-
-  cm->frame_refs[frame_idx].idx = buf_idx;
-  cm->frame_refs[frame_idx].buf = &cm->buffer_pool->frame_bufs[buf_idx].buf;
-  cm->frame_refs[frame_idx].map_idx = ref_info->map_idx;
+  cm->remapped_ref_idx[frame_idx] = ref_info->map_idx;
 }
 
 void av1_set_frame_refs(AV1_COMMON *const cm, int lst_map_idx,
                         int gld_map_idx) {
-  BufferPool *const pool = cm->buffer_pool;
-  RefCntBuffer *const frame_bufs = pool->frame_bufs;
-
   int lst_frame_sort_idx = -1;
   int gld_frame_sort_idx = -1;
 
-  assert(cm->seq_params.enable_order_hint);
-  assert(cm->seq_params.order_hint_bits_minus_1 >= 0);
-  const int cur_frame_offset = (int)cm->frame_offset;
-  const int cur_frame_sort_idx = 1 << cm->seq_params.order_hint_bits_minus_1;
+  assert(cm->seq_params.order_hint_info.enable_order_hint);
+  assert(cm->seq_params.order_hint_info.order_hint_bits_minus_1 >= 0);
+  const int cur_order_hint = (int)cm->current_frame.order_hint;
+  const int cur_frame_sort_idx =
+      1 << cm->seq_params.order_hint_info.order_hint_bits_minus_1;
 
   REF_FRAME_INFO ref_frame_info[REF_FRAMES];
   int ref_flag_list[INTER_REFS_PER_FRAME] = { 0, 0, 0, 0, 0, 0, 0 };
@@ -1380,18 +1389,21 @@ void av1_set_frame_refs(AV1_COMMON *const cm, int lst_map_idx,
     ref_frame_info[i].map_idx = map_idx;
     ref_frame_info[i].sort_idx = -1;
 
-    const int buf_idx = cm->ref_frame_map[map_idx];
-    ref_frame_info[i].buf_idx = buf_idx;
+    RefCntBuffer *const buf = cm->ref_frame_map[map_idx];
+    ref_frame_info[i].buf = buf;
 
-    if (buf_idx < 0 || buf_idx >= FRAME_BUFFERS) continue;
-    // TODO(zoeliu@google.com): To verify the checking on ref_count.
-    if (frame_bufs[buf_idx].ref_count <= 0) continue;
+    if (buf == NULL) continue;
+    // If this assertion fails, there is a reference leak.
+    assert(buf->ref_count > 0);
+    // TODO(wtc@google.com): Remove the checking on ref_count after 2019-03-01.
+    if (buf->ref_count <= 0) continue;
 
-    const int offset = (int)frame_bufs[buf_idx].cur_frame_offset;
+    const int offset = (int)buf->order_hint;
     ref_frame_info[i].sort_idx =
         (offset == -1) ? -1
                        : cur_frame_sort_idx +
-                             get_relative_dist(cm, offset, cur_frame_offset);
+                             get_relative_dist(&cm->seq_params.order_hint_info,
+                                               offset, cur_order_hint);
     assert(ref_frame_info[i].sort_idx >= -1);
 
     if (map_idx == lst_map_idx) lst_frame_sort_idx = ref_frame_info[i].sort_idx;
@@ -1414,8 +1426,8 @@ void av1_set_frame_refs(AV1_COMMON *const cm, int lst_map_idx,
         compare_ref_frame_info);
 
   // Identify forward and backward reference frames.
-  // Forward  reference: offset < cur_frame_offset
-  // Backward reference: offset >= cur_frame_offset
+  // Forward  reference: offset < order_hint
+  // Backward reference: offset >= order_hint
   int fwd_start_idx = 0, fwd_end_idx = REF_FRAMES - 1;
 
   for (int i = 0; i < REF_FRAMES; i++) {
diff --git a/third_party/aom/av1/common/mvref_common.h b/third_party/aom/av1/common/mvref_common.h
index 83f7a1ac0d..1ece84a8b1 100644
--- a/third_party/aom/av1/common/mvref_common.h
+++ b/third_party/aom/av1/common/mvref_common.h
@@ -34,10 +34,10 @@ typedef struct position {
 // clamp_mv_ref
 #define MV_BORDER (16 << 3)  // Allow 16 pels in 1/8th pel units
 
-static INLINE int get_relative_dist(const AV1_COMMON *cm, int a, int b) {
-  if (!cm->seq_params.enable_order_hint) return 0;
+static INLINE int get_relative_dist(const OrderHintInfo *oh, int a, int b) {
+  if (!oh->enable_order_hint) return 0;
 
-  const int bits = cm->seq_params.order_hint_bits_minus_1 + 1;
+  const int bits = oh->order_hint_bits_minus_1 + 1;
 
   assert(bits >= 1);
   assert(a >= 0 && a < (1 << bits));
@@ -70,18 +70,6 @@ static INLINE int_mv get_sub_block_pred_mv(const MB_MODE_INFO *candidate,
   return candidate->mv[which_mv];
 }
 
-// Performs mv sign inversion if indicated by the reference frame combination.
-static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
-                              const MV_REFERENCE_FRAME this_ref_frame,
-                              const int *ref_sign_bias) {
-  int_mv mv = mbmi->mv[ref];
-  if (ref_sign_bias[mbmi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) {
-    mv.as_mv.row *= -1;
-    mv.as_mv.col *= -1;
-  }
-  return mv;
-}
-
 // Checks that the given mi_row, mi_col and search point
 // are inside the borders of the tile.
 static INLINE int is_inside(const TileInfo *const tile, int mi_col, int mi_row,
@@ -169,14 +157,14 @@ static MV_REFERENCE_FRAME ref_frame_map[TOTAL_COMP_REFS][2] = {
 // clang-format on
 
 static INLINE void av1_set_ref_frame(MV_REFERENCE_FRAME *rf,
-                                     int8_t ref_frame_type) {
+                                     MV_REFERENCE_FRAME ref_frame_type) {
   if (ref_frame_type >= REF_FRAMES) {
     rf[0] = ref_frame_map[ref_frame_type - REF_FRAMES][0];
     rf[1] = ref_frame_map[ref_frame_type - REF_FRAMES][1];
   } else {
+    assert(ref_frame_type > NONE_FRAME);
     rf[0] = ref_frame_type;
     rf[1] = NONE_FRAME;
-    assert(ref_frame_type > NONE_FRAME);
   }
 }
 
diff --git a/third_party/aom/av1/common/obu_util.c b/third_party/aom/av1/common/obu_util.c
index 823b700b13..7d2694b89b 100644
--- a/third_party/aom/av1/common/obu_util.c
+++ b/third_party/aom/av1/common/obu_util.c
@@ -8,6 +8,8 @@
  * Media Patent License 1.0 was not distributed with this source code in the
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
+#include <assert.h>
+
 #include "av1/common/obu_util.h"
 
 #include "aom_dsp/bitreader_buffer.h"
@@ -112,36 +114,41 @@ aom_codec_err_t aom_read_obu_header_and_size(const uint8_t *data,
                                              ObuHeader *obu_header,
                                              size_t *const payload_size,
                                              size_t *const bytes_read) {
-  size_t length_field_size = 0, obu_size = 0;
+  size_t length_field_size_obu = 0;
+  size_t length_field_size_payload = 0;
+  size_t obu_size = 0;
   aom_codec_err_t status;
 
   if (is_annexb) {
     // Size field comes before the OBU header, and includes the OBU header
     status =
-        read_obu_size(data, bytes_available, &obu_size, &length_field_size);
+        read_obu_size(data, bytes_available, &obu_size, &length_field_size_obu);
 
     if (status != AOM_CODEC_OK) return status;
   }
 
-  struct aom_read_bit_buffer rb = { data + length_field_size,
+  struct aom_read_bit_buffer rb = { data + length_field_size_obu,
                                     data + bytes_available, 0, NULL, NULL };
 
   status = read_obu_header(&rb, is_annexb, obu_header);
   if (status != AOM_CODEC_OK) return status;
 
-  if (is_annexb) {
+  if (!obu_header->has_size_field) {
+    assert(is_annexb);
     // Derive the payload size from the data we've already read
     if (obu_size < obu_header->size) return AOM_CODEC_CORRUPT_FRAME;
 
     *payload_size = obu_size - obu_header->size;
   } else {
     // Size field comes after the OBU header, and is just the payload size
-    status = read_obu_size(data + obu_header->size,
-                           bytes_available - obu_header->size, payload_size,
-                           &length_field_size);
+    status = read_obu_size(
+        data + length_field_size_obu + obu_header->size,
+        bytes_available - length_field_size_obu - obu_header->size,
+        payload_size, &length_field_size_payload);
     if (status != AOM_CODEC_OK) return status;
   }
 
-  *bytes_read = length_field_size + obu_header->size;
+  *bytes_read =
+      length_field_size_obu + obu_header->size + length_field_size_payload;
   return AOM_CODEC_OK;
 }
diff --git a/third_party/aom/av1/common/onyxc_int.h b/third_party/aom/av1/common/onyxc_int.h
index ff011c89e9..56f823ec92 100644
--- a/third_party/aom/av1/common/onyxc_int.h
+++ b/third_party/aom/av1/common/onyxc_int.h
@@ -79,14 +79,14 @@ extern "C" {
 #define TXCOEFF_TIMER 0
 #define TXCOEFF_COST_TIMER 0
 
-typedef enum {
+enum {
   SINGLE_REFERENCE = 0,
   COMPOUND_REFERENCE = 1,
   REFERENCE_MODE_SELECT = 2,
   REFERENCE_MODES = 3,
-} REFERENCE_MODE;
+} UENUM1BYTE(REFERENCE_MODE);
 
-typedef enum {
+enum {
   /**
    * Frame context updates are disabled
    */
@@ -96,7 +96,7 @@ typedef enum {
    * updates based on entropy/counts in the decoded frame
    */
   REFRESH_FRAME_CONTEXT_BACKWARD,
-} REFRESH_FRAME_CONTEXT_MODE;
+} UENUM1BYTE(REFRESH_FRAME_CONTEXT_MODE);
 
 #define MFMV_STACK_SIZE 3
 typedef struct {
@@ -109,11 +109,39 @@ typedef struct {
   MV_REFERENCE_FRAME ref_frame;
 } MV_REF;
 
-typedef struct {
+// FIXME(jack.haughton@argondesign.com): This enum was originally in
+// encoder/ratectrl.h, and is encoder specific. When we move to C++, this
+// should go back there and BufferPool should be templatized.
+enum {
+  INTER_NORMAL = 0,
+  INTER_LOW = 1,
+  INTER_HIGH = 2,
+  GF_ARF_LOW = 3,
+  GF_ARF_STD = 4,
+  KF_STD = 5,
+  RATE_FACTOR_LEVELS = 6
+} UENUM1BYTE(RATE_FACTOR_LEVEL);
+
+typedef struct RefCntBuffer {
+  // For a RefCntBuffer, the following are reference-holding variables:
+  // - cm->ref_frame_map[]
+  // - cm->cur_frame
+  // - cm->scaled_ref_buf[] (encoder only)
+  // - cm->next_ref_frame_map[] (decoder only)
+  // - pbi->output_frame_index[] (decoder only)
+  // With that definition, 'ref_count' is the number of reference-holding
+  // variables that are currently referencing this buffer.
+  // For example:
+  // - suppose this buffer is at index 'k' in the buffer pool, and
+  // - Total 'n' of the variables / array elements above have value 'k' (that
+  // is, they are pointing to buffer at index 'k').
+  // Then, pool->frame_bufs[k].ref_count = n.
+  // TODO(david.turner@argondesign.com) Check whether this helpful comment is
+  // still correct after we finish restructuring
   int ref_count;
 
-  unsigned int cur_frame_offset;
-  unsigned int ref_frame_offset[INTER_REFS_PER_FRAME];
+  unsigned int order_hint;
+  unsigned int ref_order_hints[INTER_REFS_PER_FRAME];
 
   MV_REF *mvs;
   uint8_t *seg_map;
@@ -126,36 +154,29 @@ typedef struct {
   int height;
   WarpedMotionParams global_motion[REF_FRAMES];
   int showable_frame;  // frame can be used as show existing frame in future
-  int film_grain_params_present;
+  uint8_t film_grain_params_present;
   aom_film_grain_t film_grain_params;
   aom_codec_frame_buffer_t raw_frame_buffer;
   YV12_BUFFER_CONFIG buf;
   hash_table hash_table;
-  uint8_t intra_only;
   FRAME_TYPE frame_type;
-  // The Following variables will only be used in frame parallel decode.
-
-  // frame_worker_owner indicates which FrameWorker owns this buffer. NULL means
-  // that no FrameWorker owns, or is decoding, this buffer.
-  AVxWorker *frame_worker_owner;
-
-  // row and col indicate which position frame has been decoded to in real
-  // pixel unit. They are reset to -1 when decoding begins and set to INT_MAX
-  // when the frame is fully decoded.
-  int row;
-  int col;
 
   // Inter frame reference frame delta for loop filter
   int8_t ref_deltas[REF_FRAMES];
 
   // 0 = ZERO_MV, MV
   int8_t mode_deltas[MAX_MODE_LF_DELTAS];
+
+  FRAME_CONTEXT frame_context;
+  RATE_FACTOR_LEVEL frame_rf_level;
 } RefCntBuffer;
 
 typedef struct BufferPool {
 // Protect BufferPool from being accessed by several FrameWorkers at
 // the same time during frame parallel decode.
 // TODO(hkuang): Try to use atomic variable instead of locking the whole pool.
+// TODO(wtc): Remove this. See
+// https://chromium-review.googlesource.com/c/webm/libvpx/+/560630.
 #if CONFIG_MULTITHREAD
   pthread_mutex_t pool_mutex;
 #endif
@@ -172,18 +193,45 @@ typedef struct BufferPool {
   InternalFrameBufferList int_frame_buffers;
 } BufferPool;
 
-typedef struct {
-  int base_ctx_table[2 /*row*/][2 /*col*/][3 /*sig_map*/]
-                    [BASE_CONTEXT_POSITION_NUM + 1];
-} LV_MAP_CTX_TABLE;
-typedef int BASE_CTX_TABLE[2 /*col*/][3 /*sig_map*/]
-                          [BASE_CONTEXT_POSITION_NUM + 1];
-
 typedef struct BitstreamLevel {
   uint8_t major;
   uint8_t minor;
 } BitstreamLevel;
 
+typedef struct {
+  int cdef_pri_damping;
+  int cdef_sec_damping;
+  int nb_cdef_strengths;
+  int cdef_strengths[CDEF_MAX_STRENGTHS];
+  int cdef_uv_strengths[CDEF_MAX_STRENGTHS];
+  int cdef_bits;
+} CdefInfo;
+
+typedef struct {
+  int delta_q_present_flag;
+  // Resolution of delta quant
+  int delta_q_res;
+  int delta_lf_present_flag;
+  // Resolution of delta lf level
+  int delta_lf_res;
+  // This is a flag for number of deltas of loop filter level
+  // 0: use 1 delta, for y_vertical, y_horizontal, u, and v
+  // 1: use separate deltas for each filter level
+  int delta_lf_multi;
+} DeltaQInfo;
+
+typedef struct {
+  int enable_order_hint;           // 0 - disable order hint, and related tools
+  int order_hint_bits_minus_1;     // dist_wtd_comp, ref_frame_mvs,
+                                   // frame_sign_bias
+                                   // if 0, enable_dist_wtd_comp and
+                                   // enable_ref_frame_mvs must be set as 0.
+  int enable_dist_wtd_comp;        // 0 - disable dist-wtd compound modes
+                                   // 1 - enable it
+  int enable_ref_frame_mvs;        // 0 - disable ref frame mvs
+                                   // 1 - enable it
+} OrderHintInfo;
+
 // Sequence header structure.
 // Note: All syntax elements of sequence_header_obu that need to be
 // bit-identical across multiple sequence headers must be part of this struct,
@@ -193,51 +241,44 @@ typedef struct SequenceHeader {
   int num_bits_height;
   int max_frame_width;
   int max_frame_height;
-  int frame_id_numbers_present_flag;
+  uint8_t frame_id_numbers_present_flag;
   int frame_id_length;
   int delta_frame_id_length;
   BLOCK_SIZE sb_size;  // Size of the superblock used for this frame
   int mib_size;        // Size of the superblock in units of MI blocks
   int mib_size_log2;   // Log 2 of above.
-  int order_hint_bits_minus_1;
-  int force_screen_content_tools;  // 0 - force off
-                                   // 1 - force on
-                                   // 2 - adaptive
-  int force_integer_mv;            // 0 - Not to force. MV can be in 1/4 or 1/8
-                                   // 1 - force to integer
-                                   // 2 - adaptive
-  int still_picture;               // Video is a single frame still picture
-  int reduced_still_picture_hdr;   // Use reduced header for still picture
-  int enable_filter_intra;         // enables/disables filterintra
-  int enable_intra_edge_filter;    // enables/disables corner/edge/upsampling
-  int enable_interintra_compound;  // enables/disables interintra_compound
-  int enable_masked_compound;      // enables/disables masked compound
-  int enable_dual_filter;          // 0 - disable dual interpolation filter
-                                   // 1 - enable vert/horiz filter selection
-  int enable_order_hint;           // 0 - disable order hint, and related tools
-                                   // jnt_comp, ref_frame_mvs, frame_sign_bias
-                                   // if 0, enable_jnt_comp and
-                                   // enable_ref_frame_mvs must be set zs 0.
-  int enable_jnt_comp;             // 0 - disable joint compound modes
-                                   // 1 - enable it
-  int enable_ref_frame_mvs;        // 0 - disable ref frame mvs
-                                   // 1 - enable it
-  int enable_warped_motion;        // 0 - disable warped motion for sequence
-                                   // 1 - enable it for the sequence
-  int enable_superres;     // 0 - Disable superres for the sequence, and disable
-                           //     transmitting per-frame superres enabled flag.
-                           // 1 - Enable superres for the sequence, and also
-                           //     enable per-frame flag to denote if superres is
-                           //     enabled for that frame.
-  int enable_cdef;         // To turn on/off CDEF
-  int enable_restoration;  // To turn on/off loop restoration
+
+  OrderHintInfo order_hint_info;
+
+  uint8_t force_screen_content_tools;  // 0 - force off
+                                       // 1 - force on
+                                       // 2 - adaptive
+  uint8_t still_picture;               // Video is a single frame still picture
+  uint8_t reduced_still_picture_hdr;   // Use reduced header for still picture
+  uint8_t force_integer_mv;            // 0 - Don't force. MV can use subpel
+                                       // 1 - force to integer
+                                       // 2 - adaptive
+  uint8_t enable_filter_intra;         // enables/disables filterintra
+  uint8_t enable_intra_edge_filter;    // enables/disables edge upsampling
+  uint8_t enable_interintra_compound;  // enables/disables interintra_compound
+  uint8_t enable_masked_compound;      // enables/disables masked compound
+  uint8_t enable_dual_filter;          // 0 - disable dual interpolation filter
+                                       // 1 - enable vert/horz filter selection
+  uint8_t enable_warped_motion;        // 0 - disable warp for the sequence
+                                       // 1 - enable warp for the sequence
+  uint8_t enable_superres;             // 0 - Disable superres for the sequence
+                                       //     and no frame level superres flag
+                                       // 1 - Enable superres for the sequence
+                                       //     enable per-frame superres flag
+  uint8_t enable_cdef;                 // To turn on/off CDEF
+  uint8_t enable_restoration;          // To turn on/off loop restoration
   BITSTREAM_PROFILE profile;
 
   // Operating point info.
   int operating_points_cnt_minus_1;
   int operating_point_idc[MAX_NUM_OPERATING_POINTS];
-  int display_model_info_present_flag;
-  int decoder_model_info_present_flag;
+  uint8_t display_model_info_present_flag;
+  uint8_t decoder_model_info_present_flag;
   BitstreamLevel level[MAX_NUM_OPERATING_POINTS];
   uint8_t tier[MAX_NUM_OPERATING_POINTS];  // seq_tier in the spec. One bit: 0
                                            // or 1.
@@ -245,8 +286,8 @@ typedef struct SequenceHeader {
   // Color config.
   aom_bit_depth_t bit_depth;  // AOM_BITS_8 in profile 0 or 1,
                               // AOM_BITS_10 or AOM_BITS_12 in profile 2 or 3.
-  int use_highbitdepth;       // If true, we need to use 16bit frame buffers.
-  int monochrome;             // Monochorme video
+  uint8_t use_highbitdepth;   // If true, we need to use 16bit frame buffers.
+  uint8_t monochrome;         // Monochorme video
   aom_color_primaries_t color_primaries;
   aom_transfer_characteristics_t transfer_characteristics;
   aom_matrix_coefficients_t matrix_coefficients;
@@ -254,19 +295,34 @@ typedef struct SequenceHeader {
   int subsampling_x;          // Chroma subsampling for x
   int subsampling_y;          // Chroma subsampling for y
   aom_chroma_sample_position_t chroma_sample_position;
-  int separate_uv_delta_q;
-
-  int film_grain_params_present;
+  uint8_t separate_uv_delta_q;
+  uint8_t film_grain_params_present;
 } SequenceHeader;
 
+typedef struct {
+    int skip_mode_allowed;
+    int skip_mode_flag;
+    int ref_frame_idx_0;
+    int ref_frame_idx_1;
+} SkipModeInfo;
+
+typedef struct {
+  FRAME_TYPE frame_type;
+  REFERENCE_MODE reference_mode;
+
+  unsigned int order_hint;
+  unsigned int frame_number;
+  SkipModeInfo skip_mode_info;
+  int refresh_frame_flags;  // Which ref frames are overwritten by this frame
+} CurrentFrame;
+
 typedef struct AV1Common {
+  CurrentFrame current_frame;
   struct aom_internal_error_info error;
   int width;
   int height;
   int render_width;
   int render_height;
-  int last_width;
-  int last_height;
   int timing_info_present;
   aom_timing_info_t timing_info;
   int buffer_removal_time_present;
@@ -275,56 +331,59 @@ typedef struct AV1Common {
   aom_op_timing_info_t op_frame_timing[MAX_NUM_OPERATING_POINTS + 1];
   uint32_t frame_presentation_time;
 
-  int largest_tile_id;
-  size_t largest_tile_size;
   int context_update_tile_id;
 
   // Scale of the current frame with respect to itself.
   struct scale_factors sf_identity;
 
-  YV12_BUFFER_CONFIG *frame_to_show;
   RefCntBuffer *prev_frame;
 
   // TODO(hkuang): Combine this with cur_buf in macroblockd.
   RefCntBuffer *cur_frame;
 
-  int ref_frame_map[REF_FRAMES]; /* maps fb_idx to reference slot */
+  // For encoder, we have a two-level mapping from reference frame type to the
+  // corresponding buffer in the buffer pool:
+  // * 'remapped_ref_idx[i - 1]' maps reference type ‘i’ (range: LAST_FRAME ...
+  // EXTREF_FRAME) to a remapped index ‘j’ (in range: 0 ... REF_FRAMES - 1)
+  // * Later, 'cm->ref_frame_map[j]' maps the remapped index ‘j’ to a pointer to
+  // the reference counted buffer structure RefCntBuffer, taken from the buffer
+  // pool cm->buffer_pool->frame_bufs.
+  //
+  // LAST_FRAME,                        ...,      EXTREF_FRAME
+  //      |                                           |
+  //      v                                           v
+  // remapped_ref_idx[LAST_FRAME - 1],  ...,  remapped_ref_idx[EXTREF_FRAME - 1]
+  //      |                                           |
+  //      v                                           v
+  // ref_frame_map[],                   ...,     ref_frame_map[]
+  //
+  // Note: INTRA_FRAME always refers to the current frame, so there's no need to
+  // have a remapped index for the same.
+  int remapped_ref_idx[REF_FRAMES];
+
+  struct scale_factors ref_scale_factors[REF_FRAMES];
+
+  // For decoder, ref_frame_map[i] maps reference type 'i' to a pointer to
+  // the buffer in the buffer pool ‘cm->buffer_pool.frame_bufs’.
+  // For encoder, ref_frame_map[j] (where j = remapped_ref_idx[i]) maps
+  // remapped reference index 'j' (that is, original reference type 'i') to
+  // a pointer to the buffer in the buffer pool ‘cm->buffer_pool.frame_bufs’.
+  RefCntBuffer *ref_frame_map[REF_FRAMES];
 
   // Prepare ref_frame_map for the next frame.
   // Only used in frame parallel decode.
-  int next_ref_frame_map[REF_FRAMES];
-
-  // TODO(jkoleszar): could expand active_ref_idx to 4, with 0 as intra, and
-  // roll new_fb_idx into it.
-
-  // Each Inter frame can reference INTER_REFS_PER_FRAME buffers
-  RefBuffer frame_refs[INTER_REFS_PER_FRAME];
-  int is_skip_mode_allowed;
-  int skip_mode_flag;
-  int ref_frame_idx_0;
-  int ref_frame_idx_1;
-
-  int new_fb_idx;
-
+  RefCntBuffer *next_ref_frame_map[REF_FRAMES];
   FRAME_TYPE last_frame_type; /* last frame's frame type for motion search.*/
-  FRAME_TYPE frame_type;
 
   int show_frame;
   int showable_frame;  // frame can be used as show existing frame in future
-  int last_show_frame;
   int show_existing_frame;
-  // Flag for a frame used as a reference - not written to the bitstream
-  int is_reference_frame;
-  int reset_decoder_state;
 
-  // Flag signaling that the frame is encoded using only INTRA modes.
-  uint8_t intra_only;
-  uint8_t last_intra_only;
   uint8_t disable_cdf_update;
   int allow_high_precision_mv;
-  int cur_frame_force_integer_mv;  // 0 the default in AOM, 1 only integer
+  uint8_t cur_frame_force_integer_mv;  // 0 the default in AOM, 1 only integer
 
-  int allow_screen_content_tools;
+  uint8_t allow_screen_content_tools;
   int allow_intrabc;
   int allow_warped_motion;
 
@@ -349,7 +408,7 @@ typedef struct AV1Common {
   int u_ac_delta_q;
   int v_ac_delta_q;
 
-  // The dequantizers below are true dequntizers used only in the
+  // The dequantizers below are true dequantizers used only in the
   // dequantization process.  They have the same coefficient
   // shift/scale as TX.
   int16_t y_dequant_QTX[MAX_SEGMENTS][2];
@@ -372,6 +431,7 @@ typedef struct AV1Common {
   int qm_v;
   int min_qmlevel;
   int max_qmlevel;
+  int use_quant_b_adapt;
 
   /* We allocate a MB_MODE_INFO struct for each macroblock, together with
      an extra row on top and column on the left to simplify prediction. */
@@ -400,8 +460,6 @@ typedef struct AV1Common {
   int allow_ref_frame_mvs;
 
   uint8_t *last_frame_seg_map;
-  uint8_t *current_frame_seg_map;
-  int seg_map_alloc_size;
 
   InterpFilter interp_filter;
 
@@ -414,10 +472,6 @@ typedef struct AV1Common {
   int superres_upscaled_height;
   RestorationInfo rst_info[MAX_MB_PLANE];
 
-  // rst_end_stripe[i] is one more than the index of the bottom stripe
-  // for tile row i.
-  int rst_end_stripe[MAX_TILE_ROWS];
-
   // Pointer to a scratch buffer used by self-guided restoration
   int32_t *rst_tmpbuf;
   RestorationLineBuffers *rlbs;
@@ -441,25 +495,14 @@ typedef struct AV1Common {
   // Context probabilities for reference frame prediction
   MV_REFERENCE_FRAME comp_fwd_ref[FWD_REFS];
   MV_REFERENCE_FRAME comp_bwd_ref[BWD_REFS];
-  REFERENCE_MODE reference_mode;
 
   FRAME_CONTEXT *fc;              /* this frame entropy */
-  FRAME_CONTEXT *frame_contexts;  // FRAME_CONTEXTS
-  unsigned int frame_context_idx; /* Context to use/update */
-  int fb_of_context_type[REF_FRAMES];
+  FRAME_CONTEXT *default_frame_context;
   int primary_ref_frame;
 
-  unsigned int frame_offset;
-
-  unsigned int current_video_frame;
-
-  aom_bit_depth_t dequant_bit_depth;  // bit_depth of current dequantizer
-
   int error_resilient_mode;
-  int force_primary_ref_none;
 
   int tile_cols, tile_rows;
-  int last_tile_cols, last_tile_rows;
 
   int max_tile_width_sb;
   int min_log2_tile_cols;
@@ -482,14 +525,6 @@ typedef struct AV1Common {
   int skip_loop_filter;
   int skip_film_grain;
 
-  // Private data associated with the frame buffer callbacks.
-  void *cb_priv;
-  aom_get_frame_buffer_cb_fn_t get_fb_cb;
-  aom_release_frame_buffer_cb_fn_t release_fb_cb;
-
-  // Handles memory for the codec.
-  InternalFrameBufferList int_frame_buffers;
-
   // External BufferPool passed from outside.
   BufferPool *buffer_pool;
 
@@ -499,30 +534,14 @@ typedef struct AV1Common {
   WarpedMotionParams global_motion[REF_FRAMES];
   aom_film_grain_t film_grain_params;
 
-  int cdef_pri_damping;
-  int cdef_sec_damping;
-  int nb_cdef_strengths;
-  int cdef_strengths[CDEF_MAX_STRENGTHS];
-  int cdef_uv_strengths[CDEF_MAX_STRENGTHS];
-  int cdef_bits;
+  CdefInfo cdef_info;
+  DeltaQInfo delta_q_info;  // Delta Q and Delta LF parameters
 
-  int delta_q_present_flag;
-  // Resolution of delta quant
-  int delta_q_res;
-  int delta_lf_present_flag;
-  // Resolution of delta lf level
-  int delta_lf_res;
-  // This is a flag for number of deltas of loop filter level
-  // 0: use 1 delta, for y_vertical, y_horizontal, u, and v
-  // 1: use separate deltas for each filter level
-  int delta_lf_multi;
   int num_tg;
   SequenceHeader seq_params;
   int current_frame_id;
   int ref_frame_id[REF_FRAMES];
   int valid_for_referencing[REF_FRAMES];
-  int invalid_delta_frame_id_minus_1;
-  LV_MAP_CTX_TABLE coeff_ctx_table;
   TPL_MV_REF *tpl_mvs;
   int tpl_mvs_mem_size;
   // TODO(jingning): This can be combined with sign_bias later.
@@ -530,7 +549,6 @@ typedef struct AV1Common {
 
   int is_annexb;
 
-  int frame_refs_short_signaling;
   int temporal_layer_id;
   int spatial_layer_id;
   unsigned int number_temporal_layers;
@@ -551,6 +569,7 @@ typedef struct AV1Common {
   int64_t txcoeff_cost_count;
 #endif
   const cfg_options_t *options;
+  int is_decoding;
 } AV1_COMMON;
 
 // TODO(hkuang): Don't need to lock the whole pool after implementing atomic
@@ -573,14 +592,8 @@ static void unlock_buffer_pool(BufferPool *const pool) {
 
 static INLINE YV12_BUFFER_CONFIG *get_ref_frame(AV1_COMMON *cm, int index) {
   if (index < 0 || index >= REF_FRAMES) return NULL;
-  if (cm->ref_frame_map[index] < 0) return NULL;
-  assert(cm->ref_frame_map[index] < FRAME_BUFFERS);
-  return &cm->buffer_pool->frame_bufs[cm->ref_frame_map[index]].buf;
-}
-
-static INLINE YV12_BUFFER_CONFIG *get_frame_new_buffer(
-    const AV1_COMMON *const cm) {
-  return &cm->buffer_pool->frame_bufs[cm->new_fb_idx].buf;
+  if (cm->ref_frame_map[index] == NULL) return NULL;
+  return &cm->ref_frame_map[index]->buf;
 }
 
 static INLINE int get_free_fb(AV1_COMMON *cm) {
@@ -605,6 +618,9 @@ static INLINE int get_free_fb(AV1_COMMON *cm) {
 
     frame_bufs[i].ref_count = 1;
   } else {
+    // We should never run out of free buffers. If this assertion fails, there
+    // is a reference leak.
+    assert(0 && "Ran out of free frame buffers. Likely a reference leak.");
     // Reset i to be INVALID_IDX to indicate no free buffer found.
     i = INVALID_IDX;
   }
@@ -613,39 +629,90 @@ static INLINE int get_free_fb(AV1_COMMON *cm) {
   return i;
 }
 
-static INLINE void ref_cnt_fb(RefCntBuffer *bufs, int *idx, int new_idx) {
-  const int ref_index = *idx;
+static INLINE RefCntBuffer *assign_cur_frame_new_fb(AV1_COMMON *const cm) {
+  // Release the previously-used frame-buffer
+  if (cm->cur_frame != NULL) {
+    --cm->cur_frame->ref_count;
+    cm->cur_frame = NULL;
+  }
 
-  if (ref_index >= 0 && bufs[ref_index].ref_count > 0)
-    bufs[ref_index].ref_count--;
+  // Assign a new framebuffer
+  const int new_fb_idx = get_free_fb(cm);
+  if (new_fb_idx == INVALID_IDX) return NULL;
 
-  *idx = new_idx;
+  cm->cur_frame = &cm->buffer_pool->frame_bufs[new_fb_idx];
+  cm->cur_frame->buf.buf_8bit_valid = 0;
+  return cm->cur_frame;
+}
 
-  bufs[new_idx].ref_count++;
+// Modify 'lhs_ptr' to reference the buffer at 'rhs_ptr', and update the ref
+// counts accordingly.
+static INLINE void assign_frame_buffer_p(RefCntBuffer **lhs_ptr,
+                                       RefCntBuffer *rhs_ptr) {
+  RefCntBuffer *const old_ptr = *lhs_ptr;
+  if (old_ptr != NULL) {
+    assert(old_ptr->ref_count > 0);
+    // One less reference to the buffer at 'old_ptr', so decrease ref count.
+    --old_ptr->ref_count;
+  }
+
+  *lhs_ptr = rhs_ptr;
+  // One more reference to the buffer at 'rhs_ptr', so increase ref count.
+  ++rhs_ptr->ref_count;
 }
 
 static INLINE int frame_is_intra_only(const AV1_COMMON *const cm) {
-  return cm->frame_type == KEY_FRAME || cm->intra_only;
+  return cm->current_frame.frame_type == KEY_FRAME ||
+      cm->current_frame.frame_type == INTRA_ONLY_FRAME;
 }
 
 static INLINE int frame_is_sframe(const AV1_COMMON *cm) {
-  return cm->frame_type == S_FRAME;
+  return cm->current_frame.frame_type == S_FRAME;
 }
 
-static INLINE RefCntBuffer *get_prev_frame(const AV1_COMMON *const cm) {
-  if (cm->primary_ref_frame == PRIMARY_REF_NONE ||
-      cm->frame_refs[cm->primary_ref_frame].idx == INVALID_IDX) {
-    return NULL;
-  } else {
-    return &cm->buffer_pool
-                ->frame_bufs[cm->frame_refs[cm->primary_ref_frame].idx];
-  }
+// These functions take a reference frame label between LAST_FRAME and
+// EXTREF_FRAME inclusive.  Note that this is different to the indexing
+// previously used by the frame_refs[] array.
+static INLINE int get_ref_frame_map_idx(const AV1_COMMON *const cm,
+                                        const MV_REFERENCE_FRAME ref_frame) {
+  return (ref_frame >= LAST_FRAME && ref_frame <= EXTREF_FRAME)
+             ? cm->remapped_ref_idx[ref_frame - LAST_FRAME]
+             : INVALID_IDX;
+}
+
+static INLINE RefCntBuffer *get_ref_frame_buf(
+    const AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) {
+  const int map_idx = get_ref_frame_map_idx(cm, ref_frame);
+  return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : NULL;
+}
+
+// Both const and non-const versions of this function are provided so that it
+// can be used with a const AV1_COMMON if needed.
+static INLINE const struct scale_factors *get_ref_scale_factors_const(
+    const AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) {
+  const int map_idx = get_ref_frame_map_idx(cm, ref_frame);
+  return (map_idx != INVALID_IDX) ? &cm->ref_scale_factors[map_idx] : NULL;
+}
+
+static INLINE struct scale_factors *get_ref_scale_factors(
+    AV1_COMMON *const cm, const MV_REFERENCE_FRAME ref_frame) {
+  const int map_idx = get_ref_frame_map_idx(cm, ref_frame);
+  return (map_idx != INVALID_IDX) ? &cm->ref_scale_factors[map_idx] : NULL;
+}
+
+static INLINE RefCntBuffer *get_primary_ref_frame_buf(
+    const AV1_COMMON *const cm) {
+  if (cm->primary_ref_frame == PRIMARY_REF_NONE) return NULL;
+  const int map_idx = get_ref_frame_map_idx(cm, cm->primary_ref_frame + 1);
+  return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : NULL;
 }
 
 // Returns 1 if this frame might allow mvs from some reference frame.
 static INLINE int frame_might_allow_ref_frame_mvs(const AV1_COMMON *cm) {
-  return !cm->error_resilient_mode && cm->seq_params.enable_ref_frame_mvs &&
-         cm->seq_params.enable_order_hint && !frame_is_intra_only(cm);
+  return !cm->error_resilient_mode &&
+    cm->seq_params.order_hint_info.enable_ref_frame_mvs &&
+    cm->seq_params.order_hint_info.enable_order_hint &&
+    !frame_is_intra_only(cm);
 }
 
 // Returns 1 if this frame might use warped_motion
@@ -1193,8 +1260,8 @@ static INLINE TX_SIZE get_tx_size(int width, int height) {
   return TX_4X4;
 }
 
-static INLINE int txfm_partition_context(TXFM_CONTEXT *above_ctx,
-                                         TXFM_CONTEXT *left_ctx,
+static INLINE int txfm_partition_context(const TXFM_CONTEXT *const above_ctx,
+                                         const TXFM_CONTEXT *const left_ctx,
                                          BLOCK_SIZE bsize, TX_SIZE tx_size) {
   const uint8_t txw = tx_size_wide[tx_size];
   const uint8_t txh = tx_size_high[tx_size];
@@ -1292,10 +1359,6 @@ static INLINE PARTITION_TYPE get_partition(const AV1_COMMON *const cm,
   return base_partitions[split_idx];
 }
 
-static INLINE void set_use_reference_buffer(AV1_COMMON *const cm, int use) {
-  cm->seq_params.frame_id_numbers_present_flag = use;
-}
-
 static INLINE void set_sb_size(SequenceHeader *const seq_params,
                                BLOCK_SIZE sb_size) {
   seq_params->sb_size = sb_size;
diff --git a/third_party/aom/av1/common/ppc/cfl_ppc.c b/third_party/aom/av1/common/ppc/cfl_ppc.c
index 026a078095..61d8dc112c 100644
--- a/third_party/aom/av1/common/ppc/cfl_ppc.c
+++ b/third_party/aom/av1/common/ppc/cfl_ppc.c
@@ -130,21 +130,21 @@ cfl_subtract_average_fn get_subtract_average_fn_vsx(TX_SIZE tx_size) {
     subtract_average_8x8_vsx,   /* 8x8 */
     subtract_average_16x16_vsx, /* 16x16 */
     subtract_average_32x32_vsx, /* 32x32 */
-    cfl_subtract_average_null,  /* 64x64 (invalid CFL size) */
+    NULL,                       /* 64x64 (invalid CFL size) */
     subtract_average_4x8_c,     /* 4x8 */
     subtract_average_8x4_vsx,   /* 8x4 */
     subtract_average_8x16_vsx,  /* 8x16 */
     subtract_average_16x8_vsx,  /* 16x8 */
     subtract_average_16x32_vsx, /* 16x32 */
     subtract_average_32x16_vsx, /* 32x16 */
-    cfl_subtract_average_null,  /* 32x64 (invalid CFL size) */
-    cfl_subtract_average_null,  /* 64x32 (invalid CFL size) */
+    NULL,                       /* 32x64 (invalid CFL size) */
+    NULL,                       /* 64x32 (invalid CFL size) */
     subtract_average_4x16_c,    /* 4x16 */
     subtract_average_16x4_vsx,  /* 16x4 */
     subtract_average_8x32_vsx,  /* 8x32 */
     subtract_average_32x8_vsx,  /* 32x8 */
-    cfl_subtract_average_null,  /* 16x64 (invalid CFL size) */
-    cfl_subtract_average_null,  /* 64x16 (invalid CFL size) */
+    NULL,                       /* 16x64 (invalid CFL size) */
+    NULL,                       /* 64x16 (invalid CFL size) */
   };
   // Modulo TX_SIZES_ALL to ensure that an attacker won't be able to
   // index the function pointer array out of bounds.
diff --git a/third_party/aom/av1/common/pred_common.h b/third_party/aom/av1/common/pred_common.h
index 6dba2322d6..1d69ef003e 100644
--- a/third_party/aom/av1/common/pred_common.h
+++ b/third_party/aom/av1/common/pred_common.h
@@ -48,16 +48,16 @@ static INLINE int av1_get_spatial_seg_pred(const AV1_COMMON *const cm,
   int prev_l = -1;   // left segment_id
   int prev_u = -1;   // top segment_id
   if ((xd->up_available) && (xd->left_available)) {
-    prev_ul = get_segment_id(cm, cm->current_frame_seg_map, BLOCK_4X4,
-                             mi_row - 1, mi_col - 1);
+    prev_ul = get_segment_id(cm, cm->cur_frame->seg_map, BLOCK_4X4, mi_row - 1,
+                             mi_col - 1);
   }
   if (xd->up_available) {
-    prev_u = get_segment_id(cm, cm->current_frame_seg_map, BLOCK_4X4,
-                            mi_row - 1, mi_col - 0);
+    prev_u = get_segment_id(cm, cm->cur_frame->seg_map, BLOCK_4X4, mi_row - 1,
+                            mi_col - 0);
   }
   if (xd->left_available) {
-    prev_l = get_segment_id(cm, cm->current_frame_seg_map, BLOCK_4X4,
-                            mi_row - 0, mi_col - 1);
+    prev_l = get_segment_id(cm, cm->cur_frame->seg_map, BLOCK_4X4, mi_row - 0,
+                            mi_col - 1);
   }
 
   // Pick CDF index based on number of matching/out-of-bounds segment IDs.
@@ -90,18 +90,18 @@ static INLINE int av1_get_pred_context_seg_id(const MACROBLOCKD *xd) {
 static INLINE int get_comp_index_context(const AV1_COMMON *cm,
                                          const MACROBLOCKD *xd) {
   MB_MODE_INFO *mbmi = xd->mi[0];
-  int bck_idx = cm->frame_refs[mbmi->ref_frame[0] - LAST_FRAME].idx;
-  int fwd_idx = cm->frame_refs[mbmi->ref_frame[1] - LAST_FRAME].idx;
+  const RefCntBuffer *const bck_buf = get_ref_frame_buf(cm, mbmi->ref_frame[0]);
+  const RefCntBuffer *const fwd_buf = get_ref_frame_buf(cm, mbmi->ref_frame[1]);
   int bck_frame_index = 0, fwd_frame_index = 0;
-  int cur_frame_index = cm->cur_frame->cur_frame_offset;
+  int cur_frame_index = cm->cur_frame->order_hint;
 
-  if (bck_idx >= 0)
-    bck_frame_index = cm->buffer_pool->frame_bufs[bck_idx].cur_frame_offset;
+  if (bck_buf != NULL) bck_frame_index = bck_buf->order_hint;
+  if (fwd_buf != NULL) fwd_frame_index = fwd_buf->order_hint;
 
-  if (fwd_idx >= 0)
-    fwd_frame_index = cm->buffer_pool->frame_bufs[fwd_idx].cur_frame_offset;
-  int fwd = abs(get_relative_dist(cm, fwd_frame_index, cur_frame_index));
-  int bck = abs(get_relative_dist(cm, cur_frame_index, bck_frame_index));
+  int fwd = abs(get_relative_dist(&cm->seq_params.order_hint_info,
+                                  fwd_frame_index, cur_frame_index));
+  int bck = abs(get_relative_dist(&cm->seq_params.order_hint_info,
+                                  cur_frame_index, bck_frame_index));
 
   const MB_MODE_INFO *const above_mi = xd->above_mbmi;
   const MB_MODE_INFO *const left_mi = xd->left_mbmi;
@@ -109,14 +109,14 @@ static INLINE int get_comp_index_context(const AV1_COMMON *cm,
   int above_ctx = 0, left_ctx = 0;
   const int offset = (fwd == bck);
 
-  if (above_mi) {
+  if (above_mi != NULL) {
     if (has_second_ref(above_mi))
       above_ctx = above_mi->compound_idx;
     else if (above_mi->ref_frame[0] == ALTREF_FRAME)
       above_ctx = 1;
   }
 
-  if (left_mi) {
+  if (left_mi != NULL) {
     if (has_second_ref(left_mi))
       left_ctx = left_mi->compound_idx;
     else if (left_mi->ref_frame[0] == ALTREF_FRAME)
diff --git a/third_party/aom/av1/common/quant_common.c b/third_party/aom/av1/common/quant_common.c
index 0e14da7a38..d4bdb98062 100644
--- a/third_party/aom/av1/common/quant_common.c
+++ b/third_party/aom/av1/common/quant_common.c
@@ -191,10 +191,11 @@ static const int16_t ac_qlookup_12_Q3[QINDEX_RANGE] = {
 // underflow to 0 in the actual quantization routines.
 
 int16_t av1_dc_quant_Q3(int qindex, int delta, aom_bit_depth_t bit_depth) {
+  const int q_clamped = clamp(qindex + delta, 0, MAXQ);
   switch (bit_depth) {
-    case AOM_BITS_8: return dc_qlookup_Q3[clamp(qindex + delta, 0, MAXQ)];
-    case AOM_BITS_10: return dc_qlookup_10_Q3[clamp(qindex + delta, 0, MAXQ)];
-    case AOM_BITS_12: return dc_qlookup_12_Q3[clamp(qindex + delta, 0, MAXQ)];
+    case AOM_BITS_8: return dc_qlookup_Q3[q_clamped];
+    case AOM_BITS_10: return dc_qlookup_10_Q3[q_clamped];
+    case AOM_BITS_12: return dc_qlookup_12_Q3[q_clamped];
     default:
       assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
       return -1;
@@ -202,10 +203,11 @@ int16_t av1_dc_quant_Q3(int qindex, int delta, aom_bit_depth_t bit_depth) {
 }
 
 int16_t av1_ac_quant_Q3(int qindex, int delta, aom_bit_depth_t bit_depth) {
+  const int q_clamped = clamp(qindex + delta, 0, MAXQ);
   switch (bit_depth) {
-    case AOM_BITS_8: return ac_qlookup_Q3[clamp(qindex + delta, 0, MAXQ)];
-    case AOM_BITS_10: return ac_qlookup_10_Q3[clamp(qindex + delta, 0, MAXQ)];
-    case AOM_BITS_12: return ac_qlookup_12_Q3[clamp(qindex + delta, 0, MAXQ)];
+    case AOM_BITS_8: return ac_qlookup_Q3[q_clamped];
+    case AOM_BITS_10: return ac_qlookup_10_Q3[q_clamped];
+    case AOM_BITS_12: return ac_qlookup_12_Q3[q_clamped];
     default:
       assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
       return -1;
@@ -244,8 +246,10 @@ const qm_val_t *av1_qmatrix(AV1_COMMON *cm, int qmlevel, int plane,
 }
 
 #define QM_TOTAL_SIZE 3344
-static const qm_val_t wt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE];
-static const qm_val_t iwt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE];
+// We only use wt_matrix_ref[q] and iwt_matrix_ref[q]
+// for q = 0, ..., NUM_QM_LEVELS - 2.
+static const qm_val_t wt_matrix_ref[NUM_QM_LEVELS - 1][2][QM_TOTAL_SIZE];
+static const qm_val_t iwt_matrix_ref[NUM_QM_LEVELS - 1][2][QM_TOTAL_SIZE];
 
 void av1_qm_init(AV1_COMMON *cm) {
   const int num_planes = av1_num_planes(cm);
@@ -274,7 +278,7 @@ void av1_qm_init(AV1_COMMON *cm) {
   }
 }
 
-/* Provide 16 sets of quantization matrices for chroma and luma
+/* Provide 15 sets of quantization matrices for chroma and luma
    and each TX size. Matrices for different TX sizes are in fact
    sub-sampled from the 32x32 and 16x16 sizes, but explicitly
    defined here for convenience. Intra and inter matrix sets are the
@@ -283,9 +287,10 @@ void av1_qm_init(AV1_COMMON *cm) {
    frame.
    Matrices for different QM levels have been rescaled in the
    frequency domain according to different nominal viewing
-   distances.
+   distances. Matrices for QM level 15 are omitted because they are
+   not used.
  */
-static const qm_val_t iwt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE] = {
+static const qm_val_t iwt_matrix_ref[NUM_QM_LEVELS - 1][2][QM_TOTAL_SIZE] = {
   {
       { /* Luma */
         /* Size 4x4 */
@@ -6633,427 +6638,9 @@ static const qm_val_t iwt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE] = {
         31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 31, 32, 32, 32, 32, 32,
         32, 32, 32, 32 },
   },
-  {
-      { /* Luma */
-        /* Size 4x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 32x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32 },
-      { /* Chroma */
-        /* Size 4x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 32x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32 },
-  },
 };
 
-static const qm_val_t wt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE] = {
+static const qm_val_t wt_matrix_ref[NUM_QM_LEVELS - 1][2][QM_TOTAL_SIZE] = {
   {
       { /* Luma */
         /* Size 4x4 */
@@ -13255,422 +12842,4 @@ static const qm_val_t wt_matrix_ref[NUM_QM_LEVELS][2][QM_TOTAL_SIZE] = {
         33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 32, 32, 32, 32, 32,
         32, 32, 32, 32 },
   },
-  {
-      { /* Luma */
-        /* Size 4x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 32x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32 },
-      { /* Chroma */
-        /* Size 4x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32,
-        /* Size 16x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 32x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 4x16 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 16x4 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        /* Size 8x32 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32,
-        /* Size 32x8 */
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
-        32, 32, 32, 32 },
-  },
 };
diff --git a/third_party/aom/av1/common/reconinter.c b/third_party/aom/av1/common/reconinter.c
index 3203efce4b..ea351cf698 100644
--- a/third_party/aom/av1/common/reconinter.c
+++ b/third_party/aom/av1/common/reconinter.c
@@ -35,10 +35,12 @@
 int av1_allow_warp(const MB_MODE_INFO *const mbmi,
                    const WarpTypesAllowed *const warp_types,
                    const WarpedMotionParams *const gm_params,
-                   int build_for_obmc, int x_scale, int y_scale,
+                   int build_for_obmc, const struct scale_factors *const sf,
                    WarpedMotionParams *final_warp_params) {
-  if (x_scale != SCALE_SUBPEL_SHIFTS || y_scale != SCALE_SUBPEL_SHIFTS)
-    return 0;
+  // Note: As per the spec, we must test the fixed point scales here, which are
+  // at a higher precision (1 << 14) than the xs and ys in subpel_params (that
+  // have 1 << 10 precision).
+  if (av1_is_scaled(sf)) return 0;
 
   if (final_warp_params != NULL) *final_warp_params = default_warp_params;
 
@@ -75,20 +77,18 @@ void av1_make_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
   const int do_warp =
       (w >= 8 && h >= 8 &&
        av1_allow_warp(mi, warp_types, &xd->global_motion[mi->ref_frame[ref]],
-                      build_for_obmc, subpel_params->xs, subpel_params->ys,
-                      &final_warp_params));
+                      build_for_obmc, sf, &final_warp_params));
   const int is_intrabc = mi->use_intrabc;
   assert(IMPLIES(is_intrabc, !do_warp));
 
   if (do_warp && xd->cur_frame_force_integer_mv == 0) {
     const struct macroblockd_plane *const pd = &xd->plane[plane];
     const struct buf_2d *const pre_buf = &pd->pre[ref];
-    av1_warp_plane(&final_warp_params,
-                   xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
+    av1_warp_plane(&final_warp_params, is_cur_buf_hbd(xd), xd->bd,
                    pre_buf->buf0, pre_buf->width, pre_buf->height,
                    pre_buf->stride, dst, p_col, p_row, w, h, dst_stride,
                    pd->subsampling_x, pd->subsampling_y, conv_params);
-  } else if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  } else if (is_cur_buf_hbd(xd)) {
     highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_params, sf,
                            w, h, conv_params, interp_filters, is_intrabc,
                            xd->bd);
@@ -567,14 +567,15 @@ static void build_masked_compound_no_round(
   const int subh = (2 << mi_size_high_log2[sb_type]) == h;
   const int subw = (2 << mi_size_wide_log2[sb_type]) == w;
   const uint8_t *mask = av1_get_compound_type_mask(comp_data, sb_type);
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+  if (is_cur_buf_hbd(xd)) {
     aom_highbd_blend_a64_d16_mask(dst, dst_stride, src0, src0_stride, src1,
                                   src1_stride, mask, block_size_wide[sb_type],
                                   w, h, subw, subh, conv_params, xd->bd);
-  else
+  } else {
     aom_lowbd_blend_a64_d16_mask(dst, dst_stride, src0, src0_stride, src1,
                                  src1_stride, mask, block_size_wide[sb_type], w,
                                  h, subw, subh, conv_params);
+  }
 }
 
 void av1_make_masked_inter_predictor(
@@ -625,32 +626,31 @@ void av1_make_masked_inter_predictor(
                                  mi->sb_type, h, w, conv_params, xd);
 }
 
-void av1_jnt_comp_weight_assign(const AV1_COMMON *cm, const MB_MODE_INFO *mbmi,
-                                int order_idx, int *fwd_offset, int *bck_offset,
-                                int *use_jnt_comp_avg, int is_compound) {
+void av1_dist_wtd_comp_weight_assign(const AV1_COMMON *cm,
+                                     const MB_MODE_INFO *mbmi, int order_idx,
+                                     int *fwd_offset, int *bck_offset,
+                                     int *use_dist_wtd_comp_avg,
+                                     int is_compound) {
   assert(fwd_offset != NULL && bck_offset != NULL);
   if (!is_compound || mbmi->compound_idx) {
-    *use_jnt_comp_avg = 0;
+    *use_dist_wtd_comp_avg = 0;
     return;
   }
 
-  *use_jnt_comp_avg = 1;
-  const int bck_idx = cm->frame_refs[mbmi->ref_frame[0] - LAST_FRAME].idx;
-  const int fwd_idx = cm->frame_refs[mbmi->ref_frame[1] - LAST_FRAME].idx;
-  const int cur_frame_index = cm->cur_frame->cur_frame_offset;
+  *use_dist_wtd_comp_avg = 1;
+  const RefCntBuffer *const bck_buf = get_ref_frame_buf(cm, mbmi->ref_frame[0]);
+  const RefCntBuffer *const fwd_buf = get_ref_frame_buf(cm, mbmi->ref_frame[1]);
+  const int cur_frame_index = cm->cur_frame->order_hint;
   int bck_frame_index = 0, fwd_frame_index = 0;
 
-  if (bck_idx >= 0) {
-    bck_frame_index = cm->buffer_pool->frame_bufs[bck_idx].cur_frame_offset;
-  }
+  if (bck_buf != NULL) bck_frame_index = bck_buf->order_hint;
+  if (fwd_buf != NULL) fwd_frame_index = fwd_buf->order_hint;
 
-  if (fwd_idx >= 0) {
-    fwd_frame_index = cm->buffer_pool->frame_bufs[fwd_idx].cur_frame_offset;
-  }
-
-  int d0 = clamp(abs(get_relative_dist(cm, fwd_frame_index, cur_frame_index)),
+  int d0 = clamp(abs(get_relative_dist(&cm->seq_params.order_hint_info,
+                                       fwd_frame_index, cur_frame_index)),
                  0, MAX_FRAME_DISTANCE);
-  int d1 = clamp(abs(get_relative_dist(cm, cur_frame_index, bck_frame_index)),
+  int d1 = clamp(abs(get_relative_dist(&cm->seq_params.order_hint_info,
+                                       cur_frame_index, bck_frame_index)),
                  0, MAX_FRAME_DISTANCE);
 
   const int order = d0 <= d1;
@@ -708,10 +708,9 @@ void av1_setup_pre_planes(MACROBLOCKD *xd, int idx,
 
 // obmc_mask_N[overlap_position]
 static const uint8_t obmc_mask_1[1] = { 64 };
+DECLARE_ALIGNED(2, static const uint8_t, obmc_mask_2[2]) = { 45, 64 };
 
-static const uint8_t obmc_mask_2[2] = { 45, 64 };
-
-static const uint8_t obmc_mask_4[4] = { 39, 50, 59, 64 };
+DECLARE_ALIGNED(4, static const uint8_t, obmc_mask_4[4]) = { 39, 50, 59, 64 };
 
 static const uint8_t obmc_mask_8[8] = { 36, 42, 48, 53, 57, 61, 64, 64 };
 
@@ -814,7 +813,7 @@ static INLINE void build_obmc_inter_pred_above(MACROBLOCKD *xd, int rel_mi_col,
   (void)above_mi;
   struct obmc_inter_pred_ctxt *ctxt = (struct obmc_inter_pred_ctxt *)fun_ctxt;
   const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
-  const int is_hbd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+  const int is_hbd = is_cur_buf_hbd(xd);
   const int overlap =
       AOMMIN(block_size_high[bsize], block_size_high[BLOCK_64X64]) >> 1;
 
@@ -851,7 +850,7 @@ static INLINE void build_obmc_inter_pred_left(MACROBLOCKD *xd, int rel_mi_row,
   const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
   const int overlap =
       AOMMIN(block_size_wide[bsize], block_size_wide[BLOCK_64X64]) >> 1;
-  const int is_hbd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+  const int is_hbd = is_cur_buf_hbd(xd);
 
   for (int plane = 0; plane < num_planes; ++plane) {
     const struct macroblockd_plane *pd = &xd->plane[plane];
@@ -922,14 +921,15 @@ void av1_setup_build_prediction_by_above_pred(
   for (int ref = 0; ref < num_refs; ++ref) {
     const MV_REFERENCE_FRAME frame = above_mbmi->ref_frame[ref];
 
-    const RefBuffer *const ref_buf = &ctxt->cm->frame_refs[frame - LAST_FRAME];
-
-    xd->block_refs[ref] = ref_buf;
-    if ((!av1_is_valid_scale(&ref_buf->sf)))
+    const RefCntBuffer *const ref_buf = get_ref_frame_buf(ctxt->cm, frame);
+    const struct scale_factors *const sf =
+        get_ref_scale_factors_const(ctxt->cm, frame);
+    xd->block_ref_scale_factors[ref] = sf;
+    if ((!av1_is_valid_scale(sf)))
       aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
                          "Reference frame has invalid dimensions");
-    av1_setup_pre_planes(xd, ref, ref_buf->buf, ctxt->mi_row, above_mi_col,
-                         &ref_buf->sf, num_planes);
+    av1_setup_pre_planes(xd, ref, &ref_buf->buf, ctxt->mi_row, above_mi_col, sf,
+                         num_planes);
   }
 
   xd->mb_to_left_edge = 8 * MI_SIZE * (-above_mi_col);
@@ -959,14 +959,16 @@ void av1_setup_build_prediction_by_left_pred(MACROBLOCKD *xd, int rel_mi_row,
   for (int ref = 0; ref < num_refs; ++ref) {
     const MV_REFERENCE_FRAME frame = left_mbmi->ref_frame[ref];
 
-    const RefBuffer *const ref_buf = &ctxt->cm->frame_refs[frame - LAST_FRAME];
+    const RefCntBuffer *const ref_buf = get_ref_frame_buf(ctxt->cm, frame);
+    const struct scale_factors *const ref_scale_factors =
+        get_ref_scale_factors_const(ctxt->cm, frame);
 
-    xd->block_refs[ref] = ref_buf;
-    if ((!av1_is_valid_scale(&ref_buf->sf)))
+    xd->block_ref_scale_factors[ref] = ref_scale_factors;
+    if ((!av1_is_valid_scale(ref_scale_factors)))
       aom_internal_error(xd->error_info, AOM_CODEC_UNSUP_BITSTREAM,
                          "Reference frame has invalid dimensions");
-    av1_setup_pre_planes(xd, ref, ref_buf->buf, left_mi_row, ctxt->mi_col,
-                         &ref_buf->sf, num_planes);
+    av1_setup_pre_planes(xd, ref, &ref_buf->buf, left_mi_row, ctxt->mi_col,
+                         ref_scale_factors, num_planes);
   }
 
   xd->mb_to_top_edge = 8 * MI_SIZE * (-left_mi_row);
@@ -1033,12 +1035,13 @@ static void build_smooth_interintra_mask(uint8_t *mask, int stride,
   }
 }
 
-static void combine_interintra(INTERINTRA_MODE mode, int use_wedge_interintra,
-                               int wedge_index, int wedge_sign,
-                               BLOCK_SIZE bsize, BLOCK_SIZE plane_bsize,
-                               uint8_t *comppred, int compstride,
-                               const uint8_t *interpred, int interstride,
-                               const uint8_t *intrapred, int intrastride) {
+static void combine_interintra(INTERINTRA_MODE mode,
+                               int8_t use_wedge_interintra, int wedge_index,
+                               int wedge_sign, BLOCK_SIZE bsize,
+                               BLOCK_SIZE plane_bsize, uint8_t *comppred,
+                               int compstride, const uint8_t *interpred,
+                               int interstride, const uint8_t *intrapred,
+                               int intrastride) {
   const int bw = block_size_wide[plane_bsize];
   const int bh = block_size_high[plane_bsize];
 
@@ -1062,7 +1065,7 @@ static void combine_interintra(INTERINTRA_MODE mode, int use_wedge_interintra,
 }
 
 static void combine_interintra_highbd(
-    INTERINTRA_MODE mode, int use_wedge_interintra, int wedge_index,
+    INTERINTRA_MODE mode, int8_t use_wedge_interintra, int wedge_index,
     int wedge_sign, BLOCK_SIZE bsize, BLOCK_SIZE plane_bsize,
     uint8_t *comppred8, int compstride, const uint8_t *interpred8,
     int interstride, const uint8_t *intrapred8, int intrastride, int bd) {
@@ -1092,8 +1095,8 @@ static void combine_interintra_highbd(
 void av1_build_intra_predictors_for_interintra(const AV1_COMMON *cm,
                                                MACROBLOCKD *xd,
                                                BLOCK_SIZE bsize, int plane,
-                                               BUFFER_SET *ctx, uint8_t *dst,
-                                               int dst_stride) {
+                                               const BUFFER_SET *ctx,
+                                               uint8_t *dst, int dst_stride) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
   const int ssx = xd->plane[plane].subsampling_x;
   const int ssy = xd->plane[plane].subsampling_y;
@@ -1116,7 +1119,7 @@ void av1_combine_interintra(MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
   const int ssx = xd->plane[plane].subsampling_x;
   const int ssy = xd->plane[plane].subsampling_y;
   const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, ssx, ssy);
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     combine_interintra_highbd(
         xd->mi[0]->interintra_mode, xd->mi[0]->use_wedge_interintra,
         xd->mi[0]->interintra_wedge_index, xd->mi[0]->interintra_wedge_sign,
@@ -1135,9 +1138,9 @@ void av1_combine_interintra(MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
 // build interintra_predictors for one plane
 void av1_build_interintra_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                          uint8_t *pred, int stride,
-                                         BUFFER_SET *ctx, int plane,
+                                         const BUFFER_SET *ctx, int plane,
                                          BLOCK_SIZE bsize) {
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     DECLARE_ALIGNED(16, uint16_t, intrapredictor[MAX_SB_SQUARE]);
     av1_build_intra_predictors_for_interintra(
         cm, xd, bsize, plane, ctx, CONVERT_TO_BYTEPTR(intrapredictor),
@@ -1156,7 +1159,8 @@ void av1_build_interintra_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
 void av1_build_interintra_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                           uint8_t *upred, uint8_t *vpred,
                                           int ustride, int vstride,
-                                          BUFFER_SET *ctx, BLOCK_SIZE bsize) {
+                                          const BUFFER_SET *ctx,
+                                          BLOCK_SIZE bsize) {
   av1_build_interintra_predictors_sbp(cm, xd, upred, ustride, ctx, 1, bsize);
   av1_build_interintra_predictors_sbp(cm, xd, vpred, vstride, ctx, 2, bsize);
 }
diff --git a/third_party/aom/av1/common/reconinter.h b/third_party/aom/av1/common/reconinter.h
index db86c777e3..1385be50f9 100644
--- a/third_party/aom/av1/common/reconinter.h
+++ b/third_party/aom/av1/common/reconinter.h
@@ -47,7 +47,7 @@ extern "C" {
 #define WEDGE_NONE -1
 
 // Angles are with respect to horizontal anti-clockwise
-typedef enum {
+enum {
   WEDGE_HORIZONTAL = 0,
   WEDGE_VERTICAL = 1,
   WEDGE_OBLIQUE27 = 2,
@@ -55,7 +55,7 @@ typedef enum {
   WEDGE_OBLIQUE117 = 4,
   WEDGE_OBLIQUE153 = 5,
   WEDGE_DIRECTIONS
-} WedgeDirectionType;
+} UENUM1BYTE(WedgeDirectionType);
 
 // 3-tuple: {direction, x_offset, y_offset}
 typedef struct {
@@ -333,29 +333,32 @@ const uint8_t *av1_get_compound_type_mask(
 // build interintra_predictors for one plane
 void av1_build_interintra_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                          uint8_t *pred, int stride,
-                                         BUFFER_SET *ctx, int plane,
+                                         const BUFFER_SET *ctx, int plane,
                                          BLOCK_SIZE bsize);
 
 void av1_build_interintra_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                           uint8_t *upred, uint8_t *vpred,
                                           int ustride, int vstride,
-                                          BUFFER_SET *ctx, BLOCK_SIZE bsize);
+                                          const BUFFER_SET *ctx,
+                                          BLOCK_SIZE bsize);
 
 void av1_build_intra_predictors_for_interintra(
     const AV1_COMMON *cm, MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
-    BUFFER_SET *ctx, uint8_t *intra_pred, int intra_stride);
+    const BUFFER_SET *ctx, uint8_t *intra_pred, int intra_stride);
 
 void av1_combine_interintra(MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
                             const uint8_t *inter_pred, int inter_stride,
                             const uint8_t *intra_pred, int intra_stride);
 
-void av1_jnt_comp_weight_assign(const AV1_COMMON *cm, const MB_MODE_INFO *mbmi,
-                                int order_idx, int *fwd_offset, int *bck_offset,
-                                int *use_jnt_comp_avg, int is_compound);
+void av1_dist_wtd_comp_weight_assign(const AV1_COMMON *cm,
+                                     const MB_MODE_INFO *mbmi, int order_idx,
+                                     int *fwd_offset, int *bck_offset,
+                                     int *use_dist_wtd_comp_avg,
+                                     int is_compound);
 int av1_allow_warp(const MB_MODE_INFO *const mbmi,
                    const WarpTypesAllowed *const warp_types,
                    const WarpedMotionParams *const gm_params,
-                   int build_for_obmc, int x_scale, int y_scale,
+                   int build_for_obmc, const struct scale_factors *const sf,
                    WarpedMotionParams *final_warp_params);
 
 #ifdef __cplusplus
diff --git a/third_party/aom/av1/common/reconintra.c b/third_party/aom/av1/common/reconintra.c
index 71a52e73e5..559e49921b 100644
--- a/third_party/aom/av1/common/reconintra.c
+++ b/third_party/aom/av1/common/reconintra.c
@@ -556,33 +556,37 @@ void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                             const uint8_t *above, const uint8_t *left,
                             int upsample_above, int upsample_left, int dx,
                             int dy) {
-  int r, c, x, y, shift1, shift2, val, base1, base2;
-
   assert(dx > 0);
   assert(dy > 0);
 
   const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  (void)min_base_y;
   const int frac_bits_x = 6 - upsample_above;
   const int frac_bits_y = 6 - upsample_left;
-  const int base_inc_x = 1 << upsample_above;
-  x = -dx;
-  for (r = 0; r < bh; ++r, x -= dx, dst += stride) {
-    base1 = x >> frac_bits_x;
-    y = (r << 6) - dy;
-    for (c = 0; c < bw; ++c, base1 += base_inc_x, y -= dy) {
-      if (base1 >= min_base_x) {
-        shift1 = ((x * (1 << upsample_above)) & 0x3F) >> 1;
-        val = above[base1] * (32 - shift1) + above[base1 + 1] * shift1;
+
+  for (int r = 0; r < bh; ++r) {
+    for (int c = 0; c < bw; ++c) {
+      int val;
+      int y = r + 1;
+      int x = (c << 6) - y * dx;
+      const int base_x = x >> frac_bits_x;
+      if (base_x >= min_base_x) {
+        const int shift = ((x * (1 << upsample_above)) & 0x3F) >> 1;
+        val = above[base_x] * (32 - shift) + above[base_x + 1] * shift;
         val = ROUND_POWER_OF_TWO(val, 5);
       } else {
-        base2 = y >> frac_bits_y;
-        assert(base2 >= -(1 << upsample_left));
-        shift2 = ((y * (1 << upsample_left)) & 0x3F) >> 1;
-        val = left[base2] * (32 - shift2) + left[base2 + 1] * shift2;
+        x = c + 1;
+        y = (r << 6) - x * dy;
+        const int base_y = y >> frac_bits_y;
+        assert(base_y >= min_base_y);
+        const int shift = ((y * (1 << upsample_left)) & 0x3F) >> 1;
+        val = left[base_y] * (32 - shift) + left[base_y + 1] * shift;
         val = ROUND_POWER_OF_TWO(val, 5);
       }
       dst[c] = val;
     }
+    dst += stride;
   }
 }
 
@@ -688,30 +692,33 @@ void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw,
                                    int bh, const uint16_t *above,
                                    const uint16_t *left, int upsample_above,
                                    int upsample_left, int dx, int dy, int bd) {
-  int r, c, x, y, shift, val, base;
-
   (void)bd;
   assert(dx > 0);
   assert(dy > 0);
 
   const int min_base_x = -(1 << upsample_above);
+  const int min_base_y = -(1 << upsample_left);
+  (void)min_base_y;
   const int frac_bits_x = 6 - upsample_above;
   const int frac_bits_y = 6 - upsample_left;
-  for (r = 0; r < bh; ++r) {
-    for (c = 0; c < bw; ++c) {
-      y = r + 1;
-      x = (c << 6) - y * dx;
-      base = x >> frac_bits_x;
-      if (base >= min_base_x) {
-        shift = ((x * (1 << upsample_above)) & 0x3F) >> 1;
-        val = above[base] * (32 - shift) + above[base + 1] * shift;
+
+  for (int r = 0; r < bh; ++r) {
+    for (int c = 0; c < bw; ++c) {
+      int val;
+      int y = r + 1;
+      int x = (c << 6) - y * dx;
+      const int base_x = x >> frac_bits_x;
+      if (base_x >= min_base_x) {
+        const int shift = ((x * (1 << upsample_above)) & 0x3F) >> 1;
+        val = above[base_x] * (32 - shift) + above[base_x + 1] * shift;
         val = ROUND_POWER_OF_TWO(val, 5);
       } else {
         x = c + 1;
         y = (r << 6) - x * dy;
-        base = y >> frac_bits_y;
-        shift = ((y * (1 << upsample_left)) & 0x3F) >> 1;
-        val = left[base] * (32 - shift) + left[base + 1] * shift;
+        const int base_y = y >> frac_bits_y;
+        assert(base_y >= min_base_y);
+        const int shift = ((y * (1 << upsample_left)) & 0x3F) >> 1;
+        val = left[base_y] * (32 - shift) + left[base_y + 1] * shift;
         val = ROUND_POWER_OF_TWO(val, 5);
       }
       dst[c] = val;
@@ -1503,7 +1510,7 @@ void av1_predict_intra_block(
                                xd->color_index_map_offset[plane != 0];
     const uint16_t *const palette =
         mbmi->palette_mode_info.palette_colors + plane * PALETTE_MAX_SIZE;
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
       for (r = 0; r < txhpx; ++r) {
         for (c = 0; c < txwpx; ++c) {
@@ -1562,7 +1569,7 @@ void av1_predict_intra_block(
       tx_size, row_off, col_off, pd->subsampling_x, pd->subsampling_y);
 
   const int disable_edge_filter = !cm->seq_params.enable_intra_edge_filter;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     build_intra_predictors_high(
         xd, ref, ref_stride, dst, dst_stride, mode, angle_delta,
         filter_intra_mode, tx_size, disable_edge_filter,
diff --git a/third_party/aom/av1/common/reconintra.h b/third_party/aom/av1/common/reconintra.h
index 07853aba01..3c781abb52 100644
--- a/third_party/aom/av1/common/reconintra.h
+++ b/third_party/aom/av1/common/reconintra.h
@@ -110,7 +110,7 @@ static INLINE int av1_use_intra_edge_upsample(int bs0, int bs1, int delta,
                                               int type) {
   const int d = abs(delta);
   const int blk_wh = bs0 + bs1;
-  if (d <= 0 || d >= 40) return 0;
+  if (d == 0 || d >= 40) return 0;
   return type ? (blk_wh <= 8) : (blk_wh <= 16);
 }
 #ifdef __cplusplus
diff --git a/third_party/aom/av1/common/resize.c b/third_party/aom/av1/common/resize.c
index d61a20aa2d..8b24ed01b9 100644
--- a/third_party/aom/av1/common/resize.c
+++ b/third_party/aom/av1/common/resize.c
@@ -313,6 +313,91 @@ static void interpolate_core(const uint8_t *const input, int in_length,
   }
 }
 
+static void interpolate_core_double_prec(const double *const input,
+                                         int in_length, double *output,
+                                         int out_length,
+                                         const int16_t *interp_filters,
+                                         int interp_taps) {
+  const int32_t delta =
+      (((uint32_t)in_length << RS_SCALE_SUBPEL_BITS) + out_length / 2) /
+      out_length;
+  const int32_t offset =
+      in_length > out_length
+          ? (((int32_t)(in_length - out_length) << (RS_SCALE_SUBPEL_BITS - 1)) +
+             out_length / 2) /
+                out_length
+          : -(((int32_t)(out_length - in_length)
+               << (RS_SCALE_SUBPEL_BITS - 1)) +
+              out_length / 2) /
+                out_length;
+  double *optr = output;
+  int x, x1, x2, k, int_pel, sub_pel;
+  double sum;
+  int32_t y;
+
+  x = 0;
+  y = offset + RS_SCALE_EXTRA_OFF;
+  while ((y >> RS_SCALE_SUBPEL_BITS) < (interp_taps / 2 - 1)) {
+    x++;
+    y += delta;
+  }
+  x1 = x;
+  x = out_length - 1;
+  y = delta * x + offset + RS_SCALE_EXTRA_OFF;
+  while ((y >> RS_SCALE_SUBPEL_BITS) + (int32_t)(interp_taps / 2) >=
+         in_length) {
+    x--;
+    y -= delta;
+  }
+  x2 = x;
+  if (x1 > x2) {
+    for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < out_length;
+         ++x, y += delta) {
+      int_pel = y >> RS_SCALE_SUBPEL_BITS;
+      sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK;
+      const int16_t *filter = &interp_filters[sub_pel * interp_taps];
+      sum = 0;
+      for (k = 0; k < interp_taps; ++k) {
+        const int pk = int_pel - interp_taps / 2 + 1 + k;
+        sum += filter[k] * input[AOMMAX(AOMMIN(pk, in_length - 1), 0)];
+      }
+      *optr++ = sum / (1 << FILTER_BITS);
+    }
+  } else {
+    // Initial part.
+    for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < x1; ++x, y += delta) {
+      int_pel = y >> RS_SCALE_SUBPEL_BITS;
+      sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK;
+      const int16_t *filter = &interp_filters[sub_pel * interp_taps];
+      sum = 0;
+      for (k = 0; k < interp_taps; ++k)
+        sum += filter[k] * input[AOMMAX(int_pel - interp_taps / 2 + 1 + k, 0)];
+      *optr++ = sum / (1 << FILTER_BITS);
+    }
+    // Middle part.
+    for (; x <= x2; ++x, y += delta) {
+      int_pel = y >> RS_SCALE_SUBPEL_BITS;
+      sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK;
+      const int16_t *filter = &interp_filters[sub_pel * interp_taps];
+      sum = 0;
+      for (k = 0; k < interp_taps; ++k)
+        sum += filter[k] * input[int_pel - interp_taps / 2 + 1 + k];
+      *optr++ = sum / (1 << FILTER_BITS);
+    }
+    // End part.
+    for (; x < out_length; ++x, y += delta) {
+      int_pel = y >> RS_SCALE_SUBPEL_BITS;
+      sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK;
+      const int16_t *filter = &interp_filters[sub_pel * interp_taps];
+      sum = 0;
+      for (k = 0; k < interp_taps; ++k)
+        sum += filter[k] *
+               input[AOMMIN(int_pel - interp_taps / 2 + 1 + k, in_length - 1)];
+      *optr++ = sum / (1 << FILTER_BITS);
+    }
+  }
+}
+
 static void interpolate(const uint8_t *const input, int in_length,
                         uint8_t *output, int out_length) {
   const InterpKernel *interp_filters =
@@ -322,6 +407,15 @@ static void interpolate(const uint8_t *const input, int in_length,
                    SUBPEL_TAPS);
 }
 
+static void interpolate_double_prec(const double *const input, int in_length,
+                                    double *output, int out_length) {
+  const InterpKernel *interp_filters =
+      choose_interp_filter(in_length, out_length);
+
+  interpolate_core_double_prec(input, in_length, output, out_length,
+                               &interp_filters[0][0], SUBPEL_TAPS);
+}
+
 int32_t av1_get_upscale_convolve_step(int in_length, int out_length) {
   return ((in_length << RS_SCALE_SUBPEL_BITS) + out_length / 2) / out_length;
 }
@@ -505,6 +599,12 @@ static void resize_multistep(const uint8_t *const input, int length,
   }
 }
 
+static void upscale_multistep_double_prec(const double *const input, int length,
+                                          double *output, int olength) {
+  assert(length < olength);
+  interpolate_double_prec(input, length, output, olength);
+}
+
 static void fill_col_to_arr(uint8_t *img, int stride, int len, uint8_t *arr) {
   int i;
   uint8_t *iptr = img;
@@ -523,9 +623,29 @@ static void fill_arr_to_col(uint8_t *img, int stride, int len, uint8_t *arr) {
   }
 }
 
-static void resize_plane(const uint8_t *const input, int height, int width,
-                         int in_stride, uint8_t *output, int height2,
-                         int width2, int out_stride) {
+static void fill_col_to_arr_double_prec(double *img, int stride, int len,
+                                        double *arr) {
+  int i;
+  double *iptr = img;
+  double *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *aptr++ = *iptr;
+  }
+}
+
+static void fill_arr_to_col_double_prec(double *img, int stride, int len,
+                                        double *arr) {
+  int i;
+  double *iptr = img;
+  double *aptr = arr;
+  for (i = 0; i < len; ++i, iptr += stride) {
+    *iptr = *aptr++;
+  }
+}
+
+void av1_resize_plane(const uint8_t *const input, int height, int width,
+                      int in_stride, uint8_t *output, int height2, int width2,
+                      int out_stride) {
   int i;
   uint8_t *intbuf = (uint8_t *)aom_malloc(sizeof(uint8_t) * width2 * height);
   uint8_t *tmpbuf =
@@ -554,6 +674,33 @@ Error:
   aom_free(arrbuf2);
 }
 
+void av1_upscale_plane_double_prec(const double *const input, int height,
+                                   int width, int in_stride, double *output,
+                                   int height2, int width2, int out_stride) {
+  int i;
+  double *intbuf = (double *)aom_malloc(sizeof(double) * width2 * height);
+  double *arrbuf = (double *)aom_malloc(sizeof(double) * height);
+  double *arrbuf2 = (double *)aom_malloc(sizeof(double) * height2);
+  if (intbuf == NULL || arrbuf == NULL || arrbuf2 == NULL) goto Error;
+  assert(width > 0);
+  assert(height > 0);
+  assert(width2 > 0);
+  assert(height2 > 0);
+  for (i = 0; i < height; ++i)
+    upscale_multistep_double_prec(input + in_stride * i, width,
+                                  intbuf + width2 * i, width2);
+  for (i = 0; i < width2; ++i) {
+    fill_col_to_arr_double_prec(intbuf + i, width2, height, arrbuf);
+    upscale_multistep_double_prec(arrbuf, height, arrbuf2, height2);
+    fill_arr_to_col_double_prec(output + i, out_stride, height2, arrbuf2);
+  }
+
+Error:
+  aom_free(intbuf);
+  aom_free(arrbuf);
+  aom_free(arrbuf2);
+}
+
 static void upscale_normative_rect(const uint8_t *const input, int height,
                                    int width, int in_stride, uint8_t *output,
                                    int height2, int width2, int out_stride,
@@ -871,10 +1018,9 @@ static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
   }
 }
 
-static void highbd_resize_plane(const uint8_t *const input, int height,
-                                int width, int in_stride, uint8_t *output,
-                                int height2, int width2, int out_stride,
-                                int bd) {
+void av1_highbd_resize_plane(const uint8_t *const input, int height, int width,
+                             int in_stride, uint8_t *output, int height2,
+                             int width2, int out_stride, int bd) {
   int i;
   uint16_t *intbuf = (uint16_t *)aom_malloc(sizeof(uint16_t) * width2 * height);
   uint16_t *tmpbuf =
@@ -969,11 +1115,11 @@ void av1_resize_frame420(const uint8_t *const y, int y_stride,
                          int uv_stride, int height, int width, uint8_t *oy,
                          int oy_stride, uint8_t *ou, uint8_t *ov,
                          int ouv_stride, int oheight, int owidth) {
-  resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
-  resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2, owidth / 2,
-               ouv_stride);
-  resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2, owidth / 2,
-               ouv_stride);
+  av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
+  av1_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
+                   owidth / 2, ouv_stride);
+  av1_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
+                   owidth / 2, ouv_stride);
 }
 
 void av1_resize_frame422(const uint8_t *const y, int y_stride,
@@ -981,11 +1127,11 @@ void av1_resize_frame422(const uint8_t *const y, int y_stride,
                          int uv_stride, int height, int width, uint8_t *oy,
                          int oy_stride, uint8_t *ou, uint8_t *ov,
                          int ouv_stride, int oheight, int owidth) {
-  resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
-  resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
-               ouv_stride);
-  resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
-               ouv_stride);
+  av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
+  av1_resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
+                   ouv_stride);
+  av1_resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
+                   ouv_stride);
 }
 
 void av1_resize_frame444(const uint8_t *const y, int y_stride,
@@ -993,9 +1139,11 @@ void av1_resize_frame444(const uint8_t *const y, int y_stride,
                          int uv_stride, int height, int width, uint8_t *oy,
                          int oy_stride, uint8_t *ou, uint8_t *ov,
                          int ouv_stride, int oheight, int owidth) {
-  resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
-  resize_plane(u, height, width, uv_stride, ou, oheight, owidth, ouv_stride);
-  resize_plane(v, height, width, uv_stride, ov, oheight, owidth, ouv_stride);
+  av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
+  av1_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
+                   ouv_stride);
+  av1_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
+                   ouv_stride);
 }
 
 void av1_highbd_resize_frame420(const uint8_t *const y, int y_stride,
@@ -1004,12 +1152,12 @@ void av1_highbd_resize_frame420(const uint8_t *const y, int y_stride,
                                 uint8_t *oy, int oy_stride, uint8_t *ou,
                                 uint8_t *ov, int ouv_stride, int oheight,
                                 int owidth, int bd) {
-  highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
-                      oy_stride, bd);
-  highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
-                      owidth / 2, ouv_stride, bd);
-  highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
-                      owidth / 2, ouv_stride, bd);
+  av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
+                          oy_stride, bd);
+  av1_highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
+                          owidth / 2, ouv_stride, bd);
+  av1_highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
+                          owidth / 2, ouv_stride, bd);
 }
 
 void av1_highbd_resize_frame422(const uint8_t *const y, int y_stride,
@@ -1018,12 +1166,12 @@ void av1_highbd_resize_frame422(const uint8_t *const y, int y_stride,
                                 uint8_t *oy, int oy_stride, uint8_t *ou,
                                 uint8_t *ov, int ouv_stride, int oheight,
                                 int owidth, int bd) {
-  highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
-                      oy_stride, bd);
-  highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
-                      ouv_stride, bd);
-  highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
-                      ouv_stride, bd);
+  av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
+                          oy_stride, bd);
+  av1_highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight,
+                          owidth / 2, ouv_stride, bd);
+  av1_highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight,
+                          owidth / 2, ouv_stride, bd);
 }
 
 void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
@@ -1032,12 +1180,12 @@ void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
                                 uint8_t *oy, int oy_stride, uint8_t *ou,
                                 uint8_t *ov, int ouv_stride, int oheight,
                                 int owidth, int bd) {
-  highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
-                      oy_stride, bd);
-  highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
-                      ouv_stride, bd);
-  highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
-                      ouv_stride, bd);
+  av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
+                          oy_stride, bd);
+  av1_highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
+                          ouv_stride, bd);
+  av1_highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
+                          ouv_stride, bd);
 }
 
 void av1_resize_and_extend_frame(const YV12_BUFFER_CONFIG *src,
@@ -1050,15 +1198,15 @@ void av1_resize_and_extend_frame(const YV12_BUFFER_CONFIG *src,
   for (int i = 0; i < AOMMIN(num_planes, MAX_MB_PLANE); ++i) {
     const int is_uv = i > 0;
     if (src->flags & YV12_FLAG_HIGHBITDEPTH)
-      highbd_resize_plane(src->buffers[i], src->crop_heights[is_uv],
-                          src->crop_widths[is_uv], src->strides[is_uv],
-                          dst->buffers[i], dst->crop_heights[is_uv],
-                          dst->crop_widths[is_uv], dst->strides[is_uv], bd);
+      av1_highbd_resize_plane(src->buffers[i], src->crop_heights[is_uv],
+                              src->crop_widths[is_uv], src->strides[is_uv],
+                              dst->buffers[i], dst->crop_heights[is_uv],
+                              dst->crop_widths[is_uv], dst->strides[is_uv], bd);
     else
-      resize_plane(src->buffers[i], src->crop_heights[is_uv],
-                   src->crop_widths[is_uv], src->strides[is_uv],
-                   dst->buffers[i], dst->crop_heights[is_uv],
-                   dst->crop_widths[is_uv], dst->strides[is_uv]);
+      av1_resize_plane(src->buffers[i], src->crop_heights[is_uv],
+                       src->crop_widths[is_uv], src->strides[is_uv],
+                       dst->buffers[i], dst->crop_heights[is_uv],
+                       dst->crop_widths[is_uv], dst->strides[is_uv]);
   }
   aom_extend_frame_borders(dst, num_planes);
 }
@@ -1155,10 +1303,19 @@ YV12_BUFFER_CONFIG *av1_scale_if_required(AV1_COMMON *cm,
 // denominator.
 static void calculate_scaled_size_helper(int *dim, int denom) {
   if (denom != SCALE_NUMERATOR) {
+    // We need to ensure the constraint in "Appendix A" of the spec:
+    // * FrameWidth is greater than or equal to 16
+    // * FrameHeight is greater than or equal to 16
+    // For this, we clamp the downscaled dimension to at least 16. One
+    // exception: if original dimension itself was < 16, then we keep the
+    // downscaled dimension to be same as the original, to ensure that resizing
+    // is valid.
+    const int min_dim = AOMMIN(16, *dim);
     // Use this version if we need *dim to be even
     // *width = (*width * SCALE_NUMERATOR + denom) / (2 * denom);
     // *width <<= 1;
     *dim = (*dim * SCALE_NUMERATOR + denom / 2) / (denom);
+    *dim = AOMMAX(*dim, min_dim);
   }
 }
 
@@ -1205,7 +1362,7 @@ void av1_superres_upscale(AV1_COMMON *cm, BufferPool *const pool) {
   YV12_BUFFER_CONFIG copy_buffer;
   memset(&copy_buffer, 0, sizeof(copy_buffer));
 
-  YV12_BUFFER_CONFIG *const frame_to_show = get_frame_new_buffer(cm);
+  YV12_BUFFER_CONFIG *const frame_to_show = &cm->cur_frame->buf;
 
   const int aligned_width = ALIGN_POWER_OF_TWO(cm->width, 3);
   if (aom_alloc_frame_buffer(
@@ -1225,8 +1382,7 @@ void av1_superres_upscale(AV1_COMMON *cm, BufferPool *const pool) {
   // Realloc the current frame buffer at a higher resolution in place.
   if (pool != NULL) {
     // Use callbacks if on the decoder.
-    aom_codec_frame_buffer_t *fb =
-        &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer;
+    aom_codec_frame_buffer_t *fb = &cm->cur_frame->raw_frame_buffer;
     aom_release_frame_buffer_cb_fn_t release_fb_cb = pool->release_fb_cb;
     aom_get_frame_buffer_cb_fn_t cb = pool->get_fb_cb;
     void *cb_priv = pool->cb_priv;
diff --git a/third_party/aom/av1/common/resize.h b/third_party/aom/av1/common/resize.h
index 9a59a8d63b..43bea58b82 100644
--- a/third_party/aom/av1/common/resize.h
+++ b/third_party/aom/av1/common/resize.h
@@ -23,6 +23,9 @@ extern "C" {
 void av1_resize_plane(const uint8_t *const input, int height, int width,
                       int in_stride, uint8_t *output, int height2, int width2,
                       int out_stride);
+void av1_upscale_plane_double_prec(const double *const input, int height,
+                                   int width, int in_stride, double *output,
+                                   int height2, int width2, int out_stride);
 void av1_resize_frame420(const uint8_t *const y, int y_stride,
                          const uint8_t *const u, const uint8_t *const v,
                          int uv_stride, int height, int width, uint8_t *oy,
diff --git a/third_party/aom/av1/common/restoration.c b/third_party/aom/av1/common/restoration.c
index d276a915b5..9e472b8a6e 100644
--- a/third_party/aom/av1/common/restoration.c
+++ b/third_party/aom/av1/common/restoration.c
@@ -1072,13 +1072,6 @@ void av1_loop_restoration_filter_unit(
   }
 }
 
-static void filter_frame_on_tile(int tile_row, int tile_col, void *priv,
-                                 AV1_COMMON *cm) {
-  (void)tile_col;
-  FilterFrameCtxt *ctxt = (FilterFrameCtxt *)priv;
-  ctxt->tile_stripe0 = (tile_row == 0) ? 0 : cm->rst_end_stripe[tile_row - 1];
-}
-
 static void filter_frame_on_unit(const RestorationTileLimits *limits,
                                  const AV1PixelRect *tile_rect,
                                  int rest_unit_idx, void *priv, int32_t *tmpbuf,
@@ -1106,7 +1099,7 @@ void av1_loop_restoration_filter_frame_init(AV1LrStruct *lr_ctxt,
   const int frame_height = frame->crop_heights[0];
   if (aom_realloc_frame_buffer(
           lr_ctxt->dst, frame_width, frame_height, seq_params->subsampling_x,
-          seq_params->subsampling_y, highbd, AOM_BORDER_IN_PIXELS,
+          seq_params->subsampling_y, highbd, AOM_RESTORATION_FRAME_BORDER,
           cm->byte_alignment, NULL, NULL, NULL) < 0)
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate restoration dst buffer");
@@ -1141,7 +1134,7 @@ void av1_loop_restoration_filter_frame_init(AV1LrStruct *lr_ctxt,
     lr_plane_ctxt->data_stride = frame->strides[is_uv];
     lr_plane_ctxt->dst_stride = lr_ctxt->dst->strides[is_uv];
     lr_plane_ctxt->tile_rect = av1_whole_frame_rect(cm, is_uv);
-    filter_frame_on_tile(LR_TILE_ROW, LR_TILE_COL, lr_plane_ctxt, cm);
+    lr_plane_ctxt->tile_stripe0 = 0;
   }
 }
 
@@ -1150,9 +1143,9 @@ void av1_loop_restoration_copy_planes(AV1LrStruct *loop_rest_ctxt,
   typedef void (*copy_fun)(const YV12_BUFFER_CONFIG *src_ybc,
                            YV12_BUFFER_CONFIG *dst_ybc, int hstart, int hend,
                            int vstart, int vend);
-  static const copy_fun copy_funs[3] = {
-    aom_yv12_partial_copy_y, aom_yv12_partial_copy_u, aom_yv12_partial_copy_v
-  };
+  static const copy_fun copy_funs[3] = { aom_yv12_partial_coloc_copy_y,
+                                         aom_yv12_partial_coloc_copy_u,
+                                         aom_yv12_partial_coloc_copy_v };
 
   for (int plane = 0; plane < num_planes; ++plane) {
     if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue;
diff --git a/third_party/aom/av1/common/restoration.h b/third_party/aom/av1/common/restoration.h
index d834f9270f..5d196a9a59 100644
--- a/third_party/aom/av1/common/restoration.h
+++ b/third_party/aom/av1/common/restoration.h
@@ -22,6 +22,8 @@
 extern "C" {
 #endif
 
+// Boarder for Loop restoration buffer
+#define AOM_RESTORATION_FRAME_BORDER 32
 #define CLIP(x, lo, hi) ((x) < (lo) ? (lo) : (x) > (hi) ? (hi) : (x))
 #define RINT(x) ((x) < 0 ? (int)((x)-0.5) : (int)((x) + 0.5))
 
diff --git a/third_party/aom/av1/common/scale.c b/third_party/aom/av1/common/scale.c
index c525fe2296..bac7bd9cbd 100644
--- a/third_party/aom/av1/common/scale.c
+++ b/third_party/aom/av1/common/scale.c
@@ -97,13 +97,13 @@ void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
   // subpel_x_q4 != 0 && subpel_y_q4 != 0
   sf->convolve[1][1][0] = av1_convolve_2d_sr;
   // subpel_x_q4 == 0 && subpel_y_q4 == 0
-  sf->convolve[0][0][1] = av1_jnt_convolve_2d_copy;
+  sf->convolve[0][0][1] = av1_dist_wtd_convolve_2d_copy;
   // subpel_x_q4 == 0
-  sf->convolve[0][1][1] = av1_jnt_convolve_y;
+  sf->convolve[0][1][1] = av1_dist_wtd_convolve_y;
   // subpel_y_q4 == 0
-  sf->convolve[1][0][1] = av1_jnt_convolve_x;
+  sf->convolve[1][0][1] = av1_dist_wtd_convolve_x;
   // subpel_x_q4 != 0 && subpel_y_q4 != 0
-  sf->convolve[1][1][1] = av1_jnt_convolve_2d;
+  sf->convolve[1][1][1] = av1_dist_wtd_convolve_2d;
   // AV1 High BD convolve functions
   // Special case convolve functions should produce the same result as
   // av1_highbd_convolve_2d.
@@ -116,11 +116,11 @@ void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
   // subpel_x_q4 != 0 && subpel_y_q4 != 0
   sf->highbd_convolve[1][1][0] = av1_highbd_convolve_2d_sr;
   // subpel_x_q4 == 0 && subpel_y_q4 == 0
-  sf->highbd_convolve[0][0][1] = av1_highbd_jnt_convolve_2d_copy;
+  sf->highbd_convolve[0][0][1] = av1_highbd_dist_wtd_convolve_2d_copy;
   // subpel_x_q4 == 0
-  sf->highbd_convolve[0][1][1] = av1_highbd_jnt_convolve_y;
+  sf->highbd_convolve[0][1][1] = av1_highbd_dist_wtd_convolve_y;
   // subpel_y_q4 == 0
-  sf->highbd_convolve[1][0][1] = av1_highbd_jnt_convolve_x;
+  sf->highbd_convolve[1][0][1] = av1_highbd_dist_wtd_convolve_x;
   // subpel_x_q4 != 0 && subpel_y_q4 != 0
-  sf->highbd_convolve[1][1][1] = av1_highbd_jnt_convolve_2d;
+  sf->highbd_convolve[1][1][1] = av1_highbd_dist_wtd_convolve_2d;
 }
diff --git a/third_party/aom/av1/common/scan.h b/third_party/aom/av1/common/scan.h
index 233dc0efa2..f9c339215b 100644
--- a/third_party/aom/av1/common/scan.h
+++ b/third_party/aom/av1/common/scan.h
@@ -25,14 +25,14 @@ extern "C" {
 
 #define MAX_NEIGHBORS 2
 
-typedef enum SCAN_MODE {
+enum {
   SCAN_MODE_ZIG_ZAG,
   SCAN_MODE_COL_DIAG,
   SCAN_MODE_ROW_DIAG,
   SCAN_MODE_COL_1D,
   SCAN_MODE_ROW_1D,
   SCAN_MODES
-} SCAN_MODE;
+} UENUM1BYTE(SCAN_MODE);
 
 extern const SCAN_ORDER av1_default_scan_orders[TX_SIZES];
 extern const SCAN_ORDER av1_scan_orders[TX_SIZES_ALL][TX_TYPES];
diff --git a/third_party/aom/av1/common/seg_common.c b/third_party/aom/av1/common/seg_common.c
index cd189ad769..4650903b91 100644
--- a/third_party/aom/av1/common/seg_common.c
+++ b/third_party/aom/av1/common/seg_common.c
@@ -16,12 +16,19 @@
 #include "av1/common/seg_common.h"
 #include "av1/common/quant_common.h"
 
-static const int seg_feature_data_signed[SEG_LVL_MAX] = { 1, 1, 1, 1, 1, 0, 0 };
-
-static const int seg_feature_data_max[SEG_LVL_MAX] = {
-  MAXQ, MAX_LOOP_FILTER, MAX_LOOP_FILTER, MAX_LOOP_FILTER, MAX_LOOP_FILTER, 7, 0
+static const int seg_feature_data_signed[SEG_LVL_MAX] = {
+  1, 1, 1, 1, 1, 0, 0, 0
 };
 
+static const int seg_feature_data_max[SEG_LVL_MAX] = { MAXQ,
+                                                       MAX_LOOP_FILTER,
+                                                       MAX_LOOP_FILTER,
+                                                       MAX_LOOP_FILTER,
+                                                       MAX_LOOP_FILTER,
+                                                       7,
+                                                       0,
+                                                       0 };
+
 // These functions provide access to new segment level features.
 // Eventually these function may be "optimized out" but for the moment,
 // the coding mechanism is still subject to change so these provide a
diff --git a/third_party/aom/av1/common/seg_common.h b/third_party/aom/av1/common/seg_common.h
index 8c35bba86c..fa7894cf61 100644
--- a/third_party/aom/av1/common/seg_common.h
+++ b/third_party/aom/av1/common/seg_common.h
@@ -24,7 +24,7 @@ extern "C" {
 #define SEG_TEMPORAL_PRED_CTXS 3
 #define SPATIAL_PREDICTION_PROBS 3
 
-typedef enum {
+enum {
   SEG_LVL_ALT_Q,       // Use alternate Quantizer ....
   SEG_LVL_ALT_LF_Y_V,  // Use alternate loop filter value on y plane vertical
   SEG_LVL_ALT_LF_Y_H,  // Use alternate loop filter value on y plane horizontal
@@ -34,7 +34,7 @@ typedef enum {
   SEG_LVL_SKIP,        // Optional Segment (0,0) + skip mode
   SEG_LVL_GLOBALMV,
   SEG_LVL_MAX
-} SEG_LVL_FEATURES;
+} UENUM1BYTE(SEG_LVL_FEATURES);
 
 struct segmentation {
   uint8_t enabled;
diff --git a/third_party/aom/av1/common/thread_common.c b/third_party/aom/av1/common/thread_common.c
index 8df4c9a09d..14406e6444 100644
--- a/third_party/aom/av1/common/thread_common.c
+++ b/third_party/aom/av1/common/thread_common.c
@@ -205,7 +205,11 @@ static INLINE void sync_write(AV1LfSync *const lf_sync, int r, int c,
 }
 
 static void enqueue_lf_jobs(AV1LfSync *lf_sync, AV1_COMMON *cm, int start,
-                            int stop, int plane_start, int plane_end) {
+                            int stop,
+#if LOOP_FILTER_BITMASK
+                            int is_decoding,
+#endif
+                            int plane_start, int plane_end) {
   int mi_row, plane, dir;
   AV1LfMTInfo *lf_job_queue = lf_sync->job_queue;
   lf_sync->jobs_enqueued = 0;
@@ -219,7 +223,16 @@ static void enqueue_lf_jobs(AV1LfSync *lf_sync, AV1_COMMON *cm, int start,
         continue;
       else if (plane == 2 && !(cm->lf.filter_level_v))
         continue;
-      for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
+#if LOOP_FILTER_BITMASK
+      int step = MAX_MIB_SIZE;
+      if (is_decoding) {
+        step = MI_SIZE_64X64;
+      }
+      for (mi_row = start; mi_row < stop; mi_row += step)
+#else
+      for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE)
+#endif
+      {
         lf_job_queue->mi_row = mi_row;
         lf_job_queue->plane = plane;
         lf_job_queue->dir = dir;
@@ -230,7 +243,7 @@ static void enqueue_lf_jobs(AV1LfSync *lf_sync, AV1_COMMON *cm, int start,
   }
 }
 
-AV1LfMTInfo *get_lf_job_info(AV1LfSync *lf_sync) {
+static AV1LfMTInfo *get_lf_job_info(AV1LfSync *lf_sync) {
   AV1LfMTInfo *cur_job_info = NULL;
 
 #if CONFIG_MULTITHREAD
@@ -312,15 +325,94 @@ static int loop_filter_row_worker(void *arg1, void *arg2) {
   return 1;
 }
 
+#if LOOP_FILTER_BITMASK
+static INLINE void thread_loop_filter_bitmask_rows(
+    const YV12_BUFFER_CONFIG *const frame_buffer, AV1_COMMON *const cm,
+    struct macroblockd_plane *planes, MACROBLOCKD *xd,
+    AV1LfSync *const lf_sync) {
+  const int sb_cols =
+      ALIGN_POWER_OF_TWO(cm->mi_cols, MIN_MIB_SIZE_LOG2) >> MIN_MIB_SIZE_LOG2;
+  int mi_row, mi_col, plane, dir;
+  int r, c;
+  (void)xd;
+
+  while (1) {
+    AV1LfMTInfo *cur_job_info = get_lf_job_info(lf_sync);
+
+    if (cur_job_info != NULL) {
+      mi_row = cur_job_info->mi_row;
+      plane = cur_job_info->plane;
+      dir = cur_job_info->dir;
+      r = mi_row >> MIN_MIB_SIZE_LOG2;
+
+      if (dir == 0) {
+        for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_SIZE_64X64) {
+          c = mi_col >> MIN_MIB_SIZE_LOG2;
+
+          av1_setup_dst_planes(planes, BLOCK_64X64, frame_buffer, mi_row,
+                               mi_col, plane, plane + 1);
+
+          av1_filter_block_plane_bitmask_vert(cm, &planes[plane], plane, mi_row,
+                                              mi_col);
+          sync_write(lf_sync, r, c, sb_cols, plane);
+        }
+      } else if (dir == 1) {
+        for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_SIZE_64X64) {
+          c = mi_col >> MIN_MIB_SIZE_LOG2;
+
+          // Wait for vertical edge filtering of the top-right block to be
+          // completed
+          sync_read(lf_sync, r, c, plane);
+
+          // Wait for vertical edge filtering of the right block to be
+          // completed
+          sync_read(lf_sync, r + 1, c, plane);
+
+          av1_setup_dst_planes(planes, BLOCK_64X64, frame_buffer, mi_row,
+                               mi_col, plane, plane + 1);
+          av1_filter_block_plane_bitmask_horz(cm, &planes[plane], plane, mi_row,
+                                              mi_col);
+        }
+      }
+    } else {
+      break;
+    }
+  }
+}
+
+// Row-based multi-threaded loopfilter hook
+static int loop_filter_bitmask_row_worker(void *arg1, void *arg2) {
+  AV1LfSync *const lf_sync = (AV1LfSync *)arg1;
+  LFWorkerData *const lf_data = (LFWorkerData *)arg2;
+  thread_loop_filter_bitmask_rows(lf_data->frame_buffer, lf_data->cm,
+                                  lf_data->planes, lf_data->xd, lf_sync);
+  return 1;
+}
+#endif  // LOOP_FILTER_BITMASK
+
 static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
                                 MACROBLOCKD *xd, int start, int stop,
                                 int plane_start, int plane_end,
+#if LOOP_FILTER_BITMASK
+                                int is_decoding,
+#endif
                                 AVxWorker *workers, int nworkers,
                                 AV1LfSync *lf_sync) {
   const AVxWorkerInterface *const winterface = aom_get_worker_interface();
+#if LOOP_FILTER_BITMASK
+  int sb_rows;
+  if (is_decoding) {
+    sb_rows =
+        ALIGN_POWER_OF_TWO(cm->mi_rows, MIN_MIB_SIZE_LOG2) >> MIN_MIB_SIZE_LOG2;
+  } else {
+    sb_rows =
+        ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2) >> MAX_MIB_SIZE_LOG2;
+  }
+#else
   // Number of superblock rows and cols
   const int sb_rows =
       ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2) >> MAX_MIB_SIZE_LOG2;
+#endif
   const int num_workers = nworkers;
   int i;
 
@@ -336,14 +428,26 @@ static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
            sizeof(*(lf_sync->cur_sb_col[i])) * sb_rows);
   }
 
-  enqueue_lf_jobs(lf_sync, cm, start, stop, plane_start, plane_end);
+  enqueue_lf_jobs(lf_sync, cm, start, stop,
+#if LOOP_FILTER_BITMASK
+                  is_decoding,
+#endif
+                  plane_start, plane_end);
 
   // Set up loopfilter thread data.
   for (i = 0; i < num_workers; ++i) {
     AVxWorker *const worker = &workers[i];
     LFWorkerData *const lf_data = &lf_sync->lfdata[i];
 
+#if LOOP_FILTER_BITMASK
+    if (is_decoding) {
+      worker->hook = loop_filter_bitmask_row_worker;
+    } else {
+      worker->hook = loop_filter_row_worker;
+    }
+#else
     worker->hook = loop_filter_row_worker;
+#endif
     worker->data1 = lf_sync;
     worker->data2 = lf_data;
 
@@ -366,8 +470,12 @@ static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
 
 void av1_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
                               MACROBLOCKD *xd, int plane_start, int plane_end,
-                              int partial_frame, AVxWorker *workers,
-                              int num_workers, AV1LfSync *lf_sync) {
+                              int partial_frame,
+#if LOOP_FILTER_BITMASK
+                              int is_decoding,
+#endif
+                              AVxWorker *workers, int num_workers,
+                              AV1LfSync *lf_sync) {
   int start_mi_row, end_mi_row, mi_rows_to_filter;
 
   start_mi_row = 0;
@@ -380,8 +488,37 @@ void av1_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
   end_mi_row = start_mi_row + mi_rows_to_filter;
   av1_loop_filter_frame_init(cm, plane_start, plane_end);
 
+#if LOOP_FILTER_BITMASK
+  if (is_decoding) {
+    cm->is_decoding = is_decoding;
+    // TODO(chengchen): currently use one thread to build bitmasks for the
+    // frame. Make it support multi-thread later.
+    for (int plane = plane_start; plane < plane_end; plane++) {
+      if (plane == 0 && !(cm->lf.filter_level[0]) && !(cm->lf.filter_level[1]))
+        break;
+      else if (plane == 1 && !(cm->lf.filter_level_u))
+        continue;
+      else if (plane == 2 && !(cm->lf.filter_level_v))
+        continue;
+
+      // TODO(chengchen): can we remove this?
+      struct macroblockd_plane *pd = xd->plane;
+      av1_setup_dst_planes(pd, cm->seq_params.sb_size, frame, 0, 0, plane,
+                           plane + 1);
+
+      av1_build_bitmask_vert_info(cm, &pd[plane], plane);
+      av1_build_bitmask_horz_info(cm, &pd[plane], plane);
+    }
+    loop_filter_rows_mt(frame, cm, xd, start_mi_row, end_mi_row, plane_start,
+                        plane_end, 1, workers, num_workers, lf_sync);
+  } else {
+    loop_filter_rows_mt(frame, cm, xd, start_mi_row, end_mi_row, plane_start,
+                        plane_end, 0, workers, num_workers, lf_sync);
+  }
+#else
   loop_filter_rows_mt(frame, cm, xd, start_mi_row, end_mi_row, plane_start,
                       plane_end, workers, num_workers, lf_sync);
+#endif
 }
 
 static INLINE void lr_sync_read(void *const lr_sync, int r, int c, int plane) {
@@ -630,7 +767,7 @@ static void enqueue_lr_jobs(AV1LrSync *lr_sync, AV1LrStruct *lr_ctxt,
   }
 }
 
-AV1LrMTInfo *get_lr_job_info(AV1LrSync *lr_sync) {
+static AV1LrMTInfo *get_lr_job_info(AV1LrSync *lr_sync) {
   AV1LrMTInfo *cur_job_info = NULL;
 
 #if CONFIG_MULTITHREAD
@@ -664,9 +801,9 @@ static int loop_restoration_row_worker(void *arg1, void *arg2) {
   typedef void (*copy_fun)(const YV12_BUFFER_CONFIG *src_ybc,
                            YV12_BUFFER_CONFIG *dst_ybc, int hstart, int hend,
                            int vstart, int vend);
-  static const copy_fun copy_funs[3] = {
-    aom_yv12_partial_copy_y, aom_yv12_partial_copy_u, aom_yv12_partial_copy_v
-  };
+  static const copy_fun copy_funs[3] = { aom_yv12_partial_coloc_copy_y,
+                                         aom_yv12_partial_coloc_copy_u,
+                                         aom_yv12_partial_coloc_copy_v };
 
   while (1) {
     AV1LrMTInfo *cur_job_info = get_lr_job_info(lr_sync);
diff --git a/third_party/aom/av1/common/thread_common.h b/third_party/aom/av1/common/thread_common.h
index 23d61d72a3..e7dbb8bd33 100644
--- a/third_party/aom/av1/common/thread_common.h
+++ b/third_party/aom/av1/common/thread_common.h
@@ -103,6 +103,9 @@ void av1_loop_filter_dealloc(AV1LfSync *lf_sync);
 void av1_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, struct AV1Common *cm,
                               struct macroblockd *mbd, int plane_start,
                               int plane_end, int partial_frame,
+#if LOOP_FILTER_BITMASK
+                              int is_decoding,
+#endif
                               AVxWorker *workers, int num_workers,
                               AV1LfSync *lf_sync);
 void av1_loop_restoration_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
diff --git a/third_party/aom/av1/common/tile_common.c b/third_party/aom/av1/common/tile_common.c
index 1b413487f0..4d905924ac 100644
--- a/third_party/aom/av1/common/tile_common.c
+++ b/third_party/aom/av1/common/tile_common.c
@@ -143,30 +143,6 @@ int av1_get_sb_cols_in_tile(AV1_COMMON *cm, TileInfo tile) {
   return sb_cols;
 }
 
-int get_tile_size(int mi_frame_size, int log2_tile_num, int *ntiles) {
-  // Round the frame up to a whole number of max superblocks
-  mi_frame_size = ALIGN_POWER_OF_TWO(mi_frame_size, MAX_MIB_SIZE_LOG2);
-
-  // Divide by the signalled number of tiles, rounding up to the multiple of
-  // the max superblock size. To do this, shift right (and round up) to get the
-  // tile size in max super-blocks and then shift left again to convert it to
-  // mi units.
-  const int shift = log2_tile_num + MAX_MIB_SIZE_LOG2;
-  const int max_sb_tile_size =
-      ALIGN_POWER_OF_TWO(mi_frame_size, shift) >> shift;
-  const int mi_tile_size = max_sb_tile_size << MAX_MIB_SIZE_LOG2;
-
-  // The actual number of tiles is the ceiling of the frame size in mi units
-  // divided by mi_size. This is at most 1 << log2_tile_num but might be
-  // strictly less if max_sb_tile_size got rounded up significantly.
-  if (ntiles) {
-    *ntiles = (mi_frame_size + mi_tile_size - 1) / mi_tile_size;
-    assert(*ntiles <= (1 << log2_tile_num));
-  }
-
-  return mi_tile_size;
-}
-
 AV1PixelRect av1_get_tile_rect(const TileInfo *tile_info, const AV1_COMMON *cm,
                                int is_uv) {
   AV1PixelRect r;
@@ -205,3 +181,26 @@ AV1PixelRect av1_get_tile_rect(const TileInfo *tile_info, const AV1_COMMON *cm,
 
   return r;
 }
+
+void av1_get_uniform_tile_size(const AV1_COMMON *cm, int *w, int *h) {
+  if (cm->uniform_tile_spacing_flag) {
+    *w = cm->tile_width;
+    *h = cm->tile_height;
+  } else {
+    for (int i = 0; i < cm->tile_cols; ++i) {
+      const int tile_width_sb =
+          cm->tile_col_start_sb[i + 1] - cm->tile_col_start_sb[i];
+      const int tile_w = tile_width_sb * cm->seq_params.mib_size;
+      assert(i == 0 || tile_w == *w);  // ensure all tiles have same dimension
+      *w = tile_w;
+    }
+
+    for (int i = 0; i < cm->tile_rows; ++i) {
+      const int tile_height_sb =
+          cm->tile_row_start_sb[i + 1] - cm->tile_row_start_sb[i];
+      const int tile_h = tile_height_sb * cm->seq_params.mib_size;
+      assert(i == 0 || tile_h == *h);  // ensure all tiles have same dimension
+      *h = tile_h;
+    }
+  }
+}
diff --git a/third_party/aom/av1/common/tile_common.h b/third_party/aom/av1/common/tile_common.h
index c03553dc6f..b7203cef75 100644
--- a/third_party/aom/av1/common/tile_common.h
+++ b/third_party/aom/av1/common/tile_common.h
@@ -25,7 +25,6 @@ struct AV1Common;
 typedef struct TileInfo {
   int mi_row_start, mi_row_end;
   int mi_col_start, mi_col_end;
-  int tg_horz_boundary;
   int tile_row;
   int tile_col;
 } TileInfo;
@@ -37,12 +36,6 @@ void av1_tile_init(TileInfo *tile, const struct AV1Common *cm, int row,
 
 void av1_tile_set_row(TileInfo *tile, const struct AV1Common *cm, int row);
 void av1_tile_set_col(TileInfo *tile, const struct AV1Common *cm, int col);
-void av1_get_tile_n_bits(int mi_cols, int *min_log2_tile_cols,
-                         int *max_log2_tile_cols);
-
-// Calculate the correct tile size (width or height) for (1 << log2_tile_num)
-// tiles horizontally or vertically in the frame.
-int get_tile_size(int mi_frame_size, int log2_tile_num, int *ntiles);
 
 int av1_get_sb_rows_in_tile(struct AV1Common *cm, TileInfo tile);
 int av1_get_sb_cols_in_tile(struct AV1Common *cm, TileInfo tile);
@@ -61,6 +54,7 @@ AV1PixelRect av1_get_tile_rect(const TileInfo *tile_info,
 #define MAX_TILE_WIDTH (4096)        // Max Tile width in pixels
 #define MAX_TILE_AREA (4096 * 2304)  // Maximum tile area in pixels
 
+void av1_get_uniform_tile_size(const struct AV1Common *cm, int *w, int *h);
 void av1_get_tile_limits(struct AV1Common *const cm);
 void av1_calculate_tile_cols(struct AV1Common *const cm);
 void av1_calculate_tile_rows(struct AV1Common *const cm);
diff --git a/third_party/aom/av1/common/txb_common.c b/third_party/aom/av1/common/txb_common.c
index c96d37cca7..cb92bd8c6e 100644
--- a/third_party/aom/av1/common/txb_common.c
+++ b/third_party/aom/av1/common/txb_common.c
@@ -453,23 +453,6 @@ const int8_t *av1_nz_map_ctx_offset[19] = {
   av1_nz_map_ctx_offset_64x32,  // TX_64x16
 };
 
-void av1_init_lv_map(AV1_COMMON *cm) {
-  LV_MAP_CTX_TABLE *coeff_ctx_table = &cm->coeff_ctx_table;
-  for (int row = 0; row < 2; ++row) {
-    for (int col = 0; col < 2; ++col) {
-      for (int sig_mag = 0; sig_mag < 3; ++sig_mag) {
-        for (int count = 0; count < BASE_CONTEXT_POSITION_NUM + 1; ++count) {
-          if (row == 0 && col == 0 && count > 5) continue;
-          if ((row == 0 || col == 0) && count > 8) continue;
-
-          coeff_ctx_table->base_ctx_table[row][col][sig_mag][count] =
-              get_base_ctx_from_count_mag(row, col, count, sig_mag);
-        }
-      }
-    }
-  }
-}
-
 const int16_t k_eob_group_start[12] = { 0,  1,  2,  3,   5,   9,
                                         17, 33, 65, 129, 257, 513 };
 const int16_t k_eob_offset_bits[12] = { 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
diff --git a/third_party/aom/av1/common/txb_common.h b/third_party/aom/av1/common/txb_common.h
index 1dda51f8b4..8a3932d91d 100644
--- a/third_party/aom/av1/common/txb_common.h
+++ b/third_party/aom/av1/common/txb_common.h
@@ -12,6 +12,8 @@
 #ifndef AOM_AV1_COMMON_TXB_COMMON_H_
 #define AOM_AV1_COMMON_TXB_COMMON_H_
 
+#include "av1/common/onyxc_int.h"
+
 extern const int16_t k_eob_group_start[12];
 extern const int16_t k_eob_offset_bits[12];
 
@@ -157,6 +159,19 @@ static INLINE int get_br_ctx_2d(const uint8_t *const levels,
   return mag + 14;
 }
 
+static AOM_FORCE_INLINE int get_br_ctx_eob(const int c,  // raster order
+                                           const int bwl,
+                                           const TX_CLASS tx_class) {
+  const int row = c >> bwl;
+  const int col = c - (row << bwl);
+  if (c == 0) return 0;
+  if ((tx_class == TX_CLASS_2D && row < 2 && col < 2) ||
+      (tx_class == TX_CLASS_HORIZ && col == 0) ||
+      (tx_class == TX_CLASS_VERT && row == 0))
+    return 7;
+  return 14;
+}
+
 static AOM_FORCE_INLINE int get_br_ctx(const uint8_t *const levels,
                                        const int c,  // raster order
                                        const int bwl, const TX_CLASS tx_class) {
@@ -270,12 +285,10 @@ static AOM_FORCE_INLINE int get_nz_map_ctx_from_stats(
       const int row = coeff_idx >> bwl;
       const int col = coeff_idx - (row << bwl);
       return ctx + nz_map_ctx_offset_1d[col];
-      break;
     }
     case TX_CLASS_VERT: {
       const int row = coeff_idx >> bwl;
       return ctx + nz_map_ctx_offset_1d[row];
-      break;
     }
     default: break;
   }
@@ -419,6 +432,4 @@ static INLINE void get_txb_ctx(const BLOCK_SIZE plane_bsize,
 #undef MAX_TX_SIZE_UNIT
 }
 
-void av1_init_lv_map(AV1_COMMON *cm);
-
 #endif  // AOM_AV1_COMMON_TXB_COMMON_H_
diff --git a/third_party/aom/av1/common/warped_motion.c b/third_party/aom/av1/common/warped_motion.c
index 4144c43896..e232e10828 100644
--- a/third_party/aom/av1/common/warped_motion.c
+++ b/third_party/aom/av1/common/warped_motion.c
@@ -485,7 +485,7 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
               uint16_t *dst16 =
                   &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
               int32_t tmp32 = *p;
-              if (conv_params->use_jnt_comp_avg) {
+              if (conv_params->use_dist_wtd_comp_avg) {
                 tmp32 = tmp32 * conv_params->fwd_offset +
                         sum * conv_params->bck_offset;
                 tmp32 = tmp32 >> DIST_PRECISION_BITS;
@@ -563,7 +563,7 @@ static int64_t highbd_warp_error(
   uint16_t tmp[WARP_ERROR_BLOCK * WARP_ERROR_BLOCK];
 
   ConvolveParams conv_params = get_conv_params(0, 0, bd);
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
   for (int i = p_row; i < p_row + p_height; i += WARP_ERROR_BLOCK) {
     for (int j = p_col; j < p_col + p_width; j += WARP_ERROR_BLOCK) {
       // avoid warping extra 8x8 blocks in the padded region of the frame
@@ -773,7 +773,7 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
               uint8_t *dst8 =
                   &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
               int32_t tmp32 = *p;
-              if (conv_params->use_jnt_comp_avg) {
+              if (conv_params->use_dist_wtd_comp_avg) {
                 tmp32 = tmp32 * conv_params->fwd_offset +
                         sum * conv_params->bck_offset;
                 tmp32 = tmp32 >> DIST_PRECISION_BITS;
@@ -846,7 +846,7 @@ static int64_t warp_error(WarpedMotionParams *wm, const uint8_t *const ref,
   int error_bsize_h = AOMMIN(p_height, WARP_ERROR_BLOCK);
   uint8_t tmp[WARP_ERROR_BLOCK * WARP_ERROR_BLOCK];
   ConvolveParams conv_params = get_conv_params(0, 0, 8);
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
 
   for (int i = p_row; i < p_row + p_height; i += WARP_ERROR_BLOCK) {
     for (int j = p_col; j < p_col + p_width; j += WARP_ERROR_BLOCK) {
diff --git a/third_party/aom/av1/common/x86/av1_convolve_scale_sse4.c b/third_party/aom/av1/common/x86/av1_convolve_scale_sse4.c
index d9fb537856..8f442383ca 100644
--- a/third_party/aom/av1/common/x86/av1_convolve_scale_sse4.c
+++ b/third_party/aom/av1/common/x86/av1_convolve_scale_sse4.c
@@ -175,7 +175,7 @@ static void vfilter8(const int16_t *src, int src_stride, uint8_t *dst,
       if (conv_params->is_compound) {
         if (conv_params->do_average) {
           const __m128i p_16 = _mm_loadl_epi64((__m128i *)dst_16_x);
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             const __m128i p_16_lo = _mm_unpacklo_epi16(p_16, shifted_16);
             const __m128i wt_res_lo = _mm_madd_epi16(p_16_lo, wt);
             const __m128i shifted_32 =
@@ -207,7 +207,7 @@ static void vfilter8(const int16_t *src, int src_stride, uint8_t *dst,
       if (conv_params->is_compound) {
         if (conv_params->do_average) {
           int32_t tmp = dst16[y * dst16_stride + x];
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
             tmp = tmp >> DIST_PRECISION_BITS;
           } else {
@@ -408,7 +408,7 @@ static void highbd_vfilter8(const int16_t *src, int src_stride, uint16_t *dst,
           __m128i p_32 =
               _mm_cvtepu16_epi32(_mm_loadl_epi64((__m128i *)dst_16_x));
 
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             shifted = _mm_add_epi32(_mm_mullo_epi32(p_32, wt0),
                                     _mm_mullo_epi32(shifted, wt1));
             shifted = _mm_srai_epi32(shifted, DIST_PRECISION_BITS);
@@ -443,7 +443,7 @@ static void highbd_vfilter8(const int16_t *src, int src_stride, uint16_t *dst,
       if (conv_params->is_compound) {
         if (conv_params->do_average) {
           int32_t tmp = dst16[y * dst16_stride + x];
-          if (conv_params->use_jnt_comp_avg) {
+          if (conv_params->use_dist_wtd_comp_avg) {
             tmp = tmp * conv_params->fwd_offset + res * conv_params->bck_offset;
             tmp = tmp >> DIST_PRECISION_BITS;
           } else {
diff --git a/third_party/aom/av1/common/x86/av1_inv_txfm_avx2.c b/third_party/aom/av1/common/x86/av1_inv_txfm_avx2.c
index 5db2ccf6c3..cf1f947d30 100644
--- a/third_party/aom/av1/common/x86/av1_inv_txfm_avx2.c
+++ b/third_party/aom/av1/common/x86/av1_inv_txfm_avx2.c
@@ -1577,6 +1577,9 @@ static void idct64_low32_new_avx2(const __m256i *input, __m256i *output,
   idct64_stage11_avx2(output, x);
 }
 
+typedef void (*transform_1d_avx2)(const __m256i *input, __m256i *output,
+                                  int8_t cos_bit);
+
 // 1D functions process 16 pixels at one time.
 static const transform_1d_avx2
     lowbd_txfm_all_1d_zeros_w16_arr[TX_SIZES][ITX_TYPES_1D][4] = {
diff --git a/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.c b/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.c
index 995bc3da44..9841bf3e24 100644
--- a/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.c
+++ b/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.c
@@ -2437,9 +2437,10 @@ static INLINE void lowbd_inv_txfm2d_add_idtx_ssse3(const int32_t *input,
   }
 }
 
-void lowbd_inv_txfm2d_add_4x4_ssse3(const int32_t *input, uint8_t *output,
-                                    int stride, TX_TYPE tx_type,
-                                    TX_SIZE tx_size_, int eob) {
+static void lowbd_inv_txfm2d_add_4x4_ssse3(const int32_t *input,
+                                           uint8_t *output, int stride,
+                                           TX_TYPE tx_type, TX_SIZE tx_size_,
+                                           int eob) {
   (void)tx_size_;
   (void)eob;
   __m128i buf[4];
@@ -2708,9 +2709,10 @@ static INLINE void lowbd_inv_txfm2d_add_universe_ssse3(
   }
 }
 
-void lowbd_inv_txfm2d_add_4x8_ssse3(const int32_t *input, uint8_t *output,
-                                    int stride, TX_TYPE tx_type,
-                                    TX_SIZE tx_size_, int eob) {
+static void lowbd_inv_txfm2d_add_4x8_ssse3(const int32_t *input,
+                                           uint8_t *output, int stride,
+                                           TX_TYPE tx_type, TX_SIZE tx_size_,
+                                           int eob) {
   (void)tx_size_;
   (void)eob;
   __m128i buf[8];
@@ -2747,9 +2749,10 @@ void lowbd_inv_txfm2d_add_4x8_ssse3(const int32_t *input, uint8_t *output,
   lowbd_write_buffer_4xn_sse2(buf, output, stride, ud_flip, txfm_size_row);
 }
 
-void lowbd_inv_txfm2d_add_8x4_ssse3(const int32_t *input, uint8_t *output,
-                                    int stride, TX_TYPE tx_type,
-                                    TX_SIZE tx_size_, int eob) {
+static void lowbd_inv_txfm2d_add_8x4_ssse3(const int32_t *input,
+                                           uint8_t *output, int stride,
+                                           TX_TYPE tx_type, TX_SIZE tx_size_,
+                                           int eob) {
   (void)tx_size_;
   (void)eob;
   __m128i buf[8];
@@ -2786,9 +2789,10 @@ void lowbd_inv_txfm2d_add_8x4_ssse3(const int32_t *input, uint8_t *output,
   lowbd_write_buffer_8xn_sse2(buf, output, stride, ud_flip, txfm_size_row);
 }
 
-void lowbd_inv_txfm2d_add_4x16_ssse3(const int32_t *input, uint8_t *output,
-                                     int stride, TX_TYPE tx_type,
-                                     TX_SIZE tx_size_, int eob) {
+static void lowbd_inv_txfm2d_add_4x16_ssse3(const int32_t *input,
+                                            uint8_t *output, int stride,
+                                            TX_TYPE tx_type, TX_SIZE tx_size_,
+                                            int eob) {
   (void)tx_size_;
   (void)eob;
   __m128i buf[16];
@@ -2831,9 +2835,10 @@ void lowbd_inv_txfm2d_add_4x16_ssse3(const int32_t *input, uint8_t *output,
   lowbd_write_buffer_4xn_sse2(buf, output, stride, ud_flip, txfm_size_row);
 }
 
-void lowbd_inv_txfm2d_add_16x4_ssse3(const int32_t *input, uint8_t *output,
-                                     int stride, TX_TYPE tx_type,
-                                     TX_SIZE tx_size_, int eob) {
+static void lowbd_inv_txfm2d_add_16x4_ssse3(const int32_t *input,
+                                            uint8_t *output, int stride,
+                                            TX_TYPE tx_type, TX_SIZE tx_size_,
+                                            int eob) {
   (void)tx_size_;
   (void)eob;
   __m128i buf[16];
diff --git a/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.h b/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.h
index 66bd339d11..7d5055debe 100644
--- a/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.h
+++ b/third_party/aom/av1/common/x86/av1_inv_txfm_ssse3.h
@@ -72,13 +72,13 @@ static INLINE void round_shift_16bit_ssse3(__m128i *in, int size, int bit) {
 }
 
 // 1D itx types
-typedef enum ATTRIBUTE_PACKED {
+enum {
   IDCT_1D,
   IADST_1D,
   IFLIPADST_1D = IADST_1D,
   IIDENTITY_1D,
   ITX_TYPES_1D,
-} ITX_TYPE_1D;
+} UENUM1BYTE(ITX_TYPE_1D);
 
 static const ITX_TYPE_1D vitx_1d_tab[TX_TYPES] = {
   IDCT_1D,      IADST_1D,     IDCT_1D,      IADST_1D,
diff --git a/third_party/aom/av1/common/x86/av1_txfm_sse4.c b/third_party/aom/av1/common/x86/av1_txfm_sse4.c
index 90b9879cc4..65ccd1952e 100644
--- a/third_party/aom/av1/common/x86/av1_txfm_sse4.c
+++ b/third_party/aom/av1/common/x86/av1_txfm_sse4.c
@@ -9,7 +9,7 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
 
 #include "av1/common/av1_txfm.h"
 #include "av1/common/x86/av1_txfm_sse4.h"
diff --git a/third_party/aom/av1/common/x86/cfl_avx2.c b/third_party/aom/av1/common/x86/cfl_avx2.c
index a8bfdcce6b..d9bdf60320 100644
--- a/third_party/aom/av1/common/x86/cfl_avx2.c
+++ b/third_party/aom/av1/common/x86/cfl_avx2.c
@@ -27,21 +27,21 @@
       subsample_##bd##_##sub##_8x8_ssse3,   /* 8x8 */                      \
       subsample_##bd##_##sub##_16x16_ssse3, /* 16x16 */                    \
       subsample_##bd##_##sub##_32x32_avx2,  /* 32x32 */                    \
-      cfl_subsample_##bd##_null,            /* 64x64 (invalid CFL size) */ \
+      NULL,                                 /* 64x64 (invalid CFL size) */ \
       subsample_##bd##_##sub##_4x8_ssse3,   /* 4x8 */                      \
       subsample_##bd##_##sub##_8x4_ssse3,   /* 8x4 */                      \
       subsample_##bd##_##sub##_8x16_ssse3,  /* 8x16 */                     \
       subsample_##bd##_##sub##_16x8_ssse3,  /* 16x8 */                     \
       subsample_##bd##_##sub##_16x32_ssse3, /* 16x32 */                    \
       subsample_##bd##_##sub##_32x16_avx2,  /* 32x16 */                    \
-      cfl_subsample_##bd##_null,            /* 32x64 (invalid CFL size) */ \
-      cfl_subsample_##bd##_null,            /* 64x32 (invalid CFL size) */ \
+      NULL,                                 /* 32x64 (invalid CFL size) */ \
+      NULL,                                 /* 64x32 (invalid CFL size) */ \
       subsample_##bd##_##sub##_4x16_ssse3,  /* 4x16  */                    \
       subsample_##bd##_##sub##_16x4_ssse3,  /* 16x4  */                    \
       subsample_##bd##_##sub##_8x32_ssse3,  /* 8x32  */                    \
       subsample_##bd##_##sub##_32x8_avx2,   /* 32x8  */                    \
-      cfl_subsample_##bd##_null,            /* 16x64 (invalid CFL size) */ \
-      cfl_subsample_##bd##_null,            /* 64x16 (invalid CFL size) */ \
+      NULL,                                 /* 16x64 (invalid CFL size) */ \
+      NULL,                                 /* 64x16 (invalid CFL size) */ \
     };                                                                     \
     return subfn_##sub[tx_size];                                           \
   }
@@ -279,21 +279,21 @@ cfl_predict_lbd_fn get_predict_lbd_fn_avx2(TX_SIZE tx_size) {
     predict_lbd_8x8_ssse3,   /* 8x8 */
     predict_lbd_16x16_ssse3, /* 16x16 */
     predict_lbd_32x32_avx2,  /* 32x32 */
-    cfl_predict_lbd_null,    /* 64x64 (invalid CFL size) */
+    NULL,                    /* 64x64 (invalid CFL size) */
     predict_lbd_4x8_ssse3,   /* 4x8 */
     predict_lbd_8x4_ssse3,   /* 8x4 */
     predict_lbd_8x16_ssse3,  /* 8x16 */
     predict_lbd_16x8_ssse3,  /* 16x8 */
     predict_lbd_16x32_ssse3, /* 16x32 */
     predict_lbd_32x16_avx2,  /* 32x16 */
-    cfl_predict_lbd_null,    /* 32x64 (invalid CFL size) */
-    cfl_predict_lbd_null,    /* 64x32 (invalid CFL size) */
+    NULL,                    /* 32x64 (invalid CFL size) */
+    NULL,                    /* 64x32 (invalid CFL size) */
     predict_lbd_4x16_ssse3,  /* 4x16  */
     predict_lbd_16x4_ssse3,  /* 16x4  */
     predict_lbd_8x32_ssse3,  /* 8x32  */
     predict_lbd_32x8_avx2,   /* 32x8  */
-    cfl_predict_lbd_null,    /* 16x64 (invalid CFL size) */
-    cfl_predict_lbd_null,    /* 64x16 (invalid CFL size) */
+    NULL,                    /* 16x64 (invalid CFL size) */
+    NULL,                    /* 64x16 (invalid CFL size) */
   };
   // Modulo TX_SIZES_ALL to ensure that an attacker won't be able to index the
   // function pointer array out of bounds.
@@ -352,21 +352,21 @@ cfl_predict_hbd_fn get_predict_hbd_fn_avx2(TX_SIZE tx_size) {
     predict_hbd_8x8_ssse3,  /* 8x8 */
     predict_hbd_16x16_avx2, /* 16x16 */
     predict_hbd_32x32_avx2, /* 32x32 */
-    cfl_predict_hbd_null,   /* 64x64 (invalid CFL size) */
+    NULL,                   /* 64x64 (invalid CFL size) */
     predict_hbd_4x8_ssse3,  /* 4x8 */
     predict_hbd_8x4_ssse3,  /* 8x4 */
     predict_hbd_8x16_ssse3, /* 8x16 */
     predict_hbd_16x8_avx2,  /* 16x8 */
     predict_hbd_16x32_avx2, /* 16x32 */
     predict_hbd_32x16_avx2, /* 32x16 */
-    cfl_predict_hbd_null,   /* 32x64 (invalid CFL size) */
-    cfl_predict_hbd_null,   /* 64x32 (invalid CFL size) */
+    NULL,                   /* 32x64 (invalid CFL size) */
+    NULL,                   /* 64x32 (invalid CFL size) */
     predict_hbd_4x16_ssse3, /* 4x16  */
     predict_hbd_16x4_avx2,  /* 16x4  */
     predict_hbd_8x32_ssse3, /* 8x32  */
     predict_hbd_32x8_avx2,  /* 32x8  */
-    cfl_predict_hbd_null,   /* 16x64 (invalid CFL size) */
-    cfl_predict_hbd_null,   /* 64x16 (invalid CFL size) */
+    NULL,                   /* 16x64 (invalid CFL size) */
+    NULL,                   /* 64x16 (invalid CFL size) */
   };
   // Modulo TX_SIZES_ALL to ensure that an attacker won't be able to index the
   // function pointer array out of bounds.
@@ -469,21 +469,21 @@ cfl_subtract_average_fn get_subtract_average_fn_avx2(TX_SIZE tx_size) {
     subtract_average_8x8_sse2,   /* 8x8 */
     subtract_average_16x16_avx2, /* 16x16 */
     subtract_average_32x32_avx2, /* 32x32 */
-    cfl_subtract_average_null,   /* 64x64 (invalid CFL size) */
+    NULL,                        /* 64x64 (invalid CFL size) */
     subtract_average_4x8_sse2,   /* 4x8 */
     subtract_average_8x4_sse2,   /* 8x4 */
     subtract_average_8x16_sse2,  /* 8x16 */
     subtract_average_16x8_avx2,  /* 16x8 */
     subtract_average_16x32_avx2, /* 16x32 */
     subtract_average_32x16_avx2, /* 32x16 */
-    cfl_subtract_average_null,   /* 32x64 (invalid CFL size) */
-    cfl_subtract_average_null,   /* 64x32 (invalid CFL size) */
+    NULL,                        /* 32x64 (invalid CFL size) */
+    NULL,                        /* 64x32 (invalid CFL size) */
     subtract_average_4x16_sse2,  /* 4x16 */
     subtract_average_16x4_avx2,  /* 16x4 */
     subtract_average_8x32_sse2,  /* 8x32 */
     subtract_average_32x8_avx2,  /* 32x8 */
-    cfl_subtract_average_null,   /* 16x64 (invalid CFL size) */
-    cfl_subtract_average_null,   /* 64x16 (invalid CFL size) */
+    NULL,                        /* 16x64 (invalid CFL size) */
+    NULL,                        /* 64x16 (invalid CFL size) */
   };
   // Modulo TX_SIZES_ALL to ensure that an attacker won't be able to
   // index the function pointer array out of bounds.
diff --git a/third_party/aom/av1/common/x86/convolve_2d_avx2.c b/third_party/aom/av1/common/x86/convolve_2d_avx2.c
index 0acafd0446..5a69ef66a1 100644
--- a/third_party/aom/av1/common/x86/convolve_2d_avx2.c
+++ b/third_party/aom/av1/common/x86/convolve_2d_avx2.c
@@ -195,20 +195,20 @@ void av1_convolve_2d_copy_sr_avx2(const uint8_t *src, int src_stride,
 
   if (w == 2) {
     do {
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
       h -= 2;
     } while (h);
   } else if (w == 4) {
     do {
-      memcpy(dst, src, 4 * sizeof(*src));
+      memmove(dst, src, 4 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
-      memcpy(dst, src, 4 * sizeof(*src));
+      memmove(dst, src, 4 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
       h -= 2;
diff --git a/third_party/aom/av1/common/x86/convolve_2d_sse2.c b/third_party/aom/av1/common/x86/convolve_2d_sse2.c
index b1a62a4f69..369922bbdf 100644
--- a/third_party/aom/av1/common/x86/convolve_2d_sse2.c
+++ b/third_party/aom/av1/common/x86/convolve_2d_sse2.c
@@ -255,20 +255,20 @@ void av1_convolve_2d_copy_sr_sse2(const uint8_t *src, int src_stride,
 
   if (w == 2) {
     do {
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
       h -= 2;
     } while (h);
   } else if (w == 4) {
     do {
-      memcpy(dst, src, 4 * sizeof(*src));
+      memmove(dst, src, 4 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
-      memcpy(dst, src, 4 * sizeof(*src));
+      memmove(dst, src, 4 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
       h -= 2;
@@ -354,12 +354,11 @@ void av1_convolve_2d_copy_sr_sse2(const uint8_t *src, int src_stride,
   }
 }
 
-void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride,
-                                   uint8_t *dst0, int dst_stride0, int w, int h,
-                                   const InterpFilterParams *filter_params_x,
-                                   const InterpFilterParams *filter_params_y,
-                                   const int subpel_x_q4, const int subpel_y_q4,
-                                   ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_copy_sse2(
+    const uint8_t *src, int src_stride, uint8_t *dst0, int dst_stride0, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params) {
   const int bd = 8;
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
@@ -371,7 +370,7 @@ void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride,
   const int bits =
       FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const __m128i zero = _mm_setzero_si128();
   const __m128i left_shift = _mm_cvtsi32_si128(bits);
   int i, j;
@@ -411,14 +410,14 @@ void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride,
           const __m128i data_ref_0_hi =
               _mm_loadu_si128((__m128i *)(&dst[j + 8]));
 
-          const __m128i comp_avg_res_lo =
-              comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt, use_jnt_comp_avg);
+          const __m128i comp_avg_res_lo = comp_avg(
+              &data_ref_0_lo, &res_unsigned_lo, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result_lo = convolve_rounding(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
 
-          const __m128i comp_avg_res_hi =
-              comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt, use_jnt_comp_avg);
+          const __m128i comp_avg_res_hi = comp_avg(
+              &data_ref_0_hi, &res_unsigned_hi, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result_hi = convolve_rounding(
               &comp_avg_res_hi, &offset_const, &rounding_const, rounding_shift);
@@ -449,7 +448,7 @@ void av1_jnt_convolve_2d_copy_sse2(const uint8_t *src, int src_stride,
           const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)(&dst[j]));
 
           const __m128i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
diff --git a/third_party/aom/av1/common/x86/highbd_convolve_2d_avx2.c b/third_party/aom/av1/common/x86/highbd_convolve_2d_avx2.c
index ae68f0bbb3..357df12656 100644
--- a/third_party/aom/av1/common/x86/highbd_convolve_2d_avx2.c
+++ b/third_party/aom/av1/common/x86/highbd_convolve_2d_avx2.c
@@ -238,10 +238,10 @@ void av1_highbd_convolve_2d_copy_sr_avx2(
 
   if (w == 2) {
     do {
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
-      memcpy(dst, src, 2 * sizeof(*src));
+      memmove(dst, src, 2 * sizeof(*src));
       src += src_stride;
       dst += dst_stride;
       h -= 2;
diff --git a/third_party/aom/av1/common/x86/highbd_convolve_2d_sse4.c b/third_party/aom/av1/common/x86/highbd_convolve_2d_sse4.c
index 3f8dafb4b3..3c1d5d1418 100644
--- a/third_party/aom/av1/common/x86/highbd_convolve_2d_sse4.c
+++ b/third_party/aom/av1/common/x86/highbd_convolve_2d_sse4.c
@@ -21,7 +21,7 @@
 #include "aom_dsp/x86/convolve_sse4_1.h"
 #include "av1/common/convolve.h"
 
-void av1_highbd_jnt_convolve_2d_copy_sse4_1(
+void av1_highbd_dist_wtd_convolve_2d_copy_sse4_1(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -37,7 +37,7 @@ void av1_highbd_jnt_convolve_2d_copy_sse4_1(
       FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
   const __m128i left_shift = _mm_cvtsi32_si128(bits);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
   const __m128i wt0 = _mm_set1_epi32(w0);
@@ -75,15 +75,17 @@ void av1_highbd_jnt_convolve_2d_copy_sse4_1(
           const __m128i res_unsigned_lo =
               _mm_add_epi32(res_32b_lo, offset_const);
 
-          const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
-              &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+          const __m128i comp_avg_res_lo =
+              highbd_comp_avg_sse4_1(&data_ref_0_lo, &res_unsigned_lo, &wt0,
+                                     &wt1, use_dist_wtd_comp_avg);
 
           const __m128i res_32b_hi = _mm_unpackhi_epi16(res, zero);
           const __m128i res_unsigned_hi =
               _mm_add_epi32(res_32b_hi, offset_const);
 
-          const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
-              &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+          const __m128i comp_avg_res_hi =
+              highbd_comp_avg_sse4_1(&data_ref_0_hi, &res_unsigned_hi, &wt0,
+                                     &wt1, use_dist_wtd_comp_avg);
 
           const __m128i round_result_lo = highbd_convolve_rounding_sse2(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
@@ -132,9 +134,9 @@ void av1_highbd_jnt_convolve_2d_copy_sse4_1(
               _mm_add_epi32(res_32b_hi, offset_const);
 
           const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
-              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_dist_wtd_comp_avg);
           const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
-              &data_ref_1, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+              &data_ref_1, &res_unsigned_hi, &wt0, &wt1, use_dist_wtd_comp_avg);
 
           const __m128i round_result_lo = highbd_convolve_rounding_sse2(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
@@ -166,7 +168,7 @@ void av1_highbd_jnt_convolve_2d_copy_sse4_1(
   }
 }
 
-void av1_highbd_jnt_convolve_2d_sse4_1(
+void av1_highbd_dist_wtd_convolve_2d_sse4_1(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -179,7 +181,7 @@ void av1_highbd_jnt_convolve_2d_sse4_1(
   int im_stride = MAX_SB_SIZE;
   int i, j;
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int fo_vert = filter_params_y->taps / 2 - 1;
   const int fo_horiz = filter_params_x->taps / 2 - 1;
   const uint16_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
@@ -359,8 +361,9 @@ void av1_highbd_jnt_convolve_2d_sse4_1(
 
             const __m128i data_ref_0 = _mm_cvtepu16_epi32(data_0);
 
-            const __m128i comp_avg_res = highbd_comp_avg_sse4_1(
-                &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+            const __m128i comp_avg_res =
+                highbd_comp_avg_sse4_1(&data_ref_0, &res_unsigned_lo, &wt0,
+                                       &wt1, use_dist_wtd_comp_avg);
 
             const __m128i round_result = highbd_convolve_rounding_sse2(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -391,10 +394,12 @@ void av1_highbd_jnt_convolve_2d_sse4_1(
             const __m128i data_ref_0_lo = _mm_cvtepu16_epi32(data_lo);
             const __m128i data_ref_0_hi = _mm_cvtepu16_epi32(data_hi);
 
-            const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
-                &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
-            const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
-                &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+            const __m128i comp_avg_res_lo =
+                highbd_comp_avg_sse4_1(&data_ref_0_lo, &res_unsigned_lo, &wt0,
+                                       &wt1, use_dist_wtd_comp_avg);
+            const __m128i comp_avg_res_hi =
+                highbd_comp_avg_sse4_1(&data_ref_0_hi, &res_unsigned_hi, &wt0,
+                                       &wt1, use_dist_wtd_comp_avg);
 
             const __m128i round_result_lo =
                 highbd_convolve_rounding_sse2(&comp_avg_res_lo, &offset_const,
diff --git a/third_party/aom/av1/common/x86/highbd_inv_txfm_avx2.c b/third_party/aom/av1/common/x86/highbd_inv_txfm_avx2.c
index ade2af03e4..fe22465592 100644
--- a/third_party/aom/av1/common/x86/highbd_inv_txfm_avx2.c
+++ b/third_party/aom/av1/common/x86/highbd_inv_txfm_avx2.c
@@ -18,6 +18,7 @@
 #include "av1/common/idct.h"
 #include "av1/common/x86/av1_inv_txfm_ssse3.h"
 #include "av1/common/x86/highbd_txfm_utility_sse4.h"
+#include "aom_dsp/x86/txfm_common_avx2.h"
 
 // Note:
 //  Total 32x4 registers to represent 32x32 block coefficients.
@@ -72,30 +73,48 @@ static INLINE void highbd_write_buffer_16xn_avx2(__m256i *in, uint16_t *output,
     _mm256_storeu_si256((__m256i *)(output + i * stride), u);
   }
 }
-
-static INLINE __m256i av1_round_shift_32_avx2(__m256i vec, int bit) {
-  __m256i tmp, round;
-  round = _mm256_set1_epi32(1 << (bit - 1));
-  tmp = _mm256_add_epi32(vec, round);
-  return _mm256_srai_epi32(tmp, bit);
+static INLINE __m256i highbd_get_recon_8x8_avx2(const __m256i pred, __m256i res,
+                                                const int bd) {
+  __m256i x0 = pred;
+  x0 = _mm256_add_epi32(res, x0);
+  x0 = _mm256_packus_epi32(x0, x0);
+  x0 = _mm256_permute4x64_epi64(x0, 0xd8);
+  x0 = highbd_clamp_epi16_avx2(x0, bd);
+  return x0;
 }
 
-static INLINE void av1_round_shift_array_32_avx2(__m256i *input,
-                                                 __m256i *output,
-                                                 const int size,
-                                                 const int bit) {
-  if (bit > 0) {
-    int i;
-    for (i = 0; i < size; i++) {
-      output[i] = av1_round_shift_32_avx2(input[i], bit);
-    }
-  } else {
-    int i;
-    for (i = 0; i < size; i++) {
-      output[i] = _mm256_slli_epi32(input[i], -bit);
-    }
+static INLINE void highbd_write_buffer_8xn_avx2(__m256i *in, uint16_t *output,
+                                                int stride, int flipud,
+                                                int height, const int bd) {
+  int j = flipud ? (height - 1) : 0;
+  __m128i temp;
+  const int step = flipud ? -1 : 1;
+  for (int i = 0; i < height; ++i, j += step) {
+    temp = _mm_loadu_si128((__m128i const *)(output + i * stride));
+    __m256i v = _mm256_cvtepi16_epi32(temp);
+    __m256i u = highbd_get_recon_8x8_avx2(v, in[j], bd);
+    __m128i u1 = _mm256_castsi256_si128(u);
+    _mm_storeu_si128((__m128i *)(output + i * stride), u1);
   }
 }
+static void neg_shift_avx2(const __m256i in0, const __m256i in1, __m256i *out0,
+                           __m256i *out1, const __m256i *clamp_lo,
+                           const __m256i *clamp_hi, int shift) {
+  __m256i offset = _mm256_set1_epi32((1 << shift) >> 1);
+  __m256i a0 = _mm256_add_epi32(offset, in0);
+  __m256i a1 = _mm256_sub_epi32(offset, in1);
+
+  a0 = _mm256_sra_epi32(a0, _mm_cvtsi32_si128(shift));
+  a1 = _mm256_sra_epi32(a1, _mm_cvtsi32_si128(shift));
+
+  a0 = _mm256_max_epi32(a0, *clamp_lo);
+  a0 = _mm256_min_epi32(a0, *clamp_hi);
+  a1 = _mm256_max_epi32(a1, *clamp_lo);
+  a1 = _mm256_min_epi32(a1, *clamp_hi);
+
+  *out0 = a0;
+  *out1 = a1;
+}
 
 static void transpose_8x8_avx2(const __m256i *in, __m256i *out) {
   __m256i u0, u1, u2, u3, u4, u5, u6, u7;
@@ -134,6 +153,43 @@ static void transpose_8x8_avx2(const __m256i *in, __m256i *out) {
   out[7] = _mm256_permute2f128_si256(x0, x1, 0x31);
 }
 
+static void transpose_8x8_flip_avx2(const __m256i *in, __m256i *out) {
+  __m256i u0, u1, u2, u3, u4, u5, u6, u7;
+  __m256i x0, x1;
+
+  u0 = _mm256_unpacklo_epi32(in[7], in[6]);
+  u1 = _mm256_unpackhi_epi32(in[7], in[6]);
+
+  u2 = _mm256_unpacklo_epi32(in[5], in[4]);
+  u3 = _mm256_unpackhi_epi32(in[5], in[4]);
+
+  u4 = _mm256_unpacklo_epi32(in[3], in[2]);
+  u5 = _mm256_unpackhi_epi32(in[3], in[2]);
+
+  u6 = _mm256_unpacklo_epi32(in[1], in[0]);
+  u7 = _mm256_unpackhi_epi32(in[1], in[0]);
+
+  x0 = _mm256_unpacklo_epi64(u0, u2);
+  x1 = _mm256_unpacklo_epi64(u4, u6);
+  out[0] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[4] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpackhi_epi64(u0, u2);
+  x1 = _mm256_unpackhi_epi64(u4, u6);
+  out[1] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[5] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpacklo_epi64(u1, u3);
+  x1 = _mm256_unpacklo_epi64(u5, u7);
+  out[2] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[6] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpackhi_epi64(u1, u3);
+  x1 = _mm256_unpackhi_epi64(u5, u7);
+  out[3] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[7] = _mm256_permute2f128_si256(x0, x1, 0x31);
+}
+
 static void load_buffer_32x32(const int32_t *coeff, __m256i *in,
                               int input_stiride, int size) {
   int i;
@@ -1153,7 +1209,2979 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit, int do_cols, int bd,
     }
   }
 }
+static void idct16_low1_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                             int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
 
+  {
+    // stage 0
+    // stage 1
+    // stage 2
+    // stage 3
+    // stage 4
+    in[0] = _mm256_mullo_epi32(in[0], cospi32);
+    in[0] = _mm256_add_epi32(in[0], rnding);
+    in[0] = _mm256_srai_epi32(in[0], bit);
+
+    // stage 5
+    // stage 6
+    // stage 7
+    if (do_cols) {
+      in[0] = _mm256_max_epi32(in[0], clamp_lo);
+      in[0] = _mm256_min_epi32(in[0], clamp_hi);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+          -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+      const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+          (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+      __m256i offset = _mm256_set1_epi32((1 << out_shift) >> 1);
+      in[0] = _mm256_add_epi32(in[0], offset);
+      in[0] = _mm256_sra_epi32(in[0], _mm_cvtsi32_si128(out_shift));
+      in[0] = _mm256_max_epi32(in[0], clamp_lo_out);
+      in[0] = _mm256_min_epi32(in[0], clamp_hi_out);
+    }
+
+    out[0] = in[0];
+    out[1] = in[0];
+    out[2] = in[0];
+    out[3] = in[0];
+    out[4] = in[0];
+    out[5] = in[0];
+    out[6] = in[0];
+    out[7] = in[0];
+    out[8] = in[0];
+    out[9] = in[0];
+    out[10] = in[0];
+    out[11] = in[0];
+    out[12] = in[0];
+    out[13] = in[0];
+    out[14] = in[0];
+    out[15] = in[0];
+  }
+}
+
+static void idct16_low8_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                             int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u[16], x, y;
+
+  {
+    // stage 0
+    // stage 1
+    u[0] = in[0];
+    u[2] = in[4];
+    u[4] = in[2];
+    u[6] = in[6];
+    u[8] = in[1];
+    u[10] = in[5];
+    u[12] = in[3];
+    u[14] = in[7];
+
+    // stage 2
+    u[15] = half_btf_0_avx2(&cospi4, &u[8], &rnding, bit);
+    u[8] = half_btf_0_avx2(&cospi60, &u[8], &rnding, bit);
+
+    u[9] = half_btf_0_avx2(&cospim36, &u[14], &rnding, bit);
+    u[14] = half_btf_0_avx2(&cospi28, &u[14], &rnding, bit);
+
+    u[13] = half_btf_0_avx2(&cospi20, &u[10], &rnding, bit);
+    u[10] = half_btf_0_avx2(&cospi44, &u[10], &rnding, bit);
+
+    u[11] = half_btf_0_avx2(&cospim52, &u[12], &rnding, bit);
+    u[12] = half_btf_0_avx2(&cospi12, &u[12], &rnding, bit);
+
+    // stage 3
+    u[7] = half_btf_0_avx2(&cospi8, &u[4], &rnding, bit);
+    u[4] = half_btf_0_avx2(&cospi56, &u[4], &rnding, bit);
+    u[5] = half_btf_0_avx2(&cospim40, &u[6], &rnding, bit);
+    u[6] = half_btf_0_avx2(&cospi24, &u[6], &rnding, bit);
+
+    addsub_avx2(u[8], u[9], &u[8], &u[9], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[11], u[10], &u[11], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[12], u[13], &u[12], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[15], u[14], &u[15], &u[14], &clamp_lo, &clamp_hi);
+
+    // stage 4
+    x = _mm256_mullo_epi32(u[0], cospi32);
+    u[0] = _mm256_add_epi32(x, rnding);
+    u[0] = _mm256_srai_epi32(u[0], bit);
+    u[1] = u[0];
+
+    u[3] = half_btf_0_avx2(&cospi16, &u[2], &rnding, bit);
+    u[2] = half_btf_0_avx2(&cospi48, &u[2], &rnding, bit);
+
+    addsub_avx2(u[4], u[5], &u[4], &u[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[7], u[6], &u[7], &u[6], &clamp_lo, &clamp_hi);
+
+    x = half_btf_avx2(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
+    u[14] = half_btf_avx2(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
+    u[9] = x;
+    y = half_btf_avx2(&cospim48, &u[10], &cospim16, &u[13], &rnding, bit);
+    u[13] = half_btf_avx2(&cospim16, &u[10], &cospi48, &u[13], &rnding, bit);
+    u[10] = y;
+
+    // stage 5
+    addsub_avx2(u[0], u[3], &u[0], &u[3], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[2], &u[1], &u[2], &clamp_lo, &clamp_hi);
+
+    x = _mm256_mullo_epi32(u[5], cospi32);
+    y = _mm256_mullo_epi32(u[6], cospi32);
+    u[5] = _mm256_sub_epi32(y, x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    u[6] = _mm256_add_epi32(y, x);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    addsub_avx2(u[8], u[11], &u[8], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[9], u[10], &u[9], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[15], u[12], &u[15], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[14], u[13], &u[14], &u[13], &clamp_lo, &clamp_hi);
+
+    // stage 6
+    addsub_avx2(u[0], u[7], &u[0], &u[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[6], &u[1], &u[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[2], u[5], &u[2], &u[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[3], u[4], &u[3], &u[4], &clamp_lo, &clamp_hi);
+
+    x = _mm256_mullo_epi32(u[10], cospi32);
+    y = _mm256_mullo_epi32(u[13], cospi32);
+    u[10] = _mm256_sub_epi32(y, x);
+    u[10] = _mm256_add_epi32(u[10], rnding);
+    u[10] = _mm256_srai_epi32(u[10], bit);
+
+    u[13] = _mm256_add_epi32(x, y);
+    u[13] = _mm256_add_epi32(u[13], rnding);
+    u[13] = _mm256_srai_epi32(u[13], bit);
+
+    x = _mm256_mullo_epi32(u[11], cospi32);
+    y = _mm256_mullo_epi32(u[12], cospi32);
+    u[11] = _mm256_sub_epi32(y, x);
+    u[11] = _mm256_add_epi32(u[11], rnding);
+    u[11] = _mm256_srai_epi32(u[11], bit);
+
+    u[12] = _mm256_add_epi32(x, y);
+    u[12] = _mm256_add_epi32(u[12], rnding);
+    u[12] = _mm256_srai_epi32(u[12], bit);
+    // stage 7
+    if (do_cols) {
+      addsub_no_clamp_avx2(u[0], u[15], out + 0, out + 15);
+      addsub_no_clamp_avx2(u[1], u[14], out + 1, out + 14);
+      addsub_no_clamp_avx2(u[2], u[13], out + 2, out + 13);
+      addsub_no_clamp_avx2(u[3], u[12], out + 3, out + 12);
+      addsub_no_clamp_avx2(u[4], u[11], out + 4, out + 11);
+      addsub_no_clamp_avx2(u[5], u[10], out + 5, out + 10);
+      addsub_no_clamp_avx2(u[6], u[9], out + 6, out + 9);
+      addsub_no_clamp_avx2(u[7], u[8], out + 7, out + 8);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+          -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+      const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+          (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+      addsub_shift_avx2(u[0], u[15], out + 0, out + 15, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[1], u[14], out + 1, out + 14, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[2], u[13], out + 2, out + 13, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[3], u[12], out + 3, out + 12, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[4], u[11], out + 4, out + 11, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[5], u[10], out + 5, out + 10, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[6], u[9], out + 6, out + 9, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(u[7], u[8], out + 7, out + 8, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+    }
+  }
+}
+
+static void idct16_avx2(__m256i *in, __m256i *out, int bit, int do_cols, int bd,
+                        int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospim4 = _mm256_set1_epi32(-cospi[4]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospim20 = _mm256_set1_epi32(-cospi[20]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u[16], v[16], x, y;
+
+  {
+    // stage 0
+    // stage 1
+    u[0] = in[0];
+    u[1] = in[8];
+    u[2] = in[4];
+    u[3] = in[12];
+    u[4] = in[2];
+    u[5] = in[10];
+    u[6] = in[6];
+    u[7] = in[14];
+    u[8] = in[1];
+    u[9] = in[9];
+    u[10] = in[5];
+    u[11] = in[13];
+    u[12] = in[3];
+    u[13] = in[11];
+    u[14] = in[7];
+    u[15] = in[15];
+
+    // stage 2
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+    v[4] = u[4];
+    v[5] = u[5];
+    v[6] = u[6];
+    v[7] = u[7];
+
+    v[8] = half_btf_avx2(&cospi60, &u[8], &cospim4, &u[15], &rnding, bit);
+    v[9] = half_btf_avx2(&cospi28, &u[9], &cospim36, &u[14], &rnding, bit);
+    v[10] = half_btf_avx2(&cospi44, &u[10], &cospim20, &u[13], &rnding, bit);
+    v[11] = half_btf_avx2(&cospi12, &u[11], &cospim52, &u[12], &rnding, bit);
+    v[12] = half_btf_avx2(&cospi52, &u[11], &cospi12, &u[12], &rnding, bit);
+    v[13] = half_btf_avx2(&cospi20, &u[10], &cospi44, &u[13], &rnding, bit);
+    v[14] = half_btf_avx2(&cospi36, &u[9], &cospi28, &u[14], &rnding, bit);
+    v[15] = half_btf_avx2(&cospi4, &u[8], &cospi60, &u[15], &rnding, bit);
+
+    // stage 3
+    u[0] = v[0];
+    u[1] = v[1];
+    u[2] = v[2];
+    u[3] = v[3];
+    u[4] = half_btf_avx2(&cospi56, &v[4], &cospim8, &v[7], &rnding, bit);
+    u[5] = half_btf_avx2(&cospi24, &v[5], &cospim40, &v[6], &rnding, bit);
+    u[6] = half_btf_avx2(&cospi40, &v[5], &cospi24, &v[6], &rnding, bit);
+    u[7] = half_btf_avx2(&cospi8, &v[4], &cospi56, &v[7], &rnding, bit);
+    addsub_avx2(v[8], v[9], &u[8], &u[9], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[11], v[10], &u[11], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[12], v[13], &u[12], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[15], v[14], &u[15], &u[14], &clamp_lo, &clamp_hi);
+
+    // stage 4
+    x = _mm256_mullo_epi32(u[0], cospi32);
+    y = _mm256_mullo_epi32(u[1], cospi32);
+    v[0] = _mm256_add_epi32(x, y);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    v[0] = _mm256_srai_epi32(v[0], bit);
+
+    v[1] = _mm256_sub_epi32(x, y);
+    v[1] = _mm256_add_epi32(v[1], rnding);
+    v[1] = _mm256_srai_epi32(v[1], bit);
+
+    v[2] = half_btf_avx2(&cospi48, &u[2], &cospim16, &u[3], &rnding, bit);
+    v[3] = half_btf_avx2(&cospi16, &u[2], &cospi48, &u[3], &rnding, bit);
+    addsub_avx2(u[4], u[5], &v[4], &v[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[7], u[6], &v[7], &v[6], &clamp_lo, &clamp_hi);
+    v[8] = u[8];
+    v[9] = half_btf_avx2(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
+    v[10] = half_btf_avx2(&cospim48, &u[10], &cospim16, &u[13], &rnding, bit);
+    v[11] = u[11];
+    v[12] = u[12];
+    v[13] = half_btf_avx2(&cospim16, &u[10], &cospi48, &u[13], &rnding, bit);
+    v[14] = half_btf_avx2(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
+    v[15] = u[15];
+
+    // stage 5
+    addsub_avx2(v[0], v[3], &u[0], &u[3], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[1], v[2], &u[1], &u[2], &clamp_lo, &clamp_hi);
+    u[4] = v[4];
+
+    x = _mm256_mullo_epi32(v[5], cospi32);
+    y = _mm256_mullo_epi32(v[6], cospi32);
+    u[5] = _mm256_sub_epi32(y, x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    u[6] = _mm256_add_epi32(y, x);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    u[7] = v[7];
+    addsub_avx2(v[8], v[11], &u[8], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[9], v[10], &u[9], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[15], v[12], &u[15], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[14], v[13], &u[14], &u[13], &clamp_lo, &clamp_hi);
+
+    // stage 6
+    addsub_avx2(u[0], u[7], &v[0], &v[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[6], &v[1], &v[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[2], u[5], &v[2], &v[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[3], u[4], &v[3], &v[4], &clamp_lo, &clamp_hi);
+    v[8] = u[8];
+    v[9] = u[9];
+
+    x = _mm256_mullo_epi32(u[10], cospi32);
+    y = _mm256_mullo_epi32(u[13], cospi32);
+    v[10] = _mm256_sub_epi32(y, x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[13] = _mm256_add_epi32(x, y);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    x = _mm256_mullo_epi32(u[11], cospi32);
+    y = _mm256_mullo_epi32(u[12], cospi32);
+    v[11] = _mm256_sub_epi32(y, x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = _mm256_add_epi32(x, y);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+
+    v[14] = u[14];
+    v[15] = u[15];
+
+    // stage 7
+    if (do_cols) {
+      addsub_no_clamp_avx2(v[0], v[15], out + 0, out + 15);
+      addsub_no_clamp_avx2(v[1], v[14], out + 1, out + 14);
+      addsub_no_clamp_avx2(v[2], v[13], out + 2, out + 13);
+      addsub_no_clamp_avx2(v[3], v[12], out + 3, out + 12);
+      addsub_no_clamp_avx2(v[4], v[11], out + 4, out + 11);
+      addsub_no_clamp_avx2(v[5], v[10], out + 5, out + 10);
+      addsub_no_clamp_avx2(v[6], v[9], out + 6, out + 9);
+      addsub_no_clamp_avx2(v[7], v[8], out + 7, out + 8);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+          -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+      const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+          (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+      addsub_shift_avx2(v[0], v[15], out + 0, out + 15, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[1], v[14], out + 1, out + 14, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[2], v[13], out + 2, out + 13, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[3], v[12], out + 3, out + 12, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[4], v[11], out + 4, out + 11, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[5], v[10], out + 5, out + 10, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[6], v[9], out + 6, out + 9, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+      addsub_shift_avx2(v[7], v[8], out + 7, out + 8, &clamp_lo_out,
+                        &clamp_hi_out, out_shift);
+    }
+  }
+}
+
+static void iadst16_low1_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const __m256i zero = _mm256_setzero_si256();
+  __m256i v[16], x, y, temp1, temp2;
+
+  // Calculate the column 0, 1, 2, 3
+  {
+    // stage 0
+    // stage 1
+    // stage 2
+    x = _mm256_mullo_epi32(in[0], cospi62);
+    v[0] = _mm256_add_epi32(x, rnding);
+    v[0] = _mm256_srai_epi32(v[0], bit);
+
+    x = _mm256_mullo_epi32(in[0], cospi2);
+    v[1] = _mm256_sub_epi32(zero, x);
+    v[1] = _mm256_add_epi32(v[1], rnding);
+    v[1] = _mm256_srai_epi32(v[1], bit);
+
+    // stage 3
+    v[8] = v[0];
+    v[9] = v[1];
+
+    // stage 4
+    temp1 = _mm256_mullo_epi32(v[8], cospi8);
+    x = _mm256_mullo_epi32(v[9], cospi56);
+    temp1 = _mm256_add_epi32(temp1, x);
+    temp1 = _mm256_add_epi32(temp1, rnding);
+    temp1 = _mm256_srai_epi32(temp1, bit);
+
+    temp2 = _mm256_mullo_epi32(v[8], cospi56);
+    x = _mm256_mullo_epi32(v[9], cospi8);
+    temp2 = _mm256_sub_epi32(temp2, x);
+    temp2 = _mm256_add_epi32(temp2, rnding);
+    temp2 = _mm256_srai_epi32(temp2, bit);
+    v[8] = temp1;
+    v[9] = temp2;
+
+    // stage 5
+    v[4] = v[0];
+    v[5] = v[1];
+    v[12] = v[8];
+    v[13] = v[9];
+
+    // stage 6
+    temp1 = _mm256_mullo_epi32(v[4], cospi16);
+    x = _mm256_mullo_epi32(v[5], cospi48);
+    temp1 = _mm256_add_epi32(temp1, x);
+    temp1 = _mm256_add_epi32(temp1, rnding);
+    temp1 = _mm256_srai_epi32(temp1, bit);
+
+    temp2 = _mm256_mullo_epi32(v[4], cospi48);
+    x = _mm256_mullo_epi32(v[5], cospi16);
+    temp2 = _mm256_sub_epi32(temp2, x);
+    temp2 = _mm256_add_epi32(temp2, rnding);
+    temp2 = _mm256_srai_epi32(temp2, bit);
+    v[4] = temp1;
+    v[5] = temp2;
+
+    temp1 = _mm256_mullo_epi32(v[12], cospi16);
+    x = _mm256_mullo_epi32(v[13], cospi48);
+    temp1 = _mm256_add_epi32(temp1, x);
+    temp1 = _mm256_add_epi32(temp1, rnding);
+    temp1 = _mm256_srai_epi32(temp1, bit);
+
+    temp2 = _mm256_mullo_epi32(v[12], cospi48);
+    x = _mm256_mullo_epi32(v[13], cospi16);
+    temp2 = _mm256_sub_epi32(temp2, x);
+    temp2 = _mm256_add_epi32(temp2, rnding);
+    temp2 = _mm256_srai_epi32(temp2, bit);
+    v[12] = temp1;
+    v[13] = temp2;
+
+    // stage 7
+    v[2] = v[0];
+    v[3] = v[1];
+    v[6] = v[4];
+    v[7] = v[5];
+    v[10] = v[8];
+    v[11] = v[9];
+    v[14] = v[12];
+    v[15] = v[13];
+
+    // stage 8
+    y = _mm256_mullo_epi32(v[2], cospi32);
+    x = _mm256_mullo_epi32(v[3], cospi32);
+    v[2] = _mm256_add_epi32(y, x);
+    v[2] = _mm256_add_epi32(v[2], rnding);
+    v[2] = _mm256_srai_epi32(v[2], bit);
+
+    v[3] = _mm256_sub_epi32(y, x);
+    v[3] = _mm256_add_epi32(v[3], rnding);
+    v[3] = _mm256_srai_epi32(v[3], bit);
+
+    y = _mm256_mullo_epi32(v[6], cospi32);
+    x = _mm256_mullo_epi32(v[7], cospi32);
+    v[6] = _mm256_add_epi32(y, x);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    v[7] = _mm256_sub_epi32(y, x);
+    v[7] = _mm256_add_epi32(v[7], rnding);
+    v[7] = _mm256_srai_epi32(v[7], bit);
+
+    y = _mm256_mullo_epi32(v[10], cospi32);
+    x = _mm256_mullo_epi32(v[11], cospi32);
+    v[10] = _mm256_add_epi32(y, x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[11] = _mm256_sub_epi32(y, x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    y = _mm256_mullo_epi32(v[14], cospi32);
+    x = _mm256_mullo_epi32(v[15], cospi32);
+    v[14] = _mm256_add_epi32(y, x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_sub_epi32(y, x);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 9
+    if (do_cols) {
+      out[0] = v[0];
+      out[1] = _mm256_sub_epi32(_mm256_setzero_si256(), v[8]);
+      out[2] = v[12];
+      out[3] = _mm256_sub_epi32(_mm256_setzero_si256(), v[4]);
+      out[4] = v[6];
+      out[5] = _mm256_sub_epi32(_mm256_setzero_si256(), v[14]);
+      out[6] = v[10];
+      out[7] = _mm256_sub_epi32(_mm256_setzero_si256(), v[2]);
+      out[8] = v[3];
+      out[9] = _mm256_sub_epi32(_mm256_setzero_si256(), v[11]);
+      out[10] = v[15];
+      out[11] = _mm256_sub_epi32(_mm256_setzero_si256(), v[7]);
+      out[12] = v[5];
+      out[13] = _mm256_sub_epi32(_mm256_setzero_si256(), v[13]);
+      out[14] = v[9];
+      out[15] = _mm256_sub_epi32(_mm256_setzero_si256(), v[1]);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out =
+          _mm256_set1_epi32(-(1 << (log_range_out - 1)));
+      const __m256i clamp_hi_out =
+          _mm256_set1_epi32((1 << (log_range_out - 1)) - 1);
+
+      neg_shift_avx2(v[0], v[8], out + 0, out + 1, &clamp_lo_out, &clamp_hi_out,
+                     out_shift);
+      neg_shift_avx2(v[12], v[4], out + 2, out + 3, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[6], v[14], out + 4, out + 5, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[10], v[2], out + 6, out + 7, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[3], v[11], out + 8, out + 9, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[15], v[7], out + 10, out + 11, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[5], v[13], out + 12, out + 13, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[9], v[1], out + 14, out + 15, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+    }
+  }
+}
+
+static void iadst16_low8_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospi10 = _mm256_set1_epi32(cospi[10]);
+  const __m256i cospi54 = _mm256_set1_epi32(cospi[54]);
+  const __m256i cospi18 = _mm256_set1_epi32(cospi[18]);
+  const __m256i cospi46 = _mm256_set1_epi32(cospi[46]);
+  const __m256i cospi26 = _mm256_set1_epi32(cospi[26]);
+  const __m256i cospi38 = _mm256_set1_epi32(cospi[38]);
+  const __m256i cospi34 = _mm256_set1_epi32(cospi[34]);
+  const __m256i cospi30 = _mm256_set1_epi32(cospi[30]);
+  const __m256i cospi42 = _mm256_set1_epi32(cospi[42]);
+  const __m256i cospi22 = _mm256_set1_epi32(cospi[22]);
+  const __m256i cospi50 = _mm256_set1_epi32(cospi[50]);
+  const __m256i cospi14 = _mm256_set1_epi32(cospi[14]);
+  const __m256i cospi58 = _mm256_set1_epi32(cospi[58]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u[16], x, y;
+
+  {
+    // stage 0
+    // stage 1
+    // stage 2
+    __m256i zero = _mm256_setzero_si256();
+    x = _mm256_mullo_epi32(in[0], cospi62);
+    u[0] = _mm256_add_epi32(x, rnding);
+    u[0] = _mm256_srai_epi32(u[0], bit);
+
+    x = _mm256_mullo_epi32(in[0], cospi2);
+    u[1] = _mm256_sub_epi32(zero, x);
+    u[1] = _mm256_add_epi32(u[1], rnding);
+    u[1] = _mm256_srai_epi32(u[1], bit);
+
+    x = _mm256_mullo_epi32(in[2], cospi54);
+    u[2] = _mm256_add_epi32(x, rnding);
+    u[2] = _mm256_srai_epi32(u[2], bit);
+
+    x = _mm256_mullo_epi32(in[2], cospi10);
+    u[3] = _mm256_sub_epi32(zero, x);
+    u[3] = _mm256_add_epi32(u[3], rnding);
+    u[3] = _mm256_srai_epi32(u[3], bit);
+
+    x = _mm256_mullo_epi32(in[4], cospi46);
+    u[4] = _mm256_add_epi32(x, rnding);
+    u[4] = _mm256_srai_epi32(u[4], bit);
+
+    x = _mm256_mullo_epi32(in[4], cospi18);
+    u[5] = _mm256_sub_epi32(zero, x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    x = _mm256_mullo_epi32(in[6], cospi38);
+    u[6] = _mm256_add_epi32(x, rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    x = _mm256_mullo_epi32(in[6], cospi26);
+    u[7] = _mm256_sub_epi32(zero, x);
+    u[7] = _mm256_add_epi32(u[7], rnding);
+    u[7] = _mm256_srai_epi32(u[7], bit);
+
+    u[8] = _mm256_mullo_epi32(in[7], cospi34);
+    u[8] = _mm256_add_epi32(u[8], rnding);
+    u[8] = _mm256_srai_epi32(u[8], bit);
+
+    u[9] = _mm256_mullo_epi32(in[7], cospi30);
+    u[9] = _mm256_add_epi32(u[9], rnding);
+    u[9] = _mm256_srai_epi32(u[9], bit);
+
+    u[10] = _mm256_mullo_epi32(in[5], cospi42);
+    u[10] = _mm256_add_epi32(u[10], rnding);
+    u[10] = _mm256_srai_epi32(u[10], bit);
+
+    u[11] = _mm256_mullo_epi32(in[5], cospi22);
+    u[11] = _mm256_add_epi32(u[11], rnding);
+    u[11] = _mm256_srai_epi32(u[11], bit);
+
+    u[12] = _mm256_mullo_epi32(in[3], cospi50);
+    u[12] = _mm256_add_epi32(u[12], rnding);
+    u[12] = _mm256_srai_epi32(u[12], bit);
+
+    u[13] = _mm256_mullo_epi32(in[3], cospi14);
+    u[13] = _mm256_add_epi32(u[13], rnding);
+    u[13] = _mm256_srai_epi32(u[13], bit);
+
+    u[14] = _mm256_mullo_epi32(in[1], cospi58);
+    u[14] = _mm256_add_epi32(u[14], rnding);
+    u[14] = _mm256_srai_epi32(u[14], bit);
+
+    u[15] = _mm256_mullo_epi32(in[1], cospi6);
+    u[15] = _mm256_add_epi32(u[15], rnding);
+    u[15] = _mm256_srai_epi32(u[15], bit);
+
+    // stage 3
+    addsub_avx2(u[0], u[8], &u[0], &u[8], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[9], &u[1], &u[9], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[2], u[10], &u[2], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[3], u[11], &u[3], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[4], u[12], &u[4], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[5], u[13], &u[5], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[6], u[14], &u[6], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[7], u[15], &u[7], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 4
+    y = _mm256_mullo_epi32(u[8], cospi56);
+    x = _mm256_mullo_epi32(u[9], cospi56);
+    u[8] = _mm256_mullo_epi32(u[8], cospi8);
+    u[8] = _mm256_add_epi32(u[8], x);
+    u[8] = _mm256_add_epi32(u[8], rnding);
+    u[8] = _mm256_srai_epi32(u[8], bit);
+
+    x = _mm256_mullo_epi32(u[9], cospi8);
+    u[9] = _mm256_sub_epi32(y, x);
+    u[9] = _mm256_add_epi32(u[9], rnding);
+    u[9] = _mm256_srai_epi32(u[9], bit);
+
+    x = _mm256_mullo_epi32(u[11], cospi24);
+    y = _mm256_mullo_epi32(u[10], cospi24);
+    u[10] = _mm256_mullo_epi32(u[10], cospi40);
+    u[10] = _mm256_add_epi32(u[10], x);
+    u[10] = _mm256_add_epi32(u[10], rnding);
+    u[10] = _mm256_srai_epi32(u[10], bit);
+
+    x = _mm256_mullo_epi32(u[11], cospi40);
+    u[11] = _mm256_sub_epi32(y, x);
+    u[11] = _mm256_add_epi32(u[11], rnding);
+    u[11] = _mm256_srai_epi32(u[11], bit);
+
+    x = _mm256_mullo_epi32(u[13], cospi8);
+    y = _mm256_mullo_epi32(u[12], cospi8);
+    u[12] = _mm256_mullo_epi32(u[12], cospim56);
+    u[12] = _mm256_add_epi32(u[12], x);
+    u[12] = _mm256_add_epi32(u[12], rnding);
+    u[12] = _mm256_srai_epi32(u[12], bit);
+
+    x = _mm256_mullo_epi32(u[13], cospim56);
+    u[13] = _mm256_sub_epi32(y, x);
+    u[13] = _mm256_add_epi32(u[13], rnding);
+    u[13] = _mm256_srai_epi32(u[13], bit);
+
+    x = _mm256_mullo_epi32(u[15], cospi40);
+    y = _mm256_mullo_epi32(u[14], cospi40);
+    u[14] = _mm256_mullo_epi32(u[14], cospim24);
+    u[14] = _mm256_add_epi32(u[14], x);
+    u[14] = _mm256_add_epi32(u[14], rnding);
+    u[14] = _mm256_srai_epi32(u[14], bit);
+
+    x = _mm256_mullo_epi32(u[15], cospim24);
+    u[15] = _mm256_sub_epi32(y, x);
+    u[15] = _mm256_add_epi32(u[15], rnding);
+    u[15] = _mm256_srai_epi32(u[15], bit);
+
+    // stage 5
+    addsub_avx2(u[0], u[4], &u[0], &u[4], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[5], &u[1], &u[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[2], u[6], &u[2], &u[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[3], u[7], &u[3], &u[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[8], u[12], &u[8], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[9], u[13], &u[9], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[10], u[14], &u[10], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[11], u[15], &u[11], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 6
+    x = _mm256_mullo_epi32(u[5], cospi48);
+    y = _mm256_mullo_epi32(u[4], cospi48);
+    u[4] = _mm256_mullo_epi32(u[4], cospi16);
+    u[4] = _mm256_add_epi32(u[4], x);
+    u[4] = _mm256_add_epi32(u[4], rnding);
+    u[4] = _mm256_srai_epi32(u[4], bit);
+
+    x = _mm256_mullo_epi32(u[5], cospi16);
+    u[5] = _mm256_sub_epi32(y, x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    x = _mm256_mullo_epi32(u[7], cospi16);
+    y = _mm256_mullo_epi32(u[6], cospi16);
+    u[6] = _mm256_mullo_epi32(u[6], cospim48);
+    u[6] = _mm256_add_epi32(u[6], x);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    x = _mm256_mullo_epi32(u[7], cospim48);
+    u[7] = _mm256_sub_epi32(y, x);
+    u[7] = _mm256_add_epi32(u[7], rnding);
+    u[7] = _mm256_srai_epi32(u[7], bit);
+
+    x = _mm256_mullo_epi32(u[13], cospi48);
+    y = _mm256_mullo_epi32(u[12], cospi48);
+    u[12] = _mm256_mullo_epi32(u[12], cospi16);
+    u[12] = _mm256_add_epi32(u[12], x);
+    u[12] = _mm256_add_epi32(u[12], rnding);
+    u[12] = _mm256_srai_epi32(u[12], bit);
+
+    x = _mm256_mullo_epi32(u[13], cospi16);
+    u[13] = _mm256_sub_epi32(y, x);
+    u[13] = _mm256_add_epi32(u[13], rnding);
+    u[13] = _mm256_srai_epi32(u[13], bit);
+
+    x = _mm256_mullo_epi32(u[15], cospi16);
+    y = _mm256_mullo_epi32(u[14], cospi16);
+    u[14] = _mm256_mullo_epi32(u[14], cospim48);
+    u[14] = _mm256_add_epi32(u[14], x);
+    u[14] = _mm256_add_epi32(u[14], rnding);
+    u[14] = _mm256_srai_epi32(u[14], bit);
+
+    x = _mm256_mullo_epi32(u[15], cospim48);
+    u[15] = _mm256_sub_epi32(y, x);
+    u[15] = _mm256_add_epi32(u[15], rnding);
+    u[15] = _mm256_srai_epi32(u[15], bit);
+
+    // stage 7
+    addsub_avx2(u[0], u[2], &u[0], &u[2], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[1], u[3], &u[1], &u[3], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[4], u[6], &u[4], &u[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[5], u[7], &u[5], &u[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[8], u[10], &u[8], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[9], u[11], &u[9], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[12], u[14], &u[12], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[13], u[15], &u[13], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 8
+    y = _mm256_mullo_epi32(u[2], cospi32);
+    x = _mm256_mullo_epi32(u[3], cospi32);
+    u[2] = _mm256_add_epi32(y, x);
+    u[2] = _mm256_add_epi32(u[2], rnding);
+    u[2] = _mm256_srai_epi32(u[2], bit);
+
+    u[3] = _mm256_sub_epi32(y, x);
+    u[3] = _mm256_add_epi32(u[3], rnding);
+    u[3] = _mm256_srai_epi32(u[3], bit);
+    y = _mm256_mullo_epi32(u[6], cospi32);
+    x = _mm256_mullo_epi32(u[7], cospi32);
+    u[6] = _mm256_add_epi32(y, x);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    u[7] = _mm256_sub_epi32(y, x);
+    u[7] = _mm256_add_epi32(u[7], rnding);
+    u[7] = _mm256_srai_epi32(u[7], bit);
+
+    y = _mm256_mullo_epi32(u[10], cospi32);
+    x = _mm256_mullo_epi32(u[11], cospi32);
+    u[10] = _mm256_add_epi32(y, x);
+    u[10] = _mm256_add_epi32(u[10], rnding);
+    u[10] = _mm256_srai_epi32(u[10], bit);
+
+    u[11] = _mm256_sub_epi32(y, x);
+    u[11] = _mm256_add_epi32(u[11], rnding);
+    u[11] = _mm256_srai_epi32(u[11], bit);
+
+    y = _mm256_mullo_epi32(u[14], cospi32);
+    x = _mm256_mullo_epi32(u[15], cospi32);
+    u[14] = _mm256_add_epi32(y, x);
+    u[14] = _mm256_add_epi32(u[14], rnding);
+    u[14] = _mm256_srai_epi32(u[14], bit);
+
+    u[15] = _mm256_sub_epi32(y, x);
+    u[15] = _mm256_add_epi32(u[15], rnding);
+    u[15] = _mm256_srai_epi32(u[15], bit);
+
+    // stage 9
+    if (do_cols) {
+      out[0] = u[0];
+      out[1] = _mm256_sub_epi32(_mm256_setzero_si256(), u[8]);
+      out[2] = u[12];
+      out[3] = _mm256_sub_epi32(_mm256_setzero_si256(), u[4]);
+      out[4] = u[6];
+      out[5] = _mm256_sub_epi32(_mm256_setzero_si256(), u[14]);
+      out[6] = u[10];
+      out[7] = _mm256_sub_epi32(_mm256_setzero_si256(), u[2]);
+      out[8] = u[3];
+      out[9] = _mm256_sub_epi32(_mm256_setzero_si256(), u[11]);
+      out[10] = u[15];
+      out[11] = _mm256_sub_epi32(_mm256_setzero_si256(), u[7]);
+      out[12] = u[5];
+      out[13] = _mm256_sub_epi32(_mm256_setzero_si256(), u[13]);
+      out[14] = u[9];
+      out[15] = _mm256_sub_epi32(_mm256_setzero_si256(), u[1]);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out =
+          _mm256_set1_epi32(-(1 << (log_range_out - 1)));
+      const __m256i clamp_hi_out =
+          _mm256_set1_epi32((1 << (log_range_out - 1)) - 1);
+
+      neg_shift_avx2(u[0], u[8], out + 0, out + 1, &clamp_lo_out, &clamp_hi_out,
+                     out_shift);
+      neg_shift_avx2(u[12], u[4], out + 2, out + 3, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[6], u[14], out + 4, out + 5, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[10], u[2], out + 6, out + 7, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[3], u[11], out + 8, out + 9, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[15], u[7], out + 10, out + 11, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[5], u[13], out + 12, out + 13, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(u[9], u[1], out + 14, out + 15, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+    }
+  }
+}
+
+static void iadst16_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                         int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospi10 = _mm256_set1_epi32(cospi[10]);
+  const __m256i cospi54 = _mm256_set1_epi32(cospi[54]);
+  const __m256i cospi18 = _mm256_set1_epi32(cospi[18]);
+  const __m256i cospi46 = _mm256_set1_epi32(cospi[46]);
+  const __m256i cospi26 = _mm256_set1_epi32(cospi[26]);
+  const __m256i cospi38 = _mm256_set1_epi32(cospi[38]);
+  const __m256i cospi34 = _mm256_set1_epi32(cospi[34]);
+  const __m256i cospi30 = _mm256_set1_epi32(cospi[30]);
+  const __m256i cospi42 = _mm256_set1_epi32(cospi[42]);
+  const __m256i cospi22 = _mm256_set1_epi32(cospi[22]);
+  const __m256i cospi50 = _mm256_set1_epi32(cospi[50]);
+  const __m256i cospi14 = _mm256_set1_epi32(cospi[14]);
+  const __m256i cospi58 = _mm256_set1_epi32(cospi[58]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u[16], v[16], x, y;
+
+  {
+    // stage 0
+    // stage 1
+    // stage 2
+    v[0] = _mm256_mullo_epi32(in[15], cospi2);
+    x = _mm256_mullo_epi32(in[0], cospi62);
+    v[0] = _mm256_add_epi32(v[0], x);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    v[0] = _mm256_srai_epi32(v[0], bit);
+
+    v[1] = _mm256_mullo_epi32(in[15], cospi62);
+    x = _mm256_mullo_epi32(in[0], cospi2);
+    v[1] = _mm256_sub_epi32(v[1], x);
+    v[1] = _mm256_add_epi32(v[1], rnding);
+    v[1] = _mm256_srai_epi32(v[1], bit);
+
+    v[2] = _mm256_mullo_epi32(in[13], cospi10);
+    x = _mm256_mullo_epi32(in[2], cospi54);
+    v[2] = _mm256_add_epi32(v[2], x);
+    v[2] = _mm256_add_epi32(v[2], rnding);
+    v[2] = _mm256_srai_epi32(v[2], bit);
+
+    v[3] = _mm256_mullo_epi32(in[13], cospi54);
+    x = _mm256_mullo_epi32(in[2], cospi10);
+    v[3] = _mm256_sub_epi32(v[3], x);
+    v[3] = _mm256_add_epi32(v[3], rnding);
+    v[3] = _mm256_srai_epi32(v[3], bit);
+
+    v[4] = _mm256_mullo_epi32(in[11], cospi18);
+    x = _mm256_mullo_epi32(in[4], cospi46);
+    v[4] = _mm256_add_epi32(v[4], x);
+    v[4] = _mm256_add_epi32(v[4], rnding);
+    v[4] = _mm256_srai_epi32(v[4], bit);
+
+    v[5] = _mm256_mullo_epi32(in[11], cospi46);
+    x = _mm256_mullo_epi32(in[4], cospi18);
+    v[5] = _mm256_sub_epi32(v[5], x);
+    v[5] = _mm256_add_epi32(v[5], rnding);
+    v[5] = _mm256_srai_epi32(v[5], bit);
+
+    v[6] = _mm256_mullo_epi32(in[9], cospi26);
+    x = _mm256_mullo_epi32(in[6], cospi38);
+    v[6] = _mm256_add_epi32(v[6], x);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    v[7] = _mm256_mullo_epi32(in[9], cospi38);
+    x = _mm256_mullo_epi32(in[6], cospi26);
+    v[7] = _mm256_sub_epi32(v[7], x);
+    v[7] = _mm256_add_epi32(v[7], rnding);
+    v[7] = _mm256_srai_epi32(v[7], bit);
+
+    v[8] = _mm256_mullo_epi32(in[7], cospi34);
+    x = _mm256_mullo_epi32(in[8], cospi30);
+    v[8] = _mm256_add_epi32(v[8], x);
+    v[8] = _mm256_add_epi32(v[8], rnding);
+    v[8] = _mm256_srai_epi32(v[8], bit);
+
+    v[9] = _mm256_mullo_epi32(in[7], cospi30);
+    x = _mm256_mullo_epi32(in[8], cospi34);
+    v[9] = _mm256_sub_epi32(v[9], x);
+    v[9] = _mm256_add_epi32(v[9], rnding);
+    v[9] = _mm256_srai_epi32(v[9], bit);
+
+    v[10] = _mm256_mullo_epi32(in[5], cospi42);
+    x = _mm256_mullo_epi32(in[10], cospi22);
+    v[10] = _mm256_add_epi32(v[10], x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[11] = _mm256_mullo_epi32(in[5], cospi22);
+    x = _mm256_mullo_epi32(in[10], cospi42);
+    v[11] = _mm256_sub_epi32(v[11], x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = _mm256_mullo_epi32(in[3], cospi50);
+    x = _mm256_mullo_epi32(in[12], cospi14);
+    v[12] = _mm256_add_epi32(v[12], x);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+
+    v[13] = _mm256_mullo_epi32(in[3], cospi14);
+    x = _mm256_mullo_epi32(in[12], cospi50);
+    v[13] = _mm256_sub_epi32(v[13], x);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[14] = _mm256_mullo_epi32(in[1], cospi58);
+    x = _mm256_mullo_epi32(in[14], cospi6);
+    v[14] = _mm256_add_epi32(v[14], x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_mullo_epi32(in[1], cospi6);
+    x = _mm256_mullo_epi32(in[14], cospi58);
+    v[15] = _mm256_sub_epi32(v[15], x);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 3
+    addsub_avx2(v[0], v[8], &u[0], &u[8], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[1], v[9], &u[1], &u[9], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[2], v[10], &u[2], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[3], v[11], &u[3], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[4], v[12], &u[4], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[5], v[13], &u[5], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[6], v[14], &u[6], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[7], v[15], &u[7], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 4
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+    v[4] = u[4];
+    v[5] = u[5];
+    v[6] = u[6];
+    v[7] = u[7];
+
+    v[8] = _mm256_mullo_epi32(u[8], cospi8);
+    x = _mm256_mullo_epi32(u[9], cospi56);
+    v[8] = _mm256_add_epi32(v[8], x);
+    v[8] = _mm256_add_epi32(v[8], rnding);
+    v[8] = _mm256_srai_epi32(v[8], bit);
+
+    v[9] = _mm256_mullo_epi32(u[8], cospi56);
+    x = _mm256_mullo_epi32(u[9], cospi8);
+    v[9] = _mm256_sub_epi32(v[9], x);
+    v[9] = _mm256_add_epi32(v[9], rnding);
+    v[9] = _mm256_srai_epi32(v[9], bit);
+
+    v[10] = _mm256_mullo_epi32(u[10], cospi40);
+    x = _mm256_mullo_epi32(u[11], cospi24);
+    v[10] = _mm256_add_epi32(v[10], x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[11] = _mm256_mullo_epi32(u[10], cospi24);
+    x = _mm256_mullo_epi32(u[11], cospi40);
+    v[11] = _mm256_sub_epi32(v[11], x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = _mm256_mullo_epi32(u[12], cospim56);
+    x = _mm256_mullo_epi32(u[13], cospi8);
+    v[12] = _mm256_add_epi32(v[12], x);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+
+    v[13] = _mm256_mullo_epi32(u[12], cospi8);
+    x = _mm256_mullo_epi32(u[13], cospim56);
+    v[13] = _mm256_sub_epi32(v[13], x);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[14] = _mm256_mullo_epi32(u[14], cospim24);
+    x = _mm256_mullo_epi32(u[15], cospi40);
+    v[14] = _mm256_add_epi32(v[14], x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_mullo_epi32(u[14], cospi40);
+    x = _mm256_mullo_epi32(u[15], cospim24);
+    v[15] = _mm256_sub_epi32(v[15], x);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 5
+    addsub_avx2(v[0], v[4], &u[0], &u[4], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[1], v[5], &u[1], &u[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[2], v[6], &u[2], &u[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[3], v[7], &u[3], &u[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[8], v[12], &u[8], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[9], v[13], &u[9], &u[13], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[10], v[14], &u[10], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[11], v[15], &u[11], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 6
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+
+    v[4] = _mm256_mullo_epi32(u[4], cospi16);
+    x = _mm256_mullo_epi32(u[5], cospi48);
+    v[4] = _mm256_add_epi32(v[4], x);
+    v[4] = _mm256_add_epi32(v[4], rnding);
+    v[4] = _mm256_srai_epi32(v[4], bit);
+
+    v[5] = _mm256_mullo_epi32(u[4], cospi48);
+    x = _mm256_mullo_epi32(u[5], cospi16);
+    v[5] = _mm256_sub_epi32(v[5], x);
+    v[5] = _mm256_add_epi32(v[5], rnding);
+    v[5] = _mm256_srai_epi32(v[5], bit);
+
+    v[6] = _mm256_mullo_epi32(u[6], cospim48);
+    x = _mm256_mullo_epi32(u[7], cospi16);
+    v[6] = _mm256_add_epi32(v[6], x);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    v[7] = _mm256_mullo_epi32(u[6], cospi16);
+    x = _mm256_mullo_epi32(u[7], cospim48);
+    v[7] = _mm256_sub_epi32(v[7], x);
+    v[7] = _mm256_add_epi32(v[7], rnding);
+    v[7] = _mm256_srai_epi32(v[7], bit);
+
+    v[8] = u[8];
+    v[9] = u[9];
+    v[10] = u[10];
+    v[11] = u[11];
+
+    v[12] = _mm256_mullo_epi32(u[12], cospi16);
+    x = _mm256_mullo_epi32(u[13], cospi48);
+    v[12] = _mm256_add_epi32(v[12], x);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+
+    v[13] = _mm256_mullo_epi32(u[12], cospi48);
+    x = _mm256_mullo_epi32(u[13], cospi16);
+    v[13] = _mm256_sub_epi32(v[13], x);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[14] = _mm256_mullo_epi32(u[14], cospim48);
+    x = _mm256_mullo_epi32(u[15], cospi16);
+    v[14] = _mm256_add_epi32(v[14], x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_mullo_epi32(u[14], cospi16);
+    x = _mm256_mullo_epi32(u[15], cospim48);
+    v[15] = _mm256_sub_epi32(v[15], x);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 7
+    addsub_avx2(v[0], v[2], &u[0], &u[2], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[1], v[3], &u[1], &u[3], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[4], v[6], &u[4], &u[6], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[5], v[7], &u[5], &u[7], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[8], v[10], &u[8], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[9], v[11], &u[9], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[12], v[14], &u[12], &u[14], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[13], v[15], &u[13], &u[15], &clamp_lo, &clamp_hi);
+
+    // stage 8
+    v[0] = u[0];
+    v[1] = u[1];
+
+    y = _mm256_mullo_epi32(u[2], cospi32);
+    x = _mm256_mullo_epi32(u[3], cospi32);
+    v[2] = _mm256_add_epi32(y, x);
+    v[2] = _mm256_add_epi32(v[2], rnding);
+    v[2] = _mm256_srai_epi32(v[2], bit);
+
+    v[3] = _mm256_sub_epi32(y, x);
+    v[3] = _mm256_add_epi32(v[3], rnding);
+    v[3] = _mm256_srai_epi32(v[3], bit);
+
+    v[4] = u[4];
+    v[5] = u[5];
+
+    y = _mm256_mullo_epi32(u[6], cospi32);
+    x = _mm256_mullo_epi32(u[7], cospi32);
+    v[6] = _mm256_add_epi32(y, x);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    v[7] = _mm256_sub_epi32(y, x);
+    v[7] = _mm256_add_epi32(v[7], rnding);
+    v[7] = _mm256_srai_epi32(v[7], bit);
+
+    v[8] = u[8];
+    v[9] = u[9];
+
+    y = _mm256_mullo_epi32(u[10], cospi32);
+    x = _mm256_mullo_epi32(u[11], cospi32);
+    v[10] = _mm256_add_epi32(y, x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[11] = _mm256_sub_epi32(y, x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = u[12];
+    v[13] = u[13];
+
+    y = _mm256_mullo_epi32(u[14], cospi32);
+    x = _mm256_mullo_epi32(u[15], cospi32);
+    v[14] = _mm256_add_epi32(y, x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_sub_epi32(y, x);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 9
+    if (do_cols) {
+      out[0] = v[0];
+      out[1] = _mm256_sub_epi32(_mm256_setzero_si256(), v[8]);
+      out[2] = v[12];
+      out[3] = _mm256_sub_epi32(_mm256_setzero_si256(), v[4]);
+      out[4] = v[6];
+      out[5] = _mm256_sub_epi32(_mm256_setzero_si256(), v[14]);
+      out[6] = v[10];
+      out[7] = _mm256_sub_epi32(_mm256_setzero_si256(), v[2]);
+      out[8] = v[3];
+      out[9] = _mm256_sub_epi32(_mm256_setzero_si256(), v[11]);
+      out[10] = v[15];
+      out[11] = _mm256_sub_epi32(_mm256_setzero_si256(), v[7]);
+      out[12] = v[5];
+      out[13] = _mm256_sub_epi32(_mm256_setzero_si256(), v[13]);
+      out[14] = v[9];
+      out[15] = _mm256_sub_epi32(_mm256_setzero_si256(), v[1]);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out =
+          _mm256_set1_epi32(-(1 << (log_range_out - 1)));
+      const __m256i clamp_hi_out =
+          _mm256_set1_epi32((1 << (log_range_out - 1)) - 1);
+
+      neg_shift_avx2(v[0], v[8], out + 0, out + 1, &clamp_lo_out, &clamp_hi_out,
+                     out_shift);
+      neg_shift_avx2(v[12], v[4], out + 2, out + 3, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[6], v[14], out + 4, out + 5, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[10], v[2], out + 6, out + 7, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[3], v[11], out + 8, out + 9, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[15], v[7], out + 10, out + 11, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[5], v[13], out + 12, out + 13, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+      neg_shift_avx2(v[9], v[1], out + 14, out + 15, &clamp_lo_out,
+                     &clamp_hi_out, out_shift);
+    }
+  }
+}
+static void idct8x8_low1_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  __m256i x;
+
+  // stage 0
+  // stage 1
+  // stage 2
+  // stage 3
+  x = _mm256_mullo_epi32(in[0], cospi32);
+  x = _mm256_add_epi32(x, rnding);
+  x = _mm256_srai_epi32(x, bit);
+
+  // stage 4
+  // stage 5
+  if (!do_cols) {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+    __m256i offset = _mm256_set1_epi32((1 << out_shift) >> 1);
+    x = _mm256_add_epi32(x, offset);
+    x = _mm256_sra_epi32(x, _mm_cvtsi32_si128(out_shift));
+    x = _mm256_max_epi32(x, clamp_lo_out);
+    x = _mm256_min_epi32(x, clamp_hi_out);
+  }
+
+  out[0] = x;
+  out[1] = x;
+  out[2] = x;
+  out[3] = x;
+  out[4] = x;
+  out[5] = x;
+  out[6] = x;
+  out[7] = x;
+}
+static void idct8x8_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                         int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u0, u1, u2, u3, u4, u5, u6, u7;
+  __m256i v0, v1, v2, v3, v4, v5, v6, v7;
+  __m256i x, y;
+
+  // stage 0
+  // stage 1
+  // stage 2
+  u0 = in[0];
+  u1 = in[4];
+  u2 = in[2];
+  u3 = in[6];
+
+  x = _mm256_mullo_epi32(in[1], cospi56);
+  y = _mm256_mullo_epi32(in[7], cospim8);
+  u4 = _mm256_add_epi32(x, y);
+  u4 = _mm256_add_epi32(u4, rnding);
+  u4 = _mm256_srai_epi32(u4, bit);
+
+  x = _mm256_mullo_epi32(in[1], cospi8);
+  y = _mm256_mullo_epi32(in[7], cospi56);
+  u7 = _mm256_add_epi32(x, y);
+  u7 = _mm256_add_epi32(u7, rnding);
+  u7 = _mm256_srai_epi32(u7, bit);
+
+  x = _mm256_mullo_epi32(in[5], cospi24);
+  y = _mm256_mullo_epi32(in[3], cospim40);
+  u5 = _mm256_add_epi32(x, y);
+  u5 = _mm256_add_epi32(u5, rnding);
+  u5 = _mm256_srai_epi32(u5, bit);
+
+  x = _mm256_mullo_epi32(in[5], cospi40);
+  y = _mm256_mullo_epi32(in[3], cospi24);
+  u6 = _mm256_add_epi32(x, y);
+  u6 = _mm256_add_epi32(u6, rnding);
+  u6 = _mm256_srai_epi32(u6, bit);
+
+  // stage 3
+  x = _mm256_mullo_epi32(u0, cospi32);
+  y = _mm256_mullo_epi32(u1, cospi32);
+  v0 = _mm256_add_epi32(x, y);
+  v0 = _mm256_add_epi32(v0, rnding);
+  v0 = _mm256_srai_epi32(v0, bit);
+
+  v1 = _mm256_sub_epi32(x, y);
+  v1 = _mm256_add_epi32(v1, rnding);
+  v1 = _mm256_srai_epi32(v1, bit);
+
+  x = _mm256_mullo_epi32(u2, cospi48);
+  y = _mm256_mullo_epi32(u3, cospim16);
+  v2 = _mm256_add_epi32(x, y);
+  v2 = _mm256_add_epi32(v2, rnding);
+  v2 = _mm256_srai_epi32(v2, bit);
+
+  x = _mm256_mullo_epi32(u2, cospi16);
+  y = _mm256_mullo_epi32(u3, cospi48);
+  v3 = _mm256_add_epi32(x, y);
+  v3 = _mm256_add_epi32(v3, rnding);
+  v3 = _mm256_srai_epi32(v3, bit);
+
+  addsub_avx2(u4, u5, &v4, &v5, &clamp_lo, &clamp_hi);
+  addsub_avx2(u7, u6, &v7, &v6, &clamp_lo, &clamp_hi);
+
+  // stage 4
+  addsub_avx2(v0, v3, &u0, &u3, &clamp_lo, &clamp_hi);
+  addsub_avx2(v1, v2, &u1, &u2, &clamp_lo, &clamp_hi);
+  u4 = v4;
+  u7 = v7;
+
+  x = _mm256_mullo_epi32(v5, cospi32);
+  y = _mm256_mullo_epi32(v6, cospi32);
+  u6 = _mm256_add_epi32(y, x);
+  u6 = _mm256_add_epi32(u6, rnding);
+  u6 = _mm256_srai_epi32(u6, bit);
+
+  u5 = _mm256_sub_epi32(y, x);
+  u5 = _mm256_add_epi32(u5, rnding);
+  u5 = _mm256_srai_epi32(u5, bit);
+
+  // stage 5
+  if (do_cols) {
+    addsub_no_clamp_avx2(u0, u7, out + 0, out + 7);
+    addsub_no_clamp_avx2(u1, u6, out + 1, out + 6);
+    addsub_no_clamp_avx2(u2, u5, out + 2, out + 5);
+    addsub_no_clamp_avx2(u3, u4, out + 3, out + 4);
+  } else {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+    addsub_shift_avx2(u0, u7, out + 0, out + 7, &clamp_lo_out, &clamp_hi_out,
+                      out_shift);
+    addsub_shift_avx2(u1, u6, out + 1, out + 6, &clamp_lo_out, &clamp_hi_out,
+                      out_shift);
+    addsub_shift_avx2(u2, u5, out + 2, out + 5, &clamp_lo_out, &clamp_hi_out,
+                      out_shift);
+    addsub_shift_avx2(u3, u4, out + 3, out + 4, &clamp_lo_out, &clamp_hi_out,
+                      out_shift);
+  }
+}
+static void iadst8x8_low1_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                               int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const __m256i kZero = _mm256_setzero_si256();
+  __m256i u[8], x;
+
+  // stage 0
+  // stage 1
+  // stage 2
+
+  x = _mm256_mullo_epi32(in[0], cospi60);
+  u[0] = _mm256_add_epi32(x, rnding);
+  u[0] = _mm256_srai_epi32(u[0], bit);
+
+  x = _mm256_mullo_epi32(in[0], cospi4);
+  u[1] = _mm256_sub_epi32(kZero, x);
+  u[1] = _mm256_add_epi32(u[1], rnding);
+  u[1] = _mm256_srai_epi32(u[1], bit);
+
+  // stage 3
+  // stage 4
+  __m256i temp1, temp2;
+  temp1 = _mm256_mullo_epi32(u[0], cospi16);
+  x = _mm256_mullo_epi32(u[1], cospi48);
+  temp1 = _mm256_add_epi32(temp1, x);
+  temp1 = _mm256_add_epi32(temp1, rnding);
+  temp1 = _mm256_srai_epi32(temp1, bit);
+  u[4] = temp1;
+
+  temp2 = _mm256_mullo_epi32(u[0], cospi48);
+  x = _mm256_mullo_epi32(u[1], cospi16);
+  u[5] = _mm256_sub_epi32(temp2, x);
+  u[5] = _mm256_add_epi32(u[5], rnding);
+  u[5] = _mm256_srai_epi32(u[5], bit);
+
+  // stage 5
+  // stage 6
+  temp1 = _mm256_mullo_epi32(u[0], cospi32);
+  x = _mm256_mullo_epi32(u[1], cospi32);
+  u[2] = _mm256_add_epi32(temp1, x);
+  u[2] = _mm256_add_epi32(u[2], rnding);
+  u[2] = _mm256_srai_epi32(u[2], bit);
+
+  u[3] = _mm256_sub_epi32(temp1, x);
+  u[3] = _mm256_add_epi32(u[3], rnding);
+  u[3] = _mm256_srai_epi32(u[3], bit);
+
+  temp1 = _mm256_mullo_epi32(u[4], cospi32);
+  x = _mm256_mullo_epi32(u[5], cospi32);
+  u[6] = _mm256_add_epi32(temp1, x);
+  u[6] = _mm256_add_epi32(u[6], rnding);
+  u[6] = _mm256_srai_epi32(u[6], bit);
+
+  u[7] = _mm256_sub_epi32(temp1, x);
+  u[7] = _mm256_add_epi32(u[7], rnding);
+  u[7] = _mm256_srai_epi32(u[7], bit);
+
+  // stage 7
+  if (do_cols) {
+    out[0] = u[0];
+    out[1] = _mm256_sub_epi32(kZero, u[4]);
+    out[2] = u[6];
+    out[3] = _mm256_sub_epi32(kZero, u[2]);
+    out[4] = u[3];
+    out[5] = _mm256_sub_epi32(kZero, u[7]);
+    out[6] = u[5];
+    out[7] = _mm256_sub_epi32(kZero, u[1]);
+  } else {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m256i clamp_lo_out = _mm256_set1_epi32(-(1 << (log_range_out - 1)));
+    const __m256i clamp_hi_out =
+        _mm256_set1_epi32((1 << (log_range_out - 1)) - 1);
+
+    neg_shift_avx2(u[0], u[4], out + 0, out + 1, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[6], u[2], out + 2, out + 3, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[3], u[7], out + 4, out + 5, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[5], u[1], out + 6, out + 7, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+  }
+}
+
+static void iadst8x8_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                          int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const __m256i kZero = _mm256_setzero_si256();
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+  __m256i u[8], v[8], x;
+
+  // stage 0
+  // stage 1
+  // stage 2
+
+  u[0] = _mm256_mullo_epi32(in[7], cospi4);
+  x = _mm256_mullo_epi32(in[0], cospi60);
+  u[0] = _mm256_add_epi32(u[0], x);
+  u[0] = _mm256_add_epi32(u[0], rnding);
+  u[0] = _mm256_srai_epi32(u[0], bit);
+
+  u[1] = _mm256_mullo_epi32(in[7], cospi60);
+  x = _mm256_mullo_epi32(in[0], cospi4);
+  u[1] = _mm256_sub_epi32(u[1], x);
+  u[1] = _mm256_add_epi32(u[1], rnding);
+  u[1] = _mm256_srai_epi32(u[1], bit);
+
+  u[2] = _mm256_mullo_epi32(in[5], cospi20);
+  x = _mm256_mullo_epi32(in[2], cospi44);
+  u[2] = _mm256_add_epi32(u[2], x);
+  u[2] = _mm256_add_epi32(u[2], rnding);
+  u[2] = _mm256_srai_epi32(u[2], bit);
+
+  u[3] = _mm256_mullo_epi32(in[5], cospi44);
+  x = _mm256_mullo_epi32(in[2], cospi20);
+  u[3] = _mm256_sub_epi32(u[3], x);
+  u[3] = _mm256_add_epi32(u[3], rnding);
+  u[3] = _mm256_srai_epi32(u[3], bit);
+
+  u[4] = _mm256_mullo_epi32(in[3], cospi36);
+  x = _mm256_mullo_epi32(in[4], cospi28);
+  u[4] = _mm256_add_epi32(u[4], x);
+  u[4] = _mm256_add_epi32(u[4], rnding);
+  u[4] = _mm256_srai_epi32(u[4], bit);
+
+  u[5] = _mm256_mullo_epi32(in[3], cospi28);
+  x = _mm256_mullo_epi32(in[4], cospi36);
+  u[5] = _mm256_sub_epi32(u[5], x);
+  u[5] = _mm256_add_epi32(u[5], rnding);
+  u[5] = _mm256_srai_epi32(u[5], bit);
+
+  u[6] = _mm256_mullo_epi32(in[1], cospi52);
+  x = _mm256_mullo_epi32(in[6], cospi12);
+  u[6] = _mm256_add_epi32(u[6], x);
+  u[6] = _mm256_add_epi32(u[6], rnding);
+  u[6] = _mm256_srai_epi32(u[6], bit);
+
+  u[7] = _mm256_mullo_epi32(in[1], cospi12);
+  x = _mm256_mullo_epi32(in[6], cospi52);
+  u[7] = _mm256_sub_epi32(u[7], x);
+  u[7] = _mm256_add_epi32(u[7], rnding);
+  u[7] = _mm256_srai_epi32(u[7], bit);
+
+  // stage 3
+  addsub_avx2(u[0], u[4], &v[0], &v[4], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[1], u[5], &v[1], &v[5], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[2], u[6], &v[2], &v[6], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[3], u[7], &v[3], &v[7], &clamp_lo, &clamp_hi);
+
+  // stage 4
+  u[0] = v[0];
+  u[1] = v[1];
+  u[2] = v[2];
+  u[3] = v[3];
+
+  u[4] = _mm256_mullo_epi32(v[4], cospi16);
+  x = _mm256_mullo_epi32(v[5], cospi48);
+  u[4] = _mm256_add_epi32(u[4], x);
+  u[4] = _mm256_add_epi32(u[4], rnding);
+  u[4] = _mm256_srai_epi32(u[4], bit);
+
+  u[5] = _mm256_mullo_epi32(v[4], cospi48);
+  x = _mm256_mullo_epi32(v[5], cospi16);
+  u[5] = _mm256_sub_epi32(u[5], x);
+  u[5] = _mm256_add_epi32(u[5], rnding);
+  u[5] = _mm256_srai_epi32(u[5], bit);
+
+  u[6] = _mm256_mullo_epi32(v[6], cospim48);
+  x = _mm256_mullo_epi32(v[7], cospi16);
+  u[6] = _mm256_add_epi32(u[6], x);
+  u[6] = _mm256_add_epi32(u[6], rnding);
+  u[6] = _mm256_srai_epi32(u[6], bit);
+
+  u[7] = _mm256_mullo_epi32(v[6], cospi16);
+  x = _mm256_mullo_epi32(v[7], cospim48);
+  u[7] = _mm256_sub_epi32(u[7], x);
+  u[7] = _mm256_add_epi32(u[7], rnding);
+  u[7] = _mm256_srai_epi32(u[7], bit);
+
+  // stage 5
+  addsub_avx2(u[0], u[2], &v[0], &v[2], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[1], u[3], &v[1], &v[3], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[4], u[6], &v[4], &v[6], &clamp_lo, &clamp_hi);
+  addsub_avx2(u[5], u[7], &v[5], &v[7], &clamp_lo, &clamp_hi);
+
+  // stage 6
+  u[0] = v[0];
+  u[1] = v[1];
+  u[4] = v[4];
+  u[5] = v[5];
+
+  v[0] = _mm256_mullo_epi32(v[2], cospi32);
+  x = _mm256_mullo_epi32(v[3], cospi32);
+  u[2] = _mm256_add_epi32(v[0], x);
+  u[2] = _mm256_add_epi32(u[2], rnding);
+  u[2] = _mm256_srai_epi32(u[2], bit);
+
+  u[3] = _mm256_sub_epi32(v[0], x);
+  u[3] = _mm256_add_epi32(u[3], rnding);
+  u[3] = _mm256_srai_epi32(u[3], bit);
+
+  v[0] = _mm256_mullo_epi32(v[6], cospi32);
+  x = _mm256_mullo_epi32(v[7], cospi32);
+  u[6] = _mm256_add_epi32(v[0], x);
+  u[6] = _mm256_add_epi32(u[6], rnding);
+  u[6] = _mm256_srai_epi32(u[6], bit);
+
+  u[7] = _mm256_sub_epi32(v[0], x);
+  u[7] = _mm256_add_epi32(u[7], rnding);
+  u[7] = _mm256_srai_epi32(u[7], bit);
+
+  // stage 7
+  if (do_cols) {
+    out[0] = u[0];
+    out[1] = _mm256_sub_epi32(kZero, u[4]);
+    out[2] = u[6];
+    out[3] = _mm256_sub_epi32(kZero, u[2]);
+    out[4] = u[3];
+    out[5] = _mm256_sub_epi32(kZero, u[7]);
+    out[6] = u[5];
+    out[7] = _mm256_sub_epi32(kZero, u[1]);
+  } else {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m256i clamp_lo_out = _mm256_set1_epi32(-(1 << (log_range_out - 1)));
+    const __m256i clamp_hi_out =
+        _mm256_set1_epi32((1 << (log_range_out - 1)) - 1);
+
+    neg_shift_avx2(u[0], u[4], out + 0, out + 1, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[6], u[2], out + 2, out + 3, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[3], u[7], out + 4, out + 5, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+    neg_shift_avx2(u[5], u[1], out + 6, out + 7, &clamp_lo_out, &clamp_hi_out,
+                   out_shift);
+  }
+}
+static INLINE void idct64_stage8_avx2(
+    __m256i *u, const __m256i *cospim32, const __m256i *cospi32,
+    const __m256i *cospim16, const __m256i *cospi48, const __m256i *cospi16,
+    const __m256i *cospim48, const __m256i *clamp_lo, const __m256i *clamp_hi,
+    const __m256i *rnding, int bit) {
+  int i;
+  __m256i temp1, temp2, temp3, temp4;
+  temp1 = half_btf_avx2(cospim32, &u[10], cospi32, &u[13], rnding, bit);
+  u[13] = half_btf_avx2(cospi32, &u[10], cospi32, &u[13], rnding, bit);
+  u[10] = temp1;
+  temp2 = half_btf_avx2(cospim32, &u[11], cospi32, &u[12], rnding, bit);
+  u[12] = half_btf_avx2(cospi32, &u[11], cospi32, &u[12], rnding, bit);
+  u[11] = temp2;
+
+  for (i = 16; i < 20; ++i) {
+    addsub_avx2(u[i], u[i ^ 7], &u[i], &u[i ^ 7], clamp_lo, clamp_hi);
+    addsub_avx2(u[i ^ 15], u[i ^ 8], &u[i ^ 15], &u[i ^ 8], clamp_lo, clamp_hi);
+  }
+
+  temp1 = half_btf_avx2(cospim16, &u[36], cospi48, &u[59], rnding, bit);
+  temp2 = half_btf_avx2(cospim16, &u[37], cospi48, &u[58], rnding, bit);
+  temp3 = half_btf_avx2(cospim16, &u[38], cospi48, &u[57], rnding, bit);
+  temp4 = half_btf_avx2(cospim16, &u[39], cospi48, &u[56], rnding, bit);
+  u[56] = half_btf_avx2(cospi48, &u[39], cospi16, &u[56], rnding, bit);
+  u[57] = half_btf_avx2(cospi48, &u[38], cospi16, &u[57], rnding, bit);
+  u[58] = half_btf_avx2(cospi48, &u[37], cospi16, &u[58], rnding, bit);
+  u[59] = half_btf_avx2(cospi48, &u[36], cospi16, &u[59], rnding, bit);
+  u[36] = temp1;
+  u[37] = temp2;
+  u[38] = temp3;
+  u[39] = temp4;
+
+  temp1 = half_btf_avx2(cospim48, &u[40], cospim16, &u[55], rnding, bit);
+  temp2 = half_btf_avx2(cospim48, &u[41], cospim16, &u[54], rnding, bit);
+  temp3 = half_btf_avx2(cospim48, &u[42], cospim16, &u[53], rnding, bit);
+  temp4 = half_btf_avx2(cospim48, &u[43], cospim16, &u[52], rnding, bit);
+  u[52] = half_btf_avx2(cospim16, &u[43], cospi48, &u[52], rnding, bit);
+  u[53] = half_btf_avx2(cospim16, &u[42], cospi48, &u[53], rnding, bit);
+  u[54] = half_btf_avx2(cospim16, &u[41], cospi48, &u[54], rnding, bit);
+  u[55] = half_btf_avx2(cospim16, &u[40], cospi48, &u[55], rnding, bit);
+  u[40] = temp1;
+  u[41] = temp2;
+  u[42] = temp3;
+  u[43] = temp4;
+}
+
+static INLINE void idct64_stage9_avx2(__m256i *u, const __m256i *cospim32,
+                                      const __m256i *cospi32,
+                                      const __m256i *clamp_lo,
+                                      const __m256i *clamp_hi,
+                                      const __m256i *rnding, int bit) {
+  int i;
+  __m256i temp1, temp2, temp3, temp4;
+  for (i = 0; i < 8; ++i) {
+    addsub_avx2(u[i], u[15 - i], &u[i], &u[15 - i], clamp_lo, clamp_hi);
+  }
+
+  temp1 = half_btf_avx2(cospim32, &u[20], cospi32, &u[27], rnding, bit);
+  temp2 = half_btf_avx2(cospim32, &u[21], cospi32, &u[26], rnding, bit);
+  temp3 = half_btf_avx2(cospim32, &u[22], cospi32, &u[25], rnding, bit);
+  temp4 = half_btf_avx2(cospim32, &u[23], cospi32, &u[24], rnding, bit);
+  u[24] = half_btf_avx2(cospi32, &u[23], cospi32, &u[24], rnding, bit);
+  u[25] = half_btf_avx2(cospi32, &u[22], cospi32, &u[25], rnding, bit);
+  u[26] = half_btf_avx2(cospi32, &u[21], cospi32, &u[26], rnding, bit);
+  u[27] = half_btf_avx2(cospi32, &u[20], cospi32, &u[27], rnding, bit);
+  u[20] = temp1;
+  u[21] = temp2;
+  u[22] = temp3;
+  u[23] = temp4;
+  for (i = 32; i < 40; i++) {
+    addsub_avx2(u[i], u[i ^ 15], &u[i], &u[i ^ 15], clamp_lo, clamp_hi);
+  }
+
+  for (i = 48; i < 56; i++) {
+    addsub_avx2(u[i ^ 15], u[i], &u[i ^ 15], &u[i], clamp_lo, clamp_hi);
+  }
+}
+
+static INLINE void idct64_stage10_avx2(__m256i *u, const __m256i *cospim32,
+                                       const __m256i *cospi32,
+                                       const __m256i *clamp_lo,
+                                       const __m256i *clamp_hi,
+                                       const __m256i *rnding, int bit) {
+  __m256i temp1, temp2, temp3, temp4;
+  for (int i = 0; i < 16; i++) {
+    addsub_avx2(u[i], u[31 - i], &u[i], &u[31 - i], clamp_lo, clamp_hi);
+  }
+
+  temp1 = half_btf_avx2(cospim32, &u[40], cospi32, &u[55], rnding, bit);
+  temp2 = half_btf_avx2(cospim32, &u[41], cospi32, &u[54], rnding, bit);
+  temp3 = half_btf_avx2(cospim32, &u[42], cospi32, &u[53], rnding, bit);
+  temp4 = half_btf_avx2(cospim32, &u[43], cospi32, &u[52], rnding, bit);
+  u[52] = half_btf_avx2(cospi32, &u[43], cospi32, &u[52], rnding, bit);
+  u[53] = half_btf_avx2(cospi32, &u[42], cospi32, &u[53], rnding, bit);
+  u[54] = half_btf_avx2(cospi32, &u[41], cospi32, &u[54], rnding, bit);
+  u[55] = half_btf_avx2(cospi32, &u[40], cospi32, &u[55], rnding, bit);
+  u[40] = temp1;
+  u[41] = temp2;
+  u[42] = temp3;
+  u[43] = temp4;
+
+  temp1 = half_btf_avx2(cospim32, &u[44], cospi32, &u[51], rnding, bit);
+  temp2 = half_btf_avx2(cospim32, &u[45], cospi32, &u[50], rnding, bit);
+  temp3 = half_btf_avx2(cospim32, &u[46], cospi32, &u[49], rnding, bit);
+  temp4 = half_btf_avx2(cospim32, &u[47], cospi32, &u[48], rnding, bit);
+  u[48] = half_btf_avx2(cospi32, &u[47], cospi32, &u[48], rnding, bit);
+  u[49] = half_btf_avx2(cospi32, &u[46], cospi32, &u[49], rnding, bit);
+  u[50] = half_btf_avx2(cospi32, &u[45], cospi32, &u[50], rnding, bit);
+  u[51] = half_btf_avx2(cospi32, &u[44], cospi32, &u[51], rnding, bit);
+  u[44] = temp1;
+  u[45] = temp2;
+  u[46] = temp3;
+  u[47] = temp4;
+}
+
+static INLINE void idct64_stage11_avx2(__m256i *u, __m256i *out, int do_cols,
+                                       int bd, int out_shift,
+                                       const int log_range) {
+  if (do_cols) {
+    for (int i = 0; i < 32; i++) {
+      addsub_no_clamp_avx2(u[i], u[63 - i], &out[(i)], &out[(63 - i)]);
+    }
+  } else {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+    for (int i = 0; i < 32; i++) {
+      addsub_shift_avx2(u[i], u[63 - i], &out[(i)], &out[(63 - i)],
+                        &clamp_lo_out, &clamp_hi_out, out_shift);
+    }
+  }
+}
+
+static void idct64_low1_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                             int bd, int out_shift) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+
+  {
+    __m256i x;
+
+    // stage 1
+    // stage 2
+    // stage 3
+    // stage 4
+    // stage 5
+    // stage 6
+    x = half_btf_0_avx2(&cospi32, &in[0], &rnding, bit);
+
+    // stage 8
+    // stage 9
+    // stage 10
+    // stage 11
+    if (do_cols) {
+      x = _mm256_max_epi32(x, clamp_lo);
+      x = _mm256_min_epi32(x, clamp_hi);
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+          -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+      const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+          (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+      __m256i offset = _mm256_set1_epi32((1 << out_shift) >> 1);
+      x = _mm256_add_epi32(x, offset);
+      x = _mm256_sra_epi32(x, _mm_cvtsi32_si128(out_shift));
+
+      x = _mm256_max_epi32(x, clamp_lo_out);
+      x = _mm256_min_epi32(x, clamp_hi_out);
+    }
+
+    out[0] = x;
+    out[1] = x;
+    out[2] = x;
+    out[3] = x;
+    out[4] = x;
+    out[5] = x;
+    out[6] = x;
+    out[7] = x;
+    out[8] = x;
+    out[9] = x;
+    out[10] = x;
+    out[11] = x;
+    out[12] = x;
+    out[13] = x;
+    out[14] = x;
+    out[15] = x;
+    out[16] = x;
+    out[17] = x;
+    out[18] = x;
+    out[19] = x;
+    out[20] = x;
+    out[21] = x;
+    out[22] = x;
+    out[23] = x;
+    out[24] = x;
+    out[25] = x;
+    out[26] = x;
+    out[27] = x;
+    out[28] = x;
+    out[29] = x;
+    out[30] = x;
+    out[31] = x;
+    out[32] = x;
+    out[33] = x;
+    out[34] = x;
+    out[35] = x;
+    out[36] = x;
+    out[37] = x;
+    out[38] = x;
+    out[39] = x;
+    out[40] = x;
+    out[41] = x;
+    out[42] = x;
+    out[43] = x;
+    out[44] = x;
+    out[45] = x;
+    out[46] = x;
+    out[47] = x;
+    out[48] = x;
+    out[49] = x;
+    out[50] = x;
+    out[51] = x;
+    out[52] = x;
+    out[53] = x;
+    out[54] = x;
+    out[55] = x;
+    out[56] = x;
+    out[57] = x;
+    out[58] = x;
+    out[59] = x;
+    out[60] = x;
+    out[61] = x;
+    out[62] = x;
+    out[63] = x;
+  }
+}
+static void idct64_low8_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                             int bd, int out_shift) {
+  int i, j;
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+
+  const __m256i cospi1 = _mm256_set1_epi32(cospi[1]);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi3 = _mm256_set1_epi32(cospi[3]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospim4 = _mm256_set1_epi32(-cospi[4]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospim12 = _mm256_set1_epi32(-cospi[12]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim20 = _mm256_set1_epi32(-cospi[20]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospim28 = _mm256_set1_epi32(-cospi[28]);
+  const __m256i cospim32 = _mm256_set1_epi32(-cospi[32]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospi63 = _mm256_set1_epi32(cospi[63]);
+  const __m256i cospim57 = _mm256_set1_epi32(-cospi[57]);
+  const __m256i cospi7 = _mm256_set1_epi32(cospi[7]);
+  const __m256i cospi5 = _mm256_set1_epi32(cospi[5]);
+  const __m256i cospi59 = _mm256_set1_epi32(cospi[59]);
+  const __m256i cospim61 = _mm256_set1_epi32(-cospi[61]);
+  const __m256i cospim58 = _mm256_set1_epi32(-cospi[58]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+
+  {
+    __m256i u[64];
+
+    // stage 1
+    u[0] = in[0];
+    u[8] = in[4];
+    u[16] = in[2];
+    u[24] = in[6];
+    u[32] = in[1];
+    u[40] = in[5];
+    u[48] = in[3];
+    u[56] = in[7];
+
+    // stage 2
+    u[63] = half_btf_0_avx2(&cospi1, &u[32], &rnding, bit);
+    u[32] = half_btf_0_avx2(&cospi63, &u[32], &rnding, bit);
+    u[39] = half_btf_0_avx2(&cospim57, &u[56], &rnding, bit);
+    u[56] = half_btf_0_avx2(&cospi7, &u[56], &rnding, bit);
+    u[55] = half_btf_0_avx2(&cospi5, &u[40], &rnding, bit);
+    u[40] = half_btf_0_avx2(&cospi59, &u[40], &rnding, bit);
+    u[47] = half_btf_0_avx2(&cospim61, &u[48], &rnding, bit);
+    u[48] = half_btf_0_avx2(&cospi3, &u[48], &rnding, bit);
+
+    // stage 3
+    u[31] = half_btf_0_avx2(&cospi2, &u[16], &rnding, bit);
+    u[16] = half_btf_0_avx2(&cospi62, &u[16], &rnding, bit);
+    u[23] = half_btf_0_avx2(&cospim58, &u[24], &rnding, bit);
+    u[24] = half_btf_0_avx2(&cospi6, &u[24], &rnding, bit);
+    u[33] = u[32];
+    u[38] = u[39];
+    u[41] = u[40];
+    u[46] = u[47];
+    u[49] = u[48];
+    u[54] = u[55];
+    u[57] = u[56];
+    u[62] = u[63];
+
+    // stage 4
+    __m256i temp1, temp2;
+    u[15] = half_btf_0_avx2(&cospi4, &u[8], &rnding, bit);
+    u[8] = half_btf_0_avx2(&cospi60, &u[8], &rnding, bit);
+    u[17] = u[16];
+    u[22] = u[23];
+    u[25] = u[24];
+    u[30] = u[31];
+
+    temp1 = half_btf_avx2(&cospim4, &u[33], &cospi60, &u[62], &rnding, bit);
+    u[62] = half_btf_avx2(&cospi60, &u[33], &cospi4, &u[62], &rnding, bit);
+    u[33] = temp1;
+
+    temp2 = half_btf_avx2(&cospim36, &u[38], &cospi28, &u[57], &rnding, bit);
+    u[38] = half_btf_avx2(&cospim28, &u[38], &cospim36, &u[57], &rnding, bit);
+    u[57] = temp2;
+
+    temp1 = half_btf_avx2(&cospim20, &u[41], &cospi44, &u[54], &rnding, bit);
+    u[54] = half_btf_avx2(&cospi44, &u[41], &cospi20, &u[54], &rnding, bit);
+    u[41] = temp1;
+
+    temp2 = half_btf_avx2(&cospim12, &u[46], &cospim52, &u[49], &rnding, bit);
+    u[49] = half_btf_avx2(&cospim52, &u[46], &cospi12, &u[49], &rnding, bit);
+    u[46] = temp2;
+
+    // stage 5
+    u[9] = u[8];
+    u[14] = u[15];
+
+    temp1 = half_btf_avx2(&cospim8, &u[17], &cospi56, &u[30], &rnding, bit);
+    u[30] = half_btf_avx2(&cospi56, &u[17], &cospi8, &u[30], &rnding, bit);
+    u[17] = temp1;
+
+    temp2 = half_btf_avx2(&cospim24, &u[22], &cospim40, &u[25], &rnding, bit);
+    u[25] = half_btf_avx2(&cospim40, &u[22], &cospi24, &u[25], &rnding, bit);
+    u[22] = temp2;
+
+    u[35] = u[32];
+    u[34] = u[33];
+    u[36] = u[39];
+    u[37] = u[38];
+    u[43] = u[40];
+    u[42] = u[41];
+    u[44] = u[47];
+    u[45] = u[46];
+    u[51] = u[48];
+    u[50] = u[49];
+    u[52] = u[55];
+    u[53] = u[54];
+    u[59] = u[56];
+    u[58] = u[57];
+    u[60] = u[63];
+    u[61] = u[62];
+
+    // stage 6
+    temp1 = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    u[1] = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    u[0] = temp1;
+
+    temp2 = half_btf_avx2(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
+    u[14] = half_btf_avx2(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
+    u[9] = temp2;
+    u[19] = u[16];
+    u[18] = u[17];
+    u[20] = u[23];
+    u[21] = u[22];
+    u[27] = u[24];
+    u[26] = u[25];
+    u[28] = u[31];
+    u[29] = u[30];
+
+    temp1 = half_btf_avx2(&cospim8, &u[34], &cospi56, &u[61], &rnding, bit);
+    u[61] = half_btf_avx2(&cospi56, &u[34], &cospi8, &u[61], &rnding, bit);
+    u[34] = temp1;
+    temp2 = half_btf_avx2(&cospim8, &u[35], &cospi56, &u[60], &rnding, bit);
+    u[60] = half_btf_avx2(&cospi56, &u[35], &cospi8, &u[60], &rnding, bit);
+    u[35] = temp2;
+    temp1 = half_btf_avx2(&cospim56, &u[36], &cospim8, &u[59], &rnding, bit);
+    u[59] = half_btf_avx2(&cospim8, &u[36], &cospi56, &u[59], &rnding, bit);
+    u[36] = temp1;
+    temp2 = half_btf_avx2(&cospim56, &u[37], &cospim8, &u[58], &rnding, bit);
+    u[58] = half_btf_avx2(&cospim8, &u[37], &cospi56, &u[58], &rnding, bit);
+    u[37] = temp2;
+    temp1 = half_btf_avx2(&cospim40, &u[42], &cospi24, &u[53], &rnding, bit);
+    u[53] = half_btf_avx2(&cospi24, &u[42], &cospi40, &u[53], &rnding, bit);
+    u[42] = temp1;
+    temp2 = half_btf_avx2(&cospim40, &u[43], &cospi24, &u[52], &rnding, bit);
+    u[52] = half_btf_avx2(&cospi24, &u[43], &cospi40, &u[52], &rnding, bit);
+    u[43] = temp2;
+    temp1 = half_btf_avx2(&cospim24, &u[44], &cospim40, &u[51], &rnding, bit);
+    u[51] = half_btf_avx2(&cospim40, &u[44], &cospi24, &u[51], &rnding, bit);
+    u[44] = temp1;
+    temp2 = half_btf_avx2(&cospim24, &u[45], &cospim40, &u[50], &rnding, bit);
+    u[50] = half_btf_avx2(&cospim40, &u[45], &cospi24, &u[50], &rnding, bit);
+    u[45] = temp2;
+
+    // stage 7
+    u[3] = u[0];
+    u[2] = u[1];
+    u[11] = u[8];
+    u[10] = u[9];
+    u[12] = u[15];
+    u[13] = u[14];
+
+    temp1 = half_btf_avx2(&cospim16, &u[18], &cospi48, &u[29], &rnding, bit);
+    u[29] = half_btf_avx2(&cospi48, &u[18], &cospi16, &u[29], &rnding, bit);
+    u[18] = temp1;
+    temp2 = half_btf_avx2(&cospim16, &u[19], &cospi48, &u[28], &rnding, bit);
+    u[28] = half_btf_avx2(&cospi48, &u[19], &cospi16, &u[28], &rnding, bit);
+    u[19] = temp2;
+    temp1 = half_btf_avx2(&cospim48, &u[20], &cospim16, &u[27], &rnding, bit);
+    u[27] = half_btf_avx2(&cospim16, &u[20], &cospi48, &u[27], &rnding, bit);
+    u[20] = temp1;
+    temp2 = half_btf_avx2(&cospim48, &u[21], &cospim16, &u[26], &rnding, bit);
+    u[26] = half_btf_avx2(&cospim16, &u[21], &cospi48, &u[26], &rnding, bit);
+    u[21] = temp2;
+    for (i = 32; i < 64; i += 16) {
+      for (j = i; j < i + 4; j++) {
+        addsub_avx2(u[j], u[j ^ 7], &u[j], &u[j ^ 7], &clamp_lo, &clamp_hi);
+        addsub_avx2(u[j ^ 15], u[j ^ 8], &u[j ^ 15], &u[j ^ 8], &clamp_lo,
+                    &clamp_hi);
+      }
+    }
+
+    // stage 8
+    u[7] = u[0];
+    u[6] = u[1];
+    u[5] = u[2];
+    u[4] = u[3];
+    u[9] = u[9];
+
+    idct64_stage8_avx2(u, &cospim32, &cospi32, &cospim16, &cospi48, &cospi16,
+                       &cospim48, &clamp_lo, &clamp_hi, &rnding, bit);
+
+    // stage 9
+    idct64_stage9_avx2(u, &cospim32, &cospi32, &clamp_lo, &clamp_hi, &rnding,
+                       bit);
+
+    // stage 10
+    idct64_stage10_avx2(u, &cospim32, &cospi32, &clamp_lo, &clamp_hi, &rnding,
+                        bit);
+
+    // stage 11
+    idct64_stage11_avx2(u, out, do_cols, bd, out_shift, log_range);
+  }
+}
+static void idct64_low16_avx2(__m256i *in, __m256i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  int i, j;
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+
+  const __m256i cospi1 = _mm256_set1_epi32(cospi[1]);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi3 = _mm256_set1_epi32(cospi[3]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi5 = _mm256_set1_epi32(cospi[5]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospi7 = _mm256_set1_epi32(cospi[7]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi9 = _mm256_set1_epi32(cospi[9]);
+  const __m256i cospi10 = _mm256_set1_epi32(cospi[10]);
+  const __m256i cospi11 = _mm256_set1_epi32(cospi[11]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi13 = _mm256_set1_epi32(cospi[13]);
+  const __m256i cospi14 = _mm256_set1_epi32(cospi[14]);
+  const __m256i cospi15 = _mm256_set1_epi32(cospi[15]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi51 = _mm256_set1_epi32(cospi[51]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i cospi54 = _mm256_set1_epi32(cospi[54]);
+  const __m256i cospi55 = _mm256_set1_epi32(cospi[55]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi59 = _mm256_set1_epi32(cospi[59]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospi63 = _mm256_set1_epi32(cospi[63]);
+
+  const __m256i cospim4 = _mm256_set1_epi32(-cospi[4]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospim12 = _mm256_set1_epi32(-cospi[12]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim20 = _mm256_set1_epi32(-cospi[20]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospim28 = _mm256_set1_epi32(-cospi[28]);
+  const __m256i cospim32 = _mm256_set1_epi32(-cospi[32]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospim44 = _mm256_set1_epi32(-cospi[44]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospim49 = _mm256_set1_epi32(-cospi[49]);
+  const __m256i cospim50 = _mm256_set1_epi32(-cospi[50]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i cospim53 = _mm256_set1_epi32(-cospi[53]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospim57 = _mm256_set1_epi32(-cospi[57]);
+  const __m256i cospim58 = _mm256_set1_epi32(-cospi[58]);
+  const __m256i cospim60 = _mm256_set1_epi32(-cospi[60]);
+  const __m256i cospim61 = _mm256_set1_epi32(-cospi[61]);
+
+  {
+    __m256i u[64];
+    __m256i tmp1, tmp2, tmp3, tmp4;
+    // stage 1
+    u[0] = in[0];
+    u[32] = in[1];
+    u[36] = in[9];
+    u[40] = in[5];
+    u[44] = in[13];
+    u[48] = in[3];
+    u[52] = in[11];
+    u[56] = in[7];
+    u[60] = in[15];
+    u[16] = in[2];
+    u[20] = in[10];
+    u[24] = in[6];
+    u[28] = in[14];
+    u[4] = in[8];
+    u[8] = in[4];
+    u[12] = in[12];
+
+    // stage 2
+    u[63] = half_btf_0_avx2(&cospi1, &u[32], &rnding, bit);
+    u[32] = half_btf_0_avx2(&cospi63, &u[32], &rnding, bit);
+    u[35] = half_btf_0_avx2(&cospim49, &u[60], &rnding, bit);
+    u[60] = half_btf_0_avx2(&cospi15, &u[60], &rnding, bit);
+    u[59] = half_btf_0_avx2(&cospi9, &u[36], &rnding, bit);
+    u[36] = half_btf_0_avx2(&cospi55, &u[36], &rnding, bit);
+    u[39] = half_btf_0_avx2(&cospim57, &u[56], &rnding, bit);
+    u[56] = half_btf_0_avx2(&cospi7, &u[56], &rnding, bit);
+    u[55] = half_btf_0_avx2(&cospi5, &u[40], &rnding, bit);
+    u[40] = half_btf_0_avx2(&cospi59, &u[40], &rnding, bit);
+    u[43] = half_btf_0_avx2(&cospim53, &u[52], &rnding, bit);
+    u[52] = half_btf_0_avx2(&cospi11, &u[52], &rnding, bit);
+    u[47] = half_btf_0_avx2(&cospim61, &u[48], &rnding, bit);
+    u[48] = half_btf_0_avx2(&cospi3, &u[48], &rnding, bit);
+    u[51] = half_btf_0_avx2(&cospi13, &u[44], &rnding, bit);
+    u[44] = half_btf_0_avx2(&cospi51, &u[44], &rnding, bit);
+
+    // stage 3
+    u[31] = half_btf_0_avx2(&cospi2, &u[16], &rnding, bit);
+    u[16] = half_btf_0_avx2(&cospi62, &u[16], &rnding, bit);
+    u[19] = half_btf_0_avx2(&cospim50, &u[28], &rnding, bit);
+    u[28] = half_btf_0_avx2(&cospi14, &u[28], &rnding, bit);
+    u[27] = half_btf_0_avx2(&cospi10, &u[20], &rnding, bit);
+    u[20] = half_btf_0_avx2(&cospi54, &u[20], &rnding, bit);
+    u[23] = half_btf_0_avx2(&cospim58, &u[24], &rnding, bit);
+    u[24] = half_btf_0_avx2(&cospi6, &u[24], &rnding, bit);
+    u[33] = u[32];
+    u[34] = u[35];
+    u[37] = u[36];
+    u[38] = u[39];
+    u[41] = u[40];
+    u[42] = u[43];
+    u[45] = u[44];
+    u[46] = u[47];
+    u[49] = u[48];
+    u[50] = u[51];
+    u[53] = u[52];
+    u[54] = u[55];
+    u[57] = u[56];
+    u[58] = u[59];
+    u[61] = u[60];
+    u[62] = u[63];
+
+    // stage 4
+    u[15] = half_btf_0_avx2(&cospi4, &u[8], &rnding, bit);
+    u[8] = half_btf_0_avx2(&cospi60, &u[8], &rnding, bit);
+    u[11] = half_btf_0_avx2(&cospim52, &u[12], &rnding, bit);
+    u[12] = half_btf_0_avx2(&cospi12, &u[12], &rnding, bit);
+
+    u[17] = u[16];
+    u[18] = u[19];
+    u[21] = u[20];
+    u[22] = u[23];
+    u[25] = u[24];
+    u[26] = u[27];
+    u[29] = u[28];
+    u[30] = u[31];
+
+    tmp1 = half_btf_avx2(&cospim4, &u[33], &cospi60, &u[62], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim60, &u[34], &cospim4, &u[61], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim36, &u[37], &cospi28, &u[58], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim28, &u[38], &cospim36, &u[57], &rnding, bit);
+    u[57] = half_btf_avx2(&cospim36, &u[38], &cospi28, &u[57], &rnding, bit);
+    u[58] = half_btf_avx2(&cospi28, &u[37], &cospi36, &u[58], &rnding, bit);
+    u[61] = half_btf_avx2(&cospim4, &u[34], &cospi60, &u[61], &rnding, bit);
+    u[62] = half_btf_avx2(&cospi60, &u[33], &cospi4, &u[62], &rnding, bit);
+    u[33] = tmp1;
+    u[34] = tmp2;
+    u[37] = tmp3;
+    u[38] = tmp4;
+
+    tmp1 = half_btf_avx2(&cospim20, &u[41], &cospi44, &u[54], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim44, &u[42], &cospim20, &u[53], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim52, &u[45], &cospi12, &u[50], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim12, &u[46], &cospim52, &u[49], &rnding, bit);
+    u[49] = half_btf_avx2(&cospim52, &u[46], &cospi12, &u[49], &rnding, bit);
+    u[50] = half_btf_avx2(&cospi12, &u[45], &cospi52, &u[50], &rnding, bit);
+    u[53] = half_btf_avx2(&cospim20, &u[42], &cospi44, &u[53], &rnding, bit);
+    u[54] = half_btf_avx2(&cospi44, &u[41], &cospi20, &u[54], &rnding, bit);
+    u[41] = tmp1;
+    u[42] = tmp2;
+    u[45] = tmp3;
+    u[46] = tmp4;
+
+    // stage 5
+    u[7] = half_btf_0_avx2(&cospi8, &u[4], &rnding, bit);
+    u[4] = half_btf_0_avx2(&cospi56, &u[4], &rnding, bit);
+
+    u[9] = u[8];
+    u[10] = u[11];
+    u[13] = u[12];
+    u[14] = u[15];
+
+    tmp1 = half_btf_avx2(&cospim8, &u[17], &cospi56, &u[30], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim56, &u[18], &cospim8, &u[29], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim40, &u[21], &cospi24, &u[26], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim24, &u[22], &cospim40, &u[25], &rnding, bit);
+    u[25] = half_btf_avx2(&cospim40, &u[22], &cospi24, &u[25], &rnding, bit);
+    u[26] = half_btf_avx2(&cospi24, &u[21], &cospi40, &u[26], &rnding, bit);
+    u[29] = half_btf_avx2(&cospim8, &u[18], &cospi56, &u[29], &rnding, bit);
+    u[30] = half_btf_avx2(&cospi56, &u[17], &cospi8, &u[30], &rnding, bit);
+    u[17] = tmp1;
+    u[18] = tmp2;
+    u[21] = tmp3;
+    u[22] = tmp4;
+
+    for (i = 32; i < 64; i += 8) {
+      addsub_avx2(u[i + 0], u[i + 3], &u[i + 0], &u[i + 3], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 1], u[i + 2], &u[i + 1], &u[i + 2], &clamp_lo,
+                  &clamp_hi);
+
+      addsub_avx2(u[i + 7], u[i + 4], &u[i + 7], &u[i + 4], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 6], u[i + 5], &u[i + 6], &u[i + 5], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    // stage 6
+    tmp1 = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    u[1] = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    u[0] = tmp1;
+    u[5] = u[4];
+    u[6] = u[7];
+
+    tmp1 = half_btf_avx2(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
+    u[14] = half_btf_avx2(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
+    u[9] = tmp1;
+    tmp2 = half_btf_avx2(&cospim48, &u[10], &cospim16, &u[13], &rnding, bit);
+    u[13] = half_btf_avx2(&cospim16, &u[10], &cospi48, &u[13], &rnding, bit);
+    u[10] = tmp2;
+
+    for (i = 16; i < 32; i += 8) {
+      addsub_avx2(u[i + 0], u[i + 3], &u[i + 0], &u[i + 3], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 1], u[i + 2], &u[i + 1], &u[i + 2], &clamp_lo,
+                  &clamp_hi);
+
+      addsub_avx2(u[i + 7], u[i + 4], &u[i + 7], &u[i + 4], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 6], u[i + 5], &u[i + 6], &u[i + 5], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    tmp1 = half_btf_avx2(&cospim8, &u[34], &cospi56, &u[61], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim8, &u[35], &cospi56, &u[60], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim56, &u[36], &cospim8, &u[59], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim56, &u[37], &cospim8, &u[58], &rnding, bit);
+    u[58] = half_btf_avx2(&cospim8, &u[37], &cospi56, &u[58], &rnding, bit);
+    u[59] = half_btf_avx2(&cospim8, &u[36], &cospi56, &u[59], &rnding, bit);
+    u[60] = half_btf_avx2(&cospi56, &u[35], &cospi8, &u[60], &rnding, bit);
+    u[61] = half_btf_avx2(&cospi56, &u[34], &cospi8, &u[61], &rnding, bit);
+    u[34] = tmp1;
+    u[35] = tmp2;
+    u[36] = tmp3;
+    u[37] = tmp4;
+
+    tmp1 = half_btf_avx2(&cospim40, &u[42], &cospi24, &u[53], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim40, &u[43], &cospi24, &u[52], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim24, &u[44], &cospim40, &u[51], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim24, &u[45], &cospim40, &u[50], &rnding, bit);
+    u[50] = half_btf_avx2(&cospim40, &u[45], &cospi24, &u[50], &rnding, bit);
+    u[51] = half_btf_avx2(&cospim40, &u[44], &cospi24, &u[51], &rnding, bit);
+    u[52] = half_btf_avx2(&cospi24, &u[43], &cospi40, &u[52], &rnding, bit);
+    u[53] = half_btf_avx2(&cospi24, &u[42], &cospi40, &u[53], &rnding, bit);
+    u[42] = tmp1;
+    u[43] = tmp2;
+    u[44] = tmp3;
+    u[45] = tmp4;
+
+    // stage 7
+    u[3] = u[0];
+    u[2] = u[1];
+    tmp1 = half_btf_avx2(&cospim32, &u[5], &cospi32, &u[6], &rnding, bit);
+    u[6] = half_btf_avx2(&cospi32, &u[5], &cospi32, &u[6], &rnding, bit);
+    u[5] = tmp1;
+    addsub_avx2(u[8], u[11], &u[8], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[9], u[10], &u[9], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[15], u[12], &u[15], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[14], u[13], &u[14], &u[13], &clamp_lo, &clamp_hi);
+
+    tmp1 = half_btf_avx2(&cospim16, &u[18], &cospi48, &u[29], &rnding, bit);
+    tmp2 = half_btf_avx2(&cospim16, &u[19], &cospi48, &u[28], &rnding, bit);
+    tmp3 = half_btf_avx2(&cospim48, &u[20], &cospim16, &u[27], &rnding, bit);
+    tmp4 = half_btf_avx2(&cospim48, &u[21], &cospim16, &u[26], &rnding, bit);
+    u[26] = half_btf_avx2(&cospim16, &u[21], &cospi48, &u[26], &rnding, bit);
+    u[27] = half_btf_avx2(&cospim16, &u[20], &cospi48, &u[27], &rnding, bit);
+    u[28] = half_btf_avx2(&cospi48, &u[19], &cospi16, &u[28], &rnding, bit);
+    u[29] = half_btf_avx2(&cospi48, &u[18], &cospi16, &u[29], &rnding, bit);
+    u[18] = tmp1;
+    u[19] = tmp2;
+    u[20] = tmp3;
+    u[21] = tmp4;
+
+    for (i = 32; i < 64; i += 16) {
+      for (j = i; j < i + 4; j++) {
+        addsub_avx2(u[j], u[j ^ 7], &u[j], &u[j ^ 7], &clamp_lo, &clamp_hi);
+        addsub_avx2(u[j ^ 15], u[j ^ 8], &u[j ^ 15], &u[j ^ 8], &clamp_lo,
+                    &clamp_hi);
+      }
+    }
+
+    // stage 8
+    for (i = 0; i < 4; ++i) {
+      addsub_avx2(u[i], u[7 - i], &u[i], &u[7 - i], &clamp_lo, &clamp_hi);
+    }
+
+    idct64_stage8_avx2(u, &cospim32, &cospi32, &cospim16, &cospi48, &cospi16,
+                       &cospim48, &clamp_lo, &clamp_hi, &rnding, bit);
+
+    // stage 9
+    idct64_stage9_avx2(u, &cospim32, &cospi32, &clamp_lo, &clamp_hi, &rnding,
+                       bit);
+
+    // stage 10
+    idct64_stage10_avx2(u, &cospim32, &cospi32, &clamp_lo, &clamp_hi, &rnding,
+                        bit);
+
+    // stage 11
+    idct64_stage11_avx2(u, out, do_cols, bd, out_shift, log_range);
+  }
+}
+static void idct64_avx2(__m256i *in, __m256i *out, int bit, int do_cols, int bd,
+                        int out_shift) {
+  int i, j;
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m256i clamp_lo = _mm256_set1_epi32(-(1 << (log_range - 1)));
+  const __m256i clamp_hi = _mm256_set1_epi32((1 << (log_range - 1)) - 1);
+
+  const __m256i cospi1 = _mm256_set1_epi32(cospi[1]);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi3 = _mm256_set1_epi32(cospi[3]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi5 = _mm256_set1_epi32(cospi[5]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospi7 = _mm256_set1_epi32(cospi[7]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi9 = _mm256_set1_epi32(cospi[9]);
+  const __m256i cospi10 = _mm256_set1_epi32(cospi[10]);
+  const __m256i cospi11 = _mm256_set1_epi32(cospi[11]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi13 = _mm256_set1_epi32(cospi[13]);
+  const __m256i cospi14 = _mm256_set1_epi32(cospi[14]);
+  const __m256i cospi15 = _mm256_set1_epi32(cospi[15]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi17 = _mm256_set1_epi32(cospi[17]);
+  const __m256i cospi18 = _mm256_set1_epi32(cospi[18]);
+  const __m256i cospi19 = _mm256_set1_epi32(cospi[19]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi21 = _mm256_set1_epi32(cospi[21]);
+  const __m256i cospi22 = _mm256_set1_epi32(cospi[22]);
+  const __m256i cospi23 = _mm256_set1_epi32(cospi[23]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospi25 = _mm256_set1_epi32(cospi[25]);
+  const __m256i cospi26 = _mm256_set1_epi32(cospi[26]);
+  const __m256i cospi27 = _mm256_set1_epi32(cospi[27]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi29 = _mm256_set1_epi32(cospi[29]);
+  const __m256i cospi30 = _mm256_set1_epi32(cospi[30]);
+  const __m256i cospi31 = _mm256_set1_epi32(cospi[31]);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi35 = _mm256_set1_epi32(cospi[35]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospi38 = _mm256_set1_epi32(cospi[38]);
+  const __m256i cospi39 = _mm256_set1_epi32(cospi[39]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi43 = _mm256_set1_epi32(cospi[43]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi46 = _mm256_set1_epi32(cospi[46]);
+  const __m256i cospi47 = _mm256_set1_epi32(cospi[47]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi51 = _mm256_set1_epi32(cospi[51]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i cospi54 = _mm256_set1_epi32(cospi[54]);
+  const __m256i cospi55 = _mm256_set1_epi32(cospi[55]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi59 = _mm256_set1_epi32(cospi[59]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospi63 = _mm256_set1_epi32(cospi[63]);
+
+  const __m256i cospim4 = _mm256_set1_epi32(-cospi[4]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospim12 = _mm256_set1_epi32(-cospi[12]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim20 = _mm256_set1_epi32(-cospi[20]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospim28 = _mm256_set1_epi32(-cospi[28]);
+  const __m256i cospim32 = _mm256_set1_epi32(-cospi[32]);
+  const __m256i cospim33 = _mm256_set1_epi32(-cospi[33]);
+  const __m256i cospim34 = _mm256_set1_epi32(-cospi[34]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospim37 = _mm256_set1_epi32(-cospi[37]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospim41 = _mm256_set1_epi32(-cospi[41]);
+  const __m256i cospim42 = _mm256_set1_epi32(-cospi[42]);
+  const __m256i cospim44 = _mm256_set1_epi32(-cospi[44]);
+  const __m256i cospim45 = _mm256_set1_epi32(-cospi[45]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospim49 = _mm256_set1_epi32(-cospi[49]);
+  const __m256i cospim50 = _mm256_set1_epi32(-cospi[50]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i cospim53 = _mm256_set1_epi32(-cospi[53]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospim57 = _mm256_set1_epi32(-cospi[57]);
+  const __m256i cospim58 = _mm256_set1_epi32(-cospi[58]);
+  const __m256i cospim60 = _mm256_set1_epi32(-cospi[60]);
+  const __m256i cospim61 = _mm256_set1_epi32(-cospi[61]);
+
+  {
+    __m256i u[64], v[64];
+
+    // stage 1
+    u[32] = in[1];
+    u[34] = in[17];
+    u[36] = in[9];
+    u[38] = in[25];
+    u[40] = in[5];
+    u[42] = in[21];
+    u[44] = in[13];
+    u[46] = in[29];
+    u[48] = in[3];
+    u[50] = in[19];
+    u[52] = in[11];
+    u[54] = in[27];
+    u[56] = in[7];
+    u[58] = in[23];
+    u[60] = in[15];
+    u[62] = in[31];
+
+    v[16] = in[2];
+    v[18] = in[18];
+    v[20] = in[10];
+    v[22] = in[26];
+    v[24] = in[6];
+    v[26] = in[22];
+    v[28] = in[14];
+    v[30] = in[30];
+
+    u[8] = in[4];
+    u[10] = in[20];
+    u[12] = in[12];
+    u[14] = in[28];
+
+    v[4] = in[8];
+    v[6] = in[24];
+
+    u[0] = in[0];
+    u[2] = in[16];
+
+    // stage 2
+    v[32] = half_btf_0_avx2(&cospi63, &u[32], &rnding, bit);
+    v[33] = half_btf_0_avx2(&cospim33, &u[62], &rnding, bit);
+    v[34] = half_btf_0_avx2(&cospi47, &u[34], &rnding, bit);
+    v[35] = half_btf_0_avx2(&cospim49, &u[60], &rnding, bit);
+    v[36] = half_btf_0_avx2(&cospi55, &u[36], &rnding, bit);
+    v[37] = half_btf_0_avx2(&cospim41, &u[58], &rnding, bit);
+    v[38] = half_btf_0_avx2(&cospi39, &u[38], &rnding, bit);
+    v[39] = half_btf_0_avx2(&cospim57, &u[56], &rnding, bit);
+    v[40] = half_btf_0_avx2(&cospi59, &u[40], &rnding, bit);
+    v[41] = half_btf_0_avx2(&cospim37, &u[54], &rnding, bit);
+    v[42] = half_btf_0_avx2(&cospi43, &u[42], &rnding, bit);
+    v[43] = half_btf_0_avx2(&cospim53, &u[52], &rnding, bit);
+    v[44] = half_btf_0_avx2(&cospi51, &u[44], &rnding, bit);
+    v[45] = half_btf_0_avx2(&cospim45, &u[50], &rnding, bit);
+    v[46] = half_btf_0_avx2(&cospi35, &u[46], &rnding, bit);
+    v[47] = half_btf_0_avx2(&cospim61, &u[48], &rnding, bit);
+    v[48] = half_btf_0_avx2(&cospi3, &u[48], &rnding, bit);
+    v[49] = half_btf_0_avx2(&cospi29, &u[46], &rnding, bit);
+    v[50] = half_btf_0_avx2(&cospi19, &u[50], &rnding, bit);
+    v[51] = half_btf_0_avx2(&cospi13, &u[44], &rnding, bit);
+    v[52] = half_btf_0_avx2(&cospi11, &u[52], &rnding, bit);
+    v[53] = half_btf_0_avx2(&cospi21, &u[42], &rnding, bit);
+    v[54] = half_btf_0_avx2(&cospi27, &u[54], &rnding, bit);
+    v[55] = half_btf_0_avx2(&cospi5, &u[40], &rnding, bit);
+    v[56] = half_btf_0_avx2(&cospi7, &u[56], &rnding, bit);
+    v[57] = half_btf_0_avx2(&cospi25, &u[38], &rnding, bit);
+    v[58] = half_btf_0_avx2(&cospi23, &u[58], &rnding, bit);
+    v[59] = half_btf_0_avx2(&cospi9, &u[36], &rnding, bit);
+    v[60] = half_btf_0_avx2(&cospi15, &u[60], &rnding, bit);
+    v[61] = half_btf_0_avx2(&cospi17, &u[34], &rnding, bit);
+    v[62] = half_btf_0_avx2(&cospi31, &u[62], &rnding, bit);
+    v[63] = half_btf_0_avx2(&cospi1, &u[32], &rnding, bit);
+
+    // stage 3
+    u[16] = half_btf_0_avx2(&cospi62, &v[16], &rnding, bit);
+    u[17] = half_btf_0_avx2(&cospim34, &v[30], &rnding, bit);
+    u[18] = half_btf_0_avx2(&cospi46, &v[18], &rnding, bit);
+    u[19] = half_btf_0_avx2(&cospim50, &v[28], &rnding, bit);
+    u[20] = half_btf_0_avx2(&cospi54, &v[20], &rnding, bit);
+    u[21] = half_btf_0_avx2(&cospim42, &v[26], &rnding, bit);
+    u[22] = half_btf_0_avx2(&cospi38, &v[22], &rnding, bit);
+    u[23] = half_btf_0_avx2(&cospim58, &v[24], &rnding, bit);
+    u[24] = half_btf_0_avx2(&cospi6, &v[24], &rnding, bit);
+    u[25] = half_btf_0_avx2(&cospi26, &v[22], &rnding, bit);
+    u[26] = half_btf_0_avx2(&cospi22, &v[26], &rnding, bit);
+    u[27] = half_btf_0_avx2(&cospi10, &v[20], &rnding, bit);
+    u[28] = half_btf_0_avx2(&cospi14, &v[28], &rnding, bit);
+    u[29] = half_btf_0_avx2(&cospi18, &v[18], &rnding, bit);
+    u[30] = half_btf_0_avx2(&cospi30, &v[30], &rnding, bit);
+    u[31] = half_btf_0_avx2(&cospi2, &v[16], &rnding, bit);
+
+    for (i = 32; i < 64; i += 4) {
+      addsub_avx2(v[i + 0], v[i + 1], &u[i + 0], &u[i + 1], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(v[i + 3], v[i + 2], &u[i + 3], &u[i + 2], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    // stage 4
+    v[8] = half_btf_0_avx2(&cospi60, &u[8], &rnding, bit);
+    v[9] = half_btf_0_avx2(&cospim36, &u[14], &rnding, bit);
+    v[10] = half_btf_0_avx2(&cospi44, &u[10], &rnding, bit);
+    v[11] = half_btf_0_avx2(&cospim52, &u[12], &rnding, bit);
+    v[12] = half_btf_0_avx2(&cospi12, &u[12], &rnding, bit);
+    v[13] = half_btf_0_avx2(&cospi20, &u[10], &rnding, bit);
+    v[14] = half_btf_0_avx2(&cospi28, &u[14], &rnding, bit);
+    v[15] = half_btf_0_avx2(&cospi4, &u[8], &rnding, bit);
+
+    for (i = 16; i < 32; i += 4) {
+      addsub_avx2(u[i + 0], u[i + 1], &v[i + 0], &v[i + 1], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 3], u[i + 2], &v[i + 3], &v[i + 2], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    for (i = 32; i < 64; i += 4) {
+      v[i + 0] = u[i + 0];
+      v[i + 3] = u[i + 3];
+    }
+
+    v[33] = half_btf_avx2(&cospim4, &u[33], &cospi60, &u[62], &rnding, bit);
+    v[34] = half_btf_avx2(&cospim60, &u[34], &cospim4, &u[61], &rnding, bit);
+    v[37] = half_btf_avx2(&cospim36, &u[37], &cospi28, &u[58], &rnding, bit);
+    v[38] = half_btf_avx2(&cospim28, &u[38], &cospim36, &u[57], &rnding, bit);
+    v[41] = half_btf_avx2(&cospim20, &u[41], &cospi44, &u[54], &rnding, bit);
+    v[42] = half_btf_avx2(&cospim44, &u[42], &cospim20, &u[53], &rnding, bit);
+    v[45] = half_btf_avx2(&cospim52, &u[45], &cospi12, &u[50], &rnding, bit);
+    v[46] = half_btf_avx2(&cospim12, &u[46], &cospim52, &u[49], &rnding, bit);
+    v[49] = half_btf_avx2(&cospim52, &u[46], &cospi12, &u[49], &rnding, bit);
+    v[50] = half_btf_avx2(&cospi12, &u[45], &cospi52, &u[50], &rnding, bit);
+    v[53] = half_btf_avx2(&cospim20, &u[42], &cospi44, &u[53], &rnding, bit);
+    v[54] = half_btf_avx2(&cospi44, &u[41], &cospi20, &u[54], &rnding, bit);
+    v[57] = half_btf_avx2(&cospim36, &u[38], &cospi28, &u[57], &rnding, bit);
+    v[58] = half_btf_avx2(&cospi28, &u[37], &cospi36, &u[58], &rnding, bit);
+    v[61] = half_btf_avx2(&cospim4, &u[34], &cospi60, &u[61], &rnding, bit);
+    v[62] = half_btf_avx2(&cospi60, &u[33], &cospi4, &u[62], &rnding, bit);
+
+    // stage 5
+    u[4] = half_btf_0_avx2(&cospi56, &v[4], &rnding, bit);
+    u[5] = half_btf_0_avx2(&cospim40, &v[6], &rnding, bit);
+    u[6] = half_btf_0_avx2(&cospi24, &v[6], &rnding, bit);
+    u[7] = half_btf_0_avx2(&cospi8, &v[4], &rnding, bit);
+
+    for (i = 8; i < 16; i += 4) {
+      addsub_avx2(v[i + 0], v[i + 1], &u[i + 0], &u[i + 1], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(v[i + 3], v[i + 2], &u[i + 3], &u[i + 2], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    for (i = 16; i < 32; i += 4) {
+      u[i + 0] = v[i + 0];
+      u[i + 3] = v[i + 3];
+    }
+
+    u[17] = half_btf_avx2(&cospim8, &v[17], &cospi56, &v[30], &rnding, bit);
+    u[18] = half_btf_avx2(&cospim56, &v[18], &cospim8, &v[29], &rnding, bit);
+    u[21] = half_btf_avx2(&cospim40, &v[21], &cospi24, &v[26], &rnding, bit);
+    u[22] = half_btf_avx2(&cospim24, &v[22], &cospim40, &v[25], &rnding, bit);
+    u[25] = half_btf_avx2(&cospim40, &v[22], &cospi24, &v[25], &rnding, bit);
+    u[26] = half_btf_avx2(&cospi24, &v[21], &cospi40, &v[26], &rnding, bit);
+    u[29] = half_btf_avx2(&cospim8, &v[18], &cospi56, &v[29], &rnding, bit);
+    u[30] = half_btf_avx2(&cospi56, &v[17], &cospi8, &v[30], &rnding, bit);
+
+    for (i = 32; i < 64; i += 8) {
+      addsub_avx2(v[i + 0], v[i + 3], &u[i + 0], &u[i + 3], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(v[i + 1], v[i + 2], &u[i + 1], &u[i + 2], &clamp_lo,
+                  &clamp_hi);
+
+      addsub_avx2(v[i + 7], v[i + 4], &u[i + 7], &u[i + 4], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(v[i + 6], v[i + 5], &u[i + 6], &u[i + 5], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    // stage 6
+    v[0] = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    v[1] = half_btf_0_avx2(&cospi32, &u[0], &rnding, bit);
+    v[2] = half_btf_0_avx2(&cospi48, &u[2], &rnding, bit);
+    v[3] = half_btf_0_avx2(&cospi16, &u[2], &rnding, bit);
+
+    addsub_avx2(u[4], u[5], &v[4], &v[5], &clamp_lo, &clamp_hi);
+    addsub_avx2(u[7], u[6], &v[7], &v[6], &clamp_lo, &clamp_hi);
+
+    for (i = 8; i < 16; i += 4) {
+      v[i + 0] = u[i + 0];
+      v[i + 3] = u[i + 3];
+    }
+
+    v[9] = half_btf_avx2(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
+    v[10] = half_btf_avx2(&cospim48, &u[10], &cospim16, &u[13], &rnding, bit);
+    v[13] = half_btf_avx2(&cospim16, &u[10], &cospi48, &u[13], &rnding, bit);
+    v[14] = half_btf_avx2(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
+
+    for (i = 16; i < 32; i += 8) {
+      addsub_avx2(u[i + 0], u[i + 3], &v[i + 0], &v[i + 3], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 1], u[i + 2], &v[i + 1], &v[i + 2], &clamp_lo,
+                  &clamp_hi);
+
+      addsub_avx2(u[i + 7], u[i + 4], &v[i + 7], &v[i + 4], &clamp_lo,
+                  &clamp_hi);
+      addsub_avx2(u[i + 6], u[i + 5], &v[i + 6], &v[i + 5], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    for (i = 32; i < 64; i += 8) {
+      v[i + 0] = u[i + 0];
+      v[i + 1] = u[i + 1];
+      v[i + 6] = u[i + 6];
+      v[i + 7] = u[i + 7];
+    }
+
+    v[34] = half_btf_avx2(&cospim8, &u[34], &cospi56, &u[61], &rnding, bit);
+    v[35] = half_btf_avx2(&cospim8, &u[35], &cospi56, &u[60], &rnding, bit);
+    v[36] = half_btf_avx2(&cospim56, &u[36], &cospim8, &u[59], &rnding, bit);
+    v[37] = half_btf_avx2(&cospim56, &u[37], &cospim8, &u[58], &rnding, bit);
+    v[42] = half_btf_avx2(&cospim40, &u[42], &cospi24, &u[53], &rnding, bit);
+    v[43] = half_btf_avx2(&cospim40, &u[43], &cospi24, &u[52], &rnding, bit);
+    v[44] = half_btf_avx2(&cospim24, &u[44], &cospim40, &u[51], &rnding, bit);
+    v[45] = half_btf_avx2(&cospim24, &u[45], &cospim40, &u[50], &rnding, bit);
+    v[50] = half_btf_avx2(&cospim40, &u[45], &cospi24, &u[50], &rnding, bit);
+    v[51] = half_btf_avx2(&cospim40, &u[44], &cospi24, &u[51], &rnding, bit);
+    v[52] = half_btf_avx2(&cospi24, &u[43], &cospi40, &u[52], &rnding, bit);
+    v[53] = half_btf_avx2(&cospi24, &u[42], &cospi40, &u[53], &rnding, bit);
+    v[58] = half_btf_avx2(&cospim8, &u[37], &cospi56, &u[58], &rnding, bit);
+    v[59] = half_btf_avx2(&cospim8, &u[36], &cospi56, &u[59], &rnding, bit);
+    v[60] = half_btf_avx2(&cospi56, &u[35], &cospi8, &u[60], &rnding, bit);
+    v[61] = half_btf_avx2(&cospi56, &u[34], &cospi8, &u[61], &rnding, bit);
+
+    // stage 7
+    addsub_avx2(v[0], v[3], &u[0], &u[3], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[1], v[2], &u[1], &u[2], &clamp_lo, &clamp_hi);
+
+    u[4] = v[4];
+    u[7] = v[7];
+    u[5] = half_btf_avx2(&cospim32, &v[5], &cospi32, &v[6], &rnding, bit);
+    u[6] = half_btf_avx2(&cospi32, &v[5], &cospi32, &v[6], &rnding, bit);
+
+    addsub_avx2(v[8], v[11], &u[8], &u[11], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[9], v[10], &u[9], &u[10], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[15], v[12], &u[15], &u[12], &clamp_lo, &clamp_hi);
+    addsub_avx2(v[14], v[13], &u[14], &u[13], &clamp_lo, &clamp_hi);
+
+    for (i = 16; i < 32; i += 8) {
+      u[i + 0] = v[i + 0];
+      u[i + 1] = v[i + 1];
+      u[i + 6] = v[i + 6];
+      u[i + 7] = v[i + 7];
+    }
+
+    u[18] = half_btf_avx2(&cospim16, &v[18], &cospi48, &v[29], &rnding, bit);
+    u[19] = half_btf_avx2(&cospim16, &v[19], &cospi48, &v[28], &rnding, bit);
+    u[20] = half_btf_avx2(&cospim48, &v[20], &cospim16, &v[27], &rnding, bit);
+    u[21] = half_btf_avx2(&cospim48, &v[21], &cospim16, &v[26], &rnding, bit);
+    u[26] = half_btf_avx2(&cospim16, &v[21], &cospi48, &v[26], &rnding, bit);
+    u[27] = half_btf_avx2(&cospim16, &v[20], &cospi48, &v[27], &rnding, bit);
+    u[28] = half_btf_avx2(&cospi48, &v[19], &cospi16, &v[28], &rnding, bit);
+    u[29] = half_btf_avx2(&cospi48, &v[18], &cospi16, &v[29], &rnding, bit);
+
+    for (i = 32; i < 64; i += 16) {
+      for (j = i; j < i + 4; j++) {
+        addsub_avx2(v[j], v[j ^ 7], &u[j], &u[j ^ 7], &clamp_lo, &clamp_hi);
+        addsub_avx2(v[j ^ 15], v[j ^ 8], &u[j ^ 15], &u[j ^ 8], &clamp_lo,
+                    &clamp_hi);
+      }
+    }
+
+    // stage 8
+    for (i = 0; i < 4; ++i) {
+      addsub_avx2(u[i], u[7 - i], &v[i], &v[7 - i], &clamp_lo, &clamp_hi);
+    }
+
+    v[8] = u[8];
+    v[9] = u[9];
+    v[14] = u[14];
+    v[15] = u[15];
+
+    v[10] = half_btf_avx2(&cospim32, &u[10], &cospi32, &u[13], &rnding, bit);
+    v[11] = half_btf_avx2(&cospim32, &u[11], &cospi32, &u[12], &rnding, bit);
+    v[12] = half_btf_avx2(&cospi32, &u[11], &cospi32, &u[12], &rnding, bit);
+    v[13] = half_btf_avx2(&cospi32, &u[10], &cospi32, &u[13], &rnding, bit);
+
+    for (i = 16; i < 20; ++i) {
+      addsub_avx2(u[i], u[i ^ 7], &v[i], &v[i ^ 7], &clamp_lo, &clamp_hi);
+      addsub_avx2(u[i ^ 15], u[i ^ 8], &v[i ^ 15], &v[i ^ 8], &clamp_lo,
+                  &clamp_hi);
+    }
+
+    for (i = 32; i < 36; ++i) {
+      v[i] = u[i];
+      v[i + 12] = u[i + 12];
+      v[i + 16] = u[i + 16];
+      v[i + 28] = u[i + 28];
+    }
+
+    v[36] = half_btf_avx2(&cospim16, &u[36], &cospi48, &u[59], &rnding, bit);
+    v[37] = half_btf_avx2(&cospim16, &u[37], &cospi48, &u[58], &rnding, bit);
+    v[38] = half_btf_avx2(&cospim16, &u[38], &cospi48, &u[57], &rnding, bit);
+    v[39] = half_btf_avx2(&cospim16, &u[39], &cospi48, &u[56], &rnding, bit);
+    v[40] = half_btf_avx2(&cospim48, &u[40], &cospim16, &u[55], &rnding, bit);
+    v[41] = half_btf_avx2(&cospim48, &u[41], &cospim16, &u[54], &rnding, bit);
+    v[42] = half_btf_avx2(&cospim48, &u[42], &cospim16, &u[53], &rnding, bit);
+    v[43] = half_btf_avx2(&cospim48, &u[43], &cospim16, &u[52], &rnding, bit);
+    v[52] = half_btf_avx2(&cospim16, &u[43], &cospi48, &u[52], &rnding, bit);
+    v[53] = half_btf_avx2(&cospim16, &u[42], &cospi48, &u[53], &rnding, bit);
+    v[54] = half_btf_avx2(&cospim16, &u[41], &cospi48, &u[54], &rnding, bit);
+    v[55] = half_btf_avx2(&cospim16, &u[40], &cospi48, &u[55], &rnding, bit);
+    v[56] = half_btf_avx2(&cospi48, &u[39], &cospi16, &u[56], &rnding, bit);
+    v[57] = half_btf_avx2(&cospi48, &u[38], &cospi16, &u[57], &rnding, bit);
+    v[58] = half_btf_avx2(&cospi48, &u[37], &cospi16, &u[58], &rnding, bit);
+    v[59] = half_btf_avx2(&cospi48, &u[36], &cospi16, &u[59], &rnding, bit);
+
+    // stage 9
+    for (i = 0; i < 8; ++i) {
+      addsub_avx2(v[i], v[15 - i], &u[i], &u[15 - i], &clamp_lo, &clamp_hi);
+    }
+
+    for (i = 16; i < 20; ++i) {
+      u[i] = v[i];
+      u[i + 12] = v[i + 12];
+    }
+
+    u[20] = half_btf_avx2(&cospim32, &v[20], &cospi32, &v[27], &rnding, bit);
+    u[21] = half_btf_avx2(&cospim32, &v[21], &cospi32, &v[26], &rnding, bit);
+    u[22] = half_btf_avx2(&cospim32, &v[22], &cospi32, &v[25], &rnding, bit);
+    u[23] = half_btf_avx2(&cospim32, &v[23], &cospi32, &v[24], &rnding, bit);
+    u[24] = half_btf_avx2(&cospi32, &v[23], &cospi32, &v[24], &rnding, bit);
+    u[25] = half_btf_avx2(&cospi32, &v[22], &cospi32, &v[25], &rnding, bit);
+    u[26] = half_btf_avx2(&cospi32, &v[21], &cospi32, &v[26], &rnding, bit);
+    u[27] = half_btf_avx2(&cospi32, &v[20], &cospi32, &v[27], &rnding, bit);
+
+    for (i = 32; i < 40; i++) {
+      addsub_avx2(v[i], v[i ^ 15], &u[i], &u[i ^ 15], &clamp_lo, &clamp_hi);
+    }
+
+    for (i = 48; i < 56; i++) {
+      addsub_avx2(v[i ^ 15], v[i], &u[i ^ 15], &u[i], &clamp_lo, &clamp_hi);
+    }
+
+    // stage 10
+    for (i = 0; i < 16; i++) {
+      addsub_avx2(u[i], u[31 - i], &v[i], &v[31 - i], &clamp_lo, &clamp_hi);
+    }
+
+    for (i = 32; i < 40; i++) v[i] = u[i];
+
+    v[40] = half_btf_avx2(&cospim32, &u[40], &cospi32, &u[55], &rnding, bit);
+    v[41] = half_btf_avx2(&cospim32, &u[41], &cospi32, &u[54], &rnding, bit);
+    v[42] = half_btf_avx2(&cospim32, &u[42], &cospi32, &u[53], &rnding, bit);
+    v[43] = half_btf_avx2(&cospim32, &u[43], &cospi32, &u[52], &rnding, bit);
+    v[44] = half_btf_avx2(&cospim32, &u[44], &cospi32, &u[51], &rnding, bit);
+    v[45] = half_btf_avx2(&cospim32, &u[45], &cospi32, &u[50], &rnding, bit);
+    v[46] = half_btf_avx2(&cospim32, &u[46], &cospi32, &u[49], &rnding, bit);
+    v[47] = half_btf_avx2(&cospim32, &u[47], &cospi32, &u[48], &rnding, bit);
+    v[48] = half_btf_avx2(&cospi32, &u[47], &cospi32, &u[48], &rnding, bit);
+    v[49] = half_btf_avx2(&cospi32, &u[46], &cospi32, &u[49], &rnding, bit);
+    v[50] = half_btf_avx2(&cospi32, &u[45], &cospi32, &u[50], &rnding, bit);
+    v[51] = half_btf_avx2(&cospi32, &u[44], &cospi32, &u[51], &rnding, bit);
+    v[52] = half_btf_avx2(&cospi32, &u[43], &cospi32, &u[52], &rnding, bit);
+    v[53] = half_btf_avx2(&cospi32, &u[42], &cospi32, &u[53], &rnding, bit);
+    v[54] = half_btf_avx2(&cospi32, &u[41], &cospi32, &u[54], &rnding, bit);
+    v[55] = half_btf_avx2(&cospi32, &u[40], &cospi32, &u[55], &rnding, bit);
+
+    for (i = 56; i < 64; i++) v[i] = u[i];
+
+    // stage 11
+    if (do_cols) {
+      for (i = 0; i < 32; i++) {
+        addsub_no_clamp_avx2(v[i], v[63 - i], &out[(i)], &out[(63 - i)]);
+      }
+    } else {
+      const int log_range_out = AOMMAX(16, bd + 6);
+      const __m256i clamp_lo_out = _mm256_set1_epi32(AOMMAX(
+          -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+      const __m256i clamp_hi_out = _mm256_set1_epi32(AOMMIN(
+          (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+      for (i = 0; i < 32; i++) {
+        addsub_shift_avx2(v[i], v[63 - i], &out[(i)], &out[(63 - i)],
+                          &clamp_lo_out, &clamp_hi_out, out_shift);
+      }
+    }
+  }
+}
 typedef void (*transform_1d_avx2)(__m256i *in, __m256i *out, int bit,
                                   int do_cols, int bd, int out_shift);
 
@@ -1164,19 +4192,21 @@ static const transform_1d_avx2
           { NULL, NULL, NULL, NULL },
           { NULL, NULL, NULL, NULL },
       },
-      { { NULL, NULL, NULL, NULL },
-        { NULL, NULL, NULL, NULL },
-        { NULL, NULL, NULL, NULL } },
       {
+          { idct8x8_low1_avx2, idct8x8_avx2, NULL, NULL },
+          { iadst8x8_low1_avx2, iadst8x8_avx2, NULL, NULL },
           { NULL, NULL, NULL, NULL },
-          { NULL, NULL, NULL, NULL },
+      },
+      {
+          { idct16_low1_avx2, idct16_low8_avx2, idct16_avx2, NULL },
+          { iadst16_low1_avx2, iadst16_low8_avx2, iadst16_avx2, NULL },
           { NULL, NULL, NULL, NULL },
       },
       { { idct32_low1_avx2, idct32_low8_avx2, idct32_low16_avx2, idct32_avx2 },
         { NULL, NULL, NULL, NULL },
         { NULL, NULL, NULL, NULL } },
 
-      { { NULL, NULL, NULL, NULL },
+      { { idct64_low1_avx2, idct64_low8_avx2, idct64_low16_avx2, idct64_avx2 },
         { NULL, NULL, NULL, NULL },
         { NULL, NULL, NULL, NULL } }
     };
@@ -1186,7 +4216,7 @@ static void highbd_inv_txfm2d_add_no_identity_avx2(const int32_t *input,
                                                    TX_TYPE tx_type,
                                                    TX_SIZE tx_size, int eob,
                                                    const int bd) {
-  __m256i buf1[64 * 2];
+  __m256i buf1[64 * 8];
   int eobx, eoby;
   get_eobx_eoby_scan_default(&eobx, &eoby, tx_size, eob);
   const int8_t *shift = inv_txfm_shift_ls[tx_size];
@@ -1198,7 +4228,7 @@ static void highbd_inv_txfm2d_add_no_identity_avx2(const int32_t *input,
   const int buf_size_nonzero_w_div8 = (eobx + 8) >> 3;
   const int buf_size_nonzero_h_div8 = (eoby + 8) >> 3;
   const int input_stride = AOMMIN(32, txfm_size_col);
-
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
   const int fun_idx_x = lowbd_txfm_all_1d_zeros_idx[eobx];
   const int fun_idx_y = lowbd_txfm_all_1d_zeros_idx[eoby];
   const transform_1d_avx2 row_txfm =
@@ -1213,7 +4243,7 @@ static void highbd_inv_txfm2d_add_no_identity_avx2(const int32_t *input,
 
   // 1st stage: column transform
   for (int i = 0; i < buf_size_nonzero_h_div8; i++) {
-    __m256i buf0[32];
+    __m256i buf0[64];
     const int32_t *input_row = input + i * input_stride * 8;
     for (int j = 0; j < buf_size_nonzero_w_div8; ++j) {
       __m256i *buf0_cur = buf0 + j * 8;
@@ -1221,12 +4251,22 @@ static void highbd_inv_txfm2d_add_no_identity_avx2(const int32_t *input,
 
       transpose_8x8_avx2(&buf0_cur[0], &buf0_cur[0]);
     }
-
+    if (rect_type == 1 || rect_type == -1) {
+      av1_round_shift_rect_array_32_avx2(
+          buf0, buf0, buf_size_nonzero_w_div8 << 3, 0, NewInvSqrt2);
+    }
     row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
 
     __m256i *_buf1 = buf1 + i * 8;
-    for (int j = 0; j < buf_size_w_div8; ++j) {
-      transpose_8x8_avx2(&buf0[j * 8], &_buf1[j * txfm_size_row]);
+    if (lr_flip) {
+      for (int j = 0; j < buf_size_w_div8; ++j) {
+        transpose_8x8_flip_avx2(
+            &buf0[j * 8], &_buf1[(buf_size_w_div8 - 1 - j) * txfm_size_row]);
+      }
+    } else {
+      for (int j = 0; j < buf_size_w_div8; ++j) {
+        transpose_8x8_avx2(&buf0[j * 8], &_buf1[j * txfm_size_row]);
+      }
     }
   }
   // 2nd stage: column transform
@@ -1240,12 +4280,15 @@ static void highbd_inv_txfm2d_add_no_identity_avx2(const int32_t *input,
   }
 
   // write to buffer
-  {
+  if (txfm_size_col >= 16) {
     for (int i = 0; i < (txfm_size_col >> 4); i++) {
       highbd_write_buffer_16xn_avx2(buf1 + i * txfm_size_row * 2,
                                     output + 16 * i, stride, ud_flip,
                                     txfm_size_row, bd);
     }
+  } else if (txfm_size_col == 8) {
+    highbd_write_buffer_8xn_avx2(buf1, output, stride, ud_flip, txfm_size_row,
+                                 bd);
   }
 }
 
@@ -1255,95 +4298,54 @@ void av1_highbd_inv_txfm2d_add_universe_avx2(const int32_t *input,
                                              int eob, const int bd) {
   switch (tx_type) {
     case DCT_DCT:
+    case ADST_DCT:
+    case DCT_ADST:
+    case ADST_ADST:
+    case FLIPADST_DCT:
+    case DCT_FLIPADST:
+    case FLIPADST_FLIPADST:
+    case ADST_FLIPADST:
+    case FLIPADST_ADST:
       highbd_inv_txfm2d_add_no_identity_avx2(input, CONVERT_TO_SHORTPTR(output),
                                              stride, tx_type, tx_size, eob, bd);
       break;
+    case IDTX:
+    case H_DCT:
+    case H_ADST:
+    case H_FLIPADST:
+    case V_DCT:
+    case V_ADST:
+    case V_FLIPADST:
+      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, output, stride, tx_type,
+                                                tx_size, eob, bd);
+      break;
     default: assert(0); break;
   }
 }
-
-void av1_highbd_inv_txfm_add_32x32_avx2(const tran_low_t *input, uint8_t *dest,
-                                        int stride,
-                                        const TxfmParam *txfm_param) {
-  const int bd = txfm_param->bd;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  const int32_t *src = cast_to_int32(input);
-  switch (tx_type) {
-    case DCT_DCT:
-      av1_highbd_inv_txfm2d_add_universe_avx2(input, dest, stride, tx_type,
-                                              txfm_param->tx_size,
-                                              txfm_param->eob, bd);
-      break;
-      // Assembly version doesn't support IDTX, so use C version for it.
-    case IDTX:
-      av1_inv_txfm2d_add_32x32_c(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                 tx_type, bd);
-      break;
-
-    default: assert(0);
-  }
-}
-
 void av1_highbd_inv_txfm_add_avx2(const tran_low_t *input, uint8_t *dest,
                                   int stride, const TxfmParam *txfm_param) {
   assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]);
   const TX_SIZE tx_size = txfm_param->tx_size;
   switch (tx_size) {
-    case TX_32X32:
-      av1_highbd_inv_txfm_add_32x32_avx2(input, dest, stride, txfm_param);
-      break;
-    case TX_16X16:
-      av1_highbd_inv_txfm_add_16x16_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_8X8:
-      av1_highbd_inv_txfm_add_8x8_sse4_1(input, dest, stride, txfm_param);
-      break;
     case TX_4X8:
-      av1_highbd_inv_txfm_add_4x8(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x8_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_8X4:
-      av1_highbd_inv_txfm_add_8x4(input, dest, stride, txfm_param);
-      break;
-    case TX_8X16:
-      av1_highbd_inv_txfm_add_8x16_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_16X8:
-      av1_highbd_inv_txfm_add_16x8_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_16X32:
-      av1_highbd_inv_txfm_add_16x32(input, dest, stride, txfm_param);
-      break;
-    case TX_32X16:
-      av1_highbd_inv_txfm_add_32x16(input, dest, stride, txfm_param);
-      break;
-    case TX_32X64:
-      av1_highbd_inv_txfm_add_32x64(input, dest, stride, txfm_param);
-      break;
-    case TX_64X32:
-      av1_highbd_inv_txfm_add_64x32(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_8x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_4X4:
       av1_highbd_inv_txfm_add_4x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_16X4:
-      av1_highbd_inv_txfm_add_16x4(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_16x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_4X16:
-      av1_highbd_inv_txfm_add_4x16(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x16_sse4_1(input, dest, stride, txfm_param);
       break;
-    case TX_8X32:
-      av1_highbd_inv_txfm_add_8x32(input, dest, stride, txfm_param);
-      break;
-    case TX_32X8:
-      av1_highbd_inv_txfm_add_32x8(input, dest, stride, txfm_param);
-      break;
-    case TX_64X64:
-    case TX_16X64:
-    case TX_64X16:
-      av1_highbd_inv_txfm2d_add_universe_sse4_1(
+    default:
+      av1_highbd_inv_txfm2d_add_universe_avx2(
           input, dest, stride, txfm_param->tx_type, txfm_param->tx_size,
           txfm_param->eob, txfm_param->bd);
       break;
-    default: assert(0 && "Invalid transform size"); break;
   }
 }
diff --git a/third_party/aom/av1/common/x86/highbd_inv_txfm_sse4.c b/third_party/aom/av1/common/x86/highbd_inv_txfm_sse4.c
index e29e0baf50..2f0eecb8bc 100644
--- a/third_party/aom/av1/common/x86/highbd_inv_txfm_sse4.c
+++ b/third_party/aom/av1/common/x86/highbd_inv_txfm_sse4.c
@@ -17,6 +17,7 @@
 #include "av1/common/av1_inv_txfm1d_cfg.h"
 #include "av1/common/idct.h"
 #include "av1/common/x86/av1_inv_txfm_ssse3.h"
+#include "av1/common/x86/av1_txfm_sse2.h"
 #include "av1/common/x86/av1_txfm_sse4.h"
 #include "av1/common/x86/highbd_txfm_utility_sse4.h"
 
@@ -49,6 +50,29 @@ static INLINE __m128i highbd_get_recon_8x8_sse4_1(const __m128i pred,
   return x0;
 }
 
+static INLINE __m128i highbd_get_recon_4xn_sse4_1(const __m128i pred,
+                                                  __m128i res0, const int bd) {
+  __m128i x0 = _mm_cvtepi16_epi32(pred);
+
+  x0 = _mm_add_epi32(res0, x0);
+  x0 = _mm_packus_epi32(x0, x0);
+  x0 = highbd_clamp_epi16(x0, bd);
+  return x0;
+}
+
+static INLINE void highbd_write_buffer_4xn_sse4_1(__m128i *in, uint16_t *output,
+                                                  int stride, int flipud,
+                                                  int height, const int bd) {
+  int j = flipud ? (height - 1) : 0;
+  const int step = flipud ? -1 : 1;
+  for (int i = 0; i < height; ++i, j += step) {
+    __m128i v = _mm_loadl_epi64((__m128i const *)(output + i * stride));
+    __m128i u = highbd_get_recon_4xn_sse4_1(v, in[j], bd);
+
+    _mm_storel_epi64((__m128i *)(output + i * stride), u);
+  }
+}
+
 static INLINE void highbd_write_buffer_8xn_sse4_1(__m128i *in, uint16_t *output,
                                                   int stride, int flipud,
                                                   int height, const int bd) {
@@ -354,7 +378,9 @@ static void neg_shift_sse4_1(const __m128i in0, const __m128i in1,
   *out1 = a1;
 }
 
-static void idct4x4_sse4_1(__m128i *in, int bit, int do_cols, int bd) {
+static void idct4x4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                           int bd, int out_shift) {
+  (void)out_shift;
   const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
@@ -398,18 +424,20 @@ static void idct4x4_sse4_1(__m128i *in, int bit, int do_cols, int bd) {
   v3 = _mm_srai_epi32(v3, bit);
 
   if (do_cols) {
-    addsub_no_clamp_sse4_1(v0, v3, in + 0, in + 3);
-    addsub_no_clamp_sse4_1(v1, v2, in + 1, in + 2);
+    addsub_no_clamp_sse4_1(v0, v3, out + 0, out + 3);
+    addsub_no_clamp_sse4_1(v1, v2, out + 1, out + 2);
   } else {
     const int log_range = AOMMAX(16, bd + 6);
     const __m128i clamp_lo = _mm_set1_epi32(-(1 << (log_range - 1)));
     const __m128i clamp_hi = _mm_set1_epi32((1 << (log_range - 1)) - 1);
-    addsub_sse4_1(v0, v3, in + 0, in + 3, &clamp_lo, &clamp_hi);
-    addsub_sse4_1(v1, v2, in + 1, in + 2, &clamp_lo, &clamp_hi);
+    addsub_sse4_1(v0, v3, out + 0, out + 3, &clamp_lo, &clamp_hi);
+    addsub_sse4_1(v1, v2, out + 1, out + 2, &clamp_lo, &clamp_hi);
   }
 }
 
-static void iadst4x4_sse4_1(__m128i *in, int bit, int do_cols, int bd) {
+static void iadst4x4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                            int bd, int out_shift) {
+  (void)out_shift;
   const int32_t *sinpi = sinpi_arr(bit);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
   const __m128i sinpi1 = _mm_set1_epi32((int)sinpi[1]);
@@ -482,10 +510,10 @@ static void iadst4x4_sse4_1(__m128i *in, int bit, int do_cols, int bd) {
     u3 = _mm_min_epi32(u3, clamp_hi);
   }
 
-  in[0] = u0;
-  in[1] = u1;
-  in[2] = u2;
-  in[3] = u3;
+  out[0] = u0;
+  out[1] = u1;
+  out[2] = u2;
+  out[3] = u3;
 }
 
 static INLINE void round_shift_4x4(__m128i *in, int shift) {
@@ -555,7 +583,66 @@ static void write_buffer_4x4(__m128i *in, uint16_t *output, int stride,
   _mm_storel_epi64((__m128i *)(output + 2 * stride), v2);
   _mm_storel_epi64((__m128i *)(output + 3 * stride), v3);
 }
+static void highbd_clamp_epi32_sse4_1(const __m128i *in, __m128i *out,
+                                      const __m128i *clamp_lo,
+                                      const __m128i *clamp_hi, int size) {
+  __m128i a0, a1;
+  for (int i = 0; i < size; i += 4) {
+    a0 = _mm_max_epi32(in[i], *clamp_lo);
+    out[i] = _mm_min_epi32(a0, *clamp_hi);
 
+    a1 = _mm_max_epi32(in[i + 1], *clamp_lo);
+    out[i + 1] = _mm_min_epi32(a1, *clamp_hi);
+
+    a0 = _mm_max_epi32(in[i + 2], *clamp_lo);
+    out[i + 2] = _mm_min_epi32(a0, *clamp_hi);
+
+    a1 = _mm_max_epi32(in[i + 3], *clamp_lo);
+    out[i + 3] = _mm_min_epi32(a1, *clamp_hi);
+  }
+}
+static void iidentity4_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  (void)bit;
+  (void)out_shift;
+  __m128i v[4];
+  __m128i fact = _mm_set1_epi32(NewSqrt2);
+  __m128i offset = _mm_set1_epi32(1 << (NewSqrt2Bits - 1));
+  __m128i a0, a1;
+
+  a0 = _mm_mullo_epi32(in[0], fact);
+  a1 = _mm_mullo_epi32(in[1], fact);
+  a0 = _mm_add_epi32(a0, offset);
+  a1 = _mm_add_epi32(a1, offset);
+  out[0] = _mm_srai_epi32(a0, NewSqrt2Bits);
+  out[1] = _mm_srai_epi32(a1, NewSqrt2Bits);
+
+  a0 = _mm_mullo_epi32(in[2], fact);
+  a1 = _mm_mullo_epi32(in[3], fact);
+  a0 = _mm_add_epi32(a0, offset);
+  a1 = _mm_add_epi32(a1, offset);
+  out[2] = _mm_srai_epi32(a0, NewSqrt2Bits);
+  out[3] = _mm_srai_epi32(a1, NewSqrt2Bits);
+
+  if (!do_cols) {
+    const int log_range = AOMMAX(16, bd + 6);
+    const __m128i clamp_lo = _mm_set1_epi32(-(1 << (log_range - 1)));
+    const __m128i clamp_hi = _mm_set1_epi32((1 << (log_range - 1)) - 1);
+
+    highbd_clamp_epi32_sse4_1(out, out, &clamp_lo, &clamp_hi, 4);
+  }
+
+  // Transpose for 4x4
+  v[0] = _mm_unpacklo_epi32(out[0], out[1]);
+  v[1] = _mm_unpackhi_epi32(out[0], out[1]);
+  v[2] = _mm_unpacklo_epi32(out[2], out[3]);
+  v[3] = _mm_unpackhi_epi32(out[2], out[3]);
+
+  out[0] = _mm_unpacklo_epi64(v[0], v[2]);
+  out[1] = _mm_unpackhi_epi64(v[0], v[2]);
+  out[2] = _mm_unpacklo_epi64(v[1], v[3]);
+  out[3] = _mm_unpackhi_epi64(v[1], v[3]);
+}
 void av1_inv_txfm2d_add_4x4_sse4_1(const int32_t *coeff, uint16_t *output,
                                    int stride, TX_TYPE tx_type, int bd) {
   __m128i in[4];
@@ -566,58 +653,100 @@ void av1_inv_txfm2d_add_4x4_sse4_1(const int32_t *coeff, uint16_t *output,
   switch (tx_type) {
     case DCT_DCT:
       load_buffer_4x4(coeff, in);
-      idct4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      idct4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      idct4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      idct4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
       break;
     case ADST_DCT:
       load_buffer_4x4(coeff, in);
-      idct4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      idct4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
       break;
     case DCT_ADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      idct4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      idct4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
       break;
     case ADST_ADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
       break;
     case FLIPADST_DCT:
       load_buffer_4x4(coeff, in);
-      idct4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      idct4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 1, -shift[1], bd);
       break;
     case DCT_FLIPADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      idct4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      idct4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 1, 0, -shift[1], bd);
       break;
     case FLIPADST_FLIPADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 1, 1, -shift[1], bd);
       break;
     case ADST_FLIPADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 1, 0, -shift[1], bd);
       break;
     case FLIPADST_ADST:
       load_buffer_4x4(coeff, in);
-      iadst4x4_sse4_1(in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd);
-      iadst4x4_sse4_1(in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
       write_buffer_4x4(in, output, stride, 0, 1, -shift[1], bd);
       break;
+    case IDTX:
+      load_buffer_4x4(coeff, in);
+      iidentity4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iidentity4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
+      break;
+    case V_DCT:
+      load_buffer_4x4(coeff, in);
+      iidentity4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      idct4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
+      break;
+    case H_DCT:
+      load_buffer_4x4(coeff, in);
+      idct4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iidentity4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
+      break;
+    case V_ADST:
+      load_buffer_4x4(coeff, in);
+      iidentity4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
+      break;
+    case H_ADST:
+      load_buffer_4x4(coeff, in);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iidentity4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 0, -shift[1], bd);
+      break;
+    case V_FLIPADST:
+      load_buffer_4x4(coeff, in);
+      iidentity4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 0, 1, -shift[1], bd);
+      break;
+    case H_FLIPADST:
+      load_buffer_4x4(coeff, in);
+      iadst4x4_sse4_1(in, in, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, 0);
+      iidentity4_sse4_1(in, in, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+      write_buffer_4x4(in, output, stride, 1, 0, -shift[1], bd);
+      break;
     default: assert(0);
   }
 }
@@ -1088,6 +1217,61 @@ static void iadst8x8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
                      &clamp_hi_out, out_shift);
   }
 }
+static void shift_sse4_1(const __m128i *in, __m128i *out,
+                         const __m128i *clamp_lo, const __m128i *clamp_hi,
+                         int shift, int size) {
+  __m128i offset = _mm_set1_epi32((1 << shift) >> 1);
+  __m128i shift_vec = _mm_cvtsi32_si128(shift);
+  __m128i a0, a1;
+  for (int i = 0; i < size; i += 4) {
+    a0 = _mm_add_epi32(in[i], offset);
+    a1 = _mm_add_epi32(in[i + 1], offset);
+    a0 = _mm_sra_epi32(a0, shift_vec);
+    a1 = _mm_sra_epi32(a1, shift_vec);
+    a0 = _mm_max_epi32(a0, *clamp_lo);
+    a1 = _mm_max_epi32(a1, *clamp_lo);
+    out[i] = _mm_min_epi32(a0, *clamp_hi);
+    out[i + 1] = _mm_min_epi32(a1, *clamp_hi);
+
+    a0 = _mm_add_epi32(in[i + 2], offset);
+    a1 = _mm_add_epi32(in[i + 3], offset);
+    a0 = _mm_sra_epi32(a0, shift_vec);
+    a1 = _mm_sra_epi32(a1, shift_vec);
+    a0 = _mm_max_epi32(a0, *clamp_lo);
+    a1 = _mm_max_epi32(a1, *clamp_lo);
+    out[i + 2] = _mm_min_epi32(a0, *clamp_hi);
+    out[i + 3] = _mm_min_epi32(a1, *clamp_hi);
+  }
+}
+
+static void iidentity8_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                              int bd, int out_shift) {
+  (void)bit;
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m128i clamp_lo = _mm_set1_epi32(-(1 << (log_range - 1)));
+  const __m128i clamp_hi = _mm_set1_epi32((1 << (log_range - 1)) - 1);
+  __m128i v[8];
+  v[0] = _mm_add_epi32(in[0], in[0]);
+  v[1] = _mm_add_epi32(in[1], in[1]);
+  v[2] = _mm_add_epi32(in[2], in[2]);
+  v[3] = _mm_add_epi32(in[3], in[3]);
+  v[4] = _mm_add_epi32(in[4], in[4]);
+  v[5] = _mm_add_epi32(in[5], in[5]);
+  v[6] = _mm_add_epi32(in[6], in[6]);
+  v[7] = _mm_add_epi32(in[7], in[7]);
+
+  if (!do_cols) {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m128i clamp_lo_out = _mm_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m128i clamp_hi_out = _mm_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+    shift_sse4_1(v, out, &clamp_lo_out, &clamp_hi_out, out_shift, 8);
+  } else {
+    highbd_clamp_epi32_sse4_1(v, out, &clamp_lo, &clamp_hi, 8);
+  }
+}
 
 static void round_shift_8x8(__m128i *in, int shift) {
   round_shift_4x4(&in[0], shift);
@@ -2972,7 +3156,59 @@ static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
     }
   }
 }
+static void iidentity16_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                               int bd, int out_shift) {
+  (void)bit;
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m128i clamp_lo = _mm_set1_epi32(-(1 << (log_range - 1)));
+  const __m128i clamp_hi = _mm_set1_epi32((1 << (log_range - 1)) - 1);
+  __m128i v[16];
+  __m128i fact = _mm_set1_epi32(2 * NewSqrt2);
+  __m128i offset = _mm_set1_epi32(1 << (NewSqrt2Bits - 1));
+  __m128i a0, a1, a2, a3;
 
+  for (int i = 0; i < 16; i += 8) {
+    a0 = _mm_mullo_epi32(in[i], fact);
+    a1 = _mm_mullo_epi32(in[i + 1], fact);
+    a0 = _mm_add_epi32(a0, offset);
+    a1 = _mm_add_epi32(a1, offset);
+    v[i] = _mm_srai_epi32(a0, NewSqrt2Bits);
+    v[i + 1] = _mm_srai_epi32(a1, NewSqrt2Bits);
+
+    a2 = _mm_mullo_epi32(in[i + 2], fact);
+    a3 = _mm_mullo_epi32(in[i + 3], fact);
+    a2 = _mm_add_epi32(a2, offset);
+    a3 = _mm_add_epi32(a3, offset);
+    v[i + 2] = _mm_srai_epi32(a2, NewSqrt2Bits);
+    v[i + 3] = _mm_srai_epi32(a3, NewSqrt2Bits);
+
+    a0 = _mm_mullo_epi32(in[i + 4], fact);
+    a1 = _mm_mullo_epi32(in[i + 5], fact);
+    a0 = _mm_add_epi32(a0, offset);
+    a1 = _mm_add_epi32(a1, offset);
+    v[i + 4] = _mm_srai_epi32(a0, NewSqrt2Bits);
+    v[i + 5] = _mm_srai_epi32(a1, NewSqrt2Bits);
+
+    a2 = _mm_mullo_epi32(in[i + 6], fact);
+    a3 = _mm_mullo_epi32(in[i + 7], fact);
+    a2 = _mm_add_epi32(a2, offset);
+    a3 = _mm_add_epi32(a3, offset);
+    v[i + 6] = _mm_srai_epi32(a2, NewSqrt2Bits);
+    v[i + 7] = _mm_srai_epi32(a3, NewSqrt2Bits);
+  }
+
+  if (!do_cols) {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m128i clamp_lo_out = _mm_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m128i clamp_hi_out = _mm_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+
+    shift_sse4_1(v, out, &clamp_lo_out, &clamp_hi_out, out_shift, 16);
+  } else {
+    highbd_clamp_epi32_sse4_1(v, out, &clamp_lo, &clamp_hi, 16);
+  }
+}
 static INLINE void idct64_stage8_sse4_1(
     __m128i *u, const __m128i *cospim32, const __m128i *cospi32,
     const __m128i *cospim16, const __m128i *cospi48, const __m128i *cospi16,
@@ -3154,69 +3390,69 @@ static void idct64x64_low1_sse4_1(__m128i *in, __m128i *out, int bit,
     }
 
     out[0] = x;
-    out[63] = x;
     out[1] = x;
-    out[62] = x;
     out[2] = x;
-    out[61] = x;
     out[3] = x;
-    out[60] = x;
     out[4] = x;
-    out[59] = x;
     out[5] = x;
-    out[58] = x;
     out[6] = x;
-    out[57] = x;
     out[7] = x;
-    out[56] = x;
     out[8] = x;
-    out[55] = x;
     out[9] = x;
-    out[54] = x;
     out[10] = x;
-    out[53] = x;
     out[11] = x;
-    out[52] = x;
     out[12] = x;
-    out[51] = x;
     out[13] = x;
-    out[50] = x;
     out[14] = x;
-    out[49] = x;
     out[15] = x;
-    out[48] = x;
     out[16] = x;
-    out[47] = x;
     out[17] = x;
-    out[46] = x;
     out[18] = x;
-    out[45] = x;
     out[19] = x;
-    out[44] = x;
     out[20] = x;
-    out[43] = x;
     out[21] = x;
-    out[42] = x;
     out[22] = x;
-    out[41] = x;
     out[23] = x;
-    out[40] = x;
     out[24] = x;
-    out[39] = x;
     out[25] = x;
-    out[38] = x;
     out[26] = x;
-    out[37] = x;
     out[27] = x;
-    out[36] = x;
     out[28] = x;
-    out[35] = x;
     out[29] = x;
-    out[34] = x;
     out[30] = x;
-    out[33] = x;
     out[31] = x;
     out[32] = x;
+    out[33] = x;
+    out[34] = x;
+    out[35] = x;
+    out[36] = x;
+    out[37] = x;
+    out[38] = x;
+    out[39] = x;
+    out[40] = x;
+    out[41] = x;
+    out[42] = x;
+    out[43] = x;
+    out[44] = x;
+    out[45] = x;
+    out[46] = x;
+    out[47] = x;
+    out[48] = x;
+    out[49] = x;
+    out[50] = x;
+    out[51] = x;
+    out[52] = x;
+    out[53] = x;
+    out[54] = x;
+    out[55] = x;
+    out[56] = x;
+    out[57] = x;
+    out[58] = x;
+    out[59] = x;
+    out[60] = x;
+    out[61] = x;
+    out[62] = x;
+    out[63] = x;
   }
 }
 
@@ -4992,17 +5228,16 @@ void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *input, uint8_t *dest,
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int32_t *src = cast_to_int32(input);
   switch (tx_type) {
-      // Assembly version doesn't support some transform types, so use C version
-      // for those.
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
     case IDTX:
-      av1_inv_txfm2d_add_8x8_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type,
-                               bd);
+    case H_DCT:
+    case H_ADST:
+    case H_FLIPADST:
+    case V_DCT:
+    case V_ADST:
+    case V_FLIPADST:
+      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, dest, stride, tx_type,
+                                                txfm_param->tx_size,
+                                                txfm_param->eob, bd);
       break;
     default:
       av1_inv_txfm2d_add_8x8_sse4_1(src, CONVERT_TO_SHORTPTR(dest), stride,
@@ -5010,109 +5245,6 @@ void av1_highbd_inv_txfm_add_8x8_sse4_1(const tran_low_t *input, uint8_t *dest,
       break;
   }
 }
-
-void av1_highbd_inv_txfm_add_16x8_sse4_1(const tran_low_t *input, uint8_t *dest,
-                                         int stride,
-                                         const TxfmParam *txfm_param) {
-  int bd = txfm_param->bd;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  const int32_t *src = cast_to_int32(input);
-  switch (tx_type) {
-      // Assembly version doesn't support some transform types, so use C version
-      // for those.
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_inv_txfm2d_add_16x8_c(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                txfm_param->tx_type, txfm_param->bd);
-      break;
-    default:
-      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, dest, stride, tx_type,
-                                                txfm_param->tx_size,
-                                                txfm_param->eob, bd);
-      break;
-  }
-}
-
-void av1_highbd_inv_txfm_add_8x16_sse4_1(const tran_low_t *input, uint8_t *dest,
-                                         int stride,
-                                         const TxfmParam *txfm_param) {
-  int bd = txfm_param->bd;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  const int32_t *src = cast_to_int32(input);
-  switch (tx_type) {
-      // Assembly version doesn't support some transform types, so use C version
-      // for those.
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_inv_txfm2d_add_8x16_c(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                txfm_param->tx_type, txfm_param->bd);
-      break;
-    default:
-      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, dest, stride, tx_type,
-                                                txfm_param->tx_size,
-                                                txfm_param->eob, bd);
-      break;
-  }
-}
-
-void av1_highbd_inv_txfm_add_16x16_sse4_1(const tran_low_t *input,
-                                          uint8_t *dest, int stride,
-                                          const TxfmParam *txfm_param) {
-  int bd = txfm_param->bd;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  const int32_t *src = cast_to_int32(input);
-  switch (tx_type) {
-      // Assembly version doesn't support some transform types, so use C version
-      // for those.
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_inv_txfm2d_add_16x16_c(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                 tx_type, bd);
-      break;
-    default:
-      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, dest, stride, tx_type,
-                                                txfm_param->tx_size,
-                                                txfm_param->eob, bd);
-      break;
-  }
-}
-
-void av1_highbd_inv_txfm_add_32x32_sse4_1(const tran_low_t *input,
-                                          uint8_t *dest, int stride,
-                                          const TxfmParam *txfm_param) {
-  int bd = txfm_param->bd;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  const int32_t *src = cast_to_int32(input);
-  switch (tx_type) {
-    case DCT_DCT:
-      av1_highbd_inv_txfm2d_add_universe_sse4_1(input, dest, stride, tx_type,
-                                                txfm_param->tx_size,
-                                                txfm_param->eob, bd);
-      break;
-      // Assembly version doesn't support IDTX, so use C version for it.
-    case IDTX:
-      av1_inv_txfm2d_add_32x32_c(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                 tx_type, bd);
-      break;
-    default: assert(0);
-  }
-}
-
 void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *input, uint8_t *dest,
                                         int stride,
                                         const TxfmParam *txfm_param) {
@@ -5127,53 +5259,271 @@ void av1_highbd_inv_txfm_add_4x4_sse4_1(const tran_low_t *input, uint8_t *dest,
     av1_highbd_iwht4x4_add(input, dest, stride, eob, bd);
     return;
   }
-  switch (tx_type) {
-      // Assembly version doesn't support some transform types, so use C version
-      // for those.
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_inv_txfm2d_add_4x4_c(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type,
-                               bd);
-      break;
-    default:
-      av1_inv_txfm2d_add_4x4_sse4_1(src, CONVERT_TO_SHORTPTR(dest), stride,
-                                    tx_type, bd);
-      break;
+  av1_inv_txfm2d_add_4x4_sse4_1(src, CONVERT_TO_SHORTPTR(dest), stride, tx_type,
+                                bd);
+}
+static void iidentity32_sse4_1(__m128i *in, __m128i *out, int bit, int do_cols,
+                               int bd, int out_shift) {
+  (void)bit;
+  const int log_range = AOMMAX(16, bd + (do_cols ? 6 : 8));
+  const __m128i clamp_lo = _mm_set1_epi32(-(1 << (log_range - 1)));
+  const __m128i clamp_hi = _mm_set1_epi32((1 << (log_range - 1)) - 1);
+  __m128i v[32];
+  for (int i = 0; i < 32; i += 16) {
+    v[i] = _mm_slli_epi32(in[i], 2);
+    v[i + 1] = _mm_slli_epi32(in[i + 1], 2);
+    v[i + 2] = _mm_slli_epi32(in[i + 2], 2);
+    v[i + 3] = _mm_slli_epi32(in[i + 3], 2);
+    v[i + 4] = _mm_slli_epi32(in[i + 4], 2);
+    v[i + 5] = _mm_slli_epi32(in[i + 5], 2);
+    v[i + 6] = _mm_slli_epi32(in[i + 6], 2);
+    v[i + 7] = _mm_slli_epi32(in[i + 7], 2);
+    v[i + 8] = _mm_slli_epi32(in[i + 8], 2);
+    v[i + 9] = _mm_slli_epi32(in[i + 9], 2);
+    v[i + 10] = _mm_slli_epi32(in[i + 10], 2);
+    v[i + 11] = _mm_slli_epi32(in[i + 11], 2);
+    v[i + 12] = _mm_slli_epi32(in[i + 12], 2);
+    v[i + 13] = _mm_slli_epi32(in[i + 13], 2);
+    v[i + 14] = _mm_slli_epi32(in[i + 14], 2);
+    v[i + 15] = _mm_slli_epi32(in[i + 15], 2);
+  }
+
+  if (!do_cols) {
+    const int log_range_out = AOMMAX(16, bd + 6);
+    const __m128i clamp_lo_out = _mm_set1_epi32(AOMMAX(
+        -(1 << (log_range_out - 1)), -(1 << (log_range - 1 - out_shift))));
+    const __m128i clamp_hi_out = _mm_set1_epi32(AOMMIN(
+        (1 << (log_range_out - 1)) - 1, (1 << (log_range - 1 - out_shift))));
+    shift_sse4_1(v, out, &clamp_lo_out, &clamp_hi_out, out_shift, 32);
+  } else {
+    highbd_clamp_epi32_sse4_1(v, out, &clamp_lo, &clamp_hi, 32);
   }
 }
-
 static const transform_1d_sse4_1
     highbd_txfm_all_1d_zeros_w8_arr[TX_SIZES][ITX_TYPES_1D][4] = {
       {
-          { NULL, NULL, NULL, NULL },
-          { NULL, NULL, NULL, NULL },
-          { NULL, NULL, NULL, NULL },
+          { idct4x4_sse4_1, NULL, NULL, NULL },
+          { iadst4x4_sse4_1, NULL, NULL, NULL },
+          { iidentity4_sse4_1, iidentity4_sse4_1, iidentity4_sse4_1, NULL },
       },
       { { idct8x8_low1_sse4_1, idct8x8_new_sse4_1, NULL, NULL },
         { iadst8x8_low1_sse4_1, iadst8x8_new_sse4_1, NULL, NULL },
-        { NULL, NULL, NULL, NULL } },
+        { iidentity8_sse4_1, iidentity8_sse4_1, NULL, NULL } },
       {
           { idct16x16_low1_sse4_1, idct16x16_low8_sse4_1, idct16x16_sse4_1,
             NULL },
           { iadst16x16_low1_sse4_1, iadst16x16_low8_sse4_1, iadst16x16_sse4_1,
             NULL },
-          { NULL, NULL, NULL, NULL },
+          { iidentity16_sse4_1, NULL, iidentity16_sse4_1, NULL },
       },
       { { idct32x32_low1_sse4_1, idct32x32_low8_sse4_1, idct32x32_low16_sse4_1,
           idct32x32_sse4_1 },
         { NULL, NULL, NULL, NULL },
-        { NULL, NULL, NULL, NULL } },
+        { iidentity32_sse4_1, NULL, NULL, NULL } },
       { { idct64x64_low1_sse4_1, idct64x64_low8_sse4_1, idct64x64_low16_sse4_1,
           idct64x64_sse4_1 },
         { NULL, NULL, NULL, NULL },
         { NULL, NULL, NULL, NULL } }
     };
+static void highbd_inv_txfm2d_add_h_identity_ssse41(const int32_t *input,
+                                                    uint16_t *output,
+                                                    int stride, TX_TYPE tx_type,
+                                                    TX_SIZE tx_size, int eob,
+                                                    const int bd) {
+  __m128i buf1[64];
+  int eobx, eoby;
+  get_eobx_eoby_scan_v_identity(&eobx, &eoby, tx_size, eob);
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const int input_stride = AOMMIN(32, txfm_size_col);
+  const int buf_size_w_div4 = input_stride >> 2;
+  const int buf_size_h_div8 = (eoby + 8) >> 3;
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  const int fun_idx = lowbd_txfm_all_1d_zeros_idx[eoby];
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][0];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][fun_idx];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
 
+  for (int i = 0; i < (buf_size_h_div8 << 1); ++i) {
+    __m128i buf0[16];
+    const int32_t *input_row = input + i * input_stride * 4;
+    for (int j = 0; j < buf_size_w_div4; ++j) {
+      __m128i *buf0_cur = buf0 + j * 4;
+      load_buffer_32bit_input(input_row + j * 4, input_stride, buf0_cur, 4);
+    }
+    if (rect_type == 1 || rect_type == -1) {
+      av1_round_shift_rect_array_32_sse4_1(buf0, buf0, input_stride, 0,
+                                           NewInvSqrt2);
+    }
+    row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+
+    __m128i *_buf1 = buf1 + i * 4;
+
+    for (int j = 0; j < buf_size_w_div4; ++j) {
+      _buf1[j * txfm_size_row + 0] = buf0[j * 4 + 0];
+      _buf1[j * txfm_size_row + 1] = buf0[j * 4 + 1];
+      _buf1[j * txfm_size_row + 2] = buf0[j * 4 + 2];
+      _buf1[j * txfm_size_row + 3] = buf0[j * 4 + 3];
+    }
+  }
+  for (int i = 0; i < buf_size_w_div4; i++) {
+    col_txfm(buf1 + i * txfm_size_row, buf1 + i * txfm_size_row,
+             inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+
+    av1_round_shift_array_32_sse4_1(buf1 + i * txfm_size_row,
+                                    buf1 + i * txfm_size_row, txfm_size_row,
+                                    -shift[1]);
+  }
+
+  // write to buffer
+  for (int i = 0; i < (txfm_size_col >> 3); i++) {
+    highbd_write_buffer_8xn_sse4_1(buf1 + i * txfm_size_row * 2, output + 8 * i,
+                                   stride, ud_flip, txfm_size_row, bd);
+  }
+}
+static void highbd_inv_txfm2d_add_v_identity_ssse41(const int32_t *input,
+                                                    uint16_t *output,
+                                                    int stride, TX_TYPE tx_type,
+                                                    TX_SIZE tx_size, int eob,
+                                                    const int bd) {
+  __m128i buf1[64];
+  int eobx, eoby;
+  get_eobx_eoby_scan_h_identity(&eobx, &eoby, tx_size, eob);
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const int input_stride = AOMMIN(32, txfm_size_col);
+  const int buf_size_w_div8 = input_stride >> 2;
+  const int row_max = AOMMIN(32, txfm_size_row);
+  const int buf_size_nonzero_w_div8 = (eobx + 8) >> 3;
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  const int fun_idx = lowbd_txfm_all_1d_zeros_idx[eobx];
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][fun_idx];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][0];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  for (int i = 0; i < (row_max >> 2); ++i) {
+    __m128i buf0[16];
+    const int32_t *input_row = input + i * input_stride * 4;
+    for (int j = 0; j < (buf_size_nonzero_w_div8 << 1); ++j) {
+      __m128i *buf0_cur = buf0 + j * 4;
+      load_buffer_32bit_input(input_row + j * 4, input_stride, buf0_cur, 4);
+
+      TRANSPOSE_4X4(buf0_cur[0], buf0_cur[1], buf0_cur[2], buf0_cur[3],
+                    buf0_cur[0], buf0_cur[1], buf0_cur[2], buf0_cur[3]);
+    }
+    if (rect_type == 1 || rect_type == -1) {
+      av1_round_shift_rect_array_32_sse4_1(
+          buf0, buf0, (buf_size_nonzero_w_div8 << 3), 0, NewInvSqrt2);
+    }
+    row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+
+    __m128i *_buf1 = buf1 + i * 4;
+    if (lr_flip) {
+      for (int j = 0; j < buf_size_w_div8; ++j) {
+        TRANSPOSE_4X4(buf0[4 * j + 3], buf0[4 * j + 2], buf0[4 * j + 1],
+                      buf0[4 * j],
+                      _buf1[txfm_size_row * (buf_size_w_div8 - 1 - j) + 0],
+                      _buf1[txfm_size_row * (buf_size_w_div8 - 1 - j) + 1],
+                      _buf1[txfm_size_row * (buf_size_w_div8 - 1 - j) + 2],
+                      _buf1[txfm_size_row * (buf_size_w_div8 - 1 - j) + 3]);
+      }
+    } else {
+      for (int j = 0; j < buf_size_w_div8; ++j) {
+        TRANSPOSE_4X4(
+            buf0[j * 4 + 0], buf0[j * 4 + 1], buf0[j * 4 + 2], buf0[j * 4 + 3],
+            _buf1[j * txfm_size_row + 0], _buf1[j * txfm_size_row + 1],
+            _buf1[j * txfm_size_row + 2], _buf1[j * txfm_size_row + 3]);
+      }
+    }
+  }
+  for (int i = 0; i < buf_size_w_div8; i++) {
+    col_txfm(buf1 + i * txfm_size_row, buf1 + i * txfm_size_row,
+             inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+
+    av1_round_shift_array_32_sse4_1(buf1 + i * txfm_size_row,
+                                    buf1 + i * txfm_size_row, txfm_size_row,
+                                    -shift[1]);
+  }
+
+  // write to buffer
+  {
+    for (int i = 0; i < (txfm_size_col >> 3); i++) {
+      highbd_write_buffer_8xn_sse4_1(buf1 + i * txfm_size_row * 2,
+                                     output + 8 * i, stride, ud_flip,
+                                     txfm_size_row, bd);
+    }
+  }
+}
+static void highbd_inv_txfm2d_add_idtx_ssse41(const int32_t *input,
+                                              uint16_t *output, int stride,
+                                              TX_TYPE tx_type, TX_SIZE tx_size,
+                                              int eob, const int bd) {
+  (void)eob;
+  __m128i buf1[64 * 4];
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const int input_stride = AOMMIN(32, txfm_size_col);
+  const int row_max = AOMMIN(32, txfm_size_row);
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][0];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][0];
+
+  for (int i = 0; i < (row_max >> 2); ++i) {
+    __m128i buf0[32];
+    const int32_t *input_row = input + i * input_stride * 4;
+    for (int j = 0; j < (input_stride >> 2); ++j) {
+      __m128i *buf0_cur = buf0 + j * 4;
+      load_buffer_32bit_input(input_row + j * 4, input_stride, buf0_cur, 4);
+    }
+    if (rect_type == 1 || rect_type == -1) {
+      av1_round_shift_rect_array_32_sse4_1(buf0, buf0, input_stride, 0,
+                                           NewInvSqrt2);
+    }
+    row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+
+    __m128i *_buf1 = buf1 + i * 4;
+    for (int j = 0; j < (input_stride >> 2); ++j) {
+      _buf1[j * txfm_size_row + 0] = buf0[j * 4 + 0];
+      _buf1[j * txfm_size_row + 1] = buf0[j * 4 + 1];
+      _buf1[j * txfm_size_row + 2] = buf0[j * 4 + 2];
+      _buf1[j * txfm_size_row + 3] = buf0[j * 4 + 3];
+    }
+  }
+  for (int i = 0; i < (input_stride >> 2); i++) {
+    col_txfm(buf1 + i * txfm_size_row, buf1 + i * txfm_size_row,
+             inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+
+    av1_round_shift_array_32_sse4_1(buf1 + i * txfm_size_row,
+                                    buf1 + i * txfm_size_row, txfm_size_row,
+                                    -shift[1]);
+  }
+
+  // write to buffer
+  {
+    for (int i = 0; i < (txfm_size_col >> 3); i++) {
+      highbd_write_buffer_8xn_sse4_1(buf1 + i * txfm_size_row * 2,
+                                     output + 8 * i, stride, 0, txfm_size_row,
+                                     bd);
+    }
+  }
+}
 static void highbd_inv_txfm2d_add_no_identity_sse41(const int32_t *input,
                                                     uint16_t *output,
                                                     int stride, TX_TYPE tx_type,
@@ -5261,6 +5611,232 @@ static void highbd_inv_txfm2d_add_no_identity_sse41(const int32_t *input,
   }
 }
 
+static void highbd_inv_txfm2d_add_4x8_sse41(const int32_t *input,
+                                            uint16_t *output, int stride,
+                                            TX_TYPE tx_type, TX_SIZE tx_size,
+                                            int eob, const int bd) {
+  (void)eob;
+  __m128i buf1[8];
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][0];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][1];
+  const int input_stride = AOMMIN(32, txfm_size_col);
+
+  assert(col_txfm != NULL);
+  assert(row_txfm != NULL);
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  // 1st stage: column transform
+  __m128i buf0[8];
+  const int32_t *input_row = input;
+  __m128i *buf0_cur = buf0;
+  load_buffer_32bit_input(input_row, input_stride, buf0_cur, txfm_size_row);
+  av1_round_shift_rect_array_32_sse4_1(buf0, buf0, txfm_size_row, 0,
+                                       NewInvSqrt2);
+  row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+  row_txfm(buf0 + 4, buf0 + 4, inv_cos_bit_row[txw_idx][txh_idx], 0, bd,
+           -shift[0]);
+
+  if (lr_flip) {
+    TRANSPOSE_4X4(buf0[3], buf0[2], buf0[1], buf0[0], buf1[0], buf1[1], buf1[2],
+                  buf1[3]);
+
+    TRANSPOSE_4X4(buf0[7], buf0[6], buf0[5], buf0[4], buf1[4], buf1[5], buf1[6],
+                  buf1[7]);
+  } else {
+    TRANSPOSE_4X4(buf0[0], buf0[1], buf0[2], buf0[3], buf1[0], buf1[1], buf1[2],
+                  buf1[3]);
+
+    TRANSPOSE_4X4(buf0[4], buf0[5], buf0[6], buf0[7], buf1[4], buf1[5], buf1[6],
+                  buf1[7]);
+  }
+
+  // 2nd stage: column transform
+  col_txfm(buf1, buf1, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+
+  av1_round_shift_array_32_sse4_1(buf1, buf1, txfm_size_row, -shift[1]);
+
+  // write to buffer
+  highbd_write_buffer_4xn_sse4_1(buf1, output, stride, ud_flip, txfm_size_row,
+                                 bd);
+}
+
+static void highbd_inv_txfm2d_add_8x4_sse41(const int32_t *input,
+                                            uint16_t *output, int stride,
+                                            TX_TYPE tx_type, TX_SIZE tx_size,
+                                            int eob, const int bd) {
+  (void)eob;
+  __m128i buf1[8];
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][1];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][0];
+
+  assert(col_txfm != NULL);
+  assert(row_txfm != NULL);
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  // 1st stage: column transform
+  __m128i buf0[8];
+  const int32_t *input_row = input;
+  load_buffer_32bit_input(input_row, 4, buf0, txfm_size_col);
+
+  TRANSPOSE_4X4(buf0[0], buf0[2], buf0[4], buf0[6], buf1[0], buf1[1], buf1[2],
+                buf1[3]);
+  TRANSPOSE_4X4(buf0[1], buf0[3], buf0[5], buf0[7], buf1[4], buf1[5], buf1[6],
+                buf1[7]);
+
+  av1_round_shift_rect_array_32_sse4_1(buf1, buf0, txfm_size_col, 0,
+                                       NewInvSqrt2);
+  row_txfm(buf0, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+
+  __m128i *buf1_ptr;
+  if (lr_flip) {
+    flip_buf_sse2(buf0, buf1, txfm_size_col);
+    buf1_ptr = buf1;
+  } else {
+    buf1_ptr = buf0;
+  }
+
+  // 2nd stage: column transform
+  for (int i = 0; i < 2; i++) {
+    col_txfm(buf1_ptr + i * txfm_size_row, buf1_ptr + i * txfm_size_row,
+             inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+  }
+  av1_round_shift_array_32_sse4_1(buf1_ptr, buf1_ptr, txfm_size_col, -shift[1]);
+  // write to buffer
+  highbd_write_buffer_8xn_sse4_1(buf1_ptr, output, stride, ud_flip,
+                                 txfm_size_row, bd);
+}
+
+static void highbd_inv_txfm2d_add_4x16_sse4_1(const int32_t *input,
+                                              uint16_t *output, int stride,
+                                              TX_TYPE tx_type, TX_SIZE tx_size,
+                                              int eob, const int bd) {
+  (void)eob;
+  __m128i buf1[16];
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const int buf_size_h_div8 = txfm_size_row >> 2;
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][0];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][2];
+  const int input_stride = AOMMIN(32, txfm_size_col);
+
+  assert(col_txfm != NULL);
+  assert(row_txfm != NULL);
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  // 1st stage: column transform
+  __m128i buf0[16];
+  const int32_t *input_row = input;
+  __m128i *buf0_cur = buf0;
+  load_buffer_32bit_input(input_row, input_stride, buf0_cur, txfm_size_row);
+  for (int i = 0; i < (txfm_size_row >> 2); i++) {
+    row_txfm(buf0 + (i << 2), buf0 + (i << 2),
+             inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+  }
+
+  av1_round_shift_array_32_sse4_1(buf0, buf0, txfm_size_row, -shift[0]);
+
+  if (lr_flip) {
+    for (int j = 0; j < buf_size_h_div8; ++j) {
+      TRANSPOSE_4X4(buf0[4 * j + 3], buf0[4 * j + 2], buf0[4 * j + 1],
+                    buf0[4 * j], buf1[4 * j], buf1[4 * j + 1], buf1[4 * j + 2],
+                    buf1[4 * j + 3]);
+    }
+  } else {
+    for (int j = 0; j < buf_size_h_div8; ++j) {
+      TRANSPOSE_4X4(buf0[4 * j], buf0[4 * j + 1], buf0[4 * j + 2],
+                    buf0[4 * j + 3], buf1[4 * j], buf1[4 * j + 1],
+                    buf1[4 * j + 2], buf1[4 * j + 3]);
+    }
+  }
+
+  // 2nd stage: column transform
+  col_txfm(buf1, buf1, inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+
+  av1_round_shift_array_32_sse4_1(buf1, buf1, txfm_size_row, -shift[1]);
+
+  // write to buffer
+  highbd_write_buffer_4xn_sse4_1(buf1, output, stride, ud_flip, txfm_size_row,
+                                 bd);
+}
+
+static void highbd_inv_txfm2d_add_16x4_sse4_1(const int32_t *input,
+                                              uint16_t *output, int stride,
+                                              TX_TYPE tx_type, TX_SIZE tx_size,
+                                              int eob, const int bd) {
+  (void)eob;
+  __m128i buf1[16];
+  const int8_t *shift = inv_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int txfm_size_col = tx_size_wide[tx_size];
+  const int txfm_size_row = tx_size_high[tx_size];
+  const int buf_size_w_div8 = txfm_size_col >> 2;
+  const transform_1d_sse4_1 row_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txw_idx][hitx_1d_tab[tx_type]][2];
+  const transform_1d_sse4_1 col_txfm =
+      highbd_txfm_all_1d_zeros_w8_arr[txh_idx][vitx_1d_tab[tx_type]][0];
+
+  assert(col_txfm != NULL);
+  assert(row_txfm != NULL);
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  // 1st stage: column transform
+  __m128i buf0[16];
+  const int32_t *input_row = input;
+  load_buffer_32bit_input(input_row, 4, buf0, txfm_size_col);
+
+  for (int j = 0; j < buf_size_w_div8; j++) {
+    TRANSPOSE_4X4(buf0[j], buf0[j + 4], buf0[j + 8], buf0[j + 12], buf1[4 * j],
+                  buf1[4 * j + 1], buf1[4 * j + 2], buf1[4 * j + 3]);
+  }
+  row_txfm(buf1, buf0, inv_cos_bit_row[txw_idx][txh_idx], 0, bd, -shift[0]);
+
+  __m128i *buf1_ptr;
+  if (lr_flip) {
+    flip_buf_sse2(buf0, buf1, txfm_size_col);
+    buf1_ptr = buf1;
+  } else {
+    buf1_ptr = buf0;
+  }
+
+  // 2nd stage: column transform
+  for (int i = 0; i < buf_size_w_div8; i++) {
+    col_txfm(buf1_ptr + i * txfm_size_row, buf1_ptr + i * txfm_size_row,
+             inv_cos_bit_col[txw_idx][txh_idx], 1, bd, 0);
+  }
+  av1_round_shift_array_32_sse4_1(buf1_ptr, buf1_ptr, txfm_size_col, -shift[1]);
+
+  // write to buffer
+  for (int i = 0; i < (txfm_size_col >> 3); i++) {
+    highbd_write_buffer_8xn_sse4_1(buf1_ptr + i * txfm_size_row * 2,
+                                   output + 8 * i, stride, ud_flip,
+                                   txfm_size_row, bd);
+  }
+}
+
 void av1_highbd_inv_txfm2d_add_universe_sse4_1(const int32_t *input,
                                                uint8_t *output, int stride,
                                                TX_TYPE tx_type, TX_SIZE tx_size,
@@ -5279,70 +5855,99 @@ void av1_highbd_inv_txfm2d_add_universe_sse4_1(const int32_t *input,
           input, CONVERT_TO_SHORTPTR(output), stride, tx_type, tx_size, eob,
           bd);
       break;
+    case V_DCT:
+    case V_ADST:
+    case V_FLIPADST:
+      highbd_inv_txfm2d_add_h_identity_ssse41(
+          input, CONVERT_TO_SHORTPTR(output), stride, tx_type, tx_size, eob,
+          bd);
+      break;
+    case H_DCT:
+    case H_ADST:
+    case H_FLIPADST:
+      highbd_inv_txfm2d_add_v_identity_ssse41(
+          input, CONVERT_TO_SHORTPTR(output), stride, tx_type, tx_size, eob,
+          bd);
+      break;
+    case IDTX:
+      highbd_inv_txfm2d_add_idtx_ssse41(input, CONVERT_TO_SHORTPTR(output),
+                                        stride, tx_type, tx_size, eob, bd);
+      break;
     default: assert(0); break;
   }
 }
 
+void av1_highbd_inv_txfm_add_4x8_sse4_1(const tran_low_t *input, uint8_t *dest,
+                                        int stride,
+                                        const TxfmParam *txfm_param) {
+  int bd = txfm_param->bd;
+  const TX_TYPE tx_type = txfm_param->tx_type;
+  const TX_SIZE tx_size = txfm_param->tx_size;
+  int eob = txfm_param->eob;
+  highbd_inv_txfm2d_add_4x8_sse41(input, CONVERT_TO_SHORTPTR(dest), stride,
+                                  tx_type, tx_size, eob, bd);
+}
+
+void av1_highbd_inv_txfm_add_8x4_sse4_1(const tran_low_t *input, uint8_t *dest,
+                                        int stride,
+                                        const TxfmParam *txfm_param) {
+  int bd = txfm_param->bd;
+  const TX_TYPE tx_type = txfm_param->tx_type;
+  const TX_SIZE tx_size = txfm_param->tx_size;
+  int eob = txfm_param->eob;
+  highbd_inv_txfm2d_add_8x4_sse41(input, CONVERT_TO_SHORTPTR(dest), stride,
+                                  tx_type, tx_size, eob, bd);
+}
+
+void av1_highbd_inv_txfm_add_4x16_sse4_1(const tran_low_t *input, uint8_t *dest,
+                                         int stride,
+                                         const TxfmParam *txfm_param) {
+  int bd = txfm_param->bd;
+  const TX_TYPE tx_type = txfm_param->tx_type;
+  const TX_SIZE tx_size = txfm_param->tx_size;
+  int eob = txfm_param->eob;
+  highbd_inv_txfm2d_add_4x16_sse4_1(input, CONVERT_TO_SHORTPTR(dest), stride,
+                                    tx_type, tx_size, eob, bd);
+}
+
+void av1_highbd_inv_txfm_add_16x4_sse4_1(const tran_low_t *input, uint8_t *dest,
+                                         int stride,
+                                         const TxfmParam *txfm_param) {
+  int bd = txfm_param->bd;
+  const TX_TYPE tx_type = txfm_param->tx_type;
+  const TX_SIZE tx_size = txfm_param->tx_size;
+  int eob = txfm_param->eob;
+  highbd_inv_txfm2d_add_16x4_sse4_1(input, CONVERT_TO_SHORTPTR(dest), stride,
+                                    tx_type, tx_size, eob, bd);
+}
+
 void av1_highbd_inv_txfm_add_sse4_1(const tran_low_t *input, uint8_t *dest,
                                     int stride, const TxfmParam *txfm_param) {
   assert(av1_ext_tx_used[txfm_param->tx_set_type][txfm_param->tx_type]);
   const TX_SIZE tx_size = txfm_param->tx_size;
   switch (tx_size) {
-    case TX_32X32:
-      av1_highbd_inv_txfm_add_32x32_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_16X16:
-      av1_highbd_inv_txfm_add_16x16_sse4_1(input, dest, stride, txfm_param);
-      break;
     case TX_8X8:
       av1_highbd_inv_txfm_add_8x8_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_4X8:
-      av1_highbd_inv_txfm_add_4x8(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x8_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_8X4:
-      av1_highbd_inv_txfm_add_8x4(input, dest, stride, txfm_param);
-      break;
-    case TX_8X16:
-      av1_highbd_inv_txfm_add_8x16_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_16X8:
-      av1_highbd_inv_txfm_add_16x8_sse4_1(input, dest, stride, txfm_param);
-      break;
-    case TX_16X32:
-      av1_highbd_inv_txfm_add_16x32(input, dest, stride, txfm_param);
-      break;
-    case TX_32X16:
-      av1_highbd_inv_txfm_add_32x16(input, dest, stride, txfm_param);
-      break;
-    case TX_32X64:
-      av1_highbd_inv_txfm_add_32x64(input, dest, stride, txfm_param);
-      break;
-    case TX_64X32:
-      av1_highbd_inv_txfm_add_64x32(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_8x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_4X4:
       av1_highbd_inv_txfm_add_4x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_16X4:
-      av1_highbd_inv_txfm_add_16x4(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_16x4_sse4_1(input, dest, stride, txfm_param);
       break;
     case TX_4X16:
-      av1_highbd_inv_txfm_add_4x16(input, dest, stride, txfm_param);
+      av1_highbd_inv_txfm_add_4x16_sse4_1(input, dest, stride, txfm_param);
       break;
-    case TX_8X32:
-      av1_highbd_inv_txfm_add_8x32(input, dest, stride, txfm_param);
-      break;
-    case TX_32X8:
-      av1_highbd_inv_txfm_add_32x8(input, dest, stride, txfm_param);
-      break;
-    case TX_64X64:
-    case TX_16X64:
-    case TX_64X16:
+    default:
       av1_highbd_inv_txfm2d_add_universe_sse4_1(
-          input, dest, stride, txfm_param->tx_type, txfm_param->tx_size,
-          txfm_param->eob, txfm_param->bd);
+          input, dest, stride, txfm_param->tx_type, tx_size, txfm_param->eob,
+          txfm_param->bd);
       break;
-    default: assert(0 && "Invalid transform size"); break;
   }
 }
diff --git a/third_party/aom/av1/common/x86/highbd_jnt_convolve_avx2.c b/third_party/aom/av1/common/x86/highbd_jnt_convolve_avx2.c
index e298cf6531..c5040c414e 100644
--- a/third_party/aom/av1/common/x86/highbd_jnt_convolve_avx2.c
+++ b/third_party/aom/av1/common/x86/highbd_jnt_convolve_avx2.c
@@ -22,7 +22,7 @@
 #include "aom_dsp/aom_filter.h"
 #include "av1/common/convolve.h"
 
-void av1_highbd_jnt_convolve_2d_copy_avx2(
+void av1_highbd_dist_wtd_convolve_2d_copy_avx2(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -38,7 +38,7 @@ void av1_highbd_jnt_convolve_2d_copy_avx2(
       FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
   const __m128i left_shift = _mm_cvtsi32_si128(bits);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
   const __m256i wt0 = _mm256_set1_epi32(w0);
@@ -78,15 +78,17 @@ void av1_highbd_jnt_convolve_2d_copy_avx2(
           const __m256i res_unsigned_lo =
               _mm256_add_epi32(res_32b_lo, offset_const);
 
-          const __m256i comp_avg_res_lo = highbd_comp_avg(
-              &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+          const __m256i comp_avg_res_lo =
+              highbd_comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1,
+                              use_dist_wtd_comp_avg);
 
           const __m256i res_32b_hi = _mm256_unpackhi_epi16(res, zero);
           const __m256i res_unsigned_hi =
               _mm256_add_epi32(res_32b_hi, offset_const);
 
-          const __m256i comp_avg_res_hi = highbd_comp_avg(
-              &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+          const __m256i comp_avg_res_hi =
+              highbd_comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1,
+                              use_dist_wtd_comp_avg);
 
           const __m256i round_result_lo = highbd_convolve_rounding(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
@@ -135,8 +137,9 @@ void av1_highbd_jnt_convolve_2d_copy_avx2(
             const __m256i res_unsigned_lo =
                 _mm256_add_epi32(res_32b, offset_const);
 
-            const __m256i comp_avg_res = highbd_comp_avg(
-                &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res =
+                highbd_comp_avg(&data_ref_0, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result = highbd_convolve_rounding(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -179,15 +182,17 @@ void av1_highbd_jnt_convolve_2d_copy_avx2(
             const __m256i res_unsigned_lo =
                 _mm256_add_epi32(res_32b_lo, offset_const);
 
-            const __m256i comp_avg_res_lo = highbd_comp_avg(
-                &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res_lo =
+                highbd_comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i res_32b_hi = _mm256_unpackhi_epi16(res, zero);
             const __m256i res_unsigned_hi =
                 _mm256_add_epi32(res_32b_hi, offset_const);
 
-            const __m256i comp_avg_res_hi = highbd_comp_avg(
-                &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res_hi =
+                highbd_comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result_lo =
                 highbd_convolve_rounding(&comp_avg_res_lo, &offset_const,
@@ -223,7 +228,7 @@ void av1_highbd_jnt_convolve_2d_copy_avx2(
   }
 }
 
-void av1_highbd_jnt_convolve_2d_avx2(
+void av1_highbd_dist_wtd_convolve_2d_avx2(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -244,7 +249,7 @@ void av1_highbd_jnt_convolve_2d_avx2(
 
   __m256i s[8], coeffs_y[4], coeffs_x[4];
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
 
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
@@ -364,8 +369,9 @@ void av1_highbd_jnt_convolve_2d_avx2(
 
             const __m256i data_ref_0 = _mm256_unpacklo_epi16(data_01, zero);
 
-            const __m256i comp_avg_res = highbd_comp_avg(
-                &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res =
+                highbd_comp_avg(&data_ref_0, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result = highbd_convolve_rounding(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -409,10 +415,12 @@ void av1_highbd_jnt_convolve_2d_avx2(
             const __m256i data_ref_0_lo = _mm256_unpacklo_epi16(data_01, zero);
             const __m256i data_ref_0_hi = _mm256_unpackhi_epi16(data_01, zero);
 
-            const __m256i comp_avg_res_lo = highbd_comp_avg(
-                &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
-            const __m256i comp_avg_res_hi = highbd_comp_avg(
-                &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res_lo =
+                highbd_comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
+            const __m256i comp_avg_res_hi =
+                highbd_comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result_lo =
                 highbd_convolve_rounding(&comp_avg_res_lo, &offset_const,
@@ -456,7 +464,7 @@ void av1_highbd_jnt_convolve_2d_avx2(
   }
 }
 
-void av1_highbd_jnt_convolve_x_avx2(
+void av1_highbd_dist_wtd_convolve_x_avx2(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -473,7 +481,7 @@ void av1_highbd_jnt_convolve_x_avx2(
   __m256i s[4], coeffs_x[4];
 
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
   const __m256i wt0 = _mm256_set1_epi32(w0);
@@ -548,7 +556,7 @@ void av1_highbd_jnt_convolve_x_avx2(
           const __m256i data_ref_0 = _mm256_unpacklo_epi16(data_01, zero);
 
           const __m256i comp_avg_res = highbd_comp_avg(
-              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_dist_wtd_comp_avg);
 
           const __m256i round_result = highbd_convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -588,10 +596,12 @@ void av1_highbd_jnt_convolve_x_avx2(
           const __m256i data_ref_0_lo = _mm256_unpacklo_epi16(data_01, zero);
           const __m256i data_ref_0_hi = _mm256_unpackhi_epi16(data_01, zero);
 
-          const __m256i comp_avg_res_lo = highbd_comp_avg(
-              &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
-          const __m256i comp_avg_res_hi = highbd_comp_avg(
-              &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+          const __m256i comp_avg_res_lo =
+              highbd_comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1,
+                              use_dist_wtd_comp_avg);
+          const __m256i comp_avg_res_hi =
+              highbd_comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1,
+                              use_dist_wtd_comp_avg);
 
           const __m256i round_result_lo = highbd_convolve_rounding(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
@@ -623,7 +633,7 @@ void av1_highbd_jnt_convolve_x_avx2(
   }
 }
 
-void av1_highbd_jnt_convolve_y_avx2(
+void av1_highbd_dist_wtd_convolve_y_avx2(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -640,7 +650,7 @@ void av1_highbd_jnt_convolve_y_avx2(
   int i, j;
   __m256i s[8], coeffs_y[4];
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
 
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
@@ -753,8 +763,9 @@ void av1_highbd_jnt_convolve_y_avx2(
 
             const __m256i data_ref_0 = _mm256_unpacklo_epi16(data_01, zero);
 
-            const __m256i comp_avg_res = highbd_comp_avg(
-                &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res =
+                highbd_comp_avg(&data_ref_0, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result = highbd_convolve_rounding(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -799,10 +810,12 @@ void av1_highbd_jnt_convolve_y_avx2(
             const __m256i data_ref_0_lo = _mm256_unpacklo_epi16(data_01, zero);
             const __m256i data_ref_0_hi = _mm256_unpackhi_epi16(data_01, zero);
 
-            const __m256i comp_avg_res_lo = highbd_comp_avg(
-                &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
-            const __m256i comp_avg_res_hi = highbd_comp_avg(
-                &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+            const __m256i comp_avg_res_lo =
+                highbd_comp_avg(&data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
+            const __m256i comp_avg_res_hi =
+                highbd_comp_avg(&data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1,
+                                use_dist_wtd_comp_avg);
 
             const __m256i round_result_lo =
                 highbd_convolve_rounding(&comp_avg_res_lo, &offset_const,
diff --git a/third_party/aom/av1/common/x86/highbd_jnt_convolve_sse4.c b/third_party/aom/av1/common/x86/highbd_jnt_convolve_sse4.c
index 1a29985b56..7fea36a8ac 100644
--- a/third_party/aom/av1/common/x86/highbd_jnt_convolve_sse4.c
+++ b/third_party/aom/av1/common/x86/highbd_jnt_convolve_sse4.c
@@ -17,7 +17,7 @@
 #include "aom_dsp/x86/convolve_sse2.h"
 #include "aom_dsp/x86/convolve_sse4_1.h"
 
-void av1_highbd_jnt_convolve_y_sse4_1(
+void av1_highbd_dist_wtd_convolve_y_sse4_1(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -33,7 +33,7 @@ void av1_highbd_jnt_convolve_y_sse4_1(
   assert(bits >= 0);
   int i, j;
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
 
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
@@ -121,10 +121,12 @@ void av1_highbd_jnt_convolve_y_sse4_1(
             const __m128i data_ref_0 = _mm_unpacklo_epi16(data_0, zero);
             const __m128i data_ref_1 = _mm_unpacklo_epi16(data_1, zero);
 
-            const __m128i comp_avg_res_0 = highbd_comp_avg_sse4_1(
-                &data_ref_0, &res_unsigned_lo_0, &wt0, &wt1, use_jnt_comp_avg);
-            const __m128i comp_avg_res_1 = highbd_comp_avg_sse4_1(
-                &data_ref_1, &res_unsigned_lo_1, &wt0, &wt1, use_jnt_comp_avg);
+            const __m128i comp_avg_res_0 =
+                highbd_comp_avg_sse4_1(&data_ref_0, &res_unsigned_lo_0, &wt0,
+                                       &wt1, use_dist_wtd_comp_avg);
+            const __m128i comp_avg_res_1 =
+                highbd_comp_avg_sse4_1(&data_ref_1, &res_unsigned_lo_1, &wt0,
+                                       &wt1, use_dist_wtd_comp_avg);
 
             const __m128i round_result_0 =
                 highbd_convolve_rounding_sse2(&comp_avg_res_0, &offset_const,
@@ -186,16 +188,16 @@ void av1_highbd_jnt_convolve_y_sse4_1(
 
             const __m128i comp_avg_res_lo_0 =
                 highbd_comp_avg_sse4_1(&data_ref_0_lo_0, &res_unsigned_lo_0,
-                                       &wt0, &wt1, use_jnt_comp_avg);
+                                       &wt0, &wt1, use_dist_wtd_comp_avg);
             const __m128i comp_avg_res_lo_1 =
                 highbd_comp_avg_sse4_1(&data_ref_0_lo_1, &res_unsigned_lo_1,
-                                       &wt0, &wt1, use_jnt_comp_avg);
+                                       &wt0, &wt1, use_dist_wtd_comp_avg);
             const __m128i comp_avg_res_hi_0 =
                 highbd_comp_avg_sse4_1(&data_ref_0_hi_0, &res_unsigned_hi_0,
-                                       &wt0, &wt1, use_jnt_comp_avg);
+                                       &wt0, &wt1, use_dist_wtd_comp_avg);
             const __m128i comp_avg_res_hi_1 =
                 highbd_comp_avg_sse4_1(&data_ref_0_hi_1, &res_unsigned_hi_1,
-                                       &wt0, &wt1, use_jnt_comp_avg);
+                                       &wt0, &wt1, use_dist_wtd_comp_avg);
 
             const __m128i round_result_lo_0 =
                 highbd_convolve_rounding_sse2(&comp_avg_res_lo_0, &offset_const,
@@ -257,7 +259,7 @@ void av1_highbd_jnt_convolve_y_sse4_1(
   }
 }
 
-void av1_highbd_jnt_convolve_x_sse4_1(
+void av1_highbd_dist_wtd_convolve_x_sse4_1(
     const uint16_t *src, int src_stride, uint16_t *dst0, int dst_stride0, int w,
     int h, const InterpFilterParams *filter_params_x,
     const InterpFilterParams *filter_params_y, const int subpel_x_q4,
@@ -274,7 +276,7 @@ void av1_highbd_jnt_convolve_x_sse4_1(
   __m128i s[4], coeffs_x[4];
 
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int w0 = conv_params->fwd_offset;
   const int w1 = conv_params->bck_offset;
   const __m128i wt0 = _mm_set1_epi32(w0);
@@ -339,7 +341,7 @@ void av1_highbd_jnt_convolve_x_sse4_1(
           const __m128i data_ref_0 = _mm_unpacklo_epi16(data_0, zero);
 
           const __m128i comp_avg_res = highbd_comp_avg_sse4_1(
-              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
+              &data_ref_0, &res_unsigned_lo, &wt0, &wt1, use_dist_wtd_comp_avg);
           const __m128i round_result = highbd_convolve_rounding_sse2(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
 
@@ -359,10 +361,12 @@ void av1_highbd_jnt_convolve_x_sse4_1(
           const __m128i data_ref_0_lo = _mm_unpacklo_epi16(data_0, zero);
           const __m128i data_ref_0_hi = _mm_unpackhi_epi16(data_0, zero);
 
-          const __m128i comp_avg_res_lo = highbd_comp_avg_sse4_1(
-              &data_ref_0_lo, &res_unsigned_lo, &wt0, &wt1, use_jnt_comp_avg);
-          const __m128i comp_avg_res_hi = highbd_comp_avg_sse4_1(
-              &data_ref_0_hi, &res_unsigned_hi, &wt0, &wt1, use_jnt_comp_avg);
+          const __m128i comp_avg_res_lo =
+              highbd_comp_avg_sse4_1(&data_ref_0_lo, &res_unsigned_lo, &wt0,
+                                     &wt1, use_dist_wtd_comp_avg);
+          const __m128i comp_avg_res_hi =
+              highbd_comp_avg_sse4_1(&data_ref_0_hi, &res_unsigned_hi, &wt0,
+                                     &wt1, use_dist_wtd_comp_avg);
 
           const __m128i round_result_lo = highbd_convolve_rounding_sse2(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
diff --git a/third_party/aom/av1/common/x86/highbd_txfm_utility_sse4.h b/third_party/aom/av1/common/x86/highbd_txfm_utility_sse4.h
index 6f24e5948a..5734810f52 100644
--- a/third_party/aom/av1/common/x86/highbd_txfm_utility_sse4.h
+++ b/third_party/aom/av1/common/x86/highbd_txfm_utility_sse4.h
@@ -75,13 +75,20 @@ static INLINE void transpose_16x16(const __m128i *in, __m128i *out) {
                 out[63]);
 }
 
-static INLINE void transpose_32x32(const __m128i *input, __m128i *output) {
-  for (int j = 0; j < 8; j++) {
-    for (int i = 0; i < 8; i++) {
-      TRANSPOSE_4X4(input[i * 32 + j + 0], input[i * 32 + j + 8],
-                    input[i * 32 + j + 16], input[i * 32 + j + 24],
-                    output[j * 32 + i + 0], output[j * 32 + i + 8],
-                    output[j * 32 + i + 16], output[j * 32 + i + 24]);
+static INLINE void transpose_8nx8n(const __m128i *input, __m128i *output,
+                                   const int width, const int height) {
+  const int numcol = height >> 2;
+  const int numrow = width >> 2;
+  for (int j = 0; j < numrow; j++) {
+    for (int i = 0; i < numcol; i++) {
+      TRANSPOSE_4X4(input[i * width + j + (numrow * 0)],
+                    input[i * width + j + (numrow * 1)],
+                    input[i * width + j + (numrow * 2)],
+                    input[i * width + j + (numrow * 3)],
+                    output[j * height + i + (numcol * 0)],
+                    output[j * height + i + (numcol * 1)],
+                    output[j * height + i + (numcol * 2)],
+                    output[j * height + i + (numcol * 3)]);
     }
   }
 }
diff --git a/third_party/aom/av1/common/x86/highbd_warp_plane_sse4.c b/third_party/aom/av1/common/x86/highbd_warp_plane_sse4.c
index 4bcab05645..3765c5ef19 100644
--- a/third_party/aom/av1/common/x86/highbd_warp_plane_sse4.c
+++ b/third_party/aom/av1/common/x86/highbd_warp_plane_sse4.c
@@ -537,7 +537,7 @@ void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref,
             __m128i *const dst16 = (__m128i *)&pred[(i + k + 4) * p_stride + j];
             __m128i p_32 = _mm_cvtepu16_epi32(_mm_loadl_epi64(p));
 
-            if (conv_params->use_jnt_comp_avg) {
+            if (conv_params->use_dist_wtd_comp_avg) {
               res_lo = _mm_add_epi32(_mm_mullo_epi32(p_32, wt0),
                                      _mm_mullo_epi32(res_lo, wt1));
               res_lo = _mm_srai_epi32(res_lo, DIST_PRECISION_BITS);
@@ -570,7 +570,7 @@ void av1_highbd_warp_affine_sse4_1(const int32_t *mat, const uint16_t *ref,
                   (__m128i *)&pred[(i + k + 4) * p_stride + j + 4];
               __m128i p4_32 = _mm_cvtepu16_epi32(_mm_loadl_epi64(p4));
 
-              if (conv_params->use_jnt_comp_avg) {
+              if (conv_params->use_dist_wtd_comp_avg) {
                 res_hi = _mm_add_epi32(_mm_mullo_epi32(p4_32, wt0),
                                        _mm_mullo_epi32(res_hi, wt1));
                 res_hi = _mm_srai_epi32(res_hi, DIST_PRECISION_BITS);
diff --git a/third_party/aom/av1/common/x86/jnt_convolve_avx2.c b/third_party/aom/av1/common/x86/jnt_convolve_avx2.c
index 9f2e2b4578..cc46ef35bc 100644
--- a/third_party/aom/av1/common/x86/jnt_convolve_avx2.c
+++ b/third_party/aom/av1/common/x86/jnt_convolve_avx2.c
@@ -35,12 +35,12 @@ static INLINE __m256i load_line2_avx2(const void *a, const void *b) {
       _mm256_castsi128_si256(_mm_loadu_si128((__m128i *)b)), 0x20);
 }
 
-void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
-                             int dst_stride0, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_x_avx2(const uint8_t *src, int src_stride,
+                                  uint8_t *dst0, int dst_stride0, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int bd = 8;
@@ -50,7 +50,7 @@ void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   const int bits = FILTER_BITS - conv_params->round_1;
   const __m256i wt = unpack_weights_avx2(conv_params);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int offset_0 =
       bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
@@ -97,7 +97,7 @@ void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
         const __m256i data_ref_0 =
             load_line2_avx2(&dst_data[j], &dst_data[j + dst_stride]);
         const __m256i comp_avg_res =
-            comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
         const __m256i round_result = convolve_rounding(
             &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -127,12 +127,12 @@ void av1_jnt_convolve_x_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   }
 }
 
-void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
-                             int dst_stride0, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_y_avx2(const uint8_t *src, int src_stride,
+                                  uint8_t *dst0, int dst_stride0, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int bd = 8;
@@ -146,7 +146,7 @@ void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_1);
   const __m256i wt = unpack_weights_avx2(conv_params);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int offset_0 =
       bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
@@ -230,8 +230,8 @@ void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
         if (do_average) {
           const __m256i data_ref_0 = load_line2_avx2(
               &dst[i * dst_stride + j], &dst[i * dst_stride + j + dst_stride]);
-          const __m256i comp_avg_res =
-              comp_avg(&data_ref_0, &res_lo_unsigned, &wt, use_jnt_comp_avg);
+          const __m256i comp_avg_res = comp_avg(&data_ref_0, &res_lo_unsigned,
+                                                &wt, use_dist_wtd_comp_avg);
 
           const __m256i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -289,11 +289,11 @@ void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
               load_line2_avx2(&dst[i * dst_stride + j + 8],
                               &dst[i * dst_stride + j + 8 + dst_stride]);
 
-          const __m256i comp_avg_res_lo =
-              comp_avg(&data_ref_0_lo, &res_lo_unsigned, &wt, use_jnt_comp_avg);
+          const __m256i comp_avg_res_lo = comp_avg(
+              &data_ref_0_lo, &res_lo_unsigned, &wt, use_dist_wtd_comp_avg);
 
-          const __m256i comp_avg_res_hi =
-              comp_avg(&data_ref_0_hi, &res_hi_unsigned, &wt, use_jnt_comp_avg);
+          const __m256i comp_avg_res_hi = comp_avg(
+              &data_ref_0_hi, &res_hi_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m256i round_result_lo = convolve_rounding(
               &comp_avg_res_lo, &offset_const, &rounding_const, rounding_shift);
@@ -337,12 +337,12 @@ void av1_jnt_convolve_y_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   }
 }
 
-void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
-                              int dst_stride0, int w, int h,
-                              const InterpFilterParams *filter_params_x,
-                              const InterpFilterParams *filter_params_y,
-                              const int subpel_x_q4, const int subpel_y_q4,
-                              ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_avx2(const uint8_t *src, int src_stride,
+                                   uint8_t *dst0, int dst_stride0, int w, int h,
+                                   const InterpFilterParams *filter_params_x,
+                                   const InterpFilterParams *filter_params_y,
+                                   const int subpel_x_q4, const int subpel_y_q4,
+                                   ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int bd = 8;
@@ -356,7 +356,7 @@ void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
   const __m256i wt = unpack_weights_avx2(conv_params);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int offset_0 =
       bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
@@ -448,8 +448,8 @@ void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
             const __m256i data_ref_0 =
                 load_line2_avx2(&dst[i * dst_stride + j],
                                 &dst[i * dst_stride + j + dst_stride]);
-            const __m256i comp_avg_res =
-                comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            const __m256i comp_avg_res = comp_avg(&data_ref_0, &res_unsigned,
+                                                  &wt, use_dist_wtd_comp_avg);
 
             const __m256i round_result = convolve_rounding(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -479,8 +479,8 @@ void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
                 load_line2_avx2(&dst[i * dst_stride + j],
                                 &dst[i * dst_stride + j + dst_stride]);
 
-            const __m256i comp_avg_res =
-                comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            const __m256i comp_avg_res = comp_avg(&data_ref_0, &res_unsigned,
+                                                  &wt, use_dist_wtd_comp_avg);
 
             const __m256i round_result = convolve_rounding(
                 &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -517,12 +517,11 @@ void av1_jnt_convolve_2d_avx2(const uint8_t *src, int src_stride, uint8_t *dst0,
   }
 }
 
-void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride,
-                                   uint8_t *dst0, int dst_stride0, int w, int h,
-                                   const InterpFilterParams *filter_params_x,
-                                   const InterpFilterParams *filter_params_y,
-                                   const int subpel_x_q4, const int subpel_y_q4,
-                                   ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_copy_avx2(
+    const uint8_t *src, int src_stride, uint8_t *dst0, int dst_stride0, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params) {
   const int bd = 8;
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
@@ -535,7 +534,7 @@ void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride,
       FILTER_BITS * 2 - conv_params->round_1 - conv_params->round_0;
   const __m128i left_shift = _mm_cvtsi32_si128(bits);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const __m256i wt = unpack_weights_avx2(conv_params);
   const __m256i zero = _mm256_setzero_si256();
 
@@ -562,7 +561,7 @@ void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride,
               _mm256_loadu_si256((__m256i *)(&dst[i * dst_stride + j]));
 
           const __m256i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m256i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -600,7 +599,7 @@ void av1_jnt_convolve_2d_copy_avx2(const uint8_t *src, int src_stride,
           const __m256i data_ref_0 = load_line2_avx2(
               &dst[i * dst_stride + j], &dst[i * dst_stride + j + dst_stride]);
           const __m256i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m256i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
diff --git a/third_party/aom/av1/common/x86/jnt_convolve_sse2.c b/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
index 87dc3242e8..641cd02457 100644
--- a/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
+++ b/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
@@ -16,12 +16,12 @@
 #include "aom_dsp/aom_filter.h"
 #include "aom_dsp/x86/convolve_sse2.h"
 
-void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
-                             int dst_stride0, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_x_sse2(const uint8_t *src, int src_stride,
+                                  uint8_t *dst0, int dst_stride0, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   const int bd = 8;
   CONV_BUF_TYPE *dst = conv_params->dst;
   const int dst_stride = conv_params->dst_stride;
@@ -37,7 +37,7 @@ void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
   const __m128i wt1 = _mm_set1_epi16(w1);
   const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int offset_0 =
       bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
@@ -77,7 +77,7 @@ void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
         const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
 
         const __m128i comp_avg_res =
-            comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
         const __m128i round_result = convolve_rounding(
             &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -134,7 +134,7 @@ void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
               _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
 
           const __m128i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -150,12 +150,12 @@ void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
   }
 }
 
-void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
-                             int dst_stride0, int w, int h,
-                             const InterpFilterParams *filter_params_x,
-                             const InterpFilterParams *filter_params_y,
-                             const int subpel_x_q4, const int subpel_y_q4,
-                             ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride,
+                                  uint8_t *dst0, int dst_stride0, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params) {
   const int bd = 8;
   CONV_BUF_TYPE *dst = conv_params->dst;
   const int dst_stride = conv_params->dst_stride;
@@ -167,7 +167,7 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
   const __m128i wt1 = _mm_set1_epi16(conv_params->bck_offset);
   const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const int offset_0 =
       bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
   const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
@@ -225,7 +225,7 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
         const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
 
         const __m128i comp_avg_res =
-            comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
         const __m128i round_result = convolve_rounding(
             &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -254,7 +254,7 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
         const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
 
         const __m128i comp_avg_res =
-            comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+            comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
         const __m128i round_result = convolve_rounding(
             &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -331,7 +331,7 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
               _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
 
           const __m128i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -360,7 +360,7 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
               _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
 
           const __m128i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
@@ -383,3 +383,233 @@ void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
     } while (j < w);
   }
 }
+
+void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride,
+                                   uint8_t *dst0, int dst_stride0, int w, int h,
+                                   const InterpFilterParams *filter_params_x,
+                                   const InterpFilterParams *filter_params_y,
+                                   const int subpel_x_q4, const int subpel_y_q4,
+                                   ConvolveParams *conv_params) {
+  CONV_BUF_TYPE *dst = conv_params->dst;
+  int dst_stride = conv_params->dst_stride;
+  const int bd = 8;
+
+  DECLARE_ALIGNED(16, int16_t,
+                  im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE]);
+  int im_h = h + filter_params_y->taps - 1;
+  int im_stride = MAX_SB_SIZE;
+  int i, j;
+  const int fo_vert = filter_params_y->taps / 2 - 1;
+  const int fo_horiz = filter_params_x->taps / 2 - 1;
+  const int do_average = conv_params->do_average;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
+  const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
+
+  const __m128i zero = _mm_setzero_si128();
+
+  const int w0 = conv_params->fwd_offset;
+  const int w1 = conv_params->bck_offset;
+  const __m128i wt0 = _mm_set1_epi16(w0);
+  const __m128i wt1 = _mm_set1_epi16(w1);
+  const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
+
+  const int offset_0 =
+      bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
+  const __m128i offset_const = _mm_set1_epi16(offset);
+  const int rounding_shift =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const __m128i rounding_const = _mm_set1_epi16((1 << rounding_shift) >> 1);
+
+  /* Horizontal filter */
+  {
+    const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
+        filter_params_x, subpel_x_q4 & SUBPEL_MASK);
+    const __m128i coeffs_x = _mm_loadu_si128((__m128i *)x_filter);
+
+    // coeffs 0 1 0 1 2 3 2 3
+    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_x, coeffs_x);
+    // coeffs 4 5 4 5 6 7 6 7
+    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_x, coeffs_x);
+
+    // coeffs 0 1 0 1 0 1 0 1
+    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
+    // coeffs 2 3 2 3 2 3 2 3
+    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
+    // coeffs 4 5 4 5 4 5 4 5
+    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
+    // coeffs 6 7 6 7 6 7 6 7
+    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);
+
+    const __m128i round_const = _mm_set1_epi32(
+        ((1 << conv_params->round_0) >> 1) + (1 << (bd + FILTER_BITS - 1)));
+    const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_0);
+
+    for (i = 0; i < im_h; ++i) {
+      for (j = 0; j < w; j += 8) {
+        __m128i temp_lo, temp_hi;
+        const __m128i data =
+            _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
+
+        const __m128i src_lo = _mm_unpacklo_epi8(data, zero);
+        const __m128i src_hi = _mm_unpackhi_epi8(data, zero);
+
+        // Filter even-index pixels
+        const __m128i res_0 = _mm_madd_epi16(src_lo, coeff_01);
+        temp_lo = _mm_srli_si128(src_lo, 4);
+        temp_hi = _mm_slli_si128(src_hi, 12);
+        const __m128i src_2 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
+        temp_lo = _mm_srli_si128(src_lo, 8);
+        temp_hi = _mm_slli_si128(src_hi, 8);
+        const __m128i src_4 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
+        temp_lo = _mm_srli_si128(src_lo, 12);
+        temp_hi = _mm_slli_si128(src_hi, 4);
+        const __m128i src_6 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);
+
+        __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
+                                         _mm_add_epi32(res_2, res_6));
+        res_even =
+            _mm_sra_epi32(_mm_add_epi32(res_even, round_const), round_shift);
+
+        // Filter odd-index pixels
+        temp_lo = _mm_srli_si128(src_lo, 2);
+        temp_hi = _mm_slli_si128(src_hi, 14);
+        const __m128i src_1 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
+        temp_lo = _mm_srli_si128(src_lo, 6);
+        temp_hi = _mm_slli_si128(src_hi, 10);
+        const __m128i src_3 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
+        temp_lo = _mm_srli_si128(src_lo, 10);
+        temp_hi = _mm_slli_si128(src_hi, 6);
+        const __m128i src_5 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
+        temp_lo = _mm_srli_si128(src_lo, 14);
+        temp_hi = _mm_slli_si128(src_hi, 2);
+        const __m128i src_7 = _mm_or_si128(temp_hi, temp_lo);
+        const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);
+
+        __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
+                                        _mm_add_epi32(res_3, res_7));
+        res_odd =
+            _mm_sra_epi32(_mm_add_epi32(res_odd, round_const), round_shift);
+
+        // Pack in the column order 0, 2, 4, 6, 1, 3, 5, 7
+        __m128i res = _mm_packs_epi32(res_even, res_odd);
+        _mm_store_si128((__m128i *)&im_block[i * im_stride + j], res);
+      }
+    }
+  }
+
+  /* Vertical filter */
+  {
+    const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
+        filter_params_y, subpel_y_q4 & SUBPEL_MASK);
+    const __m128i coeffs_y = _mm_loadu_si128((__m128i *)y_filter);
+
+    // coeffs 0 1 0 1 2 3 2 3
+    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_y, coeffs_y);
+    // coeffs 4 5 4 5 6 7 6 7
+    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_y, coeffs_y);
+
+    // coeffs 0 1 0 1 0 1 0 1
+    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
+    // coeffs 2 3 2 3 2 3 2 3
+    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
+    // coeffs 4 5 4 5 4 5 4 5
+    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
+    // coeffs 6 7 6 7 6 7 6 7
+    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);
+
+    const __m128i round_const = _mm_set1_epi32(
+        ((1 << conv_params->round_1) >> 1) -
+        (1 << (bd + 2 * FILTER_BITS - conv_params->round_0 - 1)));
+    const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_1);
+
+    for (i = 0; i < h; ++i) {
+      for (j = 0; j < w; j += 8) {
+        // Filter even-index pixels
+        const int16_t *data = &im_block[i * im_stride + j];
+        const __m128i src_0 =
+            _mm_unpacklo_epi16(*(__m128i *)(data + 0 * im_stride),
+                               *(__m128i *)(data + 1 * im_stride));
+        const __m128i src_2 =
+            _mm_unpacklo_epi16(*(__m128i *)(data + 2 * im_stride),
+                               *(__m128i *)(data + 3 * im_stride));
+        const __m128i src_4 =
+            _mm_unpacklo_epi16(*(__m128i *)(data + 4 * im_stride),
+                               *(__m128i *)(data + 5 * im_stride));
+        const __m128i src_6 =
+            _mm_unpacklo_epi16(*(__m128i *)(data + 6 * im_stride),
+                               *(__m128i *)(data + 7 * im_stride));
+
+        const __m128i res_0 = _mm_madd_epi16(src_0, coeff_01);
+        const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
+        const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
+        const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);
+
+        const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
+                                               _mm_add_epi32(res_4, res_6));
+
+        // Filter odd-index pixels
+        const __m128i src_1 =
+            _mm_unpackhi_epi16(*(__m128i *)(data + 0 * im_stride),
+                               *(__m128i *)(data + 1 * im_stride));
+        const __m128i src_3 =
+            _mm_unpackhi_epi16(*(__m128i *)(data + 2 * im_stride),
+                               *(__m128i *)(data + 3 * im_stride));
+        const __m128i src_5 =
+            _mm_unpackhi_epi16(*(__m128i *)(data + 4 * im_stride),
+                               *(__m128i *)(data + 5 * im_stride));
+        const __m128i src_7 =
+            _mm_unpackhi_epi16(*(__m128i *)(data + 6 * im_stride),
+                               *(__m128i *)(data + 7 * im_stride));
+
+        const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
+        const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
+        const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
+        const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);
+
+        const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
+                                              _mm_add_epi32(res_5, res_7));
+
+        // Rearrange pixels back into the order 0 ... 7
+        const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
+        const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
+
+        const __m128i res_lo_round =
+            _mm_sra_epi32(_mm_add_epi32(res_lo, round_const), round_shift);
+        const __m128i res_hi_round =
+            _mm_sra_epi32(_mm_add_epi32(res_hi, round_const), round_shift);
+
+        const __m128i res_16b = _mm_packs_epi32(res_lo_round, res_hi_round);
+        const __m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
+
+        // Accumulate values into the destination buffer
+        if (do_average) {
+          const __m128i data_ref_0 =
+              _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
+
+          const __m128i comp_avg_res =
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
+
+          const __m128i round_result = convolve_rounding(
+              &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
+
+          const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
+
+          if (w > 4)
+            _mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_8);
+          else
+            *(uint32_t *)(&dst0[i * dst_stride0 + j]) =
+                _mm_cvtsi128_si32(res_8);
+        } else {
+          _mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_unsigned);
+        }
+      }
+    }
+  }
+}
diff --git a/third_party/aom/av1/common/x86/jnt_convolve_ssse3.c b/third_party/aom/av1/common/x86/jnt_convolve_ssse3.c
index 822772782b..9aeab29424 100644
--- a/third_party/aom/av1/common/x86/jnt_convolve_ssse3.c
+++ b/third_party/aom/av1/common/x86/jnt_convolve_ssse3.c
@@ -16,12 +16,11 @@
 #include "aom_dsp/aom_filter.h"
 #include "aom_dsp/x86/convolve_sse2.h"
 
-void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride,
-                               uint8_t *dst0, int dst_stride0, int w, int h,
-                               const InterpFilterParams *filter_params_x,
-                               const InterpFilterParams *filter_params_y,
-                               const int subpel_x_q4, const int subpel_y_q4,
-                               ConvolveParams *conv_params) {
+void av1_dist_wtd_convolve_2d_ssse3(
+    const uint8_t *src, int src_stride, uint8_t *dst0, int dst_stride0, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params) {
   CONV_BUF_TYPE *dst = conv_params->dst;
   int dst_stride = conv_params->dst_stride;
   const int bd = 8;
@@ -34,7 +33,7 @@ void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride,
   const int fo_vert = filter_params_y->taps / 2 - 1;
   const int fo_horiz = filter_params_x->taps / 2 - 1;
   const int do_average = conv_params->do_average;
-  const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
+  const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
   const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
 
   const __m128i zero = _mm_setzero_si128();
@@ -211,7 +210,7 @@ void av1_jnt_convolve_2d_ssse3(const uint8_t *src, int src_stride,
               _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
 
           const __m128i comp_avg_res =
-              comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
+              comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
 
           const __m128i round_result = convolve_rounding(
               &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
diff --git a/third_party/aom/av1/common/x86/warp_plane_sse4.c b/third_party/aom/av1/common/x86/warp_plane_sse4.c
index b810cea2e5..4532d17ced 100644
--- a/third_party/aom/av1/common/x86/warp_plane_sse4.c
+++ b/third_party/aom/av1/common/x86/warp_plane_sse4.c
@@ -577,7 +577,7 @@ static INLINE void store_vertical_filter_output(
       __m128i *const dst8 = (__m128i *)&pred[(i + k + 4) * p_stride + j];
       const __m128i p_16 = _mm_loadl_epi64(p);
 
-      if (conv_params->use_jnt_comp_avg) {
+      if (conv_params->use_dist_wtd_comp_avg) {
         const __m128i p_16_lo = _mm_unpacklo_epi16(p_16, temp_lo_16);
         const __m128i wt_res_lo = _mm_madd_epi16(p_16_lo, *wt);
         const __m128i shifted_32 =
@@ -610,7 +610,7 @@ static INLINE void store_vertical_filter_output(
             (__m128i *)&pred[(i + k + 4) * p_stride + j + 4];
         const __m128i p4_16 = _mm_loadl_epi64(p4);
 
-        if (conv_params->use_jnt_comp_avg) {
+        if (conv_params->use_dist_wtd_comp_avg) {
           const __m128i p_16_hi = _mm_unpacklo_epi16(p4_16, temp_hi_16);
           const __m128i wt_res_hi = _mm_madd_epi16(p_16_hi, *wt);
           const __m128i shifted_32 =
diff --git a/third_party/aom/av1/common/x86/wiener_convolve_avx2.c b/third_party/aom/av1/common/x86/wiener_convolve_avx2.c
index 87a6e12396..1f13e2f92e 100644
--- a/third_party/aom/av1/common/x86/wiener_convolve_avx2.c
+++ b/third_party/aom/av1/common/x86/wiener_convolve_avx2.c
@@ -17,6 +17,7 @@
 #include "av1/common/convolve.h"
 #include "aom_dsp/aom_dsp_common.h"
 #include "aom_dsp/aom_filter.h"
+#include "aom_dsp/x86/convolve_avx2.h"
 #include "aom_dsp/x86/synonyms.h"
 #include "aom_dsp/x86/synonyms_avx2.h"
 
@@ -25,236 +26,207 @@
 // on the left.
 // A row of, say, 8-bit pixels with values p0, p1, p2, ..., p30, p31 will be
 // loaded and stored as [ p31 ... p17 p16 ][ p15 ... p1 p0 ].
+
+// Exploiting the range of wiener filter coefficients,
+// horizontal filtering can be done in 16 bit intermediate precision.
+// The details are as follows :
+// Consider the horizontal wiener filter coefficients of the following form :
+//      [C0, C1, C2, 2^(FILTER_BITS) -2 * (C0 + C1 + C2), C2, C1, C0]
+// Subtracting  2^(FILTER_BITS) from the centre tap we get the following  :
+//      [C0, C1, C2,     -2 * (C0 + C1 + C2),             C2, C1, C0]
+// The sum of the product "C0 * p0 + C1 * p1 + C2 * p2 -2 * (C0 + C1 + C2) * p3
+// + C2 * p4 + C1 * p5 + C0 * p6" would be in the range of signed 16 bit
+// precision. Finally, after rounding the above result by round_0, we multiply
+// the centre pixel by 2^(FILTER_BITS - round_0) and add it to get the
+// horizontal filter output.
+
 void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride,
                                       uint8_t *dst, ptrdiff_t dst_stride,
                                       const int16_t *filter_x, int x_step_q4,
                                       const int16_t *filter_y, int y_step_q4,
                                       int w, int h,
                                       const ConvolveParams *conv_params) {
-  const int bd = 8;
   assert(x_step_q4 == 16 && y_step_q4 == 16);
   assert(!(w & 7));
   (void)x_step_q4;
   (void)y_step_q4;
 
-  DECLARE_ALIGNED(32, uint16_t,
-                  temp[(MAX_SB_SIZE + SUBPEL_TAPS - 1) * MAX_SB_SIZE]);
-  int intermediate_height = h + SUBPEL_TAPS - 2;
-  memset(temp + (intermediate_height * MAX_SB_SIZE), 0, MAX_SB_SIZE);
-  const int center_tap = ((SUBPEL_TAPS - 1) / 2);
+  DECLARE_ALIGNED(32, int16_t, im_block[(MAX_SB_SIZE + SUBPEL_TAPS) * 8]);
+  int im_h = h + SUBPEL_TAPS - 2;
+  int im_stride = 8;
+  memset(im_block + (im_h * im_stride), 0, MAX_SB_SIZE);
+  int i, j;
+  const int center_tap = (SUBPEL_TAPS - 1) / 2;
   const uint8_t *const src_ptr = src - center_tap * src_stride - center_tap;
 
-  const __m128i zero_128 = _mm_setzero_si128();
-  const __m256i zero_256 = _mm256_setzero_si256();
+  __m256i filt[4], coeffs_h[4], coeffs_v[4], filt_center;
+
+  assert(conv_params->round_0 > 0);
+
+  filt[0] = _mm256_load_si256((__m256i const *)filt1_global_avx2);
+  filt[1] = _mm256_load_si256((__m256i const *)filt2_global_avx2);
+  filt[2] = _mm256_load_si256((__m256i const *)filt3_global_avx2);
+  filt[3] = _mm256_load_si256((__m256i const *)filt4_global_avx2);
+
+  filt_center = _mm256_load_si256((__m256i const *)filt_center_global_avx2);
+
+  const __m128i coeffs_x = _mm_loadu_si128((__m128i *)filter_x);
+  const __m256i filter_coeffs_x = _mm256_broadcastsi128_si256(coeffs_x);
+
+  // coeffs 0 1 0 1 0 1 0 1
+  coeffs_h[0] =
+      _mm256_shuffle_epi8(filter_coeffs_x, _mm256_set1_epi16(0x0200u));
+  // coeffs 2 3 2 3 2 3 2 3
+  coeffs_h[1] =
+      _mm256_shuffle_epi8(filter_coeffs_x, _mm256_set1_epi16(0x0604u));
+  // coeffs 4 5 4 5 4 5 4 5
+  coeffs_h[2] =
+      _mm256_shuffle_epi8(filter_coeffs_x, _mm256_set1_epi16(0x0a08u));
+  // coeffs 6 7 6 7 6 7 6 7
+  coeffs_h[3] =
+      _mm256_shuffle_epi8(filter_coeffs_x, _mm256_set1_epi16(0x0e0cu));
+
+  const __m256i round_const_h =
+      _mm256_set1_epi16((1 << (conv_params->round_0 - 1)));
+  const __m128i round_shift_h = _mm_cvtsi32_si128(conv_params->round_0);
 
   // Add an offset to account for the "add_src" part of the convolve function.
-  const __m128i offset = _mm_insert_epi16(zero_128, 1 << FILTER_BITS, 3);
+  const __m128i zero_128 = _mm_setzero_si128();
+  const __m128i offset_0 = _mm_insert_epi16(zero_128, 1 << FILTER_BITS, 3);
+  const __m128i coeffs_y = _mm_add_epi16(xx_loadu_128(filter_y), offset_0);
 
-  const __m256i clamp_low = zero_256;
-  const __m256i clamp_high =
-      _mm256_set1_epi16(WIENER_CLAMP_LIMIT(conv_params->round_0, bd) - 1);
+  const __m256i filter_coeffs_y = _mm256_broadcastsi128_si256(coeffs_y);
 
-  /* Horizontal filter */
-  {
-    // coeffs [ f7 f6 f5 f4 f3 f2 f1 f0 ]
-    const __m128i coeffs_x = _mm_add_epi16(xx_loadu_128(filter_x), offset);
+  // coeffs 0 1 0 1 0 1 0 1
+  coeffs_v[0] = _mm256_shuffle_epi32(filter_coeffs_y, 0x00);
+  // coeffs 2 3 2 3 2 3 2 3
+  coeffs_v[1] = _mm256_shuffle_epi32(filter_coeffs_y, 0x55);
+  // coeffs 4 5 4 5 4 5 4 5
+  coeffs_v[2] = _mm256_shuffle_epi32(filter_coeffs_y, 0xaa);
+  // coeffs 6 7 6 7 6 7 6 7
+  coeffs_v[3] = _mm256_shuffle_epi32(filter_coeffs_y, 0xff);
 
-    // coeffs [ f3 f2 f3 f2 f1 f0 f1 f0 ]
-    const __m128i coeffs_0123 = _mm_unpacklo_epi32(coeffs_x, coeffs_x);
-    // coeffs [ f7 f6 f7 f6 f5 f4 f5 f4 ]
-    const __m128i coeffs_4567 = _mm_unpackhi_epi32(coeffs_x, coeffs_x);
+  const __m256i round_const_v =
+      _mm256_set1_epi32((1 << (conv_params->round_1 - 1)));
+  const __m128i round_shift_v = _mm_cvtsi32_si128(conv_params->round_1);
 
-    // coeffs [ f1 f0 f1 f0 f1 f0 f1 f0 ]
-    const __m128i coeffs_01_128 = _mm_unpacklo_epi64(coeffs_0123, coeffs_0123);
-    // coeffs [ f3 f2 f3 f2 f3 f2 f3 f2 ]
-    const __m128i coeffs_23_128 = _mm_unpackhi_epi64(coeffs_0123, coeffs_0123);
-    // coeffs [ f5 f4 f5 f4 f5 f4 f5 f4 ]
-    const __m128i coeffs_45_128 = _mm_unpacklo_epi64(coeffs_4567, coeffs_4567);
-    // coeffs [ f7 f6 f7 f6 f7 f6 f7 f6 ]
-    const __m128i coeffs_67_128 = _mm_unpackhi_epi64(coeffs_4567, coeffs_4567);
+  for (j = 0; j < w; j += 8) {
+    for (i = 0; i < im_h; i += 2) {
+      __m256i data = _mm256_castsi128_si256(
+          _mm_loadu_si128((__m128i *)&src_ptr[(i * src_stride) + j]));
 
-    // coeffs [ f1 f0 f1 f0 f1 f0 f1 f0 ][ f1 f0 f1 f0 f1 f0 f1 f0 ]
-    const __m256i coeffs_01 = yy_set_m128i(coeffs_01_128, coeffs_01_128);
-    // coeffs [ f3 f2 f3 f2 f3 f2 f3 f2 ][ f3 f2 f3 f2 f3 f2 f3 f2 ]
-    const __m256i coeffs_23 = yy_set_m128i(coeffs_23_128, coeffs_23_128);
-    // coeffs [ f5 f4 f5 f4 f5 f4 f5 f4 ][ f5 f4 f5 f4 f5 f4 f5 f4 ]
-    const __m256i coeffs_45 = yy_set_m128i(coeffs_45_128, coeffs_45_128);
-    // coeffs [ f7 f6 f7 f6 f7 f6 f7 f6 ][ f7 f6 f7 f6 f7 f6 f7 f6 ]
-    const __m256i coeffs_67 = yy_set_m128i(coeffs_67_128, coeffs_67_128);
+      // Load the next line
+      if (i + 1 < im_h)
+        data = _mm256_inserti128_si256(
+            data,
+            _mm_loadu_si128(
+                (__m128i *)&src_ptr[(i * src_stride) + j + src_stride]),
+            1);
 
-    const __m256i round_const = _mm256_set1_epi32(
-        (1 << (conv_params->round_0 - 1)) + (1 << (bd + FILTER_BITS - 1)));
+      __m256i res = convolve_lowbd_x(data, coeffs_h, filt);
 
-    for (int i = 0; i < intermediate_height; ++i) {
-      for (int j = 0; j < w; j += 16) {
-        const uint8_t *data_ij = src_ptr + i * src_stride + j;
+      res =
+          _mm256_sra_epi16(_mm256_add_epi16(res, round_const_h), round_shift_h);
 
-        // Load 8-bit src data
-        const __m128i data_0 = xx_loadu_128(data_ij + 0);
-        const __m128i data_1 = xx_loadu_128(data_ij + 1);
-        const __m128i data_2 = xx_loadu_128(data_ij + 2);
-        const __m128i data_3 = xx_loadu_128(data_ij + 3);
-        const __m128i data_4 = xx_loadu_128(data_ij + 4);
-        const __m128i data_5 = xx_loadu_128(data_ij + 5);
-        const __m128i data_6 = xx_loadu_128(data_ij + 6);
-        const __m128i data_7 = xx_loadu_128(data_ij + 7);
+      __m256i data_0 = _mm256_shuffle_epi8(data, filt_center);
 
-        // (Zero-)Extend 8-bit data to 16-bit data
-        const __m256i src_0 = _mm256_cvtepu8_epi16(data_0);
-        const __m256i src_1 = _mm256_cvtepu8_epi16(data_1);
-        const __m256i src_2 = _mm256_cvtepu8_epi16(data_2);
-        const __m256i src_3 = _mm256_cvtepu8_epi16(data_3);
-        const __m256i src_4 = _mm256_cvtepu8_epi16(data_4);
-        const __m256i src_5 = _mm256_cvtepu8_epi16(data_5);
-        const __m256i src_6 = _mm256_cvtepu8_epi16(data_6);
-        const __m256i src_7 = _mm256_cvtepu8_epi16(data_7);
+      // multiply the center pixel by 2^(FILTER_BITS - round_0) and add it to
+      // the result
+      data_0 = _mm256_slli_epi16(data_0, FILTER_BITS - conv_params->round_0);
+      res = _mm256_add_epi16(res, data_0);
 
-        // Multiply src data by filter coeffs and sum pairs
-        const __m256i res_0 = _mm256_madd_epi16(src_0, coeffs_01);
-        const __m256i res_1 = _mm256_madd_epi16(src_1, coeffs_01);
-        const __m256i res_2 = _mm256_madd_epi16(src_2, coeffs_23);
-        const __m256i res_3 = _mm256_madd_epi16(src_3, coeffs_23);
-        const __m256i res_4 = _mm256_madd_epi16(src_4, coeffs_45);
-        const __m256i res_5 = _mm256_madd_epi16(src_5, coeffs_45);
-        const __m256i res_6 = _mm256_madd_epi16(src_6, coeffs_67);
-        const __m256i res_7 = _mm256_madd_epi16(src_7, coeffs_67);
-
-        // Calculate scalar product for even- and odd-indices separately,
-        // increasing to 32-bit precision
-        const __m256i res_even_sum = _mm256_add_epi32(
-            _mm256_add_epi32(res_0, res_4), _mm256_add_epi32(res_2, res_6));
-        const __m256i res_odd_sum = _mm256_add_epi32(
-            _mm256_add_epi32(res_1, res_5), _mm256_add_epi32(res_3, res_7));
-
-        const __m256i res_even = _mm256_srai_epi32(
-            _mm256_add_epi32(res_even_sum, round_const), conv_params->round_0);
-        const __m256i res_odd = _mm256_srai_epi32(
-            _mm256_add_epi32(res_odd_sum, round_const), conv_params->round_0);
-
-        // Reduce to 16-bit precision and pack even- and odd-index results
-        // back into one register. The _mm256_packs_epi32 intrinsic returns
-        // a register with the pixels ordered as follows:
-        // [ 15 13 11 9 14 12 10 8 ] [ 7 5 3 1 6 4 2 0 ]
-        const __m256i res = _mm256_packs_epi32(res_even, res_odd);
-        const __m256i res_clamped =
-            _mm256_min_epi16(_mm256_max_epi16(res, clamp_low), clamp_high);
-
-        // Store in a temporary array
-        yy_storeu_256(temp + i * MAX_SB_SIZE + j, res_clamped);
-      }
+      _mm256_store_si256((__m256i *)&im_block[i * im_stride], res);
     }
-  }
 
-  /* Vertical filter */
-  {
-    // coeffs [ g7 g6 g5 g4 g3 g2 g1 g0 ]
-    const __m128i coeffs_y = _mm_add_epi16(xx_loadu_128(filter_y), offset);
+    /* Vertical filter */
+    {
+      __m256i src_0 = _mm256_loadu_si256((__m256i *)(im_block + 0 * im_stride));
+      __m256i src_1 = _mm256_loadu_si256((__m256i *)(im_block + 1 * im_stride));
+      __m256i src_2 = _mm256_loadu_si256((__m256i *)(im_block + 2 * im_stride));
+      __m256i src_3 = _mm256_loadu_si256((__m256i *)(im_block + 3 * im_stride));
+      __m256i src_4 = _mm256_loadu_si256((__m256i *)(im_block + 4 * im_stride));
+      __m256i src_5 = _mm256_loadu_si256((__m256i *)(im_block + 5 * im_stride));
 
-    // coeffs [ g3 g2 g3 g2 g1 g0 g1 g0 ]
-    const __m128i coeffs_0123 = _mm_unpacklo_epi32(coeffs_y, coeffs_y);
-    // coeffs [ g7 g6 g7 g6 g5 g4 g5 g4 ]
-    const __m128i coeffs_4567 = _mm_unpackhi_epi32(coeffs_y, coeffs_y);
+      __m256i s[8];
+      s[0] = _mm256_unpacklo_epi16(src_0, src_1);
+      s[1] = _mm256_unpacklo_epi16(src_2, src_3);
+      s[2] = _mm256_unpacklo_epi16(src_4, src_5);
 
-    // coeffs [ g1 g0 g1 g0 g1 g0 g1 g0 ]
-    const __m128i coeffs_01_128 = _mm_unpacklo_epi64(coeffs_0123, coeffs_0123);
-    // coeffs [ g3 g2 g3 g2 g3 g2 g3 g2 ]
-    const __m128i coeffs_23_128 = _mm_unpackhi_epi64(coeffs_0123, coeffs_0123);
-    // coeffs [ g5 g4 g5 g4 g5 g4 g5 g4 ]
-    const __m128i coeffs_45_128 = _mm_unpacklo_epi64(coeffs_4567, coeffs_4567);
-    // coeffs [ g7 g6 g7 g6 g7 g6 g7 g6 ]
-    const __m128i coeffs_67_128 = _mm_unpackhi_epi64(coeffs_4567, coeffs_4567);
+      s[4] = _mm256_unpackhi_epi16(src_0, src_1);
+      s[5] = _mm256_unpackhi_epi16(src_2, src_3);
+      s[6] = _mm256_unpackhi_epi16(src_4, src_5);
 
-    // coeffs [ g1 g0 g1 g0 g1 g0 g1 g0 ][ g1 g0 g1 g0 g1 g0 g1 g0 ]
-    const __m256i coeffs_01 = yy_set_m128i(coeffs_01_128, coeffs_01_128);
-    // coeffs [ g3 g2 g3 g2 g3 g2 g3 g2 ][ g3 g2 g3 g2 g3 g2 g3 g2 ]
-    const __m256i coeffs_23 = yy_set_m128i(coeffs_23_128, coeffs_23_128);
-    // coeffs [ g5 g4 g5 g4 g5 g4 g5 g4 ][ g5 g4 g5 g4 g5 g4 g5 g4 ]
-    const __m256i coeffs_45 = yy_set_m128i(coeffs_45_128, coeffs_45_128);
-    // coeffs [ g7 g6 g7 g6 g7 g6 g7 g6 ][ g7 g6 g7 g6 g7 g6 g7 g6 ]
-    const __m256i coeffs_67 = yy_set_m128i(coeffs_67_128, coeffs_67_128);
+      for (i = 0; i < h - 1; i += 2) {
+        const int16_t *data = &im_block[i * im_stride];
 
-    const __m256i round_const =
-        _mm256_set1_epi32((1 << (conv_params->round_1 - 1)) -
-                          (1 << (bd + conv_params->round_1 - 1)));
+        const __m256i s6 =
+            _mm256_loadu_si256((__m256i *)(data + 6 * im_stride));
+        const __m256i s7 =
+            _mm256_loadu_si256((__m256i *)(data + 7 * im_stride));
 
-    for (int i = 0; i < h; ++i) {
-      for (int j = 0; j < w; j += 16) {
-        const uint16_t *data_ij = temp + i * MAX_SB_SIZE + j;
+        s[3] = _mm256_unpacklo_epi16(s6, s7);
+        s[7] = _mm256_unpackhi_epi16(s6, s7);
 
-        // Load 16-bit data from the output of the horizontal filter in
-        // which the pixels are ordered as follows:
-        // [ 15 13 11 9 14 12 10 8 ] [ 7 5 3 1 6 4 2 0 ]
-        const __m256i data_0 = yy_loadu_256(data_ij + 0 * MAX_SB_SIZE);
-        const __m256i data_1 = yy_loadu_256(data_ij + 1 * MAX_SB_SIZE);
-        const __m256i data_2 = yy_loadu_256(data_ij + 2 * MAX_SB_SIZE);
-        const __m256i data_3 = yy_loadu_256(data_ij + 3 * MAX_SB_SIZE);
-        const __m256i data_4 = yy_loadu_256(data_ij + 4 * MAX_SB_SIZE);
-        const __m256i data_5 = yy_loadu_256(data_ij + 5 * MAX_SB_SIZE);
-        const __m256i data_6 = yy_loadu_256(data_ij + 6 * MAX_SB_SIZE);
-        const __m256i data_7 = yy_loadu_256(data_ij + 7 * MAX_SB_SIZE);
+        __m256i res_a = convolve(s, coeffs_v);
+        __m256i res_b = convolve(s + 4, coeffs_v);
 
-        // Filter the even-indices, increasing to 32-bit precision
-        const __m256i src_0 = _mm256_unpacklo_epi16(data_0, data_1);
-        const __m256i src_2 = _mm256_unpacklo_epi16(data_2, data_3);
-        const __m256i src_4 = _mm256_unpacklo_epi16(data_4, data_5);
-        const __m256i src_6 = _mm256_unpacklo_epi16(data_6, data_7);
+        const __m256i res_a_round = _mm256_sra_epi32(
+            _mm256_add_epi32(res_a, round_const_v), round_shift_v);
+        const __m256i res_b_round = _mm256_sra_epi32(
+            _mm256_add_epi32(res_b, round_const_v), round_shift_v);
 
-        const __m256i res_0 = _mm256_madd_epi16(src_0, coeffs_01);
-        const __m256i res_2 = _mm256_madd_epi16(src_2, coeffs_23);
-        const __m256i res_4 = _mm256_madd_epi16(src_4, coeffs_45);
-        const __m256i res_6 = _mm256_madd_epi16(src_6, coeffs_67);
+        /* rounding code */
+        // 16 bit conversion
+        const __m256i res_16bit = _mm256_packs_epi32(res_a_round, res_b_round);
+        // 8 bit conversion and saturation to uint8
+        const __m256i res_8b = _mm256_packus_epi16(res_16bit, res_16bit);
 
-        const __m256i res_even = _mm256_add_epi32(
-            _mm256_add_epi32(res_0, res_2), _mm256_add_epi32(res_4, res_6));
+        const __m128i res_0 = _mm256_castsi256_si128(res_8b);
+        const __m128i res_1 = _mm256_extracti128_si256(res_8b, 1);
 
-        // Filter the odd-indices, increasing to 32-bit precision
-        const __m256i src_1 = _mm256_unpackhi_epi16(data_0, data_1);
-        const __m256i src_3 = _mm256_unpackhi_epi16(data_2, data_3);
-        const __m256i src_5 = _mm256_unpackhi_epi16(data_4, data_5);
-        const __m256i src_7 = _mm256_unpackhi_epi16(data_6, data_7);
+        // Store values into the destination buffer
+        __m128i *const p_0 = (__m128i *)&dst[i * dst_stride + j];
+        __m128i *const p_1 = (__m128i *)&dst[i * dst_stride + j + dst_stride];
 
-        const __m256i res_1 = _mm256_madd_epi16(src_1, coeffs_01);
-        const __m256i res_3 = _mm256_madd_epi16(src_3, coeffs_23);
-        const __m256i res_5 = _mm256_madd_epi16(src_5, coeffs_45);
-        const __m256i res_7 = _mm256_madd_epi16(src_7, coeffs_67);
+        _mm_storel_epi64(p_0, res_0);
+        _mm_storel_epi64(p_1, res_1);
 
-        const __m256i res_odd = _mm256_add_epi32(
-            _mm256_add_epi32(res_1, res_3), _mm256_add_epi32(res_5, res_7));
+        s[0] = s[1];
+        s[1] = s[2];
+        s[2] = s[3];
 
-        // Pixels are currently in the following order:
-        // res_even order: [ 14 12 10 8 ] [ 6 4 2 0 ]
-        // res_odd order:  [ 15 13 11 9 ] [ 7 5 3 1 ]
-        //
-        // Rearrange the pixels into the following order:
-        // res_lo order: [ 11 10  9  8 ] [ 3 2 1 0 ]
-        // res_hi order: [ 15 14 13 12 ] [ 7 6 5 4 ]
-        const __m256i res_lo = _mm256_unpacklo_epi32(res_even, res_odd);
-        const __m256i res_hi = _mm256_unpackhi_epi32(res_even, res_odd);
+        s[4] = s[5];
+        s[5] = s[6];
+        s[6] = s[7];
+      }
+      if (h - i) {
+        s[0] = _mm256_permute2x128_si256(s[0], s[4], 0x20);
+        s[1] = _mm256_permute2x128_si256(s[1], s[5], 0x20);
+        s[2] = _mm256_permute2x128_si256(s[2], s[6], 0x20);
 
-        const __m256i res_lo_round = _mm256_srai_epi32(
-            _mm256_add_epi32(res_lo, round_const), conv_params->round_1);
-        const __m256i res_hi_round = _mm256_srai_epi32(
-            _mm256_add_epi32(res_hi, round_const), conv_params->round_1);
+        const int16_t *data = &im_block[i * im_stride];
+        const __m128i s6_ = _mm_loadu_si128((__m128i *)(data + 6 * im_stride));
+        const __m128i s7_ = _mm_loadu_si128((__m128i *)(data + 7 * im_stride));
 
-        // Reduce to 16-bit precision and pack into the correct order:
-        // [ 15 14 13 12 11 10 9 8 ][ 7 6 5 4 3 2 1 0 ]
-        const __m256i res_16bit =
-            _mm256_packs_epi32(res_lo_round, res_hi_round);
+        __m128i s3 = _mm_unpacklo_epi16(s6_, s7_);
+        __m128i s7 = _mm_unpackhi_epi16(s6_, s7_);
 
-        // Reduce to 8-bit precision. This messes up the order:
-        // [ - - - - - - - - 15 14 13 12 11 10 9 8 ]
-        // [ - - - - - - - - 7 6 5 4 3 2 1 0 ]
-        const __m256i res_8bit =
-            _mm256_packus_epi16(res_16bit, zero_256 /* don't care value */);
+        s[3] = _mm256_inserti128_si256(_mm256_castsi128_si256(s3), s7, 1);
+        __m256i convolveres = convolve(s, coeffs_v);
 
-        // Swap the two central 32-bit values to get the order:
-        // [ - - - - - - - - - - - - - - - - ]
-        // [ 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 ]
-        const __m256i res_8bit2 = _mm256_permute4x64_epi64(res_8bit, 0xd8);
+        const __m256i res_round = _mm256_sra_epi32(
+            _mm256_add_epi32(convolveres, round_const_v), round_shift_v);
 
-        // Store the lower 128-bit lane in the dst array
-        xx_storeu_128(dst + i * dst_stride + j,
-                      _mm256_castsi256_si128(res_8bit2));
+        /* rounding code */
+        // 16 bit conversion
+        __m128i reslo = _mm256_castsi256_si128(res_round);
+        __m128i reshi = _mm256_extracti128_si256(res_round, 1);
+        const __m128i res_16bit = _mm_packus_epi32(reslo, reshi);
+
+        // 8 bit conversion and saturation to uint8
+        const __m128i res_8b = _mm_packus_epi16(res_16bit, res_16bit);
+        __m128i *const p_0 = (__m128i *)&dst[i * dst_stride + j];
+        _mm_storel_epi64(p_0, res_8b);
       }
     }
   }
diff --git a/third_party/aom/av1/decoder/decodeframe.c b/third_party/aom/av1/decoder/decodeframe.c
index 31f14b531f..26e91d5ac8 100644
--- a/third_party/aom/av1/decoder/decodeframe.c
+++ b/third_party/aom/av1/decoder/decodeframe.c
@@ -64,6 +64,9 @@
 
 #define ACCT_STR __func__
 
+#define AOM_MIN_THREADS_PER_TILE 1
+#define AOM_MAX_THREADS_PER_TILE 2
+
 // This is needed by ext_tile related unit tests.
 #define EXT_TILE_DEBUG 1
 #define MC_TEMP_BUF_PELS                       \
@@ -153,13 +156,10 @@ static void inverse_transform_block(MACROBLOCKD *xd, int plane,
                                     const TX_SIZE tx_size, uint8_t *dst,
                                     int stride, int reduced_tx_set) {
   struct macroblockd_plane *const pd = &xd->plane[plane];
-  tran_low_t *const dqcoeff = pd->dqcoeff;
+  tran_low_t *const dqcoeff = pd->dqcoeff_block + xd->cb_offset[plane];
   eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
   uint16_t scan_line = eob_data->max_scan_line;
   uint16_t eob = eob_data->eob;
-
-  memcpy(dqcoeff, pd->dqcoeff_block + xd->cb_offset[plane],
-         (scan_line + 1) * sizeof(dqcoeff[0]));
   av1_inverse_transform_block(xd, dqcoeff, plane, tx_type, tx_size, dst, stride,
                               eob, reduced_tx_set);
   memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
@@ -268,8 +268,8 @@ static void inverse_transform_inter_block(const AV1_COMMON *const cm,
   int blk_h = block_size_high[bsize];
   mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, blk_col, blk_row,
                   pd->subsampling_x, pd->subsampling_y);
-  mismatch_check_block_tx(dst, pd->dst.stride, cm->frame_offset, plane, pixel_c,
-                          pixel_r, blk_w, blk_h,
+  mismatch_check_block_tx(dst, pd->dst.stride, cm->current_frame.order_hint,
+                          plane, pixel_c, pixel_r, blk_w, blk_h,
                           xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
 #endif
 }
@@ -368,8 +368,8 @@ static void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
   // as they are always compared to values that are in 1/8th pel units
   set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
-  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col, 0, num_planes);
+  av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0,
+                       num_planes);
 }
 
 static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
@@ -397,9 +397,6 @@ static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
       aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
                          "Invalid block size.");
   }
-
-  int reader_corrupted_flag = aom_reader_has_error(r);
-  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
 }
 
 typedef struct PadBlock {
@@ -699,26 +696,28 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
         assert(bw < 8 || bh < 8);
         ConvolveParams conv_params = get_conv_params_no_round(
             0, plane, xd->tmp_conv_dst, tmp_dst_stride, is_compound, xd->bd);
-        conv_params.use_jnt_comp_avg = 0;
+        conv_params.use_dist_wtd_comp_avg = 0;
         struct buf_2d *const dst_buf = &pd->dst;
         uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x;
 
         ref = 0;
-        const RefBuffer *ref_buf =
-            &cm->frame_refs[this_mbmi->ref_frame[ref] - LAST_FRAME];
+        const RefCntBuffer *ref_buf =
+            get_ref_frame_buf(cm, this_mbmi->ref_frame[ref]);
+        const struct scale_factors *ref_scale_factors =
+            get_ref_scale_factors_const(cm, this_mbmi->ref_frame[ref]);
 
         pd->pre[ref].buf0 =
-            (plane == 1) ? ref_buf->buf->u_buffer : ref_buf->buf->v_buffer;
+            (plane == 1) ? ref_buf->buf.u_buffer : ref_buf->buf.v_buffer;
         pd->pre[ref].buf =
             pd->pre[ref].buf0 + scaled_buffer_offset(pre_x, pre_y,
-                                                     ref_buf->buf->uv_stride,
-                                                     &ref_buf->sf);
-        pd->pre[ref].width = ref_buf->buf->uv_crop_width;
-        pd->pre[ref].height = ref_buf->buf->uv_crop_height;
-        pd->pre[ref].stride = ref_buf->buf->uv_stride;
+                                                     ref_buf->buf.uv_stride,
+                                                     ref_scale_factors);
+        pd->pre[ref].width = ref_buf->buf.uv_crop_width;
+        pd->pre[ref].height = ref_buf->buf.uv_crop_height;
+        pd->pre[ref].stride = ref_buf->buf.uv_stride;
 
         const struct scale_factors *const sf =
-            is_intrabc ? &cm->sf_identity : &ref_buf->sf;
+            is_intrabc ? &cm->sf_identity : ref_scale_factors;
         struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
 
         const MV mv = this_mbmi->mv[ref].as_mv;
@@ -738,7 +737,7 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
                                &scaled_mv, &subpel_x_mv, &subpel_y_mv);
         pre = pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
         src_stride = pre_buf->stride;
-        highbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+        highbd = is_cur_buf_hbd(xd);
         extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv,
                          subpel_y_mv, 0, is_intrabc, highbd, xd->mc_buf[ref],
                          &pre, &src_stride);
@@ -771,7 +770,7 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
     int src_stride[2];
     for (ref = 0; ref < 1 + is_compound; ++ref) {
       const struct scale_factors *const sf =
-          is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf;
+          is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref];
       struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
       const MV mv = mi->mv[ref].as_mv;
       PadBlock block;
@@ -784,7 +783,7 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
                              &scaled_mv, &subpel_x_mv, &subpel_y_mv);
       pre[ref] = pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
       src_stride[ref] = pre_buf->stride;
-      highbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+      highbd = is_cur_buf_hbd(xd);
 
       WarpTypesAllowed warp_types;
       warp_types.global_warp_allowed = is_global[ref];
@@ -792,8 +791,7 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
       int do_warp = (bw >= 8 && bh >= 8 &&
                      av1_allow_warp(mi, &warp_types,
                                     &xd->global_motion[mi->ref_frame[ref]],
-                                    build_for_obmc, subpel_params[ref].xs,
-                                    subpel_params[ref].ys, NULL));
+                                    build_for_obmc, sf, NULL));
       do_warp = (do_warp && xd->cur_frame_force_integer_mv == 0);
 
       extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv, subpel_y_mv,
@@ -803,13 +801,13 @@ static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
 
     ConvolveParams conv_params = get_conv_params_no_round(
         0, plane, xd->tmp_conv_dst, MAX_SB_SIZE, is_compound, xd->bd);
-    av1_jnt_comp_weight_assign(cm, mi, 0, &conv_params.fwd_offset,
-                               &conv_params.bck_offset,
-                               &conv_params.use_jnt_comp_avg, is_compound);
+    av1_dist_wtd_comp_weight_assign(
+        cm, mi, 0, &conv_params.fwd_offset, &conv_params.bck_offset,
+        &conv_params.use_dist_wtd_comp_avg, is_compound);
 
     for (ref = 0; ref < 1 + is_compound; ++ref) {
       const struct scale_factors *const sf =
-          is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf;
+          is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref];
       WarpTypesAllowed warp_types;
       warp_types.global_warp_allowed = is_global[ref];
       warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
@@ -858,7 +856,7 @@ static void dec_build_inter_predictors_for_planes(const AV1_COMMON *cm,
 
 static void dec_build_inter_predictors_sby(const AV1_COMMON *cm,
                                            MACROBLOCKD *xd, int mi_row,
-                                           int mi_col, BUFFER_SET *ctx,
+                                           int mi_col, const BUFFER_SET *ctx,
                                            BLOCK_SIZE bsize) {
   dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0, 0);
 
@@ -873,7 +871,7 @@ static void dec_build_inter_predictors_sby(const AV1_COMMON *cm,
 
 static void dec_build_inter_predictors_sbuv(const AV1_COMMON *cm,
                                             MACROBLOCKD *xd, int mi_row,
-                                            int mi_col, BUFFER_SET *ctx,
+                                            int mi_col, const BUFFER_SET *ctx,
                                             BLOCK_SIZE bsize) {
   dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 1,
                                         MAX_MB_PLANE - 1);
@@ -1018,7 +1016,7 @@ static void dec_build_obmc_inter_predictors_sb(const AV1_COMMON *cm,
   int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
   int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     int len = sizeof(uint16_t);
     dst_buf1[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0]);
     dst_buf1[1] =
@@ -1042,7 +1040,7 @@ static void dec_build_obmc_inter_predictors_sb(const AV1_COMMON *cm,
                                       dst_width1, dst_height1, dst_stride1);
   dec_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
                                      dst_width2, dst_height2, dst_stride2);
-  av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, get_frame_new_buffer(cm),
+  av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, &cm->cur_frame->buf,
                        mi_row, mi_col, 0, num_planes);
   av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, dst_stride1,
                                   dst_buf2, dst_stride2);
@@ -1066,11 +1064,13 @@ static void predict_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
       assert(frame == INTRA_FRAME);
       assert(ref == 0);
     } else {
-      RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
+      const RefCntBuffer *ref_buf = get_ref_frame_buf(cm, frame);
+      const struct scale_factors *ref_scale_factors =
+          get_ref_scale_factors_const(cm, frame);
 
-      xd->block_refs[ref] = ref_buf;
-      av1_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, &ref_buf->sf,
-                           num_planes);
+      xd->block_ref_scale_factors[ref] = ref_scale_factors;
+      av1_setup_pre_planes(xd, ref, &ref_buf->buf, mi_row, mi_col,
+                           ref_scale_factors, num_planes);
     }
   }
 
@@ -1087,8 +1087,9 @@ static void predict_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
     if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
                              pd->subsampling_y))
       continue;
-    mismatch_check_block_pre(pd->dst.buf, pd->dst.stride, cm->frame_offset,
-                             plane, pixel_c, pixel_r, pd->width, pd->height,
+    mismatch_check_block_pre(pd->dst.buf, pd->dst.stride,
+                             cm->current_frame.order_hint, plane, pixel_c,
+                             pixel_r, pd->width, pd->height,
                              xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
   }
 #endif
@@ -1238,10 +1239,26 @@ static void store_bitmask_vartx(AV1_COMMON *cm, int mi_row, int mi_col,
                                 MB_MODE_INFO *mbmi);
 #endif
 
+static void set_inter_tx_size(MB_MODE_INFO *mbmi, int stride_log2,
+                              int tx_w_log2, int tx_h_log2, int min_txs,
+                              int split_size, int txs, int blk_row,
+                              int blk_col) {
+  for (int idy = 0; idy < tx_size_high_unit[split_size];
+       idy += tx_size_high_unit[min_txs]) {
+    for (int idx = 0; idx < tx_size_wide_unit[split_size];
+         idx += tx_size_wide_unit[min_txs]) {
+      const int index = (((blk_row + idy) >> tx_h_log2) << stride_log2) +
+                        ((blk_col + idx) >> tx_w_log2);
+      mbmi->inter_tx_size[index] = txs;
+    }
+  }
+}
+
 static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
                                TX_SIZE tx_size, int depth,
 #if LOOP_FILTER_BITMASK
                                AV1_COMMON *cm, int mi_row, int mi_col,
+                               int store_bitmask,
 #endif
                                int blk_row, int blk_col, aom_reader *r) {
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
@@ -1251,15 +1268,17 @@ static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
   const int max_blocks_wide = max_block_wide(xd, bsize, 0);
   if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
   assert(tx_size > TX_4X4);
+  TX_SIZE txs = max_txsize_rect_lookup[bsize];
+  for (int level = 0; level < MAX_VARTX_DEPTH - 1; ++level)
+    txs = sub_tx_size_map[txs];
+  const int tx_w_log2 = tx_size_wide_log2[txs] - MI_SIZE_LOG2;
+  const int tx_h_log2 = tx_size_high_log2[txs] - MI_SIZE_LOG2;
+  const int bw_log2 = mi_size_wide_log2[bsize];
+  const int stride_log2 = bw_log2 - tx_w_log2;
 
   if (depth == MAX_VARTX_DEPTH) {
-    for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
-      for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
-        const int index =
-            av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
-        mbmi->inter_tx_size[index] = tx_size;
-      }
-    }
+    set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
+                      tx_size, blk_row, blk_col);
     mbmi->tx_size = tx_size;
     txfm_partition_update(xd->above_txfm_context + blk_col,
                           xd->left_txfm_context + blk_row, tx_size, tx_size);
@@ -1277,26 +1296,24 @@ static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
     const int bsh = tx_size_high_unit[sub_txs];
 
     if (sub_txs == TX_4X4) {
-      for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
-        for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
-          const int index =
-              av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
-          mbmi->inter_tx_size[index] = sub_txs;
-        }
-      }
+      set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
+                        sub_txs, blk_row, blk_col);
       mbmi->tx_size = sub_txs;
       txfm_partition_update(xd->above_txfm_context + blk_col,
                             xd->left_txfm_context + blk_row, sub_txs, tx_size);
 #if LOOP_FILTER_BITMASK
-      store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col, BLOCK_8X8,
-                          TX_4X4, mbmi);
+      if (store_bitmask) {
+        store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
+                            txsize_to_bsize[tx_size], TX_4X4, mbmi);
+      }
 #endif
       return;
     }
 #if LOOP_FILTER_BITMASK
-    if (depth + 1 == MAX_VARTX_DEPTH) {
+    if (depth + 1 == MAX_VARTX_DEPTH && store_bitmask) {
       store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
                           txsize_to_bsize[tx_size], sub_txs, mbmi);
+      store_bitmask = 0;
     }
 #endif
 
@@ -1307,25 +1324,22 @@ static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
         int offsetc = blk_col + col;
         read_tx_size_vartx(xd, mbmi, sub_txs, depth + 1,
 #if LOOP_FILTER_BITMASK
-                           cm, mi_row, mi_col,
+                           cm, mi_row, mi_col, store_bitmask,
 #endif
                            offsetr, offsetc, r);
       }
     }
   } else {
-    for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) {
-      for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) {
-        const int index =
-            av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx);
-        mbmi->inter_tx_size[index] = tx_size;
-      }
-    }
+    set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
+                      tx_size, blk_row, blk_col);
     mbmi->tx_size = tx_size;
     txfm_partition_update(xd->above_txfm_context + blk_col,
                           xd->left_txfm_context + blk_row, tx_size, tx_size);
 #if LOOP_FILTER_BITMASK
-    store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
-                        txsize_to_bsize[tx_size], tx_size, mbmi);
+    if (store_bitmask) {
+      store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
+                          txsize_to_bsize[tx_size], tx_size, mbmi);
+    }
 #endif
   }
 }
@@ -1505,7 +1519,9 @@ static void store_bitmask_univariant_tx(AV1_COMMON *cm, int mi_row, int mi_col,
 }
 
 static void store_bitmask_other_info(AV1_COMMON *cm, int mi_row, int mi_col,
-                                     BLOCK_SIZE bsize, MB_MODE_INFO *mbmi) {
+                                     BLOCK_SIZE bsize, MB_MODE_INFO *mbmi,
+                                     int is_horz_coding_block_border,
+                                     int is_vert_coding_block_border) {
   int index;
   int shift;
   int row;
@@ -1513,14 +1529,26 @@ static void store_bitmask_other_info(AV1_COMMON *cm, int mi_row, int mi_col,
   const int row_start = mi_row % MI_SIZE_64X64;
   const int col_start = mi_col % MI_SIZE_64X64;
   shift = get_index_shift(col_start, row_start, &index);
-  const uint64_t top_edge_mask =
-      ((uint64_t)1 << (shift + mi_size_wide[bsize])) - ((uint64_t)1 << shift);
-  lfm->is_horz_border.bits[index] |= top_edge_mask;
-  const int is_vert_border = mask_id_table_vert_border[bsize];
-  const int vert_shift = block_size_high[bsize] <= 8 ? shift : col_start;
-  for (int i = 0; i + index < 4; ++i) {
-    lfm->is_vert_border.bits[i + index] |=
-        (left_mask_univariant_reordered[is_vert_border].bits[i] << vert_shift);
+  if (is_horz_coding_block_border) {
+    const int block_shift = shift + mi_size_wide[bsize];
+    assert(block_shift <= 64);
+    const uint64_t right_edge_shift =
+        (block_shift == 64) ? 0xffffffffffffffff : ((uint64_t)1 << block_shift);
+    const uint64_t left_edge_shift = (block_shift == 64)
+                                         ? (((uint64_t)1 << shift) - 1)
+                                         : ((uint64_t)1 << shift);
+    assert(right_edge_shift > left_edge_shift);
+    const uint64_t top_edge_mask = right_edge_shift - left_edge_shift;
+    lfm->is_horz_border.bits[index] |= top_edge_mask;
+  }
+  if (is_vert_coding_block_border) {
+    const int is_vert_border = mask_id_table_vert_border[bsize];
+    const int vert_shift = block_size_high[bsize] <= 8 ? shift : col_start;
+    for (int i = 0; i + index < 4; ++i) {
+      lfm->is_vert_border.bits[i + index] |=
+          (left_mask_univariant_reordered[is_vert_border].bits[i]
+           << vert_shift);
+    }
   }
   const int is_skip = mbmi->skip && is_inter_block(mbmi);
   if (is_skip) {
@@ -1541,9 +1569,13 @@ static void store_bitmask_other_info(AV1_COMMON *cm, int mi_row, int mi_col,
            sizeof(uint8_t) * mi_size_wide[bsize]);
     memset(&lfm->lfl_y_hor[row][col_start], level_horz_y,
            sizeof(uint8_t) * mi_size_wide[bsize]);
-    memset(&lfm->lfl_u[row][col_start], level_u,
+    memset(&lfm->lfl_u_ver[row][col_start], level_u,
            sizeof(uint8_t) * mi_size_wide[bsize]);
-    memset(&lfm->lfl_v[row][col_start], level_v,
+    memset(&lfm->lfl_u_hor[row][col_start], level_u,
+           sizeof(uint8_t) * mi_size_wide[bsize]);
+    memset(&lfm->lfl_v_ver[row][col_start], level_v,
+           sizeof(uint8_t) * mi_size_wide[bsize]);
+    memset(&lfm->lfl_v_hor[row][col_start], level_v,
            sizeof(uint8_t) * mi_size_wide[bsize]);
   }
 }
@@ -1574,7 +1606,7 @@ static void parse_decode_block(AV1Decoder *const pbi, ThreadData *const td,
       for (int idx = 0; idx < width; idx += bw)
         read_tx_size_vartx(xd, mbmi, max_tx_size, 0,
 #if LOOP_FILTER_BITMASK
-                           cm, mi_row, mi_col,
+                           cm, mi_row, mi_col, 1,
 #endif
                            idy, idx, r);
   } else {
@@ -1602,18 +1634,18 @@ static void parse_decode_block(AV1Decoder *const pbi, ThreadData *const td,
   const int w = mi_size_wide[bsize];
   const int h = mi_size_high[bsize];
   if (w <= mi_size_wide[BLOCK_64X64] && h <= mi_size_high[BLOCK_64X64]) {
-    store_bitmask_other_info(cm, mi_row, mi_col, bsize, mbmi);
+    store_bitmask_other_info(cm, mi_row, mi_col, bsize, mbmi, 1, 1);
   } else {
     for (int row = 0; row < h; row += mi_size_high[BLOCK_64X64]) {
       for (int col = 0; col < w; col += mi_size_wide[BLOCK_64X64]) {
         store_bitmask_other_info(cm, mi_row + row, mi_col + col, BLOCK_64X64,
-                                 mbmi);
+                                 mbmi, row == 0, col == 0);
       }
     }
   }
 #endif
 
-  if (cm->delta_q_present_flag) {
+  if (cm->delta_q_info.delta_q_present_flag) {
     for (int i = 0; i < MAX_SEGMENTS; i++) {
       const int current_qindex =
           av1_get_qindex(&cm->seg, i, xd->current_qindex);
@@ -1633,9 +1665,6 @@ static void parse_decode_block(AV1Decoder *const pbi, ThreadData *const td,
   if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize, num_planes);
 
   decode_token_recon_block(pbi, td, mi_row, mi_col, r, bsize);
-
-  int reader_corrupted_flag = aom_reader_has_error(r);
-  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
 }
 
 static void set_offsets_for_pred_and_recon(AV1Decoder *const pbi,
@@ -1660,8 +1689,8 @@ static void set_offsets_for_pred_and_recon(AV1Decoder *const pbi,
   // as they are always compared to values that are in 1/8th pel units
   set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
 
-  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col, 0, num_planes);
+  av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0,
+                       num_planes);
 }
 
 static void decode_block(AV1Decoder *const pbi, ThreadData *const td,
@@ -1703,7 +1732,7 @@ static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col,
 
 // TODO(slavarnway): eliminate bsize and subsize in future commits
 static void decode_partition(AV1Decoder *const pbi, ThreadData *const td,
-                             int mi_row, int mi_col, aom_reader *r,
+                             int mi_row, int mi_col, aom_reader *reader,
                              BLOCK_SIZE bsize, int parse_decode_flag) {
   AV1_COMMON *const cm = &pbi->common;
   MACROBLOCKD *const xd = &td->xd;
@@ -1736,14 +1765,14 @@ static void decode_partition(AV1Decoder *const pbi, ThreadData *const td,
         for (int rrow = rrow0; rrow < rrow1; ++rrow) {
           for (int rcol = rcol0; rcol < rcol1; ++rcol) {
             const int runit_idx = rcol + rrow * rstride;
-            loop_restoration_read_sb_coeffs(cm, xd, r, plane, runit_idx);
+            loop_restoration_read_sb_coeffs(cm, xd, reader, plane, runit_idx);
           }
         }
       }
     }
 
     partition = (bsize < BLOCK_8X8) ? PARTITION_NONE
-                                    : read_partition(xd, mi_row, mi_col, r,
+                                    : read_partition(xd, mi_row, mi_col, reader,
                                                      has_rows, has_cols, bsize);
   } else {
     partition = get_partition(cm, mi_row, mi_col, bsize);
@@ -1762,12 +1791,12 @@ static void decode_partition(AV1Decoder *const pbi, ThreadData *const td,
 
 #define DEC_BLOCK_STX_ARG
 #define DEC_BLOCK_EPT_ARG partition,
-#define DEC_BLOCK(db_r, db_c, db_subsize)                                     \
-  block_visit[parse_decode_flag](pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), r, \
-                                 DEC_BLOCK_EPT_ARG(db_subsize))
-#define DEC_PARTITION(db_r, db_c, db_subsize)                                 \
-  decode_partition(pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), r, (db_subsize), \
-                   parse_decode_flag)
+#define DEC_BLOCK(db_r, db_c, db_subsize)                                  \
+  block_visit[parse_decode_flag](pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), \
+                                 reader, DEC_BLOCK_EPT_ARG(db_subsize))
+#define DEC_PARTITION(db_r, db_c, db_subsize)                        \
+  decode_partition(pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), reader, \
+                   (db_subsize), parse_decode_flag)
 
   switch (partition) {
     case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break;
@@ -2170,13 +2199,16 @@ static void setup_loopfilter(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
 
 static void setup_cdef(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
   const int num_planes = av1_num_planes(cm);
+  CdefInfo *const cdef_info = &cm->cdef_info;
+
   if (cm->allow_intrabc) return;
-  cm->cdef_pri_damping = cm->cdef_sec_damping = aom_rb_read_literal(rb, 2) + 3;
-  cm->cdef_bits = aom_rb_read_literal(rb, 2);
-  cm->nb_cdef_strengths = 1 << cm->cdef_bits;
-  for (int i = 0; i < cm->nb_cdef_strengths; i++) {
-    cm->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
-    cm->cdef_uv_strengths[i] =
+  cdef_info->cdef_pri_damping = aom_rb_read_literal(rb, 2) + 3;
+  cdef_info->cdef_sec_damping = cdef_info->cdef_pri_damping;
+  cdef_info->cdef_bits = aom_rb_read_literal(rb, 2);
+  cdef_info->nb_cdef_strengths = 1 << cdef_info->cdef_bits;
+  for (int i = 0; i < cdef_info->nb_cdef_strengths; i++) {
+    cdef_info->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
+    cdef_info->cdef_uv_strengths[i] =
         num_planes > 1 ? aom_rb_read_literal(rb, CDEF_STRENGTH_BITS) : 0;
   }
 }
@@ -2209,7 +2241,6 @@ static void setup_quantization(AV1_COMMON *const cm,
     cm->v_dc_delta_q = 0;
     cm->v_ac_delta_q = 0;
   }
-  cm->dequant_bit_depth = seq_params->bit_depth;
   cm->using_qmatrix = aom_rb_read_bit(rb);
   if (cm->using_qmatrix) {
     cm->qm_y = aom_rb_read_literal(rb, QM_LEVEL_BITS);
@@ -2226,14 +2257,15 @@ static void setup_quantization(AV1_COMMON *const cm,
 }
 
 // Build y/uv dequant values based on segmentation.
-static void setup_segmentation_dequant(AV1_COMMON *const cm) {
+static void setup_segmentation_dequant(AV1_COMMON *const cm,
+                                       MACROBLOCKD *const xd) {
   const int bit_depth = cm->seq_params.bit_depth;
   const int using_qm = cm->using_qmatrix;
   // When segmentation is disabled, only the first value is used.  The
   // remaining are don't cares.
   const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1;
   for (int i = 0; i < max_segments; ++i) {
-    const int qindex = av1_get_qindex(&cm->seg, i, cm->base_qindex);
+    const int qindex = xd->qindex[i];
     cm->y_dequant_QTX[i][0] =
         av1_dc_quant_QTX(qindex, cm->y_dc_delta_q, bit_depth);
     cm->y_dequant_QTX[i][1] = av1_ac_quant_QTX(qindex, 0, bit_depth);
@@ -2245,9 +2277,7 @@ static void setup_segmentation_dequant(AV1_COMMON *const cm) {
         av1_dc_quant_QTX(qindex, cm->v_dc_delta_q, bit_depth);
     cm->v_dequant_QTX[i][1] =
         av1_ac_quant_QTX(qindex, cm->v_ac_delta_q, bit_depth);
-    const int lossless = qindex == 0 && cm->y_dc_delta_q == 0 &&
-                         cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 &&
-                         cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0;
+    const int lossless = xd->lossless[i];
     // NB: depends on base index so there is only 1 set per frame
     // No quant weighting when lossless or signalled not using QM
     int qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_y;
@@ -2344,36 +2374,27 @@ static void setup_buffer_pool(AV1_COMMON *cm) {
 
   lock_buffer_pool(pool);
   if (aom_realloc_frame_buffer(
-          get_frame_new_buffer(cm), cm->width, cm->height,
-          seq_params->subsampling_x, seq_params->subsampling_y,
-          seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS,
-          cm->byte_alignment,
-          &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb,
-          pool->cb_priv)) {
+          &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x,
+          seq_params->subsampling_y, seq_params->use_highbitdepth,
+          AOM_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
+          &cm->cur_frame->raw_frame_buffer, pool->get_fb_cb, pool->cb_priv)) {
     unlock_buffer_pool(pool);
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
   }
   unlock_buffer_pool(pool);
 
-  pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x =
-      seq_params->subsampling_x;
-  pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y =
-      seq_params->subsampling_y;
-  pool->frame_bufs[cm->new_fb_idx].buf.bit_depth =
-      (unsigned int)seq_params->bit_depth;
-  pool->frame_bufs[cm->new_fb_idx].buf.color_primaries =
-      seq_params->color_primaries;
-  pool->frame_bufs[cm->new_fb_idx].buf.transfer_characteristics =
+  cm->cur_frame->buf.bit_depth = (unsigned int)seq_params->bit_depth;
+  cm->cur_frame->buf.color_primaries = seq_params->color_primaries;
+  cm->cur_frame->buf.transfer_characteristics =
       seq_params->transfer_characteristics;
-  pool->frame_bufs[cm->new_fb_idx].buf.matrix_coefficients =
-      seq_params->matrix_coefficients;
-  pool->frame_bufs[cm->new_fb_idx].buf.monochrome = seq_params->monochrome;
-  pool->frame_bufs[cm->new_fb_idx].buf.chroma_sample_position =
+  cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients;
+  cm->cur_frame->buf.monochrome = seq_params->monochrome;
+  cm->cur_frame->buf.chroma_sample_position =
       seq_params->chroma_sample_position;
-  pool->frame_bufs[cm->new_fb_idx].buf.color_range = seq_params->color_range;
-  pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width;
-  pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height;
+  cm->cur_frame->buf.color_range = seq_params->color_range;
+  cm->cur_frame->buf.render_width = cm->render_width;
+  cm->cur_frame->buf.render_height = cm->render_height;
 }
 
 static void setup_frame_size(AV1_COMMON *cm, int frame_size_override_flag,
@@ -2419,17 +2440,28 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
   int width, height;
   int found = 0;
   int has_valid_ref_frame = 0;
-  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+  for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
     if (aom_rb_read_bit(rb)) {
-      YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
-      width = buf->y_crop_width;
-      height = buf->y_crop_height;
-      cm->render_width = buf->render_width;
-      cm->render_height = buf->render_height;
-      setup_superres(cm, rb, &width, &height);
-      resize_context_buffers(cm, width, height);
-      found = 1;
-      break;
+      const RefCntBuffer *const ref_buf = get_ref_frame_buf(cm, i);
+      // This will never be NULL in a normal stream, as streams are required to
+      // have a shown keyframe before any inter frames, which would refresh all
+      // the reference buffers. However, it might be null if we're starting in
+      // the middle of a stream, and static analysis will error if we don't do
+      // a null check here.
+      if (ref_buf == NULL) {
+        aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                           "Invalid condition: invalid reference buffer");
+      } else {
+        const YV12_BUFFER_CONFIG *const buf = &ref_buf->buf;
+        width = buf->y_crop_width;
+        height = buf->y_crop_height;
+        cm->render_width = buf->render_width;
+        cm->render_height = buf->render_height;
+        setup_superres(cm, rb, &width, &height);
+        resize_context_buffers(cm, width, height);
+        found = 1;
+        break;
+      }
     }
   }
 
@@ -2450,20 +2482,20 @@ static void setup_frame_size_with_refs(AV1_COMMON *cm,
 
   // Check to make sure at least one of frames that this frame references
   // has valid dimensions.
-  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-    RefBuffer *const ref_frame = &cm->frame_refs[i];
+  for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+    const RefCntBuffer *const ref_frame = get_ref_frame_buf(cm, i);
     has_valid_ref_frame |=
-        valid_ref_frame_size(ref_frame->buf->y_crop_width,
-                             ref_frame->buf->y_crop_height, width, height);
+        valid_ref_frame_size(ref_frame->buf.y_crop_width,
+                             ref_frame->buf.y_crop_height, width, height);
   }
   if (!has_valid_ref_frame)
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                        "Referenced frame has invalid size");
-  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-    RefBuffer *const ref_frame = &cm->frame_refs[i];
+  for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+    const RefCntBuffer *const ref_frame = get_ref_frame_buf(cm, i);
     if (!valid_ref_frame_img_fmt(
-            ref_frame->buf->bit_depth, ref_frame->buf->subsampling_x,
-            ref_frame->buf->subsampling_y, seq_params->bit_depth,
+            ref_frame->buf.bit_depth, ref_frame->buf.subsampling_x,
+            ref_frame->buf.subsampling_y, seq_params->bit_depth,
             seq_params->subsampling_x, seq_params->subsampling_y))
       aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                          "Referenced frame has incompatible color format");
@@ -2546,15 +2578,18 @@ void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm) {
   cm->single_tile_decoding = 0;
   if (cm->large_scale_tile) {
     struct loopfilter *lf = &cm->lf;
+    RestorationInfo *const rst_info = cm->rst_info;
+    const CdefInfo *const cdef_info = &cm->cdef_info;
 
     // Figure out single_tile_decoding by loopfilter_level.
     const int no_loopfilter = !(lf->filter_level[0] || lf->filter_level[1]);
-    const int no_cdef = cm->cdef_bits == 0 && cm->cdef_strengths[0] == 0 &&
-                        cm->cdef_uv_strengths[0] == 0;
+    const int no_cdef = cdef_info->cdef_bits == 0 &&
+                        cdef_info->cdef_strengths[0] == 0 &&
+                        cdef_info->cdef_uv_strengths[0] == 0;
     const int no_restoration =
-        cm->rst_info[0].frame_restoration_type == RESTORE_NONE &&
-        cm->rst_info[1].frame_restoration_type == RESTORE_NONE &&
-        cm->rst_info[2].frame_restoration_type == RESTORE_NONE;
+        rst_info[0].frame_restoration_type == RESTORE_NONE &&
+        rst_info[1].frame_restoration_type == RESTORE_NONE &&
+        rst_info[2].frame_restoration_type == RESTORE_NONE;
     assert(IMPLIES(cm->coded_lossless, no_loopfilter && no_cdef));
     assert(IMPLIES(cm->all_lossless, no_restoration));
     cm->single_tile_decoding = no_loopfilter && no_cdef && no_restoration;
@@ -2694,9 +2729,10 @@ static const uint8_t *get_ls_tile_buffers(
 
     const int tile_col_size_bytes = pbi->tile_col_size_bytes;
     const int tile_size_bytes = pbi->tile_size_bytes;
+    int tile_width, tile_height;
+    av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
     const int tile_copy_mode =
-        ((AOMMAX(cm->tile_width, cm->tile_height) << MI_SIZE_LOG2) <= 256) ? 1
-                                                                           : 0;
+        ((AOMMAX(tile_width, tile_height) << MI_SIZE_LOG2) <= 256) ? 1 : 0;
     // Read tile column sizes for all columns (we need the last tile buffer)
     for (int c = 0; c < tile_cols; ++c) {
       const int is_last = c == tile_cols - 1;
@@ -3184,7 +3220,7 @@ static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
         continue;
 
       td->bit_reader = &tile_data->bit_reader;
-      av1_zero(td->dqcoeff);
+      av1_zero(td->cb_buffer_base.dqcoeff);
       av1_tile_init(&td->xd.tile, cm, row, col);
       td->xd.current_qindex = cm->base_qindex;
       setup_bool_decoder(tile_bs_buf->data, data_end, tile_bs_buf->size,
@@ -3198,7 +3234,7 @@ static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data,
         td->bit_reader->accounting = NULL;
       }
 #endif
-      av1_init_macroblockd(cm, &td->xd, td->dqcoeff);
+      av1_init_macroblockd(cm, &td->xd, NULL);
       av1_init_above_context(cm, &td->xd, row);
 
       // Initialise the tile context from the frame context
@@ -3255,7 +3291,7 @@ static void tile_worker_hook_init(AV1Decoder *const pbi,
   int tile_col = tile_data->tile_info.tile_col;
 
   td->bit_reader = &tile_data->bit_reader;
-  av1_zero(td->dqcoeff);
+  av1_zero(td->cb_buffer_base.dqcoeff);
   av1_tile_init(&td->xd.tile, cm, tile_row, tile_col);
   td->xd.current_qindex = cm->base_qindex;
   setup_bool_decoder(tile_buffer->data, thread_data->data_end,
@@ -3270,7 +3306,7 @@ static void tile_worker_hook_init(AV1Decoder *const pbi,
     td->bit_reader->accounting = NULL;
   }
 #endif
-  av1_init_macroblockd(cm, &td->xd, td->dqcoeff);
+  av1_init_macroblockd(cm, &td->xd, NULL);
   td->xd.error_info = &thread_data->error_info;
   av1_init_above_context(cm, &td->xd, tile_row);
 
@@ -3308,10 +3344,10 @@ static int tile_worker_hook(void *arg1, void *arg2) {
   set_decode_func_pointers(td, 0x3);
 
   assert(cm->tile_cols > 0);
-  while (1) {
+  while (!td->xd.corrupted) {
     TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info);
 
-    if (cur_job_info != NULL && !td->xd.corrupted) {
+    if (cur_job_info != NULL) {
       const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer;
       TileDataDec *const tile_data = cur_job_info->tile_data;
       tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data,
@@ -3328,6 +3364,30 @@ static int tile_worker_hook(void *arg1, void *arg2) {
   return !td->xd.corrupted;
 }
 
+static INLINE int get_max_row_mt_workers_per_tile(AV1_COMMON *cm,
+                                                  TileInfo tile) {
+  // NOTE: Currently value of max workers is calculated based
+  // on the parse and decode time. As per the theoretical estimate
+  // when percentage of parse time is equal to percentage of decode
+  // time, number of workers needed to parse + decode a tile can not
+  // exceed more than 2.
+  // TODO(any): Modify this value if parsing is optimized in future.
+  int sb_rows = av1_get_sb_rows_in_tile(cm, tile);
+  int max_workers =
+      sb_rows == 1 ? AOM_MIN_THREADS_PER_TILE : AOM_MAX_THREADS_PER_TILE;
+  return max_workers;
+}
+
+// The caller must hold pbi->row_mt_mutex_ when calling this function.
+// Returns 1 if either the next job is stored in *next_job_info or 1 is stored
+// in *end_of_frame.
+// NOTE: The caller waits on pbi->row_mt_cond_ if this function returns 0.
+// The return value of this function depends on the following variables:
+// - frame_row_mt_info->mi_rows_parse_done
+// - frame_row_mt_info->mi_rows_decode_started
+// - frame_row_mt_info->row_mt_exit
+// Therefore we may need to signal or broadcast pbi->row_mt_cond_ if any of
+// these variables is modified.
 static int get_next_job_info(AV1Decoder *const pbi,
                              AV1DecRowMTJobInfo *next_job_info,
                              int *end_of_frame) {
@@ -3348,8 +3408,8 @@ static int get_next_job_info(AV1Decoder *const pbi,
   int min_threads_working = INT_MAX;
   int max_mis_to_decode = 0;
   int tile_row_idx, tile_col_idx;
-  int tile_row = 0;
-  int tile_col = 0;
+  int tile_row = -1;
+  int tile_col = -1;
 
   memset(next_job_info, 0, sizeof(*next_job_info));
 
@@ -3362,9 +3422,10 @@ static int get_next_job_info(AV1Decoder *const pbi,
   }
 
   // Decoding cannot start as bit-stream parsing is not complete.
-  if (frame_row_mt_info->mi_rows_parse_done -
-          frame_row_mt_info->mi_rows_decode_started ==
-      0)
+  assert(frame_row_mt_info->mi_rows_parse_done >=
+         frame_row_mt_info->mi_rows_decode_started);
+  if (frame_row_mt_info->mi_rows_parse_done ==
+      frame_row_mt_info->mi_rows_decode_started)
     return 0;
 
   // Choose the tile to decode.
@@ -3396,7 +3457,9 @@ static int get_next_job_info(AV1Decoder *const pbi,
           max_mis_to_decode = 0;
         }
         if (num_threads_working == min_threads_working &&
-            num_mis_to_decode > max_mis_to_decode) {
+            num_mis_to_decode > max_mis_to_decode &&
+            num_threads_working <
+                get_max_row_mt_workers_per_tile(cm, tile_data->tile_info)) {
           max_mis_to_decode = num_mis_to_decode;
           tile_row = tile_row_idx;
           tile_col = tile_col_idx;
@@ -3404,6 +3467,8 @@ static int get_next_job_info(AV1Decoder *const pbi,
       }
     }
   }
+  // No job found to process
+  if (tile_row == -1 || tile_col == -1) return 0;
 
   tile_data = pbi->tile_data + tile_row * cm->tile_cols + tile_col;
   tile_info = tile_data->tile_info;
@@ -3417,6 +3482,14 @@ static int get_next_job_info(AV1Decoder *const pbi,
   dec_row_mt_sync->num_threads_working++;
   dec_row_mt_sync->mi_rows_decode_started += sb_mi_size;
   frame_row_mt_info->mi_rows_decode_started += sb_mi_size;
+  assert(frame_row_mt_info->mi_rows_parse_done >=
+         frame_row_mt_info->mi_rows_decode_started);
+#if CONFIG_MULTITHREAD
+  if (frame_row_mt_info->mi_rows_decode_started ==
+      frame_row_mt_info->mi_rows_to_decode) {
+    pthread_cond_broadcast(pbi->row_mt_cond_);
+  }
+#endif
 
   return 1;
 }
@@ -3428,21 +3501,67 @@ static INLINE void signal_parse_sb_row_done(AV1Decoder *const pbi,
 #if CONFIG_MULTITHREAD
   pthread_mutex_lock(pbi->row_mt_mutex_);
 #endif
+  assert(frame_row_mt_info->mi_rows_parse_done >=
+         frame_row_mt_info->mi_rows_decode_started);
   tile_data->dec_row_mt_sync.mi_rows_parse_done += sb_mi_size;
   frame_row_mt_info->mi_rows_parse_done += sb_mi_size;
 #if CONFIG_MULTITHREAD
-  pthread_cond_broadcast(pbi->row_mt_cond_);
+  // A new decode job is available. Wake up one worker thread to handle the
+  // new decode job.
+  // NOTE: This assumes we bump mi_rows_parse_done and mi_rows_decode_started
+  // by the same increment (sb_mi_size).
+  pthread_cond_signal(pbi->row_mt_cond_);
   pthread_mutex_unlock(pbi->row_mt_mutex_);
 #endif
 }
 
+// This function is very similar to decode_tile(). It would be good to figure
+// out how to share code.
+static void parse_tile_row_mt(AV1Decoder *pbi, ThreadData *const td,
+                              TileDataDec *const tile_data) {
+  AV1_COMMON *const cm = &pbi->common;
+  const int sb_mi_size = mi_size_wide[cm->seq_params.sb_size];
+  const int num_planes = av1_num_planes(cm);
+  TileInfo tile_info = tile_data->tile_info;
+  int tile_row = tile_info.tile_row;
+
+  av1_zero_above_context(cm, &td->xd, tile_info.mi_col_start,
+                         tile_info.mi_col_end, tile_row);
+  av1_reset_loop_filter_delta(&td->xd, num_planes);
+  av1_reset_loop_restoration(&td->xd, num_planes);
+
+  for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end;
+       mi_row += cm->seq_params.mib_size) {
+    av1_zero_left_context(&td->xd);
+
+    for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end;
+         mi_col += cm->seq_params.mib_size) {
+      set_cb_buffer(pbi, &td->xd, pbi->cb_buffer_base, num_planes, mi_row,
+                    mi_col);
+
+      // Bit-stream parsing of the superblock
+      decode_partition(pbi, td, mi_row, mi_col, td->bit_reader,
+                       cm->seq_params.sb_size, 0x1);
+
+      if (aom_reader_has_overflowed(td->bit_reader)) {
+        aom_merge_corrupted_flag(&td->xd.corrupted, 1);
+        return;
+      }
+    }
+    signal_parse_sb_row_done(pbi, tile_data, sb_mi_size);
+  }
+
+  int corrupted =
+      (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0;
+  aom_merge_corrupted_flag(&td->xd.corrupted, corrupted);
+}
+
 static int row_mt_worker_hook(void *arg1, void *arg2) {
   DecWorkerData *const thread_data = (DecWorkerData *)arg1;
   AV1Decoder *const pbi = (AV1Decoder *)arg2;
   AV1_COMMON *cm = &pbi->common;
   ThreadData *const td = thread_data->td;
   uint8_t allow_update_cdf;
-  const int sb_mi_size = mi_size_wide[cm->seq_params.sb_size];
   AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info;
   td->xd.corrupted = 0;
 
@@ -3464,55 +3583,54 @@ static int row_mt_worker_hook(void *arg1, void *arg2) {
   }
   thread_data->error_info.setjmp = 1;
 
-  const int num_planes = av1_num_planes(cm);
   allow_update_cdf = cm->large_scale_tile ? 0 : 1;
   allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update;
 
+  set_decode_func_pointers(td, 0x1);
+
   assert(cm->tile_cols > 0);
-  while (1) {
+  while (!td->xd.corrupted) {
     TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info);
 
-    if (cur_job_info != NULL && !td->xd.corrupted) {
+    if (cur_job_info != NULL) {
       const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer;
       TileDataDec *const tile_data = cur_job_info->tile_data;
       tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data,
                             allow_update_cdf);
-
-      set_decode_func_pointers(td, 0x1);
-
+#if CONFIG_MULTITHREAD
+      pthread_mutex_lock(pbi->row_mt_mutex_);
+#endif
+      tile_data->dec_row_mt_sync.num_threads_working++;
+#if CONFIG_MULTITHREAD
+      pthread_mutex_unlock(pbi->row_mt_mutex_);
+#endif
       // decode tile
-      TileInfo tile_info = tile_data->tile_info;
-      int tile_row = tile_info.tile_row;
-
-      av1_zero_above_context(cm, &td->xd, tile_info.mi_col_start,
-                             tile_info.mi_col_end, tile_row);
-      av1_reset_loop_filter_delta(&td->xd, num_planes);
-      av1_reset_loop_restoration(&td->xd, num_planes);
-
-      for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end;
-           mi_row += cm->seq_params.mib_size) {
-        av1_zero_left_context(&td->xd);
-
-        for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end;
-             mi_col += cm->seq_params.mib_size) {
-          set_cb_buffer(pbi, &td->xd, pbi->cb_buffer_base, num_planes, mi_row,
-                        mi_col);
-
-          // Bit-stream parsing of the superblock
-          decode_partition(pbi, td, mi_row, mi_col, td->bit_reader,
-                           cm->seq_params.sb_size, 0x1);
-        }
-        signal_parse_sb_row_done(pbi, tile_data, sb_mi_size);
-      }
-
-      int corrupted =
-          (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0;
-      aom_merge_corrupted_flag(&td->xd.corrupted, corrupted);
+      parse_tile_row_mt(pbi, td, tile_data);
+#if CONFIG_MULTITHREAD
+      pthread_mutex_lock(pbi->row_mt_mutex_);
+#endif
+      tile_data->dec_row_mt_sync.num_threads_working--;
+#if CONFIG_MULTITHREAD
+      pthread_mutex_unlock(pbi->row_mt_mutex_);
+#endif
     } else {
       break;
     }
   }
 
+  if (td->xd.corrupted) {
+    thread_data->error_info.setjmp = 0;
+#if CONFIG_MULTITHREAD
+    pthread_mutex_lock(pbi->row_mt_mutex_);
+#endif
+    frame_row_mt_info->row_mt_exit = 1;
+#if CONFIG_MULTITHREAD
+    pthread_cond_broadcast(pbi->row_mt_cond_);
+    pthread_mutex_unlock(pbi->row_mt_mutex_);
+#endif
+    return 0;
+  }
+
   set_decode_func_pointers(td, 0x2);
 
   while (1) {
@@ -3543,7 +3661,7 @@ static int row_mt_worker_hook(void *arg1, void *arg2) {
     TileInfo tile_info = tile_data->tile_info;
 
     av1_tile_init(&td->xd.tile, cm, tile_row, tile_col);
-    av1_init_macroblockd(cm, &td->xd, td->dqcoeff);
+    av1_init_macroblockd(cm, &td->xd, NULL);
     td->xd.error_info = &thread_data->error_info;
 
     decode_tile_sb_row(pbi, td, tile_info, mi_row);
@@ -3570,7 +3688,7 @@ static int compare_tile_buffers(const void *a, const void *b) {
 static void enqueue_tile_jobs(AV1Decoder *pbi, AV1_COMMON *cm,
                               int tile_rows_start, int tile_rows_end,
                               int tile_cols_start, int tile_cols_end,
-                              int startTile, int endTile) {
+                              int start_tile, int end_tile) {
   AV1DecTileMT *tile_mt_info = &pbi->tile_mt_info;
   TileJobsDec *tile_job_queue = tile_mt_info->job_queue;
   tile_mt_info->jobs_enqueued = 0;
@@ -3578,8 +3696,8 @@ static void enqueue_tile_jobs(AV1Decoder *pbi, AV1_COMMON *cm,
 
   for (int row = tile_rows_start; row < tile_rows_end; row++) {
     for (int col = tile_cols_start; col < tile_cols_end; col++) {
-      if (row * cm->tile_cols + col < startTile ||
-          row * cm->tile_cols + col > endTile)
+      if (row * cm->tile_cols + col < start_tile ||
+          row * cm->tile_cols + col > end_tile)
         continue;
       tile_job_queue->tile_buffer = &pbi->tile_buffers[row][col];
       tile_job_queue->tile_data = pbi->tile_data + row * cm->tile_cols + col;
@@ -3733,6 +3851,7 @@ static void decode_mt_init(AV1Decoder *pbi) {
       ++pbi->num_workers;
 
       winterface->init(worker);
+      worker->thread_name = "aom tile worker";
       if (worker_idx < num_threads - 1 && !winterface->reset(worker)) {
         aom_internal_error(&cm->error, AOM_CODEC_ERROR,
                            "Tile decoder thread creation failed");
@@ -3751,7 +3870,7 @@ static void decode_mt_init(AV1Decoder *pbi) {
       thread_data->error_info.setjmp = 0;
     }
   }
-  const int use_highbd = cm->seq_params.use_highbitdepth ? 1 : 0;
+  const int use_highbd = cm->seq_params.use_highbitdepth;
   const int buf_size = MC_TEMP_BUF_PELS << use_highbd;
   for (worker_idx = 0; worker_idx < pbi->max_threads - 1; ++worker_idx) {
     DecWorkerData *const thread_data = pbi->thread_data + worker_idx;
@@ -3882,6 +4001,7 @@ static void dec_alloc_cb_buf(AV1Decoder *pbi) {
     av1_dec_free_cb_buf(pbi);
     CHECK_MEM_ERROR(cm, pbi->cb_buffer_base,
                     aom_memalign(32, sizeof(*pbi->cb_buffer_base) * size));
+    memset(pbi->cb_buffer_base, 0, sizeof(*pbi->cb_buffer_base) * size);
     pbi->cb_buffer_alloc_size = size;
   }
 }
@@ -3969,7 +4089,8 @@ static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data,
   int tile_cols_start;
   int tile_cols_end;
   int tile_count_tg;
-  int num_workers;
+  int num_workers = 0;
+  int max_threads;
   const uint8_t *raw_data_end = NULL;
   int max_sb_rows = 0;
 
@@ -3985,7 +4106,7 @@ static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data,
     tile_cols_end = tile_cols;
   }
   tile_count_tg = end_tile - start_tile + 1;
-  num_workers = pbi->max_threads;
+  max_threads = pbi->max_threads;
 
   // No tiles to decode.
   if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start ||
@@ -3998,7 +4119,7 @@ static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data,
   assert(tile_rows <= MAX_TILE_ROWS);
   assert(tile_cols <= MAX_TILE_COLS);
   assert(tile_count_tg > 0);
-  assert(num_workers > 0);
+  assert(max_threads > 0);
   assert(start_tile <= end_tile);
   assert(start_tile >= 0 && end_tile < n_tiles);
 
@@ -4030,8 +4151,10 @@ static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data,
 
       max_sb_rows = AOMMAX(max_sb_rows,
                            av1_get_sb_rows_in_tile(cm, tile_data->tile_info));
+      num_workers += get_max_row_mt_workers_per_tile(cm, tile_data->tile_info);
     }
   }
+  num_workers = AOMMIN(num_workers, max_threads);
 
   if (pbi->allocated_row_mt_sync_rows != max_sb_rows) {
     for (int i = 0; i < n_tiles; ++i) {
@@ -4107,7 +4230,7 @@ void av1_read_film_grain_params(AV1_COMMON *cm,
   }
 
   pars->random_seed = aom_rb_read_literal(rb, 16);
-  if (cm->frame_type == INTER_FRAME)
+  if (cm->current_frame.frame_type == INTER_FRAME)
     pars->update_parameters = aom_rb_read_bit(rb);
   else
     pars->update_parameters = 1;
@@ -4116,20 +4239,19 @@ void av1_read_film_grain_params(AV1_COMMON *cm,
 
   if (!pars->update_parameters) {
     // inherit parameters from a previous reference frame
-    RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
     int film_grain_params_ref_idx = aom_rb_read_literal(rb, 3);
-    int buf_idx = cm->ref_frame_map[film_grain_params_ref_idx];
-    if (buf_idx == INVALID_IDX) {
+    RefCntBuffer *const buf = cm->ref_frame_map[film_grain_params_ref_idx];
+    if (buf == NULL) {
       aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
                          "Invalid Film grain reference idx");
     }
-    if (!frame_bufs[buf_idx].film_grain_params_present) {
+    if (!buf->film_grain_params_present) {
       aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
                          "Film grain reference parameters not available");
     }
     uint16_t random_seed = pars->random_seed;
-    *pars = frame_bufs[buf_idx].film_grain_params;  // inherit paramaters
-    pars->random_seed = random_seed;                // with new random seed
+    *pars = buf->film_grain_params;   // inherit paramaters
+    pars->random_seed = random_seed;  // with new random seed
     return;
   }
 
@@ -4286,7 +4408,6 @@ void av1_read_color_config(struct aom_read_bit_buffer *rb,
   if (seq_params->color_primaries == AOM_CICP_CP_BT_709 &&
       seq_params->transfer_characteristics == AOM_CICP_TC_SRGB &&
       seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) {
-    // It would be good to remove this dependency.
     seq_params->subsampling_y = seq_params->subsampling_x = 0;
     seq_params->color_range = 1;  // assume full color-range
     if (!(seq_params->profile == PROFILE_1 ||
@@ -4431,23 +4552,23 @@ void av1_read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb,
     seq_params->enable_masked_compound = 0;
     seq_params->enable_warped_motion = 0;
     seq_params->enable_dual_filter = 0;
-    seq_params->enable_order_hint = 0;
-    seq_params->enable_jnt_comp = 0;
-    seq_params->enable_ref_frame_mvs = 0;
+    seq_params->order_hint_info.enable_order_hint = 0;
+    seq_params->order_hint_info.enable_dist_wtd_comp = 0;
+    seq_params->order_hint_info.enable_ref_frame_mvs = 0;
     seq_params->force_screen_content_tools = 2;  // SELECT_SCREEN_CONTENT_TOOLS
     seq_params->force_integer_mv = 2;            // SELECT_INTEGER_MV
-    seq_params->order_hint_bits_minus_1 = -1;
+    seq_params->order_hint_info.order_hint_bits_minus_1 = -1;
   } else {
     seq_params->enable_interintra_compound = aom_rb_read_bit(rb);
     seq_params->enable_masked_compound = aom_rb_read_bit(rb);
     seq_params->enable_warped_motion = aom_rb_read_bit(rb);
     seq_params->enable_dual_filter = aom_rb_read_bit(rb);
 
-    seq_params->enable_order_hint = aom_rb_read_bit(rb);
-    seq_params->enable_jnt_comp =
-        seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0;
-    seq_params->enable_ref_frame_mvs =
-        seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0;
+    seq_params->order_hint_info.enable_order_hint = aom_rb_read_bit(rb);
+    seq_params->order_hint_info.enable_dist_wtd_comp =
+        seq_params->order_hint_info.enable_order_hint ? aom_rb_read_bit(rb) : 0;
+    seq_params->order_hint_info.enable_ref_frame_mvs =
+        seq_params->order_hint_info.enable_order_hint ? aom_rb_read_bit(rb) : 0;
 
     if (aom_rb_read_bit(rb)) {
       seq_params->force_screen_content_tools =
@@ -4465,8 +4586,10 @@ void av1_read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb,
     } else {
       seq_params->force_integer_mv = 2;  // SELECT_INTEGER_MV
     }
-    seq_params->order_hint_bits_minus_1 =
-        seq_params->enable_order_hint ? aom_rb_read_literal(rb, 3) : -1;
+    seq_params->order_hint_info.order_hint_bits_minus_1 =
+        seq_params->order_hint_info.enable_order_hint
+            ? aom_rb_read_literal(rb, 3)
+            : -1;
   }
 
   seq_params->enable_superres = aom_rb_read_bit(rb);
@@ -4576,7 +4699,7 @@ static void read_global_motion(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
     */
     /*
     printf("Dec Ref %d [%d/%d]: %d %d %d %d\n",
-           frame, cm->current_video_frame, cm->show_frame,
+           frame, cm->current_frame.frame_number, cm->show_frame,
            cm->global_motion[frame].wmmat[0],
            cm->global_motion[frame].wmmat[1],
            cm->global_motion[frame].wmmat[2],
@@ -4587,96 +4710,114 @@ static void read_global_motion(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
          REF_FRAMES * sizeof(WarpedMotionParams));
 }
 
-static void show_existing_frame_reset(AV1Decoder *const pbi,
-                                      int existing_frame_idx) {
+// Release the references to the frame buffers in cm->ref_frame_map and reset
+// all elements of cm->ref_frame_map to NULL.
+static void reset_ref_frame_map(AV1_COMMON *const cm) {
+  BufferPool *const pool = cm->buffer_pool;
+
+  for (int i = 0; i < REF_FRAMES; i++) {
+    decrease_ref_count(cm->ref_frame_map[i], pool);
+    cm->ref_frame_map[i] = NULL;
+  }
+}
+
+// Generate next_ref_frame_map.
+static void generate_next_ref_frame_map(AV1Decoder *const pbi) {
   AV1_COMMON *const cm = &pbi->common;
   BufferPool *const pool = cm->buffer_pool;
-  RefCntBuffer *const frame_bufs = pool->frame_bufs;
 
-  assert(cm->show_existing_frame);
-
-  cm->frame_type = KEY_FRAME;
-
-  pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
-
-  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-    cm->frame_refs[i].idx = INVALID_IDX;
-    cm->frame_refs[i].buf = NULL;
-  }
-
-  if (pbi->need_resync) {
-    memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
-    pbi->need_resync = 0;
-  }
-
-  cm->cur_frame->intra_only = 1;
-
-  if (cm->seq_params.frame_id_numbers_present_flag) {
-    /* If bitmask is set, update reference frame id values and
-       mark frames as valid for reference.
-       Note that the displayed frame be valid for referencing
-       in order to have been selected.
-    */
-    int refresh_frame_flags = pbi->refresh_frame_flags;
-    int display_frame_id = cm->ref_frame_id[existing_frame_idx];
-    for (int i = 0; i < REF_FRAMES; i++) {
-      if ((refresh_frame_flags >> i) & 1) {
-        cm->ref_frame_id[i] = display_frame_id;
-        cm->valid_for_referencing[i] = 1;
-      }
-    }
-  }
-
-  cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
-
-  // Generate next_ref_frame_map.
   lock_buffer_pool(pool);
+  // cm->next_ref_frame_map holds references to frame buffers. After storing a
+  // frame buffer index in cm->next_ref_frame_map, we need to increase the
+  // frame buffer's ref_count.
   int ref_index = 0;
-  for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
+  for (int mask = cm->current_frame.refresh_frame_flags; mask; mask >>= 1) {
     if (mask & 1) {
-      cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
-      ++frame_bufs[cm->new_fb_idx].ref_count;
+      cm->next_ref_frame_map[ref_index] = cm->cur_frame;
     } else {
       cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
     }
-    // Current thread holds the reference frame.
-    if (cm->ref_frame_map[ref_index] >= 0)
-      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
+    if (cm->next_ref_frame_map[ref_index] != NULL)
+      ++cm->next_ref_frame_map[ref_index]->ref_count;
     ++ref_index;
   }
 
   for (; ref_index < REF_FRAMES; ++ref_index) {
     cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
-
-    // Current thread holds the reference frame.
-    if (cm->ref_frame_map[ref_index] >= 0)
-      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
+    if (cm->next_ref_frame_map[ref_index] != NULL)
+      ++cm->next_ref_frame_map[ref_index]->ref_count;
   }
   unlock_buffer_pool(pool);
   pbi->hold_ref_buf = 1;
+}
+
+// If the refresh_frame_flags bitmask is set, update reference frame id values
+// and mark frames as valid for reference.
+static void update_ref_frame_id(AV1_COMMON *const cm, int frame_id) {
+  assert(cm->seq_params.frame_id_numbers_present_flag);
+  int refresh_frame_flags = cm->current_frame.refresh_frame_flags;
+  for (int i = 0; i < REF_FRAMES; i++) {
+    if ((refresh_frame_flags >> i) & 1) {
+      cm->ref_frame_id[i] = frame_id;
+      cm->valid_for_referencing[i] = 1;
+    }
+  }
+}
+
+static void show_existing_frame_reset(AV1Decoder *const pbi,
+                                      int existing_frame_idx) {
+  AV1_COMMON *const cm = &pbi->common;
+
+  assert(cm->show_existing_frame);
+
+  cm->current_frame.frame_type = KEY_FRAME;
+
+  cm->current_frame.refresh_frame_flags = (1 << REF_FRAMES) - 1;
+
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+    cm->remapped_ref_idx[i] = INVALID_IDX;
+  }
+
+  if (pbi->need_resync) {
+    reset_ref_frame_map(cm);
+    pbi->need_resync = 0;
+  }
+
+  // Note that the displayed frame must be valid for referencing in order to
+  // have been selected.
+  if (cm->seq_params.frame_id_numbers_present_flag) {
+    update_ref_frame_id(cm, cm->ref_frame_id[existing_frame_idx]);
+  }
+
+  cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
+
+  generate_next_ref_frame_map(pbi);
 
   // Reload the adapted CDFs from when we originally coded this keyframe
-  *cm->fc = cm->frame_contexts[existing_frame_idx];
+  *cm->fc = cm->next_ref_frame_map[existing_frame_idx]->frame_context;
 }
 
 static INLINE void reset_frame_buffers(AV1_COMMON *cm) {
   RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
   int i;
 
-  memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
-  memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
+  // We have not stored any references to frame buffers in
+  // cm->next_ref_frame_map, so we can directly reset it to all NULL.
+  for (i = 0; i < REF_FRAMES; ++i) {
+    cm->next_ref_frame_map[i] = NULL;
+  }
 
   lock_buffer_pool(cm->buffer_pool);
+  reset_ref_frame_map(cm);
+  assert(cm->cur_frame->ref_count == 1);
   for (i = 0; i < FRAME_BUFFERS; ++i) {
-    if (i != cm->new_fb_idx) {
-      frame_bufs[i].ref_count = 0;
-      cm->buffer_pool->release_fb_cb(cm->buffer_pool->cb_priv,
-                                     &frame_bufs[i].raw_frame_buffer);
-    } else {
-      assert(frame_bufs[i].ref_count == 1);
+    // Reset all unreferenced frame buffers. We can also reset cm->cur_frame
+    // because we are the sole owner of cm->cur_frame.
+    if (frame_bufs[i].ref_count > 0 && &frame_bufs[i] != cm->cur_frame) {
+      continue;
     }
-    frame_bufs[i].cur_frame_offset = 0;
-    av1_zero(frame_bufs[i].ref_frame_offset);
+    frame_bufs[i].order_hint = 0;
+    av1_zero(frame_bufs[i].ref_order_hints);
   }
   av1_zero_unused_internal_frame_buffers(&cm->buffer_pool->int_frame_buffers);
   unlock_buffer_pool(cm->buffer_pool);
@@ -4688,6 +4829,7 @@ static int read_uncompressed_header(AV1Decoder *pbi,
                                     struct aom_read_bit_buffer *rb) {
   AV1_COMMON *const cm = &pbi->common;
   const SequenceHeader *const seq_params = &cm->seq_params;
+  CurrentFrame *const current_frame = &cm->current_frame;
   MACROBLOCKD *const xd = &pbi->mb;
   BufferPool *const pool = cm->buffer_pool;
   RefCntBuffer *const frame_bufs = pool->frame_bufs;
@@ -4697,20 +4839,22 @@ static int read_uncompressed_header(AV1Decoder *pbi,
                        "No sequence header");
   }
 
-  cm->last_frame_type = cm->frame_type;
-  cm->last_intra_only = cm->intra_only;
-
-  // NOTE: By default all coded frames to be used as a reference
-  cm->is_reference_frame = 1;
+  cm->last_frame_type = current_frame->frame_type;
 
   if (seq_params->reduced_still_picture_hdr) {
     cm->show_existing_frame = 0;
     cm->show_frame = 1;
-    cm->frame_type = KEY_FRAME;
+    current_frame->frame_type = KEY_FRAME;
+    if (pbi->sequence_header_changed) {
+      // This is the start of a new coded video sequence.
+      pbi->sequence_header_changed = 0;
+      pbi->decoding_first_frame = 1;
+      reset_frame_buffers(cm);
+    }
     cm->error_resilient_mode = 1;
   } else {
     cm->show_existing_frame = aom_rb_read_bit(rb);
-    cm->reset_decoder_state = 0;
+    pbi->reset_decoder_state = 0;
 
     if (cm->show_existing_frame) {
       if (pbi->sequence_header_changed) {
@@ -4720,7 +4864,11 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       }
       // Show an existing frame directly.
       const int existing_frame_idx = aom_rb_read_literal(rb, 3);
-      const int frame_to_show = cm->ref_frame_map[existing_frame_idx];
+      RefCntBuffer *const frame_to_show = cm->ref_frame_map[existing_frame_idx];
+      if (frame_to_show == NULL) {
+        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                           "Buffer does not contain a decoded frame");
+      }
       if (seq_params->decoder_model_info_present_flag &&
           cm->timing_info.equal_picture_interval == 0) {
         av1_read_temporal_point_info(cm, rb);
@@ -4736,44 +4884,48 @@ static int read_uncompressed_header(AV1Decoder *pbi,
                              "Reference buffer frame ID mismatch");
       }
       lock_buffer_pool(pool);
-      if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
-        unlock_buffer_pool(pool);
-        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                           "Buffer %d does not contain a decoded frame",
-                           frame_to_show);
-      }
-      ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
-      cm->reset_decoder_state =
-          frame_bufs[frame_to_show].frame_type == KEY_FRAME;
+      assert(frame_to_show->ref_count > 0);
+      // cm->cur_frame should be the buffer referenced by the return value
+      // of the get_free_fb() call in av1_receive_compressed_data(), and
+      // generate_next_ref_frame_map() has not been called, so ref_count
+      // should still be 1.
+      assert(cm->cur_frame->ref_count == 1);
+      // assign_frame_buffer_p() decrements ref_count directly rather than
+      // call decrease_ref_count(). If cm->cur_frame->raw_frame_buffer has
+      // already been allocated, it will not be released by
+      // assign_frame_buffer_p()!
+      assert(!cm->cur_frame->raw_frame_buffer.data);
+      assign_frame_buffer_p(&cm->cur_frame, frame_to_show);
+      pbi->reset_decoder_state = frame_to_show->frame_type == KEY_FRAME;
       unlock_buffer_pool(pool);
 
       cm->lf.filter_level[0] = 0;
       cm->lf.filter_level[1] = 0;
       cm->show_frame = 1;
 
-      if (!frame_bufs[frame_to_show].showable_frame) {
+      if (!frame_to_show->showable_frame) {
         aom_merge_corrupted_flag(&xd->corrupted, 1);
       }
-      if (cm->reset_decoder_state) frame_bufs[frame_to_show].showable_frame = 0;
+      if (pbi->reset_decoder_state) frame_to_show->showable_frame = 0;
 
-      cm->film_grain_params = frame_bufs[frame_to_show].film_grain_params;
+      cm->film_grain_params = frame_to_show->film_grain_params;
 
-      if (cm->reset_decoder_state) {
+      if (pbi->reset_decoder_state) {
         show_existing_frame_reset(pbi, existing_frame_idx);
       } else {
-        pbi->refresh_frame_flags = 0;
+        current_frame->refresh_frame_flags = 0;
       }
 
       return 0;
     }
 
-    cm->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2);  // 2 bits
+    current_frame->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2);
     if (pbi->sequence_header_changed) {
-      if (pbi->common.frame_type == KEY_FRAME) {
+      if (current_frame->frame_type == KEY_FRAME) {
         // This is the start of a new coded video sequence.
         pbi->sequence_header_changed = 0;
         pbi->decoding_first_frame = 1;
-        reset_frame_buffers(&pbi->common);
+        reset_frame_buffers(cm);
       } else {
         aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                            "Sequence header has changed without a keyframe.");
@@ -4782,11 +4934,11 @@ static int read_uncompressed_header(AV1Decoder *pbi,
 
     cm->show_frame = aom_rb_read_bit(rb);
     if (seq_params->still_picture &&
-        (cm->frame_type != KEY_FRAME || !cm->show_frame)) {
+        (current_frame->frame_type != KEY_FRAME || !cm->show_frame)) {
       aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                          "Still pictures must be coded as shown keyframes");
     }
-    cm->showable_frame = cm->frame_type != KEY_FRAME;
+    cm->showable_frame = current_frame->frame_type != KEY_FRAME;
     if (cm->show_frame) {
       if (seq_params->decoder_model_info_present_flag &&
           cm->timing_info.equal_picture_interval == 0)
@@ -4796,9 +4948,9 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       cm->showable_frame = aom_rb_read_bit(rb);
     }
     cm->cur_frame->showable_frame = cm->showable_frame;
-    cm->intra_only = cm->frame_type == INTRA_ONLY_FRAME;
     cm->error_resilient_mode =
-        frame_is_sframe(cm) || (cm->frame_type == KEY_FRAME && cm->show_frame)
+        frame_is_sframe(cm) ||
+                (current_frame->frame_type == KEY_FRAME && cm->show_frame)
             ? 1
             : aom_rb_read_bit(rb);
   }
@@ -4820,7 +4972,6 @@ static int read_uncompressed_header(AV1Decoder *pbi,
     cm->cur_frame_force_integer_mv = 0;
   }
 
-  cm->frame_refs_short_signaling = 0;
   int frame_size_override_flag = 0;
   cm->allow_intrabc = 0;
   cm->primary_ref_frame = PRIMARY_REF_NONE;
@@ -4830,8 +4981,9 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       int frame_id_length = seq_params->frame_id_length;
       int diff_len = seq_params->delta_frame_id_length;
       int prev_frame_id = 0;
-      int have_prev_frame_id = !pbi->decoding_first_frame &&
-                               !(cm->frame_type == KEY_FRAME && cm->show_frame);
+      int have_prev_frame_id =
+          !pbi->decoding_first_frame &&
+          !(current_frame->frame_type == KEY_FRAME && cm->show_frame);
       if (have_prev_frame_id) {
         prev_frame_id = cm->current_frame_id;
       }
@@ -4854,7 +5006,7 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       }
       /* Check if some frames need to be marked as not valid for referencing */
       for (int i = 0; i < REF_FRAMES; i++) {
-        if (cm->frame_type == KEY_FRAME && cm->show_frame) {
+        if (current_frame->frame_type == KEY_FRAME && cm->show_frame) {
           cm->valid_for_referencing[i] = 0;
         } else if (cm->current_frame_id - (1 << diff_len) > 0) {
           if (cm->ref_frame_id[i] > cm->current_frame_id ||
@@ -4871,9 +5023,9 @@ static int read_uncompressed_header(AV1Decoder *pbi,
 
     frame_size_override_flag = frame_is_sframe(cm) ? 1 : aom_rb_read_bit(rb);
 
-    cm->frame_offset =
-        aom_rb_read_literal(rb, seq_params->order_hint_bits_minus_1 + 1);
-    cm->current_video_frame = cm->frame_offset;
+    current_frame->order_hint = aom_rb_read_literal(
+        rb, seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
+    current_frame->frame_number = current_frame->order_hint;
 
     if (!cm->error_resilient_mode && !frame_is_intra_only(cm)) {
       cm->primary_ref_frame = aom_rb_read_literal(rb, PRIMARY_REF_BITS);
@@ -4905,89 +5057,84 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       }
     }
   }
-  if (cm->frame_type == KEY_FRAME) {
-    if (!cm->show_frame)  // unshown keyframe (forward keyframe)
-      pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
-    else  // shown keyframe
-      pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
+  if (current_frame->frame_type == KEY_FRAME) {
+    if (!cm->show_frame) {  // unshown keyframe (forward keyframe)
+      current_frame->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
+    } else {  // shown keyframe
+      current_frame->refresh_frame_flags = (1 << REF_FRAMES) - 1;
+    }
 
     for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-      cm->frame_refs[i].idx = INVALID_IDX;
-      cm->frame_refs[i].buf = NULL;
+      cm->remapped_ref_idx[i] = INVALID_IDX;
     }
     if (pbi->need_resync) {
-      memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
+      reset_ref_frame_map(cm);
       pbi->need_resync = 0;
     }
   } else {
-    if (cm->intra_only) {
-      pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
-      if (pbi->refresh_frame_flags == 0xFF) {
+    if (current_frame->frame_type == INTRA_ONLY_FRAME) {
+      current_frame->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES);
+      if (current_frame->refresh_frame_flags == 0xFF) {
         aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
                            "Intra only frames cannot have refresh flags 0xFF");
       }
       if (pbi->need_resync) {
-        memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
+        reset_ref_frame_map(cm);
         pbi->need_resync = 0;
       }
     } else if (pbi->need_resync != 1) { /* Skip if need resync */
-      pbi->refresh_frame_flags =
+      current_frame->refresh_frame_flags =
           frame_is_sframe(cm) ? 0xFF : aom_rb_read_literal(rb, REF_FRAMES);
-      if (!pbi->refresh_frame_flags) {
-        // NOTE: "pbi->refresh_frame_flags == 0" indicates that the coded frame
-        //       will not be used as a reference
-        cm->is_reference_frame = 0;
-      }
     }
   }
 
-  if (!frame_is_intra_only(cm) || pbi->refresh_frame_flags != 0xFF) {
+  if (!frame_is_intra_only(cm) || current_frame->refresh_frame_flags != 0xFF) {
     // Read all ref frame order hints if error_resilient_mode == 1
-    if (cm->error_resilient_mode && seq_params->enable_order_hint) {
+    if (cm->error_resilient_mode &&
+        seq_params->order_hint_info.enable_order_hint) {
       for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) {
         // Read order hint from bit stream
-        unsigned int frame_offset =
-            aom_rb_read_literal(rb, seq_params->order_hint_bits_minus_1 + 1);
-        // Get buffer index
-        int buf_idx = cm->ref_frame_map[ref_idx];
-        assert(buf_idx < FRAME_BUFFERS);
-        if (buf_idx == -1 ||
-            frame_offset != frame_bufs[buf_idx].cur_frame_offset) {
-          if (buf_idx >= 0) {
+        unsigned int order_hint = aom_rb_read_literal(
+            rb, seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
+        // Get buffer
+        RefCntBuffer *buf = cm->ref_frame_map[ref_idx];
+        if (buf == NULL || order_hint != buf->order_hint) {
+          if (buf != NULL) {
             lock_buffer_pool(pool);
-            decrease_ref_count(buf_idx, frame_bufs, pool);
+            decrease_ref_count(buf, pool);
             unlock_buffer_pool(pool);
           }
           // If no corresponding buffer exists, allocate a new buffer with all
           // pixels set to neutral grey.
-          buf_idx = get_free_fb(cm);
+          int buf_idx = get_free_fb(cm);
           if (buf_idx == INVALID_IDX) {
             aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                                "Unable to find free frame buffer");
           }
+          buf = &frame_bufs[buf_idx];
           lock_buffer_pool(pool);
           if (aom_realloc_frame_buffer(
-                  &frame_bufs[buf_idx].buf, seq_params->max_frame_width,
+                  &buf->buf, seq_params->max_frame_width,
                   seq_params->max_frame_height, seq_params->subsampling_x,
                   seq_params->subsampling_y, seq_params->use_highbitdepth,
                   AOM_BORDER_IN_PIXELS, cm->byte_alignment,
-                  &pool->frame_bufs[buf_idx].raw_frame_buffer, pool->get_fb_cb,
-                  pool->cb_priv)) {
+                  &buf->raw_frame_buffer, pool->get_fb_cb, pool->cb_priv)) {
+            decrease_ref_count(buf, pool);
             unlock_buffer_pool(pool);
             aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                                "Failed to allocate frame buffer");
           }
           unlock_buffer_pool(pool);
-          set_planes_to_neutral_grey(seq_params, &frame_bufs[buf_idx].buf, 0);
+          set_planes_to_neutral_grey(seq_params, &buf->buf, 0);
 
-          cm->ref_frame_map[ref_idx] = buf_idx;
-          frame_bufs[buf_idx].cur_frame_offset = frame_offset;
+          cm->ref_frame_map[ref_idx] = buf;
+          buf->order_hint = order_hint;
         }
       }
     }
   }
 
-  if (cm->frame_type == KEY_FRAME) {
+  if (current_frame->frame_type == KEY_FRAME) {
     setup_frame_size(cm, frame_size_override_flag, rb);
 
     if (cm->allow_screen_content_tools && !av1_superres_scaled(cm))
@@ -4997,7 +5144,7 @@ static int read_uncompressed_header(AV1Decoder *pbi,
   } else {
     cm->allow_ref_frame_mvs = 0;
 
-    if (cm->intra_only) {
+    if (current_frame->frame_type == INTRA_ONLY_FRAME) {
       cm->cur_frame->film_grain_params_present =
           seq_params->film_grain_params_present;
       setup_frame_size(cm, frame_size_override_flag, rb);
@@ -5005,30 +5152,30 @@ static int read_uncompressed_header(AV1Decoder *pbi,
         cm->allow_intrabc = aom_rb_read_bit(rb);
 
     } else if (pbi->need_resync != 1) { /* Skip if need resync */
-
+      int frame_refs_short_signaling = 0;
       // Frame refs short signaling is off when error resilient mode is on.
-      if (seq_params->enable_order_hint)
-        cm->frame_refs_short_signaling = aom_rb_read_bit(rb);
+      if (seq_params->order_hint_info.enable_order_hint)
+        frame_refs_short_signaling = aom_rb_read_bit(rb);
 
-      if (cm->frame_refs_short_signaling) {
+      if (frame_refs_short_signaling) {
         // == LAST_FRAME ==
         const int lst_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
-        const int lst_idx = cm->ref_frame_map[lst_ref];
+        const RefCntBuffer *const lst_buf = cm->ref_frame_map[lst_ref];
 
         // == GOLDEN_FRAME ==
         const int gld_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
-        const int gld_idx = cm->ref_frame_map[gld_ref];
+        const RefCntBuffer *const gld_buf = cm->ref_frame_map[gld_ref];
 
         // Most of the time, streams start with a keyframe. In that case,
         // ref_frame_map will have been filled in at that point and will not
-        // contain any -1's. However, streams are explicitly allowed to start
+        // contain any NULLs. However, streams are explicitly allowed to start
         // with an intra-only frame, so long as they don't then signal a
         // reference to a slot that hasn't been set yet. That's what we are
         // checking here.
-        if (lst_idx == -1)
+        if (lst_buf == NULL)
           aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                              "Inter frame requests nonexistent reference");
-        if (gld_idx == -1)
+        if (gld_buf == NULL)
           aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                              "Inter frame requests nonexistent reference");
 
@@ -5037,9 +5184,8 @@ static int read_uncompressed_header(AV1Decoder *pbi,
 
       for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
         int ref = 0;
-        if (!cm->frame_refs_short_signaling) {
+        if (!frame_refs_short_signaling) {
           ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2);
-          const int idx = cm->ref_frame_map[ref];
 
           // Most of the time, streams start with a keyframe. In that case,
           // ref_frame_map will have been filled in at that point and will not
@@ -5047,16 +5193,12 @@ static int read_uncompressed_header(AV1Decoder *pbi,
           // with an intra-only frame, so long as they don't then signal a
           // reference to a slot that hasn't been set yet. That's what we are
           // checking here.
-          if (idx == -1)
+          if (cm->ref_frame_map[ref] == NULL)
             aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                                "Inter frame requests nonexistent reference");
-
-          RefBuffer *const ref_frame = &cm->frame_refs[i];
-          ref_frame->idx = idx;
-          ref_frame->buf = &frame_bufs[idx].buf;
-          ref_frame->map_idx = ref;
+          cm->remapped_ref_idx[i] = ref;
         } else {
-          ref = cm->frame_refs[i].map_idx;
+          ref = cm->remapped_ref_idx[i];
         }
 
         cm->ref_frame_sign_bias[LAST_FRAME + i] = 0;
@@ -5093,26 +5235,29 @@ static int read_uncompressed_header(AV1Decoder *pbi,
       cm->switchable_motion_mode = aom_rb_read_bit(rb);
     }
 
-    cm->prev_frame = get_prev_frame(cm);
+    cm->prev_frame = get_primary_ref_frame_buf(cm);
     if (cm->primary_ref_frame != PRIMARY_REF_NONE &&
-        cm->frame_refs[cm->primary_ref_frame].idx < 0) {
+        get_primary_ref_frame_buf(cm) == NULL) {
       aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                          "Reference frame containing this frame's initial "
                          "frame context is unavailable.");
     }
 
-    if (!cm->intra_only && pbi->need_resync != 1) {
+    if (!(current_frame->frame_type == INTRA_ONLY_FRAME) &&
+        pbi->need_resync != 1) {
       if (frame_might_allow_ref_frame_mvs(cm))
         cm->allow_ref_frame_mvs = aom_rb_read_bit(rb);
       else
         cm->allow_ref_frame_mvs = 0;
 
-      for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
-        RefBuffer *const ref_buf = &cm->frame_refs[i];
+      for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
+        const RefCntBuffer *const ref_buf = get_ref_frame_buf(cm, i);
+        struct scale_factors *const ref_scale_factors =
+            get_ref_scale_factors(cm, i);
         av1_setup_scale_factors_for_frame(
-            &ref_buf->sf, ref_buf->buf->y_crop_width,
-            ref_buf->buf->y_crop_height, cm->width, cm->height);
-        if ((!av1_is_valid_scale(&ref_buf->sf)))
+            ref_scale_factors, ref_buf->buf.y_crop_width,
+            ref_buf->buf.y_crop_height, cm->width, cm->height);
+        if ((!av1_is_valid_scale(ref_scale_factors)))
           aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
                              "Reference frame has invalid dimensions");
       }
@@ -5123,19 +5268,10 @@ static int read_uncompressed_header(AV1Decoder *pbi,
 
   av1_setup_frame_sign_bias(cm);
 
-  cm->cur_frame->intra_only = cm->frame_type == KEY_FRAME || cm->intra_only;
-  cm->cur_frame->frame_type = cm->frame_type;
+  cm->cur_frame->frame_type = current_frame->frame_type;
 
   if (seq_params->frame_id_numbers_present_flag) {
-    /* If bitmask is set, update reference frame id values and
-       mark frames as valid for reference */
-    int refresh_frame_flags = pbi->refresh_frame_flags;
-    for (int i = 0; i < REF_FRAMES; i++) {
-      if ((refresh_frame_flags >> i) & 1) {
-        cm->ref_frame_id[i] = cm->current_frame_id;
-        cm->valid_for_referencing[i] = 1;
-      }
-    }
+    update_ref_frame_id(cm, cm->current_frame_id);
   }
 
   const int might_bwd_adapt =
@@ -5148,18 +5284,17 @@ static int read_uncompressed_header(AV1Decoder *pbi,
     cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
   }
 
-  get_frame_new_buffer(cm)->bit_depth = seq_params->bit_depth;
-  get_frame_new_buffer(cm)->color_primaries = seq_params->color_primaries;
-  get_frame_new_buffer(cm)->transfer_characteristics =
+  cm->cur_frame->buf.bit_depth = seq_params->bit_depth;
+  cm->cur_frame->buf.color_primaries = seq_params->color_primaries;
+  cm->cur_frame->buf.transfer_characteristics =
       seq_params->transfer_characteristics;
-  get_frame_new_buffer(cm)->matrix_coefficients =
-      seq_params->matrix_coefficients;
-  get_frame_new_buffer(cm)->monochrome = seq_params->monochrome;
-  get_frame_new_buffer(cm)->chroma_sample_position =
+  cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients;
+  cm->cur_frame->buf.monochrome = seq_params->monochrome;
+  cm->cur_frame->buf.chroma_sample_position =
       seq_params->chroma_sample_position;
-  get_frame_new_buffer(cm)->color_range = seq_params->color_range;
-  get_frame_new_buffer(cm)->render_width = cm->render_width;
-  get_frame_new_buffer(cm)->render_height = cm->render_height;
+  cm->cur_frame->buf.color_range = seq_params->color_range;
+  cm->cur_frame->buf.render_width = cm->render_width;
+  cm->cur_frame->buf.render_height = cm->render_height;
 
   if (pbi->need_resync) {
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
@@ -5167,41 +5302,17 @@ static int read_uncompressed_header(AV1Decoder *pbi,
                        " state");
   }
 
-  // Generate next_ref_frame_map.
-  lock_buffer_pool(pool);
-  int ref_index = 0;
-  for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
-    if (mask & 1) {
-      cm->next_ref_frame_map[ref_index] = cm->new_fb_idx;
-      ++frame_bufs[cm->new_fb_idx].ref_count;
-    } else {
-      cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
-    }
-    // Current thread holds the reference frame.
-    if (cm->ref_frame_map[ref_index] >= 0)
-      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
-    ++ref_index;
-  }
-
-  for (; ref_index < REF_FRAMES; ++ref_index) {
-    cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index];
-
-    // Current thread holds the reference frame.
-    if (cm->ref_frame_map[ref_index] >= 0)
-      ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count;
-  }
-  unlock_buffer_pool(pool);
-  pbi->hold_ref_buf = 1;
+  generate_next_ref_frame_map(pbi);
 
   if (cm->allow_intrabc) {
     // Set parameters corresponding to no filtering.
     struct loopfilter *lf = &cm->lf;
     lf->filter_level[0] = 0;
     lf->filter_level[1] = 0;
-    cm->cdef_bits = 0;
-    cm->cdef_strengths[0] = 0;
-    cm->nb_cdef_strengths = 1;
-    cm->cdef_uv_strengths[0] = 0;
+    cm->cdef_info.cdef_bits = 0;
+    cm->cdef_info.cdef_strengths[0] = 0;
+    cm->cdef_info.nb_cdef_strengths = 1;
+    cm->cdef_info.cdef_uv_strengths[0] = 0;
     cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
     cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
     cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
@@ -5226,18 +5337,20 @@ static int read_uncompressed_header(AV1Decoder *pbi,
 
   setup_segmentation(cm, rb);
 
-  cm->delta_q_res = 1;
-  cm->delta_lf_res = 1;
-  cm->delta_lf_present_flag = 0;
-  cm->delta_lf_multi = 0;
-  cm->delta_q_present_flag = cm->base_qindex > 0 ? aom_rb_read_bit(rb) : 0;
-  if (cm->delta_q_present_flag) {
+  cm->delta_q_info.delta_q_res = 1;
+  cm->delta_q_info.delta_lf_res = 1;
+  cm->delta_q_info.delta_lf_present_flag = 0;
+  cm->delta_q_info.delta_lf_multi = 0;
+  cm->delta_q_info.delta_q_present_flag =
+      cm->base_qindex > 0 ? aom_rb_read_bit(rb) : 0;
+  if (cm->delta_q_info.delta_q_present_flag) {
     xd->current_qindex = cm->base_qindex;
-    cm->delta_q_res = 1 << aom_rb_read_literal(rb, 2);
-    if (!cm->allow_intrabc) cm->delta_lf_present_flag = aom_rb_read_bit(rb);
-    if (cm->delta_lf_present_flag) {
-      cm->delta_lf_res = 1 << aom_rb_read_literal(rb, 2);
-      cm->delta_lf_multi = aom_rb_read_bit(rb);
+    cm->delta_q_info.delta_q_res = 1 << aom_rb_read_literal(rb, 2);
+    if (!cm->allow_intrabc)
+      cm->delta_q_info.delta_lf_present_flag = aom_rb_read_bit(rb);
+    if (cm->delta_q_info.delta_lf_present_flag) {
+      cm->delta_q_info.delta_lf_res = 1 << aom_rb_read_literal(rb, 2);
+      cm->delta_q_info.delta_lf_multi = aom_rb_read_bit(rb);
       av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
     }
   }
@@ -5245,9 +5358,7 @@ static int read_uncompressed_header(AV1Decoder *pbi,
   xd->cur_frame_force_integer_mv = cm->cur_frame_force_integer_mv;
 
   for (int i = 0; i < MAX_SEGMENTS; ++i) {
-    const int qindex = cm->seg.enabled
-                           ? av1_get_qindex(&cm->seg, i, cm->base_qindex)
-                           : cm->base_qindex;
+    const int qindex = av1_get_qindex(&cm->seg, i, cm->base_qindex);
     xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 &&
                       cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 &&
                       cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0;
@@ -5255,15 +5366,15 @@ static int read_uncompressed_header(AV1Decoder *pbi,
   }
   cm->coded_lossless = is_coded_lossless(cm, xd);
   cm->all_lossless = cm->coded_lossless && !av1_superres_scaled(cm);
-  setup_segmentation_dequant(cm);
+  setup_segmentation_dequant(cm, xd);
   if (cm->coded_lossless) {
     cm->lf.filter_level[0] = 0;
     cm->lf.filter_level[1] = 0;
   }
   if (cm->coded_lossless || !seq_params->enable_cdef) {
-    cm->cdef_bits = 0;
-    cm->cdef_strengths[0] = 0;
-    cm->cdef_uv_strengths[0] = 0;
+    cm->cdef_info.cdef_bits = 0;
+    cm->cdef_info.cdef_strengths[0] = 0;
+    cm->cdef_info.cdef_uv_strengths[0] = 0;
   }
   if (cm->all_lossless || !seq_params->enable_restoration) {
     cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
@@ -5280,11 +5391,13 @@ static int read_uncompressed_header(AV1Decoder *pbi,
   }
 
   cm->tx_mode = read_tx_mode(cm, rb);
-  cm->reference_mode = read_frame_reference_mode(cm, rb);
-  if (cm->reference_mode != SINGLE_REFERENCE) setup_compound_reference_mode(cm);
+  current_frame->reference_mode = read_frame_reference_mode(cm, rb);
+  if (current_frame->reference_mode != SINGLE_REFERENCE)
+    setup_compound_reference_mode(cm);
 
   av1_setup_skip_mode_allowed(cm);
-  cm->skip_mode_flag = cm->is_skip_mode_allowed ? aom_rb_read_bit(rb) : 0;
+  current_frame->skip_mode_info.skip_mode_flag =
+      current_frame->skip_mode_info.skip_mode_allowed ? aom_rb_read_bit(rb) : 0;
 
   if (frame_might_allow_warped_motion(cm))
     cm->allow_warped_motion = aom_rb_read_bit(rb);
@@ -5335,7 +5448,7 @@ BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb) {
   return (BITSTREAM_PROFILE)profile;
 }
 
-void superres_post_decode(AV1Decoder *pbi) {
+static void superres_post_decode(AV1Decoder *pbi) {
   AV1_COMMON *const cm = &pbi->common;
   BufferPool *const pool = cm->buffer_pool;
 
@@ -5357,7 +5470,8 @@ uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi,
   MACROBLOCKD *const xd = &pbi->mb;
 
 #if CONFIG_BITSTREAM_DEBUG
-  bitstream_queue_set_frame_read(cm->current_video_frame * 2 + cm->show_frame);
+  bitstream_queue_set_frame_read(cm->current_frame.frame_number * 2 +
+                                 cm->show_frame);
 #endif
 #if CONFIG_MISMATCH_DEBUG
   mismatch_move_frame_idx_r();
@@ -5383,7 +5497,7 @@ uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi,
 
   const uint32_t uncomp_hdr_size =
       (uint32_t)aom_rb_bytes_read(rb);  // Size of the uncompressed header
-  YV12_BUFFER_CONFIG *new_fb = get_frame_new_buffer(cm);
+  YV12_BUFFER_CONFIG *new_fb = &cm->cur_frame->buf;
   xd->cur_buf = new_fb;
   if (av1_allow_intrabc(cm)) {
     av1_setup_scale_factors_for_frame(
@@ -5394,9 +5508,9 @@ uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi,
   if (cm->show_existing_frame) {
     // showing a frame directly
     *p_data_end = data + uncomp_hdr_size;
-    if (cm->reset_decoder_state) {
+    if (pbi->reset_decoder_state) {
       // Use the default frame context values.
-      *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS];
+      *cm->fc = *cm->default_frame_context;
       if (!cm->fc->initialized)
         aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                            "Uninitialized entropy context.");
@@ -5406,17 +5520,15 @@ uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi,
 
   cm->setup_mi(cm);
 
-  cm->current_frame_seg_map = cm->cur_frame->seg_map;
-
   av1_setup_motion_field(cm);
 
   av1_setup_block_planes(xd, cm->seq_params.subsampling_x,
                          cm->seq_params.subsampling_y, num_planes);
   if (cm->primary_ref_frame == PRIMARY_REF_NONE) {
     // use the default frame context values
-    *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS];
+    *cm->fc = *cm->default_frame_context;
   } else {
-    *cm->fc = cm->frame_contexts[cm->frame_refs[cm->primary_ref_frame].idx];
+    *cm->fc = get_primary_ref_frame_buf(cm)->frame_context;
   }
   if (!cm->fc->initialized)
     aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
@@ -5435,7 +5547,7 @@ static void setup_frame_info(AV1Decoder *pbi) {
       cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
     av1_alloc_restoration_buffers(cm);
   }
-  const int use_highbd = cm->seq_params.use_highbitdepth ? 1 : 0;
+  const int use_highbd = cm->seq_params.use_highbitdepth;
   const int buf_size = MC_TEMP_BUF_PELS << use_highbd;
   if (pbi->td.mc_buf_size != buf_size) {
     av1_free_mc_tmp_buf(&pbi->td);
@@ -5455,7 +5567,6 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
   const int num_planes = av1_num_planes(cm);
 #if LOOP_FILTER_BITMASK
   av1_loop_filter_frame_init(cm, 0, num_planes);
-  av1_zero_array(cm->lf.lfm, cm->lf.lfm_num);
 #endif
 
   if (pbi->max_threads > 1 && !(cm->large_scale_tile && !pbi->ext_tile_debug) &&
@@ -5479,19 +5590,20 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
 
   if (!cm->allow_intrabc && !cm->single_tile_decoding) {
     if (cm->lf.filter_level[0] || cm->lf.filter_level[1]) {
-#if LOOP_FILTER_BITMASK
-      av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 1, 0,
-                            num_planes, 0);
-#else
       if (pbi->num_workers > 1) {
-        av1_loop_filter_frame_mt(get_frame_new_buffer(cm), cm, &pbi->mb, 0,
-                                 num_planes, 0, pbi->tile_workers,
-                                 pbi->num_workers, &pbi->lf_row_sync);
-      } else {
-        av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 0,
-                              num_planes, 0);
-      }
+        av1_loop_filter_frame_mt(
+            &cm->cur_frame->buf, cm, &pbi->mb, 0, num_planes, 0,
+#if LOOP_FILTER_BITMASK
+            1,
 #endif
+            pbi->tile_workers, pbi->num_workers, &pbi->lf_row_sync);
+      } else {
+        av1_loop_filter_frame(&cm->cur_frame->buf, cm, &pbi->mb,
+#if LOOP_FILTER_BITMASK
+                              1,
+#endif
+                              0, num_planes, 0);
+      }
     }
 
     const int do_loop_restoration =
@@ -5500,20 +5612,23 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
         cm->rst_info[2].frame_restoration_type != RESTORE_NONE;
     const int do_cdef =
         !cm->skip_loop_filter && !cm->coded_lossless &&
-        (cm->cdef_bits || cm->cdef_strengths[0] || cm->cdef_uv_strengths[0]);
+        (cm->cdef_info.cdef_bits || cm->cdef_info.cdef_strengths[0] ||
+         cm->cdef_info.cdef_uv_strengths[0]);
     const int do_superres = av1_superres_scaled(cm);
     const int optimized_loop_restoration = !do_cdef && !do_superres;
 
     if (!optimized_loop_restoration) {
       if (do_loop_restoration)
-        av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 0);
+        av1_loop_restoration_save_boundary_lines(&pbi->common.cur_frame->buf,
+                                                 cm, 0);
 
-      if (do_cdef) av1_cdef_frame(&pbi->cur_buf->buf, cm, &pbi->mb);
+      if (do_cdef) av1_cdef_frame(&pbi->common.cur_frame->buf, cm, &pbi->mb);
 
       superres_post_decode(pbi);
 
       if (do_loop_restoration) {
-        av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 1);
+        av1_loop_restoration_save_boundary_lines(&pbi->common.cur_frame->buf,
+                                                 cm, 1);
         if (pbi->num_workers > 1) {
           av1_loop_restoration_filter_frame_mt(
               (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration,
@@ -5542,6 +5657,9 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
       }
     }
   }
+#if LOOP_FILTER_BITMASK
+  av1_zero_array(cm->lf.lfm, cm->lf.lfm_num);
+#endif
 
   if (!xd->corrupted) {
     if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
@@ -5562,6 +5680,6 @@ void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data,
 
   // Non frame parallel update frame context here.
   if (!cm->large_scale_tile) {
-    cm->frame_contexts[cm->new_fb_idx] = *cm->fc;
+    cm->cur_frame->frame_context = *cm->fc;
   }
 }
diff --git a/third_party/aom/av1/decoder/decodeframe.h b/third_party/aom/av1/decoder/decodeframe.h
index ddad273f18..13b969663b 100644
--- a/third_party/aom/av1/decoder/decodeframe.h
+++ b/third_party/aom/av1/decoder/decodeframe.h
@@ -46,8 +46,8 @@ uint32_t av1_decode_frame_headers_and_setup(struct AV1Decoder *pbi,
 
 void av1_decode_tg_tiles_and_wrapup(struct AV1Decoder *pbi, const uint8_t *data,
                                     const uint8_t *data_end,
-                                    const uint8_t **p_data_end, int startTile,
-                                    int endTile, int initialize_flag);
+                                    const uint8_t **p_data_end, int start_tile,
+                                    int end_tile, int initialize_flag);
 
 // Implements the color_config() function in the spec. Reports errors by
 // calling rb->error_handler() or aom_internal_error().
diff --git a/third_party/aom/av1/decoder/decodemv.c b/third_party/aom/av1/decoder/decodemv.c
index 551e4d5437..b7431f2489 100644
--- a/third_party/aom/av1/decoder/decodemv.c
+++ b/third_party/aom/av1/decoder/decodemv.c
@@ -41,7 +41,7 @@ static void read_cdef(AV1_COMMON *cm, aom_reader *r, MACROBLOCKD *const xd,
   MB_MODE_INFO *const mbmi = xd->mi[0];
   if (cm->coded_lossless) return;
   if (cm->allow_intrabc) {
-    assert(cm->cdef_bits == 0);
+    assert(cm->cdef_info.cdef_bits == 0);
     return;
   }
 
@@ -59,7 +59,7 @@ static void read_cdef(AV1_COMMON *cm, aom_reader *r, MACROBLOCKD *const xd,
   cm->mi_grid_visible[(mi_row & m) * cm->mi_stride + (mi_col & m)]
       ->cdef_strength = xd->cdef_preset[index] =
       xd->cdef_preset[index] == -1 && !mbmi->skip
-          ? aom_read_literal(r, cm->cdef_bits, ACCT_STR)
+          ? aom_read_literal(r, cm->cdef_info.cdef_bits, ACCT_STR)
           : xd->cdef_preset[index];
 }
 
@@ -299,7 +299,7 @@ static void set_segment_id(AV1_COMMON *cm, int mi_offset, int x_mis, int y_mis,
 
   for (int y = 0; y < y_mis; y++)
     for (int x = 0; x < x_mis; x++)
-      cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
+      cm->cur_frame->seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
 }
 
 static int read_intra_segment_id(AV1_COMMON *const cm,
@@ -355,7 +355,7 @@ static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
   if (!seg->enabled) return 0;  // Default for disabled segmentation
 
   if (!seg->update_map) {
-    copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
+    copy_segment_id(cm, cm->last_frame_seg_map, cm->cur_frame->seg_map,
                     mi_offset, x_mis, y_mis);
     return get_predicted_segment_id(cm, mi_offset, x_mis, y_mis);
   }
@@ -364,7 +364,6 @@ static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
   if (preskip) {
     if (!seg->segid_preskip) return 0;
   } else {
-    if (seg->segid_preskip) return mbmi->segment_id;
     if (mbmi->skip) {
       if (seg->temporal_update) {
         mbmi->seg_id_predicted = 0;
@@ -395,7 +394,7 @@ static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd,
 
 static int read_skip_mode(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id,
                           aom_reader *r) {
-  if (!cm->skip_mode_flag) return 0;
+  if (!cm->current_frame.skip_mode_info.skip_mode_flag) return 0;
 
   if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
     return 0;
@@ -612,9 +611,8 @@ void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd, int blk_row,
     return;
 
   // No need to read transform type for lossless mode(qindex==0).
-  const int qindex =
-      cm->seg.enabled ? xd->qindex[mbmi->segment_id] : cm->base_qindex;
-  if (qindex <= 0) return;
+  const int qindex = xd->qindex[mbmi->segment_id];
+  if (qindex == 0) return;
 
   const int inter_block = is_inter_block(mbmi);
   if (get_ext_tx_types(tx_size, inter_block, cm->reduced_tx_set_used) > 1) {
@@ -701,7 +699,8 @@ static void read_intrabc_info(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                                      mi_col, bsize, r);
     if (!valid_dv) {
       // Intra bc motion vectors are not valid - signal corrupt frame
-      aom_merge_corrupted_flag(&xd->corrupted, 1);
+      aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
+                         "Invalid intrabc dv");
     }
   }
 }
@@ -711,15 +710,17 @@ static void read_intrabc_info(AV1_COMMON *const cm, MACROBLOCKD *const xd,
 static void read_delta_q_params(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                                 const int mi_row, const int mi_col,
                                 aom_reader *r) {
-  if (cm->delta_q_present_flag) {
+  DeltaQInfo *const delta_q_info = &cm->delta_q_info;
+
+  if (delta_q_info->delta_q_present_flag) {
     MB_MODE_INFO *const mbmi = xd->mi[0];
-    xd->current_qindex +=
-        read_delta_qindex(cm, xd, r, mbmi, mi_col, mi_row) * cm->delta_q_res;
+    xd->current_qindex += read_delta_qindex(cm, xd, r, mbmi, mi_col, mi_row) *
+                          delta_q_info->delta_q_res;
     /* Normative: Clamp to [1,MAXQ] to not interfere with lossless mode */
     xd->current_qindex = clamp(xd->current_qindex, 1, MAXQ);
     FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
-    if (cm->delta_lf_present_flag) {
-      if (cm->delta_lf_multi) {
+    if (delta_q_info->delta_lf_present_flag) {
+      if (delta_q_info->delta_lf_multi) {
         const int frame_lf_count =
             av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
         for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
@@ -727,7 +728,7 @@ static void read_delta_q_params(AV1_COMMON *const cm, MACROBLOCKD *const xd,
               xd->delta_lf[lf_id] +
               read_delta_lflevel(cm, r, ec_ctx->delta_lf_multi_cdf[lf_id], mbmi,
                                  mi_col, mi_row) *
-                  cm->delta_lf_res;
+                  delta_q_info->delta_lf_res;
           mbmi->delta_lf[lf_id] = xd->delta_lf[lf_id] =
               clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
         }
@@ -735,7 +736,7 @@ static void read_delta_q_params(AV1_COMMON *const cm, MACROBLOCKD *const xd,
         const int tmp_lvl = xd->delta_lf_from_base +
                             read_delta_lflevel(cm, r, ec_ctx->delta_lf_cdf,
                                                mbmi, mi_col, mi_row) *
-                                cm->delta_lf_res;
+                                delta_q_info->delta_lf_res;
         mbmi->delta_lf_from_base = xd->delta_lf_from_base =
             clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
       }
@@ -882,14 +883,14 @@ static REFERENCE_MODE read_block_reference_mode(AV1_COMMON *cm,
                                                 const MACROBLOCKD *xd,
                                                 aom_reader *r) {
   if (!is_comp_ref_allowed(xd->mi[0]->sb_type)) return SINGLE_REFERENCE;
-  if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+  if (cm->current_frame.reference_mode == REFERENCE_MODE_SELECT) {
     const int ctx = av1_get_reference_mode_context(xd);
     const REFERENCE_MODE mode = (REFERENCE_MODE)aom_read_symbol(
         r, xd->tile_ctx->comp_inter_cdf[ctx], 2, ACCT_STR);
     return mode;  // SINGLE_REFERENCE or COMPOUND_REFERENCE
   } else {
-    assert(cm->reference_mode == SINGLE_REFERENCE);
-    return cm->reference_mode;
+    assert(cm->current_frame.reference_mode == SINGLE_REFERENCE);
+    return cm->current_frame.reference_mode;
   }
 }
 
@@ -907,8 +908,8 @@ static COMP_REFERENCE_TYPE read_comp_reference_type(const MACROBLOCKD *xd,
 
 static void set_ref_frames_for_skip_mode(AV1_COMMON *const cm,
                                          MV_REFERENCE_FRAME ref_frame[2]) {
-  ref_frame[0] = LAST_FRAME + cm->ref_frame_idx_0;
-  ref_frame[1] = LAST_FRAME + cm->ref_frame_idx_1;
+  ref_frame[0] = LAST_FRAME + cm->current_frame.skip_mode_info.ref_frame_idx_0;
+  ref_frame[1] = LAST_FRAME + cm->current_frame.skip_mode_info.ref_frame_idx_1;
 }
 
 // Read the referncence frame
@@ -1231,16 +1232,16 @@ static void dec_dump_logs(AV1_COMMON *cm, MB_MODE_INFO *const mbmi, int mi_row,
   }
 
 #define FRAME_TO_CHECK 11
-  if (cm->current_video_frame == FRAME_TO_CHECK && cm->show_frame == 1) {
+  if (cm->current_frame.frame_number == FRAME_TO_CHECK && cm->show_frame == 1) {
     printf(
         "=== DECODER ===: "
         "Frame=%d, (mi_row,mi_col)=(%d,%d), skip_mode=%d, mode=%d, bsize=%d, "
         "show_frame=%d, mv[0]=(%d,%d), mv[1]=(%d,%d), ref[0]=%d, "
         "ref[1]=%d, motion_mode=%d, mode_ctx=%d, "
         "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d, tx_size=%d\n",
-        cm->current_video_frame, mi_row, mi_col, mbmi->skip_mode, mbmi->mode,
-        mbmi->sb_type, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col,
-        mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0],
+        cm->current_frame.frame_number, mi_row, mi_col, mbmi->skip_mode,
+        mbmi->mode, mbmi->sb_type, cm->show_frame, mv[0].as_mv.row,
+        mv[0].as_mv.col, mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0],
         mbmi->ref_frame[1], mbmi->motion_mode, mode_ctx, newmv_ctx, zeromv_ctx,
         refmv_ctx, mbmi->tx_size);
   }
@@ -1351,16 +1352,12 @@ static void read_inter_block_mode_info(AV1Decoder *const pbi,
     }
   }
 
-  if (mbmi->skip_mode) {
-    assert(mbmi->mode == NEAREST_NEARESTMV);
-    mbmi->mv[0].as_int = nearestmv[0].as_int;
-    mbmi->mv[1].as_int = nearestmv[1].as_int;
-  } else {
-    int mv_corrupted_flag =
-        !assign_mv(cm, xd, mbmi->mode, mbmi->ref_frame, mbmi->mv, ref_mv,
-                   nearestmv, nearmv, mi_row, mi_col, is_compound, allow_hp, r);
-    aom_merge_corrupted_flag(&xd->corrupted, mv_corrupted_flag);
-  }
+  if (mbmi->skip_mode) assert(mbmi->mode == NEAREST_NEARESTMV);
+
+  int mv_corrupted_flag =
+      !assign_mv(cm, xd, mbmi->mode, mbmi->ref_frame, mbmi->mv, ref_mv,
+                 nearestmv, nearmv, mi_row, mi_col, is_compound, allow_hp, r);
+  aom_merge_corrupted_flag(&xd->corrupted, mv_corrupted_flag);
 
   mbmi->use_wedge_interintra = 0;
   if (cm->seq_params.enable_interintra_compound && !mbmi->skip_mode &&
@@ -1391,9 +1388,7 @@ static void read_inter_block_mode_info(AV1Decoder *const pbi,
 
   for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
     const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
-    RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
-
-    xd->block_refs[ref] = ref_buf;
+    xd->block_ref_scale_factors[ref] = get_ref_scale_factors_const(cm, frame);
   }
 
   mbmi->motion_mode = SIMPLE_TRANSLATION;
@@ -1422,7 +1417,7 @@ static void read_inter_block_mode_info(AV1Decoder *const pbi,
     }
 
     if (mbmi->comp_group_idx == 0) {
-      if (cm->seq_params.enable_jnt_comp) {
+      if (cm->seq_params.order_hint_info.enable_dist_wtd_comp) {
         const int comp_index_ctx = get_comp_index_context(cm, xd);
         mbmi->compound_idx = aom_read_symbol(
             r, ec_ctx->compound_index_cdf[comp_index_ctx], 2, ACCT_STR);
@@ -1431,7 +1426,7 @@ static void read_inter_block_mode_info(AV1Decoder *const pbi,
         mbmi->compound_idx = 1;
       }
     } else {
-      assert(cm->reference_mode != SINGLE_REFERENCE &&
+      assert(cm->current_frame.reference_mode != SINGLE_REFERENCE &&
              is_inter_compound_mode(mbmi->mode) &&
              mbmi->motion_mode == SIMPLE_TRANSLATION);
       assert(masked_compound_used);
@@ -1505,7 +1500,8 @@ static void read_inter_frame_mode_info(AV1Decoder *const pbi,
   else
     mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
 
-  mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 0, r);
+  if (!cm->seg.segid_preskip)
+    mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 0, r);
 
   read_cdef(cm, r, xd, mi_col, mi_row);
 
diff --git a/third_party/aom/av1/decoder/decoder.c b/third_party/aom/av1/decoder/decoder.c
index a5f4fd67fa..bff4b7a697 100644
--- a/third_party/aom/av1/decoder/decoder.c
+++ b/third_party/aom/av1/decoder/decoder.c
@@ -17,6 +17,7 @@
 #include "config/aom_dsp_rtcd.h"
 #include "config/aom_scale_rtcd.h"
 
+#include "aom_dsp/aom_dsp_common.h"
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/system_state.h"
 #include "aom_ports/aom_once.h"
@@ -71,12 +72,11 @@ static void dec_free_mi(AV1_COMMON *cm) {
 
 AV1Decoder *av1_decoder_create(BufferPool *const pool) {
   AV1Decoder *volatile const pbi = aom_memalign(32, sizeof(*pbi));
-  AV1_COMMON *volatile const cm = pbi ? &pbi->common : NULL;
-
-  if (!cm) return NULL;
-
+  if (!pbi) return NULL;
   av1_zero(*pbi);
 
+  AV1_COMMON *volatile const cm = &pbi->common;
+
   // The jmp_buf is valid only for the duration of the function that calls
   // setjmp(). Therefore, this function must reset the 'setjmp' field to 0
   // before it returns.
@@ -90,25 +90,26 @@ AV1Decoder *av1_decoder_create(BufferPool *const pool) {
 
   CHECK_MEM_ERROR(cm, cm->fc,
                   (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc)));
-  CHECK_MEM_ERROR(cm, cm->frame_contexts,
-                  (FRAME_CONTEXT *)aom_memalign(
-                      32, FRAME_CONTEXTS * sizeof(*cm->frame_contexts)));
+  CHECK_MEM_ERROR(
+      cm, cm->default_frame_context,
+      (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->default_frame_context)));
   memset(cm->fc, 0, sizeof(*cm->fc));
-  memset(cm->frame_contexts, 0, FRAME_CONTEXTS * sizeof(*cm->frame_contexts));
+  memset(cm->default_frame_context, 0, sizeof(*cm->default_frame_context));
 
   pbi->need_resync = 1;
   aom_once(initialize_dec);
 
   // Initialize the references to not point to any frame buffers.
-  memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
-  memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
+  for (int i = 0; i < REF_FRAMES; i++) {
+    cm->ref_frame_map[i] = NULL;
+    cm->next_ref_frame_map[i] = NULL;
+  }
 
-  cm->current_video_frame = 0;
+  cm->current_frame.frame_number = 0;
   pbi->decoding_first_frame = 1;
   pbi->common.buffer_pool = pool;
 
   cm->seq_params.bit_depth = AOM_BITS_8;
-  cm->dequant_bit_depth = AOM_BITS_8;
 
   cm->alloc_mi = av1_dec_alloc_mi;
   cm->free_mi = dec_free_mi;
@@ -126,6 +127,7 @@ AV1Decoder *av1_decoder_create(BufferPool *const pool) {
   cm->error.setjmp = 0;
 
   aom_get_worker_interface()->init(&pbi->lf_worker);
+  pbi->lf_worker.thread_name = "aom lf worker";
 
   return pbi;
 }
@@ -157,8 +159,7 @@ void av1_decoder_remove(AV1Decoder *pbi) {
   if (!pbi) return;
 
   // Free the tile list output buffer.
-  if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output);
-  pbi->tile_list_output = NULL;
+  aom_free_frame_buffer(&pbi->tile_list_outbuf);
 
   aom_get_worker_interface()->end(&pbi->lf_worker);
   aom_free(pbi->lf_worker.data1);
@@ -318,14 +319,36 @@ aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm,
   return cm->error.error_code;
 }
 
-/* If any buffer updating is signaled it should be done here.
-   Consumes a reference to cm->new_fb_idx.
-*/
+static void release_frame_buffers(AV1Decoder *pbi) {
+  AV1_COMMON *const cm = &pbi->common;
+  BufferPool *const pool = cm->buffer_pool;
+
+  cm->cur_frame->buf.corrupted = 1;
+  lock_buffer_pool(pool);
+  // Release all the reference buffers in cm->next_ref_frame_map if the worker
+  // thread is holding them.
+  if (pbi->hold_ref_buf) {
+    for (int ref_index = 0; ref_index < REF_FRAMES; ++ref_index) {
+      decrease_ref_count(cm->next_ref_frame_map[ref_index], pool);
+      cm->next_ref_frame_map[ref_index] = NULL;
+    }
+    pbi->hold_ref_buf = 0;
+  }
+  // Release current frame.
+  decrease_ref_count(cm->cur_frame, pool);
+  unlock_buffer_pool(pool);
+  cm->cur_frame = NULL;
+}
+
+// If any buffer updating is signaled it should be done here.
+// Consumes a reference to cm->cur_frame.
+//
+// This functions returns void. It reports failure by setting
+// cm->error.error_code.
 static void swap_frame_buffers(AV1Decoder *pbi, int frame_decoded) {
   int ref_index = 0, mask;
   AV1_COMMON *const cm = &pbi->common;
   BufferPool *const pool = cm->buffer_pool;
-  RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
 
   if (frame_decoded) {
     lock_buffer_pool(pool);
@@ -333,73 +356,78 @@ static void swap_frame_buffers(AV1Decoder *pbi, int frame_decoded) {
     // In ext-tile decoding, the camera frame header is only decoded once. So,
     // we don't release the references here.
     if (!pbi->camera_frame_header_ready) {
-      for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
-        const int old_idx = cm->ref_frame_map[ref_index];
-        // Current thread releases the holding of reference frame.
-        decrease_ref_count(old_idx, frame_bufs, pool);
+      // If we are not holding reference buffers in cm->next_ref_frame_map,
+      // assert that the following two for loops are no-ops.
+      assert(IMPLIES(!pbi->hold_ref_buf,
+                     cm->current_frame.refresh_frame_flags == 0));
+      assert(IMPLIES(!pbi->hold_ref_buf,
+                     cm->show_existing_frame && !pbi->reset_decoder_state));
 
-        // Release the reference frame holding in the reference map for the
-        // decoding of the next frame.
-        if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool);
+      // The following two for loops need to release the reference stored in
+      // cm->ref_frame_map[ref_index] before transferring the reference stored
+      // in cm->next_ref_frame_map[ref_index] to cm->ref_frame_map[ref_index].
+      for (mask = cm->current_frame.refresh_frame_flags; mask; mask >>= 1) {
+        decrease_ref_count(cm->ref_frame_map[ref_index], pool);
         cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
+        cm->next_ref_frame_map[ref_index] = NULL;
         ++ref_index;
       }
 
-      // Current thread releases the holding of reference frame.
       const int check_on_show_existing_frame =
-          !cm->show_existing_frame || cm->reset_decoder_state;
+          !cm->show_existing_frame || pbi->reset_decoder_state;
       for (; ref_index < REF_FRAMES && check_on_show_existing_frame;
            ++ref_index) {
-        const int old_idx = cm->ref_frame_map[ref_index];
-        decrease_ref_count(old_idx, frame_bufs, pool);
+        decrease_ref_count(cm->ref_frame_map[ref_index], pool);
         cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
+        cm->next_ref_frame_map[ref_index] = NULL;
       }
     }
 
-    YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm);
-
     if (cm->show_existing_frame || cm->show_frame) {
       if (pbi->output_all_layers) {
         // Append this frame to the output queue
         if (pbi->num_output_frames >= MAX_NUM_SPATIAL_LAYERS) {
           // We can't store the new frame anywhere, so drop it and return an
           // error
-          decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
+          cm->cur_frame->buf.corrupted = 1;
+          decrease_ref_count(cm->cur_frame, pool);
           cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM;
         } else {
-          pbi->output_frames[pbi->num_output_frames] = cur_frame;
-          pbi->output_frame_index[pbi->num_output_frames] = cm->new_fb_idx;
+          pbi->output_frames[pbi->num_output_frames] = cm->cur_frame;
           pbi->num_output_frames++;
         }
       } else {
         // Replace any existing output frame
         assert(pbi->num_output_frames == 0 || pbi->num_output_frames == 1);
         if (pbi->num_output_frames > 0) {
-          decrease_ref_count((int)pbi->output_frame_index[0], frame_bufs, pool);
+          decrease_ref_count(pbi->output_frames[0], pool);
         }
-        pbi->output_frames[0] = cur_frame;
-        pbi->output_frame_index[0] = cm->new_fb_idx;
+        pbi->output_frames[0] = cm->cur_frame;
         pbi->num_output_frames = 1;
       }
     } else {
-      decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
+      decrease_ref_count(cm->cur_frame, pool);
     }
 
     unlock_buffer_pool(pool);
   } else {
+    // The code here assumes we are not holding reference buffers in
+    // cm->next_ref_frame_map. If this assertion fails, we are leaking the
+    // frame buffer references in cm->next_ref_frame_map.
+    assert(IMPLIES(!pbi->camera_frame_header_ready, !pbi->hold_ref_buf));
     // Nothing was decoded, so just drop this frame buffer
     lock_buffer_pool(pool);
-    decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
+    decrease_ref_count(cm->cur_frame, pool);
     unlock_buffer_pool(pool);
   }
+  cm->cur_frame = NULL;
 
   if (!pbi->camera_frame_header_ready) {
     pbi->hold_ref_buf = 0;
 
     // Invalidate these references until the next frame starts.
     for (ref_index = 0; ref_index < INTER_REFS_PER_FRAME; ref_index++) {
-      cm->frame_refs[ref_index].idx = INVALID_IDX;
-      cm->frame_refs[ref_index].buf = NULL;
+      cm->remapped_ref_idx[ref_index] = INVALID_IDX;
     }
   }
 }
@@ -407,10 +435,9 @@ static void swap_frame_buffers(AV1Decoder *pbi, int frame_decoded) {
 int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
                                 const uint8_t **psource) {
   AV1_COMMON *volatile const cm = &pbi->common;
-  BufferPool *volatile const pool = cm->buffer_pool;
-  RefCntBuffer *volatile const frame_bufs = cm->buffer_pool->frame_bufs;
   const uint8_t *source = *psource;
   cm->error.error_code = AOM_CODEC_OK;
+  cm->error.has_detail = 0;
 
   if (size == 0) {
     // This is used to signal that we are missing frames.
@@ -421,29 +448,17 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
     // TODO(jkoleszar): Error concealment is undefined and non-normative
     // at this point, but if it becomes so, [0] may not always be the correct
     // thing to do here.
-    if (cm->frame_refs[0].idx > 0) {
-      assert(cm->frame_refs[0].buf != NULL);
-      cm->frame_refs[0].buf->corrupted = 1;
-    }
+    RefCntBuffer *ref_buf = get_ref_frame_buf(cm, LAST_FRAME);
+    if (ref_buf != NULL) ref_buf->buf.corrupted = 1;
   }
 
-  // Find a free buffer for the new frame, releasing the reference previously
-  // held.
-
-  // Find a free frame buffer. Return error if can not find any.
-  cm->new_fb_idx = get_free_fb(cm);
-  if (cm->new_fb_idx == INVALID_IDX) {
+  if (assign_cur_frame_new_fb(cm) == NULL) {
     cm->error.error_code = AOM_CODEC_MEM_ERROR;
     return 1;
   }
 
-  // Assign a MV array to the frame buffer.
-  cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
-
   if (!pbi->camera_frame_header_ready) pbi->hold_ref_buf = 0;
 
-  pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
-
   // The jmp_buf is valid only for the duration of the function that calls
   // setjmp(). Therefore, this function must reset the 'setjmp' field to 0
   // before it returns.
@@ -460,35 +475,7 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
       winterface->sync(&pbi->tile_workers[i]);
     }
 
-    lock_buffer_pool(pool);
-    // Release all the reference buffers if worker thread is holding them.
-    if (pbi->hold_ref_buf == 1) {
-      int ref_index = 0, mask;
-      for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
-        const int old_idx = cm->ref_frame_map[ref_index];
-        // Current thread releases the holding of reference frame.
-        decrease_ref_count(old_idx, frame_bufs, pool);
-
-        // Release the reference frame holding in the reference map for the
-        // decoding of the next frame.
-        if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool);
-        ++ref_index;
-      }
-
-      // Current thread releases the holding of reference frame.
-      const int check_on_show_existing_frame =
-          !cm->show_existing_frame || cm->reset_decoder_state;
-      for (; ref_index < REF_FRAMES && check_on_show_existing_frame;
-           ++ref_index) {
-        const int old_idx = cm->ref_frame_map[ref_index];
-        decrease_ref_count(old_idx, frame_bufs, pool);
-      }
-      pbi->hold_ref_buf = 0;
-    }
-    // Release current frame.
-    decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
-    unlock_buffer_pool(pool);
-
+    release_frame_buffers(pbi);
     aom_clear_system_state();
     return -1;
   }
@@ -498,10 +485,9 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   int frame_decoded =
       aom_decode_frame_from_obus(pbi, source, source + size, psource);
 
-  if (cm->error.error_code != AOM_CODEC_OK) {
-    lock_buffer_pool(pool);
-    decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
-    unlock_buffer_pool(pool);
+  if (frame_decoded < 0) {
+    assert(cm->error.error_code != AOM_CODEC_OK);
+    release_frame_buffers(pbi);
     cm->error.setjmp = 0;
     return 1;
   }
@@ -515,7 +501,7 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   cm->txb_count = 0;
 #endif
 
-  // Note: At this point, this function holds a reference to cm->new_fb_idx
+  // Note: At this point, this function holds a reference to cm->cur_frame
   // in the buffer pool. This reference is consumed by swap_frame_buffers().
   swap_frame_buffers(pbi, frame_decoded);
 
@@ -531,8 +517,6 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   aom_clear_system_state();
 
   if (!cm->show_existing_frame) {
-    cm->last_show_frame = cm->show_frame;
-
     if (cm->seg.enabled) {
       if (cm->prev_frame && (cm->mi_rows == cm->prev_frame->mi_rows) &&
           (cm->mi_cols == cm->prev_frame->mi_cols)) {
@@ -544,10 +528,6 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
   }
 
   // Update progress in frame parallel decode.
-  cm->last_width = cm->width;
-  cm->last_height = cm->height;
-  cm->last_tile_cols = cm->tile_cols;
-  cm->last_tile_rows = cm->tile_rows;
   cm->error.setjmp = 0;
 
   return 0;
@@ -556,11 +536,9 @@ int av1_receive_compressed_data(AV1Decoder *pbi, size_t size,
 // Get the frame at a particular index in the output queue
 int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd,
                       aom_film_grain_t **grain_params) {
-  RefCntBuffer *const frame_bufs = pbi->common.buffer_pool->frame_bufs;
-
   if (index >= pbi->num_output_frames) return -1;
-  *sd = pbi->output_frames[index];
-  *grain_params = &frame_bufs[pbi->output_frame_index[index]].film_grain_params;
+  *sd = &pbi->output_frames[index]->buf;
+  *grain_params = &pbi->output_frames[index]->film_grain_params;
   aom_clear_system_state();
   return 0;
 }
@@ -570,6 +548,6 @@ int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd,
 int av1_get_frame_to_show(AV1Decoder *pbi, YV12_BUFFER_CONFIG *frame) {
   if (pbi->num_output_frames == 0) return -1;
 
-  *frame = *pbi->output_frames[pbi->num_output_frames - 1];
+  *frame = pbi->output_frames[pbi->num_output_frames - 1]->buf;
   return 0;
 }
diff --git a/third_party/aom/av1/decoder/decoder.h b/third_party/aom/av1/decoder/decoder.h
index 5ca939c245..467c8a2fc1 100644
--- a/third_party/aom/av1/decoder/decoder.h
+++ b/third_party/aom/av1/decoder/decoder.h
@@ -50,8 +50,6 @@ typedef void (*cfl_store_inter_block_visitor_fn_t)(AV1_COMMON *const cm,
 typedef struct ThreadData {
   aom_reader *bit_reader;
   DECLARE_ALIGNED(32, MACROBLOCKD, xd);
-  /* dqcoeff are shared by all the planes. So planes must be decoded serially */
-  DECLARE_ALIGNED(32, tran_low_t, dqcoeff[MAX_TX_SQUARE]);
   CB_BUFFER cb_buffer_base;
   uint8_t *mc_buf[2];
   int32_t mc_buf_size;
@@ -97,9 +95,26 @@ typedef struct AV1DecRowMTInfo {
   int tile_cols_end;
   int start_tile;
   int end_tile;
-  int mi_rows_parse_done;
-  int mi_rows_decode_started;
   int mi_rows_to_decode;
+
+  // Invariant:
+  //   mi_rows_parse_done >= mi_rows_decode_started.
+  // mi_rows_parse_done and mi_rows_decode_started are both initialized to 0.
+  // mi_rows_parse_done is incremented freely. mi_rows_decode_started may only
+  // be incremented to catch up with mi_rows_parse_done but is not allowed to
+  // surpass mi_rows_parse_done.
+  //
+  // When mi_rows_decode_started reaches mi_rows_to_decode, there are no more
+  // decode jobs.
+
+  // Indicates the progress of the bit-stream parsing of superblocks.
+  // Initialized to 0. Incremented by sb_mi_size when parse sb row is done.
+  int mi_rows_parse_done;
+  // Indicates the progress of the decoding of superblocks.
+  // Initialized to 0. Incremented by sb_mi_size when decode sb row is started.
+  int mi_rows_decode_started;
+  // Boolean: Initialized to 0 (false). Set to 1 (true) on error to abort
+  // decoding.
   int row_mt_exit;
 } AV1DecRowMTInfo;
 
@@ -146,13 +161,6 @@ typedef struct AV1Decoder {
 
   DECLARE_ALIGNED(32, AV1_COMMON, common);
 
-  int refresh_frame_flags;
-
-  // TODO(hkuang): Combine this with cur_buf in macroblockd as they are
-  // the same.
-  RefCntBuffer *cur_buf;  //  Current decoding frame buffer.
-
-  AVxWorker *frame_worker_owner;  // frame_worker that owns this pbi.
   AVxWorker lf_worker;
   AV1LfSync lf_row_sync;
   AV1LrSync lr_row_sync;
@@ -178,8 +186,7 @@ typedef struct AV1Decoder {
   // Note: The saved buffers are released at the start of the next time the
   // application calls aom_codec_decode().
   int output_all_layers;
-  YV12_BUFFER_CONFIG *output_frames[MAX_NUM_SPATIAL_LAYERS];
-  size_t output_frame_index[MAX_NUM_SPATIAL_LAYERS];  // Buffer pool indices
+  RefCntBuffer *output_frames[MAX_NUM_SPATIAL_LAYERS];
   size_t num_output_frames;  // How many frames are queued up so far?
 
   // In order to properly support random-access decoding, we need
@@ -191,7 +198,9 @@ typedef struct AV1Decoder {
   int max_threads;
   int inv_tile_order;
   int need_resync;   // wait for key/intra-only frame.
-  int hold_ref_buf;  // hold the reference buffer.
+  int hold_ref_buf;  // Boolean: whether we are holding reference buffers in
+                     // common.next_ref_frame_map.
+  int reset_decoder_state;
 
   int tile_size_bytes;
   int tile_col_size_bytes;
@@ -225,9 +234,7 @@ typedef struct AV1Decoder {
   unsigned int ext_tile_debug;  // for ext-tile software debug & testing
   unsigned int row_mt;
   EXTERNAL_REFERENCES ext_refs;
-  size_t tile_list_size;
-  uint8_t *tile_list_output;
-  size_t buffer_sz;
+  YV12_BUFFER_CONFIG tile_list_outbuf;
 
   CB_BUFFER *cb_buffer_base;
   int cb_buffer_alloc_size;
@@ -245,7 +252,7 @@ typedef struct AV1Decoder {
 // Returns 0 on success. Sets pbi->common.error.error_code to a nonzero error
 // code and returns a nonzero value on failure.
 int av1_receive_compressed_data(struct AV1Decoder *pbi, size_t size,
-                                const uint8_t **dest);
+                                const uint8_t **psource);
 
 // Get the frame at a particular index in the output queue
 int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd,
@@ -266,23 +273,28 @@ aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm,
 struct AV1Decoder *av1_decoder_create(BufferPool *const pool);
 
 void av1_decoder_remove(struct AV1Decoder *pbi);
-void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_jobs_sync);
+void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_mt_info);
 
 void av1_dec_row_mt_dealloc(AV1DecRowMTSync *dec_row_mt_sync);
 
 void av1_dec_free_cb_buf(AV1Decoder *pbi);
 
-static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs,
+static INLINE void decrease_ref_count(RefCntBuffer *const buf,
                                       BufferPool *const pool) {
-  if (idx >= 0) {
-    --frame_bufs[idx].ref_count;
+  if (buf != NULL) {
+    --buf->ref_count;
+    // Reference counts should never become negative. If this assertion fails,
+    // there is a bug in our reference count management.
+    assert(buf->ref_count >= 0);
     // A worker may only get a free framebuffer index when calling get_free_fb.
-    // But the private buffer is not set up until finish decoding header.
-    // So any error happens during decoding header, the frame_bufs will not
-    // have valid priv buffer.
-    if (frame_bufs[idx].ref_count == 0 &&
-        frame_bufs[idx].raw_frame_buffer.priv) {
-      pool->release_fb_cb(pool->cb_priv, &frame_bufs[idx].raw_frame_buffer);
+    // But the raw frame buffer is not set up until we finish decoding header.
+    // So if any error happens during decoding header, frame_bufs[idx] will not
+    // have a valid raw frame buffer.
+    if (buf->ref_count == 0 && buf->raw_frame_buffer.data) {
+      pool->release_fb_cb(pool->cb_priv, &buf->raw_frame_buffer);
+      buf->raw_frame_buffer.data = NULL;
+      buf->raw_frame_buffer.size = 0;
+      buf->raw_frame_buffer.priv = NULL;
     }
   }
 }
diff --git a/third_party/aom/av1/decoder/decodetxb.c b/third_party/aom/av1/decoder/decodetxb.c
index f3ef2d55e4..223e32e990 100644
--- a/third_party/aom/av1/decoder/decodetxb.c
+++ b/third_party/aom/av1/decoder/decodetxb.c
@@ -136,6 +136,15 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
   uint16_t *const max_scan_line = &(eob_data->max_scan_line);
   *max_scan_line = 0;
   *eob = 0;
+
+#if CONFIG_INSPECTION
+  if (plane == 0) {
+    const int txk_type_idx =
+        av1_get_txk_type_index(mbmi->sb_type, blk_row, blk_col);
+    mbmi->tx_skip[txk_type_idx] = all_zero;
+  }
+#endif
+
   if (all_zero) {
     *max_scan_line = 0;
     if (plane == 0) {
@@ -146,9 +155,6 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
     return 0;
   }
 
-  memset(levels_buf, 0,
-         sizeof(*levels_buf) *
-             ((width + TX_PAD_HOR) * (height + TX_PAD_VER) + TX_PAD_END));
   if (plane == AOM_PLANE_Y) {
     // only y plane's tx_type is transmitted
     av1_read_tx_type(cm, xd, blk_row, blk_col, tx_size, r);
@@ -214,23 +220,30 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
       break;
   }
 
-  if (k_eob_offset_bits[eob_pt] > 0) {
+  const int eob_offset_bits = k_eob_offset_bits[eob_pt];
+  if (eob_offset_bits > 0) {
     const int eob_ctx = eob_pt - 3;
     int bit = aom_read_symbol(
         r, ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_ctx], 2, ACCT_STR);
     if (bit) {
-      eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1));
+      eob_extra += (1 << (eob_offset_bits - 1));
     }
 
-    for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) {
+    for (int i = 1; i < eob_offset_bits; i++) {
       bit = aom_read_bit(r, ACCT_STR);
       if (bit) {
-        eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1 - i));
+        eob_extra += (1 << (eob_offset_bits - 1 - i));
       }
     }
   }
   *eob = rec_eob_pos(eob_pt, eob_extra);
 
+  if (*eob > 1) {
+    memset(levels_buf, 0,
+           sizeof(*levels_buf) *
+               ((width + TX_PAD_HOR) * (height + TX_PAD_VER) + TX_PAD_END));
+  }
+
   {
     // Read the non-zero coefficient with scan index eob-1
     // TODO(angiebird): Put this into a function
@@ -242,12 +255,10 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
         ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx];
     int level = aom_read_symbol(r, cdf, nsymbs, ACCT_STR) + 1;
     if (level > NUM_BASE_LEVELS) {
-      const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
+      const int br_ctx = get_br_ctx_eob(pos, bwl, tx_class);
+      cdf = ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx];
       for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
-        const int k = aom_read_symbol(
-            r,
-            ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx],
-            BR_CDF_SIZE, ACCT_STR);
+        const int k = aom_read_symbol(r, cdf, BR_CDF_SIZE, ACCT_STR);
         level += k;
         if (k < BR_CDF_SIZE - 1) break;
       }
@@ -269,13 +280,6 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
     }
   }
 
-  int16_t num_zero_coeffs = 0;
-  for (int c = 0; c < *eob; ++c) {
-    const int pos = scan[c];
-    num_zero_coeffs = AOMMAX(num_zero_coeffs, pos);
-  }
-  memset(tcoeffs, 0, (num_zero_coeffs + 1) * sizeof(tcoeffs[0]));
-
   for (int c = 0; c < *eob; ++c) {
     const int pos = scan[c];
     uint8_t sign;
diff --git a/third_party/aom/av1/decoder/dthread.h b/third_party/aom/av1/decoder/dthread.h
index 1d264b07eb..c1b8719af8 100644
--- a/third_party/aom/av1/decoder/dthread.h
+++ b/third_party/aom/av1/decoder/dthread.h
@@ -41,40 +41,10 @@ typedef struct FrameWorkerData {
   void *user_priv;
   int worker_id;
   int received_frame;
-
-  // scratch_buffer is used in frame parallel mode only.
-  // It is used to make a copy of the compressed data.
-  uint8_t *scratch_buffer;
-  size_t scratch_buffer_size;
-
-#if CONFIG_MULTITHREAD
-  pthread_mutex_t stats_mutex;
-  pthread_cond_t stats_cond;
-#endif
-
   int frame_context_ready;  // Current frame's context is ready to read.
   int frame_decoded;        // Finished decoding current frame.
 } FrameWorkerData;
 
-void av1_frameworker_lock_stats(AVxWorker *const worker);
-void av1_frameworker_unlock_stats(AVxWorker *const worker);
-void av1_frameworker_signal_stats(AVxWorker *const worker);
-
-// Wait until ref_buf has been decoded to row in real pixel unit.
-// Note: worker may already finish decoding ref_buf and release it in order to
-// start decoding next frame. So need to check whether worker is still decoding
-// ref_buf.
-void av1_frameworker_wait(AVxWorker *const worker, RefCntBuffer *const ref_buf,
-                          int row);
-
-// FrameWorker broadcasts its decoding progress so other workers that are
-// waiting on it can resume decoding.
-void av1_frameworker_broadcast(RefCntBuffer *const buf, int row);
-
-// Copy necessary decoding context from src worker to dst worker.
-void av1_frameworker_copy_context(AVxWorker *const dst_worker,
-                                  AVxWorker *const src_worker);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/decoder/inspection.c b/third_party/aom/av1/decoder/inspection.c
index e6c89298a4..a027b64e63 100644
--- a/third_party/aom/av1/decoder/inspection.c
+++ b/third_party/aom/av1/decoder/inspection.c
@@ -33,15 +33,18 @@ void ifd_clear(insp_frame_data *fd) {
 
 /* TODO(negge) This function may be called by more than one thread when using
                a multi-threaded decoder and this may cause a data race. */
-int ifd_inspect(insp_frame_data *fd, void *decoder) {
+int ifd_inspect(insp_frame_data *fd, void *decoder, int skip_not_transform) {
   struct AV1Decoder *pbi = (struct AV1Decoder *)decoder;
   AV1_COMMON *const cm = &pbi->common;
+
   if (fd->mi_rows != cm->mi_rows || fd->mi_cols != cm->mi_cols) {
     ifd_clear(fd);
     ifd_init_mi_rc(fd, cm->mi_rows, cm->mi_cols);
   }
+  fd->show_existing_frame = cm->show_existing_frame;
+  fd->frame_number = cm->current_frame.frame_number;
   fd->show_frame = cm->show_frame;
-  fd->frame_type = cm->frame_type;
+  fd->frame_type = cm->current_frame.frame_type;
   fd->base_qindex = cm->base_qindex;
   // Set width and height of the first tile until generic support can be added
   TileInfo tile_info;
@@ -49,8 +52,8 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
   av1_tile_set_col(&tile_info, cm, 0);
   fd->tile_mi_cols = tile_info.mi_col_end - tile_info.mi_col_start;
   fd->tile_mi_rows = tile_info.mi_row_end - tile_info.mi_row_start;
-  fd->delta_q_present_flag = cm->delta_q_present_flag;
-  fd->delta_q_res = cm->delta_q_res;
+  fd->delta_q_present_flag = cm->delta_q_info.delta_q_present_flag;
+  fd->delta_q_res = cm->delta_q_info.delta_q_res;
 #if CONFIG_ACCOUNTING
   fd->accounting = &pbi->accounting;
 #endif
@@ -85,6 +88,10 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
       } else {
         mi->uv_mode = UV_MODE_INVALID;
       }
+
+      mi->motion_mode = mbmi->motion_mode;
+      mi->compound_type = mbmi->interinter_comp.type;
+
       // Block Size
       mi->sb_type = mbmi->sb_type;
       // Skip Flag
@@ -92,15 +99,31 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
       mi->filter[0] = av1_extract_interp_filter(mbmi->interp_filters, 0);
       mi->filter[1] = av1_extract_interp_filter(mbmi->interp_filters, 1);
       mi->dual_filter_type = mi->filter[0] * 3 + mi->filter[1];
+
       // Transform
       // TODO(anyone): extract tx type info from mbmi->txk_type[].
-      mi->tx_type = DCT_DCT;
-      mi->tx_size = mbmi->tx_size;
 
-      mi->cdef_level =
-          cm->cdef_strengths[mbmi->cdef_strength] / CDEF_SEC_STRENGTHS;
-      mi->cdef_strength =
-          cm->cdef_strengths[mbmi->cdef_strength] % CDEF_SEC_STRENGTHS;
+      const BLOCK_SIZE bsize = mbmi->sb_type;
+      const int c = i % mi_size_wide[bsize];
+      const int r = j % mi_size_high[bsize];
+      if (is_inter_block(mbmi) || is_intrabc_block(mbmi))
+        mi->tx_size = mbmi->inter_tx_size[av1_get_txb_size_index(bsize, r, c)];
+      else
+        mi->tx_size = mbmi->tx_size;
+
+      if (skip_not_transform && mi->skip) mi->tx_size = -1;
+
+      mi->tx_type =
+          (mi->skip ? 0 : mbmi->txk_type[av1_get_txk_type_index(bsize, r, c)]);
+      if (skip_not_transform &&
+          (mi->skip || mbmi->tx_skip[av1_get_txk_type_index(bsize, r, c)]))
+        mi->tx_type = -1;
+
+      mi->cdef_level = cm->cdef_info.cdef_strengths[mbmi->cdef_strength] /
+                       CDEF_SEC_STRENGTHS;
+      mi->cdef_strength = cm->cdef_info.cdef_strengths[mbmi->cdef_strength] %
+                          CDEF_SEC_STRENGTHS;
+
       mi->cdef_strength += mi->cdef_strength == 3;
       if (mbmi->uv_mode == UV_CFL_PRED) {
         mi->cfl_alpha_idx = mbmi->cfl_alpha_idx;
diff --git a/third_party/aom/av1/decoder/inspection.h b/third_party/aom/av1/decoder/inspection.h
index 7214a9beda..ddea4a1543 100644
--- a/third_party/aom/av1/decoder/inspection.h
+++ b/third_party/aom/av1/decoder/inspection.h
@@ -20,7 +20,7 @@ extern "C" {
 #include "av1/decoder/accounting.h"
 #endif
 
-#ifndef AOM_AOMDX_H_
+#ifndef AOM_AOM_AOMDX_H_
 typedef void (*aom_inspect_cb)(void *decoder, void *data);
 #endif
 
@@ -50,6 +50,8 @@ struct insp_mi_data {
   int16_t cfl_alpha_idx;
   int16_t cfl_alpha_sign;
   int16_t current_qindex;
+  int16_t compound_type;
+  int16_t motion_mode;
 };
 
 typedef struct insp_frame_data insp_frame_data;
@@ -59,6 +61,7 @@ struct insp_frame_data {
   Accounting *accounting;
 #endif
   insp_mi_data *mi_grid;
+  int16_t frame_number;
   int show_frame;
   int frame_type;
   int base_qindex;
@@ -72,11 +75,12 @@ struct insp_frame_data {
   // TODO(negge): add per frame CDEF data
   int delta_q_present_flag;
   int delta_q_res;
+  int show_existing_frame;
 };
 
 void ifd_init(insp_frame_data *fd, int frame_width, int frame_height);
 void ifd_clear(insp_frame_data *fd);
-int ifd_inspect(insp_frame_data *fd, void *decoder);
+int ifd_inspect(insp_frame_data *fd, void *decoder, int skip_not_transform);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/decoder/obu.c b/third_party/aom/av1/decoder/obu.c
index 44ecf818e7..58f7d2ca8e 100644
--- a/third_party/aom/av1/decoder/obu.c
+++ b/third_party/aom/av1/decoder/obu.c
@@ -12,6 +12,7 @@
 #include <assert.h>
 
 #include "config/aom_config.h"
+#include "config/aom_scale_rtcd.h"
 
 #include "aom/aom_codec.h"
 #include "aom_dsp/bitreader_buffer.h"
@@ -25,7 +26,7 @@
 #include "av1/decoder/obu.h"
 
 // Picture prediction structures (0-12 are predefined) in scalability metadata.
-typedef enum {
+enum {
   SCALABILITY_L1T2 = 0,
   SCALABILITY_L1T3 = 1,
   SCALABILITY_L2T1 = 2,
@@ -41,7 +42,7 @@ typedef enum {
   SCALABILITY_S2T2h = 12,
   SCALABILITY_S2T3h = 13,
   SCALABILITY_SS = 14
-} SCALABILITY_STRUCTURES;
+} UENUM1BYTE(SCALABILITY_STRUCTURES);
 
 aom_codec_err_t aom_get_num_layers_from_operating_point_idc(
     int operating_point_idc, unsigned int *number_spatial_layers,
@@ -292,6 +293,8 @@ static uint32_t read_frame_header_obu(AV1Decoder *pbi,
                                             trailing_bits_present);
 }
 
+// On success, returns the tile group header size. On failure, calls
+// aom_internal_error() and returns -1.
 static int32_t read_tile_group_header(AV1Decoder *pbi,
                                       struct aom_read_bit_buffer *rb,
                                       int *start_tile, int *end_tile,
@@ -303,26 +306,33 @@ static int32_t read_tile_group_header(AV1Decoder *pbi,
 
   if (!pbi->common.large_scale_tile && num_tiles > 1) {
     tile_start_and_end_present_flag = aom_rb_read_bit(rb);
+    if (tile_start_implicit && tile_start_and_end_present_flag) {
+      aom_internal_error(
+          &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+          "For OBU_FRAME type obu tile_start_and_end_present_flag must be 0");
+      return -1;
+    }
   }
   if (pbi->common.large_scale_tile || num_tiles == 1 ||
       !tile_start_and_end_present_flag) {
     *start_tile = 0;
     *end_tile = num_tiles - 1;
-    return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
+  } else {
+    int tile_bits = cm->log2_tile_rows + cm->log2_tile_cols;
+    *start_tile = aom_rb_read_literal(rb, tile_bits);
+    *end_tile = aom_rb_read_literal(rb, tile_bits);
   }
-  if (tile_start_implicit && tile_start_and_end_present_flag) {
-    aom_internal_error(
-        &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-        "For OBU_FRAME type obu tile_start_and_end_present_flag must be 0");
+  if (*start_tile > *end_tile) {
+    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
+                       "tg_end must be greater than or equal to tg_start");
     return -1;
   }
-  *start_tile =
-      aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
-  *end_tile = aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols);
 
   return ((rb->bit_offset - saved_bit_offset + 7) >> 3);
 }
 
+// On success, returns the tile group OBU size. On failure, sets
+// pbi->common.error.error_code and returns 0.
 static uint32_t read_one_tile_group_obu(
     AV1Decoder *pbi, struct aom_read_bit_buffer *rb, int is_first_tg,
     const uint8_t *data, const uint8_t *data_end, const uint8_t **p_data_end,
@@ -337,7 +347,6 @@ static uint32_t read_one_tile_group_obu(
   header_size = read_tile_group_header(pbi, rb, &start_tile, &end_tile,
                                        tile_start_implicit);
   if (header_size == -1 || byte_alignment(cm, rb)) return 0;
-  if (start_tile > end_tile) return header_size;
   data += header_size;
   av1_decode_tg_tiles_and_wrapup(pbi, data, data_end, p_data_end, start_tile,
                                  end_tile, is_first_tg);
@@ -350,87 +359,121 @@ static uint32_t read_one_tile_group_obu(
 }
 
 static void alloc_tile_list_buffer(AV1Decoder *pbi) {
-  // TODO(yunqing): for now, copy each tile's decoded YUV data directly to the
-  // output buffer. This needs to be modified according to the application
-  // requirement.
+  // The resolution of the output frame is read out from the bitstream. The data
+  // are stored in the order of Y plane, U plane and V plane. As an example, for
+  // image format 4:2:0, the output frame of U plane and V plane is 1/4 of the
+  // output frame.
   AV1_COMMON *const cm = &pbi->common;
-  const int tile_width_in_pixels = cm->tile_width * MI_SIZE;
-  const int tile_height_in_pixels = cm->tile_height * MI_SIZE;
-  const int ssy = cm->seq_params.subsampling_y;
-  const int ssx = cm->seq_params.subsampling_x;
-  const int num_planes = av1_num_planes(cm);
-  const size_t yplane_tile_size = tile_height_in_pixels * tile_width_in_pixels;
-  const size_t uvplane_tile_size =
-      (num_planes > 1)
-          ? (tile_height_in_pixels >> ssy) * (tile_width_in_pixels >> ssx)
-          : 0;
-  const size_t tile_size = (cm->seq_params.use_highbitdepth ? 2 : 1) *
-                           (yplane_tile_size + 2 * uvplane_tile_size);
-  pbi->tile_list_size = tile_size * (pbi->tile_count_minus_1 + 1);
+  int tile_width, tile_height;
+  av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
+  const int tile_width_in_pixels = tile_width * MI_SIZE;
+  const int tile_height_in_pixels = tile_height * MI_SIZE;
+  const int output_frame_width =
+      (pbi->output_frame_width_in_tiles_minus_1 + 1) * tile_width_in_pixels;
+  const int output_frame_height =
+      (pbi->output_frame_height_in_tiles_minus_1 + 1) * tile_height_in_pixels;
+  // The output frame is used to store the decoded tile list. The decoded tile
+  // list has to fit into 1 output frame.
+  assert((pbi->tile_count_minus_1 + 1) <=
+         (pbi->output_frame_width_in_tiles_minus_1 + 1) *
+             (pbi->output_frame_height_in_tiles_minus_1 + 1));
 
-  if (pbi->tile_list_size > pbi->buffer_sz) {
-    if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output);
-    pbi->tile_list_output = NULL;
+  // Allocate the tile list output buffer.
+  // Note: if cm->seq_params.use_highbitdepth is 1 and cm->seq_params.bit_depth
+  // is 8, we could allocate less memory, namely, 8 bits/pixel.
+  if (aom_alloc_frame_buffer(&pbi->tile_list_outbuf, output_frame_width,
+                             output_frame_height, cm->seq_params.subsampling_x,
+                             cm->seq_params.subsampling_y,
+                             (cm->seq_params.use_highbitdepth &&
+                              (cm->seq_params.bit_depth > AOM_BITS_8)),
+                             0, cm->byte_alignment))
+    aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                       "Failed to allocate the tile list output buffer");
+}
 
-    pbi->tile_list_output = (uint8_t *)aom_memalign(32, pbi->tile_list_size);
-    if (pbi->tile_list_output == NULL)
-      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
-                         "Failed to allocate the tile list output buffer");
-    pbi->buffer_sz = pbi->tile_list_size;
+static void yv12_tile_copy(const YV12_BUFFER_CONFIG *src, int hstart1,
+                           int hend1, int vstart1, int vend1,
+                           YV12_BUFFER_CONFIG *dst, int hstart2, int vstart2,
+                           int plane) {
+  const int src_stride = (plane > 0) ? src->strides[1] : src->strides[0];
+  const int dst_stride = (plane > 0) ? dst->strides[1] : dst->strides[0];
+  int row, col;
+
+  assert(src->flags & YV12_FLAG_HIGHBITDEPTH);
+  assert(!(dst->flags & YV12_FLAG_HIGHBITDEPTH));
+
+  const uint16_t *src16 =
+      CONVERT_TO_SHORTPTR(src->buffers[plane] + vstart1 * src_stride + hstart1);
+  uint8_t *dst8 = dst->buffers[plane] + vstart2 * dst_stride + hstart2;
+
+  for (row = vstart1; row < vend1; ++row) {
+    for (col = 0; col < (hend1 - hstart1); ++col) *dst8++ = (uint8_t)(*src16++);
+    src16 += src_stride - (hend1 - hstart1);
+    dst8 += dst_stride - (hend1 - hstart1);
   }
+  return;
 }
 
 static void copy_decoded_tile_to_tile_list_buffer(AV1Decoder *pbi,
-                                                  uint8_t **output) {
+                                                  int tile_idx) {
   AV1_COMMON *const cm = &pbi->common;
-  const int tile_width_in_pixels = cm->tile_width * MI_SIZE;
-  const int tile_height_in_pixels = cm->tile_height * MI_SIZE;
+  int tile_width, tile_height;
+  av1_get_uniform_tile_size(cm, &tile_width, &tile_height);
+  const int tile_width_in_pixels = tile_width * MI_SIZE;
+  const int tile_height_in_pixels = tile_height * MI_SIZE;
   const int ssy = cm->seq_params.subsampling_y;
   const int ssx = cm->seq_params.subsampling_x;
   const int num_planes = av1_num_planes(cm);
 
-  // Copy decoded tile to the tile list output buffer.
-  YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm);
-  const int mi_row = pbi->dec_tile_row * cm->tile_height;
-  const int mi_col = pbi->dec_tile_col * cm->tile_width;
-  const int is_hbd = (cur_frame->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
-  uint8_t *bufs[MAX_MB_PLANE] = { NULL, NULL, NULL };
-  int strides[MAX_MB_PLANE] = { 0, 0, 0 };
+  YV12_BUFFER_CONFIG *cur_frame = &cm->cur_frame->buf;
+  const int tr = tile_idx / (pbi->output_frame_width_in_tiles_minus_1 + 1);
+  const int tc = tile_idx % (pbi->output_frame_width_in_tiles_minus_1 + 1);
   int plane;
 
+  // Copy decoded tile to the tile list output buffer.
   for (plane = 0; plane < num_planes; ++plane) {
-    int shift_x = plane > 0 ? ssx : 0;
-    int shift_y = plane > 0 ? ssy : 0;
+    const int shift_x = plane > 0 ? ssx : 0;
+    const int shift_y = plane > 0 ? ssy : 0;
+    const int h = tile_height_in_pixels >> shift_y;
+    const int w = tile_width_in_pixels >> shift_x;
 
-    bufs[plane] = cur_frame->buffers[plane];
-    strides[plane] =
-        (plane > 0) ? cur_frame->strides[1] : cur_frame->strides[0];
+    // src offset
+    int vstart1 = pbi->dec_tile_row * h;
+    int vend1 = vstart1 + h;
+    int hstart1 = pbi->dec_tile_col * w;
+    int hend1 = hstart1 + w;
+    // dst offset
+    int vstart2 = tr * h;
+    int hstart2 = tc * w;
 
-    bufs[plane] += mi_row * (MI_SIZE >> shift_y) * strides[plane] +
-                   mi_col * (MI_SIZE >> shift_x);
-
-    if (is_hbd) {
-      bufs[plane] = (uint8_t *)CONVERT_TO_SHORTPTR(bufs[plane]);
-      strides[plane] *= 2;
-    }
-
-    int w, h;
-    w = (plane > 0 && shift_x > 0) ? ((tile_width_in_pixels + 1) >> shift_x)
-                                   : tile_width_in_pixels;
-    w *= (1 + is_hbd);
-    h = (plane > 0 && shift_y > 0) ? ((tile_height_in_pixels + 1) >> shift_y)
-                                   : tile_height_in_pixels;
-    int j;
-
-    for (j = 0; j < h; ++j) {
-      memcpy(*output, bufs[plane], w);
-      bufs[plane] += strides[plane];
-      *output += w;
+    if (cm->seq_params.use_highbitdepth &&
+        cm->seq_params.bit_depth == AOM_BITS_8) {
+      yv12_tile_copy(cur_frame, hstart1, hend1, vstart1, vend1,
+                     &pbi->tile_list_outbuf, hstart2, vstart2, plane);
+    } else {
+      switch (plane) {
+        case 0:
+          aom_yv12_partial_copy_y(cur_frame, hstart1, hend1, vstart1, vend1,
+                                  &pbi->tile_list_outbuf, hstart2, vstart2);
+          break;
+        case 1:
+          aom_yv12_partial_copy_u(cur_frame, hstart1, hend1, vstart1, vend1,
+                                  &pbi->tile_list_outbuf, hstart2, vstart2);
+          break;
+        case 2:
+          aom_yv12_partial_copy_v(cur_frame, hstart1, hend1, vstart1, vend1,
+                                  &pbi->tile_list_outbuf, hstart2, vstart2);
+          break;
+        default: assert(0);
+      }
     }
   }
 }
 
 // Only called while large_scale_tile = 1.
+//
+// On success, returns the tile list OBU size. On failure, sets
+// pbi->common.error.error_code and returns 0.
 static uint32_t read_and_decode_one_tile_list(AV1Decoder *pbi,
                                               struct aom_read_bit_buffer *rb,
                                               const uint8_t *data,
@@ -459,8 +502,8 @@ static uint32_t read_and_decode_one_tile_list(AV1Decoder *pbi,
   uint32_t tile_list_info_bytes = 4;
   tile_list_payload_size += tile_list_info_bytes;
   data += tile_list_info_bytes;
-  uint8_t *output = pbi->tile_list_output;
 
+  int tile_idx = 0;
   for (i = 0; i <= pbi->tile_count_minus_1; i++) {
     // Process 1 tile.
     // Reset the bit reader.
@@ -504,7 +547,8 @@ static uint32_t read_and_decode_one_tile_list(AV1Decoder *pbi,
     assert(data <= data_end);
 
     // Copy the decoded tile to the tile list output buffer.
-    copy_decoded_tile_to_tile_list_buffer(pbi, &output);
+    copy_decoded_tile_to_tile_list_buffer(pbi, tile_idx);
+    tile_idx++;
   }
 
   *frame_decoding_finished = 1;
@@ -613,6 +657,7 @@ static void read_metadata_timecode(const uint8_t *data, size_t sz) {
   }
 }
 
+// Not fully implemented. Always succeeds and returns sz.
 static size_t read_metadata(const uint8_t *data, size_t sz) {
   size_t type_length;
   uint64_t type_value;
@@ -659,7 +704,7 @@ int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
   if (!cm->large_scale_tile) pbi->camera_frame_header_ready = 0;
 
   // decode frame as a series of OBUs
-  while (!frame_decoding_finished && !cm->error.error_code) {
+  while (!frame_decoding_finished && cm->error.error_code == AOM_CODEC_OK) {
     struct aom_read_bit_buffer rb;
     size_t payload_size = 0;
     size_t decoded_payload_size = 0;
@@ -769,6 +814,7 @@ int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
         if (obu_header.type != OBU_FRAME) break;
         obu_payload_offset = frame_header_size;
         // Byte align the reader before reading the tile group.
+        // byte_alignment() has set cm->error.error_code if it returns -1.
         if (byte_alignment(cm, &rb)) return -1;
         AOM_FALLTHROUGH_INTENDED;  // fall through to read tile group.
       case OBU_TILE_GROUP:
@@ -784,6 +830,7 @@ int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
             pbi, &rb, is_first_tg_obu_received, data + obu_payload_offset,
             data + payload_size, p_data_end, &frame_decoding_finished,
             obu_header.type == OBU_FRAME);
+        if (cm->error.error_code != AOM_CODEC_OK) return -1;
         is_first_tg_obu_received = 0;
         if (frame_decoding_finished) pbi->seen_frame_header = 0;
         break;
@@ -835,5 +882,6 @@ int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
     data += payload_size;
   }
 
+  if (cm->error.error_code != AOM_CODEC_OK) return -1;
   return frame_decoding_finished;
 }
diff --git a/third_party/aom/av1/decoder/obu.h b/third_party/aom/av1/decoder/obu.h
index 5ab243fc90..d8ebe368e6 100644
--- a/third_party/aom/av1/decoder/obu.h
+++ b/third_party/aom/av1/decoder/obu.h
@@ -25,7 +25,7 @@ int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data,
                                const uint8_t **p_data_end);
 
 aom_codec_err_t aom_get_num_layers_from_operating_point_idc(
-    int operating_point_idc, unsigned int *num_spatial_layers,
-    unsigned int *num_temporal_layers);
+    int operating_point_idc, unsigned int *number_spatial_layers,
+    unsigned int *number_temporal_layers);
 
 #endif  // AOM_AV1_DECODER_OBU_H_
diff --git a/third_party/aom/av1/encoder/aq_complexity.c b/third_party/aom/av1/encoder/aq_complexity.c
index 80f8e2e66d..16edbc69a2 100644
--- a/third_party/aom/av1/encoder/aq_complexity.c
+++ b/third_party/aom/av1/encoder/aq_complexity.c
@@ -93,7 +93,7 @@ void av1_setup_in_frame_q_adj(AV1_COMP *cpi) {
       if (segment == DEFAULT_AQ2_SEG) continue;
 
       qindex_delta = av1_compute_qdelta_by_rate(
-          &cpi->rc, cm->frame_type, cm->base_qindex,
+          &cpi->rc, cm->current_frame.frame_type, cm->base_qindex,
           aq_c_q_adj_factor[aq_strength][segment], cm->seq_params.bit_depth);
 
       // For AQ complexity mode, we dont allow Q0 in a segment if the base
@@ -148,7 +148,7 @@ void av1_caq_select_segment(const AV1_COMP *cpi, MACROBLOCK *mb, BLOCK_SIZE bs,
             ? AOMMAX(exp(cpi->twopass.mb_av_energy), MIN_DEFAULT_LV_THRESH)
             : DEFAULT_LV_THRESH;
 
-    av1_setup_src_planes(mb, cpi->source, mi_row, mi_col, num_planes);
+    av1_setup_src_planes(mb, cpi->source, mi_row, mi_col, num_planes, bs);
     logvar = av1_log_block_var(cpi, mb, bs);
 
     segment = AQ_C_SEGMENTS - 1;  // Just in case no break out below.
diff --git a/third_party/aom/av1/encoder/aq_cyclicrefresh.c b/third_party/aom/av1/encoder/aq_cyclicrefresh.c
index f532d48da5..8d96b23920 100644
--- a/third_party/aom/av1/encoder/aq_cyclicrefresh.c
+++ b/third_party/aom/av1/encoder/aq_cyclicrefresh.c
@@ -140,8 +140,8 @@ static int compute_deltaq(const AV1_COMP *cpi, int q, double rate_factor) {
   const CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
   const RATE_CONTROL *const rc = &cpi->rc;
   int deltaq =
-      av1_compute_qdelta_by_rate(rc, cpi->common.frame_type, q, rate_factor,
-                                 cpi->common.seq_params.bit_depth);
+      av1_compute_qdelta_by_rate(rc, cpi->common.current_frame.frame_type, q,
+                                 rate_factor, cpi->common.seq_params.bit_depth);
   if ((-deltaq) > cr->max_qdelta_perc * q / 100) {
     deltaq = -cr->max_qdelta_perc * q / 100;
   }
@@ -166,15 +166,15 @@ int av1_cyclic_refresh_estimate_bits_at_q(const AV1_COMP *cpi,
   // Take segment weighted average for estimated bits.
   estimated_bits =
       (int)((1.0 - weight_segment1 - weight_segment2) *
-                av1_estimate_bits_at_q(cm->frame_type, cm->base_qindex, mbs,
-                                       correction_factor,
+                av1_estimate_bits_at_q(cm->current_frame.frame_type,
+                                       cm->base_qindex, mbs, correction_factor,
                                        cm->seq_params.bit_depth) +
             weight_segment1 * av1_estimate_bits_at_q(
-                                  cm->frame_type,
+                                  cm->current_frame.frame_type,
                                   cm->base_qindex + cr->qindex_delta[1], mbs,
                                   correction_factor, cm->seq_params.bit_depth) +
             weight_segment2 * av1_estimate_bits_at_q(
-                                  cm->frame_type,
+                                  cm->current_frame.frame_type,
                                   cm->base_qindex + cr->qindex_delta[2], mbs,
                                   correction_factor, cm->seq_params.bit_depth));
   return estimated_bits;
@@ -203,10 +203,11 @@ int av1_cyclic_refresh_rc_bits_per_mb(const AV1_COMP *cpi, int i,
   // Take segment weighted average for bits per mb.
   bits_per_mb =
       (int)((1.0 - weight_segment) *
-                av1_rc_bits_per_mb(cm->frame_type, i, correction_factor,
+                av1_rc_bits_per_mb(cm->current_frame.frame_type, i,
+                                   correction_factor,
                                    cm->seq_params.bit_depth) +
-            weight_segment * av1_rc_bits_per_mb(cm->frame_type, i + deltaq,
-                                                correction_factor,
+            weight_segment * av1_rc_bits_per_mb(cm->current_frame.frame_type,
+                                                i + deltaq, correction_factor,
                                                 cm->seq_params.bit_depth));
   return bits_per_mb;
 }
@@ -496,14 +497,14 @@ void av1_cyclic_refresh_setup(AV1_COMP *const cpi) {
     av1_disable_segmentation(seg);
     return;
   }
-  if (cm->current_video_frame == 0) cr->low_content_avg = 0.0;
+  if (cm->current_frame.frame_number == 0) cr->low_content_avg = 0.0;
   // Don't apply refresh on key frame or enhancement layer frames.
-  if (!apply_cyclic_refresh || cm->frame_type == KEY_FRAME) {
+  if (!apply_cyclic_refresh || cm->current_frame.frame_type == KEY_FRAME) {
     // Set segmentation map to 0 and disable.
     unsigned char *const seg_map = cpi->segmentation_map;
     memset(seg_map, 0, cm->mi_rows * cm->mi_cols);
     av1_disable_segmentation(&cm->seg);
-    if (cm->frame_type == KEY_FRAME) {
+    if (cm->current_frame.frame_type == KEY_FRAME) {
       memset(cr->last_coded_q_map, MAXQ,
              cm->mi_rows * cm->mi_cols * sizeof(*cr->last_coded_q_map));
       cr->sb_index = 0;
diff --git a/third_party/aom/av1/encoder/aq_variance.c b/third_party/aom/av1/encoder/aq_variance.c
index 58f906bdc0..d5729482fa 100644
--- a/third_party/aom/av1/encoder/aq_variance.c
+++ b/third_party/aom/av1/encoder/aq_variance.c
@@ -77,8 +77,8 @@ void av1_vaq_frame_setup(AV1_COMP *cpi) {
       // Set up avg segment id to be 1.0 and adjust the other segments around
       // it.
       int qindex_delta = av1_compute_qdelta_by_rate(
-          &cpi->rc, cm->frame_type, cm->base_qindex, rate_ratio[i] / avg_ratio,
-          cm->seq_params.bit_depth);
+          &cpi->rc, cm->current_frame.frame_type, cm->base_qindex,
+          rate_ratio[i] / avg_ratio, cm->seq_params.bit_depth);
 
       // We don't allow qindex 0 in a segment if the base value is not 0.
       // Q index 0 (lossless) implies 4x4 encoding only and in AQ mode a segment
@@ -121,7 +121,7 @@ int av1_log_block_var(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bs) {
 
   for (i = 0; i < bh; i += 4) {
     for (j = 0; j < bw; j += 4) {
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (is_cur_buf_hbd(xd)) {
         var +=
             log(1.0 + cpi->fn_ptr[BLOCK_4X4].vf(
                           x->plane[0].src.buf + i * x->plane[0].src.stride + j,
@@ -147,13 +147,13 @@ int av1_log_block_var(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bs) {
 
 #define DEFAULT_E_MIDPOINT 10.0
 
-unsigned int haar_ac_energy(MACROBLOCK *x, BLOCK_SIZE bs) {
+static unsigned int haar_ac_energy(MACROBLOCK *x, BLOCK_SIZE bs) {
   MACROBLOCKD *xd = &x->e_mbd;
   int stride = x->plane[0].src.stride;
   uint8_t *buf = x->plane[0].src.buf;
   const int bw = MI_SIZE * mi_size_wide[bs];
   const int bh = MI_SIZE * mi_size_high[bs];
-  int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+  const int hbd = is_cur_buf_hbd(xd);
 
   int var = 0;
   for (int r = 0; r < bh; r += 8)
@@ -192,8 +192,8 @@ int av1_compute_deltaq_from_energy_level(const AV1_COMP *const cpi,
     rate_level = block_var_level;
   }
   int qindex_delta = av1_compute_qdelta_by_rate(
-      &cpi->rc, cm->frame_type, cm->base_qindex, deltaq_rate_ratio[rate_level],
-      cm->seq_params.bit_depth);
+      &cpi->rc, cm->current_frame.frame_type, cm->base_qindex,
+      deltaq_rate_ratio[rate_level], cm->seq_params.bit_depth);
 
   if ((cm->base_qindex != 0) && ((cm->base_qindex + qindex_delta) == 0)) {
     qindex_delta = -cm->base_qindex + 1;
diff --git a/third_party/aom/av1/encoder/av1_fwd_txfm2d.c b/third_party/aom/av1/encoder/av1_fwd_txfm2d.c
index f25a667cf4..d282b5f4ca 100644
--- a/third_party/aom/av1/encoder/av1_fwd_txfm2d.c
+++ b/third_party/aom/av1/encoder/av1_fwd_txfm2d.c
@@ -364,8 +364,6 @@ static const int8_t fadst4_range_mult2[7] = { 0, 2, 4, 3, 3, 3, 3 };
 static const int8_t fadst8_range_mult2[8] = { 0, 0, 1, 3, 3, 5, 5, 5 };
 static const int8_t fadst16_range_mult2[10] = { 0, 0, 1, 3, 3, 5, 5, 7, 7, 7 };
 
-static const int8_t max_fwd_range_mult2_col[5] = { 3, 5, 7, 9, 11 };
-
 static const int8_t fidtx4_range_mult2[1] = { 1 };
 static const int8_t fidtx8_range_mult2[1] = { 2 };
 static const int8_t fidtx16_range_mult2[1] = { 3 };
@@ -389,14 +387,12 @@ const int8_t *fwd_txfm_range_mult2_list[TXFM_TYPES] = {
 };
 
 static INLINE void set_fwd_txfm_non_scale_range(TXFM_2D_FLIP_CFG *cfg) {
-  const int txh_idx = get_txh_idx(cfg->tx_size);
   av1_zero(cfg->stage_range_col);
   av1_zero(cfg->stage_range_row);
 
+  const int8_t *range_mult2_col = fwd_txfm_range_mult2_list[cfg->txfm_type_col];
   if (cfg->txfm_type_col != TXFM_TYPE_INVALID) {
     int stage_num_col = cfg->stage_num_col;
-    const int8_t *range_mult2_col =
-        fwd_txfm_range_mult2_list[cfg->txfm_type_col];
     for (int i = 0; i < stage_num_col; ++i)
       cfg->stage_range_col[i] = (range_mult2_col[i] + 1) >> 1;
   }
@@ -405,9 +401,11 @@ static INLINE void set_fwd_txfm_non_scale_range(TXFM_2D_FLIP_CFG *cfg) {
     int stage_num_row = cfg->stage_num_row;
     const int8_t *range_mult2_row =
         fwd_txfm_range_mult2_list[cfg->txfm_type_row];
-    for (int i = 0; i < stage_num_row; ++i)
+    for (int i = 0; i < stage_num_row; ++i) {
       cfg->stage_range_row[i] =
-          (max_fwd_range_mult2_col[txh_idx] + range_mult2_row[i] + 1) >> 1;
+          (range_mult2_col[cfg->stage_num_col - 1] + range_mult2_row[i] + 1) >>
+          1;
+    }
   }
 }
 
diff --git a/third_party/aom/av1/encoder/av1_multi_thread.c b/third_party/aom/av1/encoder/av1_multi_thread.c
new file mode 100644
index 0000000000..1260c7af7c
--- /dev/null
+++ b/third_party/aom/av1/encoder/av1_multi_thread.c
@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+
+#include "av1/encoder/encoder.h"
+#include "av1/encoder/ethread.h"
+#include "av1/encoder/av1_multi_thread.h"
+
+void av1_row_mt_mem_alloc(AV1_COMP *cpi, int max_sb_rows) {
+  struct AV1Common *cm = &cpi->common;
+  MultiThreadHandle *multi_thread_ctxt = &cpi->multi_thread_ctxt;
+  int tile_row, tile_col;
+  const int tile_cols = cm->tile_cols;
+  const int tile_rows = cm->tile_rows;
+
+  multi_thread_ctxt->allocated_tile_cols = tile_cols;
+  multi_thread_ctxt->allocated_tile_rows = tile_rows;
+  multi_thread_ctxt->allocated_sb_rows = max_sb_rows;
+
+  // Allocate memory for row based multi-threading
+  for (tile_row = 0; tile_row < multi_thread_ctxt->allocated_tile_rows;
+       tile_row++) {
+    for (tile_col = 0; tile_col < multi_thread_ctxt->allocated_tile_cols;
+         tile_col++) {
+      TileDataEnc *this_tile =
+          &cpi->tile_data[tile_row * multi_thread_ctxt->allocated_tile_cols +
+                          tile_col];
+      av1_row_mt_sync_mem_alloc(&this_tile->row_mt_sync, cm, max_sb_rows);
+      if (cpi->oxcf.cdf_update_mode)
+        CHECK_MEM_ERROR(
+            cm, this_tile->row_ctx,
+            (FRAME_CONTEXT *)aom_memalign(
+                16,
+                AOMMAX(1, (av1_get_sb_cols_in_tile(cm, this_tile->tile_info) -
+                           1)) *
+                    sizeof(*this_tile->row_ctx)));
+    }
+  }
+}
+
+void av1_row_mt_mem_dealloc(AV1_COMP *cpi) {
+  MultiThreadHandle *multi_thread_ctxt = &cpi->multi_thread_ctxt;
+  int tile_col;
+  int tile_row;
+
+  // Free row based multi-threading sync memory
+  for (tile_row = 0; tile_row < multi_thread_ctxt->allocated_tile_rows;
+       tile_row++) {
+    for (tile_col = 0; tile_col < multi_thread_ctxt->allocated_tile_cols;
+         tile_col++) {
+      TileDataEnc *this_tile =
+          &cpi->tile_data[tile_row * multi_thread_ctxt->allocated_tile_cols +
+                          tile_col];
+      av1_row_mt_sync_mem_dealloc(&this_tile->row_mt_sync);
+      if (cpi->oxcf.cdf_update_mode) aom_free(this_tile->row_ctx);
+    }
+  }
+  multi_thread_ctxt->allocated_sb_rows = 0;
+  multi_thread_ctxt->allocated_tile_cols = 0;
+  multi_thread_ctxt->allocated_tile_rows = 0;
+}
diff --git a/third_party/aom/av1/encoder/av1_multi_thread.h b/third_party/aom/av1/encoder/av1_multi_thread.h
new file mode 100644
index 0000000000..2a1cc7d6dd
--- /dev/null
+++ b/third_party/aom/av1/encoder/av1_multi_thread.h
@@ -0,0 +1,21 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AV1_ENCODER_AV1_MULTI_THREAD_H
+#define AV1_ENCODER_AV1_MULTI_THREAD_H
+
+#include "av1/encoder/encoder.h"
+
+void av1_row_mt_mem_alloc(AV1_COMP *cpi, int max_sb_rows);
+
+void av1_row_mt_mem_dealloc(AV1_COMP *cpi);
+
+#endif  // AV1_ENCODER_AV1_MULTI_THREAD_H
diff --git a/third_party/aom/av1/encoder/av1_quantize.c b/third_party/aom/av1/encoder/av1_quantize.c
index a0a9260052..162724b4c3 100644
--- a/third_party/aom/av1/encoder/av1_quantize.c
+++ b/third_party/aom/av1/encoder/av1_quantize.c
@@ -41,47 +41,37 @@ static void quantize_fp_helper_c(
     const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
     const qm_val_t *iqm_ptr, int log_scale) {
   int i, eob = -1;
+  const int rounding[2] = { ROUND_POWER_OF_TWO(round_ptr[0], log_scale),
+                            ROUND_POWER_OF_TWO(round_ptr[1], log_scale) };
   // TODO(jingning) Decide the need of these arguments after the
   // quantization process is completed.
   (void)zbin_ptr;
   (void)quant_shift_ptr;
+  (void)iscan;
 
   memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
   memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
 
   if (qm_ptr == NULL && iqm_ptr == NULL) {
-    const int rounding0 = ROUND_POWER_OF_TWO(round_ptr[0], log_scale);
-    {  // rc == 0
-      const int coeff = coeff_ptr[0];
+    for (i = 0; i < n_coeffs; i++) {
+      const int rc = scan[i];
+      const int32_t thresh = (int32_t)(dequant_ptr[rc != 0]);
+      const int coeff = coeff_ptr[rc];
       const int coeff_sign = (coeff >> 31);
       int64_t abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-      if ((abs_coeff << (1 + log_scale)) >= (int32_t)(dequant_ptr[0])) {
-        abs_coeff = clamp64(abs_coeff + rounding0, INT16_MIN, INT16_MAX);
-        const int tmp32 = (int)((abs_coeff * quant_ptr[0]) >> (16 - log_scale));
+      int tmp32 = 0;
+      if ((abs_coeff << (1 + log_scale)) >= thresh) {
+        abs_coeff =
+            clamp64(abs_coeff + rounding[rc != 0], INT16_MIN, INT16_MAX);
+        tmp32 = (int)((abs_coeff * quant_ptr[rc != 0]) >> (16 - log_scale));
         if (tmp32) {
-          qcoeff_ptr[0] = (tmp32 ^ coeff_sign) - coeff_sign;
-          const tran_low_t abs_dqcoeff = (tmp32 * dequant_ptr[0]) >> log_scale;
-          dqcoeff_ptr[0] = (abs_dqcoeff ^ coeff_sign) - coeff_sign;
-          eob = 0;
-        }
-      }
-    }
-    const int rounding1 = ROUND_POWER_OF_TWO(round_ptr[1], log_scale);
-    const int32_t thresh1 = (int32_t)(dequant_ptr[1]);
-    for (i = 1; i < n_coeffs; i++) {
-      const int coeff = coeff_ptr[i];
-      const int coeff_sign = (coeff >> 31);
-      int64_t abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-      if ((abs_coeff << (1 + log_scale)) >= thresh1) {
-        abs_coeff = clamp64(abs_coeff + rounding1, INT16_MIN, INT16_MAX);
-        const int tmp32 = (int)((abs_coeff * quant_ptr[1]) >> (16 - log_scale));
-        if (tmp32) {
-          qcoeff_ptr[i] = (tmp32 ^ coeff_sign) - coeff_sign;
-          const tran_low_t abs_dqcoeff = (tmp32 * dequant_ptr[1]) >> log_scale;
-          dqcoeff_ptr[i] = (abs_dqcoeff ^ coeff_sign) - coeff_sign;
-          eob = AOMMAX(iscan[i], eob);
+          qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+          const tran_low_t abs_dqcoeff =
+              (tmp32 * dequant_ptr[rc != 0]) >> log_scale;
+          dqcoeff_ptr[rc] = (abs_dqcoeff ^ coeff_sign) - coeff_sign;
         }
       }
+      if (tmp32) eob = i;
     }
   } else {
     // Quantization pass: All coefficients with index >= zero_flag are
@@ -99,7 +89,7 @@ static void quantize_fp_helper_c(
       int tmp32 = 0;
       if (abs_coeff * wt >=
           (dequant_ptr[rc != 0] << (AOM_QM_BITS - (1 + log_scale)))) {
-        abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale);
+        abs_coeff += rounding[rc != 0];
         abs_coeff = clamp64(abs_coeff, INT16_MIN, INT16_MAX);
         tmp32 = (int)((abs_coeff * wt * quant_ptr[rc != 0]) >>
                       (16 - log_scale + AOM_QM_BITS));
@@ -275,32 +265,65 @@ void av1_quantize_b_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                            const SCAN_ORDER *sc, const QUANT_PARAM *qparam) {
   const qm_val_t *qm_ptr = qparam->qmatrix;
   const qm_val_t *iqm_ptr = qparam->iqmatrix;
-  if (qm_ptr != NULL && iqm_ptr != NULL) {
-    quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                        p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                        dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                        sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+  if (qparam->use_quant_b_adapt) {
+    // TODO(sarahparker) These quantize_b optimizations need SIMD
+    // implementations
+    if (qm_ptr != NULL && iqm_ptr != NULL) {
+      quantize_b_adaptive_helper_c(
+          coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+          p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX, eob_ptr,
+          sc->scan, sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+    } else {
+      switch (qparam->log_scale) {
+        case 0:
+          aom_quantize_b_adaptive_c(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        case 1:
+          aom_quantize_b_32x32_adaptive_c(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        case 2:
+          aom_quantize_b_64x64_adaptive_c(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        default: assert(0);
+      }
+    }
   } else {
-    switch (qparam->log_scale) {
-      case 0:
-        aom_quantize_b(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                       p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                       dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                       sc->iscan);
-        break;
-      case 1:
-        aom_quantize_b_32x32(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                             p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                             dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                             sc->iscan);
-        break;
-      case 2:
-        aom_quantize_b_64x64(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                             p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                             dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                             sc->iscan);
-        break;
-      default: assert(0);
+    if (qm_ptr != NULL && iqm_ptr != NULL) {
+      quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+                          p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+                          dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+                          sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+    } else {
+      switch (qparam->log_scale) {
+        case 0:
+          aom_quantize_b(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+                         p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+                         dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+                         sc->iscan);
+          break;
+        case 1:
+          aom_quantize_b_32x32(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+                               p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+                               dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+                               sc->iscan);
+          break;
+        case 2:
+          aom_quantize_b_64x64(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+                               p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+                               dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+                               sc->iscan);
+          break;
+        default: assert(0);
+      }
     }
   }
 }
@@ -391,41 +414,81 @@ void av1_highbd_quantize_b_facade(const tran_low_t *coeff_ptr,
                                   const QUANT_PARAM *qparam) {
   const qm_val_t *qm_ptr = qparam->qmatrix;
   const qm_val_t *iqm_ptr = qparam->iqmatrix;
-  if (qm_ptr != NULL && iqm_ptr != NULL) {
-    highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                               p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                               dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                               sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+  if (qparam->use_quant_b_adapt) {
+    if (qm_ptr != NULL && iqm_ptr != NULL) {
+      highbd_quantize_b_adaptive_helper_c(
+          coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+          p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX, eob_ptr,
+          sc->scan, sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+    } else {
+      switch (qparam->log_scale) {
+        case 0:
+          if (LIKELY(n_coeffs >= 8)) {
+            aom_highbd_quantize_b_adaptive_c(
+                coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+                p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+                eob_ptr, sc->scan, sc->iscan);
+          } else {
+            // TODO(luoyi): Need SIMD (e.g. sse2) for smaller block size
+            // quantization
+            aom_highbd_quantize_b_adaptive_c(
+                coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+                p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+                eob_ptr, sc->scan, sc->iscan);
+          }
+          break;
+        case 1:
+          aom_highbd_quantize_b_32x32_adaptive_c(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        case 2:
+          aom_highbd_quantize_b_64x64_adaptive_c(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        default: assert(0);
+      }
+    }
   } else {
-    switch (qparam->log_scale) {
-      case 0:
-        if (LIKELY(n_coeffs >= 8)) {
-          aom_highbd_quantize_b(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
-                                p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
-                                dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
-                                sc->iscan);
-        } else {
-          // TODO(luoyi): Need SIMD (e.g. sse2) for smaller block size
-          // quantization
-          aom_highbd_quantize_b_c(coeff_ptr, n_coeffs, p->zbin_QTX,
+    if (qm_ptr != NULL && iqm_ptr != NULL) {
+      highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+                                 p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+                                 dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+                                 sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+    } else {
+      switch (qparam->log_scale) {
+        case 0:
+          if (LIKELY(n_coeffs >= 8)) {
+            aom_highbd_quantize_b(coeff_ptr, n_coeffs, p->zbin_QTX,
                                   p->round_QTX, p->quant_QTX,
                                   p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr,
                                   p->dequant_QTX, eob_ptr, sc->scan, sc->iscan);
-        }
-        break;
-      case 1:
-        aom_highbd_quantize_b_32x32(
-            coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
-            p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
-            eob_ptr, sc->scan, sc->iscan);
-        break;
-      case 2:
-        aom_highbd_quantize_b_64x64(
-            coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
-            p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
-            eob_ptr, sc->scan, sc->iscan);
-        break;
-      default: assert(0);
+          } else {
+            // TODO(luoyi): Need SIMD (e.g. sse2) for smaller block size
+            // quantization
+            aom_highbd_quantize_b_c(
+                coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+                p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+                eob_ptr, sc->scan, sc->iscan);
+          }
+          break;
+        case 1:
+          aom_highbd_quantize_b_32x32(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        case 2:
+          aom_highbd_quantize_b_64x64(
+              coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+              p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+              eob_ptr, sc->scan, sc->iscan);
+          break;
+        default: assert(0);
+      }
     }
   }
 }
@@ -697,7 +760,7 @@ void av1_frame_init_quantizer(AV1_COMP *cpi) {
 void av1_set_quantizer(AV1_COMMON *cm, int q) {
   // quantizer has to be reinitialized with av1_init_quantizer() if any
   // delta_q changes.
-  cm->base_qindex = AOMMAX(cm->delta_q_present_flag, q);
+  cm->base_qindex = AOMMAX(cm->delta_q_info.delta_q_present_flag, q);
   cm->y_dc_delta_q = 0;
   cm->u_dc_delta_q = 0;
   cm->u_ac_delta_q = 0;
diff --git a/third_party/aom/av1/encoder/av1_quantize.h b/third_party/aom/av1/encoder/av1_quantize.h
index 35af9a67ae..c709234956 100644
--- a/third_party/aom/av1/encoder/av1_quantize.h
+++ b/third_party/aom/av1/encoder/av1_quantize.h
@@ -27,6 +27,7 @@ typedef struct QUANT_PARAM {
   TX_SIZE tx_size;
   const qm_val_t *qmatrix;
   const qm_val_t *iqmatrix;
+  int use_quant_b_adapt;
 } QUANT_PARAM;
 
 typedef void (*AV1_QUANT_FACADE)(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
@@ -67,7 +68,7 @@ typedef struct {
 
 // The Dequants structure is used only for internal quantizer setup in
 // av1_quantize.c.
-// Fields are sufffixed according to whether or not they're expressed in
+// Fields are suffixed according to whether or not they're expressed in
 // the same coefficient shift/precision as TX or a fixed Q3 format.
 typedef struct {
   DECLARE_ALIGNED(16, int16_t,
diff --git a/third_party/aom/av1/encoder/bitstream.c b/third_party/aom/av1/encoder/bitstream.c
index 2c4acdb021..3140551e9c 100644
--- a/third_party/aom/av1/encoder/bitstream.c
+++ b/third_party/aom/av1/encoder/bitstream.c
@@ -145,7 +145,7 @@ static void write_inter_compound_mode(MACROBLOCKD *xd, aom_writer *w,
 static void write_tx_size_vartx(MACROBLOCKD *xd, const MB_MODE_INFO *mbmi,
                                 TX_SIZE tx_size, int depth, int blk_row,
                                 int blk_col, aom_writer *w) {
-  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
+  FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
   const int max_blocks_high = max_block_high(xd, mbmi->sb_type, 0);
   const int max_blocks_wide = max_block_wide(xd, mbmi->sb_type, 0);
 
@@ -229,7 +229,7 @@ static int write_skip(const AV1_COMMON *cm, const MACROBLOCKD *xd,
 static int write_skip_mode(const AV1_COMMON *cm, const MACROBLOCKD *xd,
                            int segment_id, const MB_MODE_INFO *mi,
                            aom_writer *w) {
-  if (!cm->skip_mode_flag) return 0;
+  if (!cm->current_frame.skip_mode_info.skip_mode_flag) return 0;
   if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
     return 0;
   }
@@ -315,7 +315,7 @@ static void write_delta_lflevel(const AV1_COMMON *cm, const MACROBLOCKD *xd,
   int smallval = abs < DELTA_LF_SMALL ? 1 : 0;
   FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
 
-  if (cm->delta_lf_multi) {
+  if (cm->delta_q_info.delta_lf_multi) {
     assert(lf_id >= 0 && lf_id < (av1_num_planes(cm) > 1 ? FRAME_LF_COUNT
                                                          : FRAME_LF_COUNT - 2));
     aom_write_symbol(w, AOMMIN(abs, DELTA_LF_SMALL),
@@ -460,7 +460,7 @@ static void write_segment_id(AV1_COMP *cpi, const MB_MODE_INFO *const mbmi,
     // changing from lossless to lossy.
     assert(is_inter_block(mbmi) || !cpi->has_lossless_segment);
 
-    set_spatial_segment_id(cm, cm->current_frame_seg_map, mbmi->sb_type, mi_row,
+    set_spatial_segment_id(cm, cm->cur_frame->seg_map, mbmi->sb_type, mi_row,
                            mi_col, pred);
     set_spatial_segment_id(cm, cpi->segmentation_map, mbmi->sb_type, mi_row,
                            mi_col, pred);
@@ -473,7 +473,7 @@ static void write_segment_id(AV1_COMP *cpi, const MB_MODE_INFO *const mbmi,
       av1_neg_interleave(mbmi->segment_id, pred, seg->last_active_segid + 1);
   aom_cdf_prob *pred_cdf = segp->spatial_pred_seg_cdf[cdf_num];
   aom_write_symbol(w, coded_id, pred_cdf, MAX_SEGMENTS);
-  set_spatial_segment_id(cm, cm->current_frame_seg_map, mbmi->sb_type, mi_row,
+  set_spatial_segment_id(cm, cm->cur_frame->seg_map, mbmi->sb_type, mi_row,
                          mi_col, mbmi->segment_id);
 }
 
@@ -500,11 +500,12 @@ static void write_ref_frames(const AV1_COMMON *cm, const MACROBLOCKD *xd,
   } else {
     // does the feature use compound prediction or not
     // (if not specified at the frame/segment level)
-    if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+    if (cm->current_frame.reference_mode == REFERENCE_MODE_SELECT) {
       if (is_comp_ref_allowed(mbmi->sb_type))
         aom_write_symbol(w, is_compound, av1_get_reference_mode_cdf(xd), 2);
     } else {
-      assert((!is_compound) == (cm->reference_mode == SINGLE_REFERENCE));
+      assert((!is_compound) ==
+             (cm->current_frame.reference_mode == SINGLE_REFERENCE));
     }
 
     if (is_compound) {
@@ -866,14 +867,7 @@ static void write_cfl_alphas(FRAME_CONTEXT *const ec_ctx, int idx,
 
 static void write_cdef(AV1_COMMON *cm, MACROBLOCKD *const xd, aom_writer *w,
                        int skip, int mi_col, int mi_row) {
-  if (cm->coded_lossless || cm->allow_intrabc) {
-    // Initialize to indicate no CDEF for safety.
-    cm->cdef_bits = 0;
-    cm->cdef_strengths[0] = 0;
-    cm->nb_cdef_strengths = 1;
-    cm->cdef_uv_strengths[0] = 0;
-    return;
-  }
+  if (cm->coded_lossless || cm->allow_intrabc) return;
 
   const int m = ~((1 << (6 - MI_SIZE_LOG2)) - 1);
   const MB_MODE_INFO *mbmi =
@@ -891,7 +885,7 @@ static void write_cdef(AV1_COMMON *cm, MACROBLOCKD *const xd, aom_writer *w,
                         ? !!(mi_col & mask) + 2 * !!(mi_row & mask)
                         : 0;
   if (xd->cdef_preset[index] == -1 && !skip) {
-    aom_write_literal(w, mbmi->cdef_strength, cm->cdef_bits);
+    aom_write_literal(w, mbmi->cdef_strength, cm->cdef_info.cdef_bits);
     xd->cdef_preset[index] = mbmi->cdef_strength;
   }
 }
@@ -902,7 +896,7 @@ static void write_inter_segment_id(AV1_COMP *cpi, aom_writer *w,
                                    int mi_row, int mi_col, int skip,
                                    int preskip) {
   MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
-  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  MB_MODE_INFO *const mbmi = xd->mi[0];
   AV1_COMMON *const cm = &cpi->common;
 
   if (seg->update_map) {
@@ -912,7 +906,7 @@ static void write_inter_segment_id(AV1_COMP *cpi, aom_writer *w,
       if (seg->segid_preskip) return;
       if (skip) {
         write_segment_id(cpi, mbmi, w, seg, segp, mi_row, mi_col, 1);
-        if (seg->temporal_update) ((MB_MODE_INFO *)mbmi)->seg_id_predicted = 0;
+        if (seg->temporal_update) mbmi->seg_id_predicted = 0;
         return;
       }
     }
@@ -924,7 +918,7 @@ static void write_inter_segment_id(AV1_COMP *cpi, aom_writer *w,
         write_segment_id(cpi, mbmi, w, seg, segp, mi_row, mi_col, 0);
       }
       if (pred_flag) {
-        set_spatial_segment_id(cm, cm->current_frame_seg_map, mbmi->sb_type,
+        set_spatial_segment_id(cm, cm->cur_frame->seg_map, mbmi->sb_type,
                                mi_row, mi_col, mbmi->segment_id);
       }
     } else {
@@ -938,7 +932,9 @@ static void write_inter_segment_id(AV1_COMP *cpi, aom_writer *w,
 static void write_delta_q_params(AV1_COMP *cpi, const int mi_row,
                                  const int mi_col, int skip, aom_writer *w) {
   AV1_COMMON *const cm = &cpi->common;
-  if (cm->delta_q_present_flag) {
+  const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
+
+  if (delta_q_info->delta_q_present_flag) {
     MACROBLOCK *const x = &cpi->td.mb;
     MACROBLOCKD *const xd = &x->e_mbd;
     const MB_MODE_INFO *const mbmi = xd->mi[0];
@@ -951,24 +947,25 @@ static void write_delta_q_params(AV1_COMP *cpi, const int mi_row,
         super_block_upper_left) {
       assert(mbmi->current_qindex > 0);
       const int reduced_delta_qindex =
-          (mbmi->current_qindex - xd->current_qindex) / cm->delta_q_res;
+          (mbmi->current_qindex - xd->current_qindex) /
+          delta_q_info->delta_q_res;
       write_delta_qindex(xd, reduced_delta_qindex, w);
       xd->current_qindex = mbmi->current_qindex;
-      if (cm->delta_lf_present_flag) {
-        if (cm->delta_lf_multi) {
+      if (delta_q_info->delta_lf_present_flag) {
+        if (delta_q_info->delta_lf_multi) {
           const int frame_lf_count =
               av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
           for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
             int reduced_delta_lflevel =
                 (mbmi->delta_lf[lf_id] - xd->delta_lf[lf_id]) /
-                cm->delta_lf_res;
+                delta_q_info->delta_lf_res;
             write_delta_lflevel(cm, xd, lf_id, reduced_delta_lflevel, w);
             xd->delta_lf[lf_id] = mbmi->delta_lf[lf_id];
           }
         } else {
           int reduced_delta_lflevel =
               (mbmi->delta_lf_from_base - xd->delta_lf_from_base) /
-              cm->delta_lf_res;
+              delta_q_info->delta_lf_res;
           write_delta_lflevel(cm, xd, -1, reduced_delta_lflevel, w);
           xd->delta_lf_from_base = mbmi->delta_lf_from_base;
         }
@@ -1105,7 +1102,7 @@ static void pack_inter_mode_mvs(AV1_COMP *cpi, const int mi_row,
       av1_encode_mv(cpi, w, &mbmi->mv[0].as_mv, &ref_mv.as_mv, nmvc, allow_hp);
     }
 
-    if (cpi->common.reference_mode != COMPOUND_REFERENCE &&
+    if (cpi->common.current_frame.reference_mode != COMPOUND_REFERENCE &&
         cpi->common.seq_params.enable_interintra_compound &&
         is_interintra_allowed(mbmi)) {
       const int interintra = mbmi->ref_frame[1] == INTRA_FRAME;
@@ -1130,7 +1127,7 @@ static void pack_inter_mode_mvs(AV1_COMP *cpi, const int mi_row,
     if (mbmi->ref_frame[1] != INTRA_FRAME) write_motion_mode(cm, xd, mbmi, w);
 
     // First write idx to indicate current compound inter prediction mode group
-    // Group A (0): jnt_comp, compound_average
+    // Group A (0): dist_wtd_comp, compound_average
     // Group B (1): interintra, compound_diffwtd, wedge
     if (has_second_ref(mbmi)) {
       const int masked_compound_used = is_any_masked_compound_used(bsize) &&
@@ -1148,7 +1145,7 @@ static void pack_inter_mode_mvs(AV1_COMP *cpi, const int mi_row,
         if (mbmi->compound_idx)
           assert(mbmi->interinter_comp.type == COMPOUND_AVERAGE);
 
-        if (cm->seq_params.enable_jnt_comp) {
+        if (cm->seq_params.order_hint_info.enable_dist_wtd_comp) {
           const int comp_index_ctx = get_comp_index_context(cm, xd);
           aom_write_symbol(w, mbmi->compound_idx,
                            ec_ctx->compound_index_cdf[comp_index_ctx], 2);
@@ -1156,7 +1153,7 @@ static void pack_inter_mode_mvs(AV1_COMP *cpi, const int mi_row,
           assert(mbmi->compound_idx == 1);
         }
       } else {
-        assert(cpi->common.reference_mode != SINGLE_REFERENCE &&
+        assert(cpi->common.current_frame.reference_mode != SINGLE_REFERENCE &&
                is_inter_compound_mode(mbmi->mode) &&
                mbmi->motion_mode == SIMPLE_TRANSLATION);
         assert(masked_compound_used);
@@ -1181,7 +1178,6 @@ static void pack_inter_mode_mvs(AV1_COMP *cpi, const int mi_row,
         }
       }
     }
-
     write_mb_interp_filter(cpi, xd, w);
   }
 }
@@ -1233,13 +1229,14 @@ static void write_mb_modes_kf(AV1_COMP *cpi, MACROBLOCKD *xd,
 }
 
 #if CONFIG_RD_DEBUG
-static void dump_mode_info(MODE_INFO *mi) {
+static void dump_mode_info(MB_MODE_INFO *mi) {
   printf("\nmi->mi_row == %d\n", mi->mi_row);
   printf("&& mi->mi_col == %d\n", mi->mi_col);
   printf("&& mi->sb_type == %d\n", mi->sb_type);
   printf("&& mi->tx_size == %d\n", mi->tx_size);
   printf("&& mi->mode == %d\n", mi->mode);
 }
+
 static int rd_token_stats_mismatch(RD_STATS *rd_stats, TOKEN_STATS *token_stats,
                                    int plane) {
   if (rd_stats->txb_coeff_cost[plane] != token_stats->cost) {
@@ -1270,27 +1267,26 @@ static int rd_token_stats_mismatch(RD_STATS *rd_stats, TOKEN_STATS *token_stats,
 #if ENC_MISMATCH_DEBUG
 static void enc_dump_logs(AV1_COMP *cpi, int mi_row, int mi_col) {
   AV1_COMMON *const cm = &cpi->common;
-  MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
-  xd->mi = cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col);
-  const MB_MODE_INFO *const *mbmi = xd->mi[0];
+  const MB_MODE_INFO *const *mbmi =
+      *(cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col));
+  const MB_MODE_INFO_EXT *const *mbmi_ext =
+      cpi->mbmi_ext_base + (mi_row * cm->mi_cols + mi_col);
   if (is_inter_block(mbmi)) {
 #define FRAME_TO_CHECK 11
-    if (cm->current_video_frame == FRAME_TO_CHECK && cm->show_frame == 1) {
+    if (cm->current_frame.frame_number == FRAME_TO_CHECK &&
+        cm->show_frame == 1) {
       const BLOCK_SIZE bsize = mbmi->sb_type;
 
-      int_mv mv[2];
-      int is_comp_ref = has_second_ref(mbmi);
-      int ref;
+      int_mv mv[2] = { 0 };
+      const int is_comp_ref = has_second_ref(mbmi);
 
-      for (ref = 0; ref < 1 + is_comp_ref; ++ref)
+      for (int ref = 0; ref < 1 + is_comp_ref; ++ref)
         mv[ref].as_mv = mbmi->mv[ref].as_mv;
 
       if (!is_comp_ref) {
         mv[1].as_int = 0;
       }
 
-      MACROBLOCK *const x = &cpi->td.mb;
-      const MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
       const int16_t mode_ctx =
           is_comp_ref ? mbmi_ext->compound_mode_context[mbmi->ref_frame[0]]
                       : av1_mode_context_analyzer(mbmi_ext->mode_context,
@@ -1312,8 +1308,8 @@ static void enc_dump_logs(AV1_COMP *cpi, int mi_row, int mi_col) {
           "show_frame=%d, mv[0]=(%d,%d), mv[1]=(%d,%d), ref[0]=%d, "
           "ref[1]=%d, motion_mode=%d, mode_ctx=%d, "
           "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d, tx_size=%d\n",
-          cm->current_video_frame, mi_row, mi_col, mbmi->skip_mode, mbmi->mode,
-          bsize, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col,
+          cm->current_frame.frame_number, mi_row, mi_col, mbmi->skip_mode,
+          mbmi->mode, bsize, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col,
           mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0],
           mbmi->ref_frame[1], mbmi->motion_mode, mode_ctx, newmv_ctx,
           zeromv_ctx, refmv_ctx, mbmi->tx_size);
@@ -1474,14 +1470,16 @@ static void write_tokens_b(AV1_COMP *cpi, const TileInfo *const tile,
                                   row, col, &block[plane], plane);
           }
         }
+      }
 #if CONFIG_RD_DEBUG
+      for (plane = 0; plane < num_planes && is_inter_block(mbmi); ++plane) {
         if (mbmi->sb_type >= BLOCK_8X8 &&
             rd_token_stats_mismatch(&mbmi->rd_stats, &token_stats, plane)) {
-          dump_mode_info(m);
+          dump_mode_info(mbmi);
           assert(0);
         }
-#endif  // CONFIG_RD_DEBUG
       }
+#endif  // CONFIG_RD_DEBUG
     }
   }
 }
@@ -1685,9 +1683,9 @@ static void write_modes(AV1_COMP *const cpi, const TileInfo *const tile,
   av1_zero_above_context(cm, xd, mi_col_start, mi_col_end, tile->tile_row);
   av1_init_above_context(cm, xd, tile->tile_row);
 
-  if (cpi->common.delta_q_present_flag) {
+  if (cpi->common.delta_q_info.delta_q_present_flag) {
     xd->current_qindex = cpi->common.base_qindex;
-    if (cpi->common.delta_lf_present_flag) {
+    if (cpi->common.delta_q_info.delta_lf_present_flag) {
       av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
     }
   }
@@ -1870,8 +1868,8 @@ static void loop_restoration_write_sb_coeffs(const AV1_COMMON *const cm,
   assert(!cm->all_lossless);
 
   const int wiener_win = (plane > 0) ? WIENER_WIN_CHROMA : WIENER_WIN;
-  WienerInfo *wiener_info = xd->wiener_info + plane;
-  SgrprojInfo *sgrproj_info = xd->sgrproj_info + plane;
+  WienerInfo *ref_wiener_info = &xd->wiener_info[plane];
+  SgrprojInfo *ref_sgrproj_info = &xd->sgrproj_info[plane];
   RestorationType unit_rtype = rui->restoration_type;
 
   if (frame_rtype == RESTORE_SWITCHABLE) {
@@ -1882,10 +1880,10 @@ static void loop_restoration_write_sb_coeffs(const AV1_COMMON *const cm,
 #endif
     switch (unit_rtype) {
       case RESTORE_WIENER:
-        write_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, w);
+        write_wiener_filter(wiener_win, &rui->wiener_info, ref_wiener_info, w);
         break;
       case RESTORE_SGRPROJ:
-        write_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, w);
+        write_sgrproj_filter(&rui->sgrproj_info, ref_sgrproj_info, w);
         break;
       default: assert(unit_rtype == RESTORE_NONE); break;
     }
@@ -1896,7 +1894,7 @@ static void loop_restoration_write_sb_coeffs(const AV1_COMMON *const cm,
     ++counts->wiener_restore[unit_rtype != RESTORE_NONE];
 #endif
     if (unit_rtype != RESTORE_NONE) {
-      write_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, w);
+      write_wiener_filter(wiener_win, &rui->wiener_info, ref_wiener_info, w);
     }
   } else if (frame_rtype == RESTORE_SGRPROJ) {
     aom_write_symbol(w, unit_rtype != RESTORE_NONE,
@@ -1905,7 +1903,7 @@ static void loop_restoration_write_sb_coeffs(const AV1_COMMON *const cm,
     ++counts->sgrproj_restore[unit_rtype != RESTORE_NONE];
 #endif
     if (unit_rtype != RESTORE_NONE) {
-      write_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, w);
+      write_sgrproj_filter(&rui->sgrproj_info, ref_sgrproj_info, w);
     }
   }
 }
@@ -1936,15 +1934,12 @@ static void encode_loopfilter(AV1_COMMON *cm, struct aom_write_bit_buffer *wb) {
     aom_wb_write_bit(wb, lf->mode_ref_delta_update);
 
     if (lf->mode_ref_delta_update) {
-      const int prime_idx = cm->primary_ref_frame;
-      const int buf_idx =
-          prime_idx == PRIMARY_REF_NONE ? -1 : cm->frame_refs[prime_idx].idx;
+      const RefCntBuffer *buf = get_primary_ref_frame_buf(cm);
       int8_t last_ref_deltas[REF_FRAMES];
-      if (prime_idx == PRIMARY_REF_NONE || buf_idx < 0) {
+      if (buf == NULL) {
         av1_set_default_ref_deltas(last_ref_deltas);
       } else {
-        memcpy(last_ref_deltas, cm->buffer_pool->frame_bufs[buf_idx].ref_deltas,
-               REF_FRAMES);
+        memcpy(last_ref_deltas, buf->ref_deltas, REF_FRAMES);
       }
       for (i = 0; i < REF_FRAMES; i++) {
         const int delta = lf->ref_deltas[i];
@@ -1954,12 +1949,10 @@ static void encode_loopfilter(AV1_COMMON *cm, struct aom_write_bit_buffer *wb) {
       }
 
       int8_t last_mode_deltas[MAX_MODE_LF_DELTAS];
-      if (prime_idx == PRIMARY_REF_NONE || buf_idx < 0) {
+      if (buf == NULL) {
         av1_set_default_mode_deltas(last_mode_deltas);
       } else {
-        memcpy(last_mode_deltas,
-               cm->buffer_pool->frame_bufs[buf_idx].mode_deltas,
-               MAX_MODE_LF_DELTAS);
+        memcpy(last_mode_deltas, buf->mode_deltas, MAX_MODE_LF_DELTAS);
       }
       for (i = 0; i < MAX_MODE_LF_DELTAS; i++) {
         const int delta = lf->mode_deltas[i];
@@ -1977,13 +1970,15 @@ static void encode_cdef(const AV1_COMMON *cm, struct aom_write_bit_buffer *wb) {
   if (cm->allow_intrabc) return;
   const int num_planes = av1_num_planes(cm);
   int i;
-  aom_wb_write_literal(wb, cm->cdef_pri_damping - 3, 2);
-  assert(cm->cdef_pri_damping == cm->cdef_sec_damping);
-  aom_wb_write_literal(wb, cm->cdef_bits, 2);
-  for (i = 0; i < cm->nb_cdef_strengths; i++) {
-    aom_wb_write_literal(wb, cm->cdef_strengths[i], CDEF_STRENGTH_BITS);
+  aom_wb_write_literal(wb, cm->cdef_info.cdef_pri_damping - 3, 2);
+  assert(cm->cdef_info.cdef_pri_damping == cm->cdef_info.cdef_sec_damping);
+  aom_wb_write_literal(wb, cm->cdef_info.cdef_bits, 2);
+  for (i = 0; i < cm->cdef_info.nb_cdef_strengths; i++) {
+    aom_wb_write_literal(wb, cm->cdef_info.cdef_strengths[i],
+                         CDEF_STRENGTH_BITS);
     if (num_planes > 1)
-      aom_wb_write_literal(wb, cm->cdef_uv_strengths[i], CDEF_STRENGTH_BITS);
+      aom_wb_write_literal(wb, cm->cdef_info.cdef_uv_strengths[i],
+                           CDEF_STRENGTH_BITS);
   }
 }
 
@@ -2070,15 +2065,6 @@ static void encode_segmentation(AV1_COMMON *cm, MACROBLOCKD *xd,
   }
 }
 
-static void write_tx_mode(AV1_COMMON *cm, TX_MODE *mode,
-                          struct aom_write_bit_buffer *wb) {
-  if (cm->coded_lossless) {
-    *mode = ONLY_4X4;
-    return;
-  }
-  aom_wb_write_bit(wb, *mode == TX_MODE_SELECT);
-}
-
 static void write_frame_interp_filter(InterpFilter filter,
                                       struct aom_write_bit_buffer *wb) {
   aom_wb_write_bit(wb, filter == SWITCHABLE);
@@ -2086,29 +2072,6 @@ static void write_frame_interp_filter(InterpFilter filter,
     aom_wb_write_literal(wb, filter, LOG_SWITCHABLE_FILTERS);
 }
 
-static void fix_interp_filter(AV1_COMMON *cm, FRAME_COUNTS *counts) {
-  if (cm->interp_filter == SWITCHABLE) {
-    // Check to see if only one of the filters is actually used
-    int count[SWITCHABLE_FILTERS];
-    int i, j, c = 0;
-    for (i = 0; i < SWITCHABLE_FILTERS; ++i) {
-      count[i] = 0;
-      for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
-        count[i] += counts->switchable_interp[j][i];
-      c += (count[i] > 0);
-    }
-    if (c == 1) {
-      // Only one filter is used. So set the filter at frame level
-      for (i = 0; i < SWITCHABLE_FILTERS; ++i) {
-        if (count[i]) {
-          if (i == EIGHTTAP_REGULAR) cm->interp_filter = i;
-          break;
-        }
-      }
-    }
-  }
-}
-
 // Same function as write_uniform but writing to uncompresses header wb
 static void wb_write_uniform(struct aom_write_bit_buffer *wb, int n, int v) {
   const int l = get_unsigned_bits(n);
@@ -2206,60 +2169,12 @@ static void write_ext_tile_info(const AV1_COMMON *const cm,
   }
 }
 
-static int get_refresh_mask(AV1_COMP *cpi) {
-  if ((cpi->common.frame_type == KEY_FRAME && cpi->common.show_frame) ||
-      frame_is_sframe(&cpi->common))
-    return 0xFF;
-
-  int refresh_mask = 0;
-
-  // NOTE(zoeliu): When LAST_FRAME is to get refreshed, the decoder will be
-  // notified to get LAST3_FRAME refreshed and then the virtual indexes for all
-  // the 3 LAST reference frames will be updated accordingly, i.e.:
-  // (1) The original virtual index for LAST3_FRAME will become the new virtual
-  //     index for LAST_FRAME; and
-  // (2) The original virtual indexes for LAST_FRAME and LAST2_FRAME will be
-  //     shifted and become the new virtual indexes for LAST2_FRAME and
-  //     LAST3_FRAME.
-  refresh_mask |=
-      (cpi->refresh_last_frame << cpi->ref_fb_idx[LAST_REF_FRAMES - 1]);
-#if USE_SYMM_MULTI_LAYER
-  refresh_mask |=
-      (cpi->new_bwdref_update_rule == 1)
-          ? (cpi->refresh_bwd_ref_frame << cpi->ref_fb_idx[EXTREF_FRAME - 1])
-          : (cpi->refresh_bwd_ref_frame << cpi->ref_fb_idx[BWDREF_FRAME - 1]);
-#else
-  refresh_mask |=
-      (cpi->refresh_bwd_ref_frame << cpi->ref_fb_idx[BWDREF_FRAME - 1]);
-#endif
-  refresh_mask |=
-      (cpi->refresh_alt2_ref_frame << cpi->ref_fb_idx[ALTREF2_FRAME - 1]);
-
-  if (av1_preserve_existing_gf(cpi)) {
-    // We have decided to preserve the previously existing golden frame as our
-    // new ARF frame. However, in the short term we leave it in the GF slot and,
-    // if we're updating the GF with the current decoded frame, we save it
-    // instead to the ARF slot.
-    // Later, in the function av1_encoder.c:av1_update_reference_frames() we
-    // will swap gld_fb_idx and alt_fb_idx to achieve our objective. We do it
-    // there so that it can be done outside of the recode loop.
-    // Note: This is highly specific to the use of ARF as a forward reference,
-    // and this needs to be generalized as other uses are implemented
-    // (like RTC/temporal scalability).
-
-    if (cpi->preserve_arf_as_gld) {
-      return refresh_mask;
-    } else {
-      return refresh_mask |
-             (cpi->refresh_golden_frame << cpi->ref_fb_idx[ALTREF_FRAME - 1]);
-    }
-  } else {
-    const int arf_idx = cpi->ref_fb_idx[ALTREF_FRAME - 1];
-    return refresh_mask |
-           (cpi->refresh_golden_frame << cpi->ref_fb_idx[GOLDEN_FRAME - 1]) |
-           (cpi->refresh_alt_ref_frame << arf_idx);
-  }
-}
+// Stores the location and size of a tile's data in the bitstream.  Used for
+// later identifying identical tiles
+typedef struct TileBufferEnc {
+  uint8_t *data;
+  size_t size;
+} TileBufferEnc;
 
 static INLINE int find_identical_tile(
     const int tile_row, const int tile_col,
@@ -2280,18 +2195,18 @@ static INLINE int find_identical_tile(
     int col_offset = candidate_offset[0].col;
     int row = tile_row - row_offset;
     int col = tile_col - col_offset;
-    uint8_t tile_hdr;
     const uint8_t *tile_data;
     TileBufferEnc *candidate;
 
     if (row < 0 || col < 0) continue;
 
-    tile_hdr = *(tile_buffers[row][col].data);
+    const uint32_t tile_hdr = mem_get_le32(tile_buffers[row][col].data);
 
-    // Read out tcm bit
-    if ((tile_hdr >> 7) == 1) {
-      // The candidate is a copy tile itself
-      row_offset += tile_hdr & 0x7f;
+    // Read out tile-copy-mode bit:
+    if ((tile_hdr >> 31) == 1) {
+      // The candidate is a copy tile itself: the offset is stored in bits
+      // 30 through 24 inclusive.
+      row_offset += (tile_hdr >> 24) & 0x7f;
       row = tile_row - row_offset;
     }
 
@@ -2361,14 +2276,13 @@ static void write_frame_size(const AV1_COMMON *cm, int frame_size_override,
   write_render_size(cm, wb);
 }
 
-static void write_frame_size_with_refs(AV1_COMP *cpi,
+static void write_frame_size_with_refs(const AV1_COMMON *const cm,
                                        struct aom_write_bit_buffer *wb) {
-  AV1_COMMON *const cm = &cpi->common;
   int found = 0;
 
   MV_REFERENCE_FRAME ref_frame;
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    YV12_BUFFER_CONFIG *cfg = get_ref_frame_buffer(cpi, ref_frame);
+    const YV12_BUFFER_CONFIG *cfg = get_ref_frame_yv12_buf(cm, ref_frame);
 
     if (cfg != NULL) {
       found = cm->superres_upscaled_width == cfg->y_crop_width &&
@@ -2431,9 +2345,7 @@ static void write_color_config(const SequenceHeader *const seq_params,
   }
   if (seq_params->color_primaries == AOM_CICP_CP_BT_709 &&
       seq_params->transfer_characteristics == AOM_CICP_TC_SRGB &&
-      seq_params->matrix_coefficients ==
-          AOM_CICP_MC_IDENTITY) {  // it would be better to remove this
-                                   // dependency too
+      seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) {
     assert(seq_params->subsampling_x == 0 && seq_params->subsampling_y == 0);
     assert(seq_params->profile == PROFILE_1 ||
            (seq_params->profile == PROFILE_2 &&
@@ -2532,34 +2444,27 @@ static void write_tu_pts_info(AV1_COMMON *const cm,
       cm->buffer_model.frame_presentation_time_length);
 }
 
-static void write_film_grain_params(AV1_COMP *cpi,
+static void write_film_grain_params(const AV1_COMP *const cpi,
                                     struct aom_write_bit_buffer *wb) {
-  AV1_COMMON *const cm = &cpi->common;
-  aom_film_grain_t *pars = &cm->film_grain_params;
-
-  cm->cur_frame->film_grain_params = *pars;
+  const AV1_COMMON *const cm = &cpi->common;
+  const aom_film_grain_t *const pars = &cm->cur_frame->film_grain_params;
 
   aom_wb_write_bit(wb, pars->apply_grain);
   if (!pars->apply_grain) return;
 
   aom_wb_write_literal(wb, pars->random_seed, 16);
 
-  pars->random_seed += 3381;  // Changing random seed for film grain
-  if (!pars->random_seed)     // Random seed should not be zero
-    pars->random_seed += 7391;
-  if (cm->frame_type == INTER_FRAME)
+  if (cm->current_frame.frame_type == INTER_FRAME)
     aom_wb_write_bit(wb, pars->update_parameters);
-  else
-    pars->update_parameters = 1;
+
   if (!pars->update_parameters) {
-    RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
-    int ref_frame, ref_idx, buf_idx;
+    int ref_frame, ref_idx;
     for (ref_frame = LAST_FRAME; ref_frame < REF_FRAMES; ref_frame++) {
-      ref_idx = get_ref_frame_map_idx(cpi, ref_frame);
+      ref_idx = get_ref_frame_map_idx(cm, ref_frame);
       assert(ref_idx != INVALID_IDX);
-      buf_idx = cm->ref_frame_map[ref_idx];
-      if (frame_bufs[buf_idx].film_grain_params_present &&
-          memcmp(pars, &frame_bufs[buf_idx].film_grain_params, sizeof(*pars))) {
+      const RefCntBuffer *const buf = cm->ref_frame_map[ref_idx];
+      if (buf->film_grain_params_present &&
+          av1_check_grain_params_equiv(pars, &buf->film_grain_params)) {
         break;
       }
     }
@@ -2575,16 +2480,16 @@ static void write_film_grain_params(AV1_COMP *cpi,
     aom_wb_write_literal(wb, pars->scaling_points_y[i][1], 8);
   }
 
-  if (!cm->seq_params.monochrome)
+  if (!cm->seq_params.monochrome) {
     aom_wb_write_bit(wb, pars->chroma_scaling_from_luma);
-  else
-    pars->chroma_scaling_from_luma = 0;  // for monochrome override to 0
+  } else {
+    assert(!pars->chroma_scaling_from_luma);
+  }
 
   if (cm->seq_params.monochrome || pars->chroma_scaling_from_luma ||
       ((cm->seq_params.subsampling_x == 1) &&
        (cm->seq_params.subsampling_y == 1) && (pars->num_y_points == 0))) {
-    pars->num_cb_points = 0;
-    pars->num_cr_points = 0;
+    assert(pars->num_cb_points == 0 && pars->num_cr_points == 0);
   } else {
     aom_wb_write_literal(wb, pars->num_cb_points, 4);  // max 10
     for (int i = 0; i < pars->num_cb_points; i++) {
@@ -2644,7 +2549,7 @@ static void write_film_grain_params(AV1_COMP *cpi,
   aom_wb_write_bit(wb, pars->clip_to_restricted_range);
 }
 
-static void write_sb_size(SequenceHeader *seq_params,
+static void write_sb_size(const SequenceHeader *const seq_params,
                           struct aom_write_bit_buffer *wb) {
   (void)seq_params;
   (void)wb;
@@ -2655,43 +2560,16 @@ static void write_sb_size(SequenceHeader *seq_params,
   aom_wb_write_bit(wb, seq_params->sb_size == BLOCK_128X128 ? 1 : 0);
 }
 
-static void write_sequence_header(AV1_COMP *cpi,
+static void write_sequence_header(const SequenceHeader *const seq_params,
                                   struct aom_write_bit_buffer *wb) {
-  AV1_COMMON *const cm = &cpi->common;
-  SequenceHeader *seq_params = &cm->seq_params;
+  aom_wb_write_literal(wb, seq_params->num_bits_width - 1, 4);
+  aom_wb_write_literal(wb, seq_params->num_bits_height - 1, 4);
+  aom_wb_write_literal(wb, seq_params->max_frame_width - 1,
+                       seq_params->num_bits_width);
+  aom_wb_write_literal(wb, seq_params->max_frame_height - 1,
+                       seq_params->num_bits_height);
 
-  int max_frame_width = cpi->oxcf.forced_max_frame_width
-                            ? cpi->oxcf.forced_max_frame_width
-                            : cpi->oxcf.width;
-  int max_frame_height = cpi->oxcf.forced_max_frame_height
-                             ? cpi->oxcf.forced_max_frame_height
-                             : cpi->oxcf.height;
-  // max((int)ceil(log2(max_frame_width)), 1)
-  const int num_bits_width =
-      (max_frame_width > 1) ? get_msb(max_frame_width - 1) + 1 : 1;
-  // max((int)ceil(log2(max_frame_height)), 1)
-  const int num_bits_height =
-      (max_frame_height > 1) ? get_msb(max_frame_height - 1) + 1 : 1;
-  assert(num_bits_width <= 16);
-  assert(num_bits_height <= 16);
-
-  seq_params->num_bits_width = num_bits_width;
-  seq_params->num_bits_height = num_bits_height;
-  seq_params->max_frame_width = max_frame_width;
-  seq_params->max_frame_height = max_frame_height;
-
-  aom_wb_write_literal(wb, num_bits_width - 1, 4);
-  aom_wb_write_literal(wb, num_bits_height - 1, 4);
-  aom_wb_write_literal(wb, max_frame_width - 1, num_bits_width);
-  aom_wb_write_literal(wb, max_frame_height - 1, num_bits_height);
-
-  /* Placeholder for actually writing to the bitstream */
   if (!seq_params->reduced_still_picture_hdr) {
-    seq_params->frame_id_numbers_present_flag =
-        cm->large_scale_tile ? 0 : cm->error_resilient_mode;
-    seq_params->frame_id_length = FRAME_ID_LENGTH;
-    seq_params->delta_frame_id_length = DELTA_FRAME_ID_LENGTH;
-
     aom_wb_write_bit(wb, seq_params->frame_id_numbers_present_flag);
     if (seq_params->frame_id_numbers_present_flag) {
       // We must always have delta_frame_id_length < frame_id_length,
@@ -2716,11 +2594,11 @@ static void write_sequence_header(AV1_COMP *cpi,
     aom_wb_write_bit(wb, seq_params->enable_warped_motion);
     aom_wb_write_bit(wb, seq_params->enable_dual_filter);
 
-    aom_wb_write_bit(wb, seq_params->enable_order_hint);
+    aom_wb_write_bit(wb, seq_params->order_hint_info.enable_order_hint);
 
-    if (seq_params->enable_order_hint) {
-      aom_wb_write_bit(wb, seq_params->enable_jnt_comp);
-      aom_wb_write_bit(wb, seq_params->enable_ref_frame_mvs);
+    if (seq_params->order_hint_info.enable_order_hint) {
+      aom_wb_write_bit(wb, seq_params->order_hint_info.enable_dist_wtd_comp);
+      aom_wb_write_bit(wb, seq_params->order_hint_info.enable_ref_frame_mvs);
     }
     if (seq_params->force_screen_content_tools == 2) {
       aom_wb_write_bit(wb, 1);
@@ -2738,8 +2616,9 @@ static void write_sequence_header(AV1_COMP *cpi,
     } else {
       assert(seq_params->force_integer_mv == 2);
     }
-    if (seq_params->enable_order_hint)
-      aom_wb_write_literal(wb, seq_params->order_hint_bits_minus_1, 3);
+    if (seq_params->order_hint_info.enable_order_hint)
+      aom_wb_write_literal(
+          wb, seq_params->order_hint_info.order_hint_bits_minus_1, 3);
   }
 
   aom_wb_write_bit(wb, seq_params->enable_superres);
@@ -2815,7 +2694,7 @@ static void write_global_motion(AV1_COMP *cpi,
     // does not work currently and causes mismatches when resize is on.
     // Fix it before turning the optimization back on.
     /*
-    YV12_BUFFER_CONFIG *ref_buf = get_ref_frame_buffer(cpi, frame);
+    YV12_BUFFER_CONFIG *ref_buf = get_ref_frame_yv12_buf(cpi, frame);
     if (cpi->source->y_crop_width == ref_buf->y_crop_width &&
         cpi->source->y_crop_height == ref_buf->y_crop_height) {
       write_global_motion_params(&cm->global_motion[frame],
@@ -2828,7 +2707,7 @@ static void write_global_motion(AV1_COMP *cpi,
     */
     /*
     printf("Frame %d/%d: Enc Ref %d: %d %d %d %d\n",
-           cm->current_video_frame, cm->show_frame, frame,
+           cm->current_frame.frame_number, cm->show_frame, frame,
            cm->global_motion[frame].wmmat[0],
            cm->global_motion[frame].wmmat[1], cm->global_motion[frame].wmmat[2],
            cm->global_motion[frame].wmmat[3]);
@@ -2836,23 +2715,22 @@ static void write_global_motion(AV1_COMP *cpi,
   }
 }
 
-static void check_frame_refs_short_signaling(AV1_COMP *const cpi) {
-  AV1_COMMON *const cm = &cpi->common;
-  if (!cm->frame_refs_short_signaling) return;
-
+static int check_frame_refs_short_signaling(AV1_COMMON *const cm) {
   // Check whether all references are distinct frames.
-  int buf_markers[FRAME_BUFFERS] = { 0 };
-  for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-    if (buf_idx != INVALID_IDX) {
-      assert(buf_idx >= 0 && buf_idx < FRAME_BUFFERS);
-      buf_markers[buf_idx] = 1;
-    }
-  }
-
+  const RefCntBuffer *seen_bufs[FRAME_BUFFERS] = { NULL };
   int num_refs = 0;
-  for (int buf_idx = 0; buf_idx < FRAME_BUFFERS; ++buf_idx) {
-    num_refs += buf_markers[buf_idx];
+  for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    if (buf != NULL) {
+      int seen = 0;
+      for (int i = 0; i < num_refs; i++) {
+        if (seen_bufs[i] == buf) {
+          seen = 1;
+          break;
+        }
+      }
+      if (!seen) seen_bufs[num_refs++] = buf;
+    }
   }
 
   // We only turn on frame_refs_short_signaling when all references are
@@ -2860,20 +2738,22 @@ static void check_frame_refs_short_signaling(AV1_COMP *const cpi) {
   if (num_refs < INTER_REFS_PER_FRAME) {
     // It indicates that there exist more than one reference frame pointing to
     // the same reference buffer, i.e. two or more references are duplicate.
-    cm->frame_refs_short_signaling = 0;
-    return;
+    return 0;
   }
 
   // Check whether the encoder side ref frame choices are aligned with that to
   // be derived at the decoder side.
-  RefBuffer frame_refs_copy[INTER_REFS_PER_FRAME];
+  int remapped_ref_idx_copy[REF_FRAMES];
+  struct scale_factors ref_scale_factors_copy[REF_FRAMES];
 
   // Backup the frame refs info
-  memcpy(frame_refs_copy, cm->frame_refs,
-         INTER_REFS_PER_FRAME * sizeof(RefBuffer));
+  memcpy(remapped_ref_idx_copy, cm->remapped_ref_idx,
+         REF_FRAMES * sizeof(*remapped_ref_idx_copy));
+  memcpy(ref_scale_factors_copy, cm->ref_scale_factors,
+         REF_FRAMES * sizeof(*ref_scale_factors_copy));
 
-  const int lst_map_idx = get_ref_frame_map_idx(cpi, LAST_FRAME);
-  const int gld_map_idx = get_ref_frame_map_idx(cpi, GOLDEN_FRAME);
+  const int lst_map_idx = get_ref_frame_map_idx(cm, LAST_FRAME);
+  const int gld_map_idx = get_ref_frame_map_idx(cm, GOLDEN_FRAME);
 
   // Set up the frame refs mapping indexes according to the
   // frame_refs_short_signaling policy.
@@ -2881,32 +2761,39 @@ static void check_frame_refs_short_signaling(AV1_COMP *const cpi) {
 
   // We only turn on frame_refs_short_signaling when the encoder side decision
   // on ref frames is identical to that at the decoder side.
+  int frame_refs_short_signaling = 1;
   for (int ref_idx = 0; ref_idx < INTER_REFS_PER_FRAME; ++ref_idx) {
     // Compare the buffer index between two reference frames indexed
     // respectively by the encoder and the decoder side decisions.
-    if (cm->frame_refs[ref_idx].idx != frame_refs_copy[ref_idx].idx) {
-      cm->frame_refs_short_signaling = 0;
+    RefCntBuffer *ref_frame_buf_copy = NULL;
+    if (remapped_ref_idx_copy[ref_idx] != INVALID_IDX)
+      ref_frame_buf_copy = cm->ref_frame_map[remapped_ref_idx_copy[ref_idx]];
+    if (get_ref_frame_buf(cm, LAST_FRAME + ref_idx) != ref_frame_buf_copy) {
+      frame_refs_short_signaling = 0;
       break;
     }
   }
 
 #if 0   // For debug
-  printf("\nFrame=%d: \n", cm->current_video_frame);
-  printf("***frame_refs_short_signaling=%d\n", cm->frame_refs_short_signaling);
+  printf("\nFrame=%d: \n", cm->current_frame.frame_number);
+  printf("***frame_refs_short_signaling=%d\n", frame_refs_short_signaling);
   for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    printf("enc_ref(map_idx=%d, buf_idx=%d)=%d, vs. "
-        "dec_ref(map_idx=%d, buf_idx=%d)=%d\n",
-        get_ref_frame_map_idx(cpi, ref_frame),
-        get_ref_frame_buf_idx(cpi, ref_frame), ref_frame,
-        cm->frame_refs[ref_frame - LAST_FRAME].map_idx,
-        cm->frame_refs[ref_frame - LAST_FRAME].idx, ref_frame);
+    printf("enc_ref(map_idx=%d)=%d, vs. "
+        "dec_ref(map_idx=%d)=%d\n",
+        get_ref_frame_map_idx(cm, ref_frame), ref_frame,
+        cm->remapped_ref_idx[ref_frame - LAST_FRAME],
+        ref_frame);
   }
 #endif  // 0
 
   // Restore the frame refs info if frame_refs_short_signaling is off.
-  if (!cm->frame_refs_short_signaling)
-    memcpy(cm->frame_refs, frame_refs_copy,
-           INTER_REFS_PER_FRAME * sizeof(RefBuffer));
+  if (!frame_refs_short_signaling) {
+    memcpy(cm->remapped_ref_idx, remapped_ref_idx_copy,
+           REF_FRAMES * sizeof(*remapped_ref_idx_copy));
+    memcpy(cm->ref_scale_factors, ref_scale_factors_copy,
+           REF_FRAMES * sizeof(*ref_scale_factors_copy));
+  }
+  return frame_refs_short_signaling;
 }
 
 // New function based on HLS R18
@@ -2916,28 +2803,15 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
   AV1_COMMON *const cm = &cpi->common;
   const SequenceHeader *const seq_params = &cm->seq_params;
   MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
-
-  // NOTE: By default all coded frames to be used as a reference
-  cm->is_reference_frame = 1;
-  cm->frame_type = cm->intra_only ? INTRA_ONLY_FRAME : cm->frame_type;
+  CurrentFrame *const current_frame = &cm->current_frame;
 
   if (seq_params->still_picture) {
     assert(cm->show_existing_frame == 0);
     assert(cm->show_frame == 1);
-    assert(cm->frame_type == KEY_FRAME);
+    assert(current_frame->frame_type == KEY_FRAME);
   }
   if (!seq_params->reduced_still_picture_hdr) {
     if (encode_show_existing_frame(cm)) {
-      RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
-      const int frame_to_show = cm->ref_frame_map[cpi->existing_fb_idx_to_show];
-
-      if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) {
-        aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                           "Buffer %d does not contain a reconstructed frame",
-                           frame_to_show);
-      }
-      ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show);
-
       aom_wb_write_bit(wb, 1);  // show_existing_frame
       aom_wb_write_literal(wb, cpi->existing_fb_idx_to_show, 3);
 
@@ -2950,20 +2824,12 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
         int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show];
         aom_wb_write_literal(wb, display_frame_id, frame_id_len);
       }
-
-      if (cm->reset_decoder_state &&
-          frame_bufs[frame_to_show].frame_type != KEY_FRAME) {
-        aom_internal_error(
-            &cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-            "show_existing_frame to reset state on KEY_FRAME only");
-      }
-
       return;
     } else {
       aom_wb_write_bit(wb, 0);  // show_existing_frame
     }
 
-    aom_wb_write_literal(wb, cm->frame_type, 2);
+    aom_wb_write_literal(wb, current_frame->frame_type, 2);
 
     aom_wb_write_bit(wb, cm->show_frame);
     if (cm->show_frame) {
@@ -2975,7 +2841,7 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
     }
     if (frame_is_sframe(cm)) {
       assert(cm->error_resilient_mode);
-    } else if (!(cm->frame_type == KEY_FRAME && cm->show_frame)) {
+    } else if (!(current_frame->frame_type == KEY_FRAME && cm->show_frame)) {
       aom_wb_write_bit(wb, cm->error_resilient_mode);
     }
   }
@@ -2998,9 +2864,7 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
     assert(cm->cur_frame_force_integer_mv == 0);
   }
 
-  cm->invalid_delta_frame_id_minus_1 = 0;
   int frame_size_override_flag = 0;
-  cm->frame_refs_short_signaling = 0;
 
   if (seq_params->reduced_still_picture_hdr) {
     assert(cm->width == seq_params->max_frame_width &&
@@ -3023,9 +2887,10 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
                                cm->height != seq_params->max_frame_height);
     if (!frame_is_sframe(cm)) aom_wb_write_bit(wb, frame_size_override_flag);
 
-    if (seq_params->enable_order_hint)
-      aom_wb_write_literal(wb, cm->frame_offset,
-                           seq_params->order_hint_bits_minus_1 + 1);
+    if (seq_params->order_hint_info.enable_order_hint)
+      aom_wb_write_literal(
+          wb, current_frame->order_hint,
+          seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
 
     if (!cm->error_resilient_mode && !frame_is_intra_only(cm)) {
       aom_wb_write_literal(wb, cm->primary_ref_frame, PRIMARY_REF_BITS);
@@ -3058,121 +2923,76 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
       }
     }
   }
-  cpi->refresh_frame_mask = get_refresh_mask(cpi);
-  if (cm->frame_type == KEY_FRAME) {
-    if (!cm->show_frame) {  // unshown keyframe (forward keyframe)
-      aom_wb_write_literal(wb, cpi->refresh_frame_mask, REF_FRAMES);
-    } else {
-      assert(cpi->refresh_frame_mask == 0xFF);
-    }
-  } else {
-    if (cm->frame_type == INTRA_ONLY_FRAME) {
-      assert(cpi->refresh_frame_mask != 0xFF);
-      int updated_fb = -1;
-      for (int i = 0; i < REF_FRAMES; i++) {
-        // If more than one frame is refreshed, it doesn't matter which one
-        // we pick, so pick the first.
-        if (cpi->refresh_frame_mask & (1 << i)) {
-          updated_fb = i;
-          break;
-        }
-      }
-      assert(updated_fb >= 0);
-      cm->fb_of_context_type[cm->frame_context_idx] = updated_fb;
-      aom_wb_write_literal(wb, cpi->refresh_frame_mask, REF_FRAMES);
-    } else if (cm->frame_type == INTER_FRAME || frame_is_sframe(cm)) {
-      if (cm->frame_type == INTER_FRAME) {
-        aom_wb_write_literal(wb, cpi->refresh_frame_mask, REF_FRAMES);
-      } else {
-        assert(frame_is_sframe(cm) && cpi->refresh_frame_mask == 0xFF);
-      }
-      int updated_fb = -1;
-      for (int i = 0; i < REF_FRAMES; i++) {
-        // If more than one frame is refreshed, it doesn't matter which one
-        // we pick, so pick the first.
-        if (cpi->refresh_frame_mask & (1 << i)) {
-          updated_fb = i;
-          break;
-        }
-      }
-      // large scale tile sometimes won't refresh any fbs
-      if (updated_fb >= 0) {
-        cm->fb_of_context_type[cm->frame_context_idx] = updated_fb;
-      }
 
-      if (!cpi->refresh_frame_mask) {
-        // NOTE: "cpi->refresh_frame_mask == 0" indicates that the coded frame
-        //       will not be used as a reference
-        cm->is_reference_frame = 0;
-      }
-    }
-  }
+  // Shown keyframes and switch-frames automatically refreshes all reference
+  // frames.  For all other frame types, we need to write refresh_frame_flags.
+  if ((current_frame->frame_type == KEY_FRAME && !cm->show_frame) ||
+      current_frame->frame_type == INTER_FRAME ||
+      current_frame->frame_type == INTRA_ONLY_FRAME)
+    aom_wb_write_literal(wb, current_frame->refresh_frame_flags, REF_FRAMES);
 
-  if (!frame_is_intra_only(cm) || cpi->refresh_frame_mask != 0xFF) {
+  if (!frame_is_intra_only(cm) || current_frame->refresh_frame_flags != 0xff) {
     // Write all ref frame order hints if error_resilient_mode == 1
-    if (cm->error_resilient_mode && seq_params->enable_order_hint) {
-      RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
+    if (cm->error_resilient_mode &&
+        seq_params->order_hint_info.enable_order_hint) {
       for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) {
-        // Get buffer index
-        const int buf_idx = cm->ref_frame_map[ref_idx];
-        assert(buf_idx >= 0 && buf_idx < FRAME_BUFFERS);
-
-        // Write order hint to bit stream
-        aom_wb_write_literal(wb, frame_bufs[buf_idx].cur_frame_offset,
-                             seq_params->order_hint_bits_minus_1 + 1);
+        aom_wb_write_literal(
+            wb, cm->ref_frame_map[ref_idx]->order_hint,
+            seq_params->order_hint_info.order_hint_bits_minus_1 + 1);
       }
     }
   }
 
-  if (cm->frame_type == KEY_FRAME) {
+  if (current_frame->frame_type == KEY_FRAME) {
     write_frame_size(cm, frame_size_override_flag, wb);
     assert(!av1_superres_scaled(cm) || !cm->allow_intrabc);
     if (cm->allow_screen_content_tools && !av1_superres_scaled(cm))
       aom_wb_write_bit(wb, cm->allow_intrabc);
-    // all eight fbs are refreshed, pick one that will live long enough
-    cm->fb_of_context_type[REGULAR_FRAME] = 0;
   } else {
-    if (cm->frame_type == INTRA_ONLY_FRAME) {
+    if (current_frame->frame_type == INTRA_ONLY_FRAME) {
       write_frame_size(cm, frame_size_override_flag, wb);
       assert(!av1_superres_scaled(cm) || !cm->allow_intrabc);
       if (cm->allow_screen_content_tools && !av1_superres_scaled(cm))
         aom_wb_write_bit(wb, cm->allow_intrabc);
-    } else if (cm->frame_type == INTER_FRAME || frame_is_sframe(cm)) {
+    } else if (current_frame->frame_type == INTER_FRAME ||
+               frame_is_sframe(cm)) {
       MV_REFERENCE_FRAME ref_frame;
 
       // NOTE: Error resilient mode turns off frame_refs_short_signaling
       //       automatically.
+      int frame_refs_short_signaling = 0;
 #define FRAME_REFS_SHORT_SIGNALING 0
 #if FRAME_REFS_SHORT_SIGNALING
-      cm->frame_refs_short_signaling = seq_params->enable_order_hint;
+      frame_refs_short_signaling =
+          seq_params->order_hint_info.enable_order_hint;
 #endif  // FRAME_REFS_SHORT_SIGNALING
 
-      if (cm->frame_refs_short_signaling) {
+      if (frame_refs_short_signaling) {
         // NOTE(zoeliu@google.com):
         //   An example solution for encoder-side implementation on frame refs
         //   short signaling, which is only turned on when the encoder side
         //   decision on ref frames is identical to that at the decoder side.
-        check_frame_refs_short_signaling(cpi);
+        frame_refs_short_signaling = check_frame_refs_short_signaling(cm);
       }
 
-      if (seq_params->enable_order_hint)
-        aom_wb_write_bit(wb, cm->frame_refs_short_signaling);
+      if (seq_params->order_hint_info.enable_order_hint)
+        aom_wb_write_bit(wb, frame_refs_short_signaling);
 
-      if (cm->frame_refs_short_signaling) {
-        const int lst_ref = get_ref_frame_map_idx(cpi, LAST_FRAME);
+      if (frame_refs_short_signaling) {
+        const int lst_ref = get_ref_frame_map_idx(cm, LAST_FRAME);
         aom_wb_write_literal(wb, lst_ref, REF_FRAMES_LOG2);
 
-        const int gld_ref = get_ref_frame_map_idx(cpi, GOLDEN_FRAME);
+        const int gld_ref = get_ref_frame_map_idx(cm, GOLDEN_FRAME);
         aom_wb_write_literal(wb, gld_ref, REF_FRAMES_LOG2);
       }
 
       for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-        assert(get_ref_frame_map_idx(cpi, ref_frame) != INVALID_IDX);
-        if (!cm->frame_refs_short_signaling)
-          aom_wb_write_literal(wb, get_ref_frame_map_idx(cpi, ref_frame),
+        assert(get_ref_frame_map_idx(cm, ref_frame) != INVALID_IDX);
+        if (!frame_refs_short_signaling)
+          aom_wb_write_literal(wb, get_ref_frame_map_idx(cm, ref_frame),
                                REF_FRAMES_LOG2);
         if (seq_params->frame_id_numbers_present_flag) {
-          int i = get_ref_frame_map_idx(cpi, ref_frame);
+          int i = get_ref_frame_map_idx(cm, ref_frame);
           int frame_id_len = seq_params->frame_id_length;
           int diff_len = seq_params->delta_frame_id_length;
           int delta_frame_id_minus_1 =
@@ -3181,24 +3001,22 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
                (1 << frame_id_len)) -
               1;
           if (delta_frame_id_minus_1 < 0 ||
-              delta_frame_id_minus_1 >= (1 << diff_len))
-            cm->invalid_delta_frame_id_minus_1 = 1;
+              delta_frame_id_minus_1 >= (1 << diff_len)) {
+            aom_internal_error(&cpi->common.error, AOM_CODEC_ERROR,
+                               "Invalid delta_frame_id_minus_1");
+          }
           aom_wb_write_literal(wb, delta_frame_id_minus_1, diff_len);
         }
       }
 
       if (!cm->error_resilient_mode && frame_size_override_flag) {
-        write_frame_size_with_refs(cpi, wb);
+        write_frame_size_with_refs(cm, wb);
       } else {
         write_frame_size(cm, frame_size_override_flag, wb);
       }
 
-      if (cm->cur_frame_force_integer_mv) {
-        cm->allow_high_precision_mv = 0;
-      } else {
+      if (!cm->cur_frame_force_integer_mv)
         aom_wb_write_bit(wb, cm->allow_high_precision_mv);
-      }
-      fix_interp_filter(cm, cpi->td.counts);
       write_frame_interp_filter(cm->interp_filter, wb);
       aom_wb_write_bit(wb, cm->switchable_motion_mode);
       if (frame_might_allow_ref_frame_mvs(cm)) {
@@ -3212,7 +3030,7 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
   const int might_bwd_adapt =
       !(seq_params->reduced_still_picture_hdr) && !(cm->disable_cdf_update);
   if (cm->large_scale_tile)
-    cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
+    assert(cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_DISABLED);
 
   if (might_bwd_adapt) {
     aom_wb_write_bit(
@@ -3223,19 +3041,20 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
   encode_quantization(cm, wb);
   encode_segmentation(cm, xd, wb);
 
-  if (cm->delta_q_present_flag) assert(cm->base_qindex > 0);
+  const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
+  if (delta_q_info->delta_q_present_flag) assert(cm->base_qindex > 0);
   if (cm->base_qindex > 0) {
-    aom_wb_write_bit(wb, cm->delta_q_present_flag);
-    if (cm->delta_q_present_flag) {
-      aom_wb_write_literal(wb, get_msb(cm->delta_q_res), 2);
+    aom_wb_write_bit(wb, delta_q_info->delta_q_present_flag);
+    if (delta_q_info->delta_q_present_flag) {
+      aom_wb_write_literal(wb, get_msb(delta_q_info->delta_q_res), 2);
       xd->current_qindex = cm->base_qindex;
       if (cm->allow_intrabc)
-        assert(cm->delta_lf_present_flag == 0);
+        assert(delta_q_info->delta_lf_present_flag == 0);
       else
-        aom_wb_write_bit(wb, cm->delta_lf_present_flag);
-      if (cm->delta_lf_present_flag) {
-        aom_wb_write_literal(wb, get_msb(cm->delta_lf_res), 2);
-        aom_wb_write_bit(wb, cm->delta_lf_multi);
+        aom_wb_write_bit(wb, delta_q_info->delta_lf_present_flag);
+      if (delta_q_info->delta_lf_present_flag) {
+        aom_wb_write_literal(wb, get_msb(delta_q_info->delta_lf_res), 2);
+        aom_wb_write_bit(wb, delta_q_info->delta_lf_multi);
         av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
       }
     }
@@ -3251,15 +3070,21 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
     encode_restoration_mode(cm, wb);
   }
 
-  write_tx_mode(cm, &cm->tx_mode, wb);
+  // Write TX mode
+  if (cm->coded_lossless)
+    assert(cm->tx_mode == ONLY_4X4);
+  else
+    aom_wb_write_bit(wb, cm->tx_mode == TX_MODE_SELECT);
 
-  if (cpi->allow_comp_inter_inter) {
-    const int use_hybrid_pred = cm->reference_mode == REFERENCE_MODE_SELECT;
+  if (!frame_is_intra_only(cm)) {
+    const int use_hybrid_pred =
+        current_frame->reference_mode == REFERENCE_MODE_SELECT;
 
     aom_wb_write_bit(wb, use_hybrid_pred);
   }
 
-  if (cm->is_skip_mode_allowed) aom_wb_write_bit(wb, cm->skip_mode_flag);
+  if (current_frame->skip_mode_info.skip_mode_allowed)
+    aom_wb_write_bit(wb, current_frame->skip_mode_info.skip_mode_flag);
 
   if (frame_might_allow_warped_motion(cm))
     aom_wb_write_bit(wb, cm->allow_warped_motion);
@@ -3271,19 +3096,9 @@ static void write_uncompressed_header_obu(AV1_COMP *cpi,
   if (!frame_is_intra_only(cm)) write_global_motion(cpi, wb);
 
   if (seq_params->film_grain_params_present &&
-      (cm->show_frame || cm->showable_frame)) {
-    int flip_back_update_parameters_flag = 0;
-    if (cm->frame_type != INTER_FRAME &&
-        cm->film_grain_params.update_parameters == 0) {
-      cm->film_grain_params.update_parameters = 1;
-      flip_back_update_parameters_flag = 1;
-    }
+      (cm->show_frame || cm->showable_frame))
     write_film_grain_params(cpi, wb);
 
-    if (flip_back_update_parameters_flag)
-      cm->film_grain_params.update_parameters = 0;
-  }
-
   if (cm->large_scale_tile) write_ext_tile_info(cm, saved_wb, wb);
 }
 
@@ -3538,7 +3353,7 @@ uint32_t write_sequence_header_obu(AV1_COMP *cpi, uint8_t *const dst) {
       }
     }
   }
-  write_sequence_header(cpi, &wb);
+  write_sequence_header(&cm->seq_params, &wb);
 
   write_color_config(&cm->seq_params, &wb);
 
@@ -3560,8 +3375,8 @@ static uint32_t write_frame_header_obu(AV1_COMP *cpi,
   return aom_wb_bytes_written(&wb);
 }
 
-static uint32_t write_tile_group_header(uint8_t *const dst, int startTile,
-                                        int endTile, int tiles_log2,
+static uint32_t write_tile_group_header(uint8_t *const dst, int start_tile,
+                                        int end_tile, int tiles_log2,
                                         int tile_start_and_end_present_flag) {
   struct aom_write_bit_buffer wb = { dst, 0 };
   uint32_t size = 0;
@@ -3571,8 +3386,8 @@ static uint32_t write_tile_group_header(uint8_t *const dst, int startTile,
   aom_wb_write_bit(&wb, tile_start_and_end_present_flag);
 
   if (tile_start_and_end_present_flag) {
-    aom_wb_write_literal(&wb, startTile, tiles_log2);
-    aom_wb_write_literal(&wb, endTile, tiles_log2);
+    aom_wb_write_literal(&wb, start_tile, tiles_log2);
+    aom_wb_write_literal(&wb, end_tile, tiles_log2);
   }
 
   size = aom_wb_bytes_written(&wb);
@@ -3588,11 +3403,13 @@ typedef struct {
 static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
                                        struct aom_write_bit_buffer *saved_wb,
                                        uint8_t obu_extension_header,
-                                       const FrameHeaderInfo *fh_info) {
+                                       const FrameHeaderInfo *fh_info,
+                                       int *const largest_tile_id) {
   AV1_COMMON *const cm = &cpi->common;
   aom_writer mode_bc;
   int tile_row, tile_col;
-  TileBufferEnc(*const tile_buffers)[MAX_TILE_COLS] = cpi->tile_buffers;
+  // Store the location and size of each tile's data in the bitstream:
+  TileBufferEnc tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
   uint32_t total_size = 0;
   const int tile_cols = cm->tile_cols;
   const int tile_rows = cm->tile_rows;
@@ -3613,7 +3430,7 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
   const int have_tiles = tile_cols * tile_rows > 1;
   int first_tg = 1;
 
-  cm->largest_tile_id = 0;
+  *largest_tile_id = 0;
 
   if (cm->large_scale_tile) {
     // For large_scale_tile case, we always have only one tile group, so it can
@@ -3631,9 +3448,9 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
 #if EXT_TILE_DEBUG
     {
       char fn[20] = "./fh";
-      fn[4] = cm->current_video_frame / 100 + '0';
-      fn[5] = (cm->current_video_frame % 100) / 10 + '0';
-      fn[6] = (cm->current_video_frame % 10) + '0';
+      fn[4] = cm->current_frame.frame_number / 100 + '0';
+      fn[5] = (cm->current_frame.frame_number % 100) / 10 + '0';
+      fn[6] = (cm->current_frame.frame_number % 10) + '0';
       fn[7] = '\0';
       av1_print_uncompressed_frame_header(data - frame_header_size,
                                           frame_header_size, fn);
@@ -3666,8 +3483,6 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
         // Is CONFIG_EXT_TILE = 1, every tile in the row has a header,
         // even for the last one, unless no tiling is used at all.
         total_size += data_offset;
-        // Initialise tile context from the frame context
-        this_tile->tctx = *cm->fc;
         cpi->td.mb.e_mbd.tile_ctx = &this_tile->tctx;
         mode_bc.allow_update_cdf = !cm->large_scale_tile;
         mode_bc.allow_update_cdf =
@@ -3681,7 +3496,7 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
         // Record the maximum tile size we see, so we can compact headers later.
         if (tile_size > max_tile_size) {
           max_tile_size = tile_size;
-          cm->largest_tile_id = tile_cols * tile_row + tile_col;
+          *largest_tile_id = tile_cols * tile_row + tile_col;
         }
 
         if (have_tiles) {
@@ -3695,10 +3510,13 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
           // If tile_copy_mode = 1, check if this tile is a copy tile.
           // Very low chances to have copy tiles on the key frames, so don't
           // search on key frames to reduce unnecessary search.
-          if (cm->frame_type != KEY_FRAME && tile_copy_mode) {
+          if (cm->current_frame.frame_type != KEY_FRAME && tile_copy_mode) {
             const int identical_tile_offset =
                 find_identical_tile(tile_row, tile_col, tile_buffers);
 
+            // Indicate a copy-tile by setting the most significant bit.
+            // The row-offset to copy from is stored in the highest byte.
+            // remux_tiles will move these around later
             if (identical_tile_offset > 0) {
               tile_size = 0;
               tile_header = identical_tile_offset | 0x80;
@@ -3804,8 +3622,6 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
       // The last tile of the tile group does not have a header.
       if (!is_last_tile_in_tg) total_size += 4;
 
-      // Initialise tile context from the frame context
-      this_tile->tctx = *cm->fc;
       cpi->td.mb.e_mbd.tile_ctx = &this_tile->tctx;
       mode_bc.allow_update_cdf = 1;
       mode_bc.allow_update_cdf =
@@ -3822,7 +3638,7 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
       curr_tg_data_size += (tile_size + (is_last_tile_in_tg ? 0 : 4));
       buf->size = tile_size;
       if (tile_size > max_tile_size) {
-        cm->largest_tile_id = tile_cols * tile_row + tile_col;
+        *largest_tile_id = tile_cols * tile_row + tile_col;
         max_tile_size = tile_size;
       }
 
@@ -3857,7 +3673,7 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
           // Force context update tile to be the first tile in error
           // resiliant mode as the duplicate frame headers will have
           // context_update_tile_id set to 0
-          cm->largest_tile_id = 0;
+          *largest_tile_id = 0;
 
           // Rewrite the OBU header to change the OBU type to Redundant Frame
           // Header.
@@ -3880,7 +3696,7 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
     // Fill in context_update_tile_id indicating the tile to use for the
     // cdf update. The encoder currently sets it to the largest tile
     // (but is up to the encoder)
-    aom_wb_overwrite_literal(saved_wb, cm->largest_tile_id,
+    aom_wb_overwrite_literal(saved_wb, *largest_tile_id,
                              cm->log2_tile_cols + cm->log2_tile_rows);
     // If more than one tile group. tile_size_bytes takes the default value 4
     // and does not need to be set. For a single tile group it is set in the
@@ -3926,7 +3742,8 @@ static uint32_t write_tiles_in_tg_obus(AV1_COMP *const cpi, uint8_t *const dst,
   return total_size;
 }
 
-int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size) {
+int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size,
+                       int *const largest_tile_id) {
   uint8_t *data = dst;
   uint32_t data_size;
   AV1_COMMON *const cm = &cpi->common;
@@ -3943,7 +3760,7 @@ int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size) {
   // The TD is now written outside the frame encode loop
 
   // write sequence header obu if KEY_FRAME, preceded by 4-byte size
-  if (cm->frame_type == KEY_FRAME && cm->show_frame) {
+  if (cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) {
     obu_header_size = write_obu_header(OBU_SEQUENCE_HEADER, 0, data);
 
     obu_payload_size = write_sequence_header_obu(cpi, data + obu_header_size);
@@ -3990,8 +3807,8 @@ int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size) {
   } else {
     //  Each tile group obu will be preceded by 4-byte size of the tile group
     //  obu
-    data_size = write_tiles_in_tg_obus(cpi, data, &saved_wb,
-                                       obu_extension_header, &fh_info);
+    data_size = write_tiles_in_tg_obus(
+        cpi, data, &saved_wb, obu_extension_header, &fh_info, largest_tile_id);
   }
   data += data_size;
   *size = data - dst;
diff --git a/third_party/aom/av1/encoder/bitstream.h b/third_party/aom/av1/encoder/bitstream.h
index 465ccaed57..a567086e95 100644
--- a/third_party/aom/av1/encoder/bitstream.h
+++ b/third_party/aom/av1/encoder/bitstream.h
@@ -33,9 +33,10 @@ uint32_t write_obu_header(OBU_TYPE obu_type, int obu_extension,
 int write_uleb_obu_size(uint32_t obu_header_size, uint32_t obu_payload_size,
                         uint8_t *dest);
 
-int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dest, size_t *size);
+int av1_pack_bitstream(AV1_COMP *const cpi, uint8_t *dst, size_t *size,
+                       int *const largest_tile_id);
 
-static INLINE int av1_preserve_existing_gf(AV1_COMP *cpi) {
+static INLINE int av1_preserve_existing_gf(const AV1_COMP *const cpi) {
   // Do not swap gf and arf indices for internal overlay frames
   return cpi->rc.is_src_frame_alt_ref && !cpi->rc.is_src_frame_ext_arf;
 }
diff --git a/third_party/aom/av1/encoder/block.h b/third_party/aom/av1/encoder/block.h
index 0bc5dea825..87761e9065 100644
--- a/third_party/aom/av1/encoder/block.h
+++ b/third_party/aom/av1/encoder/block.h
@@ -54,10 +54,10 @@ typedef struct macroblock_plane {
 typedef struct {
   int txb_skip_cost[TXB_SKIP_CONTEXTS][2];
   int base_eob_cost[SIG_COEF_CONTEXTS_EOB][3];
-  int base_cost[SIG_COEF_CONTEXTS][4];
+  int base_cost[SIG_COEF_CONTEXTS][8];
   int eob_extra_cost[EOB_COEF_CONTEXTS][2];
   int dc_sign_cost[DC_SIGN_CONTEXTS][2];
-  int lps_cost[LEVEL_CONTEXTS][COEFF_BASE_RANGE + 1];
+  int lps_cost[LEVEL_CONTEXTS][COEFF_BASE_RANGE + 1 + COEFF_BASE_RANGE + 1];
 } LV_MAP_COEFF_COST;
 
 typedef struct {
@@ -74,16 +74,16 @@ typedef struct {
 } CB_COEFF_BUFFER;
 
 typedef struct {
-  int16_t mode_context[MODE_CTX_REF_FRAMES];
   // TODO(angiebird): Reduce the buffer size according to sb_type
   tran_low_t *tcoeff[MAX_MB_PLANE];
   uint16_t *eobs[MAX_MB_PLANE];
   uint8_t *txb_skip_ctx[MAX_MB_PLANE];
   int *dc_sign_ctx[MAX_MB_PLANE];
-  uint8_t ref_mv_count[MODE_CTX_REF_FRAMES];
   CANDIDATE_MV ref_mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE];
   int_mv global_mvs[REF_FRAMES];
   int16_t compound_mode_context[MODE_CTX_REF_FRAMES];
+  int16_t mode_context[MODE_CTX_REF_FRAMES];
+  uint8_t ref_mv_count[MODE_CTX_REF_FRAMES];
 } MB_MODE_INFO_EXT;
 
 typedef struct {
@@ -140,6 +140,20 @@ typedef struct tx_size_rd_info_node {
   struct tx_size_rd_info_node *children[4];
 } TXB_RD_INFO_NODE;
 
+// Simple translation rd state for prune_comp_search_by_single_result
+typedef struct {
+  RD_STATS rd_stats;
+  RD_STATS rd_stats_y;
+  RD_STATS rd_stats_uv;
+  uint8_t blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE];
+  uint8_t skip;
+  uint8_t disable_skip;
+  uint8_t early_skipped;
+} SimpleRDState;
+
+// 4: NEAREST, NEW, NEAR, GLOBAL
+#define SINGLE_REF_MODES ((REF_FRAMES - 1) * 4)
+
 // Region size for mode decision sampling in the first pass of partition
 // search(two_pass_partition_search speed feature), in units of mi size(4).
 // Used by the mode_pruning_based_on_two_pass_partition_search speed feature.
@@ -173,6 +187,19 @@ typedef struct {
   COMPOUND_TYPE comp_type;
 } INTERPOLATION_FILTER_STATS;
 
+#define MAX_COMP_RD_STATS 64
+typedef struct {
+  int32_t rate[COMPOUND_TYPES];
+  int64_t dist[COMPOUND_TYPES];
+  int_mv mv[2];
+  MV_REFERENCE_FRAME ref_frames[2];
+  PREDICTION_MODE mode;
+  InterpFilters filter;
+  int ref_mv_idx;
+  int is_global[2];
+} COMP_RD_STATS;
+
+struct inter_modes_info;
 typedef struct macroblock MACROBLOCK;
 struct macroblock {
   struct macroblock_plane plane[MAX_MB_PLANE];
@@ -194,6 +221,9 @@ struct macroblock {
   INTERPOLATION_FILTER_STATS interp_filter_stats[2][MAX_INTERP_FILTER_STATS];
   int interp_filter_stats_idx[2];
 
+  // prune_comp_search_by_single_result (3:MAX_REF_MV_SERCH)
+  SimpleRDState simple_rd_state[SINGLE_REF_MODES][3];
+
   // Activate constrained coding block partition search range.
   int use_cb_search_range;
 
@@ -224,12 +254,16 @@ struct macroblock {
   // for sub-8x8 blocks.
   int sadperbit4;
   int rdmult;
+  int cb_rdmult;
   int mb_energy;
   int sb_energy_level;
   int *m_search_count_ptr;
   int *ex_search_count_ptr;
 
   unsigned int txb_split_count;
+#if CONFIG_SPEED_STATS
+  unsigned int tx_search_count;
+#endif  // CONFIG_SPEED_STATS
 
   // These are set to their default values at the beginning, and then adjusted
   // further in the encoding process.
@@ -241,12 +275,10 @@ struct macroblock {
   unsigned int pred_sse[REF_FRAMES];
   int pred_mv_sad[REF_FRAMES];
 
-  int *nmvjointcost;
   int nmv_vec_cost[MV_JOINTS];
   int *nmvcost[2];
   int *nmvcost_hp[2];
   int **mv_cost_stack;
-  int **mvcost;
 
   int32_t *wsrc_buf;
   int32_t *mask_buf;
@@ -258,6 +290,16 @@ struct macroblock {
   CONV_BUF_TYPE *tmp_conv_dst;
   uint8_t *tmp_obmc_bufs[2];
 
+  FRAME_CONTEXT *row_ctx;
+  // This context will be used to update color_map_cdf pointer which would be
+  // used during pack bitstream. For single thread and tile-multithreading case
+  // this ponter will be same as xd->tile_ctx, but for the case of row-mt:
+  // xd->tile_ctx will point to a temporary context while tile_pb_ctx will point
+  // to the accurate tile context.
+  FRAME_CONTEXT *tile_pb_ctx;
+
+  struct inter_modes_info *inter_modes_info;
+
   // buffer for hash value calculation of a block
   // used only in av1_get_block_hash_value()
   // [first hash/second hash]
@@ -351,6 +393,9 @@ struct macroblock {
   // Store the second best motion vector during full-pixel motion search
   int_mv second_best_mv;
 
+  // Store the fractional best motion vector during sub/Qpel-pixel motion search
+  int_mv fractional_best_mv[3];
+
   // use default transform and skip transform type search for intra modes
   int use_default_intra_tx_type;
   // use default transform and skip transform type search for inter modes
@@ -367,6 +412,14 @@ struct macroblock {
   int tx_split_prune_flag;  // Flag to skip tx split RD search.
   int recalc_luma_mc_data;  // Flag to indicate recalculation of MC data during
                             // interpolation filter search
+  // The likelihood of an edge existing in the block (using partial Canny edge
+  // detection). For reference, 556 is the value returned for a solid
+  // vertical black/white edge.
+  uint16_t edge_strength;
+
+  // [Saved stat index]
+  COMP_RD_STATS comp_rd_stats[MAX_COMP_RD_STATS];
+  int comp_rd_stats_idx;
 };
 
 static INLINE int is_rect_tx_allowed_bsize(BLOCK_SIZE bsize) {
diff --git a/third_party/aom/av1/encoder/context_tree.c b/third_party/aom/av1/encoder/context_tree.c
index 57f59f304b..40df6c1dae 100644
--- a/third_party/aom/av1/encoder/context_tree.c
+++ b/third_party/aom/av1/encoder/context_tree.c
@@ -16,8 +16,15 @@ static const BLOCK_SIZE square[MAX_SB_SIZE_LOG2 - 1] = {
   BLOCK_4X4, BLOCK_8X8, BLOCK_16X16, BLOCK_32X32, BLOCK_64X64, BLOCK_128X128,
 };
 
+typedef struct {
+  tran_low_t *coeff_buf[MAX_MB_PLANE];
+  tran_low_t *qcoeff_buf[MAX_MB_PLANE];
+  tran_low_t *dqcoeff_buf[MAX_MB_PLANE];
+} PC_TREE_SHARED_BUFFERS;
+
 static void alloc_mode_context(AV1_COMMON *cm, int num_pix,
-                               PICK_MODE_CONTEXT *ctx) {
+                               PICK_MODE_CONTEXT *ctx,
+                               PC_TREE_SHARED_BUFFERS *shared_bufs) {
   const int num_planes = av1_num_planes(cm);
   int i;
   const int num_blk = num_pix / 16;
@@ -25,12 +32,9 @@ static void alloc_mode_context(AV1_COMMON *cm, int num_pix,
 
   CHECK_MEM_ERROR(cm, ctx->blk_skip, aom_calloc(num_blk, sizeof(uint8_t)));
   for (i = 0; i < num_planes; ++i) {
-    CHECK_MEM_ERROR(cm, ctx->coeff[i],
-                    aom_memalign(32, num_pix * sizeof(*ctx->coeff[i])));
-    CHECK_MEM_ERROR(cm, ctx->qcoeff[i],
-                    aom_memalign(32, num_pix * sizeof(*ctx->qcoeff[i])));
-    CHECK_MEM_ERROR(cm, ctx->dqcoeff[i],
-                    aom_memalign(32, num_pix * sizeof(*ctx->dqcoeff[i])));
+    ctx->coeff[i] = shared_bufs->coeff_buf[i];
+    ctx->qcoeff[i] = shared_bufs->qcoeff_buf[i];
+    ctx->dqcoeff[i] = shared_bufs->dqcoeff_buf[i];
     CHECK_MEM_ERROR(cm, ctx->eobs[i],
                     aom_memalign(32, num_blk * sizeof(*ctx->eobs[i])));
     CHECK_MEM_ERROR(
@@ -52,11 +56,8 @@ static void free_mode_context(PICK_MODE_CONTEXT *ctx, const int num_planes) {
   aom_free(ctx->blk_skip);
   ctx->blk_skip = 0;
   for (i = 0; i < num_planes; ++i) {
-    aom_free(ctx->coeff[i]);
     ctx->coeff[i] = 0;
-    aom_free(ctx->qcoeff[i]);
     ctx->qcoeff[i] = 0;
-    aom_free(ctx->dqcoeff[i]);
     ctx->dqcoeff[i] = 0;
     aom_free(ctx->eobs[i]);
     ctx->eobs[i] = 0;
@@ -71,36 +72,37 @@ static void free_mode_context(PICK_MODE_CONTEXT *ctx, const int num_planes) {
 }
 
 static void alloc_tree_contexts(AV1_COMMON *cm, PC_TREE *tree, int num_pix,
-                                int is_leaf) {
-  alloc_mode_context(cm, num_pix, &tree->none);
+                                int is_leaf,
+                                PC_TREE_SHARED_BUFFERS *shared_bufs) {
+  alloc_mode_context(cm, num_pix, &tree->none, shared_bufs);
 
   if (is_leaf) return;
 
-  alloc_mode_context(cm, num_pix / 2, &tree->horizontal[0]);
-  alloc_mode_context(cm, num_pix / 2, &tree->vertical[0]);
+  alloc_mode_context(cm, num_pix / 2, &tree->horizontal[0], shared_bufs);
+  alloc_mode_context(cm, num_pix / 2, &tree->vertical[0], shared_bufs);
 
-  alloc_mode_context(cm, num_pix / 2, &tree->horizontal[1]);
-  alloc_mode_context(cm, num_pix / 2, &tree->vertical[1]);
+  alloc_mode_context(cm, num_pix / 2, &tree->horizontal[1], shared_bufs);
+  alloc_mode_context(cm, num_pix / 2, &tree->vertical[1], shared_bufs);
 
-  alloc_mode_context(cm, num_pix / 4, &tree->horizontala[0]);
-  alloc_mode_context(cm, num_pix / 4, &tree->horizontala[1]);
-  alloc_mode_context(cm, num_pix / 2, &tree->horizontala[2]);
+  alloc_mode_context(cm, num_pix / 4, &tree->horizontala[0], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->horizontala[1], shared_bufs);
+  alloc_mode_context(cm, num_pix / 2, &tree->horizontala[2], shared_bufs);
 
-  alloc_mode_context(cm, num_pix / 2, &tree->horizontalb[0]);
-  alloc_mode_context(cm, num_pix / 4, &tree->horizontalb[1]);
-  alloc_mode_context(cm, num_pix / 4, &tree->horizontalb[2]);
+  alloc_mode_context(cm, num_pix / 2, &tree->horizontalb[0], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->horizontalb[1], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->horizontalb[2], shared_bufs);
 
-  alloc_mode_context(cm, num_pix / 4, &tree->verticala[0]);
-  alloc_mode_context(cm, num_pix / 4, &tree->verticala[1]);
-  alloc_mode_context(cm, num_pix / 2, &tree->verticala[2]);
+  alloc_mode_context(cm, num_pix / 4, &tree->verticala[0], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->verticala[1], shared_bufs);
+  alloc_mode_context(cm, num_pix / 2, &tree->verticala[2], shared_bufs);
 
-  alloc_mode_context(cm, num_pix / 2, &tree->verticalb[0]);
-  alloc_mode_context(cm, num_pix / 4, &tree->verticalb[1]);
-  alloc_mode_context(cm, num_pix / 4, &tree->verticalb[2]);
+  alloc_mode_context(cm, num_pix / 2, &tree->verticalb[0], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->verticalb[1], shared_bufs);
+  alloc_mode_context(cm, num_pix / 4, &tree->verticalb[2], shared_bufs);
 
   for (int i = 0; i < 4; ++i) {
-    alloc_mode_context(cm, num_pix / 4, &tree->horizontal4[i]);
-    alloc_mode_context(cm, num_pix / 4, &tree->vertical4[i]);
+    alloc_mode_context(cm, num_pix / 4, &tree->horizontal4[i], shared_bufs);
+    alloc_mode_context(cm, num_pix / 4, &tree->vertical4[i], shared_bufs);
   }
 }
 
@@ -135,6 +137,7 @@ void av1_setup_pc_tree(AV1_COMMON *cm, ThreadData *td) {
   const int tree_nodes = tree_nodes_inc + 256 + 64 + 16 + 4 + 1;
   int pc_tree_index = 0;
   PC_TREE *this_pc;
+  PC_TREE_SHARED_BUFFERS shared_bufs;
   int square_index = 1;
   int nodes;
 
@@ -143,11 +146,24 @@ void av1_setup_pc_tree(AV1_COMMON *cm, ThreadData *td) {
                   aom_calloc(tree_nodes, sizeof(*td->pc_tree)));
   this_pc = &td->pc_tree[0];
 
+  for (i = 0; i < 3; i++) {
+    const int max_num_pix = MAX_SB_SIZE * MAX_SB_SIZE;
+    CHECK_MEM_ERROR(cm, td->tree_coeff_buf[i],
+                    aom_memalign(32, max_num_pix * sizeof(tran_low_t)));
+    CHECK_MEM_ERROR(cm, td->tree_qcoeff_buf[i],
+                    aom_memalign(32, max_num_pix * sizeof(tran_low_t)));
+    CHECK_MEM_ERROR(cm, td->tree_dqcoeff_buf[i],
+                    aom_memalign(32, max_num_pix * sizeof(tran_low_t)));
+    shared_bufs.coeff_buf[i] = td->tree_coeff_buf[i];
+    shared_bufs.qcoeff_buf[i] = td->tree_qcoeff_buf[i];
+    shared_bufs.dqcoeff_buf[i] = td->tree_dqcoeff_buf[i];
+  }
+
   // Sets up all the leaf nodes in the tree.
   for (pc_tree_index = 0; pc_tree_index < leaf_nodes; ++pc_tree_index) {
     PC_TREE *const tree = &td->pc_tree[pc_tree_index];
     tree->block_size = square[0];
-    alloc_tree_contexts(cm, tree, 16, 1);
+    alloc_tree_contexts(cm, tree, 16, 1, &shared_bufs);
   }
 
   // Each node has 4 leaf nodes, fill each block_size level of the tree
@@ -155,7 +171,7 @@ void av1_setup_pc_tree(AV1_COMMON *cm, ThreadData *td) {
   for (nodes = leaf_nodes >> 2; nodes > 0; nodes >>= 2) {
     for (i = 0; i < nodes; ++i) {
       PC_TREE *const tree = &td->pc_tree[pc_tree_index];
-      alloc_tree_contexts(cm, tree, 16 << (2 * square_index), 0);
+      alloc_tree_contexts(cm, tree, 16 << (2 * square_index), 0, &shared_bufs);
       tree->block_size = square[square_index];
       for (j = 0; j < 4; j++) tree->split[j] = this_pc++;
       ++pc_tree_index;
@@ -181,6 +197,14 @@ void av1_free_pc_tree(ThreadData *td, const int num_planes) {
     for (int i = 0; i < tree_nodes; ++i) {
       free_tree_contexts(&td->pc_tree[i], num_planes);
     }
+    for (int i = 0; i < 3; ++i) {
+      aom_free(td->tree_coeff_buf[i]);
+      aom_free(td->tree_qcoeff_buf[i]);
+      aom_free(td->tree_dqcoeff_buf[i]);
+      td->tree_coeff_buf[i] = NULL;
+      td->tree_qcoeff_buf[i] = NULL;
+      td->tree_dqcoeff_buf[i] = NULL;
+    }
     aom_free(td->pc_tree);
     td->pc_tree = NULL;
   }
diff --git a/third_party/aom/av1/encoder/context_tree.h b/third_party/aom/av1/encoder/context_tree.h
index 4efc349852..66f24f958e 100644
--- a/third_party/aom/av1/encoder/context_tree.h
+++ b/third_party/aom/av1/encoder/context_tree.h
@@ -23,7 +23,7 @@ struct AV1_COMP;
 struct AV1Common;
 struct ThreadData;
 
-typedef enum {
+enum {
   // Search all the partition types in this plane.
   SEARCH_FULL_PLANE = 0,
   // Only search none_partition coding block.
@@ -32,12 +32,14 @@ typedef enum {
   SEARCH_SAME_PLANE = 2,
   // Skip search partition on this plane. Go split directly.
   SPLIT_PLANE = 3,
-} CB_TREE_SEARCH;
+} UENUM1BYTE(CB_TREE_SEARCH);
 
 // Structure to hold snapshot of coding context during the mode picking process
 typedef struct {
   MB_MODE_INFO mic;
   MB_MODE_INFO_EXT mbmi_ext;
+  int64_t dist;
+  int64_t rdcost;
   uint8_t *color_index_map[2];
   uint8_t *blk_skip;
 
@@ -62,30 +64,42 @@ typedef struct {
   // TODO(jingning) Use RD_COST struct here instead. This involves a boarder
   // scope of refactoring.
   int rate;
-  int64_t dist;
-  int64_t rdcost;
+
   int rd_mode_is_ready;  // Flag to indicate whether rd pick mode decision has
                          // been made.
+  int mode_selected;
+#if CONFIG_ONE_PASS_SVM
+  // Features for one pass svm early term
+  int seg_feat;
+#endif
 
   // motion vector cache for adaptive motion search control in partition
   // search loop
   MV pred_mv[REF_FRAMES];
   InterpFilter pred_interp_filter;
   PARTITION_TYPE partition;
+
+  // Reference and prediction mode cache for ref/mode speedup
+  // TODO(zoeliu@gmail.com): The values of ref_selected and mode_selected will
+  // be explored for further encoder speedup, to differentiate this approach for
+  // setting skip_ref_frame_mask from others. For instance, it is possible that
+  // the underlying square block(s) share the same SIMPLE_TRANSLATION motion
+  // mode as well as the mode of GLOBALMV, more ref/mode combos could be
+  // skipped.
+  MV_REFERENCE_FRAME ref_selected[2];
 } PICK_MODE_CONTEXT;
 
 typedef struct {
+  int64_t rdcost;
+  int64_t sub_block_rdcost[4];
   int valid;
   int split;
-  int skip;
-  int64_t rdcost;
   int sub_block_split[4];
   int sub_block_skip[4];
-  int64_t sub_block_rdcost[4];
+  int skip;
 } PC_TREE_STATS;
 
 typedef struct PC_TREE {
-  int index;
   PARTITION_TYPE partitioning;
   BLOCK_SIZE block_size;
   PICK_MODE_CONTEXT none;
@@ -97,9 +111,11 @@ typedef struct PC_TREE {
   PICK_MODE_CONTEXT verticalb[3];
   PICK_MODE_CONTEXT horizontal4[4];
   PICK_MODE_CONTEXT vertical4[4];
-  CB_TREE_SEARCH cb_search_range;
   struct PC_TREE *split[4];
   PC_TREE_STATS pc_tree_stats;
+  CB_TREE_SEARCH cb_search_range;
+  int index;
+  MV mv_ref_fulls[REF_FRAMES];
 } PC_TREE;
 
 void av1_setup_pc_tree(struct AV1Common *cm, struct ThreadData *td);
diff --git a/third_party/aom/av1/encoder/cost.h b/third_party/aom/av1/encoder/cost.h
index af5b098370..be0241a820 100644
--- a/third_party/aom/av1/encoder/cost.h
+++ b/third_party/aom/av1/encoder/cost.h
@@ -30,6 +30,10 @@ extern const uint16_t av1_prob_cost[128];
 
 // Calculate the cost of a symbol with probability p15 / 2^15
 static INLINE int av1_cost_symbol(aom_cdf_prob p15) {
+  // p15 can be out of range [1, CDF_PROB_TOP - 1]. Clamping it, so that the
+  // following cost calculation works correctly. Otherwise, if p15 =
+  // CDF_PROB_TOP, shift would be -1, and "p15 << shift" would be wrong.
+  p15 = (aom_cdf_prob)clamp(p15, 1, CDF_PROB_TOP - 1);
   assert(0 < p15 && p15 < CDF_PROB_TOP);
   const int shift = CDF_PROB_BITS - 1 - get_msb(p15);
   const int prob = get_prob(p15 << shift, CDF_PROB_TOP);
diff --git a/third_party/aom/av1/encoder/encode_strategy.c b/third_party/aom/av1/encoder/encode_strategy.c
new file mode 100644
index 0000000000..cc933487d0
--- /dev/null
+++ b/third_party/aom/av1/encoder/encode_strategy.c
@@ -0,0 +1,918 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdint.h>
+
+#include "config/aom_config.h"
+#include "config/aom_scale_rtcd.h"
+
+#include "aom/aom_codec.h"
+#include "aom/aom_encoder.h"
+
+#include "aom_ports/system_state.h"
+
+#if CONFIG_MISMATCH_DEBUG
+#include "aom_util/debug_util.h"
+#endif  // CONFIG_MISMATCH_DEBUG
+
+#include "av1/common/onyxc_int.h"
+
+#include "av1/encoder/encoder.h"
+#include "av1/encoder/encode_strategy.h"
+#include "av1/encoder/firstpass.h"
+#include "av1/encoder/temporal_filter.h"
+#include "av1/encoder/tpl_model.h"
+
+// Define the reference buffers that will be updated post encode.
+void av1_configure_buffer_updates(AV1_COMP *cpi, const FRAME_UPDATE_TYPE type) {
+  // NOTE(weitinglin): Should we define another function to take care of
+  // cpi->rc.is_$Source_Type to make this function as it is in the comment?
+
+  // show_existing_frame is a flag left set from the end of encoding the
+  // previous frame.  Alongside it, is_src_frame_alt_ref may also be left
+  // set so shouldn't be cleared in this case.
+  if (!cpi->common.show_existing_frame) cpi->rc.is_src_frame_alt_ref = 0;
+
+  cpi->rc.is_bwd_ref_frame = 0;
+  cpi->rc.is_last_bipred_frame = 0;
+  cpi->rc.is_bipred_frame = 0;
+  cpi->rc.is_src_frame_ext_arf = 0;
+
+  switch (type) {
+    case KF_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_bwd_ref_frame = 1;
+      cpi->refresh_alt2_ref_frame = 1;
+      cpi->refresh_alt_ref_frame = 1;
+      break;
+
+    case LF_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+      break;
+
+    case GF_UPDATE:
+      // TODO(zoeliu): To further investigate whether 'refresh_last_frame' is
+      //               needed.
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+      break;
+
+    case OVERLAY_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 1;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+
+      cpi->rc.is_src_frame_alt_ref = 1;
+      break;
+
+    case ARF_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 0;
+      // NOTE: BWDREF does not get updated along with ALTREF_FRAME.
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 1;
+      break;
+
+    case BRF_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_bwd_ref_frame = 1;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+
+      cpi->rc.is_bwd_ref_frame = 1;
+      break;
+
+    case LAST_BIPRED_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+
+      cpi->rc.is_last_bipred_frame = 1;
+      break;
+
+    case BIPRED_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+
+      cpi->rc.is_bipred_frame = 1;
+      break;
+
+    case INTNL_OVERLAY_UPDATE:
+      cpi->refresh_last_frame = 1;
+      cpi->refresh_golden_frame = 0;
+      cpi->refresh_bwd_ref_frame = 0;
+      cpi->refresh_alt2_ref_frame = 0;
+      cpi->refresh_alt_ref_frame = 0;
+
+      cpi->rc.is_src_frame_alt_ref = 1;
+      cpi->rc.is_src_frame_ext_arf = 1;
+      break;
+
+    case INTNL_ARF_UPDATE:
+      cpi->refresh_last_frame = 0;
+      cpi->refresh_golden_frame = 0;
+      if (cpi->new_bwdref_update_rule == 1 && cpi->oxcf.pass == 2) {
+        cpi->refresh_bwd_ref_frame = 1;
+        cpi->refresh_alt2_ref_frame = 0;
+      } else {
+        cpi->refresh_bwd_ref_frame = 0;
+        cpi->refresh_alt2_ref_frame = 1;
+      }
+      cpi->refresh_alt_ref_frame = 0;
+      break;
+
+    default: assert(0); break;
+  }
+}
+
+static void set_additional_frame_flags(const AV1_COMMON *const cm,
+                                       unsigned int *const frame_flags) {
+  if (frame_is_intra_only(cm)) *frame_flags |= FRAMEFLAGS_INTRAONLY;
+  if (frame_is_sframe(cm)) *frame_flags |= FRAMEFLAGS_SWITCH;
+  if (cm->error_resilient_mode) *frame_flags |= FRAMEFLAGS_ERROR_RESILIENT;
+}
+
+static INLINE void update_keyframe_counters(AV1_COMP *cpi) {
+  // TODO(zoeliu): To investigate whether we should treat BWDREF_FRAME
+  //               differently here for rc->avg_frame_bandwidth.
+  if (cpi->common.show_frame || cpi->rc.is_bwd_ref_frame) {
+    if (!cpi->common.show_existing_frame || cpi->rc.is_src_frame_alt_ref ||
+        cpi->common.current_frame.frame_type == KEY_FRAME) {
+      // If this is a show_existing_frame with a source other than altref,
+      // or if it is not a displayed forward keyframe, the keyframe update
+      // counters were incremented when it was originally encoded.
+      cpi->rc.frames_since_key++;
+      cpi->rc.frames_to_key--;
+    }
+  }
+}
+
+static INLINE int is_frame_droppable(const AV1_COMP *const cpi) {
+  return !(cpi->refresh_alt_ref_frame || cpi->refresh_alt2_ref_frame ||
+           cpi->refresh_bwd_ref_frame || cpi->refresh_golden_frame ||
+           cpi->refresh_last_frame);
+}
+
+static INLINE void update_frames_till_gf_update(AV1_COMP *cpi) {
+  // TODO(weitinglin): Updating this counter for is_frame_droppable
+  // is a work-around to handle the condition when a frame is drop.
+  // We should fix the cpi->common.show_frame flag
+  // instead of checking the other condition to update the counter properly.
+  if (cpi->common.show_frame || is_frame_droppable(cpi)) {
+    // Decrement count down till next gf
+    if (cpi->rc.frames_till_gf_update_due > 0)
+      cpi->rc.frames_till_gf_update_due--;
+  }
+}
+
+static INLINE void update_twopass_gf_group_index(AV1_COMP *cpi) {
+  // Increment the gf group index ready for the next frame. If this is
+  // a show_existing_frame with a source other than altref, or if it is not
+  // a displayed forward keyframe, the index was incremented when it was
+  // originally encoded.
+  if (!cpi->common.show_existing_frame || cpi->rc.is_src_frame_alt_ref ||
+      cpi->common.current_frame.frame_type == KEY_FRAME) {
+    ++cpi->twopass.gf_group.index;
+  }
+}
+
+static void update_rc_counts(AV1_COMP *cpi) {
+  update_keyframe_counters(cpi);
+  update_frames_till_gf_update(cpi);
+  if (cpi->oxcf.pass == 2) update_twopass_gf_group_index(cpi);
+}
+
+static void check_show_existing_frame(AV1_COMP *cpi) {
+  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+  AV1_COMMON *const cm = &cpi->common;
+  const FRAME_UPDATE_TYPE next_frame_update_type =
+      gf_group->update_type[gf_group->index];
+  const int which_arf = (cpi->new_bwdref_update_rule == 1)
+                            ? gf_group->arf_update_idx[gf_group->index] > 0
+                            : gf_group->arf_update_idx[gf_group->index];
+
+  if (cm->show_existing_frame == 1) {
+    cm->show_existing_frame = 0;
+  } else if (cpi->rc.is_last_bipred_frame) {
+    // NOTE: When new structure is used, every bwdref will have one overlay
+    //       frame. Therefore, there is no need to find out which frame to
+    //       show in advance.
+    if (cpi->new_bwdref_update_rule == 0) {
+      // NOTE: If the current frame is a last bi-predictive frame, it is
+      //       needed next to show the BWDREF_FRAME, which is pointed by
+      //       the last_fb_idxes[0] after reference frame buffer update
+      cpi->rc.is_last_bipred_frame = 0;
+      cm->show_existing_frame = 1;
+      cpi->existing_fb_idx_to_show = cm->remapped_ref_idx[0];
+    }
+  } else if (cpi->is_arf_filter_off[which_arf] &&
+             (next_frame_update_type == OVERLAY_UPDATE ||
+              next_frame_update_type == INTNL_OVERLAY_UPDATE)) {
+    const int bwdref_to_show =
+        (cpi->new_bwdref_update_rule == 1) ? BWDREF_FRAME : ALTREF2_FRAME;
+    // Other parameters related to OVERLAY_UPDATE will be taken care of
+    // in av1_rc_get_second_pass_params(cpi)
+    cm->show_existing_frame = 1;
+    cpi->rc.is_src_frame_alt_ref = 1;
+    cpi->existing_fb_idx_to_show =
+        (next_frame_update_type == OVERLAY_UPDATE)
+            ? get_ref_frame_map_idx(cm, ALTREF_FRAME)
+            : get_ref_frame_map_idx(cm, bwdref_to_show);
+    if (cpi->new_bwdref_update_rule == 0) {
+      cpi->is_arf_filter_off[which_arf] = 0;
+    }
+  }
+  cpi->rc.is_src_frame_ext_arf = 0;
+}
+
+static void set_ext_overrides(AV1_COMP *const cpi,
+                              EncodeFrameParams *const frame_params) {
+  // Overrides the defaults with the externally supplied values with
+  // av1_update_reference() and av1_update_entropy() calls
+  // Note: The overrides are valid only for the next frame passed
+  // to av1_encode_lowlevel()
+
+  AV1_COMMON *const cm = &cpi->common;
+
+  if (cpi->ext_use_s_frame) {
+    frame_params->frame_type = S_FRAME;
+  }
+
+  if (cpi->ext_refresh_frame_context_pending) {
+    cm->refresh_frame_context = cpi->ext_refresh_frame_context;
+    cpi->ext_refresh_frame_context_pending = 0;
+  }
+  if (cpi->ext_refresh_frame_flags_pending) {
+    cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
+    cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
+    cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
+    cpi->refresh_bwd_ref_frame = cpi->ext_refresh_bwd_ref_frame;
+    cpi->refresh_alt2_ref_frame = cpi->ext_refresh_alt2_ref_frame;
+    cpi->ext_refresh_frame_flags_pending = 0;
+  }
+  cm->allow_ref_frame_mvs = cpi->ext_use_ref_frame_mvs;
+
+  frame_params->error_resilient_mode = cpi->ext_use_error_resilient;
+  // A keyframe is already error resilient and keyframes with
+  // error_resilient_mode interferes with the use of show_existing_frame
+  // when forward reference keyframes are enabled.
+  frame_params->error_resilient_mode &= frame_params->frame_type != KEY_FRAME;
+  // For bitstream conformance, s-frames must be error-resilient
+  frame_params->error_resilient_mode |= frame_params->frame_type == S_FRAME;
+}
+
+static int get_ref_frame_flags(const AV1_COMP *const cpi) {
+  const AV1_COMMON *const cm = &cpi->common;
+
+  const RefCntBuffer *last_buf = get_ref_frame_buf(cm, LAST_FRAME);
+  const RefCntBuffer *last2_buf = get_ref_frame_buf(cm, LAST2_FRAME);
+  const RefCntBuffer *last3_buf = get_ref_frame_buf(cm, LAST3_FRAME);
+  const RefCntBuffer *golden_buf = get_ref_frame_buf(cm, GOLDEN_FRAME);
+  const RefCntBuffer *bwd_buf = get_ref_frame_buf(cm, BWDREF_FRAME);
+  const RefCntBuffer *alt2_buf = get_ref_frame_buf(cm, ALTREF2_FRAME);
+  const RefCntBuffer *alt_buf = get_ref_frame_buf(cm, ALTREF_FRAME);
+
+  // No.1 Priority: LAST_FRAME
+  const int last2_is_last = (last2_buf == last_buf);
+  const int last3_is_last = (last3_buf == last_buf);
+  const int gld_is_last = (golden_buf == last_buf);
+  const int bwd_is_last = (bwd_buf == last_buf);
+  const int alt2_is_last = (alt2_buf == last_buf);
+  const int alt_is_last = (alt_buf == last_buf);
+
+  // No.2 Priority: ALTREF_FRAME
+  const int last2_is_alt = (last2_buf == alt_buf);
+  const int last3_is_alt = (last3_buf == alt_buf);
+  const int gld_is_alt = (golden_buf == alt_buf);
+  const int bwd_is_alt = (bwd_buf == alt_buf);
+  const int alt2_is_alt = (alt2_buf == alt_buf);
+
+  // No.3 Priority: LAST2_FRAME
+  const int last3_is_last2 = (last3_buf == last2_buf);
+  const int gld_is_last2 = (golden_buf == last2_buf);
+  const int bwd_is_last2 = (bwd_buf == last2_buf);
+  const int alt2_is_last2 = (alt2_buf == last2_buf);
+
+  // No.4 Priority: LAST3_FRAME
+  const int gld_is_last3 = (golden_buf == last3_buf);
+  const int bwd_is_last3 = (bwd_buf == last3_buf);
+  const int alt2_is_last3 = (alt2_buf == last3_buf);
+
+  // No.5 Priority: GOLDEN_FRAME
+  const int bwd_is_gld = (bwd_buf == golden_buf);
+  const int alt2_is_gld = (alt2_buf == golden_buf);
+
+  // No.6 Priority: BWDREF_FRAME
+  const int alt2_is_bwd = (alt2_buf == bwd_buf);
+
+  // No.7 Priority: ALTREF2_FRAME
+
+  // cpi->ext_ref_frame_flags allows certain reference types to be disabled
+  // by the external interface.  These are set by av1_apply_encoding_flags().
+  // Start with what the external interface allows, then suppress any reference
+  // types which we have found to be duplicates.
+
+  int flags = cpi->ext_ref_frame_flags;
+
+  if (cpi->rc.frames_till_gf_update_due == INT_MAX) flags &= ~AOM_GOLD_FLAG;
+
+  if (alt_is_last) flags &= ~AOM_ALT_FLAG;
+
+  if (last2_is_last || last2_is_alt) flags &= ~AOM_LAST2_FLAG;
+
+  if (last3_is_last || last3_is_alt || last3_is_last2) flags &= ~AOM_LAST3_FLAG;
+
+  if (gld_is_last || gld_is_alt || gld_is_last2 || gld_is_last3)
+    flags &= ~AOM_GOLD_FLAG;
+
+  if ((bwd_is_last || bwd_is_alt || bwd_is_last2 || bwd_is_last3 || bwd_is_gld))
+    flags &= ~AOM_BWD_FLAG;
+
+  if ((alt2_is_last || alt2_is_alt || alt2_is_last2 || alt2_is_last3 ||
+       alt2_is_gld || alt2_is_bwd))
+    flags &= ~AOM_ALT2_FLAG;
+
+  return flags;
+}
+
+static int get_current_frame_ref_type(
+    const AV1_COMP *const cpi, const EncodeFrameParams *const frame_params) {
+  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+  // We choose the reference "type" of this frame from the flags which indicate
+  // which reference frames will be refreshed by it.  More than one of these
+  // flags may be set, so the order here implies an order of precedence.
+  // This is just used to choose the primary_ref_frame (as the most recent
+  // reference buffer of the same reference-type as the current frame)
+
+  const int intra_only = frame_params->frame_type == KEY_FRAME ||
+                         frame_params->frame_type == INTRA_ONLY_FRAME;
+  if (intra_only || frame_params->error_resilient_mode ||
+      cpi->ext_use_primary_ref_none)
+    return REGULAR_FRAME;
+  else if (gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE)
+    return EXT_ARF_FRAME;
+  else if (cpi->refresh_alt_ref_frame)
+    return ARF_FRAME;
+  else if (cpi->rc.is_src_frame_alt_ref)
+    return OVERLAY_FRAME;
+  else if (cpi->refresh_golden_frame)
+    return GLD_FRAME;
+  else if (cpi->refresh_bwd_ref_frame)
+    return BRF_FRAME;
+  else
+    return REGULAR_FRAME;
+}
+
+static int choose_primary_ref_frame(
+    const AV1_COMP *const cpi, const EncodeFrameParams *const frame_params) {
+  const AV1_COMMON *const cm = &cpi->common;
+
+  const int intra_only = frame_params->frame_type == KEY_FRAME ||
+                         frame_params->frame_type == INTRA_ONLY_FRAME;
+  if (intra_only || frame_params->error_resilient_mode ||
+      cpi->ext_use_primary_ref_none) {
+    return PRIMARY_REF_NONE;
+  }
+
+  // Find the most recent reference frame with the same reference type as the
+  // current frame
+  const FRAME_CONTEXT_INDEX current_ref_type =
+      get_current_frame_ref_type(cpi, frame_params);
+  int wanted_fb = cpi->fb_of_context_type[current_ref_type];
+
+  int primary_ref_frame = PRIMARY_REF_NONE;
+  for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
+    if (get_ref_frame_map_idx(cm, ref_frame) == wanted_fb) {
+      primary_ref_frame = ref_frame - LAST_FRAME;
+    }
+  }
+  return primary_ref_frame;
+}
+
+static void update_fb_of_context_type(
+    const AV1_COMP *const cpi, const EncodeFrameParams *const frame_params,
+    int *const fb_of_context_type) {
+  const AV1_COMMON *const cm = &cpi->common;
+
+  if (frame_is_intra_only(cm) || cm->error_resilient_mode ||
+      cpi->ext_use_primary_ref_none) {
+    for (int i = 0; i < REF_FRAMES; i++) {
+      fb_of_context_type[i] = -1;
+    }
+    fb_of_context_type[REGULAR_FRAME] =
+        cm->show_frame ? get_ref_frame_map_idx(cm, GOLDEN_FRAME)
+                       : get_ref_frame_map_idx(cm, ALTREF_FRAME);
+  }
+
+  if (!encode_show_existing_frame(cm)) {
+    // Refresh fb_of_context_type[]: see encoder.h for explanation
+    // Note that we want the value of refresh_frame_flags for the frame that
+    // just happened.  If we call get_refresh_frame_flags now we will get a
+    // different answer, because update_reference_frames() has happened.
+    if (cm->current_frame.frame_type == KEY_FRAME) {
+      // All ref frames are refreshed, pick one that will live long enough
+      fb_of_context_type[REGULAR_FRAME] = 0;
+    } else {
+      // If more than one frame is refreshed, it doesn't matter which one we
+      // pick so pick the first.  LST sometimes doesn't refresh any: this is ok
+      const int current_frame_ref_type =
+          get_current_frame_ref_type(cpi, frame_params);
+      for (int i = 0; i < REF_FRAMES; i++) {
+        if (cm->current_frame.refresh_frame_flags & (1 << i)) {
+          fb_of_context_type[current_frame_ref_type] = i;
+          break;
+        }
+      }
+    }
+  }
+}
+
+static int get_order_offset(const AV1_COMP *const cpi,
+                            const EncodeFrameParams *const frame_params) {
+  // shown frame by definition has order offset 0
+  // show_existing_frame ignores order_offset and simply takes the order_hint
+  // from the reference frame being shown.
+  if (frame_params->show_frame || cpi->common.show_existing_frame) return 0;
+
+  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+  const int arf_offset =
+      AOMMIN((MAX_GF_INTERVAL - 1), gf_group->arf_src_offset[gf_group->index]);
+  const int brf_offset = gf_group->brf_src_offset[gf_group->index];
+  return AOMMIN((MAX_GF_INTERVAL - 1), arf_offset + brf_offset);
+}
+
+static void adjust_frame_rate(AV1_COMP *cpi,
+                              const struct lookahead_entry *source) {
+  int64_t this_duration;
+  int step = 0;
+
+  // Clear down mmx registers
+  aom_clear_system_state();
+
+  if (source->ts_start == cpi->first_time_stamp_ever) {
+    this_duration = source->ts_end - source->ts_start;
+    step = 1;
+  } else {
+    int64_t last_duration =
+        cpi->last_end_time_stamp_seen - cpi->last_time_stamp_seen;
+
+    this_duration = source->ts_end - cpi->last_end_time_stamp_seen;
+
+    // do a step update if the duration changes by 10%
+    if (last_duration)
+      step = (int)((this_duration - last_duration) * 10 / last_duration);
+  }
+
+  if (this_duration) {
+    if (step) {
+      av1_new_framerate(cpi, 10000000.0 / this_duration);
+    } else {
+      // Average this frame's rate into the last second's average
+      // frame rate. If we haven't seen 1 second yet, then average
+      // over the whole interval seen.
+      const double interval = AOMMIN(
+          (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
+      double avg_duration = 10000000.0 / cpi->framerate;
+      avg_duration *= (interval - avg_duration + this_duration);
+      avg_duration /= interval;
+
+      av1_new_framerate(cpi, 10000000.0 / avg_duration);
+    }
+  }
+  cpi->last_time_stamp_seen = source->ts_start;
+  cpi->last_end_time_stamp_seen = source->ts_end;
+}
+
+static void check_src_altref(AV1_COMP *cpi,
+                             const struct lookahead_entry *source) {
+  RATE_CONTROL *const rc = &cpi->rc;
+
+  // If pass == 2, the parameters set here will be reset in
+  // av1_rc_get_second_pass_params()
+
+  if (cpi->oxcf.pass == 2) {
+    const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+    rc->is_src_frame_alt_ref =
+        (gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE) ||
+        (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
+    rc->is_src_frame_ext_arf =
+        gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE;
+  } else {
+    rc->is_src_frame_alt_ref =
+        cpi->alt_ref_source && (source == cpi->alt_ref_source);
+  }
+
+  if (rc->is_src_frame_alt_ref) {
+    // Current frame is an ARF overlay frame.
+    cpi->alt_ref_source = NULL;
+
+    if (rc->is_src_frame_ext_arf && !cpi->common.show_existing_frame) {
+      // For INTNL_OVERLAY, when show_existing_frame == 0, they do need to
+      // refresh the LAST_FRAME, i.e. LAST3 gets retired, LAST2 becomes LAST3,
+      // LAST becomes LAST2, and INTNL_OVERLAY becomes LAST.
+      cpi->refresh_last_frame = 1;
+    } else {
+      // Don't refresh the last buffer for an ARF overlay frame. It will
+      // become the GF so preserve last as an alternative prediction option.
+      cpi->refresh_last_frame = 0;
+    }
+  }
+}
+
+// Returns 0 if this is not an alt ref else the offset of the source frame
+// used as the arf midpoint.
+static int get_arf_src_index(AV1_COMP *cpi) {
+  RATE_CONTROL *const rc = &cpi->rc;
+  int arf_src_index = 0;
+  if (is_altref_enabled(cpi)) {
+    if (cpi->oxcf.pass == 2) {
+      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+      if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
+        arf_src_index = gf_group->arf_src_offset[gf_group->index];
+      }
+    } else if (rc->source_alt_ref_pending) {
+      arf_src_index = rc->frames_till_gf_update_due;
+    }
+  }
+  return arf_src_index;
+}
+
+static int get_brf_src_index(AV1_COMP *cpi) {
+  int brf_src_index = 0;
+  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+
+  // TODO(zoeliu): We need to add the check on the -bwd_ref command line setup
+  //               flag.
+  if (gf_group->bidir_pred_enabled[gf_group->index]) {
+    if (cpi->oxcf.pass == 2) {
+      if (gf_group->update_type[gf_group->index] == BRF_UPDATE)
+        brf_src_index = gf_group->brf_src_offset[gf_group->index];
+    } else {
+      // TODO(zoeliu): To re-visit the setup for this scenario
+      brf_src_index = cpi->rc.bipred_group_interval - 1;
+    }
+  }
+
+  return brf_src_index;
+}
+
+// Returns 0 if this is not an alt ref else the offset of the source frame
+// used as the arf midpoint.
+static int get_arf2_src_index(AV1_COMP *cpi) {
+  int arf2_src_index = 0;
+  if (is_altref_enabled(cpi) && cpi->num_extra_arfs) {
+    if (cpi->oxcf.pass == 2) {
+      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+      if (gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE) {
+        arf2_src_index = gf_group->arf_src_offset[gf_group->index];
+      }
+    }
+  }
+  return arf2_src_index;
+}
+
+// Called if this frame is an ARF or ARF2. Also handles forward-keyframes
+// For an ARF set arf2=0, for ARF2 set arf2=1
+// temporal_filtered is set to 1 if we temporally filter the ARF frame, so that
+// the correct post-filter buffer can be used.
+static struct lookahead_entry *setup_arf_or_arf2(
+    AV1_COMP *const cpi, const int arf_src_index, const int arf2,
+    int *temporal_filtered, EncodeFrameParams *const frame_params) {
+  AV1_COMMON *const cm = &cpi->common;
+  RATE_CONTROL *const rc = &cpi->rc;
+  const AV1EncoderConfig *const oxcf = &cpi->oxcf;
+
+  assert(arf_src_index <= rc->frames_to_key);
+  *temporal_filtered = 0;
+
+  struct lookahead_entry *source =
+      av1_lookahead_peek(cpi->lookahead, arf_src_index);
+
+  if (source != NULL) {
+    cm->showable_frame = 1;
+    cpi->alt_ref_source = source;
+
+    // When arf_src_index == rc->frames_to_key, it indicates a fwd_kf
+    if (!arf2 && arf_src_index == rc->frames_to_key) {
+      // Skip temporal filtering and mark as intra_only if we have a fwd_kf
+      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+      int which_arf = gf_group->arf_update_idx[gf_group->index];
+      cpi->is_arf_filter_off[which_arf] = 1;
+      cpi->no_show_kf = 1;
+    } else {
+      if (oxcf->arnr_max_frames > 0) {
+        // Produce the filtered ARF frame.
+        av1_temporal_filter(cpi, arf_src_index);
+        aom_extend_frame_borders(&cpi->alt_ref_buffer, av1_num_planes(cm));
+        *temporal_filtered = 1;
+      }
+    }
+    frame_params->show_frame = 0;
+
+    if (oxcf->pass < 2) {
+      // In second pass, the buffer updates configure will be set
+      // in the function av1_rc_get_second_pass_params
+      av1_configure_buffer_updates(cpi, arf2 ? INTNL_ARF_UPDATE : ARF_UPDATE);
+    }
+  }
+  rc->source_alt_ref_pending = 0;
+  return source;
+}
+
+// Determine whether there is a forced keyframe pending in the lookahead buffer
+static int is_forced_keyframe_pending(struct lookahead_ctx *lookahead,
+                                      const int up_to_index) {
+  for (int i = 0; i <= up_to_index; i++) {
+    const struct lookahead_entry *e = av1_lookahead_peek(lookahead, i);
+    if (e == NULL) {
+      // We have reached the end of the lookahead buffer and not early-returned
+      // so there isn't a forced key-frame pending.
+      return 0;
+    } else if (e->flags == AOM_EFLAG_FORCE_KF) {
+      return 1;
+    } else {
+      continue;
+    }
+  }
+  return 0;  // Never reached
+}
+
+// Check if we should encode an ARF, ARF2 or BRF.  If not, try a LAST
+// Do some setup associated with the chosen source
+// Return the frame source, or NULL if we couldn't find one
+struct lookahead_entry *choose_frame_source(
+    AV1_COMP *const cpi, int *const temporal_filtered, int *const flush,
+    struct lookahead_entry **last_source,
+    EncodeFrameParams *const frame_params) {
+  AV1_COMMON *const cm = &cpi->common;
+  struct lookahead_entry *source = NULL;
+  *temporal_filtered = 0;
+
+  // Should we encode an alt-ref frame.
+  int arf_src_index = get_arf_src_index(cpi);
+  if (arf_src_index &&
+      is_forced_keyframe_pending(cpi->lookahead, arf_src_index)) {
+    arf_src_index = 0;
+    *flush = 1;
+  }
+
+  if (arf_src_index) {
+    source = setup_arf_or_arf2(cpi, arf_src_index, 0, temporal_filtered,
+                               frame_params);
+  }
+
+  // Should we encode an arf2 frame (mutually exclusive to ARF)
+  arf_src_index = get_arf2_src_index(cpi);
+  if (arf_src_index &&
+      is_forced_keyframe_pending(cpi->lookahead, arf_src_index)) {
+    arf_src_index = 0;
+    *flush = 1;
+  }
+
+  if (arf_src_index) {
+    source = setup_arf_or_arf2(cpi, arf_src_index, 1, temporal_filtered,
+                               frame_params);
+  }
+
+  cpi->rc.is_bwd_ref_frame = 0;
+  int brf_src_index = get_brf_src_index(cpi);
+  if (brf_src_index) {
+    assert(brf_src_index <= cpi->rc.frames_to_key);
+    if ((source = av1_lookahead_peek(cpi->lookahead, brf_src_index)) != NULL) {
+      cm->showable_frame = 1;
+      frame_params->show_frame = 0;
+
+      if (cpi->oxcf.pass < 2) {
+        // In second pass, the buffer updates configure will be set
+        // in the function av1_rc_get_second_pass_params
+        av1_configure_buffer_updates(cpi, BRF_UPDATE);
+      }
+    }
+  }
+
+  if (!source) {
+    // Get last frame source.
+    if (cm->current_frame.frame_number > 0) {
+      *last_source = av1_lookahead_peek(cpi->lookahead, -1);
+    }
+    // Read in the source frame.
+    source = av1_lookahead_pop(cpi->lookahead, *flush);
+
+    if (source != NULL) {
+      frame_params->show_frame = 1;
+
+      // Check to see if the frame should be encoded as an arf overlay.
+      check_src_altref(cpi, source);
+    }
+  }
+  return source;
+}
+
+int av1_encode_strategy(AV1_COMP *const cpi, size_t *const size,
+                        uint8_t *const dest, unsigned int *frame_flags,
+                        int64_t *const time_stamp, int64_t *const time_end,
+                        const aom_rational_t *const timebase, int flush) {
+  const AV1EncoderConfig *const oxcf = &cpi->oxcf;
+  AV1_COMMON *const cm = &cpi->common;
+
+  EncodeFrameInput frame_input;
+  EncodeFrameParams frame_params;
+  EncodeFrameResults frame_results;
+  memset(&frame_input, 0, sizeof(frame_input));
+  memset(&frame_params, 0, sizeof(frame_params));
+  memset(&frame_results, 0, sizeof(frame_results));
+
+  int temporal_filtered = 0;
+  struct lookahead_entry *source = NULL;
+  struct lookahead_entry *last_source = NULL;
+  if (cm->show_existing_frame) {
+    source = av1_lookahead_pop(cpi->lookahead, flush);
+  } else {
+    source = choose_frame_source(cpi, &temporal_filtered, &flush, &last_source,
+                                 &frame_params);
+  }
+
+  if (source == NULL) {  // If no source was found, we can't encode a frame.
+    if (flush && oxcf->pass == 1 && !cpi->twopass.first_pass_done) {
+      av1_end_first_pass(cpi); /* get last stats packet */
+      cpi->twopass.first_pass_done = 1;
+    }
+    return -1;
+  }
+
+  frame_input.source = temporal_filtered ? &cpi->alt_ref_buffer : &source->img;
+  frame_input.last_source = last_source != NULL ? &last_source->img : NULL;
+  frame_input.ts_duration = source->ts_end - source->ts_start;
+
+  *time_stamp = source->ts_start;
+  *time_end = source->ts_end;
+  if (source->ts_start < cpi->first_time_stamp_ever) {
+    cpi->first_time_stamp_ever = source->ts_start;
+    cpi->last_end_time_stamp_seen = source->ts_start;
+  }
+
+  av1_apply_encoding_flags(cpi, source->flags);
+  if (!cm->show_existing_frame)
+    *frame_flags = (source->flags & AOM_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
+  cpi->frame_flags = *frame_flags;
+
+  if (frame_params.show_frame ||
+      (cm->show_existing_frame && cpi->rc.is_src_frame_alt_ref)) {
+    // Shown frames and arf-overlay frames need frame-rate considering
+    adjust_frame_rate(cpi, source);
+  }
+
+  if (cm->show_existing_frame) {
+    // show_existing_frame implies this frame is shown!
+    frame_params.show_frame = 1;
+  } else {
+    // Retain the RF_LEVEL for the current newly coded frame.
+    cm->cur_frame->frame_rf_level =
+        cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index];
+
+    if (cpi->film_grain_table) {
+      cm->seq_params.film_grain_params_present = aom_film_grain_table_lookup(
+          cpi->film_grain_table, *time_stamp, *time_end, 0 /* =erase */,
+          &cm->film_grain_params);
+    }
+    cm->cur_frame->film_grain_params_present =
+        cm->seq_params.film_grain_params_present;
+
+    // only one operating point supported now
+    const int64_t pts64 = ticks_to_timebase_units(timebase, *time_stamp);
+    if (pts64 < 0 || pts64 > UINT32_MAX) return AOM_CODEC_ERROR;
+    cpi->common.frame_presentation_time = (uint32_t)pts64;
+  }
+
+  if (oxcf->pass == 2 &&
+      (!cm->show_existing_frame || cpi->rc.is_src_frame_alt_ref)) {
+    // GF_GROUP needs updating for arf overlays as well as non-show-existing
+    av1_rc_get_second_pass_params(cpi, &frame_params);
+  }
+  if (cm->show_existing_frame && frame_params.frame_type != KEY_FRAME) {
+    // Force show-existing frames to be INTER, except forward keyframes
+    frame_params.frame_type = INTER_FRAME;
+  }
+
+  if (!cm->show_existing_frame) {
+    cm->using_qmatrix = cpi->oxcf.using_qm;
+    cm->min_qmlevel = cpi->oxcf.qm_minlevel;
+    cm->max_qmlevel = cpi->oxcf.qm_maxlevel;
+    if (cpi->twopass.gf_group.index == 1 && cpi->oxcf.enable_tpl_model) {
+      av1_set_frame_size(cpi, cm->width, cm->height);
+      av1_tpl_setup_stats(cpi, &frame_input);
+    }
+  }
+
+  frame_params.frame_flags = frame_flags;
+
+  // TODO(david.turner@argondesign.com): Move all the encode strategy
+  // (largely near av1_get_compressed_data) in here
+
+  // TODO(david.turner@argondesign.com): Change all the encode strategy to
+  // modify frame_params instead of cm or cpi.
+
+  // Per-frame encode speed.  In theory this can vary, but things may have been
+  // written assuming speed-level will not change within a sequence, so this
+  // parameter should be used with caution.
+  frame_params.speed = oxcf->speed;
+
+  // Work out some encoding parameters specific to the pass:
+  if (oxcf->pass == 0) {
+    if (cpi->oxcf.rc_mode == AOM_CBR) {
+      av1_rc_get_one_pass_cbr_params(cpi, &frame_params);
+    } else {
+      av1_rc_get_one_pass_vbr_params(cpi, &frame_params);
+    }
+  } else if (oxcf->pass == 1) {
+    av1_setup_frame_size(cpi);
+    cpi->td.mb.e_mbd.lossless[0] = is_lossless_requested(&cpi->oxcf);
+    if (!cpi->refresh_alt_ref_frame && (cm->current_frame.frame_number == 0 ||
+                                        (cpi->frame_flags & FRAMEFLAGS_KEY))) {
+      frame_params.frame_type = KEY_FRAME;
+    } else {
+      frame_params.frame_type = INTER_FRAME;
+    }
+  } else if (oxcf->pass == 2) {
+#if CONFIG_MISMATCH_DEBUG
+    mismatch_move_frame_idx_w();
+#endif
+#if TXCOEFF_COST_TIMER
+    cm->txcoeff_cost_timer = 0;
+    cm->txcoeff_cost_count = 0;
+#endif
+  }
+
+  if (oxcf->pass == 0 || oxcf->pass == 2) {
+    // Apply external override flags
+    set_ext_overrides(cpi, &frame_params);
+
+    // Work out which reference frame slots may be used.
+    frame_params.ref_frame_flags = get_ref_frame_flags(cpi);
+  }
+
+  if (oxcf->pass == 0 || oxcf->pass == 2) {
+    frame_params.primary_ref_frame =
+        choose_primary_ref_frame(cpi, &frame_params);
+    frame_params.order_offset = get_order_offset(cpi, &frame_params);
+  }
+
+  if (av1_encode(cpi, dest, &frame_input, &frame_params, &frame_results) !=
+      AOM_CODEC_OK) {
+    return AOM_CODEC_ERROR;
+  }
+
+  if (oxcf->pass == 2) {
+#if TXCOEFF_COST_TIMER
+    cm->cum_txcoeff_cost_timer += cm->txcoeff_cost_timer;
+    fprintf(stderr,
+            "\ntxb coeff cost block number: %ld, frame time: %ld, cum time %ld "
+            "in us\n",
+            cm->txcoeff_cost_count, cm->txcoeff_cost_timer,
+            cm->cum_txcoeff_cost_timer);
+#endif
+    av1_twopass_postencode_update(cpi);
+  }
+
+  if (oxcf->pass == 0 || oxcf->pass == 2) {
+    update_fb_of_context_type(cpi, &frame_params, cpi->fb_of_context_type);
+    set_additional_frame_flags(cm, frame_params.frame_flags);
+    update_rc_counts(cpi);
+    check_show_existing_frame(cpi);  // Is next frame a show_existing frame?
+  }
+
+  // Unpack frame_results:
+  *size = frame_results.size;
+
+  // Leave a signal for a higher level caller about if this frame is droppable
+  if (*size > 0) {
+    cpi->droppable = is_frame_droppable(cpi);
+  }
+
+  return AOM_CODEC_OK;
+}
diff --git a/third_party/aom/av1/encoder/encode_strategy.h b/third_party/aom/av1/encoder/encode_strategy.h
new file mode 100644
index 0000000000..1671473b76
--- /dev/null
+++ b/third_party/aom/av1/encoder/encode_strategy.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_AV1_ENCODER_ENCODE_STRATEGY_H_
+#define AOM_AV1_ENCODER_ENCODE_STRATEGY_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+
+#include "aom/aom_encoder.h"
+
+#include "av1/encoder/encoder.h"
+#include "av1/encoder/firstpass.h"
+
+// This function will implement high-level encode strategy, choosing frame type,
+// frame placement, etc.  It populates an EncodeFrameParams struct with the
+// results of these decisions and then calls av1_encode()
+int av1_encode_strategy(AV1_COMP *const cpi, size_t *const size,
+                        uint8_t *const dest, unsigned int *frame_flags,
+                        int64_t *const time_stamp, int64_t *const time_end,
+                        const aom_rational_t *const timebase, int flush);
+
+// Set individual buffer update flags based on frame reference type
+void av1_configure_buffer_updates(AV1_COMP *const cpi,
+                                  const FRAME_UPDATE_TYPE type);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // AOM_AV1_ENCODER_ENCODE_STRATEGY_H_
diff --git a/third_party/aom/av1/encoder/encodeframe.c b/third_party/aom/av1/encoder/encodeframe.c
index cb226c59e2..ea8249300c 100644
--- a/third_party/aom/av1/encoder/encodeframe.c
+++ b/third_party/aom/av1/encoder/encodeframe.c
@@ -11,6 +11,7 @@
 
 #include <limits.h>
 #include <math.h>
+#include <stdbool.h>
 #include <stdio.h>
 
 #include "config/aom_config.h"
@@ -64,6 +65,10 @@ static void encode_superblock(const AV1_COMP *const cpi, TileDataEnc *tile_data,
                               ThreadData *td, TOKENEXTRA **t, RUN_TYPE dry_run,
                               int mi_row, int mi_col, BLOCK_SIZE bsize,
                               int *rate);
+static int ml_predict_breakout(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
+                               const MACROBLOCK *const x,
+                               const RD_STATS *const rd_stats,
+                               unsigned int pb_source_variance);
 
 // This is used as a reference when computing the source variance for the
 //  purposes of activity masking.
@@ -143,6 +148,15 @@ static const uint8_t num_16x16_blocks_high_lookup[BLOCK_SIZES_ALL] = {
 };
 #endif  // CONFIG_FP_MB_STATS
 
+static const uint8_t ref_frame_flag_list[REF_FRAMES] = { 0,
+                                                         AOM_LAST_FLAG,
+                                                         AOM_LAST2_FLAG,
+                                                         AOM_LAST3_FLAG,
+                                                         AOM_GOLD_FLAG,
+                                                         AOM_BWD_FLAG,
+                                                         AOM_ALT2_FLAG,
+                                                         AOM_ALT_FLAG };
+
 unsigned int av1_get_sby_perpixel_variance(const AV1_COMP *cpi,
                                            const struct buf_2d *ref,
                                            BLOCK_SIZE bs) {
@@ -183,7 +197,8 @@ static unsigned int get_sby_perpixel_diff_variance(const AV1_COMP *const cpi,
                                                    BLOCK_SIZE bs) {
   unsigned int sse, var;
   uint8_t *last_y;
-  const YV12_BUFFER_CONFIG *last = get_ref_frame_buffer(cpi, LAST_FRAME);
+  const YV12_BUFFER_CONFIG *last =
+      get_ref_frame_yv12_buf(&cpi->common, LAST_FRAME);
 
   assert(last != NULL);
   last_y =
@@ -236,8 +251,8 @@ static void set_offsets_without_segment_id(const AV1_COMP *const cpi,
       xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK);
 
   // Set up destination pointers.
-  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                       mi_col, 0, num_planes);
+  av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0,
+                       num_planes);
 
   // Set up limit values for MV components.
   // Mv beyond the range do not produce new/different prediction block.
@@ -255,32 +270,31 @@ static void set_offsets_without_segment_id(const AV1_COMP *const cpi,
                  cm->mi_cols);
 
   // Set up source buffers.
-  av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes);
+  av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes, bsize);
 
   // R/D setup.
   x->rdmult = cpi->rd.RDMULT;
 
   // required by av1_append_sub8x8_mvs_for_idx() and av1_find_best_ref_mvs()
   xd->tile = *tile;
+
+  xd->cfl.mi_row = mi_row;
+  xd->cfl.mi_col = mi_col;
 }
 
 static void set_offsets(const AV1_COMP *const cpi, const TileInfo *const tile,
                         MACROBLOCK *const x, int mi_row, int mi_col,
                         BLOCK_SIZE bsize) {
   const AV1_COMMON *const cm = &cpi->common;
+  const struct segmentation *const seg = &cm->seg;
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *mbmi;
-  const struct segmentation *const seg = &cm->seg;
 
   set_offsets_without_segment_id(cpi, tile, x, mi_row, mi_col, bsize);
 
-  mbmi = xd->mi[0];
-  xd->cfl.mi_row = mi_row;
-  xd->cfl.mi_col = mi_col;
-
-  mbmi->segment_id = 0;
-
   // Setup segment ID.
+  mbmi = xd->mi[0];
+  mbmi->segment_id = 0;
   if (seg->enabled) {
     if (seg->enabled && !cpi->vaq_refresh) {
       const uint8_t *const map =
@@ -292,15 +306,6 @@ static void set_offsets(const AV1_COMP *const cpi, const TileInfo *const tile,
   }
 }
 
-static void reset_intmv_filter_type(MB_MODE_INFO *mbmi) {
-  InterpFilter filters[2];
-
-  for (int dir = 0; dir < 2; ++dir) {
-    filters[dir] = av1_extract_interp_filter(mbmi->interp_filters, dir);
-  }
-  mbmi->interp_filters = av1_make_interp_filters(filters[0], filters[1]);
-}
-
 static void update_filter_type_count(uint8_t allow_update_cdf,
                                      FRAME_COUNTS *counts,
                                      const MACROBLOCKD *xd,
@@ -375,8 +380,6 @@ static void update_state(const AV1_COMP *const cpi,
   *mi_addr = *mi;
   *x->mbmi_ext = ctx->mbmi_ext;
 
-  reset_intmv_filter_type(mi_addr);
-
   memcpy(x->blk_skip, ctx->blk_skip, sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
 
   x->skip = ctx->skip;
@@ -475,7 +478,8 @@ static void update_state(const AV1_COMP *const cpi,
 }
 
 void av1_setup_src_planes(MACROBLOCK *x, const YV12_BUFFER_CONFIG *src,
-                          int mi_row, int mi_col, const int num_planes) {
+                          int mi_row, int mi_col, const int num_planes,
+                          BLOCK_SIZE bsize) {
   // Set current frame pointer.
   x->e_mbd.cur_buf = src;
 
@@ -483,11 +487,10 @@ void av1_setup_src_planes(MACROBLOCK *x, const YV12_BUFFER_CONFIG *src,
   // the static analysis warnings.
   for (int i = 0; i < AOMMIN(num_planes, MAX_MB_PLANE); i++) {
     const int is_uv = i > 0;
-    setup_pred_plane(&x->plane[i].src, x->e_mbd.mi[0]->sb_type, src->buffers[i],
-                     src->crop_widths[is_uv], src->crop_heights[is_uv],
-                     src->strides[is_uv], mi_row, mi_col, NULL,
-                     x->e_mbd.plane[i].subsampling_x,
-                     x->e_mbd.plane[i].subsampling_y);
+    setup_pred_plane(
+        &x->plane[i].src, bsize, src->buffers[i], src->crop_widths[is_uv],
+        src->crop_heights[is_uv], src->strides[is_uv], mi_row, mi_col, NULL,
+        x->e_mbd.plane[i].subsampling_x, x->e_mbd.plane[i].subsampling_y);
   }
 }
 
@@ -507,6 +510,19 @@ static int set_deltaq_rdmult(const AV1_COMP *const cpi, MACROBLOCKD *const xd) {
       cpi, cm->base_qindex + xd->delta_qindex + cm->y_dc_delta_q);
 }
 
+static uint16_t edge_strength(const struct buf_2d *ref, const BLOCK_SIZE bsize,
+                              const bool high_bd, const int bd) {
+  const int width = block_size_wide[bsize];
+  const int height = block_size_high[bsize];
+  // Implementation requires width to be a multiple of 8. It also requires
+  // height to be a multiple of 4, but this is always the case.
+  assert(height % 4 == 0);
+  if (width % 8 != 0) {
+    return 0;
+  }
+  return av1_edge_exists(ref->buf, ref->stride, width, height, high_bd, bd);
+}
+
 static void rd_pick_sb_modes(AV1_COMP *const cpi, TileDataEnc *tile_data,
                              MACROBLOCK *const x, int mi_row, int mi_col,
                              RD_STATS *rd_cost, PARTITION_TYPE partition,
@@ -584,14 +600,21 @@ static void rd_pick_sb_modes(AV1_COMP *const cpi, TileDataEnc *tile_data,
     return;
   }
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     x->source_variance = av1_high_get_sby_perpixel_variance(
         cpi, &x->plane[0].src, bsize, xd->bd);
   } else {
     x->source_variance =
         av1_get_sby_perpixel_variance(cpi, &x->plane[0].src, bsize);
   }
-
+  // If the threshold for disabling wedge search is zero, it means the feature
+  // should not be used. Use a value that will always succeed in the check.
+  if (cpi->sf.disable_wedge_search_edge_thresh == 0) {
+    x->edge_strength = UINT16_MAX;
+  } else {
+    x->edge_strength =
+        edge_strength(&x->plane[0].src, bsize, is_cur_buf_hbd(xd), xd->bd);
+  }
   // Save rdmult before it might be changed, so it can be restored later.
   orig_rdmult = x->rdmult;
 
@@ -609,6 +632,8 @@ static void rd_pick_sb_modes(AV1_COMP *const cpi, TileDataEnc *tile_data,
     // If segment is boosted, use rdmult for that segment.
     if (cyclic_refresh_segment_id_boosted(mbmi->segment_id))
       x->rdmult = av1_cyclic_refresh_get_rdmult(cpi->cyclic_refresh);
+  } else if (cpi->oxcf.enable_tpl_model) {
+    x->rdmult = x->cb_rdmult;
   }
 
   if (deltaq_mode > 0) x->rdmult = set_deltaq_rdmult(cpi, xd);
@@ -622,17 +647,23 @@ static void rd_pick_sb_modes(AV1_COMP *const cpi, TileDataEnc *tile_data,
     if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
       av1_rd_pick_inter_mode_sb_seg_skip(cpi, tile_data, x, mi_row, mi_col,
                                          rd_cost, bsize, ctx, best_rd);
+#if CONFIG_ONE_PASS_SVM
+      ctx->seg_feat = 1;
+#endif
     } else {
       av1_rd_pick_inter_mode_sb(cpi, tile_data, x, mi_row, mi_col, rd_cost,
                                 bsize, ctx, best_rd);
+#if CONFIG_ONE_PASS_SVM
+      ctx->seg_feat = 0;
+#endif
     }
   }
 
   // Examine the resulting rate and for AQ mode 2 make a segment choice.
   if ((rd_cost->rate != INT_MAX) && (aq_mode == COMPLEXITY_AQ) &&
       (bsize >= BLOCK_16X16) &&
-      (cm->frame_type == KEY_FRAME || cpi->refresh_alt_ref_frame ||
-       cpi->refresh_alt2_ref_frame ||
+      (cm->current_frame.frame_type == KEY_FRAME ||
+       cpi->refresh_alt_ref_frame || cpi->refresh_alt2_ref_frame ||
        (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref))) {
     av1_caq_select_segment(cpi, x, bsize, mi_row, mi_col, rd_cost->rate);
   }
@@ -865,6 +896,7 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
   MACROBLOCKD *const xd = &x->e_mbd;
   const MB_MODE_INFO *const mbmi = xd->mi[0];
   const MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
+  const CurrentFrame *const current_frame = &cm->current_frame;
   const BLOCK_SIZE bsize = mbmi->sb_type;
   FRAME_CONTEXT *fc = xd->tile_ctx;
   const uint8_t allow_update_cdf = tile_data->allow_update_cdf;
@@ -877,7 +909,8 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
   const int seg_ref_active =
       segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_REF_FRAME);
 
-  if (cm->skip_mode_flag && !seg_ref_active && is_comp_ref_allowed(bsize)) {
+  if (current_frame->skip_mode_info.skip_mode_flag && !seg_ref_active &&
+      is_comp_ref_allowed(bsize)) {
     const int skip_mode_ctx = av1_get_skip_mode_context(xd);
 #if CONFIG_ENTROPY_STATS
     td->counts->skip_mode[skip_mode_ctx][mbmi->skip_mode]++;
@@ -896,12 +929,13 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
     }
   }
 
-  if (cm->delta_q_present_flag &&
+  const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
+  if (delta_q_info->delta_q_present_flag &&
       (bsize != cm->seq_params.sb_size || !mbmi->skip) &&
       super_block_upper_left) {
 #if CONFIG_ENTROPY_STATS
     const int dq =
-        (mbmi->current_qindex - xd->current_qindex) / cm->delta_q_res;
+        (mbmi->current_qindex - xd->current_qindex) / delta_q_info->delta_q_res;
     const int absdq = abs(dq);
     for (int i = 0; i < AOMMIN(absdq, DELTA_Q_SMALL); ++i) {
       td->counts->delta_q[i][1]++;
@@ -909,14 +943,14 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
     if (absdq < DELTA_Q_SMALL) td->counts->delta_q[absdq][0]++;
 #endif
     xd->current_qindex = mbmi->current_qindex;
-    if (cm->delta_lf_present_flag) {
-      if (cm->delta_lf_multi) {
+    if (delta_q_info->delta_lf_present_flag) {
+      if (delta_q_info->delta_lf_multi) {
         const int frame_lf_count =
             av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
         for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
 #if CONFIG_ENTROPY_STATS
-          const int delta_lf =
-              (mbmi->delta_lf[lf_id] - xd->delta_lf[lf_id]) / cm->delta_lf_res;
+          const int delta_lf = (mbmi->delta_lf[lf_id] - xd->delta_lf[lf_id]) /
+                               delta_q_info->delta_lf_res;
           const int abs_delta_lf = abs(delta_lf);
           for (int i = 0; i < AOMMIN(abs_delta_lf, DELTA_LF_SMALL); ++i) {
             td->counts->delta_lf_multi[lf_id][i][1]++;
@@ -930,7 +964,7 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
 #if CONFIG_ENTROPY_STATS
         const int delta_lf =
             (mbmi->delta_lf_from_base - xd->delta_lf_from_base) /
-            cm->delta_lf_res;
+            delta_q_info->delta_lf_res;
         const int abs_delta_lf = abs(delta_lf);
         for (int i = 0; i < AOMMIN(abs_delta_lf, DELTA_LF_SMALL); ++i) {
           td->counts->delta_lf[i][1]++;
@@ -964,7 +998,7 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
 
     if (mbmi->skip_mode) {
       rdc->skip_mode_used_flag = 1;
-      if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+      if (current_frame->reference_mode == REFERENCE_MODE_SELECT) {
         assert(has_second_ref(mbmi));
         rdc->compound_ref_used_flag = 1;
       }
@@ -991,7 +1025,7 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
 
         av1_collect_neighbors_ref_counts(xd);
 
-        if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+        if (current_frame->reference_mode == REFERENCE_MODE_SELECT) {
           if (has_second_ref(mbmi))
             // This flag is also updated for 4x4 blocks
             rdc->compound_ref_used_flag = 1;
@@ -1219,7 +1253,7 @@ static void update_stats(const AV1_COMMON *const cm, TileDataEnc *tile_data,
         }
 
         if (has_second_ref(mbmi)) {
-          assert(cm->reference_mode != SINGLE_REFERENCE &&
+          assert(current_frame->reference_mode != SINGLE_REFERENCE &&
                  is_inter_compound_mode(mbmi->mode) &&
                  mbmi->motion_mode == SIMPLE_TRANSLATION);
 
@@ -1427,18 +1461,20 @@ static void encode_b(const AV1_COMP *const cpi, TileDataEnc *tile_data,
   MB_MODE_INFO *mbmi = xd->mi[0];
   mbmi->partition = partition;
   update_state(cpi, tile_data, td, ctx, mi_row, mi_col, bsize, dry_run);
+  if (cpi->oxcf.enable_tpl_model && cpi->oxcf.aq_mode == NO_AQ &&
+      cpi->oxcf.deltaq_mode == 0) {
+    x->rdmult = x->cb_rdmult;
+  }
 
   if (!dry_run) av1_set_coeff_buffer(cpi, x, mi_row, mi_col);
 
   encode_superblock(cpi, tile_data, td, tp, dry_run, mi_row, mi_col, bsize,
                     rate);
 
-  if (dry_run == 0)
-    x->cb_offset += block_size_wide[bsize] * block_size_high[bsize];
-
   if (!dry_run) {
+    x->cb_offset += block_size_wide[bsize] * block_size_high[bsize];
     if (bsize == cpi->common.seq_params.sb_size && mbmi->skip == 1 &&
-        cpi->common.delta_lf_present_flag) {
+        cpi->common.delta_q_info.delta_lf_present_flag) {
       const int frame_lf_count = av1_num_planes(&cpi->common) > 1
                                      ? FRAME_LF_COUNT
                                      : FRAME_LF_COUNT - 2;
@@ -2057,6 +2093,102 @@ static int active_edge_sb(const AV1_COMP *cpi, int mi_row, int mi_col) {
          active_v_edge(cpi, mi_col, cpi->common.seq_params.mib_size);
 }
 
+// Performs a motion search in SIMPLE_TRANSLATION mode using reference frame
+// ref. Note that this sets the offset of mbmi, so we will need to reset it
+// after calling this function.
+static void simple_motion_search(AV1_COMP *const cpi, MACROBLOCK *x, int mi_row,
+                                 int mi_col, BLOCK_SIZE bsize, int ref,
+                                 MV ref_mv_full, int num_planes,
+                                 int use_subpixel) {
+  assert(num_planes == 1 &&
+         "Currently simple_motion_search only supports luma plane");
+  AV1_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *xd = &x->e_mbd;
+
+  set_offsets(cpi, &xd->tile, x, mi_row, mi_col, bsize);
+
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  mbmi->sb_type = bsize;
+  mbmi->ref_frame[0] = ref;
+  mbmi->ref_frame[1] = NONE_FRAME;
+  mbmi->motion_mode = SIMPLE_TRANSLATION;
+
+  const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_yv12_buf(cm, ref);
+  const YV12_BUFFER_CONFIG *scaled_ref_frame =
+      av1_get_scaled_ref_frame(cpi, ref);
+  struct buf_2d backup_yv12;
+  // ref_mv is used to code the motion vector. ref_mv_full is the initial point.
+  // ref_mv is in units of 1/8 pel whereas ref_mv_full is in units of pel.
+  MV ref_mv = { 0, 0 };
+  const int step_param = cpi->mv_step_param;
+  const MvLimits tmp_mv_limits = x->mv_limits;
+  const SEARCH_METHODS search_methods = NSTEP;
+  const int do_mesh_search = 0;
+  const int sadpb = x->sadperbit16;
+  int cost_list[5];
+  const int ref_idx = 0;
+  int var;
+
+  if (scaled_ref_frame) {
+    backup_yv12 = xd->plane[AOM_PLANE_Y].pre[ref_idx];
+    av1_setup_pre_planes(xd, ref_idx, scaled_ref_frame, mi_row, mi_col, NULL,
+                         num_planes);
+  } else {
+    av1_setup_pre_planes(xd, ref_idx, yv12, mi_row, mi_col,
+                         get_ref_scale_factors(cm, ref), num_planes);
+  }
+
+  // This overwrites the mv_limits so we will need to restore it later.
+  av1_set_mv_search_range(&x->mv_limits, &ref_mv);
+  var = av1_full_pixel_search(cpi, x, bsize, &ref_mv_full, step_param,
+                              search_methods, do_mesh_search, sadpb,
+                              cond_cost_list(cpi, cost_list), &ref_mv, INT_MAX,
+                              1, mi_col * MI_SIZE, mi_row * MI_SIZE, 0);
+  // Restore
+  x->mv_limits = tmp_mv_limits;
+
+  const int use_subpel_search =
+      var < INT_MAX && !cpi->common.cur_frame_force_integer_mv && use_subpixel;
+  if (use_subpel_search) {
+    int not_used = 0;
+    if (cpi->sf.use_accurate_subpel_search) {
+      const int pw = block_size_wide[bsize];
+      const int ph = block_size_high[bsize];
+      cpi->find_fractional_mv_step(
+          x, cm, mi_row, mi_col, &ref_mv, cm->allow_high_precision_mv,
+          x->errorperbit, &cpi->fn_ptr[bsize], cpi->sf.mv.subpel_force_stop,
+          cpi->sf.mv.subpel_iters_per_step, cond_cost_list(cpi, cost_list),
+          x->nmv_vec_cost, x->mv_cost_stack, &not_used, &x->pred_sse[ref], NULL,
+          NULL, 0, 0, pw, ph, cpi->sf.use_accurate_subpel_search, 1);
+    } else {
+      cpi->find_fractional_mv_step(
+          x, cm, mi_row, mi_col, &ref_mv, cm->allow_high_precision_mv,
+          x->errorperbit, &cpi->fn_ptr[bsize], cpi->sf.mv.subpel_force_stop,
+          cpi->sf.mv.subpel_iters_per_step, cond_cost_list(cpi, cost_list),
+          x->nmv_vec_cost, x->mv_cost_stack, &not_used, &x->pred_sse[ref], NULL,
+          NULL, 0, 0, 0, 0, 0, 1);
+    }
+  } else {
+    // Manually convert from units of pixel to 1/8-pixels if we are not doing
+    // subpel search
+    x->best_mv.as_mv.row *= 8;
+    x->best_mv.as_mv.col *= 8;
+  }
+
+  mbmi->mv[0].as_mv = x->best_mv.as_mv;
+
+  // Get a copy of the prediction output
+  set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
+  av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize,
+                                AOM_PLANE_Y, AOM_PLANE_Y);
+
+  aom_clear_system_state();
+
+  if (scaled_ref_frame) {
+    xd->plane[AOM_PLANE_Y].pre[ref_idx] = backup_yv12;
+  }
+}
+
 // Look at neighboring blocks and set a min and max partition size based on
 // what they chose.
 static void rd_auto_partition_range(AV1_COMP *cpi, const TileInfo *const tile,
@@ -2074,7 +2206,8 @@ static void rd_auto_partition_range(AV1_COMP *cpi, const TileInfo *const tile,
   BLOCK_SIZE max_size = BLOCK_LARGEST;
 
   // Trap case where we do not have a prediction.
-  if (left_in_image || above_in_image || cm->frame_type != KEY_FRAME) {
+  if (left_in_image || above_in_image ||
+      cm->current_frame.frame_type != KEY_FRAME) {
     // Default "min to max" and "max to min"
     min_size = BLOCK_LARGEST;
     max_size = BLOCK_4X4;
@@ -2082,7 +2215,7 @@ static void rd_auto_partition_range(AV1_COMP *cpi, const TileInfo *const tile,
     // NOTE: each call to get_sb_partition_size_range() uses the previous
     // passed in values for min and max as a starting point.
     // Find the min and max partition used in previous frame at this location
-    if (cm->frame_type != KEY_FRAME) {
+    if (cm->current_frame.frame_type != KEY_FRAME) {
       MB_MODE_INFO **prev_mi =
           &cm->prev_mi_grid_visible[mi_row * xd->mi_stride + mi_col];
       get_sb_partition_size_range(cm, xd, prev_mi, &min_size, &max_size);
@@ -2210,14 +2343,14 @@ const int complexity_16x16_blocks_threshold[BLOCK_SIZES] = {
   8, 8, 10
 };
 
-typedef enum {
+enum {
   MV_ZERO = 0,
   MV_LEFT = 1,
   MV_UP = 2,
   MV_RIGHT = 3,
   MV_DOWN = 4,
   MV_INVALID
-} MOTION_DIRECTION;
+} UENUM1BYTE(MOTION_DIRECTION);
 
 static INLINE MOTION_DIRECTION get_motion_direction_fp(uint8_t fp_byte) {
   if (fp_byte & FPMB_MOTION_ZERO_MASK) {
@@ -2346,6 +2479,17 @@ static void reset_partition(PC_TREE *pc_tree, BLOCK_SIZE bsize) {
   pc_tree->cb_search_range = SEARCH_FULL_PLANE;
   pc_tree->none.skip = 0;
 
+  pc_tree->pc_tree_stats.valid = 0;
+  pc_tree->pc_tree_stats.split = 0;
+  pc_tree->pc_tree_stats.skip = 0;
+  pc_tree->pc_tree_stats.rdcost = INT64_MAX;
+
+  for (int i = 0; i < 4; i++) {
+    pc_tree->pc_tree_stats.sub_block_split[i] = 0;
+    pc_tree->pc_tree_stats.sub_block_skip[i] = 0;
+    pc_tree->pc_tree_stats.sub_block_rdcost[i] = INT64_MAX;
+  }
+
   if (bsize >= BLOCK_8X8) {
     BLOCK_SIZE subsize = get_partition_subsize(bsize, PARTITION_SPLIT);
     for (int idx = 0; idx < 4; ++idx)
@@ -2467,9 +2611,9 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
                     ? partition_cost[PARTITION_NONE]
                     : 0;
     }
-    int64_t partition_rd_cost = RDCOST(x->rdmult, pt_cost, 0);
-    int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
-                                     ? INT64_MAX
+    const int64_t partition_rd_cost = RDCOST(x->rdmult, pt_cost, 0);
+    const int64_t best_remain_rdcost =
+        best_rdc.rdcost == INT64_MAX ? INT64_MAX
                                      : (best_rdc.rdcost - partition_rd_cost);
     rd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &this_rdc,
                      PARTITION_NONE, bsize, ctx_none, best_remain_rdcost);
@@ -2499,15 +2643,30 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
 
         pc_tree->cb_search_range = SEARCH_FULL_PLANE;
 
-        // If all y, u, v transform blocks in this partition are skippable, and
-        // the dist & rate are within the thresholds, the partition search is
-        // terminated for current branch of the partition search tree.
-        // The dist & rate thresholds are set to 0 at speed 0 to disable the
-        // early termination at that speed.
-        if (!x->e_mbd.lossless[xd->mi[0]->segment_id] &&
-            (ctx_none->skippable && best_rdc.dist < dist_breakout_thr &&
-             best_rdc.rate < rate_breakout_thr)) {
-          do_square_split = 0;
+        if (!x->e_mbd.lossless[xd->mi[0]->segment_id] && ctx_none->skippable) {
+          const int use_ml_based_breakout =
+              bsize <= cpi->sf.use_square_partition_only_threshold &&
+              bsize > BLOCK_4X4 && xd->bd == 8;
+
+          // TODO(anyone): Currently this is using the same model and threshold
+          // values as in rd_pick_partition. Retraining the model and tuning the
+          // threshold values might be helpful to improve the speed.
+          if (use_ml_based_breakout) {
+            if (ml_predict_breakout(cpi, bsize, x, &this_rdc,
+                                    x->source_variance)) {
+              do_square_split = 0;
+            }
+          }
+
+          // If all y, u, v transform blocks in this partition are skippable,
+          // and the dist & rate are within the thresholds, the partition search
+          // is terminated for current branch of the partition search tree. The
+          // dist & rate thresholds are set to 0 at speed 0 to disable the early
+          // termination at that speed.
+          if (best_rdc.dist < dist_breakout_thr &&
+              best_rdc.rate < rate_breakout_thr) {
+            do_square_split = 0;
+          }
         }
       }
     }
@@ -2527,6 +2686,9 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
     subsize = get_partition_subsize(bsize, PARTITION_SPLIT);
     int idx;
 
+    sum_rdc.rate = partition_cost[PARTITION_SPLIT];
+    sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
+
     for (idx = 0; idx < 4 && sum_rdc.rdcost < temp_best_rdcost; ++idx) {
       const int x_idx = (idx & 1) * mi_step;
       const int y_idx = (idx >> 1) * mi_step;
@@ -2538,12 +2700,12 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
 
       pc_tree->split[idx]->index = idx;
       int64_t *p_split_rd = &split_rd[idx];
-      // TODO(Cherma) : Account for partition cost while passing best rd to
-      // rd_pick_sqr_partition()
-      rd_pick_sqr_partition(cpi, td, tile_data, tp, mi_row + y_idx,
-                            mi_col + x_idx, subsize, &this_rdc,
-                            temp_best_rdcost - sum_rdc.rdcost,
-                            pc_tree->split[idx], p_split_rd);
+      const int64_t best_remain_rdcost =
+          (temp_best_rdcost == INT64_MAX) ? INT64_MAX
+                                          : (temp_best_rdcost - sum_rdc.rdcost);
+      rd_pick_sqr_partition(
+          cpi, td, tile_data, tp, mi_row + y_idx, mi_col + x_idx, subsize,
+          &this_rdc, best_remain_rdcost, pc_tree->split[idx], p_split_rd);
 
       pc_tree->pc_tree_stats.sub_block_rdcost[idx] = this_rdc.rdcost;
       pc_tree->pc_tree_stats.sub_block_skip[idx] =
@@ -2561,7 +2723,6 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
     reached_last_index = (idx == 4);
 
     if (reached_last_index && sum_rdc.rdcost < best_rdc.rdcost) {
-      sum_rdc.rate += partition_cost[PARTITION_SPLIT];
       sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, sum_rdc.dist);
 
       if (sum_rdc.rdcost < best_rdc.rdcost) {
@@ -2626,77 +2787,6 @@ static void rd_pick_sqr_partition(AV1_COMP *const cpi, ThreadData *td,
   }
 }
 
-#define FEATURE_SIZE 19
-static const float two_pass_split_partition_weights_128[FEATURE_SIZE + 1] = {
-  2.683936f, -0.193620f, -4.106470f, -0.141320f, -0.282289f,
-  0.125296f, -1.134961f, 0.862757f,  -0.418799f, -0.637666f,
-  0.016232f, 0.345013f,  0.018823f,  -0.393394f, -1.130700f,
-  0.695357f, 0.112569f,  -0.341975f, -0.513882f, 5.7488966f,
-};
-
-static const float two_pass_split_partition_weights_64[FEATURE_SIZE + 1] = {
-  2.990993f,  0.423273f,  -0.926544f, 0.454646f,  -0.292698f,
-  -1.311632f, -0.284432f, 0.717141f,  -0.419257f, -0.574760f,
-  -0.674444f, 0.669047f,  -0.374255f, 0.380624f,  -0.804036f,
-  0.264021f,  0.004163f,  1.896802f,  0.924287f,  0.13490619f,
-};
-
-static const float two_pass_split_partition_weights_32[FEATURE_SIZE + 1] = {
-  2.795181f,  -0.136943f, -0.924842f, 0.405330f,  -0.463505f,
-  -0.584076f, -0.831472f, 0.382985f,  -0.597544f, -0.138915f,
-  -1.354350f, 0.466035f,  -0.553961f, 0.213202f,  -1.166429f,
-  0.010776f,  -0.096236f, 2.335084f,  1.699857f,  -0.58178353f,
-};
-
-static const float two_pass_split_partition_weights_16[FEATURE_SIZE + 1] = {
-  1.987888f,  -0.431100f, -1.687703f, 0.262602f,  -0.425298f,
-  -0.463870f, -1.493457f, 0.470917f,  -0.528457f, -0.087700f,
-  -1.815092f, 0.152883f,  -0.337908f, 0.093679f,  -1.548267f,
-  -0.042387f, -0.000861f, 2.556746f,  1.619192f,  0.03643292f,
-};
-
-static const float two_pass_split_partition_weights_8[FEATURE_SIZE + 1] = {
-  2.188344f,  -0.817528f, -2.119219f, 0.000000f,  -0.348167f,
-  -0.658074f, -1.960362f, 0.000000f,  -0.403080f, 0.282699f,
-  -2.061088f, 0.000000f,  -0.431919f, -0.127960f, -1.099550f,
-  0.000000f,  0.121622f,  2.017455f,  2.058228f,  -0.15475988f,
-};
-
-static const float two_pass_none_partition_weights_128[FEATURE_SIZE + 1] = {
-  -1.006689f, 0.777908f,  4.461072f,  -0.395782f, -0.014610f,
-  -0.853863f, 0.729997f,  -0.420477f, 0.282429f,  -1.194595f,
-  3.181220f,  -0.511416f, 0.117084f,  -1.149348f, 1.507990f,
-  -0.477212f, 0.202963f,  -1.469581f, 0.624461f,  -0.89081228f,
-};
-
-static const float two_pass_none_partition_weights_64[FEATURE_SIZE + 1] = {
-  -1.241117f, 0.844878f,  5.638803f,  -0.489780f, -0.108796f,
-  -4.576821f, 1.540624f,  -0.477519f, 0.227791f,  -1.443968f,
-  1.586911f,  -0.505125f, 0.140764f,  -0.464194f, 1.466658f,
-  -0.641166f, 0.195412f,  1.427905f,  2.080007f,  -1.98272777f,
-};
-
-static const float two_pass_none_partition_weights_32[FEATURE_SIZE + 1] = {
-  -2.130825f, 0.476023f,  5.907343f,  -0.516002f, -0.097471f,
-  -2.662754f, 0.614858f,  -0.576728f, 0.085261f,  -0.031901f,
-  0.727842f,  -0.600034f, 0.079326f,  0.324328f,  0.504502f,
-  -0.547105f, -0.037670f, 0.304995f,  0.369018f,  -2.66299987f,
-};
-
-static const float two_pass_none_partition_weights_16[FEATURE_SIZE + 1] = {
-  -1.626410f, 0.872047f,  5.414965f,  -0.554781f, -0.084514f,
-  -3.020550f, 0.467632f,  -0.382280f, 0.199568f,  0.426220f,
-  0.829426f,  -0.467100f, 0.153098f,  0.662994f,  0.327545f,
-  -0.560106f, -0.141610f, 0.403372f,  0.523991f,  -3.02891231f,
-};
-
-static const float two_pass_none_partition_weights_8[FEATURE_SIZE + 1] = {
-  -1.463349f, 0.375376f,  4.751430f, 0.000000f, -0.184451f,
-  -1.655447f, 0.443214f,  0.000000f, 0.127961f, 0.152435f,
-  0.083288f,  0.000000f,  0.143105f, 0.438012f, 0.073238f,
-  0.000000f,  -0.278137f, 0.186134f, 0.073737f, -1.6494962f,
-};
-
 // split_score indicates confidence of picking split partition;
 // none_score indicates confidence of picking none partition;
 static int ml_prune_2pass_split_partition(const PC_TREE_STATS *pc_tree_stats,
@@ -2819,7 +2909,7 @@ static void ml_prune_rect_partition(const AV1_COMP *const cpi,
   // Variance ratios
   const MACROBLOCKD *const xd = &x->e_mbd;
   int whole_block_variance;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     whole_block_variance = av1_high_get_sby_perpixel_variance(
         cpi, &x->plane[0].src, bsize, xd->bd);
   } else {
@@ -2837,7 +2927,7 @@ static void ml_prune_rect_partition(const AV1_COMP *const cpi,
     const int x_idx = (i & 1) * bw / 2;
     const int y_idx = (i >> 1) * bw / 2;
     buf.buf = x->plane[0].src.buf + x_idx + y_idx * buf.stride;
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       split_variance[i] =
           av1_high_get_sby_perpixel_variance(cpi, &buf, subsize, xd->bd);
     } else {
@@ -2851,6 +2941,7 @@ static void ml_prune_rect_partition(const AV1_COMP *const cpi,
   // 2. Do the prediction and prune 0-2 partitions based on their probabilities
   float raw_scores[3] = { 0.0f };
   av1_nn_predict(features, nn_config, raw_scores);
+  aom_clear_system_state();
   float probs[3] = { 0.0f };
   av1_nn_softmax(raw_scores, probs, 3);
 
@@ -2918,6 +3009,7 @@ static void ml_prune_ab_partition(BLOCK_SIZE bsize, int part_ctx, int var_ctx,
   // Calculate scores using the NN model.
   float score[16] = { 0.0f };
   av1_nn_predict(features, nn_config, score);
+  aom_clear_system_state();
   int int_score[16];
   int max_score = -1000;
   for (int i = 0; i < 16; ++i) {
@@ -3008,7 +3100,7 @@ static void ml_prune_4_partition(const AV1_COMP *const cpi, MACROBLOCK *const x,
     BLOCK_SIZE horz_4_bs = get_partition_subsize(bsize, PARTITION_HORZ_4);
     BLOCK_SIZE vert_4_bs = get_partition_subsize(bsize, PARTITION_VERT_4);
     av1_setup_src_planes(x, cpi->source, mi_row, mi_col,
-                         av1_num_planes(&cpi->common));
+                         av1_num_planes(&cpi->common), bsize);
     const int src_stride = x->plane[0].src.stride;
     const uint8_t *src = x->plane[0].src.buf;
     const MACROBLOCKD *const xd = &x->e_mbd;
@@ -3017,7 +3109,7 @@ static void ml_prune_4_partition(const AV1_COMP *const cpi, MACROBLOCK *const x,
           src + i * block_size_high[horz_4_bs] * src_stride;
       const uint8_t *vert_src = src + i * block_size_wide[vert_4_bs];
       unsigned int horz_var, vert_var, sse;
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (is_cur_buf_hbd(xd)) {
         switch (xd->bd) {
           case 10:
             horz_var = cpi->fn_ptr[horz_4_bs].vf(
@@ -3084,6 +3176,7 @@ static void ml_prune_4_partition(const AV1_COMP *const cpi, MACROBLOCK *const x,
   // Calculate scores using the NN model.
   float score[LABELS] = { 0.0f };
   av1_nn_predict(features, nn_config, score);
+  aom_clear_system_state();
   int int_score[LABELS];
   int max_score = -1000;
   for (int i = 0; i < LABELS; ++i) {
@@ -3168,18 +3261,495 @@ static int ml_predict_breakout(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
   // Calculate score using the NN model.
   float score = 0.0f;
   av1_nn_predict(features, nn_config, &score);
+  aom_clear_system_state();
 
   // Make decision.
   return (int)(score * 100) >= thresh;
 }
 #undef FEATURES
 
-// TODO(jingning,jimbankoski,rbultje): properly skip partition types that are
+#if CONFIG_ONE_PASS_SVM
+#define FEATURES 24
+static void ml_op_svm_early_term(const AV1_COMP *const cpi,
+                                 const MACROBLOCK *const x,
+                                 const MACROBLOCKD *const xd,
+                                 const PICK_MODE_CONTEXT *ctx_none,
+                                 const RD_STATS *none_rdc, int pb_source_var,
+                                 BLOCK_SIZE bsize, float *const score) {
+  const float *ml_weights = NULL, *ml_mean = NULL, *ml_std = NULL;
+  if (bsize == BLOCK_128X128) {
+    ml_weights = av1_op_svm_early_term_weights_128;
+    ml_mean = av1_op_svm_early_term_mean_128;
+    ml_std = av1_op_svm_early_term_std_128;
+  } else if (bsize == BLOCK_64X64) {
+    ml_weights = av1_op_svm_early_term_weights_64;
+    ml_mean = av1_op_svm_early_term_mean_64;
+    ml_std = av1_op_svm_early_term_std_64;
+  } else if (bsize == BLOCK_32X32) {
+    ml_weights = av1_op_svm_early_term_weights_32;
+    ml_mean = av1_op_svm_early_term_mean_32;
+    ml_std = av1_op_svm_early_term_std_32;
+  } else if (bsize == BLOCK_16X16) {
+    ml_weights = av1_op_svm_early_term_weights_16;
+    ml_mean = av1_op_svm_early_term_mean_16;
+    ml_std = av1_op_svm_early_term_std_16;
+  } else {
+    assert(bsize == BLOCK_128X128 || bsize == BLOCK_64X64 ||
+           bsize == BLOCK_32X32 || bsize == BLOCK_8X8);
+  }
+  if (ml_weights != NULL) {
+    // Compute some features
+
+    float features[FEATURES] = { 0 };
+    int f_idx = 0;
+    int r_idx = 0;
+
+    // None features
+    // Get none stats
+    features[f_idx++] = none_rdc->rate;
+    features[f_idx++] = none_rdc->dist;
+    features[f_idx++] = none_rdc->rdcost;
+    features[f_idx++] = ctx_none->skip;
+
+    // EOBS
+    features[f_idx++] = none_rdc->eob;
+    int scaled_eob = none_rdc->eob * 32 * 32;
+    features[f_idx++] = (1.0f + none_rdc->eob_0) / (4.0f + scaled_eob);
+    features[f_idx++] = (1.0f + none_rdc->eob_1) / (4.0f + scaled_eob);
+    features[f_idx++] = (1.0f + none_rdc->eob_2) / (4.0f + scaled_eob);
+    features[f_idx++] = (1.0f + none_rdc->eob_3) / (4.0f + scaled_eob);
+
+    // Y_RD
+    features[f_idx++] = none_rdc->rd;
+    int64_t scaled_rd = none_rdc->rd * 32 * 32;
+    features[f_idx++] = (1.0f + none_rdc->rd_0) / (4.0f + scaled_rd);
+    features[f_idx++] = (1.0f + none_rdc->rd_1) / (4.0f + scaled_rd);
+    features[f_idx++] = (1.0f + none_rdc->rd_2) / (4.0f + scaled_rd);
+    features[f_idx++] = (1.0f + none_rdc->rd_3) / (4.0f + scaled_rd);
+
+    // Q_SQUARED
+    features[f_idx++] =
+        (x->plane[0].dequant_QTX[0]) * (x->plane[0].dequant_QTX[0]);
+
+    // SIZE
+    // Get size of surrounding blocks
+    int above_size = 18, left_size = 18;
+    const MB_MODE_INFO *above_block = xd->above_mbmi;
+    const MB_MODE_INFO *left_block = xd->left_mbmi;
+
+    if (above_block) {
+      above_size = above_block->sb_type;
+    }
+    if (left_block) {
+      left_size = left_block->sb_type;
+    }
+
+    features[f_idx++] = left_size;
+    features[f_idx++] = left_size != 18;
+
+    features[f_idx++] = above_size;
+    features[f_idx++] = above_size != 18;
+
+    // Variance
+    // Get variance
+    int var = pb_source_var, var_reg[4] = { 0 };
+    const int bw = block_size_wide[bsize];
+    const int bh = block_size_high[bsize];
+    const BLOCK_SIZE split_size = get_partition_subsize(bsize, PARTITION_SPLIT);
+    struct buf_2d buf;
+    buf.stride = x->plane[0].src.stride;
+    for (int i = 0; i < 4; ++i) {
+      const int x_idx = (i & 1) * bw / 2;
+      const int y_idx = (i >> 1) * bh / 2;
+      buf.buf = x->plane[0].src.buf + x_idx + y_idx * buf.stride;
+      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+        var_reg[i] =
+            av1_high_get_sby_perpixel_variance(cpi, &buf, split_size, xd->bd);
+      } else {
+        var_reg[i] = av1_get_sby_perpixel_variance(cpi, &buf, split_size);
+      }
+    }
+
+    features[f_idx++] = var;
+    for (r_idx = 0; r_idx < 4; r_idx++) {
+      features[f_idx] = (var_reg[r_idx] + 1.0f) / (var + 4.0f);
+      f_idx++;
+    }
+
+    assert(f_idx == FEATURES);
+
+    // Calculate the score
+    *score = 0.0f;
+    for (f_idx = 0; f_idx < FEATURES; f_idx++) {
+      *score += ml_weights[f_idx] * (features[f_idx] - ml_mean[f_idx]) /
+                ml_std[f_idx];
+    }
+    // Dont forget the bias
+    *score += ml_weights[FEATURES];
+  }
+}
+#undef FEATURES
+#endif
+
+// Performs a full_pixel_motion_search with a single reference frame and extract
+// the variance of residues. Here features is assumed to be a length 6 array.
+// After this function is called, we will store the following in to features:
+// features[0] = log(1 + dc_q**2/256)
+// features[1] = log(1 + variance_of_residue)
+// for i in [2, 3, 4, 5]:
+//  features[i] = log(1 + variance_of_residue_in_block[i]/variance_of_residue)
+static void get_res_var_features(AV1_COMP *const cpi, MACROBLOCK *x, int mi_row,
+                                 int mi_col, BLOCK_SIZE bsize,
+                                 float *features) {
+  // TODO(chiyotsai@google.com): The data this model trained on did not also use
+  // SIMPLE_TRANSLATION to build the inter_predictor. Retraining and tuning the
+  // model with the correct data should give better performance.
+  assert(mi_size_wide[bsize] == mi_size_high[bsize]);
+
+  MACROBLOCKD *xd = &x->e_mbd;
+
+  // Perform a single motion search in Y_PLANE to make a prediction
+  const MV_REFERENCE_FRAME ref =
+      cpi->rc.is_src_frame_alt_ref ? ALTREF_FRAME : LAST_FRAME;
+  const int use_subpixel = 0;
+  const int num_planes = 1;
+
+  const MV ref_mv_full = { .row = 0, .col = 0 };
+  simple_motion_search(cpi, x, mi_row, mi_col, bsize, ref, ref_mv_full,
+                       num_planes, use_subpixel);
+  aom_clear_system_state();
+
+  // Start getting the features
+  int f_idx = 0;
+
+  // Q_INDEX
+  const int dc_q = av1_dc_quant_QTX(x->qindex, 0, xd->bd) >> (xd->bd - 8);
+  features[f_idx++] = logf(1.0f + (float)(dc_q * dc_q) / 256.0f);
+
+  // VARIANCE
+  const uint8_t *src = x->plane[0].src.buf;
+  const int src_stride = x->plane[0].src.stride;
+  const uint8_t *dst = xd->plane[0].dst.buf;
+  const int dst_stride = xd->plane[0].dst.stride;
+  unsigned int sse = 0;
+
+  // Whole block
+  const unsigned int var =
+      cpi->fn_ptr[bsize].vf(src, src_stride, dst, dst_stride, &sse);
+  features[f_idx++] = logf(1.0f + (float)var);
+
+  // Regional
+  const int bw = block_size_wide[bsize];
+  const int bh = block_size_high[bsize];
+  const BLOCK_SIZE subsize = get_partition_subsize(bsize, PARTITION_SPLIT);
+  int r_idx = 0;
+  for (r_idx = 0; r_idx < 4; r_idx++) {
+    const int x_idx = (r_idx & 1) * bw / 2;
+    const int y_idx = (r_idx >> 1) * bh / 2;
+    const int src_offset = y_idx * src_stride + x_idx;
+    const int dst_offset = y_idx * dst_stride + x_idx;
+    const unsigned int sub_var = cpi->fn_ptr[subsize].vf(
+        src + src_offset, src_stride, dst + dst_offset, dst_stride, &sse);
+    const float var_ratio = (1.0f + (float)sub_var) / (4.0f + (float)var);
+    features[f_idx++] = var_ratio;
+  }
+}
+
+static void full_pixel_motion_search_based_split(
+    AV1_COMP *const cpi, MACROBLOCK *x, int mi_row, int mi_col,
+    BLOCK_SIZE bsize, int *partition_none_allowed, int *partition_horz_allowed,
+    int *partition_vert_allowed, int *do_rectangular_split) {
+  const NN_CONFIG *nn_config = NULL;
+  float split_only_thresh = 0.0f;
+  if (bsize == BLOCK_128X128) {
+    nn_config = &full_pixel_motion_search_based_split_nn_config_128;
+    split_only_thresh = full_pixel_motion_search_based_split_thresh_128;
+  } else if (bsize == BLOCK_64X64) {
+    nn_config = &full_pixel_motion_search_based_split_nn_config_64;
+    split_only_thresh = full_pixel_motion_search_based_split_thresh_64;
+  } else if (bsize == BLOCK_32X32) {
+    nn_config = &full_pixel_motion_search_based_split_nn_config_32;
+    split_only_thresh = full_pixel_motion_search_based_split_thresh_32;
+  } else if (bsize == BLOCK_16X16) {
+    nn_config = &full_pixel_motion_search_based_split_nn_config_16;
+    split_only_thresh = full_pixel_motion_search_based_split_thresh_16;
+  } else if (bsize == BLOCK_8X8) {
+    // Disable BLOCK_8X8 for now
+#if !CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8
+    nn_config = &full_pixel_motion_search_based_split_nn_config_8;
+    split_only_thresh = full_pixel_motion_search_based_split_thresh_8;
+#endif
+  } else {
+    assert(0 && "Unexpected block size in full_pixel_motion_based_split");
+  }
+  if (nn_config) {
+    float features[6] = { 0 };
+    float score = 0;
+    get_res_var_features(cpi, x, mi_row, mi_col, bsize, features);
+    av1_nn_predict(features, nn_config, &score);
+
+    if (score > split_only_thresh) {
+      *partition_none_allowed = 0;
+      *partition_horz_allowed = 0;
+      *partition_vert_allowed = 0;
+      *do_rectangular_split = 0;
+    }
+  }
+}
+
+// Given a list of ref frames in refs, performs simple_motion_search on each of
+// the refs and returns the ref with the smallest sse. Returns -1 if none of the
+// ref in the list is available. Also stores the best sse and var in best_sse,
+// best_var, respectively. If save_mv_code is -1, don't update mv_ref_fulls in
+// pc_tree. If save_mv_code is between 0 and 3, update mv_ref_fulls under
+// pc_tree->split[i]. If save_mv_code is 4, update mv_ref_fulls under pc_tree.
+static int simple_motion_search_get_best_ref(
+    AV1_COMP *const cpi, MACROBLOCK *x, PC_TREE *pc_tree, int mi_row,
+    int mi_col, BLOCK_SIZE bsize, const int *const refs, int num_refs,
+    int use_subpixel, int save_mv_code, unsigned int *best_sse,
+    unsigned int *best_var) {
+  // TODO(chiyotsai@google.com): The calculation of variance currently uses
+  // bsize, so we might take area outside of the image into account. We need to
+  // modify the SIMD functions to fix this later.
+  const AV1_COMMON *const cm = &cpi->common;
+  int best_ref = -1;
+
+  if (mi_col >= cm->mi_cols || mi_row >= cm->mi_rows) {
+    // If the whole block is outside of the image, set the var and sse to 0.
+    *best_var = 0;
+    *best_sse = 0;
+
+    return best_ref;
+  }
+
+  // Otherwise do loop through the reference frames and find the one with the
+  // minimum SSE
+  const MACROBLOCKD *xd = &x->e_mbd;
+  const MV *mv_ref_fulls = pc_tree->mv_ref_fulls;
+
+  const int num_planes = 1;
+
+  *best_sse = INT_MAX;
+
+  for (int ref_idx = 0; ref_idx < num_refs; ref_idx++) {
+    const int ref = refs[ref_idx];
+
+    if (cpi->ref_frame_flags & ref_frame_flag_list[ref]) {
+      unsigned int curr_sse = 0, curr_var = 0;
+      simple_motion_search(cpi, x, mi_row, mi_col, bsize, ref,
+                           mv_ref_fulls[ref], num_planes, use_subpixel);
+      curr_var = cpi->fn_ptr[bsize].vf(
+          x->plane[0].src.buf, x->plane[0].src.stride, xd->plane[0].dst.buf,
+          xd->plane[0].dst.stride, &curr_sse);
+      if (curr_sse < *best_sse) {
+        *best_sse = curr_sse;
+        *best_var = curr_var;
+        best_ref = ref;
+      }
+
+      if (save_mv_code == 4) {
+        pc_tree->mv_ref_fulls[ref].row = x->best_mv.as_mv.row / 8;
+        pc_tree->mv_ref_fulls[ref].col = x->best_mv.as_mv.col / 8;
+      } else if (save_mv_code >= 0 && save_mv_code < 4) {
+        // Propagate the new motion vectors to a lower level
+        pc_tree->split[save_mv_code]->mv_ref_fulls[ref].row =
+            x->best_mv.as_mv.row / 8;
+        pc_tree->split[save_mv_code]->mv_ref_fulls[ref].col =
+            x->best_mv.as_mv.col / 8;
+      }
+    }
+  }
+
+  return best_ref;
+}
+
+// Performs fullpixel simple_motion_search with LAST_FRAME and ALTREF_FRAME on
+// each subblocks and extract the variance and sse of residues. Then store the
+// var and sse from each partition subblock to features. The DC qindex is also
+// stored in features.
+// Here features is assumed to be a length 19 array.
+// After this function is called, we will store the following to features:
+// features[0:17] = var and sse from subblocks
+// features[18] = DC q_index
+#define NUM_FEATURES 19
+static void simple_motion_search_prune_part_features(
+    AV1_COMP *const cpi, MACROBLOCK *x, PC_TREE *pc_tree, int mi_row,
+    int mi_col, BLOCK_SIZE bsize, float *features) {
+  // TODO(chiyotsai@google.com): Cache the result of the motion search from the
+  // larger bbsize.
+  const int w_mi = mi_size_wide[bsize];
+  const int h_mi = mi_size_high[bsize];
+  int f_idx = 0;
+  assert(mi_size_wide[bsize] == mi_size_high[bsize]);
+  assert(cpi->ref_frame_flags & ref_frame_flag_list[LAST_FRAME] ||
+         cpi->ref_frame_flags & ref_frame_flag_list[ALTREF_FRAME]);
+
+  // Setting up motion search
+  const int ref_list[] = { LAST_FRAME, ALTREF_FRAME };
+  const int num_refs = 2;
+  const int use_subpixel = 0;
+
+  unsigned int none_sse = 0, none_var = 0;
+  unsigned int int_features[NUM_FEATURES - 1];
+
+  // Doing whole block first to update the mv
+  simple_motion_search_get_best_ref(cpi, x, pc_tree, mi_row, mi_col, bsize,
+                                    ref_list, num_refs, use_subpixel, 4,
+                                    &none_sse, &none_var);
+
+  // Split subblocks
+  BLOCK_SIZE subsize = get_partition_subsize(bsize, PARTITION_SPLIT);
+  int r_idx = 0;
+  for (r_idx = 0; r_idx < 4; r_idx++) {
+    const int sub_mi_col = mi_col + (r_idx & 1) * w_mi / 2;
+    const int sub_mi_row = mi_row + (r_idx >> 1) * h_mi / 2;
+
+    simple_motion_search_get_best_ref(
+        cpi, x, pc_tree, sub_mi_row, sub_mi_col, subsize, ref_list, num_refs,
+        use_subpixel, r_idx, &int_features[f_idx + 1], &int_features[f_idx]);
+    f_idx += 2;
+  }
+
+  // Horz subblocks
+  subsize = get_partition_subsize(bsize, PARTITION_HORZ);
+  for (r_idx = 0; r_idx < 2; r_idx++) {
+    const int sub_mi_col = mi_col + 0;
+    const int sub_mi_row = mi_row + r_idx * h_mi / 2;
+
+    simple_motion_search_get_best_ref(
+        cpi, x, pc_tree, sub_mi_row, sub_mi_col, subsize, ref_list, num_refs,
+        use_subpixel, -1, &int_features[f_idx + 1], &int_features[f_idx]);
+
+    f_idx += 2;
+  }
+
+  // Vert subblock
+  subsize = get_partition_subsize(bsize, PARTITION_VERT);
+  for (r_idx = 0; r_idx < 2; r_idx++) {
+    const int sub_mi_col = mi_col + r_idx * w_mi / 2;
+    const int sub_mi_row = mi_row + 0;
+
+    simple_motion_search_get_best_ref(
+        cpi, x, pc_tree, sub_mi_row, sub_mi_col, subsize, ref_list, num_refs,
+        use_subpixel, -1, &int_features[f_idx + 1], &int_features[f_idx]);
+
+    f_idx += 2;
+  }
+
+  // Whole block
+  int_features[f_idx++] = none_var;
+  int_features[f_idx++] = none_sse;
+
+  aom_clear_system_state();
+  for (int idx = 0; idx < f_idx; idx++) {
+    features[idx] = logf(1.0f + (float)int_features[idx]);
+  }
+
+  const MACROBLOCKD *xd = &x->e_mbd;
+  set_offsets(cpi, &xd->tile, x, mi_row, mi_col, bsize);
+
+  // Q_INDEX
+  const int dc_q = av1_dc_quant_QTX(x->qindex, 0, xd->bd) >> (xd->bd - 8);
+  features[f_idx++] = logf(1.0f + (float)(dc_q * dc_q) / 256.0f);
+
+  assert(f_idx == NUM_FEATURES);
+}
+
+#define MAX_NUM_CLASSES 10
+static void simple_motion_search_prune_part(
+    AV1_COMP *const cpi, MACROBLOCK *x, PC_TREE *pc_tree, int mi_row,
+    int mi_col, BLOCK_SIZE bsize, int *partition_none_allowed,
+    int *partition_horz_allowed, int *partition_vert_allowed,
+    int *do_square_split, int *do_rectangular_split, int *prune_horz,
+    int *prune_vert) {
+  const AV1_COMMON *const cm = &cpi->common;
+  // Get model parameters
+  const NN_CONFIG *nn_config = NULL;
+  const float *prune_thresh = NULL, *only_thresh = NULL;
+  const float *ml_mean = NULL, *ml_std = NULL;
+
+  if (bsize == BLOCK_128X128) {
+    nn_config = &simple_motion_search_prune_part_nn_config_128;
+    ml_mean = simple_motion_search_prune_part_mean_128;
+    ml_std = simple_motion_search_prune_part_std_128;
+    prune_thresh = simple_motion_search_prune_part_prune_thresh_128;
+    only_thresh = simple_motion_search_prune_part_only_thresh_128;
+  } else if (bsize == BLOCK_64X64) {
+    nn_config = &simple_motion_search_prune_part_nn_config_64;
+    ml_mean = simple_motion_search_prune_part_mean_64;
+    ml_std = simple_motion_search_prune_part_std_64;
+    prune_thresh = simple_motion_search_prune_part_prune_thresh_64;
+    only_thresh = simple_motion_search_prune_part_only_thresh_64;
+  } else if (bsize == BLOCK_32X32) {
+    nn_config = &simple_motion_search_prune_part_nn_config_32;
+    ml_mean = simple_motion_search_prune_part_mean_32;
+    ml_std = simple_motion_search_prune_part_std_32;
+    prune_thresh = simple_motion_search_prune_part_prune_thresh_32;
+    only_thresh = simple_motion_search_prune_part_only_thresh_32;
+  } else if (bsize == BLOCK_16X16) {
+    nn_config = &simple_motion_search_prune_part_nn_config_16;
+    ml_mean = simple_motion_search_prune_part_mean_16;
+    ml_std = simple_motion_search_prune_part_std_16;
+    prune_thresh = simple_motion_search_prune_part_prune_thresh_16;
+    only_thresh = simple_motion_search_prune_part_only_thresh_16;
+  } else if (bsize == BLOCK_8X8) {
+    nn_config = &simple_motion_search_prune_part_nn_config_8;
+    ml_mean = simple_motion_search_prune_part_mean_8;
+    ml_std = simple_motion_search_prune_part_std_8;
+    prune_thresh = simple_motion_search_prune_part_prune_thresh_8;
+    only_thresh = simple_motion_search_prune_part_only_thresh_8;
+  } else {
+    assert(0 && "Unexpected block size in simple_motion_prune_part");
+  }
+
+  // If there is no valid threshold, return immediately.
+  if (!nn_config || (prune_thresh[PARTITION_HORZ] == 0.0f &&
+                     prune_thresh[PARTITION_VERT] == 0.0f)) {
+    return;
+  }
+
+  // Get features
+  float features[NUM_FEATURES] = { 0.0f };
+  simple_motion_search_prune_part_features(cpi, x, pc_tree, mi_row, mi_col,
+                                           bsize, features);
+  for (int f_idx = 0; f_idx < NUM_FEATURES; f_idx++) {
+    features[f_idx] = (features[f_idx] - ml_mean[f_idx]) / ml_std[f_idx];
+  }
+
+  // Get probabilities
+  float scores[MAX_NUM_CLASSES] = { 0.0f }, probs[MAX_NUM_CLASSES] = { 0.0f };
+  const int num_classes =
+      (bsize == BLOCK_128X128 || bsize == BLOCK_8X8) ? 4 : 10;
+
+  av1_nn_predict(features, nn_config, scores);
+  aom_clear_system_state();
+
+  av1_nn_softmax(scores, probs, num_classes);
+
+  // Determine if we should prune rectangular partitions.
+  if (cpi->sf.simple_motion_search_prune_rect && !frame_is_intra_only(cm) &&
+      (*partition_horz_allowed || *partition_vert_allowed) &&
+      bsize >= BLOCK_8X8 && !av1_superres_scaled(cm)) {
+    *prune_horz = probs[PARTITION_HORZ] <= prune_thresh[PARTITION_HORZ];
+    *prune_vert = probs[PARTITION_VERT] <= prune_thresh[PARTITION_VERT];
+  }
+
+  // Silence compiler warnings
+  (void)only_thresh;
+  (void)partition_none_allowed;
+  (void)do_square_split;
+  (void)do_rectangular_split;
+}
+#undef MAX_NUM_CLASSES
+#undef NUM_FEATURES
+
+// TODO(jinging,jimbankoski,rbultje): properly skip partition types that are
 // unlikely to be selected depending on previous rate-distortion optimization
 // results, for encoding speed-up.
 static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
                               TileDataEnc *tile_data, TOKENEXTRA **tp,
                               int mi_row, int mi_col, BLOCK_SIZE bsize,
+                              BLOCK_SIZE max_sq_part, BLOCK_SIZE min_sq_part,
                               RD_STATS *rd_cost, int64_t best_rd,
                               PC_TREE *pc_tree, int64_t *none_rd) {
   const AV1_COMMON *const cm = &cpi->common;
@@ -3202,17 +3772,29 @@ static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
   const int *partition_cost =
       pl >= 0 ? x->partition_cost[pl] : x->partition_cost[0];
 
-  int do_rectangular_split = 1;
+  int do_rectangular_split = cpi->oxcf.enable_rect_partitions;
   int64_t cur_none_rd = 0;
   int64_t split_rd[4] = { 0, 0, 0, 0 };
   int64_t horz_rd[2] = { 0, 0 };
   int64_t vert_rd[2] = { 0, 0 };
+  int prune_horz = 0;
+  int prune_vert = 0;
 
   int split_ctx_is_ready[2] = { 0, 0 };
   int horz_ctx_is_ready = 0;
   int vert_ctx_is_ready = 0;
   BLOCK_SIZE bsize2 = get_partition_subsize(bsize, PARTITION_SPLIT);
 
+  // Max and min square partition levels are defined as the partition nodes that
+  // the recursive function rd_pick_partition() can reach. To implement this:
+  // only PARTITION_SPLIT is NOT allowed if the current node equals min_sq_part,
+  // only PARTITION_SPLIT is allowed if the current node exceeds max_sq_part.
+  assert(block_size_wide[min_sq_part] == block_size_high[min_sq_part]);
+  assert(block_size_wide[max_sq_part] == block_size_high[max_sq_part]);
+  assert(min_sq_part <= max_sq_part);
+  int is_eq_min_sq_part = bsize == min_sq_part;
+  int is_gt_max_sq_part = bsize > max_sq_part;
+
   if (best_rd < 0) {
     pc_tree->none.rdcost = INT64_MAX;
     pc_tree->none.skip = 0;
@@ -3238,11 +3820,20 @@ static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
 #endif
 
   int partition_none_allowed = has_rows && has_cols;
-  int partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8;
-  int partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8;
+  int partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8 &&
+                               cpi->oxcf.enable_rect_partitions;
+  int partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8 &&
+                               cpi->oxcf.enable_rect_partitions;
 
   (void)*tp_orig;
 
+#if CONFIG_COLLECT_PARTITION_STATS
+  PartitionStats *part_stats = &cpi->partition_stats;
+  const int bsize_idx = av1_get_bsize_idx_for_part_stats(bsize);
+  int *partition_decisions = part_stats->partition_decisions[bsize_idx];
+  int *partition_attempts = part_stats->partition_attempts[bsize_idx];
+#endif
+
   // Override partition costs at the edges of the frame in the same
   // way as in read_partition (see decodeframe.c)
   if (!(has_rows && has_cols)) {
@@ -3292,7 +3883,7 @@ static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
   if (cpi->sf.cb_partition_search && bsize == BLOCK_16X16) {
     const int cb_partition_search_ctrl =
         ((pc_tree->index == 0 || pc_tree->index == 3) +
-         get_chessboard_index(cm->current_video_frame)) &
+         get_chessboard_index(cm->current_frame.frame_number)) &
         0x1;
 
     if (cb_partition_search_ctrl && bsize > min_size && bsize < max_size)
@@ -3362,8 +3953,10 @@ static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
         partition_horz_allowed == 0 && partition_vert_allowed == 0) {
       do_square_split = bsize_at_least_8x8;
       partition_none_allowed = has_rows && has_cols;
-      partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8;
-      partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8;
+      partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8 &&
+                               cpi->oxcf.enable_rect_partitions;
+      partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8 &&
+                               cpi->oxcf.enable_rect_partitions;
     }
   }
 
@@ -3435,11 +4028,41 @@ static void rd_pick_partition(AV1_COMP *const cpi, ThreadData *td,
     0,
   };
 
+  MB_MODE_INFO *split_mbmi[4] = { 0 };
+
+  // Use simple_motion_search to prune partitions. This must be done prior to
+  // PARTITION_SPLIT to propagate the initial mvs to a smaller blocksize.
+  const int try_split_only =
+      cpi->sf.simple_motion_search_split_only && bsize >= BLOCK_8X8 &&
+      do_square_split && mi_row + mi_size_high[bsize] <= cm->mi_rows &&
+      mi_col + mi_size_wide[bsize] <= cm->mi_cols && !frame_is_intra_only(cm) &&
+      !av1_superres_scaled(cm);
+
+  if (try_split_only) {
+    full_pixel_motion_search_based_split(
+        cpi, x, mi_row, mi_col, bsize, &partition_none_allowed,
+        &partition_horz_allowed, &partition_vert_allowed,
+        &do_rectangular_split);
+  }
+
+  const int try_prune_rect =
+      cpi->sf.simple_motion_search_prune_rect && !frame_is_intra_only(cm) &&
+      (partition_horz_allowed || partition_vert_allowed) && bsize >= BLOCK_8X8;
+
+  if (try_prune_rect) {
+    simple_motion_search_prune_part(
+        cpi, x, pc_tree, mi_row, mi_col, bsize, &partition_none_allowed,
+        &partition_horz_allowed, &partition_vert_allowed, &do_square_split,
+        &do_rectangular_split, &prune_horz, &prune_vert);
+  }
+
 BEGIN_PARTITION_SEARCH:
   if (x->must_find_valid_partition) {
     partition_none_allowed = has_rows && has_cols;
-    partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8;
-    partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8;
+    partition_horz_allowed = has_cols && yss <= xss && bsize_at_least_8x8 &&
+                             cpi->oxcf.enable_rect_partitions;
+    partition_vert_allowed = has_rows && xss <= yss && bsize_at_least_8x8 &&
+                             cpi->oxcf.enable_rect_partitions;
   }
 
   // Partition block source pixel variance.
@@ -3455,17 +4078,23 @@ BEGIN_PARTITION_SEARCH:
 #endif
 
   // PARTITION_NONE
-  if (partition_none_allowed) {
+  if (is_eq_min_sq_part) partition_none_allowed = 1;
+  if (partition_none_allowed && !is_gt_max_sq_part) {
     int pt_cost = 0;
     if (bsize_at_least_8x8) {
       pt_cost = partition_cost[PARTITION_NONE] < INT_MAX
                     ? partition_cost[PARTITION_NONE]
                     : 0;
     }
-    int64_t partition_rd_cost = RDCOST(x->rdmult, pt_cost, 0);
-    int64_t best_remain_rdcost = (best_rdc.rdcost == INT64_MAX)
-                                     ? INT64_MAX
-                                     : (best_rdc.rdcost - partition_rd_cost);
+    const int64_t partition_rd_cost = RDCOST(x->rdmult, pt_cost, 0);
+    const int64_t best_remain_rdcost =
+        (best_rdc.rdcost == INT64_MAX) ? INT64_MAX
+                                       : (best_rdc.rdcost - partition_rd_cost);
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_remain_rdcost >= 0) {
+      partition_attempts[PARTITION_NONE] += 1;
+    }
+#endif
     rd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &this_rdc,
                      PARTITION_NONE, bsize, ctx_none, best_remain_rdcost);
     pb_source_variance = x->source_variance;
@@ -3496,6 +4125,25 @@ BEGIN_PARTITION_SEARCH:
         best_rdc = this_rdc;
         if (bsize_at_least_8x8) pc_tree->partitioning = PARTITION_NONE;
 
+#if CONFIG_ONE_PASS_SVM
+        // Use ML if the block size is square and >= 16X16
+        if (bsize >= BLOCK_16X16 && !frame_is_intra_only(cm) &&
+            this_rdc.rate < INT_MAX && this_rdc.rate >= 0 &&
+            !ctx_none->seg_feat) {
+          // Model Prediction
+          float score = 0.0f;
+          ml_op_svm_early_term(cpi, x, xd, ctx_none, &this_rdc,
+                               pb_source_variance, bsize, &score);
+
+          // Decide if we want to terminate early
+          if (score >= 0) {
+            do_square_split = 0;
+            do_rectangular_split = 0;
+            partition_horz_allowed = 0;
+            partition_vert_allowed = 0;
+          }
+        }
+#endif
         if ((do_square_split || do_rectangular_split) &&
             !x->e_mbd.lossless[xd->mi[0]->segment_id] && ctx_none->skippable) {
           const int use_ml_based_breakout =
@@ -3576,13 +4224,19 @@ BEGIN_PARTITION_SEARCH:
   if (cpi->sf.adaptive_motion_search) store_pred_mv(x, ctx_none);
 
   // PARTITION_SPLIT
-  if (do_square_split) {
+  if (is_eq_min_sq_part) do_square_split = 0;
+  if (do_square_split || is_gt_max_sq_part) {
     av1_init_rd_stats(&sum_rdc);
     subsize = get_partition_subsize(bsize, PARTITION_SPLIT);
     sum_rdc.rate = partition_cost[PARTITION_SPLIT];
     sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
 
     int idx;
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_rdc.rdcost - sum_rdc.rdcost >= 0) {
+      partition_attempts[PARTITION_SPLIT] += 1;
+    }
+#endif
     for (idx = 0; idx < 4 && sum_rdc.rdcost < best_rdc.rdcost; ++idx) {
       const int x_idx = (idx & 1) * mi_step;
       const int y_idx = (idx >> 1) * mi_step;
@@ -3594,14 +4248,14 @@ BEGIN_PARTITION_SEARCH:
 
       pc_tree->split[idx]->index = idx;
       int64_t *p_split_rd = &split_rd[idx];
-      int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
-                                       ? INT64_MAX
+      const int64_t best_remain_rdcost =
+          best_rdc.rdcost == INT64_MAX ? INT64_MAX
                                        : (best_rdc.rdcost - sum_rdc.rdcost);
       if (cpi->sf.prune_ref_frame_for_rect_partitions)
         pc_tree->split[idx]->none.rate = INT_MAX;
       rd_pick_partition(cpi, td, tile_data, tp, mi_row + y_idx, mi_col + x_idx,
-                        subsize, &this_rdc, best_remain_rdcost,
-                        pc_tree->split[idx], p_split_rd);
+                        subsize, max_sq_part, min_sq_part, &this_rdc,
+                        best_remain_rdcost, pc_tree->split[idx], p_split_rd);
 
       if (this_rdc.rate == INT_MAX) {
         sum_rdc.rdcost = INT64_MAX;
@@ -3615,6 +4269,10 @@ BEGIN_PARTITION_SEARCH:
           const int ref_type =
               av1_ref_frame_type(pc_tree->split[idx]->none.mic.ref_frame);
           ref_frames_used[idx] |= (1 << ref_type);
+
+          if (cpi->sf.prune_ref_mode_for_partitions) {
+            split_mbmi[idx] = &pc_tree->split[idx]->none.mic;
+          }
         }
         if (idx <= 1 && (bsize <= BLOCK_8X8 ||
                          pc_tree->split[idx]->partitioning == PARTITION_NONE)) {
@@ -3662,18 +4320,93 @@ BEGIN_PARTITION_SEARCH:
     if (used_frames) pc_tree->vertical[1].skip_ref_frame_mask = ~used_frames;
   }
 
-  int prune_horz = 0;
-  int prune_vert = 0;
+  for (int i = 0; i < 2; ++i) {
+    pc_tree->horizontal[i].ref_selected[0] =
+        pc_tree->horizontal[i].ref_selected[1] = NONE_FRAME;
+    pc_tree->horizontal[i].mode_selected = -1;
+    pc_tree->vertical[i].ref_selected[0] =
+        pc_tree->vertical[i].ref_selected[1] = NONE_FRAME;
+    pc_tree->vertical[i].mode_selected = -1;
+  }
+
+  if (cpi->sf.prune_ref_mode_for_partitions) {
+    // horizontal partition
+    for (int idx = 0; idx < 4; idx += 2) {
+      const int horz_idx = idx / 2;
+      if (split_mbmi[idx] && split_mbmi[idx + 1] &&
+          split_mbmi[idx]->ref_frame[0] > INTRA_FRAME) {
+        if (!has_second_ref(split_mbmi[idx])) {
+          // Single ref
+          if (split_mbmi[idx]->ref_frame[0] ==
+                  split_mbmi[idx + 1]->ref_frame[0] &&
+              !has_second_ref(split_mbmi[idx + 1])) {
+            const int ref_type = av1_ref_frame_type(split_mbmi[idx]->ref_frame);
+            // Overwrite skip_ref_frame_mask for the current block
+            const int used_frames = (1 << ref_type);
+            pc_tree->horizontal[horz_idx].skip_ref_frame_mask = ~used_frames;
+            pc_tree->horizontal[horz_idx].ref_selected[0] =
+                split_mbmi[idx]->ref_frame[0];
+#if 0
+            // TODO(zoeliu@gmail.com): To consider the scenario of obmc
+            if (split_mbmi[idx]->motion_mode ==
+                    split_mbmi[idx + 1]->motion_mode &&
+                split_mbmi[idx]->motion_mode == SIMPLE_TRANSLATION &&
+                split_mbmi[idx]->use_wedge_interintra == 0) {
+              pc_tree->horizontal[horz_idx].mode_selected = SIMPLE_TRANSLATION;
+            }
+#endif  // 0
+          }
+        } else {
+          // TODO(zoeliu@gmail.com): To handle comp ref
+        }
+      }
+    }
+    // vertical partition
+    for (int idx = 0; idx < 2; ++idx) {
+      const int vert_idx = idx;
+      if (split_mbmi[idx] && split_mbmi[idx + 2] &&
+          split_mbmi[idx]->ref_frame[0] > INTRA_FRAME) {
+        if (!has_second_ref(split_mbmi[idx])) {
+          // Single ref
+          if (split_mbmi[idx]->ref_frame[0] ==
+                  split_mbmi[idx + 2]->ref_frame[0] &&
+              !has_second_ref(split_mbmi[idx + 2])) {
+            const int ref_type = av1_ref_frame_type(split_mbmi[idx]->ref_frame);
+            // Overwrite skip_ref_frame_mask for the current block
+            const int used_frames = (1 << ref_type);
+            pc_tree->vertical[vert_idx].skip_ref_frame_mask = ~used_frames;
+            pc_tree->vertical[vert_idx].ref_selected[0] =
+                split_mbmi[idx]->ref_frame[0];
+#if 0
+            // TODO(zoeliu@gmail.com): To consider the scenario of obmc
+            if (split_mbmi[idx]->motion_mode ==
+                    split_mbmi[idx + 2]->motion_mode &&
+                split_mbmi[idx]->motion_mode == SIMPLE_TRANSLATION &&
+                split_mbmi[idx]->use_wedge_interintra == 0) {
+              pc_tree->vertical[vert_idx].mode_selected = SIMPLE_TRANSLATION;
+            }
+#endif  // 0
+          }
+        } else {
+          // TODO(zoeliu@gmail.com): To handle comp ref
+        }
+      }
+    }
+  }
+
   if (cpi->sf.ml_prune_rect_partition && !frame_is_intra_only(cm) &&
-      (partition_horz_allowed || partition_vert_allowed)) {
-    av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes);
+      (partition_horz_allowed || partition_vert_allowed) &&
+      !(prune_horz || prune_vert)) {
+    av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes, bsize);
     ml_prune_rect_partition(cpi, x, bsize, best_rdc.rdcost, cur_none_rd,
                             split_rd, &prune_horz, &prune_vert);
   }
 
   // PARTITION_HORZ
+  assert(IMPLIES(!cpi->oxcf.enable_rect_partitions, !partition_horz_allowed));
   if (partition_horz_allowed && !prune_horz &&
-      (do_rectangular_split || active_h_edge(cpi, mi_row, mi_step))) {
+      (do_rectangular_split || active_h_edge(cpi, mi_row, mi_step)) &&
+      !is_gt_max_sq_part) {
     av1_init_rd_stats(&sum_rdc);
     subsize = get_partition_subsize(bsize, PARTITION_HORZ);
     if (cpi->sf.adaptive_motion_search) load_pred_mv(x, ctx_none);
@@ -3682,11 +4415,16 @@ BEGIN_PARTITION_SEARCH:
       pc_tree->horizontal[0].pred_interp_filter =
           av1_extract_interp_filter(ctx_none->mic.interp_filters, 0);
     }
-    int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
-                                     ? INT64_MAX
-                                     : (best_rdc.rdcost - sum_rdc.rdcost);
     sum_rdc.rate = partition_cost[PARTITION_HORZ];
     sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
+    const int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
+                                           ? INT64_MAX
+                                           : (best_rdc.rdcost - sum_rdc.rdcost);
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_remain_rdcost >= 0) {
+      partition_attempts[PARTITION_HORZ] += 1;
+    }
+#endif
     rd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &this_rdc,
                      PARTITION_HORZ, subsize, &pc_tree->horizontal[0],
                      best_remain_rdcost);
@@ -3745,8 +4483,10 @@ BEGIN_PARTITION_SEARCH:
   }
 
   // PARTITION_VERT
+  assert(IMPLIES(!cpi->oxcf.enable_rect_partitions, !partition_vert_allowed));
   if (partition_vert_allowed && !prune_vert &&
-      (do_rectangular_split || active_v_edge(cpi, mi_col, mi_step))) {
+      (do_rectangular_split || active_v_edge(cpi, mi_col, mi_step)) &&
+      !is_gt_max_sq_part) {
     av1_init_rd_stats(&sum_rdc);
     subsize = get_partition_subsize(bsize, PARTITION_VERT);
 
@@ -3759,9 +4499,14 @@ BEGIN_PARTITION_SEARCH:
     }
     sum_rdc.rate = partition_cost[PARTITION_VERT];
     sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
-    int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
-                                     ? INT64_MAX
-                                     : (best_rdc.rdcost - sum_rdc.rdcost);
+    const int64_t best_remain_rdcost = best_rdc.rdcost == INT64_MAX
+                                           ? INT64_MAX
+                                           : (best_rdc.rdcost - sum_rdc.rdcost);
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_remain_rdcost >= 0) {
+      partition_attempts[PARTITION_VERT] += 1;
+    }
+#endif
     rd_pick_sb_modes(cpi, tile_data, x, mi_row, mi_col, &this_rdc,
                      PARTITION_VERT, subsize, &pc_tree->vertical[0],
                      best_remain_rdcost);
@@ -3774,8 +4519,7 @@ BEGIN_PARTITION_SEARCH:
       sum_rdc.rdcost += this_rdc.rdcost;
     }
     vert_rd[0] = this_rdc.rdcost;
-    const int64_t vert_max_rdcost = best_rdc.rdcost;
-    if (sum_rdc.rdcost < vert_max_rdcost && has_cols) {
+    if (sum_rdc.rdcost < best_rdc.rdcost && has_cols) {
       const MB_MODE_INFO *const mbmi = &pc_tree->vertical[0].mic;
       const PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
       // Neither palette mode nor cfl predicted
@@ -3820,8 +4564,8 @@ BEGIN_PARTITION_SEARCH:
   }
 
   if (pb_source_variance == UINT_MAX) {
-    av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes);
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes, bsize);
+    if (is_cur_buf_hbd(xd)) {
       pb_source_variance = av1_high_get_sby_perpixel_variance(
           cpi, &x->plane[0].src, bsize, xd->bd);
     } else {
@@ -3830,6 +4574,8 @@ BEGIN_PARTITION_SEARCH:
     }
   }
 
+  assert(IMPLIES(!cpi->oxcf.enable_rect_partitions, !do_rectangular_split));
+
   const int ext_partition_allowed =
       do_rectangular_split && bsize > BLOCK_8X8 && partition_none_allowed;
 
@@ -3912,9 +4658,9 @@ BEGIN_PARTITION_SEARCH:
 
   if (cpi->sf.ml_prune_ab_partition && ext_partition_allowed &&
       partition_horz_allowed && partition_vert_allowed) {
-    // TODO(huisu@google.com): x->source_variance may not be the current block's
-    // variance. The correct one to use is pb_source_variance.
-    // Need to re-train the model to fix it.
+    // TODO(huisu@google.com): x->source_variance may not be the current
+    // block's variance. The correct one to use is pb_source_variance. Need to
+    // re-train the model to fix it.
     ml_prune_ab_partition(bsize, pc_tree->partitioning,
                           get_unsigned_bits(x->source_variance),
                           best_rdc.rdcost, horz_rd, vert_rd, split_rd,
@@ -3923,7 +4669,7 @@ BEGIN_PARTITION_SEARCH:
   }
 
   // PARTITION_HORZ_A
-  if (partition_horz_allowed && horza_partition_allowed) {
+  if (partition_horz_allowed && horza_partition_allowed && !is_gt_max_sq_part) {
     subsize = get_partition_subsize(bsize, PARTITION_HORZ_A);
     pc_tree->horizontala[0].rd_mode_is_ready = 0;
     pc_tree->horizontala[1].rd_mode_is_ready = 0;
@@ -3939,9 +4685,11 @@ BEGIN_PARTITION_SEARCH:
         pc_tree->horizontala[1].rd_mode_is_ready = 1;
       }
     }
-    pc_tree->horizontala[0].skip_ref_frame_mask = 0;
-    pc_tree->horizontala[1].skip_ref_frame_mask = 0;
-    pc_tree->horizontala[2].skip_ref_frame_mask = 0;
+    for (int i = 0; i < 3; ++i) {
+      pc_tree->horizontala[i].skip_ref_frame_mask = 0;
+      pc_tree->horizontala[i].ref_selected[0] =
+          pc_tree->horizontala[i].ref_selected[1] = NONE_FRAME;
+    }
     if (cpi->sf.prune_ref_frame_for_rect_partitions) {
       int used_frames;
       used_frames = ref_frames_used[0];
@@ -3954,6 +4702,42 @@ BEGIN_PARTITION_SEARCH:
       if (used_frames)
         pc_tree->horizontala[2].skip_ref_frame_mask = ~used_frames;
     }
+    if (cpi->sf.prune_ref_mode_for_partitions) {
+      // Overwrite skip_ref_frame_mask for the current block
+      if (split_mbmi[0] && split_mbmi[0]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[0])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[0]->ref_frame[0];
+        pc_tree->horizontala[0].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontala[0].ref_selected[0] = split_mbmi[0]->ref_frame[0];
+      }
+      if (split_mbmi[1] && split_mbmi[1]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[1])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[1]->ref_frame[0];
+        pc_tree->horizontala[1].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontala[1].ref_selected[0] = split_mbmi[1]->ref_frame[0];
+      }
+      if (split_mbmi[2] && split_mbmi[3] &&
+          split_mbmi[2]->ref_frame[0] > INTRA_FRAME &&
+          split_mbmi[2]->ref_frame[0] == split_mbmi[3]->ref_frame[0] &&
+          !has_second_ref(split_mbmi[2]) &&
+          !has_second_ref(split_mbmi[3])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[2]->ref_frame[0];
+        pc_tree->horizontala[2].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontala[2].ref_selected[0] = split_mbmi[2]->ref_frame[0];
+      }
+    }
+#if CONFIG_COLLECT_PARTITION_STATS
+    {
+      RD_STATS tmp_sum_rdc;
+      av1_init_rd_stats(&tmp_sum_rdc);
+      tmp_sum_rdc.rate = x->partition_cost[pl][PARTITION_HORZ_A];
+      tmp_sum_rdc.rdcost = RDCOST(x->rdmult, tmp_sum_rdc.rate, 0);
+      if (!frame_is_intra_only(cm) &&
+          best_rdc.rdcost - tmp_sum_rdc.rdcost >= 0) {
+        partition_attempts[PARTITION_HORZ_A] += 1;
+      }
+    }
+#endif
     rd_test_partition3(cpi, td, tile_data, tp, pc_tree, &best_rdc,
                        pc_tree->horizontala, ctx_none, mi_row, mi_col, bsize,
                        PARTITION_HORZ_A, mi_row, mi_col, bsize2, mi_row,
@@ -3962,7 +4746,7 @@ BEGIN_PARTITION_SEARCH:
     restore_context(x, &x_ctx, mi_row, mi_col, bsize, num_planes);
   }
   // PARTITION_HORZ_B
-  if (partition_horz_allowed && horzb_partition_allowed) {
+  if (partition_horz_allowed && horzb_partition_allowed && !is_gt_max_sq_part) {
     subsize = get_partition_subsize(bsize, PARTITION_HORZ_B);
     pc_tree->horizontalb[0].rd_mode_is_ready = 0;
     pc_tree->horizontalb[1].rd_mode_is_ready = 0;
@@ -3972,9 +4756,11 @@ BEGIN_PARTITION_SEARCH:
       pc_tree->horizontalb[0].mic.partition = PARTITION_HORZ_B;
       pc_tree->horizontalb[0].rd_mode_is_ready = 1;
     }
-    pc_tree->horizontalb[0].skip_ref_frame_mask = 0;
-    pc_tree->horizontalb[1].skip_ref_frame_mask = 0;
-    pc_tree->horizontalb[2].skip_ref_frame_mask = 0;
+    for (int i = 0; i < 3; ++i) {
+      pc_tree->horizontalb[i].skip_ref_frame_mask = 0;
+      pc_tree->horizontalb[i].ref_selected[0] =
+          pc_tree->horizontalb[i].ref_selected[1] = NONE_FRAME;
+    }
     if (cpi->sf.prune_ref_frame_for_rect_partitions) {
       int used_frames;
       used_frames = ref_frames_used[0] | ref_frames_used[1];
@@ -3987,6 +4773,42 @@ BEGIN_PARTITION_SEARCH:
       if (used_frames)
         pc_tree->horizontalb[2].skip_ref_frame_mask = ~used_frames;
     }
+    if (cpi->sf.prune_ref_mode_for_partitions) {
+      // Overwrite skip_ref_frame_mask for the current block
+      if (split_mbmi[0] && split_mbmi[1] &&
+          split_mbmi[0]->ref_frame[0] > INTRA_FRAME &&
+          split_mbmi[0]->ref_frame[0] == split_mbmi[1]->ref_frame[0] &&
+          !has_second_ref(split_mbmi[0]) &&
+          !has_second_ref(split_mbmi[1])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[0]->ref_frame[0];
+        pc_tree->horizontalb[0].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontalb[0].ref_selected[0] = split_mbmi[0]->ref_frame[0];
+      }
+      if (split_mbmi[2] && split_mbmi[2]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[2])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[2]->ref_frame[0];
+        pc_tree->horizontalb[1].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontalb[1].ref_selected[0] = split_mbmi[2]->ref_frame[0];
+      }
+      if (split_mbmi[3] && split_mbmi[3]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[3])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[3]->ref_frame[0];
+        pc_tree->horizontalb[2].skip_ref_frame_mask = ~used_frames;
+        pc_tree->horizontalb[2].ref_selected[0] = split_mbmi[3]->ref_frame[0];
+      }
+    }
+#if CONFIG_COLLECT_PARTITION_STATS
+    {
+      RD_STATS tmp_sum_rdc;
+      av1_init_rd_stats(&tmp_sum_rdc);
+      tmp_sum_rdc.rate = x->partition_cost[pl][PARTITION_HORZ_B];
+      tmp_sum_rdc.rdcost = RDCOST(x->rdmult, tmp_sum_rdc.rate, 0);
+      if (!frame_is_intra_only(cm) &&
+          best_rdc.rdcost - tmp_sum_rdc.rdcost >= 0) {
+        partition_attempts[PARTITION_HORZ_B] += 1;
+      }
+    }
+#endif
     rd_test_partition3(cpi, td, tile_data, tp, pc_tree, &best_rdc,
                        pc_tree->horizontalb, ctx_none, mi_row, mi_col, bsize,
                        PARTITION_HORZ_B, mi_row, mi_col, subsize,
@@ -3996,7 +4818,7 @@ BEGIN_PARTITION_SEARCH:
   }
 
   // PARTITION_VERT_A
-  if (partition_vert_allowed && verta_partition_allowed) {
+  if (partition_vert_allowed && verta_partition_allowed && !is_gt_max_sq_part) {
     subsize = get_partition_subsize(bsize, PARTITION_VERT_A);
     pc_tree->verticala[0].rd_mode_is_ready = 0;
     pc_tree->verticala[1].rd_mode_is_ready = 0;
@@ -4006,9 +4828,11 @@ BEGIN_PARTITION_SEARCH:
       pc_tree->verticala[0].mic.partition = PARTITION_VERT_A;
       pc_tree->verticala[0].rd_mode_is_ready = 1;
     }
-    pc_tree->verticala[0].skip_ref_frame_mask = 0;
-    pc_tree->verticala[1].skip_ref_frame_mask = 0;
-    pc_tree->verticala[2].skip_ref_frame_mask = 0;
+    for (int i = 0; i < 3; ++i) {
+      pc_tree->verticala[i].skip_ref_frame_mask = 0;
+      pc_tree->verticala[i].ref_selected[0] =
+          pc_tree->verticala[i].ref_selected[1] = NONE_FRAME;
+    }
     if (cpi->sf.prune_ref_frame_for_rect_partitions) {
       int used_frames;
       used_frames = ref_frames_used[0];
@@ -4018,6 +4842,42 @@ BEGIN_PARTITION_SEARCH:
       used_frames = ref_frames_used[1] | ref_frames_used[3];
       if (used_frames) pc_tree->verticala[2].skip_ref_frame_mask = ~used_frames;
     }
+    if (cpi->sf.prune_ref_mode_for_partitions) {
+      // Overwrite skip_ref_frame_mask for the current block
+      if (split_mbmi[0] && split_mbmi[0]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[0])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[0]->ref_frame[0];
+        pc_tree->verticala[0].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticala[0].ref_selected[0] = split_mbmi[0]->ref_frame[0];
+      }
+      if (split_mbmi[2] && split_mbmi[2]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[2])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[2]->ref_frame[0];
+        pc_tree->verticala[1].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticala[1].ref_selected[0] = split_mbmi[2]->ref_frame[0];
+      }
+      if (split_mbmi[1] && split_mbmi[3] &&
+          split_mbmi[1]->ref_frame[0] > INTRA_FRAME &&
+          split_mbmi[1]->ref_frame[0] == split_mbmi[3]->ref_frame[0] &&
+          !has_second_ref(split_mbmi[1]) &&
+          !has_second_ref(split_mbmi[3])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[1]->ref_frame[0];
+        pc_tree->verticala[2].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticala[2].ref_selected[0] = split_mbmi[1]->ref_frame[0];
+      }
+    }
+#if CONFIG_COLLECT_PARTITION_STATS
+    {
+      RD_STATS tmp_sum_rdc;
+      av1_init_rd_stats(&tmp_sum_rdc);
+      tmp_sum_rdc.rate = x->partition_cost[pl][PARTITION_VERT_A];
+      tmp_sum_rdc.rdcost = RDCOST(x->rdmult, tmp_sum_rdc.rate, 0);
+      if (!frame_is_intra_only(cm) &&
+          best_rdc.rdcost - tmp_sum_rdc.rdcost >= 0) {
+        partition_attempts[PARTITION_VERT_A] += 1;
+      }
+    }
+#endif
     rd_test_partition3(cpi, td, tile_data, tp, pc_tree, &best_rdc,
                        pc_tree->verticala, ctx_none, mi_row, mi_col, bsize,
                        PARTITION_VERT_A, mi_row, mi_col, bsize2,
@@ -4026,7 +4886,7 @@ BEGIN_PARTITION_SEARCH:
     restore_context(x, &x_ctx, mi_row, mi_col, bsize, num_planes);
   }
   // PARTITION_VERT_B
-  if (partition_vert_allowed && vertb_partition_allowed) {
+  if (partition_vert_allowed && vertb_partition_allowed && !is_gt_max_sq_part) {
     subsize = get_partition_subsize(bsize, PARTITION_VERT_B);
     pc_tree->verticalb[0].rd_mode_is_ready = 0;
     pc_tree->verticalb[1].rd_mode_is_ready = 0;
@@ -4036,9 +4896,11 @@ BEGIN_PARTITION_SEARCH:
       pc_tree->verticalb[0].mic.partition = PARTITION_VERT_B;
       pc_tree->verticalb[0].rd_mode_is_ready = 1;
     }
-    pc_tree->verticalb[0].skip_ref_frame_mask = 0;
-    pc_tree->verticalb[1].skip_ref_frame_mask = 0;
-    pc_tree->verticalb[2].skip_ref_frame_mask = 0;
+    for (int i = 0; i < 3; ++i) {
+      pc_tree->verticalb[i].skip_ref_frame_mask = 0;
+      pc_tree->verticalb[i].ref_selected[0] =
+          pc_tree->verticalb[i].ref_selected[1] = NONE_FRAME;
+    }
     if (cpi->sf.prune_ref_frame_for_rect_partitions) {
       int used_frames;
       used_frames = ref_frames_used[0] | ref_frames_used[2];
@@ -4048,6 +4910,42 @@ BEGIN_PARTITION_SEARCH:
       used_frames = ref_frames_used[3];
       if (used_frames) pc_tree->verticalb[2].skip_ref_frame_mask = ~used_frames;
     }
+    if (cpi->sf.prune_ref_mode_for_partitions) {
+      // Overwrite skip_ref_frame_mask for the current block
+      if (split_mbmi[0] && split_mbmi[2] &&
+          split_mbmi[0]->ref_frame[0] > INTRA_FRAME &&
+          split_mbmi[0]->ref_frame[0] == split_mbmi[2]->ref_frame[0] &&
+          !has_second_ref(split_mbmi[0]) &&
+          !has_second_ref(split_mbmi[2])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[0]->ref_frame[0];
+        pc_tree->verticalb[0].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticalb[0].ref_selected[0] = split_mbmi[0]->ref_frame[0];
+      }
+      if (split_mbmi[1] && split_mbmi[1]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[1])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[1]->ref_frame[0];
+        pc_tree->verticalb[1].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticalb[1].ref_selected[0] = split_mbmi[1]->ref_frame[0];
+      }
+      if (split_mbmi[3] && split_mbmi[3]->ref_frame[0] > INTRA_FRAME &&
+          !has_second_ref(split_mbmi[3])) {  // single ref
+        const int used_frames = 1 << (int)split_mbmi[3]->ref_frame[0];
+        pc_tree->verticalb[2].skip_ref_frame_mask = ~used_frames;
+        pc_tree->verticalb[2].ref_selected[0] = split_mbmi[3]->ref_frame[0];
+      }
+    }
+#if CONFIG_COLLECT_PARTITION_STATS
+    {
+      RD_STATS tmp_sum_rdc;
+      av1_init_rd_stats(&tmp_sum_rdc);
+      tmp_sum_rdc.rate = x->partition_cost[pl][PARTITION_VERT_B];
+      tmp_sum_rdc.rdcost = RDCOST(x->rdmult, tmp_sum_rdc.rate, 0);
+      if (!frame_is_intra_only(cm) &&
+          best_rdc.rdcost - tmp_sum_rdc.rdcost >= 0) {
+        partition_attempts[PARTITION_VERT_B] += 1;
+      }
+    }
+#endif
     rd_test_partition3(cpi, td, tile_data, tp, pc_tree, &best_rdc,
                        pc_tree->verticalb, ctx_none, mi_row, mi_col, bsize,
                        PARTITION_VERT_B, mi_row, mi_col, subsize, mi_row,
@@ -4058,8 +4956,8 @@ BEGIN_PARTITION_SEARCH:
 
   // partition4_allowed is 1 if we can use a PARTITION_HORZ_4 or
   // PARTITION_VERT_4 for this block. This is almost the same as
-  // ext_partition_allowed, except that we don't allow 128x32 or 32x128 blocks,
-  // so we require that bsize is not BLOCK_128X128.
+  // ext_partition_allowed, except that we don't allow 128x32 or 32x128
+  // blocks, so we require that bsize is not BLOCK_128X128.
   const int partition4_allowed =
       ext_partition_allowed && bsize != BLOCK_128X128;
   int partition_horz4_allowed = partition4_allowed && partition_horz_allowed;
@@ -4094,8 +4992,10 @@ BEGIN_PARTITION_SEARCH:
 #endif
 
   // PARTITION_HORZ_4
+  assert(IMPLIES(!cpi->oxcf.enable_rect_partitions, !partition_horz4_allowed));
   if (partition_horz4_allowed && has_rows &&
-      (do_rectangular_split || active_h_edge(cpi, mi_row, mi_step))) {
+      (do_rectangular_split || active_h_edge(cpi, mi_row, mi_step)) &&
+      !is_gt_max_sq_part) {
     av1_init_rd_stats(&sum_rdc);
     const int quarter_step = mi_size_high[bsize] / 4;
     PICK_MODE_CONTEXT *ctx_prev = ctx_none;
@@ -4104,6 +5004,11 @@ BEGIN_PARTITION_SEARCH:
     sum_rdc.rate = partition_cost[PARTITION_HORZ_4];
     sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
 
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_rdc.rdcost - sum_rdc.rdcost >= 0) {
+      partition_attempts[PARTITION_HORZ_4] += 1;
+    }
+#endif
     for (int i = 0; i < 4; ++i) {
       const int this_mi_row = mi_row + i * quarter_step;
 
@@ -4138,8 +5043,10 @@ BEGIN_PARTITION_SEARCH:
   }
 
   // PARTITION_VERT_4
+  assert(IMPLIES(!cpi->oxcf.enable_rect_partitions, !partition_vert4_allowed));
   if (partition_vert4_allowed && has_cols &&
-      (do_rectangular_split || active_v_edge(cpi, mi_row, mi_step))) {
+      (do_rectangular_split || active_v_edge(cpi, mi_row, mi_step)) &&
+      !is_gt_max_sq_part) {
     av1_init_rd_stats(&sum_rdc);
     const int quarter_step = mi_size_wide[bsize] / 4;
     PICK_MODE_CONTEXT *ctx_prev = ctx_none;
@@ -4148,6 +5055,11 @@ BEGIN_PARTITION_SEARCH:
     sum_rdc.rate = partition_cost[PARTITION_VERT_4];
     sum_rdc.rdcost = RDCOST(x->rdmult, sum_rdc.rate, 0);
 
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm) && best_rdc.rdcost - sum_rdc.rdcost >= 0) {
+      partition_attempts[PARTITION_VERT_4] += 1;
+    }
+#endif
     for (int i = 0; i < 4; ++i) {
       const int this_mi_col = mi_col + i * quarter_step;
 
@@ -4185,6 +5097,11 @@ BEGIN_PARTITION_SEARCH:
     // Did not find a valid partition, go back and search again, with less
     // constraint on which partition types to search.
     x->must_find_valid_partition = 1;
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (!frame_is_intra_only(cm)) {
+      part_stats->partition_redo += 1;
+    }
+#endif
     goto BEGIN_PARTITION_SEARCH;
   }
 
@@ -4195,6 +5112,13 @@ BEGIN_PARTITION_SEARCH:
   (void)best_rd;
   *rd_cost = best_rdc;
 
+#if CONFIG_COLLECT_PARTITION_STATS
+  if (!frame_is_intra_only(cm) && best_rdc.rate < INT_MAX &&
+      best_rdc.dist < INT64_MAX) {
+    partition_decisions[pc_tree->partitioning] += 1;
+  }
+#endif
+
   if (best_rdc.rate < INT_MAX && best_rdc.dist < INT64_MAX &&
       pc_tree->index != 3) {
     if (bsize == cm->seq_params.sb_size) {
@@ -4225,19 +5149,380 @@ static void init_first_partition_pass_stats_tables(
   }
 }
 
-// clear pc_tree_stats
-static INLINE void clear_pc_tree_stats(PC_TREE *pt) {
-  if (pt == NULL) return;
-  pt->pc_tree_stats.valid = 0;
-  for (int i = 0; i < 4; ++i) {
-    clear_pc_tree_stats(pt->split[i]);
-  }
-}
-
 // Minimum number of samples to trigger the
 // mode_pruning_based_on_two_pass_partition_search feature.
 #define FIRST_PARTITION_PASS_MIN_SAMPLES 16
 
+static int get_rdmult_delta(AV1_COMP *cpi, BLOCK_SIZE bsize, int mi_row,
+                            int mi_col, int orig_rdmult) {
+  TplDepFrame *tpl_frame = &cpi->tpl_stats[cpi->twopass.gf_group.index];
+  TplDepStats *tpl_stats = tpl_frame->tpl_stats_ptr;
+  int tpl_stride = tpl_frame->stride;
+  int64_t intra_cost = 0;
+  int64_t mc_dep_cost = 0;
+  int mi_wide = mi_size_wide[bsize];
+  int mi_high = mi_size_high[bsize];
+  int row, col;
+
+  int dr = 0;
+  int count = 0;
+  double r0, rk, beta;
+
+  if (tpl_frame->is_valid == 0) return orig_rdmult;
+
+  if (cpi->common.show_frame) return orig_rdmult;
+
+  if (cpi->twopass.gf_group.index >= MAX_LAG_BUFFERS) return orig_rdmult;
+
+  for (row = mi_row; row < mi_row + mi_high; ++row) {
+    for (col = mi_col; col < mi_col + mi_wide; ++col) {
+      TplDepStats *this_stats = &tpl_stats[row * tpl_stride + col];
+
+      if (row >= cpi->common.mi_rows || col >= cpi->common.mi_cols) continue;
+
+      intra_cost += this_stats->intra_cost;
+      mc_dep_cost += this_stats->mc_dep_cost;
+
+      ++count;
+    }
+  }
+
+  aom_clear_system_state();
+
+  r0 = cpi->rd.r0;
+  rk = (double)intra_cost / mc_dep_cost;
+  beta = r0 / rk;
+  dr = av1_get_adaptive_rdmult(cpi, beta);
+
+  dr = AOMMIN(dr, orig_rdmult * 3 / 2);
+  dr = AOMMAX(dr, orig_rdmult * 1 / 2);
+
+  dr = AOMMAX(1, dr);
+
+  return dr;
+}
+
+static void setup_delta_q(AV1_COMP *const cpi, MACROBLOCK *const x,
+                          const TileInfo *const tile_info, int mi_row,
+                          int mi_col, int num_planes) {
+  AV1_COMMON *const cm = &cpi->common;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  const DeltaQInfo *const delta_q_info = &cm->delta_q_info;
+  const BLOCK_SIZE sb_size = cm->seq_params.sb_size;
+  const int mib_size = cm->seq_params.mib_size;
+
+  // Delta-q modulation based on variance
+  av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes, sb_size);
+
+  int offset_qindex;
+  if (DELTAQ_MODULATION == 1) {
+    const int block_wavelet_energy_level =
+        av1_block_wavelet_energy_level(cpi, x, sb_size);
+    x->sb_energy_level = block_wavelet_energy_level;
+    offset_qindex =
+        av1_compute_deltaq_from_energy_level(cpi, block_wavelet_energy_level);
+  } else {
+    const int block_var_level = av1_log_block_var(cpi, x, sb_size);
+    x->sb_energy_level = block_var_level;
+    offset_qindex = av1_compute_deltaq_from_energy_level(cpi, block_var_level);
+  }
+  const int qmask = ~(delta_q_info->delta_q_res - 1);
+  int current_qindex =
+      clamp(cm->base_qindex + offset_qindex, delta_q_info->delta_q_res,
+            256 - delta_q_info->delta_q_res);
+  current_qindex =
+      ((current_qindex - cm->base_qindex + delta_q_info->delta_q_res / 2) &
+       qmask) +
+      cm->base_qindex;
+  assert(current_qindex > 0);
+
+  xd->delta_qindex = current_qindex - cm->base_qindex;
+  set_offsets(cpi, tile_info, x, mi_row, mi_col, sb_size);
+  xd->mi[0]->current_qindex = current_qindex;
+  av1_init_plane_quantizers(cpi, x, xd->mi[0]->segment_id);
+  if (cpi->oxcf.deltaq_mode == DELTA_Q_LF) {
+    const int lfmask = ~(delta_q_info->delta_lf_res - 1);
+    const int delta_lf_from_base =
+        ((offset_qindex / 2 + delta_q_info->delta_lf_res / 2) & lfmask);
+
+    // pre-set the delta lf for loop filter. Note that this value is set
+    // before mi is assigned for each block in current superblock
+    for (int j = 0; j < AOMMIN(mib_size, cm->mi_rows - mi_row); j++) {
+      for (int k = 0; k < AOMMIN(mib_size, cm->mi_cols - mi_col); k++) {
+        cm->mi[(mi_row + j) * cm->mi_stride + (mi_col + k)].delta_lf_from_base =
+            clamp(delta_lf_from_base, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
+        const int frame_lf_count =
+            av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
+        for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
+          cm->mi[(mi_row + j) * cm->mi_stride + (mi_col + k)].delta_lf[lf_id] =
+              clamp(delta_lf_from_base, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
+        }
+      }
+    }
+  }
+}
+
+// First pass of partition search only considers square partition block sizes.
+// The results will be used in the second partition search pass to prune
+// unlikely partition candidates.
+static void first_partition_search_pass(AV1_COMP *cpi, ThreadData *td,
+                                        TileDataEnc *tile_data, int mi_row,
+                                        int mi_col, TOKENEXTRA **tp) {
+  MACROBLOCK *const x = &td->mb;
+  x->cb_partition_scan = 1;
+
+  const SPEED_FEATURES *const sf = &cpi->sf;
+  // Reset the stats tables.
+  if (sf->mode_pruning_based_on_two_pass_partition_search)
+    av1_zero(x->first_partition_pass_stats);
+
+  AV1_COMMON *const cm = &cpi->common;
+  const BLOCK_SIZE sb_size = cm->seq_params.sb_size;
+  const int mib_size_log2 = cm->seq_params.mib_size_log2;
+  PC_TREE *const pc_root = td->pc_root[mib_size_log2 - MIN_MIB_SIZE_LOG2];
+  RD_STATS dummy_rdc;
+  rd_pick_sqr_partition(cpi, td, tile_data, tp, mi_row, mi_col, sb_size,
+                        &dummy_rdc, INT64_MAX, pc_root, NULL);
+  x->cb_partition_scan = 0;
+
+  x->source_variance = UINT_MAX;
+  if (sf->adaptive_pred_interp_filter) {
+    const int leaf_nodes = 256;
+    for (int i = 0; i < leaf_nodes; ++i) {
+      td->pc_tree[i].vertical[0].pred_interp_filter = SWITCHABLE;
+      td->pc_tree[i].vertical[1].pred_interp_filter = SWITCHABLE;
+      td->pc_tree[i].horizontal[0].pred_interp_filter = SWITCHABLE;
+      td->pc_tree[i].horizontal[1].pred_interp_filter = SWITCHABLE;
+    }
+  }
+
+  x->mb_rd_record.num = x->mb_rd_record.index_start = 0;
+  av1_zero(x->txb_rd_record_8X8);
+  av1_zero(x->txb_rd_record_16X16);
+  av1_zero(x->txb_rd_record_32X32);
+  av1_zero(x->txb_rd_record_64X64);
+  av1_zero(x->txb_rd_record_intra);
+  av1_zero(x->pred_mv);
+  pc_root->index = 0;
+
+  for (int idy = 0; idy < mi_size_high[sb_size]; ++idy) {
+    for (int idx = 0; idx < mi_size_wide[sb_size]; ++idx) {
+      const int offset = cm->mi_stride * (mi_row + idy) + (mi_col + idx);
+      cm->mi_grid_visible[offset] = 0;
+    }
+  }
+
+  x->use_cb_search_range = 1;
+
+  if (sf->mode_pruning_based_on_two_pass_partition_search) {
+    for (int i = 0; i < FIRST_PARTITION_PASS_STATS_TABLES; ++i) {
+      FIRST_PARTITION_PASS_STATS *const stat =
+          &x->first_partition_pass_stats[i];
+      if (stat->sample_counts < FIRST_PARTITION_PASS_MIN_SAMPLES) {
+        // If there are not enough samples collected, make all available.
+        memset(stat->ref0_counts, 0xff, sizeof(stat->ref0_counts));
+        memset(stat->ref1_counts, 0xff, sizeof(stat->ref1_counts));
+      } else if (sf->selective_ref_frame < 3) {
+        // ALTREF2_FRAME and BWDREF_FRAME may be skipped during the
+        // initial partition scan, so we don't eliminate them.
+        stat->ref0_counts[ALTREF2_FRAME] = 0xff;
+        stat->ref1_counts[ALTREF2_FRAME] = 0xff;
+        stat->ref0_counts[BWDREF_FRAME] = 0xff;
+        stat->ref1_counts[BWDREF_FRAME] = 0xff;
+      }
+    }
+  }
+}
+
+static void init_simple_motion_search_mvs(PC_TREE *pc_tree) {
+  for (int idx = 0; idx < REF_FRAMES; idx++) {
+    pc_tree->mv_ref_fulls[idx].row = 0;
+    pc_tree->mv_ref_fulls[idx].col = 0;
+  }
+  if (pc_tree->block_size >= BLOCK_8X8) {
+    init_simple_motion_search_mvs(pc_tree->split[0]);
+    init_simple_motion_search_mvs(pc_tree->split[1]);
+    init_simple_motion_search_mvs(pc_tree->split[2]);
+    init_simple_motion_search_mvs(pc_tree->split[3]);
+  }
+}
+
+#define AVG_CDF_WEIGHT_LEFT 3
+#define AVG_CDF_WEIGHT_TOP_RIGHT 1
+
+static void avg_cdf_symbol(aom_cdf_prob *cdf_ptr_left, aom_cdf_prob *cdf_ptr_tr,
+                           int num_cdfs, int cdf_stride, int nsymbs,
+                           int wt_left, int wt_tr) {
+  for (int i = 0; i < num_cdfs; i++) {
+    for (int j = 0; j <= nsymbs; j++) {
+      cdf_ptr_left[i * cdf_stride + j] =
+          (aom_cdf_prob)(((int)cdf_ptr_left[i * cdf_stride + j] * wt_left +
+                          (int)cdf_ptr_tr[i * cdf_stride + j] * wt_tr +
+                          ((wt_left + wt_tr) / 2)) /
+                         (wt_left + wt_tr));
+      assert(cdf_ptr_left[i * cdf_stride + j] >= 0 &&
+             cdf_ptr_left[i * cdf_stride + j] < CDF_PROB_TOP);
+    }
+  }
+}
+
+#define AVERAGE_CDF(cname_left, cname_tr, nsymbs) \
+  AVG_CDF_STRIDE(cname_left, cname_tr, nsymbs, CDF_SIZE(nsymbs))
+
+#define AVG_CDF_STRIDE(cname_left, cname_tr, nsymbs, cdf_stride)           \
+  do {                                                                     \
+    aom_cdf_prob *cdf_ptr_left = (aom_cdf_prob *)cname_left;               \
+    aom_cdf_prob *cdf_ptr_tr = (aom_cdf_prob *)cname_tr;                   \
+    int array_size = (int)sizeof(cname_left) / sizeof(aom_cdf_prob);       \
+    int num_cdfs = array_size / cdf_stride;                                \
+    avg_cdf_symbol(cdf_ptr_left, cdf_ptr_tr, num_cdfs, cdf_stride, nsymbs, \
+                   wt_left, wt_tr);                                        \
+  } while (0)
+
+static void avg_nmv(nmv_context *nmv_left, nmv_context *nmv_tr, int wt_left,
+                    int wt_tr) {
+  AVERAGE_CDF(nmv_left->joints_cdf, nmv_tr->joints_cdf, 4);
+  for (int i = 0; i < 2; i++) {
+    AVERAGE_CDF(nmv_left->comps[i].classes_cdf, nmv_tr->comps[i].classes_cdf,
+                MV_CLASSES);
+    AVERAGE_CDF(nmv_left->comps[i].class0_fp_cdf,
+                nmv_tr->comps[i].class0_fp_cdf, MV_FP_SIZE);
+    AVERAGE_CDF(nmv_left->comps[i].fp_cdf, nmv_tr->comps[i].fp_cdf, MV_FP_SIZE);
+    AVERAGE_CDF(nmv_left->comps[i].sign_cdf, nmv_tr->comps[i].sign_cdf, 2);
+    AVERAGE_CDF(nmv_left->comps[i].class0_hp_cdf,
+                nmv_tr->comps[i].class0_hp_cdf, 2);
+    AVERAGE_CDF(nmv_left->comps[i].hp_cdf, nmv_tr->comps[i].hp_cdf, 2);
+    AVERAGE_CDF(nmv_left->comps[i].class0_cdf, nmv_tr->comps[i].class0_cdf,
+                CLASS0_SIZE);
+    AVERAGE_CDF(nmv_left->comps[i].bits_cdf, nmv_tr->comps[i].bits_cdf, 2);
+  }
+}
+
+// In case of row-based multi-threading of encoder, since we always
+// keep a top - right sync, we can average the top - right SB's CDFs and
+// the left SB's CDFs and use the same for current SB's encoding to
+// improve the performance. This function facilitates the averaging
+// of CDF and used only when row-mt is enabled in encoder.
+static void avg_cdf_symbols(FRAME_CONTEXT *ctx_left, FRAME_CONTEXT *ctx_tr,
+                            int wt_left, int wt_tr) {
+  AVERAGE_CDF(ctx_left->txb_skip_cdf, ctx_tr->txb_skip_cdf, 2);
+  AVERAGE_CDF(ctx_left->eob_extra_cdf, ctx_tr->eob_extra_cdf, 2);
+  AVERAGE_CDF(ctx_left->dc_sign_cdf, ctx_tr->dc_sign_cdf, 2);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf16, ctx_tr->eob_flag_cdf16, 5);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf32, ctx_tr->eob_flag_cdf32, 6);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf64, ctx_tr->eob_flag_cdf64, 7);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf128, ctx_tr->eob_flag_cdf128, 8);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf256, ctx_tr->eob_flag_cdf256, 9);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf512, ctx_tr->eob_flag_cdf512, 10);
+  AVERAGE_CDF(ctx_left->eob_flag_cdf1024, ctx_tr->eob_flag_cdf1024, 11);
+  AVERAGE_CDF(ctx_left->coeff_base_eob_cdf, ctx_tr->coeff_base_eob_cdf, 3);
+  AVERAGE_CDF(ctx_left->coeff_base_cdf, ctx_tr->coeff_base_cdf, 4);
+  AVERAGE_CDF(ctx_left->coeff_br_cdf, ctx_tr->coeff_br_cdf, BR_CDF_SIZE);
+  AVERAGE_CDF(ctx_left->newmv_cdf, ctx_tr->newmv_cdf, 2);
+  AVERAGE_CDF(ctx_left->zeromv_cdf, ctx_tr->zeromv_cdf, 2);
+  AVERAGE_CDF(ctx_left->refmv_cdf, ctx_tr->refmv_cdf, 2);
+  AVERAGE_CDF(ctx_left->drl_cdf, ctx_tr->drl_cdf, 2);
+  AVERAGE_CDF(ctx_left->inter_compound_mode_cdf,
+              ctx_tr->inter_compound_mode_cdf, INTER_COMPOUND_MODES);
+  AVERAGE_CDF(ctx_left->compound_type_cdf, ctx_tr->compound_type_cdf,
+              COMPOUND_TYPES - 1);
+  AVERAGE_CDF(ctx_left->wedge_idx_cdf, ctx_tr->wedge_idx_cdf, 16);
+  AVERAGE_CDF(ctx_left->interintra_cdf, ctx_tr->interintra_cdf, 2);
+  AVERAGE_CDF(ctx_left->wedge_interintra_cdf, ctx_tr->wedge_interintra_cdf, 2);
+  AVERAGE_CDF(ctx_left->interintra_mode_cdf, ctx_tr->interintra_mode_cdf,
+              INTERINTRA_MODES);
+  AVERAGE_CDF(ctx_left->motion_mode_cdf, ctx_tr->motion_mode_cdf, MOTION_MODES);
+  AVERAGE_CDF(ctx_left->obmc_cdf, ctx_tr->obmc_cdf, 2);
+  AVERAGE_CDF(ctx_left->palette_y_size_cdf, ctx_tr->palette_y_size_cdf,
+              PALETTE_SIZES);
+  AVERAGE_CDF(ctx_left->palette_uv_size_cdf, ctx_tr->palette_uv_size_cdf,
+              PALETTE_SIZES);
+  for (int j = 0; j < PALETTE_SIZES; j++) {
+    int nsymbs = j + PALETTE_MIN_SIZE;
+    AVG_CDF_STRIDE(ctx_left->palette_y_color_index_cdf[j],
+                   ctx_tr->palette_y_color_index_cdf[j], nsymbs,
+                   CDF_SIZE(PALETTE_COLORS));
+    AVG_CDF_STRIDE(ctx_left->palette_uv_color_index_cdf[j],
+                   ctx_tr->palette_uv_color_index_cdf[j], nsymbs,
+                   CDF_SIZE(PALETTE_COLORS));
+  }
+  AVERAGE_CDF(ctx_left->palette_y_mode_cdf, ctx_tr->palette_y_mode_cdf, 2);
+  AVERAGE_CDF(ctx_left->palette_uv_mode_cdf, ctx_tr->palette_uv_mode_cdf, 2);
+  AVERAGE_CDF(ctx_left->comp_inter_cdf, ctx_tr->comp_inter_cdf, 2);
+  AVERAGE_CDF(ctx_left->single_ref_cdf, ctx_tr->single_ref_cdf, 2);
+  AVERAGE_CDF(ctx_left->comp_ref_type_cdf, ctx_tr->comp_ref_type_cdf, 2);
+  AVERAGE_CDF(ctx_left->uni_comp_ref_cdf, ctx_tr->uni_comp_ref_cdf, 2);
+  AVERAGE_CDF(ctx_left->comp_ref_cdf, ctx_tr->comp_ref_cdf, 2);
+  AVERAGE_CDF(ctx_left->comp_bwdref_cdf, ctx_tr->comp_bwdref_cdf, 2);
+  AVERAGE_CDF(ctx_left->txfm_partition_cdf, ctx_tr->txfm_partition_cdf, 2);
+  AVERAGE_CDF(ctx_left->compound_index_cdf, ctx_tr->compound_index_cdf, 2);
+  AVERAGE_CDF(ctx_left->comp_group_idx_cdf, ctx_tr->comp_group_idx_cdf, 2);
+  AVERAGE_CDF(ctx_left->skip_mode_cdfs, ctx_tr->skip_mode_cdfs, 2);
+  AVERAGE_CDF(ctx_left->skip_cdfs, ctx_tr->skip_cdfs, 2);
+  AVERAGE_CDF(ctx_left->intra_inter_cdf, ctx_tr->intra_inter_cdf, 2);
+  avg_nmv(&ctx_left->nmvc, &ctx_tr->nmvc, wt_left, wt_tr);
+  avg_nmv(&ctx_left->ndvc, &ctx_tr->ndvc, wt_left, wt_tr);
+  AVERAGE_CDF(ctx_left->intrabc_cdf, ctx_tr->intrabc_cdf, 2);
+  AVERAGE_CDF(ctx_left->seg.tree_cdf, ctx_tr->seg.tree_cdf, MAX_SEGMENTS);
+  AVERAGE_CDF(ctx_left->seg.pred_cdf, ctx_tr->seg.pred_cdf, 2);
+  AVERAGE_CDF(ctx_left->seg.spatial_pred_seg_cdf,
+              ctx_tr->seg.spatial_pred_seg_cdf, MAX_SEGMENTS);
+  AVERAGE_CDF(ctx_left->filter_intra_cdfs, ctx_tr->filter_intra_cdfs, 2);
+  AVERAGE_CDF(ctx_left->filter_intra_mode_cdf, ctx_tr->filter_intra_mode_cdf,
+              FILTER_INTRA_MODES);
+  AVERAGE_CDF(ctx_left->switchable_restore_cdf, ctx_tr->switchable_restore_cdf,
+              RESTORE_SWITCHABLE_TYPES);
+  AVERAGE_CDF(ctx_left->wiener_restore_cdf, ctx_tr->wiener_restore_cdf, 2);
+  AVERAGE_CDF(ctx_left->sgrproj_restore_cdf, ctx_tr->sgrproj_restore_cdf, 2);
+  AVERAGE_CDF(ctx_left->y_mode_cdf, ctx_tr->y_mode_cdf, INTRA_MODES);
+  AVG_CDF_STRIDE(ctx_left->uv_mode_cdf[0], ctx_tr->uv_mode_cdf[0],
+                 UV_INTRA_MODES - 1, CDF_SIZE(UV_INTRA_MODES));
+  AVERAGE_CDF(ctx_left->uv_mode_cdf[1], ctx_tr->uv_mode_cdf[1], UV_INTRA_MODES);
+  for (int i = 0; i < PARTITION_CONTEXTS; i++) {
+    if (i < 4) {
+      AVG_CDF_STRIDE(ctx_left->partition_cdf[i], ctx_tr->partition_cdf[i], 4,
+                     CDF_SIZE(10));
+    } else if (i < 16) {
+      AVERAGE_CDF(ctx_left->partition_cdf[i], ctx_tr->partition_cdf[i], 10);
+    } else {
+      AVG_CDF_STRIDE(ctx_left->partition_cdf[i], ctx_tr->partition_cdf[i], 8,
+                     CDF_SIZE(10));
+    }
+  }
+  AVERAGE_CDF(ctx_left->switchable_interp_cdf, ctx_tr->switchable_interp_cdf,
+              SWITCHABLE_FILTERS);
+  AVERAGE_CDF(ctx_left->kf_y_cdf, ctx_tr->kf_y_cdf, INTRA_MODES);
+  AVERAGE_CDF(ctx_left->angle_delta_cdf, ctx_tr->angle_delta_cdf,
+              2 * MAX_ANGLE_DELTA + 1);
+  AVG_CDF_STRIDE(ctx_left->tx_size_cdf[0], ctx_tr->tx_size_cdf[0], MAX_TX_DEPTH,
+                 CDF_SIZE(MAX_TX_DEPTH + 1));
+  AVERAGE_CDF(ctx_left->tx_size_cdf[1], ctx_tr->tx_size_cdf[1],
+              MAX_TX_DEPTH + 1);
+  AVERAGE_CDF(ctx_left->tx_size_cdf[2], ctx_tr->tx_size_cdf[2],
+              MAX_TX_DEPTH + 1);
+  AVERAGE_CDF(ctx_left->tx_size_cdf[3], ctx_tr->tx_size_cdf[3],
+              MAX_TX_DEPTH + 1);
+  AVERAGE_CDF(ctx_left->delta_q_cdf, ctx_tr->delta_q_cdf, DELTA_Q_PROBS + 1);
+  AVERAGE_CDF(ctx_left->delta_lf_cdf, ctx_tr->delta_lf_cdf, DELTA_LF_PROBS + 1);
+  for (int i = 0; i < FRAME_LF_COUNT; i++) {
+    AVERAGE_CDF(ctx_left->delta_lf_multi_cdf[i], ctx_tr->delta_lf_multi_cdf[i],
+                DELTA_LF_PROBS + 1);
+  }
+  AVG_CDF_STRIDE(ctx_left->intra_ext_tx_cdf[1], ctx_tr->intra_ext_tx_cdf[1], 7,
+                 CDF_SIZE(TX_TYPES));
+  AVG_CDF_STRIDE(ctx_left->intra_ext_tx_cdf[2], ctx_tr->intra_ext_tx_cdf[2], 5,
+                 CDF_SIZE(TX_TYPES));
+  AVG_CDF_STRIDE(ctx_left->inter_ext_tx_cdf[1], ctx_tr->inter_ext_tx_cdf[1], 16,
+                 CDF_SIZE(TX_TYPES));
+  AVG_CDF_STRIDE(ctx_left->inter_ext_tx_cdf[2], ctx_tr->inter_ext_tx_cdf[2], 12,
+                 CDF_SIZE(TX_TYPES));
+  AVG_CDF_STRIDE(ctx_left->inter_ext_tx_cdf[3], ctx_tr->inter_ext_tx_cdf[3], 2,
+                 CDF_SIZE(TX_TYPES));
+  AVERAGE_CDF(ctx_left->cfl_sign_cdf, ctx_tr->cfl_sign_cdf, CFL_JOINT_SIGNS);
+  AVERAGE_CDF(ctx_left->cfl_alpha_cdf, ctx_tr->cfl_alpha_cdf,
+              CFL_ALPHABET_SIZE);
+}
+
 static void encode_rd_sb_row(AV1_COMP *cpi, ThreadData *td,
                              TileDataEnc *tile_data, int mi_row,
                              TOKENEXTRA **tp) {
@@ -4246,25 +5531,47 @@ static void encode_rd_sb_row(AV1_COMP *cpi, ThreadData *td,
   const TileInfo *const tile_info = &tile_data->tile_info;
   MACROBLOCK *const x = &td->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
-  SPEED_FEATURES *const sf = &cpi->sf;
+  const SPEED_FEATURES *const sf = &cpi->sf;
   const int leaf_nodes = 256;
+  const int sb_cols_in_tile = av1_get_sb_cols_in_tile(cm, tile_data->tile_info);
+  const BLOCK_SIZE sb_size = cm->seq_params.sb_size;
+  const int mib_size = cm->seq_params.mib_size;
+  const int mib_size_log2 = cm->seq_params.mib_size_log2;
+  const int sb_row = (mi_row - tile_info->mi_row_start) >> mib_size_log2;
 
   // Initialize the left context for the new SB row
   av1_zero_left_context(xd);
 
   // Reset delta for every tile
   if (mi_row == tile_info->mi_row_start) {
-    if (cm->delta_q_present_flag) xd->current_qindex = cm->base_qindex;
-    if (cm->delta_lf_present_flag) {
+    if (cm->delta_q_info.delta_q_present_flag)
+      xd->current_qindex = cm->base_qindex;
+    if (cm->delta_q_info.delta_lf_present_flag) {
       av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
     }
   }
 
-  PC_TREE *const pc_root =
-      td->pc_root[cm->seq_params.mib_size_log2 - MIN_MIB_SIZE_LOG2];
   // Code each SB in the row
-  for (int mi_col = tile_info->mi_col_start; mi_col < tile_info->mi_col_end;
-       mi_col += cm->seq_params.mib_size) {
+  for (int mi_col = tile_info->mi_col_start, sb_col_in_tile = 0;
+       mi_col < tile_info->mi_col_end; mi_col += mib_size, sb_col_in_tile++) {
+    (*(cpi->row_mt_sync_read_ptr))(&tile_data->row_mt_sync, sb_row,
+                                   sb_col_in_tile);
+    if (tile_data->allow_update_cdf && (cpi->row_mt == 1) &&
+        (tile_info->mi_row_start != mi_row)) {
+      if ((tile_info->mi_col_start == mi_col)) {
+        // restore frame context of 1st column sb
+        memcpy(xd->tile_ctx, x->row_ctx, sizeof(*xd->tile_ctx));
+      } else {
+        int wt_left = AVG_CDF_WEIGHT_LEFT;
+        int wt_tr = AVG_CDF_WEIGHT_TOP_RIGHT;
+        if (tile_info->mi_col_end > (mi_col + mib_size))
+          avg_cdf_symbols(xd->tile_ctx, x->row_ctx + sb_col_in_tile, wt_left,
+                          wt_tr);
+        else
+          avg_cdf_symbols(xd->tile_ctx, x->row_ctx + sb_col_in_tile - 1,
+                          wt_left, wt_tr);
+      }
+    }
     av1_fill_coeff_costs(&td->mb, xd->tile_ctx, num_planes);
     av1_fill_mode_rates(cm, x, xd->tile_ctx);
 
@@ -4286,76 +5593,27 @@ static void encode_rd_sb_row(AV1_COMP *cpi, ThreadData *td,
     av1_zero(x->txb_rd_record_intra);
 
     av1_zero(x->pred_mv);
+    PC_TREE *const pc_root = td->pc_root[mib_size_log2 - MIN_MIB_SIZE_LOG2];
     pc_root->index = 0;
 
+    if (sf->simple_motion_search_prune_rect && !frame_is_intra_only(cm)) {
+      init_simple_motion_search_mvs(pc_root);
+    }
+
     const struct segmentation *const seg = &cm->seg;
     int seg_skip = 0;
     if (seg->enabled) {
       const uint8_t *const map =
           seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map;
       const int segment_id =
-          map ? get_segment_id(cm, map, cm->seq_params.sb_size, mi_row, mi_col)
-              : 0;
+          map ? get_segment_id(cm, map, sb_size, mi_row, mi_col) : 0;
       seg_skip = segfeature_active(seg, segment_id, SEG_LVL_SKIP);
     }
     xd->cur_frame_force_integer_mv = cm->cur_frame_force_integer_mv;
 
     x->sb_energy_level = 0;
-    if (cm->delta_q_present_flag) {
-      // Delta-q modulation based on variance
-      av1_setup_src_planes(x, cpi->source, mi_row, mi_col, num_planes);
-
-      int offset_qindex;
-      if (DELTAQ_MODULATION == 1) {
-        const int block_wavelet_energy_level =
-            av1_block_wavelet_energy_level(cpi, x, cm->seq_params.sb_size);
-        x->sb_energy_level = block_wavelet_energy_level;
-        offset_qindex = av1_compute_deltaq_from_energy_level(
-            cpi, block_wavelet_energy_level);
-      } else {
-        const int block_var_level =
-            av1_log_block_var(cpi, x, cm->seq_params.sb_size);
-        x->sb_energy_level = block_var_level;
-        offset_qindex =
-            av1_compute_deltaq_from_energy_level(cpi, block_var_level);
-      }
-      const int qmask = ~(cm->delta_q_res - 1);
-      int current_qindex = clamp(cm->base_qindex + offset_qindex,
-                                 cm->delta_q_res, 256 - cm->delta_q_res);
-      current_qindex =
-          ((current_qindex - cm->base_qindex + cm->delta_q_res / 2) & qmask) +
-          cm->base_qindex;
-      assert(current_qindex > 0);
-
-      xd->delta_qindex = current_qindex - cm->base_qindex;
-      set_offsets(cpi, tile_info, x, mi_row, mi_col, cm->seq_params.sb_size);
-      xd->mi[0]->current_qindex = current_qindex;
-      av1_init_plane_quantizers(cpi, x, xd->mi[0]->segment_id);
-      if (cpi->oxcf.deltaq_mode == DELTA_Q_LF) {
-        const int lfmask = ~(cm->delta_lf_res - 1);
-        const int delta_lf_from_base =
-            ((offset_qindex / 2 + cm->delta_lf_res / 2) & lfmask);
-
-        // pre-set the delta lf for loop filter. Note that this value is set
-        // before mi is assigned for each block in current superblock
-        for (int j = 0;
-             j < AOMMIN(cm->seq_params.mib_size, cm->mi_rows - mi_row); j++) {
-          for (int k = 0;
-               k < AOMMIN(cm->seq_params.mib_size, cm->mi_cols - mi_col); k++) {
-            cm->mi[(mi_row + j) * cm->mi_stride + (mi_col + k)]
-                .delta_lf_from_base =
-                clamp(delta_lf_from_base, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
-            const int frame_lf_count =
-                av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
-            for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) {
-              cm->mi[(mi_row + j) * cm->mi_stride + (mi_col + k)]
-                  .delta_lf[lf_id] =
-                  clamp(delta_lf_from_base, -MAX_LOOP_FILTER, MAX_LOOP_FILTER);
-            }
-          }
-        }
-      }
-    }
+    if (cm->delta_q_info.delta_q_present_flag)
+      setup_delta_q(cpi, x, tile_info, mi_row, mi_col, num_planes);
 
     int dummy_rate;
     int64_t dummy_dist;
@@ -4364,107 +5622,68 @@ static void encode_rd_sb_row(AV1_COMP *cpi, ThreadData *td,
     MB_MODE_INFO **mi = cm->mi_grid_visible + idx_str;
     x->source_variance = UINT_MAX;
     if (sf->partition_search_type == FIXED_PARTITION || seg_skip) {
-      set_offsets(cpi, tile_info, x, mi_row, mi_col, cm->seq_params.sb_size);
-      const BLOCK_SIZE bsize =
-          seg_skip ? cm->seq_params.sb_size : sf->always_this_block_size;
+      set_offsets(cpi, tile_info, x, mi_row, mi_col, sb_size);
+      const BLOCK_SIZE bsize = seg_skip ? sb_size : sf->always_this_block_size;
       set_fixed_partitioning(cpi, tile_info, mi, mi_row, mi_col, bsize);
-      rd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                       cm->seq_params.sb_size, &dummy_rate, &dummy_dist, 1,
-                       pc_root);
+      rd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col, sb_size,
+                       &dummy_rate, &dummy_dist, 1, pc_root);
     } else if (cpi->partition_search_skippable_frame) {
-      set_offsets(cpi, tile_info, x, mi_row, mi_col, cm->seq_params.sb_size);
+      set_offsets(cpi, tile_info, x, mi_row, mi_col, sb_size);
       const BLOCK_SIZE bsize =
           get_rd_var_based_fixed_partition(cpi, x, mi_row, mi_col);
       set_fixed_partitioning(cpi, tile_info, mi, mi_row, mi_col, bsize);
-      rd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col,
-                       cm->seq_params.sb_size, &dummy_rate, &dummy_dist, 1,
-                       pc_root);
+      rd_use_partition(cpi, td, tile_data, mi, tp, mi_row, mi_col, sb_size,
+                       &dummy_rate, &dummy_dist, 1, pc_root);
     } else {
+      const int orig_rdmult = cpi->rd.RDMULT;
+      x->cb_rdmult = orig_rdmult;
+      if (cpi->twopass.gf_group.index > 0 && cpi->oxcf.enable_tpl_model &&
+          cpi->oxcf.aq_mode == NO_AQ && cpi->oxcf.deltaq_mode == 0) {
+        const int dr =
+            get_rdmult_delta(cpi, BLOCK_128X128, mi_row, mi_col, orig_rdmult);
+
+        x->cb_rdmult = dr;
+        x->rdmult = x->cb_rdmult;
+      }
+
       // If required set upper and lower partition size limits
       if (sf->auto_min_max_partition_size) {
-        set_offsets(cpi, tile_info, x, mi_row, mi_col, cm->seq_params.sb_size);
+        set_offsets(cpi, tile_info, x, mi_row, mi_col, sb_size);
         rd_auto_partition_range(cpi, tile_info, xd, mi_row, mi_col,
                                 &x->min_partition_size, &x->max_partition_size);
       }
 
-      reset_partition(pc_root, cm->seq_params.sb_size);
+      reset_partition(pc_root, sb_size);
       x->use_cb_search_range = 0;
       init_first_partition_pass_stats_tables(x->first_partition_pass_stats);
       // Do the first pass if we need two pass partition search
       if (cpi->sf.two_pass_partition_search &&
           cpi->sf.use_square_partition_only_threshold > BLOCK_4X4 &&
-          mi_row + mi_size_high[cm->seq_params.sb_size] < cm->mi_rows &&
-          mi_col + mi_size_wide[cm->seq_params.sb_size] < cm->mi_cols &&
-          cm->frame_type != KEY_FRAME) {
-        x->cb_partition_scan = 1;
-        // Reset the stats tables.
-        if (sf->mode_pruning_based_on_two_pass_partition_search)
-          av1_zero(x->first_partition_pass_stats);
-        clear_pc_tree_stats(pc_root);
-        rd_pick_sqr_partition(cpi, td, tile_data, tp, mi_row, mi_col,
-                              cm->seq_params.sb_size, &dummy_rdc, INT64_MAX,
-                              pc_root, NULL);
-        x->cb_partition_scan = 0;
-
-        x->source_variance = UINT_MAX;
-        if (sf->adaptive_pred_interp_filter) {
-          for (int i = 0; i < leaf_nodes; ++i) {
-            td->pc_tree[i].vertical[0].pred_interp_filter = SWITCHABLE;
-            td->pc_tree[i].vertical[1].pred_interp_filter = SWITCHABLE;
-            td->pc_tree[i].horizontal[0].pred_interp_filter = SWITCHABLE;
-            td->pc_tree[i].horizontal[1].pred_interp_filter = SWITCHABLE;
-          }
-        }
-
-        x->mb_rd_record.num = x->mb_rd_record.index_start = 0;
-        av1_zero(x->txb_rd_record_8X8);
-        av1_zero(x->txb_rd_record_16X16);
-        av1_zero(x->txb_rd_record_32X32);
-        av1_zero(x->txb_rd_record_64X64);
-        av1_zero(x->txb_rd_record_intra);
-        av1_zero(x->pred_mv);
-        pc_root->index = 0;
-
-        for (int idy = 0; idy < mi_size_high[cm->seq_params.sb_size]; ++idy) {
-          for (int idx = 0; idx < mi_size_wide[cm->seq_params.sb_size]; ++idx) {
-            const int offset = cm->mi_stride * (mi_row + idy) + (mi_col + idx);
-            cm->mi_grid_visible[offset] = 0;
-          }
-        }
-
-        x->use_cb_search_range = 1;
-
-        if (sf->mode_pruning_based_on_two_pass_partition_search) {
-          for (int i = 0; i < FIRST_PARTITION_PASS_STATS_TABLES; ++i) {
-            FIRST_PARTITION_PASS_STATS *const stat =
-                &x->first_partition_pass_stats[i];
-            if (stat->sample_counts < FIRST_PARTITION_PASS_MIN_SAMPLES) {
-              // If there are not enough samples collected, make all available.
-              memset(stat->ref0_counts, 0xff, sizeof(stat->ref0_counts));
-              memset(stat->ref1_counts, 0xff, sizeof(stat->ref1_counts));
-            } else if (sf->selective_ref_frame < 2) {
-              // ALTREF2_FRAME and BWDREF_FRAME may be skipped during the
-              // initial partition scan, so we don't eliminate them.
-              stat->ref0_counts[ALTREF2_FRAME] = 0xff;
-              stat->ref1_counts[ALTREF2_FRAME] = 0xff;
-              stat->ref0_counts[BWDREF_FRAME] = 0xff;
-              stat->ref1_counts[BWDREF_FRAME] = 0xff;
-            }
-          }
-        }
+          mi_row + mi_size_high[sb_size] < cm->mi_rows &&
+          mi_col + mi_size_wide[sb_size] < cm->mi_cols &&
+          cm->current_frame.frame_type != KEY_FRAME) {
+        first_partition_search_pass(cpi, td, tile_data, mi_row, mi_col, tp);
       }
 
-      rd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col,
-                        cm->seq_params.sb_size, &dummy_rdc, INT64_MAX, pc_root,
+      rd_pick_partition(cpi, td, tile_data, tp, mi_row, mi_col, sb_size,
+                        sb_size, BLOCK_4X4, &dummy_rdc, INT64_MAX, pc_root,
                         NULL);
     }
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
     // TODO(angiebird): Let inter_mode_rd_model_estimation support multi-tile.
-    if (cpi->sf.inter_mode_rd_model_estimation && cm->tile_cols == 1 &&
+    if (cpi->sf.inter_mode_rd_model_estimation == 1 && cm->tile_cols == 1 &&
         cm->tile_rows == 1) {
       av1_inter_mode_data_fit(tile_data, x->rdmult);
     }
-#endif
+    if (tile_data->allow_update_cdf && (cpi->row_mt == 1) &&
+        (tile_info->mi_row_end > (mi_row + mib_size))) {
+      if (sb_cols_in_tile == 1)
+        memcpy(x->row_ctx, xd->tile_ctx, sizeof(*xd->tile_ctx));
+      else if (sb_col_in_tile >= 1)
+        memcpy(x->row_ctx + sb_col_in_tile - 1, xd->tile_ctx,
+               sizeof(*xd->tile_ctx));
+    }
+    (*(cpi->row_mt_sync_write_ptr))(&tile_data->row_mt_sync, sb_row,
+                                    sb_col_in_tile, sb_cols_in_tile);
   }
 }
 
@@ -4475,7 +5694,8 @@ static void init_encode_frame_mb_context(AV1_COMP *cpi) {
   MACROBLOCKD *const xd = &x->e_mbd;
 
   // Copy data over into macro block data structures.
-  av1_setup_src_planes(x, cpi->source, 0, 0, num_planes);
+  av1_setup_src_planes(x, cpi->source, 0, 0, num_planes,
+                       cm->seq_params.sb_size);
 
   av1_setup_block_planes(xd, cm->seq_params.subsampling_x,
                          cm->seq_params.subsampling_y, num_planes);
@@ -4527,7 +5747,6 @@ void av1_alloc_tile_data(AV1_COMP *cpi) {
       for (i = 0; i < BLOCK_SIZES_ALL; ++i) {
         for (j = 0; j < MAX_MODES; ++j) {
           tile_data->thresh_freq_fact[i][j] = 32;
-          tile_data->mode_map[i][j] = j;
         }
       }
     }
@@ -4561,6 +5780,7 @@ void av1_init_tile_data(AV1_COMP *cpi) {
       tile_data->allow_update_cdf = !cm->large_scale_tile;
       tile_data->allow_update_cdf =
           tile_data->allow_update_cdf && !cm->disable_cdf_update;
+      tile_data->tctx = *cm->fc;
     }
   }
 }
@@ -4573,14 +5793,12 @@ void av1_encode_sb_row(AV1_COMP *cpi, ThreadData *td, int tile_row,
   TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
   const TileInfo *const tile_info = &this_tile->tile_info;
   TOKENEXTRA *tok = NULL;
-  int sb_row_in_tile;
-  int tile_mb_cols = (tile_info->mi_col_end - tile_info->mi_col_start + 2) >> 2;
-
-  int num_mb_rows_in_sb =
-      ((1 << (cm->seq_params.mib_size_log2 + MI_SIZE_LOG2)) + 8) >> 4;
-
-  sb_row_in_tile =
+  const int sb_row_in_tile =
       (mi_row - tile_info->mi_row_start) >> cm->seq_params.mib_size_log2;
+  const int tile_mb_cols =
+      (tile_info->mi_col_end - tile_info->mi_col_start + 2) >> 2;
+  const int num_mb_rows_in_sb =
+      ((1 << (cm->seq_params.mib_size_log2 + MI_SIZE_LOG2)) + 8) >> 4;
 
   get_start_tok(cpi, tile_row, tile_col, mi_row, &tok,
                 cm->seq_params.mib_size_log2 + MI_SIZE_LOG2, num_planes);
@@ -4611,9 +5829,7 @@ void av1_encode_tile(AV1_COMP *cpi, ThreadData *td, int tile_row,
   const TileInfo *const tile_info = &this_tile->tile_info;
   int mi_row;
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
   av1_inter_mode_data_init(this_tile);
-#endif
 
   av1_zero_above_context(cm, &td->mb.e_mbd, tile_info->mi_col_start,
                          tile_info->mi_col_end, tile_row);
@@ -4624,15 +5840,11 @@ void av1_encode_tile(AV1_COMP *cpi, ThreadData *td, int tile_row,
   this_tile->ex_search_count = 0;  // Exhaustive mesh search hits.
   td->mb.m_search_count_ptr = &this_tile->m_search_count;
   td->mb.ex_search_count_ptr = &this_tile->ex_search_count;
-  this_tile->tctx = *cm->fc;
-  td->mb.e_mbd.tile_ctx = &this_tile->tctx;
 
   cfl_init(&td->mb.e_mbd.cfl, &cm->seq_params);
 
   av1_crc32c_calculator_init(&td->mb.mb_rd_record.crc_calculator);
 
-  td->intrabc_used_this_tile = 0;
-
   for (mi_row = tile_info->mi_row_start; mi_row < tile_info->mi_row_end;
        mi_row += cm->seq_params.mib_size) {
     av1_encode_sb_row(cpi, td, tile_row, tile_col, mi_row);
@@ -4652,8 +5864,13 @@ static void encode_tiles(AV1_COMP *cpi) {
 
   for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
     for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
+      TileDataEnc *const this_tile =
+          &cpi->tile_data[tile_row * cm->tile_cols + tile_col];
+      cpi->td.intrabc_used = 0;
+      cpi->td.mb.e_mbd.tile_ctx = &this_tile->tctx;
+      cpi->td.mb.tile_pb_ctx = &this_tile->tctx;
       av1_encode_tile(cpi, &cpi->td, tile_row, tile_col);
-      cpi->intrabc_used |= cpi->td.intrabc_used_this_tile;
+      cpi->intrabc_used |= cpi->td.intrabc_used;
     }
   }
 }
@@ -4661,8 +5878,9 @@ static void encode_tiles(AV1_COMP *cpi) {
 #if CONFIG_FP_MB_STATS
 static int input_fpmb_stats(FIRSTPASS_MB_STATS *firstpass_mb_stats,
                             AV1_COMMON *cm, uint8_t **this_frame_mb_stats) {
-  uint8_t *mb_stats_in = firstpass_mb_stats->mb_stats_start +
-                         cm->current_video_frame * cm->MBs * sizeof(uint8_t);
+  uint8_t *mb_stats_in =
+      firstpass_mb_stats->mb_stats_start +
+      cm->current_frame.frame_number * cm->MBs * sizeof(uint8_t);
 
   if (mb_stats_in > firstpass_mb_stats->mb_stats_end) return EOF;
 
@@ -4728,47 +5946,17 @@ static int do_gm_search_logic(SPEED_FEATURES *const sf, int num_refs_using_gm,
   (void)frame;
   switch (sf->gm_search_type) {
     case GM_FULL_SEARCH: return 1;
-    case GM_REDUCED_REF_SEARCH:
+    case GM_REDUCED_REF_SEARCH_SKIP_L2_L3:
       return !(frame == LAST2_FRAME || frame == LAST3_FRAME);
+    case GM_REDUCED_REF_SEARCH_SKIP_L2_L3_ARF2:
+      return !(frame == LAST2_FRAME || frame == LAST3_FRAME ||
+               (frame == ALTREF2_FRAME));
     case GM_DISABLE_SEARCH: return 0;
     default: assert(0);
   }
   return 1;
 }
 
-// Estimate if the source frame is screen content, based on the portion of
-// blocks that have no more than 4 (experimentally selected) luma colors.
-static int is_screen_content(const uint8_t *src, int use_hbd, int bd,
-                             int stride, int width, int height) {
-  assert(src != NULL);
-  int counts = 0;
-  const int blk_w = 16;
-  const int blk_h = 16;
-  const int limit = 4;
-  for (int r = 0; r + blk_h <= height; r += blk_h) {
-    for (int c = 0; c + blk_w <= width; c += blk_w) {
-      int count_buf[1 << 12];  // Maximum (1 << 12) color levels.
-      const int n_colors =
-          use_hbd ? av1_count_colors_highbd(src + r * stride + c, stride, blk_w,
-                                            blk_h, bd, count_buf)
-                  : av1_count_colors(src + r * stride + c, stride, blk_w, blk_h,
-                                     count_buf);
-      if (n_colors > 1 && n_colors <= limit) counts++;
-    }
-  }
-  // The threshold is 10%.
-  return counts * blk_h * blk_w * 10 > width * height;
-}
-
-static const uint8_t ref_frame_flag_list[REF_FRAMES] = { 0,
-                                                         AOM_LAST_FLAG,
-                                                         AOM_LAST2_FLAG,
-                                                         AOM_LAST3_FLAG,
-                                                         AOM_GOLD_FLAG,
-                                                         AOM_BWD_FLAG,
-                                                         AOM_ALT2_FLAG,
-                                                         AOM_ALT_FLAG };
-
 // Enforce the number of references for each arbitrary frame limited to
 // (INTER_REFS_PER_FRAME - 1)
 static void enforce_max_ref_frames(AV1_COMP *cpi) {
@@ -4788,40 +5976,41 @@ static void enforce_max_ref_frames(AV1_COMP *cpi) {
   //     the same quality level, remove the earliest reference frame.
 
   if (total_valid_refs == INTER_REFS_PER_FRAME) {
-    unsigned int min_ref_offset = UINT_MAX;
-    unsigned int second_min_ref_offset = UINT_MAX;
+    unsigned int min_ref_order_hint = UINT_MAX;
+    unsigned int second_min_ref_order_hint = UINT_MAX;
     MV_REFERENCE_FRAME earliest_ref_frames[2] = { LAST3_FRAME, LAST2_FRAME };
-    int earliest_buf_idxes[2] = { 0 };
+    const RefCntBuffer *earliest_bufs[2] = { NULL };
 
     // Locate the earliest two reference frames except GOLDEN/ALTREF.
     for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
       // Retain GOLDEN/ALTERF
       if (ref_frame == GOLDEN_FRAME || ref_frame == ALTREF_FRAME) continue;
 
-      const int buf_idx = cm->frame_refs[ref_frame - LAST_FRAME].idx;
-      if (buf_idx >= 0) {
-        const unsigned int ref_offset =
-            cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset;
+      const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+      if (buf != NULL) {
+        const unsigned int ref_order_hint = buf->order_hint;
 
-        if (min_ref_offset == UINT_MAX) {
-          min_ref_offset = ref_offset;
+        if (min_ref_order_hint == UINT_MAX) {
+          min_ref_order_hint = ref_order_hint;
           earliest_ref_frames[0] = ref_frame;
-          earliest_buf_idxes[0] = buf_idx;
+          earliest_bufs[0] = buf;
         } else {
-          if (get_relative_dist(cm, ref_offset, min_ref_offset) < 0) {
-            second_min_ref_offset = min_ref_offset;
+          if (get_relative_dist(&cm->seq_params.order_hint_info, ref_order_hint,
+                                min_ref_order_hint) < 0) {
+            second_min_ref_order_hint = min_ref_order_hint;
             earliest_ref_frames[1] = earliest_ref_frames[0];
-            earliest_buf_idxes[1] = earliest_buf_idxes[0];
+            earliest_bufs[1] = earliest_bufs[0];
 
-            min_ref_offset = ref_offset;
+            min_ref_order_hint = ref_order_hint;
             earliest_ref_frames[0] = ref_frame;
-            earliest_buf_idxes[0] = buf_idx;
-          } else if (second_min_ref_offset == UINT_MAX ||
-                     get_relative_dist(cm, ref_offset, second_min_ref_offset) <
-                         0) {
-            second_min_ref_offset = ref_offset;
+            earliest_bufs[0] = buf;
+          } else if (second_min_ref_order_hint == UINT_MAX ||
+                     get_relative_dist(&cm->seq_params.order_hint_info,
+                                       ref_order_hint,
+                                       second_min_ref_order_hint) < 0) {
+            second_min_ref_order_hint = ref_order_hint;
             earliest_ref_frames[1] = ref_frame;
-            earliest_buf_idxes[1] = buf_idx;
+            earliest_bufs[1] = buf;
           }
         }
       }
@@ -4830,7 +6019,7 @@ static void enforce_max_ref_frames(AV1_COMP *cpi) {
     RATE_FACTOR_LEVEL ref_rf_level[2];
     double ref_rf_deltas[2];
     for (int i = 0; i < 2; ++i) {
-      ref_rf_level[i] = cpi->frame_rf_level[earliest_buf_idxes[i]];
+      ref_rf_level[i] = earliest_bufs[i]->frame_rf_level;
       ref_rf_deltas[i] = rate_factor_deltas[ref_rf_level[i]];
     }
     (void)ref_rf_level;
@@ -4865,13 +6054,13 @@ static INLINE int av1_refs_are_one_sided(const AV1_COMMON *cm) {
   assert(!frame_is_intra_only(cm));
 
   int one_sided_refs = 1;
-  for (int ref = 0; ref < INTER_REFS_PER_FRAME; ++ref) {
-    const int buf_idx = cm->frame_refs[ref].idx;
-    if (buf_idx == INVALID_IDX) continue;
+  for (int ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref) {
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref);
+    if (buf == NULL) continue;
 
-    const int ref_offset =
-        cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset;
-    if (get_relative_dist(cm, ref_offset, (int)cm->frame_offset) > 0) {
+    const int ref_order_hint = buf->order_hint;
+    if (get_relative_dist(&cm->seq_params.order_hint_info, ref_order_hint,
+                          (int)cm->current_frame.order_hint) > 0) {
       one_sided_refs = 0;  // bwd reference
       break;
     }
@@ -4880,31 +6069,36 @@ static INLINE int av1_refs_are_one_sided(const AV1_COMMON *cm) {
 }
 
 static INLINE void get_skip_mode_ref_offsets(const AV1_COMMON *cm,
-                                             int ref_offset[2]) {
-  ref_offset[0] = ref_offset[1] = 0;
-  if (!cm->is_skip_mode_allowed) return;
+                                             int ref_order_hint[2]) {
+  const SkipModeInfo *const skip_mode_info = &cm->current_frame.skip_mode_info;
+  ref_order_hint[0] = ref_order_hint[1] = 0;
+  if (!skip_mode_info->skip_mode_allowed) return;
 
-  const int buf_idx_0 = cm->frame_refs[cm->ref_frame_idx_0].idx;
-  const int buf_idx_1 = cm->frame_refs[cm->ref_frame_idx_1].idx;
-  assert(buf_idx_0 != INVALID_IDX && buf_idx_1 != INVALID_IDX);
+  const RefCntBuffer *const buf_0 =
+      get_ref_frame_buf(cm, LAST_FRAME + skip_mode_info->ref_frame_idx_0);
+  const RefCntBuffer *const buf_1 =
+      get_ref_frame_buf(cm, LAST_FRAME + skip_mode_info->ref_frame_idx_1);
+  assert(buf_0 != NULL && buf_1 != NULL);
 
-  ref_offset[0] = cm->buffer_pool->frame_bufs[buf_idx_0].cur_frame_offset;
-  ref_offset[1] = cm->buffer_pool->frame_bufs[buf_idx_1].cur_frame_offset;
+  ref_order_hint[0] = buf_0->order_hint;
+  ref_order_hint[1] = buf_1->order_hint;
 }
 
 static int check_skip_mode_enabled(AV1_COMP *const cpi) {
   AV1_COMMON *const cm = &cpi->common;
 
   av1_setup_skip_mode_allowed(cm);
-  if (!cm->is_skip_mode_allowed) return 0;
+  if (!cm->current_frame.skip_mode_info.skip_mode_allowed) return 0;
 
   // Turn off skip mode if the temporal distances of the reference pair to the
   // current frame are different by more than 1 frame.
-  const int cur_offset = (int)cm->frame_offset;
+  const int cur_offset = (int)cm->current_frame.order_hint;
   int ref_offset[2];
   get_skip_mode_ref_offsets(cm, ref_offset);
-  const int cur_to_ref0 = get_relative_dist(cm, cur_offset, ref_offset[0]);
-  const int cur_to_ref1 = abs(get_relative_dist(cm, cur_offset, ref_offset[1]));
+  const int cur_to_ref0 = get_relative_dist(&cm->seq_params.order_hint_info,
+                                            cur_offset, ref_offset[0]);
+  const int cur_to_ref1 = abs(get_relative_dist(&cm->seq_params.order_hint_info,
+                                                cur_offset, ref_offset[1]));
   if (abs(cur_to_ref0 - cur_to_ref1) > 1) return 0;
 
   // High Latency: Turn off skip mode if all refs are fwd.
@@ -4918,8 +6112,10 @@ static int check_skip_mode_enabled(AV1_COMP *const cpi) {
                                              AOM_BWD_FLAG,
                                              AOM_ALT2_FLAG,
                                              AOM_ALT_FLAG };
-  const int ref_frame[2] = { cm->ref_frame_idx_0 + LAST_FRAME,
-                             cm->ref_frame_idx_1 + LAST_FRAME };
+  const int ref_frame[2] = {
+    cm->current_frame.skip_mode_info.ref_frame_idx_0 + LAST_FRAME,
+    cm->current_frame.skip_mode_info.ref_frame_idx_1 + LAST_FRAME
+  };
   if (!(cpi->ref_frame_flags & flag_list[ref_frame[0]]) ||
       !(cpi->ref_frame_flags & flag_list[ref_frame[1]]))
     return 0;
@@ -4933,8 +6129,9 @@ static INLINE int skip_gm_frame(AV1_COMMON *const cm, int ref_frame) {
   if ((ref_frame == LAST3_FRAME || ref_frame == LAST2_FRAME) &&
       cm->global_motion[GOLDEN_FRAME].wmtype != IDENTITY) {
     return get_relative_dist(
-               cm, cm->cur_frame->ref_frame_offset[ref_frame - LAST_FRAME],
-               cm->cur_frame->ref_frame_offset[GOLDEN_FRAME - LAST_FRAME]) <= 0;
+               &cm->seq_params.order_hint_info,
+               cm->cur_frame->ref_order_hints[ref_frame - LAST_FRAME],
+               cm->cur_frame->ref_order_hints[GOLDEN_FRAME - LAST_FRAME]) <= 0;
   }
   return 0;
 }
@@ -4968,20 +6165,6 @@ static void encode_frame_internal(AV1_COMP *cpi) {
   av1_zero(*td->counts);
   av1_zero(rdc->comp_pred_diff);
 
-  if (frame_is_intra_only(cm)) {
-    if (cm->seq_params.force_screen_content_tools == 2) {
-      cm->allow_screen_content_tools =
-          cpi->oxcf.content == AOM_CONTENT_SCREEN ||
-          is_screen_content(cpi->source->y_buffer,
-                            cpi->source->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
-                            cpi->source->y_stride, cpi->source->y_width,
-                            cpi->source->y_height);
-    } else {
-      cm->allow_screen_content_tools =
-          cm->seq_params.force_screen_content_tools;
-    }
-  }
-
   // Allow intrabc when screen content tools are enabled.
   cm->allow_intrabc = cm->allow_screen_content_tools;
   // Reset the flag.
@@ -5084,15 +6267,42 @@ static void encode_frame_internal(AV1_COMP *cpi) {
   cm->tx_mode = select_tx_mode(cpi);
 
   // Fix delta q resolution for the moment
-  cm->delta_q_res = DEFAULT_DELTA_Q_RES;
+  cm->delta_q_info.delta_q_res = DEFAULT_DELTA_Q_RES;
   // Set delta_q_present_flag before it is used for the first time
-  cm->delta_lf_res = DEFAULT_DELTA_LF_RES;
-  cm->delta_q_present_flag = cpi->oxcf.deltaq_mode != NO_DELTA_Q;
-  cm->delta_lf_present_flag = cpi->oxcf.deltaq_mode == DELTA_Q_LF;
-  cm->delta_lf_multi = DEFAULT_DELTA_LF_MULTI;
-  // update delta_q_present_flag and delta_lf_present_flag based on base_qindex
-  cm->delta_q_present_flag &= cm->base_qindex > 0;
-  cm->delta_lf_present_flag &= cm->base_qindex > 0;
+  cm->delta_q_info.delta_lf_res = DEFAULT_DELTA_LF_RES;
+  cm->delta_q_info.delta_q_present_flag = cpi->oxcf.deltaq_mode != NO_DELTA_Q;
+  cm->delta_q_info.delta_lf_present_flag = cpi->oxcf.deltaq_mode == DELTA_Q_LF;
+  cm->delta_q_info.delta_lf_multi = DEFAULT_DELTA_LF_MULTI;
+  // update delta_q_present_flag and delta_lf_present_flag based on
+  // base_qindex
+  cm->delta_q_info.delta_q_present_flag &= cm->base_qindex > 0;
+  cm->delta_q_info.delta_lf_present_flag &= cm->base_qindex > 0;
+
+  if (cpi->twopass.gf_group.index &&
+      cpi->twopass.gf_group.index < MAX_LAG_BUFFERS &&
+      cpi->oxcf.enable_tpl_model) {
+    TplDepFrame *tpl_frame = &cpi->tpl_stats[cpi->twopass.gf_group.index];
+    TplDepStats *tpl_stats = tpl_frame->tpl_stats_ptr;
+
+    int tpl_stride = tpl_frame->stride;
+    int64_t intra_cost_base = 0;
+    int64_t mc_dep_cost_base = 0;
+    int row, col;
+
+    for (row = 0; row < cm->mi_rows; ++row) {
+      for (col = 0; col < cm->mi_cols; ++col) {
+        TplDepStats *this_stats = &tpl_stats[row * tpl_stride + col];
+        intra_cost_base += this_stats->intra_cost;
+        mc_dep_cost_base += this_stats->mc_dep_cost;
+      }
+    }
+
+    aom_clear_system_state();
+
+    if (tpl_frame->is_valid)
+      cpi->rd.r0 =
+          (double)intra_cost_base / (intra_cost_base + mc_dep_cost_base);
+  }
 
   av1_frame_init_quantizer(cpi);
 
@@ -5104,7 +6314,6 @@ static void encode_frame_internal(AV1_COMP *cpi) {
     cm->last_frame_seg_map = cm->prev_frame->seg_map;
   else
     cm->last_frame_seg_map = NULL;
-  cm->current_frame_seg_map = cm->cur_frame->seg_map;
   if (cm->allow_intrabc || cm->coded_lossless) {
     av1_set_default_ref_deltas(cm->lf.ref_deltas);
     av1_set_default_mode_deltas(cm->lf.mode_deltas);
@@ -5120,27 +6329,31 @@ static void encode_frame_internal(AV1_COMP *cpi) {
   cm->prev_mi = cm->allow_ref_frame_mvs ? cm->prev_mip : NULL;
 
   x->txb_split_count = 0;
+#if CONFIG_SPEED_STATS
+  x->tx_search_count = 0;
+#endif  // CONFIG_SPEED_STATS
 
   av1_zero(rdc->global_motion_used);
   av1_zero(cpi->gmparams_cost);
-#if !CONFIG_GLOBAL_MOTION_SEARCH
-  cpi->global_motion_search_done = 1;
-#endif  // !CONFIG_GLOBAL_MOTION_SEARCH
-  if (cpi->common.frame_type == INTER_FRAME && cpi->source &&
-      !cpi->global_motion_search_done) {
+  if (cpi->common.current_frame.frame_type == INTER_FRAME && cpi->source &&
+      cpi->oxcf.enable_global_motion && !cpi->global_motion_search_done) {
     YV12_BUFFER_CONFIG *ref_buf[REF_FRAMES];
     int frame;
     double params_by_motion[RANSAC_NUM_MOTIONS * (MAX_PARAMDIM - 1)];
     const double *params_this_motion;
     int inliers_by_motion[RANSAC_NUM_MOTIONS];
     WarpedMotionParams tmp_wm_params;
+    // clang-format off
     static const double kIdentityParams[MAX_PARAMDIM - 1] = {
       0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0
     };
+    // clang-format on
     int num_refs_using_gm = 0;
 
     for (frame = ALTREF_FRAME; frame >= LAST_FRAME; --frame) {
-      ref_buf[frame] = get_ref_frame_buffer(cpi, frame);
+      ref_buf[frame] = NULL;
+      RefCntBuffer *buf = get_ref_frame_buf(cm, frame);
+      if (buf != NULL) ref_buf[frame] = &buf->buf;
       int pframe;
       cm->global_motion[frame] = default_warp_params;
       const WarpedMotionParams *ref_params =
@@ -5159,11 +6372,10 @@ static void encode_frame_internal(AV1_COMP *cpi) {
                  do_gm_search_logic(&cpi->sf, num_refs_using_gm, frame) &&
                  !(cpi->sf.selective_ref_gm && skip_gm_frame(cm, frame))) {
         TransformationType model;
-        const int64_t ref_frame_error =
-            av1_frame_error(xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
-                            ref_buf[frame]->y_buffer, ref_buf[frame]->y_stride,
-                            cpi->source->y_buffer, cpi->source->y_width,
-                            cpi->source->y_height, cpi->source->y_stride);
+        const int64_t ref_frame_error = av1_frame_error(
+            is_cur_buf_hbd(xd), xd->bd, ref_buf[frame]->y_buffer,
+            ref_buf[frame]->y_stride, cpi->source->y_buffer,
+            cpi->source->y_width, cpi->source->y_height, cpi->source->y_stride);
 
         if (ref_frame_error == 0) continue;
 
@@ -5176,30 +6388,30 @@ static void encode_frame_internal(AV1_COMP *cpi) {
                    (MAX_PARAMDIM - 1) * sizeof(*params_by_motion));
           }
 
-          compute_global_motion_feature_based(
-              model, cpi->source, ref_buf[frame],
-              cpi->common.seq_params.bit_depth, inliers_by_motion,
-              params_by_motion, RANSAC_NUM_MOTIONS);
+          av1_compute_global_motion(model, cpi->source, ref_buf[frame],
+                                    cpi->common.seq_params.bit_depth,
+                                    inliers_by_motion, params_by_motion,
+                                    RANSAC_NUM_MOTIONS);
 
           for (i = 0; i < RANSAC_NUM_MOTIONS; ++i) {
             if (inliers_by_motion[i] == 0) continue;
 
             params_this_motion = params_by_motion + (MAX_PARAMDIM - 1) * i;
-            convert_model_to_params(params_this_motion, &tmp_wm_params);
+            av1_convert_model_to_params(params_this_motion, &tmp_wm_params);
 
             if (tmp_wm_params.wmtype != IDENTITY) {
-              const int64_t warp_error = refine_integerized_param(
-                  &tmp_wm_params, tmp_wm_params.wmtype,
-                  xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
-                  ref_buf[frame]->y_buffer, ref_buf[frame]->y_width,
+              const int64_t warp_error = av1_refine_integerized_param(
+                  &tmp_wm_params, tmp_wm_params.wmtype, is_cur_buf_hbd(xd),
+                  xd->bd, ref_buf[frame]->y_buffer, ref_buf[frame]->y_width,
                   ref_buf[frame]->y_height, ref_buf[frame]->y_stride,
                   cpi->source->y_buffer, cpi->source->y_width,
                   cpi->source->y_height, cpi->source->y_stride, 5,
                   best_warp_error);
               if (warp_error < best_warp_error) {
                 best_warp_error = warp_error;
-                // Save the wm_params modified by refine_integerized_param()
-                // rather than motion index to avoid rerunning refine() below.
+                // Save the wm_params modified by
+                // av1_refine_integerized_param() rather than motion index to
+                // avoid rerunning refine() below.
                 memcpy(&(cm->global_motion[frame]), &tmp_wm_params,
                        sizeof(WarpedMotionParams));
               }
@@ -5222,7 +6434,7 @@ static void encode_frame_internal(AV1_COMP *cpi) {
 
           // If the best error advantage found doesn't meet the threshold for
           // this motion type, revert to IDENTITY.
-          if (!is_enough_erroradvantage(
+          if (!av1_is_enough_erroradvantage(
                   (double)best_warp_error / ref_frame_error,
                   gm_get_params_cost(&cm->global_motion[frame], ref_params,
                                      cm->allow_high_precision_mv),
@@ -5259,12 +6471,10 @@ static void encode_frame_internal(AV1_COMP *cpi) {
   cpi->all_one_sided_refs =
       frame_is_intra_only(cm) ? 0 : av1_refs_are_one_sided(cm);
 
-  cm->skip_mode_flag = check_skip_mode_enabled(cpi);
+  cm->current_frame.skip_mode_info.skip_mode_flag =
+      check_skip_mode_enabled(cpi);
 
   {
-    struct aom_usec_timer emr_timer;
-    aom_usec_timer_start(&emr_timer);
-
 #if CONFIG_FP_MB_STATS
     if (cpi->use_fp_mb_stats) {
       input_fpmb_stats(&cpi->twopass.firstpass_mb_stats, cm,
@@ -5272,41 +6482,34 @@ static void encode_frame_internal(AV1_COMP *cpi) {
     }
 #endif
 
-    if (cpi->row_mt && (cpi->oxcf.max_threads > 1))
-      av1_encode_tiles_mt(cpi);
-    else if (AOMMIN(cpi->oxcf.max_threads, cm->tile_cols * cm->tile_rows) > 1)
-      av1_encode_tiles_mt(cpi);
-    else
-      encode_tiles(cpi);
-
-    aom_usec_timer_mark(&emr_timer);
-    cpi->time_encode_sb_row += aom_usec_timer_elapsed(&emr_timer);
+    cpi->row_mt_sync_read_ptr = av1_row_mt_sync_read_dummy;
+    cpi->row_mt_sync_write_ptr = av1_row_mt_sync_write_dummy;
+    cpi->row_mt = 0;
+    if (cpi->oxcf.row_mt && (cpi->oxcf.max_threads > 1)) {
+      cpi->row_mt = 1;
+      cpi->row_mt_sync_read_ptr = av1_row_mt_sync_read;
+      cpi->row_mt_sync_write_ptr = av1_row_mt_sync_write;
+      av1_encode_tiles_row_mt(cpi);
+    } else {
+      if (AOMMIN(cpi->oxcf.max_threads, cm->tile_cols * cm->tile_rows) > 1)
+        av1_encode_tiles_mt(cpi);
+      else
+        encode_tiles(cpi);
+    }
   }
 
   // If intrabc is allowed but never selected, reset the allow_intrabc flag.
   if (cm->allow_intrabc && !cpi->intrabc_used) cm->allow_intrabc = 0;
-  if (cm->allow_intrabc) cm->delta_lf_present_flag = 0;
+  if (cm->allow_intrabc) cm->delta_q_info.delta_lf_present_flag = 0;
 }
 
 void av1_encode_frame(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
+  CurrentFrame *const current_frame = &cm->current_frame;
   const int num_planes = av1_num_planes(cm);
   // Indicates whether or not to use a default reduced set for ext-tx
   // rather than the potential full set of 16 transforms
-  cm->reduced_tx_set_used = 0;
-
-  if (cm->show_frame == 0) {
-    int arf_offset = AOMMIN(
-        (MAX_GF_INTERVAL - 1),
-        cpi->twopass.gf_group.arf_src_offset[cpi->twopass.gf_group.index]);
-    int brf_offset =
-        cpi->twopass.gf_group.brf_src_offset[cpi->twopass.gf_group.index];
-    arf_offset = AOMMIN((MAX_GF_INTERVAL - 1), arf_offset + brf_offset);
-    cm->frame_offset = cm->current_video_frame + arf_offset;
-  } else {
-    cm->frame_offset = cm->current_video_frame;
-  }
-  cm->frame_offset %= (1 << (cm->seq_params.order_hint_bits_minus_1 + 1));
+  cm->reduced_tx_set_used = cpi->oxcf.reduced_tx_type_set;
 
   // Make sure segment_id is no larger than last_active_segid.
   if (cm->seg.enabled && cm->seg.update_map) {
@@ -5323,7 +6526,9 @@ void av1_encode_frame(AV1_COMP *cpi) {
   }
 
   av1_setup_frame_buf_refs(cm);
-  if (cpi->sf.selective_ref_frame >= 2) enforce_max_ref_frames(cpi);
+  if (cpi->sf.selective_ref_frame >= 3 && cpi->oxcf.max_reference_frames == 7) {
+    enforce_max_ref_frames(cpi);
+  }
   av1_setup_frame_sign_bias(cm);
 
 #if CONFIG_MISMATCH_DEBUG
@@ -5332,8 +6537,6 @@ void av1_encode_frame(AV1_COMP *cpi) {
   (void)num_planes;
 #endif
 
-  cpi->allow_comp_inter_inter = !frame_is_intra_only(cm);
-
   if (cpi->sf.frame_parameter_update) {
     int i;
     RD_OPT *const rd_opt = &cpi->rd;
@@ -5355,10 +6558,10 @@ void av1_encode_frame(AV1_COMP *cpi) {
 
     /* prediction (compound, single or hybrid) mode selection */
     // NOTE: "is_alt_ref" is true only for OVERLAY/INTNL_OVERLAY frames
-    if (is_alt_ref || !cpi->allow_comp_inter_inter)
-      cm->reference_mode = SINGLE_REFERENCE;
+    if (is_alt_ref || frame_is_intra_only(cm))
+      current_frame->reference_mode = SINGLE_REFERENCE;
     else
-      cm->reference_mode = REFERENCE_MODE_SELECT;
+      current_frame->reference_mode = REFERENCE_MODE_SELECT;
 
     cm->interp_filter = SWITCHABLE;
     if (cm->large_scale_tile) cm->interp_filter = EIGHTTAP_REGULAR;
@@ -5373,22 +6576,25 @@ void av1_encode_frame(AV1_COMP *cpi) {
     for (i = 0; i < REFERENCE_MODES; ++i)
       mode_thrs[i] = (mode_thrs[i] + rdc->comp_pred_diff[i] / cm->MBs) / 2;
 
-    if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+    if (current_frame->reference_mode == REFERENCE_MODE_SELECT) {
       // Use a flag that includes 4x4 blocks
       if (rdc->compound_ref_used_flag == 0) {
-        cm->reference_mode = SINGLE_REFERENCE;
+        current_frame->reference_mode = SINGLE_REFERENCE;
 #if CONFIG_ENTROPY_STATS
         av1_zero(cpi->td.counts->comp_inter);
 #endif  // CONFIG_ENTROPY_STATS
       }
     }
-    // Re-check on the skip mode status as reference mode may have been changed.
-    if (frame_is_intra_only(cm) || cm->reference_mode == SINGLE_REFERENCE) {
-      cm->is_skip_mode_allowed = 0;
-      cm->skip_mode_flag = 0;
+    // Re-check on the skip mode status as reference mode may have been
+    // changed.
+    SkipModeInfo *const skip_mode_info = &current_frame->skip_mode_info;
+    if (frame_is_intra_only(cm) ||
+        current_frame->reference_mode == SINGLE_REFERENCE) {
+      skip_mode_info->skip_mode_allowed = 0;
+      skip_mode_info->skip_mode_flag = 0;
     }
-    if (cm->skip_mode_flag && rdc->skip_mode_used_flag == 0)
-      cm->skip_mode_flag = 0;
+    if (skip_mode_info->skip_mode_flag && rdc->skip_mode_used_flag == 0)
+      skip_mode_info->skip_mode_flag = 0;
 
     if (!cm->large_scale_tile) {
       if (cm->tx_mode == TX_MODE_SELECT && cpi->td.mb.txb_split_count == 0)
@@ -5625,15 +6831,19 @@ static void encode_superblock(const AV1_COMP *const cpi, TileDataEnc *tile_data,
 
     set_ref_ptrs(cm, xd, mbmi->ref_frame[0], mbmi->ref_frame[1]);
     for (ref = 0; ref < 1 + is_compound; ++ref) {
-      YV12_BUFFER_CONFIG *cfg = get_ref_frame_buffer(cpi, mbmi->ref_frame[ref]);
+      const YV12_BUFFER_CONFIG *cfg =
+          get_ref_frame_yv12_buf(cm, mbmi->ref_frame[ref]);
       assert(IMPLIES(!is_intrabc_block(mbmi), cfg));
       av1_setup_pre_planes(xd, ref, cfg, mi_row, mi_col,
-                           &xd->block_refs[ref]->sf, num_planes);
+                           xd->block_ref_scale_factors[ref], num_planes);
     }
 
-    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
-    if (mbmi->motion_mode == OBMC_CAUSAL)
+    av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize, 0,
+                                  av1_num_planes(cm) - 1);
+    if (mbmi->motion_mode == OBMC_CAUSAL) {
+      assert(cpi->oxcf.enable_obmc == 1);
       av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
+    }
 
 #if CONFIG_MISMATCH_DEBUG
     if (dry_run == OUTPUT_ENABLED) {
@@ -5645,9 +6855,9 @@ static void encode_superblock(const AV1_COMP *const cpi, TileDataEnc *tile_data,
         if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
                                  pd->subsampling_y))
           continue;
-        mismatch_record_block_pre(pd->dst.buf, pd->dst.stride, cm->frame_offset,
-                                  plane, pixel_c, pixel_r, pd->width,
-                                  pd->height,
+        mismatch_record_block_pre(pd->dst.buf, pd->dst.stride,
+                                  cm->current_frame.order_hint, plane, pixel_c,
+                                  pixel_r, pd->width, pd->height,
                                   xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
       }
     }
@@ -5661,8 +6871,7 @@ static void encode_superblock(const AV1_COMP *const cpi, TileDataEnc *tile_data,
   }
 
   if (!dry_run) {
-    if (av1_allow_intrabc(cm) && is_intrabc_block(mbmi))
-      td->intrabc_used_this_tile = 1;
+    if (av1_allow_intrabc(cm) && is_intrabc_block(mbmi)) td->intrabc_used = 1;
     if (cm->tx_mode == TX_MODE_SELECT && !xd->lossless[mbmi->segment_id] &&
         mbmi->sb_type > BLOCK_4X4 && !(is_inter && (mbmi->skip || seg_skip))) {
       if (is_inter) {
diff --git a/third_party/aom/av1/encoder/encodeframe.h b/third_party/aom/av1/encoder/encodeframe.h
index e8cf9b4685..3b1730d323 100644
--- a/third_party/aom/av1/encoder/encodeframe.h
+++ b/third_party/aom/av1/encoder/encodeframe.h
@@ -29,7 +29,7 @@ struct ThreadData;
 
 void av1_setup_src_planes(struct macroblock *x,
                           const struct yv12_buffer_config *src, int mi_row,
-                          int mi_col, const int num_planes);
+                          int mi_col, const int num_planes, BLOCK_SIZE bsize);
 
 void av1_encode_frame(struct AV1_COMP *cpi);
 
diff --git a/third_party/aom/av1/encoder/encodemb.c b/third_party/aom/av1/encoder/encodemb.c
index ad12577e6e..6edb0f88cd 100644
--- a/third_party/aom/av1/encoder/encodemb.c
+++ b/third_party/aom/av1/encoder/encodemb.c
@@ -43,7 +43,7 @@ static void subtract_block(const MACROBLOCKD *xd, int rows, int cols,
                            const uint8_t *src8, ptrdiff_t src_stride,
                            const uint8_t *pred8, ptrdiff_t pred_stride) {
   if (check_subtract_block_size(rows, cols)) {
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       aom_highbd_subtract_block_c(rows, cols, diff, diff_stride, src8,
                                   src_stride, pred8, pred_stride, xd->bd);
       return;
@@ -54,7 +54,7 @@ static void subtract_block(const MACROBLOCKD *xd, int rows, int cols,
     return;
   }
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     aom_highbd_subtract_block(rows, cols, diff, diff_stride, src8, src_stride,
                               pred8, pred_stride, xd->bd);
     return;
@@ -116,11 +116,11 @@ int av1_optimize_b(const struct AV1_COMP *cpi, MACROBLOCK *mb, int plane,
                               rate_cost, cpi->oxcf.sharpness);
 }
 
-typedef enum QUANT_FUNC {
+enum {
   QUANT_FUNC_LOWBD = 0,
   QUANT_FUNC_HIGHBD = 1,
   QUANT_FUNC_TYPES = 2
-} QUANT_FUNC;
+} UENUM1BYTE(QUANT_FUNC);
 
 static AV1_QUANT_FACADE
     quant_func_list[AV1_XFORM_QUANT_TYPES][QUANT_FUNC_TYPES] = {
@@ -163,6 +163,7 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
   qparam.tx_size = tx_size;
   qparam.qmatrix = qmatrix;
   qparam.iqmatrix = iqmatrix;
+  qparam.use_quant_b_adapt = cm->use_quant_b_adapt;
   TxfmParam txfm_param;
   txfm_param.tx_type = tx_type;
   txfm_param.tx_size = tx_size;
@@ -171,7 +172,7 @@ void av1_xform_quant(const AV1_COMMON *cm, MACROBLOCK *x, int plane, int block,
       txfm_param.tx_size, is_inter_block(mbmi), cm->reduced_tx_set_used);
 
   txfm_param.bd = xd->bd;
-  txfm_param.is_hbd = get_bitdepth_data_path_index(xd);
+  txfm_param.is_hbd = is_cur_buf_hbd(xd);
 
   av1_fwd_txfm(src_diff, coeff, diff_stride, &txfm_param);
 
@@ -226,7 +227,7 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
   if (!is_blk_skip(x, plane, blk_row * bw + blk_col) && !mbmi->skip_mode) {
     TX_TYPE tx_type = av1_get_tx_type(pd->plane_type, xd, blk_row, blk_col,
                                       tx_size, cm->reduced_tx_set_used);
-    if (args->enable_optimize_b) {
+    if (args->enable_optimize_b != NO_TRELLIS_OPT) {
       av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                       tx_size, tx_type, AV1_XFORM_QUANT_FP);
       TXB_CTX txb_ctx;
@@ -256,11 +257,11 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
   }
 
   if (p->eobs[block] == 0 && plane == 0) {
-  // TODO(debargha, jingning): Temporarily disable txk_type check for eob=0
-  // case. It is possible that certain collision in hash index would cause
-  // the assertion failure. To further optimize the rate-distortion
-  // performance, we need to re-visit this part and enable this assert
-  // again.
+    // TODO(debargha, jingning): Temporarily disable txk_type check for eob=0
+    // case. It is possible that certain collision in hash index would cause
+    // the assertion failure. To further optimize the rate-distortion
+    // performance, we need to re-visit this part and enable this assert
+    // again.
 #if 0
     if (args->cpi->oxcf.aq_mode == NO_AQ &&
         args->cpi->oxcf.deltaq_mode == NO_DELTA_Q) {
@@ -285,8 +286,8 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
     int blk_h = block_size_high[bsize];
     mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, blk_col, blk_row,
                     pd->subsampling_x, pd->subsampling_y);
-    mismatch_record_block_tx(dst, pd->dst.stride, cm->frame_offset, plane,
-                             pixel_c, pixel_r, blk_w, blk_h,
+    mismatch_record_block_tx(dst, pd->dst.stride, cm->current_frame.order_hint,
+                             plane, pixel_c, pixel_r, blk_w, blk_h,
                              xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
   }
 #endif
@@ -431,7 +432,7 @@ static void encode_block_pass1(int plane, int block, int blk_row, int blk_col,
 
   if (p->eobs[block] > 0) {
     txfm_param.bd = xd->bd;
-    txfm_param.is_hbd = get_bitdepth_data_path_index(xd);
+    txfm_param.is_hbd = is_cur_buf_hbd(xd);
     txfm_param.tx_type = DCT_DCT;
     txfm_param.tx_size = tx_size;
     txfm_param.eob = p->eobs[block];
@@ -578,7 +579,7 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
 
     const ENTROPY_CONTEXT *a = &args->ta[blk_col];
     const ENTROPY_CONTEXT *l = &args->tl[blk_row];
-    if (args->enable_optimize_b) {
+    if (args->enable_optimize_b != NO_TRELLIS_OPT) {
       av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                       tx_size, tx_type, AV1_XFORM_QUANT_FP);
       TXB_CTX txb_ctx;
@@ -598,11 +599,11 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
   }
 
   if (*eob == 0 && plane == 0) {
-  // TODO(jingning): Temporarily disable txk_type check for eob=0 case.
-  // It is possible that certain collision in hash index would cause
-  // the assertion failure. To further optimize the rate-distortion
-  // performance, we need to re-visit this part and enable this assert
-  // again.
+    // TODO(jingning): Temporarily disable txk_type check for eob=0 case.
+    // It is possible that certain collision in hash index would cause
+    // the assertion failure. To further optimize the rate-distortion
+    // performance, we need to re-visit this part and enable this assert
+    // again.
 #if 0
     if (args->cpi->oxcf.aq_mode == NO_AQ
         && args->cpi->oxcf.deltaq_mode == NO_DELTA_Q) {
diff --git a/third_party/aom/av1/encoder/encodemb.h b/third_party/aom/av1/encoder/encodemb.h
index 39080de599..d4394cf0be 100644
--- a/third_party/aom/av1/encoder/encodemb.h
+++ b/third_party/aom/av1/encoder/encodemb.h
@@ -37,13 +37,13 @@ struct encode_b_args {
   int8_t enable_optimize_b;
 };
 
-typedef enum AV1_XFORM_QUANT {
+enum {
   AV1_XFORM_QUANT_FP = 0,
   AV1_XFORM_QUANT_B = 1,
   AV1_XFORM_QUANT_DC = 2,
   AV1_XFORM_QUANT_SKIP_QUANT,
   AV1_XFORM_QUANT_TYPES,
-} AV1_XFORM_QUANT;
+} UENUM1BYTE(AV1_XFORM_QUANT);
 
 void av1_encode_sb(const struct AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
                    int mi_row, int mi_col, RUN_TYPE dry_run);
diff --git a/third_party/aom/av1/encoder/encoder.c b/third_party/aom/av1/encoder/encoder.c
index a2da2df899..9428a92740 100644
--- a/third_party/aom/av1/encoder/encoder.c
+++ b/third_party/aom/av1/encoder/encoder.c
@@ -33,9 +33,9 @@
 #include "aom_ports/mem.h"
 #include "aom_ports/system_state.h"
 #include "aom_scale/aom_scale.h"
-#if CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
+#if CONFIG_BITSTREAM_DEBUG
 #include "aom_util/debug_util.h"
-#endif  // CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
+#endif  // CONFIG_BITSTREAM_DEBUG
 
 #include "av1/common/alloccommon.h"
 #include "av1/common/cdef.h"
@@ -46,6 +46,7 @@
 #include "av1/common/resize.h"
 #include "av1/common/tile_common.h"
 
+#include "av1/encoder/av1_multi_thread.h"
 #include "av1/encoder/aq_complexity.h"
 #include "av1/encoder/aq_cyclicrefresh.h"
 #include "av1/encoder/aq_variance.h"
@@ -53,6 +54,7 @@
 #include "av1/encoder/context_tree.h"
 #include "av1/encoder/encodeframe.h"
 #include "av1/encoder/encodemv.h"
+#include "av1/encoder/encode_strategy.h"
 #include "av1/encoder/encoder.h"
 #include "av1/encoder/encodetxb.h"
 #include "av1/encoder/ethread.h"
@@ -65,15 +67,13 @@
 #include "av1/encoder/random.h"
 #include "av1/encoder/ratectrl.h"
 #include "av1/encoder/rd.h"
+#include "av1/encoder/rdopt.h"
 #include "av1/encoder/segmentation.h"
 #include "av1/encoder/speed_features.h"
-#include "av1/encoder/temporal_filter.h"
+#include "av1/encoder/reconinter_enc.h"
 
 #define DEFAULT_EXPLICIT_ORDER_HINT_BITS 7
 
-// av1 uses 10,000,000 ticks/second as time stamp
-#define TICKS_PER_SEC 10000000LL
-
 #if CONFIG_ENTROPY_STATS
 FRAME_COUNTS aggregate_fc;
 #endif  // CONFIG_ENTROPY_STATS
@@ -97,6 +97,30 @@ FILE *yuv_rec_file;
 #define FILE_NAME_LEN 100
 #endif
 
+// Estimate if the source frame is screen content, based on the portion of
+// blocks that have no more than 4 (experimentally selected) luma colors.
+static int is_screen_content(const uint8_t *src, int use_hbd, int bd,
+                             int stride, int width, int height) {
+  assert(src != NULL);
+  int counts = 0;
+  const int blk_w = 16;
+  const int blk_h = 16;
+  const int limit = 4;
+  for (int r = 0; r + blk_h <= height; r += blk_h) {
+    for (int c = 0; c + blk_w <= width; c += blk_w) {
+      int count_buf[1 << 12];  // Maximum (1 << 12) color levels.
+      const int n_colors =
+          use_hbd ? av1_count_colors_highbd(src + r * stride + c, stride, blk_w,
+                                            blk_h, bd, count_buf)
+                  : av1_count_colors(src + r * stride + c, stride, blk_w, blk_h,
+                                     count_buf);
+      if (n_colors > 1 && n_colors <= limit) counts++;
+    }
+  }
+  // The threshold is 10%.
+  return counts * blk_h * blk_w * 10 > width * height;
+}
+
 static INLINE void Scale2Ratio(AOM_SCALING mode, int *hr, int *hs) {
   switch (mode) {
     case NORMAL:
@@ -238,6 +262,66 @@ int av1_get_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows,
   }
 }
 
+// Compute the horizontal frequency components' energy in a frame
+// by calculuating the 16x4 Horizontal DCT. This is to be used to
+// decide the superresolution parameters.
+void analyze_hor_freq(const AV1_COMP *cpi, double *energy) {
+  uint64_t freq_energy[16] = { 0 };
+  const YV12_BUFFER_CONFIG *buf = cpi->source;
+  const int bd = cpi->td.mb.e_mbd.bd;
+  const int width = buf->y_crop_width;
+  const int height = buf->y_crop_height;
+  DECLARE_ALIGNED(16, int32_t, coeff[16 * 4]);
+  int n = 0;
+  memset(freq_energy, 0, sizeof(freq_energy));
+  if (buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    const int16_t *src16 = (const int16_t *)CONVERT_TO_SHORTPTR(buf->y_buffer);
+    for (int i = 0; i < height - 4; i += 4) {
+      for (int j = 0; j < width - 16; j += 16) {
+        av1_fwd_txfm2d_16x4(src16 + i * buf->y_stride + j, coeff, buf->y_stride,
+                            H_DCT, bd);
+        for (int k = 1; k < 16; ++k) {
+          const uint64_t this_energy =
+              ((int64_t)coeff[k] * coeff[k]) +
+              ((int64_t)coeff[k + 16] * coeff[k + 16]) +
+              ((int64_t)coeff[k + 32] * coeff[k + 32]) +
+              ((int64_t)coeff[k + 48] * coeff[k + 48]);
+          freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2 + 2 * (bd - 8));
+        }
+        n++;
+      }
+    }
+  } else {
+    assert(bd == 8);
+    DECLARE_ALIGNED(16, int16_t, src16[16 * 4]);
+    for (int i = 0; i < height - 4; i += 4) {
+      for (int j = 0; j < width - 16; j += 16) {
+        for (int ii = 0; ii < 4; ++ii)
+          for (int jj = 0; jj < 16; ++jj)
+            src16[ii * 16 + jj] =
+                buf->y_buffer[(i + ii) * buf->y_stride + (j + jj)];
+        av1_fwd_txfm2d_16x4(src16, coeff, 16, H_DCT, bd);
+        for (int k = 1; k < 16; ++k) {
+          const uint64_t this_energy =
+              ((int64_t)coeff[k] * coeff[k]) +
+              ((int64_t)coeff[k + 16] * coeff[k + 16]) +
+              ((int64_t)coeff[k + 32] * coeff[k + 32]) +
+              ((int64_t)coeff[k + 48] * coeff[k + 48]);
+          freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2);
+        }
+        n++;
+      }
+    }
+  }
+  if (n) {
+    for (int k = 1; k < 16; ++k) energy[k] = (double)freq_energy[k] / n;
+    // Convert to cumulative energy
+    for (int k = 14; k > 0; --k) energy[k] += energy[k + 1];
+  } else {
+    for (int k = 1; k < 16; ++k) energy[k] = 1e+20;
+  }
+}
+
 static void set_high_precision_mv(AV1_COMP *cpi, int allow_high_precision_mv,
                                   int cur_frame_force_integer_mv) {
   MACROBLOCK *const mb = &cpi->td.mb;
@@ -270,6 +354,9 @@ static BLOCK_SIZE select_sb_size(const AV1_COMP *const cpi) {
   // When superres / resize is on, 'cm->width / height' can change between
   // calls, so we don't apply this heuristic there. Also, this heuristic gives
   // compression gain for speed >= 2 only.
+  // Things break if superblock size changes per-frame which is why this
+  // heuristic is set based on configured speed rather than actual
+  // speed-features (which may change per-frame in future)
   if (cpi->oxcf.superres_mode == SUPERRES_NONE &&
       cpi->oxcf.resize_mode == RESIZE_NONE && cpi->oxcf.speed >= 2) {
     return (cm->width >= 480 && cm->height >= 360) ? BLOCK_128X128
@@ -287,64 +374,34 @@ static void setup_frame(AV1_COMP *cpi) {
   // other inter-frames the encoder currently uses only two contexts;
   // context 1 for ALTREF frames and context 0 for the others.
 
-  cm->primary_ref_frame = PRIMARY_REF_NONE;
   if (frame_is_intra_only(cm) || cm->error_resilient_mode ||
-      cm->force_primary_ref_none) {
+      cpi->ext_use_primary_ref_none) {
     av1_setup_past_independence(cm);
-    for (int i = 0; i < REF_FRAMES; i++) {
-      cm->fb_of_context_type[i] = -1;
-    }
-    cm->fb_of_context_type[REGULAR_FRAME] =
-        cm->show_frame ? get_ref_frame_map_idx(cpi, GOLDEN_FRAME)
-                       : get_ref_frame_map_idx(cpi, ALTREF_FRAME);
-    cm->frame_context_idx = REGULAR_FRAME;
-  } else {
-    const GF_GROUP *gf_group = &cpi->twopass.gf_group;
-    if (gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE)
-      cm->frame_context_idx = EXT_ARF_FRAME;
-    else if (cpi->refresh_alt_ref_frame)
-      cm->frame_context_idx = ARF_FRAME;
-    else if (cpi->rc.is_src_frame_alt_ref)
-      cm->frame_context_idx = OVERLAY_FRAME;
-    else if (cpi->refresh_golden_frame)
-      cm->frame_context_idx = GLD_FRAME;
-    else if (cpi->refresh_bwd_ref_frame)
-      cm->frame_context_idx = BRF_FRAME;
-    else
-      cm->frame_context_idx = REGULAR_FRAME;
-    int wanted_fb = cm->fb_of_context_type[cm->frame_context_idx];
-    for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) {
-      int fb = get_ref_frame_map_idx(cpi, ref_frame);
-      if (fb == wanted_fb) {
-        cm->primary_ref_frame = ref_frame - LAST_FRAME;
-      }
-    }
   }
 
-  if (cm->frame_type == KEY_FRAME && cm->show_frame) {
+  if (cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) {
     cpi->refresh_golden_frame = 1;
     cpi->refresh_alt_ref_frame = 1;
     av1_zero(cpi->interp_filter_selected);
     set_sb_size(&cm->seq_params, select_sb_size(cpi));
-    set_use_reference_buffer(cm, 0);
   } else if (frame_is_sframe(cm)) {
     cpi->refresh_golden_frame = 1;
     cpi->refresh_alt_ref_frame = 1;
     av1_zero(cpi->interp_filter_selected);
     set_sb_size(&cm->seq_params, select_sb_size(cpi));
   } else {
-    if (cm->primary_ref_frame == PRIMARY_REF_NONE ||
-        cm->frame_refs[cm->primary_ref_frame].idx < 0) {
+    const RefCntBuffer *const primary_ref_buf = get_primary_ref_frame_buf(cm);
+    if (primary_ref_buf == NULL) {
       av1_setup_past_independence(cm);
       cm->seg.update_map = 1;
       cm->seg.update_data = 1;
     } else {
-      *cm->fc = cm->frame_contexts[cm->frame_refs[cm->primary_ref_frame].idx];
+      *cm->fc = primary_ref_buf->frame_context;
     }
     av1_zero(cpi->interp_filter_selected[0]);
   }
 
-  cm->prev_frame = get_prev_frame(cm);
+  cm->prev_frame = get_primary_ref_frame_buf(cm);
   cpi->vaq_refresh = 0;
 }
 
@@ -438,6 +495,20 @@ static void alloc_context_buffers_ext(AV1_COMP *cpi) {
                   aom_calloc(mi_size, sizeof(*cpi->mbmi_ext_base)));
 }
 
+static void reset_film_grain_chroma_params(aom_film_grain_t *pars) {
+  pars->num_cr_points = 0;
+  pars->cr_mult = 0;
+  pars->cr_luma_mult = 0;
+  memset(pars->scaling_points_cr, 0, sizeof(pars->scaling_points_cr));
+  memset(pars->ar_coeffs_cr, 0, sizeof(pars->ar_coeffs_cr));
+  pars->num_cb_points = 0;
+  pars->cb_mult = 0;
+  pars->cb_luma_mult = 0;
+  pars->chroma_scaling_from_luma = 0;
+  memset(pars->scaling_points_cb, 0, sizeof(pars->scaling_points_cb));
+  memset(pars->ar_coeffs_cb, 0, sizeof(pars->ar_coeffs_cb));
+}
+
 static void update_film_grain_parameters(struct AV1_COMP *cpi,
                                          const AV1EncoderConfig *oxcf) {
   AV1_COMMON *const cm = &cpi->common;
@@ -451,11 +522,12 @@ static void update_film_grain_parameters(struct AV1_COMP *cpi,
 
   if (oxcf->film_grain_test_vector) {
     cm->seq_params.film_grain_params_present = 1;
-    if (cm->frame_type == KEY_FRAME) {
+    if (cm->current_frame.frame_type == KEY_FRAME) {
       memcpy(&cm->film_grain_params,
              film_grain_test_vectors + oxcf->film_grain_test_vector - 1,
              sizeof(cm->film_grain_params));
-
+      if (oxcf->monochrome)
+        reset_film_grain_chroma_params(&cm->film_grain_params);
       cm->film_grain_params.bit_depth = cm->seq_params.bit_depth;
       if (cm->seq_params.color_range == AOM_CR_FULL_RANGE) {
         cm->film_grain_params.clip_to_restricted_range = 0;
@@ -501,6 +573,9 @@ static void dealloc_compressor_data(AV1_COMP *cpi) {
   aom_free(cpi->td.mb.wsrc_buf);
   cpi->td.mb.wsrc_buf = NULL;
 
+  aom_free(cpi->td.mb.inter_modes_info);
+  cpi->td.mb.inter_modes_info = NULL;
+
   for (int i = 0; i < 2; i++)
     for (int j = 0; j < 2; j++) {
       aom_free(cpi->td.mb.hash_value_buffer[i][j]);
@@ -588,7 +663,7 @@ static void configure_static_seg_features(AV1_COMP *cpi) {
   int qi_delta;
 
   // Disable and clear down for KF
-  if (cm->frame_type == KEY_FRAME) {
+  if (cm->current_frame.frame_type == KEY_FRAME) {
     // Clear down the global segmentation map
     memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
     seg->update_map = 0;
@@ -716,7 +791,7 @@ static void configure_static_seg_features(AV1_COMP *cpi) {
 static void update_reference_segmentation_map(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
   MB_MODE_INFO **mi_4x4_ptr = cm->mi_grid_visible;
-  uint8_t *cache_ptr = cm->current_frame_seg_map;
+  uint8_t *cache_ptr = cm->cur_frame->seg_map;
   int row, col;
 
   for (row = 0; row < cm->mi_rows; row++) {
@@ -735,10 +810,10 @@ static void alloc_raw_frame_buffers(AV1_COMP *cpi) {
   const AV1EncoderConfig *oxcf = &cpi->oxcf;
 
   if (!cpi->lookahead)
-    cpi->lookahead =
-        av1_lookahead_init(oxcf->width, oxcf->height, seq_params->subsampling_x,
-                           seq_params->subsampling_y,
-                           seq_params->use_highbitdepth, oxcf->lag_in_frames);
+    cpi->lookahead = av1_lookahead_init(
+        oxcf->width, oxcf->height, seq_params->subsampling_x,
+        seq_params->subsampling_y, seq_params->use_highbitdepth,
+        oxcf->lag_in_frames, oxcf->border_in_pixels);
   if (!cpi->lookahead)
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate lag buffers");
@@ -747,7 +822,7 @@ static void alloc_raw_frame_buffers(AV1_COMP *cpi) {
   if (aom_realloc_frame_buffer(
           &cpi->alt_ref_buffer, oxcf->width, oxcf->height,
           seq_params->subsampling_x, seq_params->subsampling_y,
-          seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS,
+          seq_params->use_highbitdepth, oxcf->border_in_pixels,
           cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate altref buffer");
@@ -759,7 +834,7 @@ static void alloc_util_frame_buffers(AV1_COMP *cpi) {
   if (aom_realloc_frame_buffer(
           &cpi->last_frame_uf, cm->width, cm->height, seq_params->subsampling_x,
           seq_params->subsampling_y, seq_params->use_highbitdepth,
-          AOM_BORDER_IN_PIXELS, cm->byte_alignment, NULL, NULL, NULL))
+          cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate last frame buffer");
 
@@ -767,21 +842,21 @@ static void alloc_util_frame_buffers(AV1_COMP *cpi) {
           &cpi->trial_frame_rst, cm->superres_upscaled_width,
           cm->superres_upscaled_height, seq_params->subsampling_x,
           seq_params->subsampling_y, seq_params->use_highbitdepth,
-          AOM_BORDER_IN_PIXELS, cm->byte_alignment, NULL, NULL, NULL))
+          AOM_RESTORATION_FRAME_BORDER, cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate trial restored frame buffer");
 
   if (aom_realloc_frame_buffer(
           &cpi->scaled_source, cm->width, cm->height, seq_params->subsampling_x,
           seq_params->subsampling_y, seq_params->use_highbitdepth,
-          AOM_BORDER_IN_PIXELS, cm->byte_alignment, NULL, NULL, NULL))
+          cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate scaled source buffer");
 
   if (aom_realloc_frame_buffer(
           &cpi->scaled_last_source, cm->width, cm->height,
           seq_params->subsampling_x, seq_params->subsampling_y,
-          seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS,
+          seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels,
           cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate scaled last source buffer");
@@ -885,10 +960,9 @@ static void update_frame_size(AV1_COMP *cpi) {
 static void init_buffer_indices(AV1_COMP *cpi) {
   int fb_idx;
   for (fb_idx = 0; fb_idx < REF_FRAMES; ++fb_idx)
-    cpi->ref_fb_idx[fb_idx] = fb_idx;
+    cpi->common.remapped_ref_idx[fb_idx] = fb_idx;
   cpi->rate_index = 0;
   cpi->rate_size = 0;
-  cpi->cur_poc = -1;
 }
 
 static INLINE int does_level_match(int width, int height, double fps,
@@ -999,30 +1073,50 @@ static void init_seq_coding_tools(SequenceHeader *seq, AV1_COMMON *cm,
   seq->reduced_still_picture_hdr &= !oxcf->full_still_picture_hdr;
   seq->force_screen_content_tools = 2;
   seq->force_integer_mv = 2;
-  seq->enable_order_hint = oxcf->enable_order_hint;
-  seq->frame_id_numbers_present_flag = oxcf->large_scale_tile;
+  seq->order_hint_info.enable_order_hint = oxcf->enable_order_hint;
+  seq->frame_id_numbers_present_flag =
+      !(seq->still_picture && seq->reduced_still_picture_hdr) &&
+      !oxcf->large_scale_tile && oxcf->error_resilient_mode;
   if (seq->still_picture && seq->reduced_still_picture_hdr) {
-    seq->enable_order_hint = 0;
-    seq->frame_id_numbers_present_flag = 0;
+    seq->order_hint_info.enable_order_hint = 0;
     seq->force_screen_content_tools = 2;
     seq->force_integer_mv = 2;
   }
-  seq->order_hint_bits_minus_1 =
-      seq->enable_order_hint ? DEFAULT_EXPLICIT_ORDER_HINT_BITS - 1 : -1;
+  seq->order_hint_info.order_hint_bits_minus_1 =
+      seq->order_hint_info.enable_order_hint
+          ? DEFAULT_EXPLICIT_ORDER_HINT_BITS - 1
+          : -1;
+
+  seq->max_frame_width =
+      oxcf->forced_max_frame_width ? oxcf->forced_max_frame_width : oxcf->width;
+  seq->max_frame_height = oxcf->forced_max_frame_height
+                              ? oxcf->forced_max_frame_height
+                              : oxcf->height;
+  seq->num_bits_width =
+      (seq->max_frame_width > 1) ? get_msb(seq->max_frame_width - 1) + 1 : 1;
+  seq->num_bits_height =
+      (seq->max_frame_height > 1) ? get_msb(seq->max_frame_height - 1) + 1 : 1;
+  assert(seq->num_bits_width <= 16);
+  assert(seq->num_bits_height <= 16);
+
+  seq->frame_id_length = FRAME_ID_LENGTH;
+  seq->delta_frame_id_length = DELTA_FRAME_ID_LENGTH;
 
   seq->enable_dual_filter = oxcf->enable_dual_filter;
-  seq->enable_jnt_comp = oxcf->enable_jnt_comp;
-  seq->enable_jnt_comp &= seq->enable_order_hint;
-  seq->enable_ref_frame_mvs = oxcf->enable_ref_frame_mvs;
-  seq->enable_ref_frame_mvs &= seq->enable_order_hint;
+  seq->order_hint_info.enable_dist_wtd_comp = oxcf->enable_dist_wtd_comp;
+  seq->order_hint_info.enable_dist_wtd_comp &=
+      seq->order_hint_info.enable_order_hint;
+  seq->order_hint_info.enable_ref_frame_mvs = oxcf->enable_ref_frame_mvs;
+  seq->order_hint_info.enable_ref_frame_mvs &=
+      seq->order_hint_info.enable_order_hint;
   seq->enable_superres = oxcf->enable_superres;
   seq->enable_cdef = oxcf->enable_cdef;
   seq->enable_restoration = oxcf->enable_restoration;
   seq->enable_warped_motion = oxcf->enable_warped_motion;
-  seq->enable_interintra_compound = 1;
-  seq->enable_masked_compound = 1;
-  seq->enable_intra_edge_filter = 1;
-  seq->enable_filter_intra = 1;
+  seq->enable_interintra_compound = oxcf->enable_interintra_comp;
+  seq->enable_masked_compound = oxcf->enable_masked_comp;
+  seq->enable_intra_edge_filter = oxcf->enable_intra_edge_filter;
+  seq->enable_filter_intra = oxcf->enable_filter_intra;
 
   set_bitstream_level_tier(seq, cm, oxcf);
 
@@ -1219,14 +1313,14 @@ static void set_rc_buffer_sizes(RATE_CONTROL *rc,
   static unsigned int fnname##_bits8(                                       \
       const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr,    \
       int ref_stride, const uint8_t *second_pred,                           \
-      const JNT_COMP_PARAMS *jcp_param) {                                   \
+      const DIST_WTD_COMP_PARAMS *jcp_param) {                              \
     return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \
                   jcp_param);                                               \
   }                                                                         \
   static unsigned int fnname##_bits10(                                      \
       const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr,    \
       int ref_stride, const uint8_t *second_pred,                           \
-      const JNT_COMP_PARAMS *jcp_param) {                                   \
+      const DIST_WTD_COMP_PARAMS *jcp_param) {                              \
     return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \
                   jcp_param) >>                                             \
            2;                                                               \
@@ -1234,7 +1328,7 @@ static void set_rc_buffer_sizes(RATE_CONTROL *rc,
   static unsigned int fnname##_bits12(                                      \
       const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr,    \
       int ref_stride, const uint8_t *second_pred,                           \
-      const JNT_COMP_PARAMS *jcp_param) {                                   \
+      const DIST_WTD_COMP_PARAMS *jcp_param) {                              \
     return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \
                   jcp_param) >>                                             \
            4;                                                               \
@@ -1308,28 +1402,28 @@ MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x16)
 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x16_avg)
 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x16x4d)
 
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad128x128_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad128x64_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad64x128_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad32x16_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad16x32_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad64x32_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad32x64_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad32x32_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad64x64_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad16x16_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad16x8_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad8x16_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad8x8_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad8x4_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad4x8_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad4x4_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad4x16_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad16x4_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad8x32_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad32x8_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad16x64_avg)
-MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_jnt_sad64x16_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x128_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x64_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x128_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x16_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x32_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x32_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x64_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x32_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x64_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x16_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x8_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x16_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x8_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x4_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x8_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x4_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x16_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x4_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x32_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x8_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x64_avg)
+MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x16_avg)
 
 #define HIGHBD_MBFP(BT, MCSDF, MCSVF) \
   cpi->fn_ptr[BT].msdf = MCSDF;       \
@@ -1438,166 +1532,167 @@ static void highbd_set_var_fns(AV1_COMP *const cpi) {
                    aom_highbd_8_sub_pixel_variance64x16,
                    aom_highbd_8_sub_pixel_avg_variance64x16,
                    aom_highbd_sad64x16x4d_bits8,
-                   aom_highbd_jnt_sad64x16_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance64x16)
+                   aom_highbd_dist_wtd_sad64x16_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x16)
 
         HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits8,
                    aom_highbd_sad16x64_avg_bits8, aom_highbd_8_variance16x64,
                    aom_highbd_8_sub_pixel_variance16x64,
                    aom_highbd_8_sub_pixel_avg_variance16x64,
                    aom_highbd_sad16x64x4d_bits8,
-                   aom_highbd_jnt_sad16x64_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance16x64)
+                   aom_highbd_dist_wtd_sad16x64_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x64)
 
         HIGHBD_BFP(
             BLOCK_32X8, aom_highbd_sad32x8_bits8, aom_highbd_sad32x8_avg_bits8,
             aom_highbd_8_variance32x8, aom_highbd_8_sub_pixel_variance32x8,
             aom_highbd_8_sub_pixel_avg_variance32x8,
-            aom_highbd_sad32x8x4d_bits8, aom_highbd_jnt_sad32x8_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance32x8)
+            aom_highbd_sad32x8x4d_bits8, aom_highbd_dist_wtd_sad32x8_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x8)
 
         HIGHBD_BFP(
             BLOCK_8X32, aom_highbd_sad8x32_bits8, aom_highbd_sad8x32_avg_bits8,
             aom_highbd_8_variance8x32, aom_highbd_8_sub_pixel_variance8x32,
             aom_highbd_8_sub_pixel_avg_variance8x32,
-            aom_highbd_sad8x32x4d_bits8, aom_highbd_jnt_sad8x32_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance8x32)
+            aom_highbd_sad8x32x4d_bits8, aom_highbd_dist_wtd_sad8x32_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x32)
 
         HIGHBD_BFP(
             BLOCK_16X4, aom_highbd_sad16x4_bits8, aom_highbd_sad16x4_avg_bits8,
             aom_highbd_8_variance16x4, aom_highbd_8_sub_pixel_variance16x4,
             aom_highbd_8_sub_pixel_avg_variance16x4,
-            aom_highbd_sad16x4x4d_bits8, aom_highbd_jnt_sad16x4_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance16x4)
+            aom_highbd_sad16x4x4d_bits8, aom_highbd_dist_wtd_sad16x4_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x4)
 
         HIGHBD_BFP(
             BLOCK_4X16, aom_highbd_sad4x16_bits8, aom_highbd_sad4x16_avg_bits8,
             aom_highbd_8_variance4x16, aom_highbd_8_sub_pixel_variance4x16,
             aom_highbd_8_sub_pixel_avg_variance4x16,
-            aom_highbd_sad4x16x4d_bits8, aom_highbd_jnt_sad4x16_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance4x16)
+            aom_highbd_sad4x16x4d_bits8, aom_highbd_dist_wtd_sad4x16_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x16)
 
         HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits8,
                    aom_highbd_sad32x16_avg_bits8, aom_highbd_8_variance32x16,
                    aom_highbd_8_sub_pixel_variance32x16,
                    aom_highbd_8_sub_pixel_avg_variance32x16,
                    aom_highbd_sad32x16x4d_bits8,
-                   aom_highbd_jnt_sad32x16_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance32x16)
+                   aom_highbd_dist_wtd_sad32x16_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x16)
 
         HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits8,
                    aom_highbd_sad16x32_avg_bits8, aom_highbd_8_variance16x32,
                    aom_highbd_8_sub_pixel_variance16x32,
                    aom_highbd_8_sub_pixel_avg_variance16x32,
                    aom_highbd_sad16x32x4d_bits8,
-                   aom_highbd_jnt_sad16x32_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance16x32)
+                   aom_highbd_dist_wtd_sad16x32_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x32)
 
         HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits8,
                    aom_highbd_sad64x32_avg_bits8, aom_highbd_8_variance64x32,
                    aom_highbd_8_sub_pixel_variance64x32,
                    aom_highbd_8_sub_pixel_avg_variance64x32,
                    aom_highbd_sad64x32x4d_bits8,
-                   aom_highbd_jnt_sad64x32_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance64x32)
+                   aom_highbd_dist_wtd_sad64x32_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x32)
 
         HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits8,
                    aom_highbd_sad32x64_avg_bits8, aom_highbd_8_variance32x64,
                    aom_highbd_8_sub_pixel_variance32x64,
                    aom_highbd_8_sub_pixel_avg_variance32x64,
                    aom_highbd_sad32x64x4d_bits8,
-                   aom_highbd_jnt_sad32x64_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance32x64)
+                   aom_highbd_dist_wtd_sad32x64_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x64)
 
         HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits8,
                    aom_highbd_sad32x32_avg_bits8, aom_highbd_8_variance32x32,
                    aom_highbd_8_sub_pixel_variance32x32,
                    aom_highbd_8_sub_pixel_avg_variance32x32,
                    aom_highbd_sad32x32x4d_bits8,
-                   aom_highbd_jnt_sad32x32_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance32x32)
+                   aom_highbd_dist_wtd_sad32x32_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x32)
 
         HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits8,
                    aom_highbd_sad64x64_avg_bits8, aom_highbd_8_variance64x64,
                    aom_highbd_8_sub_pixel_variance64x64,
                    aom_highbd_8_sub_pixel_avg_variance64x64,
                    aom_highbd_sad64x64x4d_bits8,
-                   aom_highbd_jnt_sad64x64_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance64x64)
+                   aom_highbd_dist_wtd_sad64x64_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x64)
 
         HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits8,
                    aom_highbd_sad16x16_avg_bits8, aom_highbd_8_variance16x16,
                    aom_highbd_8_sub_pixel_variance16x16,
                    aom_highbd_8_sub_pixel_avg_variance16x16,
                    aom_highbd_sad16x16x4d_bits8,
-                   aom_highbd_jnt_sad16x16_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance16x16)
+                   aom_highbd_dist_wtd_sad16x16_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x16)
 
         HIGHBD_BFP(
             BLOCK_16X8, aom_highbd_sad16x8_bits8, aom_highbd_sad16x8_avg_bits8,
             aom_highbd_8_variance16x8, aom_highbd_8_sub_pixel_variance16x8,
             aom_highbd_8_sub_pixel_avg_variance16x8,
-            aom_highbd_sad16x8x4d_bits8, aom_highbd_jnt_sad16x8_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance16x8)
+            aom_highbd_sad16x8x4d_bits8, aom_highbd_dist_wtd_sad16x8_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x8)
 
         HIGHBD_BFP(
             BLOCK_8X16, aom_highbd_sad8x16_bits8, aom_highbd_sad8x16_avg_bits8,
             aom_highbd_8_variance8x16, aom_highbd_8_sub_pixel_variance8x16,
             aom_highbd_8_sub_pixel_avg_variance8x16,
-            aom_highbd_sad8x16x4d_bits8, aom_highbd_jnt_sad8x16_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance8x16)
-
-        HIGHBD_BFP(BLOCK_8X8, aom_highbd_sad8x8_bits8,
-                   aom_highbd_sad8x8_avg_bits8, aom_highbd_8_variance8x8,
-                   aom_highbd_8_sub_pixel_variance8x8,
-                   aom_highbd_8_sub_pixel_avg_variance8x8,
-                   aom_highbd_sad8x8x4d_bits8, aom_highbd_jnt_sad8x8_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance8x8)
-
-        HIGHBD_BFP(BLOCK_8X4, aom_highbd_sad8x4_bits8,
-                   aom_highbd_sad8x4_avg_bits8, aom_highbd_8_variance8x4,
-                   aom_highbd_8_sub_pixel_variance8x4,
-                   aom_highbd_8_sub_pixel_avg_variance8x4,
-                   aom_highbd_sad8x4x4d_bits8, aom_highbd_jnt_sad8x4_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance8x4)
-
-        HIGHBD_BFP(BLOCK_4X8, aom_highbd_sad4x8_bits8,
-                   aom_highbd_sad4x8_avg_bits8, aom_highbd_8_variance4x8,
-                   aom_highbd_8_sub_pixel_variance4x8,
-                   aom_highbd_8_sub_pixel_avg_variance4x8,
-                   aom_highbd_sad4x8x4d_bits8, aom_highbd_jnt_sad4x8_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance4x8)
-
-        HIGHBD_BFP(BLOCK_4X4, aom_highbd_sad4x4_bits8,
-                   aom_highbd_sad4x4_avg_bits8, aom_highbd_8_variance4x4,
-                   aom_highbd_8_sub_pixel_variance4x4,
-                   aom_highbd_8_sub_pixel_avg_variance4x4,
-                   aom_highbd_sad4x4x4d_bits8, aom_highbd_jnt_sad4x4_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance4x4)
+            aom_highbd_sad8x16x4d_bits8, aom_highbd_dist_wtd_sad8x16_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x16)
 
         HIGHBD_BFP(
-            BLOCK_128X128, aom_highbd_sad128x128_bits8,
-            aom_highbd_sad128x128_avg_bits8, aom_highbd_8_variance128x128,
-            aom_highbd_8_sub_pixel_variance128x128,
-            aom_highbd_8_sub_pixel_avg_variance128x128,
-            aom_highbd_sad128x128x4d_bits8, aom_highbd_jnt_sad128x128_avg_bits8,
-            aom_highbd_8_jnt_sub_pixel_avg_variance128x128)
+            BLOCK_8X8, aom_highbd_sad8x8_bits8, aom_highbd_sad8x8_avg_bits8,
+            aom_highbd_8_variance8x8, aom_highbd_8_sub_pixel_variance8x8,
+            aom_highbd_8_sub_pixel_avg_variance8x8, aom_highbd_sad8x8x4d_bits8,
+            aom_highbd_dist_wtd_sad8x8_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x8)
+
+        HIGHBD_BFP(
+            BLOCK_8X4, aom_highbd_sad8x4_bits8, aom_highbd_sad8x4_avg_bits8,
+            aom_highbd_8_variance8x4, aom_highbd_8_sub_pixel_variance8x4,
+            aom_highbd_8_sub_pixel_avg_variance8x4, aom_highbd_sad8x4x4d_bits8,
+            aom_highbd_dist_wtd_sad8x4_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x4)
+
+        HIGHBD_BFP(
+            BLOCK_4X8, aom_highbd_sad4x8_bits8, aom_highbd_sad4x8_avg_bits8,
+            aom_highbd_8_variance4x8, aom_highbd_8_sub_pixel_variance4x8,
+            aom_highbd_8_sub_pixel_avg_variance4x8, aom_highbd_sad4x8x4d_bits8,
+            aom_highbd_dist_wtd_sad4x8_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x8)
+
+        HIGHBD_BFP(
+            BLOCK_4X4, aom_highbd_sad4x4_bits8, aom_highbd_sad4x4_avg_bits8,
+            aom_highbd_8_variance4x4, aom_highbd_8_sub_pixel_variance4x4,
+            aom_highbd_8_sub_pixel_avg_variance4x4, aom_highbd_sad4x4x4d_bits8,
+            aom_highbd_dist_wtd_sad4x4_avg_bits8,
+            aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x4)
+
+        HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits8,
+                   aom_highbd_sad128x128_avg_bits8,
+                   aom_highbd_8_variance128x128,
+                   aom_highbd_8_sub_pixel_variance128x128,
+                   aom_highbd_8_sub_pixel_avg_variance128x128,
+                   aom_highbd_sad128x128x4d_bits8,
+                   aom_highbd_dist_wtd_sad128x128_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x128)
 
         HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits8,
                    aom_highbd_sad128x64_avg_bits8, aom_highbd_8_variance128x64,
                    aom_highbd_8_sub_pixel_variance128x64,
                    aom_highbd_8_sub_pixel_avg_variance128x64,
                    aom_highbd_sad128x64x4d_bits8,
-                   aom_highbd_jnt_sad128x64_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance128x64)
+                   aom_highbd_dist_wtd_sad128x64_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x64)
 
         HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits8,
                    aom_highbd_sad64x128_avg_bits8, aom_highbd_8_variance64x128,
                    aom_highbd_8_sub_pixel_variance64x128,
                    aom_highbd_8_sub_pixel_avg_variance64x128,
                    aom_highbd_sad64x128x4d_bits8,
-                   aom_highbd_jnt_sad64x128_avg_bits8,
-                   aom_highbd_8_jnt_sub_pixel_avg_variance64x128)
+                   aom_highbd_dist_wtd_sad64x128_avg_bits8,
+                   aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x128)
 
         HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits8,
                     aom_highbd_8_masked_sub_pixel_variance128x128)
@@ -1717,148 +1812,148 @@ static void highbd_set_var_fns(AV1_COMP *const cpi) {
                    aom_highbd_10_sub_pixel_variance64x16,
                    aom_highbd_10_sub_pixel_avg_variance64x16,
                    aom_highbd_sad64x16x4d_bits10,
-                   aom_highbd_jnt_sad64x16_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance64x16);
+                   aom_highbd_dist_wtd_sad64x16_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x16);
 
         HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits10,
                    aom_highbd_sad16x64_avg_bits10, aom_highbd_10_variance16x64,
                    aom_highbd_10_sub_pixel_variance16x64,
                    aom_highbd_10_sub_pixel_avg_variance16x64,
                    aom_highbd_sad16x64x4d_bits10,
-                   aom_highbd_jnt_sad16x64_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance16x64);
+                   aom_highbd_dist_wtd_sad16x64_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x64);
 
         HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits10,
                    aom_highbd_sad32x8_avg_bits10, aom_highbd_10_variance32x8,
                    aom_highbd_10_sub_pixel_variance32x8,
                    aom_highbd_10_sub_pixel_avg_variance32x8,
                    aom_highbd_sad32x8x4d_bits10,
-                   aom_highbd_jnt_sad32x8_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance32x8);
+                   aom_highbd_dist_wtd_sad32x8_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x8);
 
         HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits10,
                    aom_highbd_sad8x32_avg_bits10, aom_highbd_10_variance8x32,
                    aom_highbd_10_sub_pixel_variance8x32,
                    aom_highbd_10_sub_pixel_avg_variance8x32,
                    aom_highbd_sad8x32x4d_bits10,
-                   aom_highbd_jnt_sad8x32_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance8x32);
+                   aom_highbd_dist_wtd_sad8x32_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x32);
 
         HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits10,
                    aom_highbd_sad16x4_avg_bits10, aom_highbd_10_variance16x4,
                    aom_highbd_10_sub_pixel_variance16x4,
                    aom_highbd_10_sub_pixel_avg_variance16x4,
                    aom_highbd_sad16x4x4d_bits10,
-                   aom_highbd_jnt_sad16x4_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance16x4);
+                   aom_highbd_dist_wtd_sad16x4_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x4);
 
         HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits10,
                    aom_highbd_sad4x16_avg_bits10, aom_highbd_10_variance4x16,
                    aom_highbd_10_sub_pixel_variance4x16,
                    aom_highbd_10_sub_pixel_avg_variance4x16,
                    aom_highbd_sad4x16x4d_bits10,
-                   aom_highbd_jnt_sad4x16_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance4x16);
+                   aom_highbd_dist_wtd_sad4x16_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x16);
 
         HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits10,
                    aom_highbd_sad32x16_avg_bits10, aom_highbd_10_variance32x16,
                    aom_highbd_10_sub_pixel_variance32x16,
                    aom_highbd_10_sub_pixel_avg_variance32x16,
                    aom_highbd_sad32x16x4d_bits10,
-                   aom_highbd_jnt_sad32x16_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance32x16);
+                   aom_highbd_dist_wtd_sad32x16_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x16);
 
         HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits10,
                    aom_highbd_sad16x32_avg_bits10, aom_highbd_10_variance16x32,
                    aom_highbd_10_sub_pixel_variance16x32,
                    aom_highbd_10_sub_pixel_avg_variance16x32,
                    aom_highbd_sad16x32x4d_bits10,
-                   aom_highbd_jnt_sad16x32_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance16x32);
+                   aom_highbd_dist_wtd_sad16x32_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x32);
 
         HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits10,
                    aom_highbd_sad64x32_avg_bits10, aom_highbd_10_variance64x32,
                    aom_highbd_10_sub_pixel_variance64x32,
                    aom_highbd_10_sub_pixel_avg_variance64x32,
                    aom_highbd_sad64x32x4d_bits10,
-                   aom_highbd_jnt_sad64x32_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance64x32);
+                   aom_highbd_dist_wtd_sad64x32_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x32);
 
         HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits10,
                    aom_highbd_sad32x64_avg_bits10, aom_highbd_10_variance32x64,
                    aom_highbd_10_sub_pixel_variance32x64,
                    aom_highbd_10_sub_pixel_avg_variance32x64,
                    aom_highbd_sad32x64x4d_bits10,
-                   aom_highbd_jnt_sad32x64_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance32x64);
+                   aom_highbd_dist_wtd_sad32x64_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x64);
 
         HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits10,
                    aom_highbd_sad32x32_avg_bits10, aom_highbd_10_variance32x32,
                    aom_highbd_10_sub_pixel_variance32x32,
                    aom_highbd_10_sub_pixel_avg_variance32x32,
                    aom_highbd_sad32x32x4d_bits10,
-                   aom_highbd_jnt_sad32x32_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance32x32);
+                   aom_highbd_dist_wtd_sad32x32_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x32);
 
         HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits10,
                    aom_highbd_sad64x64_avg_bits10, aom_highbd_10_variance64x64,
                    aom_highbd_10_sub_pixel_variance64x64,
                    aom_highbd_10_sub_pixel_avg_variance64x64,
                    aom_highbd_sad64x64x4d_bits10,
-                   aom_highbd_jnt_sad64x64_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance64x64);
+                   aom_highbd_dist_wtd_sad64x64_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x64);
 
         HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits10,
                    aom_highbd_sad16x16_avg_bits10, aom_highbd_10_variance16x16,
                    aom_highbd_10_sub_pixel_variance16x16,
                    aom_highbd_10_sub_pixel_avg_variance16x16,
                    aom_highbd_sad16x16x4d_bits10,
-                   aom_highbd_jnt_sad16x16_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance16x16);
+                   aom_highbd_dist_wtd_sad16x16_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x16);
 
         HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits10,
                    aom_highbd_sad16x8_avg_bits10, aom_highbd_10_variance16x8,
                    aom_highbd_10_sub_pixel_variance16x8,
                    aom_highbd_10_sub_pixel_avg_variance16x8,
                    aom_highbd_sad16x8x4d_bits10,
-                   aom_highbd_jnt_sad16x8_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance16x8);
+                   aom_highbd_dist_wtd_sad16x8_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x8);
 
         HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits10,
                    aom_highbd_sad8x16_avg_bits10, aom_highbd_10_variance8x16,
                    aom_highbd_10_sub_pixel_variance8x16,
                    aom_highbd_10_sub_pixel_avg_variance8x16,
                    aom_highbd_sad8x16x4d_bits10,
-                   aom_highbd_jnt_sad8x16_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance8x16);
+                   aom_highbd_dist_wtd_sad8x16_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x16);
 
         HIGHBD_BFP(
             BLOCK_8X8, aom_highbd_sad8x8_bits10, aom_highbd_sad8x8_avg_bits10,
             aom_highbd_10_variance8x8, aom_highbd_10_sub_pixel_variance8x8,
             aom_highbd_10_sub_pixel_avg_variance8x8,
-            aom_highbd_sad8x8x4d_bits10, aom_highbd_jnt_sad8x8_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance8x8);
+            aom_highbd_sad8x8x4d_bits10, aom_highbd_dist_wtd_sad8x8_avg_bits10,
+            aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x8);
 
         HIGHBD_BFP(
             BLOCK_8X4, aom_highbd_sad8x4_bits10, aom_highbd_sad8x4_avg_bits10,
             aom_highbd_10_variance8x4, aom_highbd_10_sub_pixel_variance8x4,
             aom_highbd_10_sub_pixel_avg_variance8x4,
-            aom_highbd_sad8x4x4d_bits10, aom_highbd_jnt_sad8x4_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance8x4);
+            aom_highbd_sad8x4x4d_bits10, aom_highbd_dist_wtd_sad8x4_avg_bits10,
+            aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x4);
 
         HIGHBD_BFP(
             BLOCK_4X8, aom_highbd_sad4x8_bits10, aom_highbd_sad4x8_avg_bits10,
             aom_highbd_10_variance4x8, aom_highbd_10_sub_pixel_variance4x8,
             aom_highbd_10_sub_pixel_avg_variance4x8,
-            aom_highbd_sad4x8x4d_bits10, aom_highbd_jnt_sad4x8_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance4x8);
+            aom_highbd_sad4x8x4d_bits10, aom_highbd_dist_wtd_sad4x8_avg_bits10,
+            aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x8);
 
         HIGHBD_BFP(
             BLOCK_4X4, aom_highbd_sad4x4_bits10, aom_highbd_sad4x4_avg_bits10,
             aom_highbd_10_variance4x4, aom_highbd_10_sub_pixel_variance4x4,
             aom_highbd_10_sub_pixel_avg_variance4x4,
-            aom_highbd_sad4x4x4d_bits10, aom_highbd_jnt_sad4x4_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance4x4);
+            aom_highbd_sad4x4x4d_bits10, aom_highbd_dist_wtd_sad4x4_avg_bits10,
+            aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x4);
 
         HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits10,
                    aom_highbd_sad128x128_avg_bits10,
@@ -1866,24 +1961,26 @@ static void highbd_set_var_fns(AV1_COMP *const cpi) {
                    aom_highbd_10_sub_pixel_variance128x128,
                    aom_highbd_10_sub_pixel_avg_variance128x128,
                    aom_highbd_sad128x128x4d_bits10,
-                   aom_highbd_jnt_sad128x128_avg_bits10,
-                   aom_highbd_10_jnt_sub_pixel_avg_variance128x128);
+                   aom_highbd_dist_wtd_sad128x128_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x128);
 
-        HIGHBD_BFP(
-            BLOCK_128X64, aom_highbd_sad128x64_bits10,
-            aom_highbd_sad128x64_avg_bits10, aom_highbd_10_variance128x64,
-            aom_highbd_10_sub_pixel_variance128x64,
-            aom_highbd_10_sub_pixel_avg_variance128x64,
-            aom_highbd_sad128x64x4d_bits10, aom_highbd_jnt_sad128x64_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance128x64);
+        HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits10,
+                   aom_highbd_sad128x64_avg_bits10,
+                   aom_highbd_10_variance128x64,
+                   aom_highbd_10_sub_pixel_variance128x64,
+                   aom_highbd_10_sub_pixel_avg_variance128x64,
+                   aom_highbd_sad128x64x4d_bits10,
+                   aom_highbd_dist_wtd_sad128x64_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x64);
 
-        HIGHBD_BFP(
-            BLOCK_64X128, aom_highbd_sad64x128_bits10,
-            aom_highbd_sad64x128_avg_bits10, aom_highbd_10_variance64x128,
-            aom_highbd_10_sub_pixel_variance64x128,
-            aom_highbd_10_sub_pixel_avg_variance64x128,
-            aom_highbd_sad64x128x4d_bits10, aom_highbd_jnt_sad64x128_avg_bits10,
-            aom_highbd_10_jnt_sub_pixel_avg_variance64x128);
+        HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits10,
+                   aom_highbd_sad64x128_avg_bits10,
+                   aom_highbd_10_variance64x128,
+                   aom_highbd_10_sub_pixel_variance64x128,
+                   aom_highbd_10_sub_pixel_avg_variance64x128,
+                   aom_highbd_sad64x128x4d_bits10,
+                   aom_highbd_dist_wtd_sad64x128_avg_bits10,
+                   aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x128);
 
         HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits10,
                     aom_highbd_10_masked_sub_pixel_variance128x128)
@@ -2009,148 +2106,148 @@ static void highbd_set_var_fns(AV1_COMP *const cpi) {
                    aom_highbd_12_sub_pixel_variance64x16,
                    aom_highbd_12_sub_pixel_avg_variance64x16,
                    aom_highbd_sad64x16x4d_bits12,
-                   aom_highbd_jnt_sad64x16_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance64x16);
+                   aom_highbd_dist_wtd_sad64x16_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x16);
 
         HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits12,
                    aom_highbd_sad16x64_avg_bits12, aom_highbd_12_variance16x64,
                    aom_highbd_12_sub_pixel_variance16x64,
                    aom_highbd_12_sub_pixel_avg_variance16x64,
                    aom_highbd_sad16x64x4d_bits12,
-                   aom_highbd_jnt_sad16x64_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance16x64);
+                   aom_highbd_dist_wtd_sad16x64_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x64);
 
         HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits12,
                    aom_highbd_sad32x8_avg_bits12, aom_highbd_12_variance32x8,
                    aom_highbd_12_sub_pixel_variance32x8,
                    aom_highbd_12_sub_pixel_avg_variance32x8,
                    aom_highbd_sad32x8x4d_bits12,
-                   aom_highbd_jnt_sad32x8_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance32x8);
+                   aom_highbd_dist_wtd_sad32x8_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x8);
 
         HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits12,
                    aom_highbd_sad8x32_avg_bits12, aom_highbd_12_variance8x32,
                    aom_highbd_12_sub_pixel_variance8x32,
                    aom_highbd_12_sub_pixel_avg_variance8x32,
                    aom_highbd_sad8x32x4d_bits12,
-                   aom_highbd_jnt_sad8x32_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance8x32);
+                   aom_highbd_dist_wtd_sad8x32_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x32);
 
         HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits12,
                    aom_highbd_sad16x4_avg_bits12, aom_highbd_12_variance16x4,
                    aom_highbd_12_sub_pixel_variance16x4,
                    aom_highbd_12_sub_pixel_avg_variance16x4,
                    aom_highbd_sad16x4x4d_bits12,
-                   aom_highbd_jnt_sad16x4_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance16x4);
+                   aom_highbd_dist_wtd_sad16x4_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x4);
 
         HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits12,
                    aom_highbd_sad4x16_avg_bits12, aom_highbd_12_variance4x16,
                    aom_highbd_12_sub_pixel_variance4x16,
                    aom_highbd_12_sub_pixel_avg_variance4x16,
                    aom_highbd_sad4x16x4d_bits12,
-                   aom_highbd_jnt_sad4x16_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance4x16);
+                   aom_highbd_dist_wtd_sad4x16_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x16);
 
         HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits12,
                    aom_highbd_sad32x16_avg_bits12, aom_highbd_12_variance32x16,
                    aom_highbd_12_sub_pixel_variance32x16,
                    aom_highbd_12_sub_pixel_avg_variance32x16,
                    aom_highbd_sad32x16x4d_bits12,
-                   aom_highbd_jnt_sad32x16_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance32x16);
+                   aom_highbd_dist_wtd_sad32x16_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x16);
 
         HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits12,
                    aom_highbd_sad16x32_avg_bits12, aom_highbd_12_variance16x32,
                    aom_highbd_12_sub_pixel_variance16x32,
                    aom_highbd_12_sub_pixel_avg_variance16x32,
                    aom_highbd_sad16x32x4d_bits12,
-                   aom_highbd_jnt_sad16x32_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance16x32);
+                   aom_highbd_dist_wtd_sad16x32_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x32);
 
         HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits12,
                    aom_highbd_sad64x32_avg_bits12, aom_highbd_12_variance64x32,
                    aom_highbd_12_sub_pixel_variance64x32,
                    aom_highbd_12_sub_pixel_avg_variance64x32,
                    aom_highbd_sad64x32x4d_bits12,
-                   aom_highbd_jnt_sad64x32_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance64x32);
+                   aom_highbd_dist_wtd_sad64x32_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x32);
 
         HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits12,
                    aom_highbd_sad32x64_avg_bits12, aom_highbd_12_variance32x64,
                    aom_highbd_12_sub_pixel_variance32x64,
                    aom_highbd_12_sub_pixel_avg_variance32x64,
                    aom_highbd_sad32x64x4d_bits12,
-                   aom_highbd_jnt_sad32x64_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance32x64);
+                   aom_highbd_dist_wtd_sad32x64_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x64);
 
         HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits12,
                    aom_highbd_sad32x32_avg_bits12, aom_highbd_12_variance32x32,
                    aom_highbd_12_sub_pixel_variance32x32,
                    aom_highbd_12_sub_pixel_avg_variance32x32,
                    aom_highbd_sad32x32x4d_bits12,
-                   aom_highbd_jnt_sad32x32_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance32x32);
+                   aom_highbd_dist_wtd_sad32x32_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x32);
 
         HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits12,
                    aom_highbd_sad64x64_avg_bits12, aom_highbd_12_variance64x64,
                    aom_highbd_12_sub_pixel_variance64x64,
                    aom_highbd_12_sub_pixel_avg_variance64x64,
                    aom_highbd_sad64x64x4d_bits12,
-                   aom_highbd_jnt_sad64x64_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance64x64);
+                   aom_highbd_dist_wtd_sad64x64_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x64);
 
         HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits12,
                    aom_highbd_sad16x16_avg_bits12, aom_highbd_12_variance16x16,
                    aom_highbd_12_sub_pixel_variance16x16,
                    aom_highbd_12_sub_pixel_avg_variance16x16,
                    aom_highbd_sad16x16x4d_bits12,
-                   aom_highbd_jnt_sad16x16_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance16x16);
+                   aom_highbd_dist_wtd_sad16x16_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x16);
 
         HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits12,
                    aom_highbd_sad16x8_avg_bits12, aom_highbd_12_variance16x8,
                    aom_highbd_12_sub_pixel_variance16x8,
                    aom_highbd_12_sub_pixel_avg_variance16x8,
                    aom_highbd_sad16x8x4d_bits12,
-                   aom_highbd_jnt_sad16x8_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance16x8);
+                   aom_highbd_dist_wtd_sad16x8_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x8);
 
         HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits12,
                    aom_highbd_sad8x16_avg_bits12, aom_highbd_12_variance8x16,
                    aom_highbd_12_sub_pixel_variance8x16,
                    aom_highbd_12_sub_pixel_avg_variance8x16,
                    aom_highbd_sad8x16x4d_bits12,
-                   aom_highbd_jnt_sad8x16_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance8x16);
+                   aom_highbd_dist_wtd_sad8x16_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x16);
 
         HIGHBD_BFP(
             BLOCK_8X8, aom_highbd_sad8x8_bits12, aom_highbd_sad8x8_avg_bits12,
             aom_highbd_12_variance8x8, aom_highbd_12_sub_pixel_variance8x8,
             aom_highbd_12_sub_pixel_avg_variance8x8,
-            aom_highbd_sad8x8x4d_bits12, aom_highbd_jnt_sad8x8_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance8x8);
+            aom_highbd_sad8x8x4d_bits12, aom_highbd_dist_wtd_sad8x8_avg_bits12,
+            aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x8);
 
         HIGHBD_BFP(
             BLOCK_8X4, aom_highbd_sad8x4_bits12, aom_highbd_sad8x4_avg_bits12,
             aom_highbd_12_variance8x4, aom_highbd_12_sub_pixel_variance8x4,
             aom_highbd_12_sub_pixel_avg_variance8x4,
-            aom_highbd_sad8x4x4d_bits12, aom_highbd_jnt_sad8x4_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance8x4);
+            aom_highbd_sad8x4x4d_bits12, aom_highbd_dist_wtd_sad8x4_avg_bits12,
+            aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x4);
 
         HIGHBD_BFP(
             BLOCK_4X8, aom_highbd_sad4x8_bits12, aom_highbd_sad4x8_avg_bits12,
             aom_highbd_12_variance4x8, aom_highbd_12_sub_pixel_variance4x8,
             aom_highbd_12_sub_pixel_avg_variance4x8,
-            aom_highbd_sad4x8x4d_bits12, aom_highbd_jnt_sad4x8_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance4x8);
+            aom_highbd_sad4x8x4d_bits12, aom_highbd_dist_wtd_sad4x8_avg_bits12,
+            aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x8);
 
         HIGHBD_BFP(
             BLOCK_4X4, aom_highbd_sad4x4_bits12, aom_highbd_sad4x4_avg_bits12,
             aom_highbd_12_variance4x4, aom_highbd_12_sub_pixel_variance4x4,
             aom_highbd_12_sub_pixel_avg_variance4x4,
-            aom_highbd_sad4x4x4d_bits12, aom_highbd_jnt_sad4x4_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance4x4);
+            aom_highbd_sad4x4x4d_bits12, aom_highbd_dist_wtd_sad4x4_avg_bits12,
+            aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x4);
 
         HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits12,
                    aom_highbd_sad128x128_avg_bits12,
@@ -2158,24 +2255,26 @@ static void highbd_set_var_fns(AV1_COMP *const cpi) {
                    aom_highbd_12_sub_pixel_variance128x128,
                    aom_highbd_12_sub_pixel_avg_variance128x128,
                    aom_highbd_sad128x128x4d_bits12,
-                   aom_highbd_jnt_sad128x128_avg_bits12,
-                   aom_highbd_12_jnt_sub_pixel_avg_variance128x128);
+                   aom_highbd_dist_wtd_sad128x128_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x128);
 
-        HIGHBD_BFP(
-            BLOCK_128X64, aom_highbd_sad128x64_bits12,
-            aom_highbd_sad128x64_avg_bits12, aom_highbd_12_variance128x64,
-            aom_highbd_12_sub_pixel_variance128x64,
-            aom_highbd_12_sub_pixel_avg_variance128x64,
-            aom_highbd_sad128x64x4d_bits12, aom_highbd_jnt_sad128x64_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance128x64);
+        HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits12,
+                   aom_highbd_sad128x64_avg_bits12,
+                   aom_highbd_12_variance128x64,
+                   aom_highbd_12_sub_pixel_variance128x64,
+                   aom_highbd_12_sub_pixel_avg_variance128x64,
+                   aom_highbd_sad128x64x4d_bits12,
+                   aom_highbd_dist_wtd_sad128x64_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x64);
 
-        HIGHBD_BFP(
-            BLOCK_64X128, aom_highbd_sad64x128_bits12,
-            aom_highbd_sad64x128_avg_bits12, aom_highbd_12_variance64x128,
-            aom_highbd_12_sub_pixel_variance64x128,
-            aom_highbd_12_sub_pixel_avg_variance64x128,
-            aom_highbd_sad64x128x4d_bits12, aom_highbd_jnt_sad64x128_avg_bits12,
-            aom_highbd_12_jnt_sub_pixel_avg_variance64x128);
+        HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits12,
+                   aom_highbd_sad64x128_avg_bits12,
+                   aom_highbd_12_variance64x128,
+                   aom_highbd_12_sub_pixel_variance64x128,
+                   aom_highbd_12_sub_pixel_avg_variance64x128,
+                   aom_highbd_sad64x128x4d_bits12,
+                   aom_highbd_dist_wtd_sad64x128_avg_bits12,
+                   aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x128);
 
         HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits12,
                     aom_highbd_12_masked_sub_pixel_variance128x128)
@@ -2369,7 +2468,6 @@ void av1_change_config(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) {
 
   cpi->oxcf = *oxcf;
   cpi->common.options = oxcf->cfg;
-  cpi->row_mt = oxcf->row_mt;
   x->e_mbd.bd = (int)seq_params->bit_depth;
   x->e_mbd.global_motion = cm->global_motion;
 
@@ -2404,7 +2502,7 @@ void av1_change_config(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) {
   for (int i = 0; i < 2; ++i) {
     if (x->tmp_obmc_bufs[i] == NULL) {
       CHECK_MEM_ERROR(cm, x->tmp_obmc_bufs[i],
-                      aom_memalign(16, 2 * MAX_MB_PLANE * MAX_SB_SQUARE *
+                      aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE *
                                            sizeof(*x->tmp_obmc_bufs[i])));
       x->e_mbd.tmp_obmc_bufs[i] = x->tmp_obmc_bufs[i];
     }
@@ -2507,11 +2605,11 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
 
   CHECK_MEM_ERROR(cm, cm->fc,
                   (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc)));
-  CHECK_MEM_ERROR(cm, cm->frame_contexts,
-                  (FRAME_CONTEXT *)aom_memalign(
-                      32, FRAME_CONTEXTS * sizeof(*cm->frame_contexts)));
+  CHECK_MEM_ERROR(
+      cm, cm->default_frame_context,
+      (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->default_frame_context)));
   memset(cm->fc, 0, sizeof(*cm->fc));
-  memset(cm->frame_contexts, 0, FRAME_CONTEXTS * sizeof(*cm->frame_contexts));
+  memset(cm->default_frame_context, 0, sizeof(*cm->default_frame_context));
 
   cpi->resize_state = 0;
   cpi->resize_avg_qp = 0;
@@ -2522,12 +2620,12 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
   init_config(cpi, oxcf);
   av1_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
 
-  cm->current_video_frame = 0;
+  cm->current_frame.frame_number = 0;
+  cm->current_frame_id = -1;
   cpi->seq_params_locked = 0;
   cpi->partition_search_skippable_frame = 0;
   cpi->tile_data = NULL;
-  cpi->last_show_frame_buf_idx = INVALID_IDX;
-
+  cpi->last_show_frame_buf = NULL;
   realloc_segmentation_maps(cpi);
 
   memset(cpi->nmv_costs, 0, sizeof(cpi->nmv_costs));
@@ -2563,6 +2661,9 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
 
   cpi->count = 0;
   cpi->bytes = 0;
+#if CONFIG_SPEED_STATS
+  cpi->tx_search_count = 0;
+#endif  // CONFIG_SPEED_STATS
 
   if (cpi->b_calculate_psnr) {
     cpi->total_sq_error = 0;
@@ -2644,6 +2745,10 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
                   (int32_t *)aom_memalign(
                       16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.wsrc_buf)));
 
+  CHECK_MEM_ERROR(
+      cm, cpi->td.mb.inter_modes_info,
+      (InterModesInfo *)aom_malloc(sizeof(*cpi->td.mb.inter_modes_info)));
+
   for (int x = 0; x < 2; x++)
     for (int y = 0; y < 2; y++)
       CHECK_MEM_ERROR(
@@ -2657,8 +2762,27 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
                   (int32_t *)aom_memalign(
                       16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.mask_buf)));
 
-  av1_set_speed_features_framesize_independent(cpi);
-  av1_set_speed_features_framesize_dependent(cpi);
+  av1_set_speed_features_framesize_independent(cpi, oxcf->speed);
+  av1_set_speed_features_framesize_dependent(cpi, oxcf->speed);
+
+  for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) {
+    int mi_cols = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
+    int mi_rows = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2);
+
+    CHECK_MEM_ERROR(cm, cpi->tpl_stats[frame].tpl_stats_ptr,
+                    aom_calloc(mi_rows * mi_cols,
+                               sizeof(*cpi->tpl_stats[frame].tpl_stats_ptr)));
+    cpi->tpl_stats[frame].is_valid = 0;
+    cpi->tpl_stats[frame].width = mi_cols;
+    cpi->tpl_stats[frame].height = mi_rows;
+    cpi->tpl_stats[frame].stride = mi_cols;
+    cpi->tpl_stats[frame].mi_rows = cm->mi_rows;
+    cpi->tpl_stats[frame].mi_cols = cm->mi_cols;
+  }
+
+#if CONFIG_COLLECT_PARTITION_STATS
+  av1_zero(cpi->partition_stats);
+#endif
 
 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \
   cpi->fn_ptr[BT].sdf = SDF;                                    \
@@ -2672,103 +2796,109 @@ AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
 
   BFP(BLOCK_4X16, aom_sad4x16, aom_sad4x16_avg, aom_variance4x16,
       aom_sub_pixel_variance4x16, aom_sub_pixel_avg_variance4x16,
-      aom_sad4x16x4d, aom_jnt_sad4x16_avg, aom_jnt_sub_pixel_avg_variance4x16)
+      aom_sad4x16x4d, aom_dist_wtd_sad4x16_avg,
+      aom_dist_wtd_sub_pixel_avg_variance4x16)
 
   BFP(BLOCK_16X4, aom_sad16x4, aom_sad16x4_avg, aom_variance16x4,
       aom_sub_pixel_variance16x4, aom_sub_pixel_avg_variance16x4,
-      aom_sad16x4x4d, aom_jnt_sad16x4_avg, aom_jnt_sub_pixel_avg_variance16x4)
+      aom_sad16x4x4d, aom_dist_wtd_sad16x4_avg,
+      aom_dist_wtd_sub_pixel_avg_variance16x4)
 
   BFP(BLOCK_8X32, aom_sad8x32, aom_sad8x32_avg, aom_variance8x32,
       aom_sub_pixel_variance8x32, aom_sub_pixel_avg_variance8x32,
-      aom_sad8x32x4d, aom_jnt_sad8x32_avg, aom_jnt_sub_pixel_avg_variance8x32)
+      aom_sad8x32x4d, aom_dist_wtd_sad8x32_avg,
+      aom_dist_wtd_sub_pixel_avg_variance8x32)
 
   BFP(BLOCK_32X8, aom_sad32x8, aom_sad32x8_avg, aom_variance32x8,
       aom_sub_pixel_variance32x8, aom_sub_pixel_avg_variance32x8,
-      aom_sad32x8x4d, aom_jnt_sad32x8_avg, aom_jnt_sub_pixel_avg_variance32x8)
+      aom_sad32x8x4d, aom_dist_wtd_sad32x8_avg,
+      aom_dist_wtd_sub_pixel_avg_variance32x8)
 
   BFP(BLOCK_16X64, aom_sad16x64, aom_sad16x64_avg, aom_variance16x64,
       aom_sub_pixel_variance16x64, aom_sub_pixel_avg_variance16x64,
-      aom_sad16x64x4d, aom_jnt_sad16x64_avg,
-      aom_jnt_sub_pixel_avg_variance16x64)
+      aom_sad16x64x4d, aom_dist_wtd_sad16x64_avg,
+      aom_dist_wtd_sub_pixel_avg_variance16x64)
 
   BFP(BLOCK_64X16, aom_sad64x16, aom_sad64x16_avg, aom_variance64x16,
       aom_sub_pixel_variance64x16, aom_sub_pixel_avg_variance64x16,
-      aom_sad64x16x4d, aom_jnt_sad64x16_avg,
-      aom_jnt_sub_pixel_avg_variance64x16)
+      aom_sad64x16x4d, aom_dist_wtd_sad64x16_avg,
+      aom_dist_wtd_sub_pixel_avg_variance64x16)
 
   BFP(BLOCK_128X128, aom_sad128x128, aom_sad128x128_avg, aom_variance128x128,
       aom_sub_pixel_variance128x128, aom_sub_pixel_avg_variance128x128,
-      aom_sad128x128x4d, aom_jnt_sad128x128_avg,
-      aom_jnt_sub_pixel_avg_variance128x128)
+      aom_sad128x128x4d, aom_dist_wtd_sad128x128_avg,
+      aom_dist_wtd_sub_pixel_avg_variance128x128)
 
   BFP(BLOCK_128X64, aom_sad128x64, aom_sad128x64_avg, aom_variance128x64,
       aom_sub_pixel_variance128x64, aom_sub_pixel_avg_variance128x64,
-      aom_sad128x64x4d, aom_jnt_sad128x64_avg,
-      aom_jnt_sub_pixel_avg_variance128x64)
+      aom_sad128x64x4d, aom_dist_wtd_sad128x64_avg,
+      aom_dist_wtd_sub_pixel_avg_variance128x64)
 
   BFP(BLOCK_64X128, aom_sad64x128, aom_sad64x128_avg, aom_variance64x128,
       aom_sub_pixel_variance64x128, aom_sub_pixel_avg_variance64x128,
-      aom_sad64x128x4d, aom_jnt_sad64x128_avg,
-      aom_jnt_sub_pixel_avg_variance64x128)
+      aom_sad64x128x4d, aom_dist_wtd_sad64x128_avg,
+      aom_dist_wtd_sub_pixel_avg_variance64x128)
 
   BFP(BLOCK_32X16, aom_sad32x16, aom_sad32x16_avg, aom_variance32x16,
       aom_sub_pixel_variance32x16, aom_sub_pixel_avg_variance32x16,
-      aom_sad32x16x4d, aom_jnt_sad32x16_avg,
-      aom_jnt_sub_pixel_avg_variance32x16)
+      aom_sad32x16x4d, aom_dist_wtd_sad32x16_avg,
+      aom_dist_wtd_sub_pixel_avg_variance32x16)
 
   BFP(BLOCK_16X32, aom_sad16x32, aom_sad16x32_avg, aom_variance16x32,
       aom_sub_pixel_variance16x32, aom_sub_pixel_avg_variance16x32,
-      aom_sad16x32x4d, aom_jnt_sad16x32_avg,
-      aom_jnt_sub_pixel_avg_variance16x32)
+      aom_sad16x32x4d, aom_dist_wtd_sad16x32_avg,
+      aom_dist_wtd_sub_pixel_avg_variance16x32)
 
   BFP(BLOCK_64X32, aom_sad64x32, aom_sad64x32_avg, aom_variance64x32,
       aom_sub_pixel_variance64x32, aom_sub_pixel_avg_variance64x32,
-      aom_sad64x32x4d, aom_jnt_sad64x32_avg,
-      aom_jnt_sub_pixel_avg_variance64x32)
+      aom_sad64x32x4d, aom_dist_wtd_sad64x32_avg,
+      aom_dist_wtd_sub_pixel_avg_variance64x32)
 
   BFP(BLOCK_32X64, aom_sad32x64, aom_sad32x64_avg, aom_variance32x64,
       aom_sub_pixel_variance32x64, aom_sub_pixel_avg_variance32x64,
-      aom_sad32x64x4d, aom_jnt_sad32x64_avg,
-      aom_jnt_sub_pixel_avg_variance32x64)
+      aom_sad32x64x4d, aom_dist_wtd_sad32x64_avg,
+      aom_dist_wtd_sub_pixel_avg_variance32x64)
 
   BFP(BLOCK_32X32, aom_sad32x32, aom_sad32x32_avg, aom_variance32x32,
       aom_sub_pixel_variance32x32, aom_sub_pixel_avg_variance32x32,
-      aom_sad32x32x4d, aom_jnt_sad32x32_avg,
-      aom_jnt_sub_pixel_avg_variance32x32)
+      aom_sad32x32x4d, aom_dist_wtd_sad32x32_avg,
+      aom_dist_wtd_sub_pixel_avg_variance32x32)
 
   BFP(BLOCK_64X64, aom_sad64x64, aom_sad64x64_avg, aom_variance64x64,
       aom_sub_pixel_variance64x64, aom_sub_pixel_avg_variance64x64,
-      aom_sad64x64x4d, aom_jnt_sad64x64_avg,
-      aom_jnt_sub_pixel_avg_variance64x64)
+      aom_sad64x64x4d, aom_dist_wtd_sad64x64_avg,
+      aom_dist_wtd_sub_pixel_avg_variance64x64)
 
   BFP(BLOCK_16X16, aom_sad16x16, aom_sad16x16_avg, aom_variance16x16,
       aom_sub_pixel_variance16x16, aom_sub_pixel_avg_variance16x16,
-      aom_sad16x16x4d, aom_jnt_sad16x16_avg,
-      aom_jnt_sub_pixel_avg_variance16x16)
+      aom_sad16x16x4d, aom_dist_wtd_sad16x16_avg,
+      aom_dist_wtd_sub_pixel_avg_variance16x16)
 
   BFP(BLOCK_16X8, aom_sad16x8, aom_sad16x8_avg, aom_variance16x8,
       aom_sub_pixel_variance16x8, aom_sub_pixel_avg_variance16x8,
-      aom_sad16x8x4d, aom_jnt_sad16x8_avg, aom_jnt_sub_pixel_avg_variance16x8)
+      aom_sad16x8x4d, aom_dist_wtd_sad16x8_avg,
+      aom_dist_wtd_sub_pixel_avg_variance16x8)
 
   BFP(BLOCK_8X16, aom_sad8x16, aom_sad8x16_avg, aom_variance8x16,
       aom_sub_pixel_variance8x16, aom_sub_pixel_avg_variance8x16,
-      aom_sad8x16x4d, aom_jnt_sad8x16_avg, aom_jnt_sub_pixel_avg_variance8x16)
+      aom_sad8x16x4d, aom_dist_wtd_sad8x16_avg,
+      aom_dist_wtd_sub_pixel_avg_variance8x16)
 
   BFP(BLOCK_8X8, aom_sad8x8, aom_sad8x8_avg, aom_variance8x8,
       aom_sub_pixel_variance8x8, aom_sub_pixel_avg_variance8x8, aom_sad8x8x4d,
-      aom_jnt_sad8x8_avg, aom_jnt_sub_pixel_avg_variance8x8)
+      aom_dist_wtd_sad8x8_avg, aom_dist_wtd_sub_pixel_avg_variance8x8)
 
   BFP(BLOCK_8X4, aom_sad8x4, aom_sad8x4_avg, aom_variance8x4,
       aom_sub_pixel_variance8x4, aom_sub_pixel_avg_variance8x4, aom_sad8x4x4d,
-      aom_jnt_sad8x4_avg, aom_jnt_sub_pixel_avg_variance8x4)
+      aom_dist_wtd_sad8x4_avg, aom_dist_wtd_sub_pixel_avg_variance8x4)
 
   BFP(BLOCK_4X8, aom_sad4x8, aom_sad4x8_avg, aom_variance4x8,
       aom_sub_pixel_variance4x8, aom_sub_pixel_avg_variance4x8, aom_sad4x8x4d,
-      aom_jnt_sad4x8_avg, aom_jnt_sub_pixel_avg_variance4x8)
+      aom_dist_wtd_sad4x8_avg, aom_dist_wtd_sub_pixel_avg_variance4x8)
 
   BFP(BLOCK_4X4, aom_sad4x4, aom_sad4x4_avg, aom_variance4x4,
       aom_sub_pixel_variance4x4, aom_sub_pixel_avg_variance4x4, aom_sad4x4x4d,
-      aom_jnt_sad4x4_avg, aom_jnt_sub_pixel_avg_variance4x4)
+      aom_dist_wtd_sad4x4_avg, aom_dist_wtd_sub_pixel_avg_variance4x4)
 
 #define OBFP(BT, OSDF, OVF, OSVF) \
   cpi->fn_ptr[BT].osdf = OSDF;    \
@@ -2892,7 +3022,7 @@ void av1_remove_compressor(AV1_COMP *cpi) {
   cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
 
-  if (cm->current_video_frame > 0) {
+  if (cm->current_frame.frame_number > 0) {
 #if CONFIG_ENTROPY_STATS
     if (cpi->oxcf.pass != 1) {
       fprintf(stderr, "Writing counts.stt\n");
@@ -2966,9 +3096,25 @@ void av1_remove_compressor(AV1_COMP *cpi) {
       fclose(f);
     }
 #endif  // CONFIG_INTERNAL_STATS
+#if CONFIG_SPEED_STATS
+    if (cpi->oxcf.pass != 1) {
+      fprintf(stdout, "tx_search_count = %d\n", cpi->tx_search_count);
+    }
+#endif  // CONFIG_SPEED_STATS
+
+#if CONFIG_COLLECT_PARTITION_STATS
+    if (cpi->oxcf.pass != 1) {
+      av1_print_partition_stats(&cpi->partition_stats);
+    }
+#endif
   }
 
-  for (t = 0; t < cpi->num_workers; ++t) {
+  for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) {
+    aom_free(cpi->tpl_stats[frame].tpl_stats_ptr);
+    cpi->tpl_stats[frame].is_valid = 0;
+  }
+
+  for (t = cpi->num_workers - 1; t >= 0; --t) {
     AVxWorker *const worker = &cpi->workers[t];
     EncWorkerData *const thread_data = &cpi->tile_thr_data[t];
 
@@ -2976,7 +3122,8 @@ void av1_remove_compressor(AV1_COMP *cpi) {
     aom_get_worker_interface()->end(worker);
 
     // Deallocate allocated thread data.
-    if (t < cpi->num_workers - 1) {
+    if (cpi->row_mt == 1) aom_free(thread_data->td->tctx);
+    if (t > 0) {
       aom_free(thread_data->td->palette_buffer);
       aom_free(thread_data->td->tmp_conv_dst);
       for (int j = 0; j < 2; ++j) {
@@ -2985,6 +3132,8 @@ void av1_remove_compressor(AV1_COMP *cpi) {
       aom_free(thread_data->td->above_pred_buf);
       aom_free(thread_data->td->left_pred_buf);
       aom_free(thread_data->td->wsrc_buf);
+
+      aom_free(thread_data->td->inter_modes_info);
       for (int x = 0; x < 2; x++) {
         for (int y = 0; y < 2; y++) {
           aom_free(thread_data->td->hash_value_buffer[x][y]);
@@ -2997,6 +3146,15 @@ void av1_remove_compressor(AV1_COMP *cpi) {
       aom_free(thread_data->td);
     }
   }
+#if CONFIG_MULTITHREAD
+  if (cpi->row_mt == 1) {
+    if (cpi->row_mt_mutex_ != NULL) {
+      pthread_mutex_destroy(cpi->row_mt_mutex_);
+      aom_free(cpi->row_mt_mutex_);
+    }
+  }
+#endif
+  av1_row_mt_mem_dealloc(cpi);
   aom_free(cpi->tile_thr_data);
   aom_free(cpi->workers);
 
@@ -3043,7 +3201,7 @@ static void generate_psnr_packet(AV1_COMP *cpi) {
   struct aom_codec_cx_pkt pkt;
   int i;
   PSNR_STATS psnr;
-  aom_calc_highbd_psnr(cpi->source, cpi->common.frame_to_show, &psnr,
+  aom_calc_highbd_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr,
                        cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
 
   for (i = 0; i < 4; ++i) {
@@ -3062,15 +3220,6 @@ int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags) {
   return 0;
 }
 
-void av1_update_reference(AV1_COMP *cpi, int ref_frame_upd_flags) {
-  cpi->ext_refresh_last_frame = (ref_frame_upd_flags & AOM_LAST_FLAG) != 0;
-  cpi->ext_refresh_golden_frame = (ref_frame_upd_flags & AOM_GOLD_FLAG) != 0;
-  cpi->ext_refresh_alt_ref_frame = (ref_frame_upd_flags & AOM_ALT_FLAG) != 0;
-  cpi->ext_refresh_bwd_ref_frame = (ref_frame_upd_flags & AOM_BWD_FLAG) != 0;
-  cpi->ext_refresh_alt2_ref_frame = (ref_frame_upd_flags & AOM_ALT2_FLAG) != 0;
-  cpi->ext_refresh_frame_flags_pending = 1;
-}
-
 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) {
   AV1_COMMON *const cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
@@ -3133,61 +3282,6 @@ void aom_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) {
 }
 #endif
 
-static void check_show_existing_frame(AV1_COMP *cpi) {
-  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-  AV1_COMMON *const cm = &cpi->common;
-  const FRAME_UPDATE_TYPE next_frame_update_type =
-      gf_group->update_type[gf_group->index];
-#if USE_SYMM_MULTI_LAYER
-  const int which_arf = (cpi->new_bwdref_update_rule == 1)
-                            ? gf_group->arf_update_idx[gf_group->index] > 0
-                            : gf_group->arf_update_idx[gf_group->index];
-#else
-  const int which_arf = gf_group->arf_update_idx[gf_group->index];
-#endif
-
-  if (cm->show_existing_frame == 1) {
-    cm->show_existing_frame = 0;
-  } else if (cpi->rc.is_last_bipred_frame) {
-#if USE_SYMM_MULTI_LAYER
-    // NOTE: When new structure is used, every bwdref will have one overlay
-    //       frame. Therefore, there is no need to find out which frame to
-    //       show in advance.
-    if (cpi->new_bwdref_update_rule == 0) {
-#endif
-      // NOTE: If the current frame is a last bi-predictive frame, it is
-      //       needed next to show the BWDREF_FRAME, which is pointed by
-      //       the last_fb_idxes[0] after reference frame buffer update
-      cpi->rc.is_last_bipred_frame = 0;
-      cm->show_existing_frame = 1;
-      cpi->existing_fb_idx_to_show = cpi->ref_fb_idx[0];
-#if USE_SYMM_MULTI_LAYER
-    }
-#endif
-  } else if (cpi->is_arf_filter_off[which_arf] &&
-             (next_frame_update_type == OVERLAY_UPDATE ||
-              next_frame_update_type == INTNL_OVERLAY_UPDATE)) {
-#if USE_SYMM_MULTI_LAYER
-    const int bwdref_to_show =
-        (cpi->new_bwdref_update_rule == 1) ? BWDREF_FRAME : ALTREF2_FRAME;
-#else
-    const int bwdref_to_show = ALTREF2_FRAME;
-#endif
-    // Other parameters related to OVERLAY_UPDATE will be taken care of
-    // in av1_rc_get_second_pass_params(cpi)
-    cm->show_existing_frame = 1;
-    cpi->rc.is_src_frame_alt_ref = 1;
-    cpi->existing_fb_idx_to_show = (next_frame_update_type == OVERLAY_UPDATE)
-                                       ? cpi->ref_fb_idx[ALTREF_FRAME - 1]
-                                       : cpi->ref_fb_idx[bwdref_to_show - 1];
-#if USE_SYMM_MULTI_LAYER
-    if (cpi->new_bwdref_update_rule == 0)
-#endif
-      cpi->is_arf_filter_off[which_arf] = 0;
-  }
-  cpi->rc.is_src_frame_ext_arf = 0;
-}
-
 #ifdef OUTPUT_YUV_REC
 void aom_write_one_yuv_frame(AV1_COMMON *cm, YV12_BUFFER_CONFIG *s) {
   uint8_t *src = s->y_buffer;
@@ -3342,8 +3436,8 @@ static void dump_ref_frame_images(AV1_COMP *cpi) {
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
     char file_name[256] = "";
     snprintf(file_name, sizeof(file_name), "/tmp/enc_F%d_ref_%d.yuv",
-             cm->current_video_frame, ref_frame);
-    dump_one_image(cm, get_ref_frame_buffer(cpi, ref_frame), file_name);
+             cm->current_frame.frame_number, ref_frame);
+    dump_one_image(cm, get_ref_frame_yv12_buf(cpi, ref_frame), file_name);
   }
 }
 #endif  // DUMP_REF_FRAME_IMAGES == 1
@@ -3354,38 +3448,36 @@ static void dump_ref_frame_images(AV1_COMP *cpi) {
 // when the LAST_FRAME is updated.
 static INLINE void shift_last_ref_frames(AV1_COMP *cpi) {
   // TODO(isbs): shift the scaled indices as well
-  int ref_frame;
-  for (ref_frame = LAST_REF_FRAMES - 1; ref_frame > 0; --ref_frame) {
-    cpi->ref_fb_idx[ref_frame] = cpi->ref_fb_idx[ref_frame - 1];
+  for (int ref_frame = LAST3_FRAME; ref_frame > LAST_FRAME; --ref_frame) {
+    const int ref_idx = ref_frame - LAST_FRAME;
+    cpi->common.remapped_ref_idx[ref_idx] =
+        cpi->common.remapped_ref_idx[ref_idx - 1];
 
-    // [0] is allocated to the current coded frame. The statistics for the
-    // reference frames start at [LAST_FRAME], i.e. [1].
     if (!cpi->rc.is_src_frame_alt_ref) {
-      memcpy(cpi->interp_filter_selected[ref_frame + LAST_FRAME],
-             cpi->interp_filter_selected[ref_frame - 1 + LAST_FRAME],
-             sizeof(cpi->interp_filter_selected[ref_frame - 1 + LAST_FRAME]));
+      memcpy(cpi->interp_filter_selected[ref_frame],
+             cpi->interp_filter_selected[ref_frame - 1],
+             sizeof(cpi->interp_filter_selected[ref_frame - 1]));
     }
   }
 }
 
-#if USE_SYMM_MULTI_LAYER
 // This function is used to shift the virtual indices of bwd reference
 // frames as follows:
 // BWD_REF -> ALT2_REF -> EXT_REF
 // to clear a space to store the closest bwdref
 static INLINE void rshift_bwd_ref_frames(AV1_COMP *cpi) {
   // TODO(isbs): shift the scaled indices as well
-  static const int ordered_bwd[3] = { BWDREF_FRAME - 1, ALTREF2_FRAME - 1,
-                                      EXTREF_FRAME - 1 };
+  static const int ordered_bwd[3] = { BWDREF_FRAME, ALTREF2_FRAME,
+                                      EXTREF_FRAME };
 
   for (int i = 2; i > 0; --i) {
     // [0] is allocated to the current coded frame, i.e. bwdref
-    memcpy(
-        cpi->interp_filter_selected[ordered_bwd[i] + LAST_FRAME],
-        cpi->interp_filter_selected[ordered_bwd[i - 1] + LAST_FRAME],
-        sizeof(cpi->interp_filter_selected[ordered_bwd[i - 1] + LAST_FRAME]));
+    memcpy(cpi->interp_filter_selected[ordered_bwd[i]],
+           cpi->interp_filter_selected[ordered_bwd[i - 1]],
+           sizeof(cpi->interp_filter_selected[ordered_bwd[i - 1]]));
 
-    cpi->ref_fb_idx[ordered_bwd[i]] = cpi->ref_fb_idx[ordered_bwd[i - 1]];
+    cpi->common.remapped_ref_idx[ordered_bwd[i] - LAST_FRAME] =
+        cpi->common.remapped_ref_idx[ordered_bwd[i - 1] - LAST_FRAME];
   }
 }
 
@@ -3395,33 +3487,37 @@ static INLINE void rshift_bwd_ref_frames(AV1_COMP *cpi) {
 // to update the bwd reference frame for coding the next frame.
 static INLINE void lshift_bwd_ref_frames(AV1_COMP *cpi) {
   // TODO(isbs): shift the scaled indices as well
-  static const int ordered_bwd[3] = { BWDREF_FRAME - 1, ALTREF2_FRAME - 1,
-                                      EXTREF_FRAME - 1 };
+  static const int ordered_bwd[3] = { BWDREF_FRAME, ALTREF2_FRAME,
+                                      EXTREF_FRAME };
 
   for (int i = 0; i < 2; ++i) {
     // [0] is allocated to the current coded frame, i.e. bwdref
-    memcpy(
-        cpi->interp_filter_selected[ordered_bwd[i] + LAST_FRAME],
-        cpi->interp_filter_selected[ordered_bwd[i + 1] + LAST_FRAME],
-        sizeof(cpi->interp_filter_selected[ordered_bwd[i + 1] + LAST_FRAME]));
+    memcpy(cpi->interp_filter_selected[ordered_bwd[i]],
+           cpi->interp_filter_selected[ordered_bwd[i + 1]],
+           sizeof(cpi->interp_filter_selected[ordered_bwd[i + 1]]));
 
-    cpi->ref_fb_idx[ordered_bwd[i]] = cpi->ref_fb_idx[ordered_bwd[i + 1]];
+    cpi->common.remapped_ref_idx[ordered_bwd[i] - LAST_FRAME] =
+        cpi->common.remapped_ref_idx[ordered_bwd[i + 1] - LAST_FRAME];
   }
 }
-#endif  // USE_SYMM_MULTI_LAYER
 
 static void update_reference_frames(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
 
   // NOTE: Save the new show frame buffer index for --test-code=warn, i.e.,
   //       for the purpose to verify no mismatch between encoder and decoder.
-  if (cm->show_frame) cpi->last_show_frame_buf_idx = cm->new_fb_idx;
+  if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame;
 
   // In the case of show_existing frame, we will not send fresh flag
   // to decoder. Any change in the reference frame buffer can be done by
   // switching the virtual indices.
   if (cm->show_existing_frame) {
-    cpi->refresh_last_frame = 0;
+    // If we are not indicating to the decoder that this frame is
+    // a show_existing_frame, which occurs in error_resilient mode,
+    // we still want to refresh the LAST_FRAME when the current frame
+    // was the source of an ext_arf.
+    cpi->refresh_last_frame =
+        !encode_show_existing_frame(cm) && cpi->rc.is_src_frame_ext_arf;
     cpi->refresh_golden_frame = 0;
     cpi->refresh_bwd_ref_frame = 0;
     cpi->refresh_alt2_ref_frame = 0;
@@ -3432,18 +3528,16 @@ static void update_reference_frames(AV1_COMP *cpi) {
     cpi->rc.is_bipred_frame = 0;
   }
 
-  BufferPool *const pool = cm->buffer_pool;
-
   // At this point the new frame has been encoded.
   // If any buffer copy / swapping is signaled it should be done here.
 
   // Only update all of the reference buffers if a KEY_FRAME is also a
   // show_frame. This ensures a fwd keyframe does not update all of the buffers
-  if ((cm->frame_type == KEY_FRAME && cm->show_frame) || frame_is_sframe(cm)) {
+  if ((cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) ||
+      frame_is_sframe(cm)) {
     for (int ref_frame = 0; ref_frame < REF_FRAMES; ++ref_frame) {
-      ref_cnt_fb(pool->frame_bufs,
-                 &cm->ref_frame_map[cpi->ref_fb_idx[ref_frame]],
-                 cm->new_fb_idx);
+      assign_frame_buffer_p(&cm->ref_frame_map[cm->remapped_ref_idx[ref_frame]],
+                            cm->cur_frame);
     }
     return;
   }
@@ -3463,56 +3557,50 @@ static void update_reference_frames(AV1_COMP *cpi) {
     // reference instead of replacing it with overlay.
 
     if (!cpi->preserve_arf_as_gld) {
-      ref_cnt_fb(pool->frame_bufs,
-                 &cm->ref_frame_map[cpi->ref_fb_idx[ALTREF_FRAME - 1]],
-                 cm->new_fb_idx);
+      assign_frame_buffer_p(
+          &cm->ref_frame_map[get_ref_frame_map_idx(cm, ALTREF_FRAME)],
+          cm->cur_frame);
     }
 
-    tmp = cpi->ref_fb_idx[ALTREF_FRAME - 1];
-    cpi->ref_fb_idx[ALTREF_FRAME - 1] = cpi->ref_fb_idx[GOLDEN_FRAME - 1];
-    cpi->ref_fb_idx[GOLDEN_FRAME - 1] = tmp;
+    tmp = get_ref_frame_map_idx(cm, ALTREF_FRAME);
+    cm->remapped_ref_idx[ALTREF_FRAME - LAST_FRAME] =
+        get_ref_frame_map_idx(cm, GOLDEN_FRAME);
+    cm->remapped_ref_idx[GOLDEN_FRAME - LAST_FRAME] = tmp;
 
     // TODO(zoeliu): Do we need to copy cpi->interp_filter_selected[0] over to
     // cpi->interp_filter_selected[GOLDEN_FRAME]?
-  } else if (cpi->rc.is_src_frame_ext_arf && cm->show_existing_frame) {
+  } else if (cpi->rc.is_src_frame_ext_arf && encode_show_existing_frame(cm)) {
 #if CONFIG_DEBUG
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     assert(gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE);
-#endif
-#if USE_SYMM_MULTI_LAYER
+#endif  // CONFIG_DEBUG
     const int bwdref_to_show =
         (cpi->new_bwdref_update_rule == 1) ? BWDREF_FRAME : ALTREF2_FRAME;
-#else
-    const int bwdref_to_show = ALTREF2_FRAME;
-#endif
     // Deal with the special case for showing existing internal ALTREF_FRAME
     // Refresh the LAST_FRAME with the ALTREF_FRAME and retire the LAST3_FRAME
     // by updating the virtual indices.
-    const int tmp = cpi->ref_fb_idx[LAST_REF_FRAMES - 1];
+    const int last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME);
     shift_last_ref_frames(cpi);
 
-    cpi->ref_fb_idx[LAST_FRAME - 1] = cpi->ref_fb_idx[bwdref_to_show - 1];
+    cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] =
+        get_ref_frame_map_idx(cm, bwdref_to_show);
 
     memcpy(cpi->interp_filter_selected[LAST_FRAME],
            cpi->interp_filter_selected[bwdref_to_show],
            sizeof(cpi->interp_filter_selected[bwdref_to_show]));
-#if USE_SYMM_MULTI_LAYER
     if (cpi->new_bwdref_update_rule == 1) {
       lshift_bwd_ref_frames(cpi);
       // pass outdated forward reference frame (previous LAST3) to the
       // spared space
-      cpi->ref_fb_idx[EXTREF_FRAME - 1] = tmp;
+      cm->remapped_ref_idx[EXTREF_FRAME - LAST_FRAME] = last3_remapped_idx;
     } else {
-#endif
-      cpi->ref_fb_idx[bwdref_to_show - 1] = tmp;
-#if USE_SYMM_MULTI_LAYER
+      cm->remapped_ref_idx[bwdref_to_show - LAST_FRAME] = last3_remapped_idx;
     }
-#endif
   } else { /* For non key/golden frames */
     // === ALTREF_FRAME ===
     if (cpi->refresh_alt_ref_frame) {
-      int arf_idx = cpi->ref_fb_idx[ALTREF_FRAME - 1];
-      ref_cnt_fb(pool->frame_bufs, &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
+      int arf_idx = get_ref_frame_map_idx(cm, ALTREF_FRAME);
+      assign_frame_buffer_p(&cm->ref_frame_map[arf_idx], cm->cur_frame);
 
       memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
              cpi->interp_filter_selected[0],
@@ -3521,9 +3609,9 @@ static void update_reference_frames(AV1_COMP *cpi) {
 
     // === GOLDEN_FRAME ===
     if (cpi->refresh_golden_frame) {
-      ref_cnt_fb(pool->frame_bufs,
-                 &cm->ref_frame_map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]],
-                 cm->new_fb_idx);
+      assign_frame_buffer_p(
+          &cm->ref_frame_map[get_ref_frame_map_idx(cm, GOLDEN_FRAME)],
+          cm->cur_frame);
 
       memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
              cpi->interp_filter_selected[0],
@@ -3532,25 +3620,21 @@ static void update_reference_frames(AV1_COMP *cpi) {
 
     // === BWDREF_FRAME ===
     if (cpi->refresh_bwd_ref_frame) {
-#if USE_SYMM_MULTI_LAYER
       if (cpi->new_bwdref_update_rule) {
         // We shift the backward reference frame as follows:
         // BWDREF -> ALTREF2 -> EXTREF
         // and assign the newly coded frame to BWDREF so that it always
         // keeps the nearest future frame
-        int tmp = cpi->ref_fb_idx[EXTREF_FRAME - 1];
-        ref_cnt_fb(pool->frame_bufs, &cm->ref_frame_map[tmp], cm->new_fb_idx);
+        const int tmp = get_ref_frame_map_idx(cm, EXTREF_FRAME);
+        assign_frame_buffer_p(&cm->ref_frame_map[tmp], cm->cur_frame);
 
         rshift_bwd_ref_frames(cpi);
-        cpi->ref_fb_idx[BWDREF_FRAME - 1] = tmp;
+        cm->remapped_ref_idx[BWDREF_FRAME - LAST_FRAME] = tmp;
       } else {
-#endif  // USE_SYMM_MULTI_LAYER
-        ref_cnt_fb(pool->frame_bufs,
-                   &cm->ref_frame_map[cpi->ref_fb_idx[BWDREF_FRAME - 1]],
-                   cm->new_fb_idx);
-#if USE_SYMM_MULTI_LAYER
+        assign_frame_buffer_p(
+            &cm->ref_frame_map[get_ref_frame_map_idx(cm, BWDREF_FRAME)],
+            cm->cur_frame);
       }
-#endif
       memcpy(cpi->interp_filter_selected[BWDREF_FRAME],
              cpi->interp_filter_selected[0],
              sizeof(cpi->interp_filter_selected[0]));
@@ -3558,9 +3642,9 @@ static void update_reference_frames(AV1_COMP *cpi) {
 
     // === ALTREF2_FRAME ===
     if (cpi->refresh_alt2_ref_frame) {
-      ref_cnt_fb(pool->frame_bufs,
-                 &cm->ref_frame_map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]],
-                 cm->new_fb_idx);
+      assign_frame_buffer_p(
+          &cm->ref_frame_map[get_ref_frame_map_idx(cm, ALTREF2_FRAME)],
+          cm->cur_frame);
 
       memcpy(cpi->interp_filter_selected[ALTREF2_FRAME],
              cpi->interp_filter_selected[0],
@@ -3571,15 +3655,15 @@ static void update_reference_frames(AV1_COMP *cpi) {
   if (cpi->refresh_last_frame) {
     // NOTE(zoeliu): We have two layers of mapping (1) from the per-frame
     // reference to the reference frame buffer virtual index; and then (2) from
-    // the virtual index to the reference frame buffer physical index:
+    // the virtual index to the reference frame buffer (RefCntBuffer):
     //
-    // LAST_FRAME,      ..., LAST3_FRAME,     ..., ALTREF_FRAME
-    //      |                     |                     |
-    //      v                     v                     v
-    // ref_fb_idx[0],   ..., ref_fb_idx[2],   ..., ref_fb_idx[ALTREF_FRAME-1]
-    //      |                     |                     |
-    //      v                     v                     v
-    // ref_frame_map[], ..., ref_frame_map[], ..., ref_frame_map[]
+    // LAST_FRAME,                       ...,     EXTREF_FRAME
+    //      |                                           |
+    //      v                                           v
+    // remapped_ref_idx[LAST_FRAME - 1], ..., remapped_ref_idx[EXTREF_FRAME - 1]
+    //      |                                           |
+    //      v                                           v
+    // ref_frame_map[],                  ...,    ref_frame_map[]
     //
     // When refresh_last_frame is set, it is intended to retire LAST3_FRAME,
     // have the other 2 LAST reference frames shifted as follows:
@@ -3590,27 +3674,25 @@ static void update_reference_frames(AV1_COMP *cpi) {
     //
     // (a) To change ref_frame_map[] and have the virtual index of LAST3_FRAME
     //     to point to the newly coded frame, i.e.
-    //     ref_frame_map[lst_fb_idexes[2]] => new_fb_idx;
+    //     ref_frame_map[lst_fb_idexes[2]] => cur_frame;
     //
     // (b) To change the 1st layer mapping to have LAST_FRAME mapped to the
     //     original virtual index of LAST3_FRAME and have the other mappings
     //     shifted as follows:
-    // LAST_FRAME,      LAST2_FRAME,     LAST3_FRAME
-    //      |                |                |
-    //      v                v                v
-    // ref_fb_idx[2],   ref_fb_idx[0],   ref_fb_idx[1]
-    int tmp;
+    // LAST_FRAME,            LAST2_FRAME,             LAST3_FRAME
+    //      |                      |                        |
+    //      v                      v                        v
+    // remapped_ref_idx[2],   remapped_ref_idx[0],     remapped_ref_idx[1]
+    assign_frame_buffer_p(
+        &cm->ref_frame_map[get_ref_frame_map_idx(cm, LAST3_FRAME)],
+        cm->cur_frame);
 
-    ref_cnt_fb(pool->frame_bufs,
-               &cm->ref_frame_map[cpi->ref_fb_idx[LAST_REF_FRAMES - 1]],
-               cm->new_fb_idx);
-
-    tmp = cpi->ref_fb_idx[LAST_REF_FRAMES - 1];
+    int last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME);
 
     shift_last_ref_frames(cpi);
-    cpi->ref_fb_idx[0] = tmp;
+    cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] = last3_remapped_idx;
 
-    assert(cm->show_existing_frame == 0);
+    assert(!encode_show_existing_frame(cm));
     memcpy(cpi->interp_filter_selected[LAST_FRAME],
            cpi->interp_filter_selected[0],
            sizeof(cpi->interp_filter_selected[0]));
@@ -3618,11 +3700,7 @@ static void update_reference_frames(AV1_COMP *cpi) {
     // If the new structure is used, we will always have overlay frames coupled
     // with bwdref frames. Therefore, we won't have to perform this update
     // in advance (we do this update when the overlay frame shows up).
-#if USE_SYMM_MULTI_LAYER
     if (cpi->new_bwdref_update_rule == 0 && cpi->rc.is_last_bipred_frame) {
-#else
-    if (cpi->rc.is_last_bipred_frame) {
-#endif
       // Refresh the LAST_FRAME with the BWDREF_FRAME and retire the
       // LAST3_FRAME by updating the virtual indices.
       //
@@ -3631,11 +3709,12 @@ static void update_reference_frames(AV1_COMP *cpi) {
       //       virtual index reshuffling for BWDREF, the encoder always
       //       specifies a LAST_BIPRED right before BWDREF and completes the
       //       reshuffling job accordingly.
-      tmp = cpi->ref_fb_idx[LAST_REF_FRAMES - 1];
+      last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME);
 
       shift_last_ref_frames(cpi);
-      cpi->ref_fb_idx[0] = cpi->ref_fb_idx[BWDREF_FRAME - 1];
-      cpi->ref_fb_idx[BWDREF_FRAME - 1] = tmp;
+      cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] =
+          get_ref_frame_map_idx(cm, BWDREF_FRAME);
+      cm->remapped_ref_idx[BWDREF_FRAME - LAST_FRAME] = last3_remapped_idx;
 
       memcpy(cpi->interp_filter_selected[LAST_FRAME],
              cpi->interp_filter_selected[BWDREF_FRAME],
@@ -3649,14 +3728,6 @@ static void update_reference_frames(AV1_COMP *cpi) {
 #endif  // DUMP_REF_FRAME_IMAGES
 }
 
-static INLINE void alloc_frame_mvs(AV1_COMMON *const cm, int buffer_idx) {
-  assert(buffer_idx != INVALID_IDX);
-  RefCntBuffer *const new_fb_ptr = &cm->buffer_pool->frame_bufs[buffer_idx];
-  ensure_mv_buffer(new_fb_ptr, cm);
-  new_fb_ptr->width = cm->width;
-  new_fb_ptr->height = cm->height;
-}
-
 static void scale_references(AV1_COMP *cpi) {
   AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
@@ -3671,62 +3742,65 @@ static void scale_references(AV1_COMP *cpi) {
     if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) {
       BufferPool *const pool = cm->buffer_pool;
       const YV12_BUFFER_CONFIG *const ref =
-          get_ref_frame_buffer(cpi, ref_frame);
+          get_ref_frame_yv12_buf(cm, ref_frame);
 
       if (ref == NULL) {
-        cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
+        cpi->scaled_ref_buf[ref_frame - 1] = NULL;
         continue;
       }
 
       if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
-        RefCntBuffer *new_fb_ptr = NULL;
         int force_scaling = 0;
-        int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
-        if (new_fb == INVALID_IDX) {
-          new_fb = get_free_fb(cm);
+        RefCntBuffer *new_fb = cpi->scaled_ref_buf[ref_frame - 1];
+        if (new_fb == NULL) {
+          const int new_fb_idx = get_free_fb(cm);
+          if (new_fb_idx == INVALID_IDX) {
+            aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                               "Unable to find free frame buffer");
+          }
           force_scaling = 1;
+          new_fb = &pool->frame_bufs[new_fb_idx];
         }
-        if (new_fb == INVALID_IDX) return;
-        new_fb_ptr = &pool->frame_bufs[new_fb];
-        if (force_scaling || new_fb_ptr->buf.y_crop_width != cm->width ||
-            new_fb_ptr->buf.y_crop_height != cm->height) {
+
+        if (force_scaling || new_fb->buf.y_crop_width != cm->width ||
+            new_fb->buf.y_crop_height != cm->height) {
           if (aom_realloc_frame_buffer(
-                  &new_fb_ptr->buf, cm->width, cm->height,
+                  &new_fb->buf, cm->width, cm->height,
                   cm->seq_params.subsampling_x, cm->seq_params.subsampling_y,
-                  cm->seq_params.use_highbitdepth, AOM_BORDER_IN_PIXELS,
-                  cm->byte_alignment, NULL, NULL, NULL))
+                  cm->seq_params.use_highbitdepth, cpi->oxcf.border_in_pixels,
+                  cm->byte_alignment, NULL, NULL, NULL)) {
+            if (force_scaling) {
+              // Release the reference acquired in the get_free_fb() call above.
+              --new_fb->ref_count;
+            }
             aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                                "Failed to allocate frame buffer");
+          }
           av1_resize_and_extend_frame(
-              ref, &new_fb_ptr->buf, (int)cm->seq_params.bit_depth, num_planes);
-          cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
+              ref, &new_fb->buf, (int)cm->seq_params.bit_depth, num_planes);
+          cpi->scaled_ref_buf[ref_frame - 1] = new_fb;
           alloc_frame_mvs(cm, new_fb);
         }
       } else {
-        const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-        RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
+        RefCntBuffer *buf = get_ref_frame_buf(cm, ref_frame);
         buf->buf.y_crop_width = ref->y_crop_width;
         buf->buf.y_crop_height = ref->y_crop_height;
-        cpi->scaled_ref_idx[ref_frame - 1] = buf_idx;
+        cpi->scaled_ref_buf[ref_frame - 1] = buf;
         ++buf->ref_count;
       }
     } else {
-      if (cpi->oxcf.pass != 0) cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
+      if (cpi->oxcf.pass != 0) cpi->scaled_ref_buf[ref_frame - 1] = NULL;
     }
   }
 }
 
 static void release_scaled_references(AV1_COMP *cpi) {
-  AV1_COMMON *cm = &cpi->common;
-  int i;
   // TODO(isbs): only refresh the necessary frames, rather than all of them
-  for (i = 0; i < REF_FRAMES; ++i) {
-    const int idx = cpi->scaled_ref_idx[i];
-    RefCntBuffer *const buf =
-        idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[idx] : NULL;
+  for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) {
+    RefCntBuffer *const buf = cpi->scaled_ref_buf[i];
     if (buf != NULL) {
       --buf->ref_count;
-      cpi->scaled_ref_idx[i] = INVALID_IDX;
+      cpi->scaled_ref_buf[i] = NULL;
     }
   }
 }
@@ -3758,15 +3832,30 @@ static void set_mv_search_params(AV1_COMP *cpi) {
 
 static void set_size_independent_vars(AV1_COMP *cpi) {
   int i;
+  AV1_COMMON *cm = &cpi->common;
   for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
-    cpi->common.global_motion[i] = default_warp_params;
+    cm->global_motion[i] = default_warp_params;
   }
   cpi->global_motion_search_done = 0;
-  av1_set_speed_features_framesize_independent(cpi);
+  av1_set_speed_features_framesize_independent(cpi, cpi->speed);
   av1_set_rd_speed_thresholds(cpi);
-  av1_set_rd_speed_thresholds_sub8x8(cpi);
-  cpi->common.interp_filter = SWITCHABLE;
-  cpi->common.switchable_motion_mode = 1;
+  cm->interp_filter = SWITCHABLE;
+  cm->switchable_motion_mode = 1;
+
+  if (frame_is_intra_only(cm)) {
+    if (cm->seq_params.force_screen_content_tools == 2) {
+      cm->allow_screen_content_tools =
+          cpi->oxcf.content == AOM_CONTENT_SCREEN ||
+          is_screen_content(cpi->source->y_buffer,
+                            cpi->source->flags & YV12_FLAG_HIGHBITDEPTH,
+                            cm->seq_params.bit_depth, cpi->source->y_stride,
+                            cpi->source->y_width, cpi->source->y_height);
+    } else {
+      cm->allow_screen_content_tools =
+          cm->seq_params.force_screen_content_tools;
+    }
+  }
+  cpi->is_screen_content_type = (cm->allow_screen_content_tools != 0);
 }
 
 static void set_size_dependent_vars(AV1_COMP *cpi, int *q, int *bottom_index,
@@ -3775,7 +3864,7 @@ static void set_size_dependent_vars(AV1_COMP *cpi, int *q, int *bottom_index,
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
 
   // Setup variables that depend on the dimensions of the frame.
-  av1_set_speed_features_framesize_dependent(cpi);
+  av1_set_speed_features_framesize_dependent(cpi, cpi->speed);
 
   // Decide q and q bounds.
   *q = av1_rc_pick_q_and_bounds(cpi, cm->width, cm->height, bottom_index,
@@ -3829,9 +3918,11 @@ static void init_ref_frame_bufs(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
   int i;
   BufferPool *const pool = cm->buffer_pool;
-  cm->new_fb_idx = INVALID_IDX;
+  cm->cur_frame = NULL;
   for (i = 0; i < REF_FRAMES; ++i) {
-    cm->ref_frame_map[i] = INVALID_IDX;
+    cm->ref_frame_map[i] = NULL;
+  }
+  for (i = 0; i < FRAME_BUFFERS; ++i) {
     pool->frame_bufs[i].ref_count = 0;
   }
   if (cm->seq_params.force_screen_content_tools) {
@@ -3891,7 +3982,7 @@ static int set_size_literal(AV1_COMP *cpi, int width, int height) {
   return 0;
 }
 
-static void set_frame_size(AV1_COMP *cpi, int width, int height) {
+void av1_set_frame_size(AV1_COMP *cpi, int width, int height) {
   AV1_COMMON *const cm = &cpi->common;
   const SequenceHeader *const seq_params = &cm->seq_params;
   const int num_planes = av1_num_planes(cm);
@@ -3910,7 +4001,7 @@ static void set_frame_size(AV1_COMP *cpi, int width, int height) {
     av1_set_target_rate(cpi, cm->width, cm->height);
   }
 
-  alloc_frame_mvs(cm, cm->new_fb_idx);
+  alloc_frame_mvs(cm, cm->cur_frame);
 
   // Allocate above context buffers
   if (cm->num_allocated_above_context_planes < av1_num_planes(cm) ||
@@ -3924,10 +4015,9 @@ static void set_frame_size(AV1_COMP *cpi, int width, int height) {
 
   // Reset the frame pointers to the current frame size.
   if (aom_realloc_frame_buffer(
-          get_frame_new_buffer(cm), cm->width, cm->height,
-          seq_params->subsampling_x, seq_params->subsampling_y,
-          seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS,
-          cm->byte_alignment, NULL, NULL, NULL))
+          &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x,
+          seq_params->subsampling_y, seq_params->use_highbitdepth,
+          cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL))
     aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
                        "Failed to allocate frame buffer");
 
@@ -3940,25 +4030,17 @@ static void set_frame_size(AV1_COMP *cpi, int width, int height) {
     cm->rst_info[i].frame_restoration_type = RESTORE_NONE;
 
   av1_alloc_restoration_buffers(cm);
-  alloc_util_frame_buffers(cpi);  // TODO(afergs): Remove? Gets called anyways.
+  alloc_util_frame_buffers(cpi);
   init_motion_estimation(cpi);
 
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - LAST_FRAME];
-    const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-
-    ref_buf->idx = buf_idx;
-
-    if (buf_idx != INVALID_IDX) {
-      YV12_BUFFER_CONFIG *const buf = &cm->buffer_pool->frame_bufs[buf_idx].buf;
-      ref_buf->buf = buf;
-      av1_setup_scale_factors_for_frame(&ref_buf->sf, buf->y_crop_width,
-                                        buf->y_crop_height, cm->width,
+    RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    if (buf != NULL) {
+      struct scale_factors *sf = get_ref_scale_factors(cm, ref_frame);
+      av1_setup_scale_factors_for_frame(sf, buf->buf.y_crop_width,
+                                        buf->buf.y_crop_height, cm->width,
                                         cm->height);
-      if (av1_is_scaled(&ref_buf->sf))
-        aom_extend_frame_borders(buf, num_planes);
-    } else {
-      ref_buf->buf = NULL;
+      if (av1_is_scaled(sf)) aom_extend_frame_borders(&buf->buf, num_planes);
     }
   }
 
@@ -3979,7 +4061,7 @@ static uint8_t calculate_next_resize_scale(const AV1_COMP *cpi) {
   switch (oxcf->resize_mode) {
     case RESIZE_NONE: new_denom = SCALE_NUMERATOR; break;
     case RESIZE_FIXED:
-      if (cpi->common.frame_type == KEY_FRAME)
+      if (cpi->common.current_frame.frame_type == KEY_FRAME)
         new_denom = oxcf->resize_kf_scale_denominator;
       else
         new_denom = oxcf->resize_scale_denominator;
@@ -3990,6 +4072,35 @@ static uint8_t calculate_next_resize_scale(const AV1_COMP *cpi) {
   return new_denom;
 }
 
+#define ENERGY_BY_Q2_THRESH 0.01
+#define ENERGY_BY_AC_THRESH 0.2
+
+static uint8_t get_superres_denom_from_qindex_energy(int qindex, double *energy,
+                                                     double threshq,
+                                                     double threshp) {
+  const double q = av1_convert_qindex_to_q(qindex, AOM_BITS_8);
+  const double tq = threshq * q * q;
+  const double tp = threshp * energy[1];
+  const double thresh = AOMMIN(tq, tp);
+  int k;
+  for (k = 16; k > 8; --k) {
+    if (energy[k - 1] > thresh) break;
+  }
+  return 3 * SCALE_NUMERATOR - k;
+}
+
+static uint8_t get_superres_denom_for_qindex(const AV1_COMP *cpi, int qindex) {
+  double energy[16];
+  analyze_hor_freq(cpi, energy);
+  /*
+  printf("\nenergy = [");
+  for (int k = 1; k < 16; ++k) printf("%f, ", energy[k]);
+  printf("]\n");
+  */
+  return get_superres_denom_from_qindex_energy(
+      qindex, energy, ENERGY_BY_Q2_THRESH, ENERGY_BY_AC_THRESH);
+}
+
 static uint8_t calculate_next_superres_scale(AV1_COMP *cpi) {
   // Choose an arbitrary random number
   static unsigned int seed = 34567;
@@ -4007,38 +4118,28 @@ static uint8_t calculate_next_superres_scale(AV1_COMP *cpi) {
   switch (oxcf->superres_mode) {
     case SUPERRES_NONE: new_denom = SCALE_NUMERATOR; break;
     case SUPERRES_FIXED:
-      if (cpi->common.frame_type == KEY_FRAME)
+      if (cpi->common.current_frame.frame_type == KEY_FRAME)
         new_denom = oxcf->superres_kf_scale_denominator;
       else
         new_denom = oxcf->superres_scale_denominator;
       break;
     case SUPERRES_RANDOM: new_denom = lcg_rand16(&seed) % 9 + 8; break;
     case SUPERRES_QTHRESH: {
-      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-      const RATE_FACTOR_LEVEL rf_level = gf_group->rf_level[gf_group->index];
-      const double rate_factor_delta = rate_factor_deltas[rf_level];
-      const int qthresh = (rate_factor_delta <= 1.0)
-                              ? oxcf->superres_qthresh
-                              : oxcf->superres_kf_qthresh;
-      av1_set_target_rate(cpi, cpi->oxcf.width, cpi->oxcf.height);
+      // Do not use superres when screen content tools are used.
+      if (cpi->common.allow_screen_content_tools) break;
+      if (oxcf->rc_mode == AOM_VBR || oxcf->rc_mode == AOM_CQ)
+        av1_set_target_rate(cpi, cpi->oxcf.width, cpi->oxcf.height);
       int bottom_index, top_index;
       const int q = av1_rc_pick_q_and_bounds(
           cpi, cpi->oxcf.width, cpi->oxcf.height, &bottom_index, &top_index);
-      if (q < qthresh) {
+
+      const int qthresh = (frame_is_intra_only(&cpi->common))
+                              ? oxcf->superres_kf_qthresh
+                              : oxcf->superres_qthresh;
+      if (q <= qthresh) {
         new_denom = SCALE_NUMERATOR;
       } else {
-        const uint8_t min_denom = SCALE_NUMERATOR + 1;
-        const uint8_t denom_step = (MAXQ - qthresh + 1) >> 3;
-
-        if (q == qthresh) {
-          new_denom = min_denom;
-        } else if (denom_step == 0) {
-          new_denom = SCALE_NUMERATOR << 1;
-        } else {
-          const uint8_t additional_denom = (q - qthresh) / denom_step;
-          new_denom =
-              AOMMIN(min_denom + additional_denom, SCALE_NUMERATOR << 1);
-        }
+        new_denom = get_superres_denom_for_qindex(cpi, q);
       }
       break;
     }
@@ -4115,7 +4216,7 @@ static int validate_size_scales(RESIZE_MODE resize_mode,
 }
 
 // Calculates resize and superres params for next frame
-size_params_type av1_calculate_next_size_params(AV1_COMP *cpi) {
+static size_params_type calculate_next_size_params(AV1_COMP *cpi) {
   const AV1EncoderConfig *oxcf = &cpi->oxcf;
   size_params_type rsz = { oxcf->width, oxcf->height, SCALE_NUMERATOR };
   int resize_denom;
@@ -4138,7 +4239,8 @@ size_params_type av1_calculate_next_size_params(AV1_COMP *cpi) {
   return rsz;
 }
 
-static void setup_frame_size_from_params(AV1_COMP *cpi, size_params_type *rsz) {
+static void setup_frame_size_from_params(AV1_COMP *cpi,
+                                         const size_params_type *rsz) {
   int encode_width = rsz->resize_width;
   int encode_height = rsz->resize_height;
 
@@ -4148,11 +4250,13 @@ static void setup_frame_size_from_params(AV1_COMP *cpi, size_params_type *rsz) {
   cm->superres_scale_denominator = rsz->superres_denom;
   av1_calculate_scaled_superres_size(&encode_width, &encode_height,
                                      rsz->superres_denom);
-  set_frame_size(cpi, encode_width, encode_height);
+  av1_set_frame_size(cpi, encode_width, encode_height);
 }
 
-static void setup_frame_size(AV1_COMP *cpi) {
-  size_params_type rsz = av1_calculate_next_size_params(cpi);
+void av1_setup_frame_size(AV1_COMP *cpi) {
+  // Reset superres params from previous frame.
+  cpi->common.superres_scale_denominator = SCALE_NUMERATOR;
+  const size_params_type rsz = calculate_next_size_params(cpi);
   setup_frame_size_from_params(cpi, &rsz);
 }
 
@@ -4227,33 +4331,36 @@ static void loopfilter_frame(AV1_COMP *cpi, AV1_COMMON *cm) {
   }
 
   if (lf->filter_level[0] || lf->filter_level[1]) {
-#if LOOP_FILTER_BITMASK
-    av1_loop_filter_frame(cm->frame_to_show, cm, xd, 0, 0, num_planes, 0);
-#else
     if (cpi->num_workers > 1)
-      av1_loop_filter_frame_mt(cm->frame_to_show, cm, xd, 0, num_planes, 0,
+      av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, xd, 0, num_planes, 0,
+#if LOOP_FILTER_BITMASK
+                               0,
+#endif
                                cpi->workers, cpi->num_workers,
                                &cpi->lf_row_sync);
     else
-      av1_loop_filter_frame(cm->frame_to_show, cm, xd, 0, num_planes, 0);
+      av1_loop_filter_frame(&cm->cur_frame->buf, cm, xd,
+#if LOOP_FILTER_BITMASK
+                            0,
 #endif
+                            0, num_planes, 0);
   }
 
   if (!no_restoration)
-    av1_loop_restoration_save_boundary_lines(cm->frame_to_show, cm, 0);
+    av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 0);
 
   if (no_cdef) {
-    cm->cdef_bits = 0;
-    cm->cdef_strengths[0] = 0;
-    cm->nb_cdef_strengths = 1;
-    cm->cdef_uv_strengths[0] = 0;
+    cm->cdef_info.cdef_bits = 0;
+    cm->cdef_info.cdef_strengths[0] = 0;
+    cm->cdef_info.nb_cdef_strengths = 1;
+    cm->cdef_info.cdef_uv_strengths[0] = 0;
   } else {
     // Find CDEF parameters
-    av1_cdef_search(cm->frame_to_show, cpi->source, cm, xd,
+    av1_cdef_search(&cm->cur_frame->buf, cpi->source, cm, xd,
                     cpi->sf.fast_cdef_search);
 
     // Apply the filter
-    av1_cdef_frame(cm->frame_to_show, cm, xd);
+    av1_cdef_frame(&cm->cur_frame->buf, cm, xd);
   }
 
   superres_post_encode(cpi);
@@ -4263,170 +4370,365 @@ static void loopfilter_frame(AV1_COMP *cpi, AV1_COMMON *cm) {
     cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
     cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
   } else {
-    av1_loop_restoration_save_boundary_lines(cm->frame_to_show, cm, 1);
+    av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 1);
     av1_pick_filter_restoration(cpi->source, cpi);
     if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE ||
         cm->rst_info[1].frame_restoration_type != RESTORE_NONE ||
         cm->rst_info[2].frame_restoration_type != RESTORE_NONE) {
       if (cpi->num_workers > 1)
-        av1_loop_restoration_filter_frame_mt(cm->frame_to_show, cm, 0,
+        av1_loop_restoration_filter_frame_mt(&cm->cur_frame->buf, cm, 0,
                                              cpi->workers, cpi->num_workers,
                                              &cpi->lr_row_sync, &cpi->lr_ctxt);
       else
-        av1_loop_restoration_filter_frame(cm->frame_to_show, cm, 0,
+        av1_loop_restoration_filter_frame(&cm->cur_frame->buf, cm, 0,
                                           &cpi->lr_ctxt);
     }
   }
 }
 
-static int encode_without_recode_loop(AV1_COMP *cpi) {
-  AV1_COMMON *const cm = &cpi->common;
-  int q = 0, bottom_index = 0, top_index = 0;  // Dummy variables.
+static int get_refresh_frame_flags(const AV1_COMP *const cpi) {
+  const AV1_COMMON *const cm = &cpi->common;
 
-  aom_clear_system_state();
+  // Switch frames and shown key-frames overwrite all reference slots
+  if ((cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) ||
+      frame_is_sframe(cm))
+    return 0xFF;
 
-  set_size_independent_vars(cpi);
+  int refresh_mask = 0;
 
-  setup_frame_size(cpi);
+  // NOTE(zoeliu): When LAST_FRAME is to get refreshed, the decoder will be
+  // notified to get LAST3_FRAME refreshed and then the virtual indexes for all
+  // the 3 LAST reference frames will be updated accordingly, i.e.:
+  // (1) The original virtual index for LAST3_FRAME will become the new virtual
+  //     index for LAST_FRAME; and
+  // (2) The original virtual indexes for LAST_FRAME and LAST2_FRAME will be
+  //     shifted and become the new virtual indexes for LAST2_FRAME and
+  //     LAST3_FRAME.
+  refresh_mask |=
+      (cpi->refresh_last_frame << get_ref_frame_map_idx(cm, LAST3_FRAME));
 
-  assert(cm->width == cpi->scaled_source.y_crop_width);
-  assert(cm->height == cpi->scaled_source.y_crop_height);
+  const int bwd_ref_frame =
+      (cpi->new_bwdref_update_rule == 1) ? EXTREF_FRAME : BWDREF_FRAME;
+  refresh_mask |=
+      (cpi->refresh_bwd_ref_frame << get_ref_frame_map_idx(cm, bwd_ref_frame));
 
-  set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
+  refresh_mask |=
+      (cpi->refresh_alt2_ref_frame << get_ref_frame_map_idx(cm, ALTREF2_FRAME));
 
-  cpi->source =
-      av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source);
-  if (cpi->unscaled_last_source != NULL)
-    cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source,
-                                             &cpi->scaled_last_source);
-  cpi->source->buf_8bit_valid = 0;
-  if (frame_is_intra_only(cm) == 0) {
-    scale_references(cpi);
-  }
+  if (av1_preserve_existing_gf(cpi)) {
+    // We have decided to preserve the previously existing golden frame as our
+    // new ARF frame. However, in the short term we leave it in the GF slot and,
+    // if we're updating the GF with the current decoded frame, we save it
+    // instead to the ARF slot.
+    // Later, in the function av1_encoder.c:av1_update_reference_frames() we
+    // will swap gld_fb_idx and alt_fb_idx to achieve our objective. We do it
+    // there so that it can be done outside of the recode loop.
+    // Note: This is highly specific to the use of ARF as a forward reference,
+    // and this needs to be generalized as other uses are implemented
+    // (like RTC/temporal scalability).
 
-  av1_set_quantizer(cm, q);
-  setup_frame(cpi);
-  suppress_active_map(cpi);
-
-  // Variance adaptive and in frame q adjustment experiments are mutually
-  // exclusive.
-  if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
-    av1_vaq_frame_setup(cpi);
-  } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
-    av1_setup_in_frame_q_adj(cpi);
-  } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
-    av1_cyclic_refresh_setup(cpi);
-  }
-  apply_active_map(cpi);
-  if (cm->seg.enabled) {
-    if (!cm->seg.update_data && cm->prev_frame) {
-      segfeatures_copy(&cm->seg, &cm->prev_frame->seg);
-    } else {
-      calculate_segdata(&cm->seg);
+    if (!cpi->preserve_arf_as_gld) {
+      refresh_mask |= (cpi->refresh_golden_frame
+                       << get_ref_frame_map_idx(cm, ALTREF_FRAME));
     }
   } else {
-    memset(&cm->seg, 0, sizeof(cm->seg));
+    refresh_mask |=
+        (cpi->refresh_golden_frame << get_ref_frame_map_idx(cm, GOLDEN_FRAME));
+    refresh_mask |=
+        (cpi->refresh_alt_ref_frame << get_ref_frame_map_idx(cm, ALTREF_FRAME));
   }
-  segfeatures_copy(&cm->cur_frame->seg, &cm->seg);
+  return refresh_mask;
+}
 
-  // transform / motion compensation build reconstruction frame
-  av1_encode_frame(cpi);
+static void fix_interp_filter(InterpFilter *const interp_filter,
+                              const FRAME_COUNTS *const counts) {
+  if (*interp_filter == SWITCHABLE) {
+    // Check to see if only one of the filters is actually used
+    int count[SWITCHABLE_FILTERS] = { 0 };
+    int num_filters_used = 0;
+    for (int i = 0; i < SWITCHABLE_FILTERS; ++i) {
+      for (int j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
+        count[i] += counts->switchable_interp[j][i];
+      num_filters_used += (count[i] > 0);
+    }
+    if (num_filters_used == 1) {
+      // Only one filter is used. So set the filter at frame level
+      for (int i = 0; i < SWITCHABLE_FILTERS; ++i) {
+        if (count[i]) {
+          if (i == EIGHTTAP_REGULAR) *interp_filter = i;
+          break;
+        }
+      }
+    }
+  }
+}
 
-  // Update some stats from cyclic refresh, and check if we should not update
-  // golden reference, for 1 pass CBR.
-  if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->frame_type != KEY_FRAME &&
-      (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == AOM_CBR))
-    av1_cyclic_refresh_check_golden_update(cpi);
+static void finalize_encoded_frame(AV1_COMP *const cpi) {
+  AV1_COMMON *const cm = &cpi->common;
+  CurrentFrame *const current_frame = &cm->current_frame;
 
-  // Update the skip mb flag probabilities based on the distribution
-  // seen in the last encoder iteration.
-  // update_base_skip_probs(cpi);
-  aom_clear_system_state();
-  return AOM_CODEC_OK;
+  // This bitfield indicates which reference frame slots will be overwritten by
+  // the current frame
+  current_frame->refresh_frame_flags = get_refresh_frame_flags(cpi);
+
+  if (!cm->seq_params.reduced_still_picture_hdr &&
+      encode_show_existing_frame(cm)) {
+    RefCntBuffer *const frame_to_show =
+        cm->ref_frame_map[cpi->existing_fb_idx_to_show];
+
+    if (frame_to_show == NULL) {
+      aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
+                         "Buffer does not contain a reconstructed frame");
+    }
+    assert(frame_to_show->ref_count > 0);
+    assign_frame_buffer_p(&cm->cur_frame, frame_to_show);
+  }
+
+  if (!encode_show_existing_frame(cm) &&
+      cm->seq_params.film_grain_params_present &&
+      (cm->show_frame || cm->showable_frame)) {
+    // Copy the current frame's film grain params to the its corresponding
+    // RefCntBuffer slot.
+    cm->cur_frame->film_grain_params = cm->film_grain_params;
+
+    // We must update the parameters if this is not an INTER_FRAME
+    if (current_frame->frame_type != INTER_FRAME)
+      cm->cur_frame->film_grain_params.update_parameters = 1;
+
+    // Iterate the random seed for the next frame.
+    cm->film_grain_params.random_seed += 3381;
+    if (cm->film_grain_params.random_seed == 0)
+      cm->film_grain_params.random_seed = 7391;
+  }
+
+  // Initialise all tiles' contexts from the global frame context
+  for (int tile_col = 0; tile_col < cm->tile_cols; tile_col++) {
+    for (int tile_row = 0; tile_row < cm->tile_rows; tile_row++) {
+      const int tile_idx = tile_row * cm->tile_cols + tile_col;
+      cpi->tile_data[tile_idx].tctx = *cm->fc;
+    }
+  }
+
+  fix_interp_filter(&cm->interp_filter, cpi->td.counts);
+}
+
+// Called after encode_with_recode_loop() has just encoded a frame and packed
+// its bitstream.  This function works out whether we under- or over-shot
+// our bitrate target and adjusts q as appropriate.  Also decides whether
+// or not we should do another recode loop, indicated by *loop
+static void recode_loop_update_q(AV1_COMP *const cpi, int *const loop,
+                                 int *const q, int *const q_low,
+                                 int *const q_high, const int top_index,
+                                 const int bottom_index,
+                                 int *const undershoot_seen,
+                                 int *const overshoot_seen,
+                                 const int loop_at_this_size) {
+  AV1_COMMON *const cm = &cpi->common;
+  RATE_CONTROL *const rc = &cpi->rc;
+
+  int frame_over_shoot_limit = 0, frame_under_shoot_limit = 0;
+  av1_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
+                                   &frame_under_shoot_limit,
+                                   &frame_over_shoot_limit);
+  if (frame_over_shoot_limit == 0) frame_over_shoot_limit = 1;
+
+  if ((cm->current_frame.frame_type == KEY_FRAME) &&
+      rc->this_key_frame_forced &&
+      (rc->projected_frame_size < rc->max_frame_bandwidth)) {
+    int last_q = *q;
+    int64_t kf_err;
+
+    int64_t high_err_target = cpi->ambient_err;
+    int64_t low_err_target = cpi->ambient_err >> 1;
+
+    if (cm->seq_params.use_highbitdepth) {
+      kf_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf);
+    } else {
+      kf_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
+    }
+    // Prevent possible divide by zero error below for perfect KF
+    kf_err += !kf_err;
+
+    // The key frame is not good enough or we can afford
+    // to make it better without undue risk of popping.
+    if ((kf_err > high_err_target &&
+         rc->projected_frame_size <= frame_over_shoot_limit) ||
+        (kf_err > low_err_target &&
+         rc->projected_frame_size <= frame_under_shoot_limit)) {
+      // Lower q_high
+      *q_high = *q > *q_low ? *q - 1 : *q_low;
+
+      // Adjust Q
+      *q = (int)((*q * high_err_target) / kf_err);
+      *q = AOMMIN(*q, (*q_high + *q_low) >> 1);
+    } else if (kf_err < low_err_target &&
+               rc->projected_frame_size >= frame_under_shoot_limit) {
+      // The key frame is much better than the previous frame
+      // Raise q_low
+      *q_low = *q < *q_high ? *q + 1 : *q_high;
+
+      // Adjust Q
+      *q = (int)((*q * low_err_target) / kf_err);
+      *q = AOMMIN(*q, (*q_high + *q_low + 1) >> 1);
+    }
+
+    // Clamp Q to upper and lower limits:
+    *q = clamp(*q, *q_low, *q_high);
+
+    *loop = *q != last_q;
+  } else if (recode_loop_test(cpi, frame_over_shoot_limit,
+                              frame_under_shoot_limit, *q,
+                              AOMMAX(*q_high, top_index), bottom_index)) {
+    // Is the projected frame size out of range and are we allowed
+    // to attempt to recode.
+    int last_q = *q;
+    int retries = 0;
+
+    // Frame size out of permitted range:
+    // Update correction factor & compute new Q to try...
+    // Frame is too large
+    if (rc->projected_frame_size > rc->this_frame_target) {
+      // Special case if the projected size is > the max allowed.
+      if (rc->projected_frame_size >= rc->max_frame_bandwidth)
+        *q_high = rc->worst_quality;
+
+      // Raise Qlow as to at least the current value
+      *q_low = *q < *q_high ? *q + 1 : *q_high;
+
+      if (*undershoot_seen || loop_at_this_size > 1) {
+        // Update rate_correction_factor unless
+        av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+
+        *q = (*q_high + *q_low + 1) / 2;
+      } else {
+        // Update rate_correction_factor unless
+        av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+
+        *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
+                               AOMMAX(*q_high, top_index), cm->width,
+                               cm->height);
+
+        while (*q < *q_low && retries < 10) {
+          av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+          *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
+                                 AOMMAX(*q_high, top_index), cm->width,
+                                 cm->height);
+          retries++;
+        }
+      }
+
+      *overshoot_seen = 1;
+    } else {
+      // Frame is too small
+      *q_high = *q > *q_low ? *q - 1 : *q_low;
+
+      if (*overshoot_seen || loop_at_this_size > 1) {
+        av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+        *q = (*q_high + *q_low) / 2;
+      } else {
+        av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+        *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
+                               top_index, cm->width, cm->height);
+        // Special case reset for qlow for constrained quality.
+        // This should only trigger where there is very substantial
+        // undershoot on a frame and the auto cq level is above
+        // the user passsed in value.
+        if (cpi->oxcf.rc_mode == AOM_CQ && *q < *q_low) {
+          *q_low = *q;
+        }
+
+        while (*q > *q_high && retries < 10) {
+          av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
+          *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
+                                 top_index, cm->width, cm->height);
+          retries++;
+        }
+      }
+
+      *undershoot_seen = 1;
+    }
+
+    // Clamp Q to upper and lower limits:
+    *q = clamp(*q, *q_low, *q_high);
+
+    *loop = (*q != last_q);
+  } else {
+    *loop = 0;
+  }
 }
 
 static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
   AV1_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
-  int bottom_index, top_index;
-  int loop_count = 0;
-  int loop_at_this_size = 0;
-  int loop = 0;
-  int overshoot_seen = 0;
-  int undershoot_seen = 0;
-  int frame_over_shoot_limit;
-  int frame_under_shoot_limit;
-  int q = 0, q_low = 0, q_high = 0;
+  const int allow_recode = cpi->sf.recode_loop != DISALLOW_RECODE;
 
   set_size_independent_vars(cpi);
 
   cpi->source->buf_8bit_valid = 0;
 
-  aom_clear_system_state();
-  setup_frame_size(cpi);
-  set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
+  av1_setup_frame_size(cpi);
 
+  int top_index = 0, bottom_index = 0;
+  int q = 0, q_low = 0, q_high = 0;
+  set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
+  q_low = bottom_index;
+  q_high = top_index;
+
+  // Loop variables
+  int loop_count = 0;
+  int loop_at_this_size = 0;
+  int loop = 0;
+  int overshoot_seen = 0;
+  int undershoot_seen = 0;
   do {
     aom_clear_system_state();
 
-    if (loop_count == 0) {
-      // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
-      set_mv_search_params(cpi);
-
-      // Reset the loop state for new frame size.
-      overshoot_seen = 0;
-      undershoot_seen = 0;
-
-      q_low = bottom_index;
-      q_high = top_index;
-
-      loop_at_this_size = 0;
-
-      // Decide frame size bounds first time through.
-      av1_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
-                                       &frame_under_shoot_limit,
-                                       &frame_over_shoot_limit);
-    }
-
     // if frame was scaled calculate global_motion_search again if already
     // done
-    if (loop_count > 0 && cpi->source && cpi->global_motion_search_done)
+    if (loop_count > 0 && cpi->source && cpi->global_motion_search_done) {
       if (cpi->source->y_crop_width != cm->width ||
-          cpi->source->y_crop_height != cm->height)
+          cpi->source->y_crop_height != cm->height) {
         cpi->global_motion_search_done = 0;
+      }
+    }
     cpi->source =
         av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source);
-    if (cpi->unscaled_last_source != NULL)
+    if (cpi->unscaled_last_source != NULL) {
       cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source,
                                                &cpi->scaled_last_source);
+    }
 
-    if (frame_is_intra_only(cm) == 0) {
+    if (!frame_is_intra_only(cm)) {
       if (loop_count > 0) {
         release_scaled_references(cpi);
       }
       scale_references(cpi);
     }
     av1_set_quantizer(cm, q);
-    // printf("Frame %d/%d: q = %d, frame_type = %d\n", cm->current_video_frame,
-    //        cm->show_frame, q, cm->frame_type);
+    av1_init_quantizer(cpi);
+    // printf("Frame %d/%d: q = %d, frame_type = %d superres_denom = %d\n",
+    //        cm->current_frame.frame_number, cm->show_frame, q,
+    //        cm->current_frame.frame_type, cm->superres_scale_denominator);
 
-    if (loop_count == 0) setup_frame(cpi);
-
-    // Base q-index may have changed, so we need to assign proper default coef
-    // probs before every iteration.
-    if (cm->primary_ref_frame == PRIMARY_REF_NONE ||
-        cm->frame_refs[cm->primary_ref_frame].idx < 0) {
+    if (loop_count == 0) {
+      setup_frame(cpi);
+    } else if (get_primary_ref_frame_buf(cm) == NULL) {
+      // Base q-index may have changed, so we need to assign proper default coef
+      // probs before every iteration.
       av1_default_coef_probs(cm);
       av1_setup_frame_contexts(cm);
     }
 
-    // Variance adaptive and in frame q adjustment experiments are mutually
-    // exclusive.
     if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
       av1_vaq_frame_setup(cpi);
     } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
       av1_setup_in_frame_q_adj(cpi);
+    } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && !allow_recode) {
+      suppress_active_map(cpi);
+      av1_cyclic_refresh_setup(cpi);
+      apply_active_map(cpi);
     }
+
     if (cm->seg.enabled) {
       if (!cm->seg.update_data && cm->prev_frame) {
         segfeatures_copy(&cm->seg, &cm->prev_frame->seg);
@@ -4438,13 +4740,18 @@ static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
     }
     segfeatures_copy(&cm->cur_frame->seg, &cm->seg);
 
+    if (allow_recode) save_coding_context(cpi);
+
     // transform / motion compensation build reconstruction frame
-    save_coding_context(cpi);
     av1_encode_frame(cpi);
 
-    // Update the skip mb flag probabilities based on the distribution
-    // seen in the last encoder iteration.
-    // update_base_skip_probs(cpi);
+    // Update some stats from cyclic refresh, and check if we should not update
+    // golden reference, for 1 pass CBR.
+    if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
+        cm->current_frame.frame_type != KEY_FRAME &&
+        (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == AOM_CBR)) {
+      av1_cyclic_refresh_check_golden_update(cpi);
+    }
 
     aom_clear_system_state();
 
@@ -4454,140 +4761,20 @@ static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
     if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
       restore_coding_context(cpi);
 
-      if (av1_pack_bitstream(cpi, dest, size) != AOM_CODEC_OK)
+      finalize_encoded_frame(cpi);
+      int largest_tile_id = 0;  // Output from bitstream: unused here
+      if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK)
         return AOM_CODEC_ERROR;
 
       rc->projected_frame_size = (int)(*size) << 3;
       restore_coding_context(cpi);
-
-      if (frame_over_shoot_limit == 0) frame_over_shoot_limit = 1;
     }
 
-    if (cpi->oxcf.rc_mode == AOM_Q) {
-      loop = 0;
-    } else {
-      if ((cm->frame_type == KEY_FRAME) && rc->this_key_frame_forced &&
-          (rc->projected_frame_size < rc->max_frame_bandwidth)) {
-        int last_q = q;
-        int64_t kf_err;
-
-        int64_t high_err_target = cpi->ambient_err;
-        int64_t low_err_target = cpi->ambient_err >> 1;
-
-        if (cm->seq_params.use_highbitdepth) {
-          kf_err = aom_highbd_get_y_sse(cpi->source, get_frame_new_buffer(cm));
-        } else {
-          kf_err = aom_get_y_sse(cpi->source, get_frame_new_buffer(cm));
-        }
-        // Prevent possible divide by zero error below for perfect KF
-        kf_err += !kf_err;
-
-        // The key frame is not good enough or we can afford
-        // to make it better without undue risk of popping.
-        if ((kf_err > high_err_target &&
-             rc->projected_frame_size <= frame_over_shoot_limit) ||
-            (kf_err > low_err_target &&
-             rc->projected_frame_size <= frame_under_shoot_limit)) {
-          // Lower q_high
-          q_high = q > q_low ? q - 1 : q_low;
-
-          // Adjust Q
-          q = (int)((q * high_err_target) / kf_err);
-          q = AOMMIN(q, (q_high + q_low) >> 1);
-        } else if (kf_err < low_err_target &&
-                   rc->projected_frame_size >= frame_under_shoot_limit) {
-          // The key frame is much better than the previous frame
-          // Raise q_low
-          q_low = q < q_high ? q + 1 : q_high;
-
-          // Adjust Q
-          q = (int)((q * low_err_target) / kf_err);
-          q = AOMMIN(q, (q_high + q_low + 1) >> 1);
-        }
-
-        // Clamp Q to upper and lower limits:
-        q = clamp(q, q_low, q_high);
-
-        loop = q != last_q;
-      } else if (recode_loop_test(cpi, frame_over_shoot_limit,
-                                  frame_under_shoot_limit, q,
-                                  AOMMAX(q_high, top_index), bottom_index)) {
-        // Is the projected frame size out of range and are we allowed
-        // to attempt to recode.
-        int last_q = q;
-        int retries = 0;
-
-        // Frame size out of permitted range:
-        // Update correction factor & compute new Q to try...
-        // Frame is too large
-        if (rc->projected_frame_size > rc->this_frame_target) {
-          // Special case if the projected size is > the max allowed.
-          if (rc->projected_frame_size >= rc->max_frame_bandwidth)
-            q_high = rc->worst_quality;
-
-          // Raise Qlow as to at least the current value
-          q_low = q < q_high ? q + 1 : q_high;
-
-          if (undershoot_seen || loop_at_this_size > 1) {
-            // Update rate_correction_factor unless
-            av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-
-            q = (q_high + q_low + 1) / 2;
-          } else {
-            // Update rate_correction_factor unless
-            av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-
-            q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
-                                  AOMMAX(q_high, top_index), cm->width,
-                                  cm->height);
-
-            while (q < q_low && retries < 10) {
-              av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-              q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
-                                    AOMMAX(q_high, top_index), cm->width,
-                                    cm->height);
-              retries++;
-            }
-          }
-
-          overshoot_seen = 1;
-        } else {
-          // Frame is too small
-          q_high = q > q_low ? q - 1 : q_low;
-
-          if (overshoot_seen || loop_at_this_size > 1) {
-            av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-            q = (q_high + q_low) / 2;
-          } else {
-            av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-            q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
-                                  top_index, cm->width, cm->height);
-            // Special case reset for qlow for constrained quality.
-            // This should only trigger where there is very substantial
-            // undershoot on a frame and the auto cq level is above
-            // the user passsed in value.
-            if (cpi->oxcf.rc_mode == AOM_CQ && q < q_low) {
-              q_low = q;
-            }
-
-            while (q > q_high && retries < 10) {
-              av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
-              q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index,
-                                    top_index, cm->width, cm->height);
-              retries++;
-            }
-          }
-
-          undershoot_seen = 1;
-        }
-
-        // Clamp Q to upper and lower limits:
-        q = clamp(q, q_low, q_high);
-
-        loop = (q != last_q);
-      } else {
-        loop = 0;
-      }
+    if (allow_recode && cpi->oxcf.rc_mode != AOM_Q) {
+      // Update q and decide whether to do a recode loop
+      recode_loop_update_q(cpi, &loop, &q, &q_low, &q_high, top_index,
+                           bottom_index, &undershoot_seen, &overshoot_seen,
+                           loop_at_this_size);
     }
 
     // Special case for overlay frame.
@@ -4595,8 +4782,9 @@ static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
         rc->projected_frame_size < rc->max_frame_bandwidth)
       loop = 0;
 
-    if (!cpi->sf.gm_disable_recode) {
-      if (recode_loop_test_global_motion(cpi)) loop = 1;
+    if (allow_recode && !cpi->sf.gm_disable_recode &&
+        recode_loop_test_global_motion(cpi)) {
+      loop = 1;
     }
 
     if (loop) {
@@ -4612,128 +4800,17 @@ static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) {
   return AOM_CODEC_OK;
 }
 
-static int get_ref_frame_flags(const AV1_COMP *cpi) {
-  const int *const map = cpi->common.ref_frame_map;
-
-  // No.1 Priority: LAST_FRAME
-  const int last2_is_last = map[cpi->ref_fb_idx[1]] == map[cpi->ref_fb_idx[0]];
-  const int last3_is_last = map[cpi->ref_fb_idx[2]] == map[cpi->ref_fb_idx[0]];
-  const int gld_is_last =
-      map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]] == map[cpi->ref_fb_idx[0]];
-  const int bwd_is_last =
-      map[cpi->ref_fb_idx[BWDREF_FRAME - 1]] == map[cpi->ref_fb_idx[0]];
-  const int alt2_is_last =
-      map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] == map[cpi->ref_fb_idx[0]];
-  const int alt_is_last =
-      map[cpi->ref_fb_idx[ALTREF_FRAME - 1]] == map[cpi->ref_fb_idx[0]];
-
-  // No.2 Priority: ALTREF_FRAME
-  const int last2_is_alt =
-      map[cpi->ref_fb_idx[1]] == map[cpi->ref_fb_idx[ALTREF_FRAME - 1]];
-  const int last3_is_alt =
-      map[cpi->ref_fb_idx[2]] == map[cpi->ref_fb_idx[ALTREF_FRAME - 1]];
-  const int gld_is_alt = map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]] ==
-                         map[cpi->ref_fb_idx[ALTREF_FRAME - 1]];
-  const int bwd_is_alt = map[cpi->ref_fb_idx[BWDREF_FRAME - 1]] ==
-                         map[cpi->ref_fb_idx[ALTREF_FRAME - 1]];
-  const int alt2_is_alt = map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] ==
-                          map[cpi->ref_fb_idx[ALTREF_FRAME - 1]];
-
-  // No.3 Priority: LAST2_FRAME
-  const int last3_is_last2 = map[cpi->ref_fb_idx[2]] == map[cpi->ref_fb_idx[1]];
-  const int gld_is_last2 =
-      map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]] == map[cpi->ref_fb_idx[1]];
-  const int bwd_is_last2 =
-      map[cpi->ref_fb_idx[BWDREF_FRAME - 1]] == map[cpi->ref_fb_idx[1]];
-  const int alt2_is_last2 =
-      map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] == map[cpi->ref_fb_idx[1]];
-
-  // No.4 Priority: LAST3_FRAME
-  const int gld_is_last3 =
-      map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]] == map[cpi->ref_fb_idx[2]];
-  const int bwd_is_last3 =
-      map[cpi->ref_fb_idx[BWDREF_FRAME - 1]] == map[cpi->ref_fb_idx[2]];
-  const int alt2_is_last3 =
-      map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] == map[cpi->ref_fb_idx[2]];
-
-  // No.5 Priority: GOLDEN_FRAME
-  const int bwd_is_gld = map[cpi->ref_fb_idx[BWDREF_FRAME - 1]] ==
-                         map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]];
-  const int alt2_is_gld = map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] ==
-                          map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]];
-
-  // No.6 Priority: BWDREF_FRAME
-  const int alt2_is_bwd = map[cpi->ref_fb_idx[ALTREF2_FRAME - 1]] ==
-                          map[cpi->ref_fb_idx[BWDREF_FRAME - 1]];
-
-  // No.7 Priority: ALTREF2_FRAME
-
-  // After av1_apply_encoding_flags() is called, cpi->ref_frame_flags might be
-  // adjusted according to external encoder flags.
-  int flags = cpi->ext_ref_frame_flags;
-
-  if (cpi->rc.frames_till_gf_update_due == INT_MAX) flags &= ~AOM_GOLD_FLAG;
-
-  if (alt_is_last) flags &= ~AOM_ALT_FLAG;
-
-  if (last2_is_last || last2_is_alt) flags &= ~AOM_LAST2_FLAG;
-
-  if (last3_is_last || last3_is_alt || last3_is_last2) flags &= ~AOM_LAST3_FLAG;
-
-  if (gld_is_last || gld_is_alt || gld_is_last2 || gld_is_last3)
-    flags &= ~AOM_GOLD_FLAG;
-
-  if ((bwd_is_last || bwd_is_alt || bwd_is_last2 || bwd_is_last3 ||
-       bwd_is_gld) &&
-      (flags & AOM_BWD_FLAG))
-    flags &= ~AOM_BWD_FLAG;
-
-  if ((alt2_is_last || alt2_is_alt || alt2_is_last2 || alt2_is_last3 ||
-       alt2_is_gld || alt2_is_bwd) &&
-      (flags & AOM_ALT2_FLAG))
-    flags &= ~AOM_ALT2_FLAG;
-
-  return flags;
-}
-
-static void set_ext_overrides(AV1_COMP *cpi) {
-  // Overrides the defaults with the externally supplied values with
-  // av1_update_reference() and av1_update_entropy() calls
-  // Note: The overrides are valid only for the next frame passed
-  // to encode_frame_to_data_rate() function
-  if (cpi->ext_use_s_frame) cpi->common.frame_type = S_FRAME;
-  cpi->common.force_primary_ref_none = cpi->ext_use_primary_ref_none;
-
-  if (cpi->ext_refresh_frame_context_pending) {
-    cpi->common.refresh_frame_context = cpi->ext_refresh_frame_context;
-    cpi->ext_refresh_frame_context_pending = 0;
-  }
-  if (cpi->ext_refresh_frame_flags_pending) {
-    cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
-    cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
-    cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
-    cpi->refresh_bwd_ref_frame = cpi->ext_refresh_bwd_ref_frame;
-    cpi->refresh_alt2_ref_frame = cpi->ext_refresh_alt2_ref_frame;
-    cpi->ext_refresh_frame_flags_pending = 0;
-  }
-  cpi->common.allow_ref_frame_mvs = cpi->ext_use_ref_frame_mvs;
-  // A keyframe is already error resilient and keyframes with
-  // error_resilient_mode interferes with the use of show_existing_frame
-  // when forward reference keyframes are enabled.
-  cpi->common.error_resilient_mode =
-      cpi->ext_use_error_resilient && cpi->common.frame_type != KEY_FRAME;
-}
-
 #define DUMP_RECON_FRAMES 0
 
 #if DUMP_RECON_FRAMES == 1
 // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
 static void dump_filtered_recon_frames(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
-  const YV12_BUFFER_CONFIG *recon_buf = cm->frame_to_show;
+  const CurrentFrame *const current_frame = &cm->current_frame;
+  const YV12_BUFFER_CONFIG *recon_buf = &cm->cur_frame->buf;
 
   if (recon_buf == NULL) {
-    printf("Frame %d is not ready.\n", cm->current_video_frame);
+    printf("Frame %d is not ready.\n", current_frame->frame_number);
     return;
   }
 
@@ -4749,25 +4826,21 @@ static void dump_filtered_recon_frames(AV1_COMP *cpi) {
       "\n***Frame=%d (frame_offset=%d, show_frame=%d, "
       "show_existing_frame=%d) "
       "[LAST LAST2 LAST3 GOLDEN BWD ALT2 ALT]=[",
-      cm->current_video_frame, cm->frame_offset, cm->show_frame,
+      current_frame->frame_number, current_frame->order_hint, cm->show_frame,
       cm->show_existing_frame);
   for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    const int buf_idx = cm->frame_refs[ref_frame - LAST_FRAME].idx;
-    const int ref_offset =
-        (buf_idx >= 0)
-            ? (int)cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset
-            : -1;
-    printf(
-        " %d(%c-%d-%4.2f)", ref_offset,
-        (cpi->ref_frame_flags & flag_list[ref_frame]) ? 'Y' : 'N',
-        (buf_idx >= 0) ? (int)cpi->frame_rf_level[buf_idx] : -1,
-        (buf_idx >= 0) ? rate_factor_deltas[cpi->frame_rf_level[buf_idx]] : -1);
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    const int ref_offset = buf != NULL ? (int)buf->order_hint : -1;
+    printf(" %d(%c-%d-%4.2f)", ref_offset,
+           (cpi->ref_frame_flags & flag_list[ref_frame]) ? 'Y' : 'N',
+           buf ? (int)buf->frame_rf_level : -1,
+           buf ? rate_factor_deltas[buf->frame_rf_level] : -1);
   }
   printf(" ]\n");
 
   if (!cm->show_frame) {
     printf("Frame %d is a no show frame, so no image dump.\n",
-           cm->current_video_frame);
+           current_frame->frame_number);
     return;
   }
 
@@ -4775,7 +4848,7 @@ static void dump_filtered_recon_frames(AV1_COMP *cpi) {
   char file_name[256] = "/tmp/enc_filtered_recon.yuv";
   FILE *f_recon = NULL;
 
-  if (cm->current_video_frame == 0) {
+  if (current_frame->frame_number == 0) {
     if ((f_recon = fopen(file_name, "wb")) == NULL) {
       printf("Unable to open file %s to write.\n", file_name);
       return;
@@ -4791,9 +4864,9 @@ static void dump_filtered_recon_frames(AV1_COMP *cpi) {
       "show_frame=%d, show_existing_frame=%d, source_alt_ref_active=%d, "
       "refresh_alt_ref_frame=%d, rf_level=%d, "
       "y_stride=%4d, uv_stride=%4d, cm->width=%4d, cm->height=%4d\n\n",
-      cm->current_video_frame, cpi->twopass.gf_group.index,
+      current_frame->frame_number, cpi->twopass.gf_group.index,
       cpi->twopass.gf_group.update_type[cpi->twopass.gf_group.index],
-      cm->frame_offset, cm->show_frame, cm->show_existing_frame,
+      current_frame->order_hint, cm->show_frame, cm->show_existing_frame,
       cpi->rc.source_alt_ref_active, cpi->refresh_alt_ref_frame,
       cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index],
       recon_buf->y_stride, recon_buf->uv_stride, cm->width, cm->height);
@@ -4801,13 +4874,7 @@ static void dump_filtered_recon_frames(AV1_COMP *cpi) {
   int ref_frame;
   printf("get_ref_frame_map_idx: [");
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame)
-    printf(" %d", get_ref_frame_map_idx(cpi, ref_frame));
-  printf(" ]\n");
-  printf("cm->new_fb_idx = %d\n", cm->new_fb_idx);
-  printf("cm->ref_frame_map = [");
-  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    printf(" %d", cm->ref_frame_map[ref_frame - LAST_FRAME]);
-  }
+    printf(" %d", get_ref_frame_map_idx(cm, ref_frame));
   printf(" ]\n");
 #endif  // 0
 
@@ -4831,811 +4898,36 @@ static void dump_filtered_recon_frames(AV1_COMP *cpi) {
 }
 #endif  // DUMP_RECON_FRAMES
 
-static INLINE int is_frame_droppable(AV1_COMP *cpi) {
-  return !(cpi->refresh_alt_ref_frame || cpi->refresh_alt2_ref_frame ||
-           cpi->refresh_bwd_ref_frame || cpi->refresh_golden_frame ||
-           cpi->refresh_last_frame);
+static int setup_interp_filter_search_mask(AV1_COMP *cpi) {
+  InterpFilters ifilter;
+  int ref_total[REF_FRAMES] = { 0 };
+  MV_REFERENCE_FRAME ref;
+  int mask = 0;
+  int arf_idx = ALTREF_FRAME;
+  if (cpi->common.last_frame_type == KEY_FRAME || cpi->refresh_alt_ref_frame)
+    return mask;
+  for (ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref)
+    for (ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; ++ifilter)
+      ref_total[ref] += cpi->interp_filter_selected[ref][ifilter];
+
+  for (ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; ++ifilter) {
+    if ((ref_total[LAST_FRAME] &&
+         cpi->interp_filter_selected[LAST_FRAME][ifilter] * 30 <=
+             ref_total[LAST_FRAME]) &&
+        (((cpi->interp_filter_selected[LAST2_FRAME][ifilter] * 20) +
+          (cpi->interp_filter_selected[LAST3_FRAME][ifilter] * 20) +
+          (cpi->interp_filter_selected[GOLDEN_FRAME][ifilter] * 20) +
+          (cpi->interp_filter_selected[BWDREF_FRAME][ifilter] * 10) +
+          (cpi->interp_filter_selected[ALTREF2_FRAME][ifilter] * 10) +
+          (cpi->interp_filter_selected[arf_idx][ifilter] * 10)) <
+         (ref_total[LAST2_FRAME] + ref_total[LAST3_FRAME] +
+          ref_total[GOLDEN_FRAME] + ref_total[BWDREF_FRAME] +
+          ref_total[ALTREF2_FRAME] + ref_total[ALTREF_FRAME])))
+      mask |= 1 << ifilter;
+  }
+  return mask;
 }
 
-static int encode_frame_to_data_rate(AV1_COMP *cpi, size_t *size, uint8_t *dest,
-                                     int skip_adapt,
-                                     unsigned int *frame_flags) {
-  AV1_COMMON *const cm = &cpi->common;
-  SequenceHeader *const seq_params = &cm->seq_params;
-  const AV1EncoderConfig *const oxcf = &cpi->oxcf;
-  struct segmentation *const seg = &cm->seg;
-
-  set_ext_overrides(cpi);
-  aom_clear_system_state();
-
-  // frame type has been decided outside of this function call
-  cm->cur_frame->intra_only = frame_is_intra_only(cm);
-  cm->cur_frame->frame_type = cm->frame_type;
-
-  // S_FRAMEs are always error resilient
-  cm->error_resilient_mode |= frame_is_sframe(cm);
-
-  cm->large_scale_tile = cpi->oxcf.large_scale_tile;
-  cm->single_tile_decoding = cpi->oxcf.single_tile_decoding;
-  if (cm->large_scale_tile) seq_params->frame_id_numbers_present_flag = 0;
-
-  cm->allow_ref_frame_mvs &= frame_might_allow_ref_frame_mvs(cm);
-  // cm->allow_ref_frame_mvs needs to be written into the frame header while
-  // cm->large_scale_tile is 1, therefore, "cm->large_scale_tile=1" case is
-  // separated from frame_might_allow_ref_frame_mvs().
-  cm->allow_ref_frame_mvs &= !cm->large_scale_tile;
-
-  cm->allow_warped_motion =
-      cpi->oxcf.allow_warped_motion && frame_might_allow_warped_motion(cm);
-
-  // Reset the frame packet stamp index.
-  if (cm->frame_type == KEY_FRAME && cm->show_frame)
-    cm->current_video_frame = 0;
-
-  // NOTE:
-  // (1) Move the setup of the ref_frame_flags upfront as it would be
-  //     determined by the current frame properties;
-  // (2) The setup of the ref_frame_flags applies to both
-  // show_existing_frame's
-  //     and the other cases.
-  if (cm->current_video_frame > 0)
-    cpi->ref_frame_flags = get_ref_frame_flags(cpi);
-
-  if (encode_show_existing_frame(cm)) {
-    // NOTE(zoeliu): In BIDIR_PRED, the existing frame to show is the current
-    //               BWDREF_FRAME in the reference frame buffer.
-    if (cm->frame_type == KEY_FRAME) {
-      cm->reset_decoder_state = 1;
-    } else {
-      cm->frame_type = INTER_FRAME;
-    }
-    cm->show_frame = 1;
-    cpi->frame_flags = *frame_flags;
-
-    restore_coding_context(cpi);
-
-    // Build the bitstream
-    if (av1_pack_bitstream(cpi, dest, size) != AOM_CODEC_OK)
-      return AOM_CODEC_ERROR;
-
-    cpi->seq_params_locked = 1;
-
-    // Set up frame to show to get ready for stats collection.
-    cm->frame_to_show = get_frame_new_buffer(cm);
-
-    // Update current frame offset.
-    cm->frame_offset =
-        cm->buffer_pool->frame_bufs[cm->new_fb_idx].cur_frame_offset;
-
-#if DUMP_RECON_FRAMES == 1
-    // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
-    dump_filtered_recon_frames(cpi);
-#endif  // DUMP_RECON_FRAMES
-
-    // Update the LAST_FRAME in the reference frame buffer.
-    // NOTE:
-    // (1) For BWDREF_FRAME as the show_existing_frame, the reference frame
-    //     update has been done previously when handling the LAST_BIPRED_FRAME
-    //     right before BWDREF_FRAME (in the display order);
-    // (2) For INTNL_OVERLAY as the show_existing_frame, the reference frame
-    //     update will be done when the following is called, which will
-    //     exchange
-    //     the virtual indexes between LAST_FRAME and ALTREF2_FRAME, so that
-    //     LAST3 will get retired, LAST2 becomes LAST3, LAST becomes LAST2,
-    //     and
-    //     ALTREF2_FRAME will serve as the new LAST_FRAME.
-    update_reference_frames(cpi);
-
-    // Update frame flags
-    cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
-    cpi->frame_flags &= ~FRAMEFLAGS_BWDREF;
-    cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
-
-    *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
-
-    // Update the frame type
-    cm->last_frame_type = cm->frame_type;
-
-    // Since we allocate a spot for the OVERLAY frame in the gf group, we need
-    // to do post-encoding update accordingly.
-    if (cpi->rc.is_src_frame_alt_ref) {
-      av1_set_target_rate(cpi, cm->width, cm->height);
-      av1_rc_postencode_update(cpi, *size);
-    }
-
-    ++cm->current_video_frame;
-
-    return AOM_CODEC_OK;
-  }
-
-  // Set default state for segment based loop filter update flags.
-  cm->lf.mode_ref_delta_update = 0;
-
-  // Set various flags etc to special state if it is a key frame.
-  if (frame_is_intra_only(cm) || frame_is_sframe(cm)) {
-    // Reset the loop filter deltas and segmentation map.
-    av1_reset_segment_features(cm);
-
-    // If segmentation is enabled force a map update for key frames.
-    if (seg->enabled) {
-      seg->update_map = 1;
-      seg->update_data = 1;
-    }
-
-    // The alternate reference frame cannot be active for a key frame.
-    cpi->rc.source_alt_ref_active = 0;
-  }
-  if (cpi->oxcf.mtu == 0) {
-    cm->num_tg = cpi->oxcf.num_tile_groups;
-  } else {
-    // Use a default value for the purposes of weighting costs in probability
-    // updates
-    cm->num_tg = DEFAULT_MAX_NUM_TG;
-  }
-
-  // For 1 pass CBR, check if we are dropping this frame.
-  // Never drop on key frame.
-  if (oxcf->pass == 0 && oxcf->rc_mode == AOM_CBR &&
-      cm->frame_type != KEY_FRAME) {
-    if (av1_rc_drop_frame(cpi)) {
-      av1_rc_postencode_update_drop_frame(cpi);
-      return AOM_CODEC_OK;
-    }
-  }
-
-  aom_clear_system_state();
-
-#if CONFIG_INTERNAL_STATS
-  memset(cpi->mode_chosen_counts, 0,
-         MAX_MODES * sizeof(*cpi->mode_chosen_counts));
-#endif
-
-  if (seq_params->frame_id_numbers_present_flag) {
-    /* Non-normative definition of current_frame_id ("frame counter" with
-     * wraparound) */
-    const int frame_id_length = FRAME_ID_LENGTH;
-    if (cm->current_frame_id == -1) {
-      int lsb, msb;
-      /* quasi-random initialization of current_frame_id for a key frame */
-      if (cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) {
-        lsb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[0] & 0xff;
-        msb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[1] & 0xff;
-      } else {
-        lsb = cpi->source->y_buffer[0] & 0xff;
-        msb = cpi->source->y_buffer[1] & 0xff;
-      }
-      cm->current_frame_id = ((msb << 8) + lsb) % (1 << frame_id_length);
-
-      // S_frame is meant for stitching different streams of different
-      // resolutions together, so current_frame_id must be the
-      // same across different streams of the same content current_frame_id
-      // should be the same and not random. 0x37 is a chosen number as start
-      // point
-      if (cpi->oxcf.sframe_enabled) cm->current_frame_id = 0x37;
-    } else {
-      cm->current_frame_id =
-          (cm->current_frame_id + 1 + (1 << frame_id_length)) %
-          (1 << frame_id_length);
-    }
-  }
-
-  switch (cpi->oxcf.cdf_update_mode) {
-    case 0:  // No CDF update for any frames(4~6% compression loss).
-      cm->disable_cdf_update = 1;
-      break;
-    case 1:  // Enable CDF update for all frames.
-      cm->disable_cdf_update = 0;
-      break;
-    case 2:
-      // Strategically determine at which frames to do CDF update.
-      // Currently only enable CDF update for all-intra and no-show frames(1.5%
-      // compression loss).
-      // TODO(huisu@google.com): design schemes for various trade-offs between
-      // compression quality and decoding speed.
-      cm->disable_cdf_update =
-          (frame_is_intra_only(cm) || !cm->show_frame) ? 0 : 1;
-      break;
-  }
-  cm->timing_info_present &= !seq_params->reduced_still_picture_hdr;
-
-  if (cpi->sf.recode_loop == DISALLOW_RECODE) {
-    if (encode_without_recode_loop(cpi) != AOM_CODEC_OK) return AOM_CODEC_ERROR;
-  } else {
-    if (encode_with_recode_loop(cpi, size, dest) != AOM_CODEC_OK)
-      return AOM_CODEC_ERROR;
-  }
-
-  cm->last_tile_cols = cm->tile_cols;
-  cm->last_tile_rows = cm->tile_rows;
-
-#ifdef OUTPUT_YUV_SKINMAP
-  if (cpi->common.current_video_frame > 1) {
-    av1_compute_skin_map(cpi, yuv_skinmap_file);
-  }
-#endif  // OUTPUT_YUV_SKINMAP
-
-  // Special case code to reduce pulsing when key frames are forced at a
-  // fixed interval. Note the reconstruction error if it is the frame before
-  // the force key frame
-  if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
-    if (seq_params->use_highbitdepth) {
-      cpi->ambient_err =
-          aom_highbd_get_y_sse(cpi->source, get_frame_new_buffer(cm));
-    } else {
-      cpi->ambient_err = aom_get_y_sse(cpi->source, get_frame_new_buffer(cm));
-    }
-  }
-
-  // If the encoder forced a KEY_FRAME decision or if frame is an S_FRAME
-  if ((cm->frame_type == KEY_FRAME && cm->show_frame) || frame_is_sframe(cm)) {
-    cpi->refresh_last_frame = 1;
-  }
-
-  cm->frame_to_show = get_frame_new_buffer(cm);
-  cm->frame_to_show->color_primaries = seq_params->color_primaries;
-  cm->frame_to_show->transfer_characteristics =
-      seq_params->transfer_characteristics;
-  cm->frame_to_show->matrix_coefficients = seq_params->matrix_coefficients;
-  cm->frame_to_show->monochrome = seq_params->monochrome;
-  cm->frame_to_show->chroma_sample_position =
-      seq_params->chroma_sample_position;
-  cm->frame_to_show->color_range = seq_params->color_range;
-  cm->frame_to_show->render_width = cm->render_width;
-  cm->frame_to_show->render_height = cm->render_height;
-
-  // TODO(zoeliu): For non-ref frames, loop filtering may need to be turned
-  // off.
-
-  // Pick the loop filter level for the frame.
-  if (!cm->allow_intrabc) {
-    loopfilter_frame(cpi, cm);
-  } else {
-    cm->lf.filter_level[0] = 0;
-    cm->lf.filter_level[1] = 0;
-    cm->cdef_bits = 0;
-    cm->cdef_strengths[0] = 0;
-    cm->nb_cdef_strengths = 1;
-    cm->cdef_uv_strengths[0] = 0;
-    cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
-    cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
-    cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
-  }
-
-  // TODO(debargha): Fix mv search range on encoder side
-  // aom_extend_frame_inner_borders(cm->frame_to_show, av1_num_planes(cm));
-  aom_extend_frame_borders(cm->frame_to_show, av1_num_planes(cm));
-
-#ifdef OUTPUT_YUV_REC
-  aom_write_one_yuv_frame(cm, cm->frame_to_show);
-#endif
-
-  // Build the bitstream
-  if (av1_pack_bitstream(cpi, dest, size) != AOM_CODEC_OK)
-    return AOM_CODEC_ERROR;
-
-  cpi->seq_params_locked = 1;
-
-  if (skip_adapt) return AOM_CODEC_OK;
-
-  if (seq_params->frame_id_numbers_present_flag) {
-    int i;
-    // Update reference frame id values based on the value of refresh_frame_mask
-    for (i = 0; i < REF_FRAMES; i++) {
-      if ((cpi->refresh_frame_mask >> i) & 1) {
-        cm->ref_frame_id[i] = cm->current_frame_id;
-      }
-    }
-  }
-
-#if DUMP_RECON_FRAMES == 1
-  // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
-  dump_filtered_recon_frames(cpi);
-#endif  // DUMP_RECON_FRAMES
-
-  if (cm->seg.enabled) {
-    if (cm->seg.update_map) {
-      update_reference_segmentation_map(cpi);
-    } else if (cm->last_frame_seg_map) {
-      memcpy(cm->current_frame_seg_map, cm->last_frame_seg_map,
-             cm->mi_cols * cm->mi_rows * sizeof(uint8_t));
-    }
-  }
-
-  if (frame_is_intra_only(cm) == 0) {
-    release_scaled_references(cpi);
-  }
-
-  update_reference_frames(cpi);
-
-#if CONFIG_ENTROPY_STATS
-  av1_accumulate_frame_counts(&aggregate_fc, &cpi->counts);
-#endif  // CONFIG_ENTROPY_STATS
-
-  if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
-    *cm->fc = cpi->tile_data[cm->largest_tile_id].tctx;
-    av1_reset_cdf_symbol_counters(cm->fc);
-  }
-
-  if (cpi->refresh_golden_frame == 1)
-    cpi->frame_flags |= FRAMEFLAGS_GOLDEN;
-  else
-    cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
-
-  if (cpi->refresh_alt_ref_frame == 1)
-    cpi->frame_flags |= FRAMEFLAGS_ALTREF;
-  else
-    cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
-
-  if (cpi->refresh_bwd_ref_frame == 1)
-    cpi->frame_flags |= FRAMEFLAGS_BWDREF;
-  else
-    cpi->frame_flags &= ~FRAMEFLAGS_BWDREF;
-
-  cm->last_frame_type = cm->frame_type;
-
-  av1_rc_postencode_update(cpi, *size);
-
-  if (cm->frame_type == KEY_FRAME) {
-    // Tell the caller that the frame was coded as a key frame
-    *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
-  } else {
-    *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
-  }
-
-  // Clear the one shot update flags for segmentation map and mode/ref loop
-  // filter deltas.
-  cm->seg.update_map = 0;
-  cm->seg.update_data = 0;
-  cm->lf.mode_ref_delta_update = 0;
-
-  // A droppable frame might not be shown but it always
-  // takes a space in the gf group. Therefore, even when
-  // it is not shown, we still need update the count down.
-
-  if (cm->show_frame) {
-    // TODO(zoeliu): We may only swamp mi and prev_mi for those frames that
-    // are
-    // being used as reference.
-    swap_mi_and_prev_mi(cm);
-    // Don't increment frame counters if this was an altref buffer
-    // update not a real frame
-
-    ++cm->current_video_frame;
-  }
-
-  // NOTE: Shall not refer to any frame not used as reference.
-  if (cm->is_reference_frame) {
-    // keep track of the last coded dimensions
-    cm->last_width = cm->width;
-    cm->last_height = cm->height;
-
-    // reset to normal state now that we are done.
-    cm->last_show_frame = cm->show_frame;
-  }
-
-  return AOM_CODEC_OK;
-}
-
-static INLINE void update_keyframe_counters(AV1_COMP *cpi) {
-  // TODO(zoeliu): To investigate whether we should treat BWDREF_FRAME
-  //               differently here for rc->avg_frame_bandwidth.
-  if (cpi->common.show_frame || cpi->rc.is_bwd_ref_frame) {
-    if (!cpi->common.show_existing_frame || cpi->rc.is_src_frame_alt_ref ||
-        cpi->common.frame_type == KEY_FRAME) {
-      // If this is a show_existing_frame with a source other than altref,
-      // or if it is not a displayed forward keyframe, the keyframe update
-      // counters were incremented when it was originally encoded.
-      cpi->rc.frames_since_key++;
-      cpi->rc.frames_to_key--;
-    }
-  }
-}
-
-static INLINE void update_frames_till_gf_update(AV1_COMP *cpi) {
-  // TODO(weitinglin): Updating this counter for is_frame_droppable
-  // is a work-around to handle the condition when a frame is drop.
-  // We should fix the cpi->common.show_frame flag
-  // instead of checking the other condition to update the counter properly.
-  if (cpi->common.show_frame || is_frame_droppable(cpi)) {
-    // Decrement count down till next gf
-    if (cpi->rc.frames_till_gf_update_due > 0)
-      cpi->rc.frames_till_gf_update_due--;
-  }
-}
-
-static INLINE void update_twopass_gf_group_index(AV1_COMP *cpi) {
-  // Increment the gf group index ready for the next frame. If this is
-  // a show_existing_frame with a source other than altref, or if it is not
-  // a displayed forward keyframe, the index was incremented when it was
-  // originally encoded.
-  if (!cpi->common.show_existing_frame || cpi->rc.is_src_frame_alt_ref ||
-      cpi->common.frame_type == KEY_FRAME) {
-    ++cpi->twopass.gf_group.index;
-  }
-}
-
-static void update_rc_counts(AV1_COMP *cpi) {
-  update_keyframe_counters(cpi);
-  update_frames_till_gf_update(cpi);
-  if (cpi->oxcf.pass == 2) update_twopass_gf_group_index(cpi);
-}
-
-static int Pass0Encode(AV1_COMP *cpi, size_t *size, uint8_t *dest,
-                       int skip_adapt, unsigned int *frame_flags) {
-  if (cpi->oxcf.rc_mode == AOM_CBR) {
-    av1_rc_get_one_pass_cbr_params(cpi);
-  } else {
-    av1_rc_get_one_pass_vbr_params(cpi);
-  }
-  if (encode_frame_to_data_rate(cpi, size, dest, skip_adapt, frame_flags) !=
-      AOM_CODEC_OK) {
-    return AOM_CODEC_ERROR;
-  }
-  update_rc_counts(cpi);
-  check_show_existing_frame(cpi);
-  return AOM_CODEC_OK;
-}
-
-static int Pass2Encode(AV1_COMP *cpi, size_t *size, uint8_t *dest,
-                       unsigned int *frame_flags) {
-#if CONFIG_MISMATCH_DEBUG
-  mismatch_move_frame_idx_w();
-#endif
-#if TXCOEFF_COST_TIMER
-  AV1_COMMON *cm = &cpi->common;
-  cm->txcoeff_cost_timer = 0;
-  cm->txcoeff_cost_count = 0;
-#endif
-
-  if (encode_frame_to_data_rate(cpi, size, dest, 0, frame_flags) !=
-      AOM_CODEC_OK) {
-    return AOM_CODEC_ERROR;
-  }
-
-#if TXCOEFF_COST_TIMER
-  cm->cum_txcoeff_cost_timer += cm->txcoeff_cost_timer;
-  fprintf(stderr,
-          "\ntxb coeff cost block number: %ld, frame time: %ld, cum time %ld "
-          "in us\n",
-          cm->txcoeff_cost_count, cm->txcoeff_cost_timer,
-          cm->cum_txcoeff_cost_timer);
-#endif
-
-  av1_twopass_postencode_update(cpi);
-  update_rc_counts(cpi);
-  check_show_existing_frame(cpi);
-  return AOM_CODEC_OK;
-}
-
-#if CONFIG_DENOISE
-static int apply_denoise_2d(AV1_COMP *cpi, YV12_BUFFER_CONFIG *sd,
-                            int block_size, float noise_level,
-                            int64_t time_stamp, int64_t end_time) {
-  AV1_COMMON *const cm = &cpi->common;
-  if (!cpi->denoise_and_model) {
-    cpi->denoise_and_model = aom_denoise_and_model_alloc(
-        cm->seq_params.bit_depth, block_size, noise_level);
-    if (!cpi->denoise_and_model) {
-      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
-                         "Error allocating denoise and model");
-      return -1;
-    }
-  }
-  if (!cpi->film_grain_table) {
-    cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table));
-    if (!cpi->film_grain_table) {
-      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
-                         "Error allocating grain table");
-      return -1;
-    }
-    memset(cpi->film_grain_table, 0, sizeof(*cpi->film_grain_table));
-  }
-  if (aom_denoise_and_model_run(cpi->denoise_and_model, sd,
-                                &cm->film_grain_params)) {
-    if (cm->film_grain_params.apply_grain) {
-      aom_film_grain_table_append(cpi->film_grain_table, time_stamp, end_time,
-                                  &cm->film_grain_params);
-    }
-  }
-  return 0;
-}
-#endif
-
-int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
-                          YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
-                          int64_t end_time) {
-  AV1_COMMON *const cm = &cpi->common;
-  const SequenceHeader *const seq_params = &cm->seq_params;
-  struct aom_usec_timer timer;
-  int res = 0;
-  const int subsampling_x = sd->subsampling_x;
-  const int subsampling_y = sd->subsampling_y;
-  const int use_highbitdepth = (sd->flags & YV12_FLAG_HIGHBITDEPTH) != 0;
-
-  check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
-
-  aom_usec_timer_start(&timer);
-
-#if CONFIG_DENOISE
-  if (cpi->oxcf.noise_level > 0)
-    if (apply_denoise_2d(cpi, sd, cpi->oxcf.noise_block_size,
-                         cpi->oxcf.noise_level, time_stamp, end_time) < 0)
-      res = -1;
-#endif  //  CONFIG_DENOISE
-
-  if (av1_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
-                         use_highbitdepth, frame_flags))
-    res = -1;
-  aom_usec_timer_mark(&timer);
-  cpi->time_receive_data += aom_usec_timer_elapsed(&timer);
-
-  if ((seq_params->profile == PROFILE_0) && !seq_params->monochrome &&
-      (subsampling_x != 1 || subsampling_y != 1)) {
-    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
-                       "Non-4:2:0 color format requires profile 1 or 2");
-    res = -1;
-  }
-  if ((seq_params->profile == PROFILE_1) &&
-      !(subsampling_x == 0 && subsampling_y == 0)) {
-    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
-                       "Profile 1 requires 4:4:4 color format");
-    res = -1;
-  }
-  if ((seq_params->profile == PROFILE_2) &&
-      (seq_params->bit_depth <= AOM_BITS_10) &&
-      !(subsampling_x == 1 && subsampling_y == 0)) {
-    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
-                       "Profile 2 bit-depth < 10 requires 4:2:2 color format");
-    res = -1;
-  }
-
-  return res;
-}
-
-static int frame_is_reference(const AV1_COMP *cpi) {
-  const AV1_COMMON *cm = &cpi->common;
-
-  return cm->frame_type == KEY_FRAME || cpi->refresh_last_frame ||
-         cpi->refresh_golden_frame || cpi->refresh_bwd_ref_frame ||
-         cpi->refresh_alt2_ref_frame || cpi->refresh_alt_ref_frame ||
-         !cm->error_resilient_mode || cm->lf.mode_ref_delta_update ||
-         cm->seg.update_map || cm->seg.update_data;
-}
-
-static void adjust_frame_rate(AV1_COMP *cpi,
-                              const struct lookahead_entry *source) {
-  int64_t this_duration;
-  int step = 0;
-
-  if (source->ts_start == cpi->first_time_stamp_ever) {
-    this_duration = source->ts_end - source->ts_start;
-    step = 1;
-  } else {
-    int64_t last_duration =
-        cpi->last_end_time_stamp_seen - cpi->last_time_stamp_seen;
-
-    this_duration = source->ts_end - cpi->last_end_time_stamp_seen;
-
-    // do a step update if the duration changes by 10%
-    if (last_duration)
-      step = (int)((this_duration - last_duration) * 10 / last_duration);
-  }
-
-  if (this_duration) {
-    if (step) {
-      av1_new_framerate(cpi, 10000000.0 / this_duration);
-    } else {
-      // Average this frame's rate into the last second's average
-      // frame rate. If we haven't seen 1 second yet, then average
-      // over the whole interval seen.
-      const double interval = AOMMIN(
-          (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
-      double avg_duration = 10000000.0 / cpi->framerate;
-      avg_duration *= (interval - avg_duration + this_duration);
-      avg_duration /= interval;
-
-      av1_new_framerate(cpi, 10000000.0 / avg_duration);
-    }
-  }
-  cpi->last_time_stamp_seen = source->ts_start;
-  cpi->last_end_time_stamp_seen = source->ts_end;
-}
-
-// Returns 0 if this is not an alt ref else the offset of the source frame
-// used as the arf midpoint.
-static int get_arf_src_index(AV1_COMP *cpi) {
-  RATE_CONTROL *const rc = &cpi->rc;
-  int arf_src_index = 0;
-  if (is_altref_enabled(cpi)) {
-    if (cpi->oxcf.pass == 2) {
-      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-      if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
-        arf_src_index = gf_group->arf_src_offset[gf_group->index];
-      }
-    } else if (rc->source_alt_ref_pending) {
-      arf_src_index = rc->frames_till_gf_update_due;
-    }
-  }
-  return arf_src_index;
-}
-
-static int get_brf_src_index(AV1_COMP *cpi) {
-  int brf_src_index = 0;
-  const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-
-  // TODO(zoeliu): We need to add the check on the -bwd_ref command line setup
-  //               flag.
-  if (gf_group->bidir_pred_enabled[gf_group->index]) {
-    if (cpi->oxcf.pass == 2) {
-      if (gf_group->update_type[gf_group->index] == BRF_UPDATE)
-        brf_src_index = gf_group->brf_src_offset[gf_group->index];
-    } else {
-      // TODO(zoeliu): To re-visit the setup for this scenario
-      brf_src_index = cpi->rc.bipred_group_interval - 1;
-    }
-  }
-
-  return brf_src_index;
-}
-
-// Returns 0 if this is not an alt ref else the offset of the source frame
-// used as the arf midpoint.
-static int get_arf2_src_index(AV1_COMP *cpi) {
-  int arf2_src_index = 0;
-  if (is_altref_enabled(cpi) && cpi->num_extra_arfs) {
-    if (cpi->oxcf.pass == 2) {
-      const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-      if (gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE) {
-        arf2_src_index = gf_group->arf_src_offset[gf_group->index];
-      }
-    }
-  }
-  return arf2_src_index;
-}
-
-static void check_src_altref(AV1_COMP *cpi,
-                             const struct lookahead_entry *source) {
-  RATE_CONTROL *const rc = &cpi->rc;
-
-  // If pass == 2, the parameters set here will be reset in
-  // av1_rc_get_second_pass_params()
-
-  if (cpi->oxcf.pass == 2) {
-    const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-    rc->is_src_frame_alt_ref =
-        (gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE) ||
-        (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
-    rc->is_src_frame_ext_arf =
-        gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE;
-  } else {
-    rc->is_src_frame_alt_ref =
-        cpi->alt_ref_source && (source == cpi->alt_ref_source);
-  }
-
-  if (rc->is_src_frame_alt_ref) {
-    // Current frame is an ARF overlay frame.
-    cpi->alt_ref_source = NULL;
-
-    if (rc->is_src_frame_ext_arf && !cpi->common.show_existing_frame) {
-      // For INTNL_OVERLAY, when show_existing_frame == 0, they do need to
-      // refresh the LAST_FRAME, i.e. LAST3 gets retired, LAST2 becomes LAST3,
-      // LAST becomes LAST2, and INTNL_OVERLAY becomes LAST.
-      cpi->refresh_last_frame = 1;
-    } else {
-      // Don't refresh the last buffer for an ARF overlay frame. It will
-      // become the GF so preserve last as an alternative prediction option.
-      cpi->refresh_last_frame = 0;
-    }
-  }
-}
-
-#if CONFIG_INTERNAL_STATS
-extern double av1_get_blockiness(const unsigned char *img1, int img1_pitch,
-                                 const unsigned char *img2, int img2_pitch,
-                                 int width, int height);
-
-static void adjust_image_stat(double y, double u, double v, double all,
-                              ImageStat *s) {
-  s->stat[STAT_Y] += y;
-  s->stat[STAT_U] += u;
-  s->stat[STAT_V] += v;
-  s->stat[STAT_ALL] += all;
-  s->worst = AOMMIN(s->worst, all);
-}
-
-static void compute_internal_stats(AV1_COMP *cpi, int frame_bytes) {
-  AV1_COMMON *const cm = &cpi->common;
-  double samples = 0.0;
-  uint32_t in_bit_depth = 8;
-  uint32_t bit_depth = 8;
-
-#if CONFIG_INTER_STATS_ONLY
-  if (cm->frame_type == KEY_FRAME) return;  // skip key frame
-#endif
-  cpi->bytes += frame_bytes;
-
-  if (cm->seq_params.use_highbitdepth) {
-    in_bit_depth = cpi->oxcf.input_bit_depth;
-    bit_depth = cm->seq_params.bit_depth;
-  }
-  if (cm->show_frame) {
-    const YV12_BUFFER_CONFIG *orig = cpi->source;
-    const YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
-    double y, u, v, frame_all;
-
-    cpi->count++;
-    if (cpi->b_calculate_psnr) {
-      PSNR_STATS psnr;
-      double frame_ssim2 = 0.0, weight = 0.0;
-      aom_clear_system_state();
-      // TODO(yaowu): unify these two versions into one.
-      aom_calc_highbd_psnr(orig, recon, &psnr, bit_depth, in_bit_depth);
-
-      adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3], psnr.psnr[0],
-                        &cpi->psnr);
-      cpi->total_sq_error += psnr.sse[0];
-      cpi->total_samples += psnr.samples[0];
-      samples = psnr.samples[0];
-      // TODO(yaowu): unify these two versions into one.
-      if (cm->seq_params.use_highbitdepth)
-        frame_ssim2 =
-            aom_highbd_calc_ssim(orig, recon, &weight, bit_depth, in_bit_depth);
-      else
-        frame_ssim2 = aom_calc_ssim(orig, recon, &weight);
-
-      cpi->worst_ssim = AOMMIN(cpi->worst_ssim, frame_ssim2);
-      cpi->summed_quality += frame_ssim2 * weight;
-      cpi->summed_weights += weight;
-
-#if 0
-      {
-        FILE *f = fopen("q_used.stt", "a");
-        double y2 = psnr.psnr[1];
-        double u2 = psnr.psnr[2];
-        double v2 = psnr.psnr[3];
-        double frame_psnr2 = psnr.psnr[0];
-        fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
-                cm->current_video_frame, y2, u2, v2,
-                frame_psnr2, frame_ssim2);
-        fclose(f);
-      }
-#endif
-    }
-    if (cpi->b_calculate_blockiness) {
-      if (!cm->seq_params.use_highbitdepth) {
-        const double frame_blockiness =
-            av1_get_blockiness(orig->y_buffer, orig->y_stride, recon->y_buffer,
-                               recon->y_stride, orig->y_width, orig->y_height);
-        cpi->worst_blockiness = AOMMAX(cpi->worst_blockiness, frame_blockiness);
-        cpi->total_blockiness += frame_blockiness;
-      }
-
-      if (cpi->b_calculate_consistency) {
-        if (!cm->seq_params.use_highbitdepth) {
-          const double this_inconsistency = aom_get_ssim_metrics(
-              orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride,
-              orig->y_width, orig->y_height, cpi->ssim_vars, &cpi->metrics, 1);
-
-          const double peak = (double)((1 << in_bit_depth) - 1);
-          const double consistency =
-              aom_sse_to_psnr(samples, peak, cpi->total_inconsistency);
-          if (consistency > 0.0)
-            cpi->worst_consistency =
-                AOMMIN(cpi->worst_consistency, consistency);
-          cpi->total_inconsistency += this_inconsistency;
-        }
-      }
-    }
-
-    frame_all =
-        aom_calc_fastssim(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
-    adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
-    frame_all = aom_psnrhvs(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
-    adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
-  }
-}
-#endif  // CONFIG_INTERNAL_STATS
-
 static int is_integer_mv(AV1_COMP *cpi, const YV12_BUFFER_CONFIG *cur_picture,
                          const YV12_BUFFER_CONFIG *last_picture,
                          hash_table *last_hash_table) {
@@ -5769,31 +5061,661 @@ static int is_integer_mv(AV1_COMP *cpi, const YV12_BUFFER_CONFIG *cur_picture,
   return 0;
 }
 
+static int encode_frame_to_data_rate(AV1_COMP *cpi, size_t *size, uint8_t *dest,
+                                     unsigned int *frame_flags) {
+  AV1_COMMON *const cm = &cpi->common;
+  SequenceHeader *const seq_params = &cm->seq_params;
+  CurrentFrame *const current_frame = &cm->current_frame;
+  const AV1EncoderConfig *const oxcf = &cpi->oxcf;
+  struct segmentation *const seg = &cm->seg;
+
+  // frame type has been decided outside of this function call
+  cm->cur_frame->frame_type = current_frame->frame_type;
+
+  cm->large_scale_tile = cpi->oxcf.large_scale_tile;
+  cm->single_tile_decoding = cpi->oxcf.single_tile_decoding;
+
+  cm->allow_ref_frame_mvs &= frame_might_allow_ref_frame_mvs(cm);
+  // cm->allow_ref_frame_mvs needs to be written into the frame header while
+  // cm->large_scale_tile is 1, therefore, "cm->large_scale_tile=1" case is
+  // separated from frame_might_allow_ref_frame_mvs().
+  cm->allow_ref_frame_mvs &= !cm->large_scale_tile;
+
+  cm->allow_warped_motion =
+      cpi->oxcf.allow_warped_motion && frame_might_allow_warped_motion(cm);
+
+  cm->last_frame_type = current_frame->frame_type;
+  if (cpi->oxcf.pass == 2 && cpi->sf.adaptive_interp_filter_search)
+    cpi->sf.interp_filter_search_mask = setup_interp_filter_search_mask(cpi);
+
+  if (encode_show_existing_frame(cm)) {
+    // NOTE(zoeliu): In BIDIR_PRED, the existing frame to show is the current
+    //               BWDREF_FRAME in the reference frame buffer.
+    cpi->frame_flags = *frame_flags;
+
+    restore_coding_context(cpi);
+
+    finalize_encoded_frame(cpi);
+    // Build the bitstream
+    int largest_tile_id = 0;  // Output from bitstream: unused here
+    if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK)
+      return AOM_CODEC_ERROR;
+
+    if (seq_params->frame_id_numbers_present_flag &&
+        current_frame->frame_type == KEY_FRAME) {
+      // Displaying a forward key-frame, so reset the ref buffer IDs
+      int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show];
+      for (int i = 0; i < REF_FRAMES; i++)
+        cm->ref_frame_id[i] = display_frame_id;
+    }
+
+    cpi->seq_params_locked = 1;
+
+#if DUMP_RECON_FRAMES == 1
+    // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
+    dump_filtered_recon_frames(cpi);
+#endif  // DUMP_RECON_FRAMES
+
+    // Update the LAST_FRAME in the reference frame buffer.
+    // NOTE:
+    // (1) For BWDREF_FRAME as the show_existing_frame, the reference frame
+    //     update has been done previously when handling the LAST_BIPRED_FRAME
+    //     right before BWDREF_FRAME (in the display order);
+    // (2) For INTNL_OVERLAY as the show_existing_frame, the reference frame
+    //     update will be done when the following is called, which will
+    //     exchange
+    //     the virtual indexes between LAST_FRAME and ALTREF2_FRAME, so that
+    //     LAST3 will get retired, LAST2 becomes LAST3, LAST becomes LAST2,
+    //     and
+    //     ALTREF2_FRAME will serve as the new LAST_FRAME.
+    update_reference_frames(cpi);
+
+    // Update frame flags
+    cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
+    cpi->frame_flags &= ~FRAMEFLAGS_BWDREF;
+    cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
+
+    *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
+
+    // Since we allocate a spot for the OVERLAY frame in the gf group, we need
+    // to do post-encoding update accordingly.
+    if (cpi->rc.is_src_frame_alt_ref) {
+      av1_set_target_rate(cpi, cm->width, cm->height);
+      av1_rc_postencode_update(cpi, *size);
+    }
+
+    ++current_frame->frame_number;
+
+    return AOM_CODEC_OK;
+  }
+
+  // Work out whether to force_integer_mv this frame
+  if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools &&
+      !frame_is_intra_only(cm)) {
+    if (cpi->common.seq_params.force_integer_mv == 2) {
+      // Adaptive mode: see what previous frame encoded did
+      if (cpi->unscaled_last_source != NULL) {
+        cm->cur_frame_force_integer_mv =
+            is_integer_mv(cpi, cpi->source, cpi->unscaled_last_source,
+                          cpi->previous_hash_table);
+      } else {
+        cpi->common.cur_frame_force_integer_mv = 0;
+      }
+    } else {
+      cpi->common.cur_frame_force_integer_mv =
+          cpi->common.seq_params.force_integer_mv;
+    }
+  } else {
+    cpi->common.cur_frame_force_integer_mv = 0;
+  }
+
+  // Set default state for segment based loop filter update flags.
+  cm->lf.mode_ref_delta_update = 0;
+
+  // Set various flags etc to special state if it is a key frame.
+  if (frame_is_intra_only(cm) || frame_is_sframe(cm)) {
+    // Reset the loop filter deltas and segmentation map.
+    av1_reset_segment_features(cm);
+
+    // If segmentation is enabled force a map update for key frames.
+    if (seg->enabled) {
+      seg->update_map = 1;
+      seg->update_data = 1;
+    }
+
+    // The alternate reference frame cannot be active for a key frame.
+    cpi->rc.source_alt_ref_active = 0;
+  }
+  if (cpi->oxcf.mtu == 0) {
+    cm->num_tg = cpi->oxcf.num_tile_groups;
+  } else {
+    // Use a default value for the purposes of weighting costs in probability
+    // updates
+    cm->num_tg = DEFAULT_MAX_NUM_TG;
+  }
+
+  // For 1 pass CBR, check if we are dropping this frame.
+  // Never drop on key frame.
+  if (oxcf->pass == 0 && oxcf->rc_mode == AOM_CBR &&
+      current_frame->frame_type != KEY_FRAME) {
+    if (av1_rc_drop_frame(cpi)) {
+      av1_rc_postencode_update_drop_frame(cpi);
+      release_scaled_references(cpi);
+      return AOM_CODEC_OK;
+    }
+  }
+
+  aom_clear_system_state();
+
+#if CONFIG_INTERNAL_STATS
+  memset(cpi->mode_chosen_counts, 0,
+         MAX_MODES * sizeof(*cpi->mode_chosen_counts));
+#endif
+
+  if (seq_params->frame_id_numbers_present_flag) {
+    /* Non-normative definition of current_frame_id ("frame counter" with
+     * wraparound) */
+    if (cm->current_frame_id == -1) {
+      int lsb, msb;
+      /* quasi-random initialization of current_frame_id for a key frame */
+      if (cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) {
+        lsb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[0] & 0xff;
+        msb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[1] & 0xff;
+      } else {
+        lsb = cpi->source->y_buffer[0] & 0xff;
+        msb = cpi->source->y_buffer[1] & 0xff;
+      }
+      cm->current_frame_id =
+          ((msb << 8) + lsb) % (1 << seq_params->frame_id_length);
+
+      // S_frame is meant for stitching different streams of different
+      // resolutions together, so current_frame_id must be the
+      // same across different streams of the same content current_frame_id
+      // should be the same and not random. 0x37 is a chosen number as start
+      // point
+      if (cpi->oxcf.sframe_enabled) cm->current_frame_id = 0x37;
+    } else {
+      cm->current_frame_id =
+          (cm->current_frame_id + 1 + (1 << seq_params->frame_id_length)) %
+          (1 << seq_params->frame_id_length);
+    }
+  }
+
+  switch (cpi->oxcf.cdf_update_mode) {
+    case 0:  // No CDF update for any frames(4~6% compression loss).
+      cm->disable_cdf_update = 1;
+      break;
+    case 1:  // Enable CDF update for all frames.
+      cm->disable_cdf_update = 0;
+      break;
+    case 2:
+      // Strategically determine at which frames to do CDF update.
+      // Currently only enable CDF update for all-intra and no-show frames(1.5%
+      // compression loss).
+      // TODO(huisu@google.com): design schemes for various trade-offs between
+      // compression quality and decoding speed.
+      cm->disable_cdf_update =
+          (frame_is_intra_only(cm) || !cm->show_frame) ? 0 : 1;
+      break;
+  }
+  cm->timing_info_present &= !seq_params->reduced_still_picture_hdr;
+
+  if (encode_with_recode_loop(cpi, size, dest) != AOM_CODEC_OK)
+    return AOM_CODEC_ERROR;
+
+#ifdef OUTPUT_YUV_SKINMAP
+  if (cpi->common.current_frame.frame_number > 1) {
+    av1_compute_skin_map(cpi, yuv_skinmap_file);
+  }
+#endif  // OUTPUT_YUV_SKINMAP
+
+  // Special case code to reduce pulsing when key frames are forced at a
+  // fixed interval. Note the reconstruction error if it is the frame before
+  // the force key frame
+  if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
+    if (seq_params->use_highbitdepth) {
+      cpi->ambient_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf);
+    } else {
+      cpi->ambient_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf);
+    }
+  }
+
+  // If the encoder forced a KEY_FRAME decision or if frame is an S_FRAME
+  if ((current_frame->frame_type == KEY_FRAME && cm->show_frame) ||
+      frame_is_sframe(cm)) {
+    cpi->refresh_last_frame = 1;
+  }
+
+  cm->cur_frame->buf.color_primaries = seq_params->color_primaries;
+  cm->cur_frame->buf.transfer_characteristics =
+      seq_params->transfer_characteristics;
+  cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients;
+  cm->cur_frame->buf.monochrome = seq_params->monochrome;
+  cm->cur_frame->buf.chroma_sample_position =
+      seq_params->chroma_sample_position;
+  cm->cur_frame->buf.color_range = seq_params->color_range;
+  cm->cur_frame->buf.render_width = cm->render_width;
+  cm->cur_frame->buf.render_height = cm->render_height;
+
+  // TODO(zoeliu): For non-ref frames, loop filtering may need to be turned
+  // off.
+
+  // Pick the loop filter level for the frame.
+  if (!cm->allow_intrabc) {
+    loopfilter_frame(cpi, cm);
+  } else {
+    cm->lf.filter_level[0] = 0;
+    cm->lf.filter_level[1] = 0;
+    cm->cdef_info.cdef_bits = 0;
+    cm->cdef_info.cdef_strengths[0] = 0;
+    cm->cdef_info.nb_cdef_strengths = 1;
+    cm->cdef_info.cdef_uv_strengths[0] = 0;
+    cm->rst_info[0].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[1].frame_restoration_type = RESTORE_NONE;
+    cm->rst_info[2].frame_restoration_type = RESTORE_NONE;
+  }
+
+  // TODO(debargha): Fix mv search range on encoder side
+  // aom_extend_frame_inner_borders(&cm->cur_frame->buf, av1_num_planes(cm));
+  aom_extend_frame_borders(&cm->cur_frame->buf, av1_num_planes(cm));
+
+#ifdef OUTPUT_YUV_REC
+  aom_write_one_yuv_frame(cm, &cm->cur_frame->buf);
+#endif
+
+  finalize_encoded_frame(cpi);
+  // Build the bitstream
+  int largest_tile_id = 0;  // Output from pack_bitstream
+  if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK)
+    return AOM_CODEC_ERROR;
+
+  cpi->seq_params_locked = 1;
+
+  // Update reference frame ids for reference frames this frame will overwrite
+  if (seq_params->frame_id_numbers_present_flag) {
+    for (int i = 0; i < REF_FRAMES; i++) {
+      if ((current_frame->refresh_frame_flags >> i) & 1) {
+        cm->ref_frame_id[i] = cm->current_frame_id;
+      }
+    }
+  }
+
+#if DUMP_RECON_FRAMES == 1
+  // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
+  dump_filtered_recon_frames(cpi);
+#endif  // DUMP_RECON_FRAMES
+
+  if (cm->seg.enabled) {
+    if (cm->seg.update_map) {
+      update_reference_segmentation_map(cpi);
+    } else if (cm->last_frame_seg_map) {
+      memcpy(cm->cur_frame->seg_map, cm->last_frame_seg_map,
+             cm->mi_cols * cm->mi_rows * sizeof(uint8_t));
+    }
+  }
+
+  if (frame_is_intra_only(cm) == 0) {
+    release_scaled_references(cpi);
+  }
+
+  update_reference_frames(cpi);
+
+#if CONFIG_ENTROPY_STATS
+  av1_accumulate_frame_counts(&aggregate_fc, &cpi->counts);
+#endif  // CONFIG_ENTROPY_STATS
+
+  if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
+    *cm->fc = cpi->tile_data[largest_tile_id].tctx;
+    av1_reset_cdf_symbol_counters(cm->fc);
+  }
+  if (!cm->large_scale_tile) {
+    cm->cur_frame->frame_context = *cm->fc;
+  }
+#define EXT_TILE_DEBUG 0
+#if EXT_TILE_DEBUG
+  if (cm->large_scale_tile && oxcf->pass == 2) {
+    char fn[20] = "./fc";
+    fn[4] = current_frame->frame_number / 100 + '0';
+    fn[5] = (current_frame->frame_number % 100) / 10 + '0';
+    fn[6] = (current_frame->frame_number % 10) + '0';
+    fn[7] = '\0';
+    av1_print_frame_contexts(cm->fc, fn);
+  }
+#endif  // EXT_TILE_DEBUG
+#undef EXT_TILE_DEBUG
+
+  if (cpi->refresh_golden_frame == 1)
+    cpi->frame_flags |= FRAMEFLAGS_GOLDEN;
+  else
+    cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
+
+  if (cpi->refresh_alt_ref_frame == 1)
+    cpi->frame_flags |= FRAMEFLAGS_ALTREF;
+  else
+    cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
+
+  if (cpi->refresh_bwd_ref_frame == 1)
+    cpi->frame_flags |= FRAMEFLAGS_BWDREF;
+  else
+    cpi->frame_flags &= ~FRAMEFLAGS_BWDREF;
+  cm->last_frame_type = current_frame->frame_type;
+
+  av1_rc_postencode_update(cpi, *size);
+
+  if (current_frame->frame_type == KEY_FRAME) {
+    // Tell the caller that the frame was coded as a key frame
+    *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
+  } else {
+    *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
+  }
+
+  // Store encoded frame's hash table for is_integer_mv() next time
+  if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools) {
+    cpi->previous_hash_table = &cm->cur_frame->hash_table;
+  }
+
+  // Clear the one shot update flags for segmentation map and mode/ref loop
+  // filter deltas.
+  cm->seg.update_map = 0;
+  cm->seg.update_data = 0;
+  cm->lf.mode_ref_delta_update = 0;
+
+  // A droppable frame might not be shown but it always
+  // takes a space in the gf group. Therefore, even when
+  // it is not shown, we still need update the count down.
+
+  if (cm->show_frame) {
+    // TODO(zoeliu): We may only swamp mi and prev_mi for those frames that
+    // are
+    // being used as reference.
+    swap_mi_and_prev_mi(cm);
+    // Don't increment frame counters if this was an altref buffer
+    // update not a real frame
+
+    ++current_frame->frame_number;
+  }
+
+  return AOM_CODEC_OK;
+}
+
+int av1_encode(AV1_COMP *const cpi, uint8_t *const dest,
+               const EncodeFrameInput *const frame_input,
+               const EncodeFrameParams *const frame_params,
+               EncodeFrameResults *const frame_results) {
+  AV1_COMMON *const cm = &cpi->common;
+  CurrentFrame *const current_frame = &cm->current_frame;
+
+  cpi->unscaled_source = frame_input->source;
+  cpi->source = frame_input->source;
+  cpi->unscaled_last_source = frame_input->last_source;
+
+  cm->error_resilient_mode = frame_params->error_resilient_mode;
+  cm->primary_ref_frame = frame_params->primary_ref_frame;
+  cm->current_frame.frame_type = frame_params->frame_type;
+  cm->show_frame = frame_params->show_frame;
+  cpi->ref_frame_flags = frame_params->ref_frame_flags;
+  cpi->speed = frame_params->speed;
+
+  if (current_frame->frame_type == KEY_FRAME && cm->show_frame)
+    current_frame->frame_number = 0;
+
+  if (cm->show_existing_frame) {
+    current_frame->order_hint = cm->cur_frame->order_hint;
+  } else {
+    current_frame->order_hint =
+        current_frame->frame_number + frame_params->order_offset;
+    current_frame->order_hint %=
+        (1 << (cm->seq_params.order_hint_info.order_hint_bits_minus_1 + 1));
+  }
+
+  if (cpi->oxcf.pass == 1) {
+    av1_first_pass(cpi, frame_input->ts_duration);
+  } else if (cpi->oxcf.pass == 0 || cpi->oxcf.pass == 2) {
+    if (encode_frame_to_data_rate(cpi, &frame_results->size, dest,
+                                  frame_params->frame_flags) != AOM_CODEC_OK) {
+      return AOM_CODEC_ERROR;
+    }
+  } else {
+    return AOM_CODEC_ERROR;
+  }
+
+  return AOM_CODEC_OK;
+}
+
+#if CONFIG_DENOISE
+static int apply_denoise_2d(AV1_COMP *cpi, YV12_BUFFER_CONFIG *sd,
+                            int block_size, float noise_level,
+                            int64_t time_stamp, int64_t end_time) {
+  AV1_COMMON *const cm = &cpi->common;
+  if (!cpi->denoise_and_model) {
+    cpi->denoise_and_model = aom_denoise_and_model_alloc(
+        cm->seq_params.bit_depth, block_size, noise_level);
+    if (!cpi->denoise_and_model) {
+      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                         "Error allocating denoise and model");
+      return -1;
+    }
+  }
+  if (!cpi->film_grain_table) {
+    cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table));
+    if (!cpi->film_grain_table) {
+      aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                         "Error allocating grain table");
+      return -1;
+    }
+    memset(cpi->film_grain_table, 0, sizeof(*cpi->film_grain_table));
+  }
+  if (aom_denoise_and_model_run(cpi->denoise_and_model, sd,
+                                &cm->film_grain_params)) {
+    if (cm->film_grain_params.apply_grain) {
+      aom_film_grain_table_append(cpi->film_grain_table, time_stamp, end_time,
+                                  &cm->film_grain_params);
+    }
+  }
+  return 0;
+}
+#endif
+
+int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
+                          YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
+                          int64_t end_time) {
+  AV1_COMMON *const cm = &cpi->common;
+  const SequenceHeader *const seq_params = &cm->seq_params;
+  struct aom_usec_timer timer;
+  int res = 0;
+  const int subsampling_x = sd->subsampling_x;
+  const int subsampling_y = sd->subsampling_y;
+  const int use_highbitdepth = (sd->flags & YV12_FLAG_HIGHBITDEPTH) != 0;
+
+  check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
+
+  aom_usec_timer_start(&timer);
+
+#if CONFIG_DENOISE
+  if (cpi->oxcf.noise_level > 0)
+    if (apply_denoise_2d(cpi, sd, cpi->oxcf.noise_block_size,
+                         cpi->oxcf.noise_level, time_stamp, end_time) < 0)
+      res = -1;
+#endif  //  CONFIG_DENOISE
+
+  if (av1_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
+                         use_highbitdepth, frame_flags))
+    res = -1;
+  aom_usec_timer_mark(&timer);
+  cpi->time_receive_data += aom_usec_timer_elapsed(&timer);
+
+  if ((seq_params->profile == PROFILE_0) && !seq_params->monochrome &&
+      (subsampling_x != 1 || subsampling_y != 1)) {
+    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
+                       "Non-4:2:0 color format requires profile 1 or 2");
+    res = -1;
+  }
+  if ((seq_params->profile == PROFILE_1) &&
+      !(subsampling_x == 0 && subsampling_y == 0)) {
+    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
+                       "Profile 1 requires 4:4:4 color format");
+    res = -1;
+  }
+  if ((seq_params->profile == PROFILE_2) &&
+      (seq_params->bit_depth <= AOM_BITS_10) &&
+      !(subsampling_x == 1 && subsampling_y == 0)) {
+    aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM,
+                       "Profile 2 bit-depth < 10 requires 4:2:2 color format");
+    res = -1;
+  }
+
+  return res;
+}
+
+#if CONFIG_INTERNAL_STATS
+extern double av1_get_blockiness(const unsigned char *img1, int img1_pitch,
+                                 const unsigned char *img2, int img2_pitch,
+                                 int width, int height);
+
+static void adjust_image_stat(double y, double u, double v, double all,
+                              ImageStat *s) {
+  s->stat[STAT_Y] += y;
+  s->stat[STAT_U] += u;
+  s->stat[STAT_V] += v;
+  s->stat[STAT_ALL] += all;
+  s->worst = AOMMIN(s->worst, all);
+}
+
+static void compute_internal_stats(AV1_COMP *cpi, int frame_bytes) {
+  AV1_COMMON *const cm = &cpi->common;
+  double samples = 0.0;
+  uint32_t in_bit_depth = 8;
+  uint32_t bit_depth = 8;
+
+#if CONFIG_INTER_STATS_ONLY
+  if (cm->current_frame.frame_type == KEY_FRAME) return;  // skip key frame
+#endif
+  cpi->bytes += frame_bytes;
+
+  if (cm->seq_params.use_highbitdepth) {
+    in_bit_depth = cpi->oxcf.input_bit_depth;
+    bit_depth = cm->seq_params.bit_depth;
+  }
+  if (cm->show_frame) {
+    const YV12_BUFFER_CONFIG *orig = cpi->source;
+    const YV12_BUFFER_CONFIG *recon = &cpi->common.cur_frame->buf;
+    double y, u, v, frame_all;
+
+    cpi->count++;
+    if (cpi->b_calculate_psnr) {
+      PSNR_STATS psnr;
+      double frame_ssim2 = 0.0, weight = 0.0;
+      aom_clear_system_state();
+      // TODO(yaowu): unify these two versions into one.
+      aom_calc_highbd_psnr(orig, recon, &psnr, bit_depth, in_bit_depth);
+
+      adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3], psnr.psnr[0],
+                        &cpi->psnr);
+      cpi->total_sq_error += psnr.sse[0];
+      cpi->total_samples += psnr.samples[0];
+      samples = psnr.samples[0];
+      // TODO(yaowu): unify these two versions into one.
+      if (cm->seq_params.use_highbitdepth)
+        frame_ssim2 =
+            aom_highbd_calc_ssim(orig, recon, &weight, bit_depth, in_bit_depth);
+      else
+        frame_ssim2 = aom_calc_ssim(orig, recon, &weight);
+
+      cpi->worst_ssim = AOMMIN(cpi->worst_ssim, frame_ssim2);
+      cpi->summed_quality += frame_ssim2 * weight;
+      cpi->summed_weights += weight;
+
+#if 0
+      {
+        FILE *f = fopen("q_used.stt", "a");
+        double y2 = psnr.psnr[1];
+        double u2 = psnr.psnr[2];
+        double v2 = psnr.psnr[3];
+        double frame_psnr2 = psnr.psnr[0];
+        fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
+                cm->current_frame.frame_number, y2, u2, v2,
+                frame_psnr2, frame_ssim2);
+        fclose(f);
+      }
+#endif
+    }
+    if (cpi->b_calculate_blockiness) {
+      if (!cm->seq_params.use_highbitdepth) {
+        const double frame_blockiness =
+            av1_get_blockiness(orig->y_buffer, orig->y_stride, recon->y_buffer,
+                               recon->y_stride, orig->y_width, orig->y_height);
+        cpi->worst_blockiness = AOMMAX(cpi->worst_blockiness, frame_blockiness);
+        cpi->total_blockiness += frame_blockiness;
+      }
+
+      if (cpi->b_calculate_consistency) {
+        if (!cm->seq_params.use_highbitdepth) {
+          const double this_inconsistency = aom_get_ssim_metrics(
+              orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride,
+              orig->y_width, orig->y_height, cpi->ssim_vars, &cpi->metrics, 1);
+
+          const double peak = (double)((1 << in_bit_depth) - 1);
+          const double consistency =
+              aom_sse_to_psnr(samples, peak, cpi->total_inconsistency);
+          if (consistency > 0.0)
+            cpi->worst_consistency =
+                AOMMIN(cpi->worst_consistency, consistency);
+          cpi->total_inconsistency += this_inconsistency;
+        }
+      }
+    }
+
+    frame_all =
+        aom_calc_fastssim(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
+    adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
+    frame_all = aom_psnrhvs(orig, recon, &y, &u, &v, bit_depth, in_bit_depth);
+    adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
+  }
+}
+#endif  // CONFIG_INTERNAL_STATS
+
+// Don't allow a show_existing_frame to coincide with an error resilient or
+// S-Frame. An exception can be made in the case of a keyframe, since it does
+// not depend on any previous frames.
+static int allow_show_existing(const AV1_COMP *const cpi) {
+  if (cpi->common.current_frame.frame_number == 0) return 0;
+
+  const struct lookahead_entry *lookahead_src =
+      av1_lookahead_peek(cpi->lookahead, 0);
+  if (lookahead_src == NULL) return 1;
+
+  const int is_error_resilient =
+      cpi->oxcf.error_resilient_mode ||
+      (lookahead_src->flags & AOM_EFLAG_ERROR_RESILIENT);
+  const int is_s_frame =
+      cpi->oxcf.s_frame_mode || (lookahead_src->flags & AOM_EFLAG_SET_S_FRAME);
+  const int is_key_frame =
+      (cpi->rc.frames_to_key == 0) || (cpi->frame_flags & FRAMEFLAGS_KEY);
+  return !(is_error_resilient || is_s_frame) || is_key_frame;
+}
+
 int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags,
                             size_t *size, uint8_t *dest, int64_t *time_stamp,
                             int64_t *time_end, int flush,
                             const aom_rational_t *timebase) {
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
   AV1_COMMON *const cm = &cpi->common;
-  const int num_planes = av1_num_planes(cm);
-  BufferPool *const pool = cm->buffer_pool;
-  RATE_CONTROL *const rc = &cpi->rc;
   struct aom_usec_timer cmptimer;
-  YV12_BUFFER_CONFIG *force_src_buffer = NULL;
-  struct lookahead_entry *last_source = NULL;
-  struct lookahead_entry *source = NULL;
-  int arf_src_index;
-  int brf_src_index;
-  int i;
 
 #if CONFIG_BITSTREAM_DEBUG
   assert(cpi->oxcf.max_threads == 0 &&
          "bitstream debug tool does not support multithreading");
   bitstream_queue_record_write();
-  bitstream_queue_set_frame_write(cm->current_video_frame * 2 + cm->show_frame);
+  bitstream_queue_set_frame_write(current_frame->frame_number * 2 +
+                                  cm->show_frame);
 #endif
 
+  // Indicates whether or not to use an adaptive quantize b rather than
+  // the traditional version
+  cm->use_quant_b_adapt = cpi->oxcf.quant_b_adapt;
+
   cm->showable_frame = 0;
+  *size = 0;
   aom_usec_timer_start(&cmptimer);
 
   set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV, 0);
@@ -5806,381 +5728,44 @@ int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags,
     cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED;
 
   // default reference buffers update config
-  av1_configure_buffer_updates_firstpass(cpi, LF_UPDATE);
+  av1_configure_buffer_updates(cpi, LF_UPDATE);
 
   // Initialize fields related to forward keyframes
   cpi->no_show_kf = 0;
-  cm->reset_decoder_state = 0;
 
-  // Don't allow a show_existing_frame to coincide with an error resilient or
-  // S-Frame. An exception can be made in the case of a keyframe, since it
-  // does not depend on any previous frames. We must make this exception here
-  // because of the use of show_existing_frame with forward coded keyframes.
-  struct lookahead_entry *lookahead_src = NULL;
-  if (cm->current_video_frame > 0)
-    lookahead_src = av1_lookahead_peek(cpi->lookahead, 0);
+  cm->show_existing_frame &= allow_show_existing(cpi);
 
-  int use_show_existing = 1;
-  if (lookahead_src != NULL) {
-    const int is_error_resilient =
-        cpi->oxcf.error_resilient_mode ||
-        (lookahead_src->flags & AOM_EFLAG_ERROR_RESILIENT);
-    const int is_s_frame = cpi->oxcf.s_frame_mode ||
-                           (lookahead_src->flags & AOM_EFLAG_SET_S_FRAME);
-    const int is_key_frame =
-        (rc->frames_to_key == 0) || (cpi->frame_flags & FRAMEFLAGS_KEY);
-    use_show_existing = !(is_error_resilient || is_s_frame) || is_key_frame;
-  }
+  if (assign_cur_frame_new_fb(cm) == NULL) return AOM_CODEC_ERROR;
 
-  if (oxcf->pass == 2 && cm->show_existing_frame && use_show_existing) {
-    // Manage the source buffer and flush out the source frame that has been
-    // coded already; Also get prepared for PSNR calculation if needed.
-    if ((source = av1_lookahead_pop(cpi->lookahead, flush)) == NULL) {
-      *size = 0;
-      return -1;
-    }
-    av1_apply_encoding_flags(cpi, source->flags);
-    cpi->source = &source->img;
-    // TODO(zoeliu): To track down to determine whether it's needed to adjust
-    // the frame rate.
-    *time_stamp = source->ts_start;
-    *time_end = source->ts_end;
-
-    // We need to adjust frame rate for an overlay frame
-    if (cpi->rc.is_src_frame_alt_ref) adjust_frame_rate(cpi, source);
-
-    // Find a free buffer for the new frame, releasing the reference
-    // previously
-    // held.
-    if (cm->new_fb_idx != INVALID_IDX) {
-      --pool->frame_bufs[cm->new_fb_idx].ref_count;
-    }
-    cm->new_fb_idx = get_free_fb(cm);
-
-    if (cm->new_fb_idx == INVALID_IDX) return -1;
-
-    // Clear down mmx registers
-    aom_clear_system_state();
-
-    // Start with a 0 size frame.
-    *size = 0;
-
-    // We need to update the gf_group for show_existing overlay frame
-    if (cpi->rc.is_src_frame_alt_ref) av1_rc_get_second_pass_params(cpi);
-
-    if (Pass2Encode(cpi, size, dest, frame_flags) != AOM_CODEC_OK)
-      return AOM_CODEC_ERROR;
-
-    if (cpi->b_calculate_psnr) generate_psnr_packet(cpi);
-
-#if CONFIG_INTERNAL_STATS
-    compute_internal_stats(cpi, (int)(*size));
-#endif  // CONFIG_INTERNAL_STATS
-
-    // Clear down mmx registers
-    aom_clear_system_state();
-
-    cm->show_existing_frame = 0;
-    return 0;
-  }
-
-  // Should we encode an arf frame.
-  arf_src_index = get_arf_src_index(cpi);
-  if (arf_src_index) {
-    for (i = 0; i <= arf_src_index; ++i) {
-      struct lookahead_entry *e = av1_lookahead_peek(cpi->lookahead, i);
-      // Avoid creating an alt-ref if there's a forced keyframe pending.
-      if (e == NULL) {
-        break;
-      } else if (e->flags == AOM_EFLAG_FORCE_KF) {
-        arf_src_index = 0;
-        flush = 1;
-        break;
-      }
-    }
-  }
-
-  if (arf_src_index) {
-    assert(arf_src_index <= rc->frames_to_key);
-
-    if ((source = av1_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
-      cm->showable_frame = 1;
-      cpi->alt_ref_source = source;
-      // When arf_src_index == rc->frames_to_key, it indicates a fwd_kf
-      if (arf_src_index == rc->frames_to_key) {
-        // Skip temporal filtering and mark as intra_only if we have a fwd_kf
-        const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
-        int which_arf = gf_group->arf_update_idx[gf_group->index];
-        cpi->is_arf_filter_off[which_arf] = 1;
-        cpi->no_show_kf = 1;
-      } else {
-        if (oxcf->arnr_max_frames > 0) {
-          // Produce the filtered ARF frame.
-          av1_temporal_filter(cpi, arf_src_index);
-          aom_extend_frame_borders(&cpi->alt_ref_buffer, num_planes);
-          force_src_buffer = &cpi->alt_ref_buffer;
-        }
-      }
-      cm->show_frame = 0;
-      cm->intra_only = 0;
-
-      if (oxcf->pass < 2) {
-        // In second pass, the buffer updates configure will be set
-        // in the function av1_rc_get_second_pass_params
-        av1_configure_buffer_updates_firstpass(cpi, ARF_UPDATE);
-      }
-    }
-    rc->source_alt_ref_pending = 0;
-  }
-
-  // Should we encode an arf2 frame.
-  arf_src_index = get_arf2_src_index(cpi);
-  if (arf_src_index) {
-    for (i = 0; i <= arf_src_index; ++i) {
-      struct lookahead_entry *e = av1_lookahead_peek(cpi->lookahead, i);
-      // Avoid creating an alt-ref if there's a forced keyframe pending.
-      if (e == NULL) {
-        break;
-      } else if (e->flags == AOM_EFLAG_FORCE_KF) {
-        arf_src_index = 0;
-        flush = 1;
-        break;
-      }
-    }
-  }
-
-  if (arf_src_index) {
-    assert(arf_src_index <= rc->frames_to_key);
-
-    if ((source = av1_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
-      cm->showable_frame = 1;
-      cpi->alt_ref_source = source;
-
-      if (oxcf->arnr_max_frames > 0) {
-        // Produce the filtered ARF frame.
-        av1_temporal_filter(cpi, arf_src_index);
-        aom_extend_frame_borders(&cpi->alt_ref_buffer, num_planes);
-        force_src_buffer = &cpi->alt_ref_buffer;
-      }
-
-      cm->show_frame = 0;
-      cm->intra_only = 0;
-
-      if (oxcf->pass < 2) {
-        // In second pass, the buffer updates configure will be set
-        // in the function av1_rc_get_second_pass_params
-        av1_configure_buffer_updates_firstpass(cpi, INTNL_ARF_UPDATE);
-      }
-    }
-    rc->source_alt_ref_pending = 0;
-  }
-
-  rc->is_bwd_ref_frame = 0;
-  brf_src_index = get_brf_src_index(cpi);
-  if (brf_src_index) {
-    assert(brf_src_index <= rc->frames_to_key);
-    if ((source = av1_lookahead_peek(cpi->lookahead, brf_src_index)) != NULL) {
-      cm->showable_frame = 1;
-      cm->show_frame = 0;
-      cm->intra_only = 0;
-
-      if (oxcf->pass < 2) {
-        // In second pass, the buffer updates configure will be set
-        // in the function av1_rc_get_second_pass_params
-        av1_configure_buffer_updates_firstpass(cpi, BIPRED_UPDATE);
-      }
-    }
-  }
-
-  if (!source) {
-    // Get last frame source.
-    if (cm->current_video_frame > 0) {
-      if ((last_source = av1_lookahead_peek(cpi->lookahead, -1)) == NULL)
-        return -1;
-    }
-    if (cm->current_video_frame > 0) assert(last_source != NULL);
-    // Read in the source frame.
-    source = av1_lookahead_pop(cpi->lookahead, flush);
-
-    if (source != NULL) {
-      cm->show_frame = 1;
-      cm->intra_only = 0;
-
-      // Check to see if the frame should be encoded as an arf overlay.
-      check_src_altref(cpi, source);
-    }
-  }
-  if (source) {
-    cpi->unscaled_source = cpi->source =
-        force_src_buffer ? force_src_buffer : &source->img;
-    cpi->unscaled_last_source = last_source != NULL ? &last_source->img : NULL;
-
-    *time_stamp = source->ts_start;
-    *time_end = source->ts_end;
-    av1_apply_encoding_flags(cpi, source->flags);
-    *frame_flags = (source->flags & AOM_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
-
-  } else {
-    *size = 0;
-    if (flush && oxcf->pass == 1 && !cpi->twopass.first_pass_done) {
-      av1_end_first_pass(cpi); /* get last stats packet */
-      cpi->twopass.first_pass_done = 1;
-    }
+  const int result = av1_encode_strategy(cpi, size, dest, frame_flags,
+                                         time_stamp, time_end, timebase, flush);
+  if (result != AOM_CODEC_OK && result != -1) {
+    return AOM_CODEC_ERROR;
+  } else if (result == -1) {
+    // Returning -1 indicates no frame encoded; more input is required
     return -1;
   }
 
-  if (source->ts_start < cpi->first_time_stamp_ever) {
-    cpi->first_time_stamp_ever = source->ts_start;
-    cpi->last_end_time_stamp_seen = source->ts_start;
-  }
-
-  // Clear down mmx registers
-  aom_clear_system_state();
-
-  // adjust frame rates based on timestamps given
-  if (cm->show_frame) adjust_frame_rate(cpi, source);
-
-  // Find a free buffer for the new frame, releasing the reference previously
-  // held.
-  if (cm->new_fb_idx != INVALID_IDX) {
-    --pool->frame_bufs[cm->new_fb_idx].ref_count;
-  }
-  cm->new_fb_idx = get_free_fb(cm);
-
-  if (cm->new_fb_idx == INVALID_IDX) return -1;
-
-  // Retain the RF_LEVEL for the current newly coded frame.
-  cpi->frame_rf_level[cm->new_fb_idx] =
-      cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index];
-
-  cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
-  cm->cur_frame->buf.buf_8bit_valid = 0;
-
-  if (cpi->film_grain_table) {
-    cm->seq_params.film_grain_params_present = aom_film_grain_table_lookup(
-        cpi->film_grain_table, *time_stamp, *time_end, 0 /* =erase */,
-        &cm->film_grain_params);
-  }
-  cm->cur_frame->film_grain_params_present =
-      cm->seq_params.film_grain_params_present;
-
-  // only one operating point supported now
-  const int64_t pts64 = ticks_to_timebase_units(timebase, *time_stamp);
-  if (pts64 < 0 || pts64 > UINT32_MAX) return AOM_CODEC_ERROR;
-  cpi->common.frame_presentation_time = (uint32_t)pts64;
-
-  // Start with a 0 size frame.
-  *size = 0;
-
-  cpi->frame_flags = *frame_flags;
-
-  if (oxcf->pass == 2) {
-    av1_rc_get_second_pass_params(cpi);
-  } else if (oxcf->pass == 1) {
-    setup_frame_size(cpi);
-  }
-
-  if (cpi->oxcf.pass != 0 || frame_is_intra_only(cm) == 1) {
-    for (i = 0; i < REF_FRAMES; ++i) cpi->scaled_ref_idx[i] = INVALID_IDX;
-  }
-
-  cm->using_qmatrix = cpi->oxcf.using_qm;
-  cm->min_qmlevel = cpi->oxcf.qm_minlevel;
-  cm->max_qmlevel = cpi->oxcf.qm_maxlevel;
-
-  if (cm->seq_params.frame_id_numbers_present_flag) {
-    if (*time_stamp == 0) {
-      cpi->common.current_frame_id = -1;
-    }
-  }
-
-  cpi->cur_poc++;
-  if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools &&
-      !frame_is_intra_only(cm)) {
-    if (cpi->common.seq_params.force_integer_mv == 2) {
-      struct lookahead_entry *previous_entry =
-          av1_lookahead_peek(cpi->lookahead, cpi->previous_index);
-      if (!previous_entry)
-        cpi->common.cur_frame_force_integer_mv = 0;
-      else
-        cpi->common.cur_frame_force_integer_mv = is_integer_mv(
-            cpi, cpi->source, &previous_entry->img, cpi->previous_hash_table);
-    } else {
-      cpi->common.cur_frame_force_integer_mv =
-          cpi->common.seq_params.force_integer_mv;
-    }
-  } else {
-    cpi->common.cur_frame_force_integer_mv = 0;
-  }
-
-  if (oxcf->pass == 1) {
-    cpi->td.mb.e_mbd.lossless[0] = is_lossless_requested(oxcf);
-    av1_first_pass(cpi, source);
-  } else if (oxcf->pass == 2) {
-    if (Pass2Encode(cpi, size, dest, frame_flags) != AOM_CODEC_OK)
-      return AOM_CODEC_ERROR;
-  } else {
-    // One pass encode
-    if (Pass0Encode(cpi, size, dest, 0, frame_flags) != AOM_CODEC_OK)
-      return AOM_CODEC_ERROR;
-  }
-  if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools) {
-    cpi->previous_hash_table = &cm->cur_frame->hash_table;
-    {
-      int l;
-      for (l = -MAX_PRE_FRAMES; l < cpi->lookahead->max_sz; l++) {
-        if ((cpi->lookahead->buf + l) == source) {
-          cpi->previous_index = l;
-          break;
-        }
-      }
-
-      if (l == cpi->lookahead->max_sz) {
-        aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
-                           "Failed to find last frame original buffer");
-      }
-    }
-  }
-
-  if (!cm->large_scale_tile) {
-    cm->frame_contexts[cm->new_fb_idx] = *cm->fc;
-  }
-
-#define EXT_TILE_DEBUG 0
-#if EXT_TILE_DEBUG
-  if (cm->large_scale_tile && oxcf->pass == 2) {
-    char fn[20] = "./fc";
-    fn[4] = cm->current_video_frame / 100 + '0';
-    fn[5] = (cm->current_video_frame % 100) / 10 + '0';
-    fn[6] = (cm->current_video_frame % 10) + '0';
-    fn[7] = '\0';
-    av1_print_frame_contexts(cm->fc, fn);
-  }
-#endif  // EXT_TILE_DEBUG
-#undef EXT_TILE_DEBUG
-
-  cm->showable_frame = !cm->show_frame && cm->showable_frame;
-
-  // No frame encoded, or frame was dropped, release scaled references.
-  if ((*size == 0) && (frame_is_intra_only(cm) == 0)) {
-    release_scaled_references(cpi);
-  }
-
-  if (*size > 0) {
-    cpi->droppable = !frame_is_reference(cpi);
-  }
-
   aom_usec_timer_mark(&cmptimer);
   cpi->time_compress_data += aom_usec_timer_elapsed(&cmptimer);
 
-  if (cpi->b_calculate_psnr && oxcf->pass != 1 && cm->show_frame)
-    generate_psnr_packet(cpi);
+  if (cpi->b_calculate_psnr) {
+    if (cm->show_existing_frame || (oxcf->pass != 1 && cm->show_frame)) {
+      generate_psnr_packet(cpi);
+    }
+  }
 
 #if CONFIG_INTERNAL_STATS
   if (oxcf->pass != 1) {
     compute_internal_stats(cpi, (int)(*size));
   }
 #endif  // CONFIG_INTERNAL_STATS
+#if CONFIG_SPEED_STATS
+  if (cpi->oxcf.pass != 1 && !cm->show_existing_frame) {
+    cpi->tx_search_count += cpi->td.mb.tx_search_count;
+    cpi->td.mb.tx_search_count = 0;
+  }
+#endif  // CONFIG_SPEED_STATS
 
   aom_clear_system_state();
 
@@ -6193,8 +5778,8 @@ int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest) {
     return -1;
   } else {
     int ret;
-    if (cm->frame_to_show) {
-      *dest = *cm->frame_to_show;
+    if (cm->cur_frame != NULL) {
+      *dest = cm->cur_frame->buf;
       dest->y_width = cm->width;
       dest->y_height = cm->height;
       dest->uv_width = cm->width >> cm->seq_params.subsampling_x;
@@ -6209,10 +5794,9 @@ int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest) {
 }
 
 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame) {
-  if (cpi->last_show_frame_buf_idx == INVALID_IDX) return -1;
+  if (cpi->last_show_frame_buf == NULL) return -1;
 
-  *frame =
-      cpi->common.buffer_pool->frame_bufs[cpi->last_show_frame_buf_idx].buf;
+  *frame = cpi->last_show_frame_buf->buf;
   return 0;
 }
 
@@ -6364,7 +5948,12 @@ void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags) {
       upd ^= AOM_ALT2_FLAG;
     }
 
-    av1_update_reference(cpi, upd);
+    cpi->ext_refresh_last_frame = (upd & AOM_LAST_FLAG) != 0;
+    cpi->ext_refresh_golden_frame = (upd & AOM_GOLD_FLAG) != 0;
+    cpi->ext_refresh_alt_ref_frame = (upd & AOM_ALT_FLAG) != 0;
+    cpi->ext_refresh_bwd_ref_frame = (upd & AOM_BWD_FLAG) != 0;
+    cpi->ext_refresh_alt2_ref_frame = (upd & AOM_ALT2_FLAG) != 0;
+    cpi->ext_refresh_frame_flags_pending = 1;
   }
 
   cpi->ext_use_ref_frame_mvs = cpi->oxcf.allow_ref_frame_mvs &
@@ -6380,15 +5969,6 @@ void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags) {
   }
 }
 
-int64_t timebase_units_to_ticks(const aom_rational_t *timebase, int64_t n) {
-  return n * TICKS_PER_SEC * timebase->num / timebase->den;
-}
-
-int64_t ticks_to_timebase_units(const aom_rational_t *timebase, int64_t n) {
-  const int64_t round = TICKS_PER_SEC * timebase->num / 2 - 1;
-  return (n * timebase->den + round) / timebase->num / TICKS_PER_SEC;
-}
-
 aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi) {
   if (!cpi) return NULL;
 
diff --git a/third_party/aom/av1/encoder/encoder.h b/third_party/aom/av1/encoder/encoder.h
index ee7fc46379..dec531fa88 100644
--- a/third_party/aom/av1/encoder/encoder.h
+++ b/third_party/aom/av1/encoder/encoder.h
@@ -59,7 +59,7 @@ typedef struct {
   FRAME_CONTEXT fc;
 } CODING_CONTEXT;
 
-typedef enum {
+enum {
   // regular inter frame
   REGULAR_FRAME = 0,
   // alternate reference frame
@@ -73,51 +73,54 @@ typedef enum {
   // extra alternate reference frame
   EXT_ARF_FRAME = 5,
   FRAME_CONTEXT_INDEXES
-} FRAME_CONTEXT_INDEX;
+} UENUM1BYTE(FRAME_CONTEXT_INDEX);
 
-typedef enum {
+enum {
   NORMAL = 0,
   FOURFIVE = 1,
   THREEFIVE = 2,
   ONETWO = 3
-} AOM_SCALING;
+} UENUM1BYTE(AOM_SCALING);
 
-typedef enum {
+enum {
   // Good Quality Fast Encoding. The encoder balances quality with the amount of
   // time it takes to encode the output. Speed setting controls how fast.
   GOOD
-} MODE;
+} UENUM1BYTE(MODE);
 
-typedef enum {
+enum {
   FRAMEFLAGS_KEY = 1 << 0,
   FRAMEFLAGS_GOLDEN = 1 << 1,
   FRAMEFLAGS_BWDREF = 1 << 2,
   // TODO(zoeliu): To determine whether a frame flag is needed for ALTREF2_FRAME
   FRAMEFLAGS_ALTREF = 1 << 3,
-} FRAMETYPE_FLAGS;
+  FRAMEFLAGS_INTRAONLY = 1 << 4,
+  FRAMEFLAGS_SWITCH = 1 << 5,
+  FRAMEFLAGS_ERROR_RESILIENT = 1 << 6,
+} UENUM1BYTE(FRAMETYPE_FLAGS);
 
-typedef enum {
+enum {
   NO_AQ = 0,
   VARIANCE_AQ = 1,
   COMPLEXITY_AQ = 2,
   CYCLIC_REFRESH_AQ = 3,
   AQ_MODE_COUNT  // This should always be the last member of the enum
-} AQ_MODE;
-typedef enum {
+} UENUM1BYTE(AQ_MODE);
+enum {
   NO_DELTA_Q = 0,
   DELTA_Q_ONLY = 1,
   DELTA_Q_LF = 2,
   DELTAQ_MODE_COUNT  // This should always be the last member of the enum
-} DELTAQ_MODE;
+} UENUM1BYTE(DELTAQ_MODE);
 
-typedef enum {
+enum {
   RESIZE_NONE = 0,    // No frame resizing allowed.
   RESIZE_FIXED = 1,   // All frames are coded at the specified scale.
   RESIZE_RANDOM = 2,  // All frames are coded at a random scale.
   RESIZE_MODES
-} RESIZE_MODE;
+} UENUM1BYTE(RESIZE_MODE);
 
-typedef enum {
+enum {
   SUPERRES_NONE = 0,     // No frame superres allowed
   SUPERRES_FIXED = 1,    // All frames are coded at the specified scale,
                          // and super-resolved.
@@ -126,7 +129,31 @@ typedef enum {
   SUPERRES_QTHRESH = 3,  // Superres scale for a frame is determined based on
                          // q_index
   SUPERRES_MODES
-} SUPERRES_MODE;
+} UENUM1BYTE(SUPERRES_MODE);
+
+typedef struct TplDepStats {
+  int64_t intra_cost;
+  int64_t inter_cost;
+  int64_t mc_flow;
+  int64_t mc_dep_cost;
+  int64_t mc_ref_cost;
+
+  int ref_frame_index;
+  int_mv mv;
+} TplDepStats;
+
+typedef struct TplDepFrame {
+  uint8_t is_valid;
+  TplDepStats *tpl_stats_ptr;
+  int stride;
+  int width;
+  int height;
+  int mi_rows;
+  int mi_cols;
+  int base_qindex;
+} TplDepFrame;
+
+#define TPL_DEP_COST_SCALE_LOG2 4
 
 typedef struct AV1EncoderConfig {
   BITSTREAM_PROFILE profile;
@@ -188,6 +215,7 @@ typedef struct AV1EncoderConfig {
   DELTAQ_MODE deltaq_mode;
   int enable_cdef;
   int enable_restoration;
+  int enable_obmc;
   int disable_trellis_quant;
   int using_qm;
   int qm_y;
@@ -247,6 +275,7 @@ typedef struct AV1EncoderConfig {
 
   int min_gf_interval;
   int max_gf_interval;
+  int gf_max_pyr_height;
 
   int row_mt;
   int tile_columns;
@@ -256,10 +285,11 @@ typedef struct AV1EncoderConfig {
   int tile_widths[MAX_TILE_COLS];
   int tile_heights[MAX_TILE_ROWS];
 
+  int enable_tpl_model;
+
   int max_threads;
 
   aom_fixed_buf_t two_pass_stats_in;
-  struct aom_codec_pkt_list *output_pkt_list;
 
 #if CONFIG_FP_MB_STATS
   aom_fixed_buf_t firstpass_mb_stats_in;
@@ -275,15 +305,12 @@ typedef struct AV1EncoderConfig {
   int color_range;
   int render_width;
   int render_height;
-  aom_timing_info_type_t timing_info_type;
   int timing_info_present;
   aom_timing_info_t timing_info;
   int decoder_model_info_present_flag;
   int display_model_info_present_flag;
   int buffer_removal_time_present;
   aom_dec_model_info_t buffer_model;
-  aom_dec_model_op_parameters_t op_params[MAX_NUM_OPERATING_POINTS + 1];
-  aom_op_timing_info_t op_frame_timing[MAX_NUM_OPERATING_POINTS + 1];
   int film_grain_test_vector;
   const char *film_grain_table_filename;
 
@@ -291,18 +318,36 @@ typedef struct AV1EncoderConfig {
   aom_superblock_size_t superblock_size;
   unsigned int large_scale_tile;
   unsigned int single_tile_decoding;
-  int monochrome;
+  uint8_t monochrome;
   unsigned int full_still_picture_hdr;
   int enable_dual_filter;
   unsigned int motion_vector_unit_test;
   const cfg_options_t *cfg;
+  int enable_rect_partitions;
+  int enable_intra_edge_filter;
+  int enable_tx64;
   int enable_order_hint;
-  int enable_jnt_comp;
+  int enable_dist_wtd_comp;
   int enable_ref_frame_mvs;
+  unsigned int max_reference_frames;
   unsigned int allow_ref_frame_mvs;
+  int enable_masked_comp;
+  int enable_interintra_comp;
+  int enable_smooth_interintra;
+  int enable_diff_wtd_comp;
+  int enable_interinter_wedge;
+  int enable_interintra_wedge;
+  int enable_global_motion;
   int enable_warped_motion;
   int allow_warped_motion;
+  int enable_filter_intra;
+  int enable_smooth_intra;
+  int enable_paeth_intra;
+  int enable_cfl_intra;
   int enable_superres;
+  int enable_palette;
+  int enable_intrabc;
+  int enable_angle_delta;
   unsigned int save_as_annexb;
 
 #if CONFIG_DENOISE
@@ -312,6 +357,11 @@ typedef struct AV1EncoderConfig {
 
   unsigned int chroma_subsampling_x;
   unsigned int chroma_subsampling_y;
+  int reduced_tx_type_set;
+  int use_intra_dct_only;
+  int use_inter_dct_only;
+  int quant_b_adapt;
+  int border_in_pixels;
 } AV1EncoderConfig;
 
 static INLINE int is_lossless_requested(const AV1EncoderConfig *cfg) {
@@ -404,7 +454,6 @@ typedef struct FRAME_COUNTS {
                                 [SWITCHABLE_FILTERS];
 } FRAME_COUNTS;
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
 #define INTER_MODE_RD_DATA_OVERALL_SIZE 6400
 
 typedef struct {
@@ -439,22 +488,37 @@ typedef struct inter_modes_info {
   int64_t est_rd_arr[MAX_INTER_MODES];
   RdIdxPair rd_idx_pair_arr[MAX_INTER_MODES];
 } InterModesInfo;
+
+// Encoder row synchronization
+typedef struct AV1RowMTSyncData {
+#if CONFIG_MULTITHREAD
+  pthread_mutex_t *mutex_;
+  pthread_cond_t *cond_;
 #endif
+  // Allocate memory to store the sb/mb block index in each row.
+  int *cur_col;
+  int sync_range;
+  int rows;
+} AV1RowMTSync;
+
+typedef struct AV1RowMTInfo {
+  int current_mi_row;
+  int num_threads_working;
+} AV1RowMTInfo;
 
 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
 typedef struct TileDataEnc {
   TileInfo tile_info;
   int thresh_freq_fact[BLOCK_SIZES_ALL][MAX_MODES];
-  int mode_map[BLOCK_SIZES_ALL][MAX_MODES];
   int m_search_count;
   int ex_search_count;
   CFL_CTX cfl;
   DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
+  FRAME_CONTEXT *row_ctx;
   uint8_t allow_update_cdf;
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
   InterModeRdModel inter_mode_rd_models[BLOCK_SIZES_ALL];
-  InterModesInfo inter_modes_info;
-#endif
+  AV1RowMTSync row_mt_sync;
+  AV1RowMTInfo row_mt_info;
 } TileDataEnc;
 
 typedef struct {
@@ -463,6 +527,13 @@ typedef struct {
   unsigned int count;
 } TOKENLIST;
 
+typedef struct MultiThreadHandle {
+  int allocated_tile_rows;
+  int allocated_tile_cols;
+  int allocated_sb_rows;
+  int thread_id_to_tile_id[MAX_NUM_THREADS];  // Mapping of threads to tiles
+} MultiThreadHandle;
+
 typedef struct RD_COUNTS {
   int64_t comp_pred_diff[REFERENCE_MODES];
   // Stores number of 4x4 blocks using global motion per reference frame.
@@ -477,6 +548,10 @@ typedef struct ThreadData {
   FRAME_COUNTS *counts;
   PC_TREE *pc_tree;
   PC_TREE *pc_root[MAX_MIB_SIZE_LOG2 - MIN_MIB_SIZE_LOG2 + 1];
+  tran_low_t *tree_coeff_buf[MAX_MB_PLANE];
+  tran_low_t *tree_qcoeff_buf[MAX_MB_PLANE];
+  tran_low_t *tree_dqcoeff_buf[MAX_MB_PLANE];
+  InterModesInfo *inter_modes_info;
   uint32_t *hash_value_buffer[2][2];
   int32_t *wsrc_buf;
   int32_t *mask_buf;
@@ -485,7 +560,8 @@ typedef struct ThreadData {
   PALETTE_BUFFER *palette_buffer;
   CONV_BUF_TYPE *tmp_conv_dst;
   uint8_t *tmp_obmc_bufs[2];
-  int intrabc_used_this_tile;
+  int intrabc_used;
+  FRAME_CONTEXT *tctx;
 } ThreadData;
 
 struct EncWorkerData;
@@ -498,13 +574,13 @@ typedef struct ActiveMap {
 
 #if CONFIG_INTERNAL_STATS
 // types of stats
-typedef enum {
+enum {
   STAT_Y,
   STAT_U,
   STAT_V,
   STAT_ALL,
   NUM_STAT_TYPES  // This should always be the last member of the enum
-} StatType;
+} UENUM1BYTE(StatType);
 
 typedef struct IMAGE_STAT {
   double stat[NUM_STAT_TYPES];
@@ -517,10 +593,14 @@ typedef struct {
   YV12_BUFFER_CONFIG buf;
 } EncRefCntBuffer;
 
-typedef struct TileBufferEnc {
-  uint8_t *data;
-  size_t size;
-} TileBufferEnc;
+#if CONFIG_COLLECT_PARTITION_STATS
+typedef struct PartitionStats {
+  int partition_decisions[6][EXT_PARTITION_TYPES];
+  int partition_attempts[6][EXT_PARTITION_TYPES];
+
+  int partition_redo;
+} PartitionStats;
+#endif
 
 typedef struct AV1_COMP {
   QUANTS quants;
@@ -545,6 +625,9 @@ typedef struct AV1_COMP {
   YV12_BUFFER_CONFIG *unscaled_last_source;
   YV12_BUFFER_CONFIG scaled_last_source;
 
+  TplDepFrame tpl_stats[MAX_LAG_BUFFERS];
+  YV12_BUFFER_CONFIG *tpl_recon_frames[INTER_REFS_PER_FRAME + 1];
+
   // For a still frame, this flag is set to 1 to skip partition search.
   int partition_search_skippable_frame;
   double csm_rate_array[32];
@@ -553,23 +636,41 @@ typedef struct AV1_COMP {
   int rate_index;
   hash_table *previous_hash_table;
   int previous_index;
-  int cur_poc;  // DebugInfo
 
   unsigned int row_mt;
-  int scaled_ref_idx[REF_FRAMES];
-  int ref_fb_idx[REF_FRAMES];
-  int refresh_fb_idx;  // ref frame buffer index to refresh
+  RefCntBuffer *scaled_ref_buf[INTER_REFS_PER_FRAME];
 
-  int last_show_frame_buf_idx;  // last show frame buffer index
+  RefCntBuffer *last_show_frame_buf;  // last show frame buffer
 
+  // refresh_*_frame are boolean flags. If 'refresh_xyz_frame' is true, then
+  // after the current frame is encoded, the XYZ reference frame gets refreshed
+  // (updated) to be the current frame.
+  //
+  // Special case: 'refresh_last_frame' specifies that:
+  // - LAST_FRAME reference should be updated to be the current frame (as usual)
+  // - Also, LAST2_FRAME and LAST3_FRAME references are implicitly updated to be
+  // the two past reference frames just before LAST_FRAME that are available.
+  //
+  // Note: Usually at most one of these refresh flags is true at a time.
+  // But a key-frame is special, for which all the flags are true at once.
   int refresh_last_frame;
   int refresh_golden_frame;
   int refresh_bwd_ref_frame;
   int refresh_alt2_ref_frame;
   int refresh_alt_ref_frame;
-#if USE_SYMM_MULTI_LAYER
+
+  // For each type of reference frame, this contains the index of a reference
+  // frame buffer for a reference frame of the same type.  We use this to
+  // choose our primary reference frame (which is the most recent reference
+  // frame of the same type as the current frame).
+  int fb_of_context_type[REF_FRAMES];
+
+  // When true, a new rule for backward (future) reference frames is in effect:
+  // - BWDREF_FRAME is always the closest future frame available
+  // - ALTREF2_FRAME is always the 2nd closest future frame available
+  // - 'refresh_bwd_ref_frame' flag is used for updating both the BWDREF_FRAME
+  // and ALTREF2_FRAME. ('refresh_alt2_ref_frame' flag is irrelevant).
   int new_bwdref_update_rule;
-#endif
 
   int ext_refresh_frame_flags_pending;
   int ext_refresh_last_frame;
@@ -608,9 +709,10 @@ typedef struct AV1_COMP {
   RATE_CONTROL rc;
   double framerate;
 
-  // NOTE(zoeliu): Any inter frame allows maximum of REF_FRAMES inter
-  // references; Plus the currently coded frame itself, it is needed to allocate
-  // sufficient space to the size of the maximum possible number of frames.
+  // Relevant for an inter frame.
+  // - Index '0' corresponds to the values for the currently coded frame.
+  // - Indices LAST_FRAME ... EXTREF_FRAMES are used to store values for all the
+  // possible inter reference frames.
   int interp_filter_selected[REF_FRAMES + 1][SWITCHABLE];
 
   struct aom_codec_pkt_list *output_pkt_list;
@@ -620,14 +722,15 @@ typedef struct AV1_COMP {
   int static_mb_pct;     // % forced skip mbs by segmentation
   int ref_frame_flags;
   int ext_ref_frame_flags;
-  RATE_FACTOR_LEVEL frame_rf_level[FRAME_BUFFERS];
 
+  // speed is passed as a per-frame parameter into the encoder
+  int speed;
+  // sf contains fine-grained config set internally based on speed
   SPEED_FEATURES sf;
 
   unsigned int max_mv_magnitude;
   int mv_step_param;
 
-  int allow_comp_inter_inter;
   int all_one_sided_refs;
 
   uint8_t *segmentation_map;
@@ -641,7 +744,6 @@ typedef struct AV1_COMP {
   uint64_t time_receive_data;
   uint64_t time_compress_data;
   uint64_t time_pick_lpf;
-  uint64_t time_encode_sb_row;
 
 #if CONFIG_FP_MB_STATS
   int use_fp_mb_stats;
@@ -680,6 +782,9 @@ typedef struct AV1_COMP {
   Metrics metrics;
 #endif
   int b_calculate_psnr;
+#if CONFIG_SPEED_STATS
+  unsigned int tx_search_count;
+#endif  // CONFIG_SPEED_STATS
 
   int droppable;
 
@@ -705,15 +810,11 @@ typedef struct AV1_COMP {
   int allocated_tiles;  // Keep track of memory allocated for tiles.
 
   TOKENEXTRA *tile_tok[MAX_TILE_ROWS][MAX_TILE_COLS];
-  unsigned int tok_count[MAX_TILE_ROWS][MAX_TILE_COLS];
   TOKENLIST *tplist[MAX_TILE_ROWS][MAX_TILE_COLS];
 
-  TileBufferEnc tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
-
   int resize_state;
   int resize_avg_qp;
   int resize_buffer_underflow;
-  int resize_count;
 
   // Sequence parameters have been transmitted already and locked
   // or not. Once locked av1_change_config cannot change the seq
@@ -727,14 +828,12 @@ typedef struct AV1_COMP {
   int num_workers;
   AVxWorker *workers;
   struct EncWorkerData *tile_thr_data;
-  int refresh_frame_mask;
   int existing_fb_idx_to_show;
   int is_arf_filter_off[MAX_EXT_ARFS + 1];
   int num_extra_arfs;
   int arf_pos_in_gf[MAX_EXT_ARFS + 1];
   int arf_pos_for_ovrly[MAX_EXT_ARFS + 1];
   int global_motion_search_done;
-  tran_low_t *tcoeff_buf[MAX_MB_PLANE];
   int extra_arf_allowed;
   // A flag to indicate if intrabc is ever used in current frame.
   int intrabc_used;
@@ -743,10 +842,16 @@ typedef struct AV1_COMP {
   int dv_joint_cost[MV_JOINTS];
   int has_lossless_segment;
 
-  // For frame refs short signaling:
-  //   A mapping of each reference frame from its encoder side value to the
-  //   decoder side value obtained following the short signaling procedure.
-  int ref_conv[REF_FRAMES];
+  // Factors to control gating of compound type selection based on best
+  // approximate rd so far
+  int max_comp_type_rd_threshold_mul;
+  int max_comp_type_rd_threshold_div;
+
+  unsigned int tx_domain_dist_threshold;
+
+  // Factor to control R-D optimization of coeffs based on block
+  // mse.
+  unsigned int coeff_opt_dist_threshold;
 
   AV1LfSync lf_row_sync;
   AV1LrSync lr_row_sync;
@@ -760,8 +865,50 @@ typedef struct AV1_COMP {
   // Set as 1 for monochrome and 3 for other color formats
   int default_interp_skip_flags;
   int preserve_arf_as_gld;
+  MultiThreadHandle multi_thread_ctxt;
+  void (*row_mt_sync_read_ptr)(AV1RowMTSync *const, int, int);
+  void (*row_mt_sync_write_ptr)(AV1RowMTSync *const, int, int, const int);
+#if CONFIG_MULTITHREAD
+  pthread_mutex_t *row_mt_mutex_;
+#endif
+  // Set if screen content is set or relevant tools are enabled
+  int is_screen_content_type;
+#if CONFIG_COLLECT_PARTITION_STATS
+  PartitionStats partition_stats;
+#endif
 } AV1_COMP;
 
+typedef struct {
+  YV12_BUFFER_CONFIG *source;
+  YV12_BUFFER_CONFIG *last_source;
+  int64_t ts_duration;
+} EncodeFrameInput;
+
+// EncodeFrameParams contains per-frame encoding parameters decided upon by
+// av1_encode_strategy() and passed down to av1_encode()
+struct EncodeFrameParams {
+  int error_resilient_mode;
+  FRAME_TYPE frame_type;
+  int primary_ref_frame;
+  int order_offset;
+  int show_frame;
+
+  // This is a bitmask of which reference slots can be used in this frame
+  int ref_frame_flags;
+
+  // Speed level to use for this frame: Bigger number means faster.
+  int speed;
+
+  unsigned int *frame_flags;
+};
+typedef struct EncodeFrameParams EncodeFrameParams;
+
+// EncodeFrameResults contains information about the result of encoding a
+// single frame
+typedef struct {
+  size_t size;  // Size of resulting bitstream
+} EncodeFrameResults;
+
 // Must not be called more than once.
 void av1_initialize_enc(void);
 
@@ -782,6 +929,11 @@ int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags,
                             int64_t *time_end, int flush,
                             const aom_rational_t *timebase);
 
+int av1_encode(AV1_COMP *const cpi, uint8_t *const dest,
+               const EncodeFrameInput *const frame_input,
+               const EncodeFrameParams *const frame_params,
+               EncodeFrameResults *const frame_results);
+
 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest);
 
 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame);
@@ -792,12 +944,12 @@ aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm,
 
 int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags);
 
-void av1_update_reference(AV1_COMP *cpi, int ref_frame_flags);
-
 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
 
 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
 
+void av1_set_frame_size(AV1_COMP *cpi, int width, int height);
+
 int av1_update_entropy(AV1_COMP *cpi, int update);
 
 int av1_set_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
@@ -811,59 +963,62 @@ int av1_get_quantizer(struct AV1_COMP *cpi);
 
 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *input_size);
 
-int64_t timebase_units_to_ticks(const aom_rational_t *timebase, int64_t n);
-int64_t ticks_to_timebase_units(const aom_rational_t *timebase, int64_t n);
+// av1 uses 10,000,000 ticks/second as time stamp
+#define TICKS_PER_SEC 10000000LL
+
+static INLINE int64_t timebase_units_to_ticks(const aom_rational_t *timebase,
+                                              int64_t n) {
+  return n * TICKS_PER_SEC * timebase->num / timebase->den;
+}
+
+static INLINE int64_t ticks_to_timebase_units(const aom_rational_t *timebase,
+                                              int64_t n) {
+  const int64_t round = TICKS_PER_SEC * timebase->num / 2 - 1;
+  return (n * timebase->den + round) / timebase->num / TICKS_PER_SEC;
+}
 
 static INLINE int frame_is_kf_gf_arf(const AV1_COMP *cpi) {
   return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
          (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
 }
 
-static INLINE int get_ref_frame_map_idx(const AV1_COMP *cpi,
-                                        MV_REFERENCE_FRAME ref_frame) {
-  return (ref_frame >= 1) ? cpi->ref_fb_idx[ref_frame - 1] : INVALID_IDX;
-}
-
-static INLINE int get_ref_frame_buf_idx(const AV1_COMP *cpi,
-                                        MV_REFERENCE_FRAME ref_frame) {
-  const AV1_COMMON *const cm = &cpi->common;
-  const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
-  return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
-}
-
 // TODO(huisu@google.com, youzhou@microsoft.com): enable hash-me for HBD.
 static INLINE int av1_use_hash_me(const AV1_COMMON *const cm) {
   return cm->allow_screen_content_tools;
 }
 
 static INLINE hash_table *av1_get_ref_frame_hash_map(
-    const AV1_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
-  const AV1_COMMON *const cm = &cpi->common;
-  const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-  return buf_idx != INVALID_IDX
-             ? &cm->buffer_pool->frame_bufs[buf_idx].hash_table
-             : NULL;
+    const AV1_COMMON *cm, MV_REFERENCE_FRAME ref_frame) {
+  const int map_idx = get_ref_frame_map_idx(cm, ref_frame);
+  RefCntBuffer *buf =
+      (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : NULL;
+  return buf ? &buf->hash_table : NULL;
 }
 
-static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
-    const AV1_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
-  const AV1_COMMON *const cm = &cpi->common;
-  const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-  return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
-                                : NULL;
+static INLINE const YV12_BUFFER_CONFIG *get_ref_frame_yv12_buf(
+    const AV1_COMMON *const cm, MV_REFERENCE_FRAME ref_frame) {
+  const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+  return buf != NULL ? &buf->buf : NULL;
 }
 
-static INLINE int enc_is_ref_frame_buf(AV1_COMP *cpi, RefCntBuffer *frame_buf) {
+static INLINE int enc_is_ref_frame_buf(const AV1_COMMON *const cm,
+                                       const RefCntBuffer *const frame_buf) {
   MV_REFERENCE_FRAME ref_frame;
-  AV1_COMMON *const cm = &cpi->common;
   for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-    if (buf_idx == INVALID_IDX) continue;
-    if (frame_buf == &cm->buffer_pool->frame_bufs[buf_idx]) break;
+    const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
+    if (buf == NULL) continue;
+    if (frame_buf == buf) break;
   }
   return (ref_frame <= ALTREF_FRAME);
 }
 
+static INLINE void alloc_frame_mvs(AV1_COMMON *const cm, RefCntBuffer *buf) {
+  assert(buf != NULL);
+  ensure_mv_buffer(buf, cm);
+  buf->width = cm->width;
+  buf->height = cm->height;
+}
+
 // Token buffer is only used for palette tokens.
 static INLINE unsigned int get_token_alloc(int mb_rows, int mb_cols,
                                            int sb_size_log2,
@@ -919,10 +1074,10 @@ static INLINE int is_altref_enabled(const AV1_COMP *const cpi) {
 static INLINE void set_ref_ptrs(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                 MV_REFERENCE_FRAME ref0,
                                 MV_REFERENCE_FRAME ref1) {
-  xd->block_refs[0] =
-      &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
-  xd->block_refs[1] =
-      &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
+  xd->block_ref_scale_factors[0] =
+      get_ref_scale_factors_const(cm, ref0 >= LAST_FRAME ? ref0 : 1);
+  xd->block_ref_scale_factors[1] =
+      get_ref_scale_factors_const(cm, ref1 >= LAST_FRAME ? ref1 : 1);
 }
 
 static INLINE int get_chessboard_index(int frame_index) {
@@ -935,20 +1090,10 @@ static INLINE int *cond_cost_list(const struct AV1_COMP *cpi, int *cost_list) {
 
 void av1_new_framerate(AV1_COMP *cpi, double framerate);
 
+void av1_setup_frame_size(AV1_COMP *cpi);
+
 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
 
-// Update up-sampled reference frame index.
-static INLINE void uref_cnt_fb(EncRefCntBuffer *ubufs, int *uidx,
-                               int new_uidx) {
-  const int ref_index = *uidx;
-
-  if (ref_index >= 0 && ubufs[ref_index].ref_count > 0)
-    ubufs[ref_index].ref_count--;
-
-  *uidx = new_uidx;
-  ubufs[new_uidx].ref_count++;
-}
-
 // Returns 1 if a frame is scaled and 0 otherwise.
 static INLINE int av1_resize_scaled(const AV1_COMMON *cm) {
   return !(cm->superres_upscaled_width == cm->render_width &&
@@ -963,8 +1108,8 @@ static INLINE int av1_frame_scaled(const AV1_COMMON *cm) {
 // frame. An exception can be made for a forward keyframe since it has no
 // previous dependencies.
 static INLINE int encode_show_existing_frame(const AV1_COMMON *cm) {
-  return cm->show_existing_frame &&
-         (!cm->error_resilient_mode || cm->frame_type == KEY_FRAME);
+  return cm->show_existing_frame && (!cm->error_resilient_mode ||
+                                     cm->current_frame.frame_type == KEY_FRAME);
 }
 
 // Returns a Sequence Header OBU stored in an aom_fixed_buf_t, or NULL upon
@@ -978,6 +1123,52 @@ static INLINE int encode_show_existing_frame(const AV1_COMMON *cm) {
 // field.
 aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi);
 
+#if CONFIG_COLLECT_PARTITION_STATS
+static INLINE void av1_print_partition_stats(PartitionStats *part_stats) {
+  FILE *f = fopen("partition_stats.csv", "w");
+  if (!f) {
+    return;
+  }
+
+  fprintf(f, "bsize,redo,");
+  for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
+    fprintf(f, "decision_%d,", part);
+  }
+  for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
+    fprintf(f, "attempt_%d,", part);
+  }
+  fprintf(f, "\n");
+
+  const int bsizes[6] = { 128, 64, 32, 16, 8, 4 };
+
+  for (int bsize_idx = 0; bsize_idx < 6; bsize_idx++) {
+    fprintf(f, "%d,%d,", bsizes[bsize_idx], part_stats->partition_redo);
+    for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
+      fprintf(f, "%d,", part_stats->partition_decisions[bsize_idx][part]);
+    }
+    for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
+      fprintf(f, "%d,", part_stats->partition_attempts[bsize_idx][part]);
+    }
+    fprintf(f, "\n");
+  }
+  fclose(f);
+}
+
+static INLINE int av1_get_bsize_idx_for_part_stats(BLOCK_SIZE bsize) {
+  assert(bsize == BLOCK_128X128 || bsize == BLOCK_64X64 ||
+         bsize == BLOCK_32X32 || bsize == BLOCK_16X16 || bsize == BLOCK_8X8);
+  switch (bsize) {
+    case BLOCK_128X128: return 0;
+    case BLOCK_64X64: return 1;
+    case BLOCK_32X32: return 2;
+    case BLOCK_16X16: return 3;
+    case BLOCK_8X8: return 4;
+    case BLOCK_4X4: return 5;
+    default: assert(0 && "Invalid bsize for partition_stats."); return -1;
+  }
+}
+#endif
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/encodetxb.c b/third_party/aom/av1/encoder/encodetxb.c
index 5a31d93d73..0f1a2666c4 100644
--- a/third_party/aom/av1/encoder/encodetxb.c
+++ b/third_party/aom/av1/encoder/encodetxb.c
@@ -284,20 +284,16 @@ static INLINE int get_sign_bit_cost(tran_low_t qc, int coeff_idx,
   return av1_cost_literal(1);
 }
 
-static INLINE int get_br_cost(tran_low_t abs_qc, int ctx,
-                              const int *coeff_lps) {
-  const tran_low_t min_level = 1 + NUM_BASE_LEVELS;
-  const tran_low_t max_level = 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE;
-  (void)ctx;
-  if (abs_qc >= min_level) {
-    if (abs_qc >= max_level) {
-      return coeff_lps[COEFF_BASE_RANGE];  // COEFF_BASE_RANGE * cost0;
-    } else {
-      return coeff_lps[(abs_qc - min_level)];  //  * cost0 + cost1;
-    }
-  }
-  return 0;
-}
+static const int golomb_bits_cost[32] = {
+  0,       512,     512 * 3, 512 * 3, 512 * 5, 512 * 5, 512 * 5, 512 * 5,
+  512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7, 512 * 7,
+  512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9,
+  512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9, 512 * 9
+};
+static const int golomb_cost_diff[32] = {
+  0,       512, 512 * 2, 0, 512 * 2, 0, 0, 0, 512 * 2, 0, 0, 0, 0, 0, 0, 0,
+  512 * 2, 0,   0,       0, 0,       0, 0, 0, 0,       0, 0, 0, 0, 0, 0, 0
+};
 
 static INLINE int get_golomb_cost(int abs_qc) {
   if (abs_qc >= 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
@@ -308,6 +304,32 @@ static INLINE int get_golomb_cost(int abs_qc) {
   return 0;
 }
 
+static INLINE int get_br_cost_with_diff(tran_low_t level, const int *coeff_lps,
+                                        int *diff) {
+  const int base_range = AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
+  int golomb_bits = 0;
+  if (level <= COEFF_BASE_RANGE + 1 + NUM_BASE_LEVELS)
+    *diff += coeff_lps[base_range + COEFF_BASE_RANGE + 1];
+
+  if (level >= COEFF_BASE_RANGE + 1 + NUM_BASE_LEVELS) {
+    int r = level - COEFF_BASE_RANGE - NUM_BASE_LEVELS;
+    if (r < 32) {
+      golomb_bits = golomb_bits_cost[r];
+      *diff += golomb_cost_diff[r];
+    } else {
+      golomb_bits = get_golomb_cost(level);
+      *diff += (r & (r - 1)) == 0 ? 1024 : 0;
+    }
+  }
+
+  return coeff_lps[base_range] + golomb_bits;
+}
+
+static INLINE int get_br_cost(tran_low_t level, const int *coeff_lps) {
+  const int base_range = AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
+  return coeff_lps[base_range] + get_golomb_cost(level);
+}
+
 static int get_coeff_cost(const tran_low_t qc, const int scan_idx,
                           const int is_eob, const TxbInfo *const txb_info,
                           const LV_MAP_COEFF_COST *const txb_costs,
@@ -331,8 +353,7 @@ static int get_coeff_cost(const tran_low_t qc, const int scan_idx,
     if (abs_qc > NUM_BASE_LEVELS) {
       const int ctx =
           get_br_ctx(txb_info->levels, pos, txb_info->bwl, tx_class);
-      cost += get_br_cost(abs_qc, ctx, txb_costs->lps_cost[ctx]);
-      cost += get_golomb_cost(abs_qc);
+      cost += get_br_cost(abs_qc, txb_costs->lps_cost[ctx]);
     }
   }
   return cost;
@@ -464,8 +485,6 @@ void av1_txb_init_levels_c(const tran_low_t *const coeff, const int width,
   const int stride = width + TX_PAD_HOR;
   uint8_t *ls = levels;
 
-  memset(levels - TX_PAD_TOP * stride, 0,
-         sizeof(*levels) * TX_PAD_TOP * stride);
   memset(levels + stride * height, 0,
          sizeof(*levels) * (TX_PAD_BOTTOM * stride + TX_PAD_END));
 
@@ -554,14 +573,15 @@ void av1_write_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
       break;
   }
 
-  if (k_eob_offset_bits[eob_pt] > 0) {
+  const int eob_offset_bits = k_eob_offset_bits[eob_pt];
+  if (eob_offset_bits > 0) {
     const int eob_ctx = eob_pt - 3;
-    int eob_shift = k_eob_offset_bits[eob_pt] - 1;
+    int eob_shift = eob_offset_bits - 1;
     int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
     aom_write_symbol(w, bit,
                      ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_ctx], 2);
-    for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) {
-      eob_shift = k_eob_offset_bits[eob_pt] - 1 - i;
+    for (int i = 1; i < eob_offset_bits; i++) {
+      eob_shift = eob_offset_bits - 1 - i;
       bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
       aom_write_bit(w, bit);
     }
@@ -588,12 +608,11 @@ void av1_write_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
       // level is above 1.
       const int base_range = level - 1 - NUM_BASE_LEVELS;
       const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class);
+      aom_cdf_prob *cdf =
+          ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx];
       for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
         const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
-        aom_write_symbol(
-            w, k,
-            ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx],
-            BR_CDF_SIZE);
+        aom_write_symbol(w, k, cdf, BR_CDF_SIZE);
         if (k < BR_CDF_SIZE - 1) break;
       }
     }
@@ -745,7 +764,8 @@ static AOM_FORCE_INLINE int warehouse_efficients_txb(
 
   av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_class, coeff_contexts);
 
-  const int(*lps_cost)[COEFF_BASE_RANGE + 1] = coeff_costs->lps_cost;
+  const int(*lps_cost)[COEFF_BASE_RANGE + 1 + COEFF_BASE_RANGE + 1] =
+      coeff_costs->lps_cost;
   int c = eob - 1;
   {
     const int pos = scan[c];
@@ -758,11 +778,8 @@ static AOM_FORCE_INLINE int warehouse_efficients_txb(
     if (v) {
       // sign bit cost
       if (level > NUM_BASE_LEVELS) {
-        const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
-        const int base_range =
-            AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
-        cost += lps_cost[ctx][base_range];
-        cost += get_golomb_cost(level);
+        const int ctx = get_br_ctx_eob(pos, bwl, tx_class);
+        cost += get_br_cost(level, lps_cost[ctx]);
       }
       if (c) {
         cost += av1_cost_literal(1);
@@ -774,7 +791,7 @@ static AOM_FORCE_INLINE int warehouse_efficients_txb(
       }
     }
   }
-  const int(*base_cost)[4] = coeff_costs->base_cost;
+  const int(*base_cost)[8] = coeff_costs->base_cost;
   for (c = eob - 2; c >= 1; --c) {
     const int pos = scan[c];
     const int coeff_ctx = coeff_contexts[pos];
@@ -786,10 +803,7 @@ static AOM_FORCE_INLINE int warehouse_efficients_txb(
       cost += av1_cost_literal(1);
       if (level > NUM_BASE_LEVELS) {
         const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
-        const int base_range =
-            AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
-        cost += lps_cost[ctx][base_range];
-        cost += get_golomb_cost(level);
+        cost += get_br_cost(level, lps_cost[ctx]);
       }
     }
     cost += cost0;
@@ -809,10 +823,7 @@ static AOM_FORCE_INLINE int warehouse_efficients_txb(
       cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign01];
       if (level > NUM_BASE_LEVELS) {
         const int ctx = get_br_ctx(levels, pos, bwl, tx_class);
-        const int base_range =
-            AOMMIN(level - 1 - NUM_BASE_LEVELS, COEFF_BASE_RANGE);
-        cost += lps_cost[ctx][base_range];
-        cost += get_golomb_cost(level);
+        cost += get_br_cost(level, lps_cost[ctx]);
       }
     }
   }
@@ -1014,13 +1025,7 @@ static int optimize_txb(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
   return update;
 }
 
-// These numbers are empirically obtained.
-static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
-  { 17, 13 },
-  { 16, 10 },
-};
-
-void hbt_init() {
+static void hbt_init() {
   hbt_hash_table =
       aom_malloc(sizeof(OptTxbQcoeff) * HBT_TABLE_SIZE * HBT_ARRAY_LENGTH);
   memset(hbt_hash_table, 0,
@@ -1032,11 +1037,11 @@ void hbt_init() {
 
 void hbt_destroy() { aom_free(hbt_hash_table); }
 
-int hbt_hash_miss(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
-                  TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
-                  const LV_MAP_EOB_COST *txb_eob_costs,
-                  const struct macroblock_plane *p, int block, int fast_mode,
-                  int *rate_cost) {
+static int hbt_hash_miss(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
+                         TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
+                         const LV_MAP_EOB_COST *txb_eob_costs,
+                         const struct macroblock_plane *p, int block,
+                         int fast_mode, int *rate_cost) {
   (void)fast_mode;
   const int16_t *scan = txb_info->scan_order->scan;
   int prev_eob = txb_info->eob;
@@ -1098,9 +1103,9 @@ int hbt_hash_miss(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
   return txb_info->eob;
 }
 
-int hbt_hash_hit(uint32_t hbt_table_index, int hbt_array_index,
-                 TxbInfo *txb_info, const struct macroblock_plane *p, int block,
-                 int *rate_cost) {
+static int hbt_hash_hit(uint32_t hbt_table_index, int hbt_array_index,
+                        TxbInfo *txb_info, const struct macroblock_plane *p,
+                        int block, int *rate_cost) {
   const int16_t *scan = txb_info->scan_order->scan;
   int new_eob = 0;
   int update = 0;
@@ -1140,11 +1145,12 @@ int hbt_hash_hit(uint32_t hbt_table_index, int hbt_array_index,
   return txb_info->eob;
 }
 
-int hbt_search_match(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
-                     TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
-                     const LV_MAP_EOB_COST *txb_eob_costs,
-                     const struct macroblock_plane *p, int block, int fast_mode,
-                     int *rate_cost) {
+static int hbt_search_match(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
+                            TxbInfo *txb_info,
+                            const LV_MAP_COEFF_COST *txb_costs,
+                            const LV_MAP_EOB_COST *txb_eob_costs,
+                            const struct macroblock_plane *p, int block,
+                            int fast_mode, int *rate_cost) {
   // Check for qcoeff match
   int hbt_array_index = hbt_qc_hash % HBT_ARRAY_LENGTH;
   int hbt_table_index = hbt_ctx_hash % HBT_TABLE_SIZE;
@@ -1163,10 +1169,11 @@ int hbt_search_match(uint32_t hbt_ctx_hash, uint32_t hbt_qc_hash,
   }
 }
 
-int hbt_create_hashes(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
-                      const LV_MAP_EOB_COST *txb_eob_costs,
-                      const struct macroblock_plane *p, int block,
-                      int fast_mode, int *rate_cost) {
+static int hbt_create_hashes(TxbInfo *txb_info,
+                             const LV_MAP_COEFF_COST *txb_costs,
+                             const LV_MAP_EOB_COST *txb_eob_costs,
+                             const struct macroblock_plane *p, int block,
+                             int fast_mode, int *rate_cost) {
   // Initialize hash table if needed.
   if (hbt_needs_init) {
     hbt_init();
@@ -1288,20 +1295,47 @@ int hbt_create_hashes(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
                           txb_eob_costs, p, block, fast_mode, rate_cost);
 }
 
-static AOM_FORCE_INLINE int get_coeff_cost_simple(
+static AOM_FORCE_INLINE int get_two_coeff_cost_simple(
     int ci, tran_low_t abs_qc, int coeff_ctx,
     const LV_MAP_COEFF_COST *txb_costs, int bwl, TX_CLASS tx_class,
-    const uint8_t *levels) {
+    const uint8_t *levels, int *cost_low) {
   // this simple version assumes the coeff's scan_idx is not DC (scan_idx != 0)
   // and not the last (scan_idx != eob - 1)
   assert(ci > 0);
   int cost = txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
+  int diff = 0;
+  if (abs_qc <= 3) diff = txb_costs->base_cost[coeff_ctx][abs_qc + 4];
   if (abs_qc) {
     cost += av1_cost_literal(1);
     if (abs_qc > NUM_BASE_LEVELS) {
       const int br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
-      cost += get_br_cost(abs_qc, br_ctx, txb_costs->lps_cost[br_ctx]);
-      cost += get_golomb_cost(abs_qc);
+      int brcost_diff = 0;
+      cost += get_br_cost_with_diff(abs_qc, txb_costs->lps_cost[br_ctx],
+                                    &brcost_diff);
+      diff += brcost_diff;
+    }
+  }
+  *cost_low = cost - diff;
+
+  return cost;
+}
+
+static INLINE int get_coeff_cost_eob(int ci, tran_low_t abs_qc, int sign,
+                                     int coeff_ctx, int dc_sign_ctx,
+                                     const LV_MAP_COEFF_COST *txb_costs,
+                                     int bwl, TX_CLASS tx_class) {
+  int cost = 0;
+  cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
+  if (abs_qc != 0) {
+    if (ci == 0) {
+      cost += txb_costs->dc_sign_cost[dc_sign_ctx][sign];
+    } else {
+      cost += av1_cost_literal(1);
+    }
+    if (abs_qc > NUM_BASE_LEVELS) {
+      int br_ctx;
+      br_ctx = get_br_ctx_eob(ci, bwl, tx_class);
+      cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
     }
   }
   return cost;
@@ -1326,9 +1360,12 @@ static INLINE int get_coeff_cost_general(int is_last, int ci, tran_low_t abs_qc,
       cost += av1_cost_literal(1);
     }
     if (abs_qc > NUM_BASE_LEVELS) {
-      const int br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
-      cost += get_br_cost(abs_qc, br_ctx, txb_costs->lps_cost[br_ctx]);
-      cost += get_golomb_cost(abs_qc);
+      int br_ctx;
+      if (is_last)
+        br_ctx = get_br_ctx_eob(ci, bwl, tx_class);
+      else
+        br_ctx = get_br_ctx(levels, ci, bwl, tx_class);
+      cost += get_br_cost(abs_qc, txb_costs->lps_cost[br_ctx]);
     }
   }
   return cost;
@@ -1372,13 +1409,23 @@ static INLINE void update_coeff_general(
     const int64_t rd = RDCOST(rdmult, rate, dist);
 
     tran_low_t qc_low, dqc_low;
-    get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
-    const tran_low_t abs_qc_low = abs_qc - 1;
-    const int64_t dist_low = get_coeff_dist(tqc, dqc_low, shift);
-    const int rate_low =
-        get_coeff_cost_general(is_last, ci, abs_qc_low, sign, coeff_ctx,
-                               dc_sign_ctx, txb_costs, bwl, tx_class, levels);
-    const int64_t rd_low = RDCOST(rdmult, rate_low, dist_low);
+    tran_low_t abs_qc_low;
+    int64_t dist_low, rd_low;
+    int rate_low;
+    if (abs_qc == 1) {
+      abs_qc_low = qc_low = dqc_low = 0;
+      dist_low = dist0;
+      rate_low = txb_costs->base_cost[coeff_ctx][0];
+    } else {
+      get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
+      abs_qc_low = abs_qc - 1;
+      dist_low = get_coeff_dist(tqc, dqc_low, shift);
+      rate_low =
+          get_coeff_cost_general(is_last, ci, abs_qc_low, sign, coeff_ctx,
+                                 dc_sign_ctx, txb_costs, bwl, tx_class, levels);
+    }
+
+    rd_low = RDCOST(rdmult, rate_low, dist_low);
     if (rd_low < rd) {
       qcoeff[ci] = qc_low;
       dqcoeff[ci] = dqc_low;
@@ -1412,28 +1459,28 @@ static AOM_FORCE_INLINE void update_coeff_simple(
     *accu_rate += txb_costs->base_cost[coeff_ctx][0];
   } else {
     const tran_low_t abs_qc = abs(qc);
-    const tran_low_t tqc = tcoeff[ci];
-    const tran_low_t dqc = dqcoeff[ci];
-    const int rate = get_coeff_cost_simple(ci, abs_qc, coeff_ctx, txb_costs,
-                                           bwl, tx_class, levels);
-    if (abs(dqc) < abs(tqc)) {
+    const tran_low_t abs_tqc = abs(tcoeff[ci]);
+    const tran_low_t abs_dqc = abs(dqcoeff[ci]);
+    int rate_low = 0;
+    const int rate = get_two_coeff_cost_simple(
+        ci, abs_qc, coeff_ctx, txb_costs, bwl, tx_class, levels, &rate_low);
+    if (abs_dqc < abs_tqc) {
       *accu_rate += rate;
       return;
     }
-    const int64_t dist = get_coeff_dist(tqc, dqc, shift);
+
+    const int64_t dist = get_coeff_dist(abs_tqc, abs_dqc, shift);
     const int64_t rd = RDCOST(rdmult, rate, dist);
 
-    const int sign = (qc < 0) ? 1 : 0;
-    tran_low_t qc_low, dqc_low;
-    get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
     const tran_low_t abs_qc_low = abs_qc - 1;
-    const int64_t dist_low = get_coeff_dist(tqc, dqc_low, shift);
-    const int rate_low = get_coeff_cost_simple(
-        ci, abs_qc_low, coeff_ctx, txb_costs, bwl, tx_class, levels);
+    const tran_low_t abs_dqc_low = (abs_qc_low * dqv) >> shift;
+    const int64_t dist_low = get_coeff_dist(abs_tqc, abs_dqc_low, shift);
     const int64_t rd_low = RDCOST(rdmult, rate_low, dist_low);
+
     if (rd_low < rd) {
-      qcoeff[ci] = qc_low;
-      dqcoeff[ci] = dqc_low;
+      const int sign = (qc < 0) ? 1 : 0;
+      qcoeff[ci] = (-sign ^ abs_qc_low) + sign;
+      dqcoeff[ci] = (-sign ^ abs_dqc_low) + sign;
       levels[get_padded_idx(ci, bwl)] = AOMMIN(abs_qc_low, INT8_MAX);
       *accu_rate += rate_low;
     } else {
@@ -1471,40 +1518,42 @@ static AOM_FORCE_INLINE void update_coeff_eob(
     int64_t rd = RDCOST(rdmult, *accu_rate + rate, *accu_dist + dist);
 
     tran_low_t qc_low, dqc_low;
-    get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
-    const tran_low_t abs_qc_low = abs_qc - 1;
-    const int64_t dist_low = get_coeff_dist(tqc, dqc_low, shift) - dist0;
-    const int rate_low =
-        get_coeff_cost_general(0, ci, abs_qc_low, sign, coeff_ctx, dc_sign_ctx,
-                               txb_costs, bwl, tx_class, levels);
-    const int64_t rd_low =
-        RDCOST(rdmult, *accu_rate + rate_low, *accu_dist + dist_low);
+    tran_low_t abs_qc_low;
+    int64_t dist_low, rd_low;
+    int rate_low;
+    if (abs_qc == 1) {
+      abs_qc_low = 0;
+      dqc_low = qc_low = 0;
+      dist_low = 0;
+      rate_low = txb_costs->base_cost[coeff_ctx][0];
+      rd_low = RDCOST(rdmult, *accu_rate + rate_low, *accu_dist);
+    } else {
+      get_qc_dqc_low(abs_qc, sign, dqv, shift, &qc_low, &dqc_low);
+      abs_qc_low = abs_qc - 1;
+      dist_low = get_coeff_dist(tqc, dqc_low, shift) - dist0;
+      rate_low =
+          get_coeff_cost_general(0, ci, abs_qc_low, sign, coeff_ctx,
+                                 dc_sign_ctx, txb_costs, bwl, tx_class, levels);
+      rd_low = RDCOST(rdmult, *accu_rate + rate_low, *accu_dist + dist_low);
+    }
 
     int lower_level_new_eob = 0;
     const int new_eob = si + 1;
-    uint8_t tmp_levels[3];
-    for (int ni = 0; ni < *nz_num; ++ni) {
-      const int last_ci = nz_ci[ni];
-      tmp_levels[ni] = levels[get_padded_idx(last_ci, bwl)];
-      levels[get_padded_idx(last_ci, bwl)] = 0;
-    }
-
-    const int coeff_ctx_new_eob = get_lower_levels_ctx_general(
-        1, si, bwl, height, levels, ci, tx_size, tx_class);
+    const int coeff_ctx_new_eob = get_lower_levels_ctx_eob(bwl, height, si);
     const int new_eob_cost =
         get_eob_cost(new_eob, txb_eob_costs, txb_costs, tx_class);
     int rate_coeff_eob =
-        new_eob_cost + get_coeff_cost_general(1, ci, abs_qc, sign,
-                                              coeff_ctx_new_eob, dc_sign_ctx,
-                                              txb_costs, bwl, tx_class, levels);
+        new_eob_cost + get_coeff_cost_eob(ci, abs_qc, sign, coeff_ctx_new_eob,
+                                          dc_sign_ctx, txb_costs, bwl,
+                                          tx_class);
     int64_t dist_new_eob = dist;
     int64_t rd_new_eob = RDCOST(rdmult, rate_coeff_eob, dist_new_eob);
 
     if (abs_qc_low > 0) {
       const int rate_coeff_eob_low =
-          new_eob_cost +
-          get_coeff_cost_general(1, ci, abs_qc_low, sign, coeff_ctx_new_eob,
-                                 dc_sign_ctx, txb_costs, bwl, tx_class, levels);
+          new_eob_cost + get_coeff_cost_eob(ci, abs_qc_low, sign,
+                                            coeff_ctx_new_eob, dc_sign_ctx,
+                                            txb_costs, bwl, tx_class);
       const int64_t dist_new_eob_low = dist_low;
       const int64_t rd_new_eob_low =
           RDCOST(rdmult, rate_coeff_eob_low, dist_new_eob_low);
@@ -1526,7 +1575,7 @@ static AOM_FORCE_INLINE void update_coeff_eob(
     if (sharpness == 0 && rd_new_eob < rd) {
       for (int ni = 0; ni < *nz_num; ++ni) {
         int last_ci = nz_ci[ni];
-        // levels[get_padded_idx(last_ci, bwl)] = 0;
+        levels[get_padded_idx(last_ci, bwl)] = 0;
         qcoeff[last_ci] = 0;
         dqcoeff[last_ci] = 0;
       }
@@ -1536,10 +1585,6 @@ static AOM_FORCE_INLINE void update_coeff_eob(
       *accu_dist = dist_new_eob;
       lower_level = lower_level_new_eob;
     } else {
-      for (int ni = 0; ni < *nz_num; ++ni) {
-        const int last_ci = nz_ci[ni];
-        levels[get_padded_idx(last_ci, bwl)] = tmp_levels[ni];
-      }
       *accu_rate += rate;
       *accu_dist += dist;
     }
@@ -1619,14 +1664,14 @@ int av1_optimize_txb_new(const struct AV1_COMP *cpi, MACROBLOCK *x, int plane,
 
   uint8_t levels_buf[TX_PAD_2D];
   uint8_t *const levels = set_levels(levels_buf, width);
+  int eob = p->eobs[block];
 
-  av1_txb_init_levels(qcoeff, width, height, levels);
+  if (eob > 1) av1_txb_init_levels(qcoeff, width, height, levels);
 
   // TODO(angirbird): check iqmatrix
 
   const int non_skip_cost = txb_costs->txb_skip_cost[txb_ctx->txb_skip_ctx][0];
   const int skip_cost = txb_costs->txb_skip_cost[txb_ctx->txb_skip_ctx][1];
-  int eob = p->eobs[block];
   const int eob_cost = get_eob_cost(eob, txb_eob_costs, txb_costs, tx_class);
   int accu_rate = eob_cost;
   int64_t accu_dist = 0;
@@ -1646,11 +1691,10 @@ int av1_optimize_txb_new(const struct AV1_COMP *cpi, MACROBLOCK *x, int plane,
     --si;
   } else {
     assert(abs_qc == 1);
-    const int coeff_ctx = get_lower_levels_ctx_general(
-        1, si, bwl, height, levels, ci, tx_size, tx_class);
-    accu_rate += get_coeff_cost_general(1, ci, abs_qc, sign, coeff_ctx,
-                                        txb_ctx->dc_sign_ctx, txb_costs, bwl,
-                                        tx_class, levels);
+    const int coeff_ctx = get_lower_levels_ctx_eob(bwl, height, si);
+    accu_rate +=
+        get_coeff_cost_eob(ci, abs_qc, sign, coeff_ctx, txb_ctx->dc_sign_ctx,
+                           txb_costs, bwl, tx_class);
     const tran_low_t tqc = tcoeff[ci];
     const tran_low_t dqc = dqcoeff[ci];
     const int64_t dist = get_coeff_dist(tqc, dqc, shift);
@@ -1767,10 +1811,9 @@ int av1_optimize_txb(const struct AV1_COMP *cpi, MACROBLOCK *x, int plane,
   assert(width == (1 << bwl));
   const int tx_type_cost = get_tx_type_cost(cm, x, xd, plane, tx_size, tx_type);
   TxbInfo txb_info = {
-    qcoeff,   levels,       dqcoeff,    tcoeff,  dequant, shift,
-    tx_size,  txs_ctx,      tx_type,    bwl,     width,   height,
-    eob,      seg_eob,      scan_order, txb_ctx, rdmult,  &cm->coeff_ctx_table,
-    iqmatrix, tx_type_cost,
+    qcoeff,     levels,  dqcoeff, tcoeff,   dequant,      shift, tx_size,
+    txs_ctx,    tx_type, bwl,     width,    height,       eob,   seg_eob,
+    scan_order, txb_ctx, rdmult,  iqmatrix, tx_type_cost,
   };
 
   // Hash based trellis (hbt) speed feature: avoid expensive optimize_txb calls
diff --git a/third_party/aom/av1/encoder/encodetxb.h b/third_party/aom/av1/encoder/encodetxb.h
index 40ae343b0d..bfc49dea8c 100644
--- a/third_party/aom/av1/encoder/encodetxb.h
+++ b/third_party/aom/av1/encoder/encodetxb.h
@@ -42,7 +42,6 @@ typedef struct TxbInfo {
   const SCAN_ORDER *scan_order;
   TXB_CTX *txb_ctx;
   int64_t rdmult;
-  const LV_MAP_CTX_TABLE *coeff_ctx_table;
   const qm_val_t *iqmatrix;
   int tx_type_cost;
 } TxbInfo;
@@ -80,6 +79,13 @@ int av1_optimize_txb_new(const struct AV1_COMP *cpi, MACROBLOCK *x, int plane,
                          int block, TX_SIZE tx_size, TX_TYPE tx_type,
                          const TXB_CTX *const txb_ctx, int *rate_cost,
                          int sharpness);
+
+// These numbers are empirically obtained.
+static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
+  { 17, 13 },
+  { 16, 10 },
+};
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/third_party/aom/av1/encoder/ethread.c b/third_party/aom/av1/encoder/ethread.c
index e8ac30bb52..77fb811b14 100644
--- a/third_party/aom/av1/encoder/ethread.c
+++ b/third_party/aom/av1/encoder/ethread.c
@@ -9,9 +9,11 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include "av1/encoder/av1_multi_thread.h"
 #include "av1/encoder/encodeframe.h"
 #include "av1/encoder/encoder.h"
 #include "av1/encoder/ethread.h"
+#include "av1/encoder/rdopt.h"
 #include "aom_dsp/aom_dsp_common.h"
 
 static void accumulate_rd_opt(ThreadData *td, ThreadData *td_t) {
@@ -27,6 +29,318 @@ static void accumulate_rd_opt(ThreadData *td, ThreadData *td_t) {
   td->rd_counts.skip_mode_used_flag |= td_t->rd_counts.skip_mode_used_flag;
 }
 
+static void update_delta_lf_for_row_mt(AV1_COMP *cpi) {
+  AV1_COMMON *cm = &cpi->common;
+  MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
+  const int mib_size = cm->seq_params.mib_size;
+  const int frame_lf_count =
+      av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
+  for (int row = 0; row < cm->tile_rows; row++) {
+    for (int col = 0; col < cm->tile_cols; col++) {
+      TileDataEnc *tile_data = &cpi->tile_data[row * cm->tile_cols + col];
+      const TileInfo *const tile_info = &tile_data->tile_info;
+      for (int mi_row = tile_info->mi_row_start; mi_row < tile_info->mi_row_end;
+           mi_row += mib_size) {
+        if (mi_row == tile_info->mi_row_start)
+          av1_reset_loop_filter_delta(xd, av1_num_planes(cm));
+        for (int mi_col = tile_info->mi_col_start;
+             mi_col < tile_info->mi_col_end; mi_col += mib_size) {
+          const int idx_str = cm->mi_stride * mi_row + mi_col;
+          MB_MODE_INFO **mi = cm->mi_grid_visible + idx_str;
+          MB_MODE_INFO *mbmi = mi[0];
+          if (mbmi->skip == 1 && (mbmi->sb_type == cm->seq_params.sb_size)) {
+            for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id)
+              mbmi->delta_lf[lf_id] = xd->delta_lf[lf_id];
+            mbmi->delta_lf_from_base = xd->delta_lf_from_base;
+          } else {
+            if (cm->delta_q_info.delta_lf_multi) {
+              for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id)
+                xd->delta_lf[lf_id] = mbmi->delta_lf[lf_id];
+            } else {
+              xd->delta_lf_from_base = mbmi->delta_lf_from_base;
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+void av1_row_mt_sync_read_dummy(struct AV1RowMTSyncData *const row_mt_sync,
+                                int r, int c) {
+  (void)row_mt_sync;
+  (void)r;
+  (void)c;
+  return;
+}
+
+void av1_row_mt_sync_write_dummy(struct AV1RowMTSyncData *const row_mt_sync,
+                                 int r, int c, const int cols) {
+  (void)row_mt_sync;
+  (void)r;
+  (void)c;
+  (void)cols;
+  return;
+}
+
+void av1_row_mt_sync_read(AV1RowMTSync *const row_mt_sync, int r, int c) {
+#if CONFIG_MULTITHREAD
+  const int nsync = row_mt_sync->sync_range;
+
+  if (r) {
+    pthread_mutex_t *const mutex = &row_mt_sync->mutex_[r - 1];
+    pthread_mutex_lock(mutex);
+
+    while (c > row_mt_sync->cur_col[r - 1] - nsync) {
+      pthread_cond_wait(&row_mt_sync->cond_[r - 1], mutex);
+    }
+    pthread_mutex_unlock(mutex);
+  }
+#else
+  (void)row_mt_sync;
+  (void)r;
+  (void)c;
+#endif  // CONFIG_MULTITHREAD
+}
+
+void av1_row_mt_sync_write(AV1RowMTSync *const row_mt_sync, int r, int c,
+                           const int cols) {
+#if CONFIG_MULTITHREAD
+  const int nsync = row_mt_sync->sync_range;
+  int cur;
+  // Only signal when there are enough encoded blocks for next row to run.
+  int sig = 1;
+
+  if (c < cols - 1) {
+    cur = c;
+    if (c % nsync) sig = 0;
+  } else {
+    cur = cols + nsync;
+  }
+
+  if (sig) {
+    pthread_mutex_lock(&row_mt_sync->mutex_[r]);
+
+    row_mt_sync->cur_col[r] = cur;
+
+    pthread_cond_signal(&row_mt_sync->cond_[r]);
+    pthread_mutex_unlock(&row_mt_sync->mutex_[r]);
+  }
+#else
+  (void)row_mt_sync;
+  (void)r;
+  (void)c;
+  (void)cols;
+#endif  // CONFIG_MULTITHREAD
+}
+
+// Allocate memory for row synchronization
+void av1_row_mt_sync_mem_alloc(AV1RowMTSync *row_mt_sync, AV1_COMMON *cm,
+                               int rows) {
+  row_mt_sync->rows = rows;
+#if CONFIG_MULTITHREAD
+  {
+    int i;
+
+    CHECK_MEM_ERROR(cm, row_mt_sync->mutex_,
+                    aom_malloc(sizeof(*row_mt_sync->mutex_) * rows));
+    if (row_mt_sync->mutex_) {
+      for (i = 0; i < rows; ++i) {
+        pthread_mutex_init(&row_mt_sync->mutex_[i], NULL);
+      }
+    }
+
+    CHECK_MEM_ERROR(cm, row_mt_sync->cond_,
+                    aom_malloc(sizeof(*row_mt_sync->cond_) * rows));
+    if (row_mt_sync->cond_) {
+      for (i = 0; i < rows; ++i) {
+        pthread_cond_init(&row_mt_sync->cond_[i], NULL);
+      }
+    }
+  }
+#endif  // CONFIG_MULTITHREAD
+
+  CHECK_MEM_ERROR(cm, row_mt_sync->cur_col,
+                  aom_malloc(sizeof(*row_mt_sync->cur_col) * rows));
+
+  // Set up nsync.
+  row_mt_sync->sync_range = 1;
+}
+
+// Deallocate row based multi-threading synchronization related mutex and data
+void av1_row_mt_sync_mem_dealloc(AV1RowMTSync *row_mt_sync) {
+  if (row_mt_sync != NULL) {
+#if CONFIG_MULTITHREAD
+    int i;
+
+    if (row_mt_sync->mutex_ != NULL) {
+      for (i = 0; i < row_mt_sync->rows; ++i) {
+        pthread_mutex_destroy(&row_mt_sync->mutex_[i]);
+      }
+      aom_free(row_mt_sync->mutex_);
+    }
+    if (row_mt_sync->cond_ != NULL) {
+      for (i = 0; i < row_mt_sync->rows; ++i) {
+        pthread_cond_destroy(&row_mt_sync->cond_[i]);
+      }
+      aom_free(row_mt_sync->cond_);
+    }
+#endif  // CONFIG_MULTITHREAD
+    aom_free(row_mt_sync->cur_col);
+    // clear the structure as the source of this call may be dynamic change
+    // in tiles in which case this call will be followed by an _alloc()
+    // which may fail.
+    av1_zero(*row_mt_sync);
+  }
+}
+
+static void assign_tile_to_thread(MultiThreadHandle *multi_thread_ctxt,
+                                  int num_tiles, int num_workers) {
+  int tile_id = 0;
+  int i;
+
+  for (i = 0; i < num_workers; i++) {
+    multi_thread_ctxt->thread_id_to_tile_id[i] = tile_id++;
+    if (tile_id == num_tiles) tile_id = 0;
+  }
+}
+
+static int get_next_job(AV1_COMP *const cpi, int *current_mi_row,
+                        int cur_tile_id) {
+  AV1_COMMON *const cm = &cpi->common;
+  TileDataEnc *const this_tile = &cpi->tile_data[cur_tile_id];
+  AV1RowMTInfo *row_mt_info = &this_tile->row_mt_info;
+
+  if (row_mt_info->current_mi_row < this_tile->tile_info.mi_row_end) {
+    *current_mi_row = row_mt_info->current_mi_row;
+    row_mt_info->num_threads_working++;
+    row_mt_info->current_mi_row += cm->seq_params.mib_size;
+    return 1;
+  }
+  return 0;
+}
+
+static void switch_tile_and_get_next_job(AV1_COMP *const cpi, int *cur_tile_id,
+                                         int *current_mi_row,
+                                         int *end_of_frame) {
+  AV1_COMMON *const cm = &cpi->common;
+  const int tile_cols = cm->tile_cols;
+  const int tile_rows = cm->tile_rows;
+
+  int tile_id = -1;  // Stores the tile ID with minimum proc done
+  int max_mis_to_encode = 0;
+  int min_num_threads_working = INT_MAX;
+
+  for (int tile_row = 0; tile_row < tile_rows; tile_row++) {
+    for (int tile_col = 0; tile_col < tile_cols; tile_col++) {
+      int tile_index = tile_row * tile_cols + tile_col;
+      TileDataEnc *this_tile = &cpi->tile_data[tile_index];
+      AV1RowMTInfo *row_mt_info = &this_tile->row_mt_info;
+      int num_mis_to_encode =
+          this_tile->tile_info.mi_row_end - row_mt_info->current_mi_row;
+
+      // Tile to be processed by this thread is selected on the basis of
+      // availability of jobs:
+      // 1) If jobs are available, tile to be processed is chosen on the
+      // basis of minimum number of threads working for that tile. If two or
+      // more tiles have same number of threads working for them, then the tile
+      // with maximum number of jobs available will be chosen.
+      // 2) If no jobs are available, then end_of_frame is reached.
+      if (num_mis_to_encode > 0) {
+        int num_threads_working = row_mt_info->num_threads_working;
+        if (num_threads_working < min_num_threads_working) {
+          min_num_threads_working = num_threads_working;
+          max_mis_to_encode = 0;
+        }
+        if (num_threads_working == min_num_threads_working &&
+            num_mis_to_encode > max_mis_to_encode) {
+          tile_id = tile_index;
+          max_mis_to_encode = num_mis_to_encode;
+        }
+      }
+    }
+  }
+  if (tile_id == -1) {
+    *end_of_frame = 1;
+  } else {
+    // Update the cur ID to the next tile ID that will be processed,
+    // which will be the least processed tile
+    *cur_tile_id = tile_id;
+    get_next_job(cpi, current_mi_row, *cur_tile_id);
+  }
+}
+
+static int enc_row_mt_worker_hook(void *arg1, void *unused) {
+  EncWorkerData *const thread_data = (EncWorkerData *)arg1;
+  AV1_COMP *const cpi = thread_data->cpi;
+  AV1_COMMON *const cm = &cpi->common;
+
+  MultiThreadHandle *multi_thread_ctxt = &cpi->multi_thread_ctxt;
+  int thread_id = thread_data->thread_id;
+  int cur_tile_id = multi_thread_ctxt->thread_id_to_tile_id[thread_id];
+  (void)unused;
+
+  assert(cur_tile_id != -1);
+
+  int end_of_frame = 0;
+  while (1) {
+    int current_mi_row = -1;
+#if CONFIG_MULTITHREAD
+    pthread_mutex_lock(cpi->row_mt_mutex_);
+#endif
+    if (!get_next_job(cpi, &current_mi_row, cur_tile_id)) {
+      // No jobs are available for the current tile. Query for the status of
+      // other tiles and get the next job if available
+      switch_tile_and_get_next_job(cpi, &cur_tile_id, &current_mi_row,
+                                   &end_of_frame);
+    }
+#if CONFIG_MULTITHREAD
+    pthread_mutex_unlock(cpi->row_mt_mutex_);
+#endif
+    if (end_of_frame == 1) break;
+
+    TileDataEnc *const this_tile = &cpi->tile_data[cur_tile_id];
+    int tile_row = this_tile->tile_info.tile_row;
+    int tile_col = this_tile->tile_info.tile_col;
+
+    assert(current_mi_row != -1 &&
+           current_mi_row <= this_tile->tile_info.mi_row_end);
+
+    ThreadData *td = thread_data->td;
+
+    td->mb.e_mbd.tile_ctx = td->tctx;
+    td->mb.tile_pb_ctx = &this_tile->tctx;
+    if (this_tile->allow_update_cdf) {
+      td->mb.row_ctx = this_tile->row_ctx;
+      if (current_mi_row == this_tile->tile_info.mi_row_start)
+        memcpy(td->mb.e_mbd.tile_ctx, &this_tile->tctx, sizeof(FRAME_CONTEXT));
+    } else {
+      memcpy(td->mb.e_mbd.tile_ctx, &this_tile->tctx, sizeof(FRAME_CONTEXT));
+    }
+
+    av1_init_above_context(cm, &td->mb.e_mbd, tile_row);
+
+    // Disable exhaustive search speed features for row based multi-threading of
+    // encoder.
+    td->mb.m_search_count_ptr = NULL;
+    td->mb.ex_search_count_ptr = NULL;
+
+    cfl_init(&td->mb.e_mbd.cfl, &cm->seq_params);
+    av1_crc32c_calculator_init(&td->mb.mb_rd_record.crc_calculator);
+
+    av1_encode_sb_row(cpi, td, tile_row, tile_col, current_mi_row);
+#if CONFIG_MULTITHREAD
+    pthread_mutex_lock(cpi->row_mt_mutex_);
+#endif
+    this_tile->row_mt_info.num_threads_working--;
+#if CONFIG_MULTITHREAD
+    pthread_mutex_unlock(cpi->row_mt_mutex_);
+#endif
+  }
+
+  return 1;
+}
+
 static int enc_worker_hook(void *arg1, void *unused) {
   EncWorkerData *const thread_data = (EncWorkerData *)arg1;
   AV1_COMP *const cpi = thread_data->cpi;
@@ -42,6 +356,10 @@ static int enc_worker_hook(void *arg1, void *unused) {
     int tile_row = t / tile_cols;
     int tile_col = t % tile_cols;
 
+    TileDataEnc *const this_tile =
+        &cpi->tile_data[tile_row * cm->tile_cols + tile_col];
+    thread_data->td->mb.e_mbd.tile_ctx = &this_tile->tctx;
+    thread_data->td->mb.tile_pb_ctx = &this_tile->tctx;
     av1_encode_tile(cpi, thread_data->td, tile_row, tile_col);
   }
 
@@ -58,16 +376,28 @@ static void create_enc_workers(AV1_COMP *cpi, int num_workers) {
   CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
                   aom_calloc(num_workers, sizeof(*cpi->tile_thr_data)));
 
-  for (int i = 0; i < num_workers; i++) {
+#if CONFIG_MULTITHREAD
+  if (cpi->row_mt == 1) {
+    if (cpi->row_mt_mutex_ == NULL) {
+      CHECK_MEM_ERROR(cm, cpi->row_mt_mutex_,
+                      aom_malloc(sizeof(*(cpi->row_mt_mutex_))));
+      if (cpi->row_mt_mutex_) pthread_mutex_init(cpi->row_mt_mutex_, NULL);
+    }
+  }
+#endif
+
+  for (int i = num_workers - 1; i >= 0; i--) {
     AVxWorker *const worker = &cpi->workers[i];
     EncWorkerData *const thread_data = &cpi->tile_thr_data[i];
 
     ++cpi->num_workers;
     winterface->init(worker);
+    worker->thread_name = "aom enc worker";
 
     thread_data->cpi = cpi;
+    thread_data->thread_id = i;
 
-    if (i < num_workers - 1) {
+    if (i > 0) {
       // Allocate thread data.
       CHECK_MEM_ERROR(cm, thread_data->td,
                       aom_memalign(32, sizeof(*thread_data->td)));
@@ -91,6 +421,10 @@ static void create_enc_workers(AV1_COMP *cpi, int num_workers) {
           (int32_t *)aom_memalign(
               16, MAX_SB_SQUARE * sizeof(*thread_data->td->wsrc_buf)));
 
+      CHECK_MEM_ERROR(cm, thread_data->td->inter_modes_info,
+                      (InterModesInfo *)aom_malloc(
+                          sizeof(*thread_data->td->inter_modes_info)));
+
       for (int x = 0; x < 2; x++)
         for (int y = 0; y < 2; y++)
           CHECK_MEM_ERROR(
@@ -119,7 +453,7 @@ static void create_enc_workers(AV1_COMP *cpi, int num_workers) {
       for (int j = 0; j < 2; ++j) {
         CHECK_MEM_ERROR(
             cm, thread_data->td->tmp_obmc_bufs[j],
-            aom_memalign(16, 2 * MAX_MB_PLANE * MAX_SB_SQUARE *
+            aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE *
                                  sizeof(*thread_data->td->tmp_obmc_bufs[j])));
       }
 
@@ -131,6 +465,10 @@ static void create_enc_workers(AV1_COMP *cpi, int num_workers) {
       // Main thread acts as a worker and uses the thread data in cpi.
       thread_data->td = &cpi->td;
     }
+    if (cpi->row_mt == 1)
+      CHECK_MEM_ERROR(
+          cm, thread_data->td->tctx,
+          (FRAME_CONTEXT *)aom_memalign(16, sizeof(*thread_data->td->tctx)));
     winterface->sync(worker);
   }
 }
@@ -138,14 +476,14 @@ static void create_enc_workers(AV1_COMP *cpi, int num_workers) {
 static void launch_enc_workers(AV1_COMP *cpi, int num_workers) {
   const AVxWorkerInterface *const winterface = aom_get_worker_interface();
   // Encode a frame
-  for (int i = 0; i < num_workers; i++) {
+  for (int i = num_workers - 1; i >= 0; i--) {
     AVxWorker *const worker = &cpi->workers[i];
     EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
 
     // Set the starting tile for each thread.
     thread_data->start = i;
 
-    if (i == cpi->num_workers - 1)
+    if (i == 0)
       winterface->execute(worker);
     else
       winterface->launch(worker);
@@ -154,31 +492,39 @@ static void launch_enc_workers(AV1_COMP *cpi, int num_workers) {
 
 static void sync_enc_workers(AV1_COMP *cpi, int num_workers) {
   const AVxWorkerInterface *const winterface = aom_get_worker_interface();
+  int had_error = 0;
 
   // Encoding ends.
-  for (int i = 0; i < num_workers; i++) {
+  for (int i = num_workers - 1; i >= 0; i--) {
     AVxWorker *const worker = &cpi->workers[i];
-    winterface->sync(worker);
+    had_error |= !winterface->sync(worker);
   }
+
+  if (had_error)
+    aom_internal_error(&cpi->common.error, AOM_CODEC_ERROR,
+                       "Failed to encode tile data");
 }
 
 static void accumulate_counters_enc_workers(AV1_COMP *cpi, int num_workers) {
-  for (int i = 0; i < num_workers; i++) {
+  for (int i = num_workers - 1; i >= 0; i--) {
     AVxWorker *const worker = &cpi->workers[i];
     EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
-    cpi->intrabc_used |= thread_data->td->intrabc_used_this_tile;
+    cpi->intrabc_used |= thread_data->td->intrabc_used;
     // Accumulate counters.
-    if (i < cpi->num_workers - 1) {
+    if (i > 0) {
       av1_accumulate_frame_counts(&cpi->counts, thread_data->td->counts);
       accumulate_rd_opt(&cpi->td, thread_data->td);
       cpi->td.mb.txb_split_count += thread_data->td->mb.txb_split_count;
+#if CONFIG_SPEED_STATS
+      cpi->td.mb.tx_search_count += thread_data->td->mb.tx_search_count;
+#endif  // CONFIG_SPEED_STATS
     }
   }
 }
 
 static void prepare_enc_workers(AV1_COMP *cpi, AVxWorkerHook hook,
                                 int num_workers) {
-  for (int i = 0; i < num_workers; i++) {
+  for (int i = num_workers - 1; i >= 0; i--) {
     AVxWorker *const worker = &cpi->workers[i];
     EncWorkerData *const thread_data = &cpi->tile_thr_data[i];
 
@@ -186,6 +532,8 @@ static void prepare_enc_workers(AV1_COMP *cpi, AVxWorkerHook hook,
     worker->data1 = thread_data;
     worker->data2 = NULL;
 
+    thread_data->td->intrabc_used = 0;
+
     // Before encoding a frame, copy the thread data from cpi.
     if (thread_data->td != &cpi->td) {
       thread_data->td->mb = cpi->td.mb;
@@ -193,6 +541,8 @@ static void prepare_enc_workers(AV1_COMP *cpi, AVxWorkerHook hook,
       thread_data->td->mb.above_pred_buf = thread_data->td->above_pred_buf;
       thread_data->td->mb.left_pred_buf = thread_data->td->left_pred_buf;
       thread_data->td->mb.wsrc_buf = thread_data->td->wsrc_buf;
+
+      thread_data->td->mb.inter_modes_info = thread_data->td->inter_modes_info;
       for (int x = 0; x < 2; x++) {
         for (int y = 0; y < 2; y++) {
           memcpy(thread_data->td->hash_value_buffer[x][y],
@@ -209,7 +559,7 @@ static void prepare_enc_workers(AV1_COMP *cpi, AVxWorkerHook hook,
       memcpy(thread_data->td->counts, &cpi->counts, sizeof(cpi->counts));
     }
 
-    if (i < num_workers - 1) {
+    if (i > 0) {
       thread_data->td->mb.palette_buffer = thread_data->td->palette_buffer;
       thread_data->td->mb.tmp_conv_dst = thread_data->td->tmp_conv_dst;
       for (int j = 0; j < 2; ++j) {
@@ -259,3 +609,74 @@ void av1_accumulate_frame_counts(FRAME_COUNTS *acc_counts,
 
   for (unsigned int i = 0; i < n_counts; i++) acc[i] += cnt[i];
 }
+
+void av1_encode_tiles_row_mt(AV1_COMP *cpi) {
+  AV1_COMMON *const cm = &cpi->common;
+  const int tile_cols = cm->tile_cols;
+  const int tile_rows = cm->tile_rows;
+  MultiThreadHandle *multi_thread_ctxt = &cpi->multi_thread_ctxt;
+  int num_workers = 0;
+  int total_num_sb_rows = 0;
+  int max_sb_rows = 0;
+
+  if (cpi->tile_data == NULL || cpi->allocated_tiles < tile_cols * tile_rows) {
+    av1_row_mt_mem_dealloc(cpi);
+    av1_alloc_tile_data(cpi);
+  }
+
+  av1_init_tile_data(cpi);
+
+  for (int row = 0; row < tile_rows; row++) {
+    for (int col = 0; col < tile_cols; col++) {
+      TileDataEnc *tile_data = &cpi->tile_data[row * cm->tile_cols + col];
+      int num_sb_rows_in_tile =
+          av1_get_sb_rows_in_tile(cm, tile_data->tile_info);
+      total_num_sb_rows += num_sb_rows_in_tile;
+      max_sb_rows = AOMMAX(max_sb_rows, num_sb_rows_in_tile);
+    }
+  }
+  num_workers = AOMMIN(cpi->oxcf.max_threads, total_num_sb_rows);
+
+  if (multi_thread_ctxt->allocated_tile_cols != tile_cols ||
+      multi_thread_ctxt->allocated_tile_rows != tile_rows ||
+      multi_thread_ctxt->allocated_sb_rows != max_sb_rows) {
+    av1_row_mt_mem_dealloc(cpi);
+    av1_row_mt_mem_alloc(cpi, max_sb_rows);
+  }
+
+  memset(multi_thread_ctxt->thread_id_to_tile_id, -1,
+         sizeof(*multi_thread_ctxt->thread_id_to_tile_id) * MAX_NUM_THREADS);
+
+  for (int tile_row = 0; tile_row < tile_rows; tile_row++) {
+    for (int tile_col = 0; tile_col < tile_cols; tile_col++) {
+      int tile_id = tile_row * tile_cols + tile_col;
+      TileDataEnc *this_tile = &cpi->tile_data[tile_id];
+
+      // Initialize cur_col to -1 for all rows.
+      memset(this_tile->row_mt_sync.cur_col, -1,
+             sizeof(*this_tile->row_mt_sync.cur_col) * max_sb_rows);
+      this_tile->row_mt_info.current_mi_row = this_tile->tile_info.mi_row_start;
+      this_tile->row_mt_info.num_threads_working = 0;
+
+      av1_inter_mode_data_init(this_tile);
+      av1_zero_above_context(cm, &cpi->td.mb.e_mbd,
+                             this_tile->tile_info.mi_col_start,
+                             this_tile->tile_info.mi_col_end, tile_row);
+      this_tile->m_search_count = 0;   // Count of motion search hits.
+      this_tile->ex_search_count = 0;  // Exhaustive mesh search hits.
+    }
+  }
+
+  // Only run once to create threads and allocate thread data.
+  if (cpi->num_workers == 0) {
+    create_enc_workers(cpi, num_workers);
+  } else {
+    num_workers = AOMMIN(num_workers, cpi->num_workers);
+  }
+  assign_tile_to_thread(multi_thread_ctxt, tile_cols * tile_rows, num_workers);
+  prepare_enc_workers(cpi, enc_row_mt_worker_hook, num_workers);
+  launch_enc_workers(cpi, num_workers);
+  sync_enc_workers(cpi, num_workers);
+  if (cm->delta_q_info.delta_lf_present_flag) update_delta_lf_for_row_mt(cpi);
+  accumulate_counters_enc_workers(cpi, num_workers);
+}
diff --git a/third_party/aom/av1/encoder/ethread.h b/third_party/aom/av1/encoder/ethread.h
index 5de4b48038..1830759504 100644
--- a/third_party/aom/av1/encoder/ethread.h
+++ b/third_party/aom/av1/encoder/ethread.h
@@ -18,14 +18,31 @@ extern "C" {
 
 struct AV1_COMP;
 struct ThreadData;
+struct AV1RowMTSyncData;
 
 typedef struct EncWorkerData {
   struct AV1_COMP *cpi;
   struct ThreadData *td;
   int start;
+  int thread_id;
 } EncWorkerData;
 
+void av1_row_mt_sync_read(AV1RowMTSync *const row_mt_sync, int r, int c);
+void av1_row_mt_sync_write(AV1RowMTSync *const row_mt_sync, int r, int c,
+                           const int cols);
+
+void av1_row_mt_sync_read_dummy(struct AV1RowMTSyncData *const row_mt_sync,
+                                int r, int c);
+void av1_row_mt_sync_write_dummy(struct AV1RowMTSyncData *const row_mt_sync,
+                                 int r, int c, const int cols);
+
+void av1_row_mt_sync_mem_dealloc(AV1RowMTSync *row_mt_sync);
+// Allocate memory for row based multi-threading synchronization.
+void av1_row_mt_sync_mem_alloc(AV1RowMTSync *row_mt_sync, struct AV1Common *cm,
+                               int rows);
+
 void av1_encode_tiles_mt(struct AV1_COMP *cpi);
+void av1_encode_tiles_row_mt(struct AV1_COMP *cpi);
 
 void av1_accumulate_frame_counts(struct FRAME_COUNTS *acc_counts,
                                  const struct FRAME_COUNTS *counts);
diff --git a/third_party/aom/av1/encoder/firstpass.c b/third_party/aom/av1/encoder/firstpass.c
index 69dd20c526..8b9513b645 100644
--- a/third_party/aom/av1/encoder/firstpass.c
+++ b/third_party/aom/av1/encoder/firstpass.c
@@ -36,6 +36,7 @@
 #include "av1/encoder/encodemb.h"
 #include "av1/encoder/encodemv.h"
 #include "av1/encoder/encoder.h"
+#include "av1/encoder/encode_strategy.h"
 #include "av1/encoder/extend.h"
 #include "av1/encoder/firstpass.h"
 #include "av1/encoder/mcomp.h"
@@ -58,7 +59,8 @@
 #define KF_MAX_FRAME_BOOST 128.0
 #define MIN_ARF_GF_BOOST 240
 #define MIN_DECAY_FACTOR 0.01
-#define MIN_KF_BOOST 300
+#define MIN_KF_BOOST 300          // Minimum boost for non-static KF interval
+#define MIN_STATIC_KF_BOOST 5400  // Minimum boost for static KF interval
 #define NEW_MV_MODE_PENALTY 32
 #define DARK_THRESH 64
 #define DEFAULT_GRP_WEIGHT 1.0
@@ -379,7 +381,7 @@ static void first_pass_motion_search(AV1_COMP *cpi, MACROBLOCK *x,
 
   // Override the default variance function to use MSE.
   v_fn_ptr.vf = get_block_variance_fn(bsize);
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     v_fn_ptr.vf = highbd_get_block_variance_fn(bsize, xd->bd);
   }
 
@@ -448,18 +450,6 @@ static int find_fp_qindex(aom_bit_depth_t bit_depth) {
   return i;
 }
 
-static void set_first_pass_params(AV1_COMP *cpi) {
-  AV1_COMMON *const cm = &cpi->common;
-  if (!cpi->refresh_alt_ref_frame &&
-      (cm->current_video_frame == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY))) {
-    cm->frame_type = KEY_FRAME;
-  } else {
-    cm->frame_type = INTER_FRAME;
-  }
-  // Do not use periodic key frames.
-  cpi->rc.frames_to_key = INT_MAX;
-}
-
 static double raw_motion_error_stdev(int *raw_motion_err_list,
                                      int raw_motion_err_counts) {
   int64_t sum_raw_err = 0;
@@ -485,10 +475,11 @@ static double raw_motion_error_stdev(int *raw_motion_err_list,
 
 #define UL_INTRA_THRESH 50
 #define INVALID_ROW -1
-void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
+void av1_first_pass(AV1_COMP *cpi, const int64_t ts_duration) {
   int mb_row, mb_col;
   MACROBLOCK *const x = &cpi->td.mb;
   AV1_COMMON *const cm = &cpi->common;
+  CurrentFrame *const current_frame = &cm->current_frame;
   const SequenceHeader *const seq_params = &cm->seq_params;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *const xd = &x->e_mbd;
@@ -521,13 +512,13 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
   TWO_PASS *twopass = &cpi->twopass;
   int recon_y_stride, recon_uv_stride, uv_mb_height;
 
-  YV12_BUFFER_CONFIG *const lst_yv12 = get_ref_frame_buffer(cpi, LAST_FRAME);
-  YV12_BUFFER_CONFIG *gld_yv12 = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
-  YV12_BUFFER_CONFIG *const new_yv12 = get_frame_new_buffer(cm);
+  const YV12_BUFFER_CONFIG *const lst_yv12 =
+      get_ref_frame_yv12_buf(cm, LAST_FRAME);
+  const YV12_BUFFER_CONFIG *gld_yv12 = get_ref_frame_yv12_buf(cm, GOLDEN_FRAME);
+  YV12_BUFFER_CONFIG *const new_yv12 = &cm->cur_frame->buf;
   const YV12_BUFFER_CONFIG *first_ref_buf = lst_yv12;
   double intra_factor;
   double brightness_factor;
-  BufferPool *const pool = cm->buffer_pool;
   const int qindex = find_fp_qindex(seq_params->bit_depth);
   const int mb_scale = mi_size_wide[BLOCK_16X16];
 
@@ -556,13 +547,16 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
   brightness_factor = 0.0;
   neutral_count = 0.0;
 
-  set_first_pass_params(cpi);
+  // Do not use periodic key frames.
+  cpi->rc.frames_to_key = INT_MAX;
+
   av1_set_quantizer(cm, qindex);
 
   av1_setup_block_planes(&x->e_mbd, seq_params->subsampling_x,
                          seq_params->subsampling_y, num_planes);
 
-  av1_setup_src_planes(x, cpi->source, 0, 0, num_planes);
+  av1_setup_src_planes(x, cpi->source, 0, 0, num_planes,
+                       x->e_mbd.mi[0]->sb_type);
   av1_setup_dst_planes(xd->plane, seq_params->sb_size, new_yv12, 0, 0, 0,
                        num_planes);
 
@@ -586,7 +580,6 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
   }
 
   av1_init_mv_probs(cm);
-  av1_init_lv_map(cm);
   av1_initialize_rd_consts(cpi);
 
   // Tiling is ignored in the first pass.
@@ -699,14 +692,15 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
       // Accumulate the intra error.
       intra_error += (int64_t)this_error;
 
-      int stride = x->plane[0].src.stride;
+      const int hbd = is_cur_buf_hbd(xd);
+      const int stride = x->plane[0].src.stride;
       uint8_t *buf = x->plane[0].src.buf;
-      for (int r8 = 0; r8 < 2; ++r8)
+      for (int r8 = 0; r8 < 2; ++r8) {
         for (int c8 = 0; c8 < 2; ++c8) {
-          int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
           frame_avg_wavelet_energy += av1_haar_ac_sad_8x8_uint8_input(
               buf + c8 * 8 + r8 * 8 * stride, stride, hbd);
         }
+      }
 
 #if CONFIG_FP_MB_STATS
       if (cpi->use_fp_mb_stats) {
@@ -728,7 +722,7 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
         struct buf_2d unscaled_last_source_buf_2d;
 
         xd->plane[0].pre[0].buf = first_ref_buf->y_buffer + recon_yoffset;
-        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+        if (is_cur_buf_hbd(xd)) {
           motion_error = highbd_get_prediction_error(
               bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
         } else {
@@ -743,7 +737,7 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
             cpi->unscaled_last_source->y_buffer + recon_yoffset;
         unscaled_last_source_buf_2d.stride =
             cpi->unscaled_last_source->y_stride;
-        if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+        if (is_cur_buf_hbd(xd)) {
           raw_motion_error = highbd_get_prediction_error(
               bsize, &x->plane[0].src, &unscaled_last_source_buf_2d, xd->bd);
         } else {
@@ -770,12 +764,12 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
           }
 
           // Search in an older reference frame.
-          if ((cm->current_video_frame > 1) && gld_yv12 != NULL) {
+          if ((current_frame->frame_number > 1) && gld_yv12 != NULL) {
             // Assume 0,0 motion with no mv overhead.
             int gf_motion_error;
 
             xd->plane[0].pre[0].buf = gld_yv12->y_buffer + recon_yoffset;
-            if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+            if (is_cur_buf_hbd(xd)) {
               gf_motion_error = highbd_get_prediction_error(
                   bsize, &x->plane[0].src, &xd->plane[0].pre[0], xd->bd);
             } else {
@@ -852,8 +846,9 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
           xd->mi[0]->tx_size = TX_4X4;
           xd->mi[0]->ref_frame[0] = LAST_FRAME;
           xd->mi[0]->ref_frame[1] = NONE_FRAME;
-          av1_build_inter_predictors_sby(cm, xd, mb_row * mb_scale,
-                                         mb_col * mb_scale, NULL, bsize);
+          av1_enc_build_inter_predictor(cm, xd, mb_row * mb_scale,
+                                        mb_col * mb_scale, NULL, bsize,
+                                        AOM_PLANE_Y, AOM_PLANE_Y);
           av1_encode_sby_pass1(cm, x, bsize);
           sum_mvr += mv.row;
           sum_mvr_abs += abs(mv.row);
@@ -995,7 +990,7 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
     brightness_factor = brightness_factor / (double)num_mbs;
     fps.weight = intra_factor * brightness_factor;
 
-    fps.frame = cm->current_video_frame;
+    fps.frame = current_frame->frame_number;
     fps.coded_error = (double)(coded_error >> 8) + min_err;
     fps.sr_coded_error = (double)(sr_coded_error >> 8) + min_err;
     fps.intra_error = (double)(intra_error >> 8) + min_err;
@@ -1036,7 +1031,7 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
     // TODO(paulwilkins):  Handle the case when duration is set to 0, or
     // something less than the full time between subsequent values of
     // cpi->source_time_stamp.
-    fps.duration = (double)(source->ts_end - source->ts_start);
+    fps.duration = (double)ts_duration;
 
     // Don't want to do output stats with a stack variable!
     twopass->this_frame_stats = fps;
@@ -1054,14 +1049,14 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
   // Copy the previous Last Frame back into gf and and arf buffers if
   // the prediction is good enough... but also don't allow it to lag too far.
   if ((twopass->sr_update_lag > 3) ||
-      ((cm->current_video_frame > 0) &&
+      ((current_frame->frame_number > 0) &&
        (twopass->this_frame_stats.pcnt_inter > 0.20) &&
        ((twopass->this_frame_stats.intra_error /
          DOUBLE_DIVIDE_CHECK(twopass->this_frame_stats.coded_error)) > 2.0))) {
     if (gld_yv12 != NULL) {
-      ref_cnt_fb(pool->frame_bufs,
-                 &cm->ref_frame_map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]],
-                 cm->ref_frame_map[cpi->ref_fb_idx[LAST_FRAME - 1]]);
+      assign_frame_buffer_p(
+          &cm->ref_frame_map[get_ref_frame_map_idx(cm, GOLDEN_FRAME)],
+          cm->ref_frame_map[get_ref_frame_map_idx(cm, LAST_FRAME)]);
     }
     twopass->sr_update_lag = 1;
   } else {
@@ -1071,17 +1066,16 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
   aom_extend_frame_borders(new_yv12, num_planes);
 
   // The frame we just compressed now becomes the last frame.
-  ref_cnt_fb(pool->frame_bufs,
-             &cm->ref_frame_map[cpi->ref_fb_idx[LAST_FRAME - 1]],
-             cm->new_fb_idx);
+  assign_frame_buffer_p(
+      &cm->ref_frame_map[get_ref_frame_map_idx(cm, LAST_FRAME)], cm->cur_frame);
 
   // Special case for the first frame. Copy into the GF buffer as a second
   // reference.
-  if (cm->current_video_frame == 0 &&
-      cpi->ref_fb_idx[GOLDEN_FRAME - 1] != INVALID_IDX) {
-    ref_cnt_fb(pool->frame_bufs,
-               &cm->ref_frame_map[cpi->ref_fb_idx[GOLDEN_FRAME - 1]],
-               cm->ref_frame_map[cpi->ref_fb_idx[LAST_FRAME - 1]]);
+  if (current_frame->frame_number == 0 &&
+      get_ref_frame_map_idx(cm, GOLDEN_FRAME) != INVALID_IDX) {
+    assign_frame_buffer_p(
+        &cm->ref_frame_map[get_ref_frame_map_idx(cm, GOLDEN_FRAME)],
+        cm->ref_frame_map[get_ref_frame_map_idx(cm, LAST_FRAME)]);
   }
 
   // Use this to see what the first pass reconstruction looks like.
@@ -1089,9 +1083,9 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
     char filename[512];
     FILE *recon_file;
     snprintf(filename, sizeof(filename), "enc%04d.yuv",
-             (int)cm->current_video_frame);
+             (int)current_frame->frame_number);
 
-    if (cm->current_video_frame == 0)
+    if (current_frame->frame_number == 0)
       recon_file = fopen(filename, "wb");
     else
       recon_file = fopen(filename, "ab");
@@ -1100,7 +1094,7 @@ void av1_first_pass(AV1_COMP *cpi, const struct lookahead_entry *source) {
     fclose(recon_file);
   }
 
-  ++cm->current_video_frame;
+  ++current_frame->frame_number;
 }
 
 static double calc_correction_factor(double err_per_mb, double err_divisor,
@@ -1565,7 +1559,6 @@ static int calculate_boost_bits(int frame_count, int boost,
                 0);
 }
 
-#if USE_SYMM_MULTI_LAYER
 // #define CHCEK_GF_PARAMETER
 #ifdef CHCEK_GF_PARAMETER
 void check_frame_params(GF_GROUP *const gf_group, int gf_interval,
@@ -1587,7 +1580,7 @@ void check_frame_params(GF_GROUP *const gf_group, int gf_interval,
   }
 
   fprintf(fid, "number of nodes in each level: \n");
-  for (int i = 0; i < MAX_PYRAMID_LVL; ++i) {
+  for (int i = 0; i < gf_group->pyramid_height; ++i) {
     fprintf(fid, "lvl %d: %d ", i, gf_group->pyramid_lvl_nodes[i]);
   }
   fprintf(fid, "\n");
@@ -1652,7 +1645,7 @@ static void set_multi_layer_params(GF_GROUP *const gf_group, int l, int r,
 }
 
 static INLINE unsigned char get_pyramid_height(int pyramid_width) {
-  assert(pyramid_width <= 16 && pyramid_width >= 4 &&
+  assert(pyramid_width <= MAX_GF_INTERVAL && pyramid_width >= MIN_GF_INTERVAL &&
          "invalid gf interval for pyramid structure");
 
   return pyramid_width > 12 ? 4 : (pyramid_width > 6 ? 3 : 2);
@@ -1689,14 +1682,46 @@ static int construct_multi_layer_gf_structure(GF_GROUP *const gf_group,
   return frame_index;
 }
 
-void define_customized_gf_group_structure(AV1_COMP *cpi) {
+// Given the maximum allowed height of the pyramid structure, return the maximum
+// GF length supported by the same.
+static INLINE int get_max_gf_length(int max_pyr_height) {
+  // We allow a frame to have at most two left/right descendants before changing
+  // them into to a subtree, i.e., we allow the following structure:
+  /*                    OUT_OF_ORDER_FRAME
+                       / /              \ \
+  (two left children) F F                F F (two right children) */
+  // For example, the max gf size supported by 4 layer structure is:
+  // 1 (KEY/OVERLAY) + 1 + 2 + 4 + 16 (two children on both side of their
+  // parent)
+  switch (max_pyr_height) {
+    case 2: return 6;   // = 1 (KEY/OVERLAY) + 1 + 4
+    case 3: return 12;  // = 1 (KEY/OVERLAY) + 1 + 2 + 8
+    case 4: return 24;  // = 1 (KEY/OVERLAY) + 1 + 2 + 4 + 16
+    case 1:
+      return MAX_GF_INTERVAL;  // Special case: uses the old pyramid structure.
+    default: assert(0 && "Invalid max_pyr_height"); return -1;
+  }
+}
+
+// Given the maximum allowed height of the pyramid structure, return the fixed
+// GF length to be used.
+int av1_rc_get_fixed_gf_length(int max_pyr_height) {
+  const int max_gf_length_allowed = get_max_gf_length(max_pyr_height);
+  return AOMMIN(max_gf_length_allowed, MAX_GF_INTERVAL);
+}
+
+static void define_customized_gf_group_structure(
+    AV1_COMP *cpi, const EncodeFrameParams *const frame_params) {
   RATE_CONTROL *const rc = &cpi->rc;
   TWO_PASS *const twopass = &cpi->twopass;
   GF_GROUP *const gf_group = &twopass->gf_group;
-  const int key_frame = cpi->common.frame_type == KEY_FRAME;
+  const int key_frame = frame_params->frame_type == KEY_FRAME;
 
-  assert(rc->baseline_gf_interval >= 4 &&
-         rc->baseline_gf_interval <= MAX_PYRAMID_SIZE);
+  assert(rc->baseline_gf_interval >= MIN_GF_INTERVAL &&
+         rc->baseline_gf_interval <=
+             get_max_gf_length(cpi->oxcf.gf_max_pyr_height));
+  assert(cpi->oxcf.gf_max_pyr_height >= MIN_PYRAMID_LVL &&
+         cpi->oxcf.gf_max_pyr_height <= MAX_PYRAMID_LVL);
 
   const int gf_update_frames =
       construct_multi_layer_gf_structure(gf_group, rc->baseline_gf_interval);
@@ -1760,166 +1785,31 @@ void define_customized_gf_group_structure(AV1_COMP *cpi) {
 // It is an example of how to define a GF stucture manually. The function will
 // result in exactly the same GF group structure as
 // define_customized_gf_group_structure() when rc->baseline_gf_interval == 4
-#if USE_MANUAL_GF4_STRUCT
-#define GF_INTERVAL_4 4
-static const unsigned char gf4_multi_layer_params[][GF_FRAME_PARAMS] = {
-  {
-      // gf_group->index == 0 (Frame 0)
-      // It can also be KEY frame. Will assign the proper value
-      // in define_gf_group_structure
-      OVERLAY_UPDATE,  // update_type (default value)
-      0,               // arf_src_offset
-      0,               // arf_pos_in_gf
-      0                // arf_update_idx
-  },
-  {
-      // gf_group->index == 1 (Frame 4)
-      ARF_UPDATE,         // update_type
-      GF_INTERVAL_4 - 1,  // arf_src_offset
-      0,                  // arf_pos_in_gf
-      0                   // arf_update_idx
-  },
-  {
-      // gf_group->index == 2 (Frame 2)
-      INTNL_ARF_UPDATE,          // update_type
-      (GF_INTERVAL_4 >> 1) - 1,  // arf_src_offset
-      0,                         // arf_pos_in_gf
-      0                          // arf_update_idx
-  },
-  {
-      // gf_group->index == 3 (Frame 1)
-      LAST_BIPRED_UPDATE,  // update_type
-      0,                   // arf_src_offset
-      0,                   // arf_pos_in_gf
-      0                    // arf_update_idx
-  },
 
-  {
-      // gf_group->index == 4 (Frame 2 - OVERLAY)
-      INTNL_OVERLAY_UPDATE,  // update_type
-      0,                     // arf_src_offset
-      2,                     // arf_pos_in_gf
-      0                      // arf_update_idx
-  },
-  {
-      // gf_group->index == 5 (Frame 3)
-      LF_UPDATE,  // update_type
-      0,          // arf_src_offset
-      0,          // arf_pos_in_gf
-      1           // arf_update_idx
-  }
-};
-
-static int define_gf_group_structure_4(AV1_COMP *cpi) {
-  RATE_CONTROL *const rc = &cpi->rc;
-  TWO_PASS *const twopass = &cpi->twopass;
-  GF_GROUP *const gf_group = &twopass->gf_group;
-  const int key_frame = cpi->common.frame_type == KEY_FRAME;
-
-  assert(rc->baseline_gf_interval == GF_INTERVAL_4);
-
-  const int gf_update_frames = rc->baseline_gf_interval + 2;
-  int frame_index;
-
-  for (frame_index = 0; frame_index < gf_update_frames; ++frame_index) {
-    int param_idx = 0;
-
-    gf_group->bidir_pred_enabled[frame_index] = 0;
-
-    if (frame_index == 0) {
-      // gf_group->arf_src_offset[frame_index] = 0;
-      gf_group->brf_src_offset[frame_index] = 0;
-      gf_group->bidir_pred_enabled[frame_index] = 0;
-
-      // For key frames the frame target rate is already set and it
-      // is also the golden frame.
-      if (key_frame) continue;
-
-      gf_group->update_type[frame_index] =
-          gf4_multi_layer_params[frame_index][param_idx++];
-
-      if (rc->source_alt_ref_active) {
-        gf_group->update_type[frame_index] = OVERLAY_UPDATE;
-      } else {
-        gf_group->update_type[frame_index] = GF_UPDATE;
-      }
-      param_idx++;
-    } else {
-      gf_group->update_type[frame_index] =
-          gf4_multi_layer_params[frame_index][param_idx++];
-    }
-
-    // setup other parameters
-    gf_group->rf_level[frame_index] =
-        update_type_2_rf_level(gf_group->update_type[frame_index]);
-
-    // == arf_src_offset ==
-    gf_group->arf_src_offset[frame_index] =
-        gf4_multi_layer_params[frame_index][param_idx++];
-
-    // == arf_pos_in_gf ==
-    gf_group->arf_pos_in_gf[frame_index] =
-        gf4_multi_layer_params[frame_index][param_idx++];
-
-    // == arf_update_idx ==
-    gf_group->brf_src_offset[frame_index] =
-        gf4_multi_layer_params[frame_index][param_idx];
-  }
-
-  // NOTE: We need to configure the frame at the end of the sequence + 1 that
-  //       will be the start frame for the next group. Otherwise prior to the
-  //       call to av1_rc_get_second_pass_params() the data will be undefined.
-  gf_group->arf_update_idx[frame_index] = 0;
-  gf_group->arf_ref_idx[frame_index] = 0;
-
-  if (rc->source_alt_ref_pending) {
-    gf_group->update_type[frame_index] = OVERLAY_UPDATE;
-    gf_group->rf_level[frame_index] = INTER_NORMAL;
-
-  } else {
-    gf_group->update_type[frame_index] = GF_UPDATE;
-    gf_group->rf_level[frame_index] = GF_ARF_STD;
-  }
-
-  gf_group->bidir_pred_enabled[frame_index] = 0;
-  gf_group->brf_src_offset[frame_index] = 0;
-
-  // This value is only used for INTNL_OVERLAY_UPDATE
-  gf_group->arf_pos_in_gf[frame_index] = 0;
-
-  return gf_update_frames;
-}
-#endif  // USE_MANUAL_GF4_STRUCT
-#endif  // USE_SYMM_MULTI_LAYER
-
-static void define_gf_group_structure(AV1_COMP *cpi) {
+static void define_gf_group_structure(
+    AV1_COMP *cpi, const EncodeFrameParams *const frame_params) {
   RATE_CONTROL *const rc = &cpi->rc;
 
-#if USE_SYMM_MULTI_LAYER
+  const int max_pyr_height = cpi->oxcf.gf_max_pyr_height;
   const int valid_customized_gf_length =
-      rc->baseline_gf_interval >= 4 &&
-      rc->baseline_gf_interval <= MAX_PYRAMID_SIZE;
+      max_pyr_height >= MIN_PYRAMID_LVL && max_pyr_height <= MAX_PYRAMID_LVL &&
+      rc->baseline_gf_interval >= MIN_GF_INTERVAL &&
+      rc->baseline_gf_interval <= get_max_gf_length(max_pyr_height);
   // used the new structure only if extra_arf is allowed
   if (valid_customized_gf_length && rc->source_alt_ref_pending &&
       cpi->extra_arf_allowed > 0) {
-#if USE_MANUAL_GF4_STRUCT
-    if (rc->baseline_gf_interval == 4)
-      define_gf_group_structure_4(cpi);
-    else
-#endif
-      define_customized_gf_group_structure(cpi);
+    define_customized_gf_group_structure(cpi, frame_params);
     cpi->new_bwdref_update_rule = 1;
     return;
   } else {
     cpi->new_bwdref_update_rule = 0;
   }
-#endif
 
   TWO_PASS *const twopass = &cpi->twopass;
   GF_GROUP *const gf_group = &twopass->gf_group;
   int i;
   int frame_index = 0;
-  const int key_frame = cpi->common.frame_type == KEY_FRAME;
+  const int key_frame = frame_params->frame_type == KEY_FRAME;
 
   // The use of bi-predictive frames are only enabled when following 3
   // conditions are met:
@@ -2015,7 +1905,10 @@ static void define_gf_group_structure(AV1_COMP *cpi) {
     accumulative_subgroup_interval += subgroup_interval[cpi->num_extra_arfs];
   }
 
-  for (i = 0; i < rc->baseline_gf_interval - rc->source_alt_ref_pending; ++i) {
+  const int normal_frames =
+      rc->baseline_gf_interval - (key_frame || rc->source_alt_ref_pending);
+
+  for (i = 0; i < normal_frames; ++i) {
     gf_group->arf_update_idx[frame_index] = which_arf;
     gf_group->arf_ref_idx[frame_index] = which_arf;
 
@@ -2127,41 +2020,28 @@ static void define_gf_group_structure(AV1_COMP *cpi) {
   gf_group->brf_src_offset[frame_index] = 0;
 }
 
-#if USE_SYMM_MULTI_LAYER
-#define LEAF_REDUCTION_FACTOR 0.75f
-#define LVL_3_BOOST_FACTOR 0.8f
-#define LVL_2_BOOST_FACTOR 0.3f
-
-static float_t lvl_budget_factor[MAX_PYRAMID_LVL - 1][MAX_PYRAMID_LVL - 1] = {
-  { 1, 0, 0 },
-  { LVL_3_BOOST_FACTOR, 0, 0 },  // Leaking budget works better
-  { LVL_3_BOOST_FACTOR, (1 - LVL_3_BOOST_FACTOR) * LVL_2_BOOST_FACTOR,
-    (1 - LVL_3_BOOST_FACTOR) * (1 - LVL_2_BOOST_FACTOR) }
+#define LEAF_REDUCTION_FACTOR 0.75
+static double lvl_budget_factor[MAX_PYRAMID_LVL - 1][MAX_PYRAMID_LVL - 1] = {
+  { 1.0, 0.0, 0.0 }, { 0.6, 0.4, 0 }, { 0.45, 0.35, 0.20 }
 };
-#endif  // USE_SYMM_MULTI_LAYER
-static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
-                                   double group_error, int gf_arf_bits) {
+static void allocate_gf_group_bits(
+    AV1_COMP *cpi, int64_t gf_group_bits, double group_error, int gf_arf_bits,
+    const EncodeFrameParams *const frame_params) {
   RATE_CONTROL *const rc = &cpi->rc;
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
   TWO_PASS *const twopass = &cpi->twopass;
   GF_GROUP *const gf_group = &twopass->gf_group;
-  FIRSTPASS_STATS frame_stats;
   int i;
   int frame_index = 0;
-  int target_frame_size;
-  int key_frame;
+  const int key_frame = frame_params->frame_type == KEY_FRAME;
   const int max_bits = frame_max_bits(&cpi->rc, &cpi->oxcf);
   int64_t total_group_bits = gf_group_bits;
-  double modified_err = 0.0;
-  double err_fraction;
   int ext_arf_boost[MAX_EXT_ARFS];
 
-  define_gf_group_structure(cpi);
+  define_gf_group_structure(cpi, frame_params);
 
   av1_zero_array(ext_arf_boost, MAX_EXT_ARFS);
 
-  key_frame = cpi->common.frame_type == KEY_FRAME;
-
   // For key frames the frame target rate is already set and it
   // is also the golden frame.
   // === [frame_index == 0] ===
@@ -2172,6 +2052,7 @@ static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
       gf_group->bit_allocation[frame_index] = gf_arf_bits;
 
     // Step over the golden frame / overlay frame
+    FIRSTPASS_STATS frame_stats;
     if (EOF == input_stats(twopass, &frame_stats)) return;
   }
 
@@ -2196,21 +2077,26 @@ static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
     }
   }
 
+  // Save.
+  const int tmp_frame_index = frame_index;
+  int budget_reduced_from_leaf_level = 0;
+
   // Allocate bits to the other frames in the group.
-  for (i = 0; i < rc->baseline_gf_interval - rc->source_alt_ref_pending; ++i) {
+  const int normal_frames =
+      rc->baseline_gf_interval - (key_frame || rc->source_alt_ref_pending);
+
+  for (i = 0; i < normal_frames; ++i) {
+    FIRSTPASS_STATS frame_stats;
     if (EOF == input_stats(twopass, &frame_stats)) break;
 
-    modified_err = calculate_modified_err(cpi, twopass, oxcf, &frame_stats);
-
-    if (group_error > 0)
-      err_fraction = modified_err / DOUBLE_DIVIDE_CHECK(group_error);
-    else
-      err_fraction = 0.0;
-
-    target_frame_size = (int)((double)total_group_bits * err_fraction);
-
-    target_frame_size =
-        clamp(target_frame_size, 0, AOMMIN(max_bits, (int)total_group_bits));
+    const double modified_err =
+        calculate_modified_err(cpi, twopass, oxcf, &frame_stats);
+    const double err_fraction =
+        (group_error > 0) ? modified_err / DOUBLE_DIVIDE_CHECK(group_error)
+                          : 0.0;
+    const int target_frame_size =
+        clamp((int)((double)total_group_bits * err_fraction), 0,
+              AOMMIN(max_bits, (int)total_group_bits));
 
     if (gf_group->update_type[frame_index] == BRF_UPDATE) {
       // Boost up the allocated bits on BWDREF_FRAME
@@ -2224,40 +2110,26 @@ static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
       // TODO(zoeliu): To investigate whether the allocated bits on
       // BIPRED_UPDATE frames need to be further adjusted.
       gf_group->bit_allocation[frame_index] = target_frame_size;
-#if USE_SYMM_MULTI_LAYER
     } else if (cpi->new_bwdref_update_rule &&
                gf_group->update_type[frame_index] == INTNL_OVERLAY_UPDATE) {
       assert(gf_group->pyramid_height <= MAX_PYRAMID_LVL &&
-             gf_group->pyramid_height >= 0 &&
              "non-valid height for a pyramid structure");
 
-      int arf_pos = gf_group->arf_pos_in_gf[frame_index];
+      const int arf_pos = gf_group->arf_pos_in_gf[frame_index];
       gf_group->bit_allocation[frame_index] = 0;
 
       gf_group->bit_allocation[arf_pos] = target_frame_size;
-#if MULTI_LVL_BOOST_VBR_CQ
-      const int pyr_h = gf_group->pyramid_height - 2;
-      const int this_lvl = gf_group->pyramid_level[arf_pos];
-      const int dist2top = gf_group->pyramid_height - 1 - this_lvl;
-
-      const float_t budget =
-          LEAF_REDUCTION_FACTOR * gf_group->pyramid_lvl_nodes[0];
-      const float_t lvl_boost = budget * lvl_budget_factor[pyr_h][dist2top] /
-                                gf_group->pyramid_lvl_nodes[this_lvl];
-
-      gf_group->bit_allocation[arf_pos] += (int)(target_frame_size * lvl_boost);
-#endif  // MULTI_LVL_BOOST_VBR_CQ
-#endif  // USE_SYMM_MULTI_LAYER
+      // Note: Boost, if needed, is added in the next loop.
     } else {
       assert(gf_group->update_type[frame_index] == LF_UPDATE ||
              gf_group->update_type[frame_index] == INTNL_OVERLAY_UPDATE);
       gf_group->bit_allocation[frame_index] = target_frame_size;
-#if MULTI_LVL_BOOST_VBR_CQ
       if (cpi->new_bwdref_update_rule) {
-        gf_group->bit_allocation[frame_index] -=
+        const int this_budget_reduction =
             (int)(target_frame_size * LEAF_REDUCTION_FACTOR);
+        gf_group->bit_allocation[frame_index] -= this_budget_reduction;
+        budget_reduced_from_leaf_level += this_budget_reduction;
       }
-#endif  // MULTI_LVL_BOOST_VBR_CQ
     }
 
     ++frame_index;
@@ -2269,11 +2141,37 @@ static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
     }
   }
 
-#if USE_SYMM_MULTI_LAYER
+  if (budget_reduced_from_leaf_level > 0) {
+    // Restore.
+    frame_index = tmp_frame_index;
+
+    // Re-distribute this extra budget to overlay frames in the group.
+    for (i = 0; i < normal_frames; ++i) {
+      if (cpi->new_bwdref_update_rule &&
+          gf_group->update_type[frame_index] == INTNL_OVERLAY_UPDATE) {
+        assert(gf_group->pyramid_height <= MAX_PYRAMID_LVL &&
+               "non-valid height for a pyramid structure");
+        const int arf_pos = gf_group->arf_pos_in_gf[frame_index];
+        const int this_lvl = gf_group->pyramid_level[arf_pos];
+        const int dist2top = gf_group->pyramid_height - 1 - this_lvl;
+        const double lvl_boost_factor =
+            lvl_budget_factor[gf_group->pyramid_height - 2][dist2top];
+        const int extra_size =
+            (int)(budget_reduced_from_leaf_level * lvl_boost_factor /
+                  gf_group->pyramid_lvl_nodes[this_lvl]);
+        gf_group->bit_allocation[arf_pos] += extra_size;
+      }
+      ++frame_index;
+
+      // Skip all the extra-ARF's.
+      if (cpi->num_extra_arfs) {
+        while (gf_group->update_type[frame_index] == INTNL_ARF_UPDATE)
+          ++frame_index;
+      }
+    }
+  }
+
   if (cpi->new_bwdref_update_rule == 0 && rc->source_alt_ref_pending) {
-#else
-  if (rc->source_alt_ref_pending) {
-#endif
     if (cpi->num_extra_arfs) {
       // NOTE: For bit allocation, move the allocated bits associated with
       //       INTNL_OVERLAY_UPDATE to the corresponding INTNL_ARF_UPDATE.
@@ -2295,8 +2193,17 @@ static void allocate_gf_group_bits(AV1_COMP *cpi, int64_t gf_group_bits,
   }
 }
 
+// Returns true if KF group and GF group both are almost completely static.
+static INLINE int is_almost_static(double gf_zero_motion, int kf_zero_motion) {
+  return (gf_zero_motion >= 0.995) &&
+         (kf_zero_motion >= STATIC_KF_GROUP_THRESH);
+}
+
+#define ARF_ABS_ZOOM_THRESH 4.4
+
 // Analyse and define a gf/arf group.
-static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
+static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame,
+                            const EncodeFrameParams *const frame_params) {
   AV1_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
   AV1EncoderConfig *const oxcf = &cpi->oxcf;
@@ -2306,12 +2213,7 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   int i;
 
   double boost_score = 0.0;
-#if !CONFIG_FIX_GF_LENGTH
-  double old_boost_score = 0.0;
-  double mv_ratio_accumulator_thresh;
-  int active_max_gf_interval;
   int active_min_gf_interval;
-#endif
   double gf_group_err = 0.0;
 #if GROUP_ADAPTIVE_MAXQ
   double gf_group_raw_error = 0.0;
@@ -2340,14 +2242,15 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   int64_t gf_group_bits;
   double gf_group_error_left;
   int gf_arf_bits;
-  const int is_key_frame = frame_is_intra_only(cm);
-  const int arf_active_or_kf = is_key_frame || rc->source_alt_ref_active;
+  const int is_intra_only = frame_params->frame_type == KEY_FRAME ||
+                            frame_params->frame_type == INTRA_ONLY_FRAME;
+  const int arf_active_or_kf = is_intra_only || rc->source_alt_ref_active;
 
   cpi->extra_arf_allowed = 1;
 
   // Reset the GF group data structures unless this is a key
   // frame in which case it will already have been done.
-  if (is_key_frame == 0) {
+  if (!is_intra_only) {
     av1_zero(twopass->gf_group);
   }
 
@@ -2371,35 +2274,13 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
     gf_group_skip_pct -= this_frame->intra_skip_pct;
     gf_group_inactive_zone_rows -= this_frame->inactive_zone_rows;
   }
-#if !CONFIG_FIX_GF_LENGTH
   // Motion breakout threshold for loop below depends on image size.
-  mv_ratio_accumulator_thresh =
+  const double mv_ratio_accumulator_thresh =
       (cpi->initial_height + cpi->initial_width) / 4.0;
-  // Set a maximum and minimum interval for the GF group.
-  // If the image appears almost completely static we can extend beyond this.
-  {
-    int int_max_q = (int)(av1_convert_qindex_to_q(
-        twopass->active_worst_quality, cpi->common.seq_params.bit_depth));
-    int int_lbq = (int)(av1_convert_qindex_to_q(
-        rc->last_boosted_qindex, cpi->common.seq_params.bit_depth));
 
-    active_min_gf_interval = rc->min_gf_interval + AOMMIN(2, int_max_q / 200);
-    if (active_min_gf_interval > rc->max_gf_interval)
-      active_min_gf_interval = rc->max_gf_interval;
+  // TODO(urvang): Try logic to vary min and max interval based on q.
+  active_min_gf_interval = rc->min_gf_interval;
 
-    // The value chosen depends on the active Q range. At low Q we have
-    // bits to spare and are better with a smaller interval and smaller boost.
-    // At high Q when there are few bits to spare we are better with a longer
-    // interval to spread the cost of the GF.
-    active_max_gf_interval = 12 + AOMMIN(4, (int_lbq / 6));
-
-    // We have: active_min_gf_interval <= rc->max_gf_interval
-    if (active_max_gf_interval < active_min_gf_interval)
-      active_max_gf_interval = active_min_gf_interval;
-    else if (active_max_gf_interval > rc->max_gf_interval)
-      active_max_gf_interval = rc->max_gf_interval;
-  }
-#endif  // !CONFIG_FIX_GF_LENGTH
   double avg_sr_coded_error = 0;
   double avg_raw_err_stdev = 0;
   int non_zero_stdev_count = 0;
@@ -2442,8 +2323,10 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
       decay_accumulator = decay_accumulator * loop_decay_rate;
 
       // Monitor for static sections.
-      zero_motion_accumulator = AOMMIN(
-          zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
+      if ((rc->frames_since_key + i - 1) > 1) {
+        zero_motion_accumulator = AOMMIN(
+            zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
+      }
 
       // Break clause to detect very still sections after motion. For example,
       // a static image after a fade or other transition.
@@ -2458,34 +2341,23 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
     boost_score +=
         decay_accumulator *
         calc_frame_boost(cpi, &next_frame, this_frame_mv_in_out, GF_MAX_BOOST);
-#if CONFIG_FIX_GF_LENGTH
-    if (i == (FIXED_GF_LENGTH + 1)) break;
-#else
-    // Skip breaking condition for CONFIG_FIX_GF_LENGTH
-    // Break out conditions.
-    if (
-        // Break at active_max_gf_interval unless almost totally static.
-        (i >= (active_max_gf_interval + arf_active_or_kf) &&
-         zero_motion_accumulator < 0.995) ||
-        (
-            // Don't break out with a very short interval.
-            (i >= active_min_gf_interval + arf_active_or_kf) &&
-            (!flash_detected) &&
-            ((mv_ratio_accumulator > mv_ratio_accumulator_thresh) ||
-             (abs_mv_in_out_accumulator > 3.0) ||
-             (mv_in_out_accumulator < -2.0) ||
-             ((boost_score - old_boost_score) < BOOST_BREAKOUT)))) {
-      // If GF group interval is < 12, we force it to be 8. Otherwise,
-      // if it is >= 12, we keep it as is.
-      // NOTE: 'i' is 1 more than the GF group interval candidate that is being
-      //       checked.
-      if (i == (8 + 1) || i >= (12 + 1)) {
-        boost_score = old_boost_score;
-        break;
-      }
+    // If almost totally static, we will not use the the fixed GF length later,
+    // so we can continue for more frames.
+    if (i >= (av1_rc_get_fixed_gf_length(oxcf->gf_max_pyr_height) + 1) &&
+        !is_almost_static(zero_motion_accumulator,
+                          twopass->kf_zeromotion_pct)) {
+      break;
+    }
+
+    // Some conditions to breakout after min interval.
+    if (i >= active_min_gf_interval &&
+        // If possible don't break very close to a kf
+        (rc->frames_to_key - i >= rc->min_gf_interval) && (i & 0x01) &&
+        !flash_detected &&
+        (mv_ratio_accumulator > mv_ratio_accumulator_thresh ||
+         abs_mv_in_out_accumulator > ARF_ABS_ZOOM_THRESH)) {
+      break;
     }
-    old_boost_score = boost_score;
-#endif  // CONFIG_FIX_GF_LENGTH
     *this_frame = next_frame;
   }
   twopass->gf_zeromotion_pct = (int)(zero_motion_accumulator * 1000.0);
@@ -2512,23 +2384,23 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
 
   if (disable_bwd_extarf) cpi->extra_arf_allowed = 0;
 
+  const int use_alt_ref =
+      !is_almost_static(zero_motion_accumulator, twopass->kf_zeromotion_pct) &&
+      allow_alt_ref && (i < cpi->oxcf.lag_in_frames) &&
+      (i >= rc->min_gf_interval);
+
 #define REDUCE_GF_LENGTH_THRESH 4
 #define REDUCE_GF_LENGTH_TO_KEY_THRESH 9
 #define REDUCE_GF_LENGTH_BY 1
   int alt_offset = 0;
-#if REDUCE_LAST_GF_LENGTH
-  // TODO(weitinglin): The length reduction stretagy is tweaking using AOM_Q
-  // mode, and hurting the performance of VBR mode. We need to investigate how
-  // to adjust GF length for other modes.
+  // The length reduction strategy is tweaked using AOM_Q mode, and doesn't work
+  // for VBR mode.
+  // Also, we don't have do adjustment for lossless mode.
+  const int allow_gf_length_reduction =
+      (cpi->oxcf.rc_mode == AOM_Q || cpi->extra_arf_allowed == 0) &&
+      !is_lossless_requested(&cpi->oxcf);
 
-  int allow_gf_length_reduction =
-      cpi->oxcf.rc_mode == AOM_Q || cpi->extra_arf_allowed == 0;
-
-  // We are going to have an alt ref, but we don't have do adjustment for
-  // lossless mode
-  if (allow_alt_ref && allow_gf_length_reduction &&
-      (i < cpi->oxcf.lag_in_frames) && (i >= rc->min_gf_interval) &&
-      !is_lossless_requested(&cpi->oxcf)) {
+  if (allow_gf_length_reduction && use_alt_ref) {
     // adjust length of this gf group if one of the following condition met
     // 1: only one overlay frame left and this gf is too long
     // 2: next gf group is too short to have arf compared to the current gf
@@ -2552,11 +2424,9 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
       i -= roll_back;
     }
   }
-#endif
 
   // Should we use the alternate reference frame.
-  if (allow_alt_ref && (i < cpi->oxcf.lag_in_frames) &&
-      (i >= rc->min_gf_interval)) {
+  if (use_alt_ref) {
     // Calculate the boost for alt ref.
     rc->gfu_boost =
         calc_arf_boost(cpi, alt_offset, (i - 1), (i - 1), &f_boost, &b_boost);
@@ -2582,7 +2452,7 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
                 AOMMAX(MIN_FWD_KF_INTERVAL, rc->min_gf_interval)) &&
                (rc->frames_to_key != i)) {
       // if possible, merge the last two gf groups
-      if (rc->frames_to_key <= MAX_PYRAMID_SIZE) {
+      if (rc->frames_to_key <= get_max_gf_length(oxcf->gf_max_pyr_height)) {
         rc->baseline_gf_interval = rc->frames_to_key;
         // if merging the last two gf groups creates a group that is too long,
         // split them and force the last gf group to be the MIN_FWD_KF_INTERVAL
@@ -2590,14 +2460,12 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
         rc->baseline_gf_interval = rc->frames_to_key - MIN_FWD_KF_INTERVAL;
       }
     } else {
-      rc->baseline_gf_interval =
-          i - (is_key_frame || rc->source_alt_ref_pending);
+      rc->baseline_gf_interval = i - rc->source_alt_ref_pending;
     }
   } else {
-    rc->baseline_gf_interval = i - (is_key_frame || rc->source_alt_ref_pending);
+    rc->baseline_gf_interval = i - rc->source_alt_ref_pending;
   }
 
-#if REDUCE_LAST_ALT_BOOST
 #define LAST_ALR_BOOST_FACTOR 0.2f
   rc->arf_boost_factor = 1.0;
   if (rc->source_alt_ref_pending && !is_lossless_requested(&cpi->oxcf)) {
@@ -2607,29 +2475,24 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
       rc->arf_boost_factor = LAST_ALR_BOOST_FACTOR;
     }
   }
-#endif
 
   if (!cpi->extra_arf_allowed) {
     cpi->num_extra_arfs = 0;
   } else {
-#if USE_SYMM_MULTI_LAYER
-    if (rc->baseline_gf_interval == 4 && rc->source_alt_ref_pending)
+    // Calculate 'num_extra_arfs' (internal alt-refs) that we are allowed.
+    // Note: When new pyramid structure is used through
+    // 'define_customized_gf_group_structure()' function, this value is
+    // overridden.
+    if (rc->baseline_gf_interval == MIN_GF_INTERVAL &&
+        rc->source_alt_ref_pending) {
       cpi->num_extra_arfs = 1;
-    else
-      cpi->num_extra_arfs = get_number_of_extra_arfs(
-          rc->baseline_gf_interval, rc->source_alt_ref_pending);
-#else
-    // Compute how many extra alt_refs we can have
-    cpi->num_extra_arfs = get_number_of_extra_arfs(rc->baseline_gf_interval,
-                                                   rc->source_alt_ref_pending);
-#endif  // USE_SYMM_MULTI_LAYER
+    } else {
+      cpi->num_extra_arfs = get_number_of_extra_arfs(rc->baseline_gf_interval,
+                                                     rc->source_alt_ref_pending,
+                                                     oxcf->gf_max_pyr_height);
+    }
   }
 
-#if !USE_SYMM_MULTI_LAYER
-  // Currently at maximum two extra ARFs' are allowed
-  assert(cpi->num_extra_arfs <= MAX_EXT_ARFS);
-#endif
-
   rc->frames_till_gf_update_due = rc->baseline_gf_interval;
 
   rc->bipred_group_interval = BFG_INTERVAL;
@@ -2691,20 +2554,21 @@ static void define_gf_group(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   // also a key frame in which case it has already been accounted for.
   if (rc->source_alt_ref_pending) {
     gf_group_error_left = gf_group_err - mod_frame_err;
-  } else if (is_key_frame == 0) {
+  } else if (!is_intra_only) {
     gf_group_error_left = gf_group_err - gf_first_frame_err;
   } else {
     gf_group_error_left = gf_group_err;
   }
 
   // Allocate bits to each of the frames in the GF group.
-  allocate_gf_group_bits(cpi, gf_group_bits, gf_group_error_left, gf_arf_bits);
+  allocate_gf_group_bits(cpi, gf_group_bits, gf_group_error_left, gf_arf_bits,
+                         frame_params);
 
   // Reset the file position.
   reset_fpf_position(twopass, start_pos);
 
   // Calculate a section intra ratio used in setting max loop filter.
-  if (cpi->common.frame_type != KEY_FRAME) {
+  if (frame_params->frame_type != KEY_FRAME) {
     twopass->section_intra_rating = calculate_section_intra_ratio(
         start_pos, twopass->stats_in_end, rc->baseline_gf_interval);
   }
@@ -2843,7 +2707,7 @@ static void find_next_key_frame(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
 
   av1_zero(next_frame);
 
-  cpi->common.frame_type = KEY_FRAME;
+  rc->frames_since_key = 0;
 
   // Reset the GF group data structures.
   av1_zero(*gf_group);
@@ -2988,8 +2852,13 @@ static void find_next_key_frame(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
     if (EOF == input_stats(twopass, &next_frame)) break;
 
     // Monitor for static sections.
-    zero_motion_accumulator = AOMMIN(zero_motion_accumulator,
-                                     get_zero_motion_factor(cpi, &next_frame));
+    // For the first frame in kf group, the second ref indicator is invalid.
+    if (i > 0) {
+      zero_motion_accumulator = AOMMIN(
+          zero_motion_accumulator, get_zero_motion_factor(cpi, &next_frame));
+    } else {
+      zero_motion_accumulator = next_frame.pcnt_inter - next_frame.pcnt_motion;
+    }
 
     // Not all frames in the group are necessarily used in calculating boost.
     if ((i <= rc->max_gf_interval) ||
@@ -3021,10 +2890,18 @@ static void find_next_key_frame(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   twopass->section_intra_rating = calculate_section_intra_ratio(
       start_position, twopass->stats_in_end, rc->frames_to_key);
 
-  // Apply various clamps for min and max boost
   rc->kf_boost = (int)(av_decay_accumulator * boost_score);
-  rc->kf_boost = AOMMAX(rc->kf_boost, (rc->frames_to_key * 3));
-  rc->kf_boost = AOMMAX(rc->kf_boost, MIN_KF_BOOST);
+
+  // Special case for static / slide show content but don't apply
+  // if the kf group is very short.
+  if ((zero_motion_accumulator > STATIC_KF_GROUP_FLOAT_THRESH) &&
+      (rc->frames_to_key > 8)) {
+    rc->kf_boost = AOMMAX(rc->kf_boost, MIN_STATIC_KF_BOOST);
+  } else {
+    // Apply various clamps for min and max boost
+    rc->kf_boost = AOMMAX(rc->kf_boost, (rc->frames_to_key * 3));
+    rc->kf_boost = AOMMAX(rc->kf_boost, MIN_KF_BOOST);
+  }
 
   // Work out how many bits to allocate for the key frame itself.
   kf_bits = calculate_boost_bits((rc->frames_to_key - 1), rc->kf_boost,
@@ -3058,172 +2935,6 @@ static void find_next_key_frame(AV1_COMP *cpi, FIRSTPASS_STATS *this_frame) {
   twopass->modified_error_left -= kf_group_err;
 }
 
-// Define the reference buffers that will be updated post encode.
-static void configure_buffer_updates(AV1_COMP *cpi) {
-  TWO_PASS *const twopass = &cpi->twopass;
-
-  // NOTE(weitinglin): Should we define another function to take care of
-  // cpi->rc.is_$Source_Type to make this function as it is in the comment?
-
-  cpi->rc.is_src_frame_alt_ref = 0;
-  cpi->rc.is_bwd_ref_frame = 0;
-  cpi->rc.is_last_bipred_frame = 0;
-  cpi->rc.is_bipred_frame = 0;
-  cpi->rc.is_src_frame_ext_arf = 0;
-
-  switch (twopass->gf_group.update_type[twopass->gf_group.index]) {
-    case KF_UPDATE:
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 1;
-      cpi->refresh_bwd_ref_frame = 1;
-      cpi->refresh_alt2_ref_frame = 1;
-      cpi->refresh_alt_ref_frame = 1;
-      break;
-
-    case LF_UPDATE:
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-      break;
-
-    case GF_UPDATE:
-      // TODO(zoeliu): To further investigate whether 'refresh_last_frame' is
-      //               needed.
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 1;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-      break;
-
-    case OVERLAY_UPDATE:
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 1;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      cpi->rc.is_src_frame_alt_ref = 1;
-      break;
-
-    case ARF_UPDATE:
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-      // NOTE: BWDREF does not get updated along with ALTREF_FRAME.
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 1;
-      break;
-
-    case BRF_UPDATE:
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 1;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      cpi->rc.is_bwd_ref_frame = 1;
-      break;
-
-    case LAST_BIPRED_UPDATE:
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      cpi->rc.is_last_bipred_frame = 1;
-      break;
-
-    case BIPRED_UPDATE:
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      cpi->rc.is_bipred_frame = 1;
-      break;
-
-    case INTNL_OVERLAY_UPDATE:
-      cpi->refresh_last_frame = 1;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      cpi->rc.is_src_frame_alt_ref = 1;
-      cpi->rc.is_src_frame_ext_arf = 1;
-      break;
-
-    case INTNL_ARF_UPDATE:
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-#if USE_SYMM_MULTI_LAYER
-      if (cpi->new_bwdref_update_rule == 1) {
-        cpi->refresh_bwd_ref_frame = 1;
-        cpi->refresh_alt2_ref_frame = 0;
-      } else {
-#endif
-        cpi->refresh_bwd_ref_frame = 0;
-        cpi->refresh_alt2_ref_frame = 1;
-#if USE_SYMM_MULTI_LAYER
-      }
-#endif
-      cpi->refresh_alt_ref_frame = 0;
-      break;
-
-    default: assert(0); break;
-  }
-}
-
-void av1_configure_buffer_updates_firstpass(AV1_COMP *cpi,
-                                            FRAME_UPDATE_TYPE update_type) {
-  RATE_CONTROL *rc = &cpi->rc;
-
-  cpi->refresh_last_frame = 1;
-  cpi->refresh_golden_frame = 0;
-  cpi->refresh_bwd_ref_frame = 0;
-  cpi->refresh_alt2_ref_frame = 0;
-  cpi->refresh_alt_ref_frame = 0;
-
-  rc->is_bwd_ref_frame = 0;
-
-  switch (update_type) {
-    case ARF_UPDATE:
-      cpi->refresh_alt_ref_frame = 1;
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-
-      rc->is_src_frame_alt_ref = 0;
-      break;
-    case INTNL_ARF_UPDATE:
-      cpi->refresh_alt2_ref_frame = 1;
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_bwd_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-      rc->is_src_frame_alt_ref = 0;
-      rc->is_src_frame_ext_arf = 0;
-
-      break;
-    case BIPRED_UPDATE:
-      cpi->refresh_bwd_ref_frame = 1;
-      cpi->refresh_last_frame = 0;
-      cpi->refresh_golden_frame = 0;
-      cpi->refresh_alt2_ref_frame = 0;
-      cpi->refresh_alt_ref_frame = 0;
-
-      rc->is_bwd_ref_frame = 1;
-      break;
-    default: break;
-  }
-}
-
 static int is_skippable_frame(const AV1_COMP *cpi) {
   // If the current frame does not have non-zero motion vector detected in the
   // first  pass, and so do its previous and forward frames, then this frame
@@ -3243,8 +2954,10 @@ static int is_skippable_frame(const AV1_COMP *cpi) {
           twopass->stats_in->pcnt_inter - twopass->stats_in->pcnt_motion == 1);
 }
 
-void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
+void av1_rc_get_second_pass_params(AV1_COMP *cpi,
+                                   EncodeFrameParams *const frame_params) {
   AV1_COMMON *const cm = &cpi->common;
+  CurrentFrame *const current_frame = &cm->current_frame;
   RATE_CONTROL *const rc = &cpi->rc;
   TWO_PASS *const twopass = &cpi->twopass;
   GF_GROUP *const gf_group = &twopass->gf_group;
@@ -3253,7 +2966,7 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
 
   int target_rate;
 
-  frames_left = (int)(twopass->total_stats.count - cm->current_video_frame);
+  frames_left = (int)(twopass->total_stats.count - current_frame->frame_number);
 
   if (!twopass->stats_in) return;
 
@@ -3261,16 +2974,16 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
   // advance the input pointer as we already have what we need.
   if (gf_group->update_type[gf_group->index] == ARF_UPDATE ||
       gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE) {
-    configure_buffer_updates(cpi);
+    av1_configure_buffer_updates(cpi, gf_group->update_type[gf_group->index]);
     target_rate = gf_group->bit_allocation[gf_group->index];
     target_rate = av1_rc_clamp_pframe_target_size(cpi, target_rate);
     rc->base_frame_target = target_rate;
 
     if (cpi->no_show_kf) {
       assert(gf_group->update_type[gf_group->index] == ARF_UPDATE);
-      cm->frame_type = KEY_FRAME;
+      frame_params->frame_type = KEY_FRAME;
     } else {
-      cm->frame_type = INTER_FRAME;
+      frame_params->frame_type = INTER_FRAME;
     }
 
     // Do the firstpass stats indicate that this frame is skippable for the
@@ -3286,7 +2999,7 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
 
   if (cpi->oxcf.rc_mode == AOM_Q) {
     twopass->active_worst_quality = cpi->oxcf.cq_level;
-  } else if (cm->current_video_frame == 0) {
+  } else if (current_frame->frame_number == 0) {
     // Special case code for first frame.
     const int section_target_bandwidth =
         (int)(twopass->bits_left / frames_left);
@@ -3325,16 +3038,17 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
   if (rc->frames_to_key == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY)) {
     FIRSTPASS_STATS this_frame_copy;
     this_frame_copy = this_frame;
+    frame_params->frame_type = KEY_FRAME;
     // Define next KF group and assign bits to it.
     find_next_key_frame(cpi, &this_frame);
     this_frame = this_frame_copy;
   } else {
-    cm->frame_type = INTER_FRAME;
+    frame_params->frame_type = INTER_FRAME;
   }
 
   // Define a new GF/ARF group. (Should always enter here for key frames).
   if (rc->frames_till_gf_update_due == 0) {
-    define_gf_group(cpi, &this_frame);
+    define_gf_group(cpi, &this_frame, frame_params);
 
     rc->frames_till_gf_update_due = rc->baseline_gf_interval;
 
@@ -3343,7 +3057,7 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
       FILE *fpfile;
       fpfile = fopen("arf.stt", "a");
       ++arf_count;
-      fprintf(fpfile, "%10d %10d %10d %10d %10d\n", cm->current_video_frame,
+      fprintf(fpfile, "%10d %10d %10d %10d %10d\n", current_frame->frame_number,
               rc->frames_till_gf_update_due, rc->kf_boost, arf_count,
               rc->gfu_boost);
 
@@ -3352,7 +3066,7 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
 #endif
   }
 
-  configure_buffer_updates(cpi);
+  av1_configure_buffer_updates(cpi, gf_group->update_type[gf_group->index]);
 
   // Do the firstpass stats indicate that this frame is skippable for the
   // partition search?
@@ -3362,7 +3076,7 @@ void av1_rc_get_second_pass_params(AV1_COMP *cpi) {
 
   target_rate = gf_group->bit_allocation[gf_group->index];
 
-  if (cpi->common.frame_type == KEY_FRAME)
+  if (frame_params->frame_type == KEY_FRAME)
     target_rate = av1_rc_clamp_iframe_target_size(cpi, target_rate);
   else
     target_rate = av1_rc_clamp_pframe_target_size(cpi, target_rate);
@@ -3409,7 +3123,7 @@ void av1_twopass_postencode_update(AV1_COMP *cpi) {
     rc->rate_error_estimate = 0;
   }
 
-  if (cpi->common.frame_type != KEY_FRAME) {
+  if (cpi->common.current_frame.frame_type != KEY_FRAME) {
     twopass->kf_group_bits -= bits_used;
     twopass->last_kfgroup_zeromotion_pct = twopass->kf_zeromotion_pct;
   }
diff --git a/third_party/aom/av1/encoder/firstpass.h b/third_party/aom/av1/encoder/firstpass.h
index 4b7325ae21..be1101a4b0 100644
--- a/third_party/aom/av1/encoder/firstpass.h
+++ b/third_party/aom/av1/encoder/firstpass.h
@@ -86,7 +86,7 @@ typedef struct {
   double raw_error_stdev;
 } FIRSTPASS_STATS;
 
-typedef enum {
+enum {
   KF_UPDATE = 0,
   LF_UPDATE = 1,
   GF_UPDATE = 2,
@@ -98,34 +98,29 @@ typedef enum {
   INTNL_OVERLAY_UPDATE = 8,  // Internal Overlay Frame
   INTNL_ARF_UPDATE = 9,      // Internal Altref Frame (candidate for ALTREF2)
   FRAME_UPDATE_TYPES = 10
-} FRAME_UPDATE_TYPE;
+} UENUM1BYTE(FRAME_UPDATE_TYPE);
 
 #define FC_ANIMATION_THRESH 0.15
-typedef enum {
+enum {
   FC_NORMAL = 0,
   FC_GRAPHICS_ANIMATION = 1,
   FRAME_CONTENT_TYPES = 2
-} FRAME_CONTENT_TYPE;
+} UENUM1BYTE(FRAME_CONTENT_TYPE);
 
 typedef struct {
   unsigned char index;
-  RATE_FACTOR_LEVEL rf_level[(MAX_LAG_BUFFERS * 2) + 1];
-  FRAME_UPDATE_TYPE update_type[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char arf_src_offset[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char arf_update_idx[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char arf_ref_idx[(MAX_LAG_BUFFERS * 2) + 1];
-#if USE_SYMM_MULTI_LAYER
-  unsigned char arf_pos_in_gf[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char pyramid_level[(MAX_LAG_BUFFERS * 2) + 1];
+  RATE_FACTOR_LEVEL rf_level[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  FRAME_UPDATE_TYPE update_type[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char arf_src_offset[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char arf_update_idx[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char arf_ref_idx[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char arf_pos_in_gf[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char pyramid_level[MAX_STATIC_GF_GROUP_LENGTH + 1];
   unsigned char pyramid_height;
   unsigned char pyramid_lvl_nodes[MAX_PYRAMID_LVL];
-#endif
-  unsigned char brf_src_offset[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char bidir_pred_enabled[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char ref_fb_idx_map[(MAX_LAG_BUFFERS * 2) + 1][REF_FRAMES];
-  unsigned char refresh_idx[(MAX_LAG_BUFFERS * 2) + 1];
-  unsigned char refresh_flag[(MAX_LAG_BUFFERS * 2) + 1];
-  int bit_allocation[(MAX_LAG_BUFFERS * 2) + 1];
+  unsigned char brf_src_offset[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  unsigned char bidir_pred_enabled[MAX_STATIC_GF_GROUP_LENGTH + 1];
+  int bit_allocation[MAX_STATIC_GF_GROUP_LENGTH + 1];
 } GF_GROUP;
 
 typedef struct {
@@ -176,29 +171,37 @@ typedef struct {
 } TWO_PASS;
 
 struct AV1_COMP;
+struct EncodeFrameParams;
 
 void av1_init_first_pass(struct AV1_COMP *cpi);
 void av1_rc_get_first_pass_params(struct AV1_COMP *cpi);
-void av1_first_pass(struct AV1_COMP *cpi, const struct lookahead_entry *source);
+void av1_first_pass(struct AV1_COMP *cpi, const int64_t ts_duration);
 void av1_end_first_pass(struct AV1_COMP *cpi);
 
 void av1_init_second_pass(struct AV1_COMP *cpi);
-void av1_rc_get_second_pass_params(struct AV1_COMP *cpi);
-void av1_configure_buffer_updates_firstpass(struct AV1_COMP *cpi,
-                                            FRAME_UPDATE_TYPE update_type);
+void av1_rc_get_second_pass_params(
+    struct AV1_COMP *cpi, struct EncodeFrameParams *const frame_params);
 
 // Post encode update of the rate control parameters for 2-pass
 void av1_twopass_postencode_update(struct AV1_COMP *cpi);
 
-static INLINE int get_number_of_extra_arfs(int interval, int arf_pending) {
-  if (arf_pending && MAX_EXT_ARFS > 0)
-    return interval >= MIN_EXT_ARF_INTERVAL * (MAX_EXT_ARFS + 1)
-               ? MAX_EXT_ARFS
-               : interval >= MIN_EXT_ARF_INTERVAL * MAX_EXT_ARFS
-                     ? MAX_EXT_ARFS - 1
-                     : 0;
-  else
-    return 0;
+static INLINE int get_number_of_extra_arfs(int interval, int arf_pending,
+                                           int max_pyr_height) {
+  // Max extra (internal) alt-refs allowed based on interval.
+  int extra_arfs_from_interval = 0;
+  if (arf_pending && MAX_EXT_ARFS > 0) {
+    extra_arfs_from_interval =
+        (interval >= MIN_EXT_ARF_INTERVAL * (MAX_EXT_ARFS + 1))
+            ? MAX_EXT_ARFS
+            : (interval >= MIN_EXT_ARF_INTERVAL * MAX_EXT_ARFS)
+                  ? MAX_EXT_ARFS - 1
+                  : 0;
+  }
+  // Max extra (internal) alt-refs allowed based on max pyramid height.
+  assert(max_pyr_height >= 1);
+  const int ext_arfs_from_max_height = max_pyr_height - 1;
+  // Finally, min of the two above is our actual max allowance.
+  return AOMMIN(extra_arfs_from_interval, ext_arfs_from_max_height);
 }
 
 #ifdef __cplusplus
diff --git a/third_party/aom/av1/encoder/global_motion.c b/third_party/aom/av1/encoder/global_motion.c
index e9f8b0bb47..142e44d804 100644
--- a/third_party/aom/av1/encoder/global_motion.c
+++ b/third_party/aom/av1/encoder/global_motion.c
@@ -15,8 +15,12 @@
 #include <math.h>
 #include <assert.h>
 
+#include "config/aom_dsp_rtcd.h"
+
 #include "av1/encoder/global_motion.h"
 
+#include "av1/common/convolve.h"
+#include "av1/common/resize.h"
 #include "av1/common/warped_motion.h"
 
 #include "av1/encoder/segmentation.h"
@@ -28,15 +32,45 @@
 #define MIN_INLIER_PROB 0.1
 
 #define MIN_TRANS_THRESH (1 * GM_TRANS_DECODE_FACTOR)
+#define USE_GM_FEATURE_BASED 1
 
 // Border over which to compute the global motion
 #define ERRORADV_BORDER 0
 
+// Number of pyramid levels in disflow computation
+#define N_LEVELS 2
+// Size of square patches in the disflow dense grid
+#define PATCH_SIZE 8
+// Center point of square patch
+#define PATCH_CENTER ((PATCH_SIZE + 1) >> 1)
+// Step size between patches, lower value means greater patch overlap
+#define PATCH_STEP 1
+// Minimum size of border padding for disflow
+#define MIN_PAD 7
+// Warp error convergence threshold for disflow
+#define DISFLOW_ERROR_TR 0.01
+// Max number of iterations if warp convergence is not found
+#define DISFLOW_MAX_ITR 10
+
+// Struct for an image pyramid
+typedef struct {
+  int n_levels;
+  int pad_size;
+  int has_gradient;
+  int widths[N_LEVELS];
+  int heights[N_LEVELS];
+  int strides[N_LEVELS];
+  int level_loc[N_LEVELS];
+  unsigned char *level_buffer;
+  double *level_dx_buffer;
+  double *level_dy_buffer;
+} ImagePyramid;
+
 static const double erroradv_tr[] = { 0.65, 0.60, 0.55 };
 static const double erroradv_prod_tr[] = { 20000, 18000, 16000 };
 
-int is_enough_erroradvantage(double best_erroradvantage, int params_cost,
-                             int erroradv_type) {
+int av1_is_enough_erroradvantage(double best_erroradvantage, int params_cost,
+                                 int erroradv_type) {
   assert(erroradv_type < GM_ERRORADV_TR_TYPES);
   return best_erroradvantage < erroradv_tr[erroradv_type] &&
          best_erroradvantage * params_cost < erroradv_prod_tr[erroradv_type];
@@ -75,9 +109,10 @@ static void convert_to_params(const double *params, int32_t *model) {
   }
 }
 
-void convert_model_to_params(const double *params, WarpedMotionParams *model) {
+void av1_convert_model_to_params(const double *params,
+                                 WarpedMotionParams *model) {
   convert_to_params(params, model->wmmat);
-  model->wmtype = get_gmtype(model);
+  model->wmtype = get_wmtype(model);
   model->invalid = 0;
 }
 
@@ -131,12 +166,13 @@ static void force_wmtype(WarpedMotionParams *wm, TransformationType wmtype) {
   wm->wmtype = wmtype;
 }
 
-int64_t refine_integerized_param(WarpedMotionParams *wm,
-                                 TransformationType wmtype, int use_hbd, int bd,
-                                 uint8_t *ref, int r_width, int r_height,
-                                 int r_stride, uint8_t *dst, int d_width,
-                                 int d_height, int d_stride, int n_refinements,
-                                 int64_t best_frame_error) {
+int64_t av1_refine_integerized_param(WarpedMotionParams *wm,
+                                     TransformationType wmtype, int use_hbd,
+                                     int bd, uint8_t *ref, int r_width,
+                                     int r_height, int r_stride, uint8_t *dst,
+                                     int d_width, int d_height, int d_stride,
+                                     int n_refinements,
+                                     int64_t best_frame_error) {
   static const int max_trans_model_params[TRANS_TYPES] = { 0, 2, 4, 6 };
   const int border = ERRORADV_BORDER;
   int i = 0, p;
@@ -209,7 +245,7 @@ int64_t refine_integerized_param(WarpedMotionParams *wm,
     }
   }
   force_wmtype(wm, wmtype);
-  wm->wmtype = get_gmtype(wm);
+  wm->wmtype = get_wmtype(wm);
   return best_error;
 }
 
@@ -240,12 +276,11 @@ static unsigned char *downconvert_frame(YV12_BUFFER_CONFIG *frm,
   return buf_8bit;
 }
 
-int compute_global_motion_feature_based(TransformationType type,
-                                        YV12_BUFFER_CONFIG *frm,
-                                        YV12_BUFFER_CONFIG *ref, int bit_depth,
-                                        int *num_inliers_by_motion,
-                                        double *params_by_motion,
-                                        int num_motions) {
+#if USE_GM_FEATURE_BASED
+static int compute_global_motion_feature_based(
+    TransformationType type, YV12_BUFFER_CONFIG *frm, YV12_BUFFER_CONFIG *ref,
+    int bit_depth, int *num_inliers_by_motion, double *params_by_motion,
+    int num_motions) {
   int i;
   int num_frm_corners, num_ref_corners;
   int num_correspondences;
@@ -296,3 +331,557 @@ int compute_global_motion_feature_based(TransformationType type,
   }
   return 0;
 }
+#else
+static INLINE RansacFuncDouble
+get_ransac_double_prec_type(TransformationType type) {
+  switch (type) {
+    case AFFINE: return ransac_affine_double_prec;
+    case ROTZOOM: return ransac_rotzoom_double_prec;
+    case TRANSLATION: return ransac_translation_double_prec;
+    default: assert(0); return NULL;
+  }
+}
+
+// Don't use points around the frame border since they are less reliable
+static INLINE int valid_point(int x, int y, int width, int height) {
+  return (x > (PATCH_SIZE + PATCH_CENTER)) &&
+         (x < (width - PATCH_SIZE - PATCH_CENTER)) &&
+         (y > (PATCH_SIZE + PATCH_CENTER)) &&
+         (y < (height - PATCH_SIZE - PATCH_CENTER));
+}
+
+static int determine_disflow_correspondence(int *frm_corners,
+                                            int num_frm_corners, double *flow_u,
+                                            double *flow_v, int width,
+                                            int height, int stride,
+                                            double *correspondences) {
+  int num_correspondences = 0;
+  int x, y;
+  for (int i = 0; i < num_frm_corners; ++i) {
+    x = frm_corners[2 * i];
+    y = frm_corners[2 * i + 1];
+    if (valid_point(x, y, width, height)) {
+      correspondences[4 * num_correspondences] = x;
+      correspondences[4 * num_correspondences + 1] = y;
+      correspondences[4 * num_correspondences + 2] = x + flow_u[y * stride + x];
+      correspondences[4 * num_correspondences + 3] = y + flow_v[y * stride + x];
+      num_correspondences++;
+    }
+  }
+  return num_correspondences;
+}
+
+double getCubicValue(double p[4], double x) {
+  return p[1] + 0.5 * x *
+                    (p[2] - p[0] +
+                     x * (2.0 * p[0] - 5.0 * p[1] + 4.0 * p[2] - p[3] +
+                          x * (3.0 * (p[1] - p[2]) + p[3] - p[0])));
+}
+
+void get_subcolumn(unsigned char *ref, double col[4], int stride, int x,
+                   int y_start) {
+  int i;
+  for (i = 0; i < 4; ++i) {
+    col[i] = ref[(i + y_start) * stride + x];
+  }
+}
+
+double bicubic(unsigned char *ref, double x, double y, int stride) {
+  double arr[4];
+  int k;
+  int i = (int)x;
+  int j = (int)y;
+  for (k = 0; k < 4; ++k) {
+    double arr_temp[4];
+    get_subcolumn(ref, arr_temp, stride, i + k - 1, j - 1);
+    arr[k] = getCubicValue(arr_temp, y - j);
+  }
+  return getCubicValue(arr, x - i);
+}
+
+// Interpolate a warped block using bicubic interpolation when possible
+unsigned char interpolate(unsigned char *ref, double x, double y, int width,
+                          int height, int stride) {
+  if (x < 0 && y < 0)
+    return ref[0];
+  else if (x < 0 && y > height - 1)
+    return ref[(height - 1) * stride];
+  else if (x > width - 1 && y < 0)
+    return ref[width - 1];
+  else if (x > width - 1 && y > height - 1)
+    return ref[(height - 1) * stride + (width - 1)];
+  else if (x < 0) {
+    int v;
+    int i = (int)y;
+    double a = y - i;
+    if (y > 1 && y < height - 2) {
+      double arr[4];
+      get_subcolumn(ref, arr, stride, 0, i - 1);
+      return clamp((int)(getCubicValue(arr, a) + 0.5), 0, 255);
+    }
+    v = (int)(ref[i * stride] * (1 - a) + ref[(i + 1) * stride] * a + 0.5);
+    return clamp(v, 0, 255);
+  } else if (y < 0) {
+    int v;
+    int j = (int)x;
+    double b = x - j;
+    if (x > 1 && x < width - 2) {
+      double arr[4] = { ref[j - 1], ref[j], ref[j + 1], ref[j + 2] };
+      return clamp((int)(getCubicValue(arr, b) + 0.5), 0, 255);
+    }
+    v = (int)(ref[j] * (1 - b) + ref[j + 1] * b + 0.5);
+    return clamp(v, 0, 255);
+  } else if (x > width - 1) {
+    int v;
+    int i = (int)y;
+    double a = y - i;
+    if (y > 1 && y < height - 2) {
+      double arr[4];
+      get_subcolumn(ref, arr, stride, width - 1, i - 1);
+      return clamp((int)(getCubicValue(arr, a) + 0.5), 0, 255);
+    }
+    v = (int)(ref[i * stride + width - 1] * (1 - a) +
+              ref[(i + 1) * stride + width - 1] * a + 0.5);
+    return clamp(v, 0, 255);
+  } else if (y > height - 1) {
+    int v;
+    int j = (int)x;
+    double b = x - j;
+    if (x > 1 && x < width - 2) {
+      int row = (height - 1) * stride;
+      double arr[4] = { ref[row + j - 1], ref[row + j], ref[row + j + 1],
+                        ref[row + j + 2] };
+      return clamp((int)(getCubicValue(arr, b) + 0.5), 0, 255);
+    }
+    v = (int)(ref[(height - 1) * stride + j] * (1 - b) +
+              ref[(height - 1) * stride + j + 1] * b + 0.5);
+    return clamp(v, 0, 255);
+  } else if (x > 1 && y > 1 && x < width - 2 && y < height - 2) {
+    return clamp((int)(bicubic(ref, x, y, stride) + 0.5), 0, 255);
+  } else {
+    int i = (int)y;
+    int j = (int)x;
+    double a = y - i;
+    double b = x - j;
+    int v = (int)(ref[i * stride + j] * (1 - a) * (1 - b) +
+                  ref[i * stride + j + 1] * (1 - a) * b +
+                  ref[(i + 1) * stride + j] * a * (1 - b) +
+                  ref[(i + 1) * stride + j + 1] * a * b);
+    return clamp(v, 0, 255);
+  }
+}
+
+// Warps a block using flow vector [u, v] and computes the mse
+double compute_warp_and_error(unsigned char *ref, unsigned char *frm, int width,
+                              int height, int stride, int x, int y, double u,
+                              double v, int16_t *dt) {
+  int i, j;
+  unsigned char warped;
+  double x_w, y_w;
+  double mse = 0;
+  int16_t err = 0;
+  for (i = y; i < y + PATCH_SIZE; ++i)
+    for (j = x; j < x + PATCH_SIZE; ++j) {
+      x_w = (double)j + u;
+      y_w = (double)i + v;
+      warped = interpolate(ref, x_w, y_w, width, height, stride);
+      err = warped - frm[j + i * stride];
+      mse += err * err;
+      dt[(i - y) * PATCH_SIZE + (j - x)] = err;
+    }
+
+  mse /= (PATCH_SIZE * PATCH_SIZE);
+  return mse;
+}
+
+// Computes the components of the system of equations used to solve for
+// a flow vector. This includes:
+// 1.) The hessian matrix for optical flow. This matrix is in the
+// form of:
+//
+//       M = |sum(dx * dx)  sum(dx * dy)|
+//           |sum(dx * dy)  sum(dy * dy)|
+//
+// 2.)   b = |sum(dx * dt)|
+//           |sum(dy * dt)|
+// Where the sums are computed over a square window of PATCH_SIZE.
+static INLINE void compute_flow_system(const double *dx, int dx_stride,
+                                       const double *dy, int dy_stride,
+                                       const int16_t *dt, int dt_stride,
+                                       double *M, double *b) {
+  for (int i = 0; i < PATCH_SIZE; i++) {
+    for (int j = 0; j < PATCH_SIZE; j++) {
+      M[0] += dx[i * dx_stride + j] * dx[i * dx_stride + j];
+      M[1] += dx[i * dx_stride + j] * dy[i * dy_stride + j];
+      M[3] += dy[i * dy_stride + j] * dy[i * dy_stride + j];
+
+      b[0] += dx[i * dx_stride + j] * dt[i * dt_stride + j];
+      b[1] += dy[i * dy_stride + j] * dt[i * dt_stride + j];
+    }
+  }
+
+  M[2] = M[1];
+}
+
+// Solves a general Mx = b where M is a 2x2 matrix and b is a 2x1 matrix
+static INLINE void solve_2x2_system(const double *M, const double *b,
+                                    double *output_vec) {
+  double M_0 = M[0];
+  double M_3 = M[3];
+  double det = (M_0 * M_3) - (M[1] * M[2]);
+  if (det < 1e-5) {
+    // Handle singular matrix
+    // TODO(sarahparker) compare results using pseudo inverse instead
+    M_0 += 1e-10;
+    M_3 += 1e-10;
+    det = (M_0 * M_3) - (M[1] * M[2]);
+  }
+  const double det_inv = 1 / det;
+  const double mult_b0 = det_inv * b[0];
+  const double mult_b1 = det_inv * b[1];
+  output_vec[0] = M_3 * mult_b0 - M[1] * mult_b1;
+  output_vec[1] = -M[2] * mult_b0 + M_0 * mult_b1;
+}
+
+static INLINE void image_difference(const uint8_t *src, int src_stride,
+                                    const uint8_t *ref, int ref_stride,
+                                    int16_t *dst, int dst_stride, int height,
+                                    int width) {
+  const int block_unit = 8;
+  // Take difference in 8x8 blocks to make use of optimized diff function
+  for (int i = 0; i < height; i += block_unit) {
+    for (int j = 0; j < width; j += block_unit) {
+      aom_subtract_block(block_unit, block_unit, dst + i * dst_stride + j,
+                         dst_stride, src + i * src_stride + j, src_stride,
+                         ref + i * ref_stride + j, ref_stride);
+    }
+  }
+}
+
+// Compute an image gradient using a sobel filter.
+// If dir == 1, compute the x gradient. If dir == 0, compute y. This function
+// assumes the images have been padded so that they can be processed in units
+// of 8.
+static INLINE void sobel_xy_image_gradient(const uint8_t *src, int src_stride,
+                                           double *dst, int dst_stride,
+                                           int height, int width, int dir) {
+  double norm = 1.0;
+  // TODO(sarahparker) experiment with doing this over larger block sizes
+  const int block_unit = 8;
+  // Filter in 8x8 blocks to eventually make use of optimized convolve function
+  for (int i = 0; i < height; i += block_unit) {
+    for (int j = 0; j < width; j += block_unit) {
+      av1_convolve_2d_sobel_y_c(src + i * src_stride + j, src_stride,
+                                dst + i * dst_stride + j, dst_stride,
+                                block_unit, block_unit, dir, norm);
+    }
+  }
+}
+
+static ImagePyramid *alloc_pyramid(int width, int height, int pad_size,
+                                   int compute_gradient) {
+  ImagePyramid *pyr = aom_malloc(sizeof(*pyr));
+  pyr->has_gradient = compute_gradient;
+  // 2 * width * height is the upper bound for a buffer that fits
+  // all pyramid levels + padding for each level
+  const int buffer_size = sizeof(*pyr->level_buffer) * 2 * width * height +
+                          (width + 2 * pad_size) * 2 * pad_size * N_LEVELS;
+  pyr->level_buffer = aom_malloc(buffer_size);
+  memset(pyr->level_buffer, 0, buffer_size);
+
+  if (compute_gradient) {
+    const int gradient_size =
+        sizeof(*pyr->level_dx_buffer) * 2 * width * height +
+        (width + 2 * pad_size) * 2 * pad_size * N_LEVELS;
+    pyr->level_dx_buffer = aom_malloc(gradient_size);
+    pyr->level_dy_buffer = aom_malloc(gradient_size);
+    memset(pyr->level_dx_buffer, 0, gradient_size);
+    memset(pyr->level_dy_buffer, 0, gradient_size);
+  }
+  return pyr;
+}
+
+static void free_pyramid(ImagePyramid *pyr) {
+  aom_free(pyr->level_buffer);
+  if (pyr->has_gradient) {
+    aom_free(pyr->level_dx_buffer);
+    aom_free(pyr->level_dy_buffer);
+  }
+  aom_free(pyr);
+}
+
+static INLINE void update_level_dims(ImagePyramid *frm_pyr, int level) {
+  frm_pyr->widths[level] = frm_pyr->widths[level - 1] >> 1;
+  frm_pyr->heights[level] = frm_pyr->heights[level - 1] >> 1;
+  frm_pyr->strides[level] = frm_pyr->widths[level] + 2 * frm_pyr->pad_size;
+  // Point the beginning of the next level buffer to the correct location inside
+  // the padded border
+  frm_pyr->level_loc[level] =
+      frm_pyr->level_loc[level - 1] +
+      frm_pyr->strides[level - 1] *
+          (2 * frm_pyr->pad_size + frm_pyr->heights[level - 1]);
+}
+
+// Compute coarse to fine pyramids for a frame
+static void compute_flow_pyramids(unsigned char *frm, const int frm_width,
+                                  const int frm_height, const int frm_stride,
+                                  int n_levels, int pad_size, int compute_grad,
+                                  ImagePyramid *frm_pyr) {
+  int cur_width, cur_height, cur_stride, cur_loc;
+  assert((frm_width >> n_levels) > 0);
+  assert((frm_height >> n_levels) > 0);
+
+  // Initialize first level
+  frm_pyr->n_levels = n_levels;
+  frm_pyr->pad_size = pad_size;
+  frm_pyr->widths[0] = frm_width;
+  frm_pyr->heights[0] = frm_height;
+  frm_pyr->strides[0] = frm_width + 2 * frm_pyr->pad_size;
+  // Point the beginning of the level buffer to the location inside
+  // the padded border
+  frm_pyr->level_loc[0] =
+      frm_pyr->strides[0] * frm_pyr->pad_size + frm_pyr->pad_size;
+  // This essentially copies the original buffer into the pyramid buffer
+  // without the original padding
+  av1_resize_plane(frm, frm_height, frm_width, frm_stride,
+                   frm_pyr->level_buffer + frm_pyr->level_loc[0],
+                   frm_pyr->heights[0], frm_pyr->widths[0],
+                   frm_pyr->strides[0]);
+
+  if (compute_grad) {
+    cur_width = frm_pyr->widths[0];
+    cur_height = frm_pyr->heights[0];
+    cur_stride = frm_pyr->strides[0];
+    cur_loc = frm_pyr->level_loc[0];
+    assert(frm_pyr->has_gradient && frm_pyr->level_dx_buffer != NULL &&
+           frm_pyr->level_dy_buffer != NULL);
+    // Computation x gradient
+    sobel_xy_image_gradient(frm_pyr->level_buffer + cur_loc, cur_stride,
+                            frm_pyr->level_dx_buffer + cur_loc, cur_stride,
+                            cur_height, cur_width, 1);
+
+    // Computation y gradient
+    sobel_xy_image_gradient(frm_pyr->level_buffer + cur_loc, cur_stride,
+                            frm_pyr->level_dy_buffer + cur_loc, cur_stride,
+                            cur_height, cur_width, 0);
+  }
+
+  // Start at the finest level and resize down to the coarsest level
+  for (int level = 1; level < n_levels; ++level) {
+    update_level_dims(frm_pyr, level);
+    cur_width = frm_pyr->widths[level];
+    cur_height = frm_pyr->heights[level];
+    cur_stride = frm_pyr->strides[level];
+    cur_loc = frm_pyr->level_loc[level];
+
+    av1_resize_plane(frm_pyr->level_buffer + frm_pyr->level_loc[level - 1],
+                     frm_pyr->heights[level - 1], frm_pyr->widths[level - 1],
+                     frm_pyr->strides[level - 1],
+                     frm_pyr->level_buffer + cur_loc, cur_height, cur_width,
+                     cur_stride);
+
+    if (compute_grad) {
+      assert(frm_pyr->has_gradient && frm_pyr->level_dx_buffer != NULL &&
+             frm_pyr->level_dy_buffer != NULL);
+      // Computation x gradient
+      sobel_xy_image_gradient(frm_pyr->level_buffer + cur_loc, cur_stride,
+                              frm_pyr->level_dx_buffer + cur_loc, cur_stride,
+                              cur_height, cur_width, 1);
+
+      // Computation y gradient
+      sobel_xy_image_gradient(frm_pyr->level_buffer + cur_loc, cur_stride,
+                              frm_pyr->level_dy_buffer + cur_loc, cur_stride,
+                              cur_height, cur_width, 0);
+    }
+  }
+}
+
+static INLINE void compute_flow_at_point(unsigned char *frm, unsigned char *ref,
+                                         double *dx, double *dy, int x, int y,
+                                         int width, int height, int stride,
+                                         double *u, double *v) {
+  double M[4] = { 0 };
+  double b[2] = { 0 };
+  double tmp_output_vec[2] = { 0 };
+  double error = 0;
+  int16_t dt[PATCH_SIZE * PATCH_SIZE];
+  double o_u = *u;
+  double o_v = *v;
+
+  for (int itr = 0; itr < DISFLOW_MAX_ITR; itr++) {
+    error = compute_warp_and_error(ref, frm, width, height, stride, x, y, *u,
+                                   *v, dt);
+    if (error <= DISFLOW_ERROR_TR) break;
+    compute_flow_system(dx, stride, dy, stride, dt, PATCH_SIZE, M, b);
+    solve_2x2_system(M, b, tmp_output_vec);
+    *u += tmp_output_vec[0];
+    *v += tmp_output_vec[1];
+  }
+  if (fabs(*u - o_u) > PATCH_SIZE || fabs(*v - o_u) > PATCH_SIZE) {
+    *u = o_u;
+    *v = o_v;
+  }
+}
+
+// make sure flow_u and flow_v start at 0
+static void compute_flow_field(ImagePyramid *frm_pyr, ImagePyramid *ref_pyr,
+                               double *flow_u, double *flow_v) {
+  int cur_width, cur_height, cur_stride, cur_loc, patch_loc, patch_center;
+  double *u_upscale =
+      aom_malloc(frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_u));
+  double *v_upscale =
+      aom_malloc(frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_v));
+
+  assert(frm_pyr->n_levels == ref_pyr->n_levels);
+
+  // Compute flow field from coarsest to finest level of the pyramid
+  for (int level = frm_pyr->n_levels - 1; level >= 0; --level) {
+    cur_width = frm_pyr->widths[level];
+    cur_height = frm_pyr->heights[level];
+    cur_stride = frm_pyr->strides[level];
+    cur_loc = frm_pyr->level_loc[level];
+
+    for (int i = PATCH_SIZE; i < cur_height - PATCH_SIZE; i += PATCH_STEP) {
+      for (int j = PATCH_SIZE; j < cur_width - PATCH_SIZE; j += PATCH_STEP) {
+        patch_loc = i * cur_stride + j;
+        patch_center = patch_loc + PATCH_CENTER * cur_stride + PATCH_CENTER;
+        compute_flow_at_point(frm_pyr->level_buffer + cur_loc,
+                              ref_pyr->level_buffer + cur_loc,
+                              frm_pyr->level_dx_buffer + cur_loc + patch_loc,
+                              frm_pyr->level_dy_buffer + cur_loc + patch_loc, j,
+                              i, cur_width, cur_height, cur_stride,
+                              flow_u + patch_center, flow_v + patch_center);
+      }
+    }
+    // TODO(sarahparker) Replace this with upscale function in resize.c
+    if (level > 0) {
+      int h_upscale = frm_pyr->heights[level - 1];
+      int w_upscale = frm_pyr->widths[level - 1];
+      int s_upscale = frm_pyr->strides[level - 1];
+      for (int i = 0; i < h_upscale; ++i) {
+        for (int j = 0; j < w_upscale; ++j) {
+          u_upscale[j + i * s_upscale] =
+              flow_u[(int)(j >> 1) + (int)(i >> 1) * cur_stride];
+          v_upscale[j + i * s_upscale] =
+              flow_v[(int)(j >> 1) + (int)(i >> 1) * cur_stride];
+        }
+      }
+      memcpy(flow_u, u_upscale,
+             frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_u));
+      memcpy(flow_v, v_upscale,
+             frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_v));
+    }
+  }
+  aom_free(u_upscale);
+  aom_free(v_upscale);
+}
+
+static int compute_global_motion_disflow_based(
+    TransformationType type, YV12_BUFFER_CONFIG *frm, YV12_BUFFER_CONFIG *ref,
+    int bit_depth, int *num_inliers_by_motion, double *params_by_motion,
+    int num_motions) {
+  unsigned char *frm_buffer = frm->y_buffer;
+  unsigned char *ref_buffer = ref->y_buffer;
+  const int frm_width = frm->y_width;
+  const int frm_height = frm->y_height;
+  const int ref_width = ref->y_width;
+  const int ref_height = ref->y_height;
+  const int pad_size = AOMMAX(PATCH_SIZE, MIN_PAD);
+  int num_frm_corners;
+  int num_correspondences;
+  double *correspondences;
+  int frm_corners[2 * MAX_CORNERS];
+  RansacFuncDouble ransac = get_ransac_double_prec_type(type);
+  assert(frm_width == ref_width);
+  assert(frm_height == ref_height);
+
+  // Ensure the number of pyramid levels will work with the frame resolution
+  const int msb =
+      frm_width < frm_height ? get_msb(frm_width) : get_msb(frm_height);
+  const int n_levels = AOMMIN(msb, N_LEVELS);
+
+  if (frm->flags & YV12_FLAG_HIGHBITDEPTH) {
+    // The frame buffer is 16-bit, so we need to convert to 8 bits for the
+    // following code. We cache the result until the frame is released.
+    frm_buffer = downconvert_frame(frm, bit_depth);
+  }
+  if (ref->flags & YV12_FLAG_HIGHBITDEPTH) {
+    ref_buffer = downconvert_frame(ref, bit_depth);
+  }
+
+  // TODO(sarahparker) We will want to do the source pyramid computation
+  // outside of this function so it doesn't get recomputed for every
+  // reference. We also don't need to compute every pyramid level for the
+  // reference in advance, since lower levels can be overwritten once their
+  // flow field is computed and upscaled. I'll add these optimizations
+  // once the full implementation is working.
+  // Allocate frm image pyramids
+  int compute_gradient = 1;
+  ImagePyramid *frm_pyr =
+      alloc_pyramid(frm_width, frm_height, pad_size, compute_gradient);
+  compute_flow_pyramids(frm_buffer, frm_width, frm_height, frm->y_stride,
+                        n_levels, pad_size, compute_gradient, frm_pyr);
+  // Allocate ref image pyramids
+  compute_gradient = 0;
+  ImagePyramid *ref_pyr =
+      alloc_pyramid(ref_width, ref_height, pad_size, compute_gradient);
+  compute_flow_pyramids(ref_buffer, ref_width, ref_height, ref->y_stride,
+                        n_levels, pad_size, compute_gradient, ref_pyr);
+
+  double *flow_u =
+      aom_malloc(frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_u));
+  double *flow_v =
+      aom_malloc(frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_v));
+
+  memset(flow_u, 0,
+         frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_u));
+  memset(flow_v, 0,
+         frm_pyr->strides[0] * frm_pyr->heights[0] * sizeof(*flow_v));
+
+  compute_flow_field(frm_pyr, ref_pyr, flow_u, flow_v);
+
+  // compute interest points in images using FAST features
+  num_frm_corners = fast_corner_detect(frm_buffer, frm_width, frm_height,
+                                       frm->y_stride, frm_corners, MAX_CORNERS);
+  // find correspondences between the two images using the flow field
+  correspondences = aom_malloc(num_frm_corners * 4 * sizeof(*correspondences));
+  num_correspondences = determine_disflow_correspondence(
+      frm_corners, num_frm_corners, flow_u, flow_v, frm_width, frm_height,
+      frm_pyr->strides[0], correspondences);
+  ransac(correspondences, num_correspondences, num_inliers_by_motion,
+         params_by_motion, num_motions);
+
+  free_pyramid(frm_pyr);
+  free_pyramid(ref_pyr);
+  aom_free(correspondences);
+  aom_free(flow_u);
+  aom_free(flow_v);
+  // Set num_inliers = 0 for motions with too few inliers so they are ignored.
+  for (int i = 0; i < num_motions; ++i) {
+    if (num_inliers_by_motion[i] < MIN_INLIER_PROB * num_correspondences) {
+      num_inliers_by_motion[i] = 0;
+    }
+  }
+
+  // Return true if any one of the motions has inliers.
+  for (int i = 0; i < num_motions; ++i) {
+    if (num_inliers_by_motion[i] > 0) return 1;
+  }
+  return 0;
+}
+#endif
+
+int av1_compute_global_motion(TransformationType type, YV12_BUFFER_CONFIG *frm,
+                              YV12_BUFFER_CONFIG *ref, int bit_depth,
+                              int *num_inliers_by_motion,
+                              double *params_by_motion, int num_motions) {
+#if USE_GM_FEATURE_BASED
+  return compute_global_motion_feature_based(type, frm, ref, bit_depth,
+                                             num_inliers_by_motion,
+                                             params_by_motion, num_motions);
+#else
+  return compute_global_motion_disflow_based(type, frm, ref, bit_depth,
+                                             num_inliers_by_motion,
+                                             params_by_motion, num_motions);
+#endif
+}
diff --git a/third_party/aom/av1/encoder/global_motion.h b/third_party/aom/av1/encoder/global_motion.h
index c7c016c430..42cf221697 100644
--- a/third_party/aom/av1/encoder/global_motion.h
+++ b/third_party/aom/av1/encoder/global_motion.h
@@ -22,20 +22,22 @@ extern "C" {
 
 #define RANSAC_NUM_MOTIONS 1
 
-void convert_model_to_params(const double *params, WarpedMotionParams *model);
+void av1_convert_model_to_params(const double *params,
+                                 WarpedMotionParams *model);
 
-int is_enough_erroradvantage(double best_erroradvantage, int params_cost,
-                             int erroradv_type);
+int av1_is_enough_erroradvantage(double best_erroradvantage, int params_cost,
+                                 int erroradv_type);
 
 // Returns the av1_warp_error between "dst" and the result of applying the
 // motion params that result from fine-tuning "wm" to "ref". Note that "wm" is
 // modified in place.
-int64_t refine_integerized_param(WarpedMotionParams *wm,
-                                 TransformationType wmtype, int use_hbd, int bd,
-                                 uint8_t *ref, int r_width, int r_height,
-                                 int r_stride, uint8_t *dst, int d_width,
-                                 int d_height, int d_stride, int n_refinements,
-                                 int64_t best_frame_error);
+int64_t av1_refine_integerized_param(WarpedMotionParams *wm,
+                                     TransformationType wmtype, int use_hbd,
+                                     int bd, uint8_t *ref, int r_width,
+                                     int r_height, int r_stride, uint8_t *dst,
+                                     int d_width, int d_height, int d_stride,
+                                     int n_refinements,
+                                     int64_t best_frame_error);
 
 /*
   Computes "num_motions" candidate global motion parameters between two frames.
@@ -52,12 +54,10 @@ int64_t refine_integerized_param(WarpedMotionParams *wm,
   number of inlier feature points for each motion. Params for which the
   num_inliers entry is 0 should be ignored by the caller.
 */
-int compute_global_motion_feature_based(TransformationType type,
-                                        YV12_BUFFER_CONFIG *frm,
-                                        YV12_BUFFER_CONFIG *ref, int bit_depth,
-                                        int *num_inliers_by_motion,
-                                        double *params_by_motion,
-                                        int num_motions);
+int av1_compute_global_motion(TransformationType type, YV12_BUFFER_CONFIG *frm,
+                              YV12_BUFFER_CONFIG *ref, int bit_depth,
+                              int *num_inliers_by_motion,
+                              double *params_by_motion, int num_motions);
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/hash_motion.c b/third_party/aom/av1/encoder/hash_motion.c
index e85a516e85..4c872f15aa 100644
--- a/third_party/aom/av1/encoder/hash_motion.c
+++ b/third_party/aom/av1/encoder/hash_motion.c
@@ -147,7 +147,8 @@ static void hash_table_add_to_table(hash_table *p_hash_table,
   }
 }
 
-int32_t av1_hash_table_count(hash_table *p_hash_table, uint32_t hash_value) {
+int32_t av1_hash_table_count(const hash_table *p_hash_table,
+                             uint32_t hash_value) {
   if (p_hash_table->p_lookup_table[hash_value] == NULL) {
     return 0;
   } else {
diff --git a/third_party/aom/av1/encoder/hash_motion.h b/third_party/aom/av1/encoder/hash_motion.h
index df3ec32158..ed9bb6e52d 100644
--- a/third_party/aom/av1/encoder/hash_motion.h
+++ b/third_party/aom/av1/encoder/hash_motion.h
@@ -37,7 +37,8 @@ typedef struct _hash_table {
 void av1_hash_table_init(hash_table *p_hash_table, struct macroblock *x);
 void av1_hash_table_destroy(hash_table *p_hash_table);
 void av1_hash_table_create(hash_table *p_hash_table);
-int32_t av1_hash_table_count(hash_table *p_hash_table, uint32_t hash_value);
+int32_t av1_hash_table_count(const hash_table *p_hash_table,
+                             uint32_t hash_value);
 Iterator av1_hash_get_first_iterator(hash_table *p_hash_table,
                                      uint32_t hash_value);
 int32_t av1_has_exact_match(hash_table *p_hash_table, uint32_t hash_value1,
diff --git a/third_party/aom/av1/encoder/hybrid_fwd_txfm.c b/third_party/aom/av1/encoder/hybrid_fwd_txfm.c
index 67898fd184..06990857a0 100644
--- a/third_party/aom/av1/encoder/hybrid_fwd_txfm.c
+++ b/third_party/aom/av1/encoder/hybrid_fwd_txfm.c
@@ -87,35 +87,21 @@ static void highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
     av1_highbd_fwht4x4(src_diff, coeff, diff_stride);
     return;
   }
-  switch (tx_type) {
-    // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_4x4_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_4x4(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_4x4(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_4x8(const int16_t *src_diff, tran_low_t *coeff,
                                 int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_4x8_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                       txfm_param->bd);
+  av1_fwd_txfm2d_4x8(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                     txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_8x4(const int16_t *src_diff, tran_low_t *coeff,
                                 int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_8x4_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                       txfm_param->bd);
+  av1_fwd_txfm2d_8x4(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                     txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_8x16(const int16_t *src_diff, tran_low_t *coeff,
@@ -123,21 +109,7 @@ static void highbd_fwd_txfm_8x16(const int16_t *src_diff, tran_low_t *coeff,
   int32_t *dst_coeff = (int32_t *)coeff;
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int bd = txfm_param->bd;
-  switch (tx_type) {
-      // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_8x16_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_8x16(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_8x16(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_16x8(const int16_t *src_diff, tran_low_t *coeff,
@@ -145,63 +117,49 @@ static void highbd_fwd_txfm_16x8(const int16_t *src_diff, tran_low_t *coeff,
   int32_t *dst_coeff = (int32_t *)coeff;
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int bd = txfm_param->bd;
-  switch (tx_type) {
-      // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_16x8_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_16x8(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_16x8(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_16x32(const int16_t *src_diff, tran_low_t *coeff,
                                   int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_16x32_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                         txfm_param->bd);
+  av1_fwd_txfm2d_16x32(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                       txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_32x16(const int16_t *src_diff, tran_low_t *coeff,
                                   int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_32x16_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                         txfm_param->bd);
+  av1_fwd_txfm2d_32x16(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                       txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_16x4(const int16_t *src_diff, tran_low_t *coeff,
                                  int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_16x4_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                        txfm_param->bd);
+  av1_fwd_txfm2d_16x4(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                      txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_4x16(const int16_t *src_diff, tran_low_t *coeff,
                                  int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_4x16_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                        txfm_param->bd);
+  av1_fwd_txfm2d_4x16(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                      txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_32x8(const int16_t *src_diff, tran_low_t *coeff,
                                  int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_32x8_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                        txfm_param->bd);
+  av1_fwd_txfm2d_32x8(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                      txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_8x32(const int16_t *src_diff, tran_low_t *coeff,
                                  int diff_stride, TxfmParam *txfm_param) {
   int32_t *dst_coeff = (int32_t *)coeff;
-  av1_fwd_txfm2d_8x32_c(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
-                        txfm_param->bd);
+  av1_fwd_txfm2d_8x32(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                      txfm_param->bd);
 }
 
 static void highbd_fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
@@ -209,21 +167,7 @@ static void highbd_fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
   int32_t *dst_coeff = (int32_t *)coeff;
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int bd = txfm_param->bd;
-  switch (tx_type) {
-    // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_8x8_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_8x8(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_8x8(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
@@ -231,21 +175,7 @@ static void highbd_fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
   int32_t *dst_coeff = (int32_t *)coeff;
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int bd = txfm_param->bd;
-  switch (tx_type) {
-    // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_16x16_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_16x16(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_16x16(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_32x32(const int16_t *src_diff, tran_low_t *coeff,
@@ -253,21 +183,7 @@ static void highbd_fwd_txfm_32x32(const int16_t *src_diff, tran_low_t *coeff,
   int32_t *dst_coeff = (int32_t *)coeff;
   const TX_TYPE tx_type = txfm_param->tx_type;
   const int bd = txfm_param->bd;
-  switch (tx_type) {
-    // use the c version for anything including identity for now
-    case V_DCT:
-    case H_DCT:
-    case V_ADST:
-    case H_ADST:
-    case V_FLIPADST:
-    case H_FLIPADST:
-    case IDTX:
-      av1_fwd_txfm2d_32x32_c(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-    default:
-      av1_fwd_txfm2d_32x32(src_diff, dst_coeff, diff_stride, tx_type, bd);
-      break;
-  }
+  av1_fwd_txfm2d_32x32(src_diff, dst_coeff, diff_stride, tx_type, bd);
 }
 
 static void highbd_fwd_txfm_32x64(const int16_t *src_diff, tran_low_t *coeff,
@@ -275,7 +191,8 @@ static void highbd_fwd_txfm_32x64(const int16_t *src_diff, tran_low_t *coeff,
   assert(txfm_param->tx_type == DCT_DCT);
   int32_t *dst_coeff = (int32_t *)coeff;
   const int bd = txfm_param->bd;
-  av1_fwd_txfm2d_32x64_c(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
+  av1_fwd_txfm2d_32x64(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                       bd);
 }
 
 static void highbd_fwd_txfm_64x32(const int16_t *src_diff, tran_low_t *coeff,
@@ -283,7 +200,8 @@ static void highbd_fwd_txfm_64x32(const int16_t *src_diff, tran_low_t *coeff,
   assert(txfm_param->tx_type == DCT_DCT);
   int32_t *dst_coeff = (int32_t *)coeff;
   const int bd = txfm_param->bd;
-  av1_fwd_txfm2d_64x32_c(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
+  av1_fwd_txfm2d_64x32(src_diff, dst_coeff, diff_stride, txfm_param->tx_type,
+                       bd);
 }
 
 static void highbd_fwd_txfm_16x64(const int16_t *src_diff, tran_low_t *coeff,
@@ -291,7 +209,7 @@ static void highbd_fwd_txfm_16x64(const int16_t *src_diff, tran_low_t *coeff,
   assert(txfm_param->tx_type == DCT_DCT);
   int32_t *dst_coeff = (int32_t *)coeff;
   const int bd = txfm_param->bd;
-  av1_fwd_txfm2d_16x64_c(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
+  av1_fwd_txfm2d_16x64(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
 }
 
 static void highbd_fwd_txfm_64x16(const int16_t *src_diff, tran_low_t *coeff,
@@ -299,7 +217,7 @@ static void highbd_fwd_txfm_64x16(const int16_t *src_diff, tran_low_t *coeff,
   assert(txfm_param->tx_type == DCT_DCT);
   int32_t *dst_coeff = (int32_t *)coeff;
   const int bd = txfm_param->bd;
-  av1_fwd_txfm2d_64x16_c(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
+  av1_fwd_txfm2d_64x16(src_diff, dst_coeff, diff_stride, DCT_DCT, bd);
 }
 
 static void highbd_fwd_txfm_64x64(const int16_t *src_diff, tran_low_t *coeff,
diff --git a/third_party/aom/av1/encoder/lookahead.c b/third_party/aom/av1/encoder/lookahead.c
index 1bf8ecbacb..fe3bc20512 100644
--- a/third_party/aom/av1/encoder/lookahead.c
+++ b/third_party/aom/av1/encoder/lookahead.c
@@ -43,7 +43,8 @@ void av1_lookahead_destroy(struct lookahead_ctx *ctx) {
 
 struct lookahead_ctx *av1_lookahead_init(
     unsigned int width, unsigned int height, unsigned int subsampling_x,
-    unsigned int subsampling_y, int use_highbitdepth, unsigned int depth) {
+    unsigned int subsampling_y, int use_highbitdepth, unsigned int depth,
+    const int border_in_pixels) {
   struct lookahead_ctx *ctx = NULL;
 
   // Clamp the lookahead queue depth
@@ -63,7 +64,7 @@ struct lookahead_ctx *av1_lookahead_init(
     for (i = 0; i < depth; i++)
       if (aom_alloc_frame_buffer(&ctx->buf[i].img, width, height, subsampling_x,
                                  subsampling_y, use_highbitdepth,
-                                 AOM_BORDER_IN_PIXELS, legacy_byte_alignment))
+                                 border_in_pixels, legacy_byte_alignment))
         goto bail;
   }
   return ctx;
diff --git a/third_party/aom/av1/encoder/lookahead.h b/third_party/aom/av1/encoder/lookahead.h
index e55224cf75..e5d9702443 100644
--- a/third_party/aom/av1/encoder/lookahead.h
+++ b/third_party/aom/av1/encoder/lookahead.h
@@ -46,7 +46,8 @@ struct lookahead_ctx {
  */
 struct lookahead_ctx *av1_lookahead_init(
     unsigned int width, unsigned int height, unsigned int subsampling_x,
-    unsigned int subsampling_y, int use_highbitdepth, unsigned int depth);
+    unsigned int subsampling_y, int use_highbitdepth, unsigned int depth,
+    const int border_in_pixels);
 
 /**\brief Destroys the lookahead stage
  */
diff --git a/third_party/aom/av1/encoder/mbgraph.c b/third_party/aom/av1/encoder/mbgraph.c
index 1a35ff77c7..0cb628682f 100644
--- a/third_party/aom/av1/encoder/mbgraph.c
+++ b/third_party/aom/av1/encoder/mbgraph.c
@@ -60,7 +60,7 @@ static unsigned int do_16x16_motion_iteration(AV1_COMP *cpi, const MV *ref_mv,
         x, &cpi->common, mb_row, mb_col, ref_mv,
         cpi->common.allow_high_precision_mv, x->errorperbit, &v_fn_ptr, 0,
         mv_sf->subpel_iters_per_step, cond_cost_list(cpi, cost_list), NULL,
-        NULL, &distortion, &sse, NULL, NULL, 0, 0, 0, 0, 0);
+        NULL, &distortion, &sse, NULL, NULL, 0, 0, 0, 0, 0, 1);
   }
 
   if (has_second_ref(xd->mi[0]))
@@ -71,8 +71,8 @@ static unsigned int do_16x16_motion_iteration(AV1_COMP *cpi, const MV *ref_mv,
   xd->mi[0]->mv[0] = x->best_mv;
   xd->mi[0]->ref_frame[1] = NONE_FRAME;
 
-  av1_build_inter_predictors_sby(&cpi->common, xd, mb_row, mb_col, NULL,
-                                 BLOCK_16X16);
+  av1_enc_build_inter_predictor(&cpi->common, xd, mb_row, mb_col, NULL,
+                                BLOCK_16X16, AOM_PLANE_Y, AOM_PLANE_Y);
 
   /* restore UMV window */
   x->mv_limits = tmp_mv_limits;
@@ -179,8 +179,8 @@ static void update_mbgraph_mb_stats(AV1_COMP *cpi, MBGRAPH_MB_STATS *stats,
   x->plane[0].src.buf = buf->y_buffer + mb_y_offset;
   x->plane[0].src.stride = buf->y_stride;
 
-  xd->plane[0].dst.buf = get_frame_new_buffer(cm)->y_buffer + mb_y_offset;
-  xd->plane[0].dst.stride = get_frame_new_buffer(cm)->y_stride;
+  xd->plane[0].dst.buf = cm->cur_frame->buf.y_buffer + mb_y_offset;
+  xd->plane[0].dst.stride = cm->cur_frame->buf.y_stride;
 
   // do intra 16x16 prediction
   intra_error = find_best_16x16_intra(cpi, &stats->ref[INTRA_FRAME].m.mode);
@@ -364,7 +364,7 @@ static void separate_arf_mbs(AV1_COMP *cpi) {
 void av1_update_mbgraph_stats(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
   int i, n_frames = av1_lookahead_depth(cpi->lookahead);
-  YV12_BUFFER_CONFIG *golden_ref = get_ref_frame_buffer(cpi, GOLDEN_FRAME);
+  YV12_BUFFER_CONFIG *golden_ref = &get_ref_frame_buf(cm, GOLDEN_FRAME)->buf;
 
   assert(golden_ref != NULL);
 
diff --git a/third_party/aom/av1/encoder/mcomp.c b/third_party/aom/av1/encoder/mcomp.c
index 8f6de9b532..1bab45c7f1 100644
--- a/third_party/aom/av1/encoder/mcomp.c
+++ b/third_party/aom/av1/encoder/mcomp.c
@@ -112,7 +112,7 @@ static int mvsad_err_cost(const MACROBLOCK *x, const MV *mv, const MV *ref,
                           int sad_per_bit) {
   const MV diff = { (mv->row - ref->row) * 8, (mv->col - ref->col) * 8 };
   return ROUND_POWER_OF_TWO(
-      (unsigned)mv_cost(&diff, x->nmvjointcost, x->mvcost) * sad_per_bit,
+      (unsigned)mv_cost(&diff, x->nmv_vec_cost, x->mv_cost_stack) * sad_per_bit,
       AV1_PROB_COST_SHIFT);
 }
 
@@ -179,36 +179,31 @@ static INLINE const uint8_t *pre(const uint8_t *buf, int stride, int r, int c) {
 }
 
 /* checks if (r, c) has better score than previous best */
-#define CHECK_BETTER(v, r, c)                                                \
-  if (c >= minc && c <= maxc && r >= minr && r <= maxr) {                    \
-    MV this_mv = { r, c };                                                   \
-    v = mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);       \
-    if (second_pred == NULL) {                                               \
-      thismse = vfp->svf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r),     \
-                         src_address, src_stride, &sse);                     \
-    } else if (mask) {                                                       \
-      thismse = vfp->msvf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r),    \
-                          src_address, src_stride, second_pred, mask,        \
-                          mask_stride, invert_mask, &sse);                   \
-    } else {                                                                 \
-      if (xd->jcp_param.use_jnt_comp_avg)                                    \
-        thismse = vfp->jsvaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
-                             src_address, src_stride, &sse, second_pred,     \
-                             &xd->jcp_param);                                \
-      else                                                                   \
-        thismse = vfp->svaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r),  \
-                            src_address, src_stride, &sse, second_pred);     \
-    }                                                                        \
-    v += thismse;                                                            \
-    if (v < besterr) {                                                       \
-      besterr = v;                                                           \
-      br = r;                                                                \
-      bc = c;                                                                \
-      *distortion = thismse;                                                 \
-      *sse1 = sse;                                                           \
-    }                                                                        \
-  } else {                                                                   \
-    v = INT_MAX;                                                             \
+#define CHECK_BETTER(v, r, c)                                             \
+  if (c >= minc && c <= maxc && r >= minr && r <= maxr) {                 \
+    MV this_mv = { r, c };                                                \
+    v = mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);    \
+    if (second_pred == NULL) {                                            \
+      thismse = vfp->svf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r),  \
+                         src_address, src_stride, &sse);                  \
+    } else if (mask) {                                                    \
+      thismse = vfp->msvf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
+                          src_address, src_stride, second_pred, mask,     \
+                          mask_stride, invert_mask, &sse);                \
+    } else {                                                              \
+      thismse = vfp->svaf(pre(y, y_stride, r, c), y_stride, sp(c), sp(r), \
+                          src_address, src_stride, &sse, second_pred);    \
+    }                                                                     \
+    v += thismse;                                                         \
+    if (v < besterr) {                                                    \
+      besterr = v;                                                        \
+      br = r;                                                             \
+      bc = c;                                                             \
+      *distortion = thismse;                                              \
+      *sse1 = sse;                                                        \
+    }                                                                     \
+  } else {                                                                \
+    v = INT_MAX;                                                          \
   }
 
 #define CHECK_BETTER0(v, r, c) CHECK_BETTER(v, r, c)
@@ -341,19 +336,15 @@ static unsigned int setup_center_error(
     int *mvcost[2], unsigned int *sse1, int *distortion) {
   unsigned int besterr;
   if (second_pred != NULL) {
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       DECLARE_ALIGNED(16, uint16_t, comp_pred16[MAX_SB_SQUARE]);
       uint8_t *comp_pred = CONVERT_TO_BYTEPTR(comp_pred16);
       if (mask) {
         aom_highbd_comp_mask_pred(comp_pred, second_pred, w, h, y + offset,
                                   y_stride, mask, mask_stride, invert_mask);
       } else {
-        if (xd->jcp_param.use_jnt_comp_avg)
-          aom_highbd_jnt_comp_avg_pred(comp_pred, second_pred, w, h, y + offset,
-                                       y_stride, &xd->jcp_param);
-        else
-          aom_highbd_comp_avg_pred(comp_pred, second_pred, w, h, y + offset,
-                                   y_stride);
+        aom_highbd_comp_avg_pred(comp_pred, second_pred, w, h, y + offset,
+                                 y_stride);
       }
       besterr = vfp->vf(comp_pred, w, src, src_stride, sse1);
     } else {
@@ -362,11 +353,7 @@ static unsigned int setup_center_error(
         aom_comp_mask_pred(comp_pred, second_pred, w, h, y + offset, y_stride,
                            mask, mask_stride, invert_mask);
       } else {
-        if (xd->jcp_param.use_jnt_comp_avg)
-          aom_jnt_comp_avg_pred(comp_pred, second_pred, w, h, y + offset,
-                                y_stride, &xd->jcp_param);
-        else
-          aom_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
+        aom_comp_avg_pred(comp_pred, second_pred, w, h, y + offset, y_stride);
       }
       besterr = vfp->vf(comp_pred, w, src, src_stride, sse1);
     }
@@ -409,7 +396,7 @@ int av1_find_best_sub_pixel_tree_pruned_evenmore(
     int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
     unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
     int mask_stride, int invert_mask, int w, int h,
-    int use_accurate_subpel_search) {
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   SETUP_SUBPEL_SEARCH;
   besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                                src_address, src_stride, y, y_stride,
@@ -426,6 +413,7 @@ int av1_find_best_sub_pixel_tree_pruned_evenmore(
   (void)cm;
   (void)mi_row;
   (void)mi_col;
+  (void)do_reset_fractional_mv;
 
   if (cost_list && cost_list[0] != INT_MAX && cost_list[1] != INT_MAX &&
       cost_list[2] != INT_MAX && cost_list[3] != INT_MAX &&
@@ -481,12 +469,13 @@ int av1_find_best_sub_pixel_tree_pruned_more(
     int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
     unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
     int mask_stride, int invert_mask, int w, int h,
-    int use_accurate_subpel_search) {
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   SETUP_SUBPEL_SEARCH;
   (void)use_accurate_subpel_search;
   (void)cm;
   (void)mi_row;
   (void)mi_col;
+  (void)do_reset_fractional_mv;
 
   besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                                src_address, src_stride, y, y_stride,
@@ -549,12 +538,13 @@ int av1_find_best_sub_pixel_tree_pruned(
     int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
     unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
     int mask_stride, int invert_mask, int w, int h,
-    int use_accurate_subpel_search) {
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   SETUP_SUBPEL_SEARCH;
   (void)use_accurate_subpel_search;
   (void)cm;
   (void)mi_row;
   (void)mi_col;
+  (void)do_reset_fractional_mv;
 
   besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
                                src_address, src_stride, y, y_stride,
@@ -651,7 +641,7 @@ static int upsampled_pref_error(MACROBLOCKD *xd, const AV1_COMMON *const cm,
                                 int mask_stride, int invert_mask, int w, int h,
                                 unsigned int *sse, int subpel_search) {
   unsigned int besterr;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     DECLARE_ALIGNED(16, uint16_t, pred16[MAX_SB_SQUARE]);
     uint8_t *pred8 = CONVERT_TO_BYTEPTR(pred16);
     if (second_pred != NULL) {
@@ -661,14 +651,9 @@ static int upsampled_pref_error(MACROBLOCKD *xd, const AV1_COMMON *const cm,
             subpel_y_q3, y, y_stride, mask, mask_stride, invert_mask, xd->bd,
             subpel_search);
       } else {
-        if (xd->jcp_param.use_jnt_comp_avg)
-          aom_highbd_jnt_comp_avg_upsampled_pred(
-              xd, cm, mi_row, mi_col, mv, pred8, second_pred, w, h, subpel_x_q3,
-              subpel_y_q3, y, y_stride, xd->bd, &xd->jcp_param, subpel_search);
-        else
-          aom_highbd_comp_avg_upsampled_pred(
-              xd, cm, mi_row, mi_col, mv, pred8, second_pred, w, h, subpel_x_q3,
-              subpel_y_q3, y, y_stride, xd->bd, subpel_search);
+        aom_highbd_comp_avg_upsampled_pred(
+            xd, cm, mi_row, mi_col, mv, pred8, second_pred, w, h, subpel_x_q3,
+            subpel_y_q3, y, y_stride, xd->bd, subpel_search);
       }
     } else {
       aom_highbd_upsampled_pred(xd, cm, mi_row, mi_col, mv, pred8, w, h,
@@ -685,14 +670,9 @@ static int upsampled_pref_error(MACROBLOCKD *xd, const AV1_COMMON *const cm,
                                      subpel_y_q3, y, y_stride, mask,
                                      mask_stride, invert_mask, subpel_search);
       } else {
-        if (xd->jcp_param.use_jnt_comp_avg)
-          aom_jnt_comp_avg_upsampled_pred(
-              xd, cm, mi_row, mi_col, mv, pred, second_pred, w, h, subpel_x_q3,
-              subpel_y_q3, y, y_stride, &xd->jcp_param, subpel_search);
-        else
-          aom_comp_avg_upsampled_pred(xd, cm, mi_row, mi_col, mv, pred,
-                                      second_pred, w, h, subpel_x_q3,
-                                      subpel_y_q3, y, y_stride, subpel_search);
+        aom_comp_avg_upsampled_pred(xd, cm, mi_row, mi_col, mv, pred,
+                                    second_pred, w, h, subpel_x_q3, subpel_y_q3,
+                                    y, y_stride, subpel_search);
       }
     } else {
       aom_upsampled_pred(xd, cm, mi_row, mi_col, mv, pred, w, h, subpel_x_q3,
@@ -723,7 +703,7 @@ static unsigned int upsampled_setup_center_error(
 
 // when use_accurate_subpel_search == 0
 static INLINE unsigned int estimate_upsampled_pref_error(
-    MACROBLOCKD *xd, const aom_variance_fn_ptr_t *vfp, const uint8_t *const src,
+    const aom_variance_fn_ptr_t *vfp, const uint8_t *const src,
     const int src_stride, const uint8_t *const pre, int y_stride,
     int subpel_x_q3, int subpel_y_q3, const uint8_t *second_pred,
     const uint8_t *mask, int mask_stride, int invert_mask, unsigned int *sse) {
@@ -734,12 +714,8 @@ static INLINE unsigned int estimate_upsampled_pref_error(
     return vfp->msvf(pre, y_stride, subpel_x_q3, subpel_y_q3, src, src_stride,
                      second_pred, mask, mask_stride, invert_mask, sse);
   } else {
-    if (xd->jcp_param.use_jnt_comp_avg)
-      return vfp->jsvaf(pre, y_stride, subpel_x_q3, subpel_y_q3, src,
-                        src_stride, sse, second_pred, &xd->jcp_param);
-    else
-      return vfp->svaf(pre, y_stride, subpel_x_q3, subpel_y_q3, src, src_stride,
-                       sse, second_pred);
+    return vfp->svaf(pre, y_stride, subpel_x_q3, subpel_y_q3, src, src_stride,
+                     sse, second_pred);
   }
 }
 
@@ -750,7 +726,7 @@ int av1_find_best_sub_pixel_tree(
     int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
     unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
     int mask_stride, int invert_mask, int w, int h,
-    int use_accurate_subpel_search) {
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   const uint8_t *const src_address = x->plane[0].src.buf;
   const int src_stride = x->plane[0].src.stride;
   MACROBLOCKD *xd = &x->e_mbd;
@@ -796,7 +772,16 @@ int av1_find_best_sub_pixel_tree(
 
   (void)cost_list;  // to silence compiler warning
 
+  if (do_reset_fractional_mv) {
+    av1_set_fractional_mv(x->fractional_best_mv);
+  }
+
   for (iter = 0; iter < round; ++iter) {
+    if ((x->fractional_best_mv[iter].as_mv.row == br) &&
+        (x->fractional_best_mv[iter].as_mv.col == bc))
+      return INT_MAX;
+    x->fractional_best_mv[iter].as_mv.row = br;
+    x->fractional_best_mv[iter].as_mv.col = bc;
     // Check vertical and horizontal sub-pixel positions.
     for (idx = 0; idx < 4; ++idx) {
       tr = br + search_step[idx].row;
@@ -812,9 +797,9 @@ int av1_find_best_sub_pixel_tree(
               use_accurate_subpel_search);
         } else {
           thismse = estimate_upsampled_pref_error(
-              xd, vfp, src_address, src_stride, pre(y, y_stride, tr, tc),
-              y_stride, sp(tc), sp(tr), second_pred, mask, mask_stride,
-              invert_mask, &sse);
+              vfp, src_address, src_stride, pre(y, y_stride, tr, tc), y_stride,
+              sp(tc), sp(tr), second_pred, mask, mask_stride, invert_mask,
+              &sse);
         }
 
         cost_array[idx] = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost,
@@ -848,9 +833,8 @@ int av1_find_best_sub_pixel_tree(
             use_accurate_subpel_search);
       } else {
         thismse = estimate_upsampled_pref_error(
-            xd, vfp, src_address, src_stride, pre(y, y_stride, tr, tc),
-            y_stride, sp(tc), sp(tr), second_pred, mask, mask_stride,
-            invert_mask, &sse);
+            vfp, src_address, src_stride, pre(y, y_stride, tr, tc), y_stride,
+            sp(tc), sp(tr), second_pred, mask, mask_stride, invert_mask, &sse);
       }
 
       cost_array[4] = thismse + mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost,
@@ -915,9 +899,10 @@ unsigned int av1_compute_motion_cost(const AV1_COMP *cpi, MACROBLOCK *const x,
   unsigned int mse;
   unsigned int sse;
 
-  av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, NULL, bsize);
+  av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize,
+                                AOM_PLANE_Y, AOM_PLANE_Y);
   mse = vfp->vf(dst, dst_stride, src, src_stride, &sse);
-  mse += mv_err_cost(this_mv, &ref_mv.as_mv, x->nmvjointcost, x->mvcost,
+  mse += mv_err_cost(this_mv, &ref_mv.as_mv, x->nmv_vec_cost, x->mv_cost_stack,
                      x->errorperbit);
   return mse;
 }
@@ -1057,8 +1042,8 @@ static INLINE void calc_int_cost_list(const MACROBLOCK *x,
       cost_list[i + 1] = fn_ptr->vf(what->buf, what->stride,
                                     get_buf_from_mv(in_what, &neighbor_mv),
                                     in_what->stride, &sse) +
-                         mv_err_cost(&neighbor_mv, &fcenter_mv, x->nmvjointcost,
-                                     x->mvcost, x->errorperbit);
+                         mv_err_cost(&neighbor_mv, &fcenter_mv, x->nmv_vec_cost,
+                                     x->mv_cost_stack, x->errorperbit);
     }
   } else {
     for (i = 0; i < 4; i++) {
@@ -1070,8 +1055,8 @@ static INLINE void calc_int_cost_list(const MACROBLOCK *x,
             fn_ptr->vf(what->buf, what->stride,
                        get_buf_from_mv(in_what, &neighbor_mv), in_what->stride,
                        &sse) +
-            mv_err_cost(&neighbor_mv, &fcenter_mv, x->nmvjointcost, x->mvcost,
-                        x->errorperbit);
+            mv_err_cost(&neighbor_mv, &fcenter_mv, x->nmv_vec_cost,
+                        x->mv_cost_stack, x->errorperbit);
     }
   }
 }
@@ -1393,8 +1378,8 @@ int av1_get_mvpred_var(const MACROBLOCK *x, const MV *best_mv,
 
   return vfp->vf(what->buf, what->stride, get_buf_from_mv(in_what, best_mv),
                  in_what->stride, &unused) +
-         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmvjointcost, x->mvcost,
-                                   x->errorperbit)
+         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmv_vec_cost,
+                                   x->mv_cost_stack, x->errorperbit)
                      : 0);
 }
 
@@ -1407,19 +1392,11 @@ int av1_get_mvpred_av_var(const MACROBLOCK *x, const MV *best_mv,
   const MV mv = { best_mv->row * 8, best_mv->col * 8 };
   unsigned int unused;
 
-  if (xd->jcp_param.use_jnt_comp_avg)
-    return vfp->jsvaf(get_buf_from_mv(in_what, best_mv), in_what->stride, 0, 0,
-                      what->buf, what->stride, &unused, second_pred,
-                      &xd->jcp_param) +
-           (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmvjointcost, x->mvcost,
-                                     x->errorperbit)
-                       : 0);
-  else
-    return vfp->svaf(get_buf_from_mv(in_what, best_mv), in_what->stride, 0, 0,
-                     what->buf, what->stride, &unused, second_pred) +
-           (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmvjointcost, x->mvcost,
-                                     x->errorperbit)
-                       : 0);
+  return vfp->svaf(get_buf_from_mv(in_what, best_mv), in_what->stride, 0, 0,
+                   what->buf, what->stride, &unused, second_pred) +
+         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmv_vec_cost,
+                                   x->mv_cost_stack, x->errorperbit)
+                     : 0);
 }
 
 int av1_get_mvpred_mask_var(const MACROBLOCK *x, const MV *best_mv,
@@ -1436,8 +1413,8 @@ int av1_get_mvpred_mask_var(const MACROBLOCK *x, const MV *best_mv,
   return vfp->msvf(what->buf, what->stride, 0, 0,
                    get_buf_from_mv(in_what, best_mv), in_what->stride,
                    second_pred, mask, mask_stride, invert_mask, &unused) +
-         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmvjointcost, x->mvcost,
-                                   x->errorperbit)
+         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmv_vec_cost,
+                                   x->mv_cost_stack, x->errorperbit)
                      : 0);
 }
 
@@ -1897,7 +1874,7 @@ static int full_pixel_exhaustive(const AV1_COMP *const cpi, MACROBLOCK *x,
   int baseline_interval_divisor;
 
   // Keep track of number of exhaustive calls (this frame in this thread).
-  ++(*x->ex_search_count_ptr);
+  if (x->ex_search_count_ptr != NULL) ++(*x->ex_search_count_ptr);
 
   // Trap illegal values for interval and range for this function.
   if ((range < MIN_RANGE) || (range > MAX_RANGE) || (interval < MIN_INTERVAL) ||
@@ -2051,16 +2028,10 @@ int av1_refining_search_8p_c(MACROBLOCK *x, int error_per_bit, int search_range,
                             second_pred, mask, mask_stride, invert_mask) +
                mvsad_err_cost(x, best_mv, &fcenter_mv, error_per_bit);
   } else {
-    if (xd->jcp_param.use_jnt_comp_avg)
-      best_sad = fn_ptr->jsdaf(what->buf, what->stride,
-                               get_buf_from_mv(in_what, best_mv),
-                               in_what->stride, second_pred, &xd->jcp_param) +
-                 mvsad_err_cost(x, best_mv, &fcenter_mv, error_per_bit);
-    else
-      best_sad = fn_ptr->sdaf(what->buf, what->stride,
-                              get_buf_from_mv(in_what, best_mv),
-                              in_what->stride, second_pred) +
-                 mvsad_err_cost(x, best_mv, &fcenter_mv, error_per_bit);
+    best_sad =
+        fn_ptr->sdaf(what->buf, what->stride, get_buf_from_mv(in_what, best_mv),
+                     in_what->stride, second_pred) +
+        mvsad_err_cost(x, best_mv, &fcenter_mv, error_per_bit);
   }
 
   do_refine_search_grid[grid_coord] = 1;
@@ -2084,14 +2055,9 @@ int av1_refining_search_8p_c(MACROBLOCK *x, int error_per_bit, int search_range,
                              get_buf_from_mv(in_what, &mv), in_what->stride,
                              second_pred, mask, mask_stride, invert_mask);
         } else {
-          if (xd->jcp_param.use_jnt_comp_avg)
-            sad = fn_ptr->jsdaf(what->buf, what->stride,
-                                get_buf_from_mv(in_what, &mv), in_what->stride,
-                                second_pred, &xd->jcp_param);
-          else
-            sad = fn_ptr->sdaf(what->buf, what->stride,
-                               get_buf_from_mv(in_what, &mv), in_what->stride,
-                               second_pred);
+          sad = fn_ptr->sdaf(what->buf, what->stride,
+                             get_buf_from_mv(in_what, &mv), in_what->stride,
+                             second_pred);
         }
         if (sad < best_sad) {
           sad += mvsad_err_cost(x, &mv, &fcenter_mv, error_per_bit);
@@ -2117,13 +2083,16 @@ int av1_refining_search_8p_c(MACROBLOCK *x, int error_per_bit, int search_range,
 #define MIN_EX_SEARCH_LIMIT 128
 static int is_exhaustive_allowed(const AV1_COMP *const cpi, MACROBLOCK *x) {
   const SPEED_FEATURES *const sf = &cpi->sf;
-  const int max_ex =
-      AOMMAX(MIN_EX_SEARCH_LIMIT,
-             (*x->m_search_count_ptr * sf->max_exaustive_pct) / 100);
-
-  return sf->allow_exhaustive_searches &&
-         (sf->exhaustive_searches_thresh < INT_MAX) &&
-         (*x->ex_search_count_ptr <= max_ex) && !cpi->rc.is_src_frame_alt_ref;
+  int is_allowed = sf->allow_exhaustive_searches &&
+                   (sf->exhaustive_searches_thresh < INT_MAX) &&
+                   !cpi->rc.is_src_frame_alt_ref;
+  if (x->m_search_count_ptr != NULL && x->ex_search_count_ptr != NULL) {
+    const int max_ex =
+        AOMMAX(MIN_EX_SEARCH_LIMIT,
+               (*x->m_search_count_ptr * sf->max_exaustive_pct) / 100);
+    is_allowed = *x->ex_search_count_ptr <= max_ex && is_allowed;
+  }
+  return is_allowed;
 }
 
 int av1_full_pixel_search(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
@@ -2144,7 +2113,7 @@ int av1_full_pixel_search(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
   }
 
   // Keep track of number of searches (this frame in this thread).
-  ++(*x->m_search_count_ptr);
+  if (x->m_search_count_ptr != NULL) ++(*x->m_search_count_ptr);
 
   switch (method) {
     case FAST_DIAMOND:
@@ -2241,13 +2210,12 @@ int av1_full_pixel_search(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
 
         // for the hashMap
         hash_table *ref_frame_hash =
-            intra
-                ? &cpi->common.cur_frame->hash_table
-                : av1_get_ref_frame_hash_map(cpi, x->e_mbd.mi[0]->ref_frame[0]);
+            intra ? &cpi->common.cur_frame->hash_table
+                  : av1_get_ref_frame_hash_map(&cpi->common,
+                                               x->e_mbd.mi[0]->ref_frame[0]);
 
-        av1_get_block_hash_value(
-            what, what_stride, block_width, &hash_value1, &hash_value2,
-            x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, x);
+        av1_get_block_hash_value(what, what_stride, block_width, &hash_value1,
+                                 &hash_value2, is_cur_buf_hbd(&x->e_mbd), x);
 
         const int count = av1_hash_table_count(ref_frame_hash, hash_value1);
         // for intra, at lest one matching can be found, itself.
@@ -2333,8 +2301,9 @@ int av1_full_pixel_search(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
                                         mask, vfp, z, pre(y, y_stride, r, c), \
                                         y_stride, sp(c), sp(r), w, h, &sse,   \
                                         use_accurate_subpel_search);          \
-    if ((v = MVC(r, c) + thismse) < besterr) {                                \
-      besterr = v;                                                            \
+    v = mv_err_cost(&this_mv, ref_mv, mvjcost, mvcost, error_per_bit);        \
+    if ((v + thismse) < besterr) {                                            \
+      besterr = v + thismse;                                                  \
       br = r;                                                                 \
       bc = c;                                                                 \
       *distortion = thismse;                                                  \
@@ -2365,7 +2334,7 @@ static int upsampled_obmc_pref_error(
   unsigned int besterr;
 
   DECLARE_ALIGNED(16, uint8_t, pred[2 * MAX_SB_SQUARE]);
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     uint8_t *pred8 = CONVERT_TO_BYTEPTR(pred);
     aom_highbd_upsampled_pred(xd, cm, mi_row, mi_col, mv, pred8, w, h,
                               subpel_x_q3, subpel_y_q3, y, y_stride, xd->bd,
@@ -2568,16 +2537,16 @@ static int get_obmc_mvpred_var(const MACROBLOCK *x, const int32_t *wsrc,
 
   return vfp->ovf(get_buf_from_mv(in_what, best_mv), in_what->stride, wsrc,
                   mask, &unused) +
-         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmvjointcost, x->mvcost,
-                                   x->errorperbit)
+         (use_mvcost ? mv_err_cost(&mv, center_mv, x->nmv_vec_cost,
+                                   x->mv_cost_stack, x->errorperbit)
                      : 0);
 }
 
-int obmc_refining_search_sad(const MACROBLOCK *x, const int32_t *wsrc,
-                             const int32_t *mask, MV *ref_mv, int error_per_bit,
-                             int search_range,
-                             const aom_variance_fn_ptr_t *fn_ptr,
-                             const MV *center_mv, int is_second) {
+static int obmc_refining_search_sad(const MACROBLOCK *x, const int32_t *wsrc,
+                                    const int32_t *mask, MV *ref_mv,
+                                    int error_per_bit, int search_range,
+                                    const aom_variance_fn_ptr_t *fn_ptr,
+                                    const MV *center_mv, int is_second) {
   const MV neighbors[4] = { { -1, 0 }, { 0, -1 }, { 0, 1 }, { 1, 0 } };
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const in_what = &xd->plane[0].pre[is_second];
@@ -2616,12 +2585,13 @@ int obmc_refining_search_sad(const MACROBLOCK *x, const int32_t *wsrc,
   return best_sad;
 }
 
-int obmc_diamond_search_sad(const MACROBLOCK *x, const search_site_config *cfg,
-                            const int32_t *wsrc, const int32_t *mask,
-                            MV *ref_mv, MV *best_mv, int search_param,
-                            int sad_per_bit, int *num00,
-                            const aom_variance_fn_ptr_t *fn_ptr,
-                            const MV *center_mv, int is_second) {
+static int obmc_diamond_search_sad(const MACROBLOCK *x,
+                                   const search_site_config *cfg,
+                                   const int32_t *wsrc, const int32_t *mask,
+                                   MV *ref_mv, MV *best_mv, int search_param,
+                                   int sad_per_bit, int *num00,
+                                   const aom_variance_fn_ptr_t *fn_ptr,
+                                   const MV *center_mv, int is_second) {
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct buf_2d *const in_what = &xd->plane[0].pre[is_second];
   // search_param determines the length of the initial step and hence the number
@@ -2824,16 +2794,14 @@ int av1_obmc_full_pixel_search(const AV1_COMP *cpi, MACROBLOCK *x, MV *mvp_full,
   (void)thismse;                    \
   (void)cost_list;
 // Return the maximum MV.
-int av1_return_max_sub_pixel_mv(MACROBLOCK *x, const AV1_COMMON *const cm,
-                                int mi_row, int mi_col, const MV *ref_mv,
-                                int allow_hp, int error_per_bit,
-                                const aom_variance_fn_ptr_t *vfp,
-                                int forced_stop, int iters_per_step,
-                                int *cost_list, int *mvjcost, int *mvcost[2],
-                                int *distortion, unsigned int *sse1,
-                                const uint8_t *second_pred, const uint8_t *mask,
-                                int mask_stride, int invert_mask, int w, int h,
-                                int use_accurate_subpel_search) {
+int av1_return_max_sub_pixel_mv(
+    MACROBLOCK *x, const AV1_COMMON *const cm, int mi_row, int mi_col,
+    const MV *ref_mv, int allow_hp, int error_per_bit,
+    const aom_variance_fn_ptr_t *vfp, int forced_stop, int iters_per_step,
+    int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
+    unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
+    int mask_stride, int invert_mask, int w, int h,
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   COMMON_MV_TEST;
   (void)mask;
   (void)mask_stride;
@@ -2844,6 +2812,7 @@ int av1_return_max_sub_pixel_mv(MACROBLOCK *x, const AV1_COMMON *const cm,
   (void)cm;
   (void)mi_row;
   (void)mi_col;
+  (void)do_reset_fractional_mv;
 
   bestmv->row = maxr;
   bestmv->col = maxc;
@@ -2854,16 +2823,14 @@ int av1_return_max_sub_pixel_mv(MACROBLOCK *x, const AV1_COMMON *const cm,
   return besterr;
 }
 // Return the minimum MV.
-int av1_return_min_sub_pixel_mv(MACROBLOCK *x, const AV1_COMMON *const cm,
-                                int mi_row, int mi_col, const MV *ref_mv,
-                                int allow_hp, int error_per_bit,
-                                const aom_variance_fn_ptr_t *vfp,
-                                int forced_stop, int iters_per_step,
-                                int *cost_list, int *mvjcost, int *mvcost[2],
-                                int *distortion, unsigned int *sse1,
-                                const uint8_t *second_pred, const uint8_t *mask,
-                                int mask_stride, int invert_mask, int w, int h,
-                                int use_accurate_subpel_search) {
+int av1_return_min_sub_pixel_mv(
+    MACROBLOCK *x, const AV1_COMMON *const cm, int mi_row, int mi_col,
+    const MV *ref_mv, int allow_hp, int error_per_bit,
+    const aom_variance_fn_ptr_t *vfp, int forced_stop, int iters_per_step,
+    int *cost_list, int *mvjcost, int *mvcost[2], int *distortion,
+    unsigned int *sse1, const uint8_t *second_pred, const uint8_t *mask,
+    int mask_stride, int invert_mask, int w, int h,
+    int use_accurate_subpel_search, const int do_reset_fractional_mv) {
   COMMON_MV_TEST;
   (void)maxr;
   (void)maxc;
@@ -2874,6 +2841,7 @@ int av1_return_min_sub_pixel_mv(MACROBLOCK *x, const AV1_COMMON *const cm,
   (void)cm;
   (void)mi_row;
   (void)mi_col;
+  (void)do_reset_fractional_mv;
 
   bestmv->row = minr;
   bestmv->col = minc;
diff --git a/third_party/aom/av1/encoder/mcomp.h b/third_party/aom/av1/encoder/mcomp.h
index a975218b09..3f8b3b1fd3 100644
--- a/third_party/aom/av1/encoder/mcomp.h
+++ b/third_party/aom/av1/encoder/mcomp.h
@@ -103,7 +103,7 @@ typedef int(fractional_mv_step_fp)(
     int iters_per_step, int *cost_list, int *mvjcost, int *mvcost[2],
     int *distortion, unsigned int *sse1, const uint8_t *second_pred,
     const uint8_t *mask, int mask_stride, int invert_mask, int w, int h,
-    int use_accurate_subpel_search);
+    int use_accurate_subpel_search, const int do_reset_fractional_mv);
 
 extern fractional_mv_step_fp av1_find_best_sub_pixel_tree;
 extern fractional_mv_step_fp av1_find_best_sub_pixel_tree_pruned;
@@ -154,6 +154,12 @@ unsigned int av1_refine_warped_mv(const struct AV1_COMP *cpi,
                                   int mi_row, int mi_col, int *pts0,
                                   int *pts_inref0, int total_samples);
 
+static INLINE void av1_set_fractional_mv(int_mv *fractional_best_mv) {
+  for (int z = 0; z < 3; z++) {
+    fractional_best_mv[z].as_int = INVALID_MV;
+  }
+}
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/mips/msa/temporal_filter_msa.c b/third_party/aom/av1/encoder/mips/msa/temporal_filter_msa.c
index 531ae090a3..effa75b83f 100644
--- a/third_party/aom/av1/encoder/mips/msa/temporal_filter_msa.c
+++ b/third_party/aom/av1/encoder/mips/msa/temporal_filter_msa.c
@@ -267,6 +267,7 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr, uint32_t stride,
   }
 }
 
+// TODO(yunqing) The following optimization is not used since c code changes.
 void av1_temporal_filter_apply_msa(uint8_t *frame1_ptr, uint32_t stride,
                                    uint8_t *frame2_ptr, uint32_t blk_w,
                                    uint32_t blk_h, int32_t strength,
diff --git a/third_party/aom/av1/encoder/ml.c b/third_party/aom/av1/encoder/ml.c
index d21def43a8..579900afd3 100644
--- a/third_party/aom/av1/encoder/ml.c
+++ b/third_party/aom/av1/encoder/ml.c
@@ -15,31 +15,31 @@
 #include "aom_dsp/aom_dsp_common.h"
 #include "av1/encoder/ml.h"
 
-void av1_nn_predict(const float *features, const NN_CONFIG *nn_config,
-                    float *output) {
+// Calculate prediction based on the given input features and neural net config.
+// Assume there are no more than NN_MAX_NODES_PER_LAYER nodes in each hidden
+// layer.
+void av1_nn_predict_c(const float *input_nodes,
+                      const NN_CONFIG *const nn_config, float *const output) {
   int num_input_nodes = nn_config->num_inputs;
   int buf_index = 0;
   float buf[2][NN_MAX_NODES_PER_LAYER];
-  const float *input_nodes = features;
 
   // Propagate hidden layers.
   const int num_layers = nn_config->num_hidden_layers;
   assert(num_layers <= NN_MAX_HIDDEN_LAYERS);
   for (int layer = 0; layer < num_layers; ++layer) {
-    const float *weights = nn_config->weights[layer];
-    const float *bias = nn_config->bias[layer];
+    const float *layer_weights = nn_config->weights[layer];
+    const float *layer_bias = nn_config->bias[layer];
     float *output_nodes = buf[buf_index];
     const int num_output_nodes = nn_config->num_hidden_nodes[layer];
     assert(num_output_nodes < NN_MAX_NODES_PER_LAYER);
     for (int node = 0; node < num_output_nodes; ++node) {
-      float val = 0.0f;
+      float val = layer_bias[node];
       for (int i = 0; i < num_input_nodes; ++i)
-        val += weights[i] * input_nodes[i];
-      val += bias[node];
+        val += layer_weights[node * num_input_nodes + i] * input_nodes[i];
       // ReLU as activation function.
       val = val > 0.0f ? val : 0.0f;  // Could use AOMMAX().
       output_nodes[node] = val;
-      weights += num_input_nodes;
     }
     num_input_nodes = num_output_nodes;
     input_nodes = output_nodes;
@@ -47,14 +47,13 @@ void av1_nn_predict(const float *features, const NN_CONFIG *nn_config,
   }
 
   // Final output layer.
-  const float *weights = nn_config->weights[num_layers];
+  const float *layer_weights = nn_config->weights[num_layers];
+  const float *layer_bias = nn_config->bias[num_layers];
   for (int node = 0; node < nn_config->num_outputs; ++node) {
-    const float *bias = nn_config->bias[num_layers];
-    float val = 0.0f;
+    float val = layer_bias[node];
     for (int i = 0; i < num_input_nodes; ++i)
-      val += weights[i] * input_nodes[i];
-    output[node] = val + bias[node];
-    weights += num_input_nodes;
+      val += layer_weights[node * num_input_nodes + i] * input_nodes[i];
+    output[node] = val;
   }
 }
 
@@ -66,7 +65,9 @@ void av1_nn_softmax(const float *input, float *output, int n) {
   for (int i = 1; i < n; i++) max_inp = AOMMAX(max_inp, input[i]);
   float sum_out = 0.0f;
   for (int i = 0; i < n; i++) {
-    output[i] = (float)exp(input[i] - max_inp);
+    // Clamp to range [-10.0, 0.0] to prevent FE_UNDERFLOW errors.
+    const float normalized_input = AOMMAX(input[i] - max_inp, -10.0f);
+    output[i] = (float)exp(normalized_input);
     sum_out += output[i];
   }
   for (int i = 0; i < n; i++) output[i] /= sum_out;
diff --git a/third_party/aom/av1/encoder/ml.h b/third_party/aom/av1/encoder/ml.h
index cb8ef2871b..7f2750b31d 100644
--- a/third_party/aom/av1/encoder/ml.h
+++ b/third_party/aom/av1/encoder/ml.h
@@ -16,10 +16,12 @@
 extern "C" {
 #endif
 
+#include "config/av1_rtcd.h"
+
 #define NN_MAX_HIDDEN_LAYERS 10
 #define NN_MAX_NODES_PER_LAYER 128
 
-typedef struct {
+struct NN_CONFIG {
   int num_inputs;         // Number of input nodes, i.e. features.
   int num_outputs;        // Number of output nodes.
   int num_hidden_layers;  // Number of hidden layers, maximum 10.
@@ -29,13 +31,8 @@ typedef struct {
   const float *weights[NN_MAX_HIDDEN_LAYERS + 1];
   // Bias parameters, indexed by layer.
   const float *bias[NN_MAX_HIDDEN_LAYERS + 1];
-} NN_CONFIG;
-
-// Calculate prediction based on the given input features and neural net config.
-// Assume there are no more than NN_MAX_NODES_PER_LAYER nodes in each hidden
-// layer.
-void av1_nn_predict(const float *features, const NN_CONFIG *nn_config,
-                    float *output);
+};
+// Typedef from struct NN_CONFIG to NN_CONFIG is in rtcd_defs
 
 // Applies the softmax normalization function to the input
 // to get a valid probability distribution in the output:
diff --git a/third_party/aom/av1/encoder/partition_model_weights.h b/third_party/aom/av1/encoder/partition_model_weights.h
index 437ea43f9e..2e7b2c62af 100644
--- a/third_party/aom/av1/encoder/partition_model_weights.h
+++ b/third_party/aom/av1/encoder/partition_model_weights.h
@@ -2441,6 +2441,1420 @@ static const NN_CONFIG av1_rect_partition_nnconfig_128 = {
 #undef NUM_NODES
 #undef LABEL_SIZE
 
+#if CONFIG_ONE_PASS_SVM
+#define FEATURE_SIZE 24
+static const float av1_op_svm_early_term_weights_128[FEATURE_SIZE + 1] = {
+  -4.5893036051f, 6.9065208136f,  -9.1579514692f, 0.1353151366f,
+  -1.0271889653f, -0.0020988254f, -0.0094355949f, 0.0040209656f,
+  0.0073014747f,  0.7939705382f,  0.0254545714f,  0.0557559708f,
+  -0.0339662064f, -0.0496818300f, 0.3053600283f,  0.3699486845f,
+  0.0848271391f,  0.4091075988f,  0.1196729398f,  -0.0038137193f,
+  -0.0773495909f, -0.0651630642f, -0.0123704995f, -0.0036697401f,
+  -4.1930227095f,
+};
+
+static const float av1_op_svm_early_term_weights_64[FEATURE_SIZE + 1] = {
+  -2.7600454480f, 5.6822046712f,  -6.7576830133f, 0.1326457117f,
+  -1.0541818372f, 0.0107782654f,  0.0050469147f,  -0.0021362631f,
+  -0.0135151040f, -0.1020115005f, -0.0283409957f, -0.0176311233f,
+  0.0250648204f,  0.0196228570f,  0.5441528594f,  0.2767320141f,
+  0.1261231351f,  0.2998476408f,  0.1336215695f,  -0.1107823946f,
+  -0.0697279598f, -0.0577520545f, -0.0558441075f, -0.0699750617f,
+  -2.6995991503f,
+};
+
+static const float av1_op_svm_early_term_weights_32[FEATURE_SIZE + 1] = {
+  -0.8950734172f, 1.3559565008f,  -2.6733642653f, 0.2661361319f,
+  -0.0314731140f, 0.0044943456f,  0.0006438044f,  -0.0029066686f,
+  -0.0021903213f, 0.5845049496f,  -0.0003629350f, 0.0006982840f,
+  0.0014157386f,  -0.0017427528f, 0.7078456733f,  0.1600998068f,
+  0.0933852747f,  0.2822125876f,  0.1923826165f,  -0.0905903459f,
+  -0.0564717590f, -0.0591007486f, -0.0692268554f, -0.0677411981f,
+  -0.7101853206f,
+};
+
+static const float av1_op_svm_early_term_weights_16[FEATURE_SIZE + 1] = {
+  -0.1719124013f, -0.3192305362f, -1.1714597182f, 0.4437770294f,
+  -0.0042344643f, 0.0000027764f,  0.0018827450f,  -0.0015555613f,
+  -0.0003250050f, 0.9413693294f,  0.0076188418f,  -0.0067870352f,
+  0.0006329246f,  -0.0013059613f, 0.8596697254f,  0.0635558018f,
+  0.0447224598f,  0.0915706321f,  0.0741662273f,  -0.0269096547f,
+  -0.0244610614f, -0.0281113318f, -0.0326108845f, -0.0350908892f,
+  -0.0307521675f,
+};
+
+static const float av1_op_svm_early_term_mean_128[FEATURE_SIZE] = {
+  940540.3259649610f,    3988285.5905584921f, 575475302.3545289040f,
+  0.5775348803f,         866.9828469502f,     0.2503762393f,
+  0.2501466215f,         0.2513213770f,       0.2481557622f,
+  521994448.3219169378f, 0.2666920631f,       0.2535864361f,
+  0.2481589186f,         0.2315625823f,       100519.1049708007f,
+  12.1299754840f,        0.8279971004f,       12.6664603305f,
+  0.7313258998f,         935.8233056680f,     0.7436563032f,
+  0.7710055018f,         0.7376516970f,       0.6859818720f,
+};
+
+static const float av1_op_svm_early_term_mean_64[FEATURE_SIZE] = {
+  420419.7529613562f,    839754.4414347620f, 129360420.5256031156f,
+  0.6525652037f,         548.8972009954f,    0.2506918565f,
+  0.2488349076f,         0.2501724146f,      0.2503008213f,
+  113132974.7944754064f, 0.2479344278f,      0.2471446791f,
+  0.2524478512f,         0.2524730419f,      91147.9854189453f,
+  10.9642508460f,        0.8936554428f,      11.3877865621f,
+  0.8307555282f,         752.7787491956f,    0.7243363939f,
+  0.7198362119f,         0.7329432336f,      0.7245090283f,
+};
+
+static const float av1_op_svm_early_term_mean_32[FEATURE_SIZE] = {
+  105111.0236438536f,   184296.0939716828f, 29117017.6751756854f,
+  0.6402298612f,        140.2223339218f,    0.2495860872f,
+  0.2496407600f,        0.2506238629f,      0.2501492900f,
+  24480304.9390618578f, 0.2494442027f,      0.2496080963f,
+  0.2504881563f,        0.2504595447f,      60297.6762059058f,
+  9.4279752138f,        0.9287901132f,      9.6516813792f,
+  0.9009173677f,        591.5406335030f,    0.6944486917f,
+  0.6983941982f,        0.6927236901f,      0.6921613649f,
+};
+
+static const float av1_op_svm_early_term_mean_16[FEATURE_SIZE] = {
+  34080.7994802934f, 44108.1176228864f,   7494288.4946180154f, 0.6240636218f,
+  36.4539515827f,    0.2490867417f,       0.2499231014f,       0.2505361492f,
+  0.2504540077f,     5913397.2957480755f, 0.2487482536f,       0.2495500728f,
+  0.2503693302f,     0.2513323434f,       36574.9686737814f,   7.4345592768f,
+  0.9592429205f,     7.6001764585f,       0.9459867777f,       490.4635033056f,
+  0.6626215237f,     0.6580791886f,       0.6655481064f,       0.6589010119f,
+};
+
+static const float av1_op_svm_early_term_std_128[FEATURE_SIZE] = {
+  2054266.2732957317f,    7550554.6241466375f, 1078688147.1656334400f,
+  0.4939517611f,          1414.3139592985f,    0.1504634077f,
+  0.1515907199f,          0.1590329744f,       0.1515653324f,
+  1006422867.8989596367f, 0.1168668155f,       0.1195725959f,
+  0.1195825693f,          0.1123065533f,       195261.0940245980f,
+  4.5876675121f,          0.3773829648f,       4.8017339769f,
+  0.4432700397f,          973.7532938848f,     0.4790027843f,
+  0.5056275222f,          0.5262278749f,       0.4685586148f,
+};
+
+static const float av1_op_svm_early_term_std_64[FEATURE_SIZE] = {
+  1093636.0522712648f,   1749863.5221569177f, 255168612.8025657237f,
+  0.4761552884f,         1084.7927994662f,    0.1099344646f,
+  0.1100619440f,         0.1090853225f,       0.1115303745f,
+  232084513.1365262568f, 0.0759732385f,       0.0762942913f,
+  0.0785624106f,         0.0779284747f,       185687.9441778057f,
+  4.4371901245f,         0.3082781088f,       4.6670562831f,
+  0.3749677061f,         854.3212307408f,     0.4920531348f,
+  0.5073919158f,         0.5054698298f,       0.4904895620f,
+};
+
+static const float av1_op_svm_early_term_std_32[FEATURE_SIZE] = {
+  238229.7484988807f,   400136.8703966461f, 60267828.4581554681f,
+  0.4799328974f,        268.9377064297f,    0.1122938575f,
+  0.1126479260f,        0.1137018559f,      0.1126389337f,
+  52174139.1477040648f, 0.0715628767f,      0.0720997035f,
+  0.0728961434f,        0.0732065300f,      147785.0049793872f,
+  4.2092341484f,        0.2571751131f,      4.3893075417f,
+  0.2987729310f,        769.0253148602f,    0.5027558039f,
+  0.4982811444f,        0.5092312751f,      0.4991214994f,
+};
+
+static const float av1_op_svm_early_term_std_16[FEATURE_SIZE] = {
+  64177.9527087587f,    103729.9987511119f, 16632490.8146969266f,
+  0.4843637247f,        65.8114470725f,     0.0884226846f,
+  0.0912638659f,        0.0914771167f,      0.0916078800f,
+  13364581.3877149168f, 0.0677468925f,      0.0689631274f,
+  0.0689915367f,        0.0702648469f,      111397.2620676765f,
+  3.7858187888f,        0.1977269328f,      3.9420183951f,
+  0.2260437881f,        717.5336868275f,    0.5017939514f,
+  0.5066633533f,        0.5086806985f,      0.5085585987f,
+};
+
+#undef FEATURE_SIZE
+#endif  // CONFIG_ONE_PASS_SVM
+
+// Below are the models used for full_pixel_motion_search_based_split
+static const float full_pixel_motion_search_based_split_thresh_128 = 2.0f;
+static const float full_pixel_motion_search_based_split_thresh_64 = 2.0f;
+static const float full_pixel_motion_search_based_split_thresh_32 = 2.0f;
+static const float full_pixel_motion_search_based_split_thresh_16 = 2.0f;
+static const float full_pixel_motion_search_based_split_thresh_8 = 2.0f;
+
+// BLOCK_128X128
+#define NUM_HIDDEN_LAYERS_128 1
+#define NUM_FEATURES_128 6
+#define NUM_LAYER_0_UNITS_128 16
+#define NUM_LOGITS_128 1
+
+static const float full_pixel_motion_search_based_split_layer_0_kernel_128[] = {
+  -0.807346f,  0.242298f,   12.9862f,   -1.19161f,  5.21734f,    -1.1363f,
+  -2.39127f,   0.930915f,   -2.44285f,  -2.42966f,  5.73476f,    0.0506879f,
+  -0.234878f,  -0.317875f,  0.361322f,  0.431648f,  -0.39105f,   -0.110225f,
+  -2.46236f,   0.979713f,   -10.5596f,  -7.76653f,  -3.06518f,   2.42554f,
+  0.0492961f,  -0.467176f,  0.130746f,  0.494527f,  -0.0336645f, 0.501755f,
+  0.176486f,   -0.869541f,  7.77757f,   6.81303f,   6.00771f,    7.35696f,
+  0.150731f,   -0.307017f,  -0.437639f, -0.082924f, 0.379107f,   0.452278f,
+  -0.0143836f, -0.183691f,  -0.604698f, -9.2681f,   -2.06087f,   11.0256f,
+  0.0487599f,  -0.249168f,  -0.180407f, 0.304772f,  0.218642f,   -0.406073f,
+  -0.0289919f, -0.794381f,  5.45092f,   5.38374f,   3.25745f,    5.32903f,
+  1.12718f,    -0.0215478f, 2.78552f,   4.8951f,    -0.959671f,  0.694264f,
+  -0.0611219f, -0.331937f,  0.258252f,  -0.495331f, -0.285923f,  0.294713f,
+  -0.119947f,  0.0753204f,  10.2021f,   -5.82147f,  -12.0137f,   3.0365f,
+  0.366697f,   0.142683f,   -3.29731f,  -5.76651f,  -5.62578f,   10.9462f,
+  -0.325459f,  0.092602f,   -0.868027f, -0.691768f, -0.292017f,  -0.00841203f,
+  0.702545f,   -0.612227f,  -7.68881f,  9.52225f,   -1.18581f,   -2.56762f
+};
+
+static const float full_pixel_motion_search_based_split_logits_kernel_128[] = {
+  0.364895f,    0.577553f,  0.115758f,  -0.999496f, 0.124885f, 3.23193f,
+  -0.00386642f, 0.970794f,  0.136637f,  -4.28052f,  -1.49234f, 0.370436f,
+  0.576981f,    -0.469656f, -0.124071f, 1.07669f
+};
+
+static const float full_pixel_motion_search_based_split_layer_0_bias_128[] = {
+  1.32916f,    0.817212f,  0.0f,       -0.921066f, 0.0f,      3.57649f,
+  -0.0204517f, 2.97286f,   0.0f,       5.49957f,   -8.14518f, 0.0f,
+  1.30826f,    -0.349536f, -0.638933f, 5.4496f
+};
+
+static const float full_pixel_motion_search_based_split_logits_bias_128[] = {
+  0.683442f
+};
+
+static const NN_CONFIG full_pixel_motion_search_based_split_nn_config_128 = {
+  NUM_FEATURES_128,
+  NUM_LOGITS_128,
+  NUM_HIDDEN_LAYERS_128,
+  {
+      NUM_LAYER_0_UNITS_128,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_kernel_128,
+      full_pixel_motion_search_based_split_logits_kernel_128,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_bias_128,
+      full_pixel_motion_search_based_split_logits_bias_128,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_128
+#undef NUM_FEATURES_128
+#undef NUM_LAYER_0_UNITS_128
+#undef NUM_LOGITS_128
+
+// BLOCK_64X64
+#define NUM_HIDDEN_LAYERS_64 1
+#define NUM_FEATURES_64 6
+#define NUM_LAYER_0_UNITS_64 16
+#define NUM_LOGITS_64 1
+
+static const float full_pixel_motion_search_based_split_layer_0_kernel_64[] = {
+  0.0345945f,  -0.394064f,  0.0919978f, 0.270358f,  -0.384502f, -0.504608f,
+  -0.25759f,   0.155981f,   2.62567f,   -10.7204f,  -0.709802f, 8.15948f,
+  0.589866f,   -0.445645f,  -1.68232f,  10.0061f,   -3.17671f,  4.87259f,
+  -0.448886f,  -0.205568f,  -0.462388f, 0.385001f,  -0.451687f, 0.49602f,
+  -0.256708f,  0.803322f,   3.25594f,   0.38541f,   -1.83867f,  -2.15132f,
+  0.936059f,   -0.203056f,  -5.92959f,  -6.24554f,  -6.68631f,  -6.85977f,
+  -0.0407565f, -0.258902f,  0.195053f,  -0.366515f, 0.339543f,  -0.433017f,
+  -2.67026f,   0.385457f,   1.86683f,   1.9501f,    0.0381398f, 1.086f,
+  -0.153729f,  0.173772f,   -42.9029f,  -36.8934f,  -2.892f,    -0.0540691f,
+  0.77469f,    -0.380145f,  2.2689f,    -9.53332f,  1.15712f,   2.86601f,
+  -0.437036f,  0.247132f,   -8.51058f,  -3.62972f,  -8.99449f,  -0.638738f,
+  0.0609263f,  -0.0614603f, 5.42307f,   5.35926f,   5.27437f,   5.26599f,
+  -0.0729677f, 0.0306104f,  -7.77867f,  5.03598f,   -8.17832f,  5.85461f,
+  -0.253269f,  0.164582f,   -4.49713f,  3.83265f,   9.04851f,   -2.85668f,
+  1.22618f,    0.166904f,   -1.51975f,  -4.01576f,  -1.44374f,  -2.22147f,
+  -0.217072f,  -0.0984913f, -0.265515f, 0.360021f,  0.0779512f, 0.361516f
+};
+
+static const float full_pixel_motion_search_based_split_logits_kernel_64[] = {
+  0.470821f, 0.474747f, -0.571292f, 0.403221f,  0.628966f,  -0.617029f,
+  0.501105f, 0.499962f, -1.5451f,   -0.473518f, -0.730568f, -5.55817f,
+  0.776761f, 0.42569f,  0.311925f,  0.469968f
+};
+
+static const float full_pixel_motion_search_based_split_layer_0_bias_64[] = {
+  -0.134085f, 0.0758715f, 1.10419f,  0.0f,       -5.75737f, 1.65494f,
+  0.0f,       3.44047f,   0.394852f, 3.43858f,   3.65871f,  -4.84987f,
+  1.21207f,   -1.7705f,   -5.46469f, -0.0889634f
+};
+
+static const float full_pixel_motion_search_based_split_logits_bias_64[] = {
+  -0.479491f
+};
+
+static const NN_CONFIG full_pixel_motion_search_based_split_nn_config_64 = {
+  NUM_FEATURES_64,
+  NUM_LOGITS_64,
+  NUM_HIDDEN_LAYERS_64,
+  {
+      NUM_LAYER_0_UNITS_64,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_kernel_64,
+      full_pixel_motion_search_based_split_logits_kernel_64,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_bias_64,
+      full_pixel_motion_search_based_split_logits_bias_64,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_64
+#undef NUM_FEATURES_64
+#undef NUM_LAYER_0_UNITS_64
+#undef NUM_LOGITS_64
+
+// BLOCK_32X32
+#define NUM_HIDDEN_LAYERS_32 1
+#define NUM_FEATURES_32 6
+#define NUM_LAYER_0_UNITS_32 16
+#define NUM_LOGITS_32 1
+
+static const float full_pixel_motion_search_based_split_layer_0_kernel_32[] = {
+  -1.61796f,   0.0585128f,  1.57904f,   1.52703f,   0.367779f, 0.220434f,
+  1.66652f,    -1.77782f,   6.41118f,   4.16976f,   4.97299f,  4.84111f,
+  -0.0956536f, -0.163284f,  -0.143662f, 0.129329f,  0.449659f, -0.528844f,
+  -1.00067f,   1.17203f,    -4.26777f,  -4.78521f,  8.45658f,  -3.49498f,
+  -1.78386f,   0.111488f,   4.176f,     6.31911f,   -10.5369f, 6.26983f,
+  -1.32233f,   1.22999f,    -4.1666f,   -10.0359f,  -4.14779f, -10.4695f,
+  1.83011f,    -0.333152f,  -9.87986f,  -8.11992f,  -8.2775f,  -7.79918f,
+  -0.101404f,  0.00401393f, 8.89046f,   -7.32186f,  -6.59597f, 9.66257f,
+  -1.1492f,    1.23067f,    -3.6341f,   6.59275f,   -3.2373f,  -3.42564f,
+  0.371736f,   -0.140902f,  -2.75715f,  5.92487f,   -7.9185f,  9.13743f,
+  -3.52698f,   -0.191044f,  5.96691f,   6.26327f,   4.36378f,  5.69354f,
+  -0.608845f,  -0.191236f,  -0.482191f, -0.180474f, -3.8838f,  -3.92934f,
+  -1.03191f,   0.994568f,   7.95516f,   -4.0035f,   -2.86266f, -4.96105f,
+  1.75022f,    0.125058f,   -1.52159f,  -3.59304f,  -2.82634f, -2.49556f,
+  -2.05557f,   -0.222577f,  3.7608f,    5.50475f,   2.7046f,   5.25952f,
+  -1.91327f,   -0.0356497f, 1.47611f,   1.27499f,   -1.76108f, -0.578954f
+};
+
+static const float full_pixel_motion_search_based_split_logits_kernel_32[] = {
+  -0.220382f, -0.693902f, 0.424827f, 0.379952f, -0.413791f, -0.326785f,
+  -0.455086f, 0.242402f,  0.307986f, 0.175746f, 0.498901f,  -0.628053f,
+  0.285447f,  0.230052f,  0.415151f, -0.842946f
+};
+
+static const float full_pixel_motion_search_based_split_layer_0_bias_32[] = {
+  -1.80751f, 6.40356f,   -0.0512058f, -4.59163f, -0.369933f, -0.195755f,
+  -0.16648f, -0.599755f, -5.35975f,   -1.21349f, 2.48414f,   1.07096f,
+  -3.66684f, -6.17761f,  4.2159f,     -1.05286f
+};
+
+static const float full_pixel_motion_search_based_split_logits_bias_32[] = {
+  -2.58676f
+};
+
+static const NN_CONFIG full_pixel_motion_search_based_split_nn_config_32 = {
+  NUM_FEATURES_32,
+  NUM_LOGITS_32,
+  NUM_HIDDEN_LAYERS_32,
+  {
+      NUM_LAYER_0_UNITS_32,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_kernel_32,
+      full_pixel_motion_search_based_split_logits_kernel_32,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_bias_32,
+      full_pixel_motion_search_based_split_logits_bias_32,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_32
+#undef NUM_FEATURES_32
+#undef NUM_LAYER_0_UNITS_32
+#undef NUM_LOGITS_32
+
+// BLOCK_16X16
+#define NUM_HIDDEN_LAYERS_16 1
+#define NUM_FEATURES_16 6
+#define NUM_LAYER_0_UNITS_16 16
+#define NUM_LOGITS_16 1
+
+static const float full_pixel_motion_search_based_split_layer_0_kernel_16[] = {
+  -0.611497f,  -0.0422086f, -0.555957f,   -0.632451f, -0.144179f, -0.152722f,
+  -0.330265f,  -0.419866f,  0.287343f,    0.385295f,  -0.424486f, 0.424281f,
+  2.27442f,    -2.47933f,   5.24731f,     4.33827f,   4.73215f,   3.41909f,
+  1.16058f,    -0.364505f,  0.12207f,     -0.287749f, 0.0509783f, -0.0200119f,
+  1.52907f,    -1.1905f,    -2.56978f,    -3.00186f,  -3.56084f,  -3.89276f,
+  0.00365657f, 1.57125f,    -4.421f,      -2.48803f,  -2.51531f,  -4.28646f,
+  2.52248f,    -1.03377f,   -1.09607f,    -1.44633f,  -1.58736f,  -1.25927f,
+  -1.45841f,   -0.566619f,  -0.246166f,   -0.182289f, -0.238156f, 0.177991f,
+  0.0112509f,  -0.17677f,   -0.485877f,   0.0812852f, 0.104975f,  0.222793f,
+  -0.372858f,  -0.48624f,   -0.00870389f, -0.385019f, 0.405842f,  0.288523f,
+  0.167374f,   -0.204208f,  -8.74148f,    -8.59267f,  -8.42492f,  -8.3778f,
+  -5.57063f,   -0.406818f,  -0.873199f,   -0.896224f, -0.701479f, -0.985736f,
+  -0.625956f,  -0.0446202f, -0.509987f,   -0.321804f, -0.470759f, -0.248556f,
+  -0.369436f,  -0.160828f,  0.0591148f,   0.405218f,  0.142584f,  -0.130106f,
+  0.125321f,   0.0888179f,  7.34822f,     -6.71488f,  -7.06592f,  6.33224f,
+  0.0333619f,  -0.377782f,  0.160767f,    -0.128169f, -0.484818f, -0.311973f
+};
+
+static const float full_pixel_motion_search_based_split_logits_kernel_16[] = {
+  -0.132207f,   0.15176f,   -0.680086f, 0.605921f, -0.43294f,  0.485811f,
+  -0.306286f,   0.551368f,  0.413904f,  0.548748f, -0.437391f, 0.560778f,
+  -0.00685266f, -0.558657f, 0.122127f,  0.260165f
+};
+
+static const float full_pixel_motion_search_based_split_layer_0_bias_16[] = {
+  -0.200928f, -0.074132f, 8.69963f,    -9.00807f,  9.08983f, -6.83586f,
+  -3.89329f,  10.4881f,   -0.0670618f, 0.0f,       9.21614f, 8.41773f,
+  -0.145851f, 0.0f,       -1.43038f,   -0.0460311f
+};
+
+static const float full_pixel_motion_search_based_split_logits_bias_16[] = {
+  -4.19885f
+};
+
+static const NN_CONFIG full_pixel_motion_search_based_split_nn_config_16 = {
+  NUM_FEATURES_16,
+  NUM_LOGITS_16,
+  NUM_HIDDEN_LAYERS_16,
+  {
+      NUM_LAYER_0_UNITS_16,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_kernel_16,
+      full_pixel_motion_search_based_split_logits_kernel_16,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_bias_16,
+      full_pixel_motion_search_based_split_logits_bias_16,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_16
+#undef NUM_FEATURES_16
+#undef NUM_LAYER_0_UNITS_16
+#undef NUM_LOGITS_16
+
+#if !CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8
+// BLOCK_8X8
+#define NUM_HIDDEN_LAYERS_8 1
+#define NUM_FEATURES_8 6
+#define NUM_LAYER_0_UNITS_8 16
+#define NUM_LOGITS_8 1
+
+static const float full_pixel_motion_search_based_split_layer_0_kernel_8[] = {
+  0.0370236f,   -0.580211f,  2.0134f,    1.69637f,    2.43181f,   -0.521648f,
+  -0.00375187f, 0.122712f,   -4.74411f,  7.36187f,    5.42574f,   -5.53557f,
+  0.0993344f,   -0.358843f,  0.0765453f, -0.615987f,  -0.754633f, -0.175846f,
+  0.714976f,    0.492862f,   0.346604f,  -1.23922f,   -2.67031f,  2.12749f,
+  1.71511f,     -1.4239f,    2.09396f,   2.42478f,    2.40151f,   2.90487f,
+  0.540813f,    -0.0954257f, -4.57571f,  -4.88078f,   -4.62386f,  -5.75167f,
+  1.35351f,     -1.08114f,   1.43744f,   1.44333f,    0.608153f,  0.193742f,
+  -0.405512f,   -0.155164f,  0.0771456f, -0.473182f,  -0.057984f, 0.140435f,
+  0.743021f,    -0.418589f,  -0.377622f, -0.531411f,  -0.668025f, -0.826607f,
+  1.37834f,     -1.07753f,   0.870466f,  0.516756f,   0.708689f,  0.286795f,
+  -3.97895f,    -0.338629f,  2.79427f,   1.80561f,    1.46275f,   1.50438f,
+  0.0232533f,   -0.43174f,   -0.348251f, 0.0863006f,  0.0321103f, 0.129674f,
+  -1.12024f,    -0.0990596f, -0.283472f, -0.238713f,  -0.239175f, -0.40816f,
+  -0.00106566f, 0.0972736f,  5.19284f,   -3.70862f,   6.39657f,   -5.27588f,
+  -2.08003f,    0.38825f,    2.38771f,   -1.27501f,   -2.45619f,  3.07324f,
+  0.616966f,    -0.451472f,  -0.319365f, 0.00807278f, -0.303261f, -0.351679f
+};
+
+static const float full_pixel_motion_search_based_split_logits_kernel_8[] = {
+  -0.625847f, 0.381323f, 0.342475f, 0.526161f,  -0.665965f, -0.515317f,
+  -0.406218f, 0.568007f, 0.479397f, -0.426116f, 0.615638f,  0.338572f,
+  0.185583f,  0.308031f, 0.260748f, 0.531619f
+};
+
+static const float full_pixel_motion_search_based_split_layer_0_bias_8[] = {
+  4.73775f,  -1.12658f, -0.258038f, -6.06696f, 1.79131f, 2.49609f,
+  4.28388f,  0.0f,      -4.63598f,  3.06034f,  5.31994f, -0.152142f,
+  0.514738f, -1.30098f, 3.00296f,   -3.83481f
+};
+
+static const float full_pixel_motion_search_based_split_logits_bias_8[] = {
+  -3.44508f
+};
+
+static const NN_CONFIG full_pixel_motion_search_based_split_nn_config_8 = {
+  NUM_FEATURES_8,
+  NUM_LOGITS_8,
+  NUM_HIDDEN_LAYERS_8,
+  {
+      NUM_LAYER_0_UNITS_8,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_kernel_8,
+      full_pixel_motion_search_based_split_logits_kernel_8,
+  },
+  {
+      full_pixel_motion_search_based_split_layer_0_bias_8,
+      full_pixel_motion_search_based_split_logits_bias_8,
+  },
+};
+
+#endif
+
+// Model based on simple_motion_search
+
+// Thresholds for doing a single type of partition
+// TODO(chiyotsai@google.com): Set the thresholds for PARTITION_SPLIT.
+static const float simple_motion_search_prune_part_only_thresh_128[10] = {
+  1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+};
+static const float simple_motion_search_prune_part_only_thresh_64[10] = {
+  1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+};
+static const float simple_motion_search_prune_part_only_thresh_32[10] = {
+  1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+};
+static const float simple_motion_search_prune_part_only_thresh_16[10] = {
+  1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+};
+static const float simple_motion_search_prune_part_only_thresh_8[10] = {
+  1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f
+};
+
+// Thresholds for pruning a partition type
+// TODO(chiyotsai@google.com): Retune the thresholds for rectangular partition.
+static const float simple_motion_search_prune_part_prune_thresh_128[10] = {
+  0.0f, 0.0110f, 0.0110f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f
+};
+static const float simple_motion_search_prune_part_prune_thresh_64[10] = {
+  0.0f, 0.0110f, 0.0110f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f
+};
+static const float simple_motion_search_prune_part_prune_thresh_32[10] = {
+  0.0f, 0.0110f, 0.0110f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f
+};
+static const float simple_motion_search_prune_part_prune_thresh_16[10] = {
+  0.0f, 0.0110f, 0.0110f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f
+};
+static const float simple_motion_search_prune_part_prune_thresh_8[10] = {
+  0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f
+};
+
+// Mean and std
+static const float simple_motion_search_prune_part_mean_128[19] = {
+  11.561261f, 11.626691f, 10.820066f, 10.880652f, 11.193911f,
+  11.269741f, 10.509796f, 10.584754f, 12.659430f, 12.724323f,
+  12.254200f, 12.326454f, 12.632568f, 12.694887f, 11.814313f,
+  11.872458f, 13.690346f, 13.753730f, 5.165726f,
+};
+static const float simple_motion_search_prune_part_std_128[19] = {
+  2.434728f, 2.429037f, 3.701918f, 3.709713f, 3.288111f, 3.286657f, 4.174337f,
+  4.185339f, 2.531916f, 2.540137f, 3.488878f, 3.496263f, 2.459336f, 2.460739f,
+  3.919058f, 3.930766f, 2.511685f, 2.524923f, 2.257989f,
+};
+static const float simple_motion_search_prune_part_mean_64[19] = {
+  10.073195f, 10.158249f, 9.627341f,  9.708094f,  9.002025f,
+  9.076430f,  8.597153f,  8.668316f,  11.021699f, 11.097507f,
+  9.852887f,  9.919513f,  11.007151f, 11.085718f, 10.525035f,
+  10.599575f, 11.936800f, 12.007580f, 4.860905f,
+};
+static const float simple_motion_search_prune_part_std_64[19] = {
+  2.272113f, 2.256767f, 3.023892f, 3.027094f, 3.794186f, 3.808349f, 4.143885f,
+  4.164291f, 2.262401f, 2.249636f, 4.039632f, 4.053496f, 2.234799f, 2.219867f,
+  3.136350f, 3.139975f, 2.199973f, 2.187191f, 2.185985f,
+};
+static const float simple_motion_search_prune_part_mean_32[19] = {
+  8.499073f, 8.600074f, 8.508158f,  8.611221f,  8.498978f, 8.599821f, 8.506126f,
+  8.609448f, 9.481358f, 9.569901f,  9.481697f,  9.569918f, 9.469805f, 9.557789f,
+  9.483047f, 9.572816f, 10.416696f, 10.496502f, 4.413157f,
+};
+static const float simple_motion_search_prune_part_std_32[19] = {
+  2.129560f, 2.117602f, 2.115962f, 2.104043f, 2.136564f, 2.125870f, 2.125247f,
+  2.114250f, 2.146261f, 2.134374f, 2.160289f, 2.149360f, 2.116179f, 2.103744f,
+  2.099453f, 2.087138f, 2.115011f, 2.101419f, 2.024279f,
+};
+static const float simple_motion_search_prune_part_mean_16[19] = {
+  6.990048f, 7.116367f, 6.883459f, 7.007796f, 6.993151f, 7.119445f, 6.886147f,
+  7.010345f, 7.994230f, 8.099907f, 7.998478f, 8.104088f, 7.988468f, 8.092865f,
+  7.868373f, 7.971079f, 8.946122f, 9.037547f, 3.976610f,
+};
+static const float simple_motion_search_prune_part_std_16[19] = {
+  1.963746f, 1.954729f, 2.127820f, 2.126451f, 1.963392f, 1.955016f, 2.126736f,
+  2.126004f, 1.991209f, 1.981070f, 1.992683f, 1.983143f, 1.986913f, 1.977408f,
+  2.200119f, 2.197786f, 1.997332f, 1.986091f, 1.829684f,
+};
+static const float simple_motion_search_prune_part_mean_8[19] = {
+  5.362649f, 5.561670f, 5.362880f, 5.564151f, 5.365307f, 5.565139f, 5.365830f,
+  5.567452f, 6.453082f, 6.598073f, 6.458814f, 6.604363f, 6.454421f, 6.599441f,
+  6.458431f, 6.604804f, 7.469143f, 7.585015f, 3.625800f,
+};
+static const float simple_motion_search_prune_part_std_8[19] = {
+  1.846273f, 1.835266f, 1.846272f, 1.834365f, 1.836967f, 1.826428f, 1.837247f,
+  1.825985f, 1.886601f, 1.874138f, 1.874622f, 1.862656f, 1.877218f, 1.865398f,
+  1.874494f, 1.862057f, 1.896421f, 1.883210f, 1.670200f,
+};
+
+// BLOCK_128X128
+#define NUM_HIDDEN_LAYERS_128 1
+#define NUM_FEATURES_128 19
+#define NUM_LAYER_0_UNITS_128 24
+#define NUM_LOGITS_128 4
+
+static const float simple_motion_search_prune_part_logits_kernel_128[] = {
+  -0.12815f,   -1.2107f,    0.525183f,  0.379155f,   -2.04966f,  -0.212818f,
+  -0.0609484f, -0.119338f,  1.07054f,   -0.0304129f, 0.563817f,  -0.0204729f,
+  0.757013f,   0.464791f,   0.191185f,  0.183903f,   -2.71738f,  1.41455f,
+  -0.449089f,  -0.0344047f, 0.349133f,  -0.945292f,  0.63729f,   -0.973444f,
+  -0.203447f,  -19.7875f,   -12.7536f,  -0.260806f,  0.934733f,  -3.17648f,
+  -1.11806f,   -4.63145f,   -16.0947f,  -24.203f,    -2.94649f,  -8.34109f,
+  -0.720804f,  -5.16623f,   -0.443715f, 0.510014f,   -4.2846f,   -11.2272f,
+  0.187633f,   -10.0189f,   -0.915574f, -0.235434f,  -2.57912f,  0.151642f,
+  -0.180428f,  -0.110439f,  0.349015f,  -1.83906f,   -3.38461f,  -0.309185f,
+  -9.3706f,    -5.54824f,   0.45334f,   -8.48801f,   0.392325f,  -4.19948f,
+  -0.535814f,  -4.92133f,   0.0159878f, 0.275078f,   -2.65358f,  1.32524f,
+  -0.761795f,  -14.782f,    -4.15338f,  -0.177581f,  -0.158376f, -0.0832793f,
+  0.226712f,   -0.349409f,  0.0529463f, -0.412873f,  -0.241807f, 0.19054f,
+  0.79237f,    0.218937f,   -1.60466f,  0.845653f,   -0.287058f, 0.766271f,
+  -0.455829f,  0.00261991f, 0.506805f,  -0.0593004f, 2.25348f,   -2.29282f,
+  0.0491916f,  0.38261f,    -0.281848f, -0.17802f,   -0.472124f, 0.204326f
+};
+
+static const float simple_motion_search_prune_part_layer_0_kernel_128[] = {
+  0.485797f,   -0.326149f,  0.238042f,     -1.41255f,     -0.00899515f,
+  -0.715324f,  0.623049f,   0.502229f,     0.54064f,      -0.0667834f,
+  0.0591777f,  -0.435869f,  0.221466f,     -0.222184f,    0.696493f,
+  -0.514433f,  -0.771882f,  -0.080914f,    -1.58162f,     -0.214871f,
+  -0.0141856f, -0.566004f,  -1.01071f,     -0.0400122f,   -0.212581f,
+  0.144009f,   0.156065f,   0.419812f,     0.351579f,     -0.0363798f,
+  0.272286f,   0.177348f,   0.643484f,     -0.686943f,    -1.5021f,
+  -0.283203f,  -0.317958f,  0.834615f,     1.18195f,      0.704402f,
+  0.723567f,   0.0765064f,  0.0998969f,    0.358114f,     -0.471958f,
+  -0.33568f,   -0.0455057f, 0.0659417f,    0.717502f,     1.40203f,
+  -1.78214f,   -1.00743f,   -0.217992f,    -0.244456f,    -0.681988f,
+  0.199185f,   1.27797f,    0.744351f,     0.853728f,     0.976386f,
+  1.5212f,     0.246886f,   0.458125f,     -0.278508f,    -0.601164f,
+  -0.874409f,  -0.0882617f, -0.93093f,     -0.0279819f,   0.289613f,
+  -0.795015f,  1.2224f,     0.464495f,     -0.996107f,    -1.05131f,
+  0.780663f,   1.70779f,    -0.122481f,    2.64976f,      0.284897f,
+  -0.129159f,  -0.0472128f, 0.190105f,     0.549306f,     -0.829239f,
+  -2.87392f,   0.0735607f,  -1.00692f,     -0.0171315f,   0.217386f,
+  0.715055f,   1.2976f,     -0.387801f,    -0.301424f,    -0.265159f,
+  0.381381f,   -0.637641f,  0.424966f,     0.317794f,     1.99832f,
+  0.604094f,   0.444838f,   0.354898f,     -0.2603f,      -0.625206f,
+  1.37372f,    0.474128f,   1.47492f,      -0.318267f,    -0.32122f,
+  -1.29434f,   0.426303f,   -0.604944f,    -2.33438f,     -1.33476f,
+  -2.13254f,   2.28472f,    0.844804f,     0.366394f,     0.148218f,
+  -0.0996413f, -0.665787f,  1.30951f,      0.447868f,     -0.160892f,
+  0.336539f,   -0.461251f,  -0.940939f,    1.9886f,       0.774578f,
+  -0.0274317f, -0.288034f,  -0.188082f,    0.296018f,     -0.171846f,
+  0.245583f,   0.39913f,    -0.0544446f,   0.823534f,     0.215969f,
+  -0.393239f,  -0.208693f,  -0.336752f,    0.258191f,     1.03315f,
+  -0.0356378f, -0.35999f,   0.48529f,      1.09123f,      -0.243654f,
+  0.131594f,   -2.5832f,    1.26485f,      -0.700618f,    0.466223f,
+  -0.525984f,  2.19872f,    0.685663f,     1.40228f,      1.10528f,
+  0.0451285f,  -0.39981f,   -0.644025f,    -1.31137f,     0.836056f,
+  -0.884238f,  0.394441f,   -0.16017f,     -1.45246f,     -1.24252f,
+  0.195623f,   0.716485f,   -0.0522381f,   0.271321f,     0.560211f,
+  0.685537f,   -0.117774f,  -0.227596f,    0.346762f,     -0.402137f,
+  0.00528979f, 0.756388f,   0.181091f,     -0.513235f,    -0.354307f,
+  1.04355f,    -0.114563f,  0.457349f,     -0.525175f,    -1.306f,
+  -1.57685f,   -1.69919f,   0.393799f,     1.35822f,      -0.496799f,
+  0.102504f,   -0.0950497f, 0.518029f,     0.211844f,     0.385925f,
+  0.366614f,   0.52377f,    -0.22419f,     -0.209355f,    0.343361f,
+  0.0535484f,  0.320409f,   0.467987f,     -0.00542184f,  -1.02011f,
+  0.488822f,   -0.248647f,  0.113298f,     -0.665206f,    0.374419f,
+  0.731859f,   1.20004f,    -0.000319085f, -0.000491633f, -0.260543f,
+  0.804927f,   -0.410925f,  0.960751f,     0.281493f,     0.607288f,
+  -0.681745f,  -0.0925707f, -2.06876f,     -0.52926f,     0.864695f,
+  -0.712922f,  0.706643f,   -0.651177f,    -0.231887f,    -0.318515f,
+  0.0635488f,  -0.183294f,  -0.589271f,    0.288583f,     0.38702f,
+  0.582912f,   0.225519f,   1.23027f,      0.108471f,     1.3328f,
+  0.689896f,   -2.21988f,   -0.337789f,    -0.762238f,    1.79149f,
+  0.840519f,   0.825442f,   0.460811f,     0.434437f,     2.13268f,
+  -0.341805f,  -0.139746f,  -1.66773f,     -0.583511f,    -0.238547f,
+  0.156871f,   0.719404f,   1.05882f,      -0.173445f,    -0.867368f,
+  0.170469f,   1.57755f,    1.36095f,      -0.418987f,    -0.396689f,
+  0.289496f,   -0.292606f,  0.688943f,     0.176553f,     0.853828f,
+  1.19608f,    -0.557087f,  -0.58118f,     1.12805f,      0.232023f,
+  -0.167864f,  -0.222265f,  -0.719284f,    -0.802184f,    -2.85383f,
+  -1.01816f,   -1.344f,     -0.154875f,    0.308999f,     -0.62822f,
+  0.389045f,   -0.498046f,  -0.0646747f,   0.056343f,     0.188288f,
+  0.493475f,   0.241337f,   0.122423f,     -0.506522f,    0.0492923f,
+  0.0642163f,  -0.508636f,  0.0328815f,    0.78567f,      0.224673f,
+  0.627972f,   0.759423f,   1.16089f,      1.28274f,      1.19646f,
+  1.63401f,    1.56657f,    -0.131442f,    -0.423053f,    -0.599536f,
+  -1.20933f,   0.0438751f,  -0.538444f,    -0.749136f,    -0.769951f,
+  -1.06683f,   -1.66474f,   0.136208f,     1.35745f,      -0.960854f,
+  2.32612f,    0.349384f,   1.04506f,      -0.888194f,    1.28114f,
+  0.579831f,   0.53927f,    -1.07214f,     0.456708f,     -0.884062f,
+  0.789767f,   -0.776746f,  -0.0143946f,   -0.930974f,    -0.632568f,
+  -1.487f,     0.0284456f,  -0.865849f,    -1.51263f,     0.117335f,
+  -0.671304f,  -0.367558f,  -0.650224f,    1.09562f,      0.246097f,
+  0.461775f,   -0.722606f,  0.0886829f,    -0.490978f,    1.92718f,
+  0.864959f,   0.0705385f,  -0.543902f,    1.4704f,       -0.643494f,
+  0.194486f,   0.890657f,   0.901154f,     -0.458931f,    0.209679f,
+  0.303869f,   -0.167078f,  1.60197f,      1.05005f,      -0.499959f,
+  0.930484f,   -0.555236f,  -0.778482f,    -0.772935f,    -0.590988f,
+  -0.122397f,  -0.346605f,  0.100747f,     -0.237732f,    -0.180927f,
+  0.136491f,   -0.230076f,  1.52049f,      1.70212f,      0.415758f,
+  -0.811552f,  0.570249f,   1.73167f,      0.808393f,     -1.28608f,
+  0.518307f,   0.601205f,   0.258751f,     -0.498631f,    -0.790802f,
+  -0.465072f,  -0.507542f,  0.312956f,     -0.485421f,    -0.0525576f,
+  0.24136f,    -0.201595f,  -0.721587f,    0.190955f,     -0.123108f,
+  -0.263243f,  -0.36142f,   -0.105093f,    0.0651603f,    0.240065f,
+  0.241104f,   -0.238277f,  -1.11714f,     1.50723f,      0.443715f,
+  1.83051f,    1.29719f,    -0.0367089f,   1.57883f,      1.96011f,
+  0.0373268f,  0.480469f,   0.497726f,     1.47078f,      -0.359254f,
+  0.120578f,   -0.651964f,  -0.537215f,    -1.20527f,     -0.374681f,
+  -0.443443f,  -0.39421f,   0.688228f,     0.0989643f,    -1.28785f,
+  -1.24392f,   -0.165236f,  1.1812f,       -0.76329f,     1.44085f,
+  -0.540544f,  0.0958765f,  -0.31224f,     0.471211f,     -0.997358f,
+  1.16174f,    -0.948466f,  0.660749f,     -0.574517f,    1.42706f,
+  -0.477758f,  0.509914f,   -0.930534f,    0.722869f,     -0.955477f,
+  -0.383073f
+};
+
+static const float simple_motion_search_prune_part_logits_bias_128[] = {
+  0.814703f, -1.18485f, -1.49788f, -0.0054509f
+};
+
+static const float simple_motion_search_prune_part_layer_0_bias_128[] = {
+  -0.283185f, -2.5867f,  -2.07531f, -2.31513f,  -1.85869f,  -0.310685f,
+  -0.802324f, 0.056964f, -1.43881f, 2.84016f,   -1.15494f,  -2.30652f,
+  0.015622f,  0.303776f, 0.774322f, -1.25979f,  -2.37144f,  -1.35687f,
+  0.743607f,  0.768569f, -1.09908f, -0.470101f, -0.615497f, 1.04214f
+};
+
+static const NN_CONFIG simple_motion_search_prune_part_nn_config_128 = {
+  NUM_FEATURES_128,
+  NUM_LOGITS_128,
+  NUM_HIDDEN_LAYERS_128,
+  {
+      NUM_LAYER_0_UNITS_128,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_kernel_128,
+      simple_motion_search_prune_part_logits_kernel_128,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_bias_128,
+      simple_motion_search_prune_part_logits_bias_128,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_128
+#undef NUM_FEATURES_128
+#undef NUM_LAYER_0_UNITS_128
+#undef NUM_LOGITS_128
+
+// BLOCK_64X64
+#define NUM_HIDDEN_LAYERS_64 1
+#define NUM_FEATURES_64 19
+#define NUM_LAYER_0_UNITS_64 24
+#define NUM_LOGITS_64 10
+
+static const float simple_motion_search_prune_part_logits_kernel_64[] = {
+  -2.71432f,    0.853994f,   0.740156f,   -0.30982f,   -0.448512f,  0.0984177f,
+  -0.664821f,   -0.150188f,  0.104798f,   1.27746f,    0.217043f,   -0.553914f,
+  -0.985639f,   1.99904f,    -0.994877f,  0.0697683f,  0.64856f,    -0.412276f,
+  -0.740292f,   0.93851f,    0.647461f,   0.874707f,   -1.06881f,   0.704589f,
+  0.641148f,    -0.23997f,   -0.0955951f, 0.249869f,   -0.240726f,  0.420737f,
+  -0.558423f,   -0.863005f,  0.981276f,   -0.520796f,  0.206416f,   -0.0512005f,
+  -0.76665f,    -0.341038f,  -0.727661f,  0.348269f,   -0.0786858f, -0.370214f,
+  -0.292226f,   -0.106107f,  0.262638f,   1.0065f,     0.558744f,   -0.0492801f,
+  -1.22877f,    0.225909f,   0.485238f,   -0.009513f,  -0.299224f,  -0.405938f,
+  -0.379131f,   0.0190173f,  -0.66387f,   0.846164f,   0.298606f,   -0.811507f,
+  -0.0853147f,  -0.146277f,  0.0860954f,  0.151524f,   0.239613f,   -0.0141985f,
+  0.684708f,    -0.0732401f, -0.709585f,  -1.86211f,   -0.622393f,  -0.503456f,
+  0.482376f,    -0.10075f,   -0.449184f,  0.999368f,   0.442391f,   -0.60335f,
+  0.631041f,    0.377315f,   -0.141438f,  -0.395021f,  -0.750439f,  0.79456f,
+  0.29555f,     -1.51727f,   0.964882f,   -0.256708f,  -0.182677f,  0.727967f,
+  0.411337f,    -0.252901f,  -0.546165f,  -0.620733f,  0.948828f,   0.187498f,
+  -6.20879f,    0.1671f,     -0.131667f,  -0.0719152f, -0.101839f,  -0.326827f,
+  0.00418824f,  -0.819253f,  -0.168854f,  -0.588437f,  -0.192699f,  0.126879f,
+  -0.0287044f,  -0.686708f,  -0.445826f,  -0.0993483f, -0.682881f,  0.0685043f,
+  -0.472371f,   -0.399788f,  0.170985f,   -0.164692f,  -0.380181f,  -0.268889f,
+  -3.16193f,    -0.965943f,  -0.178737f,  -0.517274f,  -0.102274f,  -0.359742f,
+  0.0117419f,   -0.0690246f, 0.206539f,   -1.29449f,   0.0976349f,  -0.188444f,
+  0.160939f,    -0.387867f,  -0.34513f,   -0.221073f,  -0.326615f,  0.0302811f,
+  0.245349f,    -0.381586f,  0.0414228f,  -0.414076f,  0.0232891f,  -0.118837f,
+  -2.81659f,    -0.288328f,  -0.0328096f, -0.61535f,   -0.23593f,   -1.16019f,
+  -0.0619081f,  0.0646133f,  -0.313036f,  -0.390373f,  -0.0481636f, 0.0450287f,
+  0.196415f,    -0.717419f,  -0.0707908f, -0.2078f,    -0.459761f,  0.230835f,
+  -0.0703036f,  -0.112402f,  -0.170908f,  0.0485359f,  -0.237293f,  -0.989058f,
+  -2.05295f,    0.134268f,   -0.173105f,  -0.493955f,  -0.28408f,   -0.91465f,
+  -0.00206022f, -0.297255f,  0.161636f,   -0.52414f,   -0.324677f,  -0.321039f,
+  0.188412f,    -0.333399f,  -0.325067f,  -0.024484f,  0.128572f,   0.20541f,
+  -0.0743476f,  0.0762302f,  -0.0370672f, -1.53757f,   -0.199054f,  -0.318512f,
+  -3.41239f,    -0.50942f,   0.100088f,   0.395587f,   0.00633016f, 0.851201f,
+  -0.0700995f,  -0.685436f,  0.0817234f,  -0.632139f,  -0.41931f,   0.15613f,
+  -0.201166f,   -0.0721472f, -0.239791f,  0.656797f,   -0.501886f,  -0.160969f,
+  -0.561536f,   -0.491154f,  -0.213447f,  0.21926f,    -0.542261f,  0.594975f,
+  -1.75303f,    0.284615f,   -0.0128245f, -0.0316972f, -0.300138f,  -0.980628f,
+  -0.332859f,   0.19298f,    -0.248332f,  0.445278f,   -0.470188f,  -0.483686f,
+  0.361407f,    -0.363551f,  0.175335f,   -0.401025f,  0.494573f,   0.366983f,
+  0.302351f,    -0.368503f,  -0.443071f,  -0.562775f,  -0.375816f,  -0.657459f
+};
+
+static const float simple_motion_search_prune_part_layer_0_kernel_64[] = {
+  0.615272f,   -0.151592f,   0.346172f,   -0.0656032f,  0.212875f,
+  0.0798838f,  -0.146093f,   -0.230484f,  -0.620322f,   0.0277852f,
+  -0.400931f,  -0.777044f,   -0.0871401f, -0.248128f,   0.471518f,
+  -0.34174f,   -0.149884f,   -0.0763207f, 0.900372f,    0.233323f,
+  -0.788699f,  0.28221f,     0.327214f,   -0.244877f,   -0.414568f,
+  0.0472988f,  0.168193f,    0.64126f,    0.56686f,     -0.318429f,
+  0.530612f,   -0.750709f,   0.0400333f,  -0.023951f,   -0.0218727f,
+  -0.0449836f, -0.0764245f,  -0.070088f,  -0.0427561f,  -0.0873884f,
+  -0.216423f,  0.622787f,    0.0297492f,  -0.233673f,   0.510639f,
+  0.481654f,   -0.234274f,   0.347734f,   -0.480302f,   0.832741f,
+  0.228223f,   0.247019f,    -0.32416f,   0.352315f,    -1.1491f,
+  0.00806435f, -0.439654f,   0.151723f,   -0.0719021f,  0.463499f,
+  0.00134313f, 0.0553619f,   -0.582163f,  -0.215118f,   -0.291413f,
+  0.889738f,   0.284575f,    0.00450409f, -0.896799f,   0.0303598f,
+  0.187935f,   -0.160993f,   0.089854f,   0.0450689f,   -0.548899f,
+  -0.10556f,   0.29384f,     -0.154158f,  0.423591f,    0.322065f,
+  0.244633f,   -0.320866f,   0.739022f,   -0.500495f,   0.595557f,
+  -0.62001f,   0.259688f,    0.114985f,   -0.13888f,    -0.0737616f,
+  0.801863f,   -0.0606944f,  0.238712f,   -0.0636556f,  0.04317f,
+  0.359843f,   -0.273265f,   0.782092f,   0.341857f,    0.884373f,
+  0.378388f,   0.632049f,    0.633597f,   -0.32103f,    -1.1014f,
+  -0.852525f,  -0.867999f,   0.228625f,   0.444601f,    0.246595f,
+  0.386288f,   -0.341772f,   -0.335849f,  -0.482717f,   0.00928882f,
+  -0.262618f,  0.497003f,    0.579316f,   0.371681f,    -0.440324f,
+  0.115786f,   -0.112295f,   0.07564f,    -0.326419f,   -0.195152f,
+  0.315485f,   -0.368989f,   0.281852f,   0.418865f,    0.807047f,
+  0.293759f,   -0.129749f,   -1.26096f,   -0.0526466f,  -0.618876f,
+  0.330563f,   0.26479f,     0.364751f,   0.68361f,     0.500237f,
+  -0.564011f,  0.158277f,    -0.214105f,  0.50558f,     0.69422f,
+  -0.172224f,  -0.280568f,   0.17241f,    -0.500658f,   -0.00418558f,
+  -0.420356f,  0.0453594f,   0.00471851f, -0.273979f,   0.355633f,
+  0.515393f,   0.0699701f,   -0.413576f,  0.487255f,    -0.388845f,
+  -0.192441f,  -0.353274f,   -0.1773f,    0.256746f,    -0.380197f,
+  0.023061f,   0.431125f,    1.0368f,     -0.0915799f,  -0.268898f,
+  0.0408016f,  0.593356f,    0.308486f,   0.508452f,    -0.0623963f,
+  0.307066f,   0.180916f,    0.326197f,   0.246641f,    0.270665f,
+  -0.145655f,  -0.341277f,   0.0218018f,  -0.326939f,   -0.203528f,
+  -0.25496f,   -0.772783f,   -0.306114f,  -0.79778f,    0.690016f,
+  -0.183109f,  -0.076211f,   -0.474873f,  -0.0219729f,  -0.152253f,
+  -0.0315833f, -0.134908f,   -0.111871f,  0.0409385f,   -0.282586f,
+  -0.101171f,  0.351942f,    0.30023f,    0.369243f,    0.0619639f,
+  -0.170842f,  -0.235501f,   -0.615927f,  0.883574f,    0.150949f,
+  -0.338489f,  -0.213431f,   -0.6075f,    -0.107739f,   -0.251963f,
+  0.143906f,   0.254205f,    -0.260084f,  0.224897f,    0.609902f,
+  0.5606f,     0.795273f,    0.117204f,   0.345592f,    0.243912f,
+  0.0626369f,  -0.166448f,   -0.886364f,  -0.280833f,   -0.0891074f,
+  0.0372486f,  -0.138729f,   0.120189f,   0.345857f,    0.176341f,
+  0.481862f,   -0.137464f,   -0.208001f,  0.621677f,    -0.121462f,
+  0.00445475f, 0.159351f,    0.149918f,   0.198679f,    -0.489814f,
+  -0.30931f,   -0.39161f,    0.268615f,   0.323864f,    0.770061f,
+  0.27962f,    0.536173f,    0.122727f,   0.571712f,    0.48351f,
+  -0.13558f,   -0.426532f,   -0.100426f,  -0.209695f,   0.470483f,
+  -0.424789f,  0.0973791f,   -0.296483f,  -0.302057f,   -0.918591f,
+  0.16328f,    0.0335916f,   0.415762f,   -0.235873f,   -0.00934998f,
+  -0.0967718f, 0.616418f,    0.0931619f,  -0.447975f,   0.156388f,
+  -0.40139f,   0.382807f,    -0.578455f,  1.01437f,     0.37469f,
+  -0.986411f,  -0.827175f,   0.0344976f,  -0.00244486f, -0.142117f,
+  -0.344406f,  -0.0124787f,  -0.226427f,  -0.301835f,   -0.705097f,
+  0.214542f,   -0.0346878f,  -0.255193f,  0.552685f,    -0.104125f,
+  0.77408f,    0.00984395f,  0.0476041f,  -0.244548f,   -0.35082f,
+  -0.487322f,  0.700091f,    -0.138812f,  -0.0110718f,  -0.0400376f,
+  0.254488f,   -0.0867923f,  -0.215463f,  1.03355f,     0.214947f,
+  0.385536f,   0.140597f,    -0.234973f,  -0.0611124f,  1.0974f,
+  0.463592f,   -0.393788f,   -0.428132f,  -0.14664f,    -0.156533f,
+  -0.113426f,  -0.111081f,   -0.0228256f, 0.354909f,    -0.0232978f,
+  0.177394f,   0.0503901f,   -0.220314f,  0.170207f,    0.236719f,
+  0.681525f,   -0.245544f,   -0.119278f,  -0.258491f,   -0.285309f,
+  0.186384f,   -0.119177f,   -0.279398f,  -0.097341f,   -0.324634f,
+  -0.296797f,  -1.3325f,     0.147225f,   0.0158408f,   -0.0365257f,
+  -0.166737f,  -0.159859f,   -0.0765398f, 0.0973026f,   0.294529f,
+  -0.16425f,   0.241506f,    0.561487f,   0.799885f,    -0.565667f,
+  -0.254387f,  -0.360831f,   -0.138432f,  0.406999f,    0.463259f,
+  -0.497314f,  -0.00549591f, -0.0313973f, -0.386042f,   -0.364537f,
+  -0.079987f,  0.148503f,    0.0532473f,  -0.0196838f,  0.232687f,
+  0.212998f,   0.2142f,      0.409001f,   -0.104871f,   -0.217793f,
+  0.0161162f,  -0.371663f,   0.887427f,   0.420187f,    -0.649711f,
+  -0.398999f,  0.448972f,    -0.116108f,  0.0444676f,   -0.501289f,
+  -0.0410878f, 0.00447241f,  -0.402102f,  -0.0926268f,  -0.254362f,
+  -0.395526f,  0.165273f,    0.339347f,   0.91932f,     -0.280113f,
+  0.320112f,   -0.102728f,   -0.131682f,  -0.566621f,   0.386611f,
+  0.146975f,   0.333248f,    -0.179748f,  0.655328f,    0.222658f,
+  0.349512f,   -0.135588f,   -0.22506f,   -0.353001f,   -0.400244f,
+  -0.815125f,  0.72407f,     0.281178f,   -0.109426f,   -0.0977383f,
+  -0.679406f,  -0.401318f,   0.308941f,   -0.594461f,   0.482289f,
+  -0.168467f,  0.76529f,     -0.155056f,  0.0874018f,   0.266124f,
+  -0.0808725f, 0.078792f,    -0.33271f,   -1.0637f,     -0.172125f,
+  0.427025f,   0.332865f,    0.625523f,   0.264166f,    -0.843882f,
+  -0.275777f,  -0.657372f,   0.172814f,   0.0490389f,   -0.561928f,
+  0.0998399f,  0.106047f,    0.39061f,    0.422161f,    0.46369f,
+  -0.364271f,  -0.326968f,   -0.527243f,  -0.618535f,   -0.287804f,
+  -0.211365f,  0.349277f,    0.456535f,   -0.0815191f,  -0.156835f,
+  0.57273f
+};
+
+static const float simple_motion_search_prune_part_logits_bias_64[] = {
+  0.272852f,  -0.260088f, 0.11881f,   0.146764f,  0.00289019f,
+  -0.559007f, -0.203668f, -0.268267f, -0.218921f, -0.260717f
+};
+
+static const float simple_motion_search_prune_part_layer_0_bias_64[] = {
+  -2.12804f,   0.445147f,  -0.0210471f, 0.533662f,   0.662023f,  0.020519f,
+  -0.0452071f, -0.343995f, 0.112379f,   -0.16776f,   0.112004f,  0.932147f,
+  -1.22944f,   -0.822248f, 0.370083f,   -0.0951491f, 0.0631258f, -0.875801f,
+  0.996276f,   0.116103f,  0.925914f,   0.054293f,   -1.02697f,  -0.359516f
+};
+
+static const NN_CONFIG simple_motion_search_prune_part_nn_config_64 = {
+  NUM_FEATURES_64,
+  NUM_LOGITS_64,
+  NUM_HIDDEN_LAYERS_64,
+  {
+      NUM_LAYER_0_UNITS_64,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_kernel_64,
+      simple_motion_search_prune_part_logits_kernel_64,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_bias_64,
+      simple_motion_search_prune_part_logits_bias_64,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_64
+#undef NUM_FEATURES_64
+#undef NUM_LAYER_0_UNITS_64
+#undef NUM_LOGITS_64
+
+// BLOCK_32X32
+#define NUM_HIDDEN_LAYERS_32 1
+#define NUM_FEATURES_32 19
+#define NUM_LAYER_0_UNITS_32 24
+#define NUM_LOGITS_32 10
+
+static const float simple_motion_search_prune_part_logits_kernel_32[] = {
+  -1.22965f,   0.49978f,    -0.376271f,  0.0615445f,  -0.212491f,  0.58427f,
+  0.142683f,   2.62142f,    -0.459751f,  -0.570143f,  0.72257f,    -0.64127f,
+  0.266919f,   0.3235f,     0.64405f,    -0.0398269f, -1.82822f,   -0.366573f,
+  0.753387f,   -0.274241f,  -0.484971f,  -0.503907f,  -0.911176f,  -0.455827f,
+  -0.706345f,  0.0687611f,  -0.362788f,  -0.369416f,  -0.530408f,  0.0812615f,
+  -0.863012f,  -0.613333f,  -0.268833f,  0.0982181f,  0.19374f,    -0.147078f,
+  -0.0200551f, 0.363817f,   0.179229f,   0.429671f,   0.905826f,   -0.441191f,
+  0.48874f,    0.539192f,   -0.418992f,  -0.468762f,  -0.482158f,  -0.770778f,
+  -0.743739f,  0.0175955f,  -0.40662f,   0.0276107f,  0.0110658f,  -0.119298f,
+  0.698777f,   -0.187602f,  -0.300483f,  -0.147972f,  0.324042f,   -0.0265165f,
+  -0.375496f,  0.245762f,   -0.111271f,  -0.829879f,  -0.434606f,  -0.367099f,
+  0.113229f,   -0.755101f,  -0.0623393f, -0.13764f,   -0.205864f,  -0.332027f,
+  0.966638f,   0.477078f,   0.262501f,   -0.110232f,  0.141242f,   -0.699602f,
+  -0.14943f,   -2.36868f,   0.556236f,   0.345883f,   -0.466765f,  0.279446f,
+  -0.56809f,   -0.356351f,  -0.503855f,  -0.607959f,  -0.0328514f, 0.313352f,
+  -0.644447f,  0.221808f,   0.254575f,   0.476802f,   1.40433f,    0.317169f,
+  0.109761f,   -0.0105197f, -0.22896f,   -0.389234f,  0.423948f,   -0.272114f,
+  -1.02704f,   -1.50706f,   0.0563541f,  -0.0371609f, -0.306187f,  0.0570354f,
+  -0.367998f,  0.0949103f,  -0.254703f,  -0.0853516f, 1.56471f,    -0.594372f,
+  -0.23582f,   0.267108f,   0.015297f,   -0.318317f,  0.287599f,   -0.687022f,
+  0.360804f,   -0.177637f,  -0.118144f,  -0.418225f,  -0.985658f,  -0.391334f,
+  -0.772539f,  -0.44002f,   0.0722112f,  -0.0825626f, 0.0466844f,  0.0525818f,
+  -0.788047f,  0.120401f,   -0.785688f,  -0.346177f,  0.240363f,   0.0248826f,
+  -0.0815654f, 0.39317f,    -0.122372f,  0.199969f,   0.202777f,   -0.1979f,
+  0.313558f,   -0.180202f,  0.191204f,   -0.313598f,  0.155464f,   -0.600106f,
+  0.0230461f,  -0.718126f,  0.266874f,   -0.313614f,  0.0352148f,  0.131058f,
+  -0.221112f,  0.115419f,   -1.41928f,   -0.909453f,  0.341747f,   -0.25318f,
+  -0.312476f,  -0.288919f,  -0.0135884f, -0.278496f,  0.381449f,   -0.0725586f,
+  0.254479f,   -0.0658647f, -0.694657f,  -0.0653643f, -0.0233378f, -0.484902f,
+  -0.104919f,  -1.52446f,   0.0140919f,  -0.142396f,  -0.11836f,   0.113235f,
+  -1.21564f,   0.0618471f,  0.0421833f,  -0.883143f,  0.288008f,   -0.278322f,
+  -0.185151f,  -0.616223f,  0.334549f,   0.370898f,   0.24646f,    -0.027439f,
+  0.366905f,   -0.40628f,   0.0260464f,  -0.506151f,  0.0191214f,  -0.104617f,
+  -0.848722f,  -1.66357f,   -0.194607f,  0.0346545f,  -0.412502f,  0.200471f,
+  -0.285588f,  0.0309609f,  -0.150546f,  0.569096f,   0.210382f,   0.14322f,
+  0.908182f,   1.23553f,    -0.432935f,  -0.0963189f, -0.610778f,  -0.532583f,
+  -0.302028f,  -0.600906f,  -0.0777476f, 0.0230625f,  0.375802f,   -0.466292f,
+  1.01104f,    -1.16706f,   0.0121855f,  0.0100479f,  -0.147181f,  0.255058f,
+  -0.285898f,  -0.167804f,  -0.240171f,  -1.15287f,   -1.51181f,   0.0859809f,
+  0.0811324f,  -0.829305f,  0.272793f,   0.407484f,   0.324943f,   0.230197f
+};
+
+static const float simple_motion_search_prune_part_layer_0_kernel_32[] = {
+  -0.0711848f, 0.187893f,     -0.0532872f,  -0.0403218f, 0.132223f,
+  0.0150309f,  0.132917f,     -0.00883543f, -0.0614934f, 0.207838f,
+  -0.102772f,  0.127833f,     -0.332145f,   0.00237219f, -0.338402f,
+  0.123288f,   -0.248341f,    -0.0182967f,  -1.16283f,   -0.166611f,
+  -0.220962f,  -0.268939f,    -0.207455f,   -0.214904f,  -0.0639896f,
+  -0.0729676f, -0.000552725f, 0.217902f,    0.350537f,   -0.0217492f,
+  0.136239f,   0.0641448f,    0.515814f,    0.525507f,   0.240539f,
+  0.109375f,   0.322722f,     -1.20791f,    0.323287f,   0.0231209f,
+  -0.194629f,  0.143504f,     0.193943f,    -0.15059f,   -0.0156431f,
+  -0.210354f,  0.13426f,      -0.379795f,   0.0585494f,  0.0947257f,
+  0.948153f,   0.492681f,     -0.977907f,   -0.30367f,   0.216922f,
+  -0.366185f,  0.00734769f,   0.528627f,    0.334327f,   0.0222445f,
+  0.400606f,   -0.0313966f,   -0.047664f,   -0.512993f,  -0.433836f,
+  0.340769f,   -0.131347f,    -0.518235f,   -0.220933f,  -0.379338f,
+  -0.116978f,  0.143766f,     0.456281f,    -0.0921022f, -0.0109533f,
+  -0.107467f,  0.335672f,     -0.313228f,   0.175586f,   0.108769f,
+  0.129646f,   -0.053502f,    0.272574f,    0.220172f,   1.51046f,
+  0.255032f,   -0.162058f,    -0.770721f,   0.238623f,   -0.65159f,
+  -0.0354393f, -0.577323f,    0.0954972f,   -0.661719f,  0.227893f,
+  -0.0584787f, 0.234849f,     -0.16515f,    -0.247194f,  0.12563f,
+  0.39043f,    -0.178219f,    -0.0629665f,  0.395761f,   0.24224f,
+  -0.336032f,  0.552893f,     -0.300073f,   -0.235499f,  -0.33557f,
+  0.0398952f,  -0.853526f,    -0.550247f,   1.13965f,    0.197325f,
+  -0.0750203f, 0.145096f,     0.128642f,    0.113606f,   0.0972738f,
+  0.219874f,   0.252284f,     0.833398f,    0.398332f,   0.802753f,
+  0.650573f,   -0.725374f,    -0.548798f,   -0.668723f,  -0.705524f,
+  -0.929613f,  -0.138145f,    0.161306f,    0.578878f,   0.0605056f,
+  0.720406f,   0.163571f,     0.707962f,    -0.0777347f, 0.52085f,
+  0.279096f,   0.180012f,     -0.459535f,   0.219678f,   -0.273058f,
+  -0.030387f,  -0.262664f,    0.367791f,    -0.899145f,  -1.07191f,
+  -1.15497f,   0.460442f,     0.020009f,    -0.329794f,  0.846777f,
+  0.287855f,   0.603544f,     0.645424f,    0.00753003f, -0.0666031f,
+  -0.141114f,  -0.149588f,    -0.431897f,   -0.591811f,  0.134026f,
+  0.0933469f,  -0.130808f,    -0.0335611f,  -0.488446f,  -0.275566f,
+  -0.937488f,  0.245793f,     -0.548915f,   0.0108641f,  -0.216292f,
+  0.00680863f, -0.113137f,    -0.137064f,   0.083321f,   0.689742f,
+  -0.213193f,  0.172408f,     -0.123794f,   0.383401f,   -0.496176f,
+  0.427836f,   -0.644447f,    0.967995f,    0.181371f,   0.301512f,
+  -0.376098f,  -0.0571203f,   0.0012891f,   0.467542f,   -0.400791f,
+  -0.204521f,  0.253847f,     0.107368f,    -0.434741f,  -0.389665f,
+  0.30953f,    0.053949f,     -0.0415711f,  0.0432576f,  -0.319656f,
+  -0.228939f,  -0.0928546f,   0.164408f,    1.26655f,    0.991928f,
+  -0.0596439f, 0.0786943f,    0.001457f,    0.282831f,   0.0975484f,
+  0.13724f,    0.0891527f,    -0.0471925f,  0.255467f,   -0.469073f,
+  -0.0759757f, 0.331197f,     0.139434f,    -0.0118866f, 0.359463f,
+  -0.774947f,  -0.474647f,    -1.36563f,    0.24901f,    0.776025f,
+  0.150621f,   0.192407f,     0.752499f,    0.0435322f,  -0.0334263f,
+  0.470741f,   -0.781907f,    -0.149471f,   -0.649484f,  -0.866079f,
+  0.536967f,   0.541611f,     -0.00237101f, -0.119688f,  -0.594591f,
+  -0.967503f,  0.545025f,     -0.335037f,   -0.172105f,  -0.532514f,
+  -0.442211f,  -0.0864824f,   0.038418f,    -0.337492f,  -0.412653f,
+  0.0778177f,  0.0824839f,    -0.232478f,   0.553077f,   0.179073f,
+  0.222848f,   -0.261019f,    0.854277f,    -0.643185f,  0.476021f,
+  -0.0174448f, 0.621669f,     -0.527978f,   0.45279f,    -0.286967f,
+  0.328291f,   0.0627781f,    0.495794f,    0.0156424f,  -0.170515f,
+  -0.435815f,  -0.264061f,    -0.964473f,   0.259361f,   -0.297514f,
+  0.667044f,   0.719631f,     -0.812974f,   -0.350331f,  0.594795f,
+  -0.106071f,  -0.409225f,    0.283481f,    0.5735f,     -0.231666f,
+  -0.270059f,  0.687725f,     0.349693f,    0.0637911f,  0.208738f,
+  0.299087f,   -0.249188f,    -0.426365f,   -1.01906f,   0.0919539f,
+  -0.198362f,  0.256663f,     -0.302723f,   0.457283f,   0.433322f,
+  0.242223f,   0.740515f,     0.247248f,    0.436274f,   0.588906f,
+  0.852493f,   0.385838f,     0.243232f,    -0.501191f,  0.212121f,
+  0.297253f,   -0.664029f,    -0.470196f,   0.0175681f,  -1.01849f,
+  -0.619865f,  -1.26982f,     -0.088134f,   -0.29233f,   -0.196112f,
+  -0.141886f,  0.0602684f,    0.469327f,    0.0227135f,  0.275863f,
+  0.161156f,   -0.864189f,    -0.72576f,    0.62214f,    0.804845f,
+  0.183776f,   -0.155777f,    -0.0826676f,  -0.23302f,   0.0845181f,
+  -0.126365f,  0.0925764f,    0.889037f,    -0.208894f,  0.441273f,
+  0.476529f,   0.510144f,     0.321464f,    0.764386f,   0.198938f,
+  -0.669809f,  -0.13214f,     -0.426596f,   -0.547804f,  -0.0953238f,
+  0.164997f,   0.275626f,     -0.0469242f,  -0.923263f,  -1.05855f,
+  -0.487197f,  0.261922f,     -0.108634f,   0.409852f,   0.142348f,
+  0.153656f,   -0.119714f,    0.490507f,    -0.26725f,   -0.0651784f,
+  -0.962803f,  -0.596098f,    -0.686329f,   0.66723f,    0.290971f,
+  0.692586f,   0.56926f,      -0.0494205f,  -0.498585f,  -0.1832f,
+  -0.571507f,  -0.438308f,    0.607958f,    0.985154f,   0.131991f,
+  0.0469712f,  0.415769f,     -0.518799f,   -0.353338f,  -0.219197f,
+  -0.159141f,  0.085484f,     -0.405673f,   -0.222816f,  0.813276f,
+  0.323344f,   -0.235178f,    -0.40799f,    -0.083177f,  0.514834f,
+  0.177916f,   0.0197744f,    -0.523574f,   -0.445548f,  -0.21085f,
+  0.760991f,   0.170654f,     -0.00312635f, -0.175464f,  0.665364f,
+  -0.713881f,  -0.701069f,    -0.628863f,   1.03196f,    -0.0414137f,
+  -0.0163186f, 0.16796f,      -0.0263498f,  0.484873f,   0.153922f,
+  0.0593931f,  0.171615f,     0.330269f,    -0.0728948f, 0.42332f,
+  0.294273f,   -0.0185935f,   -0.609248f,   -0.305841f,  -0.293534f,
+  0.0766015f,  0.130506f,     0.16716f,     0.0168997f,  -0.429074f,
+  -0.744995f,  -0.972633f,    0.50621f,     -0.275645f,  0.402252f,
+  -0.0190824f, 0.637343f,     -0.410791f,   0.546844f,   -0.555088f,
+  0.491182f,   -0.0369695f,   0.974657f,    -0.142378f,  -0.148588f,
+  -0.217919f,  0.0146815f,    -0.419912f,   0.111356f,   -0.640505f,
+  -0.352691f
+};
+
+static const float simple_motion_search_prune_part_logits_bias_32[] = {
+  0.0868818f, 0.370904f,  0.126136f,  0.57768f,   -0.316245f,
+  -0.526958f, -0.451775f, -0.595569f, -0.805166f, -0.592605f
+};
+
+static const float simple_motion_search_prune_part_layer_0_bias_32[] = {
+  -1.85942f, 0.727049f,   -0.0421259f, 0.759194f,   0.241057f,  0.0752973f,
+  0.321934f, -0.593116f,  -0.873202f,  -0.238779f,  0.851339f,  -0.536633f,
+  -0.18552f, -0.0941714f, 0.0153855f,  -1.34736f,   -0.870722f, 0.0973251f,
+  0.491483f, 0.650638f,   -0.205986f,  -0.0499654f, -1.00592f,  0.65992f
+};
+
+static const NN_CONFIG simple_motion_search_prune_part_nn_config_32 = {
+  NUM_FEATURES_32,
+  NUM_LOGITS_32,
+  NUM_HIDDEN_LAYERS_32,
+  {
+      NUM_LAYER_0_UNITS_32,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_kernel_32,
+      simple_motion_search_prune_part_logits_kernel_32,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_bias_32,
+      simple_motion_search_prune_part_logits_bias_32,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_32
+#undef NUM_FEATURES_32
+#undef NUM_LAYER_0_UNITS_32
+#undef NUM_LOGITS_32
+
+// BLOCK_16X16
+#define NUM_HIDDEN_LAYERS_16 1
+#define NUM_FEATURES_16 19
+#define NUM_LAYER_0_UNITS_16 16
+#define NUM_LOGITS_16 10
+
+static const float simple_motion_search_prune_part_logits_kernel_16[] = {
+  0.274123f,   -0.551439f,  0.578938f,   -0.213001f,  -0.310021f,
+  0.0301311f,  -0.377388f,  -3.16339f,   0.434616f,   0.264216f,
+  0.320385f,   -1.41657f,   1.05919f,    -0.396075f,  -0.855458f,
+  0.418105f,   -0.0769093f, -0.54773f,   -0.215041f,  -0.13885f,
+  -0.209062f,  -0.526914f,  0.211082f,   -2.34741f,   -0.000587922f,
+  -0.431294f,  0.757511f,   -0.184199f,  -0.623552f,  -0.221362f,
+  -0.628871f,  0.646393f,   -0.192193f,  0.0166809f,  -0.061691f,
+  -0.214283f,  -0.35264f,   0.361452f,   -0.0160958f, -0.0475499f,
+  0.196822f,   0.44674f,    -0.408064f,  -0.559148f,  -0.67489f,
+  -0.202201f,  -0.358012f,  -0.341735f,  -0.325682f,  0.220407f,
+  -0.642416f,  0.350069f,   0.430505f,   -0.0997745f, -0.300435f,
+  1.25364f,    -0.869717f,  -0.0728051f, -0.172433f,  0.385328f,
+  -2.13737f,   0.673629f,   0.580102f,   -0.0745359f, -0.442542f,
+  -0.208393f,  -0.44303f,   0.318855f,   0.0557663f,  0.268257f,
+  -0.0941132f, -4.52876f,   -0.638897f,  -0.907808f,  0.28328f,
+  0.619328f,   -1.64249f,   0.0241431f,  -0.0880383f, 0.299287f,
+  -0.248316f,  0.039253f,   -0.544645f,  -0.141389f,  0.0185283f,
+  -1.01973f,   -0.269349f,  -4.88803f,   -0.632892f,  0.181151f,
+  0.193291f,   0.592713f,   -1.71466f,   0.398923f,   0.00355734f,
+  0.212158f,   -0.949188f,  0.386679f,   -0.243317f,  -0.17458f,
+  0.0576348f,  0.096034f,   -0.282317f,  -4.13556f,   -0.516689f,
+  0.169427f,   -0.541422f,  0.529281f,   -1.45161f,   -0.0417835f,
+  0.0174299f,  -0.26404f,   -0.105773f,  0.102701f,   -0.65039f,
+  0.308121f,   -0.185395f,  0.273611f,   -0.408464f,  -6.09866f,
+  -0.512249f,  0.0668662f,  -0.547454f,  0.62565f,    -1.63952f,
+  0.510116f,   0.127461f,   -0.506263f,  0.113016f,   -0.258605f,
+  -0.308066f,  0.0957733f,  0.059395f,   -0.915338f,  0.499407f,
+  -10.3199f,   -0.32224f,   -0.677307f,  1.13218f,    -0.0888689f,
+  -0.108029f,  -0.156533f,  -0.391431f,  0.576607f,   -0.0833897f,
+  0.536752f,   -0.0832957f, -0.0957492f, -0.135149f,  0.423881f,
+  0.317691f,   -7.29755f,   0.104775f,   0.534134f,   -0.797288f,
+  -0.386335f,  -0.20785f,   -0.0507494f, -0.338127f,  -0.868801f
+};
+
+static const float simple_motion_search_prune_part_layer_0_kernel_16[] = {
+  -0.244782f,   0.00425059f, -0.337534f,  -0.423125f,   -0.349549f,
+  -0.00261207f, -0.0845866f, -0.259278f,  -0.375872f,   0.00052185f,
+  -0.427483f,   0.175625f,   -0.357737f,  0.129946f,    0.97588f,
+  1.19281f,     -0.51292f,   -0.100629f,  1.04044f,     0.197375f,
+  -0.210977f,   -0.20334f,   0.173408f,   0.431999f,    0.083979f,
+  -0.334416f,   -0.171408f,  0.160035f,   -0.0464933f,  0.164574f,
+  -0.162781f,   1.01648f,    0.624844f,   -1.20372f,    -0.698742f,
+  0.335747f,    -0.362217f,  -0.0817772f, -0.0366522f,  -0.249627f,
+  -0.3135f,     0.256151f,   -0.257869f,  0.0816391f,   -0.256592f,
+  0.071219f,    -0.295956f,  0.287213f,   -0.375107f,   -0.149275f,
+  0.0449292f,   0.49489f,    -0.958944f,  0.386166f,    -0.76147f,
+  -0.223786f,   1.91139f,    -0.0866129f, -0.516556f,   0.871813f,
+  0.389086f,    0.3205f,     0.605997f,   -0.531384f,   -0.693121f,
+  0.728219f,    0.223733f,   -0.58534f,   -0.410249f,   0.0271334f,
+  -0.43904f,    0.325998f,   -0.123988f,  0.103988f,    -0.2227f,
+  0.0779071f,   0.100073f,   -0.924956f,  -0.00472742f, -0.21932f,
+  -0.220187f,   -0.719278f,  0.099118f,   -0.0891471f,  0.743678f,
+  -0.173308f,   0.588225f,   -0.31524f,   1.1415f,      0.408762f,
+  -0.213088f,   -0.291621f,  0.655754f,   0.531725f,    -0.013178f,
+  -0.100753f,   -0.0775942f, -0.181806f,  -0.441297f,   -0.00589338f,
+  0.14217f,     -0.0592671f, 0.10907f,    1.47619f,     0.813817f,
+  0.1996f,      0.178263f,   -0.569958f,  -0.374858f,   -0.743922f,
+  -0.269343f,   -0.118752f,  0.0797114f,  0.027829f,    -0.133402f,
+  -0.164884f,   -0.42919f,   0.0682661f,  -0.206438f,   0.0184534f,
+  -0.141316f,   -0.138817f,  0.0305671f,  0.295578f,    -0.236413f,
+  0.172918f,    -0.35082f,   -0.0986917f, -0.650484f,   0.13655f,
+  0.212195f,    0.549501f,   1.64941f,    0.152883f,    0.0117464f,
+  -0.200688f,   -0.0412075f, -0.0296324f, 0.122624f,    0.1871f,
+  -0.515693f,   0.20351f,    0.333617f,   0.254752f,    0.293802f,
+  -0.428904f,   -0.0312053f, -0.879481f,  -0.516353f,   -0.417837f,
+  -0.230442f,   0.229367f,   0.615339f,   0.327347f,    0.288747f,
+  0.596166f,    0.497054f,   0.536953f,   0.462426f,    0.517022f,
+  -0.287359f,   -0.162824f,  -0.249174f,  -0.364887f,   -0.538065f,
+  -0.506783f,   -0.163213f,  -0.617007f,  -0.70433f,    -0.759151f,
+  -0.235183f,   -0.196496f,  0.0925943f,  -0.0786896f,  0.135073f,
+  -0.106243f,   -0.190647f,  -0.311848f,  -0.428009f,   0.467477f,
+  0.259139f,    1.4948f,     0.986061f,   -0.534924f,   -0.193525f,
+  -0.66125f,    -0.693799f,  -0.489594f,  0.238264f,    0.249978f,
+  -0.365372f,   -0.197998f,  -0.664471f,  -0.0312145f,  -0.297293f,
+  0.0278792f,   -0.877716f,  0.0473106f,  0.913044f,    0.815569f,
+  0.789972f,    0.641374f,   -0.122848f,  -0.353137f,   -0.381102f,
+  -0.960138f,   -0.166101f,  0.230364f,   0.534173f,    0.398882f,
+  0.0599076f,   0.00176035f, 0.291062f,   0.021087f,    0.374562f,
+  0.402516f,    0.144006f,   -0.149598f,  0.065063f,    -0.0774656f,
+  -0.18804f,    -0.21123f,   0.0119723f,  -0.244286f,   0.109912f,
+  -0.98136f,    -0.583136f,  -1.3897f,    0.33196f,     0.319226f,
+  0.490015f,    0.316438f,   0.565049f,   0.163159f,    0.523771f,
+  0.325115f,    -0.0515304f, -0.0827355f, 0.112763f,    -0.517608f,
+  -0.161398f,   -0.124513f,  -0.0384277f, -0.49491f,    -1.53384f,
+  -1.3086f,     1.20601f,    0.399591f,   0.233774f,    0.0154149f,
+  -0.431196f,   0.187116f,   0.033398f,   0.655468f,    -0.390814f,
+  0.418144f,    -0.572053f,  0.936493f,   -0.116933f,   -0.20803f,
+  -0.908834f,   1.2524f,     -0.285888f,  0.227437f,    -0.317727f,
+  -0.6331f,     0.159656f,   0.244652f,   0.169768f,    0.379279f,
+  0.37089f,     0.118264f,   0.150946f,   0.536513f,    -0.261398f,
+  -0.15329f,    -0.413112f,  -0.243792f,  -0.215096f,   0.288741f,
+  -0.193763f,   -0.130652f,  -0.328159f,  -0.719996f,   -1.66213f,
+  0.0815627f,   0.129643f,   -0.0974464f, -0.0660102f,  0.349602f,
+  0.120356f,    -0.144144f,  -0.13641f,   -0.154818f,   -0.468552f,
+  -0.442585f,   -0.312206f,  0.879662f,   0.846103f,    0.0485413f,
+  0.18676f,     -0.477959f,  -0.112404f,  -0.366477f
+};
+
+static const float simple_motion_search_prune_part_logits_bias_16[] = {
+  0.791815f,  0.888054f,  0.406071f, -0.217564f, -0.504713f,
+  -0.514339f, -0.200842f, -1.09239f, -1.78297f,  -2.41754f
+};
+
+static const float simple_motion_search_prune_part_layer_0_bias_16[] = {
+  0.89469f,   0.10871f,  0.664404f,  -0.00639695f, -0.745647f, 0.910444f,
+  0.168436f,  -4.05854f, 1.00981f,   0.743352f,    -0.73947f,  -2.5654f,
+  -0.654753f, 0.602252f, -0.973004f, 1.1113f
+};
+
+static const NN_CONFIG simple_motion_search_prune_part_nn_config_16 = {
+  NUM_FEATURES_16,
+  NUM_LOGITS_16,
+  NUM_HIDDEN_LAYERS_16,
+  {
+      NUM_LAYER_0_UNITS_16,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_kernel_16,
+      simple_motion_search_prune_part_logits_kernel_16,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_bias_16,
+      simple_motion_search_prune_part_logits_bias_16,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_16
+#undef NUM_FEATURES_16
+#undef NUM_LAYER_0_UNITS_16
+#undef NUM_LOGITS_16
+
+// BLOCK_8X8
+#define NUM_HIDDEN_LAYERS_8 1
+#define NUM_FEATURES_8 19
+#define NUM_LAYER_0_UNITS_8 24
+#define NUM_LOGITS_8 4
+
+static const float simple_motion_search_prune_part_logits_kernel_8[] = {
+  -0.806717f,  0.0244927f,  -0.0280725f, 0.210426f,   0.0820434f,  -0.0222324f,
+  -0.0728984f, -0.424785f,  -0.0568035f, -0.0744017f, 0.204282f,   0.43611f,
+  0.319456f,   0.119903f,   -0.333224f,  -0.932499f,  0.105907f,   0.0905287f,
+  -0.612538f,  0.0978191f,  -0.0858663f, 0.0468925f,  -0.312509f,  0.0527469f,
+  0.308466f,   -0.139768f,  -0.468023f,  -0.069522f,  0.248949f,   0.0408582f,
+  0.199935f,   -0.448705f,  0.242315f,   0.00472276f, -0.468445f,  -0.591226f,
+  0.107975f,   0.261228f,   -0.10724f,   0.104639f,   -0.211234f,  -0.304658f,
+  -0.261689f,  -0.259529f,  0.0331369f,  -0.468468f,  -0.0555605f, -0.0495414f,
+  0.247612f,   -0.112518f,  -0.393122f,  0.302172f,   -0.252151f,  0.076605f,
+  -0.34155f,   -0.182731f,  0.154079f,   -0.464538f,  -0.396327f,  -0.654858f,
+  0.0921198f,  0.202867f,   -0.0712573f, 0.0349611f,  0.243043f,   -0.177452f,
+  -0.205536f,  -0.268024f,  0.0358315f,  -0.560347f,  -0.204207f,  -0.0890628f,
+  -1.013f,     -0.461748f,  -0.429801f,  -0.100352f,  -0.34369f,   0.115318f,
+  -0.556285f,  -0.0751575f, 0.183418f,   -0.285223f,  0.197177f,   -0.825216f,
+  -0.125709f,  -0.100452f,  0.180481f,   0.764462f,   0.136609f,   -0.660743f,
+  0.212887f,   -0.410408f,  -0.330839f,  -0.714646f,  -0.115863f,  -0.389509f
+};
+
+static const float simple_motion_search_prune_part_layer_0_kernel_8[] = {
+  0.0561017f,   0.530692f,    0.129047f,   0.296138f,   0.407593f,
+  0.151159f,    0.180472f,    0.298995f,   -0.0265362f, -0.380313f,
+  -0.136716f,   -0.0644648f,  0.00808564f, 0.0875876f,  -0.0719683f,
+  -0.0996071f,  -1.49722f,    -1.31176f,   1.54169f,    -0.50717f,
+  -0.0186767f,  -0.525924f,   0.0402954f,  -0.553913f,  0.139015f,
+  0.171752f,    -0.376497f,   0.485084f,   0.182424f,   -0.0236465f,
+  0.380048f,    -0.185493f,   0.554435f,   0.241565f,   -0.225917f,
+  -1.08637f,    -0.305121f,   1.3603f,     0.0806333f,  -0.0697466f,
+  0.216192f,    0.0493694f,   0.0620039f,  -0.0042489f, 0.0707283f,
+  0.0998331f,   -0.0252022f,  0.342077f,   -0.200664f,  0.335261f,
+  -0.100772f,   0.212881f,    -0.374258f,  0.310442f,   -1.95981f,
+  -1.16615f,    1.88243f,     0.503536f,   -0.0919037f, 0.392323f,
+  0.33149f,     0.563121f,    -0.0898854f, 0.130267f,   0.404959f,
+  1.06767f,     0.75134f,     0.810919f,   0.865237f,   -1.08001f,
+  -0.949809f,   -1.24941f,    -0.95081f,   -0.539575f,  -0.355324f,
+  -0.110592f,   -0.311277f,   0.023022f,   -0.039287f,  0.491298f,
+  -0.248853f,   -0.294741f,   0.288811f,   0.470638f,   0.330094f,
+  0.146603f,    0.0958486f,   0.308326f,   -0.87134f,   -1.06074f,
+  -0.315428f,   0.717577f,    -0.0174599f, -0.491904f,  0.977014f,
+  0.922693f,    0.0265707f,   -0.743468f,  -0.273647f,  -0.430625f,
+  -0.473789f,   0.119983f,    0.476441f,   0.639803f,   -0.0753475f,
+  -0.0746323f,  -0.0422539f,  0.013454f,   -0.0174412f, -0.253015f,
+  -0.279585f,   0.11562f,     -0.396919f,  0.306388f,   0.430346f,
+  0.275945f,    -0.196508f,   0.00180146f, 0.396652f,   0.324022f,
+  -0.142132f,   0.1768f,      -0.0746152f, 0.314189f,   -0.031924f,
+  0.254196f,    -0.253449f,   0.1537f,     -0.98635f,   -0.850926f,
+  0.0762147f,   -0.652507f,   1.16528f,    -0.122595f,  -0.00561333f,
+  0.424954f,    -0.00258127f, -0.421957f,  -0.0714059f, 0.782031f,
+  -0.0129133f,  -0.00544328f, -0.193527f,  -0.0936037f, -0.00490983f,
+  -0.852396f,   -0.649662f,   0.820325f,   0.484966f,   0.265405f,
+  -0.33032f,    0.104917f,    -0.364478f,  0.26858f,    0.068847f,
+  -0.0751046f,  0.238931f,    -0.151554f,  0.0296385f,  -0.14869f,
+  -0.173547f,   -0.397285f,   0.0561029f,  -0.0681254f, 0.222631f,
+  -0.00800993f, 0.0686761f,   0.196754f,   0.249358f,   0.348125f,
+  -2.22757f,    0.186675f,    -0.0426998f, 0.123301f,   -0.0196412f,
+  0.104882f,    0.138482f,    0.340663f,   0.0684671f,  -1.22841f,
+  -1.08539f,    -1.1055f,     -0.732678f,  0.766626f,   0.727742f,
+  1.26906f,     0.812594f,    -0.446312f,  -0.166752f,  0.396628f,
+  0.00141587f,  0.0137288f,   -0.0920456f, 0.123027f,   -0.118517f,
+  -0.0412321f,  -0.124952f,   0.00875945f, 0.0149883f,  -0.0218859f,
+  -0.233299f,   -0.0812722f,  0.269917f,   0.0246593f,  0.0915655f,
+  -0.019122f,   0.706718f,    -0.172051f,  -1.85551f,   0.0578997f,
+  0.100269f,    0.131412f,    -0.0952751f, 0.181419f,   -0.0106305f,
+  0.106179f,    -0.142344f,   0.0916649f,  -0.41298f,   0.113846f,
+  -0.322459f,   0.0415754f,   -0.2205f,    -0.224231f,  0.147907f,
+  -0.135093f,   -0.398757f,   1.93381f,    0.20456f,    -0.205381f,
+  -0.263999f,   0.364463f,    -0.424052f,  -0.0166924f, -0.319968f,
+  0.0439144f,   -0.692733f,   0.678982f,   -0.445194f,  -0.527451f,
+  0.383248f,    0.330927f,    -0.0367725f, 0.572847f,   -0.620804f,
+  -0.286471f,   1.39367f,     0.470732f,   0.0751676f,  0.637958f,
+  0.574865f,    0.411091f,    0.263806f,   0.555064f,   0.047788f,
+  0.201792f,    -0.0821822f,  -0.0226181f, -0.176679f,  -0.612642f,
+  -0.992254f,   -0.292099f,   -0.562201f,  -0.567948f,  -0.767681f,
+  1.33645f,     0.0932564f,   -0.421675f,  0.198262f,   -0.401979f,
+  0.244609f,    -0.261484f,   0.0304534f,  -0.200004f,  0.00951232f,
+  -0.085364f,   0.555152f,    -0.243778f,  0.371369f,   0.152997f,
+  0.154443f,    -0.281229f,   0.967176f,   1.0212f,     0.276169f,
+  0.0962747f,   0.449737f,    0.300859f,   0.376659f,   0.165669f,
+  0.332263f,    0.238315f,    0.376668f,   -0.139238f,  -0.0694765f,
+  -0.256137f,   -0.262146f,   -0.246265f,  0.0261978f,  -0.118975f,
+  -0.207184f,   -0.614041f,   -0.866987f,  -1.75604f,   -0.04917f,
+  0.0794673f,   -0.355224f,   0.0035685f,  -0.340216f,  0.398567f,
+  -0.314826f,   -0.12464f,    0.693428f,   0.701697f,   0.511653f,
+  0.861303f,    -0.64449f,    -1.06257f,   -0.773944f,  -0.838084f,
+  0.802429f,    -0.24573f,    0.720913f,   -0.31002f,   0.615694f,
+  0.287994f,    0.18495f,     0.0188525f,  0.374338f,   -0.12243f,
+  0.613532f,    0.132456f,    0.177285f,   -0.187238f,  0.0872563f,
+  -0.425068f,   0.154711f,    -0.24283f,   -0.193267f,  -1.14008f,
+  -1.17759f,    0.517845f,    0.458782f,   -0.161137f,  0.142772f,
+  0.2811f,      0.171871f,    0.276129f,   0.018275f,   0.390464f,
+  -0.156414f,   0.108248f,    -0.0909085f, -0.11346f,   0.0815138f,
+  0.0729377f,   -0.0769677f,  0.0110251f,  -0.46098f,   -0.421538f,
+  -2.18003f,    0.6639f,      0.541278f,   0.704337f,   0.480695f,
+  0.727474f,    0.473937f,    0.770863f,   0.41774f,    -0.112943f,
+  -0.107412f,   0.083639f,    -0.351127f,  -0.149221f,  0.0705735f,
+  -0.0908294f,  -0.0789572f,  -1.62533f,   -2.14051f,   -0.188871f,
+  0.796224f,    0.00144302f,  0.467569f,   -0.113915f,  0.660595f,
+  0.101025f,    0.020895f,    0.604505f,   -0.26598f,   -0.149282f,
+  -0.548818f,   -0.140463f,   0.13669f,    -0.0955724f, -0.694851f,
+  -0.13438f,    -0.384164f,   -0.600513f,  1.48183f,    0.16481f,
+  0.340894f,    -0.190813f,   0.487458f,   0.0769957f,  0.192901f,
+  0.039483f,    0.213366f,    -0.508968f,  -0.494034f,  -0.987906f,
+  -0.188502f,   -0.907188f,   0.0109838f,  -0.275008f,  -0.331017f,
+  -0.447202f,   0.0521897f,   0.837215f,   0.344802f,   0.216526f,
+  0.456105f,    0.139706f,    -0.373955f,  0.0232853f,  -0.365632f,
+  -0.0586727f,  0.828841f,    0.737974f,   -0.933899f,  -0.0620807f,
+  -0.262087f,   -0.0551274f,  -0.374378f,  0.118385f,   -0.0375933f,
+  -0.39549f,    -0.0844191f,  0.274343f,   0.21814f,    0.52751f,
+  0.449546f,    0.075518f,    0.369501f,   0.345248f,   0.403972f,
+  -0.465168f,   -0.551238f,   -0.354305f,  -0.233872f,  0.664137f,
+  -0.214848f,   -0.0363864f,  0.446154f,   -0.500073f,  -1.07883f,
+  -1.07772f
+};
+
+static const float simple_motion_search_prune_part_logits_bias_8[] = {
+  1.09425f, -0.345875f, -0.56098f, -1.84859f
+};
+
+static const float simple_motion_search_prune_part_layer_0_bias_8[] = {
+  -1.75332f, 0.419404f,  -0.0423004f, 0.345601f, -2.40097f,  -0.574857f,
+  -2.53366f, -0.207096f, -1.08866f,   0.801763f, -2.72428f,  -2.39884f,
+  1.04072f,  -0.782749f, -1.47003f,   -2.98207f, 0.8829f,    1.81741f,
+  -2.02369f, 0.241717f,  -0.178749f,  -3.00588f, -0.232007f, 0.21717f
+};
+
+static const NN_CONFIG simple_motion_search_prune_part_nn_config_8 = {
+  NUM_FEATURES_8,
+  NUM_LOGITS_8,
+  NUM_HIDDEN_LAYERS_8,
+  {
+      NUM_LAYER_0_UNITS_8,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_kernel_8,
+      simple_motion_search_prune_part_logits_kernel_8,
+  },
+  {
+      simple_motion_search_prune_part_layer_0_bias_8,
+      simple_motion_search_prune_part_logits_bias_8,
+  },
+};
+
+#undef NUM_HIDDEN_LAYERS_8
+#undef NUM_FEATURES_8
+#undef NUM_LAYER_0_UNITS_8
+#undef NUM_LOGITS_8
+
+#define FEATURE_SIZE 19
+static const float two_pass_split_partition_weights_128[FEATURE_SIZE + 1] = {
+  2.683936f, -0.193620f, -4.106470f, -0.141320f, -0.282289f,
+  0.125296f, -1.134961f, 0.862757f,  -0.418799f, -0.637666f,
+  0.016232f, 0.345013f,  0.018823f,  -0.393394f, -1.130700f,
+  0.695357f, 0.112569f,  -0.341975f, -0.513882f, 5.7488966f,
+};
+
+static const float two_pass_split_partition_weights_64[FEATURE_SIZE + 1] = {
+  2.990993f,  0.423273f,  -0.926544f, 0.454646f,  -0.292698f,
+  -1.311632f, -0.284432f, 0.717141f,  -0.419257f, -0.574760f,
+  -0.674444f, 0.669047f,  -0.374255f, 0.380624f,  -0.804036f,
+  0.264021f,  0.004163f,  1.896802f,  0.924287f,  0.13490619f,
+};
+
+static const float two_pass_split_partition_weights_32[FEATURE_SIZE + 1] = {
+  2.795181f,  -0.136943f, -0.924842f, 0.405330f,  -0.463505f,
+  -0.584076f, -0.831472f, 0.382985f,  -0.597544f, -0.138915f,
+  -1.354350f, 0.466035f,  -0.553961f, 0.213202f,  -1.166429f,
+  0.010776f,  -0.096236f, 2.335084f,  1.699857f,  -0.58178353f,
+};
+
+static const float two_pass_split_partition_weights_16[FEATURE_SIZE + 1] = {
+  1.987888f,  -0.431100f, -1.687703f, 0.262602f,  -0.425298f,
+  -0.463870f, -1.493457f, 0.470917f,  -0.528457f, -0.087700f,
+  -1.815092f, 0.152883f,  -0.337908f, 0.093679f,  -1.548267f,
+  -0.042387f, -0.000861f, 2.556746f,  1.619192f,  0.03643292f,
+};
+
+static const float two_pass_split_partition_weights_8[FEATURE_SIZE + 1] = {
+  2.188344f,  -0.817528f, -2.119219f, 0.000000f,  -0.348167f,
+  -0.658074f, -1.960362f, 0.000000f,  -0.403080f, 0.282699f,
+  -2.061088f, 0.000000f,  -0.431919f, -0.127960f, -1.099550f,
+  0.000000f,  0.121622f,  2.017455f,  2.058228f,  -0.15475988f,
+};
+
+static const float two_pass_none_partition_weights_128[FEATURE_SIZE + 1] = {
+  -1.006689f, 0.777908f,  4.461072f,  -0.395782f, -0.014610f,
+  -0.853863f, 0.729997f,  -0.420477f, 0.282429f,  -1.194595f,
+  3.181220f,  -0.511416f, 0.117084f,  -1.149348f, 1.507990f,
+  -0.477212f, 0.202963f,  -1.469581f, 0.624461f,  -0.89081228f,
+};
+
+static const float two_pass_none_partition_weights_64[FEATURE_SIZE + 1] = {
+  -1.241117f, 0.844878f,  5.638803f,  -0.489780f, -0.108796f,
+  -4.576821f, 1.540624f,  -0.477519f, 0.227791f,  -1.443968f,
+  1.586911f,  -0.505125f, 0.140764f,  -0.464194f, 1.466658f,
+  -0.641166f, 0.195412f,  1.427905f,  2.080007f,  -1.98272777f,
+};
+
+static const float two_pass_none_partition_weights_32[FEATURE_SIZE + 1] = {
+  -2.130825f, 0.476023f,  5.907343f,  -0.516002f, -0.097471f,
+  -2.662754f, 0.614858f,  -0.576728f, 0.085261f,  -0.031901f,
+  0.727842f,  -0.600034f, 0.079326f,  0.324328f,  0.504502f,
+  -0.547105f, -0.037670f, 0.304995f,  0.369018f,  -2.66299987f,
+};
+
+static const float two_pass_none_partition_weights_16[FEATURE_SIZE + 1] = {
+  -1.626410f, 0.872047f,  5.414965f,  -0.554781f, -0.084514f,
+  -3.020550f, 0.467632f,  -0.382280f, 0.199568f,  0.426220f,
+  0.829426f,  -0.467100f, 0.153098f,  0.662994f,  0.327545f,
+  -0.560106f, -0.141610f, 0.403372f,  0.523991f,  -3.02891231f,
+};
+
+static const float two_pass_none_partition_weights_8[FEATURE_SIZE + 1] = {
+  -1.463349f, 0.375376f,  4.751430f, 0.000000f, -0.184451f,
+  -1.655447f, 0.443214f,  0.000000f, 0.127961f, 0.152435f,
+  0.083288f,  0.000000f,  0.143105f, 0.438012f, 0.073238f,
+  0.000000f,  -0.278137f, 0.186134f, 0.073737f, -1.6494962f,
+};
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/pickcdef.c b/third_party/aom/av1/encoder/pickcdef.c
index 6d154a7d22..fb07056b84 100644
--- a/third_party/aom/av1/encoder/pickcdef.c
+++ b/third_party/aom/av1/encoder/pickcdef.c
@@ -281,6 +281,7 @@ uint64_t compute_cdef_dist(uint16_t *dst, int dstride, uint16_t *src,
 
 void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
                      AV1_COMMON *cm, MACROBLOCKD *xd, int fast) {
+  CdefInfo *const cdef_info = &cm->cdef_info;
   int r, c;
   int fbr, fbc;
   uint16_t *src[3];
@@ -475,23 +476,23 @@ void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
       best_tot_mse = tot_mse;
       nb_strength_bits = i;
       for (j = 0; j < 1 << nb_strength_bits; j++) {
-        cm->cdef_strengths[j] = best_lev0[j];
-        cm->cdef_uv_strengths[j] = best_lev1[j];
+        cdef_info->cdef_strengths[j] = best_lev0[j];
+        cdef_info->cdef_uv_strengths[j] = best_lev1[j];
       }
     }
   }
   nb_strengths = 1 << nb_strength_bits;
 
-  cm->cdef_bits = nb_strength_bits;
-  cm->nb_cdef_strengths = nb_strengths;
+  cdef_info->cdef_bits = nb_strength_bits;
+  cdef_info->nb_cdef_strengths = nb_strengths;
   for (i = 0; i < sb_count; i++) {
     int gi;
     int best_gi;
     uint64_t best_mse = (uint64_t)1 << 63;
     best_gi = 0;
-    for (gi = 0; gi < cm->nb_cdef_strengths; gi++) {
-      uint64_t curr = mse[0][i][cm->cdef_strengths[gi]];
-      if (num_planes >= 3) curr += mse[1][i][cm->cdef_uv_strengths[gi]];
+    for (gi = 0; gi < cdef_info->nb_cdef_strengths; gi++) {
+      uint64_t curr = mse[0][i][cdef_info->cdef_strengths[gi]];
+      if (num_planes >= 3) curr += mse[1][i][cdef_info->cdef_uv_strengths[gi]];
       if (curr < best_mse) {
         best_gi = gi;
         best_mse = curr;
@@ -503,18 +504,18 @@ void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
 
   if (fast) {
     for (int j = 0; j < nb_strengths; j++) {
-      cm->cdef_strengths[j] =
-          priconv[cm->cdef_strengths[j] / CDEF_SEC_STRENGTHS] *
+      cdef_info->cdef_strengths[j] =
+          priconv[cm->cdef_info.cdef_strengths[j] / CDEF_SEC_STRENGTHS] *
               CDEF_SEC_STRENGTHS +
-          (cm->cdef_strengths[j] % CDEF_SEC_STRENGTHS);
-      cm->cdef_uv_strengths[j] =
-          priconv[cm->cdef_uv_strengths[j] / CDEF_SEC_STRENGTHS] *
+          (cdef_info->cdef_strengths[j] % CDEF_SEC_STRENGTHS);
+      cdef_info->cdef_uv_strengths[j] =
+          priconv[cdef_info->cdef_uv_strengths[j] / CDEF_SEC_STRENGTHS] *
               CDEF_SEC_STRENGTHS +
-          (cm->cdef_uv_strengths[j] % CDEF_SEC_STRENGTHS);
+          (cdef_info->cdef_uv_strengths[j] % CDEF_SEC_STRENGTHS);
     }
   }
-  cm->cdef_pri_damping = pri_damping;
-  cm->cdef_sec_damping = sec_damping;
+  cdef_info->cdef_pri_damping = pri_damping;
+  cdef_info->cdef_sec_damping = sec_damping;
   aom_free(mse[0]);
   aom_free(mse[1]);
   for (pli = 0; pli < num_planes; pli++) {
diff --git a/third_party/aom/av1/encoder/picklpf.c b/third_party/aom/av1/encoder/picklpf.c
index c5508e25c4..8aeec0c08a 100644
--- a/third_party/aom/av1/encoder/picklpf.c
+++ b/third_party/aom/av1/encoder/picklpf.c
@@ -67,26 +67,27 @@ static int64_t try_filter_frame(const YV12_BUFFER_CONFIG *sd,
     case 2: cm->lf.filter_level_v = filter_level[0]; break;
   }
 
-      // TODO(any): please enable multi-thread and remove the flag when loop
-      // filter mask is compatible with multi-thread.
-#if LOOP_FILTER_BITMASK
-  av1_loop_filter_frame(cm->frame_to_show, cm, &cpi->td.mb.e_mbd, 0, plane,
-                        plane + 1, partial_frame);
-#else
+  // TODO(any): please enable multi-thread and remove the flag when loop
+  // filter mask is compatible with multi-thread.
   if (cpi->num_workers > 1)
-    av1_loop_filter_frame_mt(cm->frame_to_show, cm, &cpi->td.mb.e_mbd, plane,
-                             plane + 1, partial_frame, cpi->workers,
-                             cpi->num_workers, &cpi->lf_row_sync);
-  else
-    av1_loop_filter_frame(cm->frame_to_show, cm, &cpi->td.mb.e_mbd, plane,
-                          plane + 1, partial_frame);
+    av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, &cpi->td.mb.e_mbd, plane,
+                             plane + 1, partial_frame,
+#if LOOP_FILTER_BITMASK
+                             0,
 #endif
+                             cpi->workers, cpi->num_workers, &cpi->lf_row_sync);
+  else
+    av1_loop_filter_frame(&cm->cur_frame->buf, cm, &cpi->td.mb.e_mbd,
+#if LOOP_FILTER_BITMASK
+                          0,
+#endif
+                          plane, plane + 1, partial_frame);
 
-  filt_err = aom_get_sse_plane(sd, cm->frame_to_show, plane,
+  filt_err = aom_get_sse_plane(sd, &cm->cur_frame->buf, plane,
                                cm->seq_params.use_highbitdepth);
 
   // Re-instate the unfiltered frame
-  yv12_copy_plane(&cpi->last_frame_uf, cm->frame_to_show, plane);
+  yv12_copy_plane(&cpi->last_frame_uf, &cm->cur_frame->buf, plane);
 
   return filt_err;
 }
@@ -107,7 +108,17 @@ static int search_filter_level(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi,
   // range.
   int lvl;
   switch (plane) {
-    case 0: lvl = last_frame_filter_level[dir]; break;
+    case 0:
+      switch (dir) {
+        case 2:
+          lvl = (last_frame_filter_level[0] + last_frame_filter_level[1] + 1) >>
+                1;
+          break;
+        case 0:
+        case 1: lvl = last_frame_filter_level[dir]; break;
+        default: assert(dir >= 0 && dir <= 2); return 0;
+      }
+      break;
     case 1: lvl = last_frame_filter_level[2]; break;
     case 2: lvl = last_frame_filter_level[3]; break;
     default: assert(plane >= 0 && plane <= 2); return 0;
@@ -119,7 +130,7 @@ static int search_filter_level(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi,
 
   // Set each entry to -1
   memset(ss_err, 0xFF, sizeof(ss_err));
-  yv12_copy_plane(cm->frame_to_show, &cpi->last_frame_uf, plane);
+  yv12_copy_plane(&cm->cur_frame->buf, &cpi->last_frame_uf, plane);
   best_err = try_filter_frame(sd, cpi, filt_mid, partial_frame, plane, dir);
   filt_best = filt_mid;
   ss_err[filt_mid] = best_err;
@@ -212,7 +223,7 @@ void av1_pick_filter_level(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi,
     int filt_guess;
     switch (cm->seq_params.bit_depth) {
       case AOM_BITS_8:
-        filt_guess = (cm->frame_type == KEY_FRAME)
+        filt_guess = (cm->current_frame.frame_type == KEY_FRAME)
                          ? ROUND_POWER_OF_TWO(q * 17563 - 421574, 18)
                          : ROUND_POWER_OF_TWO(q * 6017 + 650707, 18);
         break;
@@ -228,7 +239,8 @@ void av1_pick_filter_level(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi,
                "or AOM_BITS_12");
         return;
     }
-    if (cm->seq_params.bit_depth != AOM_BITS_8 && cm->frame_type == KEY_FRAME)
+    if (cm->seq_params.bit_depth != AOM_BITS_8 &&
+        cm->current_frame.frame_type == KEY_FRAME)
       filt_guess -= 4;
     // TODO(chengchen): retrain the model for Y, U, V filter levels
     lf->filter_level[0] = clamp(filt_guess, min_filter_level, max_filter_level);
diff --git a/third_party/aom/av1/encoder/pickrst.c b/third_party/aom/av1/encoder/pickrst.c
index e7804f6b44..39db3d2ea8 100644
--- a/third_party/aom/av1/encoder/pickrst.c
+++ b/third_party/aom/av1/encoder/pickrst.c
@@ -43,6 +43,9 @@ static const RestorationType force_restore_type = RESTORE_TYPES;
 // Penalty factor for use of dual sgr
 #define DUAL_SGR_PENALTY_MULT 0.01
 
+// Working precision for Wiener filter coefficients
+#define WIENER_TAP_SCALE_FACTOR ((int64_t)1 << 16)
+
 const int frame_level_restore_bits[RESTORE_TYPES] = { 2, 2, 2, 2 };
 
 typedef int64_t (*sse_extractor_type)(const YV12_BUFFER_CONFIG *a,
@@ -113,15 +116,11 @@ typedef struct {
   AV1PixelRect tile_rect;
 } RestSearchCtxt;
 
-static void rsc_on_tile(int tile_row, int tile_col, void *priv) {
-  (void)tile_col;
-
+static void rsc_on_tile(void *priv) {
   RestSearchCtxt *rsc = (RestSearchCtxt *)priv;
   set_default_sgrproj(&rsc->sgrproj);
   set_default_wiener(&rsc->wiener);
-
-  rsc->tile_stripe0 =
-      (tile_row == 0) ? 0 : rsc->cm->rst_end_stripe[tile_row - 1];
+  rsc->tile_stripe0 = 0;
 }
 
 static void reset_rsc(RestSearchCtxt *rsc) {
@@ -141,7 +140,7 @@ static void init_rsc(const YV12_BUFFER_CONFIG *src, const AV1_COMMON *cm,
   rsc->rusi = rusi;
   rsc->sf = sf;
 
-  const YV12_BUFFER_CONFIG *dgd = cm->frame_to_show;
+  const YV12_BUFFER_CONFIG *dgd = &cm->cur_frame->buf;
   const int is_uv = plane != AOM_PLANE_Y;
   rsc->plane_width = src->crop_widths[is_uv];
   rsc->plane_height = src->crop_heights[is_uv];
@@ -166,7 +165,7 @@ static int64_t try_restoration_unit(const RestSearchCtxt *rsc,
   const int bit_depth = cm->seq_params.bit_depth;
   const int highbd = cm->seq_params.use_highbitdepth;
 
-  const YV12_BUFFER_CONFIG *fts = cm->frame_to_show;
+  const YV12_BUFFER_CONFIG *fts = &cm->cur_frame->buf;
   // TODO(yunqing): For now, only use optimized LR filter in decoder. Can be
   // also used in encoder.
   const int optimized_lr = 0;
@@ -201,7 +200,7 @@ int64_t av1_lowbd_pixel_proj_error_c(const uint8_t *src8, int width, int height,
         v += xq[0] * (flt0[j] - u) + xq[1] * (flt1[j] - u);
         const int32_t e =
             ROUND_POWER_OF_TWO(v, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS) - src[j];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -217,7 +216,7 @@ int64_t av1_lowbd_pixel_proj_error_c(const uint8_t *src8, int width, int height,
         v += xq[0] * (flt0[j] - u);
         const int32_t e =
             ROUND_POWER_OF_TWO(v, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS) - src[j];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -232,7 +231,7 @@ int64_t av1_lowbd_pixel_proj_error_c(const uint8_t *src8, int width, int height,
         v += xq[1] * (flt1[j] - u);
         const int32_t e =
             ROUND_POWER_OF_TWO(v, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS) - src[j];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -242,7 +241,7 @@ int64_t av1_lowbd_pixel_proj_error_c(const uint8_t *src8, int width, int height,
     for (i = 0; i < height; ++i) {
       for (j = 0; j < width; ++j) {
         const int32_t e = (int32_t)(dat[j]) - src[j];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -252,88 +251,97 @@ int64_t av1_lowbd_pixel_proj_error_c(const uint8_t *src8, int width, int height,
   return err;
 }
 
+int64_t av1_highbd_pixel_proj_error_c(const uint8_t *src8, int width,
+                                      int height, int src_stride,
+                                      const uint8_t *dat8, int dat_stride,
+                                      int32_t *flt0, int flt0_stride,
+                                      int32_t *flt1, int flt1_stride, int xq[2],
+                                      const sgr_params_type *params) {
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
+  int i, j;
+  int64_t err = 0;
+  const int32_t half = 1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1);
+  if (params->r[0] > 0 && params->r[1] > 0) {
+    int xq0 = xq[0];
+    int xq1 = xq[1];
+    for (i = 0; i < height; ++i) {
+      for (j = 0; j < width; ++j) {
+        const int32_t d = dat[j];
+        const int32_t s = src[j];
+        const int32_t u = (int32_t)(d << SGRPROJ_RST_BITS);
+        int32_t v0 = flt0[j] - u;
+        int32_t v1 = flt1[j] - u;
+        int32_t v = half;
+        v += xq0 * v0;
+        v += xq1 * v1;
+        const int32_t e = (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + d - s;
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      flt0 += flt0_stride;
+      flt1 += flt1_stride;
+      src += src_stride;
+    }
+  } else if (params->r[0] > 0 || params->r[1] > 0) {
+    int exq;
+    int32_t *flt;
+    int flt_stride;
+    if (params->r[0] > 0) {
+      exq = xq[0];
+      flt = flt0;
+      flt_stride = flt0_stride;
+    } else {
+      exq = xq[1];
+      flt = flt1;
+      flt_stride = flt1_stride;
+    }
+    for (i = 0; i < height; ++i) {
+      for (j = 0; j < width; ++j) {
+        const int32_t d = dat[j];
+        const int32_t s = src[j];
+        const int32_t u = (int32_t)(d << SGRPROJ_RST_BITS);
+        int32_t v = half;
+        v += exq * (flt[j] - u);
+        const int32_t e = (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + d - s;
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      flt += flt_stride;
+      src += src_stride;
+    }
+  } else {
+    for (i = 0; i < height; ++i) {
+      for (j = 0; j < width; ++j) {
+        const int32_t d = dat[j];
+        const int32_t s = src[j];
+        const int32_t e = d - s;
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+    }
+  }
+  return err;
+}
+
 static int64_t get_pixel_proj_error(const uint8_t *src8, int width, int height,
                                     int src_stride, const uint8_t *dat8,
                                     int dat_stride, int use_highbitdepth,
                                     int32_t *flt0, int flt0_stride,
                                     int32_t *flt1, int flt1_stride, int *xqd,
                                     const sgr_params_type *params) {
-  int i, j;
-  int64_t err = 0;
   int xq[2];
   decode_xq(xqd, xq, params);
   if (!use_highbitdepth) {
-    err = av1_lowbd_pixel_proj_error(src8, width, height, src_stride, dat8,
-                                     dat_stride, flt0, flt0_stride, flt1,
-                                     flt1_stride, xq, params);
+    return av1_lowbd_pixel_proj_error(src8, width, height, src_stride, dat8,
+                                      dat_stride, flt0, flt0_stride, flt1,
+                                      flt1_stride, xq, params);
   } else {
-    const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
-    const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
-    const int32_t half = 1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1);
-    if (params->r[0] > 0 && params->r[1] > 0) {
-      int xq0 = xq[0];
-      int xq1 = xq[1];
-      for (i = 0; i < height; ++i) {
-        for (j = 0; j < width; ++j) {
-          const int32_t d = dat[j];
-          const int32_t s = src[j];
-          const int32_t u = (int32_t)(d << SGRPROJ_RST_BITS);
-          int32_t v0 = flt0[j] - u;
-          int32_t v1 = flt1[j] - u;
-          int32_t v = half;
-          v += xq0 * v0;
-          v += xq1 * v1;
-          const int32_t e =
-              (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + d - s;
-          err += e * e;
-        }
-        dat += dat_stride;
-        flt0 += flt0_stride;
-        flt1 += flt1_stride;
-        src += src_stride;
-      }
-    } else if (params->r[0] > 0 || params->r[1] > 0) {
-      int exq;
-      int32_t *flt;
-      int flt_stride;
-      if (params->r[0] > 0) {
-        exq = xq[0];
-        flt = flt0;
-        flt_stride = flt0_stride;
-      } else {
-        exq = xq[1];
-        flt = flt1;
-        flt_stride = flt1_stride;
-      }
-      for (i = 0; i < height; ++i) {
-        for (j = 0; j < width; ++j) {
-          const int32_t d = dat[j];
-          const int32_t s = src[j];
-          const int32_t u = (int32_t)(d << SGRPROJ_RST_BITS);
-          int32_t v = half;
-          v += exq * (flt[j] - u);
-          const int32_t e =
-              (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + d - s;
-          err += e * e;
-        }
-        dat += dat_stride;
-        flt += flt_stride;
-        src += src_stride;
-      }
-    } else {
-      for (i = 0; i < height; ++i) {
-        for (j = 0; j < width; ++j) {
-          const int32_t d = dat[j];
-          const int32_t s = src[j];
-          const int32_t e = d - s;
-          err += e * e;
-        }
-        dat += dat_stride;
-        src += src_stride;
-      }
-    }
+    return av1_highbd_pixel_proj_error(src8, width, height, src_stride, dat8,
+                                       dat_stride, flt0, flt0_stride, flt1,
+                                       flt1_stride, xq, params);
   }
-  return err;
 }
 
 #define USE_SGRPROJ_REFINEMENT_SEARCH 1
@@ -398,6 +406,13 @@ static int64_t finer_search_pixel_proj_error(
   return err;
 }
 
+static int64_t signed_rounded_divide(int64_t dividend, int64_t divisor) {
+  if (dividend < 0)
+    return (dividend - divisor / 2) / divisor;
+  else
+    return (dividend + divisor / 2) / divisor;
+}
+
 static void get_proj_subspace(const uint8_t *src8, int width, int height,
                               int src_stride, const uint8_t *dat8,
                               int dat_stride, int use_highbitdepth,
@@ -405,34 +420,32 @@ static void get_proj_subspace(const uint8_t *src8, int width, int height,
                               int flt1_stride, int *xq,
                               const sgr_params_type *params) {
   int i, j;
-  double H[2][2] = { { 0, 0 }, { 0, 0 } };
-  double C[2] = { 0, 0 };
-  double Det;
-  double x[2];
+  int64_t H[2][2] = { { 0, 0 }, { 0, 0 } };
+  int64_t C[2] = { 0, 0 };
   const int size = width * height;
 
-  aom_clear_system_state();
-
-  // Default
+  // Default values to be returned if the problem becomes ill-posed
   xq[0] = 0;
   xq[1] = 0;
+
   if (!use_highbitdepth) {
     const uint8_t *src = src8;
     const uint8_t *dat = dat8;
     for (i = 0; i < height; ++i) {
       for (j = 0; j < width; ++j) {
-        const double u = (double)(dat[i * dat_stride + j] << SGRPROJ_RST_BITS);
-        const double s =
-            (double)(src[i * src_stride + j] << SGRPROJ_RST_BITS) - u;
-        const double f1 =
-            (params->r[0] > 0) ? (double)flt0[i * flt0_stride + j] - u : 0;
-        const double f2 =
-            (params->r[1] > 0) ? (double)flt1[i * flt1_stride + j] - u : 0;
-        H[0][0] += f1 * f1;
-        H[1][1] += f2 * f2;
-        H[0][1] += f1 * f2;
-        C[0] += f1 * s;
-        C[1] += f2 * s;
+        const int32_t u =
+            (int32_t)(dat[i * dat_stride + j] << SGRPROJ_RST_BITS);
+        const int32_t s =
+            (int32_t)(src[i * src_stride + j] << SGRPROJ_RST_BITS) - u;
+        const int32_t f1 =
+            (params->r[0] > 0) ? (int32_t)flt0[i * flt0_stride + j] - u : 0;
+        const int32_t f2 =
+            (params->r[1] > 0) ? (int32_t)flt1[i * flt1_stride + j] - u : 0;
+        H[0][0] += (int64_t)f1 * f1;
+        H[1][1] += (int64_t)f2 * f2;
+        H[0][1] += (int64_t)f1 * f2;
+        C[0] += (int64_t)f1 * s;
+        C[1] += (int64_t)f2 * s;
       }
     }
   } else {
@@ -440,18 +453,19 @@ static void get_proj_subspace(const uint8_t *src8, int width, int height,
     const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
     for (i = 0; i < height; ++i) {
       for (j = 0; j < width; ++j) {
-        const double u = (double)(dat[i * dat_stride + j] << SGRPROJ_RST_BITS);
-        const double s =
-            (double)(src[i * src_stride + j] << SGRPROJ_RST_BITS) - u;
-        const double f1 =
-            (params->r[0] > 0) ? (double)flt0[i * flt0_stride + j] - u : 0;
-        const double f2 =
-            (params->r[1] > 0) ? (double)flt1[i * flt1_stride + j] - u : 0;
-        H[0][0] += f1 * f1;
-        H[1][1] += f2 * f2;
-        H[0][1] += f1 * f2;
-        C[0] += f1 * s;
-        C[1] += f2 * s;
+        const int32_t u =
+            (int32_t)(dat[i * dat_stride + j] << SGRPROJ_RST_BITS);
+        const int32_t s =
+            (int32_t)(src[i * src_stride + j] << SGRPROJ_RST_BITS) - u;
+        const int32_t f1 =
+            (params->r[0] > 0) ? (int32_t)flt0[i * flt0_stride + j] - u : 0;
+        const int32_t f2 =
+            (params->r[1] > 0) ? (int32_t)flt1[i * flt1_stride + j] - u : 0;
+        H[0][0] += (int64_t)f1 * f1;
+        H[1][1] += (int64_t)f2 * f2;
+        H[0][1] += (int64_t)f1 * f2;
+        C[0] += (int64_t)f1 * s;
+        C[1] += (int64_t)f2 * s;
       }
     }
   }
@@ -464,35 +478,39 @@ static void get_proj_subspace(const uint8_t *src8, int width, int height,
   if (params->r[0] == 0) {
     // H matrix is now only the scalar H[1][1]
     // C vector is now only the scalar C[1]
-    Det = H[1][1];
-    if (Det < 1e-8) return;  // ill-posed, return default values
-    x[0] = 0;
-    x[1] = C[1] / Det;
-
+    const int64_t Det = H[1][1];
+    if (Det == 0) return;  // ill-posed, return default values
     xq[0] = 0;
-    xq[1] = (int)rint(x[1] * (1 << SGRPROJ_PRJ_BITS));
+    xq[1] = (int)signed_rounded_divide(C[1] * (1 << SGRPROJ_PRJ_BITS), Det);
   } else if (params->r[1] == 0) {
     // H matrix is now only the scalar H[0][0]
     // C vector is now only the scalar C[0]
-    Det = H[0][0];
-    if (Det < 1e-8) return;  // ill-posed, return default values
-    x[0] = C[0] / Det;
-    x[1] = 0;
-
-    xq[0] = (int)rint(x[0] * (1 << SGRPROJ_PRJ_BITS));
+    const int64_t Det = H[0][0];
+    if (Det == 0) return;  // ill-posed, return default values
+    xq[0] = (int)signed_rounded_divide(C[0] * (1 << SGRPROJ_PRJ_BITS), Det);
     xq[1] = 0;
   } else {
-    Det = (H[0][0] * H[1][1] - H[0][1] * H[1][0]);
-    if (Det < 1e-8) return;  // ill-posed, return default values
-    x[0] = (H[1][1] * C[0] - H[0][1] * C[1]) / Det;
-    x[1] = (H[0][0] * C[1] - H[1][0] * C[0]) / Det;
+    const int64_t Det = H[0][0] * H[1][1] - H[0][1] * H[1][0];
+    if (Det == 0) return;  // ill-posed, return default values
 
-    xq[0] = (int)rint(x[0] * (1 << SGRPROJ_PRJ_BITS));
-    xq[1] = (int)rint(x[1] * (1 << SGRPROJ_PRJ_BITS));
+    // If scaling up dividend would overflow, instead scale down the divisor
+    const int64_t div1 = H[1][1] * C[0] - H[0][1] * C[1];
+    if ((div1 > 0 && INT64_MAX / (1 << SGRPROJ_PRJ_BITS) < div1) ||
+        (div1 < 0 && INT64_MIN / (1 << SGRPROJ_PRJ_BITS) > div1))
+      xq[0] = (int)signed_rounded_divide(div1, Det / (1 << SGRPROJ_PRJ_BITS));
+    else
+      xq[0] = (int)signed_rounded_divide(div1 * (1 << SGRPROJ_PRJ_BITS), Det);
+
+    const int64_t div2 = H[0][0] * C[1] - H[1][0] * C[0];
+    if ((div2 > 0 && INT64_MAX / (1 << SGRPROJ_PRJ_BITS) < div2) ||
+        (div2 < 0 && INT64_MIN / (1 << SGRPROJ_PRJ_BITS) > div2))
+      xq[1] = (int)signed_rounded_divide(div2, Det / (1 << SGRPROJ_PRJ_BITS));
+    else
+      xq[1] = (int)signed_rounded_divide(div2 * (1 << SGRPROJ_PRJ_BITS), Det);
   }
 }
 
-void encode_xq(int *xq, int *xqd, const sgr_params_type *params) {
+static void encode_xq(int *xq, int *xqd, const sgr_params_type *params) {
   if (params->r[0] == 0) {
     xqd[0] = 0;
     xqd[1] = clamp((1 << SGRPROJ_PRJ_BITS) - xq[1], SGRPROJ_PRJ_MIN1,
@@ -651,34 +669,34 @@ static void search_sgrproj(const RestorationTileLimits *limits,
 
 void av1_compute_stats_c(int wiener_win, const uint8_t *dgd, const uint8_t *src,
                          int h_start, int h_end, int v_start, int v_end,
-                         int dgd_stride, int src_stride, double *M, double *H) {
+                         int dgd_stride, int src_stride, int64_t *M,
+                         int64_t *H) {
   int i, j, k, l;
-  double Y[WIENER_WIN2];
+  int16_t Y[WIENER_WIN2];
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
-      find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+  uint8_t avg = find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   memset(M, 0, sizeof(*M) * wiener_win2);
   memset(H, 0, sizeof(*H) * wiener_win2 * wiener_win2);
   for (i = v_start; i < v_end; i++) {
     for (j = h_start; j < h_end; j++) {
-      const double X = (double)src[i * src_stride + j] - avg;
+      const int16_t X = (int16_t)src[i * src_stride + j] - (int16_t)avg;
       int idx = 0;
       for (k = -wiener_halfwin; k <= wiener_halfwin; k++) {
         for (l = -wiener_halfwin; l <= wiener_halfwin; l++) {
-          Y[idx] = (double)dgd[(i + l) * dgd_stride + (j + k)] - avg;
+          Y[idx] = (int16_t)dgd[(i + l) * dgd_stride + (j + k)] - (int16_t)avg;
           idx++;
         }
       }
       assert(idx == wiener_win2);
       for (k = 0; k < wiener_win2; ++k) {
-        M[k] += Y[k] * X;
+        M[k] += (int32_t)Y[k] * X;
         for (l = k; l < wiener_win2; ++l) {
           // H is a symmetric matrix, so we only need to fill out the upper
           // triangle here. We can copy it down to the lower triangle outside
           // the (i, j) loops.
-          H[k * wiener_win2 + l] += Y[k] * Y[l];
+          H[k * wiener_win2 + l] += (int32_t)Y[k] * Y[l];
         }
       }
     }
@@ -690,60 +708,55 @@ void av1_compute_stats_c(int wiener_win, const uint8_t *dgd, const uint8_t *src,
   }
 }
 
-static double find_average_highbd(const uint16_t *src, int h_start, int h_end,
-                                  int v_start, int v_end, int stride) {
-  uint64_t sum = 0;
-  double avg = 0;
-  int i, j;
-  aom_clear_system_state();
-  for (i = v_start; i < v_end; i++)
-    for (j = h_start; j < h_end; j++) sum += src[i * stride + j];
-  avg = (double)sum / ((v_end - v_start) * (h_end - h_start));
-  return avg;
-}
-
-static AOM_FORCE_INLINE void compute_stats_highbd(
-    int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start,
-    int h_end, int v_start, int v_end, int dgd_stride, int src_stride,
-    double *M, double *H) {
+void av1_compute_stats_highbd_c(int wiener_win, const uint8_t *dgd8,
+                                const uint8_t *src8, int h_start, int h_end,
+                                int v_start, int v_end, int dgd_stride,
+                                int src_stride, int64_t *M, int64_t *H,
+                                aom_bit_depth_t bit_depth) {
   int i, j, k, l;
-  double Y[WIENER_WIN2];
+  int32_t Y[WIENER_WIN2];
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
   const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
   const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
-  const double avg =
+  uint16_t avg =
       find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
   memset(M, 0, sizeof(*M) * wiener_win2);
   memset(H, 0, sizeof(*H) * wiener_win2 * wiener_win2);
   for (i = v_start; i < v_end; i++) {
     for (j = h_start; j < h_end; j++) {
-      const double X = (double)src[i * src_stride + j] - avg;
+      const int32_t X = (int32_t)src[i * src_stride + j] - (int32_t)avg;
       int idx = 0;
       for (k = -wiener_halfwin; k <= wiener_halfwin; k++) {
         for (l = -wiener_halfwin; l <= wiener_halfwin; l++) {
-          Y[idx] = (double)dgd[(i + l) * dgd_stride + (j + k)] - avg;
+          Y[idx] = (int32_t)dgd[(i + l) * dgd_stride + (j + k)] - (int32_t)avg;
           idx++;
         }
       }
       assert(idx == wiener_win2);
       for (k = 0; k < wiener_win2; ++k) {
-        double Yk = Y[k];
-        M[k] += Yk * X;
-        double *H2 = &H[k * wiener_win2];
-        H2[k] += Yk * Yk;
-        for (l = k + 1; l < wiener_win2; ++l) {
+        M[k] += (int64_t)Y[k] * X;
+        for (l = k; l < wiener_win2; ++l) {
           // H is a symmetric matrix, so we only need to fill out the upper
           // triangle here. We can copy it down to the lower triangle outside
           // the (i, j) loops.
-          H2[l] += Yk * Y[l];
+          H[k * wiener_win2 + l] += (int64_t)Y[k] * Y[l];
         }
       }
     }
   }
   for (k = 0; k < wiener_win2; ++k) {
+    M[k] /= bit_depth_divider;
+    H[k * wiener_win2 + k] /= bit_depth_divider;
     for (l = k + 1; l < wiener_win2; ++l) {
+      H[k * wiener_win2 + l] /= bit_depth_divider;
       H[l * wiener_win2 + k] = H[k * wiener_win2 + l];
     }
   }
@@ -754,12 +767,56 @@ static INLINE int wrap_index(int i, int wiener_win) {
   return (i >= wiener_halfwin1 ? wiener_win - 1 - i : i);
 }
 
+// Solve linear equations to find Wiener filter tap values
+// Taps are output scaled by WIENER_FILT_STEP
+static int linsolve_wiener(int n, int64_t *A, int stride, int64_t *b,
+                           int32_t *x) {
+  for (int k = 0; k < n - 1; k++) {
+    // Partial pivoting: bring the row with the largest pivot to the top
+    for (int i = n - 1; i > k; i--) {
+      // If row i has a better (bigger) pivot than row (i-1), swap them
+      if (llabs(A[(i - 1) * stride + k]) < llabs(A[i * stride + k])) {
+        for (int j = 0; j < n; j++) {
+          const int64_t c = A[i * stride + j];
+          A[i * stride + j] = A[(i - 1) * stride + j];
+          A[(i - 1) * stride + j] = c;
+        }
+        const int64_t c = b[i];
+        b[i] = b[i - 1];
+        b[i - 1] = c;
+      }
+    }
+    // Forward elimination (convert A to row-echelon form)
+    for (int i = k; i < n - 1; i++) {
+      if (A[k * stride + k] == 0) return 0;
+      const int64_t c = A[(i + 1) * stride + k];
+      const int64_t cd = A[k * stride + k];
+      for (int j = 0; j < n; j++) {
+        A[(i + 1) * stride + j] -= c / 256 * A[k * stride + j] / cd * 256;
+      }
+      b[i + 1] -= c * b[k] / cd;
+    }
+  }
+  // Back-substitution
+  for (int i = n - 1; i >= 0; i--) {
+    if (A[i * stride + i] == 0) return 0;
+    int64_t c = 0;
+    for (int j = i + 1; j <= n - 1; j++) {
+      c += A[i * stride + j] * x[j] / WIENER_TAP_SCALE_FACTOR;
+    }
+    // Store filter taps x in scaled form.
+    x[i] = (int32_t)(WIENER_TAP_SCALE_FACTOR * (b[i] - c) / A[i * stride + i]);
+  }
+
+  return 1;
+}
+
 // Fix vector b, update vector a
-static void update_a_sep_sym(int wiener_win, double **Mc, double **Hc,
-                             double *a, double *b) {
+static void update_a_sep_sym(int wiener_win, int64_t **Mc, int64_t **Hc,
+                             int32_t *a, int32_t *b) {
   int i, j;
-  double S[WIENER_WIN];
-  double A[WIENER_HALFWIN1], B[WIENER_HALFWIN1 * WIENER_HALFWIN1];
+  int32_t S[WIENER_WIN];
+  int64_t A[WIENER_HALFWIN1], B[WIENER_HALFWIN1 * WIENER_HALFWIN1];
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin1 = (wiener_win >> 1) + 1;
   memset(A, 0, sizeof(A));
@@ -767,36 +824,41 @@ static void update_a_sep_sym(int wiener_win, double **Mc, double **Hc,
   for (i = 0; i < wiener_win; i++) {
     for (j = 0; j < wiener_win; ++j) {
       const int jj = wrap_index(j, wiener_win);
-      A[jj] += Mc[i][j] * b[i];
+      A[jj] += Mc[i][j] * b[i] / WIENER_TAP_SCALE_FACTOR;
     }
   }
   for (i = 0; i < wiener_win; i++) {
     for (j = 0; j < wiener_win; j++) {
       int k, l;
-      for (k = 0; k < wiener_win; ++k)
+      for (k = 0; k < wiener_win; ++k) {
         for (l = 0; l < wiener_win; ++l) {
           const int kk = wrap_index(k, wiener_win);
           const int ll = wrap_index(l, wiener_win);
           B[ll * wiener_halfwin1 + kk] +=
-              Hc[j * wiener_win + i][k * wiener_win2 + l] * b[i] * b[j];
+              Hc[j * wiener_win + i][k * wiener_win2 + l] * b[i] /
+              WIENER_TAP_SCALE_FACTOR * b[j] / WIENER_TAP_SCALE_FACTOR;
         }
+      }
     }
   }
   // Normalization enforcement in the system of equations itself
-  for (i = 0; i < wiener_halfwin1 - 1; ++i)
+  for (i = 0; i < wiener_halfwin1 - 1; ++i) {
     A[i] -=
         A[wiener_halfwin1 - 1] * 2 +
         B[i * wiener_halfwin1 + wiener_halfwin1 - 1] -
         2 * B[(wiener_halfwin1 - 1) * wiener_halfwin1 + (wiener_halfwin1 - 1)];
-  for (i = 0; i < wiener_halfwin1 - 1; ++i)
-    for (j = 0; j < wiener_halfwin1 - 1; ++j)
+  }
+  for (i = 0; i < wiener_halfwin1 - 1; ++i) {
+    for (j = 0; j < wiener_halfwin1 - 1; ++j) {
       B[i * wiener_halfwin1 + j] -=
           2 * (B[i * wiener_halfwin1 + (wiener_halfwin1 - 1)] +
                B[(wiener_halfwin1 - 1) * wiener_halfwin1 + j] -
                2 * B[(wiener_halfwin1 - 1) * wiener_halfwin1 +
                      (wiener_halfwin1 - 1)]);
-  if (linsolve(wiener_halfwin1 - 1, B, wiener_halfwin1, A, S)) {
-    S[wiener_halfwin1 - 1] = 1.0;
+    }
+  }
+  if (linsolve_wiener(wiener_halfwin1 - 1, B, wiener_halfwin1, A, S)) {
+    S[wiener_halfwin1 - 1] = WIENER_TAP_SCALE_FACTOR;
     for (i = wiener_halfwin1; i < wiener_win; ++i) {
       S[i] = S[wiener_win - 1 - i];
       S[wiener_halfwin1 - 1] -= 2 * S[i];
@@ -806,18 +868,20 @@ static void update_a_sep_sym(int wiener_win, double **Mc, double **Hc,
 }
 
 // Fix vector a, update vector b
-static void update_b_sep_sym(int wiener_win, double **Mc, double **Hc,
-                             double *a, double *b) {
+static void update_b_sep_sym(int wiener_win, int64_t **Mc, int64_t **Hc,
+                             int32_t *a, int32_t *b) {
   int i, j;
-  double S[WIENER_WIN];
-  double A[WIENER_HALFWIN1], B[WIENER_HALFWIN1 * WIENER_HALFWIN1];
+  int32_t S[WIENER_WIN];
+  int64_t A[WIENER_HALFWIN1], B[WIENER_HALFWIN1 * WIENER_HALFWIN1];
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin1 = (wiener_win >> 1) + 1;
   memset(A, 0, sizeof(A));
   memset(B, 0, sizeof(B));
   for (i = 0; i < wiener_win; i++) {
     const int ii = wrap_index(i, wiener_win);
-    for (j = 0; j < wiener_win; j++) A[ii] += Mc[i][j] * a[j];
+    for (j = 0; j < wiener_win; j++) {
+      A[ii] += Mc[i][j] * a[j] / WIENER_TAP_SCALE_FACTOR;
+    }
   }
 
   for (i = 0; i < wiener_win; i++) {
@@ -825,27 +889,33 @@ static void update_b_sep_sym(int wiener_win, double **Mc, double **Hc,
       const int ii = wrap_index(i, wiener_win);
       const int jj = wrap_index(j, wiener_win);
       int k, l;
-      for (k = 0; k < wiener_win; ++k)
-        for (l = 0; l < wiener_win; ++l)
+      for (k = 0; k < wiener_win; ++k) {
+        for (l = 0; l < wiener_win; ++l) {
           B[jj * wiener_halfwin1 + ii] +=
-              Hc[i * wiener_win + j][k * wiener_win2 + l] * a[k] * a[l];
+              Hc[i * wiener_win + j][k * wiener_win2 + l] * a[k] /
+              WIENER_TAP_SCALE_FACTOR * a[l] / WIENER_TAP_SCALE_FACTOR;
+        }
+      }
     }
   }
   // Normalization enforcement in the system of equations itself
-  for (i = 0; i < wiener_halfwin1 - 1; ++i)
+  for (i = 0; i < wiener_halfwin1 - 1; ++i) {
     A[i] -=
         A[wiener_halfwin1 - 1] * 2 +
         B[i * wiener_halfwin1 + wiener_halfwin1 - 1] -
         2 * B[(wiener_halfwin1 - 1) * wiener_halfwin1 + (wiener_halfwin1 - 1)];
-  for (i = 0; i < wiener_halfwin1 - 1; ++i)
-    for (j = 0; j < wiener_halfwin1 - 1; ++j)
+  }
+  for (i = 0; i < wiener_halfwin1 - 1; ++i) {
+    for (j = 0; j < wiener_halfwin1 - 1; ++j) {
       B[i * wiener_halfwin1 + j] -=
           2 * (B[i * wiener_halfwin1 + (wiener_halfwin1 - 1)] +
                B[(wiener_halfwin1 - 1) * wiener_halfwin1 + j] -
                2 * B[(wiener_halfwin1 - 1) * wiener_halfwin1 +
                      (wiener_halfwin1 - 1)]);
-  if (linsolve(wiener_halfwin1 - 1, B, wiener_halfwin1, A, S)) {
-    S[wiener_halfwin1 - 1] = 1.0;
+    }
+  }
+  if (linsolve_wiener(wiener_halfwin1 - 1, B, wiener_halfwin1, A, S)) {
+    S[wiener_halfwin1 - 1] = WIENER_TAP_SCALE_FACTOR;
     for (i = wiener_halfwin1; i < wiener_win; ++i) {
       S[i] = S[wiener_win - 1 - i];
       S[wiener_halfwin1 - 1] -= 2 * S[i];
@@ -854,20 +924,21 @@ static void update_b_sep_sym(int wiener_win, double **Mc, double **Hc,
   }
 }
 
-static int wiener_decompose_sep_sym(int wiener_win, double *M, double *H,
-                                    double *a, double *b) {
-  static const int init_filt[WIENER_WIN] = {
+static int wiener_decompose_sep_sym(int wiener_win, int64_t *M, int64_t *H,
+                                    int32_t *a, int32_t *b) {
+  static const int32_t init_filt[WIENER_WIN] = {
     WIENER_FILT_TAP0_MIDV, WIENER_FILT_TAP1_MIDV, WIENER_FILT_TAP2_MIDV,
     WIENER_FILT_TAP3_MIDV, WIENER_FILT_TAP2_MIDV, WIENER_FILT_TAP1_MIDV,
     WIENER_FILT_TAP0_MIDV,
   };
-  double *Hc[WIENER_WIN2];
-  double *Mc[WIENER_WIN];
+  int64_t *Hc[WIENER_WIN2];
+  int64_t *Mc[WIENER_WIN];
   int i, j, iter;
   const int plane_off = (WIENER_WIN - wiener_win) >> 1;
   const int wiener_win2 = wiener_win * wiener_win;
   for (i = 0; i < wiener_win; i++) {
-    a[i] = b[i] = (double)init_filt[i + plane_off] / WIENER_FILT_STEP;
+    a[i] = b[i] =
+        WIENER_TAP_SCALE_FACTOR / WIENER_FILT_STEP * init_filt[i + plane_off];
   }
   for (i = 0; i < wiener_win; i++) {
     Mc[i] = M + i * wiener_win;
@@ -889,23 +960,23 @@ static int wiener_decompose_sep_sym(int wiener_win, double *M, double *H,
 // Computes the function x'*H*x - x'*M for the learned 2D filter x, and compares
 // against identity filters; Final score is defined as the difference between
 // the function values
-static double compute_score(int wiener_win, double *M, double *H,
-                            InterpKernel vfilt, InterpKernel hfilt) {
-  double ab[WIENER_WIN * WIENER_WIN];
+static int64_t compute_score(int wiener_win, int64_t *M, int64_t *H,
+                             InterpKernel vfilt, InterpKernel hfilt) {
+  int32_t ab[WIENER_WIN * WIENER_WIN];
+  int16_t a[WIENER_WIN], b[WIENER_WIN];
+  int64_t P = 0, Q = 0;
+  int64_t iP = 0, iQ = 0;
+  int64_t Score, iScore;
   int i, k, l;
-  double P = 0, Q = 0;
-  double iP = 0, iQ = 0;
-  double Score, iScore;
-  double a[WIENER_WIN], b[WIENER_WIN];
   const int plane_off = (WIENER_WIN - wiener_win) >> 1;
   const int wiener_win2 = wiener_win * wiener_win;
 
   aom_clear_system_state();
 
-  a[WIENER_HALFWIN] = b[WIENER_HALFWIN] = 1.0;
+  a[WIENER_HALFWIN] = b[WIENER_HALFWIN] = WIENER_FILT_STEP;
   for (i = 0; i < WIENER_HALFWIN; ++i) {
-    a[i] = a[WIENER_WIN - i - 1] = (double)vfilt[i] / WIENER_FILT_STEP;
-    b[i] = b[WIENER_WIN - i - 1] = (double)hfilt[i] / WIENER_FILT_STEP;
+    a[i] = a[WIENER_WIN - i - 1] = vfilt[i];
+    b[i] = b[WIENER_WIN - i - 1] = hfilt[i];
     a[WIENER_HALFWIN] -= 2 * a[i];
     b[WIENER_HALFWIN] -= 2 * b[i];
   }
@@ -915,9 +986,11 @@ static double compute_score(int wiener_win, double *M, double *H,
       ab[k * wiener_win + l] = a[l + plane_off] * b[k + plane_off];
   }
   for (k = 0; k < wiener_win2; ++k) {
-    P += ab[k] * M[k];
-    for (l = 0; l < wiener_win2; ++l)
-      Q += ab[k] * H[k * wiener_win2 + l] * ab[l];
+    P += ab[k] * M[k] / WIENER_FILT_STEP / WIENER_FILT_STEP;
+    for (l = 0; l < wiener_win2; ++l) {
+      Q += ab[k] * H[k * wiener_win2 + l] * ab[l] / WIENER_FILT_STEP /
+           WIENER_FILT_STEP / WIENER_FILT_STEP / WIENER_FILT_STEP;
+    }
   }
   Score = Q - 2 * P;
 
@@ -928,11 +1001,19 @@ static double compute_score(int wiener_win, double *M, double *H,
   return Score - iScore;
 }
 
-static void quantize_sym_filter(int wiener_win, double *f, InterpKernel fi) {
+static void finalize_sym_filter(int wiener_win, int32_t *f, InterpKernel fi) {
   int i;
   const int wiener_halfwin = (wiener_win >> 1);
+
   for (i = 0; i < wiener_halfwin; ++i) {
-    fi[i] = RINT(f[i] * WIENER_FILT_STEP);
+    const int64_t dividend = f[i] * WIENER_FILT_STEP;
+    const int64_t divisor = WIENER_TAP_SCALE_FACTOR;
+    // Perform this division with proper rounding rather than truncation
+    if (dividend < 0) {
+      fi[i] = (int16_t)((dividend - (divisor / 2)) / divisor);
+    } else {
+      fi[i] = (int16_t)((dividend + (divisor / 2)) / divisor);
+    }
   }
   // Specialize for 7-tap filter
   if (wiener_win == WIENER_WIN) {
@@ -1110,15 +1191,16 @@ static void search_wiener(const RestorationTileLimits *limits,
   const int wiener_win =
       (rsc->plane == AOM_PLANE_Y) ? WIENER_WIN : WIENER_WIN_CHROMA;
 
-  double M[WIENER_WIN2];
-  double H[WIENER_WIN2 * WIENER_WIN2];
-  double vfilterd[WIENER_WIN], hfilterd[WIENER_WIN];
+  int64_t M[WIENER_WIN2];
+  int64_t H[WIENER_WIN2 * WIENER_WIN2];
+  int32_t vfilter[WIENER_WIN], hfilter[WIENER_WIN];
 
   const AV1_COMMON *const cm = rsc->cm;
   if (cm->seq_params.use_highbitdepth) {
-    compute_stats_highbd(wiener_win, rsc->dgd_buffer, rsc->src_buffer,
-                         limits->h_start, limits->h_end, limits->v_start,
-                         limits->v_end, rsc->dgd_stride, rsc->src_stride, M, H);
+    av1_compute_stats_highbd(wiener_win, rsc->dgd_buffer, rsc->src_buffer,
+                             limits->h_start, limits->h_end, limits->v_start,
+                             limits->v_end, rsc->dgd_stride, rsc->src_stride, M,
+                             H, cm->seq_params.bit_depth);
   } else {
     av1_compute_stats(wiener_win, rsc->dgd_buffer, rsc->src_buffer,
                       limits->h_start, limits->h_end, limits->v_start,
@@ -1128,7 +1210,7 @@ static void search_wiener(const RestorationTileLimits *limits,
   const MACROBLOCK *const x = rsc->x;
   const int64_t bits_none = x->wiener_restore_cost[0];
 
-  if (!wiener_decompose_sep_sym(wiener_win, M, H, vfilterd, hfilterd)) {
+  if (!wiener_decompose_sep_sym(wiener_win, M, H, vfilter, hfilter)) {
     rsc->bits += bits_none;
     rsc->sse += rusi->sse[RESTORE_NONE];
     rusi->best_rtype[RESTORE_WIENER - 1] = RESTORE_NONE;
@@ -1139,8 +1221,8 @@ static void search_wiener(const RestorationTileLimits *limits,
   RestorationUnitInfo rui;
   memset(&rui, 0, sizeof(rui));
   rui.restoration_type = RESTORE_WIENER;
-  quantize_sym_filter(wiener_win, vfilterd, rui.wiener_info.vfilter);
-  quantize_sym_filter(wiener_win, hfilterd, rui.wiener_info.hfilter);
+  finalize_sym_filter(wiener_win, vfilter, rui.wiener_info.vfilter);
+  finalize_sym_filter(wiener_win, hfilter, rui.wiener_info.hfilter);
 
   // Filter score computes the value of the function x'*A*x - x'*b for the
   // learned filter and compares it against identity filer. If there is no
@@ -1199,7 +1281,7 @@ static void search_norestore(const RestorationTileLimits *limits,
 
   const int highbd = rsc->cm->seq_params.use_highbitdepth;
   rusi->sse[RESTORE_NONE] = sse_restoration_unit(
-      limits, rsc->src, rsc->cm->frame_to_show, rsc->plane, highbd);
+      limits, rsc->src, &rsc->cm->cur_frame->buf, rsc->plane, highbd);
 
   rsc->sse += rusi->sse[RESTORE_NONE];
 }
@@ -1283,7 +1365,8 @@ static double search_rest_type(RestSearchCtxt *rsc, RestorationType rtype) {
   };
 
   reset_rsc(rsc);
-  rsc_on_tile(LR_TILE_ROW, LR_TILE_COL, rsc);
+  rsc_on_tile(rsc);
+
   av1_foreach_rest_unit_in_plane(rsc->cm, rsc->plane, funs[rtype], rsc,
                                  &rsc->tile_rect, rsc->cm->rst_tmpbuf, NULL);
   return RDCOST_DBL(rsc->x->rdmult, rsc->bits >> 4, rsc->sse);
diff --git a/third_party/aom/av1/encoder/pickrst.h b/third_party/aom/av1/encoder/pickrst.h
index 3fec0c34b3..f34359c1fc 100644
--- a/third_party/aom/av1/encoder/pickrst.h
+++ b/third_party/aom/av1/encoder/pickrst.h
@@ -25,16 +25,34 @@ static const uint8_t g_shuffle_stats_data[16] = {
   0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
 };
 
-static INLINE double find_average(const uint8_t *src, int h_start, int h_end,
-                                  int v_start, int v_end, int stride) {
+static const uint8_t g_shuffle_stats_highbd_data[32] = {
+  0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9,
+  0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9,
+};
+
+static INLINE uint8_t find_average(const uint8_t *src, int h_start, int h_end,
+                                   int v_start, int v_end, int stride) {
   uint64_t sum = 0;
-  double avg = 0;
-  int i, j;
-  aom_clear_system_state();
-  for (i = v_start; i < v_end; i++)
-    for (j = h_start; j < h_end; j++) sum += src[i * stride + j];
-  avg = (double)sum / ((v_end - v_start) * (h_end - h_start));
-  return avg;
+  for (int i = v_start; i < v_end; i++) {
+    for (int j = h_start; j < h_end; j++) {
+      sum += src[i * stride + j];
+    }
+  }
+  uint64_t avg = sum / ((v_end - v_start) * (h_end - h_start));
+  return (uint8_t)avg;
+}
+
+static INLINE uint16_t find_average_highbd(const uint16_t *src, int h_start,
+                                           int h_end, int v_start, int v_end,
+                                           int stride) {
+  uint64_t sum = 0;
+  for (int i = v_start; i < v_end; i++) {
+    for (int j = h_start; j < h_end; j++) {
+      sum += src[i * stride + j];
+    }
+  }
+  uint64_t avg = sum / ((v_end - v_start) * (h_end - h_start));
+  return (uint16_t)avg;
 }
 
 void av1_pick_filter_restoration(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi);
diff --git a/third_party/aom/av1/encoder/ransac.c b/third_party/aom/av1/encoder/ransac.c
index 781f528ebf..6a8854c58c 100644
--- a/third_party/aom/av1/encoder/ransac.c
+++ b/third_party/aom/av1/encoder/ransac.c
@@ -23,7 +23,7 @@
 #define MAX_DEGENERATE_ITER 10
 #define MINPTS_MULTIPLIER 5
 
-#define INLIER_THRESHOLD 1.0
+#define INLIER_THRESHOLD 1.25
 #define MIN_TRIALS 20
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -376,10 +376,7 @@ static int ransac(const int *matched_points, int npoints,
                   IsDegenerateFunc is_degenerate,
                   FindTransformationFunc find_transformation,
                   ProjectPointsDoubleFunc projectpoints) {
-  static const double PROBABILITY_REQUIRED = 0.9;
-  static const double EPS = 1e-12;
-
-  int N = 10000, trial_count = 0;
+  int trial_count = 0;
   int i = 0;
   int ret_val = 0;
 
@@ -444,7 +441,7 @@ static int ransac(const int *matched_points, int npoints,
     *(cnp2++) = *(matched_points++);
   }
 
-  while (N > trial_count) {
+  while (MIN_TRIALS > trial_count) {
     double sum_distance = 0.0;
     double sum_distance_squared = 0.0;
 
@@ -491,8 +488,7 @@ static int ransac(const int *matched_points, int npoints,
 
     if (current_motion.num_inliers >= worst_kept_motion->num_inliers &&
         current_motion.num_inliers > 1) {
-      int temp;
-      double fracinliers, pNoOutliers, mean_distance, dtemp;
+      double mean_distance;
       mean_distance = sum_distance / ((double)current_motion.num_inliers);
       current_motion.variance =
           sum_distance_squared / ((double)current_motion.num_inliers - 1.0) -
@@ -506,21 +502,183 @@ static int ransac(const int *matched_points, int npoints,
         worst_kept_motion->variance = current_motion.variance;
         memcpy(worst_kept_motion->inlier_indices, current_motion.inlier_indices,
                sizeof(*current_motion.inlier_indices) * npoints);
-
         assert(npoints > 0);
-        fracinliers = (double)current_motion.num_inliers / (double)npoints;
-        pNoOutliers = 1 - pow(fracinliers, minpts);
-        pNoOutliers = fmax(EPS, pNoOutliers);
-        pNoOutliers = fmin(1 - EPS, pNoOutliers);
-        dtemp = log(1.0 - PROBABILITY_REQUIRED) / log(pNoOutliers);
-        temp = (dtemp > (double)INT32_MAX)
-                   ? INT32_MAX
-                   : dtemp < (double)INT32_MIN ? INT32_MIN : (int)dtemp;
-
-        if (temp > 0 && temp < N) {
-          N = AOMMAX(temp, MIN_TRIALS);
+        // Determine the new worst kept motion and its num_inliers and variance.
+        for (i = 0; i < num_desired_motions; ++i) {
+          if (is_better_motion(worst_kept_motion, &motions[i])) {
+            worst_kept_motion = &motions[i];
+          }
         }
+      }
+    }
+    trial_count++;
+  }
 
+  // Sort the motions, best first.
+  qsort(motions, num_desired_motions, sizeof(RANSAC_MOTION), compare_motions);
+
+  // Recompute the motions using only the inliers.
+  for (i = 0; i < num_desired_motions; ++i) {
+    if (motions[i].num_inliers >= minpts) {
+      copy_points_at_indices(points1, corners1, motions[i].inlier_indices,
+                             motions[i].num_inliers);
+      copy_points_at_indices(points2, corners2, motions[i].inlier_indices,
+                             motions[i].num_inliers);
+
+      find_transformation(motions[i].num_inliers, points1, points2,
+                          params_by_motion + (MAX_PARAMDIM - 1) * i);
+    }
+    num_inliers_by_motion[i] = motions[i].num_inliers;
+  }
+
+finish_ransac:
+  aom_free(points1);
+  aom_free(points2);
+  aom_free(corners1);
+  aom_free(corners2);
+  aom_free(image1_coord);
+  aom_free(current_motion.inlier_indices);
+  for (i = 0; i < num_desired_motions; ++i) {
+    aom_free(motions[i].inlier_indices);
+  }
+  aom_free(motions);
+
+  return ret_val;
+}
+
+static int ransac_double_prec(const double *matched_points, int npoints,
+                              int *num_inliers_by_motion,
+                              double *params_by_motion, int num_desired_motions,
+                              const int minpts, IsDegenerateFunc is_degenerate,
+                              FindTransformationFunc find_transformation,
+                              ProjectPointsDoubleFunc projectpoints) {
+  int trial_count = 0;
+  int i = 0;
+  int ret_val = 0;
+
+  unsigned int seed = (unsigned int)npoints;
+
+  int indices[MAX_MINPTS] = { 0 };
+
+  double *points1, *points2;
+  double *corners1, *corners2;
+  double *image1_coord;
+
+  // Store information for the num_desired_motions best transformations found
+  // and the worst motion among them, as well as the motion currently under
+  // consideration.
+  RANSAC_MOTION *motions, *worst_kept_motion = NULL;
+  RANSAC_MOTION current_motion;
+
+  // Store the parameters and the indices of the inlier points for the motion
+  // currently under consideration.
+  double params_this_motion[MAX_PARAMDIM];
+
+  double *cnp1, *cnp2;
+
+  for (i = 0; i < num_desired_motions; ++i) {
+    num_inliers_by_motion[i] = 0;
+  }
+  if (npoints < minpts * MINPTS_MULTIPLIER || npoints == 0) {
+    return 1;
+  }
+
+  points1 = (double *)aom_malloc(sizeof(*points1) * npoints * 2);
+  points2 = (double *)aom_malloc(sizeof(*points2) * npoints * 2);
+  corners1 = (double *)aom_malloc(sizeof(*corners1) * npoints * 2);
+  corners2 = (double *)aom_malloc(sizeof(*corners2) * npoints * 2);
+  image1_coord = (double *)aom_malloc(sizeof(*image1_coord) * npoints * 2);
+
+  motions =
+      (RANSAC_MOTION *)aom_malloc(sizeof(RANSAC_MOTION) * num_desired_motions);
+  for (i = 0; i < num_desired_motions; ++i) {
+    motions[i].inlier_indices =
+        (int *)aom_malloc(sizeof(*motions->inlier_indices) * npoints);
+    clear_motion(motions + i, npoints);
+  }
+  current_motion.inlier_indices =
+      (int *)aom_malloc(sizeof(*current_motion.inlier_indices) * npoints);
+  clear_motion(&current_motion, npoints);
+
+  worst_kept_motion = motions;
+
+  if (!(points1 && points2 && corners1 && corners2 && image1_coord && motions &&
+        current_motion.inlier_indices)) {
+    ret_val = 1;
+    goto finish_ransac;
+  }
+
+  cnp1 = corners1;
+  cnp2 = corners2;
+  for (i = 0; i < npoints; ++i) {
+    *(cnp1++) = *(matched_points++);
+    *(cnp1++) = *(matched_points++);
+    *(cnp2++) = *(matched_points++);
+    *(cnp2++) = *(matched_points++);
+  }
+
+  while (MIN_TRIALS > trial_count) {
+    double sum_distance = 0.0;
+    double sum_distance_squared = 0.0;
+
+    clear_motion(&current_motion, npoints);
+
+    int degenerate = 1;
+    int num_degenerate_iter = 0;
+
+    while (degenerate) {
+      num_degenerate_iter++;
+      if (!get_rand_indices(npoints, minpts, indices, &seed)) {
+        ret_val = 1;
+        goto finish_ransac;
+      }
+
+      copy_points_at_indices(points1, corners1, indices, minpts);
+      copy_points_at_indices(points2, corners2, indices, minpts);
+
+      degenerate = is_degenerate(points1);
+      if (num_degenerate_iter > MAX_DEGENERATE_ITER) {
+        ret_val = 1;
+        goto finish_ransac;
+      }
+    }
+
+    if (find_transformation(minpts, points1, points2, params_this_motion)) {
+      trial_count++;
+      continue;
+    }
+
+    projectpoints(params_this_motion, corners1, image1_coord, npoints, 2, 2);
+
+    for (i = 0; i < npoints; ++i) {
+      double dx = image1_coord[i * 2] - corners2[i * 2];
+      double dy = image1_coord[i * 2 + 1] - corners2[i * 2 + 1];
+      double distance = sqrt(dx * dx + dy * dy);
+
+      if (distance < INLIER_THRESHOLD) {
+        current_motion.inlier_indices[current_motion.num_inliers++] = i;
+        sum_distance += distance;
+        sum_distance_squared += distance * distance;
+      }
+    }
+
+    if (current_motion.num_inliers >= worst_kept_motion->num_inliers &&
+        current_motion.num_inliers > 1) {
+      double mean_distance;
+      mean_distance = sum_distance / ((double)current_motion.num_inliers);
+      current_motion.variance =
+          sum_distance_squared / ((double)current_motion.num_inliers - 1.0) -
+          mean_distance * mean_distance * ((double)current_motion.num_inliers) /
+              ((double)current_motion.num_inliers - 1.0);
+      if (is_better_motion(&current_motion, worst_kept_motion)) {
+        // This motion is better than the worst currently kept motion. Remember
+        // the inlier points and variance. The parameters for each kept motion
+        // will be recomputed later using only the inliers.
+        worst_kept_motion->num_inliers = current_motion.num_inliers;
+        worst_kept_motion->variance = current_motion.variance;
+        memcpy(worst_kept_motion->inlier_indices, current_motion.inlier_indices,
+               sizeof(*current_motion.inlier_indices) * npoints);
+        assert(npoints > 0);
         // Determine the new worst kept motion and its num_inliers and variance.
         for (i = 0; i < num_desired_motions; ++i) {
           if (is_better_motion(worst_kept_motion, &motions[i])) {
@@ -601,3 +759,33 @@ int ransac_affine(int *matched_points, int npoints, int *num_inliers_by_motion,
                 params_by_motion, num_desired_motions, 3, is_degenerate_affine,
                 find_affine, project_points_double_affine);
 }
+
+int ransac_translation_double_prec(double *matched_points, int npoints,
+                                   int *num_inliers_by_motion,
+                                   double *params_by_motion,
+                                   int num_desired_motions) {
+  return ransac_double_prec(matched_points, npoints, num_inliers_by_motion,
+                            params_by_motion, num_desired_motions, 3,
+                            is_degenerate_translation, find_translation,
+                            project_points_double_translation);
+}
+
+int ransac_rotzoom_double_prec(double *matched_points, int npoints,
+                               int *num_inliers_by_motion,
+                               double *params_by_motion,
+                               int num_desired_motions) {
+  return ransac_double_prec(matched_points, npoints, num_inliers_by_motion,
+                            params_by_motion, num_desired_motions, 3,
+                            is_degenerate_affine, find_rotzoom,
+                            project_points_double_rotzoom);
+}
+
+int ransac_affine_double_prec(double *matched_points, int npoints,
+                              int *num_inliers_by_motion,
+                              double *params_by_motion,
+                              int num_desired_motions) {
+  return ransac_double_prec(matched_points, npoints, num_inliers_by_motion,
+                            params_by_motion, num_desired_motions, 3,
+                            is_degenerate_affine, find_affine,
+                            project_points_double_affine);
+}
diff --git a/third_party/aom/av1/encoder/ransac.h b/third_party/aom/av1/encoder/ransac.h
index c429f2ce5e..b754bacdf9 100644
--- a/third_party/aom/av1/encoder/ransac.h
+++ b/third_party/aom/av1/encoder/ransac.h
@@ -22,6 +22,9 @@
 typedef int (*RansacFunc)(int *matched_points, int npoints,
                           int *num_inliers_by_motion, double *params_by_motion,
                           int num_motions);
+typedef int (*RansacFuncDouble)(double *matched_points, int npoints,
+                                int *num_inliers_by_motion,
+                                double *params_by_motion, int num_motions);
 
 /* Each of these functions fits a motion model from a set of
    corresponding points in 2 frames using RANSAC. */
@@ -32,4 +35,16 @@ int ransac_rotzoom(int *matched_points, int npoints, int *num_inliers_by_motion,
 int ransac_translation(int *matched_points, int npoints,
                        int *num_inliers_by_motion, double *params_by_motion,
                        int num_motions);
+int ransac_translation_double_prec(double *matched_points, int npoints,
+                                   int *num_inliers_by_motion,
+                                   double *params_by_motion,
+                                   int num_desired_motions);
+int ransac_rotzoom_double_prec(double *matched_points, int npoints,
+                               int *num_inliers_by_motion,
+                               double *params_by_motion,
+                               int num_desired_motions);
+int ransac_affine_double_prec(double *matched_points, int npoints,
+                              int *num_inliers_by_motion,
+                              double *params_by_motion,
+                              int num_desired_motions);
 #endif  // AOM_AV1_ENCODER_RANSAC_H_
diff --git a/third_party/aom/av1/encoder/ratectrl.c b/third_party/aom/av1/encoder/ratectrl.c
index 2597fb9908..6a7637154d 100644
--- a/third_party/aom/av1/encoder/ratectrl.c
+++ b/third_party/aom/av1/encoder/ratectrl.c
@@ -29,6 +29,7 @@
 #include "av1/common/seg_common.h"
 
 #include "av1/encoder/encodemv.h"
+#include "av1/encoder/encode_strategy.h"
 #include "av1/encoder/random.h"
 #include "av1/encoder/ratectrl.h"
 
@@ -250,12 +251,10 @@ int av1_rc_get_default_min_gf_interval(int width, int height,
   // 4K60: 12
 }
 
-int av1_rc_get_default_max_gf_interval(double framerate, int min_gf_interval) {
+int av1_rc_get_default_max_gf_interval(double framerate, int min_gf_interval,
+                                       int max_pyr_height) {
   int interval = AOMMIN(MAX_GF_INTERVAL, (int)(framerate * 0.75));
-  interval += (interval & 0x01);  // Round to even value
-#if CONFIG_FIX_GF_LENGTH
-  interval = AOMMAX(FIXED_GF_LENGTH, interval);
-#endif
+  interval = AOMMAX(av1_rc_get_fixed_gf_length(max_pyr_height), interval);
   return AOMMAX(interval, min_gf_interval);
 }
 
@@ -312,7 +311,7 @@ void av1_rc_init(const AV1EncoderConfig *oxcf, int pass, RATE_CONTROL *rc) {
         oxcf->width, oxcf->height, oxcf->init_framerate);
   if (rc->max_gf_interval == 0)
     rc->max_gf_interval = av1_rc_get_default_max_gf_interval(
-        oxcf->init_framerate, rc->min_gf_interval);
+        oxcf->init_framerate, rc->min_gf_interval, oxcf->gf_max_pyr_height);
   rc->baseline_gf_interval = (rc->min_gf_interval + rc->max_gf_interval) / 2;
 }
 
@@ -357,7 +356,7 @@ static double get_rate_correction_factor(const AV1_COMP *cpi, int width,
   const RATE_CONTROL *const rc = &cpi->rc;
   double rcf;
 
-  if (cpi->common.frame_type == KEY_FRAME) {
+  if (cpi->common.current_frame.frame_type == KEY_FRAME) {
     rcf = rc->rate_correction_factors[KF_STD];
   } else if (cpi->oxcf.pass == 2) {
     RATE_FACTOR_LEVEL rf_lvl =
@@ -384,7 +383,7 @@ static void set_rate_correction_factor(AV1_COMP *cpi, double factor, int width,
 
   factor = fclamp(factor, MIN_BPB_FACTOR, MAX_BPB_FACTOR);
 
-  if (cpi->common.frame_type == KEY_FRAME) {
+  if (cpi->common.current_frame.frame_type == KEY_FRAME) {
     rc->rate_correction_factors[KF_STD] = factor;
   } else if (cpi->oxcf.pass == 2) {
     RATE_FACTOR_LEVEL rf_lvl =
@@ -425,8 +424,8 @@ void av1_rc_update_rate_correction_factors(AV1_COMP *cpi, int width,
         av1_cyclic_refresh_estimate_bits_at_q(cpi, rate_correction_factor);
   } else {
     projected_size_based_on_q = av1_estimate_bits_at_q(
-        cpi->common.frame_type, cm->base_qindex, MBs, rate_correction_factor,
-        cm->seq_params.bit_depth);
+        cpi->common.current_frame.frame_type, cm->base_qindex, MBs,
+        rate_correction_factor, cm->seq_params.bit_depth);
   }
   // Work out a size correction factor.
   if (projected_size_based_on_q > FRAME_OVERHEAD_BITS)
@@ -497,8 +496,9 @@ int av1_rc_regulate_q(const AV1_COMP *cpi, int target_bits_per_frame,
       bits_per_mb_at_this_q =
           (int)av1_cyclic_refresh_rc_bits_per_mb(cpi, i, correction_factor);
     } else {
-      bits_per_mb_at_this_q = (int)av1_rc_bits_per_mb(
-          cm->frame_type, i, correction_factor, cm->seq_params.bit_depth);
+      bits_per_mb_at_this_q =
+          (int)av1_rc_bits_per_mb(cm->current_frame.frame_type, i,
+                                  correction_factor, cm->seq_params.bit_depth);
     }
 
     if (bits_per_mb_at_this_q <= target_bits_per_mb) {
@@ -559,20 +559,18 @@ static int get_gf_active_quality(const RATE_CONTROL *const rc, int q,
                             arfgf_low_motion_minq, arfgf_high_motion_minq);
 }
 
-#if REDUCE_LAST_ALT_BOOST
 static int get_gf_high_motion_quality(int q, aom_bit_depth_t bit_depth) {
   int *arfgf_high_motion_minq;
   ASSIGN_MINQ_TABLE(bit_depth, arfgf_high_motion_minq);
   return arfgf_high_motion_minq[q];
 }
-#endif
 
 static int calc_active_worst_quality_one_pass_vbr(const AV1_COMP *cpi) {
   const RATE_CONTROL *const rc = &cpi->rc;
-  const unsigned int curr_frame = cpi->common.current_video_frame;
+  const unsigned int curr_frame = cpi->common.current_frame.frame_number;
   int active_worst_quality;
 
-  if (cpi->common.frame_type == KEY_FRAME) {
+  if (cpi->common.current_frame.frame_type == KEY_FRAME) {
     active_worst_quality =
         curr_frame == 0 ? rc->worst_quality : rc->last_q[KEY_FRAME] * 2;
   } else {
@@ -604,13 +602,13 @@ static int calc_active_worst_quality_one_pass_cbr(const AV1_COMP *cpi) {
   int adjustment = 0;
   int active_worst_quality;
   int ambient_qp;
-  if (cm->frame_type == KEY_FRAME) return rc->worst_quality;
+  if (cm->current_frame.frame_type == KEY_FRAME) return rc->worst_quality;
   // For ambient_qp we use minimum of avg_frame_qindex[KEY_FRAME/INTER_FRAME]
   // for the first few frames following key frame. These are both initialized
   // to worst_quality and updated with (3/4, 1/4) average in postencode_update.
   // So for first few frames following key, the qp of that key frame is weighted
   // into the active_worst_quality setting.
-  ambient_qp = (cm->current_video_frame < 5)
+  ambient_qp = (cm->current_frame.frame_number < 5)
                    ? AOMMIN(rc->avg_frame_qindex[INTER_FRAME],
                             rc->avg_frame_qindex[KEY_FRAME])
                    : rc->avg_frame_qindex[INTER_FRAME];
@@ -650,6 +648,7 @@ static int rc_pick_q_and_bounds_one_pass_cbr(const AV1_COMP *cpi, int width,
                                              int *top_index) {
   const AV1_COMMON *const cm = &cpi->common;
   const RATE_CONTROL *const rc = &cpi->rc;
+  const CurrentFrame *const current_frame = &cm->current_frame;
   int active_best_quality;
   int active_worst_quality = calc_active_worst_quality_one_pass_cbr(cpi);
   int q;
@@ -668,7 +667,7 @@ static int rc_pick_q_and_bounds_one_pass_cbr(const AV1_COMP *cpi, int width,
       int delta_qindex = av1_compute_qdelta(rc, last_boosted_q,
                                             (last_boosted_q * 0.75), bit_depth);
       active_best_quality = AOMMAX(qindex + delta_qindex, rc->best_quality);
-    } else if (cm->current_video_frame > 0) {
+    } else if (current_frame->frame_number > 0) {
       // not first frame of one pass and kf_boost is set
       double q_adj_factor = 1.0;
       double q_val;
@@ -701,7 +700,7 @@ static int rc_pick_q_and_bounds_one_pass_cbr(const AV1_COMP *cpi, int width,
     active_best_quality = get_gf_active_quality(rc, q, bit_depth);
   } else {
     // Use the lower of active_worst_quality and recent/average Q.
-    if (cm->current_video_frame > 1) {
+    if (current_frame->frame_number > 1) {
       if (rc->avg_frame_qindex[INTER_FRAME] < active_worst_quality)
         active_best_quality = rtc_minq[rc->avg_frame_qindex[INTER_FRAME]];
       else
@@ -724,18 +723,18 @@ static int rc_pick_q_and_bounds_one_pass_cbr(const AV1_COMP *cpi, int width,
   *bottom_index = active_best_quality;
 
   // Limit Q range for the adaptive loop.
-  if (cm->frame_type == KEY_FRAME && !rc->this_key_frame_forced &&
-      !(cm->current_video_frame == 0)) {
+  if (current_frame->frame_type == KEY_FRAME && !rc->this_key_frame_forced &&
+      !(current_frame->frame_number == 0)) {
     int qdelta = 0;
     aom_clear_system_state();
-    qdelta = av1_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
+    qdelta = av1_compute_qdelta_by_rate(&cpi->rc, current_frame->frame_type,
                                         active_worst_quality, 2.0, bit_depth);
     *top_index = active_worst_quality + qdelta;
     *top_index = AOMMAX(*top_index, *bottom_index);
   }
 
   // Special case code to try and match quality with forced key frames
-  if (cm->frame_type == KEY_FRAME && rc->this_key_frame_forced) {
+  if (current_frame->frame_type == KEY_FRAME && rc->this_key_frame_forced) {
     q = rc->last_boosted_qindex;
   } else {
     q = av1_rc_regulate_q(cpi, rc->this_frame_target, active_best_quality,
@@ -756,10 +755,28 @@ static int rc_pick_q_and_bounds_one_pass_cbr(const AV1_COMP *cpi, int width,
   return q;
 }
 
+static int gf_group_pyramid_level(const AV1_COMP *cpi) {
+  const GF_GROUP *gf_group = &cpi->twopass.gf_group;
+  int this_height = gf_group->pyramid_level[gf_group->index];
+  return this_height;
+}
+
 static int get_active_cq_level(const RATE_CONTROL *rc,
-                               const AV1EncoderConfig *const oxcf) {
+                               const AV1EncoderConfig *const oxcf,
+                               int intra_only, int superres_denom) {
   static const double cq_adjust_threshold = 0.1;
   int active_cq_level = oxcf->cq_level;
+  (void)intra_only;
+  if (oxcf->rc_mode == AOM_CQ || oxcf->rc_mode == AOM_Q) {
+    // printf("Superres %d %d %d = %d\n", superres_denom, intra_only,
+    //        rc->frames_to_key, !(intra_only && rc->frames_to_key <= 1));
+    if (oxcf->superres_mode == SUPERRES_QTHRESH &&
+        superres_denom != SCALE_NUMERATOR &&
+        !(intra_only && rc->frames_to_key <= 1)) {
+      active_cq_level =
+          AOMMAX(active_cq_level - ((superres_denom - SCALE_NUMERATOR) * 4), 0);
+    }
+  }
   if (oxcf->rc_mode == AOM_CQ && rc->total_target_bits > 0) {
     const double x = (double)rc->total_actual_bits / rc->total_target_bits;
     if (x < cq_adjust_threshold) {
@@ -774,8 +791,10 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const AV1_COMP *cpi, int width,
                                              int *top_index) {
   const AV1_COMMON *const cm = &cpi->common;
   const RATE_CONTROL *const rc = &cpi->rc;
+  const CurrentFrame *const current_frame = &cm->current_frame;
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
-  const int cq_level = get_active_cq_level(rc, oxcf);
+  const int cq_level = get_active_cq_level(rc, oxcf, frame_is_intra_only(cm),
+                                           cm->superres_scale_denominator);
   int active_best_quality;
   int active_worst_quality = calc_active_worst_quality_one_pass_vbr(cpi);
   int q;
@@ -849,12 +868,12 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const AV1_COMP *cpi, int width,
                                                      0.70, 1.0, 0.85, 1.0 };
       const int delta_qindex = av1_compute_qdelta(
           rc, q_val,
-          q_val * delta_rate[cm->current_video_frame % FIXED_GF_INTERVAL],
+          q_val * delta_rate[current_frame->frame_number % FIXED_GF_INTERVAL],
           bit_depth);
       active_best_quality = AOMMAX(qindex + delta_qindex, rc->best_quality);
     } else {
       // Use the lower of active_worst_quality and recent/average Q.
-      active_best_quality = (cm->current_video_frame > 1)
+      active_best_quality = (current_frame->frame_number > 1)
                                 ? inter_minq[rc->avg_frame_qindex[INTER_FRAME]]
                                 : inter_minq[rc->avg_frame_qindex[KEY_FRAME]];
       // For the constrained quality mode we don't want
@@ -878,14 +897,15 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const AV1_COMP *cpi, int width,
   {
     int qdelta = 0;
     aom_clear_system_state();
-    if (cm->frame_type == KEY_FRAME && !rc->this_key_frame_forced &&
-        !(cm->current_video_frame == 0)) {
-      qdelta = av1_compute_qdelta_by_rate(&cpi->rc, cm->frame_type,
+    if (current_frame->frame_type == KEY_FRAME && !rc->this_key_frame_forced &&
+        !(current_frame->frame_number == 0)) {
+      qdelta = av1_compute_qdelta_by_rate(&cpi->rc, current_frame->frame_type,
                                           active_worst_quality, 2.0, bit_depth);
     } else if (!rc->is_src_frame_alt_ref &&
                (cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
-      qdelta = av1_compute_qdelta_by_rate(
-          &cpi->rc, cm->frame_type, active_worst_quality, 1.75, bit_depth);
+      qdelta =
+          av1_compute_qdelta_by_rate(&cpi->rc, current_frame->frame_type,
+                                     active_worst_quality, 1.75, bit_depth);
     }
     *top_index = active_worst_quality + qdelta;
     *top_index = AOMMAX(*top_index, *bottom_index);
@@ -894,7 +914,8 @@ static int rc_pick_q_and_bounds_one_pass_vbr(const AV1_COMP *cpi, int width,
   if (oxcf->rc_mode == AOM_Q) {
     q = active_best_quality;
     // Special case code to try and match quality with forced key frames
-  } else if ((cm->frame_type == KEY_FRAME) && rc->this_key_frame_forced) {
+  } else if ((current_frame->frame_type == KEY_FRAME) &&
+             rc->this_key_frame_forced) {
     q = rc->last_boosted_qindex;
   } else {
     q = av1_rc_regulate_q(cpi, rc->this_frame_target, active_best_quality,
@@ -934,7 +955,8 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
   const RATE_CONTROL *const rc = &cpi->rc;
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
   const GF_GROUP *gf_group = &cpi->twopass.gf_group;
-  const int cq_level = get_active_cq_level(rc, oxcf);
+  const int cq_level = get_active_cq_level(rc, oxcf, frame_is_intra_only(cm),
+                                           cm->superres_scale_denominator);
   int active_best_quality;
   int active_worst_quality = cpi->twopass.active_worst_quality;
   int q;
@@ -942,18 +964,20 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
   const int bit_depth = cm->seq_params.bit_depth;
   ASSIGN_MINQ_TABLE(bit_depth, inter_minq);
 
-#if CUSTOMIZED_GF
   const int is_intrl_arf_boost =
       gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE;
-#else
-  const int is_intrl_arf_boost = cpi->refresh_alt2_ref_frame;
-#endif  // CUSTOMIZED_GF
 
   if (frame_is_intra_only(cm)) {
-    // Handle the special case for key frames forced when we have reached
-    // the maximum key frame interval. Here force the Q to a range
-    // based on the ambient Q to reduce the risk of popping.
-    if (rc->this_key_frame_forced) {
+    if (rc->frames_to_key == 1 && oxcf->rc_mode == AOM_Q) {
+      // If the next frame is also a key frame or the current frame is the
+      // only frame in the sequence in AOM_Q mode, just use the cq_level
+      // as q.
+      active_best_quality = cq_level;
+      active_worst_quality = cq_level;
+    } else if (rc->this_key_frame_forced) {
+      // Handle the special case for key frames forced when we have reached
+      // the maximum key frame interval. Here force the Q to a range
+      // based on the ambient Q to reduce the risk of popping.
       double last_boosted_q;
       int delta_qindex;
       int qindex;
@@ -967,10 +991,13 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
         active_worst_quality =
             AOMMIN(qindex + delta_qindex, active_worst_quality);
       } else {
+        // Increase the boost if the forced keyframe is a forward reference.
+        // These numbers were derived empirically.
+        const double boost_factor = cpi->oxcf.fwd_kf_enabled ? 0.25 : 0.50;
         qindex = rc->last_boosted_qindex;
         last_boosted_q = av1_convert_qindex_to_q(qindex, bit_depth);
-        delta_qindex = av1_compute_qdelta(rc, last_boosted_q,
-                                          last_boosted_q * 0.5, bit_depth);
+        delta_qindex = av1_compute_qdelta(
+            rc, last_boosted_q, last_boosted_q * boost_factor, bit_depth);
         active_best_quality = AOMMAX(qindex + delta_qindex, rc->best_quality);
       }
     } else {
@@ -982,6 +1009,10 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
       active_best_quality =
           get_kf_active_quality(rc, active_worst_quality, bit_depth);
 
+      if (cpi->twopass.kf_zeromotion_pct >= STATIC_KF_GROUP_THRESH) {
+        active_best_quality /= 3;
+      }
+
       // Allow somewhat lower kf minq with small image formats.
       if ((width * height) <= (352 * 288)) {
         q_adj_factor -= 0.25;
@@ -1011,33 +1042,30 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
     // For constrained quality dont allow Q less than the cq level
     if (oxcf->rc_mode == AOM_CQ) {
       if (q < cq_level) q = cq_level;
-#if USE_SYMM_MULTI_LAYER && MULTI_LVL_BOOST_VBR_CQ
+
+      active_best_quality = get_gf_active_quality(rc, q, bit_depth);
+
+      // Constrained quality use slightly lower active best.
+      active_best_quality = active_best_quality * 15 / 16;
+
       if (gf_group->update_type[gf_group->index] == ARF_UPDATE ||
           (is_intrl_arf_boost && !cpi->new_bwdref_update_rule)) {
-#endif  // USE_SYMM_MULTI_LAYER && MULTI_LVL_BOOST_VBR_CQ
-        active_best_quality = get_gf_active_quality(rc, q, bit_depth);
-
-        // Constrained quality use slightly lower active best.
-        active_best_quality = active_best_quality * 15 / 16;
-#if REDUCE_LAST_ALT_BOOST
         if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
           const int min_boost = get_gf_high_motion_quality(q, bit_depth);
           const int boost = min_boost - active_best_quality;
 
           active_best_quality = min_boost - (int)(boost * rc->arf_boost_factor);
         }
-#endif
         *arf_q = active_best_quality;
-#if USE_SYMM_MULTI_LAYER && MULTI_LVL_BOOST_VBR_CQ
-      } else {
+      } else if (cpi->new_bwdref_update_rule && is_intrl_arf_boost) {
+        assert(rc->arf_q >= 0);  // Ensure it is set to a valid value.
         active_best_quality = rc->arf_q;
-        int this_height = gf_group->pyramid_level[gf_group->index];
+        int this_height = gf_group_pyramid_level(cpi);
         while (this_height < gf_group->pyramid_height) {
           active_best_quality = (active_best_quality + cq_level + 1) / 2;
           ++this_height;
         }
       }
-#endif  // USE_SYMM_MULTI_LAYER && MULTI_LVL_BOOST_VBR_CQ
     } else if (oxcf->rc_mode == AOM_Q) {
       if (!cpi->refresh_alt_ref_frame && !is_intrl_arf_boost) {
         active_best_quality = cq_level;
@@ -1045,50 +1073,41 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
         if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
           active_best_quality = get_gf_active_quality(rc, q, bit_depth);
           *arf_q = active_best_quality;
-#if REDUCE_LAST_ALT_BOOST
           const int min_boost = get_gf_high_motion_quality(q, bit_depth);
           const int boost = min_boost - active_best_quality;
 
           active_best_quality = min_boost - (int)(boost * rc->arf_boost_factor);
-#endif
         } else {
+          assert(rc->arf_q >= 0);  // Ensure it is set to a valid value.
           active_best_quality = rc->arf_q;
         }
-#if USE_SYMM_MULTI_LAYER
         if (cpi->new_bwdref_update_rule && is_intrl_arf_boost) {
-          int this_height = gf_group->pyramid_level[gf_group->index];
+          int this_height = gf_group_pyramid_level(cpi);
           while (this_height < gf_group->pyramid_height) {
             active_best_quality = (active_best_quality + cq_level + 1) / 2;
             ++this_height;
           }
         } else {
-#endif
           // Modify best quality for second level arfs. For mode AOM_Q this
           // becomes the baseline frame q.
           if (gf_group->rf_level[gf_group->index] == GF_ARF_LOW)
             active_best_quality = (active_best_quality + cq_level + 1) / 2;
-#if USE_SYMM_MULTI_LAYER
         }
-#endif
       }
     } else {
       active_best_quality = get_gf_active_quality(rc, q, bit_depth);
-#if REDUCE_LAST_ALT_BOOST
       const int min_boost = get_gf_high_motion_quality(q, bit_depth);
       const int boost = min_boost - active_best_quality;
 
       active_best_quality = min_boost - (int)(boost * rc->arf_boost_factor);
-#endif
-#if USE_SYMM_MULTI_LAYER
       if (cpi->new_bwdref_update_rule && is_intrl_arf_boost) {
-        int this_height = gf_group->pyramid_level[gf_group->index];
+        int this_height = gf_group_pyramid_level(cpi);
         while (this_height < gf_group->pyramid_height) {
           active_best_quality =
               (active_best_quality + active_worst_quality + 1) / 2;
           ++this_height;
         }
       }
-#endif
     }
   } else {
     if (oxcf->rc_mode == AOM_Q) {
@@ -1135,7 +1154,7 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
   // Modify active_best_quality for downscaled normal frames.
   if (av1_frame_scaled(cm) && !frame_is_kf_gf_arf(cpi)) {
     int qdelta = av1_compute_qdelta_by_rate(
-        rc, cm->frame_type, active_best_quality, 2.0, bit_depth);
+        rc, cm->current_frame.frame_type, active_best_quality, 2.0, bit_depth);
     active_best_quality =
         AOMMAX(active_best_quality + qdelta, rc->best_quality);
   }
@@ -1145,7 +1164,9 @@ static int rc_pick_q_and_bounds_two_pass(const AV1_COMP *cpi, int width,
   active_worst_quality =
       clamp(active_worst_quality, active_best_quality, rc->worst_quality);
 
-  if (oxcf->rc_mode == AOM_Q) {
+  if (oxcf->rc_mode == AOM_Q ||
+      (frame_is_intra_only(cm) && !rc->this_key_frame_forced &&
+       cpi->twopass.kf_zeromotion_pct >= STATIC_KF_GROUP_THRESH)) {
     q = active_best_quality;
     // Special case code to try and match quality with forced key frames.
   } else if (frame_is_intra_only(cm) && rc->this_key_frame_forced) {
@@ -1193,7 +1214,7 @@ int av1_rc_pick_q_and_bounds(AV1_COMP *cpi, int width, int height,
     assert(cpi->oxcf.pass == 2 && "invalid encode pass");
 
     GF_GROUP *gf_group = &cpi->twopass.gf_group;
-    int arf_q = 0;
+    int arf_q = -1;  // Initialize to invalid value, for sanity check later.
 
     q = rc_pick_q_and_bounds_two_pass(cpi, width, height, bottom_index,
                                       top_index, &arf_q);
@@ -1253,16 +1274,12 @@ static void update_alt_ref_frame_stats(AV1_COMP *cpi) {
 
 static void update_golden_frame_stats(AV1_COMP *cpi) {
   RATE_CONTROL *const rc = &cpi->rc;
-#if CUSTOMIZED_GF
   const TWO_PASS *const twopass = &cpi->twopass;
   const GF_GROUP *const gf_group = &twopass->gf_group;
   const int is_intrnl_arf =
       cpi->oxcf.pass == 2
           ? gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE
           : cpi->refresh_alt2_ref_frame;
-#else
-  const int is_intnl_arf = cpi->refresh_alt2_ref_frame;
-#endif
 
   // Update the Golden frame usage counts.
   // NOTE(weitinglin): If we use show_existing_frame for an OVERLAY frame,
@@ -1288,19 +1305,42 @@ static void update_golden_frame_stats(AV1_COMP *cpi) {
   }
 }
 
+void av1_estimate_qp_gop(AV1_COMP *cpi) {
+  AV1_COMMON *const cm = &cpi->common;
+  int gop_length = cpi->rc.baseline_gf_interval;
+  int bottom_index, top_index;
+  int idx;
+  const int gf_index = cpi->twopass.gf_group.index;
+
+  for (idx = 1; idx <= gop_length + 1 && idx < MAX_LAG_BUFFERS; ++idx) {
+    TplDepFrame *tpl_frame = &cpi->tpl_stats[idx];
+    int target_rate = cpi->twopass.gf_group.bit_allocation[idx];
+    int arf_q = 0;
+
+    cpi->twopass.gf_group.index = idx;
+    rc_set_frame_target(cpi, target_rate, cm->width, cm->height);
+    av1_configure_buffer_updates(
+        cpi, cpi->twopass.gf_group.update_type[cpi->twopass.gf_group.index]);
+    tpl_frame->base_qindex = rc_pick_q_and_bounds_two_pass(
+        cpi, cm->width, cm->height, &bottom_index, &top_index, &arf_q);
+    tpl_frame->base_qindex = AOMMAX(tpl_frame->base_qindex, 1);
+  }
+  // Reset the actual index and frame update
+  cpi->twopass.gf_group.index = gf_index;
+  av1_configure_buffer_updates(
+      cpi, cpi->twopass.gf_group.update_type[cpi->twopass.gf_group.index]);
+}
+
 void av1_rc_postencode_update(AV1_COMP *cpi, uint64_t bytes_used) {
   const AV1_COMMON *const cm = &cpi->common;
+  const CurrentFrame *const current_frame = &cm->current_frame;
   RATE_CONTROL *const rc = &cpi->rc;
-#if CUSTOMIZED_GF
   const TWO_PASS *const twopass = &cpi->twopass;
   const GF_GROUP *const gf_group = &twopass->gf_group;
   const int is_intrnl_arf =
       cpi->oxcf.pass == 2
           ? gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE
           : cpi->refresh_alt2_ref_frame;
-#else
-  const int is_intrnl_arf = cpi->refresh_alt2_ref_frame;
-#endif
 
   const int qindex = cm->base_qindex;
 
@@ -1315,7 +1355,7 @@ void av1_rc_postencode_update(AV1_COMP *cpi, uint64_t bytes_used) {
   av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height);
 
   // Keep a record of last Q and ambient average Q.
-  if (cm->frame_type == KEY_FRAME) {
+  if (current_frame->frame_type == KEY_FRAME) {
     rc->last_q[KEY_FRAME] = qindex;
     rc->avg_frame_qindex[KEY_FRAME] =
         ROUND_POWER_OF_TWO(3 * rc->avg_frame_qindex[KEY_FRAME] + qindex, 2);
@@ -1341,13 +1381,14 @@ void av1_rc_postencode_update(AV1_COMP *cpi, uint64_t bytes_used) {
   // If all mbs in this group are skipped only update if the Q value is
   // better than that already stored.
   // This is used to help set quality in forced key frames to reduce popping
-  if ((qindex < rc->last_boosted_qindex) || (cm->frame_type == KEY_FRAME) ||
+  if ((qindex < rc->last_boosted_qindex) ||
+      (current_frame->frame_type == KEY_FRAME) ||
       (!rc->constrained_gf_group &&
        (cpi->refresh_alt_ref_frame || is_intrnl_arf ||
         (cpi->refresh_golden_frame && !rc->is_src_frame_alt_ref)))) {
     rc->last_boosted_qindex = qindex;
   }
-  if (cm->frame_type == KEY_FRAME) rc->last_kf_qindex = qindex;
+  if (current_frame->frame_type == KEY_FRAME) rc->last_kf_qindex = qindex;
 
   update_buffer_level(cpi, rc->projected_frame_size);
 
@@ -1357,7 +1398,7 @@ void av1_rc_postencode_update(AV1_COMP *cpi, uint64_t bytes_used) {
     rc->this_frame_target =
         (int)(rc->this_frame_target /
               resize_rate_factor(cpi, cm->width, cm->height));
-  if (cm->frame_type != KEY_FRAME) {
+  if (current_frame->frame_type != KEY_FRAME) {
     rc->rolling_target_bits = ROUND_POWER_OF_TWO(
         rc->rolling_target_bits * 3 + rc->this_frame_target, 2);
     rc->rolling_actual_bits = ROUND_POWER_OF_TWO(
@@ -1378,15 +1419,15 @@ void av1_rc_postencode_update(AV1_COMP *cpi, uint64_t bytes_used) {
   rc->total_target_vs_actual = rc->total_actual_bits - rc->total_target_bits;
 
   if (is_altref_enabled(cpi) && cpi->refresh_alt_ref_frame &&
-      (cm->frame_type != KEY_FRAME))
+      (current_frame->frame_type != KEY_FRAME))
     // Update the alternate reference frame stats as appropriate.
     update_alt_ref_frame_stats(cpi);
   else
     // Update the Golden frame stats as appropriate.
     update_golden_frame_stats(cpi);
 
-  if (cm->frame_type == KEY_FRAME) rc->frames_since_key = 0;
-  // if (cm->current_video_frame == 1 && cm->show_frame)
+  if (current_frame->frame_type == KEY_FRAME) rc->frames_since_key = 0;
+  // if (current_frame->frame_number == 1 && cm->show_frame)
   /*
   rc->this_frame_target =
       (int)(rc->this_frame_target / resize_rate_factor(cpi, cm->width,
@@ -1431,9 +1472,11 @@ static int calc_iframe_target_size_one_pass_vbr(const AV1_COMP *const cpi) {
   return av1_rc_clamp_iframe_target_size(cpi, target);
 }
 
-void av1_rc_get_one_pass_vbr_params(AV1_COMP *cpi) {
+void av1_rc_get_one_pass_vbr_params(AV1_COMP *cpi,
+                                    EncodeFrameParams *const frame_params) {
   AV1_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
+  CurrentFrame *const current_frame = &cm->current_frame;
   int target;
   int altref_enabled = is_altref_enabled(cpi);
   int sframe_dist = cpi->oxcf.sframe_dist;
@@ -1441,44 +1484,44 @@ void av1_rc_get_one_pass_vbr_params(AV1_COMP *cpi) {
   int sframe_enabled = cpi->oxcf.sframe_enabled;
   // TODO(yaowu): replace the "auto_key && 0" below with proper decision logic.
   if (!cpi->refresh_alt_ref_frame &&
-      (cm->current_video_frame == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY) ||
-       rc->frames_to_key == 0 || (cpi->oxcf.auto_key && 0))) {
-    cm->frame_type = KEY_FRAME;
+      (current_frame->frame_number == 0 ||
+       (cpi->frame_flags & FRAMEFLAGS_KEY) || rc->frames_to_key == 0 ||
+       (cpi->oxcf.auto_key && 0))) {
+    frame_params->frame_type = KEY_FRAME;
     rc->this_key_frame_forced =
-        cm->current_video_frame != 0 && rc->frames_to_key == 0;
+        current_frame->frame_number != 0 && rc->frames_to_key == 0;
     rc->frames_to_key = cpi->oxcf.key_freq;
     rc->kf_boost = DEFAULT_KF_BOOST;
     rc->source_alt_ref_active = 0;
   } else {
-    cm->frame_type = INTER_FRAME;
+    frame_params->frame_type = INTER_FRAME;
     if (sframe_enabled) {
       if (altref_enabled) {
         if (sframe_mode == 1) {
           // sframe_mode == 1: insert sframe if it matches altref frame.
 
-          if (cm->current_video_frame % sframe_dist == 0 &&
-              cm->frame_type != KEY_FRAME && cm->current_video_frame != 0 &&
-              cpi->refresh_alt_ref_frame) {
-            cm->frame_type = S_FRAME;
+          if (current_frame->frame_number % sframe_dist == 0 &&
+              current_frame->frame_number != 0 && cpi->refresh_alt_ref_frame) {
+            frame_params->frame_type = S_FRAME;
           }
         } else {
           // sframe_mode != 1: if sframe will be inserted at the next available
           // altref frame
 
-          if (cm->current_video_frame % sframe_dist == 0 &&
-              cm->frame_type != KEY_FRAME && cm->current_video_frame != 0) {
+          if (current_frame->frame_number % sframe_dist == 0 &&
+              current_frame->frame_number != 0) {
             rc->sframe_due = 1;
           }
 
           if (rc->sframe_due && cpi->refresh_alt_ref_frame) {
-            cm->frame_type = S_FRAME;
+            frame_params->frame_type = S_FRAME;
             rc->sframe_due = 0;
           }
         }
       } else {
-        if (cm->current_video_frame % sframe_dist == 0 &&
-            cm->frame_type != KEY_FRAME && cm->current_video_frame != 0) {
-          cm->frame_type = S_FRAME;
+        if (current_frame->frame_number % sframe_dist == 0 &&
+            current_frame->frame_number != 0) {
+          frame_params->frame_type = S_FRAME;
         }
       }
     }
@@ -1501,7 +1544,7 @@ void av1_rc_get_one_pass_vbr_params(AV1_COMP *cpi) {
   if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ)
     av1_cyclic_refresh_update_parameters(cpi);
 
-  if (cm->frame_type == KEY_FRAME)
+  if (frame_params->frame_type == KEY_FRAME)
     target = calc_iframe_target_size_one_pass_vbr(cpi);
   else
     target = calc_pframe_target_size_one_pass_vbr(cpi);
@@ -1550,7 +1593,7 @@ static int calc_pframe_target_size_one_pass_cbr(const AV1_COMP *cpi) {
 static int calc_iframe_target_size_one_pass_cbr(const AV1_COMP *cpi) {
   const RATE_CONTROL *rc = &cpi->rc;
   int target;
-  if (cpi->common.current_video_frame == 0) {
+  if (cpi->common.current_frame.frame_number == 0) {
     target = ((rc->starting_buffer_level / 2) > INT_MAX)
                  ? INT_MAX
                  : (int)(rc->starting_buffer_level / 2);
@@ -1567,21 +1610,24 @@ static int calc_iframe_target_size_one_pass_cbr(const AV1_COMP *cpi) {
   return av1_rc_clamp_iframe_target_size(cpi, target);
 }
 
-void av1_rc_get_one_pass_cbr_params(AV1_COMP *cpi) {
+void av1_rc_get_one_pass_cbr_params(AV1_COMP *cpi,
+                                    EncodeFrameParams *const frame_params) {
   AV1_COMMON *const cm = &cpi->common;
   RATE_CONTROL *const rc = &cpi->rc;
+  CurrentFrame *const current_frame = &cm->current_frame;
   int target;
   // TODO(yaowu): replace the "auto_key && 0" below with proper decision logic.
-  if ((cm->current_video_frame == 0 || (cpi->frame_flags & FRAMEFLAGS_KEY) ||
-       rc->frames_to_key == 0 || (cpi->oxcf.auto_key && 0))) {
-    cm->frame_type = KEY_FRAME;
+  if ((current_frame->frame_number == 0 ||
+       (cpi->frame_flags & FRAMEFLAGS_KEY) || rc->frames_to_key == 0 ||
+       (cpi->oxcf.auto_key && 0))) {
+    frame_params->frame_type = KEY_FRAME;
     rc->this_key_frame_forced =
-        cm->current_video_frame != 0 && rc->frames_to_key == 0;
+        current_frame->frame_number != 0 && rc->frames_to_key == 0;
     rc->frames_to_key = cpi->oxcf.key_freq;
     rc->kf_boost = DEFAULT_KF_BOOST;
     rc->source_alt_ref_active = 0;
   } else {
-    cm->frame_type = INTER_FRAME;
+    frame_params->frame_type = INTER_FRAME;
   }
   if (rc->frames_till_gf_update_due == 0) {
     if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ)
@@ -1602,7 +1648,7 @@ void av1_rc_get_one_pass_cbr_params(AV1_COMP *cpi) {
   if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ)
     av1_cyclic_refresh_update_parameters(cpi);
 
-  if (cm->frame_type == KEY_FRAME)
+  if (frame_params->frame_type == KEY_FRAME)
     target = calc_iframe_target_size_one_pass_cbr(cpi);
   else
     target = calc_pframe_target_size_one_pass_cbr(cpi);
@@ -1674,15 +1720,10 @@ void av1_rc_set_gf_interval_range(const AV1_COMP *const cpi,
           oxcf->width, oxcf->height, cpi->framerate);
     if (rc->max_gf_interval == 0)
       rc->max_gf_interval = av1_rc_get_default_max_gf_interval(
-          cpi->framerate, rc->min_gf_interval);
+          cpi->framerate, rc->min_gf_interval, oxcf->gf_max_pyr_height);
 
-    // Extended interval for genuinely static scenes
-    rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
-
-    if (is_altref_enabled(cpi)) {
-      if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
-        rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
-    }
+    // Extended max interval for genuinely static scenes like slide shows.
+    rc->static_scene_max_gf_interval = MAX_STATIC_GF_GROUP_LENGTH;
 
     if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
       rc->max_gf_interval = rc->static_scene_max_gf_interval;
@@ -1733,7 +1774,7 @@ static void vbr_rate_correction(AV1_COMP *cpi, int *this_frame_target) {
   // This number is used to damp the per frame rate correction.
   // Range 0 - 1.0
   if (cpi->twopass.total_stats.count != 0.) {
-    position_factor = sqrt((double)cpi->common.current_video_frame /
+    position_factor = sqrt((double)cpi->common.current_frame.frame_number /
                            cpi->twopass.total_stats.count);
   }
   max_delta = (int)(position_factor *
diff --git a/third_party/aom/av1/encoder/ratectrl.h b/third_party/aom/av1/encoder/ratectrl.h
index 198ecab975..36c738c8cc 100644
--- a/third_party/aom/av1/encoder/ratectrl.h
+++ b/third_party/aom/av1/encoder/ratectrl.h
@@ -16,6 +16,7 @@
 #include "aom/aom_integer.h"
 
 #include "av1/common/blockd.h"
+#include "av1/common/onyxc_int.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -24,48 +25,24 @@ extern "C" {
 // Bits Per MB at different Q (Multiplied by 512)
 #define BPER_MB_NORMBITS 9
 
-#define CUSTOMIZED_GF 1
+// Threshold used to define if a KF group is static (e.g. a slide show).
+// Essentially, this means that no frame in the group has more than 1% of MBs
+// that are not marked as coded with 0,0 motion in the first pass.
+#define STATIC_KF_GROUP_THRESH 99
+#define STATIC_KF_GROUP_FLOAT_THRESH 0.99
 
-#if CONFIG_FIX_GF_LENGTH
-#define FIXED_GF_LENGTH 16
+// The maximum duration of a GF group that is static (e.g. a slide show).
+#define MAX_STATIC_GF_GROUP_LENGTH 250
+
+// Minimum and maximum height for the new pyramid structure.
+// (Old structure supports height = 1, but does NOT support height = 4).
+#define MIN_PYRAMID_LVL 2
 #define MAX_PYRAMID_LVL 4
-// We allow a frame to have at most two left/right descendants before changing
-// them into to a subtree, i.e., we allow the following structure:
-/*                    OUT_OF_ORDER_FRAME
-                     / /              \ \
-(two left children) F F                F F (two right children) */
-// Therefore the max gf size supported by 4 layer structure is
-// 1 (KEY/OVERLAY) + 1 + 2 + 4 + 16 (two children on both side of their parent)
-#define MAX_PYRAMID_SIZE 24
-#define USE_SYMM_MULTI_LAYER 1
-#define REDUCE_LAST_ALT_BOOST 1
-#define REDUCE_LAST_GF_LENGTH 1
-#define MULTI_LVL_BOOST_VBR_CQ 1
-#else
-#define USE_SYMM_MULTI_LAYER 0
-#define REDUCE_LAST_ALT_BOOST 0
-#define REDUCE_LAST_GF_LENGTH 0
-#define MULTI_LVL_BOOST_VBR_CQ 0
-#endif
-
-#if USE_SYMM_MULTI_LAYER
-#define USE_MANUAL_GF4_STRUCT 0
-#endif
 
 #define MIN_GF_INTERVAL 4
 #define MAX_GF_INTERVAL 16
 #define FIXED_GF_INTERVAL 8  // Used in some testing modes only
 
-typedef enum {
-  INTER_NORMAL = 0,
-  INTER_LOW = 1,
-  INTER_HIGH = 2,
-  GF_ARF_LOW = 3,
-  GF_ARF_STD = 4,
-  KF_STD = 5,
-  RATE_FACTOR_LEVELS = 6
-} RATE_FACTOR_LEVEL;
-
 static const double rate_factor_deltas[RATE_FACTOR_LEVELS] = {
   1.00,  // INTER_NORMAL
   0.80,  // INTER_LOW
@@ -93,7 +70,6 @@ typedef struct {
   int last_kf_qindex;       // Q index of the last key frame coded.
 
   int gfu_boost;
-  int last_boost;
   int kf_boost;
 
   double rate_correction_factors[RATE_FACTOR_LEVELS];
@@ -195,7 +171,10 @@ int av1_rc_get_default_min_gf_interval(int width, int height, double framerate);
 // Note av1_rc_get_default_max_gf_interval() requires the min_gf_interval to
 // be passed in to ensure that the max_gf_interval returned is at least as bis
 // as that.
-int av1_rc_get_default_max_gf_interval(double framerate, int min_frame_rate);
+int av1_rc_get_default_max_gf_interval(double framerate, int min_frame_rate,
+                                       int max_pyr_height);
+
+int av1_rc_get_fixed_gf_length(int max_pyr_height);
 
 // Generally at the high level, the following flow is expected
 // to be enforced for rate control:
@@ -220,8 +199,11 @@ int av1_rc_get_default_max_gf_interval(double framerate, int min_frame_rate);
 
 // Functions to set parameters for encoding before the actual
 // encode_frame_to_data_rate() function.
-void av1_rc_get_one_pass_vbr_params(struct AV1_COMP *cpi);
-void av1_rc_get_one_pass_cbr_params(struct AV1_COMP *cpi);
+struct EncodeFrameParams;
+void av1_rc_get_one_pass_vbr_params(
+    struct AV1_COMP *cpi, struct EncodeFrameParams *const frame_params);
+void av1_rc_get_one_pass_cbr_params(
+    struct AV1_COMP *cpi, struct EncodeFrameParams *const frame_params);
 
 // Post encode update of the rate control parameters based
 // on bytes used
@@ -262,9 +244,6 @@ int av1_rc_clamp_iframe_target_size(const struct AV1_COMP *const cpi,
                                     int target);
 int av1_rc_clamp_pframe_target_size(const struct AV1_COMP *const cpi,
                                     int target);
-// Utility to set frame_target into the RATE_CONTROL structure
-// This function is called only from the av1_rc_get_..._params() functions.
-void av1_rc_set_frame_target(struct AV1_COMP *cpi, int target);
 
 // Computes a q delta (in "q index" terms) to get from a starting q value
 // to a target q value
@@ -288,6 +267,8 @@ void av1_set_target_rate(struct AV1_COMP *cpi, int width, int height);
 
 int av1_resize_one_pass_cbr(struct AV1_COMP *cpi);
 
+void av1_estimate_qp_gop(struct AV1_COMP *cpi);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/rd.c b/third_party/aom/av1/encoder/rd.c
index b87d89e50b..f71ec8033b 100644
--- a/third_party/aom/av1/encoder/rd.c
+++ b/third_party/aom/av1/encoder/rd.c
@@ -91,7 +91,7 @@ void av1_fill_mode_rates(AV1_COMMON *const cm, MACROBLOCK *x,
   for (i = 0; i < PARTITION_CONTEXTS; ++i)
     av1_cost_tokens_from_cdf(x->partition_cost[i], fc->partition_cdf[i], NULL);
 
-  if (cm->skip_mode_flag) {
+  if (cm->current_frame.skip_mode_info.skip_mode_flag) {
     for (i = 0; i < SKIP_CONTEXTS; ++i) {
       av1_cost_tokens_from_cdf(x->skip_mode_cost[i], fc->skip_mode_cdfs[i],
                                NULL);
@@ -353,19 +353,54 @@ static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = {
   128   // INTNL_ARF_UPDATE
 };
 
-int av1_compute_rd_mult(const AV1_COMP *cpi, int qindex) {
-  const int64_t q =
-      av1_dc_quant_Q3(qindex, 0, cpi->common.seq_params.bit_depth);
-  int64_t rdmult = 0;
+int av1_compute_rd_mult_based_on_qindex(const AV1_COMP *cpi, int qindex) {
+  const int q = av1_dc_quant_Q3(qindex, 0, cpi->common.seq_params.bit_depth);
+  int rdmult = q * q;
+  rdmult = rdmult * 3 + (rdmult * 2 / 3);
   switch (cpi->common.seq_params.bit_depth) {
-    case AOM_BITS_8: rdmult = 88 * q * q / 24; break;
-    case AOM_BITS_10: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4); break;
-    case AOM_BITS_12: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8); break;
+    case AOM_BITS_8: break;
+    case AOM_BITS_10: rdmult = ROUND_POWER_OF_TWO(rdmult, 4); break;
+    case AOM_BITS_12: rdmult = ROUND_POWER_OF_TWO(rdmult, 8); break;
     default:
       assert(0 && "bit_depth should be AOM_BITS_8, AOM_BITS_10 or AOM_BITS_12");
       return -1;
   }
-  if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
+  return rdmult > 0 ? rdmult : 1;
+}
+
+int av1_compute_rd_mult(const AV1_COMP *cpi, int qindex) {
+  int64_t rdmult = av1_compute_rd_mult_based_on_qindex(cpi, qindex);
+  if (cpi->oxcf.pass == 2 &&
+      (cpi->common.current_frame.frame_type != KEY_FRAME)) {
+    const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+    const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
+    const int boost_index = AOMMIN(15, (cpi->rc.gfu_boost / 100));
+
+    rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
+    rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
+  }
+  return (int)rdmult;
+}
+
+int av1_get_adaptive_rdmult(const AV1_COMP *cpi, double beta) {
+  const AV1_COMMON *cm = &cpi->common;
+  int64_t q =
+      av1_dc_quant_Q3(cm->base_qindex, 0, cpi->common.seq_params.bit_depth);
+  int64_t rdmult = 0;
+
+  switch (cpi->common.seq_params.bit_depth) {
+    case AOM_BITS_8: rdmult = (int)((88 * q * q / beta) / 24); break;
+    case AOM_BITS_10:
+      rdmult = ROUND_POWER_OF_TWO((int)((88 * q * q / beta) / 24), 4);
+      break;
+    default:
+      assert(cpi->common.seq_params.bit_depth == AOM_BITS_12);
+      rdmult = ROUND_POWER_OF_TWO((int)((88 * q * q / beta) / 24), 8);
+      break;
+  }
+
+  if (cpi->oxcf.pass == 2 &&
+      (cpi->common.current_frame.frame_type != KEY_FRAME)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
     const int boost_index = AOMMIN(15, (cpi->rc.gfu_boost / 100));
@@ -434,13 +469,6 @@ static void set_block_thresholds(const AV1_COMMON *cm, RD_OPT *rd) {
   }
 }
 
-void av1_set_mvcost(MACROBLOCK *x, int ref, int ref_mv_idx) {
-  (void)ref;
-  (void)ref_mv_idx;
-  x->mvcost = x->mv_cost_stack;
-  x->nmvjointcost = x->nmv_vec_cost;
-}
-
 void av1_fill_coeff_costs(MACROBLOCK *x, FRAME_CONTEXT *fc,
                           const int num_planes) {
   const int nplanes = AOMMIN(num_planes, PLANE_TYPES);
@@ -480,6 +508,17 @@ void av1_fill_coeff_costs(MACROBLOCK *x, FRAME_CONTEXT *fc,
         av1_cost_tokens_from_cdf(pcost->base_cost[ctx],
                                  fc->coeff_base_cdf[tx_size][plane][ctx], NULL);
 
+      for (int ctx = 0; ctx < SIG_COEF_CONTEXTS; ++ctx) {
+        pcost->base_cost[ctx][4] = 0;
+        pcost->base_cost[ctx][5] = pcost->base_cost[ctx][1] +
+                                   av1_cost_literal(1) -
+                                   pcost->base_cost[ctx][0];
+        pcost->base_cost[ctx][6] =
+            pcost->base_cost[ctx][2] - pcost->base_cost[ctx][1];
+        pcost->base_cost[ctx][7] =
+            pcost->base_cost[ctx][3] - pcost->base_cost[ctx][2];
+      }
+
       for (int ctx = 0; ctx < EOB_COEF_CONTEXTS; ++ctx)
         av1_cost_tokens_from_cdf(pcost->eob_extra_cost[ctx],
                                  fc->eob_extra_cdf[tx_size][plane][ctx], NULL);
@@ -510,10 +549,30 @@ void av1_fill_coeff_costs(MACROBLOCK *x, FRAME_CONTEXT *fc,
         //  printf("%5d ", pcost->lps_cost[ctx][i]);
         // printf("\n");
       }
+      for (int ctx = 0; ctx < LEVEL_CONTEXTS; ++ctx) {
+        pcost->lps_cost[ctx][0 + COEFF_BASE_RANGE + 1] =
+            pcost->lps_cost[ctx][0];
+        for (int i = 1; i <= COEFF_BASE_RANGE; ++i) {
+          pcost->lps_cost[ctx][i + COEFF_BASE_RANGE + 1] =
+              pcost->lps_cost[ctx][i] - pcost->lps_cost[ctx][i - 1];
+        }
+      }
     }
   }
 }
 
+void av1_initialize_cost_tables(const AV1_COMMON *const cm, MACROBLOCK *x) {
+  if (cm->cur_frame_force_integer_mv) {
+    av1_build_nmv_cost_table(x->nmv_vec_cost, x->nmvcost, &cm->fc->nmvc,
+                             MV_SUBPEL_NONE);
+  } else {
+    av1_build_nmv_cost_table(
+        x->nmv_vec_cost,
+        cm->allow_high_precision_mv ? x->nmvcost_hp : x->nmvcost, &cm->fc->nmvc,
+        cm->allow_high_precision_mv);
+  }
+}
+
 void av1_initialize_rd_consts(AV1_COMP *cpi) {
   AV1_COMMON *const cm = &cpi->common;
   MACROBLOCK *const x = &cpi->td.mb;
@@ -527,18 +586,7 @@ void av1_initialize_rd_consts(AV1_COMP *cpi) {
 
   set_block_thresholds(cm, rd);
 
-  if (cm->cur_frame_force_integer_mv) {
-    av1_build_nmv_cost_table(x->nmv_vec_cost, x->nmvcost, &cm->fc->nmvc,
-                             MV_SUBPEL_NONE);
-  } else {
-    av1_build_nmv_cost_table(
-        x->nmv_vec_cost,
-        cm->allow_high_precision_mv ? x->nmvcost_hp : x->nmvcost, &cm->fc->nmvc,
-        cm->allow_high_precision_mv);
-  }
-
-  x->mvcost = x->mv_cost_stack;
-  x->nmvjointcost = x->nmv_vec_cost;
+  av1_initialize_cost_tables(cm, x);
 
   if (frame_is_intra_only(cm) && cm->allow_screen_content_tools &&
       cpi->oxcf.pass != 1) {
@@ -665,385 +713,601 @@ static double interp_bicubic(const double *p, int p_stride, double x,
   return interp_cubic(q, y);
 }
 
-static const double interp_rgrid_surf[65 * 18] = {
-  0.104019,    0.245714,    0.293686,    0.358635,    0.382167,    0.412446,
-  0.419955,    0.421388,    0.426672,    0.427990,    0.428531,    0.456868,
-  0.569880,    0.638822,    1.016319,    2.143453,    3.565229,    4.720880,
-  0.124618,    0.294211,    0.352023,    0.429991,    0.458206,    0.494510,
-  0.503513,    0.505232,    0.511566,    0.513234,    0.519365,    0.570225,
-  0.697373,    0.840624,    1.462198,    3.289054,    6.256517,    6.852788,
-  0.118630,    0.269669,    0.346620,    0.430999,    0.459385,    0.495783,
-  0.504808,    0.506532,    0.512884,    0.514988,    0.543437,    0.662772,
-  0.795876,    1.313596,    2.403841,    4.163098,    7.440589,    8.616275,
-  0.093329,    0.168205,    0.321320,    0.430607,    0.459385,    0.495783,
-  0.504813,    0.506548,    0.512975,    0.520662,    0.571659,    0.701841,
-  1.010727,    2.138851,    3.460626,    6.317955,    10.098127,   14.418553,
-  0.087021,    0.142905,    0.315011,    0.430509,    0.459385,    0.495787,
-  0.505075,    0.507599,    0.513584,    0.543182,    0.669941,    0.825620,
-  1.362800,    2.572187,    4.205047,    7.498399,    12.303118,   16.641735,
-  0.086923,    0.142513,    0.314913,    0.430508,    0.459385,    0.495803,
-  0.506126,    0.511816,    0.514810,    0.549705,    0.725350,    1.127334,
-  2.168597,    3.463686,    6.318605,    10.162284,   18.556041,   19.847042,
-  0.086923,    0.142513,    0.314913,    0.430506,    0.459376,    0.495805,
-  0.506388,    0.512954,    0.520772,    0.580215,    0.810474,    1.391548,
-  2.579442,    4.205160,    7.498399,    12.381597,   21.703618,   24.015457,
-  0.086923,    0.142513,    0.314911,    0.430353,    0.458765,    0.495652,
-  0.506391,    0.513406,    0.544098,    0.702950,    1.121860,    2.168961,
-  3.463798,    6.318607,    10.162284,   18.685361,   28.188192,   37.638872,
-  0.086923,    0.142513,    0.314901,    0.429742,    0.456313,    0.495045,
-  0.506484,    0.519195,    0.580104,    0.810126,    1.391462,    2.579441,
-  4.205160,    7.498399,    12.381597,   21.848607,   33.367199,   42.623190,
-  0.086923,    0.142513,    0.314899,    0.429589,    0.455706,    0.495155,
-  0.507882,    0.542426,    0.702360,    1.119921,    2.168478,    3.463791,
-  6.318607,    10.162284,   18.685361,   28.345760,   47.802028,   49.163533,
-  0.086924,    0.142548,    0.315086,    0.429842,    0.455870,    0.496336,
-  0.512412,    0.556953,    0.773373,    1.266396,    2.548277,    4.204676,
-  7.498399,    12.381597,   21.848607,   33.548250,   54.301011,   56.262859,
-  0.087067,    0.144957,    0.327436,    0.446616,    0.466362,    0.505706,
-  0.522077,    0.610747,    0.972543,    1.666916,    3.338812,    6.316669,
-  10.162284,   18.685361,   28.345760,   48.065311,   66.145302,   78.396020,
-  0.094295,    0.164235,    0.393722,    0.534219,    0.530922,    0.579308,
-  0.603889,    0.760870,    1.229961,    2.423214,    4.173513,    7.497916,
-  12.381597,   21.848607,   33.548250,   54.589585,   74.875848,   86.468182,
-  0.124096,    0.213005,    0.497188,    0.665176,    0.685973,    0.800200,
-  0.911394,    1.077971,    1.677290,    3.332129,    6.314960,    10.162257,
-  18.685361,   28.345760,   48.065311,   66.453506,   98.275189,   96.862588,
-  0.140999,    0.270140,    0.658212,    0.867661,    0.970183,    1.149516,
-  1.480599,    1.664833,    2.421893,    3.857981,    7.418830,    12.380371,
-  21.848607,   33.548250,   54.589585,   75.188867,   106.657971,  99.762997,
-  0.178353,    0.398001,    0.988462,    1.241473,    1.340967,    1.713568,
-  2.335030,    2.701432,    3.348532,    5.077158,    9.829903,    18.676528,
-  28.345700,   48.065311,   66.453506,   98.588283,   117.057193,  101.130722,
-  0.281079,    0.548300,    1.395825,    1.780770,    2.000508,    2.702964,
-  3.638454,    4.573843,    5.051641,    7.079129,    11.293332,   21.594861,
-  33.544335,   54.589585,   75.188867,   106.971065,  119.957601,  101.466632,
-  0.476762,    0.842189,    2.019678,    2.723895,    3.188467,    4.011610,
-  5.545111,    7.508984,    8.176339,    9.774504,    14.720782,   27.334416,
-  48.049609,   66.453506,   98.588283,   117.370357,  121.329855,  101.509242,
-  0.993999,    1.520111,    3.013605,    4.203530,    4.982992,    6.074944,
-  8.583581,    11.818375,   14.192544,   14.937517,   21.258160,   33.305953,
-  54.585735,   75.188867,   106.971135,  120.279824,  121.976055,  102.690130,
-  1.776487,    2.613655,    4.356487,    6.161726,    7.622196,    9.464193,
-  13.077233,   18.051656,   23.221051,   24.080068,   30.085038,   48.345269,
-  66.457698,   98.588353,   117.379415,  121.976128,  124.356210,  107.713202,
-  3.191085,    4.495201,    5.686033,    8.365566,    11.275339,   14.706437,
-  20.300969,   28.152237,   35.688355,   39.341382,   41.030743,   55.752262,
-  75.211764,   106.980285,  120.608403,  124.680746,  130.222528,  112.260098,
-  6.136611,    7.305215,    7.272532,    10.646713,   15.630815,   22.383168,
-  31.349131,   42.419822,   52.301680,   58.983454,   58.915405,   69.161305,
-  98.992460,   117.713855,  124.344836,  130.623638,  138.442401,  127.846670,
-  11.707980,   13.490761,   11.640845,   14.176132,   22.131124,   33.776462,
-  47.365711,   61.603834,   75.281056,   83.463985,   85.510533,   86.026513,
-  108.787480,  123.031136,  130.607284,  138.954406,  160.867784,  158.958882,
-  27.062874,   32.195139,   24.147297,   22.114632,   35.580506,   52.551674,
-  71.652956,   88.606776,   102.107193,  110.703186,  114.398733,  111.118539,
-  121.503578,  132.455924,  139.490806,  161.412674,  193.563210,  172.203945,
-  35.625692,   47.953028,   42.639820,   42.276254,   58.815664,   84.977282,
-  110.656412,  126.168446,  134.658126,  140.604482,  144.006012,  141.702382,
-  140.125323,  153.122630,  164.748041,  194.156197,  206.854650,  174.013079,
-  49.516447,   65.335381,   71.738306,   81.872819,   98.400740,   136.840488,
-  163.775802,  169.440078,  172.747876,  171.222919,  171.679604,  172.173550,
-  168.200129,  187.617133,  199.683394,  207.768200,  210.062520,  175.478356,
-  60.341673,   92.487135,   119.907299,  136.068010,  144.778950,  189.443534,
-  220.120077,  219.641635,  214.616503,  205.894657,  198.453924,  200.013069,
-  195.938103,  206.118661,  210.447375,  212.061379,  216.078218,  181.162805,
-  78.422159,   112.242899,  158.416312,  181.404320,  193.188690,  229.296967,
-  270.461799,  275.168977,  256.511701,  244.706786,  231.344608,  226.065087,
-  222.248618,  218.662324,  217.966722,  218.248574,  218.818588,  182.740573,
-  88.713664,   123.594164,  172.928179,  213.781414,  245.800351,  252.063414,
-  313.283141,  331.703831,  305.866639,  285.177142,  269.759635,  251.988739,
-  245.998388,  232.688076,  230.588702,  230.882657,  230.319053,  192.120741,
-  102.540561,  152.905927,  189.137131,  241.806756,  273.868497,  284.258017,
-  339.689853,  373.561104,  362.657463,  326.291984,  311.922687,  290.460189,
-  276.774381,  273.012072,  277.751792,  279.123748,  278.820447,  233.813798,
-  132.983118,  176.307242,  197.415684,  243.307787,  280.893995,  332.922370,
-  340.329043,  404.530166,  419.475405,  375.775209,  351.300889,  340.042759,
-  315.683832,  306.123530,  306.359319,  306.733063,  307.609556,  261.647847,
-  149.579109,  185.925581,  207.937033,  245.159084,  301.890957,  350.040480,
-  352.250771,  418.742329,  458.112686,  430.125208,  386.460441,  380.346839,
-  354.679150,  337.305620,  334.504124,  335.889932,  341.060725,  286.898578,
-  153.576812,  202.105624,  219.366967,  248.524506,  314.255692,  350.607526,
-  390.567688,  408.629209,  488.000213,  480.563823,  432.461799,  410.412624,
-  398.607371,  400.188740,  402.780916,  408.853470,  430.449735,  363.777088,
-  161.353129,  214.848904,  231.549852,  258.536466,  313.163177,  368.140577,
-  412.136393,  413.409032,  499.838438,  519.571063,  485.833867,  444.562715,
-  435.738129,  442.358549,  450.166531,  453.208524,  458.424358,  385.823139,
-  175.109034,  227.608058,  250.069563,  286.101747,  312.256740,  378.421485,
-  413.344147,  435.058646,  476.960941,  542.448886,  530.189154,  495.408402,
-  475.326752,  465.017144,  464.694045,  465.144689,  466.905382,  398.669138,
-  184.750180,  240.766694,  283.240772,  305.480150,  322.409001,  374.526162,
-  427.141326,  452.840323,  472.604139,  545.366105,  567.676694,  541.666203,
-  509.591873,  492.044219,  492.778569,  493.765684,  493.235693,  413.684325,
-  194.728357,  254.928927,  289.991157,  300.193195,  324.194589,  371.563147,
-  439.226438,  468.295088,  495.654854,  533.506353,  587.476353,  578.298989,
-  548.041942,  527.393885,  538.965146,  545.070442,  544.295454,  454.012211,
-  205.195287,  283.135677,  297.921431,  319.295927,  355.621830,  392.466463,
-  446.696167,  485.053519,  516.426615,  532.264584,  588.481600,  615.906737,
-  589.319634,  555.754316,  558.389367,  569.094521,  569.779764,  475.384946,
-  218.552054,  298.511016,  319.188338,  351.781666,  372.789510,  412.827434,
-  464.569387,  506.270203,  533.049810,  553.347364,  580.644599,  632.759854,
-  622.235843,  569.960552,  580.799340,  586.553714,  579.488366,  491.826482,
-  244.803348,  299.790203,  324.187975,  363.280782,  403.710443,  441.724083,
-  492.732682,  534.722691,  552.193622,  575.112647,  586.097705,  635.224970,
-  644.642944,  606.017786,  640.321218,  642.316989,  616.397020,  548.300111,
-  256.957358,  318.638991,  355.063346,  389.889307,  433.607315,  468.209001,
-  515.178157,  573.556591,  578.113115,  587.246475,  601.762801,  638.454644,
-  656.574853,  641.184609,  676.908189,  684.198162,  678.387412,  574.805864,
-  251.211502,  323.448532,  364.227424,  411.792704,  462.226488,  503.572288,
-  549.299249,  599.124071,  601.227977,  597.118176,  613.247552,  633.278532,
-  658.074755,  664.930719,  685.731531,  693.632845,  693.076350,  578.326477,
-  267.695377,  354.273736,  389.976833,  438.518178,  493.332686,  544.343027,
-  588.895829,  620.206193,  628.327410,  606.067827,  620.998532,  657.985256,
-  683.936059,  691.345257,  693.894723,  695.175306,  693.618786,  578.517148,
-  274.290725,  363.465288,  411.808596,  463.369805,  515.310226,  581.009306,
-  613.070738,  636.638714,  647.333929,  629.867603,  644.646319,  687.796202,
-  702.859596,  713.495479,  704.068069,  704.991807,  704.188594,  587.283658,
-  302.538449,  389.174737,  438.518422,  493.398902,  547.662399,  601.981814,
-  624.773046,  641.629484,  644.699451,  645.848784,  668.033340,  703.643523,
-  707.422408,  717.329600,  726.298973,  744.127507,  745.365167,  617.954068,
-  310.328188,  410.984766,  463.369805,  515.315010,  581.309832,  613.787792,
-  634.988538,  654.145284,  662.632978,  668.413496,  706.494057,  750.545471,
-  730.724808,  730.002100,  743.625262,  750.801609,  745.308457,  606.505800,
-  329.948756,  437.600191,  493.398902,  547.661910,  601.917884,  622.557745,
-  633.244395,  644.055898,  648.224221,  665.062911,  763.555733,  812.391078,
-  769.063582,  744.865168,  727.579796,  724.950408,  722.179707,  598.564510,
-  350.848328,  462.437458,  515.315010,  581.309823,  613.779123,  634.465309,
-  652.056257,  662.179143,  671.466297,  726.881256,  819.824030,  880.232789,
-  810.371672,  754.246481,  725.053473,  724.253390,  723.503395,  603.394909,
-  373.704088,  492.408266,  547.661910,  601.917884,  622.557620,  633.236320,
-  644.023513,  648.232514,  666.381639,  785.498283,  929.441612,  999.772800,
-  890.339033,  775.852504,  731.840181,  726.905100,  725.251844,  604.899901,
-  394.473422,  514.261306,  581.309823,  613.779123,  634.465309,  652.056257,
-  662.179143,  671.466557,  727.134512,  835.764144,  981.747089,  1018.462934,
-  939.686967,  811.276731,  739.398459,  727.365647,  725.285425,  604.923525,
-  419.976505,  546.538939,  601.917884,  622.557620,  633.236320,  644.023513,
-  648.232514,  666.381639,  785.545191,  932.841398,  1036.609617, 1026.945092,
-  963.822765,  840.827315,  755.532423,  730.241865,  725.366847,  604.924155,
-  437.281359,  580.116337,  613.779123,  634.465309,  652.056257,  662.179143,
-  671.466557,  727.134512,  835.764859,  981.996194,  1031.896881, 1002.544732,
-  881.157178,  828.151494,  799.340975,  751.314325,  728.316587,  605.005504,
-  464.713920,  600.649281,  622.557620,  633.236320,  644.023513,  648.232514,
-  666.381639,  785.545191,  932.841398,  1036.735329, 1035.037004, 995.478339,
-  858.093733,  823.471976,  819.881754,  798.749289,  749.440463,  607.955244,
-  495.880237,  612.473139,  634.465309,  652.056257,  662.179143,  671.466557,
-  727.134512,  835.764859,  981.996194,  1032.339788, 1031.105117, 995.303259,
-  857.733663,  823.435877,  822.822791,  819.873050,  796.882480,  629.038445,
-  510.391280,  621.158273,  633.236320,  644.023513,  648.232514,  666.381639,
-  785.545191,  932.841398,  1036.735329, 1035.566013, 1029.599350, 994.926093,
-  857.645648,  823.435143,  822.904139,  822.822791,  817.965681,  673.856962,
-  514.588176,  632.947715,  652.056257,  662.179143,  671.466557,  727.134512,
-  835.764859,  981.996194,  1032.339788, 1031.547475, 1023.835377, 972.158629,
-  851.968626,  823.347128,  822.904770,  822.904139,  820.752301,  684.418900,
-  520.013294,  631.668183,  644.023513,  648.232514,  666.381639,  785.545191,
-  932.841398,  1036.735329, 1035.567378, 1029.776746, 1001.044108, 880.853721,
-  829.201546,  822.994150,  822.904770,  822.904770,  820.792975,  684.582020,
-  531.253628,  650.479606,  662.179143,  671.466557,  727.134512,  835.764859,
-  981.996194,  1032.339788, 1031.636855, 1029.601779, 995.366703,  858.086641,
-  823.524524,  822.906135,  822.904770,  822.904770,  820.792975,  684.582020,
-  528.531744,  642.424501,  648.232514,  666.381639,  785.545191,  932.841398,
-  1036.735329, 1035.567378, 1030.219103, 1029.576226, 995.278687,  857.733663,
-  823.436508,  822.904770,  822.904770,  822.904770,  820.792975,  684.582020,
-  545.401164,  660.550678,  671.508859,  727.304161,  835.807162,  981.996850,
-  1032.339788, 1031.636855, 1030.130788, 1029.487827, 994.925709,  857.645648,
-  823.435143,  822.904770,  822.904770,  822.904770,  820.792975,  684.582020,
-  537.684760,  646.650947,  669.110131,  796.487512,  935.569890,  1036.777631,
-  1035.567378, 1030.219103, 1030.018584, 1023.810805, 972.158629,  851.968626,
-  823.347128,  822.904770,  822.904770,  822.904770,  820.792975,  684.582020,
-  552.408370,  670.001885,  738.246482,  879.690154,  992.939171,  1032.509436,
-  1031.636855, 1030.132153, 1029.665223, 1001.043724, 880.853721,  829.201546,
-  822.994150,  822.904770,  822.904770,  822.904770,  820.792975,  684.582020,
-  539.835902,  667.496388,  799.216004,  946.512211,  1039.506123, 1035.609680,
-  1030.219103, 1030.107964, 1029.577207, 995.366703,  858.086641,  823.524524,
-  822.906135,  822.904770,  822.904770,  822.904770,  820.792975,  684.582020,
-  558.362529,  734.277451,  877.197218,  990.478243,  1029.908393, 1028.993978,
-  1027.488620, 1027.464048, 1026.933674, 992.724534,  855.532488,  821.323349,
-  820.792975,  820.792975,  820.792975,  820.792975,  818.686600,  682.825198,
-  453.127195,  649.075095,  780.278390,  867.165890,  862.469711,  857.067460,
-  856.956321,  856.955937,  856.513579,  827.981461,  713.556496,  685.024378,
-  684.582020,  684.582020,  684.582020,  684.582020,  682.825198,  569.510056,
+static const uint8_t bsize_model_cat_lookup[BLOCK_SIZES_ALL] = {
+  0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 0, 0, 1, 1, 2, 2
 };
 
-static const double interp_dgrid_surf[65 * 18] = {
-  10.650434, 12.204694, 12.040917, 11.843008, 11.845578, 12.051535, 12.103583,
-  12.136780, 12.266709, 12.299107, 12.299673, 12.303120, 12.316337, 12.293431,
-  12.092165, 11.602421, 11.141559, 8.864495,  12.770003, 14.634889, 14.437149,
-  14.199413, 14.202487, 14.449423, 14.511827, 14.551629, 14.707410, 14.746265,
-  14.747610, 14.753705, 14.762194, 14.699395, 14.390525, 13.690970, 12.874168,
-  10.367121, 12.832328, 14.790730, 14.503765, 14.236403, 14.239028, 14.486600,
-  14.549164, 14.589069, 14.745250, 14.784258, 14.788320, 14.801930, 14.762798,
-  14.499088, 14.021544, 13.469684, 12.661560, 10.108384, 12.950520, 15.264726,
-  14.621957, 14.238236, 14.239028, 14.486601, 14.549264, 14.589469, 14.745361,
-  14.784949, 14.791572, 14.798652, 14.660251, 14.119394, 13.651131, 12.935657,
-  12.176082, 9.228999,  12.979992, 15.382918, 14.651428, 14.238693, 14.239028,
-  14.486701, 14.555710, 14.615321, 14.751849, 14.787700, 14.797104, 14.743189,
-  14.475057, 13.944406, 13.450468, 12.687876, 11.824993, 8.906683,  12.980449,
-  15.384750, 14.651885, 14.238700, 14.239028, 14.487102, 14.581562, 14.718998,
-  14.777721, 14.788445, 14.778661, 14.582790, 14.099785, 13.649637, 12.935359,
-  12.201859, 10.891931, 8.482221,  12.980449, 15.384750, 14.651886, 14.238801,
-  14.239434, 14.487303, 14.588010, 14.744860, 14.784773, 14.786094, 14.735647,
-  14.455704, 13.939591, 13.450393, 12.687876, 11.849334, 10.476658, 8.043672,
-  12.980449, 15.384750, 14.651987, 14.245320, 14.265579, 14.493824, 14.588211,
-  14.745312, 14.787263, 14.775934, 14.582036, 14.099475, 13.649563, 12.935358,
-  12.201859, 10.911285, 9.730570,  6.696921,  12.980449, 15.384750, 14.652393,
-  14.271466, 14.370434, 14.520069, 14.589027, 14.746028, 14.785482, 14.735605,
-  14.455693, 13.939590, 13.450393, 12.687876, 11.849334, 10.494514, 9.195398,
-  6.215460,  12.980449, 15.384750, 14.652494, 14.277985, 14.396679, 14.533035,
-  14.615021, 14.754825, 14.775610, 14.582796, 14.099664, 13.649565, 12.935358,
-  12.201859, 10.911285, 9.747361,  7.779960,  5.617541,  12.980448, 15.384731,
-  14.652415, 14.278078, 14.397578, 14.559053, 14.718657, 14.776398, 14.747044,
-  14.504690, 13.951810, 13.450583, 12.687876, 11.849334, 10.494514, 9.210817,
-  7.210003,  5.164575,  12.980446, 15.383448, 14.647073, 14.277541, 14.403813,
-  14.569546, 14.744956, 14.765103, 14.629073, 14.296161, 13.698573, 12.936118,
-  12.201859, 10.911285, 9.747361,  7.790897,  6.322998,  3.931551,  12.981550,
-  15.376916, 14.615597, 14.274820, 14.437479, 14.575942, 14.707492, 14.734111,
-  14.515975, 14.000806, 13.462803, 12.688066, 11.849334, 10.494514, 9.210817,
-  7.219566,  5.781392,  3.486081,  12.991899, 15.376201, 14.579444, 14.296898,
-  14.473361, 14.522910, 14.491600, 14.543267, 14.288580, 13.700311, 12.936579,
-  12.201867, 10.911285, 9.747361,  7.790897,  6.331506,  4.480348,  2.923138,
-  13.019848, 15.383477, 14.582260, 14.385262, 14.452673, 14.436019, 14.238174,
-  14.255993, 13.977481, 13.532342, 12.705591, 11.849605, 10.494514, 9.210817,
-  7.219566,  5.789642,  4.018194,  2.766222,  13.028558, 15.315782, 14.439141,
-  14.326286, 14.452429, 14.311731, 14.033235, 13.922587, 13.665868, 13.207897,
-  12.274375, 10.912967, 9.747371,  7.790897,  6.331506,  4.488594,  3.454993,
-  2.692682,  12.992752, 15.321471, 14.409573, 14.236340, 14.322969, 14.049072,
-  13.764823, 13.479242, 13.250105, 12.759133, 12.019174, 10.532951, 9.211409,
-  7.219566,  5.789642,  4.026440,  3.298077,  2.674624,  12.945493, 15.276596,
-  14.315745, 14.026198, 14.085774, 13.844563, 13.447576, 12.964935, 12.735525,
-  12.288592, 11.511693, 9.900227,  7.793270,  6.331506,  4.488594,  3.463236,
-  3.224318,  2.672433,  12.757570, 15.056661, 14.095011, 13.722362, 13.812624,
-  13.608480, 13.021206, 12.367627, 11.937931, 11.581049, 10.599552, 9.247860,
-  7.220151,  5.789642,  4.026437,  3.305882,  3.191260,  2.615317,  12.581293,
-  14.824658, 13.909074, 13.496158, 13.491402, 13.221550, 12.514140, 11.677229,
-  10.936895, 10.619912, 9.634779,  7.763570,  6.331082,  4.488590,  3.462798,
-  3.216460,  3.076315,  2.373499,  12.283499, 14.455760, 13.890593, 13.427587,
-  13.183783, 12.763833, 11.861006, 10.740618, 9.820756,  9.354945,  8.669862,
-  7.123268,  5.787860,  4.025994,  3.290000,  3.084410,  2.810905,  2.222916,
-  12.010893, 14.300919, 13.986624, 13.484026, 13.025385, 12.224281, 11.064265,
-  9.631040,  8.594396,  8.003736,  7.561587,  6.274418,  4.466637,  3.446574,
-  3.102467,  2.816989,  2.598688,  1.951541,  11.581477, 13.831132, 13.632027,
-  13.380414, 12.807880, 11.665651, 10.218236, 8.562237,  7.222614,  6.611808,
-  6.261676,  5.402793,  3.938544,  3.174375,  2.818166,  2.602758,  2.213911,
-  1.434763,  11.050735, 12.893449, 12.363152, 12.712829, 12.012961, 10.887854,
-  9.109699,  7.421701,  5.965603,  5.272129,  4.991435,  4.423000,  3.369988,
-  2.800371,  2.593901,  2.217431,  1.670917,  1.215265,  10.641194, 11.766277,
-  10.777082, 10.972917, 10.689298, 9.701545,  7.719947,  6.145654,  4.872442,
-  4.099600,  3.880934,  3.514159,  2.786474,  2.368963,  2.162376,  1.673670,
-  1.450770,  1.185424,  10.071964, 11.107701, 9.172361,  8.551313,  8.412080,
-  7.641397,  6.174246,  4.853916,  3.904549,  3.246810,  2.959903,  2.785066,
-  2.240001,  1.793166,  1.585520,  1.449824,  1.405368,  1.168856,  9.213182,
-  9.173278,  7.219231,  6.242951,  5.626013,  5.768007,  4.908666,  3.809589,
-  3.115109,  2.617899,  2.274793,  2.172960,  1.838597,  1.505915,  1.414333,
-  1.392666,  1.338173,  1.105611,  7.365015,  7.471370,  5.622346,  4.520127,
-  3.936272,  4.208822,  3.623024,  2.977794,  2.450003,  2.097261,  1.824090,
-  1.643270,  1.473525,  1.351388,  1.327504,  1.323865,  1.307894,  1.088234,
-  6.198210,  6.580712,  4.682511,  3.416952,  2.941929,  2.766637,  2.650686,
-  2.315439,  1.925838,  1.659784,  1.464419,  1.252806,  1.162722,  1.197518,
-  1.199875,  1.197365,  1.194040,  0.995797,  5.402507,  5.055466,  3.728724,
-  2.624359,  2.165810,  1.943189,  1.918190,  1.738078,  1.516328,  1.290520,
-  1.155793,  1.015962,  0.881900,  0.807203,  0.754242,  0.743378,  0.740288,
-  0.614158,  3.937867,  3.862507,  2.884664,  2.088147,  1.648496,  1.473584,
-  1.340123,  1.291769,  1.165381,  1.000224,  0.893316,  0.821333,  0.691363,
-  0.610501,  0.586766,  0.583762,  0.577840,  0.468733,  3.104660,  3.181078,
-  2.420208,  1.747442,  1.297956,  1.109835,  0.970385,  0.943229,  0.876923,
-  0.777584,  0.678183,  0.628623,  0.553745,  0.523430,  0.519490,  0.514394,
-  0.492259,  0.403172,  2.593833,  2.533720,  2.010452,  1.480944,  1.060302,
-  0.846383,  0.738703,  0.673144,  0.658010,  0.592449,  0.518236,  0.470335,
-  0.425088,  0.393168,  0.378116,  0.355846,  0.275469,  0.213128,  2.176988,
-  2.089575,  1.671284,  1.225008,  0.895382,  0.672008,  0.566241,  0.496746,
-  0.488005,  0.449874,  0.400899,  0.354002,  0.318150,  0.281533,  0.238545,
-  0.224159,  0.202399,  0.160681,  1.874679,  1.769165,  1.430124,  1.068727,
-  0.780272,  0.557801,  0.441643,  0.377256,  0.352957,  0.338452,  0.304965,
-  0.273172,  0.240052,  0.208724,  0.193431,  0.190845,  0.185025,  0.138166,
-  1.590226,  1.502830,  1.193127,  0.917885,  0.670432,  0.474546,  0.355420,
-  0.292305,  0.259035,  0.249937,  0.232079,  0.208943,  0.181936,  0.160038,
-  0.152257,  0.151235,  0.149583,  0.120747,  1.331730,  1.255907,  1.012871,
-  0.778422,  0.578977,  0.412432,  0.293155,  0.231824,  0.197187,  0.183921,
-  0.174876,  0.157252,  0.140263,  0.127050,  0.110244,  0.105041,  0.104323,
-  0.086944,  1.153994,  1.118771,  0.822355,  0.612321,  0.478249,  0.348222,
-  0.247408,  0.186141,  0.152714,  0.135445,  0.129810,  0.119994,  0.115619,
-  0.131626,  0.095612,  0.079343,  0.077502,  0.064550,  0.946317,  0.925894,
-  0.677969,  0.499906,  0.397101,  0.297931,  0.214467,  0.152333,  0.120731,
-  0.102686,  0.095062,  0.090361,  0.122319,  0.240194,  0.112687,  0.070690,
-  0.070461,  0.054194,  0.824155,  0.787241,  0.581856,  0.419228,  0.313167,
-  0.245582,  0.183500,  0.128101,  0.096577,  0.080267,  0.071022,  0.066851,
-  0.085754,  0.154163,  0.075884,  0.052401,  0.054270,  0.026656,  0.716310,
-  0.671378,  0.489580,  0.349569,  0.256155,  0.206343,  0.157853,  0.111950,
-  0.079271,  0.062518,  0.053441,  0.049660,  0.051400,  0.063778,  0.039993,
-  0.029133,  0.023382,  0.013725,  0.614125,  0.579096,  0.417126,  0.299465,
-  0.217849,  0.165515,  0.129040,  0.093127,  0.065612,  0.049543,  0.041429,
-  0.036850,  0.034416,  0.033989,  0.024216,  0.017377,  0.014833,  0.011987,
-  0.520407,  0.487239,  0.349473,  0.251741,  0.184897,  0.135813,  0.107098,
-  0.073607,  0.053938,  0.040531,  0.032931,  0.028876,  0.025759,  0.022168,
-  0.016739,  0.014638,  0.014333,  0.011947,  0.449954,  0.415124,  0.299452,
-  0.216942,  0.158874,  0.115334,  0.088821,  0.060105,  0.042610,  0.032566,
-  0.026903,  0.023123,  0.019913,  0.016835,  0.014306,  0.013625,  0.013535,
-  0.011284,  0.377618,  0.347773,  0.251741,  0.184839,  0.132857,  0.095439,
-  0.070462,  0.052244,  0.036078,  0.026025,  0.021518,  0.018487,  0.015361,
-  0.012905,  0.011470,  0.010569,  0.010283,  0.008297,  0.319953,  0.297976,
-  0.216942,  0.158842,  0.113280,  0.080426,  0.057367,  0.041987,  0.030135,
-  0.022295,  0.017901,  0.015121,  0.012224,  0.010035,  0.009353,  0.009108,
-  0.008695,  0.006139,  0.267864,  0.250502,  0.184839,  0.132851,  0.095039,
-  0.068220,  0.049135,  0.035315,  0.025144,  0.018237,  0.013857,  0.012094,
-  0.009715,  0.007743,  0.006937,  0.006446,  0.006243,  0.004929,  0.230449,
-  0.215895,  0.158842,  0.113280,  0.080417,  0.057174,  0.041304,  0.029959,
-  0.021866,  0.015673,  0.012133,  0.010083,  0.007801,  0.006053,  0.005401,
-  0.003834,  0.003429,  0.002851,  0.193984,  0.183963,  0.132851,  0.095039,
-  0.068220,  0.049133,  0.035305,  0.025140,  0.018150,  0.013175,  0.010422,
-  0.008491,  0.006397,  0.004567,  0.003494,  0.002933,  0.002825,  0.002355,
-  0.167298,  0.158088,  0.113280,  0.080417,  0.057174,  0.041304,  0.029959,
-  0.021866,  0.015669,  0.011955,  0.009257,  0.007051,  0.005543,  0.003905,
-  0.002984,  0.002825,  0.002814,  0.002347,  0.143228,  0.132220,  0.095039,
-  0.068220,  0.049133,  0.035305,  0.025140,  0.018150,  0.013174,  0.010394,
-  0.008403,  0.006661,  0.005378,  0.003545,  0.002876,  0.002818,  0.002814,
-  0.002347,  0.122934,  0.112735,  0.080417,  0.057174,  0.041304,  0.029959,
-  0.021866,  0.015669,  0.011955,  0.009258,  0.007182,  0.006012,  0.003762,
-  0.002866,  0.002739,  0.002788,  0.002810,  0.002347,  0.101934,  0.094569,
-  0.068220,  0.049133,  0.035305,  0.025140,  0.018150,  0.013174,  0.010394,
-  0.008405,  0.006797,  0.005845,  0.003333,  0.002703,  0.002695,  0.002723,
-  0.002781,  0.002343,  0.086702,  0.080014,  0.057174,  0.041304,  0.029959,
-  0.021866,  0.015669,  0.011955,  0.009258,  0.007190,  0.006533,  0.005839,
-  0.003326,  0.002700,  0.002690,  0.002694,  0.002716,  0.002314,  0.073040,
-  0.067886,  0.049133,  0.035305,  0.025140,  0.018150,  0.013174,  0.010394,
-  0.008405,  0.006807,  0.006468,  0.005831,  0.003325,  0.002700,  0.002690,
-  0.002690,  0.002687,  0.002253,  0.061685,  0.056890,  0.041304,  0.029959,
-  0.021866,  0.015669,  0.011955,  0.009258,  0.007190,  0.006542,  0.006360,
-  0.005416,  0.003221,  0.002698,  0.002690,  0.002690,  0.002683,  0.002238,
-  0.052465,  0.048894,  0.035305,  0.025140,  0.018150,  0.013174,  0.010394,
-  0.008405,  0.006807,  0.006472,  0.005943,  0.003748,  0.002805,  0.002692,
-  0.002690,  0.002690,  0.002683,  0.002238,  0.043838,  0.041101,  0.029959,
-  0.021866,  0.015669,  0.011955,  0.009258,  0.007190,  0.006543,  0.006465,
-  0.005839,  0.003333,  0.002702,  0.002690,  0.002690,  0.002690,  0.002683,
-  0.002238,  0.037824,  0.035133,  0.025140,  0.018150,  0.013174,  0.010394,
-  0.008405,  0.006807,  0.006480,  0.006464,  0.005838,  0.003326,  0.002700,
-  0.002690,  0.002690,  0.002690,  0.002683,  0.002238,  0.031865,  0.029815,
-  0.021866,  0.015668,  0.011955,  0.009258,  0.007190,  0.006543,  0.006475,
-  0.006462,  0.005831,  0.003325,  0.002700,  0.002690,  0.002690,  0.002690,
-  0.002683,  0.002238,  0.027150,  0.025016,  0.018128,  0.013083,  0.010371,
-  0.008405,  0.006807,  0.006480,  0.006472,  0.006359,  0.005416,  0.003221,
-  0.002698,  0.002690,  0.002690,  0.002690,  0.002683,  0.002238,  0.023094,
-  0.021760,  0.015577,  0.011590,  0.009167,  0.007188,  0.006543,  0.006475,
-  0.006466,  0.005943,  0.003748,  0.002805,  0.002692,  0.002690,  0.002690,
-  0.002690,  0.002683,  0.002238,  0.019269,  0.018038,  0.013060,  0.010280,
-  0.008382,  0.006806,  0.006480,  0.006474,  0.006464,  0.005839,  0.003333,
-  0.002702,  0.002690,  0.002690,  0.002690,  0.002690,  0.002683,  0.002238,
-  0.016874,  0.015472,  0.011566,  0.009148,  0.007171,  0.006527,  0.006458,
-  0.006457,  0.006447,  0.005823,  0.003318,  0.002693,  0.002683,  0.002683,
-  0.002683,  0.002683,  0.002676,  0.002232,  0.011968,  0.011056,  0.008762,
-  0.007219,  0.005717,  0.005391,  0.005386,  0.005386,  0.005377,  0.004856,
-  0.002767,  0.002246,  0.002238,  0.002238,  0.002238,  0.002238,  0.002232,
-  0.001862,
+static int sse_norm_model_cat_lookup(double sse_norm) {
+  return (sse_norm > 16.0);
+}
+
+static const double interp_rgrid_surf[4][33 * 18] = {
+  {
+      29.726102,   30.738006,   25.294088,   25.736759,   41.255961,
+      41.937828,   41.937901,   42.110487,   46.038554,   46.593284,
+      55.290144,   55.672257,   55.672289,   55.672284,   55.641633,
+      54.251766,   38.404522,   36.186584,   30.977201,   32.031695,
+      26.358675,   26.819956,   42.992269,   43.702888,   43.702966,
+      43.882830,   47.992288,   48.920311,   57.633255,   58.015369,
+      58.015396,   57.984745,   56.594876,   40.717227,   39.327227,
+      37.709563,   30.977308,   32.041872,   26.587891,   26.801416,
+      42.339251,   43.674340,   43.710377,   44.048333,   48.365747,
+      57.251292,   57.999483,   58.015595,   57.984977,   56.595076,
+      40.717426,   39.327559,   39.313319,   38.085580,   30.977328,
+      32.271000,   31.802769,   26.377378,   27.473270,   43.021187,
+      43.875643,   47.811034,   48.547182,   57.617338,   58.015563,
+      58.001989,   56.982277,   40.734438,   39.327560,   39.313487,
+      40.065111,   46.657660,   30.977329,   32.281067,   32.031899,
+      26.358835,   26.820108,   42.992490,   43.882906,   47.992438,
+      48.920477,   57.633450,   58.001989,   57.369478,   49.547092,
+      39.714761,   39.313520,   40.065295,   48.834535,   51.525122,
+      30.977329,   32.281072,   32.077511,   27.396520,   26.865699,
+      42.992474,   43.882906,   48.358483,   57.251290,   57.985906,
+      57.369511,   49.564105,   40.101962,   39.330532,   40.065297,
+      48.835069,   57.817665,   145.394655,  30.977369,   32.281197,
+      33.076074,   50.151877,   28.497451,   42.792931,   43.512994,
+      48.377350,   57.604551,   57.386489,   49.951306,   40.118976,
+      39.330567,   40.065298,   48.835069,   57.818203,   151.867755,
+      153.625733,  31.447893,   32.750906,   32.266221,   32.302627,
+      42.268290,   38.854272,   35.574619,   49.427510,   56.878041,
+      50.330896,   48.931678,   39.717770,   40.065331,   48.835069,
+      57.818191,   151.733328,  156.892368,  150.600520,  42.125514,
+      43.448292,   33.901056,   49.284440,   47.710751,   52.761179,
+      46.446220,   71.330907,   47.520016,   49.126258,   40.678046,
+      40.107037,   48.835077,   57.818191,   151.733068,  153.820869,
+      83.953859,   77.615621,   42.078145,   44.460837,   53.413056,
+      53.860750,   51.935802,   60.688133,   54.022549,   59.058236,
+      61.487243,   49.511225,   53.558701,   50.619860,   57.871196,
+      151.733230,  153.821045,  83.819613,   77.759808,   76.318016,
+      47.334708,   55.115021,   63.088351,   65.025437,   66.447005,
+      66.685816,   72.840860,   70.204772,   75.247159,   75.769300,
+      62.167236,   83.948064,   149.878947,  155.679346,  86.031186,
+      83.012511,   86.265353,   124.454706,  64.496779,   78.702984,
+      76.659370,   83.931191,   87.076287,   91.454837,   94.323380,
+      104.708654,  109.932924,  118.596035,  126.154595,  114.409068,
+      89.213630,   131.569970,  134.824613,  205.866155,  239.628810,
+      233.070258,  72.308997,   94.104934,   101.404120,  121.159064,
+      139.921349,  156.802796,  164.908506,  170.895200,  184.099064,
+      198.250167,  207.100063,  207.922921,  207.455763,  210.915553,
+      241.757520,  246.227265,  247.706317,  237.758893,  138.726610,
+      135.459290,  198.494473,  240.094885,  275.356590,  295.461068,
+      304.435273,  303.110796,  296.273053,  317.439936,  319.323966,
+      331.200883,  330.246858,  256.379818,  247.994095,  247.767952,
+      247.767210,  237.760453,  243.406633,  292.096487,  381.480937,
+      421.494757,  458.357268,  483.808573,  487.800297,  481.212962,
+      472.226349,  472.996935,  480.732289,  475.694812,  418.477476,
+      412.710307,  268.338558,  261.989349,  261.987913,  251.406810,
+      469.068582,  568.835648,  608.315967,  642.717253,  653.879813,
+      681.008793,  688.659203,  677.706251,  668.348079,  652.995530,
+      636.214467,  632.005555,  550.175654,  522.412011,  588.754096,
+      591.721361,  591.719586,  567.821360,  667.228892,  787.133995,
+      830.015474,  848.841400,  863.280871,  882.475988,  896.321308,
+      889.205975,  878.059084,  871.467549,  860.073920,  814.054753,
+      726.521001,  747.583633,  751.922326,  752.065525,  752.062970,
+      721.688835,  795.516527,  916.651129,  1045.662316, 1064.160855,
+      1070.930169, 1060.177543, 1080.056283, 1096.204211, 1095.337870,
+      1098.401980, 1087.336698, 1022.507849, 916.586633,  938.424837,
+      942.750167,  945.352203,  945.457385,  907.272493,  943.653980,
+      1021.248655, 1129.544563, 1267.035138, 1281.163505, 1235.783117,
+      1258.747831, 1285.425350, 1293.324247, 1314.802791, 1320.786990,
+      1253.910457, 1138.645930, 1091.248129, 1161.397763, 1220.617170,
+      1223.081462, 1173.684183, 1077.194216, 1216.741493, 1278.148407,
+      1383.489545, 1497.622061, 1505.771495, 1422.262568, 1462.870558,
+      1491.544736, 1494.922359, 1533.495237, 1509.573542, 1356.235452,
+      1322.878872, 1324.869188, 1327.897500, 1337.708123, 1284.090364,
+      1177.946473, 1335.104679, 1414.612819, 1508.772208, 1656.204257,
+      1636.726625, 1693.515385, 1658.176544, 1626.718453, 1699.881842,
+      1721.127481, 1715.540667, 1610.027798, 1528.475428, 1525.090509,
+      1534.987574, 1760.238672, 1698.643936, 1285.821939, 1417.714456,
+      1527.788112, 1660.663143, 1648.660011, 1838.036905, 1838.669356,
+      1803.685279, 1895.404145, 1878.456820, 1907.761437, 1891.806178,
+      1798.474576, 1772.695413, 1777.922135, 1782.961020, 1891.441373,
+      1819.624352, 1363.652278, 1525.975687, 1664.733346, 1725.645678,
+      1842.181220, 1845.935374, 1847.645643, 2084.649919, 2156.327082,
+      2070.132265, 2093.920913, 2066.856209, 1935.233655, 1757.519179,
+      1894.998543, 1904.433642, 1909.668939, 1832.757449, 1466.924321,
+      1622.799826, 1723.952077, 1846.075992, 1857.802996, 1849.640165,
+      2127.098845, 2370.422248, 2375.277013, 2471.317955, 2282.670767,
+      2242.894713, 2110.737592, 1924.405669, 2022.847422, 2092.614827,
+      2113.100659, 2028.502155, 1561.024406, 1719.325091, 1845.872940,
+      1857.825889, 1850.155079, 2127.184115, 2372.221523, 2397.560183,
+      2793.802030, 2525.955468, 2512.515727, 2465.117634, 2335.991284,
+      2104.381407, 2124.161214, 2128.328755, 2129.221941, 2043.260174,
+      1651.007418, 1782.604815, 1855.046326, 1850.154844, 2127.184127,
+      2372.221529, 2397.560385, 2794.384764, 2544.781302, 2651.443272,
+      2739.000803, 2702.474497, 2522.919941, 2276.563499, 2274.919319,
+      2275.308573, 2275.300889, 2183.406597, 1713.105119, 1852.260885,
+      1850.032717, 2127.168115, 2371.857312, 2397.544382, 2794.384763,
+      2544.781370, 2651.673334, 2746.255820, 2763.494638, 2872.508140,
+      2846.367495, 2305.137739, 2281.712444, 2281.725436, 2281.717621,
+      2189.564169, 1780.286046, 1850.026376, 2127.168073, 2371.493095,
+      2389.254845, 2794.020546, 2544.781339, 2651.673334, 2746.255845,
+      2763.564597, 2874.688737, 2884.399707, 2860.263639, 2306.077266,
+      2281.728045, 2281.725981, 2281.718165, 2189.564691, 1775.067091,
+      2127.161715, 2371.493094, 2389.238842, 2793.656329, 2544.765337,
+      2651.673332, 2746.255845, 2763.564597, 2874.688743, 2884.401957,
+      2884.614099, 2860.264919, 2306.079244, 2281.728130, 2281.725981,
+      2281.718165, 2189.564691, 2052.948444, 2371.486799, 2389.238842,
+      2793.656328, 2544.765306, 2651.673331, 2746.255845, 2763.564597,
+      2874.688743, 2884.401957, 2884.616164, 2884.614218, 2861.248329,
+      2328.461556, 2282.711540, 2281.726065, 2281.718165, 2189.564691,
+      2285.514142, 2389.231550, 2793.656328, 2544.765306, 2651.673331,
+      2746.255845, 2763.564597, 2874.688743, 2884.401957, 2884.616164,
+      2884.616200, 2884.616117, 2883.630641, 2837.880625, 2305.093852,
+      2281.727963, 2281.718165, 2189.564691, 2293.011609, 2793.639080,
+      2544.755886, 2651.663879, 2746.245982, 2763.554715, 2874.678861,
+      2884.392075, 2884.606282, 2884.606318, 2884.606318, 2884.606318,
+      2884.604170, 2860.253139, 2306.069362, 2281.720230, 2281.710349,
+      2189.557190, 2600.476153, 2437.607663, 2540.606072, 2634.813199,
+      2647.265343, 2758.176106, 2767.889302, 2768.103509, 2768.103545,
+      2768.103545, 2768.103545, 2768.103545, 2768.101564, 2744.735758,
+      2212.932478, 2189.566673, 2189.557190, 2101.125891,
+  },
+  {
+      8.370360,    8.509764,    8.942497,    8.962541,    8.993650,
+      9.401615,    9.414854,    9.308823,    9.304164,    9.304163,
+      9.304164,    9.304609,    9.324803,    9.555049,    9.565277,
+      9.571194,    9.647370,    9.451428,    8.722648,    8.867100,
+      9.300225,    9.352116,    9.672213,    9.809754,    9.794381,
+      9.699874,    9.695753,    9.695753,    9.696199,    9.716392,
+      9.947083,    9.967407,    9.973767,    10.050124,   10.454503,
+      14.384681,   8.722677,    8.848489,    8.875968,    9.334087,
+      9.685450,    9.793645,    9.413799,    9.683184,    9.695785,
+      9.696231,    9.716425,    9.947116,    9.967440,    9.973802,
+      10.050158,   10.454591,   15.023561,   15.153937,   8.723107,
+      8.857467,    8.857756,    9.333268,    9.685451,    9.792911,
+      9.397076,    9.682449,    9.695786,    9.706329,    9.946672,
+      9.967440,    9.973802,    10.050158,   10.454591,   15.023647,
+      16.163497,   23.616388,   8.732904,    9.080445,    8.867552,
+      9.333269,    9.685451,    9.792910,    9.397075,    9.682449,
+      9.696231,    9.716869,    9.957343,    9.973801,    10.050158,
+      10.454591,   15.023647,   16.163589,   24.700350,   25.680478,
+      8.733342,    9.090251,    8.868006,    9.333704,    9.685477,
+      9.792913,    9.397086,    9.682895,    9.716424,    9.947246,
+      9.973357,    10.050158,   10.454591,   15.023647,   16.163589,
+      24.700554,   28.087109,   55.858717,   8.827122,    9.186580,
+      8.932093,    9.387586,    9.780396,    9.827339,    9.522726,
+      9.715771,    9.936594,    9.972878,    10.050158,   10.454591,
+      15.023647,   16.163589,   24.700554,   28.087318,   58.323463,
+      58.504090,   11.020305,   11.418729,   10.432068,   10.462985,
+      11.980204,   10.705178,   12.392570,   10.364495,   9.829351,
+      10.102947,   10.457228,   15.023647,   16.163589,   24.700554,
+      28.087319,   58.323704,   60.984846,   58.926665,   12.554530,
+      12.511580,   11.828923,   12.308025,   13.298760,   13.044522,
+      12.945689,   13.556029,   12.673484,   12.067338,   15.194674,
+      16.168241,   24.700556,   28.087321,   58.339708,   61.349323,
+      61.787659,   67.232638,   15.008518,   15.485683,   15.872869,
+      16.107583,   15.835820,   16.877645,   17.678106,   18.822352,
+      20.208378,   19.665073,   19.090588,   25.203257,   28.104179,
+      58.339741,   61.713565,   70.078019,   70.458033,   67.961299,
+      17.761908,   35.462897,   20.615440,   19.929266,   20.679593,
+      22.692816,   25.941528,   29.215386,   34.956494,   38.908872,
+      35.866805,   37.824252,   59.112479,   61.729892,   70.173287,
+      72.702736,   72.797652,   69.857555,   20.506234,   39.710259,
+      24.054829,   25.703729,   31.621917,   38.509103,   44.229053,
+      56.544042,   67.468427,   79.581678,   80.144809,   78.556111,
+      74.399792,   73.852202,   76.027762,   117.654284,  119.482115,
+      114.656650,  25.548773,   48.434418,   42.617322,   59.026359,
+      82.885084,   103.416136,  110.596012,  118.893312,  139.741780,
+      150.511338,  161.901080,  167.647532,  162.429576,  152.184833,
+      152.445359,  168.398896,  169.099625,  162.269514,  82.673806,
+      70.110218,   132.312464,  183.332020,  215.910819,  237.790158,
+      239.207817,  229.469490,  230.174506,  250.717349,  253.902836,
+      250.549622,  207.721430,  175.337781,  174.022082,  174.844559,
+      179.095719,  164.620978,  109.303621,  158.273825,  282.689646,
+      346.940666,  386.914531,  410.989383,  397.460378,  370.331321,
+      364.107004,  366.410030,  380.333542,  363.501310,  294.254304,
+      262.048447,  260.790777,  265.975674,  362.177173,  183.336338,
+      404.520239,  473.542827,  507.606249,  529.861510,  559.718099,
+      588.632991,  577.691379,  533.799138,  522.780262,  528.340923,
+      525.905829,  507.879853,  415.994095,  425.708669,  426.932731,
+      441.248848,  446.721366,  435.160437,  604.741350,  709.039378,
+      780.416591,  732.829427,  733.032326,  766.737728,  758.413295,
+      720.856125,  705.746779,  717.211527,  703.753940,  661.340782,
+      552.231698,  523.727433,  522.707356,  523.327476,  523.380946,
+      502.845082,  745.431184,  827.490462,  962.723712,  998.756854,
+      911.671845,  907.248783,  925.966430,  901.879365,  900.337680,
+      919.439892,  914.334537,  842.878262,  704.639674,  702.883177,
+      703.067185,  703.067254,  703.064850,  674.669690,  894.169844,
+      1025.903412, 1134.439771, 1219.193918, 1158.526825, 1079.781195,
+      1077.325043, 1076.106019, 1075.934290, 1117.511500, 1135.469693,
+      1063.699731, 901.900714,  920.553711,  921.681648,  921.681860,
+      921.679125,  884.454592,  994.327035,  1175.512444, 1231.834682,
+      1307.830164, 1407.421778, 1326.963114, 1253.371428, 1250.787362,
+      1269.653825, 1282.857814, 1343.429863, 1307.501250, 1114.048857,
+      983.551449,  978.228429,  979.601009,  984.578211,  945.021162,
+      1135.950879, 1278.106415, 1310.418579, 1419.457570, 1488.462247,
+      1521.009066, 1542.754625, 1434.182316, 1454.788272, 1471.133321,
+      1506.092266, 1521.256433, 1327.843116, 1309.574134, 1310.302024,
+      1341.725430, 1459.021892, 1404.992306, 1232.172010, 1349.471091,
+      1428.291402, 1487.206606, 1535.398237, 1746.731849, 1716.656287,
+      1635.940921, 1639.303784, 1673.047286, 1693.682270, 1671.177816,
+      1517.793233, 1479.700025, 1478.369392, 1483.694229, 1578.610159,
+      1518.853600, 1297.751330, 1428.849626, 1485.708645, 1493.412575,
+      1741.940596, 1728.475751, 1735.779511, 1858.578642, 1918.863689,
+      1863.724239, 1891.004802, 1842.815568, 1675.715043, 1588.071911,
+      1589.814363, 1592.306606, 1596.345158, 1532.039001, 1373.297412,
+      1461.343441, 1492.261194, 1736.797392, 1653.192795, 1733.476694,
+      1888.263820, 2082.789610, 1991.468137, 1883.805569, 2116.637406,
+      2071.956368, 1871.698996, 1644.776129, 1756.448731, 1809.269770,
+      1811.354640, 1738.198277, 1403.551810, 1491.185946, 1736.750346,
+      1653.047461, 1730.172539, 1888.230584, 2085.624280, 1993.438043,
+      1786.471183, 2337.838134, 2448.464613, 2336.935884, 2097.581279,
+      1808.183187, 1828.142934, 1821.170551, 1820.795051, 1747.257197,
+      1427.896429, 1736.558511, 1653.047441, 1730.172527, 1888.230448,
+      2087.322730, 2032.094869, 1787.954651, 2341.066895, 2641.531977,
+      2539.340235, 2507.302262, 2309.116599, 2030.685235, 2078.550644,
+      1842.720183, 1832.216691, 1758.216718, 1580.198506, 1648.792285,
+      1730.172167, 1888.230448, 2087.322736, 2032.169494, 1789.653110,
+      2341.141530, 2641.873005, 2555.239313, 2709.221434, 2715.218691,
+      2682.937371, 2124.893499, 2102.967545, 2092.601766, 2092.133383,
+      2007.636776, 1578.033231, 1729.979789, 1888.230432, 2087.322736,
+      2032.169494, 1789.653117, 2341.141675, 2641.873012, 2555.239371,
+      2709.566764, 2723.749115, 2724.064349, 2699.003101, 2128.626010,
+      2103.563660, 2103.560569, 2103.553324, 2018.595528, 1663.710423,
+      1888.224796, 2087.322736, 2032.169494, 1789.653117, 2341.141675,
+      2641.873012, 2555.239371, 2709.566764, 2723.749145, 2724.067198,
+      2724.065212, 2700.016608, 2151.662616, 2104.575854, 2103.561585,
+      2103.554293, 2018.596458, 1818.473656, 2087.316820, 2032.169494,
+      1789.653117, 2341.141675, 2641.873012, 2555.239371, 2709.566764,
+      2723.749145, 2724.067198, 2724.067252, 2724.067166, 2723.052897,
+      2675.966135, 2127.612143, 2103.563538, 2103.554293, 2018.596458,
+      2011.142209, 2032.163033, 1789.653117, 2341.141675, 2641.873012,
+      2555.239371, 2709.566764, 2723.749145, 2724.067198, 2724.067252,
+      2724.067252, 2724.067252, 2724.065041, 2699.002510, 2128.626241,
+      2103.563710, 2103.554293, 2018.596458, 1947.414707, 1789.636821,
+      2341.132961, 2641.863707, 2555.230039, 2709.557432, 2723.739813,
+      2724.057867, 2724.057920, 2724.057920, 2724.057920, 2724.057920,
+      2724.055881, 2700.007276, 2151.653370, 2104.568562, 2103.547173,
+      2018.589543, 1614.950117, 2233.740132, 2539.133260, 2445.528333,
+      2599.538876, 2613.721229, 2614.039283, 2614.039337, 2614.039337,
+      2614.039337, 2614.039337, 2614.039337, 2614.039251, 2613.025067,
+      2566.950536, 2041.634786, 2018.591496, 1937.063243,
+  },
+  {
+      2.147514,    2.237898,    2.237803,    2.237151,    2.275320,
+      2.279324,    2.332179,    2.334501,    2.322470,    2.048664,
+      2.036634,    2.036633,    2.043731,    2.205267,    2.213955,
+      2.286115,    3.114672,    3.150236,    2.237900,    2.332085,
+      2.329658,    2.278309,    2.368754,    2.375255,    2.430335,
+      2.432248,    2.408693,    2.134381,    2.122351,    2.122650,
+      2.136546,    2.299971,    2.379228,    3.209393,    3.546162,
+      6.294600,    2.271800,    2.333582,    2.329564,    2.275984,
+      2.368873,    2.380129,    2.430556,    2.420732,    2.146419,
+      2.122864,    2.122657,    2.136254,    2.293180,    2.378937,
+      3.209400,    3.546195,    6.591552,    7.028860,    3.043184,
+      2.367477,    2.329567,    2.275984,    2.373739,    2.490860,
+      2.435421,    2.420225,    2.134895,    2.122658,    2.136254,
+      2.293180,    2.378936,    3.209400,    3.546195,    6.591617,
+      7.795421,    17.743823,   3.078566,    2.402859,    2.331056,
+      2.275985,    2.373953,    2.495725,    2.435129,    2.408702,
+      2.134688,    2.136255,    2.293180,    2.378936,    3.209400,
+      3.546195,    6.591617,    7.795492,    18.578087,   19.754344,
+      3.112459,    3.174471,    2.370323,    2.279235,    2.375398,
+      2.494530,    2.423939,    2.146738,    2.136763,    2.293180,
+      2.378939,    3.209400,    3.546195,    6.591617,    7.795492,
+      18.578249,   21.659892,   44.204400,   3.089303,    3.189642,
+      2.509317,    2.372448,    2.432176,    2.489915,    2.461894,
+      2.170933,    2.301935,    2.380378,    3.234280,    3.547288,
+      6.591617,    7.795492,    18.578249,   21.660057,   46.157016,
+      46.346972,   2.545327,    2.662725,    2.893299,    2.840660,
+      3.006570,    3.048065,    3.224317,    2.907751,    2.627415,
+      3.268959,    4.113541,    6.616496,    7.795494,    18.578249,
+      21.660057,   46.157182,   48.299592,   46.394222,   2.942757,
+      3.155386,    3.688593,    3.481851,    3.715109,    4.145502,
+      4.681446,    5.080585,    4.600294,    4.278096,    6.726591,
+      7.800142,    18.578250,   21.660057,   46.157182,   48.299758,
+      48.346843,   46.394230,   3.793838,    5.473983,    4.608110,
+      4.665834,    4.593326,    5.621563,    7.102711,    9.025169,
+      10.122234,   10.159723,   10.023976,   18.857332,   21.668316,
+      46.157261,   48.299839,   48.347091,   48.346956,   46.394491,
+      4.271176,    6.466779,    5.484235,    6.027476,    7.301091,
+      9.352794,    12.715546,   16.192475,   22.560070,   27.958500,
+      27.052873,   26.766614,   46.406220,   49.221059,   49.307058,
+      49.315280,   49.587875,   49.469333,   4.312908,    7.475387,
+      7.833860,    10.223714,   16.775457,   24.234924,   32.175390,
+      45.584094,   54.342692,   64.941124,   66.654416,   62.626466,
+      52.266425,   72.572219,   73.485826,   73.672773,   79.880600,
+      119.538616,  9.063034,    10.103318,   19.539238,   36.472083,
+      62.703355,   82.709235,   91.988082,   100.285999,  115.687111,
+      124.329614,  125.634719,  122.315209,  112.379054,  130.322056,
+      131.130875,  131.139174,  131.411947,  127.988734,  19.796341,
+      25.882925,   93.253237,   153.548489,  189.401787,  213.347763,
+      206.988813,  189.758327,  189.286642,  204.628607,  202.531100,
+      201.472079,  186.352604,  214.273591,  215.531911,  215.769339,
+      221.170008,  212.465343,  169.309448,  175.824536,  224.720662,
+      303.438505,  343.063029,  370.535703,  343.624360,  300.549614,
+      289.296503,  298.082336,  310.854450,  294.466802,  270.163254,
+      249.832086,  248.983097,  254.384429,  377.318513,  367.262701,
+      353.359003,  463.883442,  447.419222,  453.002989,  489.006379,
+      528.921013,  500.655371,  439.860852,  423.703334,  423.144668,
+      431.190760,  422.628418,  346.619232,  383.306337,  385.067653,
+      385.305123,  390.705188,  375.153218,  539.087109,  664.575466,
+      708.656277,  662.482491,  618.723324,  674.195070,  656.531281,
+      596.084694,  576.463478,  584.628279,  577.147684,  562.515800,
+      453.416064,  428.149599,  427.247359,  427.247313,  427.246307,
+      409.990807,  687.292802,  777.058828,  905.030245,  912.381642,
+      788.119446,  769.561998,  795.174180,  761.722895,  738.637994,
+      753.718308,  761.364923,  712.753474,  603.465105,  591.662313,
+      594.079856,  594.200005,  594.197958,  570.199152,  779.060173,
+      951.796204,  1069.175314, 1170.755086, 1067.610429, 949.009502,
+      934.793193,  914.719926,  899.055222,  929.548018,  948.443491,
+      883.741844,  754.604267,  719.256051,  780.076337,  782.811403,
+      782.547500,  745.230787,  922.165691,  1108.252645, 1185.274121,
+      1237.276156, 1266.012843, 1226.812751, 1102.465568, 1085.286013,
+      1075.433883, 1091.747391, 1130.914031, 1079.716598, 927.463157,
+      942.209973,  945.879069,  946.263934,  946.332394,  778.375752,
+      1069.516638, 1181.232895, 1244.763295, 1280.392765, 1419.980489,
+      1443.722749, 1374.005679, 1247.118477, 1252.786860, 1265.078694,
+      1280.267151, 1273.804486, 1085.050290, 1037.883124, 1036.171983,
+      1042.457032, 1185.241611, 1137.692305, 1136.326843, 1246.301355,
+      1321.499026, 1422.581245, 1459.231744, 1574.624247, 1593.556714,
+      1554.290382, 1474.539686, 1480.100956, 1464.679108, 1412.507057,
+      1251.278472, 1197.245685, 1195.175062, 1201.168704, 1337.571298,
+      1289.300549, 1198.629451, 1322.569335, 1424.892277, 1470.063106,
+      1576.317080, 1629.881049, 1759.007438, 1685.455960, 1823.183401,
+      1696.763728, 1720.897219, 1582.701816, 1409.816846, 1352.692221,
+      1350.761108, 1356.950874, 1362.935844, 1308.131482, 1273.245632,
+      1447.544983, 1471.079493, 1576.789596, 1629.891022, 1759.113129,
+      1704.167080, 2077.760859, 1944.393655, 2008.282739, 1951.682502,
+      1905.896173, 1653.621076, 1524.210363, 1526.775296, 1662.009621,
+      1667.942643, 1600.578688, 1394.223453, 1472.070445, 1576.833336,
+      1629.890970, 1757.978972, 1678.353822, 2077.085726, 1963.548056,
+      2214.617152, 2129.636312, 2170.262437, 2204.652056, 1892.295487,
+      1652.520236, 1673.779173, 1681.078895, 1681.334159, 1613.428869,
+      1413.634564, 1576.837954, 1629.890974, 1757.978960, 1678.303800,
+      2075.951574, 1963.498706, 2214.984962, 2140.104142, 2224.748512,
+      2358.489855, 2338.333779, 2147.326062, 1959.161343, 1704.142189,
+      1693.264072, 1693.257371, 1624.870507, 1522.455341, 1630.102341,
+      1757.978970, 1678.205343, 2073.710713, 1963.613482, 2219.840292,
+      2140.317509, 2224.830630, 2367.413225, 2514.706448, 2519.715697,
+      2490.578281, 1997.672226, 1965.092163, 1964.608831, 1964.602061,
+      1885.256194, 1566.312385, 1757.983049, 1678.205154, 2071.469838,
+      1912.611809, 2222.454760, 2250.824415, 2229.685973, 2367.413669,
+      2515.007972, 2527.205354, 2527.466626, 2505.215096, 1998.786468,
+      1976.532889, 1976.530885, 1976.524114, 1896.696743, 1692.302152,
+      1678.199584, 2071.469829, 1912.513351, 2220.213903, 2250.939287,
+      2234.541308, 2367.626998, 2515.007990, 2527.205379, 2527.469148,
+      2527.467382, 2506.115004, 2019.240121, 1977.432526, 1976.531972,
+      1976.525125, 1896.697713, 1606.783592, 2071.463772, 1912.513351,
+      2220.213894, 2250.939097, 2234.541317, 2367.627410, 2515.008008,
+      2527.205379, 2527.469148, 2527.469193, 2527.469116, 2526.568562,
+      2484.760968, 1997.886085, 1976.533707, 1976.525125, 1896.697713,
+      2004.589287, 1912.507679, 2220.213894, 2250.939097, 2234.541317,
+      2367.627410, 2515.008008, 2527.205379, 2527.469148, 2527.469193,
+      2527.469193, 2527.469193, 2527.467229, 2505.214602, 1998.786486,
+      1976.533860, 1976.525125, 1896.697713, 1827.779426, 2220.199076,
+      2250.930988, 2234.532682, 2367.618752, 2514.999350, 2527.196721,
+      2527.460490, 2527.460534, 2527.460534, 2527.460534, 2527.460534,
+      2527.458723, 2506.106346, 2019.231539, 1977.425679, 1976.518431,
+      1896.691216, 2056.541124, 2160.944440, 2138.934044, 2265.803052,
+      2412.920916, 2425.118264, 2425.382033, 2425.382078, 2425.382078,
+      2425.382078, 2425.382078, 2425.382078, 2425.382002, 2424.481524,
+      2383.572672, 1917.153083, 1896.692951, 1820.088116,
+  },
+  {
+      0.254129,    0.265575,    0.267769,    0.316933,    0.320396,
+      0.350337,    0.358368,    0.358663,    0.358663,    0.358675,
+      0.359199,    0.365181,    0.366036,    0.392895,    0.701286,
+      0.812014,    1.926722,    2.011743,    0.247749,    0.276002,
+      0.279039,    0.330191,    0.332015,    0.365000,    0.373450,
+      0.373758,    0.373759,    0.374032,    0.380276,    0.381405,
+      0.408276,    0.717257,    0.841415,    1.958868,    2.298681,
+      4.856667,    0.246998,    0.275970,    0.279040,    0.328326,
+      0.289563,    0.363136,    0.373451,    0.373760,    0.373771,
+      0.374308,    0.381144,    0.408277,    0.717258,    0.841416,
+      1.958870,    2.298703,    5.088461,    5.484975,    0.246965,
+      0.275218,    0.279007,    0.328244,    0.287698,    0.363054,
+      0.373451,    0.373771,    0.374296,    0.380881,    0.408266,
+      0.717258,    0.841416,    1.958870,    2.298703,    5.088514,
+      6.102534,    14.287379,   0.246213,    0.258111,    0.278255,
+      0.328244,    0.287698,    0.363054,    0.373452,    0.374034,
+      0.380870,    0.408644,    0.725849,    0.841794,    1.958870,
+      2.298703,    5.088514,    6.102591,    14.959455,   15.913957,
+      0.246180,    0.257402,    0.279186,    0.328526,    0.290138,
+      0.364877,    0.376133,    0.375014,    0.408382,    0.734439,
+      1.037308,    1.967461,    2.298704,    5.088514,    6.102591,
+      14.959584,   17.441072,   35.429097,   0.250019,    0.262303,
+      0.303891,    0.340011,    0.350106,    0.413735,    0.446832,
+      0.410301,    0.734390,    1.038841,    2.002354,    2.300237,
+      5.088514,    6.102591,    14.959584,   17.441204,   36.992539,
+      37.110962,   0.339750,    0.354696,    0.373261,    0.461008,
+      0.462495,    0.598987,    0.695421,    0.760530,    1.068705,
+      2.029958,    2.898902,    5.114817,    6.102593,    14.959584,
+      17.441205,   36.992675,   38.648954,   36.568097,   0.400795,
+      0.421342,    0.508307,    0.588014,    0.677109,    0.972744,
+      1.364551,    2.005727,    2.968255,    3.091537,    5.221096,
+      6.106993,    14.959585,   17.441202,   36.995441,   38.712050,
+      38.170575,   37.947398,   0.489106,    0.499732,    0.682198,
+      0.797985,    0.928841,    1.529889,    3.407135,    6.047262,
+      9.020716,    9.566838,    8.225222,    15.250390,   17.449015,
+      36.967227,   38.773771,   39.603706,   39.612863,   38.072105,
+      0.782539,    0.400752,    1.185871,    1.379942,    2.131362,
+      4.169824,    9.556239,    12.949414,   21.418888,   24.451152,
+      23.409241,   22.939144,   37.166430,   38.131684,   39.591435,
+      39.985395,   39.999803,   38.384307,   1.616514,    0.576268,
+      3.109295,    3.930083,    8.248113,    15.543679,   28.108995,
+      37.139521,   46.111565,   52.306471,   54.245700,   51.814241,
+      40.882242,   41.810659,   42.231519,   48.976557,   49.261637,
+      47.271630,   5.094715,    1.500966,    10.566421,   19.002838,
+      38.783971,   59.976285,   70.642488,   79.940911,   88.863960,
+      91.805239,   95.624856,   92.381504,   94.587340,   93.842585,
+      93.403266,   93.670318,   93.432487,   89.648397,   0.664794,
+      7.849196,    61.154449,   96.443290,   137.609372,  162.556446,
+      148.037064,  138.292519,  136.715840,  145.772994,  149.246615,
+      151.612049,  131.470885,  109.201663,  101.316335,  101.904288,
+      101.932270,  97.814994,   6.925953,    50.062367,   155.094765,
+      216.414869,  252.438993,  271.718839,  232.337598,  204.289605,
+      199.194140,  205.331210,  215.677297,  214.590331,  188.109722,
+      172.081271,  228.048207,  250.892300,  251.787562,  241.618501,
+      134.674271,  296.106246,  298.718911,  312.356451,  323.413855,
+      342.631007,  307.749299,  283.067516,  265.987870,  265.460375,
+      276.156298,  267.602576,  244.127989,  255.341106,  257.373708,
+      258.333723,  258.372207,  247.937137,  342.349647,  510.603738,
+      581.005531,  478.386434,  373.143790,  378.600404,  377.561939,
+      356.141074,  338.631477,  331.842551,  340.566852,  327.567230,
+      297.260507,  287.191935,  263.781497,  262.767950,  262.766967,
+      252.154399,  475.742336,  629.567678,  578.444958,  564.378757,
+      438.113186,  396.420987,  414.959234,  431.302969,  421.606208,
+      403.786618,  405.430259,  389.856615,  363.613524,  366.303795,
+      365.460005,  365.415394,  365.414130,  350.655776,  624.537781,
+      642.232055,  604.750655,  630.663539,  588.111844,  475.625654,
+      468.814426,  477.664040,  494.124835,  483.676981,  477.873259,
+      452.815795,  423.548326,  431.959314,  432.336669,  431.259561,
+      431.158392,  412.599201,  599.136090,  596.383540,  649.766837,
+      714.641904,  647.024168,  546.886099,  544.047374,  542.897551,
+      558.506982,  555.978025,  571.819113,  532.351774,  510.677525,
+      495.125872,  493.402635,  468.981876,  468.871284,  423.950331,
+      572.225475,  705.639818,  1024.735591, 667.143308,  595.280675,
+      762.720257,  552.209148,  613.853858,  630.029121,  653.039002,
+      643.569732,  623.030443,  573.888949,  571.441956,  571.380435,
+      572.467453,  621.621663,  597.446835,  681.815611,  1037.097688,
+      905.438670,  599.972561,  762.219180,  517.929098,  688.053947,
+      744.181205,  679.524546,  682.569374,  713.780286,  707.985178,
+      646.983619,  636.062414,  635.699032,  635.794133,  637.956277,
+      612.281811,  1009.018046, 905.871087,  604.028246,  629.564692,
+      511.991374,  689.885845,  836.027416,  766.534552,  881.860194,
+      735.849782,  783.725360,  809.653496,  750.636979,  811.765460,
+      814.639525,  816.244480,  818.253804,  785.288396,  867.862571,
+      604.231326,  629.308626,  506.143862,  689.628928,  836.193031,
+      774.742632,  1007.244245, 1293.627165, 897.984304,  931.611215,
+      973.236387,  945.068518,  947.276414,  950.892263,  987.500595,
+      1033.288662, 993.423200,  674.214249,  634.009759,  506.144239,
+      689.628432,  836.193010,  774.742663,  1007.434892, 1306.991055,
+      1109.422389, 1067.304686, 1126.861498, 1124.180515, 989.502672,
+      997.732672,  1042.608506, 1046.158446, 1048.166963, 1005.915872,
+      633.151925,  507.355996,  689.628570,  836.193810,  774.742698,
+      1007.434892, 1306.991101, 1109.775470, 1079.300073, 1239.065289,
+      1628.963638, 1322.149598, 1152.072478, 1133.642807, 1135.074392,
+      1135.159950, 1135.156239, 1089.309834, 1002.301724, 712.557708,
+      836.610527,  784.181750,  1007.849615, 1306.991136, 1109.775470,
+      1079.300117, 1239.231971, 1644.540785, 1609.385289, 1575.587962,
+      1175.113940, 1139.762609, 1138.978460, 1138.978401, 1138.974499,
+      1092.973876, 693.249751,  850.650934,  794.193638,  1222.681498,
+      1316.430188, 1109.776271, 1079.300117, 1239.231971, 1644.540839,
+      1609.866572, 1604.515037, 1603.618487, 1568.302641, 1157.017568,
+      1138.980254, 1138.978724, 1138.974823, 1092.974187, 834.755585,
+      1090.883669, 1236.131974, 1325.870346, 1110.190994, 1079.300152,
+      1239.231971, 1644.540839, 1609.866572, 1604.515142, 1604.380213,
+      1604.378532, 1585.580675, 1157.776710, 1138.980319, 1138.978724,
+      1138.974823, 1092.974187, 1056.722338, 1249.164781, 1326.443132,
+      1110.191843, 1079.300187, 1239.231971, 1644.540839, 1609.866572,
+      1604.515142, 1604.380213, 1604.380190, 1604.378596, 1585.582205,
+      1157.778240, 1138.980383, 1138.978724, 1138.974823, 1092.974187,
+      1204.957557, 1326.440488, 1110.191892, 1079.300187, 1239.231971,
+      1644.540839, 1609.866572, 1604.515142, 1604.380213, 1604.380190,
+      1604.380190, 1604.378661, 1586.341349, 1175.056279, 1139.739527,
+      1138.978789, 1138.974823, 1092.974187, 1275.872102, 1110.183396,
+      1079.294485, 1239.226464, 1644.535343, 1609.861076, 1604.509646,
+      1604.374717, 1604.374695, 1604.374695, 1604.374695, 1604.374630,
+      1603.613892, 1568.297204, 1157.013600, 1138.976353, 1138.970921,
+      1092.970443, 1014.576161, 1029.697894, 1172.002606, 1579.603765,
+      1545.063972, 1539.712554, 1539.577625, 1539.577603, 1539.577603,
+      1539.577603, 1539.577603, 1539.577603, 1539.576008, 1521.537360,
+      1111.012900, 1092.975717, 1092.970443, 1048.827820,
+  },
 };
 
-void av1_model_rd_surffit(double xm, double yl, double *rate_f,
-                          double *dist_f) {
+static const double interp_dgrid_surf[33 * 18] = {
+  15.491252, 15.496413, 15.496379, 15.394447, 15.394431, 15.446941, 15.480480,
+  15.480491, 15.480491, 15.480491, 15.480573, 15.603112, 15.603153, 15.602880,
+  15.196621, 15.195983, 14.447585, 14.442076, 15.599420, 15.501644, 15.501575,
+  15.399610, 15.399669, 15.452121, 15.485671, 15.485682, 15.485682, 15.485724,
+  15.608303, 15.608385, 15.608113, 15.201854, 15.201079, 14.452681, 14.451350,
+  13.210484, 15.599455, 15.501644, 15.501575, 15.399686, 15.626744, 15.452197,
+  15.485671, 15.485682, 15.485682, 15.485765, 15.608344, 15.608113, 15.201854,
+  15.201079, 14.452681, 14.451350, 13.214913, 13.208705, 15.599455, 15.501679,
+  15.501575, 15.399686, 15.626820, 15.452197, 15.485671, 15.485682, 15.485765,
+  15.608303, 15.608113, 15.201854, 15.201079, 14.452681, 14.451350, 13.214913,
+  13.212184, 10.377543, 15.599489, 15.604652, 15.501610, 15.399686, 15.626820,
+  15.452197, 15.485671, 15.485724, 15.608303, 15.608113, 15.201794, 15.201079,
+  14.452681, 14.451350, 13.214913, 13.212184, 10.381022, 10.373758, 15.599489,
+  15.604686, 15.501645, 15.399727, 15.626807, 15.452213, 15.485673, 15.485765,
+  15.608072, 15.201734, 15.022022, 14.452621, 14.451350, 13.214913, 13.212184,
+  10.381022, 10.375351, 4.752366,  15.599553, 15.604785, 15.606745, 15.523109,
+  15.588440, 15.498592, 15.490939, 15.608113, 15.201734, 15.022022, 14.452398,
+  14.451350, 13.214913, 13.212184, 10.381022, 10.375351, 4.753958,  4.748594,
+  15.788491, 15.793763, 15.727178, 15.593404, 15.576185, 15.541155, 15.528638,
+  15.443234, 15.022046, 14.452234, 13.963908, 13.214750, 13.212184, 10.381022,
+  10.375351, 4.753958,  4.750187,  4.748513,  15.783667, 15.788604, 15.715040,
+  15.574608, 15.604897, 15.345511, 15.322323, 15.003579, 14.280362, 13.963611,
+  13.214382, 13.212184, 10.381022, 10.375351, 4.753958,  4.750106,  4.750025,
+  4.508804,  15.747709, 14.765834, 15.706678, 15.619092, 15.637946, 15.437393,
+  14.705235, 13.886464, 13.092309, 12.499506, 12.601594, 10.380302, 10.375350,
+  4.753958,  4.750026,  4.510253,  4.510173,  4.508580,  15.641901, 14.968111,
+  15.663074, 15.583828, 15.420627, 14.983936, 13.912901, 13.425841, 11.601603,
+  10.600757, 10.110971, 8.839811,  4.753363,  4.749963,  4.510190,  4.322919,
+  4.322857,  4.321407,  15.545263, 14.859883, 15.429167, 15.185696, 14.527376,
+  13.650042, 12.244185, 10.755706, 9.356137,  8.229212,  8.166627,  7.324414,
+  4.511146,  4.322709,  4.322566,  4.322503,  4.322503,  4.321054,  15.151031,
+  14.579126, 14.592456, 13.733213, 11.980508, 10.685709, 9.143695,  7.466555,
+  6.434770,  5.659204,  5.592407,  5.356211,  4.597869,  3.456439,  3.456056,
+  3.456176,  3.456176,  3.455018,  13.478598, 13.969965, 10.921203, 9.274388,
+  7.530261,  6.614303,  5.617812,  4.558886,  4.008929,  3.739688,  3.597707,
+  3.510418,  3.041638,  3.096611,  3.096750,  3.455008,  3.455038,  3.453880,
+  11.982351, 11.102072, 6.384662,  5.366178,  4.365376,  3.782955,  3.285890,
+  2.758904,  2.439067,  2.338593,  2.257332,  2.190385,  2.088145,  1.867744,
+  1.867445,  1.196878,  0.929332,  0.928931,  6.882587,  5.102224,  3.928701,
+  3.021205,  2.614349,  2.162491,  1.870802,  1.628127,  1.448370,  1.352404,
+  1.349332,  1.277236,  1.128399,  0.928668,  0.928601,  0.928485,  0.928395,
+  0.928084,  4.066496,  3.233175,  2.413117,  1.830193,  1.473572,  1.273461,
+  1.065156,  0.929824,  0.844729,  0.777037,  0.754300,  0.759737,  0.658871,
+  0.603118,  0.603209,  0.928096,  0.928205,  0.927893,  2.747433,  2.039019,
+  1.400016,  1.123698,  0.918772,  0.744824,  0.627536,  0.528263,  0.480634,
+  0.446093,  0.423444,  0.401394,  0.394353,  0.360914,  0.360903,  0.361012,
+  0.361012,  0.360891,  1.709038,  1.258373,  0.985529,  0.673582,  0.559577,
+  0.456411,  0.372560,  0.312622,  0.268425,  0.251698,  0.238132,  0.222477,
+  0.209643,  0.165264,  0.165254,  0.165254,  0.165254,  0.165199,  1.088160,
+  0.806677,  0.673453,  0.490758,  0.350173,  0.277724,  0.229342,  0.190245,
+  0.158717,  0.138414,  0.132405,  0.123800,  0.120550,  0.128355,  0.142808,
+  0.142812,  0.142800,  0.142752,  0.806227,  0.588621,  0.490714,  0.350136,
+  0.236497,  0.188583,  0.143991,  0.115039,  0.097317,  0.081406,  0.071159,
+  0.067980,  0.062769,  0.066575,  0.066581,  0.066569,  0.030318,  0.030291,
+  0.588339,  0.456572,  0.302189,  0.236473,  0.188564,  0.126491,  0.101248,
+  0.075242,  0.059591,  0.051942,  0.042824,  0.037390,  0.034523,  0.030277,
+  0.030275,  0.030275,  0.030258,  0.014677,  0.456374,  0.302172,  0.236448,
+  0.164278,  0.126481,  0.094667,  0.068489,  0.053835,  0.038513,  0.031335,
+  0.028947,  0.022753,  0.019347,  0.014676,  0.014675,  0.014674,  0.014674,
+  0.014664,  0.302014,  0.221625,  0.164273,  0.126473,  0.094667,  0.068483,
+  0.053831,  0.034298,  0.027445,  0.022220,  0.016315,  0.015692,  0.012949,
+  0.009448,  0.009442,  0.007914,  0.007913,  0.007910,  0.221524,  0.164268,
+  0.126473,  0.094667,  0.068480,  0.044800,  0.034295,  0.027442,  0.019666,
+  0.014389,  0.011041,  0.011550,  0.009117,  0.021326,  0.007915,  0.007911,
+  0.007911,  0.007908,  0.164193,  0.126473,  0.094667,  0.068480,  0.044800,
+  0.034291,  0.027442,  0.019665,  0.014388,  0.008515,  0.008320,  0.009117,
+  0.007357,  0.004348,  0.004343,  0.007908,  0.007910,  0.007907,  0.125623,
+  0.094667,  0.068480,  0.044800,  0.034289,  0.023364,  0.019663,  0.011887,
+  0.008514,  0.008318,  0.007208,  0.007208,  0.007206,  0.004342,  0.004341,
+  0.004342,  0.004342,  0.004341,  0.094624,  0.068480,  0.044800,  0.034289,
+  0.023364,  0.019662,  0.011887,  0.008514,  0.008318,  0.007208,  0.007207,
+  0.007207,  0.007206,  0.004342,  0.004341,  0.004341,  0.004341,  0.004340,
+  0.068449,  0.044800,  0.034289,  0.023364,  0.019662,  0.011887,  0.008514,
+  0.008318,  0.007208,  0.007207,  0.007207,  0.007207,  0.007206,  0.004343,
+  0.004341,  0.004341,  0.004341,  0.004340,  0.044777,  0.034289,  0.023364,
+  0.019662,  0.011887,  0.008514,  0.008318,  0.007208,  0.007207,  0.007207,
+  0.007207,  0.007207,  0.007207,  0.007205,  0.004342,  0.004341,  0.004341,
+  0.004340,  0.034274,  0.023364,  0.019662,  0.011887,  0.008514,  0.008318,
+  0.007208,  0.007207,  0.007207,  0.007207,  0.007207,  0.007207,  0.007207,
+  0.007206,  0.004342,  0.004341,  0.004341,  0.004340,  0.023353,  0.019662,
+  0.011887,  0.008514,  0.008318,  0.007208,  0.007207,  0.007207,  0.007207,
+  0.007207,  0.007207,  0.007207,  0.007207,  0.007206,  0.004343,  0.004341,
+  0.004341,  0.004340,  0.019650,  0.011884,  0.008511,  0.008316,  0.007205,
+  0.007205,  0.007205,  0.007205,  0.007205,  0.007205,  0.007205,  0.007205,
+  0.007205,  0.007205,  0.007203,  0.004341,  0.004340,  0.004338,
+};
+
+void av1_model_rd_surffit(BLOCK_SIZE bsize, double sse_norm, double xm,
+                          double yl, double *rate_f, double *dist_f) {
+  (void)sse_norm;
   const double x_start = -0.5;
   const double x_end = 16.5;
-  const double x_step = 1;
+  const double x_step = 1.0;
   const double y_start = -15.5;
   const double y_end = 16.5;
-  const double y_step = 0.5;
+  const double y_step = 1.0;
   const double epsilon = 1e-6;
   const int stride = (int)rint((x_end - x_start) / x_step) + 1;
+  const int rcat = bsize_model_cat_lookup[bsize];
   (void)y_end;
 
   xm = AOMMAX(xm, x_start + x_step + epsilon);
@@ -1061,44 +1325,112 @@ void av1_model_rd_surffit(double xm, double yl, double *rate_f,
 
   const double yo = y - yi;
   const double xo = x - xi;
-  const double *prate = &interp_rgrid_surf[(yi - 1) * stride + (xi - 1)];
+  const double *prate = &interp_rgrid_surf[rcat][(yi - 1) * stride + (xi - 1)];
   const double *pdist = &interp_dgrid_surf[(yi - 1) * stride + (xi - 1)];
   *rate_f = interp_bicubic(prate, stride, xo, yo);
   *dist_f = interp_bicubic(pdist, stride, xo, yo);
 }
 
-static const double interp_rgrid_curv[65] = {
-  0.000000,    0.000000,    0.000000,    0.000000,    0.000000,     0.000000,
-  0.000000,    0.000000,    0.000000,    0.000000,    0.000000,     0.000000,
-  0.000000,    0.000000,    0.000000,    0.000000,    0.000000,     4.759876,
-  8.132086,    13.651828,   21.908271,   33.522054,   48.782376,    71.530983,
-  106.728649,  151.942795,  199.893011,  242.850965,  283.933923,   322.154203,
-  360.684608,  394.801656,  426.879017,  460.234313,  484.103987,   508.261495,
-  536.486763,  558.196737,  586.285894,  614.764511,  634.166333,   647.706472,
-  658.211478,  681.360407,  701.052141,  727.007310,  768.663973,   804.407660,
-  884.627751,  1065.658131, 1238.875214, 1440.185176, 1678.377931,  1962.243390,
-  2300.571467, 2702.152072, 3175.775119, 3730.230519, 4374.308184,  5116.798028,
-  5966.489961, 6932.173897, 8022.639747, 9246.677424, 10613.076839,
+static const double interp_rgrid_curv[4][65] = {
+  {
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    23.801499,   28.387688,   33.388795,   42.298282,
+      41.525408,   51.597692,   49.566271,   54.632979,   60.321507,
+      67.730678,   75.766165,   85.324032,   96.600012,   120.839562,
+      173.917577,  255.974908,  354.107573,  458.063476,  562.345966,
+      668.568424,  772.072881,  878.598490,  982.202274,  1082.708946,
+      1188.037853, 1287.702240, 1395.588773, 1490.825830, 1584.231230,
+      1691.386090, 1766.822555, 1869.630904, 1926.743565, 2002.949495,
+      2047.431137, 2138.486068, 2154.743767, 2209.242472, 2277.593051,
+      2290.996432, 2307.452938, 2343.567091, 2397.654644, 2469.425868,
+      2558.591037, 2664.860422, 2787.944296, 2927.552932, 3083.396602,
+      3255.185579, 3442.630134, 3645.440541, 3863.327072, 4096.000000,
+  },
+  {
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    8.998436,    9.439592,    9.731837,    10.865931,
+      11.561347,   12.578139,   14.205101,   16.770584,   19.094853,
+      21.330863,   23.298907,   26.901921,   34.501017,   57.891733,
+      112.234763,  194.853189,  288.302032,  380.499422,  472.625309,
+      560.226809,  647.928463,  734.155122,  817.489721,  906.265783,
+      999.260562,  1094.489206, 1197.062998, 1293.296825, 1378.926484,
+      1472.760990, 1552.663779, 1635.196884, 1692.451951, 1759.741063,
+      1822.162720, 1916.515921, 1966.686071, 2031.647506, 2033.700134,
+      2087.847688, 2161.688858, 2242.536028, 2334.023491, 2436.337802,
+      2549.665519, 2674.193198, 2810.107395, 2957.594666, 3116.841567,
+      3288.034655, 3471.360486, 3667.005616, 3875.156602, 4096.000000,
+  },
+  {
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    2.377584,    2.557185,    2.732445,    2.851114,
+      3.281800,    3.765589,    4.342578,    5.145582,    5.611038,
+      6.642238,    7.945977,    11.800522,   17.346624,   37.501413,
+      87.216800,   165.860942,  253.865564,  332.039345,  408.518863,
+      478.120452,  547.268590,  616.067676,  680.022540,  753.863541,
+      834.529973,  919.489191,  1008.264989, 1092.230318, 1173.971886,
+      1249.514122, 1330.510941, 1399.523249, 1466.923387, 1530.533471,
+      1586.515722, 1695.197774, 1746.648696, 1837.136959, 1909.075485,
+      1975.074651, 2060.159200, 2155.335095, 2259.762505, 2373.710437,
+      2497.447898, 2631.243895, 2775.367434, 2930.087523, 3095.673170,
+      3272.393380, 3460.517161, 3660.313520, 3872.051464, 4096.000000,
+  },
+  {
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    0.000000,    0.000000,    0.000000,    0.000000,
+      0.000000,    0.296997,    0.342545,    0.403097,    0.472889,
+      0.614483,    0.842937,    1.050824,    1.326663,    1.717750,
+      2.530591,    3.582302,    6.995373,    9.973335,    24.042464,
+      56.598240,   113.680735,  180.018689,  231.050567,  266.101082,
+      294.957934,  323.326511,  349.434429,  380.443211,  408.171987,
+      441.214916,  475.716772,  512.900000,  551.186939,  592.364455,
+      624.527378,  661.940693,  679.185473,  724.800679,  764.781792,
+      873.050019,  950.299001,  939.292954,  1052.406153, 1033.893184,
+      1112.182406, 1219.174326, 1337.296681, 1471.648357, 1622.492809,
+      1790.093491, 1974.713858, 2176.617364, 2396.067465, 2633.327614,
+      2888.661266, 3162.331876, 3454.602899, 3765.737789, 4096.000000,
+  },
 };
 
-static const double interp_dgrid_curv[65] = {
-  14.604855, 14.604855, 14.604855, 14.604855, 14.604855, 14.604855, 14.604855,
-  14.604855, 14.604855, 14.604855, 14.604855, 14.604855, 14.555776, 14.533692,
-  14.439920, 14.257791, 13.977230, 13.623229, 13.064884, 12.355411, 11.560773,
-  10.728960, 9.861975,  8.643612,  6.916021,  5.154769,  3.734940,  2.680051,
-  1.925506,  1.408410,  1.042223,  0.767641,  0.565392,  0.420116,  0.310427,
-  0.231711,  0.172999,  0.128293,  0.094992,  0.072171,  0.052972,  0.039354,
-  0.029555,  0.022857,  0.016832,  0.013297,  0.000000,  0.000000,  0.000000,
-  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,
-  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,  0.000000,
-  0.000000,  0.000000,
+static const double interp_dgrid_curv[2][65] = {
+  {
+      16.000000, 15.962891, 15.925174, 15.886888, 15.848074, 15.808770,
+      15.769015, 15.728850, 15.688313, 15.647445, 15.606284, 15.564870,
+      15.525918, 15.483820, 15.373330, 15.126844, 14.637442, 14.184387,
+      13.560070, 12.880717, 12.165995, 11.378144, 10.438769, 9.130790,
+      7.487633,  5.688649,  4.267515,  3.196300,  2.434201,  1.834064,
+      1.369920,  1.035921,  0.775279,  0.574895,  0.427232,  0.314123,
+      0.233236,  0.171440,  0.128188,  0.092762,  0.067569,  0.049324,
+      0.036330,  0.027008,  0.019853,  0.015539,  0.011093,  0.008733,
+      0.007624,  0.008105,  0.005427,  0.004065,  0.003427,  0.002848,
+      0.002328,  0.001865,  0.001457,  0.001103,  0.000801,  0.000550,
+      0.000348,  0.000193,  0.000085,  0.000021,  0.000000,
+  },
+  {
+      16.000000, 15.996116, 15.984769, 15.966413, 15.941505, 15.910501,
+      15.873856, 15.832026, 15.785466, 15.734633, 15.679981, 15.621967,
+      15.560961, 15.460157, 15.288367, 15.052462, 14.466922, 13.921212,
+      13.073692, 12.222005, 11.237799, 9.985848,  8.898823,  7.423519,
+      5.995325,  4.773152,  3.744032,  2.938217,  2.294526,  1.762412,
+      1.327145,  1.020728,  0.765535,  0.570548,  0.425833,  0.313825,
+      0.232959,  0.171324,  0.128174,  0.092750,  0.067558,  0.049319,
+      0.036330,  0.027008,  0.019853,  0.015539,  0.011093,  0.008733,
+      0.007624,  0.008105,  0.005427,  0.004065,  0.003427,  0.002848,
+      0.002328,  0.001865,  0.001457,  0.001103,  0.000801,  0.000550,
+      0.000348,  0.000193,  0.000085,  0.000021,  -0.000000,
+  },
 };
 
-void av1_model_rd_curvfit(double xqr, double *rate_f, double *distbysse_f) {
+void av1_model_rd_curvfit(BLOCK_SIZE bsize, double sse_norm, double xqr,
+                          double *rate_f, double *distbysse_f) {
   const double x_start = -15.5;
   const double x_end = 16.5;
   const double x_step = 0.5;
   const double epsilon = 1e-6;
+  const int rcat = bsize_model_cat_lookup[bsize];
+  const int dcat = sse_norm_model_cat_lookup(sse_norm);
   (void)x_end;
 
   xqr = AOMMAX(xqr, x_start + x_step + epsilon);
@@ -1109,9 +1441,9 @@ void av1_model_rd_curvfit(double xqr, double *rate_f, double *distbysse_f) {
 
   assert(xi > 0);
 
-  const double *prate = &interp_rgrid_curv[(xi - 1)];
-  const double *pdist = &interp_dgrid_curv[(xi - 1)];
+  const double *prate = &interp_rgrid_curv[rcat][(xi - 1)];
   *rate_f = interp_cubic(prate, xo);
+  const double *pdist = &interp_dgrid_curv[dcat][(xi - 1)];
   *distbysse_f = interp_cubic(pdist, xo);
 }
 
@@ -1228,12 +1560,12 @@ int16_t *av1_raster_block_offset_int16(BLOCK_SIZE plane_bsize, int raster_block,
 
 YV12_BUFFER_CONFIG *av1_get_scaled_ref_frame(const AV1_COMP *cpi,
                                              int ref_frame) {
-  const AV1_COMMON *const cm = &cpi->common;
-  const int scaled_idx = cpi->scaled_ref_idx[ref_frame - 1];
-  const int ref_idx = get_ref_frame_buf_idx(cpi, ref_frame);
-  return (scaled_idx != ref_idx && scaled_idx != INVALID_IDX)
-             ? &cm->buffer_pool->frame_bufs[scaled_idx].buf
-             : NULL;
+  assert(ref_frame >= LAST_FRAME && ref_frame <= ALTREF_FRAME);
+  RefCntBuffer *const scaled_buf = cpi->scaled_ref_buf[ref_frame - 1];
+  const RefCntBuffer *const ref_buf =
+      get_ref_frame_buf(&cpi->common, ref_frame);
+  return (scaled_buf != ref_buf && scaled_buf != NULL) ? &scaled_buf->buf
+                                                       : NULL;
 }
 
 int av1_get_switchable_rate(const AV1_COMMON *const cm, MACROBLOCK *x,
@@ -1274,7 +1606,7 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   } else {
     rd->thresh_mult[THR_NEARESTMV] = 0;
     rd->thresh_mult[THR_NEARESTL2] = 0;
-    rd->thresh_mult[THR_NEARESTL3] = 0;
+    rd->thresh_mult[THR_NEARESTL3] = 100;
     rd->thresh_mult[THR_NEARESTB] = 0;
     rd->thresh_mult[THR_NEARESTA2] = 0;
     rd->thresh_mult[THR_NEARESTA] = 0;
@@ -1285,7 +1617,7 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_NEWL2] += 1000;
   rd->thresh_mult[THR_NEWL3] += 1000;
   rd->thresh_mult[THR_NEWB] += 1000;
-  rd->thresh_mult[THR_NEWA2] = 1000;
+  rd->thresh_mult[THR_NEWA2] = 1100;
   rd->thresh_mult[THR_NEWA] += 1000;
   rd->thresh_mult[THR_NEWG] += 1000;
 
@@ -1297,18 +1629,18 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_NEARA] += 1000;
   rd->thresh_mult[THR_NEARG] += 1000;
 
-  rd->thresh_mult[THR_GLOBALMV] += 2000;
+  rd->thresh_mult[THR_GLOBALMV] += 2200;
   rd->thresh_mult[THR_GLOBALL2] += 2000;
   rd->thresh_mult[THR_GLOBALL3] += 2000;
-  rd->thresh_mult[THR_GLOBALB] += 2000;
+  rd->thresh_mult[THR_GLOBALB] += 2400;
   rd->thresh_mult[THR_GLOBALA2] = 2000;
   rd->thresh_mult[THR_GLOBALG] += 2000;
-  rd->thresh_mult[THR_GLOBALA] += 2000;
+  rd->thresh_mult[THR_GLOBALA] += 2400;
 
-  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLA] += 1000;
+  rd->thresh_mult[THR_COMP_NEAREST_NEARESTLA] += 1100;
   rd->thresh_mult[THR_COMP_NEAREST_NEARESTL2A] += 1000;
-  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3A] += 1000;
-  rd->thresh_mult[THR_COMP_NEAREST_NEARESTGA] += 1000;
+  rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3A] += 800;
+  rd->thresh_mult[THR_COMP_NEAREST_NEARESTGA] += 900;
   rd->thresh_mult[THR_COMP_NEAREST_NEARESTLB] += 1000;
   rd->thresh_mult[THR_COMP_NEAREST_NEARESTL2B] += 1000;
   rd->thresh_mult[THR_COMP_NEAREST_NEARESTL3B] += 1000;
@@ -1326,17 +1658,17 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEAR_NEARLA] += 1200;
   rd->thresh_mult[THR_COMP_NEAREST_NEWLA] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTLA] += 1500;
-  rd->thresh_mult[THR_COMP_NEAR_NEWLA] += 1700;
-  rd->thresh_mult[THR_COMP_NEW_NEARLA] += 1700;
-  rd->thresh_mult[THR_COMP_NEW_NEWLA] += 2000;
-  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLA] += 2500;
+  rd->thresh_mult[THR_COMP_NEAR_NEWLA] += 1530;
+  rd->thresh_mult[THR_COMP_NEW_NEARLA] += 1870;
+  rd->thresh_mult[THR_COMP_NEW_NEWLA] += 2400;
+  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLA] += 2750;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARL2A] += 1200;
   rd->thresh_mult[THR_COMP_NEAREST_NEWL2A] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTL2A] += 1500;
-  rd->thresh_mult[THR_COMP_NEAR_NEWL2A] += 1700;
+  rd->thresh_mult[THR_COMP_NEAR_NEWL2A] += 1870;
   rd->thresh_mult[THR_COMP_NEW_NEARL2A] += 1700;
-  rd->thresh_mult[THR_COMP_NEW_NEWL2A] += 2000;
+  rd->thresh_mult[THR_COMP_NEW_NEWL2A] += 1800;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALL2A] += 2500;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARL3A] += 1200;
@@ -1345,23 +1677,23 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEAR_NEWL3A] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEARL3A] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEWL3A] += 2000;
-  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALL3A] += 2500;
+  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALL3A] += 3000;
 
-  rd->thresh_mult[THR_COMP_NEAR_NEARGA] += 1200;
+  rd->thresh_mult[THR_COMP_NEAR_NEARGA] += 1320;
   rd->thresh_mult[THR_COMP_NEAREST_NEWGA] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTGA] += 1500;
-  rd->thresh_mult[THR_COMP_NEAR_NEWGA] += 1700;
+  rd->thresh_mult[THR_COMP_NEAR_NEWGA] += 2040;
   rd->thresh_mult[THR_COMP_NEW_NEARGA] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEWGA] += 2000;
-  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALGA] += 2500;
+  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALGA] += 2250;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARLB] += 1200;
   rd->thresh_mult[THR_COMP_NEAREST_NEWLB] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTLB] += 1500;
-  rd->thresh_mult[THR_COMP_NEAR_NEWLB] += 1700;
+  rd->thresh_mult[THR_COMP_NEAR_NEWLB] += 1360;
   rd->thresh_mult[THR_COMP_NEW_NEARLB] += 1700;
-  rd->thresh_mult[THR_COMP_NEW_NEWLB] += 2000;
-  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLB] += 2500;
+  rd->thresh_mult[THR_COMP_NEW_NEWLB] += 2400;
+  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLB] += 2250;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARL2B] += 1200;
   rd->thresh_mult[THR_COMP_NEAREST_NEWL2B] += 1500;
@@ -1374,7 +1706,7 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEAR_NEARL3B] += 1200;
   rd->thresh_mult[THR_COMP_NEAREST_NEWL3B] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTL3B] += 1500;
-  rd->thresh_mult[THR_COMP_NEAR_NEWL3B] += 1700;
+  rd->thresh_mult[THR_COMP_NEAR_NEWL3B] += 1870;
   rd->thresh_mult[THR_COMP_NEW_NEARL3B] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEWL3B] += 2000;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALL3B] += 2500;
@@ -1388,7 +1720,7 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALGB] += 2500;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARLA2] += 1200;
-  rd->thresh_mult[THR_COMP_NEAREST_NEWLA2] += 1500;
+  rd->thresh_mult[THR_COMP_NEAREST_NEWLA2] += 1800;
   rd->thresh_mult[THR_COMP_NEW_NEARESTLA2] += 1500;
   rd->thresh_mult[THR_COMP_NEAR_NEWLA2] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEARLA2] += 1700;
@@ -1403,7 +1735,7 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEW_NEWL2A2] += 2000;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALL2A2] += 2500;
 
-  rd->thresh_mult[THR_COMP_NEAR_NEARL3A2] += 1200;
+  rd->thresh_mult[THR_COMP_NEAR_NEARL3A2] += 1440;
   rd->thresh_mult[THR_COMP_NEAREST_NEWL3A2] += 1500;
   rd->thresh_mult[THR_COMP_NEW_NEARESTL3A2] += 1500;
   rd->thresh_mult[THR_COMP_NEAR_NEWL3A2] += 1700;
@@ -1417,29 +1749,29 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEAR_NEWGA2] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEARGA2] += 1700;
   rd->thresh_mult[THR_COMP_NEW_NEWGA2] += 2000;
-  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALGA2] += 2500;
+  rd->thresh_mult[THR_COMP_GLOBAL_GLOBALGA2] += 2750;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARLL2] += 1600;
   rd->thresh_mult[THR_COMP_NEAREST_NEWLL2] += 2000;
   rd->thresh_mult[THR_COMP_NEW_NEARESTLL2] += 2000;
-  rd->thresh_mult[THR_COMP_NEAR_NEWLL2] += 2200;
+  rd->thresh_mult[THR_COMP_NEAR_NEWLL2] += 2640;
   rd->thresh_mult[THR_COMP_NEW_NEARLL2] += 2200;
   rd->thresh_mult[THR_COMP_NEW_NEWLL2] += 2400;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLL2] += 3200;
 
   rd->thresh_mult[THR_COMP_NEAR_NEARLL3] += 1600;
   rd->thresh_mult[THR_COMP_NEAREST_NEWLL3] += 2000;
-  rd->thresh_mult[THR_COMP_NEW_NEARESTLL3] += 2000;
+  rd->thresh_mult[THR_COMP_NEW_NEARESTLL3] += 1800;
   rd->thresh_mult[THR_COMP_NEAR_NEWLL3] += 2200;
   rd->thresh_mult[THR_COMP_NEW_NEARLL3] += 2200;
   rd->thresh_mult[THR_COMP_NEW_NEWLL3] += 2400;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLL3] += 3200;
 
-  rd->thresh_mult[THR_COMP_NEAR_NEARLG] += 1600;
-  rd->thresh_mult[THR_COMP_NEAREST_NEWLG] += 2000;
+  rd->thresh_mult[THR_COMP_NEAR_NEARLG] += 1760;
+  rd->thresh_mult[THR_COMP_NEAREST_NEWLG] += 2400;
   rd->thresh_mult[THR_COMP_NEW_NEARESTLG] += 2000;
-  rd->thresh_mult[THR_COMP_NEAR_NEWLG] += 2200;
-  rd->thresh_mult[THR_COMP_NEW_NEARLG] += 2200;
+  rd->thresh_mult[THR_COMP_NEAR_NEWLG] += 1760;
+  rd->thresh_mult[THR_COMP_NEW_NEARLG] += 2640;
   rd->thresh_mult[THR_COMP_NEW_NEWLG] += 2400;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALLG] += 3200;
 
@@ -1447,34 +1779,25 @@ void av1_set_rd_speed_thresholds(AV1_COMP *cpi) {
   rd->thresh_mult[THR_COMP_NEAREST_NEWBA] += 2000;
   rd->thresh_mult[THR_COMP_NEW_NEARESTBA] += 2000;
   rd->thresh_mult[THR_COMP_NEAR_NEWBA] += 2200;
-  rd->thresh_mult[THR_COMP_NEW_NEARBA] += 2200;
-  rd->thresh_mult[THR_COMP_NEW_NEWBA] += 2400;
+  rd->thresh_mult[THR_COMP_NEW_NEARBA] += 1980;
+  rd->thresh_mult[THR_COMP_NEW_NEWBA] += 2640;
   rd->thresh_mult[THR_COMP_GLOBAL_GLOBALBA] += 3200;
 
   rd->thresh_mult[THR_DC] += 1000;
   rd->thresh_mult[THR_PAETH] += 1000;
-  rd->thresh_mult[THR_SMOOTH] += 2000;
+  rd->thresh_mult[THR_SMOOTH] += 2200;
   rd->thresh_mult[THR_SMOOTH_V] += 2000;
   rd->thresh_mult[THR_SMOOTH_H] += 2000;
   rd->thresh_mult[THR_H_PRED] += 2000;
-  rd->thresh_mult[THR_V_PRED] += 2000;
+  rd->thresh_mult[THR_V_PRED] += 1800;
   rd->thresh_mult[THR_D135_PRED] += 2500;
-  rd->thresh_mult[THR_D203_PRED] += 2500;
+  rd->thresh_mult[THR_D203_PRED] += 2000;
   rd->thresh_mult[THR_D157_PRED] += 2500;
-  rd->thresh_mult[THR_D67_PRED] += 2500;
+  rd->thresh_mult[THR_D67_PRED] += 2000;
   rd->thresh_mult[THR_D113_PRED] += 2500;
   rd->thresh_mult[THR_D45_PRED] += 2500;
 }
 
-void av1_set_rd_speed_thresholds_sub8x8(AV1_COMP *cpi) {
-  static const int thresh_mult[MAX_REFS] = { 2500, 2500, 2500, 2500, 2500,
-                                             2500, 2500, 4500, 4500, 4500,
-                                             4500, 4500, 4500, 4500, 4500,
-                                             4500, 4500, 4500, 4500, 2500 };
-  RD_OPT *const rd = &cpi->rd;
-  memcpy(rd->thresh_mult_sub8x8, thresh_mult, sizeof(thresh_mult));
-}
-
 void av1_update_rd_thresh_fact(const AV1_COMMON *const cm,
                                int (*factor_buf)[MAX_MODES], int rd_thresh,
                                int bsize, int best_mode_index) {
diff --git a/third_party/aom/av1/encoder/rd.h b/third_party/aom/av1/encoder/rd.h
index 755b61df50..350eeb6d5f 100644
--- a/third_party/aom/av1/encoder/rd.h
+++ b/third_party/aom/av1/encoder/rd.h
@@ -48,7 +48,7 @@ extern "C" {
 
 // This enumerator type needs to be kept aligned with the mode order in
 // const MODE_DEFINITION av1_mode_order[MAX_MODES] used in the rd code.
-typedef enum {
+enum {
   THR_NEARESTMV,
   THR_NEARESTL2,
   THR_NEARESTL3,
@@ -246,9 +246,9 @@ typedef enum {
   MAX_SINGLE_REF_MODES = LAST_SINGLE_REF_MODES + 1,
   LAST_COMP_REF_MODES = THR_COMP_GLOBAL_GLOBALBA,
   MAX_COMP_REF_MODES = LAST_COMP_REF_MODES + 1
-} THR_MODES;
+} UENUM1BYTE(THR_MODES);
 
-typedef enum {
+enum {
   THR_LAST,
   THR_LAST2,
   THR_LAST3,
@@ -275,7 +275,7 @@ typedef enum {
   THR_INTRA,
 
   MAX_REFS
-} THR_MODES_SUB8X8;
+} UENUM1BYTE(THR_MODES_SUB8X8);
 
 typedef struct RD_OPT {
   // Thresh_mult is used to set a threshold for the rd score. A higher value
@@ -283,13 +283,14 @@ typedef struct RD_OPT {
   // is used in combination with the current block size, and thresh_freq_fact
   // to pick a threshold.
   int thresh_mult[MAX_MODES];
-  int thresh_mult_sub8x8[MAX_REFS];
 
   int threshes[MAX_SEGMENTS][BLOCK_SIZES_ALL][MAX_MODES];
 
   int64_t prediction_type_threshes[REF_FRAMES][REFERENCE_MODES];
 
   int RDMULT;
+
+  double r0;
 } RD_OPT;
 
 static INLINE void av1_init_rd_stats(RD_STATS *rd_stats) {
@@ -317,6 +318,25 @@ static INLINE void av1_init_rd_stats(RD_STATS *rd_stats) {
     }
   }
 #endif
+#if CONFIG_ONE_PASS_SVM
+  rd_stats->eob = 0;
+  rd_stats->eob_0 = 0;
+  rd_stats->eob_1 = 0;
+  rd_stats->eob_2 = 0;
+  rd_stats->eob_3 = 0;
+
+  rd_stats->rd = 0;
+  rd_stats->rd_0 = 0;
+  rd_stats->rd_1 = 0;
+  rd_stats->rd_2 = 0;
+  rd_stats->rd_3 = 0;
+
+  rd_stats->y_sse = 0;
+  rd_stats->sse_0 = 0;
+  rd_stats->sse_1 = 0;
+  rd_stats->sse_2 = 0;
+  rd_stats->sse_3 = 0;
+#endif
 }
 
 static INLINE void av1_invalid_rd_stats(RD_STATS *rd_stats) {
@@ -344,6 +364,30 @@ static INLINE void av1_invalid_rd_stats(RD_STATS *rd_stats) {
     }
   }
 #endif
+#if CONFIG_ONE_PASS_SVM
+  // TODO(chiyotsai@google.com): Change invalid values to INT_MAX and
+  // INT64_MAX. Currently there are some code paths where rd_stats's properties
+  // are set directly without calling av1_init_rd_stats, so changing it now will
+  // break this speed feature. Need to hunt down all places where rd_stats is
+  // used without initialized.
+  rd_stats->eob = 0;
+  rd_stats->eob_0 = 0;
+  rd_stats->eob_1 = 0;
+  rd_stats->eob_2 = 0;
+  rd_stats->eob_3 = 0;
+
+  rd_stats->rd = 0;
+  rd_stats->rd_0 = 0;
+  rd_stats->rd_1 = 0;
+  rd_stats->rd_2 = 0;
+  rd_stats->rd_3 = 0;
+
+  rd_stats->y_sse = 0;
+  rd_stats->sse_0 = 0;
+  rd_stats->sse_1 = 0;
+  rd_stats->sse_2 = 0;
+  rd_stats->sse_3 = 0;
+#endif
 }
 
 static INLINE void av1_merge_rd_stats(RD_STATS *rd_stats_dst,
@@ -377,26 +421,245 @@ static INLINE void av1_merge_rd_stats(RD_STATS *rd_stats_dst,
     }
   }
 #endif
+#if CONFIG_ONE_PASS_SVM
+  rd_stats_dst->eob += rd_stats_src->eob;
+  rd_stats_dst->eob_0 += rd_stats_src->eob_0;
+  rd_stats_dst->eob_1 += rd_stats_src->eob_1;
+  rd_stats_dst->eob_2 += rd_stats_src->eob_2;
+  rd_stats_dst->eob_3 += rd_stats_src->eob_3;
+
+  rd_stats_dst->rd += rd_stats_src->rd;
+  rd_stats_dst->rd_0 += rd_stats_src->rd_0;
+  rd_stats_dst->rd_1 += rd_stats_src->rd_1;
+  rd_stats_dst->rd_2 += rd_stats_src->rd_2;
+  rd_stats_dst->rd_3 += rd_stats_src->rd_3;
+
+  rd_stats_dst->y_sse += rd_stats_src->y_sse;
+  rd_stats_dst->sse_0 += rd_stats_src->sse_0;
+  rd_stats_dst->sse_1 += rd_stats_src->sse_1;
+  rd_stats_dst->sse_2 += rd_stats_src->sse_2;
+  rd_stats_dst->sse_3 += rd_stats_src->sse_3;
+#endif
 }
 
+#if CONFIG_ONE_PASS_SVM
+static INLINE void av1_add_reg_stat(RD_STATS *rd_stats, int eob, int64_t rd,
+                                    int64_t sse, int blk_row, int blk_col,
+                                    BLOCK_SIZE bsize, BLOCK_SIZE crop_bsize) {
+  // NOTE: Currently the calculation of regional features works by assuming
+  // bsize is square so that each transform block of size crop_bsize either
+  // 1. locates completely within a quadrant or
+  // 2. is exactly half of bsize or
+  // 3. is the entire prediction block
+  // Size of TX block and SB
+  const int block_width_mi = mi_size_wide[bsize];
+  const int block_height_mi = mi_size_high[bsize];
+  const int crop_width_mi = mi_size_wide[crop_bsize];
+  const int crop_height_mi = mi_size_high[crop_bsize];
+
+  // Increment the eob proportionally to how much the tx_block overlaps with
+  // each quadrant. We will scale it by MAX_MIB_SIZE * MAX_MIB_SIZE to avoid
+  // being truncated.
+  const int max_scaling_factor = MAX_MIB_SIZE * MAX_MIB_SIZE;
+
+  // Update the stats
+  rd_stats->eob = eob;
+  rd_stats->rd = rd;
+  rd_stats->y_sse = sse;
+
+  if (crop_width_mi <= block_width_mi / 2 &&
+      crop_height_mi <= block_width_mi / 2) {
+    // The transform block lies completely in a quadrant.
+    const int scaling_factor = max_scaling_factor;
+    const int r_eob = eob * scaling_factor, r_rd = rd * scaling_factor,
+              r_sse = sse * scaling_factor;
+
+    if (blk_row < block_height_mi / 2 && blk_col < block_width_mi / 2) {
+      rd_stats->eob_0 = r_eob;
+      rd_stats->rd_0 = r_rd;
+      rd_stats->sse_0 = r_sse;
+    } else if (blk_row < block_height_mi / 2 && blk_col >= block_width_mi / 2) {
+      rd_stats->eob_1 = r_eob;
+      rd_stats->rd_1 = r_rd;
+      rd_stats->sse_1 = r_sse;
+    } else if (blk_row >= block_height_mi / 2 && blk_col < block_width_mi / 2) {
+      rd_stats->eob_2 = r_eob;
+      rd_stats->rd_2 = r_rd;
+      rd_stats->sse_2 = r_sse;
+    } else {
+      rd_stats->eob_3 = r_eob;
+      rd_stats->rd_3 = r_rd;
+      rd_stats->sse_3 = r_sse;
+    }
+  } else if (crop_height_mi == block_height_mi &&
+             crop_width_mi == block_width_mi) {
+    // The transform block is the whole prediction block
+    const int scaling_factor = max_scaling_factor;
+    const int r_eob = eob * scaling_factor, r_rd = rd * scaling_factor,
+              r_sse = sse * scaling_factor;
+
+    rd_stats->eob_0 = r_eob;
+    rd_stats->rd_0 = r_rd;
+    rd_stats->sse_0 = r_sse;
+
+    rd_stats->eob_1 = r_eob;
+    rd_stats->rd_1 = r_rd;
+    rd_stats->sse_1 = r_sse;
+
+    rd_stats->eob_2 = r_eob;
+    rd_stats->rd_2 = r_rd;
+    rd_stats->sse_2 = r_sse;
+
+    rd_stats->eob_3 = r_eob;
+    rd_stats->rd_3 = r_rd;
+    rd_stats->sse_3 = r_sse;
+  } else if (crop_height_mi == block_height_mi) {
+    // The tranform block is a vertical block
+    const int scaling_factor = max_scaling_factor / 2;
+    const int r_eob = eob * scaling_factor, r_rd = rd * scaling_factor,
+              r_sse = sse * scaling_factor;
+
+    if (blk_col < block_width_mi / 2) {
+      rd_stats->eob_0 = r_eob;
+      rd_stats->rd_0 = r_rd;
+      rd_stats->sse_0 = r_sse;
+
+      rd_stats->eob_2 = r_eob;
+      rd_stats->rd_2 = r_rd;
+      rd_stats->sse_2 = r_sse;
+    } else {
+      rd_stats->eob_1 = r_eob;
+      rd_stats->rd_1 = r_rd;
+      rd_stats->sse_1 = r_sse;
+
+      rd_stats->eob_3 = r_eob;
+      rd_stats->rd_3 = r_rd;
+      rd_stats->sse_3 = r_sse;
+    }
+  } else if (crop_width_mi == block_width_mi) {
+    // The tranform block is a horizontal block half the size of predition block
+    const int scaling_factor = max_scaling_factor / 2;
+    const int r_eob = eob * scaling_factor, r_rd = rd * scaling_factor,
+              r_sse = sse * scaling_factor;
+
+    if (blk_row < block_height_mi / 2) {
+      rd_stats->eob_0 = r_eob;
+      rd_stats->rd_0 = r_rd;
+      rd_stats->sse_0 = r_sse;
+
+      rd_stats->eob_1 = r_eob;
+      rd_stats->rd_1 = r_rd;
+      rd_stats->sse_1 = r_sse;
+    } else {
+      rd_stats->eob_2 = r_eob;
+      rd_stats->rd_2 = r_rd;
+      rd_stats->sse_2 = r_sse;
+
+      rd_stats->eob_3 = r_eob;
+      rd_stats->rd_3 = r_rd;
+      rd_stats->sse_3 = r_sse;
+    }
+  } else {
+    assert(0 && "Unexpected transform size");
+  }
+}
+
+static INLINE void av1_reg_stat_skipmode_update(RD_STATS *rd_stats,
+                                                int rdmult) {
+  // Update the stats
+  rd_stats->eob = 0;
+  rd_stats->eob_0 = 0;
+  rd_stats->eob_1 = 0;
+  rd_stats->eob_2 = 0;
+  rd_stats->eob_3 = 0;
+
+  rd_stats->rd = RDCOST(rdmult, 0, rd_stats->sse);
+  rd_stats->rd_0 = RDCOST(rdmult, 0, rd_stats->sse_0);
+  rd_stats->rd_1 = RDCOST(rdmult, 0, rd_stats->sse_1);
+  rd_stats->rd_2 = RDCOST(rdmult, 0, rd_stats->sse_2);
+  rd_stats->rd_3 = RDCOST(rdmult, 0, rd_stats->sse_3);
+}
+
+static INLINE void av1_copy_reg_stat(RD_STATS *rd_stats_dst,
+                                     RD_STATS *rd_stats_src) {
+  rd_stats_dst->eob = rd_stats_src->eob;
+  rd_stats_dst->eob_0 = rd_stats_src->eob_0;
+  rd_stats_dst->eob_1 = rd_stats_src->eob_1;
+  rd_stats_dst->eob_2 = rd_stats_src->eob_2;
+  rd_stats_dst->eob_3 = rd_stats_src->eob_3;
+
+  rd_stats_dst->rd = rd_stats_src->rd;
+  rd_stats_dst->rd_0 = rd_stats_src->rd_0;
+  rd_stats_dst->rd_1 = rd_stats_src->rd_1;
+  rd_stats_dst->rd_2 = rd_stats_src->rd_2;
+  rd_stats_dst->rd_3 = rd_stats_src->rd_3;
+
+  rd_stats_dst->y_sse = rd_stats_src->y_sse;
+  rd_stats_dst->sse_0 = rd_stats_src->sse_0;
+  rd_stats_dst->sse_1 = rd_stats_src->sse_1;
+  rd_stats_dst->sse_2 = rd_stats_src->sse_2;
+  rd_stats_dst->sse_3 = rd_stats_src->sse_3;
+}
+
+static INLINE void av1_unpack_reg_stat(RD_STATS *rd_stats, int *eob, int *eob_0,
+                                       int *eob_1, int *eob_2, int *eob_3,
+                                       int64_t *rd, int64_t *rd_0,
+                                       int64_t *rd_1, int64_t *rd_2,
+                                       int64_t *rd_3) {
+  *rd = rd_stats->rd;
+  *rd_0 = rd_stats->rd_0;
+  *rd_1 = rd_stats->rd_1;
+  *rd_2 = rd_stats->rd_2;
+  *rd_3 = rd_stats->rd_3;
+
+  *eob = rd_stats->eob;
+  *eob_0 = rd_stats->eob_0;
+  *eob_1 = rd_stats->eob_1;
+  *eob_2 = rd_stats->eob_2;
+  *eob_3 = rd_stats->eob_3;
+}
+
+static INLINE void av1_set_reg_stat(RD_STATS *rd_stats, int eob, int eob_0,
+                                    int eob_1, int eob_2, int eob_3, int64_t rd,
+                                    int64_t rd_0, int64_t rd_1, int64_t rd_2,
+                                    int64_t rd_3) {
+  rd_stats->rd = rd;
+  rd_stats->rd_0 = rd_0;
+  rd_stats->rd_1 = rd_1;
+  rd_stats->rd_2 = rd_2;
+  rd_stats->rd_3 = rd_3;
+
+  rd_stats->eob = eob;
+  rd_stats->eob_0 = eob_0;
+  rd_stats->eob_1 = eob_1;
+  rd_stats->eob_2 = eob_2;
+  rd_stats->eob_3 = eob_3;
+}
+#endif
+
 struct TileInfo;
 struct TileDataEnc;
 struct AV1_COMP;
 struct macroblock;
 
+int av1_compute_rd_mult_based_on_qindex(const struct AV1_COMP *cpi, int qindex);
+
 int av1_compute_rd_mult(const struct AV1_COMP *cpi, int qindex);
 
 void av1_initialize_rd_consts(struct AV1_COMP *cpi);
 
+void av1_initialize_cost_tables(const AV1_COMMON *const cm, MACROBLOCK *x);
+
 void av1_initialize_me_consts(const struct AV1_COMP *cpi, MACROBLOCK *x,
                               int qindex);
 
 void av1_model_rd_from_var_lapndz(int64_t var, unsigned int n,
                                   unsigned int qstep, int *rate, int64_t *dist);
 
-void av1_model_rd_curvfit(double xqr, double *rate_f, double *distbysse_f);
-void av1_model_rd_surffit(double xm, double yl, double *rate_f,
-                          double *distbysse_f);
+void av1_model_rd_curvfit(BLOCK_SIZE bsize, double sse_norm, double xqr,
+                          double *rate_f, double *distbysse_f);
+void av1_model_rd_surffit(BLOCK_SIZE bsize, double sse_norm, double xm,
+                          double yl, double *rate_f, double *distbysse_f);
 
 int av1_get_switchable_rate(const AV1_COMMON *const cm, MACROBLOCK *x,
                             const MACROBLOCKD *xd);
@@ -421,8 +684,6 @@ void av1_get_entropy_contexts(BLOCK_SIZE bsize,
 
 void av1_set_rd_speed_thresholds(struct AV1_COMP *cpi);
 
-void av1_set_rd_speed_thresholds_sub8x8(struct AV1_COMP *cpi);
-
 void av1_update_rd_thresh_fact(const AV1_COMMON *const cm,
                                int (*fact)[MAX_MODES], int rd_thresh, int bsize,
                                int best_mode_index);
@@ -457,6 +718,8 @@ void av1_fill_mode_rates(AV1_COMMON *const cm, MACROBLOCK *x,
 void av1_fill_coeff_costs(MACROBLOCK *x, FRAME_CONTEXT *fc,
                           const int num_planes);
 
+int av1_get_adaptive_rdmult(const struct AV1_COMP *cpi, double beta);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/av1/encoder/rdopt.c b/third_party/aom/av1/encoder/rdopt.c
index c2d15534f0..0be76e4cb2 100644
--- a/third_party/aom/av1/encoder/rdopt.c
+++ b/third_party/aom/av1/encoder/rdopt.c
@@ -11,6 +11,7 @@
 
 #include <assert.h>
 #include <math.h>
+#include <stdbool.h>
 
 #include "config/aom_dsp_rtcd.h"
 #include "config/av1_rtcd.h"
@@ -30,6 +31,7 @@
 #include "av1/common/idct.h"
 #include "av1/common/mvref_common.h"
 #include "av1/common/obmc.h"
+#include "av1/common/onyxc_int.h"
 #include "av1/common/pred_common.h"
 #include "av1/common/quant_common.h"
 #include "av1/common/reconinter.h"
@@ -59,6 +61,13 @@
 #include "av1/encoder/tokenize.h"
 #include "av1/encoder/tx_prune_model_weights.h"
 
+// Set this macro as 1 to collect data about tx size selection.
+#define COLLECT_TX_SIZE_DATA 0
+
+#if COLLECT_TX_SIZE_DATA
+static const char av1_tx_size_data_output_file[] = "tx_size_data.txt";
+#endif
+
 typedef void (*model_rd_for_sb_type)(
     const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
     int plane_from, int plane_to, int mi_row, int mi_col, int *out_rate_sum,
@@ -116,14 +125,14 @@ static void model_rd_with_surffit(const AV1_COMP *const cpi,
                                   int64_t sse, int num_samples, int *rate,
                                   int64_t *dist);
 
-typedef enum {
+enum {
   MODELRD_LEGACY,
   MODELRD_CURVFIT,
   MODELRD_SUFFIT,
   MODELRD_DNN,
   MODELRD_FULLRDY,
   MODELRD_TYPES
-} ModelRdType;
+} UENUM1BYTE(ModelRdType);
 
 static model_rd_for_sb_type model_rd_sb_fn[MODELRD_TYPES] = {
   model_rd_for_sb, model_rd_for_sb_with_curvfit, model_rd_for_sb_with_surffit,
@@ -141,11 +150,12 @@ static model_rd_from_sse_type model_rd_sse_fn[MODELRD_TYPES] = {
 // 3: DNN regression model
 // 4: Full rd model
 #define MODELRD_TYPE_INTERP_FILTER 1
-#define MODELRD_TYPE_TX_SEARCH_PRUNE 2
+#define MODELRD_TYPE_TX_SEARCH_PRUNE 1
 #define MODELRD_TYPE_MASKED_COMPOUND 1
 #define MODELRD_TYPE_INTERINTRA 1
 #define MODELRD_TYPE_INTRA 1
-#define MODELRD_TYPE_JNT_COMPOUND 1
+#define MODELRD_TYPE_DIST_WTD_COMPOUND 1
+#define MODELRD_TYPE_MOTION_MODE_RD 1
 
 #define DUAL_FILTER_SET_SIZE (SWITCHABLE_FILTERS * SWITCHABLE_FILTERS)
 static const InterpFilters filter_sets[DUAL_FILTER_SET_SIZE] = {
@@ -158,8 +168,6 @@ static const InterpFilters filter_sets[DUAL_FILTER_SET_SIZE] = {
   ((1 << ALTREF_FRAME) | (1 << ALTREF2_FRAME) | (1 << BWDREF_FRAME) | \
    (1 << GOLDEN_FRAME) | (1 << LAST2_FRAME) | 0x01)
 
-#define ANGLE_SKIP_THRESH 10
-
 static const double ADST_FLIP_SVM[8] = {
   /* vertical */
   -6.6623, -2.8062, -3.2531, 3.1671,
@@ -176,22 +184,12 @@ typedef struct {
   MV_REFERENCE_FRAME ref_frame[2];
 } REF_DEFINITION;
 
-typedef enum {
+enum {
   FTXS_NONE = 0,
   FTXS_DCT_AND_1D_DCT_ONLY = 1 << 0,
   FTXS_DISABLE_TRELLIS_OPT = 1 << 1,
   FTXS_USE_TRANSFORM_DOMAIN = 1 << 2
-} FAST_TX_SEARCH_MODE;
-
-static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
-                               RD_STATS *rd_stats, BLOCK_SIZE bsize, int mi_row,
-                               int mi_col, int64_t ref_best_rd);
-
-static int inter_block_uvrd(const AV1_COMP *cpi, MACROBLOCK *x,
-                            RD_STATS *rd_stats, BLOCK_SIZE bsize,
-                            int64_t non_skip_ref_best_rd,
-                            int64_t skip_ref_best_rd,
-                            FAST_TX_SEARCH_MODE ftxs_mode);
+} UENUM1BYTE(FAST_TX_SEARCH_MODE);
 
 struct rdcost_block_args {
   const AV1_COMP *cpi;
@@ -406,32 +404,36 @@ static const MODE_DEFINITION av1_mode_order[MAX_MODES] = {
 };
 
 static const int16_t intra_to_mode_idx[INTRA_MODE_NUM] = {
-  7,    // DC_PRED,
-  134,  // V_PRED,
-  133,  // H_PRED,
-  140,  // D45_PRED,
-  135,  // D135_PRED,
-  139,  // D113_PRED,
-  137,  // D157_PRED,
-  136,  // D203_PRED,
-  138,  // D67_PRED,
-  46,   // SMOOTH_PRED,
-  47,   // SMOOTH_V_PRED,
-  48,   // SMOOTH_H_PRED,
-  45,   // PAETH_PRED,
+  THR_DC,         // DC_PRED,
+  THR_V_PRED,     // V_PRED,
+  THR_H_PRED,     // H_PRED,
+  THR_D45_PRED,   // D45_PRED,
+  THR_D135_PRED,  // D135_PRED,
+  THR_D113_PRED,  // D113_PRED,
+  THR_D157_PRED,  // D157_PRED,
+  THR_D203_PRED,  // D203_PRED,
+  THR_D67_PRED,   // D67_PRED,
+  THR_SMOOTH,     // SMOOTH_PRED,
+  THR_SMOOTH_V,   // SMOOTH_V_PRED,
+  THR_SMOOTH_H,   // SMOOTH_H_PRED,
+  THR_PAETH,      // PAETH_PRED,
 };
 
 /* clang-format off */
 static const int16_t single_inter_to_mode_idx[SINGLE_INTER_MODE_NUM]
                                              [REF_FRAMES] = {
   // NEARESTMV,
-  { -1, 0, 1, 2, 6, 3, 4, 5, },
+  { -1, THR_NEARESTMV, THR_NEARESTL2, THR_NEARESTL3,
+    THR_NEARESTG, THR_NEARESTB, THR_NEARESTA2, THR_NEARESTA, },
   // NEARMV,
-  { -1, 15, 16, 17, 21, 18, 19, 20, },
+  { -1, THR_NEARMV, THR_NEARL2, THR_NEARL3,
+    THR_NEARG, THR_NEARB, THR_NEARA2, THR_NEARA, },
   // GLOBALMV,
-  { -1, 22, 23, 24, 27, 25, 26, 28, },
+  { -1, THR_GLOBALMV, THR_GLOBALL2, THR_GLOBALL3,
+    THR_GLOBALG, THR_GLOBALB, THR_GLOBALA2, THR_GLOBALA, },
   // NEWMV,
-  { -1, 8, 9, 10, 14, 11, 12, 13, },
+  { -1, THR_NEWMV, THR_NEWL2, THR_NEWL3,
+    THR_NEWG, THR_NEWB, THR_NEWA2, THR_NEWA, },
 };
 /* clang-format on */
 
@@ -440,91 +442,211 @@ static const int16_t comp_inter_to_mode_idx[COMP_INTER_MODE_NUM][REF_FRAMES]
                                      [REF_FRAMES] = {
   // NEAREST_NEARESTMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 41, 42, 43, 33, 37, 29, },
-      { -1, -1, -1, -1, -1, 34, 38, 30, },
-      { -1, -1, -1, -1, -1, 35, 39, 31, },
-      { -1, -1, -1, -1, -1, 36, 40, 32, },
-      { -1, -1, -1, -1, -1, -1, -1, 44, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEAREST_NEARESTLL2, THR_COMP_NEAREST_NEARESTLL3,
+      THR_COMP_NEAREST_NEARESTLG, THR_COMP_NEAREST_NEARESTLB,
+      THR_COMP_NEAREST_NEARESTLA2, THR_COMP_NEAREST_NEARESTLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEARESTL2B,
+      THR_COMP_NEAREST_NEARESTL2A2, THR_COMP_NEAREST_NEARESTL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEARESTL3B,
+      THR_COMP_NEAREST_NEARESTL3A2, THR_COMP_NEAREST_NEARESTL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEARESTGB,
+      THR_COMP_NEAREST_NEARESTGA2, THR_COMP_NEAREST_NEARESTGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEARESTBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEAR_NEARMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 141, 148, 155, 77, 105, 49, },
-      { -1, -1, -1, -1, -1, 84, 112, 56, },
-      { -1, -1, -1, -1, -1, 91, 119, 63, },
-      { -1, -1, -1, -1, -1, 98, 126, 70, },
-      { -1, -1, -1, -1, -1, -1, -1, 162, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEAR_NEARLL2, THR_COMP_NEAR_NEARLL3,
+      THR_COMP_NEAR_NEARLG, THR_COMP_NEAR_NEARLB,
+      THR_COMP_NEAR_NEARLA2, THR_COMP_NEAR_NEARLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEARL2B,
+      THR_COMP_NEAR_NEARL2A2, THR_COMP_NEAR_NEARL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEARL3B,
+      THR_COMP_NEAR_NEARL3A2, THR_COMP_NEAR_NEARL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEARGB,
+      THR_COMP_NEAR_NEARGA2, THR_COMP_NEAR_NEARGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEARBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEAREST_NEWMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 143, 150, 157, 79, 107, 51, },
-      { -1, -1, -1, -1, -1, 86, 114, 58, },
-      { -1, -1, -1, -1, -1, 93, 121, 65, },
-      { -1, -1, -1, -1, -1, 100, 128, 72, },
-      { -1, -1, -1, -1, -1, -1, -1, 164, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEAREST_NEWLL2, THR_COMP_NEAREST_NEWLL3,
+      THR_COMP_NEAREST_NEWLG, THR_COMP_NEAREST_NEWLB,
+      THR_COMP_NEAREST_NEWLA2, THR_COMP_NEAREST_NEWLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEWL2B,
+      THR_COMP_NEAREST_NEWL2A2, THR_COMP_NEAREST_NEWL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEWL3B,
+      THR_COMP_NEAREST_NEWL3A2, THR_COMP_NEAREST_NEWL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEWGB,
+      THR_COMP_NEAREST_NEWGA2, THR_COMP_NEAREST_NEWGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAREST_NEWBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEW_NEARESTMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 142, 149, 156, 78, 106, 50, },
-      { -1, -1, -1, -1, -1, 85, 113, 57, },
-      { -1, -1, -1, -1, -1, 92, 120, 64, },
-      { -1, -1, -1, -1, -1, 99, 127, 71, },
-      { -1, -1, -1, -1, -1, -1, -1, 163, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEW_NEARESTLL2, THR_COMP_NEW_NEARESTLL3,
+      THR_COMP_NEW_NEARESTLG, THR_COMP_NEW_NEARESTLB,
+      THR_COMP_NEW_NEARESTLA2, THR_COMP_NEW_NEARESTLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARESTL2B,
+      THR_COMP_NEW_NEARESTL2A2, THR_COMP_NEW_NEARESTL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARESTL3B,
+      THR_COMP_NEW_NEARESTL3A2, THR_COMP_NEW_NEARESTL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARESTGB,
+      THR_COMP_NEW_NEARESTGA2, THR_COMP_NEW_NEARESTGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARESTBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEAR_NEWMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 145, 152, 159, 81, 109, 53, },
-      { -1, -1, -1, -1, -1, 88, 116, 60, },
-      { -1, -1, -1, -1, -1, 95, 123, 67, },
-      { -1, -1, -1, -1, -1, 102, 130, 74, },
-      { -1, -1, -1, -1, -1, -1, -1, 166, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEAR_NEWLL2, THR_COMP_NEAR_NEWLL3,
+      THR_COMP_NEAR_NEWLG, THR_COMP_NEAR_NEWLB,
+      THR_COMP_NEAR_NEWLA2, THR_COMP_NEAR_NEWLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEWL2B,
+      THR_COMP_NEAR_NEWL2A2, THR_COMP_NEAR_NEWL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEWL3B,
+      THR_COMP_NEAR_NEWL3A2, THR_COMP_NEAR_NEWL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEWGB,
+      THR_COMP_NEAR_NEWGA2, THR_COMP_NEAR_NEWGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEAR_NEWBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEW_NEARMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 144, 151, 158, 80, 108, 52, },
-      { -1, -1, -1, -1, -1, 87, 115, 59, },
-      { -1, -1, -1, -1, -1, 94, 122, 66, },
-      { -1, -1, -1, -1, -1, 101, 129, 73, },
-      { -1, -1, -1, -1, -1, -1, -1, 165, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEW_NEARLL2, THR_COMP_NEW_NEARLL3,
+      THR_COMP_NEW_NEARLG, THR_COMP_NEW_NEARLB,
+      THR_COMP_NEW_NEARLA2, THR_COMP_NEW_NEARLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARL2B,
+      THR_COMP_NEW_NEARL2A2, THR_COMP_NEW_NEARL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARL3B,
+      THR_COMP_NEW_NEARL3A2, THR_COMP_NEW_NEARL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARGB,
+      THR_COMP_NEW_NEARGA2, THR_COMP_NEW_NEARGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEARBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // GLOBAL_GLOBALMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 147, 154, 161, 83, 111, 55, },
-      { -1, -1, -1, -1, -1, 90, 118, 62, },
-      { -1, -1, -1, -1, -1, 97, 125, 69, },
-      { -1, -1, -1, -1, -1, 104, 132, 76, },
-      { -1, -1, -1, -1, -1, -1, -1, 168, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_GLOBAL_GLOBALLL2, THR_COMP_GLOBAL_GLOBALLL3,
+      THR_COMP_GLOBAL_GLOBALLG, THR_COMP_GLOBAL_GLOBALLB,
+      THR_COMP_GLOBAL_GLOBALLA2, THR_COMP_GLOBAL_GLOBALLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_GLOBAL_GLOBALL2B,
+      THR_COMP_GLOBAL_GLOBALL2A2, THR_COMP_GLOBAL_GLOBALL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_GLOBAL_GLOBALL3B,
+      THR_COMP_GLOBAL_GLOBALL3A2, THR_COMP_GLOBAL_GLOBALL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_GLOBAL_GLOBALGB,
+      THR_COMP_GLOBAL_GLOBALGA2, THR_COMP_GLOBAL_GLOBALGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_GLOBAL_GLOBALBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
   // NEW_NEWMV,
   {
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, 146, 153, 160, 82, 110, 54, },
-      { -1, -1, -1, -1, -1, 89, 117, 61, },
-      { -1, -1, -1, -1, -1, 96, 124, 68, },
-      { -1, -1, -1, -1, -1, 103, 131, 75, },
-      { -1, -1, -1, -1, -1, -1, -1, 167, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
-      { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1,
+      THR_COMP_NEW_NEWLL2, THR_COMP_NEW_NEWLL3,
+      THR_COMP_NEW_NEWLG, THR_COMP_NEW_NEWLB,
+      THR_COMP_NEW_NEWLA2, THR_COMP_NEW_NEWLA, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEWL2B,
+      THR_COMP_NEW_NEWL2A2, THR_COMP_NEW_NEWL2A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEWL3B,
+      THR_COMP_NEW_NEWL3A2, THR_COMP_NEW_NEWL3A, },
+    { -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEWGB,
+      THR_COMP_NEW_NEWGA2, THR_COMP_NEW_NEWGA, },
+    { -1, -1,
+      -1, -1,
+      -1, -1,
+      -1, THR_COMP_NEW_NEWBA, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
+    { -1, -1, -1, -1, -1, -1, -1, -1, },
   },
 };
 /* clang-format on */
@@ -618,12 +740,12 @@ typedef struct InterModeSearchState {
   MV_REFERENCE_FRAME single_rd_order[2][SINGLE_INTER_MODE_NUM][FWD_REFS];
 } InterModeSearchState;
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-int inter_mode_data_block_idx(BLOCK_SIZE bsize) {
-  if (bsize == BLOCK_8X8) return 1;
-  if (bsize == BLOCK_16X16) return 2;
-  if (bsize == BLOCK_32X32) return 3;
-  return -1;
+static int inter_mode_data_block_idx(BLOCK_SIZE bsize) {
+  if (bsize == BLOCK_4X4 || bsize == BLOCK_4X8 || bsize == BLOCK_8X4 ||
+      bsize == BLOCK_4X16 || bsize == BLOCK_16X4) {
+    return -1;
+  }
+  return 1;
 }
 
 void av1_inter_mode_data_init(TileDataEnc *tile_data) {
@@ -639,37 +761,41 @@ void av1_inter_mode_data_init(TileDataEnc *tile_data) {
   }
 }
 
-static int get_est_rate_dist(TileDataEnc *tile_data, BLOCK_SIZE bsize,
+static int get_est_rate_dist(const TileDataEnc *tile_data, BLOCK_SIZE bsize,
                              int64_t sse, int *est_residue_cost,
                              int64_t *est_dist) {
   aom_clear_system_state();
   const InterModeRdModel *md = &tile_data->inter_mode_rd_models[bsize];
   if (md->ready) {
-    const double est_ld = md->a * sse + md->b;
     if (sse < md->dist_mean) {
       *est_residue_cost = 0;
       *est_dist = sse;
     } else {
-      *est_residue_cost = (int)round((sse - md->dist_mean) / est_ld);
       *est_dist = (int64_t)round(md->dist_mean);
+      const double est_ld = md->a * sse + md->b;
+      // Clamp estimated rate cost by INT_MAX / 2.
+      // TODO(angiebird@google.com): find better solution than clamping.
+      if (fabs(est_ld) < 1e-2) {
+        *est_residue_cost = INT_MAX / 2;
+      } else {
+        double est_residue_cost_dbl = ((sse - md->dist_mean) / est_ld);
+        if (est_residue_cost_dbl < 0) {
+          *est_residue_cost = 0;
+        } else {
+          *est_residue_cost =
+              (int)AOMMIN((int64_t)round(est_residue_cost_dbl), INT_MAX / 2);
+        }
+      }
+      if (*est_residue_cost <= 0) {
+        *est_residue_cost = 0;
+        *est_dist = sse;
+      }
     }
     return 1;
   }
   return 0;
 }
 
-static int64_t get_est_rd(TileDataEnc *tile_data, BLOCK_SIZE bsize, int rdmult,
-                          int64_t sse, int curr_cost) {
-  int est_residue_cost;
-  int64_t est_dist;
-  if (get_est_rate_dist(tile_data, bsize, sse, &est_residue_cost, &est_dist)) {
-    int rate = est_residue_cost + curr_cost;
-    int64_t est_rd = RDCOST(rdmult, rate, est_dist);
-    return est_rd;
-  }
-  return 0;
-}
-
 void av1_inter_mode_data_fit(TileDataEnc *tile_data, int rdmult) {
   aom_clear_system_state();
   for (int bsize = 0; bsize < BLOCK_SIZES_ALL; ++bsize) {
@@ -734,7 +860,7 @@ static void inter_mode_data_push(TileDataEnc *tile_data, BLOCK_SIZE bsize,
     rd_model->dist_sum += dist;
     rd_model->ld_sum += ld;
     rd_model->sse_sum += sse;
-    rd_model->sse_sse_sum += sse * sse;
+    rd_model->sse_sse_sum += (double)sse * (double)sse;
     rd_model->sse_ld_sum += sse * ld;
   }
 }
@@ -773,7 +899,6 @@ static void inter_modes_info_sort(const InterModesInfo *inter_modes_info,
   qsort(rd_idx_pair_arr, inter_modes_info->num, sizeof(rd_idx_pair_arr[0]),
         compare_rd_idx_pair);
 }
-#endif  // CONFIG_COLLECT_INTER_MODE_RD_STATS
 
 static INLINE int write_uniform_cost(int n, int v) {
   const int l = get_unsigned_bits(n);
@@ -830,7 +955,7 @@ static unsigned pixel_dist_visible_only(
   }
   const MACROBLOCKD *xd = &x->e_mbd;
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     uint64_t sse64 = aom_highbd_sse_odd_size(src, src_stride, dst, dst_stride,
                                              visible_cols, visible_rows);
     return (unsigned int)ROUND_POWER_OF_TWO(sse64, (xd->bd - 8) * 2);
@@ -1086,7 +1211,7 @@ int64_t av1_dist_8x8(const AV1_COMP *const cpi, const MACROBLOCK *x,
 
   if (x->tune_metric == AOM_TUNE_CDEF_DIST ||
       x->tune_metric == AOM_TUNE_DAALA_DIST) {
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       for (j = 0; j < bsh; j++)
         for (i = 0; i < bsw; i++)
           orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
@@ -1150,8 +1275,7 @@ int64_t av1_dist_8x8(const AV1_COMP *const cpi, const MACROBLOCK *x,
                                  bsw, coeff_shift);
       }
     }
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-      d = ((uint64_t)d) >> 2 * coeff_shift;
+    if (is_cur_buf_hbd(xd)) d = ((uint64_t)d) >> 2 * coeff_shift;
   } else {
     // Otherwise, MSE by default
     d = pixel_dist_visible_only(cpi, x, src, src_stride, dst, dst_stride,
@@ -1179,7 +1303,7 @@ static int64_t dist_8x8_diff(const MACROBLOCK *x, const uint8_t *src,
 
   if (x->tune_metric == AOM_TUNE_CDEF_DIST ||
       x->tune_metric == AOM_TUNE_DAALA_DIST) {
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       for (j = 0; j < bsh; j++)
         for (i = 0; i < bsw; i++)
           orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i];
@@ -1372,54 +1496,10 @@ static int adst_vs_flipadst(const AV1_COMP *cpi, BLOCK_SIZE bsize,
   return prune_bitmask;
 }
 
-static void get_horver_correlation(const int16_t *diff, int stride, int w,
-                                   int h, double *hcorr, double *vcorr) {
-  // Returns hor/ver correlation coefficient
-  const int num = (h - 1) * (w - 1);
-  double num_r;
-  int i, j;
-  int64_t xy_sum = 0, xz_sum = 0;
-  int64_t x_sum = 0, y_sum = 0, z_sum = 0;
-  int64_t x2_sum = 0, y2_sum = 0, z2_sum = 0;
-  double x_var_n, y_var_n, z_var_n, xy_var_n, xz_var_n;
-  *hcorr = *vcorr = 1;
-
-  assert(num > 0);
-  num_r = 1.0 / num;
-  for (i = 1; i < h; ++i) {
-    for (j = 1; j < w; ++j) {
-      const int16_t x = diff[i * stride + j];
-      const int16_t y = diff[i * stride + j - 1];
-      const int16_t z = diff[(i - 1) * stride + j];
-      xy_sum += x * y;
-      xz_sum += x * z;
-      x_sum += x;
-      y_sum += y;
-      z_sum += z;
-      x2_sum += x * x;
-      y2_sum += y * y;
-      z2_sum += z * z;
-    }
-  }
-  x_var_n = x2_sum - (x_sum * x_sum) * num_r;
-  y_var_n = y2_sum - (y_sum * y_sum) * num_r;
-  z_var_n = z2_sum - (z_sum * z_sum) * num_r;
-  xy_var_n = xy_sum - (x_sum * y_sum) * num_r;
-  xz_var_n = xz_sum - (x_sum * z_sum) * num_r;
-  if (x_var_n > 0 && y_var_n > 0) {
-    *hcorr = xy_var_n / sqrt(x_var_n * y_var_n);
-    *hcorr = *hcorr < 0 ? 0 : *hcorr;
-  }
-  if (x_var_n > 0 && z_var_n > 0) {
-    *vcorr = xz_var_n / sqrt(x_var_n * z_var_n);
-    *vcorr = *vcorr < 0 ? 0 : *vcorr;
-  }
-}
-
 static int dct_vs_idtx(const int16_t *diff, int stride, int w, int h) {
-  double hcorr, vcorr;
+  float hcorr, vcorr;
   int prune_bitmask = 0;
-  get_horver_correlation(diff, stride, w, h, &hcorr, &vcorr);
+  av1_get_horver_correlation_full(diff, stride, w, h, &hcorr, &vcorr);
 
   if (vcorr > FAST_EXT_TX_CORR_MID + FAST_EXT_TX_CORR_MARGIN)
     prune_bitmask |= 1 << IDTX_1D;
@@ -1539,69 +1619,100 @@ static void get_energy_distribution_finer(const int16_t *diff, int stride,
 
 // Similar to get_horver_correlation, but also takes into account first
 // row/column, when computing horizontal/vertical correlation.
-static void get_horver_correlation_full(const int16_t *diff, int stride, int w,
-                                        int h, float *hcorr, float *vcorr) {
-  const float num_hor = (float)(h * (w - 1));
-  const float num_ver = (float)((h - 1) * w);
-  int i, j;
-
+void av1_get_horver_correlation_full_c(const int16_t *diff, int stride,
+                                       int width, int height, float *hcorr,
+                                       float *vcorr) {
   // The following notation is used:
   // x - current pixel
   // y - left neighbor pixel
   // z - top neighbor pixel
-  int64_t xy_sum = 0, xz_sum = 0;
-  int64_t xhor_sum = 0, xver_sum = 0, y_sum = 0, z_sum = 0;
-  int64_t x2hor_sum = 0, x2ver_sum = 0, y2_sum = 0, z2_sum = 0;
+  int64_t x_sum = 0, x2_sum = 0, xy_sum = 0, xz_sum = 0;
+  int64_t x_firstrow = 0, x_finalrow = 0, x_firstcol = 0, x_finalcol = 0;
+  int64_t x2_firstrow = 0, x2_finalrow = 0, x2_firstcol = 0, x2_finalcol = 0;
 
-  int16_t x, y, z;
-  for (j = 1; j < w; ++j) {
-    x = diff[j];
-    y = diff[j - 1];
+  // First, process horizontal correlation on just the first row
+  x_sum += diff[0];
+  x2_sum += diff[0] * diff[0];
+  x_firstrow += diff[0];
+  x2_firstrow += diff[0] * diff[0];
+  for (int j = 1; j < width; ++j) {
+    const int16_t x = diff[j];
+    const int16_t y = diff[j - 1];
+    x_sum += x;
+    x_firstrow += x;
+    x2_sum += x * x;
+    x2_firstrow += x * x;
     xy_sum += x * y;
-    xhor_sum += x;
-    y_sum += y;
-    x2hor_sum += x * x;
-    y2_sum += y * y;
   }
-  for (i = 1; i < h; ++i) {
-    x = diff[i * stride];
-    z = diff[(i - 1) * stride];
+
+  // Process vertical correlation in the first column
+  x_firstcol += diff[0];
+  x2_firstcol += diff[0] * diff[0];
+  for (int i = 1; i < height; ++i) {
+    const int16_t x = diff[i * stride];
+    const int16_t z = diff[(i - 1) * stride];
+    x_sum += x;
+    x_firstcol += x;
+    x2_sum += x * x;
+    x2_firstcol += x * x;
     xz_sum += x * z;
-    xver_sum += x;
-    z_sum += z;
-    x2ver_sum += x * x;
-    z2_sum += z * z;
-    for (j = 1; j < w; ++j) {
-      x = diff[i * stride + j];
-      y = diff[i * stride + j - 1];
-      z = diff[(i - 1) * stride + j];
+  }
+
+  // Now process horiz and vert correlation through the rest unit
+  for (int i = 1; i < height; ++i) {
+    for (int j = 1; j < width; ++j) {
+      const int16_t x = diff[i * stride + j];
+      const int16_t y = diff[i * stride + j - 1];
+      const int16_t z = diff[(i - 1) * stride + j];
+      x_sum += x;
+      x2_sum += x * x;
       xy_sum += x * y;
       xz_sum += x * z;
-      xhor_sum += x;
-      xver_sum += x;
-      y_sum += y;
-      z_sum += z;
-      x2hor_sum += x * x;
-      x2ver_sum += x * x;
-      y2_sum += y * y;
-      z2_sum += z * z;
     }
   }
+
+  for (int j = 0; j < width; ++j) {
+    x_finalrow += diff[(height - 1) * stride + j];
+    x2_finalrow +=
+        diff[(height - 1) * stride + j] * diff[(height - 1) * stride + j];
+  }
+  for (int i = 0; i < height; ++i) {
+    x_finalcol += diff[i * stride + width - 1];
+    x2_finalcol += diff[i * stride + width - 1] * diff[i * stride + width - 1];
+  }
+
+  int64_t xhor_sum = x_sum - x_finalcol;
+  int64_t xver_sum = x_sum - x_finalrow;
+  int64_t y_sum = x_sum - x_firstcol;
+  int64_t z_sum = x_sum - x_firstrow;
+  int64_t x2hor_sum = x2_sum - x2_finalcol;
+  int64_t x2ver_sum = x2_sum - x2_finalrow;
+  int64_t y2_sum = x2_sum - x2_firstcol;
+  int64_t z2_sum = x2_sum - x2_firstrow;
+
+  const float num_hor = (float)(height * (width - 1));
+  const float num_ver = (float)((height - 1) * width);
+
   const float xhor_var_n = x2hor_sum - (xhor_sum * xhor_sum) / num_hor;
-  const float y_var_n = y2_sum - (y_sum * y_sum) / num_hor;
-  const float xy_var_n = xy_sum - (xhor_sum * y_sum) / num_hor;
   const float xver_var_n = x2ver_sum - (xver_sum * xver_sum) / num_ver;
+
+  const float y_var_n = y2_sum - (y_sum * y_sum) / num_hor;
   const float z_var_n = z2_sum - (z_sum * z_sum) / num_ver;
+
+  const float xy_var_n = xy_sum - (xhor_sum * y_sum) / num_hor;
   const float xz_var_n = xz_sum - (xver_sum * z_sum) / num_ver;
 
-  *hcorr = *vcorr = 1;
   if (xhor_var_n > 0 && y_var_n > 0) {
     *hcorr = xy_var_n / sqrtf(xhor_var_n * y_var_n);
     *hcorr = *hcorr < 0 ? 0 : *hcorr;
+  } else {
+    *hcorr = 1.0;
   }
   if (xver_var_n > 0 && z_var_n > 0) {
     *vcorr = xz_var_n / sqrtf(xver_var_n * z_var_n);
     *vcorr = *vcorr < 0 ? 0 : *vcorr;
+  } else {
+    *vcorr = 1.0;
   }
 }
 
@@ -1609,16 +1720,19 @@ static void get_horver_correlation_full(const int16_t *diff, int stride, int w,
 static void score_2D_transform_pow8(float *scores_2D, float shift) {
   float sum = 0.0f;
   int i;
-
   for (i = 0; i < 16; i++) {
-    float v, v2, v4;
-    v = AOMMAX(scores_2D[i] + shift, 0.0f);
-    v2 = v * v;
-    v4 = v2 * v2;
+    const float v = AOMMIN(AOMMAX(scores_2D[i] + shift, 0.0f), 100.0f);
+    const float v2 = v * v;
+    const float v4 = v2 * v2;
     scores_2D[i] = v4 * v4;
     sum += scores_2D[i];
   }
-  for (i = 0; i < 16; i++) scores_2D[i] /= sum;
+  for (i = 0; i < 16; i++) {
+    if (scores_2D[i] < sum * 1e-4)
+      scores_2D[i] = 0.0f;
+    else
+      scores_2D[i] /= sum;
+  }
 }
 
 // These thresholds were calibrated to provide a certain number of TX types
@@ -1714,11 +1828,12 @@ static uint16_t prune_tx_2D(MACROBLOCK *x, BLOCK_SIZE bsize, TX_SIZE tx_size,
   const int16_t *diff = p->src_diff + 4 * blk_row * diff_stride + 4 * blk_col;
   get_energy_distribution_finer(diff, diff_stride, bw, bh, hfeatures,
                                 vfeatures);
-  get_horver_correlation_full(diff, diff_stride, bw, bh,
-                              &hfeatures[hfeatures_num - 1],
-                              &vfeatures[vfeatures_num - 1]);
+  av1_get_horver_correlation_full(diff, diff_stride, bw, bh,
+                                  &hfeatures[hfeatures_num - 1],
+                                  &vfeatures[vfeatures_num - 1]);
   av1_nn_predict(hfeatures, nn_config_hor, hscores);
   av1_nn_predict(vfeatures, nn_config_ver, vscores);
+  aom_clear_system_state();
 
   float score_2D_average = 0.0f;
   for (int i = 0; i < 4; i++) {
@@ -1790,7 +1905,13 @@ static void prune_tx(const AV1_COMP *cpi, BLOCK_SIZE bsize, MACROBLOCK *x,
   x->tx_search_prune[tx_set_type] = 0;
   x->tx_split_prune_flag = 0;
   const MB_MODE_INFO *mbmi = xd->mi[0];
-  if (!is_inter_block(mbmi) || cpi->sf.tx_type_search.prune_mode == NO_PRUNE ||
+  const int is_inter = is_inter_block(mbmi);
+  if ((is_inter && cpi->oxcf.use_inter_dct_only) ||
+      (!is_inter && cpi->oxcf.use_intra_dct_only)) {
+    x->tx_search_prune[tx_set_type] = ~(1 << DCT_DCT);
+    return;
+  }
+  if (!is_inter || cpi->sf.tx_type_search.prune_mode == NO_PRUNE ||
       x->use_default_inter_tx_type || xd->lossless[mbmi->segment_id] ||
       x->cb_partition_scan)
     return;
@@ -1829,8 +1950,7 @@ static void model_rd_from_sse(const AV1_COMP *const cpi,
   (void)num_samples;
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
 
   // Fast approximate the modelling function.
   if (cpi->sf.simple_model_rd_from_var) {
@@ -1852,7 +1972,6 @@ static void model_rd_from_sse(const AV1_COMP *const cpi,
   *dist <<= 4;
 }
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
 static int64_t get_sse(const AV1_COMP *cpi, const MACROBLOCK *x) {
   const AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
@@ -1875,7 +1994,6 @@ static int64_t get_sse(const AV1_COMP *cpi, const MACROBLOCK *x) {
   total_sse <<= 4;
   return total_sse;
 }
-#endif
 
 static void model_rd_for_sb(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
                             MACROBLOCK *x, MACROBLOCKD *xd, int plane_from,
@@ -1909,7 +2027,7 @@ static void model_rd_for_sb(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
 
     if (x->skip_chroma_rd && plane) continue;
 
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
                            pd->dst.stride, bw, bh);
     } else {
@@ -1938,43 +2056,6 @@ static void model_rd_for_sb(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
   *out_dist_sum = dist_sum;
 }
 
-static void check_block_skip(const AV1_COMP *const cpi, BLOCK_SIZE bsize,
-                             MACROBLOCK *x, MACROBLOCKD *xd, int plane_from,
-                             int plane_to, int *skip_txfm_sb) {
-  *skip_txfm_sb = 1;
-  for (int plane = plane_from; plane <= plane_to; ++plane) {
-    struct macroblock_plane *const p = &x->plane[plane];
-    struct macroblockd_plane *const pd = &xd->plane[plane];
-    const BLOCK_SIZE bs =
-        get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
-    unsigned int sse;
-
-    if (x->skip_chroma_rd && plane) continue;
-
-    // Since fast HBD variance functions scale down sse by 4 bit, we first use
-    // fast vf implementation to rule out blocks with non-zero scaled sse. Then,
-    // only if the source is HBD and the scaled sse is 0, accurate sse
-    // computation is applied to determine if the sse is really 0. This step is
-    // necessary for HBD lossless coding.
-    cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
-                       &sse);
-    if (sse) {
-      *skip_txfm_sb = 0;
-      return;
-    } else if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      uint64_t sse64 = aom_highbd_sse_odd_size(
-          p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
-          block_size_wide[bs], block_size_high[bs]);
-
-      if (sse64) {
-        *skip_txfm_sb = 0;
-        return;
-      }
-    }
-  }
-  return;
-}
-
 int64_t av1_block_error_c(const tran_low_t *coeff, const tran_low_t *dqcoeff,
                           intptr_t block_size, int64_t *ssz) {
   int i;
@@ -2076,7 +2157,8 @@ static unsigned pixel_dist(const AV1_COMP *const cpi, const MACROBLOCK *x,
 static INLINE int64_t pixel_diff_dist(const MACROBLOCK *x, int plane,
                                       int blk_row, int blk_col,
                                       const BLOCK_SIZE plane_bsize,
-                                      const BLOCK_SIZE tx_bsize) {
+                                      const BLOCK_SIZE tx_bsize,
+                                      unsigned int *block_mse_q8) {
   int visible_rows, visible_cols;
   const MACROBLOCKD *xd = &x->e_mbd;
   get_txb_dimensions(xd, plane, plane_bsize, blk_row, blk_col, tx_bsize, NULL,
@@ -2099,7 +2181,11 @@ static INLINE int64_t pixel_diff_dist(const MACROBLOCK *x, int plane,
   }
 #endif
   diff += ((blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]);
-  return aom_sum_squares_2d_i16(diff, diff_stride, visible_cols, visible_rows);
+  uint64_t sse =
+      aom_sum_squares_2d_i16(diff, diff_stride, visible_cols, visible_rows);
+  if (block_mse_q8 != NULL)
+    *block_mse_q8 = (unsigned int)((256 * sse) / (visible_cols * visible_rows));
+  return sse;
 }
 
 int av1_count_colors(const uint8_t *src, int stride, int rows, int cols,
@@ -2199,7 +2285,7 @@ static INLINE void dist_block_tx_domain(MACROBLOCK *x, int plane, int block,
   tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
   tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+  if (is_cur_buf_hbd(xd))
     *out_dist = av1_highbd_block_error(coeff, dqcoeff, buffer_length, &this_sse,
                                        xd->bd);
   else
@@ -2235,7 +2321,7 @@ static INLINE int64_t dist_block_px_domain(const AV1_COMP *cpi, MACROBLOCK *x,
   uint8_t *recon;
   DECLARE_ALIGNED(16, uint16_t, recon16[MAX_TX_SQUARE]);
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     recon = CONVERT_TO_BYTEPTR(recon16);
     av1_highbd_convolve_2d_copy_sr(CONVERT_TO_SHORTPTR(dst), dst_stride,
                                    CONVERT_TO_SHORTPTR(recon), MAX_TX_SIZE, bsw,
@@ -2257,11 +2343,29 @@ static INLINE int64_t dist_block_px_domain(const AV1_COMP *cpi, MACROBLOCK *x,
                          blk_row, blk_col, plane_bsize, tx_bsize);
 }
 
-static double get_mean(const int16_t *diff, int stride, int w, int h) {
+static double get_diff_mean(const uint8_t *src, int src_stride,
+                            const uint8_t *dst, int dst_stride, int w, int h) {
   double sum = 0.0;
   for (int j = 0; j < h; ++j) {
     for (int i = 0; i < w; ++i) {
-      sum += diff[j * stride + i];
+      const int diff = src[j * src_stride + i] - dst[j * dst_stride + i];
+      sum += diff;
+    }
+  }
+  assert(w > 0 && h > 0);
+  return sum / (w * h);
+}
+
+static double get_highbd_diff_mean(const uint8_t *src8, int src_stride,
+                                   const uint8_t *dst8, int dst_stride, int w,
+                                   int h) {
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+  double sum = 0.0;
+  for (int j = 0; j < h; ++j) {
+    for (int i = 0; i < w; ++i) {
+      const int diff = src[j * src_stride + i] - dst[j * dst_stride + i];
+      sum += diff;
     }
   }
   assert(w > 0 && h > 0);
@@ -2350,6 +2454,17 @@ static void get_2x2_normalized_sses_and_sads(
 #if CONFIG_COLLECT_RD_STATS
 
 #if CONFIG_COLLECT_RD_STATS == 1
+static double get_mean(const int16_t *diff, int stride, int w, int h) {
+  double sum = 0.0;
+  for (int j = 0; j < h; ++j) {
+    for (int i = 0; i < w; ++i) {
+      sum += diff[j * stride + i];
+    }
+  }
+  assert(w > 0 && h > 0);
+  return sum / (w * h);
+}
+
 static void PrintTransformUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
                                     const RD_STATS *const rd_stats, int blk_row,
                                     int blk_col, BLOCK_SIZE plane_bsize,
@@ -2372,10 +2487,9 @@ static void PrintTransformUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   const int txw = tx_size_wide[tx_size];
   const int txh = tx_size_high[tx_size];
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
   const int q_step = pd->dequant_Q3[1] >> dequant_shift;
-  const double num_samples = txw * txh;
+  const int num_samples = txw * txh;
 
   const double rate_norm = (double)rd_stats->rate / num_samples;
   const double dist_norm = (double)rd_stats->dist / num_samples;
@@ -2428,9 +2542,9 @@ static void PrintTransformUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   fprintf(fout, " %g %g", model_rate_norm, model_dist_norm);
 
   const double mean = get_mean(src_diff, diff_stride, txw, txh);
-  double hor_corr, vert_corr;
-  get_horver_correlation(src_diff, diff_stride, txw, txh, &hor_corr,
-                         &vert_corr);
+  float hor_corr, vert_corr;
+  av1_get_horver_correlation_full(src_diff, diff_stride, txw, txh, &hor_corr,
+                                  &vert_corr);
   fprintf(fout, " %g %g %g", mean, hor_corr, vert_corr);
 
   double hdist[4] = { 0 }, vdist[4] = { 0 };
@@ -2447,15 +2561,25 @@ static void PrintTransformUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
 #endif  // CONFIG_COLLECT_RD_STATS == 1
 
 #if CONFIG_COLLECT_RD_STATS >= 2
-static void PrintPredictionUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
+static void PrintPredictionUnitStats(const AV1_COMP *const cpi,
+                                     const TileDataEnc *tile_data,
+                                     MACROBLOCK *x,
                                      const RD_STATS *const rd_stats,
                                      BLOCK_SIZE plane_bsize) {
   if (rd_stats->invalid_rate) return;
   if (rd_stats->rate == INT_MAX || rd_stats->dist == INT64_MAX) return;
 
+  if (cpi->sf.inter_mode_rd_model_estimation == 1 &&
+      (tile_data == NULL ||
+       !tile_data->inter_mode_rd_models[plane_bsize].ready))
+    return;
+  (void)tile_data;
   // Generate small sample to restrict output size.
   static unsigned int seed = 95014;
-  if (lcg_rand16(&seed) % 256 > 0) return;
+
+  if ((lcg_rand16(&seed) % (1 << (14 - num_pels_log2_lookup[plane_bsize]))) !=
+      1)
+    return;
 
   const char output_file[] = "pu_stats.txt";
   FILE *fout = fopen(output_file, "a");
@@ -2470,8 +2594,7 @@ static void PrintPredictionUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   get_txb_dimensions(xd, plane, plane_bsize, 0, 0, plane_bsize, NULL, NULL, &bw,
                      &bh);
   const int num_samples = bw * bh;
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
   const int q_step = pd->dequant_Q3[1] >> dequant_shift;
 
   const double rate_norm = (double)rd_stats->rate / num_samples;
@@ -2488,7 +2611,14 @@ static void PrintPredictionUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   const int16_t *const src_diff = p->src_diff;
   const int shift = (xd->bd - 8);
 
-  int64_t sse = aom_sum_squares_2d_i16(src_diff, diff_stride, bw, bh);
+  int64_t sse;
+  if (is_cur_buf_hbd(xd)) {
+    sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
+                         bw, bh);
+  } else {
+    sse =
+        aom_sse(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride, bw, bh);
+  }
   sse = ROUND_POWER_OF_TWO(sse, shift * 2);
   const double sse_norm = (double)sse / num_samples;
 
@@ -2527,10 +2657,18 @@ static void PrintPredictionUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   fprintf(fout, " %g %g %g", model_rate_norm, model_dist_norm,
           model_rdcost_norm);
 
-  double mean = get_mean(src_diff, diff_stride, bw, bh);
+  double mean;
+  if (is_cur_buf_hbd(xd)) {
+    mean = get_highbd_diff_mean(p->src.buf, p->src.stride, pd->dst.buf,
+                                pd->dst.stride, bw, bh);
+  } else {
+    mean = get_diff_mean(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride,
+                         bw, bh);
+  }
   mean /= (1 << shift);
-  double hor_corr, vert_corr;
-  get_horver_correlation(src_diff, diff_stride, bw, bh, &hor_corr, &vert_corr);
+  float hor_corr, vert_corr;
+  av1_get_horver_correlation_full(src_diff, diff_stride, bw, bh, &hor_corr,
+                                  &vert_corr);
   fprintf(fout, " %g %g %g", mean, hor_corr, vert_corr);
 
   double hdist[4] = { 0 }, vdist[4] = { 0 };
@@ -2539,6 +2677,21 @@ static void PrintPredictionUnitStats(const AV1_COMP *const cpi, MACROBLOCK *x,
   fprintf(fout, " %g %g %g %g %g %g %g %g", hdist[0], hdist[1], hdist[2],
           hdist[3], vdist[0], vdist[1], vdist[2], vdist[3]);
 
+  if (cpi->sf.inter_mode_rd_model_estimation == 1) {
+    assert(tile_data->inter_mode_rd_models[plane_bsize].ready);
+    const int64_t overall_sse = get_sse(cpi, x);
+    int est_residue_cost = 0;
+    int64_t est_dist = 0;
+    get_est_rate_dist(tile_data, plane_bsize, overall_sse, &est_residue_cost,
+                      &est_dist);
+    const double est_residue_cost_norm = (double)est_residue_cost / num_samples;
+    const double est_dist_norm = (double)est_dist / num_samples;
+    const double est_rdcost_norm =
+        (double)RDCOST(x->rdmult, est_residue_cost, est_dist) / num_samples;
+    fprintf(fout, " %g %g %g", est_residue_cost_norm, est_dist_norm,
+            est_rdcost_norm);
+  }
+
   fprintf(fout, "\n");
   fclose(fout);
 }
@@ -2553,8 +2706,7 @@ static void model_rd_with_dnn(const AV1_COMP *const cpi,
   const struct macroblockd_plane *const pd = &xd->plane[plane];
   const int log_numpels = num_pels_log2_lookup[plane_bsize];
 
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
   const int q_step = AOMMAX(pd->dequant_Q3[1] >> dequant_shift, 1);
 
   const struct macroblock_plane *const p = &x->plane[plane];
@@ -2591,7 +2743,12 @@ static void model_rd_with_dnn(const AV1_COMP *const cpi,
   get_2x2_normalized_sses_and_sads(cpi, plane_bsize, src, src_stride, dst,
                                    dst_stride, src_diff, diff_stride,
                                    sse_norm_arr, NULL);
-  double mean = get_mean(src_diff, bw, bw, bh);
+  double mean;
+  if (is_cur_buf_hbd(xd)) {
+    mean = get_highbd_diff_mean(src, src_stride, dst, dst_stride, bw, bh);
+  } else {
+    mean = get_diff_mean(src, src_stride, dst, dst_stride, bw, bh);
+  }
   if (shift) {
     for (int k = 0; k < 4; ++k) sse_norm_arr[k] /= (1 << (2 * shift));
     mean /= (1 << shift);
@@ -2604,8 +2761,9 @@ static void model_rd_with_dnn(const AV1_COMP *const cpi,
   const double q_sqr = (double)(q_step * q_step);
   const double q_sqr_by_sse_norm = q_sqr / (sse_norm + 1.0);
   const double mean_sqr_by_sse_norm = mean * mean / (sse_norm + 1.0);
-  double hor_corr, vert_corr;
-  get_horver_correlation(src_diff, diff_stride, bw, bh, &hor_corr, &vert_corr);
+  float hor_corr, vert_corr;
+  av1_get_horver_correlation_full(src_diff, diff_stride, bw, bh, &hor_corr,
+                                  &vert_corr);
 
   float features[NUM_FEATURES_PUSTATS];
   features[0] = (float)hor_corr;
@@ -2620,10 +2778,10 @@ static void model_rd_with_dnn(const AV1_COMP *const cpi,
   float rate_f, dist_by_sse_norm_f;
   av1_nn_predict(features, &av1_pustats_dist_nnconfig, &dist_by_sse_norm_f);
   av1_nn_predict(features, &av1_pustats_rate_nnconfig, &rate_f);
+  aom_clear_system_state();
   const float dist_f = (float)((double)dist_by_sse_norm_f * (1.0 + sse_norm));
   int rate_i = (int)(AOMMAX(0.0, rate_f * num_samples) + 0.5);
   int64_t dist_i = (int64_t)(AOMMAX(0.0, dist_f * num_samples) + 0.5);
-  aom_clear_system_state();
 
   // Check if skip is better
   if (rate_i == 0) {
@@ -2669,7 +2827,7 @@ static void model_rd_for_sb_with_dnn(
     int bw, bh;
     get_txb_dimensions(xd, plane, plane_bsize, 0, 0, plane_bsize, NULL, NULL,
                        &bw, &bh);
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
                            pd->dst.stride, bw, bh);
     } else {
@@ -2708,8 +2866,7 @@ static void model_rd_with_surffit(const AV1_COMP *const cpi,
   (void)plane_bsize;
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
   const int qstep = AOMMAX(pd->dequant_Q3[1] >> dequant_shift, 1);
   if (sse == 0) {
     if (rate) *rate = 0;
@@ -2723,7 +2880,8 @@ static void model_rd_with_surffit(const AV1_COMP *const cpi,
   const double yl = log(sse_norm / qstepsqr) / log(2.0);
   double rate_f, dist_by_sse_norm_f;
 
-  av1_model_rd_surffit(xm, yl, &rate_f, &dist_by_sse_norm_f);
+  av1_model_rd_surffit(plane_bsize, sse_norm, xm, yl, &rate_f,
+                       &dist_by_sse_norm_f);
 
   const double dist_f = dist_by_sse_norm_f * sse_norm;
   int rate_i = (int)(AOMMAX(0.0, rate_f * num_samples) + 0.5);
@@ -2773,7 +2931,7 @@ static void model_rd_for_sb_with_surffit(
     const int shift = (xd->bd - 8);
     get_txb_dimensions(xd, plane, plane_bsize, 0, 0, plane_bsize, NULL, NULL,
                        &bw, &bh);
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
                            pd->dst.stride, bw, bh);
     } else {
@@ -2813,8 +2971,7 @@ static void model_rd_with_curvfit(const AV1_COMP *const cpi,
   (void)plane_bsize;
   const MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblockd_plane *const pd = &xd->plane[plane];
-  const int dequant_shift =
-      (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? xd->bd - 5 : 3;
+  const int dequant_shift = (is_cur_buf_hbd(xd)) ? xd->bd - 5 : 3;
   const int qstep = AOMMAX(pd->dequant_Q3[1] >> dequant_shift, 1);
 
   if (sse == 0) {
@@ -2828,7 +2985,8 @@ static void model_rd_with_curvfit(const AV1_COMP *const cpi,
   const double xqr = log(sse_norm / qstepsqr) / log(2.0);
 
   double rate_f, dist_by_sse_norm_f;
-  av1_model_rd_curvfit(xqr, &rate_f, &dist_by_sse_norm_f);
+  av1_model_rd_curvfit(plane_bsize, sse_norm, xqr, &rate_f,
+                       &dist_by_sse_norm_f);
 
   const double dist_f = dist_by_sse_norm_f * sse_norm;
   int rate_i = (int)(AOMMAX(0.0, rate_f * num_samples) + 0.5);
@@ -2879,7 +3037,7 @@ static void model_rd_for_sb_with_curvfit(
     get_txb_dimensions(xd, plane, plane_bsize, 0, 0, plane_bsize, NULL, NULL,
                        &bw, &bh);
 
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+    if (is_cur_buf_hbd(xd)) {
       sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
                            pd->dst.stride, bw, bh);
     } else {
@@ -2908,71 +3066,6 @@ static void model_rd_for_sb_with_curvfit(
   *out_dist_sum = dist_sum;
 }
 
-static void model_rd_for_sb_with_fullrdy(
-    const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
-    int plane_from, int plane_to, int mi_row, int mi_col, int *out_rate_sum,
-    int64_t *out_dist_sum, int *skip_txfm_sb, int64_t *skip_sse_sb,
-    int *plane_rate, int64_t *plane_sse, int64_t *plane_dist) {
-  const int ref = xd->mi[0]->ref_frame[0];
-
-  int64_t rate_sum = 0;
-  int64_t dist_sum = 0;
-  int64_t total_sse = 0;
-
-  for (int plane = plane_from; plane <= plane_to; ++plane) {
-    struct macroblock_plane *const p = &x->plane[plane];
-    struct macroblockd_plane *const pd = &xd->plane[plane];
-    const BLOCK_SIZE plane_bsize =
-        get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
-    const int bw = block_size_wide[plane_bsize];
-    const int bh = block_size_high[plane_bsize];
-    int64_t sse;
-    int rate;
-    int64_t dist;
-
-    if (x->skip_chroma_rd && plane) continue;
-
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-      sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
-                           pd->dst.stride, bw, bh);
-    } else {
-      sse = aom_sse(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride, bw,
-                    bh);
-    }
-    sse = ROUND_POWER_OF_TWO(sse, (xd->bd - 8) * 2);
-
-    RD_STATS rd_stats;
-    if (plane == 0) {
-      select_tx_type_yrd(cpi, x, &rd_stats, bsize, mi_row, mi_col, INT64_MAX);
-      if (rd_stats.invalid_rate) {
-        rate = 0;
-        dist = sse << 4;
-      } else {
-        rate = rd_stats.rate;
-        dist = rd_stats.dist;
-      }
-    } else {
-      model_rd_with_curvfit(cpi, x, plane_bsize, plane, sse, bw * bh, &rate,
-                            &dist);
-    }
-
-    if (plane == 0) x->pred_sse[ref] = (unsigned int)AOMMIN(sse, UINT_MAX);
-
-    total_sse += sse;
-    rate_sum += rate;
-    dist_sum += dist;
-
-    if (plane_rate) plane_rate[plane] = rate;
-    if (plane_sse) plane_sse[plane] = sse;
-    if (plane_dist) plane_dist[plane] = dist;
-  }
-
-  if (skip_txfm_sb) *skip_txfm_sb = total_sse == 0;
-  if (skip_sse_sb) *skip_sse_sb = total_sse << 4;
-  *out_rate_sum = (int)rate_sum;
-  *out_dist_sum = dist_sum;
-}
-
 static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
                                int block, int blk_row, int blk_col,
                                BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
@@ -2997,6 +3090,7 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
   tran_low_t *best_dqcoeff = this_dqcoeff;
   const int txk_type_idx =
       av1_get_txk_type_index(plane_bsize, blk_row, blk_col);
+  int perform_block_coeff_opt;
   av1_invalid_rd_stats(best_rd_stats);
 
   TXB_RD_INFO *intra_txb_rd_info = NULL;
@@ -3047,7 +3141,8 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
   TX_TYPE txk_end = TX_TYPES - 1;
   if ((!is_inter && x->use_default_intra_tx_type) ||
       (is_inter && x->use_default_inter_tx_type)) {
-    txk_start = txk_end = get_default_tx_type(0, xd, tx_size);
+    txk_start = txk_end =
+        get_default_tx_type(0, xd, tx_size, cpi->is_screen_content_type);
   } else if (x->rd_model == LOW_TXFM_RD || x->cb_partition_scan) {
     if (plane == 0) txk_end = DCT_DCT;
   }
@@ -3065,7 +3160,9 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
   }
   const uint16_t ext_tx_used_flag = av1_ext_tx_used_flag[tx_set_type];
   if (xd->lossless[mbmi->segment_id] || txsize_sqr_up_map[tx_size] > TX_32X32 ||
-      ext_tx_used_flag == 0x0001) {
+      ext_tx_used_flag == 0x0001 ||
+      (is_inter && cpi->oxcf.use_inter_dct_only) ||
+      (!is_inter && cpi->oxcf.use_intra_dct_only)) {
     txk_start = txk_end = DCT_DCT;
   }
   uint16_t allowed_tx_mask = 0;  // 1: allow; 0: skip.
@@ -3097,8 +3194,22 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
     allowed_tx_mask = (1 << txk_start);
   }
 
+  const BLOCK_SIZE tx_bsize = txsize_to_bsize[tx_size];
+  int64_t block_sse = 0;
+  unsigned int block_mse_q8 = UINT_MAX;
+  block_sse = pixel_diff_dist(x, plane, blk_row, blk_col, plane_bsize, tx_bsize,
+                              &block_mse_q8);
+  assert(block_mse_q8 != UINT_MAX);
+  if (is_cur_buf_hbd(xd)) {
+    block_sse = ROUND_POWER_OF_TWO(block_sse, (xd->bd - 8) * 2);
+    block_mse_q8 = ROUND_POWER_OF_TWO(block_mse_q8, (xd->bd - 8) * 2);
+  }
+  block_sse *= 16;
+  // Tranform domain distortion is accurate for higher residuals.
+  // TODO(any): Experiment with variance and mean based thresholds
   int use_transform_domain_distortion =
       (cpi->sf.use_transform_domain_distortion > 0) &&
+      (block_mse_q8 >= cpi->tx_domain_dist_threshold) &&
       // Any 64-pt transforms only preserves half the coefficients.
       // Therefore transform domain distortion is not valid for these
       // transform sizes.
@@ -3116,20 +3227,19 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
 
   const uint16_t *eobs_ptr = x->plane[plane].eobs;
 
-  const BLOCK_SIZE tx_bsize = txsize_to_bsize[tx_size];
-  int64_t block_sse =
-      pixel_diff_dist(x, plane, blk_row, blk_col, plane_bsize, tx_bsize);
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-    block_sse = ROUND_POWER_OF_TWO(block_sse, (xd->bd - 8) * 2);
-  block_sse *= 16;
+  // Used mse based threshold logic to take decision of R-D of optimization of
+  // coeffs. For snaller residuals, coeff optimization would be helpful. For
+  // larger residuals, R-D optimization may not be effective.
+  // TODO(any): Experiment with variance and mean based thresholds
+  perform_block_coeff_opt = (block_mse_q8 <= cpi->coeff_opt_dist_threshold);
 
   for (TX_TYPE tx_type = txk_start; tx_type <= txk_end; ++tx_type) {
     if (!(allowed_tx_mask & (1 << tx_type))) continue;
     if (plane == 0) mbmi->txk_type[txk_type_idx] = tx_type;
     RD_STATS this_rd_stats;
     av1_invalid_rd_stats(&this_rd_stats);
-
-    if (!cpi->optimize_seg_arr[mbmi->segment_id]) {
+    if ((cpi->optimize_seg_arr[mbmi->segment_id] != FULL_TRELLIS_OPT) ||
+        (!perform_block_coeff_opt)) {
       av1_xform_quant(
           cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size, tx_type,
           USE_B_QUANT_NO_TRELLIS ? AV1_XFORM_QUANT_B : AV1_XFORM_QUANT_FP);
@@ -3148,13 +3258,6 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
         const int64_t dist_cost_estimate =
             RDCOST(x->rdmult, 0, AOMMIN(this_rd_stats.dist, this_rd_stats.sse));
         if (dist_cost_estimate - (dist_cost_estimate >> 3) > best_rd_) continue;
-
-        rate_cost = av1_cost_coeffs(cm, x, plane, block, tx_size, tx_type,
-                                    txb_ctx, use_fast_coef_costing);
-        const int64_t rd_estimate =
-            AOMMIN(RDCOST(x->rdmult, rate_cost, this_rd_stats.dist),
-                   RDCOST(x->rdmult, 0, this_rd_stats.sse));
-        if (rd_estimate - (rd_estimate >> 3) > best_rd_) continue;
       }
       av1_optimize_b(cpi, x, plane, block, tx_size, tx_type, txb_ctx, 1,
                      &rate_cost);
@@ -3197,6 +3300,41 @@ static int64_t search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
     }
 #endif  // CONFIG_COLLECT_RD_STATS == 1
 
+#if COLLECT_TX_SIZE_DATA
+    // Generate small sample to restrict output size.
+    static unsigned int seed = 21743;
+    if (lcg_rand16(&seed) % 200 == 0) {
+      FILE *fp = NULL;
+
+      if (within_border) {
+        fp = fopen(av1_tx_size_data_output_file, "a");
+      }
+
+      if (fp) {
+        // Transform info and RD
+        const int txb_w = tx_size_wide[tx_size];
+        const int txb_h = tx_size_high[tx_size];
+
+        // Residue signal.
+        const int diff_stride = block_size_wide[plane_bsize];
+        struct macroblock_plane *const p = &x->plane[plane];
+        const int16_t *src_diff =
+            &p->src_diff[(blk_row * diff_stride + blk_col) * 4];
+
+        for (int r = 0; r < txb_h; ++r) {
+          for (int c = 0; c < txb_w; ++c) {
+            fprintf(fp, "%d,", src_diff[c]);
+          }
+          src_diff += diff_stride;
+        }
+
+        fprintf(fp, "%d,%d,%d,%" PRId64, txb_w, txb_h, tx_type, rd);
+        fprintf(fp, "\n");
+        fclose(fp);
+      }
+    }
+#endif  // COLLECT_TX_SIZE_DATA
+
     if (cpi->sf.adaptive_txb_search_level) {
       if ((best_rd - (best_rd >> cpi->sf.adaptive_txb_search_level)) >
           ref_best_rd) {
@@ -3247,7 +3385,7 @@ RECON_INTRA:
     // if the last search tx_type is the best tx_type, we don't need to
     // do this again
     if (best_tx_type != last_tx_type) {
-      if (!cpi->optimize_seg_arr[mbmi->segment_id]) {
+      if (cpi->optimize_seg_arr[mbmi->segment_id] != FULL_TRELLIS_OPT) {
         av1_xform_quant(
             cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
             best_tx_type,
@@ -3283,12 +3421,11 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
   struct rdcost_block_args *args = arg;
   MACROBLOCK *const x = args->x;
   MACROBLOCKD *const xd = &x->e_mbd;
-  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  const int is_inter = is_inter_block(xd->mi[0]);
   const AV1_COMP *cpi = args->cpi;
   ENTROPY_CONTEXT *a = args->t_above + blk_col;
   ENTROPY_CONTEXT *l = args->t_left + blk_row;
   const AV1_COMMON *cm = &cpi->common;
-  int64_t rd1, rd2, rd;
   RD_STATS this_rd_stats;
 
   av1_init_rd_stats(&this_rd_stats);
@@ -3298,7 +3435,7 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
     return;
   }
 
-  if (!is_inter_block(mbmi)) {
+  if (!is_inter) {
     av1_predict_intra_block_facade(cm, xd, plane, blk_col, blk_row, tx_size);
     av1_subtract_txb(x, plane, plane_bsize, blk_col, blk_row, tx_size);
   }
@@ -3309,7 +3446,7 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
                   args->best_rd - args->this_rd, &this_rd_stats);
 
   if (plane == AOM_PLANE_Y && xd->cfl.store_y) {
-    assert(!is_inter_block(mbmi) || plane_bsize < BLOCK_8X8);
+    assert(!is_inter || plane_bsize < BLOCK_8X8);
     cfl_store_tx(xd, blk_row, blk_col, tx_size, plane_bsize);
   }
 
@@ -3328,28 +3465,33 @@ static void block_rd_txfm(int plane, int block, int blk_row, int blk_col,
   else
     set_blk_skip(x, plane, blk_idx, 0);
 
-  rd1 = RDCOST(x->rdmult, this_rd_stats.rate, this_rd_stats.dist);
-  rd2 = RDCOST(x->rdmult, 0, this_rd_stats.sse);
+  const int64_t rd1 = RDCOST(x->rdmult, this_rd_stats.rate, this_rd_stats.dist);
+  const int64_t rd2 = RDCOST(x->rdmult, 0, this_rd_stats.sse);
 
   // TODO(jingning): temporarily enabled only for luma component
-  rd = AOMMIN(rd1, rd2);
+  const int64_t rd = AOMMIN(rd1, rd2);
 
   this_rd_stats.skip &= !x->plane[plane].eobs[block];
 
+#if CONFIG_ONE_PASS_SVM
+  if (plane == AOM_PLANE_Y && plane_bsize >= BLOCK_8X8) {
+    const int eob = x->plane[plane].eobs[block];
+    av1_add_reg_stat(&this_rd_stats, eob, rd, this_rd_stats.sse, blk_row,
+                     blk_col, plane_bsize, txsize_to_bsize[tx_size]);
+  }
+#endif
+
   av1_merge_rd_stats(&args->rd_stats, &this_rd_stats);
 
   args->this_rd += rd;
 
-  if (args->this_rd > args->best_rd) {
-    args->exit_early = 1;
-    return;
-  }
+  if (args->this_rd > args->best_rd) args->exit_early = 1;
 }
 
 static void txfm_rd_in_plane(MACROBLOCK *x, const AV1_COMP *cpi,
-                             RD_STATS *rd_stats, int64_t ref_best_rd, int plane,
-                             BLOCK_SIZE bsize, TX_SIZE tx_size,
-                             int use_fast_coef_casting,
+                             RD_STATS *rd_stats, int64_t ref_best_rd,
+                             int64_t this_rd, int plane, BLOCK_SIZE bsize,
+                             TX_SIZE tx_size, int use_fast_coef_casting,
                              FAST_TX_SEARCH_MODE ftxs_mode) {
   MACROBLOCKD *const xd = &x->e_mbd;
   const struct macroblockd_plane *const pd = &xd->plane[plane];
@@ -3360,12 +3502,22 @@ static void txfm_rd_in_plane(MACROBLOCK *x, const AV1_COMP *cpi,
   args.best_rd = ref_best_rd;
   args.use_fast_coef_costing = use_fast_coef_casting;
   args.ftxs_mode = ftxs_mode;
+  args.this_rd = this_rd;
   av1_init_rd_stats(&args.rd_stats);
 
+  if (!cpi->oxcf.enable_tx64 && txsize_sqr_up_map[tx_size] == TX_64X64) {
+    av1_invalid_rd_stats(rd_stats);
+    return;
+  }
+
   if (plane == 0) xd->mi[0]->tx_size = tx_size;
 
   av1_get_entropy_contexts(bsize, pd, args.t_above, args.t_left);
 
+  if (args.this_rd > args.best_rd) {
+    args.exit_early = 1;
+  }
+
   av1_foreach_transformed_block_in_plane(xd, bsize, plane, block_rd_txfm,
                                          &args);
 
@@ -3382,18 +3534,14 @@ static void txfm_rd_in_plane(MACROBLOCK *x, const AV1_COMP *cpi,
 
 static int tx_size_cost(const AV1_COMMON *const cm, const MACROBLOCK *const x,
                         BLOCK_SIZE bsize, TX_SIZE tx_size) {
-  const MACROBLOCKD *const xd = &x->e_mbd;
-  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  assert(bsize == x->e_mbd.mi[0]->sb_type);
+  if (cm->tx_mode != TX_MODE_SELECT || !block_signals_txsize(bsize)) return 0;
 
-  if (cm->tx_mode == TX_MODE_SELECT && block_signals_txsize(mbmi->sb_type)) {
-    const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize);
-    const int depth = tx_size_to_depth(tx_size, bsize);
-    const int tx_size_ctx = get_tx_size_context(xd);
-    int r_tx_size = x->tx_size_cost[tx_size_cat][tx_size_ctx][depth];
-    return r_tx_size;
-  } else {
-    return 0;
-  }
+  const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize);
+  const int depth = tx_size_to_depth(tx_size, bsize);
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const int tx_size_ctx = get_tx_size_context(xd);
+  return x->tx_size_cost[tx_size_cat][tx_size_ctx][depth];
 }
 
 static int64_t txfm_yrd(const AV1_COMP *const cpi, MACROBLOCK *x,
@@ -3418,43 +3566,79 @@ static int64_t txfm_yrd(const AV1_COMP *const cpi, MACROBLOCK *x,
   s0 = x->skip_cost[skip_ctx][0];
   s1 = x->skip_cost[skip_ctx][1];
 
+  int64_t skip_rd;
+  int64_t this_rd;
+
+  if (is_inter) {
+    skip_rd = RDCOST(x->rdmult, s1, 0);
+    this_rd = RDCOST(x->rdmult, s0 + r_tx_size * tx_select, 0);
+  } else {
+    skip_rd = RDCOST(x->rdmult, s1 + r_tx_size * tx_select, 0);
+    this_rd = RDCOST(x->rdmult, s0 + r_tx_size * tx_select, 0);
+  }
+
   mbmi->tx_size = tx_size;
-  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, AOM_PLANE_Y, bs, tx_size,
-                   cpi->sf.use_fast_coef_costing, ftxs_mode);
+  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, AOMMIN(this_rd, skip_rd),
+                   AOM_PLANE_Y, bs, tx_size, cpi->sf.use_fast_coef_costing,
+                   ftxs_mode);
   if (rd_stats->rate == INT_MAX) return INT64_MAX;
 
+  // rdstats->rate should include all the rate except skip/non-skip cost as the
+  // same is accounted in the caller functions after rd evaluation of all
+  // planes. However the decisions should be done after considering the
+  // skip/non-skip header cost
   if (rd_stats->skip) {
     if (is_inter) {
       rd = RDCOST(x->rdmult, s1, rd_stats->sse);
+#if CONFIG_ONE_PASS_SVM
+      // TODO(chiyotsai@google.com): Investigate if these updates are really
+      // needed.
+      av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+#endif
     } else {
       rd = RDCOST(x->rdmult, s1 + r_tx_size * tx_select, rd_stats->sse);
+#if CONFIG_ONE_PASS_SVM
+      av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+#endif
+      rd_stats->rate += r_tx_size * tx_select;
     }
   } else {
     rd = RDCOST(x->rdmult, rd_stats->rate + s0 + r_tx_size * tx_select,
                 rd_stats->dist);
+    rd_stats->rate += r_tx_size * tx_select;
+  }
+  if (is_inter && !xd->lossless[xd->mi[0]->segment_id]) {
+    int64_t temp_skip_rd = RDCOST(x->rdmult, s1, rd_stats->sse);
+    if (temp_skip_rd <= rd) {
+      rd = temp_skip_rd;
+      rd_stats->rate = 0;
+      rd_stats->dist = rd_stats->sse;
+      rd_stats->skip = 1;
+    }
   }
-
-  if (tx_select) rd_stats->rate += r_tx_size;
-
-  if (is_inter && !xd->lossless[xd->mi[0]->segment_id] && !(rd_stats->skip))
-    rd = AOMMIN(rd, RDCOST(x->rdmult, s1, rd_stats->sse));
 
   return rd;
 }
 
 static int64_t estimate_yrd_for_sb(const AV1_COMP *const cpi, BLOCK_SIZE bs,
-                                   MACROBLOCK *x, int *r, int64_t *d, int *s,
-                                   int64_t *sse, int64_t ref_best_rd) {
-  RD_STATS rd_stats;
+                                   MACROBLOCK *x, int64_t ref_best_rd,
+                                   RD_STATS *rd_stats) {
   av1_subtract_plane(x, bs, 0);
   x->rd_model = LOW_TXFM_RD;
-  int64_t rd = txfm_yrd(cpi, x, &rd_stats, ref_best_rd, bs,
-                        max_txsize_rect_lookup[bs], FTXS_NONE);
+  const int64_t rd = txfm_yrd(cpi, x, rd_stats, ref_best_rd, bs,
+                              max_txsize_rect_lookup[bs], FTXS_NONE);
   x->rd_model = FULL_TXFM_RD;
-  *r = rd_stats.rate;
-  *d = rd_stats.dist;
-  *s = rd_stats.skip;
-  *sse = rd_stats.sse;
+  if (rd != INT64_MAX) {
+    MACROBLOCKD *const xd = &x->e_mbd;
+    const int skip_ctx = av1_get_skip_context(xd);
+    if (rd_stats->skip) {
+      const int s1 = x->skip_cost[skip_ctx][1];
+      rd_stats->rate = s1;
+    } else {
+      const int s0 = x->skip_cost[skip_ctx][0];
+      rd_stats->rate += s0;
+    }
+  }
   return rd;
 }
 
@@ -3469,8 +3653,18 @@ static void choose_largest_tx_size(const AV1_COMP *const cpi, MACROBLOCK *x,
   const TxSetType tx_set_type =
       av1_get_ext_tx_set_type(mbmi->tx_size, is_inter, cm->reduced_tx_set_used);
   prune_tx(cpi, bs, x, xd, tx_set_type);
-  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, AOM_PLANE_Y, bs,
-                   mbmi->tx_size, cpi->sf.use_fast_coef_costing, FTXS_NONE);
+  const int skip_ctx = av1_get_skip_context(xd);
+  int s0, s1;
+
+  s0 = x->skip_cost[skip_ctx][0];
+  s1 = x->skip_cost[skip_ctx][1];
+
+  int64_t skip_rd = RDCOST(x->rdmult, s1, 0);
+  int64_t this_rd = RDCOST(x->rdmult, s0, 0);
+
+  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, AOMMIN(this_rd, skip_rd),
+                   AOM_PLANE_Y, bs, mbmi->tx_size,
+                   cpi->sf.use_fast_coef_costing, FTXS_NONE);
   // Reset the pruning flags.
   av1_zero(x->tx_search_prune);
   x->tx_split_prune_flag = 0;
@@ -3483,7 +3677,8 @@ static void choose_smallest_tx_size(const AV1_COMP *const cpi, MACROBLOCK *x,
   MB_MODE_INFO *const mbmi = xd->mi[0];
 
   mbmi->tx_size = TX_4X4;
-  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, 0, bs, mbmi->tx_size,
+  // TODO(any) : Pass this_rd based on skip/non-skip cost
+  txfm_rd_in_plane(x, cpi, rd_stats, ref_best_rd, 0, 0, bs, mbmi->tx_size,
                    cpi->sf.use_fast_coef_costing, FTXS_NONE);
 }
 
@@ -3514,22 +3709,15 @@ static int get_search_init_depth(int mi_width, int mi_height, int is_inter,
 static void choose_tx_size_type_from_rd(const AV1_COMP *const cpi,
                                         MACROBLOCK *x, RD_STATS *rd_stats,
                                         int64_t ref_best_rd, BLOCK_SIZE bs) {
+  av1_invalid_rd_stats(rd_stats);
+
   const AV1_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
-  int64_t rd = INT64_MAX;
-  int n;
+  const TX_SIZE max_rect_tx_size = max_txsize_rect_lookup[bs];
+  const int tx_select = cm->tx_mode == TX_MODE_SELECT;
   int start_tx;
   int depth;
-  int64_t best_rd = INT64_MAX;
-  const TX_SIZE max_rect_tx_size = max_txsize_rect_lookup[bs];
-  TX_SIZE best_tx_size = max_rect_tx_size;
-  TX_TYPE best_txk_type[TXK_TYPE_BUF_LEN];
-  uint8_t best_blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE];
-  const int n4 = bsize_to_num_blk(bs);
-  const int tx_select = cm->tx_mode == TX_MODE_SELECT;
-
-  av1_invalid_rd_stats(rd_stats);
 
   if (tx_select) {
     start_tx = max_rect_tx_size;
@@ -3543,16 +3731,24 @@ static void choose_tx_size_type_from_rd(const AV1_COMP *const cpi,
 
   prune_tx(cpi, bs, x, xd, EXT_TX_SET_ALL16);
 
-  for (n = start_tx; depth <= MAX_TX_DEPTH; depth++, n = sub_tx_size_map[n]) {
+  TX_TYPE best_txk_type[TXK_TYPE_BUF_LEN];
+  uint8_t best_blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE];
+  TX_SIZE best_tx_size = max_rect_tx_size;
+  int64_t best_rd = INT64_MAX;
+  const int n4 = bsize_to_num_blk(bs);
+  x->rd_model = FULL_TXFM_RD;
+  for (int n = start_tx; depth <= MAX_TX_DEPTH;
+       depth++, n = sub_tx_size_map[n]) {
 #if CONFIG_DIST_8X8
     if (x->using_dist_8x8) {
       if (tx_size_wide[n] < 8 || tx_size_high[n] < 8) continue;
     }
 #endif
+    if (!cpi->oxcf.enable_tx64 && txsize_sqr_up_map[n] == TX_64X64) continue;
+
     RD_STATS this_rd_stats;
-    if (mbmi->ref_mv_idx > 0) x->rd_model = LOW_TXFM_RD;
-    rd = txfm_yrd(cpi, x, &this_rd_stats, ref_best_rd, bs, n, FTXS_NONE);
-    x->rd_model = FULL_TXFM_RD;
+    const int64_t rd =
+        txfm_yrd(cpi, x, &this_rd_stats, ref_best_rd, bs, n, FTXS_NONE);
 
     if (rd < best_rd) {
       memcpy(best_txk_type, mbmi->txk_type,
@@ -4196,21 +4392,13 @@ static const uint8_t mode_to_angle_bin[INTRA_MODES] = {
 };
 /* clang-format on */
 
-static void angle_estimation(const uint8_t *src, int src_stride, int rows,
-                             int cols, BLOCK_SIZE bsize,
-                             uint8_t *directional_mode_skip_mask) {
-  memset(directional_mode_skip_mask, 0,
-         INTRA_MODES * sizeof(*directional_mode_skip_mask));
-  // Check if angle_delta is used
-  if (!av1_use_angle_delta(bsize)) return;
-  uint64_t hist[DIRECTIONAL_MODES];
-  memset(hist, 0, DIRECTIONAL_MODES * sizeof(hist[0]));
+static void get_gradient_hist(const uint8_t *src, int src_stride, int rows,
+                              int cols, uint64_t *hist) {
   src += src_stride;
-  int r, c, dx, dy;
-  for (r = 1; r < rows; ++r) {
-    for (c = 1; c < cols; ++c) {
-      dx = src[c] - src[c - 1];
-      dy = src[c] - src[c - src_stride];
+  for (int r = 1; r < rows; ++r) {
+    for (int c = 1; c < cols; ++c) {
+      int dx = src[c] - src[c - 1];
+      int dy = src[c] - src[c - src_stride];
       int index;
       const int temp = dx * dx + dy * dy;
       if (dy == 0) {
@@ -4227,45 +4415,16 @@ static void angle_estimation(const uint8_t *src, int src_stride, int rows,
     }
     src += src_stride;
   }
-
-  int i;
-  uint64_t hist_sum = 0;
-  for (i = 0; i < DIRECTIONAL_MODES; ++i) hist_sum += hist[i];
-  for (i = 0; i < INTRA_MODES; ++i) {
-    if (av1_is_directional_mode(i)) {
-      const uint8_t angle_bin = mode_to_angle_bin[i];
-      uint64_t score = 2 * hist[angle_bin];
-      int weight = 2;
-      if (angle_bin > 0) {
-        score += hist[angle_bin - 1];
-        ++weight;
-      }
-      if (angle_bin < DIRECTIONAL_MODES - 1) {
-        score += hist[angle_bin + 1];
-        ++weight;
-      }
-      if (score * ANGLE_SKIP_THRESH < hist_sum * weight)
-        directional_mode_skip_mask[i] = 1;
-    }
-  }
 }
 
-static void highbd_angle_estimation(const uint8_t *src8, int src_stride,
-                                    int rows, int cols, BLOCK_SIZE bsize,
-                                    uint8_t *directional_mode_skip_mask) {
-  memset(directional_mode_skip_mask, 0,
-         INTRA_MODES * sizeof(*directional_mode_skip_mask));
-  // Check if angle_delta is used
-  if (!av1_use_angle_delta(bsize)) return;
+static void get_highbd_gradient_hist(const uint8_t *src8, int src_stride,
+                                     int rows, int cols, uint64_t *hist) {
   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
-  uint64_t hist[DIRECTIONAL_MODES];
-  memset(hist, 0, DIRECTIONAL_MODES * sizeof(hist[0]));
   src += src_stride;
-  int r, c, dx, dy;
-  for (r = 1; r < rows; ++r) {
-    for (c = 1; c < cols; ++c) {
-      dx = src[c] - src[c - 1];
-      dy = src[c] - src[c - src_stride];
+  for (int r = 1; r < rows; ++r) {
+    for (int c = 1; c < cols; ++c) {
+      int dx = src[c] - src[c - 1];
+      int dy = src[c] - src[c - src_stride];
       int index;
       const int temp = dx * dx + dy * dy;
       if (dy == 0) {
@@ -4282,6 +4441,19 @@ static void highbd_angle_estimation(const uint8_t *src8, int src_stride,
     }
     src += src_stride;
   }
+}
+
+static void angle_estimation(const uint8_t *src, int src_stride, int rows,
+                             int cols, BLOCK_SIZE bsize, int is_hbd,
+                             uint8_t *directional_mode_skip_mask) {
+  // Check if angle_delta is used
+  if (!av1_use_angle_delta(bsize)) return;
+
+  uint64_t hist[DIRECTIONAL_MODES] = { 0 };
+  if (is_hbd)
+    get_highbd_gradient_hist(src, src_stride, rows, cols, hist);
+  else
+    get_gradient_hist(src, src_stride, rows, cols, hist);
 
   int i;
   uint64_t hist_sum = 0;
@@ -4299,8 +4471,8 @@ static void highbd_angle_estimation(const uint8_t *src8, int src_stride,
         score += hist[angle_bin + 1];
         ++weight;
       }
-      if (score * ANGLE_SKIP_THRESH < hist_sum * weight)
-        directional_mode_skip_mask[i] = 1;
+      const int thresh = 10;
+      if (score * thresh < hist_sum * weight) directional_mode_skip_mask[i] = 1;
     }
   }
 }
@@ -4353,13 +4525,12 @@ static int64_t rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
   const int rows = block_size_high[bsize];
   const int cols = block_size_wide[bsize];
   int is_directional_mode;
-  uint8_t directional_mode_skip_mask[INTRA_MODES];
-  const int src_stride = x->plane[0].src.stride;
-  const uint8_t *src = x->plane[0].src.buf;
+  uint8_t directional_mode_skip_mask[INTRA_MODES] = { 0 };
   int beat_best_rd = 0;
   const int *bmode_costs;
   PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
   const int try_palette =
+      cpi->oxcf.enable_palette &&
       av1_allow_palette(cpi->common.allow_screen_content_tools, mbmi->sb_type);
   uint8_t *best_palette_color_map =
       try_palette ? x->palette_buffer->best_palette_color_map : NULL;
@@ -4372,12 +4543,12 @@ static int64_t rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
   bmode_costs = x->y_mode_costs[above_ctx][left_ctx];
 
   mbmi->angle_delta[PLANE_TYPE_Y] = 0;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-    highbd_angle_estimation(src, src_stride, rows, cols, bsize,
-                            directional_mode_skip_mask);
-  else
-    angle_estimation(src, src_stride, rows, cols, bsize,
+  if (cpi->sf.intra_angle_estimation) {
+    const int src_stride = x->plane[0].src.stride;
+    const uint8_t *src = x->plane[0].src.buf;
+    angle_estimation(src, src_stride, rows, cols, bsize, is_cur_buf_hbd(xd),
                      directional_mode_skip_mask);
+  }
   mbmi->filter_intra_mode_info.use_filter_intra = 0;
   pmi->palette_size[0] = 0;
 
@@ -4393,6 +4564,11 @@ static int64_t rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
     int this_rate, this_rate_tokenonly, s;
     int64_t this_distortion, this_rd, this_model_rd;
     mbmi->mode = intra_rd_search_mode_order[mode_idx];
+    if (!cpi->oxcf.enable_smooth_intra &&
+        (mbmi->mode == SMOOTH_PRED || mbmi->mode == SMOOTH_H_PRED ||
+         mbmi->mode == SMOOTH_V_PRED))
+      continue;
+    if (!cpi->oxcf.enable_paeth_intra && mbmi->mode == PAETH_PRED) continue;
     mbmi->angle_delta[PLANE_TYPE_Y] = 0;
     this_model_rd =
         intra_model_yrd(cpi, x, bsize, bmode_costs[mbmi->mode], mi_row, mi_col);
@@ -4402,7 +4578,8 @@ static int64_t rd_pick_intra_sby_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
     if (this_model_rd < best_model_rd) best_model_rd = this_model_rd;
     is_directional_mode = av1_is_directional_mode(mbmi->mode);
     if (is_directional_mode && directional_mode_skip_mask[mbmi->mode]) continue;
-    if (is_directional_mode && av1_use_angle_delta(bsize)) {
+    if (is_directional_mode && av1_use_angle_delta(bsize) &&
+        cpi->oxcf.enable_angle_delta) {
       this_rd_stats.rate = INT_MAX;
       rd_pick_intra_angle_sby(cpi, x, mi_row, mi_col, &this_rate,
                               &this_rd_stats, bsize, bmode_costs[mbmi->mode],
@@ -4497,7 +4674,7 @@ static int super_block_uvrd(const AV1_COMP *const cpi, MACROBLOCK *x,
   if (is_cost_valid) {
     for (plane = 1; plane < MAX_MB_PLANE; ++plane) {
       RD_STATS pn_rd_stats;
-      txfm_rd_in_plane(x, cpi, &pn_rd_stats, ref_best_rd, plane, bsize,
+      txfm_rd_in_plane(x, cpi, &pn_rd_stats, ref_best_rd, 0, plane, bsize,
                        uv_tx_size, cpi->sf.use_fast_coef_costing, FTXS_NONE);
       if (pn_rd_stats.rate == INT_MAX) {
         is_cost_valid = 0;
@@ -4520,11 +4697,12 @@ static int super_block_uvrd(const AV1_COMP *const cpi, MACROBLOCK *x,
   return is_cost_valid;
 }
 
-static void tx_block_rd_b(const AV1_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
-                          int blk_row, int blk_col, int plane, int block,
-                          int plane_bsize, TXB_CTX *txb_ctx, RD_STATS *rd_stats,
-                          FAST_TX_SEARCH_MODE ftxs_mode, int64_t ref_rdcost,
-                          TXB_RD_INFO *rd_info_array) {
+// Pick transform type for a transform block of tx_size.
+static void tx_type_rd(const AV1_COMP *cpi, MACROBLOCK *x, TX_SIZE tx_size,
+                       int blk_row, int blk_col, int plane, int block,
+                       int plane_bsize, TXB_CTX *txb_ctx, RD_STATS *rd_stats,
+                       FAST_TX_SEARCH_MODE ftxs_mode, int64_t ref_rdcost,
+                       TXB_RD_INFO *rd_info_array) {
   const struct macroblock_plane *const p = &x->plane[plane];
   const uint16_t cur_joint_ctx =
       (txb_ctx->dc_sign_ctx << 8) + txb_ctx->txb_skip_ctx;
@@ -4675,6 +4853,7 @@ static int ml_predict_tx_split(MACROBLOCK *x, BLOCK_SIZE bsize, int blk_row,
 
   float score = 0.0f;
   av1_nn_predict(features, nn_config, &score);
+  aom_clear_system_state();
   if (score > 8.0f) return 100;
   if (score < -8.0f) return 0;
   score = 1.0f / (1.0f + (float)exp(-score));
@@ -4716,9 +4895,9 @@ static void try_tx_block_no_split(
   rd_stats->zero_rate = zero_blk_rate;
   const int index = av1_get_txb_size_index(plane_bsize, blk_row, blk_col);
   mbmi->inter_tx_size[index] = tx_size;
-  tx_block_rd_b(cpi, x, tx_size, blk_row, blk_col, 0, block, plane_bsize,
-                &txb_ctx, rd_stats, ftxs_mode, ref_best_rd,
-                rd_info_node != NULL ? rd_info_node->rd_info_array : NULL);
+  tx_type_rd(cpi, x, tx_size, blk_row, blk_col, 0, block, plane_bsize, &txb_ctx,
+             rd_stats, ftxs_mode, ref_best_rd,
+             rd_info_node != NULL ? rd_info_node->rd_info_array : NULL);
   assert(rd_stats->rate < INT_MAX);
 
   if ((RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist) >=
@@ -4726,7 +4905,7 @@ static void try_tx_block_no_split(
        rd_stats->skip == 1) &&
       !xd->lossless[mbmi->segment_id]) {
 #if CONFIG_RD_DEBUG
-    av1_update_txb_coeff_cost(rd_stats, plane, tx_size, blk_row, blk_col,
+    av1_update_txb_coeff_cost(rd_stats, 0, tx_size, blk_row, blk_col,
                               zero_blk_rate - rd_stats->rate);
 #endif  // CONFIG_RD_DEBUG
     rd_stats->rate = zero_blk_rate;
@@ -4749,6 +4928,13 @@ static void try_tx_block_no_split(
   const int txk_type_idx =
       av1_get_txk_type_index(plane_bsize, blk_row, blk_col);
   no_split->tx_type = mbmi->txk_type[txk_type_idx];
+
+#if CONFIG_ONE_PASS_SVM
+  if (plane_bsize >= BLOCK_8X8) {
+    av1_add_reg_stat(rd_stats, p->eobs[block], no_split->rd, rd_stats->sse,
+                     blk_row, blk_col, plane_bsize, txsize_to_bsize[tx_size]);
+  }
+#endif
 }
 
 static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
@@ -4756,8 +4942,8 @@ static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
                             BLOCK_SIZE plane_bsize, ENTROPY_CONTEXT *ta,
                             ENTROPY_CONTEXT *tl, TXFM_CONTEXT *tx_above,
                             TXFM_CONTEXT *tx_left, RD_STATS *rd_stats,
-                            int64_t ref_best_rd, int *is_cost_valid,
-                            FAST_TX_SEARCH_MODE ftxs_mode,
+                            int64_t prev_level_rd, int64_t ref_best_rd,
+                            int *is_cost_valid, FAST_TX_SEARCH_MODE ftxs_mode,
                             TXB_RD_INFO_NODE *rd_info_node);
 
 static void try_tx_block_split(
@@ -4767,6 +4953,7 @@ static void try_tx_block_split(
     int txfm_partition_ctx, int64_t no_split_rd, int64_t ref_best_rd,
     FAST_TX_SEARCH_MODE ftxs_mode, TXB_RD_INFO_NODE *rd_info_node,
     RD_STATS *split_rd_stats, int64_t *split_rd) {
+  assert(tx_size < TX_SIZES_ALL);
   MACROBLOCKD *const xd = &x->e_mbd;
   const int max_blocks_high = max_block_high(xd, plane_bsize, 0);
   const int max_blocks_wide = max_block_wide(xd, plane_bsize, 0);
@@ -4774,44 +4961,37 @@ static void try_tx_block_split(
   const int bsw = tx_size_wide_unit[sub_txs];
   const int bsh = tx_size_high_unit[sub_txs];
   const int sub_step = bsw * bsh;
-  RD_STATS this_rd_stats;
-  int this_cost_valid = 1;
+  const int nblks =
+      (tx_size_high_unit[tx_size] / bsh) * (tx_size_wide_unit[tx_size] / bsw);
+  assert(nblks > 0);
+  int blk_idx = 0;
   int64_t tmp_rd = 0;
-
+  *split_rd = INT64_MAX;
   split_rd_stats->rate = x->txfm_partition_cost[txfm_partition_ctx][1];
 
-  assert(tx_size < TX_SIZES_ALL);
-
-  int blk_idx = 0;
   for (int r = 0; r < tx_size_high_unit[tx_size]; r += bsh) {
     for (int c = 0; c < tx_size_wide_unit[tx_size]; c += bsw, ++blk_idx) {
+      assert(blk_idx < 4);
       const int offsetr = blk_row + r;
       const int offsetc = blk_col + c;
       if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
-      assert(blk_idx < 4);
+
+      RD_STATS this_rd_stats;
+      int this_cost_valid = 1;
       select_tx_block(
           cpi, x, offsetr, offsetc, block, sub_txs, depth + 1, plane_bsize, ta,
-          tl, tx_above, tx_left, &this_rd_stats, ref_best_rd - tmp_rd,
-          &this_cost_valid, ftxs_mode,
+          tl, tx_above, tx_left, &this_rd_stats, no_split_rd / nblks,
+          ref_best_rd - tmp_rd, &this_cost_valid, ftxs_mode,
           (rd_info_node != NULL) ? rd_info_node->children[blk_idx] : NULL);
-
-      if (!this_cost_valid) goto LOOP_EXIT;
-
+      if (!this_cost_valid) return;
       av1_merge_rd_stats(split_rd_stats, &this_rd_stats);
-
       tmp_rd = RDCOST(x->rdmult, split_rd_stats->rate, split_rd_stats->dist);
-
-      if (no_split_rd < tmp_rd) {
-        this_cost_valid = 0;
-        goto LOOP_EXIT;
-      }
+      if (no_split_rd < tmp_rd) return;
       block += sub_step;
     }
   }
 
-LOOP_EXIT : {}
-
-  if (this_cost_valid) *split_rd = tmp_rd;
+  *split_rd = tmp_rd;
 }
 
 // Search for the best tx partition/type for a given luma block.
@@ -4820,8 +5000,8 @@ static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
                             BLOCK_SIZE plane_bsize, ENTROPY_CONTEXT *ta,
                             ENTROPY_CONTEXT *tl, TXFM_CONTEXT *tx_above,
                             TXFM_CONTEXT *tx_left, RD_STATS *rd_stats,
-                            int64_t ref_best_rd, int *is_cost_valid,
-                            FAST_TX_SEARCH_MODE ftxs_mode,
+                            int64_t prev_level_rd, int64_t ref_best_rd,
+                            int *is_cost_valid, FAST_TX_SEARCH_MODE ftxs_mode,
                             TXB_RD_INFO_NODE *rd_info_node) {
   assert(tx_size < TX_SIZES_ALL);
   av1_init_rd_stats(rd_stats);
@@ -4841,7 +5021,8 @@ static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
                                          mbmi->sb_type, tx_size);
   struct macroblock_plane *const p = &x->plane[0];
 
-  const int try_no_split = 1;
+  const int try_no_split =
+      cpi->oxcf.enable_tx64 || txsize_sqr_up_map[tx_size] != TX_64X64;
   int try_split = tx_size > TX_4X4 && depth < MAX_VARTX_DEPTH;
 #if CONFIG_DIST_8X8
   if (x->using_dist_8x8)
@@ -4866,6 +5047,13 @@ static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
     if (cpi->sf.txb_split_cap) {
       if (p->eobs[block] == 0) try_split = 0;
     }
+
+    if (cpi->sf.adaptive_txb_search_level &&
+        (no_split.rd -
+         (no_split.rd >> (2 + cpi->sf.adaptive_txb_search_level))) >
+            prev_level_rd) {
+      try_split = 0;
+    }
   }
 
   if (x->e_mbd.bd == 8 && !x->cb_partition_scan && try_split) {
@@ -4913,95 +5101,12 @@ static void select_tx_block(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
   }
 }
 
-static void select_inter_block_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
-                                   RD_STATS *rd_stats, BLOCK_SIZE bsize,
-                                   int64_t ref_best_rd,
-                                   FAST_TX_SEARCH_MODE ftxs_mode,
-                                   TXB_RD_INFO_NODE *rd_info_tree) {
-  MACROBLOCKD *const xd = &x->e_mbd;
-  int is_cost_valid = 1;
-  int64_t this_rd = 0, skip_rd = 0;
-
-  if (ref_best_rd < 0) is_cost_valid = 0;
-
-  av1_init_rd_stats(rd_stats);
-
-  if (is_cost_valid) {
-    const struct macroblockd_plane *const pd = &xd->plane[0];
-    const BLOCK_SIZE plane_bsize =
-        get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
-    const int mi_width = mi_size_wide[plane_bsize];
-    const int mi_height = mi_size_high[plane_bsize];
-    const TX_SIZE max_tx_size = max_txsize_rect_lookup[plane_bsize];
-    const int bh = tx_size_high_unit[max_tx_size];
-    const int bw = tx_size_wide_unit[max_tx_size];
-    int idx, idy;
-    int block = 0;
-    int step = tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];
-    ENTROPY_CONTEXT ctxa[MAX_MIB_SIZE];
-    ENTROPY_CONTEXT ctxl[MAX_MIB_SIZE];
-    TXFM_CONTEXT tx_above[MAX_MIB_SIZE];
-    TXFM_CONTEXT tx_left[MAX_MIB_SIZE];
-
-    RD_STATS pn_rd_stats;
-    const int init_depth =
-        get_search_init_depth(mi_width, mi_height, 1, &cpi->sf);
-    av1_init_rd_stats(&pn_rd_stats);
-
-    av1_get_entropy_contexts(bsize, pd, ctxa, ctxl);
-    memcpy(tx_above, xd->above_txfm_context, sizeof(TXFM_CONTEXT) * mi_width);
-    memcpy(tx_left, xd->left_txfm_context, sizeof(TXFM_CONTEXT) * mi_height);
-    const int skip_ctx = av1_get_skip_context(xd);
-    const int s0 = x->skip_cost[skip_ctx][0];
-    const int s1 = x->skip_cost[skip_ctx][1];
-
-    skip_rd = RDCOST(x->rdmult, s1, 0);
-    this_rd = RDCOST(x->rdmult, s0, 0);
-    for (idy = 0; idy < mi_height; idy += bh) {
-      for (idx = 0; idx < mi_width; idx += bw) {
-        int64_t best_rd_sofar = (ref_best_rd - (AOMMIN(skip_rd, this_rd)));
-        select_tx_block(cpi, x, idy, idx, block, max_tx_size, init_depth,
-                        plane_bsize, ctxa, ctxl, tx_above, tx_left,
-                        &pn_rd_stats, best_rd_sofar, &is_cost_valid, ftxs_mode,
-                        rd_info_tree);
-        if (!is_cost_valid || pn_rd_stats.rate == INT_MAX) {
-          av1_invalid_rd_stats(rd_stats);
-          return;
-        }
-        av1_merge_rd_stats(rd_stats, &pn_rd_stats);
-        skip_rd = RDCOST(x->rdmult, s1, rd_stats->sse);
-        this_rd = RDCOST(x->rdmult, rd_stats->rate + s0, rd_stats->dist);
-        block += step;
-        if (rd_info_tree != NULL) rd_info_tree += 1;
-      }
-    }
-    if (skip_rd <= this_rd) {
-      rd_stats->rate = 0;
-      rd_stats->dist = rd_stats->sse;
-      rd_stats->skip = 1;
-    } else {
-      rd_stats->skip = 0;
-    }
-  }
-
-  if (!is_cost_valid) {
-    // reset cost value
-    av1_invalid_rd_stats(rd_stats);
-  }
-}
-
-static int64_t select_tx_size_fix_type(const AV1_COMP *cpi, MACROBLOCK *x,
+static int64_t select_tx_size_and_type(const AV1_COMP *cpi, MACROBLOCK *x,
                                        RD_STATS *rd_stats, BLOCK_SIZE bsize,
                                        int64_t ref_best_rd,
                                        TXB_RD_INFO_NODE *rd_info_tree) {
-  const int fast_tx_search = cpi->sf.tx_size_search_method > USE_FULL_RD;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = xd->mi[0];
-  const int is_inter = is_inter_block(mbmi);
-  const int skip_ctx = av1_get_skip_context(xd);
-  int s0 = x->skip_cost[skip_ctx][0];
-  int s1 = x->skip_cost[skip_ctx][1];
-  int64_t rd;
+  assert(is_inter_block(xd->mi[0]));
 
   // TODO(debargha): enable this as a speed feature where the
   // select_inter_block_yrd() function above will use a simplified search
@@ -5009,16 +5114,74 @@ static int64_t select_tx_size_fix_type(const AV1_COMP *cpi, MACROBLOCK *x,
   // will use more complex search given that the transform partitions have
   // already been decided.
 
+  const int fast_tx_search = cpi->sf.tx_size_search_method > USE_FULL_RD;
   int64_t rd_thresh = ref_best_rd;
   if (fast_tx_search && rd_thresh < INT64_MAX) {
     if (INT64_MAX - rd_thresh > (rd_thresh >> 3)) rd_thresh += (rd_thresh >> 3);
   }
   assert(rd_thresh > 0);
 
-  FAST_TX_SEARCH_MODE ftxs_mode =
+  const FAST_TX_SEARCH_MODE ftxs_mode =
       fast_tx_search ? FTXS_DCT_AND_1D_DCT_ONLY : FTXS_NONE;
-  select_inter_block_yrd(cpi, x, rd_stats, bsize, rd_thresh, ftxs_mode,
-                         rd_info_tree);
+  const struct macroblockd_plane *const pd = &xd->plane[0];
+  const BLOCK_SIZE plane_bsize =
+      get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
+  const int mi_width = mi_size_wide[plane_bsize];
+  const int mi_height = mi_size_high[plane_bsize];
+  ENTROPY_CONTEXT ctxa[MAX_MIB_SIZE];
+  ENTROPY_CONTEXT ctxl[MAX_MIB_SIZE];
+  TXFM_CONTEXT tx_above[MAX_MIB_SIZE];
+  TXFM_CONTEXT tx_left[MAX_MIB_SIZE];
+  av1_get_entropy_contexts(bsize, pd, ctxa, ctxl);
+  memcpy(tx_above, xd->above_txfm_context, sizeof(TXFM_CONTEXT) * mi_width);
+  memcpy(tx_left, xd->left_txfm_context, sizeof(TXFM_CONTEXT) * mi_height);
+
+  const int skip_ctx = av1_get_skip_context(xd);
+  const int s0 = x->skip_cost[skip_ctx][0];
+  const int s1 = x->skip_cost[skip_ctx][1];
+  const int init_depth =
+      get_search_init_depth(mi_width, mi_height, 1, &cpi->sf);
+  const TX_SIZE max_tx_size = max_txsize_rect_lookup[plane_bsize];
+  const int bh = tx_size_high_unit[max_tx_size];
+  const int bw = tx_size_wide_unit[max_tx_size];
+  const int step = bw * bh;
+  int64_t skip_rd = RDCOST(x->rdmult, s1, 0);
+  int64_t this_rd = RDCOST(x->rdmult, s0, 0);
+  int block = 0;
+
+  av1_init_rd_stats(rd_stats);
+  for (int idy = 0; idy < mi_height; idy += bh) {
+    for (int idx = 0; idx < mi_width; idx += bw) {
+      const int64_t best_rd_sofar =
+          (rd_thresh == INT64_MAX) ? INT64_MAX
+                                   : (rd_thresh - (AOMMIN(skip_rd, this_rd)));
+      int is_cost_valid = 1;
+      RD_STATS pn_rd_stats;
+      select_tx_block(cpi, x, idy, idx, block, max_tx_size, init_depth,
+                      plane_bsize, ctxa, ctxl, tx_above, tx_left, &pn_rd_stats,
+                      INT64_MAX, best_rd_sofar, &is_cost_valid, ftxs_mode,
+                      rd_info_tree);
+      if (!is_cost_valid || pn_rd_stats.rate == INT_MAX) {
+        av1_invalid_rd_stats(rd_stats);
+        return INT64_MAX;
+      }
+      av1_merge_rd_stats(rd_stats, &pn_rd_stats);
+      skip_rd = RDCOST(x->rdmult, s1, rd_stats->sse);
+      this_rd = RDCOST(x->rdmult, rd_stats->rate + s0, rd_stats->dist);
+      block += step;
+      if (rd_info_tree != NULL) rd_info_tree += 1;
+    }
+  }
+
+  if (skip_rd <= this_rd) {
+    rd_stats->skip = 1;
+#if CONFIG_ONE_PASS_SVM
+    av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+#endif
+  } else {
+    rd_stats->skip = 0;
+  }
+
   if (rd_stats->rate == INT_MAX) return INT64_MAX;
 
   // If fast_tx_search is true, only DCT and 1D DCT were tested in
@@ -5029,13 +5192,19 @@ static int64_t select_tx_size_fix_type(const AV1_COMP *cpi, MACROBLOCK *x,
       return INT64_MAX;
   }
 
-  if (rd_stats->skip)
+  int64_t rd;
+  if (rd_stats->skip) {
     rd = RDCOST(x->rdmult, s1, rd_stats->sse);
-  else
+#if CONFIG_ONE_PASS_SVM
+    // TODO(chiyotsai@google.com): Investigate if these updates are really
+    // needed.
+    av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+#endif
+  } else {
     rd = RDCOST(x->rdmult, rd_stats->rate + s0, rd_stats->dist);
-
-  if (is_inter && !xd->lossless[xd->mi[0]->segment_id] && !(rd_stats->skip))
-    rd = AOMMIN(rd, RDCOST(x->rdmult, s1, rd_stats->sse));
+    if (!xd->lossless[xd->mi[0]->segment_id])
+      rd = AOMMIN(rd, RDCOST(x->rdmult, s1, rd_stats->sse));
+  }
 
   return rd;
 }
@@ -5075,8 +5244,8 @@ static void tx_block_yrd(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
                                   .txb_skip_cost[txb_ctx.txb_skip_ctx][1];
     rd_stats->zero_rate = zero_blk_rate;
     rd_stats->ref_rdcost = ref_best_rd;
-    tx_block_rd_b(cpi, x, tx_size, blk_row, blk_col, 0, block, plane_bsize,
-                  &txb_ctx, rd_stats, ftxs_mode, ref_best_rd, NULL);
+    tx_type_rd(cpi, x, tx_size, blk_row, blk_col, 0, block, plane_bsize,
+               &txb_ctx, rd_stats, ftxs_mode, ref_best_rd, NULL);
     const int mi_width = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
     if (RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist) >=
             RDCOST(x->rdmult, zero_blk_rate, rd_stats->sse) ||
@@ -5089,9 +5258,20 @@ static void tx_block_yrd(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
       x->plane[0].txb_entropy_ctx[block] = 0;
       update_txk_array(mbmi->txk_type, plane_bsize, blk_row, blk_col, tx_size,
                        DCT_DCT);
+#if CONFIG_ONE_PASS_SVM
+      av1_add_reg_stat(rd_stats, 0, RDCOST(x->rdmult, 0, rd_stats->sse),
+                       rd_stats->sse, blk_row, blk_col, plane_bsize,
+                       txsize_to_bsize[tx_size]);
+#endif
     } else {
       rd_stats->skip = 0;
       set_blk_skip(x, 0, blk_row * mi_width + blk_col, 0);
+#if CONFIG_ONE_PASS_SVM
+      av1_add_reg_stat(rd_stats, x->plane[0].eobs[block],
+                       RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist),
+                       rd_stats->sse, blk_row, blk_col, plane_bsize,
+                       txsize_to_bsize[tx_size]);
+#endif
     }
     if (tx_size > TX_4X4 && depth < MAX_VARTX_DEPTH)
       rd_stats->rate += x->txfm_partition_cost[ctx][0];
@@ -5199,6 +5379,11 @@ static int inter_block_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
     rd_stats->rate = 0;
     rd_stats->dist = rd_stats->sse;
     rd_stats->skip = 1;
+#if CONFIG_ONE_PASS_SVM
+    // TODO(chiyotasi@google.com): Investigate if these updates are really
+    // needed.
+    av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+#endif
   }
   if (this_rd > ref_best_rd) is_cost_valid = 0;
 
@@ -5542,7 +5727,8 @@ static int predict_skip_flag(MACROBLOCK *x, BLOCK_SIZE bsize, int64_t *dist,
   const MACROBLOCKD *xd = &x->e_mbd;
   const int16_t dc_q = av1_dc_quant_QTX(x->qindex, 0, xd->bd);
 
-  *dist = pixel_diff_dist(x, 0, 0, 0, bsize, bsize);
+  *dist = pixel_diff_dist(x, 0, 0, 0, bsize, bsize, NULL);
+
   const int64_t mse = *dist / bw / bh;
   // Normalized quantizer takes the transform upscaling factor (8 for tx size
   // smaller than 32) into account.
@@ -5559,7 +5745,7 @@ static int predict_skip_flag(MACROBLOCK *x, BLOCK_SIZE bsize, int64_t *dist,
   param.tx_type = DCT_DCT;
   param.tx_size = max_tx_size;
   param.bd = xd->bd;
-  param.is_hbd = get_bitdepth_data_path_index(xd);
+  param.is_hbd = is_cur_buf_hbd(xd);
   param.lossless = 0;
   param.tx_set_type = av1_get_ext_tx_set_type(
       param.tx_size, is_inter_block(xd->mi[0]), reduced_tx_set);
@@ -5586,6 +5772,42 @@ static int predict_skip_flag(MACROBLOCK *x, BLOCK_SIZE bsize, int64_t *dist,
   return 1;
 }
 
+#if CONFIG_ONE_PASS_SVM
+static void calc_regional_sse(MACROBLOCK *x, BLOCK_SIZE bsize, int64_t dist,
+                              RD_STATS *rd_stats) {
+  // TODO(chiyotsai@google.com): Don't need regional sse's unless we are doing
+  // none.
+  const int bw = block_size_wide[bsize];
+  const int bw_mi = bw >> tx_size_wide_log2[0];
+  const int bh_mi = bw >> tx_size_high_log2[0];
+  const BLOCK_SIZE split_size = get_partition_subsize(bsize, PARTITION_SPLIT);
+  int64_t dist_0, dist_1, dist_2, dist_3;
+  MACROBLOCKD *xd = &x->e_mbd;
+  dist_0 = pixel_diff_dist(x, AOM_PLANE_Y, 0, 0, bsize, split_size, NULL);
+  dist_1 =
+      pixel_diff_dist(x, AOM_PLANE_Y, 0, bw_mi / 2, bsize, split_size, NULL);
+  dist_2 =
+      pixel_diff_dist(x, AOM_PLANE_Y, bh_mi / 2, 0, bsize, split_size, NULL);
+  dist_3 = pixel_diff_dist(x, AOM_PLANE_Y, bh_mi / 2, bw_mi / 2, bsize,
+                           split_size, NULL);
+
+  if (is_cur_buf_hbd(xd)) {
+    dist = ROUND_POWER_OF_TWO(dist, (xd->bd - 8) * 2);
+    dist_0 = ROUND_POWER_OF_TWO(dist_0, (xd->bd - 8) * 2);
+    dist_1 = ROUND_POWER_OF_TWO(dist_1, (xd->bd - 8) * 2);
+    dist_2 = ROUND_POWER_OF_TWO(dist_2, (xd->bd - 8) * 2);
+    dist_3 = ROUND_POWER_OF_TWO(dist_3, (xd->bd - 8) * 2);
+  }
+  const int scaling_factor = MAX_MIB_SIZE * MAX_MIB_SIZE;
+  rd_stats->y_sse = (dist << 4);
+  rd_stats->sse_0 = (dist_0 << 4) * scaling_factor;
+  rd_stats->sse_1 = (dist_1 << 4) * scaling_factor;
+  rd_stats->sse_2 = (dist_2 << 4) * scaling_factor;
+  rd_stats->sse_3 = (dist_3 << 4) * scaling_factor;
+  av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+}
+#endif
+
 // Used to set proper context for early termination with skip = 1.
 static void set_skip_flag(MACROBLOCK *x, RD_STATS *rd_stats, int bsize,
                           int64_t dist) {
@@ -5598,31 +5820,41 @@ static void set_skip_flag(MACROBLOCK *x, RD_STATS *rd_stats, int bsize,
   mbmi->tx_size = tx_size;
   for (int i = 0; i < n4; ++i) set_blk_skip(x, 0, i, 1);
   rd_stats->skip = 1;
-  rd_stats->rate = 0;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-    dist = ROUND_POWER_OF_TWO(dist, (xd->bd - 8) * 2);
+  if (is_cur_buf_hbd(xd)) dist = ROUND_POWER_OF_TWO(dist, (xd->bd - 8) * 2);
   rd_stats->dist = rd_stats->sse = (dist << 4);
+  // Though decision is to make the block as skip based on luma stats,
+  // it is possible that block becomes non skip after chroma rd. In addition
+  // intermediate non skip costs calculated by caller function will be
+  // incorrect, if rate is set as  zero (i.e., if zero_blk_rate is not
+  // accounted). Hence intermediate rate is populated to code the luma tx blks
+  // as skip, the caller function based on final rd decision (i.e., skip vs
+  // non-skip) sets the final rate accordingly. Here the rate populated
+  // corresponds to coding all the tx blocks with zero_blk_rate (based on max tx
+  // size possible) in the current block. Eg: For 128*128 block, rate would be
+  // 4 * zero_blk_rate where zero_blk_rate corresponds to coding of one 64x64 tx
+  // block as 'all zeros'
+  ENTROPY_CONTEXT ctxa[MAX_MIB_SIZE];
+  ENTROPY_CONTEXT ctxl[MAX_MIB_SIZE];
+  av1_get_entropy_contexts(bsize, &xd->plane[0], ctxa, ctxl);
+  ENTROPY_CONTEXT *ta = ctxa;
+  ENTROPY_CONTEXT *tl = ctxl;
+  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
+  TXB_CTX txb_ctx;
+  get_txb_ctx(bsize, tx_size, 0, ta, tl, &txb_ctx);
+  const int zero_blk_rate = x->coeff_costs[txs_ctx][PLANE_TYPE_Y]
+                                .txb_skip_cost[txb_ctx.txb_skip_ctx][1];
+  rd_stats->rate = zero_blk_rate *
+                   (block_size_wide[bsize] >> tx_size_wide_log2[tx_size]) *
+                   (block_size_high[bsize] >> tx_size_high_log2[tx_size]);
 }
 
-static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
-                               RD_STATS *rd_stats, BLOCK_SIZE bsize, int mi_row,
-                               int mi_col, int64_t ref_best_rd) {
+// Search for best transform size and type for luma inter blocks.
+static void pick_tx_size_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
+                                  RD_STATS *rd_stats, BLOCK_SIZE bsize,
+                                  int mi_row, int mi_col, int64_t ref_best_rd) {
   const AV1_COMMON *cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = xd->mi[0];
-  int64_t rd = INT64_MAX;
-  int64_t best_rd = INT64_MAX;
-  const int is_inter = is_inter_block(mbmi);
-  const int n4 = bsize_to_num_blk(bsize);
-  // Get the tx_size 1 level down
-  const TX_SIZE min_tx_size = sub_tx_size_map[max_txsize_rect_lookup[bsize]];
-  const TxSetType tx_set_type =
-      av1_get_ext_tx_set_type(min_tx_size, is_inter, cm->reduced_tx_set_used);
-  const int within_border =
-      mi_row >= xd->tile.mi_row_start &&
-      (mi_row + mi_size_high[bsize] < xd->tile.mi_row_end) &&
-      mi_col >= xd->tile.mi_col_start &&
-      (mi_col + mi_size_wide[bsize] < xd->tile.mi_col_end);
+  assert(is_inter_block(xd->mi[0]));
 
   av1_invalid_rd_stats(rd_stats);
 
@@ -5639,8 +5871,7 @@ static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
     // tighter.
     assert(cpi->sf.model_based_prune_tx_search_level >= 0 &&
            cpi->sf.model_based_prune_tx_search_level <= 2);
-    static const int prune_factor_by8[] = { 2 + MODELRD_TYPE_TX_SEARCH_PRUNE,
-                                            4 + MODELRD_TYPE_TX_SEARCH_PRUNE };
+    static const int prune_factor_by8[] = { 3, 5 };
     if (!model_skip &&
         ((model_rd *
           prune_factor_by8[cpi->sf.model_based_prune_tx_search_level - 1]) >>
@@ -5648,18 +5879,29 @@ static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
       return;
   }
 
-  const uint32_t hash = get_block_residue_hash(x, bsize);
-  MB_RD_RECORD *mb_rd_record = &x->mb_rd_record;
-
-  if (ref_best_rd != INT64_MAX && within_border && cpi->sf.use_mb_rd_hash) {
-    for (int i = 0; i < mb_rd_record->num; ++i) {
-      const int index = (mb_rd_record->index_start + i) % RD_RECORD_BUFFER_LEN;
-      // If there is a match in the tx_rd_record, fetch the RD decision and
-      // terminate early.
-      if (mb_rd_record->tx_rd_info[index].hash_value == hash) {
-        MB_RD_INFO *tx_rd_info = &mb_rd_record->tx_rd_info[index];
-        fetch_tx_rd_info(n4, tx_rd_info, rd_stats, x);
-        return;
+  uint32_t hash = 0;
+  MB_RD_RECORD *mb_rd_record = NULL;
+  const int within_border =
+      mi_row >= xd->tile.mi_row_start &&
+      (mi_row + mi_size_high[bsize] < xd->tile.mi_row_end) &&
+      mi_col >= xd->tile.mi_col_start &&
+      (mi_col + mi_size_wide[bsize] < xd->tile.mi_col_end);
+  const int is_mb_rd_hash_enabled = (within_border && cpi->sf.use_mb_rd_hash);
+  const int n4 = bsize_to_num_blk(bsize);
+  if (is_mb_rd_hash_enabled) {
+    hash = get_block_residue_hash(x, bsize);
+    mb_rd_record = &x->mb_rd_record;
+    if (ref_best_rd != INT64_MAX) {
+      for (int i = 0; i < mb_rd_record->num; ++i) {
+        const int index =
+            (mb_rd_record->index_start + i) % RD_RECORD_BUFFER_LEN;
+        // If there is a match in the tx_rd_record, fetch the RD decision and
+        // terminate early.
+        if (mb_rd_record->tx_rd_info[index].hash_value == hash) {
+          MB_RD_INFO *tx_rd_info = &mb_rd_record->tx_rd_info[index];
+          fetch_tx_rd_info(n4, tx_rd_info, rd_stats, x);
+          return;
+        }
       }
     }
   }
@@ -5667,13 +5909,23 @@ static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
   // If we predict that skip is the optimal RD decision - set the respective
   // context and terminate early.
   int64_t dist;
-  if (is_inter && cpi->sf.tx_type_search.use_skip_flag_prediction &&
+  if (cpi->sf.tx_type_search.use_skip_flag_prediction &&
       predict_skip_flag(x, bsize, &dist, cm->reduced_tx_set_used)) {
     set_skip_flag(x, rd_stats, bsize, dist);
+#if CONFIG_ONE_PASS_SVM
+    if (bsize >= BLOCK_8X8 && mi_size_wide[bsize] == mi_size_high[bsize] &&
+        xd->mi[0]->partition == PARTITION_NONE) {
+      calc_regional_sse(x, bsize, dist, rd_stats);
+    }
+#endif
     // Save the RD search results into tx_rd_record.
-    if (within_border) save_tx_rd_info(n4, hash, x, rd_stats, mb_rd_record);
+    if (is_mb_rd_hash_enabled)
+      save_tx_rd_info(n4, hash, x, rd_stats, mb_rd_record);
     return;
   }
+#if CONFIG_SPEED_STATS
+  ++x->tx_search_count;
+#endif  // CONFIG_SPEED_STATS
 
   // Precompute residual hashes and find existing or add new RD records to
   // store and reuse rate and distortion values to speed up TX size search.
@@ -5684,20 +5936,20 @@ static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
         find_tx_size_rd_records(x, bsize, mi_row, mi_col, matched_rd_info);
   }
 
+  // Get the tx_size 1 level down
+  const TX_SIZE min_tx_size = sub_tx_size_map[max_txsize_rect_lookup[bsize]];
+  const TxSetType tx_set_type =
+      av1_get_ext_tx_set_type(min_tx_size, 1, cm->reduced_tx_set_used);
   prune_tx(cpi, bsize, x, xd, tx_set_type);
 
   int found = 0;
-
   RD_STATS this_rd_stats;
   av1_init_rd_stats(&this_rd_stats);
+  const int64_t rd =
+      select_tx_size_and_type(cpi, x, &this_rd_stats, bsize, ref_best_rd,
+                              found_rd_info ? matched_rd_info : NULL);
 
-  rd = select_tx_size_fix_type(cpi, x, &this_rd_stats, bsize, ref_best_rd,
-                               found_rd_info ? matched_rd_info : NULL);
-  assert(IMPLIES(this_rd_stats.skip && !this_rd_stats.invalid_rate,
-                 this_rd_stats.rate == 0));
-
-  ref_best_rd = AOMMIN(rd, ref_best_rd);
-  if (rd < best_rd) {
+  if (rd < INT64_MAX) {
     *rd_stats = this_rd_stats;
     found = 1;
   }
@@ -5713,127 +5965,76 @@ static void select_tx_type_yrd(const AV1_COMP *cpi, MACROBLOCK *x,
   if (!found) return;
 
   // Save the RD search results into tx_rd_record.
-  if (within_border && cpi->sf.use_mb_rd_hash)
+  if (is_mb_rd_hash_enabled) {
+    assert(mb_rd_record != NULL);
     save_tx_rd_info(n4, hash, x, rd_stats, mb_rd_record);
+  }
 }
 
-static void tx_block_uvrd(const AV1_COMP *cpi, MACROBLOCK *x, int blk_row,
-                          int blk_col, int plane, int block, TX_SIZE tx_size,
-                          BLOCK_SIZE plane_bsize, ENTROPY_CONTEXT *above_ctx,
-                          ENTROPY_CONTEXT *left_ctx, RD_STATS *rd_stats,
-                          FAST_TX_SEARCH_MODE ftxs_mode) {
-  assert(plane > 0);
-  assert(tx_size < TX_SIZES_ALL);
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = xd->mi[0];
-  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
-  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
-  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
+static void model_rd_for_sb_with_fullrdy(
+    const AV1_COMP *const cpi, BLOCK_SIZE bsize, MACROBLOCK *x, MACROBLOCKD *xd,
+    int plane_from, int plane_to, int mi_row, int mi_col, int *out_rate_sum,
+    int64_t *out_dist_sum, int *skip_txfm_sb, int64_t *skip_sse_sb,
+    int *plane_rate, int64_t *plane_sse, int64_t *plane_dist) {
+  const int ref = xd->mi[0]->ref_frame[0];
 
-  ENTROPY_CONTEXT *ta = above_ctx + blk_col;
-  ENTROPY_CONTEXT *tl = left_ctx + blk_row;
-  TXB_CTX txb_ctx;
-  get_txb_ctx(plane_bsize, tx_size, plane, ta, tl, &txb_ctx);
-  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
-  const int zero_blk_rate = x->coeff_costs[txs_ctx][PLANE_TYPE_UV]
-                                .txb_skip_cost[txb_ctx.txb_skip_ctx][1];
-  tx_block_rd_b(cpi, x, tx_size, blk_row, blk_col, plane, block, plane_bsize,
-                &txb_ctx, rd_stats, ftxs_mode, INT64_MAX, NULL);
+  int64_t rate_sum = 0;
+  int64_t dist_sum = 0;
+  int64_t total_sse = 0;
 
-  const int mi_width = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
-  const int blk_idx = blk_row * mi_width + blk_col;
+  for (int plane = plane_from; plane <= plane_to; ++plane) {
+    struct macroblock_plane *const p = &x->plane[plane];
+    struct macroblockd_plane *const pd = &xd->plane[plane];
+    const BLOCK_SIZE plane_bsize =
+        get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
+    const int bw = block_size_wide[plane_bsize];
+    const int bh = block_size_high[plane_bsize];
+    int64_t sse;
+    int rate;
+    int64_t dist;
 
-  av1_set_txb_context(x, plane, block, tx_size, ta, tl);
-  if ((RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist) >=
-           RDCOST(x->rdmult, zero_blk_rate, rd_stats->sse) ||
-       rd_stats->skip == 1) &&
-      !xd->lossless[mbmi->segment_id]) {
-    rd_stats->rate = zero_blk_rate;
-    rd_stats->dist = rd_stats->sse;
-  }
+    if (x->skip_chroma_rd && plane) continue;
 
-  // Set chroma blk_skip to 0
-  set_blk_skip(x, plane, blk_idx, 0);
-}
-
-// Return value 0: early termination triggered, no valid rd cost available;
-//              1: rd cost values are valid.
-static int inter_block_uvrd(const AV1_COMP *cpi, MACROBLOCK *x,
-                            RD_STATS *rd_stats, BLOCK_SIZE bsize,
-                            int64_t non_skip_ref_best_rd,
-                            int64_t skip_ref_best_rd,
-                            FAST_TX_SEARCH_MODE ftxs_mode) {
-  MACROBLOCKD *const xd = &x->e_mbd;
-  MB_MODE_INFO *const mbmi = xd->mi[0];
-  int plane;
-  int is_cost_valid = 1;
-  int64_t this_rd = 0;
-  int64_t skip_rd = 0;
-
-  if ((non_skip_ref_best_rd < 0) && (skip_ref_best_rd < 0)) is_cost_valid = 0;
-
-  av1_init_rd_stats(rd_stats);
-
-  if (x->skip_chroma_rd) {
-    if (!is_cost_valid) av1_invalid_rd_stats(rd_stats);
-
-    return is_cost_valid;
-  }
-
-  const BLOCK_SIZE bsizec = scale_chroma_bsize(
-      bsize, xd->plane[1].subsampling_x, xd->plane[1].subsampling_y);
-
-  if (is_inter_block(mbmi) && is_cost_valid) {
-    for (plane = 1; plane < MAX_MB_PLANE; ++plane)
-      av1_subtract_plane(x, bsizec, plane);
-  }
-
-  if (is_cost_valid) {
-    for (plane = 1; plane < MAX_MB_PLANE; ++plane) {
-      const struct macroblockd_plane *const pd = &xd->plane[plane];
-      const BLOCK_SIZE plane_bsize =
-          get_plane_block_size(bsizec, pd->subsampling_x, pd->subsampling_y);
-      const int mi_width = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
-      const int mi_height =
-          block_size_high[plane_bsize] >> tx_size_high_log2[0];
-      const TX_SIZE max_tx_size = get_vartx_max_txsize(xd, plane_bsize, plane);
-      const int bh = tx_size_high_unit[max_tx_size];
-      const int bw = tx_size_wide_unit[max_tx_size];
-      int idx, idy;
-      int block = 0;
-      const int step = bh * bw;
-      ENTROPY_CONTEXT ta[MAX_MIB_SIZE];
-      ENTROPY_CONTEXT tl[MAX_MIB_SIZE];
-      av1_get_entropy_contexts(bsizec, pd, ta, tl);
-
-      for (idy = 0; idy < mi_height; idy += bh) {
-        for (idx = 0; idx < mi_width; idx += bw) {
-          RD_STATS pn_rd_stats;
-          av1_init_rd_stats(&pn_rd_stats);
-          tx_block_uvrd(cpi, x, idy, idx, plane, block, max_tx_size,
-                        plane_bsize, ta, tl, &pn_rd_stats, ftxs_mode);
-          if (pn_rd_stats.rate == INT_MAX) {
-            av1_invalid_rd_stats(rd_stats);
-            return 0;
-          }
-          av1_merge_rd_stats(rd_stats, &pn_rd_stats);
-          this_rd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
-          skip_rd = RDCOST(x->rdmult, 0, rd_stats->sse);
-          if ((this_rd > non_skip_ref_best_rd) &&
-              (skip_rd > skip_ref_best_rd)) {
-            av1_invalid_rd_stats(rd_stats);
-            return 0;
-          }
-          block += step;
-        }
-      }
+    if (is_cur_buf_hbd(xd)) {
+      sse = aom_highbd_sse(p->src.buf, p->src.stride, pd->dst.buf,
+                           pd->dst.stride, bw, bh);
+    } else {
+      sse = aom_sse(p->src.buf, p->src.stride, pd->dst.buf, pd->dst.stride, bw,
+                    bh);
     }
-  } else {
-    // reset cost value
-    av1_invalid_rd_stats(rd_stats);
+    sse = ROUND_POWER_OF_TWO(sse, (xd->bd - 8) * 2);
+
+    RD_STATS rd_stats;
+    if (plane == 0) {
+      pick_tx_size_type_yrd(cpi, x, &rd_stats, bsize, mi_row, mi_col,
+                            INT64_MAX);
+      if (rd_stats.invalid_rate) {
+        rate = 0;
+        dist = sse << 4;
+      } else {
+        rate = rd_stats.rate;
+        dist = rd_stats.dist;
+      }
+    } else {
+      model_rd_with_curvfit(cpi, x, plane_bsize, plane, sse, bw * bh, &rate,
+                            &dist);
+    }
+
+    if (plane == 0) x->pred_sse[ref] = (unsigned int)AOMMIN(sse, UINT_MAX);
+
+    total_sse += sse;
+    rate_sum += rate;
+    dist_sum += dist;
+
+    if (plane_rate) plane_rate[plane] = rate;
+    if (plane_sse) plane_sse[plane] = sse;
+    if (plane_dist) plane_dist[plane] = dist;
   }
 
-  return is_cost_valid;
+  if (skip_txfm_sb) *skip_txfm_sb = total_sse == 0;
+  if (skip_sse_sb) *skip_sse_sb = total_sse << 4;
+  *out_rate_sum = (int)rate_sum;
+  *out_dist_sum = dist_sum;
 }
 
 static void rd_pick_palette_intra_sbuv(const AV1_COMP *const cpi, MACROBLOCK *x,
@@ -6081,7 +6282,7 @@ static int cfl_rd_pick_alpha(MACROBLOCK *const x, const AV1_COMP *const cpi,
 
   const BLOCK_SIZE bsize = mbmi->sb_type;
 #if CONFIG_DEBUG
-  assert(is_cfl_allowed(xd));
+  assert(is_cfl_allowed(xd) && cpi->oxcf.enable_cfl_intra);
   const int ssx = xd->plane[AOM_PLANE_U].subsampling_x;
   const int ssy = xd->plane[AOM_PLANE_U].subsampling_y;
   const BLOCK_SIZE plane_bsize = get_plane_block_size(mbmi->sb_type, ssx, ssy);
@@ -6117,8 +6318,8 @@ static int cfl_rd_pick_alpha(MACROBLOCK *const x, const AV1_COMP *const cpi,
       if (i == CFL_SIGN_NEG) {
         mbmi->cfl_alpha_idx = 0;
         mbmi->cfl_alpha_signs = joint_sign;
-        txfm_rd_in_plane(x, cpi, &rd_stats, best_rd, plane + 1, bsize, tx_size,
-                         cpi->sf.use_fast_coef_costing, FTXS_NONE);
+        txfm_rd_in_plane(x, cpi, &rd_stats, best_rd, 0, plane + 1, bsize,
+                         tx_size, cpi->sf.use_fast_coef_costing, FTXS_NONE);
         if (rd_stats.rate == INT_MAX) break;
       }
       const int alpha_rate = x->cfl_cost[joint_sign][plane][0];
@@ -6145,7 +6346,7 @@ static int cfl_rd_pick_alpha(MACROBLOCK *const x, const AV1_COMP *const cpi,
           if (i == 0) {
             mbmi->cfl_alpha_idx = (c << CFL_ALPHABET_SIZE_LOG2) + c;
             mbmi->cfl_alpha_signs = joint_sign;
-            txfm_rd_in_plane(x, cpi, &rd_stats, best_rd, plane + 1, bsize,
+            txfm_rd_in_plane(x, cpi, &rd_stats, best_rd, 0, plane + 1, bsize,
                              tx_size, cpi->sf.use_fast_coef_costing, FTXS_NONE);
             if (rd_stats.rate == INT_MAX) break;
           }
@@ -6223,14 +6424,15 @@ static int64_t rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
     mbmi->uv_mode = mode;
     int cfl_alpha_rate = 0;
     if (mode == UV_CFL_PRED) {
-      if (!is_cfl_allowed(xd)) continue;
+      if (!is_cfl_allowed(xd) || !cpi->oxcf.enable_cfl_intra) continue;
       assert(!is_directional_mode);
       const TX_SIZE uv_tx_size = av1_get_tx_size(AOM_PLANE_U, xd);
       cfl_alpha_rate = cfl_rd_pick_alpha(x, cpi, uv_tx_size, best_rd);
       if (cfl_alpha_rate == INT_MAX) continue;
     }
     mbmi->angle_delta[PLANE_TYPE_UV] = 0;
-    if (is_directional_mode && av1_use_angle_delta(mbmi->sb_type)) {
+    if (is_directional_mode && av1_use_angle_delta(mbmi->sb_type) &&
+        cpi->oxcf.enable_angle_delta) {
       const int rate_overhead =
           x->intra_uv_mode_cost[is_cfl_allowed(xd)][mbmi->mode][mode];
       if (!rd_pick_intra_angle_sbuv(cpi, x, bsize, rate_overhead, best_rd,
@@ -6247,7 +6449,7 @@ static int64_t rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
     this_rate = tokenonly_rd_stats.rate +
                 intra_mode_info_cost_uv(cpi, x, mbmi, bsize, mode_cost);
     if (mode == UV_CFL_PRED) {
-      assert(is_cfl_allowed(xd));
+      assert(is_cfl_allowed(xd) && cpi->oxcf.enable_cfl_intra);
 #if CONFIG_DEBUG
       if (!xd->lossless[mbmi->segment_id])
         assert(xd->cfl.rate == tokenonly_rd_stats.rate + mode_cost);
@@ -6266,6 +6468,7 @@ static int64_t rd_pick_intra_sbuv_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
   }
 
   const int try_palette =
+      cpi->oxcf.enable_palette &&
       av1_allow_palette(cpi->common.allow_screen_content_tools, mbmi->sb_type);
   if (try_palette) {
     uint8_t *best_palette_color_map = x->palette_buffer->best_palette_color_map;
@@ -6443,7 +6646,7 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
   const int p_row = ((mi_row * MI_SIZE) >> pd->subsampling_y) + 4 * ir;
 
   ConvolveParams conv_params = get_conv_params(0, plane, xd->bd);
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
   WarpTypesAllowed warp_types[2];
   for (ref = 0; ref < 2; ++ref) {
     const WarpedMotionParams *const wm =
@@ -6466,8 +6669,6 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
   uint8_t *second_pred = get_buf_by_bd(xd, second_pred16);
   (void)ref_mv_sub8x8;
 
-  const int have_newmv = have_nearmv_in_inter_mode(mbmi->mode);
-  const int ref_mv_idx = mbmi->ref_mv_idx + (have_newmv ? 1 : 0);
   MV *const best_mv = &x->best_mv.as_mv;
   const int search_range = SEARCH_RANGE_8P;
   const int sadpb = x->sadperbit16;
@@ -6486,7 +6687,7 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
       } else {
         int_mv cur_int_mv, init_int_mv;
         cur_int_mv.as_mv.col = cur_mv[id].as_mv.col >> 3;
-        cur_int_mv.as_mv.row = cur_mv[id].as_mv.col >> 3;
+        cur_int_mv.as_mv.row = cur_mv[id].as_mv.row >> 3;
         init_int_mv.as_mv.row = init_mv[id].as_mv.row >> 3;
         init_int_mv.as_mv.col = init_mv[id].as_mv.col >> 3;
         if (cur_int_mv.as_int == init_int_mv.as_int) {
@@ -6532,9 +6733,9 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
                               mi_row * MI_SIZE, xd, cm->allow_warped_motion);
 
     const int order_idx = id != 0;
-    av1_jnt_comp_weight_assign(cm, mbmi, order_idx, &xd->jcp_param.fwd_offset,
-                               &xd->jcp_param.bck_offset,
-                               &xd->jcp_param.use_jnt_comp_avg, 1);
+    av1_dist_wtd_comp_weight_assign(
+        cm, mbmi, order_idx, &xd->jcp_param.fwd_offset,
+        &xd->jcp_param.bck_offset, &xd->jcp_param.use_dist_wtd_comp_avg, 1);
 
     // Do full-pixel compound motion search on the current reference frame.
     if (id) xd->plane[plane].pre[0] = ref_yv12[id];
@@ -6546,8 +6747,6 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
     best_mv->col >>= 3;
     best_mv->row >>= 3;
 
-    av1_set_mvcost(x, id, ref_mv_idx);
-
     // Small-range full-pixel motion search.
     bestsme = av1_refining_search_8p_c(x, sadpb, search_range,
                                        &cpi->fn_ptr[bsize], mask, mask_stride,
@@ -6592,8 +6791,8 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
           x, cm, mi_row, mi_col, &ref_mv[id].as_mv,
           cpi->common.allow_high_precision_mv, x->errorperbit,
           &cpi->fn_ptr[bsize], 0, cpi->sf.mv.subpel_iters_per_step, NULL,
-          x->nmvjointcost, x->mvcost, &dis, &sse, second_pred, mask,
-          mask_stride, id, pw, ph, cpi->sf.use_accurate_subpel_search);
+          x->nmv_vec_cost, x->mv_cost_stack, &dis, &sse, second_pred, mask,
+          mask_stride, id, pw, ph, cpi->sf.use_accurate_subpel_search, 1);
     }
 
     // Restore the pointer to the first prediction buffer.
@@ -6609,10 +6808,10 @@ static void joint_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
   *rate_mv = 0;
 
   for (ref = 0; ref < 2; ++ref) {
-    av1_set_mvcost(x, ref, ref_mv_idx);
     const int_mv curr_ref_mv = av1_get_ref_mv(x, ref);
-    *rate_mv += av1_mv_bit_cost(&cur_mv[ref].as_mv, &curr_ref_mv.as_mv,
-                                x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
+    *rate_mv +=
+        av1_mv_bit_cost(&cur_mv[ref].as_mv, &curr_ref_mv.as_mv, x->nmv_vec_cost,
+                        x->mv_cost_stack, MV_COST_WEIGHT);
   }
 }
 
@@ -6679,7 +6878,7 @@ static void estimate_ref_frame_costs(
     ref_costs_single[BWDREF_FRAME] += x->single_ref_cost[ctx_p6][5][0];
     ref_costs_single[ALTREF2_FRAME] += x->single_ref_cost[ctx_p6][5][1];
 
-    if (cm->reference_mode != SINGLE_REFERENCE) {
+    if (cm->current_frame.reference_mode != SINGLE_REFERENCE) {
       // Similar to single ref, determine cost of compound ref frames.
       // cost_compound_refs = cost_first_ref + cost_second_ref
       const int bwdref_comp_ctx_p = av1_get_pred_context_comp_bwdref_p(xd);
@@ -6790,18 +6989,25 @@ static void setup_buffer_ref_mvs_inter(
     struct buf_2d yv12_mb[REF_FRAMES][MAX_MB_PLANE]) {
   const AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
-  const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
+  const YV12_BUFFER_CONFIG *scaled_ref_frame =
+      av1_get_scaled_ref_frame(cpi, ref_frame);
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
-  const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf;
   MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
-
+  const struct scale_factors *const sf =
+      get_ref_scale_factors_const(cm, ref_frame);
+  const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_yv12_buf(cm, ref_frame);
   assert(yv12 != NULL);
 
-  // TODO(jkoleszar): Is the UV buffer ever used here? If so, need to make this
-  // use the UV scaling factors.
-  av1_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf,
-                       num_planes);
+  if (scaled_ref_frame) {
+    // Setup pred block based on scaled reference, because av1_mv_pred() doesn't
+    // support scaling.
+    av1_setup_pred_block(xd, yv12_mb[ref_frame], scaled_ref_frame, mi_row,
+                         mi_col, NULL, NULL, num_planes);
+  } else {
+    av1_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf,
+                         num_planes);
+  }
 
   // Gets an initial list of candidate vectors from neighbours and orders them
   av1_find_mv_refs(cm, xd, mbmi, ref_frame, mbmi_ext->ref_mv_count,
@@ -6809,11 +7015,18 @@ static void setup_buffer_ref_mvs_inter(
                    mi_col, mbmi_ext->mode_context);
 
   // Further refinement that is encode side only to test the top few candidates
-  // in full and choose the best as the centre point for subsequent searches.
+  // in full and choose the best as the center point for subsequent searches.
   // The current implementation doesn't support scaling.
-  (void)block_size;
-  av1_mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12->y_stride, ref_frame,
-              block_size);
+  av1_mv_pred(cpi, x, yv12_mb[ref_frame][0].buf, yv12_mb[ref_frame][0].stride,
+              ref_frame, block_size);
+
+  // Go back to unscaled reference.
+  if (scaled_ref_frame) {
+    // We had temporarily setup pred block based on scaled reference above. Go
+    // back to unscaled reference now, for subsequent use.
+    av1_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf,
+                         num_planes);
+  }
 }
 
 static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
@@ -6849,10 +7062,6 @@ static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
                          num_planes);
   }
 
-  av1_set_mvcost(
-      x, ref_idx,
-      mbmi->ref_mv_idx + (have_nearmv_in_inter_mode(mbmi->mode) ? 1 : 0));
-
   // Work out the size of the first step in the mv step search.
   // 0 here is maximum length first step. 1 is AOMMAX >> 1 etc.
   if (cpi->sf.mv.auto_mv_step_size && cm->show_frame) {
@@ -6958,13 +7167,12 @@ static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
                                  x->second_best_mv.as_int != x->best_mv.as_int;
           const int pw = block_size_wide[bsize];
           const int ph = block_size_high[bsize];
-
           best_mv_var = cpi->find_fractional_mv_step(
               x, cm, mi_row, mi_col, &ref_mv, cm->allow_high_precision_mv,
               x->errorperbit, &cpi->fn_ptr[bsize], cpi->sf.mv.subpel_force_stop,
               cpi->sf.mv.subpel_iters_per_step, cond_cost_list(cpi, cost_list),
-              x->nmvjointcost, x->mvcost, &dis, &x->pred_sse[ref], NULL, NULL,
-              0, 0, pw, ph, 1);
+              x->nmv_vec_cost, x->mv_cost_stack, &dis, &x->pred_sse[ref], NULL,
+              NULL, 0, 0, pw, ph, cpi->sf.use_accurate_subpel_search, 1);
 
           if (try_second) {
             const int minc =
@@ -6988,8 +7196,9 @@ static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
                   x->errorperbit, &cpi->fn_ptr[bsize],
                   cpi->sf.mv.subpel_force_stop,
                   cpi->sf.mv.subpel_iters_per_step,
-                  cond_cost_list(cpi, cost_list), x->nmvjointcost, x->mvcost,
-                  &dis, &x->pred_sse[ref], NULL, NULL, 0, 0, pw, ph, 1);
+                  cond_cost_list(cpi, cost_list), x->nmv_vec_cost,
+                  x->mv_cost_stack, &dis, &x->pred_sse[ref], NULL, NULL, 0, 0,
+                  pw, ph, cpi->sf.use_accurate_subpel_search, 0);
               if (this_var < best_mv_var) best_mv = x->best_mv.as_mv;
               x->best_mv.as_mv = best_mv;
             }
@@ -6999,8 +7208,8 @@ static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
               x, cm, mi_row, mi_col, &ref_mv, cm->allow_high_precision_mv,
               x->errorperbit, &cpi->fn_ptr[bsize], cpi->sf.mv.subpel_force_stop,
               cpi->sf.mv.subpel_iters_per_step, cond_cost_list(cpi, cost_list),
-              x->nmvjointcost, x->mvcost, &dis, &x->pred_sse[ref], NULL, NULL,
-              0, 0, 0, 0, 0);
+              x->nmv_vec_cost, x->mv_cost_stack, &dis, &x->pred_sse[ref], NULL,
+              NULL, 0, 0, 0, 0, 0, 1);
         }
         break;
       case OBMC_CAUSAL:
@@ -7008,23 +7217,22 @@ static void single_motion_search(const AV1_COMP *const cpi, MACROBLOCK *x,
             x, cm, mi_row, mi_col, &x->best_mv.as_mv, &ref_mv,
             cm->allow_high_precision_mv, x->errorperbit, &cpi->fn_ptr[bsize],
             cpi->sf.mv.subpel_force_stop, cpi->sf.mv.subpel_iters_per_step,
-            x->nmvjointcost, x->mvcost, &dis, &x->pred_sse[ref], 0,
+            x->nmv_vec_cost, x->mv_cost_stack, &dis, &x->pred_sse[ref], 0,
             cpi->sf.use_accurate_subpel_search);
         break;
       default: assert(0 && "Invalid motion mode!\n");
     }
   }
-  *rate_mv = av1_mv_bit_cost(&x->best_mv.as_mv, &ref_mv, x->nmvjointcost,
-                             x->mvcost, MV_COST_WEIGHT);
+  *rate_mv = av1_mv_bit_cost(&x->best_mv.as_mv, &ref_mv, x->nmv_vec_cost,
+                             x->mv_cost_stack, MV_COST_WEIGHT);
 
   if (cpi->sf.adaptive_motion_search && mbmi->motion_mode == SIMPLE_TRANSLATION)
     x->pred_mv[ref] = x->best_mv.as_mv;
 }
 
-static INLINE void restore_dst_buf(MACROBLOCKD *xd, BUFFER_SET dst,
+static INLINE void restore_dst_buf(MACROBLOCKD *xd, const BUFFER_SET dst,
                                    const int num_planes) {
-  int i;
-  for (i = 0; i < num_planes; i++) {
+  for (int i = 0; i < num_planes; i++) {
     xd->plane[i].dst.buf = dst.plane[i];
     xd->plane[i].dst.stride = dst.stride[i];
   }
@@ -7071,9 +7279,9 @@ static void build_second_inter_pred(const AV1_COMP *cpi, MACROBLOCK *x,
                             plane, !ref_idx, MV_PRECISION_Q3, mi_col * MI_SIZE,
                             mi_row * MI_SIZE, xd, cm->allow_warped_motion);
 
-  av1_jnt_comp_weight_assign(cm, mbmi, 0, &xd->jcp_param.fwd_offset,
-                             &xd->jcp_param.bck_offset,
-                             &xd->jcp_param.use_jnt_comp_avg, 1);
+  av1_dist_wtd_comp_weight_assign(cm, mbmi, 0, &xd->jcp_param.fwd_offset,
+                                  &xd->jcp_param.bck_offset,
+                                  &xd->jcp_param.use_dist_wtd_comp_avg, 1);
 }
 
 // Search for the best mv for one component of a compound,
@@ -7135,10 +7343,6 @@ static void compound_single_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
   best_mv->col >>= 3;
   best_mv->row >>= 3;
 
-  av1_set_mvcost(
-      x, ref_idx,
-      mbmi->ref_mv_idx + (have_nearmv_in_inter_mode(mbmi->mode) ? 1 : 0));
-
   // Small-range full-pixel motion search.
   bestsme = av1_refining_search_8p_c(x, sadpb, search_range,
                                      &cpi->fn_ptr[bsize], mask, mask_stride,
@@ -7175,8 +7379,8 @@ static void compound_single_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
         x, cm, mi_row, mi_col, &ref_mv.as_mv,
         cpi->common.allow_high_precision_mv, x->errorperbit,
         &cpi->fn_ptr[bsize], 0, cpi->sf.mv.subpel_iters_per_step, NULL,
-        x->nmvjointcost, x->mvcost, &dis, &sse, second_pred, mask, mask_stride,
-        ref_idx, pw, ph, cpi->sf.use_accurate_subpel_search);
+        x->nmv_vec_cost, x->mv_cost_stack, &dis, &sse, second_pred, mask,
+        mask_stride, ref_idx, pw, ph, cpi->sf.use_accurate_subpel_search, 1);
   }
 
   // Restore the pointer to the first unscaled prediction buffer.
@@ -7186,11 +7390,8 @@ static void compound_single_motion_search(const AV1_COMP *cpi, MACROBLOCK *x,
 
   *rate_mv = 0;
 
-  av1_set_mvcost(
-      x, ref_idx,
-      mbmi->ref_mv_idx + (have_nearmv_in_inter_mode(mbmi->mode) ? 1 : 0));
-  *rate_mv += av1_mv_bit_cost(this_mv, &ref_mv.as_mv, x->nmvjointcost,
-                              x->mvcost, MV_COST_WEIGHT);
+  *rate_mv += av1_mv_bit_cost(this_mv, &ref_mv.as_mv, x->nmv_vec_cost,
+                              x->mv_cost_stack, MV_COST_WEIGHT);
 }
 
 // Wrapper for compound_single_motion_search, for the common case
@@ -7206,7 +7407,7 @@ static void compound_single_motion_search_interinter(
   // Prediction buffer from second frame.
   DECLARE_ALIGNED(16, uint16_t, second_pred_alloc_16[MAX_SB_SQUARE]);
   uint8_t *second_pred;
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+  if (is_cur_buf_hbd(xd))
     second_pred = CONVERT_TO_BYTEPTR(second_pred_alloc_16);
   else
     second_pred = (uint8_t *)second_pred_alloc_16;
@@ -7336,7 +7537,7 @@ static int estimate_wedge_sign(const AV1_COMP *cpi, const MACROBLOCK *x,
   const BLOCK_SIZE f_index = split_qtr[bsize];
   assert(f_index != BLOCK_INVALID);
 
-  if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(&x->e_mbd)) {
     pred0 = CONVERT_TO_BYTEPTR(pred0);
     pred1 = CONVERT_TO_BYTEPTR(pred1);
   }
@@ -7386,7 +7587,7 @@ static int64_t pick_wedge(const AV1_COMP *const cpi, const MACROBLOCK *const x,
   int wedge_types = (1 << get_wedge_bits_lookup(bsize));
   const uint8_t *mask;
   uint64_t sse;
-  const int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+  const int hbd = is_cur_buf_hbd(xd);
   const int bd_round = hbd ? (xd->bd - 8) * 2 : 0;
 
   DECLARE_ALIGNED(32, int16_t, residual0[MAX_SB_SQUARE]);  // src - pred0
@@ -7457,7 +7658,7 @@ static int64_t pick_wedge_fixed_sign(const AV1_COMP *const cpi,
   int wedge_types = (1 << get_wedge_bits_lookup(bsize));
   const uint8_t *mask;
   uint64_t sse;
-  const int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+  const int hbd = is_cur_buf_hbd(xd);
   const int bd_round = hbd ? (xd->bd - 8) * 2 : 0;
   for (wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
     mask = av1_get_contiguous_soft_mask(wedge_index, wedge_sign, bsize);
@@ -7523,7 +7724,7 @@ static int64_t pick_interinter_seg(const AV1_COMP *const cpi,
   DIFFWTD_MASK_TYPE cur_mask_type;
   int64_t best_rd = INT64_MAX;
   DIFFWTD_MASK_TYPE best_mask_type = 0;
-  const int hbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH;
+  const int hbd = is_cur_buf_hbd(xd);
   const int bd_round = hbd ? (xd->bd - 8) * 2 : 0;
   DECLARE_ALIGNED(16, uint8_t, seg_mask[2 * MAX_SB_SQUARE]);
   uint8_t *tmp_mask[2] = { xd->seg_mask, seg_mask };
@@ -7574,7 +7775,7 @@ static int64_t pick_interintra_wedge(const AV1_COMP *const cpi,
   const int bh = block_size_high[bsize];
   DECLARE_ALIGNED(32, int16_t, residual1[MAX_SB_SQUARE]);  // src - pred1
   DECLARE_ALIGNED(32, int16_t, diff10[MAX_SB_SQUARE]);     // pred1 - pred0
-  if (get_bitdepth_data_path_index(xd)) {
+  if (is_cur_buf_hbd(xd)) {
     aom_highbd_subtract_block(bh, bw, residual1, bw, src->buf, src->stride,
                               CONVERT_TO_BYTEPTR(p1), bw, xd->bd);
     aom_highbd_subtract_block(bh, bw, diff10, bw, CONVERT_TO_BYTEPTR(p1), bw,
@@ -7653,7 +7854,7 @@ static void get_inter_predictors_masked_compound(
   av1_build_inter_predictors_for_planes_single_buf(
       xd, bsize, 0, 0, mi_row, mi_col, 1, preds1, strides, can_use_previous);
   const struct buf_2d *const src = &x->plane[0].src;
-  if (get_bitdepth_data_path_index(xd)) {
+  if (is_cur_buf_hbd(xd)) {
     aom_highbd_subtract_block(bh, bw, residual1, bw, src->buf, src->stride,
                               CONVERT_TO_BYTEPTR(*preds1), bw, xd->bd);
     aom_highbd_subtract_block(bh, bw, diff10, bw, CONVERT_TO_BYTEPTR(*preds1),
@@ -7668,26 +7869,38 @@ static void get_inter_predictors_masked_compound(
 static int64_t build_and_cost_compound_type(
     const AV1_COMP *const cpi, MACROBLOCK *x, const int_mv *const cur_mv,
     const BLOCK_SIZE bsize, const PREDICTION_MODE this_mode, int *rs2,
-    int rate_mv, BUFFER_SET *ctx, int *out_rate_mv, uint8_t **preds0,
+    int rate_mv, const BUFFER_SET *ctx, int *out_rate_mv, uint8_t **preds0,
     uint8_t **preds1, int16_t *residual1, int16_t *diff10, int *strides,
     int mi_row, int mi_col, int mode_rate, int64_t ref_best_rd,
-    int *calc_pred_masked_compound) {
+    int *calc_pred_masked_compound, int32_t *comp_rate, int64_t *comp_dist) {
   const AV1_COMMON *const cm = &cpi->common;
   MACROBLOCKD *xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
-  int rate_sum;
-  int64_t dist_sum;
   int64_t best_rd_cur = INT64_MAX;
   int64_t rd = INT64_MAX;
-  int tmp_skip_txfm_sb;
-  int64_t tmp_skip_sse_sb;
   const COMPOUND_TYPE compound_type = mbmi->interinter_comp.type;
 
+  // TODO(any): Save pred and mask calculation as well into records. However
+  // this may increase memory requirements as compound segment mask needs to be
+  // stored in each record.
   if (*calc_pred_masked_compound) {
     get_inter_predictors_masked_compound(cpi, x, bsize, mi_row, mi_col, preds0,
                                          preds1, residual1, diff10, strides);
     *calc_pred_masked_compound = 0;
   }
+  if (cpi->sf.prune_wedge_pred_diff_based && compound_type == COMPOUND_WEDGE) {
+    unsigned int sse;
+    if (is_cur_buf_hbd(xd))
+      (void)cpi->fn_ptr[bsize].vf(CONVERT_TO_BYTEPTR(*preds0), *strides,
+                                  CONVERT_TO_BYTEPTR(*preds1), *strides, &sse);
+    else
+      (void)cpi->fn_ptr[bsize].vf(*preds0, *strides, *preds1, *strides, &sse);
+    const unsigned int mse =
+        ROUND_POWER_OF_TWO(sse, num_pels_log2_lookup[bsize]);
+    // If two predictors are very similar, skip wedge compound mode search
+    if (mse < 8 || (!have_newmv_in_inter_mode(this_mode) && mse < 64))
+      return INT64_MAX;
+  }
 
   best_rd_cur =
       pick_interinter_mask(cpi, x, bsize, *preds0, *preds1, residual1, diff10);
@@ -7700,37 +7913,53 @@ static int64_t build_and_cost_compound_type(
   int64_t mode_rd = RDCOST(x->rdmult, *rs2 + mode_rate, 0);
   if (mode_rd > ref_best_rd) return INT64_MAX;
 
-  if (have_newmv_in_inter_mode(this_mode) && compound_type == COMPOUND_WEDGE) {
-    *out_rate_mv = interinter_compound_motion_search(cpi, x, cur_mv, bsize,
-                                                     this_mode, mi_row, mi_col);
-    av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, ctx, bsize);
-    model_rd_sb_fn[MODELRD_TYPE_MASKED_COMPOUND](
-        cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
-        &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
-    rd = RDCOST(x->rdmult, *rs2 + *out_rate_mv + rate_sum, dist_sum);
-    if (rd >= best_rd_cur) {
-      mbmi->mv[0].as_int = cur_mv[0].as_int;
-      mbmi->mv[1].as_int = cur_mv[1].as_int;
+  // Reuse data if matching record is found
+  if (comp_rate[compound_type] == INT_MAX) {
+    if (have_newmv_in_inter_mode(this_mode) &&
+        compound_type == COMPOUND_WEDGE &&
+        !cpi->sf.disable_interinter_wedge_newmv_search) {
+      *out_rate_mv = interinter_compound_motion_search(
+          cpi, x, cur_mv, bsize, this_mode, mi_row, mi_col);
+      av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, ctx, bsize,
+                                    AOM_PLANE_Y, AOM_PLANE_Y);
+      int rate_sum, tmp_skip_txfm_sb;
+      int64_t dist_sum, tmp_skip_sse_sb;
+      model_rd_sb_fn[MODELRD_TYPE_MASKED_COMPOUND](
+          cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
+          &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
+      rd = RDCOST(x->rdmult, *rs2 + *out_rate_mv + rate_sum, dist_sum);
+      if (rd >= best_rd_cur) {
+        mbmi->mv[0].as_int = cur_mv[0].as_int;
+        mbmi->mv[1].as_int = cur_mv[1].as_int;
+        *out_rate_mv = rate_mv;
+        av1_build_wedge_inter_predictor_from_buf(xd, bsize, 0, 0, preds0,
+                                                 strides, preds1, strides);
+      }
+    } else {
       *out_rate_mv = rate_mv;
       av1_build_wedge_inter_predictor_from_buf(xd, bsize, 0, 0, preds0, strides,
                                                preds1, strides);
     }
-    rd = estimate_yrd_for_sb(cpi, bsize, x, &rate_sum, &dist_sum,
-                             &tmp_skip_txfm_sb, &tmp_skip_sse_sb, INT64_MAX);
-    if (rd != INT64_MAX)
-      rd = RDCOST(x->rdmult, *rs2 + *out_rate_mv + rate_sum, dist_sum);
-    best_rd_cur = rd;
 
+    RD_STATS rd_stats;
+    rd = estimate_yrd_for_sb(cpi, bsize, x, INT64_MAX, &rd_stats);
+    if (rd != INT64_MAX) {
+      rd =
+          RDCOST(x->rdmult, *rs2 + *out_rate_mv + rd_stats.rate, rd_stats.dist);
+      // Backup rate and distortion for future reuse
+      comp_rate[compound_type] = rd_stats.rate;
+      comp_dist[compound_type] = rd_stats.dist;
+    }
   } else {
-    av1_build_wedge_inter_predictor_from_buf(xd, bsize, 0, 0, preds0, strides,
-                                             preds1, strides);
-    rd = estimate_yrd_for_sb(cpi, bsize, x, &rate_sum, &dist_sum,
-                             &tmp_skip_txfm_sb, &tmp_skip_sse_sb, INT64_MAX);
-    if (rd != INT64_MAX)
-      rd = RDCOST(x->rdmult, *rs2 + rate_mv + rate_sum, dist_sum);
-    best_rd_cur = rd;
+    assert(comp_dist[compound_type] != INT64_MAX);
+    // Rate is populated assuming no refinement is done=> not NEWMV case
+    assert(!have_newmv_in_inter_mode(this_mode));
+    *out_rate_mv = rate_mv;
+    // Calculate RD cost based on stored stats
+    rd = RDCOST(x->rdmult, *rs2 + *out_rate_mv + comp_rate[compound_type],
+                comp_dist[compound_type]);
   }
-  return best_rd_cur;
+  return rd;
 }
 
 typedef struct {
@@ -7753,6 +7982,8 @@ typedef struct {
   int64_t (*simple_rd)[MAX_REF_MV_SERCH][REF_FRAMES];
   int skip_motion_mode;
   INTERINTRA_MODE *inter_intra_mode;
+  int single_ref_first_pass;
+  SimpleRDState *simple_rd_state;
 } HandleInterModeArgs;
 
 /* If the current mode shares the same mv with other modes with higher cost,
@@ -7851,10 +8082,9 @@ static int64_t handle_newmv(const AV1_COMP *const cpi, MACROBLOCK *const x,
         *rate_mv = 0;
         for (i = 0; i < 2; ++i) {
           const int_mv ref_mv = av1_get_ref_mv(x, i);
-          av1_set_mvcost(x, i, mbmi->ref_mv_idx);
           *rate_mv +=
-              av1_mv_bit_cost(&cur_mv[i].as_mv, &ref_mv.as_mv, x->nmvjointcost,
-                              x->mvcost, MV_COST_WEIGHT);
+              av1_mv_bit_cost(&cur_mv[i].as_mv, &ref_mv.as_mv, x->nmv_vec_cost,
+                              x->mv_cost_stack, MV_COST_WEIGHT);
         }
       }
     } else if (this_mode == NEAREST_NEWMV || this_mode == NEAR_NEWMV) {
@@ -7863,11 +8093,10 @@ static int64_t handle_newmv(const AV1_COMP *const cpi, MACROBLOCK *const x,
         compound_single_motion_search_interinter(
             cpi, x, bsize, cur_mv, mi_row, mi_col, NULL, 0, rate_mv, 0, 1);
       } else {
-        av1_set_mvcost(x, 1,
-                       mbmi->ref_mv_idx + (this_mode == NEAR_NEWMV ? 1 : 0));
         const int_mv ref_mv = av1_get_ref_mv(x, 1);
-        *rate_mv = av1_mv_bit_cost(&cur_mv[1].as_mv, &ref_mv.as_mv,
-                                   x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
+        *rate_mv =
+            av1_mv_bit_cost(&cur_mv[1].as_mv, &ref_mv.as_mv, x->nmv_vec_cost,
+                            x->mv_cost_stack, MV_COST_WEIGHT);
       }
     } else {
       assert(this_mode == NEW_NEARESTMV || this_mode == NEW_NEARMV);
@@ -7877,10 +8106,9 @@ static int64_t handle_newmv(const AV1_COMP *const cpi, MACROBLOCK *const x,
             cpi, x, bsize, cur_mv, mi_row, mi_col, NULL, 0, rate_mv, 0, 0);
       } else {
         const int_mv ref_mv = av1_get_ref_mv(x, 0);
-        av1_set_mvcost(x, 0,
-                       mbmi->ref_mv_idx + (this_mode == NEW_NEARMV ? 1 : 0));
-        *rate_mv = av1_mv_bit_cost(&cur_mv[0].as_mv, &ref_mv.as_mv,
-                                   x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
+        *rate_mv =
+            av1_mv_bit_cost(&cur_mv[0].as_mv, &ref_mv.as_mv, x->nmv_vec_cost,
+                            x->mv_cost_stack, MV_COST_WEIGHT);
       }
     }
   } else {
@@ -7928,8 +8156,9 @@ static INLINE int get_switchable_rate(MACROBLOCK *const x,
 
 // calculate the rdcost of given interpolation_filter
 static INLINE int64_t interpolation_filter_rd(
-    MACROBLOCK *const x, const AV1_COMP *const cpi, BLOCK_SIZE bsize,
-    int mi_row, int mi_col, BUFFER_SET *const orig_dst, int64_t *const rd,
+    MACROBLOCK *const x, const AV1_COMP *const cpi,
+    const TileDataEnc *tile_data, BLOCK_SIZE bsize, int mi_row, int mi_col,
+    const BUFFER_SET *const orig_dst, int64_t *const rd,
     int *const switchable_rate, int *const skip_txfm_sb,
     int64_t *const skip_sse_sb, const BUFFER_SET *dst_bufs[2], int filter_idx,
     const int switchable_ctx[2], const int skip_pred, int *rate,
@@ -7946,6 +8175,14 @@ static INLINE int64_t interpolation_filter_rd(
   const int tmp_rs =
       get_switchable_rate(x, mbmi->interp_filters, switchable_ctx);
 
+  int64_t min_rd = RDCOST(x->rdmult, tmp_rs, 0);
+  if (min_rd > *rd) {
+    mbmi->interp_filters = last_best;
+    return 0;
+  }
+
+  (void)tile_data;
+
   assert(skip_pred != 2);
   assert((skip_pred >= 0) && (skip_pred <= cpi->default_interp_skip_flags));
   assert(rate[0] >= 0);
@@ -7959,11 +8196,13 @@ static INLINE int64_t interpolation_filter_rd(
 
   if (skip_pred != cpi->default_interp_skip_flags) {
     if (skip_pred != DEFAULT_LUMA_INTERP_SKIP_FLAG) {
-      av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, orig_dst, bsize);
+      av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                    AOM_PLANE_Y, AOM_PLANE_Y);
 #if CONFIG_COLLECT_RD_STATS == 3
       RD_STATS rd_stats_y;
-      select_tx_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col, INT64_MAX);
-      PrintPredictionUnitStats(cpi, x, &rd_stats_y, bsize);
+      pick_tx_size_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col,
+                            INT64_MAX);
+      PrintPredictionUnitStats(cpi, tile_data, x, &rd_stats_y, bsize);
 #endif  // CONFIG_COLLECT_RD_STATS == 3
       model_rd_sb_fn[MODELRD_TYPE_INTERP_FILTER](
           cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &tmp_rate[0], &tmp_dist[0],
@@ -7984,8 +8223,8 @@ static INLINE int64_t interpolation_filter_rd(
           mbmi->interp_filters = last_best;
           return 0;
         }
-        av1_build_inter_predictors_sbp(cm, xd, mi_row, mi_col, orig_dst, bsize,
-                                       plane);
+        av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                      plane, plane);
         model_rd_sb_fn[MODELRD_TYPE_INTERP_FILTER](
             cpi, bsize, x, xd, plane, plane, mi_row, mi_col, &tmp_rate_uv,
             &tmp_dist_uv, &tmp_skip_sb_uv, &tmp_skip_sse_uv, NULL, NULL, NULL);
@@ -8037,73 +8276,119 @@ static INLINE int64_t interpolation_filter_rd(
   return 0;
 }
 
-// Find the best rd filter in horizontal direction
-static INLINE int find_best_horiz_interp_filter_rd(
-    MACROBLOCK *const x, const AV1_COMP *const cpi, BLOCK_SIZE bsize,
-    int mi_row, int mi_col, BUFFER_SET *const orig_dst, int64_t *const rd,
+static INLINE void pred_dual_interp_filter_rd(
+    MACROBLOCK *const x, const AV1_COMP *const cpi,
+    const TileDataEnc *tile_data, BLOCK_SIZE bsize, int mi_row, int mi_col,
+    const BUFFER_SET *const orig_dst, int64_t *const rd,
     int *const switchable_rate, int *const skip_txfm_sb,
     int64_t *const skip_sse_sb, const BUFFER_SET *dst_bufs[2],
-    const int switchable_ctx[2], const int skip_hor, int *rate, int64_t *dist,
-    int best_dual_mode) {
-  int i;
-  const int bw = block_size_wide[bsize];
-  assert(best_dual_mode == 0);
-  if ((bw <= 4) && (skip_hor != cpi->default_interp_skip_flags)) {
-    int skip_pred = cpi->default_interp_skip_flags;
-    // Process the filters in reverse order to enable reusing rate and
-    // distortion (calcuated during EIGHTTAP_REGULAR) for MULTITAP_SHARP
-    for (i = (SWITCHABLE_FILTERS - 1); i >= 1; --i) {
-      if (interpolation_filter_rd(x, cpi, bsize, mi_row, mi_col, orig_dst, rd,
-                                  switchable_rate, skip_txfm_sb, skip_sse_sb,
-                                  dst_bufs, i, switchable_ctx, skip_pred, rate,
-                                  dist)) {
-        best_dual_mode = i;
+    InterpFilters filter_idx, const int switchable_ctx[2], const int skip_pred,
+    int *rate, int64_t *dist, InterpFilters af_horiz, InterpFilters af_vert,
+    InterpFilters lf_horiz, InterpFilters lf_vert) {
+  if ((af_horiz == lf_horiz) && (af_horiz != SWITCHABLE)) {
+    if (((af_vert == lf_vert) && (af_vert != SWITCHABLE))) {
+      filter_idx = af_horiz + (af_vert * SWITCHABLE_FILTERS);
+      if (filter_idx) {
+        interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                                orig_dst, rd, switchable_rate, skip_txfm_sb,
+                                skip_sse_sb, dst_bufs, filter_idx,
+                                switchable_ctx, skip_pred, rate, dist);
+      }
+    } else {
+      for (filter_idx = af_horiz; filter_idx < (DUAL_FILTER_SET_SIZE);
+           filter_idx += SWITCHABLE_FILTERS) {
+        if (filter_idx) {
+          interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                                  orig_dst, rd, switchable_rate, skip_txfm_sb,
+                                  skip_sse_sb, dst_bufs, filter_idx,
+                                  switchable_ctx, skip_pred, rate, dist);
+        }
       }
-      skip_pred = skip_hor;
     }
-  } else {
-    for (i = 1; i < SWITCHABLE_FILTERS; ++i) {
-      if (interpolation_filter_rd(x, cpi, bsize, mi_row, mi_col, orig_dst, rd,
-                                  switchable_rate, skip_txfm_sb, skip_sse_sb,
-                                  dst_bufs, i, switchable_ctx, skip_hor, rate,
-                                  dist)) {
-        best_dual_mode = i;
+  } else if ((af_vert == lf_vert) && (af_vert != SWITCHABLE)) {
+    for (filter_idx = (af_vert * SWITCHABLE_FILTERS);
+         filter_idx <= ((af_vert * SWITCHABLE_FILTERS) + 2); filter_idx += 1) {
+      if (filter_idx) {
+        interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                                orig_dst, rd, switchable_rate, skip_txfm_sb,
+                                skip_sse_sb, dst_bufs, filter_idx,
+                                switchable_ctx, skip_pred, rate, dist);
       }
     }
   }
-  return best_dual_mode;
 }
 
-// Find the best rd filter in vertical direction
-static INLINE void find_best_vert_interp_filter_rd(
-    MACROBLOCK *const x, const AV1_COMP *const cpi, BLOCK_SIZE bsize,
-    int mi_row, int mi_col, BUFFER_SET *const orig_dst, int64_t *const rd,
+// Find the best interp filter if dual_interp_filter = 0
+static INLINE void find_best_non_dual_interp_filter(
+    MACROBLOCK *const x, const AV1_COMP *const cpi,
+    const TileDataEnc *tile_data, BLOCK_SIZE bsize, int mi_row, int mi_col,
+    const BUFFER_SET *const orig_dst, int64_t *const rd,
     int *const switchable_rate, int *const skip_txfm_sb,
     int64_t *const skip_sse_sb, const BUFFER_SET *dst_bufs[2],
-    const int switchable_ctx[2], const int skip_ver, int *rate, int64_t *dist,
-    int best_dual_mode, int filter_set_size) {
-  int i;
-  const int bh = block_size_high[bsize];
-  if ((bh <= 4) && (skip_ver != cpi->default_interp_skip_flags)) {
+    const int switchable_ctx[2], const int skip_ver, const int skip_hor,
+    int *rate, int64_t *dist, int filter_set_size) {
+  int16_t i;
+
+  // Regular filter evaluation should have been done and hence the same should
+  // be the winner
+  assert(x->e_mbd.mi[0]->interp_filters == filter_sets[0]);
+  assert(filter_set_size == DUAL_FILTER_SET_SIZE);
+
+  // Reuse regular filter's modeled rd data for sharp filter for following
+  // cases
+  // 1) When bsize is 4x4
+  // 2) When block width is 4 (i.e. 4x8/4x16 blocks) and MV in vertical
+  // direction is full-pel
+  // 3) When block height is 4 (i.e. 8x4/16x4 blocks) and MV in horizontal
+  // direction is full-pel
+  // TODO(any): Optimize cases 2 and 3 further if luma MV in relavant direction
+  // alone is full-pel
+
+  if ((bsize == BLOCK_4X4) ||
+      (block_size_wide[bsize] == 4 &&
+       skip_ver == cpi->default_interp_skip_flags) ||
+      (block_size_high[bsize] == 4 &&
+       skip_hor == cpi->default_interp_skip_flags)) {
     int skip_pred = cpi->default_interp_skip_flags;
-    // Process the filters in reverse order to enable reusing rate and
-    // distortion (calcuated during EIGHTTAP_REGULAR) for MULTITAP_SHARP
-    assert(filter_set_size == DUAL_FILTER_SET_SIZE);
-    for (i = (filter_set_size - SWITCHABLE_FILTERS + best_dual_mode);
-         i >= (best_dual_mode + SWITCHABLE_FILTERS); i -= SWITCHABLE_FILTERS) {
-      interpolation_filter_rd(x, cpi, bsize, mi_row, mi_col, orig_dst, rd,
-                              switchable_rate, skip_txfm_sb, skip_sse_sb,
-                              dst_bufs, i, switchable_ctx, skip_pred, rate,
-                              dist);
-      skip_pred = skip_ver;
+    for (i = filter_set_size - 1; i > 0; i -= (SWITCHABLE_FILTERS + 1)) {
+      // This assert tells that (filter_x == filter_y) for non-dual filter case
+      assert((filter_sets[i] & 0xffff) == (filter_sets[i] >> 16));
+      if (cpi->sf.adaptive_interp_filter_search &&
+          (cpi->sf.interp_filter_search_mask & (1 << (i >> 2)))) {
+        continue;
+      }
+      interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                              orig_dst, rd, switchable_rate, skip_txfm_sb,
+                              skip_sse_sb, dst_bufs, i, switchable_ctx,
+                              skip_pred, rate, dist);
+      skip_pred = (skip_hor & skip_ver);
     }
   } else {
-    for (i = best_dual_mode + SWITCHABLE_FILTERS; i < filter_set_size;
-         i += SWITCHABLE_FILTERS) {
-      interpolation_filter_rd(x, cpi, bsize, mi_row, mi_col, orig_dst, rd,
-                              switchable_rate, skip_txfm_sb, skip_sse_sb,
-                              dst_bufs, i, switchable_ctx, skip_ver, rate,
-                              dist);
+    int skip_pred = (skip_hor & skip_ver);
+    for (i = (SWITCHABLE_FILTERS + 1); i < filter_set_size;
+         i += (SWITCHABLE_FILTERS + 1)) {
+      // This assert tells that (filter_x == filter_y) for non-dual filter case
+      assert((filter_sets[i] & 0xffff) == (filter_sets[i] >> 16));
+      if (cpi->sf.adaptive_interp_filter_search &&
+          (cpi->sf.interp_filter_search_mask & (1 << (i >> 2)))) {
+        continue;
+      }
+      interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                              orig_dst, rd, switchable_rate, skip_txfm_sb,
+                              skip_sse_sb, dst_bufs, i, switchable_ctx,
+                              skip_pred, rate, dist);
+      // In first iteration, smooth filter is evaluated. If smooth filter
+      // (which is less sharper) is the winner among regular and smooth filters,
+      // sharp filter evaluation is skipped
+      // TODO(any): Refine this gating based on modelled rd only (i.e., by not
+      // accounting switchable filter rate)
+      if (cpi->sf.skip_sharp_interp_filter_search &&
+          skip_pred != cpi->default_interp_skip_flags) {
+        MACROBLOCKD *const xd = &x->e_mbd;
+        MB_MODE_INFO *const mbmi = xd->mi[0];
+        if (mbmi->interp_filters == filter_sets[(SWITCHABLE_FILTERS + 1)])
+          break;
+      }
     }
   }
 }
@@ -8121,6 +8406,45 @@ static INLINE int is_interp_filter_match(const INTERPOLATION_FILTER_STATS *st,
   return 1;
 }
 
+// Checks if characteristics of search match
+static INLINE int is_comp_rd_match(const MACROBLOCK *const x,
+                                   const COMP_RD_STATS *st,
+                                   const MB_MODE_INFO *const mi,
+                                   int32_t *comp_rate, int64_t *comp_dist) {
+  // TODO(ranjit): Ensure that compound type search use regular filter always
+  // and check if following check can be removed
+  // Check if interp filter matches with previous case
+  if (st->filter != mi->interp_filters) return 0;
+
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  // Match MV and reference indices
+  for (int i = 0; i < 2; ++i) {
+    if ((st->ref_frames[i] != mi->ref_frame[i]) ||
+        (st->mv[i].as_int != mi->mv[i].as_int)) {
+      return 0;
+    }
+    const WarpedMotionParams *const wm = &xd->global_motion[mi->ref_frame[i]];
+    if (is_global_mv_block(mi, wm->wmtype) != st->is_global[i]) return 0;
+  }
+
+  // Store the stats for compound average
+  comp_rate[0] = st->rate[0];
+  comp_dist[0] = st->dist[0];
+
+  // For compound wedge/segment, reuse data only if NEWMV is not present in
+  // either of the directions
+  if (!have_newmv_in_inter_mode(mi->mode) &&
+      !have_newmv_in_inter_mode(st->mode)) {
+    memcpy(&comp_rate[1], &st->rate[1], sizeof(comp_rate[1]) * 2);
+    memcpy(&comp_dist[1], &st->dist[1], sizeof(comp_dist[1]) * 2);
+  }
+
+  // TODO(ranjit) : Check if compound wedge/segment can reuse data for NEWMV
+  // case as well based on speed feature
+
+  return 1;
+}
+
 static INLINE int find_interp_filter_in_stats(MACROBLOCK *x,
                                               MB_MODE_INFO *const mbmi) {
   const int comp_idx = mbmi->compound_idx;
@@ -8134,6 +8458,19 @@ static INLINE int find_interp_filter_in_stats(MACROBLOCK *x,
   }
   return -1;  // no match result found
 }
+// Checks if similar compound type search case is accounted earlier
+// If found, returns relevant rd data
+static INLINE int find_comp_rd_in_stats(const MACROBLOCK *x,
+                                        const MB_MODE_INFO *const mbmi,
+                                        int32_t *comp_rate,
+                                        int64_t *comp_dist) {
+  for (int j = 0; j < x->comp_rd_stats_idx; ++j) {
+    if (is_comp_rd_match(x, &x->comp_rd_stats[j], mbmi, comp_rate, comp_dist)) {
+      return 1;
+    }
+  }
+  return 0;  // no match result found
+}
 
 static INLINE void save_interp_filter_search_stat(MACROBLOCK *x,
                                                   MB_MODE_INFO *const mbmi) {
@@ -8150,11 +8487,37 @@ static INLINE void save_interp_filter_search_stat(MACROBLOCK *x,
   }
 }
 
+static INLINE void save_comp_rd_search_stat(MACROBLOCK *x,
+                                            const MB_MODE_INFO *const mbmi,
+                                            const int32_t *comp_rate,
+                                            const int64_t *comp_dist,
+                                            const int_mv *cur_mv) {
+  const int offset = x->comp_rd_stats_idx;
+  if (offset < MAX_COMP_RD_STATS) {
+    COMP_RD_STATS *const rd_stats = x->comp_rd_stats + offset;
+    memcpy(rd_stats->rate, comp_rate, sizeof(rd_stats->rate));
+    memcpy(rd_stats->dist, comp_dist, sizeof(rd_stats->dist));
+    memcpy(rd_stats->mv, cur_mv, sizeof(rd_stats->mv));
+    memcpy(rd_stats->ref_frames, mbmi->ref_frame, sizeof(rd_stats->ref_frames));
+    rd_stats->mode = mbmi->mode;
+    rd_stats->filter = mbmi->interp_filters;
+    rd_stats->ref_mv_idx = mbmi->ref_mv_idx;
+    const MACROBLOCKD *const xd = &x->e_mbd;
+    for (int i = 0; i < 2; ++i) {
+      const WarpedMotionParams *const wm =
+          &xd->global_motion[mbmi->ref_frame[i]];
+      rd_stats->is_global[i] = is_global_mv_block(mbmi, wm->wmtype);
+    }
+    ++x->comp_rd_stats_idx;
+  }
+}
+
 static int64_t interpolation_filter_search(
-    MACROBLOCK *const x, const AV1_COMP *const cpi, BLOCK_SIZE bsize,
-    int mi_row, int mi_col, const BUFFER_SET *const tmp_dst,
-    BUFFER_SET *const orig_dst, InterpFilter (*const single_filter)[REF_FRAMES],
-    int64_t *const rd, int *const switchable_rate, int *const skip_txfm_sb,
+    MACROBLOCK *const x, const AV1_COMP *const cpi,
+    const TileDataEnc *tile_data, BLOCK_SIZE bsize, int mi_row, int mi_col,
+    const BUFFER_SET *const tmp_dst, const BUFFER_SET *const orig_dst,
+    InterpFilter (*const single_filter)[REF_FRAMES], int64_t *const rd,
+    int *const switchable_rate, int *const skip_txfm_sb,
     int64_t *const skip_sse_sb, const int skip_build_pred,
     HandleInterModeArgs *args, int64_t ref_best_rd) {
   const AV1_COMMON *cm = &cpi->common;
@@ -8186,13 +8549,15 @@ static int64_t interpolation_filter_search(
   switchable_ctx[1] = av1_get_pred_context_switchable_interp(xd, 1);
   *switchable_rate =
       get_switchable_rate(x, mbmi->interp_filters, switchable_ctx);
-  if (!skip_build_pred)
-    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, orig_dst, bsize);
+  if (!skip_build_pred) {
+    av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize, 0,
+                                  av1_num_planes(cm) - 1);
+  }
 
 #if CONFIG_COLLECT_RD_STATS == 3
   RD_STATS rd_stats_y;
-  select_tx_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col, INT64_MAX);
-  PrintPredictionUnitStats(cpi, x, &rd_stats_y, bsize);
+  pick_tx_size_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col, INT64_MAX);
+  PrintPredictionUnitStats(cpi, tile_data, x, &rd_stats_y, bsize);
 #endif  // CONFIG_COLLECT_RD_STATS == 3
   model_rd_sb_fn[MODELRD_TYPE_INTERP_FILTER](
       cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &tmp_rate[0], &tmp_dist[0],
@@ -8248,8 +8613,8 @@ static int64_t interpolation_filter_search(
   const int is_compound = has_second_ref(mbmi);
   assert(is_intrabc_block(mbmi) == 0);
   for (int j = 0; j < 1 + is_compound; ++j) {
-    const RefBuffer *ref_buf = &cm->frame_refs[mbmi->ref_frame[j] - LAST_FRAME];
-    const struct scale_factors *const sf = &ref_buf->sf;
+    const struct scale_factors *const sf =
+        get_ref_scale_factors_const(cm, mbmi->ref_frame[j]);
     // TODO(any): Refine skip flag calculation considering scaling
     if (av1_is_scaled(sf)) {
       skip_hor = 0;
@@ -8296,30 +8661,75 @@ static int64_t interpolation_filter_search(
     // default to (R,R): EIGHTTAP_REGULARxEIGHTTAP_REGULAR
     int best_dual_mode = 0;
     // Find best of {R}x{R,Sm,Sh}
-    // EIGHTTAP_REGULAR mode is calculated beforehand
-    best_dual_mode = find_best_horiz_interp_filter_rd(
-        x, cpi, bsize, mi_row, mi_col, orig_dst, rd, switchable_rate,
-        best_skip_txfm_sb, best_skip_sse_sb, dst_bufs, switchable_ctx, skip_hor,
-        tmp_rate, tmp_dist, best_dual_mode);
-
-    // From best of horizontal EIGHTTAP_REGULAR modes, check vertical modes
-    find_best_vert_interp_filter_rd(
-        x, cpi, bsize, mi_row, mi_col, orig_dst, rd, switchable_rate,
+    const int bw = block_size_wide[bsize];
+    const int bh = block_size_high[bsize];
+    int skip_pred;
+    int bsl, pred_filter_search;
+    InterpFilters af_horiz = SWITCHABLE, af_vert = SWITCHABLE,
+                  lf_horiz = SWITCHABLE, lf_vert = SWITCHABLE, filter_idx = 0;
+    const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+    const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+    bsl = mi_size_wide_log2[bsize];
+    pred_filter_search =
+        cpi->sf.cb_pred_filter_search
+            ? (((mi_row + mi_col) >> bsl) +
+               get_chessboard_index(cm->current_frame.frame_number)) &
+                  0x1
+            : 0;
+    if (above_mbmi && is_inter_block(above_mbmi)) {
+      af_horiz = av1_extract_interp_filter(above_mbmi->interp_filters, 1);
+      af_vert = av1_extract_interp_filter(above_mbmi->interp_filters, 0);
+    }
+    if (left_mbmi && is_inter_block(left_mbmi)) {
+      lf_horiz = av1_extract_interp_filter(left_mbmi->interp_filters, 1);
+      lf_vert = av1_extract_interp_filter(left_mbmi->interp_filters, 0);
+    }
+    pred_filter_search &= !have_newmv_in_inter_mode(mbmi->mode);
+    pred_filter_search &=
+        ((af_horiz == lf_horiz) && (af_horiz != SWITCHABLE)) ||
+        ((af_vert == lf_vert) && (af_vert != SWITCHABLE));
+    if (pred_filter_search) {
+      pred_dual_interp_filter_rd(
+          x, cpi, tile_data, bsize, mi_row, mi_col, orig_dst, rd,
+          switchable_rate, best_skip_txfm_sb, best_skip_sse_sb, dst_bufs,
+          filter_idx, switchable_ctx, (skip_hor & skip_ver), tmp_rate, tmp_dist,
+          af_horiz, af_vert, lf_horiz, lf_vert);
+    } else {
+      skip_pred = bw <= 4 ? cpi->default_interp_skip_flags : skip_hor;
+      for (i = (SWITCHABLE_FILTERS - 1); i >= 1; --i) {
+        if (interpolation_filter_rd(
+                x, cpi, tile_data, bsize, mi_row, mi_col, orig_dst, rd,
+                switchable_rate, best_skip_txfm_sb, best_skip_sse_sb, dst_bufs,
+                i, switchable_ctx, skip_pred, tmp_rate, tmp_dist)) {
+          best_dual_mode = i;
+        }
+        skip_pred = skip_hor;
+      }
+      // From best of horizontal EIGHTTAP_REGULAR modes, check vertical modes
+      skip_pred = bh <= 4 ? cpi->default_interp_skip_flags : skip_ver;
+      assert(filter_set_size == DUAL_FILTER_SET_SIZE);
+      for (i = (best_dual_mode + (SWITCHABLE_FILTERS * 2));
+           i >= (best_dual_mode + SWITCHABLE_FILTERS);
+           i -= SWITCHABLE_FILTERS) {
+        interpolation_filter_rd(
+            x, cpi, tile_data, bsize, mi_row, mi_col, orig_dst, rd,
+            switchable_rate, best_skip_txfm_sb, best_skip_sse_sb, dst_bufs, i,
+            switchable_ctx, skip_pred, tmp_rate, tmp_dist);
+        skip_pred = skip_ver;
+      }
+    }
+  } else if (cm->seq_params.enable_dual_filter == 0) {
+    find_best_non_dual_interp_filter(
+        x, cpi, tile_data, bsize, mi_row, mi_col, orig_dst, rd, switchable_rate,
         best_skip_txfm_sb, best_skip_sse_sb, dst_bufs, switchable_ctx, skip_ver,
-        tmp_rate, tmp_dist, best_dual_mode, filter_set_size);
+        skip_hor, tmp_rate, tmp_dist, filter_set_size);
   } else {
     // EIGHTTAP_REGULAR mode is calculated beforehand
     for (i = 1; i < filter_set_size; ++i) {
-      if (cm->seq_params.enable_dual_filter == 0) {
-        const int16_t filter_y = filter_sets[i] & 0xffff;
-        const int16_t filter_x = filter_sets[i] >> 16;
-        if (filter_x != filter_y) continue;
-      }
-      interpolation_filter_rd(x, cpi, bsize, mi_row, mi_col, orig_dst, rd,
-                              switchable_rate, best_skip_txfm_sb,
-                              best_skip_sse_sb, dst_bufs, i, switchable_ctx, 0,
-                              tmp_rate, tmp_dist);
-      assert(x->recalc_luma_mc_data == 0);
+      interpolation_filter_rd(x, cpi, tile_data, bsize, mi_row, mi_col,
+                              orig_dst, rd, switchable_rate, best_skip_txfm_sb,
+                              best_skip_sse_sb, dst_bufs, i, switchable_ctx,
+                              (skip_hor & skip_ver), tmp_rate, tmp_dist);
     }
   }
   swap_dst_buf(xd, dst_bufs, num_planes);
@@ -8329,7 +8739,8 @@ static int64_t interpolation_filter_search(
     // in either of the directions  Condition below is necessary, but not
     // sufficient
     assert((skip_hor == 1) || (skip_ver == 1));
-    av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, orig_dst, bsize);
+    av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                  AOM_PLANE_Y, AOM_PLANE_Y);
   }
   *skip_txfm_sb = best_skip_txfm_sb[1];
   *skip_sse_sb = best_skip_sse_sb[1];
@@ -8343,173 +8754,172 @@ static int64_t interpolation_filter_search(
   return 0;
 }
 
-static int txfm_search(const AV1_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
-                       int mi_row, int mi_col, RD_STATS *rd_stats,
-                       RD_STATS *rd_stats_y, RD_STATS *rd_stats_uv,
-                       int mode_rate, int64_t ref_best_rd) {
+static int txfm_search(const AV1_COMP *cpi, const TileDataEnc *tile_data,
+                       MACROBLOCK *x, BLOCK_SIZE bsize, int mi_row, int mi_col,
+                       RD_STATS *rd_stats, RD_STATS *rd_stats_y,
+                       RD_STATS *rd_stats_uv, int mode_rate,
+                       int64_t ref_best_rd) {
   /*
    * This function combines y and uv planes' transform search processes
-   * together, when the prediction is generated. It first does subtration to
+   * together, when the prediction is generated. It first does subtraction to
    * obtain the prediction error. Then it calls
-   * select_tx_type_yrd/super_block_yrd and inter_block_uvrd sequentially and
-   * handles the early terminations happen in those functions. At the end, it
+   * pick_tx_size_type_yrd/super_block_yrd and super_block_uvrd sequentially and
+   * handles the early terminations happening in those functions. At the end, it
    * computes the rd_stats/_y/_uv accordingly.
    */
   const AV1_COMMON *cm = &cpi->common;
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
-  int skip_txfm_sb = 0;
-  const int num_planes = av1_num_planes(cm);
   const int ref_frame_1 = mbmi->ref_frame[1];
   const int64_t mode_rd = RDCOST(x->rdmult, mode_rate, 0);
   const int64_t rd_thresh =
       ref_best_rd == INT64_MAX ? INT64_MAX : ref_best_rd - mode_rd;
   const int skip_ctx = av1_get_skip_context(xd);
+  const int skip_flag_cost[2] = { x->skip_cost[skip_ctx][0],
+                                  x->skip_cost[skip_ctx][1] };
   const int64_t min_header_rate =
-      mode_rate + AOMMIN(x->skip_cost[skip_ctx][0], x->skip_cost[skip_ctx][1]);
+      mode_rate + AOMMIN(skip_flag_cost[0], skip_flag_cost[1]);
   // Account for minimum skip and non_skip rd.
   // Eventually either one of them will be added to mode_rate
   const int64_t min_header_rd_possible = RDCOST(x->rdmult, min_header_rate, 0);
+  (void)tile_data;
 
   if (min_header_rd_possible > ref_best_rd) {
     av1_invalid_rd_stats(rd_stats_y);
-    av1_invalid_rd_stats(rd_stats);
     return 0;
   }
 
   av1_init_rd_stats(rd_stats);
   av1_init_rd_stats(rd_stats_y);
-  av1_init_rd_stats(rd_stats_uv);
   rd_stats->rate = mode_rate;
 
-  if (!cpi->common.all_lossless)
-    check_block_skip(cpi, bsize, x, xd, 0, num_planes - 1, &skip_txfm_sb);
-  if (!skip_txfm_sb) {
-    int64_t non_skip_rdcosty = INT64_MAX;
-    int64_t skip_rdcosty = INT64_MAX;
-    int64_t min_rdcosty = INT64_MAX;
-    int is_cost_valid_uv = 0;
-
-    // cost and distortion
-    av1_subtract_plane(x, bsize, 0);
-    if (cm->tx_mode == TX_MODE_SELECT && !xd->lossless[mbmi->segment_id]) {
-      // Motion mode
-      select_tx_type_yrd(cpi, x, rd_stats_y, bsize, mi_row, mi_col, rd_thresh);
+  // cost and distortion
+  av1_subtract_plane(x, bsize, 0);
+  if (cm->tx_mode == TX_MODE_SELECT && !xd->lossless[mbmi->segment_id]) {
+    pick_tx_size_type_yrd(cpi, x, rd_stats_y, bsize, mi_row, mi_col, rd_thresh);
 #if CONFIG_COLLECT_RD_STATS == 2
-      PrintPredictionUnitStats(cpi, x, rd_stats_y, bsize);
+    PrintPredictionUnitStats(cpi, tile_data, x, rd_stats_y, bsize);
 #endif  // CONFIG_COLLECT_RD_STATS == 2
-    } else {
-      super_block_yrd(cpi, x, rd_stats_y, bsize, rd_thresh);
-      memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size));
-      for (int i = 0; i < xd->n4_h * xd->n4_w; ++i)
-        set_blk_skip(x, 0, i, rd_stats_y->skip);
-    }
-
-    if (rd_stats_y->rate == INT_MAX) {
-      av1_invalid_rd_stats(rd_stats);
-      // TODO(angiebird): check if we need this
-      // restore_dst_buf(xd, *orig_dst, num_planes);
-      mbmi->ref_frame[1] = ref_frame_1;
-      return 0;
-    }
-
-    av1_merge_rd_stats(rd_stats, rd_stats_y);
-
-    non_skip_rdcosty = RDCOST(
-        x->rdmult, rd_stats->rate + x->skip_cost[skip_ctx][0], rd_stats->dist);
-    skip_rdcosty =
-        RDCOST(x->rdmult, mode_rate + x->skip_cost[skip_ctx][1], rd_stats->sse);
-    min_rdcosty = AOMMIN(non_skip_rdcosty, skip_rdcosty);
-
-    if (min_rdcosty > ref_best_rd) {
-      int64_t tokenonly_rdy =
-          AOMMIN(RDCOST(x->rdmult, rd_stats_y->rate, rd_stats_y->dist),
-                 RDCOST(x->rdmult, 0, rd_stats_y->sse));
-      // Invalidate rd_stats_y to skip the rest of the motion modes search
-      if (tokenonly_rdy - (tokenonly_rdy >> cpi->sf.adaptive_txb_search_level) >
-          rd_thresh)
-        av1_invalid_rd_stats(rd_stats_y);
-      mbmi->ref_frame[1] = ref_frame_1;
-      return 0;
-    }
-
-    if (num_planes > 1) {
-      /* clang-format off */
-      is_cost_valid_uv =
-          inter_block_uvrd(cpi, x, rd_stats_uv, bsize,
-                           ref_best_rd - non_skip_rdcosty,
-                           ref_best_rd - skip_rdcosty, FTXS_NONE);
-      if (!is_cost_valid_uv) {
-        mbmi->ref_frame[1] = ref_frame_1;
-        return 0;
-      }
-      /* clang-format on */
-      av1_merge_rd_stats(rd_stats, rd_stats_uv);
-    } else {
-      av1_init_rd_stats(rd_stats_uv);
-    }
-    if (rd_stats->skip) {
-      rd_stats->rate -= rd_stats_uv->rate + rd_stats_y->rate;
-      rd_stats_y->rate = 0;
-      rd_stats_uv->rate = 0;
-      rd_stats->rate += x->skip_cost[skip_ctx][1];
-      mbmi->skip = 0;
-      // here mbmi->skip temporarily plays a role as what this_skip2 does
-
-      int64_t tmprd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
-      if (tmprd > ref_best_rd) {
-        mbmi->ref_frame[1] = ref_frame_1;
-        return 0;
-      }
-    } else if (!xd->lossless[mbmi->segment_id] &&
-               (RDCOST(x->rdmult,
-                       rd_stats_y->rate + rd_stats_uv->rate +
-                           x->skip_cost[skip_ctx][0],
-                       rd_stats->dist) >=
-                RDCOST(x->rdmult, x->skip_cost[skip_ctx][1], rd_stats->sse))) {
-      rd_stats->rate -= rd_stats_uv->rate + rd_stats_y->rate;
-      rd_stats->rate += x->skip_cost[skip_ctx][1];
-      rd_stats->dist = rd_stats->sse;
-      rd_stats_y->rate = 0;
-      rd_stats_uv->rate = 0;
-      mbmi->skip = 1;
-    } else {
-      rd_stats->rate += x->skip_cost[skip_ctx][0];
-      mbmi->skip = 0;
-    }
   } else {
-    x->skip = 1;
-    mbmi->tx_size = tx_size_from_tx_mode(bsize, cm->tx_mode);
-    // The cost of skip bit needs to be added.
-    mbmi->skip = 0;
-    rd_stats->rate += x->skip_cost[skip_ctx][1];
+    super_block_yrd(cpi, x, rd_stats_y, bsize, rd_thresh);
+    memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size));
+    for (int i = 0; i < xd->n4_h * xd->n4_w; ++i)
+      set_blk_skip(x, 0, i, rd_stats_y->skip);
+  }
 
-    rd_stats->dist = 0;
-    rd_stats->sse = 0;
+  if (rd_stats_y->rate == INT_MAX) {
+    // TODO(angiebird): check if we need this
+    // restore_dst_buf(xd, *orig_dst, num_planes);
+    mbmi->ref_frame[1] = ref_frame_1;
+    return 0;
+  }
+
+  av1_merge_rd_stats(rd_stats, rd_stats_y);
+
+  const int64_t non_skip_rdcosty =
+      RDCOST(x->rdmult, rd_stats->rate + skip_flag_cost[0], rd_stats->dist);
+  const int64_t skip_rdcosty =
+      RDCOST(x->rdmult, mode_rate + skip_flag_cost[1], rd_stats->sse);
+  const int64_t min_rdcosty = AOMMIN(non_skip_rdcosty, skip_rdcosty);
+  if (min_rdcosty > ref_best_rd) {
+    const int64_t tokenonly_rdy =
+        AOMMIN(RDCOST(x->rdmult, rd_stats_y->rate, rd_stats_y->dist),
+               RDCOST(x->rdmult, 0, rd_stats_y->sse));
+    // Invalidate rd_stats_y to skip the rest of the motion modes search
+    if (tokenonly_rdy - (tokenonly_rdy >> cpi->sf.prune_motion_mode_level) >
+        rd_thresh)
+      av1_invalid_rd_stats(rd_stats_y);
+    mbmi->ref_frame[1] = ref_frame_1;
+    return 0;
+  }
+
+  av1_init_rd_stats(rd_stats_uv);
+  const int num_planes = av1_num_planes(cm);
+  if (num_planes > 1) {
+    const int is_cost_valid_uv =
+        super_block_uvrd(cpi, x, rd_stats_uv, bsize, ref_best_rd);
+    if (!is_cost_valid_uv) {
+      mbmi->ref_frame[1] = ref_frame_1;
+      return 0;
+    }
+    av1_merge_rd_stats(rd_stats, rd_stats_uv);
+  }
+
+  if (rd_stats->skip) {
+    rd_stats->rate -= rd_stats_uv->rate + rd_stats_y->rate;
     rd_stats_y->rate = 0;
     rd_stats_uv->rate = 0;
-    rd_stats->skip = 1;
-    int64_t tmprd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
+    rd_stats->dist = rd_stats->sse;
+    rd_stats_y->dist = rd_stats_y->sse;
+    rd_stats_uv->dist = rd_stats_uv->sse;
+    rd_stats->rate += skip_flag_cost[1];
+    mbmi->skip = 1;
+    // here mbmi->skip temporarily plays a role as what this_skip2 does
+
+    const int64_t tmprd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
     if (tmprd > ref_best_rd) {
       mbmi->ref_frame[1] = ref_frame_1;
       return 0;
     }
+#if CONFIG_ONE_PASS_SVM
+    av1_reg_stat_skipmode_update(rd_stats_y, x->rdmult);
+#endif
+  } else if (!xd->lossless[mbmi->segment_id] &&
+             (RDCOST(x->rdmult,
+                     rd_stats_y->rate + rd_stats_uv->rate + skip_flag_cost[0],
+                     rd_stats->dist) >=
+              RDCOST(x->rdmult, skip_flag_cost[1], rd_stats->sse))) {
+    rd_stats->rate -= rd_stats_uv->rate + rd_stats_y->rate;
+    rd_stats->rate += skip_flag_cost[1];
+    rd_stats->dist = rd_stats->sse;
+    rd_stats_y->dist = rd_stats_y->sse;
+    rd_stats_uv->dist = rd_stats_uv->sse;
+    rd_stats_y->rate = 0;
+    rd_stats_uv->rate = 0;
+    mbmi->skip = 1;
+#if CONFIG_ONE_PASS_SVM
+    av1_reg_stat_skipmode_update(rd_stats_y, x->rdmult);
+#endif
+  } else {
+    rd_stats->rate += skip_flag_cost[0];
+    mbmi->skip = 0;
   }
+
   return 1;
 }
 
+static INLINE bool enable_wedge_search(MACROBLOCK *const x,
+                                       const AV1_COMP *const cpi) {
+  // Enable wedge search if source variance and edge strength are above
+  // the thresholds.
+  return x->source_variance > cpi->sf.disable_wedge_search_var_thresh &&
+         x->edge_strength > cpi->sf.disable_wedge_search_edge_thresh;
+}
+
+static INLINE bool enable_wedge_interinter_search(MACROBLOCK *const x,
+                                                  const AV1_COMP *const cpi) {
+  return enable_wedge_search(x, cpi) && cpi->oxcf.enable_interinter_wedge;
+}
+
+static INLINE bool enable_wedge_interintra_search(MACROBLOCK *const x,
+                                                  const AV1_COMP *const cpi) {
+  return enable_wedge_search(x, cpi) && cpi->oxcf.enable_interintra_wedge &&
+         !cpi->sf.disable_wedge_interintra_search;
+}
+
 static int handle_inter_intra_mode(const AV1_COMP *const cpi,
                                    MACROBLOCK *const x, BLOCK_SIZE bsize,
                                    int mi_row, int mi_col, MB_MODE_INFO *mbmi,
                                    HandleInterModeArgs *args,
                                    int64_t ref_best_rd, int *rate_mv,
-                                   int *tmp_rate2, BUFFER_SET *orig_dst) {
+                                   int *tmp_rate2, const BUFFER_SET *orig_dst) {
   const AV1_COMMON *const cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *xd = &x->e_mbd;
 
   INTERINTRA_MODE best_interintra_mode = II_DC_PRED;
-  int64_t rd, best_interintra_rd = INT64_MAX;
+  int64_t rd = INT64_MAX;
+  int64_t best_interintra_rd = INT64_MAX;
   int rmode, rate_sum;
   int64_t dist_sum;
   int tmp_rate_mv = 0;
@@ -8528,61 +8938,115 @@ static int handle_inter_intra_mode(const AV1_COMP *const cpi,
   mbmi->ref_frame[1] = NONE_FRAME;
   xd->plane[0].dst.buf = tmp_buf;
   xd->plane[0].dst.stride = bw;
-  av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, NULL, bsize);
+  av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize,
+                                AOM_PLANE_Y, AOM_PLANE_Y);
 
   restore_dst_buf(xd, *orig_dst, num_planes);
   mbmi->ref_frame[1] = INTRA_FRAME;
-  mbmi->use_wedge_interintra = 0;
   best_interintra_mode = args->inter_intra_mode[mbmi->ref_frame[0]];
-  int j = 0;
-  if (cpi->sf.reuse_inter_intra_mode == 0 ||
-      best_interintra_mode == INTERINTRA_MODES) {
-    for (j = 0; j < INTERINTRA_MODES; ++j) {
-      mbmi->interintra_mode = (INTERINTRA_MODE)j;
-      rmode = interintra_mode_cost[mbmi->interintra_mode];
+
+  if (cpi->oxcf.enable_smooth_interintra) {
+    mbmi->use_wedge_interintra = 0;
+    int j = 0;
+    if (cpi->sf.reuse_inter_intra_mode == 0 ||
+        best_interintra_mode == INTERINTRA_MODES) {
+      for (j = 0; j < INTERINTRA_MODES; ++j) {
+        if (!cpi->oxcf.enable_smooth_intra &&
+            (INTERINTRA_MODE)j == II_SMOOTH_PRED)
+          continue;
+        mbmi->interintra_mode = (INTERINTRA_MODE)j;
+        rmode = interintra_mode_cost[mbmi->interintra_mode];
+        av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0, orig_dst,
+                                                  intrapred, bw);
+        av1_combine_interintra(xd, bsize, 0, tmp_buf, bw, intrapred, bw);
+        model_rd_sb_fn[MODELRD_TYPE_INTERINTRA](
+            cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
+            &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
+        rd = RDCOST(x->rdmult, tmp_rate_mv + rate_sum + rmode, dist_sum);
+        if (rd < best_interintra_rd) {
+          best_interintra_rd = rd;
+          best_interintra_mode = mbmi->interintra_mode;
+        }
+      }
+      args->inter_intra_mode[mbmi->ref_frame[0]] = best_interintra_mode;
+    }
+    assert(IMPLIES(!cpi->oxcf.enable_smooth_interintra,
+                   best_interintra_mode != II_SMOOTH_PRED));
+    rmode = interintra_mode_cost[best_interintra_mode];
+    if (j == 0 || best_interintra_mode != II_SMOOTH_PRED) {
+      mbmi->interintra_mode = best_interintra_mode;
       av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0, orig_dst,
                                                 intrapred, bw);
       av1_combine_interintra(xd, bsize, 0, tmp_buf, bw, intrapred, bw);
-      model_rd_sb_fn[MODELRD_TYPE_INTERINTRA](
-          cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
-          &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
-      rd = RDCOST(x->rdmult, tmp_rate_mv + rate_sum + rmode, dist_sum);
-      if (rd < best_interintra_rd) {
-        best_interintra_rd = rd;
-        best_interintra_mode = mbmi->interintra_mode;
-      }
     }
-    args->inter_intra_mode[mbmi->ref_frame[0]] = best_interintra_mode;
-  }
-  if (j == 0 || best_interintra_mode != II_SMOOTH_PRED) {
-    mbmi->interintra_mode = best_interintra_mode;
-    rmode = interintra_mode_cost[mbmi->interintra_mode];
-    av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0, orig_dst,
-                                              intrapred, bw);
-    av1_combine_interintra(xd, bsize, 0, tmp_buf, bw, intrapred, bw);
-  }
-  rd = estimate_yrd_for_sb(cpi, bsize, x, &rate_sum, &dist_sum,
-                           &tmp_skip_txfm_sb, &tmp_skip_sse_sb, INT64_MAX);
-  if (rd != INT64_MAX)
-    rd = RDCOST(x->rdmult, *rate_mv + rmode + rate_sum + rwedge, dist_sum);
-  best_interintra_rd = rd;
-  if (ref_best_rd < INT64_MAX && (best_interintra_rd >> 1) > ref_best_rd) {
-    return -1;
+
+    RD_STATS rd_stats;
+    rd = estimate_yrd_for_sb(cpi, bsize, x, INT64_MAX, &rd_stats);
+    if (rd != INT64_MAX) {
+      rd = RDCOST(x->rdmult, *rate_mv + rmode + rd_stats.rate + rwedge,
+                  rd_stats.dist);
+    }
+    best_interintra_rd = rd;
+    if (ref_best_rd < INT64_MAX &&
+        ((best_interintra_rd >> 4) * 9) > ref_best_rd) {
+      return -1;
+    }
   }
   if (is_wedge_used) {
     int64_t best_interintra_rd_nowedge = rd;
     int64_t best_interintra_rd_wedge = INT64_MAX;
     int_mv tmp_mv;
-    // Disable wedge search if source variance is small
-    if (x->source_variance > cpi->sf.disable_wedge_search_var_thresh) {
+    if (enable_wedge_interintra_search(x, cpi)) {
       mbmi->use_wedge_interintra = 1;
 
       rwedge = av1_cost_literal(get_interintra_wedge_bits(bsize)) +
                x->wedge_interintra_cost[bsize][1];
 
-      best_interintra_rd_wedge =
-          pick_interintra_wedge(cpi, x, bsize, intrapred_, tmp_buf_);
+      if (!cpi->oxcf.enable_smooth_interintra) {
+        if (best_interintra_mode == INTERINTRA_MODES) {
+          mbmi->interintra_mode = II_SMOOTH_PRED;
+          best_interintra_mode = II_SMOOTH_PRED;
+          av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0, orig_dst,
+                                                    intrapred, bw);
+          best_interintra_rd_wedge =
+              pick_interintra_wedge(cpi, x, bsize, intrapred_, tmp_buf_);
 
+          int j = 0;
+          for (j = 0; j < INTERINTRA_MODES; ++j) {
+            mbmi->interintra_mode = (INTERINTRA_MODE)j;
+            rmode = interintra_mode_cost[mbmi->interintra_mode];
+            av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0,
+                                                      orig_dst, intrapred, bw);
+            av1_combine_interintra(xd, bsize, 0, tmp_buf, bw, intrapred, bw);
+            model_rd_sb_fn[MODELRD_TYPE_INTERINTRA](
+                cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
+                &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
+            rd = RDCOST(x->rdmult, tmp_rate_mv + rate_sum + rmode, dist_sum);
+            if (rd < best_interintra_rd) {
+              best_interintra_rd_wedge = rd;
+              best_interintra_mode = mbmi->interintra_mode;
+            }
+          }
+          args->inter_intra_mode[mbmi->ref_frame[0]] = best_interintra_mode;
+          mbmi->interintra_mode = best_interintra_mode;
+
+          if (best_interintra_mode != II_SMOOTH_PRED) {
+            av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0,
+                                                      orig_dst, intrapred, bw);
+          }
+        } else {
+          mbmi->interintra_mode = best_interintra_mode;
+          av1_build_intra_predictors_for_interintra(cm, xd, bsize, 0, orig_dst,
+                                                    intrapred, bw);
+          best_interintra_rd_wedge =
+              pick_interintra_wedge(cpi, x, bsize, intrapred_, tmp_buf_);
+        }
+      } else {
+        best_interintra_rd_wedge =
+            pick_interintra_wedge(cpi, x, bsize, intrapred_, tmp_buf_);
+      }
+
+      rmode = interintra_mode_cost[mbmi->interintra_mode];
       best_interintra_rd_wedge +=
           RDCOST(x->rdmult, rmode + *rate_mv + rwedge, 0);
       rd = INT64_MAX;
@@ -8597,8 +9061,8 @@ static int handle_inter_intra_mode(const AV1_COMP *const cpi,
                                       0);
         if (mbmi->mv[0].as_int != tmp_mv.as_int) {
           mbmi->mv[0].as_int = tmp_mv.as_int;
-          av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, orig_dst,
-                                         bsize);
+          av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                        AOM_PLANE_Y, AOM_PLANE_Y);
           model_rd_sb_fn[MODELRD_TYPE_MASKED_COMPOUND](
               cpi, bsize, x, xd, 0, 0, mi_row, mi_col, &rate_sum, &dist_sum,
               &tmp_skip_txfm_sb, &tmp_skip_sse_sb, NULL, NULL, NULL);
@@ -8612,12 +9076,16 @@ static int handle_inter_intra_mode(const AV1_COMP *const cpi,
         av1_combine_interintra(xd, bsize, 0, tmp_buf, bw, intrapred, bw);
       }
       // Evaluate closer to true rd
-      rd = estimate_yrd_for_sb(cpi, bsize, x, &rate_sum, &dist_sum,
-                               &tmp_skip_txfm_sb, &tmp_skip_sse_sb, INT64_MAX);
-      if (rd != INT64_MAX)
-        rd = RDCOST(x->rdmult, rmode + tmp_rate_mv + rwedge + rate_sum,
-                    dist_sum);
+      RD_STATS rd_stats;
+      rd = estimate_yrd_for_sb(cpi, bsize, x, INT64_MAX, &rd_stats);
+      if (rd != INT64_MAX) {
+        rd = RDCOST(x->rdmult, rmode + tmp_rate_mv + rwedge + rd_stats.rate,
+                    rd_stats.dist);
+      }
       best_interintra_rd_wedge = rd;
+      if (!cpi->oxcf.enable_smooth_interintra &&
+          best_interintra_rd_wedge == INT64_MAX)
+        return -1;
       if (best_interintra_rd_wedge < best_interintra_rd_nowedge) {
         mbmi->use_wedge_interintra = 1;
         mbmi->mv[0].as_int = tmp_mv.as_int;
@@ -8626,33 +9094,100 @@ static int handle_inter_intra_mode(const AV1_COMP *const cpi,
       } else {
         mbmi->use_wedge_interintra = 0;
         mbmi->mv[0].as_int = mv0.as_int;
-        av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, orig_dst, bsize);
+        av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                      AOM_PLANE_Y, AOM_PLANE_Y);
       }
     } else {
+      if (!cpi->oxcf.enable_smooth_interintra) return -1;
       mbmi->use_wedge_interintra = 0;
     }
-  }  // if (is_interintra_wedge_used(bsize))
+  }  // if (is_wedge_used)
   if (num_planes > 1) {
-    av1_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, orig_dst, bsize);
+    av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                  AOM_PLANE_U, num_planes - 1);
   }
   return 0;
 }
 
+// If number of valid neighbours is 1,
+// 1) ROTZOOM parameters can be obtained reliably (2 parameters from
+// one neighbouring MV)
+// 2) For IDENTITY/TRANSLATION cases, warp can perform better due to
+// a different interpolation filter being used. However the quality
+// gains (due to the same) may not be much
+// For above 2 cases warp evaluation is skipped
+
+static int check_if_optimal_warp(const AV1_COMP *cpi,
+                                 WarpedMotionParams *wm_params,
+                                 int num_proj_ref) {
+  int is_valid_warp = 1;
+  if (cpi->sf.prune_warp_using_wmtype) {
+    TransformationType wmtype = get_wmtype(wm_params);
+    if (num_proj_ref == 1) {
+      if (wmtype != ROTZOOM) is_valid_warp = 0;
+    } else {
+      if (wmtype < ROTZOOM) is_valid_warp = 0;
+    }
+  }
+  return is_valid_warp;
+}
+
+struct obmc_check_mv_field_ctxt {
+  MB_MODE_INFO *current_mi;
+  int mv_field_check_result;
+};
+
+static INLINE void obmc_check_identical_mv(MACROBLOCKD *xd, int rel_mi_col,
+                                           uint8_t nb_mi_width,
+                                           MB_MODE_INFO *nb_mi, void *fun_ctxt,
+                                           const int num_planes) {
+  (void)xd;
+  (void)rel_mi_col;
+  (void)nb_mi_width;
+  (void)num_planes;
+  struct obmc_check_mv_field_ctxt *ctxt =
+      (struct obmc_check_mv_field_ctxt *)fun_ctxt;
+  const MB_MODE_INFO *current_mi = ctxt->current_mi;
+
+  if (ctxt->mv_field_check_result == 0) return;
+
+  if (nb_mi->ref_frame[0] != current_mi->ref_frame[0] ||
+      nb_mi->mv[0].as_int != current_mi->mv[0].as_int ||
+      nb_mi->interp_filters != current_mi->interp_filters) {
+    ctxt->mv_field_check_result = 0;
+  }
+  return;
+}
+
+// Check if the neighbors' motions used by obmc have same parameters as for
+// the current block. If all the parameters are identical, obmc will produce
+// the same prediction as from regular bmc, therefore we can skip the
+// overlapping operations for less complexity. The parameters checked include
+// reference frame, motion vector, and interpolation filter.
+int check_identical_obmc_mv_field(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                  int mi_row, int mi_col) {
+  const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
+  struct obmc_check_mv_field_ctxt mv_field_check_ctxt = { xd->mi[0], 1 };
+
+  foreach_overlappable_nb_above(cm, xd, mi_col,
+                                max_neighbor_obmc[mi_size_wide_log2[bsize]],
+                                obmc_check_identical_mv, &mv_field_check_ctxt);
+  foreach_overlappable_nb_left(cm, xd, mi_row,
+                               max_neighbor_obmc[mi_size_high_log2[bsize]],
+                               obmc_check_identical_mv, &mv_field_check_ctxt);
+
+  return mv_field_check_ctxt.mv_field_check_result;
+}
+
 // TODO(afergs): Refactor the MBMI references in here - there's four
 // TODO(afergs): Refactor optional args - add them to a struct or remove
-static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
-                              BLOCK_SIZE bsize, RD_STATS *rd_stats,
-                              RD_STATS *rd_stats_y, RD_STATS *rd_stats_uv,
-                              int *disable_skip, int mi_row, int mi_col,
-                              HandleInterModeArgs *const args,
-                              int64_t ref_best_rd, const int *refs,
-                              int *rate_mv, BUFFER_SET *orig_dst
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-                              ,
-                              TileDataEnc *tile_data, int64_t *best_est_rd,
-                              int do_tx_search, InterModesInfo *inter_modes_info
-#endif
-) {
+static int64_t motion_mode_rd(
+    const AV1_COMP *const cpi, TileDataEnc *tile_data, MACROBLOCK *const x,
+    BLOCK_SIZE bsize, RD_STATS *rd_stats, RD_STATS *rd_stats_y,
+    RD_STATS *rd_stats_uv, int *disable_skip, int mi_row, int mi_col,
+    HandleInterModeArgs *const args, int64_t ref_best_rd, const int *refs,
+    int *rate_mv, const BUFFER_SET *orig_dst, int64_t *best_est_rd,
+    int do_tx_search, InterModesInfo *inter_modes_info) {
   const AV1_COMMON *const cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *xd = &x->e_mbd;
@@ -8660,18 +9195,18 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
   const int is_comp_pred = has_second_ref(mbmi);
   const PREDICTION_MODE this_mode = mbmi->mode;
   const int rate2_nocoeff = rd_stats->rate;
-  int best_xskip, best_disable_skip = 0;
+  int best_xskip = 0, best_disable_skip = 0;
   RD_STATS best_rd_stats, best_rd_stats_y, best_rd_stats_uv;
-  MB_MODE_INFO base_mbmi, best_mbmi;
   uint8_t best_blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE];
   const int rate_mv0 = *rate_mv;
-
-  int interintra_allowed = cm->seq_params.enable_interintra_compound &&
-                           is_interintra_allowed(mbmi) && mbmi->compound_idx;
+  const int interintra_allowed = cm->seq_params.enable_interintra_compound &&
+                                 is_interintra_allowed(mbmi) &&
+                                 mbmi->compound_idx;
   int pts0[SAMPLES_ARRAY_SIZE], pts_inref0[SAMPLES_ARRAY_SIZE];
 
   assert(mbmi->ref_frame[1] != INTRA_FRAME);
   const MV_REFERENCE_FRAME ref_frame_1 = mbmi->ref_frame[1];
+  (void)tile_data;
   av1_invalid_rd_stats(&best_rd_stats);
   aom_clear_system_state();
   mbmi->num_proj_ref = 1;  // assume num_proj_ref >=1
@@ -8683,24 +9218,32 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
   if (last_motion_mode_allowed == WARPED_CAUSAL) {
     mbmi->num_proj_ref = findSamples(cm, xd, mi_row, mi_col, pts0, pts_inref0);
   }
-  int total_samples = mbmi->num_proj_ref;
+  const int total_samples = mbmi->num_proj_ref;
   if (total_samples == 0) {
     last_motion_mode_allowed = OBMC_CAUSAL;
   }
-  base_mbmi = *mbmi;
 
+  const MB_MODE_INFO base_mbmi = *mbmi;
+  MB_MODE_INFO best_mbmi;
+  SimpleRDState *const simple_states = &args->simple_rd_state[mbmi->ref_mv_idx];
   const int switchable_rate =
       av1_is_interp_needed(xd) ? av1_get_switchable_rate(cm, x, xd) : 0;
   int64_t best_rd = INT64_MAX;
   int best_rate_mv = rate_mv0;
+  const int identical_obmc_mv_field_detected =
+      (cpi->sf.skip_obmc_in_uniform_mv_field ||
+       cpi->sf.skip_wm_in_uniform_mv_field)
+          ? check_identical_obmc_mv_field(cm, xd, mi_row, mi_col)
+          : 0;
   for (int mode_index = (int)SIMPLE_TRANSLATION;
        mode_index <= (int)last_motion_mode_allowed + interintra_allowed;
        mode_index++) {
     if (args->skip_motion_mode && mode_index) continue;
-    int64_t tmp_rd = INT64_MAX;
+    if (cpi->sf.prune_single_motion_modes_by_simple_trans &&
+        args->single_ref_first_pass && mode_index)
+      break;
     int tmp_rate2 = rate2_nocoeff;
-    int is_interintra_mode = mode_index > (int)last_motion_mode_allowed;
-    int skip_txfm_sb = 0;
+    const int is_interintra_mode = mode_index > (int)last_motion_mode_allowed;
     int tmp_rate_mv = rate_mv0;
 
     *mbmi = base_mbmi;
@@ -8711,12 +9254,44 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
       assert(mbmi->ref_frame[1] != INTRA_FRAME);
     }
 
+    if (cpi->oxcf.enable_obmc == 0 && mbmi->motion_mode == OBMC_CAUSAL)
+      continue;
+
+    if (identical_obmc_mv_field_detected) {
+      if (cpi->sf.skip_obmc_in_uniform_mv_field &&
+          mbmi->motion_mode == OBMC_CAUSAL)
+        continue;
+      if (cpi->sf.skip_wm_in_uniform_mv_field &&
+          mbmi->motion_mode == WARPED_CAUSAL)
+        continue;
+    }
+
     if (mbmi->motion_mode == SIMPLE_TRANSLATION && !is_interintra_mode) {
       // SIMPLE_TRANSLATION mode: no need to recalculate.
       // The prediction is calculated before motion_mode_rd() is called in
       // handle_inter_mode()
+      if (cpi->sf.prune_single_motion_modes_by_simple_trans &&
+          args->single_ref_first_pass == 0 && !is_comp_pred) {
+        if (simple_states->early_skipped) {
+          assert(simple_states->rd_stats.rdcost == INT64_MAX);
+          return INT64_MAX;
+        }
+        if (simple_states->rd_stats.rdcost != INT64_MAX) {
+          best_rd = simple_states->rd_stats.rdcost;
+          best_rd_stats = simple_states->rd_stats;
+          best_rd_stats_y = simple_states->rd_stats_y;
+          best_rd_stats_uv = simple_states->rd_stats_uv;
+          memcpy(best_blk_skip, simple_states->blk_skip,
+                 sizeof(x->blk_skip[0]) * xd->n4_h * xd->n4_w);
+          best_xskip = simple_states->skip;
+          best_disable_skip = simple_states->disable_skip;
+          best_mbmi = *mbmi;
+        }
+        continue;
+      }
+      simple_states->early_skipped = 0;
     } else if (mbmi->motion_mode == OBMC_CAUSAL) {
-      uint32_t cur_mv = mbmi->mv[0].as_int;
+      const uint32_t cur_mv = mbmi->mv[0].as_int;
       assert(!is_comp_pred);
       if (have_newmv_in_inter_mode(this_mode)) {
         single_motion_search(cpi, x, bsize, mi_row, mi_col, 0, &tmp_rate_mv);
@@ -8729,7 +9304,8 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
         tmp_rate2 = rate2_nocoeff - rate_mv0 + tmp_rate_mv;
       }
       if (mbmi->mv[0].as_int != cur_mv) {
-        av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, orig_dst, bsize);
+        av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                      0, av1_num_planes(cm) - 1);
       }
       av1_build_obmc_inter_prediction(
           cm, xd, mi_row, mi_col, args->above_pred_buf, args->above_pred_stride,
@@ -8757,7 +9333,12 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
         if (have_newmv_in_inter_mode(this_mode)) {
           const int_mv mv0 = mbmi->mv[0];
           const WarpedMotionParams wm_params0 = mbmi->wm_params;
-          int num_proj_ref0 = mbmi->num_proj_ref;
+          const int num_proj_ref0 = mbmi->num_proj_ref;
+
+          if (cpi->sf.prune_warp_using_wmtype) {
+            TransformationType wmtype = get_wmtype(&mbmi->wm_params);
+            if (wmtype < ROTZOOM) continue;
+          }
 
           // Refine MV in a small range.
           av1_refine_warped_mv(cpi, x, bsize, mi_row, mi_col, pts0, pts_inref0,
@@ -8767,9 +9348,9 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
           if (mv0.as_int != mbmi->mv[0].as_int) {
             const int ref = refs[0];
             const int_mv ref_mv = av1_get_ref_mv(x, 0);
-            tmp_rate_mv =
-                av1_mv_bit_cost(&mbmi->mv[0].as_mv, &ref_mv.as_mv,
-                                x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
+            tmp_rate_mv = av1_mv_bit_cost(&mbmi->mv[0].as_mv, &ref_mv.as_mv,
+                                          x->nmv_vec_cost, x->mv_cost_stack,
+                                          MV_COST_WEIGHT);
 
             if (cpi->sf.adaptive_motion_search)
               x->pred_mv[ref] = mbmi->mv[0].as_mv;
@@ -8786,9 +9367,13 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
             mbmi->wm_params = wm_params0;
             mbmi->num_proj_ref = num_proj_ref0;
           }
+        } else {
+          if (!check_if_optimal_warp(cpi, &mbmi->wm_params, mbmi->num_proj_ref))
+            continue;
         }
 
-        av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
+        av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize, 0,
+                                      av1_num_planes(cm) - 1);
       } else {
         continue;
       }
@@ -8799,11 +9384,7 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
       if (ret < 0) continue;
     }
 
-    if (!cpi->common.all_lossless)
-      check_block_skip(cpi, bsize, x, xd, 0, num_planes - 1, &skip_txfm_sb);
-
     x->skip = 0;
-
     rd_stats->dist = 0;
     rd_stats->sse = 0;
     rd_stats->skip = 1;
@@ -8834,45 +9415,47 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
       }
     }
 
-    if (!skip_txfm_sb) {
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-      int64_t est_rd = 0;
-      int est_skip = 0;
-      if (cpi->sf.inter_mode_rd_model_estimation && cm->tile_cols == 1 &&
-          cm->tile_rows == 1) {
-        InterModeRdModel *md = &tile_data->inter_mode_rd_models[mbmi->sb_type];
-        if (md->ready) {
-          const int64_t curr_sse = get_sse(cpi, x);
-          est_rd = get_est_rd(tile_data, mbmi->sb_type, x->rdmult, curr_sse,
-                              rd_stats->rate);
-          est_skip = est_rd * 0.8 > *best_est_rd;
-          if (est_skip) {
-            mbmi->ref_frame[1] = ref_frame_1;
-            continue;
-          } else {
-            if (est_rd < *best_est_rd) {
-              *best_est_rd = est_rd;
-            }
-          }
-        }
+    if (cpi->sf.model_based_motion_mode_rd_breakout && do_tx_search) {
+      int model_rate;
+      int64_t model_dist;
+      model_rd_sb_fn[MODELRD_TYPE_MOTION_MODE_RD](
+          cpi, mbmi->sb_type, x, xd, 0, num_planes - 1, mi_row, mi_col,
+          &model_rate, &model_dist, NULL, NULL, NULL, NULL, NULL);
+      const int64_t est_rd =
+          RDCOST(x->rdmult, rd_stats->rate + model_rate, model_dist);
+      if ((est_rd >> 3) * 6 > ref_best_rd) {
+        mbmi->ref_frame[1] = ref_frame_1;
+        continue;
       }
-#endif  // CONFIG_COLLECT_INTER_MODE_RD_STATS
     }
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
     if (!do_tx_search) {
-      const int64_t curr_sse = get_sse(cpi, x);
+      int64_t curr_sse;
       int est_residue_cost = 0;
       int64_t est_dist = 0;
-      const int has_est_rd = get_est_rate_dist(tile_data, bsize, curr_sse,
-                                               &est_residue_cost, &est_dist);
-      (void)has_est_rd;
-      assert(has_est_rd);
+      int64_t est_rd = 0;
+      if (cpi->sf.inter_mode_rd_model_estimation == 1) {
+        curr_sse = get_sse(cpi, x);
+        const int has_est_rd = get_est_rate_dist(tile_data, bsize, curr_sse,
+                                                 &est_residue_cost, &est_dist);
+        (void)has_est_rd;
+        assert(has_est_rd);
+      } else if (cpi->sf.inter_mode_rd_model_estimation == 2) {
+        model_rd_sb_fn[MODELRD_TYPE_MOTION_MODE_RD](
+            cpi, bsize, x, xd, 0, num_planes - 1, mi_row, mi_col,
+            &est_residue_cost, &est_dist, NULL, &curr_sse, NULL, NULL, NULL);
+      }
+      est_rd = RDCOST(x->rdmult, rd_stats->rate + est_residue_cost, est_dist);
+      if (est_rd * 0.8 > *best_est_rd) {
+        mbmi->ref_frame[1] = ref_frame_1;
+        continue;
+      }
       const int mode_rate = rd_stats->rate;
       rd_stats->rate += est_residue_cost;
       rd_stats->dist = est_dist;
-      rd_stats->rdcost = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
-      if (cm->reference_mode == SINGLE_REFERENCE) {
+      rd_stats->rdcost = est_rd;
+      *best_est_rd = AOMMIN(*best_est_rd, rd_stats->rdcost);
+      if (cm->current_frame.reference_mode == SINGLE_REFERENCE) {
         if (!is_comp_pred) {
           inter_modes_info_push(inter_modes_info, mode_rate, curr_sse,
                                 rd_stats->rdcost, mbmi);
@@ -8882,37 +9465,26 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
                               rd_stats->rdcost, mbmi);
       }
     } else {
-#endif
-      int mode_rate = rd_stats->rate;
-      if (!txfm_search(cpi, x, bsize, mi_row, mi_col, rd_stats, rd_stats_y,
-                       rd_stats_uv, mode_rate, ref_best_rd)) {
+      if (!txfm_search(cpi, tile_data, x, bsize, mi_row, mi_col, rd_stats,
+                       rd_stats_y, rd_stats_uv, rd_stats->rate, ref_best_rd)) {
         if (rd_stats_y->rate == INT_MAX && mode_index == 0) {
+          simple_states->early_skipped = 1;
           return INT64_MAX;
         }
         continue;
       }
-      if (!skip_txfm_sb) {
-        const int64_t curr_rd =
-            RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
-        if (curr_rd < ref_best_rd) {
-          ref_best_rd = curr_rd;
-        }
-        *disable_skip = 0;
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-        if (cpi->sf.inter_mode_rd_model_estimation) {
-          const int skip_ctx = av1_get_skip_context(xd);
-          inter_mode_data_push(tile_data, mbmi->sb_type, rd_stats->sse,
-                               rd_stats->dist,
-                               rd_stats_y->rate + rd_stats_uv->rate +
-                                   x->skip_cost[skip_ctx][mbmi->skip]);
-        }
-#endif  // CONFIG_COLLECT_INTER_MODE_RD_STATS
-      } else {
-        *disable_skip = 1;
+
+      const int64_t curr_rd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
+      ref_best_rd = AOMMIN(ref_best_rd, curr_rd);
+      *disable_skip = 0;
+      if (cpi->sf.inter_mode_rd_model_estimation == 1) {
+        const int skip_ctx = av1_get_skip_context(xd);
+        inter_mode_data_push(tile_data, mbmi->sb_type, rd_stats->sse,
+                             rd_stats->dist,
+                             rd_stats_y->rate + rd_stats_uv->rate +
+                                 x->skip_cost[skip_ctx][mbmi->skip]);
       }
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
     }
-#endif
 
     if (this_mode == GLOBALMV || this_mode == GLOBAL_GLOBALMV) {
       if (is_nontrans_global_motion(xd, xd->mi[0])) {
@@ -8921,10 +9493,21 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
       }
     }
 
-    tmp_rd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
-    if (mode_index == 0)
+    const int64_t tmp_rd = RDCOST(x->rdmult, rd_stats->rate, rd_stats->dist);
+    if (mode_index == 0) {
       args->simple_rd[this_mode][mbmi->ref_mv_idx][mbmi->ref_frame[0]] = tmp_rd;
-    if ((mode_index == 0) || (tmp_rd < best_rd)) {
+      if (!is_comp_pred) {
+        simple_states->rd_stats = *rd_stats;
+        simple_states->rd_stats.rdcost = tmp_rd;
+        simple_states->rd_stats_y = *rd_stats_y;
+        simple_states->rd_stats_uv = *rd_stats_uv;
+        memcpy(simple_states->blk_skip, x->blk_skip,
+               sizeof(x->blk_skip[0]) * xd->n4_h * xd->n4_w);
+        simple_states->skip = x->skip;
+        simple_states->disable_skip = *disable_skip;
+      }
+    }
+    if (mode_index == 0 || tmp_rd < best_rd) {
       best_mbmi = *mbmi;
       best_rd = tmp_rd;
       best_rd_stats = *rd_stats;
@@ -8960,11 +9543,12 @@ static int64_t motion_mode_rd(const AV1_COMP *const cpi, MACROBLOCK *const x,
 
 static int64_t skip_mode_rd(RD_STATS *rd_stats, const AV1_COMP *const cpi,
                             MACROBLOCK *const x, BLOCK_SIZE bsize, int mi_row,
-                            int mi_col, BUFFER_SET *const orig_dst) {
+                            int mi_col, const BUFFER_SET *const orig_dst) {
   const AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *const xd = &x->e_mbd;
-  av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, orig_dst, bsize);
+  av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize, 0,
+                                av1_num_planes(cm) - 1);
 
   int64_t total_sse = 0;
   for (int plane = 0; plane < num_planes; ++plane) {
@@ -8976,9 +9560,44 @@ static int64_t skip_mode_rd(RD_STATS *rd_stats, const AV1_COMP *const cpi,
     const int bh = block_size_high[plane_bsize];
 
     av1_subtract_plane(x, bsize, plane);
-    int64_t sse = aom_sum_squares_2d_i16(p->src_diff, bw, bw, bh);
+    int64_t sse;
+#if CONFIG_ONE_PASS_SVM
+    if (plane == AOM_PLANE_Y && bsize >= BLOCK_8X8 && bw == bh) {
+      rd_stats->sse_0 = aom_sum_squares_2d_i16(p->src_diff, bw, bw / 2, bh / 2)
+                        << 4;
+      rd_stats->sse_1 =
+          aom_sum_squares_2d_i16(p->src_diff + bw / 2, bw, bw / 2, bh / 2) << 4;
+      rd_stats->sse_2 =
+          aom_sum_squares_2d_i16(p->src_diff + bh / 2 * bw, bw, bw / 2, bh / 2)
+          << 4;
+      rd_stats->sse_3 =
+          aom_sum_squares_2d_i16(p->src_diff + bh / 2 * bw + bw / 2, bw, bw / 2,
+                                 bh / 2)
+          << 4;
+
+      sse =
+          rd_stats->sse_0 + rd_stats->sse_1 + rd_stats->sse_2 + rd_stats->sse_3;
+      total_sse += sse;
+
+      const int scaling_factor = MAX_MIB_SIZE * MAX_MIB_SIZE;
+      rd_stats->sse = sse;
+      rd_stats->sse_0 = rd_stats->sse_0 * scaling_factor;
+      rd_stats->sse_1 = rd_stats->sse_1 * scaling_factor;
+      rd_stats->sse_2 = rd_stats->sse_2 * scaling_factor;
+      rd_stats->sse_3 = rd_stats->sse_3 * scaling_factor;
+      rd_stats->y_sse = sse;
+      // TODO(chiyotsai@google.com): Don't manually set the flags
+      av1_reg_stat_skipmode_update(rd_stats, x->rdmult);
+    } else {
+      sse = aom_sum_squares_2d_i16(p->src_diff, bw, bw, bh);
+      sse = sse << 4;
+      total_sse += sse;
+    }
+#else
+    sse = aom_sum_squares_2d_i16(p->src_diff, bw, bw, bh);
     sse = sse << 4;
     total_sse += sse;
+#endif
   }
   const int skip_mode_ctx = av1_get_skip_mode_context(xd);
   rd_stats->dist = rd_stats->sse = total_sse;
@@ -9101,22 +9720,19 @@ typedef struct {
 static int compound_type_rd(const AV1_COMP *const cpi, MACROBLOCK *x,
                             BLOCK_SIZE bsize, int mi_col, int mi_row,
                             int_mv *cur_mv, int masked_compound_used,
-                            BUFFER_SET *orig_dst, const BUFFER_SET *tmp_dst,
+                            const BUFFER_SET *orig_dst,
+                            const BUFFER_SET *tmp_dst,
                             CompoundTypeRdBuffers *buffers, int *rate_mv,
                             int64_t *rd, RD_STATS *rd_stats,
-                            int64_t ref_best_rd) {
+                            int64_t ref_best_rd, int *is_luma_interp_done) {
   const AV1_COMMON *cm = &cpi->common;
   MACROBLOCKD *xd = &x->e_mbd;
   MB_MODE_INFO *mbmi = xd->mi[0];
   const PREDICTION_MODE this_mode = mbmi->mode;
   const int bw = block_size_wide[bsize];
-  int rate_sum, rs2;
-  int64_t dist_sum;
-
+  int rs2;
   int_mv best_mv[2];
   int best_tmp_rate_mv = *rate_mv;
-  int tmp_skip_txfm_sb;
-  int64_t tmp_skip_sse_sb;
   INTERINTER_COMPOUND_DATA best_compound_data;
   best_compound_data.type = COMPOUND_AVERAGE;
   uint8_t *preds0[1] = { buffers->pred0 };
@@ -9128,7 +9744,9 @@ static int compound_type_rd(const AV1_COMP *const cpi, MACROBLOCK *x,
   COMPOUND_TYPE cur_type;
   int best_compmode_interinter_cost = 0;
   int calc_pred_masked_compound = 1;
-
+  int64_t comp_dist[COMPOUND_TYPES] = { INT64_MAX, INT64_MAX, INT64_MAX };
+  int32_t comp_rate[COMPOUND_TYPES] = { INT_MAX, INT_MAX, INT_MAX };
+  const int match_found = find_comp_rd_in_stats(x, mbmi, comp_rate, comp_dist);
   best_mv[0].as_int = cur_mv[0].as_int;
   best_mv[1].as_int = cur_mv[1].as_int;
   *rd = INT64_MAX;
@@ -9152,12 +9770,31 @@ static int compound_type_rd(const AV1_COMP *const cpi, MACROBLOCK *x,
       rs2 = masked_type_cost;
       const int64_t mode_rd = RDCOST(x->rdmult, rs2 + rd_stats->rate, 0);
       if (mode_rd < ref_best_rd) {
-        av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, orig_dst, bsize);
-        int64_t est_rd =
-            estimate_yrd_for_sb(cpi, bsize, x, &rate_sum, &dist_sum,
-                                &tmp_skip_txfm_sb, &tmp_skip_sse_sb, INT64_MAX);
-        if (est_rd != INT64_MAX)
-          best_rd_cur = RDCOST(x->rdmult, rs2 + *rate_mv + rate_sum, dist_sum);
+        // Reuse data if matching record is found
+        if (comp_rate[0] == INT_MAX) {
+          av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, orig_dst, bsize,
+                                        AOM_PLANE_Y, AOM_PLANE_Y);
+          *is_luma_interp_done = 1;
+          RD_STATS est_rd_stats;
+          const int64_t est_rd =
+              estimate_yrd_for_sb(cpi, bsize, x, INT64_MAX, &est_rd_stats);
+          if (comp_rate[0] != INT_MAX) {
+            assert(comp_rate[0] == est_rd_stats.rate);
+            assert(comp_dist[0] == est_rd_stats.dist);
+          }
+          if (est_rd != INT64_MAX) {
+            best_rd_cur = RDCOST(x->rdmult, rs2 + *rate_mv + est_rd_stats.rate,
+                                 est_rd_stats.dist);
+            // Backup rate and distortion for future reuse
+            comp_rate[0] = est_rd_stats.rate;
+            comp_dist[0] = est_rd_stats.dist;
+          }
+        } else {
+          // Calculate RD cost based on stored stats
+          assert(comp_dist[0] != INT64_MAX);
+          best_rd_cur =
+              RDCOST(x->rdmult, rs2 + *rate_mv + comp_rate[0], comp_dist[0]);
+        }
       }
       // use spare buffer for following compound type try
       restore_dst_buf(xd, *tmp_dst, 1);
@@ -9166,13 +9803,20 @@ static int compound_type_rd(const AV1_COMP *const cpi, MACROBLOCK *x,
       masked_type_cost += x->comp_group_idx_cost[comp_group_idx_ctx][1];
       masked_type_cost += x->compound_type_cost[bsize][cur_type - 1];
       rs2 = masked_type_cost;
-      if (x->source_variance > cpi->sf.disable_wedge_search_var_thresh &&
-          *rd / 3 < ref_best_rd) {
-        best_rd_cur = build_and_cost_compound_type(
-            cpi, x, cur_mv, bsize, this_mode, &rs2, *rate_mv, orig_dst,
-            &tmp_rate_mv, preds0, preds1, buffers->residual1, buffers->diff10,
-            strides, mi_row, mi_col, rd_stats->rate, ref_best_rd,
-            &calc_pred_masked_compound);
+
+      if (((*rd / cpi->max_comp_type_rd_threshold_div) *
+           cpi->max_comp_type_rd_threshold_mul) < ref_best_rd) {
+        const COMPOUND_TYPE compound_type = mbmi->interinter_comp.type;
+
+        if (!((compound_type == COMPOUND_WEDGE &&
+               !enable_wedge_interinter_search(x, cpi)) ||
+              (compound_type == COMPOUND_DIFFWTD &&
+               !cpi->oxcf.enable_diff_wtd_comp)))
+          best_rd_cur = build_and_cost_compound_type(
+              cpi, x, cur_mv, bsize, this_mode, &rs2, *rate_mv, orig_dst,
+              &tmp_rate_mv, preds0, preds1, buffers->residual1, buffers->diff10,
+              strides, mi_row, mi_col, rd_stats->rate, ref_best_rd,
+              &calc_pred_masked_compound, comp_rate, comp_dist);
       }
     }
     if (best_rd_cur < *rd) {
@@ -9212,6 +9856,8 @@ static int compound_type_rd(const AV1_COMP *const cpi, MACROBLOCK *x,
     }
   }
   restore_dst_buf(xd, *orig_dst, 1);
+  if (!match_found)
+    save_comp_rd_search_stat(x, mbmi, comp_rate, comp_dist, cur_mv);
   return best_compmode_interinter_cost;
 }
 
@@ -9252,20 +9898,13 @@ typedef struct {
   int_mv mv;
 } inter_mode_info;
 
-static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
-                                 BLOCK_SIZE bsize, RD_STATS *rd_stats,
-                                 RD_STATS *rd_stats_y, RD_STATS *rd_stats_uv,
-                                 int *disable_skip, int mi_row, int mi_col,
-                                 HandleInterModeArgs *args, int64_t ref_best_rd,
-                                 uint8_t *const tmp_buf,
-                                 CompoundTypeRdBuffers *rd_buffers
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-                                 ,
-                                 TileDataEnc *tile_data, int64_t *best_est_rd,
-                                 const int do_tx_search,
-                                 InterModesInfo *inter_modes_info
-#endif
-) {
+static int64_t handle_inter_mode(
+    const AV1_COMP *const cpi, TileDataEnc *tile_data, MACROBLOCK *x,
+    BLOCK_SIZE bsize, RD_STATS *rd_stats, RD_STATS *rd_stats_y,
+    RD_STATS *rd_stats_uv, int *disable_skip, int mi_row, int mi_col,
+    HandleInterModeArgs *args, int64_t ref_best_rd, uint8_t *const tmp_buf,
+    CompoundTypeRdBuffers *rd_buffers, int64_t *best_est_rd,
+    const int do_tx_search, InterModesInfo *inter_modes_info) {
   const AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *xd = &x->e_mbd;
@@ -9285,7 +9924,7 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
   // one for future predictions. In the end, copy from tmp_buf to
   // dst if necessary.
   struct macroblockd_plane *p = xd->plane;
-  BUFFER_SET orig_dst = {
+  const BUFFER_SET orig_dst = {
     { p[0].dst.buf, p[1].dst.buf, p[2].dst.buf },
     { p[0].dst.stride, p[1].dst.stride, p[2].dst.stride },
   };
@@ -9312,8 +9951,10 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
   inter_mode_info mode_info[MAX_REF_MV_SERCH];
 
   int comp_idx;
-  const int search_jnt_comp = is_comp_pred & cm->seq_params.enable_jnt_comp &
-                              (mbmi->mode != GLOBAL_GLOBALMV);
+  const int search_dist_wtd_comp =
+      is_comp_pred & cm->seq_params.order_hint_info.enable_dist_wtd_comp &
+      (mbmi->mode != GLOBAL_GLOBALMV) &
+      (cpi->sf.use_dist_wtd_comp_flag != DIST_WTD_COMP_DISABLED);
 
   // TODO(jingning): This should be deprecated shortly.
   const int has_nearmv = have_nearmv_in_inter_mode(mbmi->mode) ? 1 : 0;
@@ -9334,7 +9975,12 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
         }
       }
     }
-
+    if (cpi->sf.prune_single_motion_modes_by_simple_trans && !is_comp_pred &&
+        args->single_ref_first_pass == 0) {
+      if (args->simple_rd_state[ref_mv_idx].early_skipped) {
+        continue;
+      }
+    }
     av1_init_rd_stats(rd_stats);
 
     mbmi->interinter_comp.type = COMPOUND_AVERAGE;
@@ -9364,28 +10010,20 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
       continue;
     }
 
-    int64_t best_rd2 = INT64_MAX;
-
     const RD_STATS backup_rd_stats = *rd_stats;
-    // If !search_jnt_comp, we need to force mbmi->compound_idx = 1.
-    for (comp_idx = 1; comp_idx >= !search_jnt_comp; --comp_idx) {
+    // If !search_dist_wtd_comp, we need to force mbmi->compound_idx = 1.
+    for (comp_idx = 1; comp_idx >= !search_dist_wtd_comp; --comp_idx) {
       int rs = 0;
       int compmode_interinter_cost = 0;
       mbmi->compound_idx = comp_idx;
       if (is_comp_pred && comp_idx == 0) {
         *rd_stats = backup_rd_stats;
         mbmi->interinter_comp.type = COMPOUND_AVERAGE;
-        if (mbmi->ref_frame[1] == INTRA_FRAME) mbmi->ref_frame[1] = NONE_FRAME;
         mbmi->num_proj_ref = 0;
         mbmi->motion_mode = SIMPLE_TRANSLATION;
         mbmi->comp_group_idx = 0;
 
-        const int comp_group_idx_ctx = get_comp_group_idx_context(xd);
         const int comp_index_ctx = get_comp_index_context(cm, xd);
-        if (masked_compound_used) {
-          compmode_interinter_cost +=
-              x->comp_group_idx_cost[comp_group_idx_ctx][0];
-        }
         compmode_interinter_cost += x->comp_idx_cost[comp_index_ctx][0];
       }
 
@@ -9400,9 +10038,16 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
           rate_mv = backup_rate_mv;
         }
 
-        // when jnt_comp_skip_mv_search flag is on, new mv will be searched once
-        if (!(search_jnt_comp && cpi->sf.jnt_comp_skip_mv_search &&
-              comp_idx == 0)) {
+        if (cpi->sf.prune_single_motion_modes_by_simple_trans &&
+            args->single_ref_first_pass == 0 && !is_comp_pred) {
+          const int ref0 = mbmi->ref_frame[0];
+          newmv_ret_val = args->single_newmv_valid[ref_mv_idx][ref0] ? 0 : 1;
+          cur_mv[0] = args->single_newmv[ref_mv_idx][ref0];
+          rate_mv = args->single_newmv_rate[ref_mv_idx][ref0];
+        } else if (!(search_dist_wtd_comp &&
+                     (cpi->sf.use_dist_wtd_comp_flag ==
+                      DIST_WTD_COMP_SKIP_MV_SEARCH) &&
+                     comp_idx == 0)) {
           newmv_ret_val = handle_newmv(cpi, x, bsize, cur_mv, mi_row, mi_col,
                                        &rate_mv, args);
 
@@ -9437,9 +10082,9 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
                   const int compare_cost =
                       mode_info[i].rate_mv + mode_info[i].drl_cost;
                   const int_mv ref_mv = av1_get_ref_mv(x, 0);
-                  this_rate_mv = av1_mv_bit_cost(&mode_info[i].mv.as_mv,
-                                                 &ref_mv.as_mv, x->nmvjointcost,
-                                                 x->mvcost, MV_COST_WEIGHT);
+                  this_rate_mv = av1_mv_bit_cost(
+                      &mode_info[i].mv.as_mv, &ref_mv.as_mv, x->nmv_vec_cost,
+                      x->mv_cost_stack, MV_COST_WEIGHT);
                   const int this_cost = this_rate_mv + drl_cost;
 
                   if (compare_cost < this_cost) {
@@ -9455,7 +10100,6 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
                       best_rd = RDCOST(x->rdmult, best_rd_stats.rate,
                                        best_rd_stats.dist);
                       if (best_rd < ref_best_rd) ref_best_rd = best_rd;
-
                       skip = 1;
                       break;
                     }
@@ -9514,6 +10158,7 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
             av1_is_interp_needed(xd) && av1_is_interp_search_needed(xd);
         int match_found = -1;
         const InterpFilter assign_filter = cm->interp_filter;
+        int is_luma_interp_done = 0;
         if (cpi->sf.skip_repeat_interpolation_filter_search && need_search) {
           match_found = find_interp_filter_in_stats(x, mbmi);
         }
@@ -9525,25 +10170,28 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
         compmode_interinter_cost = compound_type_rd(
             cpi, x, bsize, mi_col, mi_row, cur_mv, masked_compound_used,
             &orig_dst, &tmp_dst, rd_buffers, &rate_mv, &best_rd_compound,
-            rd_stats, ref_best_rd);
-        if (ref_best_rd < INT64_MAX && best_rd_compound / 3 > ref_best_rd) {
+            rd_stats, ref_best_rd, &is_luma_interp_done);
+        if (ref_best_rd < INT64_MAX &&
+            (best_rd_compound >> 4) * 13 > ref_best_rd) {
           restore_dst_buf(xd, orig_dst, num_planes);
           continue;
         }
-        // No need to call av1_build_inter_predictors_sby if
+        // No need to call av1_enc_build_inter_predictor for luma if
         // COMPOUND_AVERAGE is selected because it is the first
         // candidate in compound_type_rd, and the following
         // compound types searching uses tmp_dst buffer
-        if (mbmi->interinter_comp.type == COMPOUND_AVERAGE) {
-          if (num_planes > 1)
-            av1_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, &orig_dst,
-                                            bsize);
+        if (mbmi->interinter_comp.type == COMPOUND_AVERAGE &&
+            is_luma_interp_done) {
+          if (num_planes > 1) {
+            av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, &orig_dst,
+                                          bsize, AOM_PLANE_U, num_planes - 1);
+          }
           skip_build_pred = 1;
         }
       }
 
       ret_val = interpolation_filter_search(
-          x, cpi, bsize, mi_row, mi_col, &tmp_dst, &orig_dst,
+          x, cpi, tile_data, bsize, mi_row, mi_col, &tmp_dst, &orig_dst,
           args->single_filter, &rd, &rs, &skip_txfm_sb, &skip_sse_sb,
           skip_build_pred, args, ref_best_rd);
       if (args->modelled_rd != NULL && !is_comp_pred) {
@@ -9555,17 +10203,7 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
       } else if (cpi->sf.model_based_post_interp_filter_breakout &&
                  ref_best_rd != INT64_MAX && (rd >> 3) * 3 > ref_best_rd) {
         restore_dst_buf(xd, orig_dst, num_planes);
-        if ((rd >> 3) * 2 > ref_best_rd) break;
-        continue;
-      }
-
-      if (search_jnt_comp) {
-        // if 1/2 model rd is larger than best_rd in jnt_comp mode,
-        // use jnt_comp mode, save additional search
-        if ((rd >> 3) * 4 > best_rd) {
-          restore_dst_buf(xd, orig_dst, num_planes);
-          continue;
-        }
+        break;
       }
 
       if (!is_comp_pred)
@@ -9587,7 +10225,8 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
       }
       rd_stats->rate += compmode_interinter_cost;
 
-      if (search_jnt_comp && cpi->sf.jnt_comp_fast_tx_search && comp_idx == 0) {
+      if (search_dist_wtd_comp && cpi->sf.dist_wtd_comp_fast_tx_search &&
+          comp_idx == 0) {
         // TODO(chengchen): this speed feature introduces big loss.
         // Need better estimation of rate distortion.
         int dummy_rate;
@@ -9596,7 +10235,7 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
         int64_t plane_sse[MAX_MB_PLANE] = { 0 };
         int64_t plane_dist[MAX_MB_PLANE] = { 0 };
 
-        model_rd_sb_fn[MODELRD_TYPE_JNT_COMPOUND](
+        model_rd_sb_fn[MODELRD_TYPE_DIST_WTD_COMPOUND](
             cpi, bsize, x, xd, 0, num_planes - 1, mi_row, mi_col, &dummy_rate,
             &dummy_dist, &skip_txfm_sb, &skip_sse_sb, plane_rate, plane_sse,
             plane_dist);
@@ -9612,16 +10251,10 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
         rd_stats_y->dist = plane_dist[0];
         rd_stats_uv->dist = plane_dist[1] + plane_dist[2];
       } else {
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
-        ret_val = motion_mode_rd(
-            cpi, x, bsize, rd_stats, rd_stats_y, rd_stats_uv, disable_skip,
-            mi_row, mi_col, args, ref_best_rd, refs, &rate_mv, &orig_dst,
-            tile_data, best_est_rd, do_tx_search, inter_modes_info);
-#else
-        ret_val = motion_mode_rd(cpi, x, bsize, rd_stats, rd_stats_y,
+        ret_val = motion_mode_rd(cpi, tile_data, x, bsize, rd_stats, rd_stats_y,
                                  rd_stats_uv, disable_skip, mi_row, mi_col,
-                                 args, ref_best_rd, refs, &rate_mv, &orig_dst);
-#endif
+                                 args, ref_best_rd, refs, &rate_mv, &orig_dst,
+                                 best_est_rd, do_tx_search, inter_modes_info);
       }
       mode_info[ref_mv_idx].mv.as_int = mbmi->mv[0].as_int;
       mode_info[ref_mv_idx].rate_mv = rate_mv;
@@ -9640,10 +10273,6 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
                  sizeof(best_blk_skip[0]) * xd->n4_h * xd->n4_w);
         }
 
-        if (tmp_rd < best_rd2) {
-          best_rd2 = tmp_rd;
-        }
-
         if (tmp_rd < ref_best_rd) {
           ref_best_rd = tmp_rd;
         }
@@ -9670,10 +10299,10 @@ static int64_t handle_inter_mode(const AV1_COMP *const cpi, MACROBLOCK *x,
 }
 
 static int64_t rd_pick_intrabc_mode_sb(const AV1_COMP *cpi, MACROBLOCK *x,
-                                       RD_STATS *rd_cost, BLOCK_SIZE bsize,
+                                       RD_STATS *rd_stats, BLOCK_SIZE bsize,
                                        int64_t best_rd) {
   const AV1_COMMON *const cm = &cpi->common;
-  if (!av1_allow_intrabc(cm)) return INT64_MAX;
+  if (!av1_allow_intrabc(cm) || !cpi->oxcf.enable_intrabc) return INT64_MAX;
   const int num_planes = av1_num_planes(cm);
 
   MACROBLOCKD *const xd = &x->e_mbd;
@@ -9725,7 +10354,7 @@ static int64_t rd_pick_intrabc_mode_sb(const AV1_COMP *cpi, MACROBLOCK *x,
   };
 
   MB_MODE_INFO best_mbmi = *mbmi;
-  RD_STATS best_rdcost = *rd_cost;
+  RD_STATS best_rdstats = *rd_stats;
   int best_skip = x->skip;
 
   uint8_t best_blk_skip[MAX_MIB_SIZE * MAX_MIB_SIZE] = { 0 };
@@ -9769,9 +10398,9 @@ static int64_t rd_pick_intrabc_mode_sb(const AV1_COMP *cpi, MACROBLOCK *x,
     MV mvp_full = dv_ref.as_mv;
     mvp_full.col >>= 3;
     mvp_full.row >>= 3;
-    int sadpb = x->sadperbit16;
+    const int sadpb = x->sadperbit16;
     int cost_list[5];
-    int bestsme = av1_full_pixel_search(
+    const int bestsme = av1_full_pixel_search(
         cpi, x, bsize, &mvp_full, step_param, cpi->sf.mv.search_method, 0,
         sadpb, cond_cost_list(cpi, cost_list), &dv_ref.as_mv, INT_MAX, 1,
         (MI_SIZE * mi_col), (MI_SIZE * mi_row), 1);
@@ -9779,7 +10408,7 @@ static int64_t rd_pick_intrabc_mode_sb(const AV1_COMP *cpi, MACROBLOCK *x,
     x->mv_limits = tmp_mv_limits;
     if (bestsme == INT_MAX) continue;
     mvp_full = x->best_mv.as_mv;
-    MV dv = { .row = mvp_full.row * 8, .col = mvp_full.col * 8 };
+    const MV dv = { .row = mvp_full.row * 8, .col = mvp_full.col * 8 };
     if (mv_check_bounds(&x->mv_limits, &dv)) continue;
     if (!av1_is_dv_valid(dv, cm, xd, mi_row, mi_col, bsize,
                          cm->seq_params.mib_size_log2))
@@ -9798,74 +10427,39 @@ static int64_t rd_pick_intrabc_mode_sb(const AV1_COMP *cpi, MACROBLOCK *x,
     mbmi->interp_filters = av1_broadcast_interp_filter(BILINEAR);
     mbmi->skip = 0;
     x->skip = 0;
-    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
+    av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize, 0,
+                                  av1_num_planes(cm) - 1);
 
     int *dvcost[2] = { (int *)&cpi->dv_cost[0][MV_MAX],
                        (int *)&cpi->dv_cost[1][MV_MAX] };
     // TODO(aconverse@google.com): The full motion field defining discount
     // in MV_COST_WEIGHT is too large. Explore other values.
-    int rate_mv = av1_mv_bit_cost(&dv, &dv_ref.as_mv, cpi->dv_joint_cost,
-                                  dvcost, MV_COST_WEIGHT_SUB);
+    const int rate_mv = av1_mv_bit_cost(&dv, &dv_ref.as_mv, cpi->dv_joint_cost,
+                                        dvcost, MV_COST_WEIGHT_SUB);
     const int rate_mode = x->intrabc_cost[1];
-    RD_STATS rd_stats, rd_stats_uv;
-    av1_subtract_plane(x, bsize, 0);
-    if (cm->tx_mode == TX_MODE_SELECT && !xd->lossless[mbmi->segment_id]) {
-      // Intrabc
-      select_tx_type_yrd(cpi, x, &rd_stats, bsize, mi_row, mi_col, INT64_MAX);
-    } else {
-      super_block_yrd(cpi, x, &rd_stats, bsize, INT64_MAX);
-      memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size));
-      for (int i = 0; i < xd->n4_h * xd->n4_w; ++i)
-        set_blk_skip(x, 0, i, rd_stats.skip);
-    }
-    if (num_planes > 1) {
-      super_block_uvrd(cpi, x, &rd_stats_uv, bsize, INT64_MAX);
-      av1_merge_rd_stats(&rd_stats, &rd_stats_uv);
-    }
-#if CONFIG_RD_DEBUG
-    mbmi->rd_stats = rd_stats;
-#endif
-
-    const int skip_ctx = av1_get_skip_context(xd);
-
-    RD_STATS rdc_noskip;
-    av1_init_rd_stats(&rdc_noskip);
-    rdc_noskip.rate =
-        rate_mode + rate_mv + rd_stats.rate + x->skip_cost[skip_ctx][0];
-    rdc_noskip.dist = rd_stats.dist;
-    rdc_noskip.rdcost = RDCOST(x->rdmult, rdc_noskip.rate, rdc_noskip.dist);
-    if (rdc_noskip.rdcost < best_rd) {
-      best_rd = rdc_noskip.rdcost;
+    RD_STATS rd_stats_yuv, rd_stats_y, rd_stats_uv;
+    if (!txfm_search(cpi, NULL, x, bsize, mi_row, mi_col, &rd_stats_yuv,
+                     &rd_stats_y, &rd_stats_uv, rate_mode + rate_mv, INT64_MAX))
+      continue;
+    rd_stats_yuv.rdcost =
+        RDCOST(x->rdmult, rd_stats_yuv.rate, rd_stats_yuv.dist);
+    if (rd_stats_yuv.rdcost < best_rd) {
+      best_rd = rd_stats_yuv.rdcost;
       best_mbmi = *mbmi;
-      best_skip = x->skip;
-      best_rdcost = rdc_noskip;
+      best_skip = mbmi->skip;
+      best_rdstats = rd_stats_yuv;
       memcpy(best_blk_skip, x->blk_skip,
              sizeof(x->blk_skip[0]) * xd->n4_h * xd->n4_w);
     }
-
-    if (!xd->lossless[mbmi->segment_id]) {
-      x->skip = 1;
-      mbmi->skip = 1;
-      RD_STATS rdc_skip;
-      av1_init_rd_stats(&rdc_skip);
-      rdc_skip.rate = rate_mode + rate_mv + x->skip_cost[skip_ctx][1];
-      rdc_skip.dist = rd_stats.sse;
-      rdc_skip.rdcost = RDCOST(x->rdmult, rdc_skip.rate, rdc_skip.dist);
-      if (rdc_skip.rdcost < best_rd) {
-        best_rd = rdc_skip.rdcost;
-        best_mbmi = *mbmi;
-        best_skip = x->skip;
-        best_rdcost = rdc_skip;
-        memcpy(best_blk_skip, x->blk_skip,
-               sizeof(x->blk_skip[0]) * xd->n4_h * xd->n4_w);
-      }
-    }
   }
   *mbmi = best_mbmi;
-  *rd_cost = best_rdcost;
+  *rd_stats = best_rdstats;
   x->skip = best_skip;
   memcpy(x->blk_skip, best_blk_skip,
          sizeof(x->blk_skip[0]) * xd->n4_h * xd->n4_w);
+#if CONFIG_RD_DEBUG
+  mbmi->rd_stats = *rd_stats;
+#endif
   return best_rd;
 }
 
@@ -10006,6 +10600,7 @@ static void rd_pick_skip_mode(RD_STATS *rd_cost,
                               BLOCK_SIZE bsize, int mi_row, int mi_col,
                               struct buf_2d yv12_mb[REF_FRAMES][MAX_MB_PLANE]) {
   const AV1_COMMON *const cm = &cpi->common;
+  const SkipModeInfo *const skip_mode_info = &cm->current_frame.skip_mode_info;
   const int num_planes = av1_num_planes(cm);
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
@@ -10014,13 +10609,15 @@ static void rd_pick_skip_mode(RD_STATS *rd_cost,
   RD_STATS skip_mode_rd_stats;
   av1_invalid_rd_stats(&skip_mode_rd_stats);
 
-  if (cm->ref_frame_idx_0 == INVALID_IDX ||
-      cm->ref_frame_idx_1 == INVALID_IDX) {
+  if (skip_mode_info->ref_frame_idx_0 == INVALID_IDX ||
+      skip_mode_info->ref_frame_idx_1 == INVALID_IDX) {
     return;
   }
 
-  const MV_REFERENCE_FRAME ref_frame = LAST_FRAME + cm->ref_frame_idx_0;
-  const MV_REFERENCE_FRAME second_ref_frame = LAST_FRAME + cm->ref_frame_idx_1;
+  const MV_REFERENCE_FRAME ref_frame =
+      LAST_FRAME + skip_mode_info->ref_frame_idx_0;
+  const MV_REFERENCE_FRAME second_ref_frame =
+      LAST_FRAME + skip_mode_info->ref_frame_idx_1;
   const PREDICTION_MODE this_mode = NEAREST_NEARESTMV;
   const int mode_index =
       get_prediction_mode_idx(this_mode, ref_frame, second_ref_frame);
@@ -10085,7 +10682,8 @@ static void rd_pick_skip_mode(RD_STATS *rd_cost,
                    rd_cost->dist)
           : INT64_MAX;
 
-  if (skip_mode_rd_stats.rdcost <= best_intra_inter_mode_cost) {
+  if (skip_mode_rd_stats.rdcost <= best_intra_inter_mode_cost &&
+      (!xd->lossless[mbmi->segment_id] || skip_mode_rd_stats.dist == 0)) {
     assert(mode_index != -1);
     search_state->best_mbmode.skip_mode = 1;
     search_state->best_mbmode = *mbmi;
@@ -10131,9 +10729,16 @@ static void rd_pick_skip_mode(RD_STATS *rd_cost,
     rd_cost->dist = rd_cost->sse = skip_mode_rd_stats.dist;
     rd_cost->rdcost = skip_mode_rd_stats.rdcost;
 
+#if CONFIG_ONE_PASS_SVM
+    if (bsize >= BLOCK_8X8 &&
+        block_size_high[bsize] == block_size_wide[bsize]) {
+      av1_copy_reg_stat(rd_cost, &skip_mode_rd_stats);
+    }
+#endif
+
     search_state->best_rd = rd_cost->rdcost;
     search_state->best_skip2 = 1;
-    search_state->best_mode_skippable = (skip_mode_rd_stats.sse == 0);
+    search_state->best_mode_skippable = 1;
 
     x->skip = 1;
   }
@@ -10163,6 +10768,7 @@ static void sf_refine_fast_tx_type_search(
         !is_inter_mode(best_mbmode->mode)))) {
     int skip_blk = 0;
     RD_STATS rd_stats_y, rd_stats_uv;
+    const int skip_ctx = av1_get_skip_context(xd);
 
     x->use_default_inter_tx_type = 0;
     x->use_default_intra_tx_type = 0;
@@ -10179,15 +10785,15 @@ static void sf_refine_fast_tx_type_search(
     }
 
     if (is_inter_mode(mbmi->mode)) {
-      av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
+      av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize, 0,
+                                    av1_num_planes(cm) - 1);
       if (mbmi->motion_mode == OBMC_CAUSAL)
         av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
 
       av1_subtract_plane(x, bsize, 0);
       if (cm->tx_mode == TX_MODE_SELECT && !xd->lossless[mbmi->segment_id]) {
-        // av1_rd_pick_inter_mode_sb
-        select_tx_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col,
-                           INT64_MAX);
+        pick_tx_size_type_yrd(cpi, x, &rd_stats_y, bsize, mi_row, mi_col,
+                              INT64_MAX);
         assert(rd_stats_y.rate != INT_MAX);
       } else {
         super_block_yrd(cpi, x, &rd_stats_y, bsize, INT64_MAX);
@@ -10195,32 +10801,29 @@ static void sf_refine_fast_tx_type_search(
         for (int i = 0; i < xd->n4_h * xd->n4_w; ++i)
           set_blk_skip(x, 0, i, rd_stats_y.skip);
       }
-      if (num_planes > 1) {
-        inter_block_uvrd(cpi, x, &rd_stats_uv, bsize, INT64_MAX, INT64_MAX,
-                         FTXS_NONE);
-      } else {
-        av1_init_rd_stats(&rd_stats_uv);
-      }
     } else {
       super_block_yrd(cpi, x, &rd_stats_y, bsize, INT64_MAX);
-      if (num_planes > 1) {
-        super_block_uvrd(cpi, x, &rd_stats_uv, bsize, INT64_MAX);
-      } else {
-        av1_init_rd_stats(&rd_stats_uv);
-      }
     }
 
-    if (RDCOST(x->rdmult, rd_stats_y.rate + rd_stats_uv.rate,
+    if (num_planes > 1) {
+      super_block_uvrd(cpi, x, &rd_stats_uv, bsize, INT64_MAX);
+    } else {
+      av1_init_rd_stats(&rd_stats_uv);
+    }
+
+    if (RDCOST(x->rdmult,
+               x->skip_cost[skip_ctx][0] + rd_stats_y.rate + rd_stats_uv.rate,
                (rd_stats_y.dist + rd_stats_uv.dist)) >
-        RDCOST(x->rdmult, 0, (rd_stats_y.sse + rd_stats_uv.sse))) {
+        RDCOST(x->rdmult, x->skip_cost[skip_ctx][1],
+               (rd_stats_y.sse + rd_stats_uv.sse))) {
       skip_blk = 1;
-      rd_stats_y.rate = x->skip_cost[av1_get_skip_context(xd)][1];
+      rd_stats_y.rate = x->skip_cost[skip_ctx][1];
       rd_stats_uv.rate = 0;
       rd_stats_y.dist = rd_stats_y.sse;
       rd_stats_uv.dist = rd_stats_uv.sse;
     } else {
       skip_blk = 0;
-      rd_stats_y.rate += x->skip_cost[av1_get_skip_context(xd)][0];
+      rd_stats_y.rate += x->skip_cost[skip_ctx][0];
     }
 
     if (RDCOST(x->rdmult, best_rate_y + best_rate_uv, rd_cost->dist) >
@@ -10240,13 +10843,137 @@ static void sf_refine_fast_tx_type_search(
   }
 }
 
+// Contains information on which modes to skip. The last entry contains
+// information on whether the reference frames should should be skipped.
+typedef struct mode_skip_mask_struct {
+  uint32_t mode[REF_FRAMES];
+  uint16_t ref_frame1;
+  uint16_t ref_frame2;
+} mode_skip_mask_t;
+
+static void init_mode_skip_mask(mode_skip_mask_t *mask, const AV1_COMP *cpi,
+                                MACROBLOCK *x, BLOCK_SIZE bsize) {
+  const AV1_COMMON *const cm = &cpi->common;
+  const struct segmentation *const seg = &cm->seg;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MB_MODE_INFO *const mbmi = xd->mi[0];
+  unsigned char segment_id = mbmi->segment_id;
+  const SPEED_FEATURES *const sf = &cpi->sf;
+
+  memset(mask, 0, sizeof(*mask));
+
+  int min_pred_mv_sad = INT_MAX;
+  MV_REFERENCE_FRAME ref_frame;
+  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame)
+    min_pred_mv_sad = AOMMIN(min_pred_mv_sad, x->pred_mv_sad[ref_frame]);
+
+  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
+    if (!(cpi->ref_frame_flags & ref_frame_flag_list[ref_frame])) {
+      // Skip checking missing references in both single and compound reference
+      // modes. Note that a mode will be skipped iff both reference frames
+      // are masked out.
+      mask->ref_frame1 |= (1 << ref_frame);
+      mask->ref_frame2 |= SECOND_REF_FRAME_MASK;
+    } else {
+      // Skip fixed mv modes for poor references
+      if ((x->pred_mv_sad[ref_frame] >> 2) > min_pred_mv_sad) {
+        mask->mode[ref_frame] |= INTER_NEAREST_NEAR_ZERO;
+      }
+    }
+    // If the segment reference frame feature is enabled....
+    // then do nothing if the current ref frame is not allowed..
+    if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
+        get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame) {
+      mask->ref_frame1 |= (1 << ref_frame);
+      mask->ref_frame2 |= SECOND_REF_FRAME_MASK;
+    }
+  }
+  // Disable this drop out case if the ref frame
+  // segment level feature is enabled for this segment. This is to
+  // prevent the possibility that we end up unable to pick any mode.
+  if (!segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) {
+    // Only consider GLOBALMV/ALTREF_FRAME for alt ref frame,
+    // unless ARNR filtering is enabled in which case we want
+    // an unfiltered alternative. We allow near/nearest as well
+    // because they may result in zero-zero MVs but be cheaper.
+    if (cpi->rc.is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
+      mask->ref_frame1 = (1 << LAST_FRAME) | (1 << LAST2_FRAME) |
+                         (1 << LAST3_FRAME) | (1 << BWDREF_FRAME) |
+                         (1 << ALTREF2_FRAME) | (1 << GOLDEN_FRAME);
+      mask->ref_frame2 = SECOND_REF_FRAME_MASK;
+      // TODO(zoeliu): To further explore whether following needs to be done for
+      //               BWDREF_FRAME as well.
+      mask->mode[ALTREF_FRAME] = ~INTER_NEAREST_NEAR_ZERO;
+      const MV_REFERENCE_FRAME tmp_ref_frames[2] = { ALTREF_FRAME, NONE_FRAME };
+      int_mv near_mv, nearest_mv, global_mv;
+      get_this_mv(&nearest_mv, NEARESTMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
+      get_this_mv(&near_mv, NEARMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
+      get_this_mv(&global_mv, GLOBALMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
+
+      if (near_mv.as_int != global_mv.as_int)
+        mask->mode[ALTREF_FRAME] |= (1 << NEARMV);
+      if (nearest_mv.as_int != global_mv.as_int)
+        mask->mode[ALTREF_FRAME] |= (1 << NEARESTMV);
+    }
+  }
+
+  if (cpi->rc.is_src_frame_alt_ref) {
+    if (sf->alt_ref_search_fp) {
+      assert(cpi->ref_frame_flags & ref_frame_flag_list[ALTREF_FRAME]);
+      mask->mode[ALTREF_FRAME] = 0;
+      mask->ref_frame1 = ~(1 << ALTREF_FRAME);
+      mask->ref_frame2 = SECOND_REF_FRAME_MASK;
+    }
+  }
+
+  if (sf->alt_ref_search_fp)
+    if (!cm->show_frame && x->pred_mv_sad[GOLDEN_FRAME] < INT_MAX)
+      if (x->pred_mv_sad[ALTREF_FRAME] > (x->pred_mv_sad[GOLDEN_FRAME] << 1))
+        mask->mode[ALTREF_FRAME] |= INTER_ALL;
+
+  if (sf->adaptive_mode_search) {
+    if (cm->show_frame && !cpi->rc.is_src_frame_alt_ref &&
+        cpi->rc.frames_since_golden >= 3)
+      if ((x->pred_mv_sad[GOLDEN_FRAME] >> 1) > x->pred_mv_sad[LAST_FRAME])
+        mask->mode[GOLDEN_FRAME] |= INTER_ALL;
+  }
+
+  if (bsize > sf->max_intra_bsize) {
+    mask->ref_frame1 |= (1 << INTRA_FRAME);
+    mask->ref_frame2 |= (1 << INTRA_FRAME);
+  }
+
+  mask->mode[INTRA_FRAME] |= ~(sf->intra_y_mode_mask[max_txsize_lookup[bsize]]);
+
+  // Disable some frames to enforce max number of reference frames requested.
+  // TODO(urvang): Should this logic be moved to enforce_max_ref_frames()?
+  const MV_REFERENCE_FRAME disable_order[] = {
+    LAST3_FRAME, LAST2_FRAME, ALTREF2_FRAME, BWDREF_FRAME,
+    // GOLDEN_FRAME,
+  };
+  assert(cpi->oxcf.max_reference_frames <= INTER_REFS_PER_FRAME);
+  const int num_frames_to_disable =
+      INTER_REFS_PER_FRAME - cpi->oxcf.max_reference_frames;
+  for (int i = 0; i < num_frames_to_disable; ++i) {
+    const MV_REFERENCE_FRAME ref_frame_to_disable = disable_order[i];
+
+    mask->ref_frame1 |= (1 << ref_frame_to_disable);
+    mask->ref_frame2 |= (1 << ref_frame_to_disable);
+
+    if (ref_frame_to_disable == BWDREF_FRAME &&
+        (cpi->ref_frame_flags & AOM_BWD_FLAG)) {
+      mask->ref_frame1 |= (1 << ALTREF_FRAME);
+      mask->ref_frame2 |= (1 << ALTREF_FRAME);
+    }
+  }
+}
+
 // Please add/modify parameter setting in this function, making it consistent
 // and easy to read and maintain.
 static void set_params_rd_pick_inter_mode(
     const AV1_COMP *cpi, MACROBLOCK *x, HandleInterModeArgs *args,
-    BLOCK_SIZE bsize, int mi_row, int mi_col, uint16_t ref_frame_skip_mask[2],
-    uint32_t mode_skip_mask[REF_FRAMES], int skip_ref_frame_mask,
-    unsigned int ref_costs_single[REF_FRAMES],
+    BLOCK_SIZE bsize, int mi_row, int mi_col, mode_skip_mask_t *mode_skip_mask,
+    int skip_ref_frame_mask, unsigned int ref_costs_single[REF_FRAMES],
     unsigned int ref_costs_comp[REF_FRAMES][REF_FRAMES],
     struct buf_2d yv12_mb[REF_FRAMES][MAX_MB_PLANE]) {
   const AV1_COMMON *const cm = &cpi->common;
@@ -10254,8 +10981,6 @@ static void set_params_rd_pick_inter_mode(
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
   MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
-  const struct segmentation *const seg = &cm->seg;
-  const SPEED_FEATURES *const sf = &cpi->sf;
   unsigned char segment_id = mbmi->segment_id;
   int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
   int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE >> 1, MAX_SB_SIZE >> 1,
@@ -10267,7 +10992,7 @@ static void set_params_rd_pick_inter_mode(
   for (int i = 0; i < MB_MODE_COUNT; ++i)
     for (int k = 0; k < REF_FRAMES; ++k) args->single_filter[i][k] = SWITCHABLE;
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     int len = sizeof(uint16_t);
     args->above_pred_buf[0] = CONVERT_TO_BYTEPTR(x->above_pred_buf);
     args->above_pred_buf[1] =
@@ -10316,7 +11041,7 @@ static void set_params_rd_pick_inter_mode(
           if (skip) continue;
         }
       }
-      assert(get_ref_frame_buffer(cpi, ref_frame) != NULL);
+      assert(get_ref_frame_yv12_buf(cm, ref_frame) != NULL);
       setup_buffer_ref_mvs_inter(cpi, x, ref_frame, bsize, mi_row, mi_col,
                                  yv12_mb);
     }
@@ -10352,101 +11077,15 @@ static void set_params_rd_pick_inter_mode(
     av1_build_prediction_by_left_preds(cm, xd, mi_row, mi_col,
                                        args->left_pred_buf, dst_width2,
                                        dst_height2, args->left_pred_stride);
-    av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
-                         mi_col, 0, num_planes);
+    av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col,
+                         0, num_planes);
     calc_target_weighted_pred(
         cm, x, xd, mi_row, mi_col, args->above_pred_buf[0],
         args->above_pred_stride[0], args->left_pred_buf[0],
         args->left_pred_stride[0]);
   }
 
-  int min_pred_mv_sad = INT_MAX;
-  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame)
-    min_pred_mv_sad = AOMMIN(min_pred_mv_sad, x->pred_mv_sad[ref_frame]);
-
-  for (int i = 0; i < 2; ++i) {
-    ref_frame_skip_mask[i] = 0;
-  }
-  memset(mode_skip_mask, 0, REF_FRAMES * sizeof(*mode_skip_mask));
-  for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
-    if (!(cpi->ref_frame_flags & ref_frame_flag_list[ref_frame])) {
-      // Skip checking missing references in both single and compound reference
-      // modes. Note that a mode will be skipped iff both reference frames
-      // are masked out.
-      ref_frame_skip_mask[0] |= (1 << ref_frame);
-      ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
-    } else {
-      // Skip fixed mv modes for poor references
-      if ((x->pred_mv_sad[ref_frame] >> 2) > min_pred_mv_sad) {
-        mode_skip_mask[ref_frame] |= INTER_NEAREST_NEAR_ZERO;
-      }
-    }
-    // If the segment reference frame feature is enabled....
-    // then do nothing if the current ref frame is not allowed..
-    if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
-        get_segdata(seg, segment_id, SEG_LVL_REF_FRAME) != (int)ref_frame) {
-      ref_frame_skip_mask[0] |= (1 << ref_frame);
-      ref_frame_skip_mask[1] |= SECOND_REF_FRAME_MASK;
-    }
-  }
-
-  // Disable this drop out case if the ref frame
-  // segment level feature is enabled for this segment. This is to
-  // prevent the possibility that we end up unable to pick any mode.
-  if (!segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) {
-    // Only consider GLOBALMV/ALTREF_FRAME for alt ref frame,
-    // unless ARNR filtering is enabled in which case we want
-    // an unfiltered alternative. We allow near/nearest as well
-    // because they may result in zero-zero MVs but be cheaper.
-    if (cpi->rc.is_src_frame_alt_ref && (cpi->oxcf.arnr_max_frames == 0)) {
-      ref_frame_skip_mask[0] = (1 << LAST_FRAME) | (1 << LAST2_FRAME) |
-                               (1 << LAST3_FRAME) | (1 << BWDREF_FRAME) |
-                               (1 << ALTREF2_FRAME) | (1 << GOLDEN_FRAME);
-      ref_frame_skip_mask[1] = SECOND_REF_FRAME_MASK;
-      // TODO(zoeliu): To further explore whether following needs to be done for
-      //               BWDREF_FRAME as well.
-      mode_skip_mask[ALTREF_FRAME] = ~INTER_NEAREST_NEAR_ZERO;
-      const MV_REFERENCE_FRAME tmp_ref_frames[2] = { ALTREF_FRAME, NONE_FRAME };
-      int_mv near_mv, nearest_mv, global_mv;
-      get_this_mv(&nearest_mv, NEARESTMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
-      get_this_mv(&near_mv, NEARMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
-      get_this_mv(&global_mv, GLOBALMV, 0, 0, tmp_ref_frames, x->mbmi_ext);
-
-      if (near_mv.as_int != global_mv.as_int)
-        mode_skip_mask[ALTREF_FRAME] |= (1 << NEARMV);
-      if (nearest_mv.as_int != global_mv.as_int)
-        mode_skip_mask[ALTREF_FRAME] |= (1 << NEARESTMV);
-    }
-  }
-
-  if (cpi->rc.is_src_frame_alt_ref) {
-    if (sf->alt_ref_search_fp) {
-      assert(cpi->ref_frame_flags & ref_frame_flag_list[ALTREF_FRAME]);
-      mode_skip_mask[ALTREF_FRAME] = 0;
-      ref_frame_skip_mask[0] = ~(1 << ALTREF_FRAME);
-      ref_frame_skip_mask[1] = SECOND_REF_FRAME_MASK;
-    }
-  }
-
-  if (sf->alt_ref_search_fp)
-    if (!cm->show_frame && x->pred_mv_sad[GOLDEN_FRAME] < INT_MAX)
-      if (x->pred_mv_sad[ALTREF_FRAME] > (x->pred_mv_sad[GOLDEN_FRAME] << 1))
-        mode_skip_mask[ALTREF_FRAME] |= INTER_ALL;
-
-  if (sf->adaptive_mode_search) {
-    if (cm->show_frame && !cpi->rc.is_src_frame_alt_ref &&
-        cpi->rc.frames_since_golden >= 3)
-      if ((x->pred_mv_sad[GOLDEN_FRAME] >> 1) > x->pred_mv_sad[LAST_FRAME])
-        mode_skip_mask[GOLDEN_FRAME] |= INTER_ALL;
-  }
-
-  if (bsize > sf->max_intra_bsize) {
-    ref_frame_skip_mask[0] |= (1 << INTRA_FRAME);
-    ref_frame_skip_mask[1] |= (1 << INTRA_FRAME);
-  }
-
-  mode_skip_mask[INTRA_FRAME] |=
-      ~(sf->intra_y_mode_mask[max_txsize_lookup[bsize]]);
+  init_mode_skip_mask(mode_skip_mask, cpi, x, bsize);
 
   if (cpi->sf.tx_type_search.fast_intra_tx_type_search)
     x->use_default_intra_tx_type = 1;
@@ -10461,6 +11100,7 @@ static void set_params_rd_pick_inter_mode(
     x->interp_filter_stats_idx[0] = 0;
     x->interp_filter_stats_idx[1] = 0;
   }
+  x->comp_rd_stats_idx = 0;
 }
 
 static void search_palette_mode(const AV1_COMP *cpi, MACROBLOCK *x, int mi_row,
@@ -10538,6 +11178,9 @@ static void search_palette_mode(const AV1_COMP *cpi, MACROBLOCK *x, int mi_row,
     rate2 -= rd_stats_y.rate;
     if (num_planes > 1) rate2 -= search_state->rate_uv_tokenonly[uv_tx];
     rate2 += x->skip_cost[av1_get_skip_context(xd)][1];
+#if CONFIG_ONE_PASS_SVM
+    av1_reg_stat_skipmode_update(&rd_stats_y, x->rdmult);
+#endif
   } else {
     rate2 += x->skip_cost[av1_get_skip_context(xd)][0];
   }
@@ -10554,6 +11197,9 @@ static void search_palette_mode(const AV1_COMP *cpi, MACROBLOCK *x, int mi_row,
     search_state->best_mode_skippable = skippable;
     memcpy(ctx->blk_skip, x->blk_skip,
            sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
+#if CONFIG_ONE_PASS_SVM
+    av1_copy_reg_stat(rd_cost, &rd_stats_y);
+#endif
   }
 }
 
@@ -10598,6 +11244,7 @@ static void init_inter_mode_search_state(InterModeSearchState *search_state,
   search_state->best_intra_rd = INT64_MAX;
 
   search_state->angle_stats_ready = 0;
+  av1_zero(search_state->directional_mode_skip_mask);
 
   search_state->best_pred_sse = UINT_MAX;
 
@@ -10647,23 +11294,81 @@ static void init_inter_mode_search_state(InterModeSearchState *search_state,
   av1_zero(search_state->single_state_modelled_cnt);
 }
 
+bool mask_says_skip(const mode_skip_mask_t *mode_skip_mask,
+                    const MV_REFERENCE_FRAME *ref_frame,
+                    const PREDICTION_MODE this_mode) {
+  if (mode_skip_mask->mode[ref_frame[0]] & (1 << this_mode)) {
+    return true;
+  }
+
+  return (mode_skip_mask->ref_frame1 & (1 << ref_frame[0])) &&
+         (mode_skip_mask->ref_frame2 & (1 << AOMMAX(0, ref_frame[1])));
+}
+
+static int inter_mode_compatible_skip(const AV1_COMP *cpi, const MACROBLOCK *x,
+                                      BLOCK_SIZE bsize, int mode_index) {
+  const AV1_COMMON *const cm = &cpi->common;
+  const struct segmentation *const seg = &cm->seg;
+  const MV_REFERENCE_FRAME *ref_frame = av1_mode_order[mode_index].ref_frame;
+  const PREDICTION_MODE this_mode = av1_mode_order[mode_index].mode;
+  const CurrentFrame *const current_frame = &cm->current_frame;
+  const MACROBLOCKD *const xd = &x->e_mbd;
+  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  const unsigned char segment_id = mbmi->segment_id;
+  const int comp_pred = ref_frame[1] > INTRA_FRAME;
+
+  if (comp_pred) {
+    if (frame_is_intra_only(cm)) return 1;
+
+    if (current_frame->reference_mode == SINGLE_REFERENCE) return 1;
+
+    // Skip compound inter modes if ARF is not available.
+    if (!(cpi->ref_frame_flags & ref_frame_flag_list[ref_frame[1]])) return 1;
+
+    // Do not allow compound prediction if the segment level reference frame
+    // feature is in use as in this case there can only be one reference.
+    if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) return 1;
+
+    if (!is_comp_ref_allowed(bsize)) return 1;
+  }
+
+  if (ref_frame[0] > INTRA_FRAME && ref_frame[1] == INTRA_FRAME) {
+    // Mode must by compatible
+    if (!is_interintra_allowed_mode(this_mode)) return 1;
+    if (!is_interintra_allowed_bsize(bsize)) return 1;
+  }
+
+  return 0;
+}
+
 // Case 1: return 0, means don't skip this mode
 // Case 2: return 1, means skip this mode completely
 // Case 3: return 2, means skip compound only, but still try single motion modes
 static int inter_mode_search_order_independent_skip(
     const AV1_COMP *cpi, const PICK_MODE_CONTEXT *ctx, const MACROBLOCK *x,
     BLOCK_SIZE bsize, int mode_index, int mi_row, int mi_col,
-    uint32_t *mode_skip_mask, uint16_t *ref_frame_skip_mask,
-    InterModeSearchState *search_state) {
+    mode_skip_mask_t *mode_skip_mask, InterModeSearchState *search_state) {
   const SPEED_FEATURES *const sf = &cpi->sf;
   const AV1_COMMON *const cm = &cpi->common;
-  const struct segmentation *const seg = &cm->seg;
+  const OrderHintInfo *const order_hint_info = &cm->seq_params.order_hint_info;
+  const CurrentFrame *const current_frame = &cm->current_frame;
   const MACROBLOCKD *const xd = &x->e_mbd;
   const MB_MODE_INFO *const mbmi = xd->mi[0];
-  const unsigned char segment_id = mbmi->segment_id;
   const MV_REFERENCE_FRAME *ref_frame = av1_mode_order[mode_index].ref_frame;
   const PREDICTION_MODE this_mode = av1_mode_order[mode_index].mode;
+  const int comp_pred = ref_frame[1] > INTRA_FRAME;
   int skip_motion_mode = 0;
+
+  if (mask_says_skip(mode_skip_mask, ref_frame, this_mode)) {
+    return 1;
+  }
+
+  // If no valid mode has been found so far in PARTITION_NONE when finding a
+  // valid partition is required, do not skip mode.
+  if (search_state->best_rd == INT64_MAX && mbmi->partition == PARTITION_NONE &&
+      x->must_find_valid_partition)
+    return 0;
+
   if (mbmi->partition != PARTITION_NONE && mbmi->partition != PARTITION_SPLIT) {
     const int ref_type = av1_ref_frame_type(ref_frame);
     int skip_ref = ctx->skip_ref_frame_mask & (1 << ref_type);
@@ -10714,12 +11419,6 @@ static int inter_mode_search_order_independent_skip(
     if (!found) return 1;
   }
 
-  if (ref_frame[0] > INTRA_FRAME && ref_frame[1] == INTRA_FRAME) {
-    // Mode must by compatible
-    if (!is_interintra_allowed_mode(this_mode)) return 1;
-    if (!is_interintra_allowed_bsize(bsize)) return 1;
-  }
-
   // This is only used in motion vector unit test.
   if (cpi->oxcf.motion_vector_unit_test && ref_frame[0] == INTRA_FRAME)
     return 1;
@@ -10734,71 +11433,94 @@ static int inter_mode_search_order_independent_skip(
           x->source_variance < skip_intra_var_thresh)
         return 1;
     }
-  } else {
-    if (!is_comp_ref_allowed(bsize) && ref_frame[1] > INTRA_FRAME) return 1;
-  }
-
-  const int comp_pred = ref_frame[1] > INTRA_FRAME;
-  if (comp_pred) {
-    if (!cpi->allow_comp_inter_inter) return 1;
-
-    if (cm->reference_mode == SINGLE_REFERENCE) return 1;
-
-    // Skip compound inter modes if ARF is not available.
-    if (!(cpi->ref_frame_flags & ref_frame_flag_list[ref_frame[1]])) return 1;
-
-    // Do not allow compound prediction if the segment level reference frame
-    // feature is in use as in this case there can only be one reference.
-    if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) return 1;
   }
 
   if (sf->selective_ref_frame) {
-    if (sf->selective_ref_frame >= 2 || x->cb_partition_scan) {
+    if (sf->selective_ref_frame >= 3 || x->cb_partition_scan) {
       if (ref_frame[0] == ALTREF2_FRAME || ref_frame[1] == ALTREF2_FRAME)
         if (get_relative_dist(
-                cm, cm->cur_frame->ref_frame_offset[ALTREF2_FRAME - LAST_FRAME],
-                cm->frame_offset) < 0)
+                order_hint_info,
+                cm->cur_frame->ref_order_hints[ALTREF2_FRAME - LAST_FRAME],
+                current_frame->order_hint) < 0)
           return 1;
       if (ref_frame[0] == BWDREF_FRAME || ref_frame[1] == BWDREF_FRAME)
         if (get_relative_dist(
-                cm, cm->cur_frame->ref_frame_offset[BWDREF_FRAME - LAST_FRAME],
-                cm->frame_offset) < 0)
+                order_hint_info,
+                cm->cur_frame->ref_order_hints[BWDREF_FRAME - LAST_FRAME],
+                current_frame->order_hint) < 0)
+          return 1;
+    }
+
+    if (sf->selective_ref_frame >= 2 ||
+        (sf->selective_ref_frame == 1 && comp_pred)) {
+      if (ref_frame[0] == LAST3_FRAME || ref_frame[1] == LAST3_FRAME)
+        if (get_relative_dist(
+                order_hint_info,
+                cm->cur_frame->ref_order_hints[LAST3_FRAME - LAST_FRAME],
+                cm->cur_frame->ref_order_hints[GOLDEN_FRAME - LAST_FRAME]) <= 0)
+          return 1;
+      if (ref_frame[0] == LAST2_FRAME || ref_frame[1] == LAST2_FRAME)
+        if (get_relative_dist(
+                order_hint_info,
+                cm->cur_frame->ref_order_hints[LAST2_FRAME - LAST_FRAME],
+                cm->cur_frame->ref_order_hints[GOLDEN_FRAME - LAST_FRAME]) <= 0)
           return 1;
     }
-    if (ref_frame[0] == LAST3_FRAME || ref_frame[1] == LAST3_FRAME)
-      if (get_relative_dist(
-              cm, cm->cur_frame->ref_frame_offset[LAST3_FRAME - LAST_FRAME],
-              cm->cur_frame->ref_frame_offset[GOLDEN_FRAME - LAST_FRAME]) <= 0)
-        return 1;
-    if (ref_frame[0] == LAST2_FRAME || ref_frame[1] == LAST2_FRAME)
-      if (get_relative_dist(
-              cm, cm->cur_frame->ref_frame_offset[LAST2_FRAME - LAST_FRAME],
-              cm->cur_frame->ref_frame_offset[GOLDEN_FRAME - LAST_FRAME]) <= 0)
-        return 1;
   }
 
   // One-sided compound is used only when all reference frames are one-sided.
-  if (sf->selective_ref_frame && comp_pred && !cpi->all_one_sided_refs) {
+  if ((sf->selective_ref_frame >= 2) && comp_pred && !cpi->all_one_sided_refs) {
     unsigned int ref_offsets[2];
     for (int i = 0; i < 2; ++i) {
-      const int buf_idx = cm->frame_refs[ref_frame[i] - LAST_FRAME].idx;
-      assert(buf_idx >= 0);
-      ref_offsets[i] = cm->buffer_pool->frame_bufs[buf_idx].cur_frame_offset;
+      const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame[i]);
+      assert(buf != NULL);
+      ref_offsets[i] = buf->order_hint;
     }
-    if ((get_relative_dist(cm, ref_offsets[0], cm->frame_offset) <= 0 &&
-         get_relative_dist(cm, ref_offsets[1], cm->frame_offset) <= 0) ||
-        (get_relative_dist(cm, ref_offsets[0], cm->frame_offset) > 0 &&
-         get_relative_dist(cm, ref_offsets[1], cm->frame_offset) > 0))
+    if ((get_relative_dist(order_hint_info, ref_offsets[0],
+                           current_frame->order_hint) <= 0 &&
+         get_relative_dist(order_hint_info, ref_offsets[1],
+                           current_frame->order_hint) <= 0) ||
+        (get_relative_dist(order_hint_info, ref_offsets[0],
+                           current_frame->order_hint) > 0 &&
+         get_relative_dist(order_hint_info, ref_offsets[1],
+                           current_frame->order_hint) > 0))
       return 1;
   }
 
-  if (mode_skip_mask[ref_frame[0]] & (1 << this_mode)) {
-    return 1;
-  }
-
-  if ((ref_frame_skip_mask[0] & (1 << ref_frame[0])) &&
-      (ref_frame_skip_mask[1] & (1 << AOMMAX(0, ref_frame[1])))) {
-    return 1;
+  if (sf->selective_ref_frame >= 4 && comp_pred) {
+    // Check if one of the reference is ALTREF2_FRAME and BWDREF_FRAME is a
+    // valid reference.
+    if ((ref_frame[0] == ALTREF2_FRAME || ref_frame[1] == ALTREF2_FRAME) &&
+        (cpi->ref_frame_flags & ref_frame_flag_list[BWDREF_FRAME])) {
+      // Check if both ALTREF2_FRAME and BWDREF_FRAME are future references.
+      if ((get_relative_dist(
+               order_hint_info,
+               cm->cur_frame->ref_order_hints[ALTREF2_FRAME - LAST_FRAME],
+               current_frame->order_hint) > 0) &&
+          (get_relative_dist(
+               order_hint_info,
+               cm->cur_frame->ref_order_hints[BWDREF_FRAME - LAST_FRAME],
+               current_frame->order_hint) > 0)) {
+        // Drop ALTREF2_FRAME as a reference if BWDREF_FRAME is a closer
+        // reference to the current frame than ALTREF2_FRAME
+        if (get_relative_dist(
+                order_hint_info,
+                cm->cur_frame->ref_order_hints[ALTREF2_FRAME - LAST_FRAME],
+                cm->cur_frame->ref_order_hints[BWDREF_FRAME - LAST_FRAME]) >=
+            0) {
+          const RefCntBuffer *const buf_arf2 =
+              get_ref_frame_buf(cm, ALTREF2_FRAME);
+          assert(buf_arf2 != NULL);
+          const RefCntBuffer *const buf_bwd =
+              get_ref_frame_buf(cm, BWDREF_FRAME);
+          assert(buf_bwd != NULL);
+          assert(buf_arf2->frame_rf_level == buf_bwd->frame_rf_level);
+          (void)buf_arf2;
+          (void)buf_bwd;
+          return 1;
+        }
+      }
+    }
   }
 
   if (skip_repeated_mv(cm, x, this_mode, ref_frame, search_state)) {
@@ -10864,18 +11586,15 @@ static int64_t handle_intra_mode(InterModeSearchState *search_state,
 
   TX_SIZE uv_tx;
   int is_directional_mode = av1_is_directional_mode(mbmi->mode);
-  if (is_directional_mode && av1_use_angle_delta(bsize)) {
+  if (is_directional_mode && av1_use_angle_delta(bsize) &&
+      cpi->oxcf.enable_angle_delta) {
     int rate_dummy;
     int64_t model_rd = INT64_MAX;
-    if (!search_state->angle_stats_ready) {
+    if (sf->intra_angle_estimation && !search_state->angle_stats_ready) {
       const int src_stride = x->plane[0].src.stride;
       const uint8_t *src = x->plane[0].src.buf;
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-        highbd_angle_estimation(src, src_stride, rows, cols, bsize,
-                                search_state->directional_mode_skip_mask);
-      else
-        angle_estimation(src, src_stride, rows, cols, bsize,
-                         search_state->directional_mode_skip_mask);
+      angle_estimation(src, src_stride, rows, cols, bsize, is_cur_buf_hbd(xd),
+                       search_state->directional_mode_skip_mask);
       search_state->angle_stats_ready = 1;
     }
     if (search_state->directional_mode_skip_mask[mbmi->mode]) return INT64_MAX;
@@ -11110,6 +11829,8 @@ static void analyze_single_states(const AV1_COMP *cpi,
     for (dir = 0; dir < 2; ++dir) {
       int64_t best_rd;
       SingleInterModeState(*state)[FWD_REFS];
+      const int prune_factor =
+          cpi->sf.prune_comp_search_by_single_result >= 2 ? 6 : 5;
 
       // Use the best rd of GLOBALMV or NEWMV to prune the unlikely
       // reference frames for all the modes (NEARESTMV and NEARMV may not
@@ -11121,7 +11842,7 @@ static void analyze_single_states(const AV1_COMP *cpi,
       for (mode = 0; mode < SINGLE_INTER_MODE_NUM; ++mode) {
         for (i = 1; i < search_state->single_state_cnt[dir][mode]; ++i) {
           if (state[mode][i].rd != INT64_MAX &&
-              (state[mode][i].rd >> 1) > best_rd) {
+              (state[mode][i].rd >> 3) * prune_factor > best_rd) {
             state[mode][i].valid = 0;
           }
         }
@@ -11134,7 +11855,7 @@ static void analyze_single_states(const AV1_COMP *cpi,
         for (i = 1; i < search_state->single_state_modelled_cnt[dir][mode];
              ++i) {
           if (state[mode][i].rd != INT64_MAX &&
-              (state[mode][i].rd >> 1) > best_rd) {
+              (state[mode][i].rd >> 3) * prune_factor > best_rd) {
             state[mode][i].valid = 0;
           }
         }
@@ -11338,6 +12059,44 @@ static INLINE void sf_drop_ref_analyze(InterModeSearchState *search_state,
   }
 }
 
+// sf->prune_single_motion_modes_by_simple_trans
+static int analyze_simple_trans_states(const AV1_COMP *cpi, MACROBLOCK *x) {
+  (void)cpi;
+  int64_t rdcosts[REF_FRAMES] = { INT64_MAX, INT64_MAX, INT64_MAX, INT64_MAX,
+                                  INT64_MAX, INT64_MAX, INT64_MAX, INT64_MAX };
+  int skip_ref = 0;
+  int64_t min_rd = INT64_MAX;
+  for (int i = 0; i < SINGLE_REF_MODES; ++i) {
+    const MODE_DEFINITION *mode_order = &av1_mode_order[i];
+    const MV_REFERENCE_FRAME ref_frame = mode_order->ref_frame[0];
+    for (int k = 0; k < MAX_REF_MV_SERCH; ++k) {
+      const int64_t rd = x->simple_rd_state[i][k].rd_stats.rdcost;
+      rdcosts[ref_frame] = AOMMIN(rdcosts[ref_frame], rd);
+      min_rd = AOMMIN(min_rd, rd);
+    }
+  }
+  int valid_cnt = 0;
+  for (int i = 1; i < REF_FRAMES; ++i) {
+    if (rdcosts[i] == INT64_MAX) {
+      skip_ref |= (1 << i);
+    } else {
+      valid_cnt++;
+    }
+  }
+  if (valid_cnt < 2) {
+    return 0;
+  }
+  min_rd += (min_rd >> 1);
+  if (valid_cnt > 2) {
+    for (int i = 1; i < REF_FRAMES; ++i) {
+      if (rdcosts[i] > min_rd) {
+        skip_ref |= (1 << i);
+      }
+    }
+  }
+  return skip_ref;
+}
+
 static void alloc_compound_type_rd_buffers(AV1_COMMON *const cm,
                                            CompoundTypeRdBuffers *const bufs) {
   CHECK_MEM_ERROR(
@@ -11377,6 +12136,7 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
   const int try_palette =
+      cpi->oxcf.enable_palette &&
       av1_allow_palette(cm->allow_screen_content_tools, mbmi->sb_type);
   PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info;
   const struct segmentation *const seg = &cm->seg;
@@ -11387,9 +12147,14 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
   unsigned int ref_costs_single[REF_FRAMES];
   unsigned int ref_costs_comp[REF_FRAMES][REF_FRAMES];
   int *comp_inter_cost = x->comp_inter_cost[av1_get_reference_mode_context(xd)];
-  int *mode_map = tile_data->mode_map[bsize];
-  uint32_t mode_skip_mask[REF_FRAMES];
-  uint16_t ref_frame_skip_mask[2];
+  mode_skip_mask_t mode_skip_mask;
+  uint8_t motion_mode_skip_mask = 0;  // second pass of single ref modes
+#if CONFIG_ONE_PASS_SVM
+  int temp_y_eob = 0, temp_y_eob_0 = 0, temp_y_eob_1 = 0, temp_y_eob_2 = 0,
+      temp_y_eob_3 = 0;
+  int64_t temp_y_rd = 0, temp_y_rd_0 = 0, temp_y_rd_1 = 0, temp_y_rd_2 = 0,
+          temp_y_rd_3 = 0;
+#endif
 
   InterModeSearchState search_state;
   init_inter_mode_search_state(&search_state, cpi, tile_data, x, bsize,
@@ -11405,64 +12170,71 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     NULL,      search_state.modelled_rd,
     { { 0 } }, INT_MAX,
     INT_MAX,   search_state.simple_rd,
-    0,         interintra_modes
+    0,         interintra_modes,
+    1,         NULL
   };
   for (i = 0; i < REF_FRAMES; ++i) x->pred_sse[i] = INT_MAX;
 
   av1_invalid_rd_stats(rd_cost);
 
   // init params, set frame modes, speed features
-  set_params_rd_pick_inter_mode(
-      cpi, x, &args, bsize, mi_row, mi_col, ref_frame_skip_mask, mode_skip_mask,
-      ctx->skip_ref_frame_mask, ref_costs_single, ref_costs_comp, yv12_mb);
+  set_params_rd_pick_inter_mode(cpi, x, &args, bsize, mi_row, mi_col,
+                                &mode_skip_mask, ctx->skip_ref_frame_mask,
+                                ref_costs_single, ref_costs_comp, yv12_mb);
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
   int64_t best_est_rd = INT64_MAX;
   // TODO(angiebird): Turn this on when this speed feature is well tested
-#if 1
   const InterModeRdModel *md = &tile_data->inter_mode_rd_models[bsize];
-  const int do_tx_search = !md->ready;
-#else
-  const int do_tx_search = 1;
-#endif
-  InterModesInfo *inter_modes_info = &tile_data->inter_modes_info;
+  const int do_tx_search =
+      !((cpi->sf.inter_mode_rd_model_estimation == 1 && md->ready) ||
+        (cpi->sf.inter_mode_rd_model_estimation == 2 &&
+         x->source_variance < 512));
+  InterModesInfo *inter_modes_info = x->inter_modes_info;
   inter_modes_info->num = 0;
-#endif
 
   int intra_mode_num = 0;
   int intra_mode_idx_ls[MAX_MODES];
   int reach_first_comp_mode = 0;
 
   // Temporary buffers used by handle_inter_mode().
-  // We allocate them once and reuse it in every call to that function.
-  // Note: Must be allocated on the heap due to large size of the arrays.
-  uint8_t *tmp_buf_orig;
-  CHECK_MEM_ERROR(
-      cm, tmp_buf_orig,
-      (uint8_t *)aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE));
-  uint8_t *const tmp_buf = get_buf_by_bd(xd, tmp_buf_orig);
+  uint8_t *const tmp_buf = get_buf_by_bd(xd, x->tmp_obmc_bufs[0]);
 
   CompoundTypeRdBuffers rd_buffers;
   alloc_compound_type_rd_buffers(cm, &rd_buffers);
 
   for (int midx = 0; midx < MAX_MODES; ++midx) {
-    int mode_index = mode_map[midx];
-    int64_t this_rd = INT64_MAX;
-    int disable_skip = 0;
-    int rate2 = 0, rate_y = 0, rate_uv = 0;
-    int64_t distortion2 = 0;
-    int skippable = 0;
-    int this_skip2 = 0;
-    const MODE_DEFINITION *mode_order = &av1_mode_order[mode_index];
+    const MODE_DEFINITION *mode_order = &av1_mode_order[midx];
+    this_mode = mode_order->mode;
     const MV_REFERENCE_FRAME ref_frame = mode_order->ref_frame[0];
     const MV_REFERENCE_FRAME second_ref_frame = mode_order->ref_frame[1];
     const int comp_pred = second_ref_frame > INTRA_FRAME;
-    this_mode = mode_order->mode;
 
-    init_mbmi(mbmi, mode_index, cm);
-
-    x->skip = 0;
-    set_ref_ptrs(cm, xd, ref_frame, second_ref_frame);
+    // When single ref motion search ends:
+    // 1st pass: To evaluate single ref RD results and rewind to the beginning;
+    // 2nd pass: To continue with compound ref search.
+    if (sf->prune_single_motion_modes_by_simple_trans) {
+      if (comp_pred && args.single_ref_first_pass) {
+        args.single_ref_first_pass = 0;
+        // Reach the first comp ref mode
+        // Reset midx to start the 2nd pass for single ref motion search
+        midx = -1;
+        motion_mode_skip_mask = analyze_simple_trans_states(cpi, x);
+        continue;
+      }
+      if (!comp_pred) {  // single ref mode
+        if (args.single_ref_first_pass) {
+          // clear stats
+          for (int k = 0; k < MAX_REF_MV_SERCH; ++k) {
+            x->simple_rd_state[midx][k].rd_stats.rdcost = INT64_MAX;
+            x->simple_rd_state[midx][k].early_skipped = 0;
+          }
+        } else {
+          if (motion_mode_skip_mask & (1 << ref_frame)) {
+            continue;
+          }
+        }
+      }
+    }
 
     // Reach the first compound prediction mode
     if (sf->prune_comp_search_by_single_result > 0 && comp_pred &&
@@ -11470,9 +12242,23 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
       analyze_single_states(cpi, &search_state);
       reach_first_comp_mode = 1;
     }
+    int64_t this_rd = INT64_MAX;
+    int disable_skip = 0;
+    int rate2 = 0, rate_y = 0, rate_uv = 0;
+    int64_t distortion2 = 0;
+    int skippable = 0;
+    int this_skip2 = 0;
+
+    init_mbmi(mbmi, midx, cm);
+
+    x->skip = 0;
+    set_ref_ptrs(cm, xd, ref_frame, second_ref_frame);
+
+    if (inter_mode_compatible_skip(cpi, x, bsize, midx)) continue;
+
     const int ret = inter_mode_search_order_independent_skip(
-        cpi, ctx, x, bsize, mode_index, mi_row, mi_col, mode_skip_mask,
-        ref_frame_skip_mask, &search_state);
+        cpi, ctx, x, bsize, midx, mi_row, mi_col, &mode_skip_mask,
+        &search_state);
     if (ret == 1) continue;
     args.skip_motion_mode = (ret == 2);
 
@@ -11482,8 +12268,7 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
       }
     }
 
-    if (search_state.best_rd < search_state.mode_threshold[mode_index])
-      continue;
+    if (search_state.best_rd < search_state.mode_threshold[midx]) continue;
 
     if (sf->prune_comp_search_by_single_result > 0 && comp_pred) {
       if (compound_skip_by_single_states(cpi, &search_state, this_mode,
@@ -11497,7 +12282,9 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     const int compmode_cost =
         is_comp_ref_allowed(mbmi->sb_type) ? comp_inter_cost[comp_pred] : 0;
     const int real_compmode_cost =
-        cm->reference_mode == REFERENCE_MODE_SELECT ? compmode_cost : 0;
+        cm->current_frame.reference_mode == REFERENCE_MODE_SELECT
+            ? compmode_cost
+            : 0;
 
     if (comp_pred) {
       if ((sf->mode_search_skip_flags & FLAG_SKIP_COMP_BESTINTRA) &&
@@ -11507,7 +12294,12 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     }
 
     if (ref_frame == INTRA_FRAME) {
-      if (sf->adaptive_mode_search)
+      if (!cpi->oxcf.enable_smooth_intra &&
+          (mbmi->mode == SMOOTH_PRED || mbmi->mode == SMOOTH_H_PRED ||
+           mbmi->mode == SMOOTH_V_PRED))
+        continue;
+      if (!cpi->oxcf.enable_paeth_intra && mbmi->mode == PAETH_PRED) continue;
+      if (sf->adaptive_mode_search > 1)
         if ((x->source_variance << num_pels_log2_lookup[bsize]) >
             search_state.best_pred_sse)
           continue;
@@ -11535,7 +12327,7 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     }
 
     if (ref_frame == INTRA_FRAME) {
-      intra_mode_idx_ls[intra_mode_num++] = mode_index;
+      intra_mode_idx_ls[intra_mode_num++] = midx;
       continue;
     } else {
       mbmi->angle_delta[PLANE_TYPE_Y] = 0;
@@ -11554,25 +12346,28 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
         args.single_newmv_valid = search_state.single_newmv_valid;
         args.single_comp_cost = real_compmode_cost;
         args.ref_frame_cost = ref_frame_cost;
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
+        if (midx < MAX_SINGLE_REF_MODES) {
+          args.simple_rd_state = x->simple_rd_state[midx];
+        }
         this_rd = handle_inter_mode(
-            cpi, x, bsize, &rd_stats, &rd_stats_y, &rd_stats_uv, &disable_skip,
-            mi_row, mi_col, &args, ref_best_rd, tmp_buf, &rd_buffers, tile_data,
-            &best_est_rd, do_tx_search, inter_modes_info);
-#else
-        this_rd = handle_inter_mode(cpi, x, bsize, &rd_stats, &rd_stats_y,
-                                    &rd_stats_uv, &disable_skip, mi_row, mi_col,
-                                    &args, ref_best_rd, tmp_buf, &rd_buffers);
-#endif
+            cpi, tile_data, x, bsize, &rd_stats, &rd_stats_y, &rd_stats_uv,
+            &disable_skip, mi_row, mi_col, &args, ref_best_rd, tmp_buf,
+            &rd_buffers, &best_est_rd, do_tx_search, inter_modes_info);
         rate2 = rd_stats.rate;
         skippable = rd_stats.skip;
         distortion2 = rd_stats.dist;
         rate_y = rd_stats_y.rate;
         rate_uv = rd_stats_uv.rate;
+#if CONFIG_ONE_PASS_SVM
+        av1_unpack_reg_stat(&rd_stats_y, &temp_y_eob, &temp_y_eob_0,
+                            &temp_y_eob_1, &temp_y_eob_2, &temp_y_eob_3,
+                            &temp_y_rd, &temp_y_rd_0, &temp_y_rd_1,
+                            &temp_y_rd_2, &temp_y_rd_3);
+#endif
       }
 
       if (sf->prune_comp_search_by_single_result > 0 &&
-          is_inter_singleref_mode(this_mode)) {
+          is_inter_singleref_mode(this_mode) && args.single_ref_first_pass) {
         collect_single_states(x, &search_state, mbmi);
       }
 
@@ -11590,11 +12385,11 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     if (this_rd < search_state.best_rd || x->skip) {
       int mode_excluded = 0;
       if (comp_pred) {
-        mode_excluded = cm->reference_mode == SINGLE_REFERENCE;
+        mode_excluded = cm->current_frame.reference_mode == SINGLE_REFERENCE;
       }
       if (!mode_excluded) {
         // Note index of best mode so far
-        search_state.best_mode_index = mode_index;
+        search_state.best_mode_index = midx;
 
         if (ref_frame == INTRA_FRAME) {
           /* required for left and above block mv */
@@ -11610,7 +12405,6 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
         search_state.best_mbmode = *mbmi;
         search_state.best_skip2 = this_skip2;
         search_state.best_mode_skippable = skippable;
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
         if (do_tx_search) {
           // When do_tx_search == 0, handle_inter_mode won't provide correct
           // rate_y and rate_uv because txfm_search process is replaced by
@@ -11621,13 +12415,13 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
               rate_y +
               x->skip_cost[av1_get_skip_context(xd)][this_skip2 || skippable];
           search_state.best_rate_uv = rate_uv;
+
+#if CONFIG_ONE_PASS_SVM
+          av1_set_reg_stat(rd_cost, temp_y_eob, temp_y_eob_0, temp_y_eob_1,
+                           temp_y_eob_2, temp_y_eob_3, temp_y_rd, temp_y_rd_0,
+                           temp_y_rd_1, temp_y_rd_2, temp_y_rd_3);
+#endif
         }
-#else   // CONFIG_COLLECT_INTER_MODE_RD_STATS
-        search_state.best_rate_y =
-            rate_y +
-            x->skip_cost[av1_get_skip_context(xd)][this_skip2 || skippable];
-        search_state.best_rate_uv = rate_uv;
-#endif  // CONFIG_COLLECT_INTER_MODE_RD_STATS
         memcpy(ctx->blk_skip, x->blk_skip,
                sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
       }
@@ -11637,7 +12431,7 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     if (!disable_skip && ref_frame != INTRA_FRAME) {
       int64_t single_rd, hybrid_rd, single_rate, hybrid_rate;
 
-      if (cm->reference_mode == REFERENCE_MODE_SELECT) {
+      if (cm->current_frame.reference_mode == REFERENCE_MODE_SELECT) {
         single_rate = rate2 - compmode_cost;
         hybrid_rate = rate2;
       } else {
@@ -11666,24 +12460,23 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
     if (x->skip && !comp_pred) break;
   }
 
-  aom_free(tmp_buf_orig);
-  tmp_buf_orig = NULL;
   release_compound_type_rd_buffers(&rd_buffers);
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
   if (!do_tx_search) {
     inter_modes_info_sort(inter_modes_info, inter_modes_info->rd_idx_pair_arr);
     search_state.best_rd = INT64_MAX;
 
     int64_t top_est_rd =
-        inter_modes_info->est_rd_arr[inter_modes_info->rd_idx_pair_arr[0].idx];
+        inter_modes_info->num > 0
+            ? inter_modes_info
+                  ->est_rd_arr[inter_modes_info->rd_idx_pair_arr[0].idx]
+            : INT64_MAX;
     for (int j = 0; j < inter_modes_info->num; ++j) {
       const int data_idx = inter_modes_info->rd_idx_pair_arr[j].idx;
       *mbmi = inter_modes_info->mbmi_arr[data_idx];
       int64_t curr_est_rd = inter_modes_info->est_rd_arr[data_idx];
-      if (curr_est_rd * 0.9 > top_est_rd) {
-        continue;
-      }
+      if (curr_est_rd * 0.80 > top_est_rd) break;
+
       const int mode_rate = inter_modes_info->mode_rate_arr[data_idx];
 
       x->skip = 0;
@@ -11700,14 +12493,16 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
       RD_STATS rd_stats_y;
       RD_STATS rd_stats_uv;
 
-      av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
+      av1_enc_build_inter_predictor(cm, xd, mi_row, mi_col, NULL, bsize, 0,
+                                    av1_num_planes(cm) - 1);
       if (mbmi->motion_mode == OBMC_CAUSAL)
         av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
 
-      if (!txfm_search(cpi, x, bsize, mi_row, mi_col, &rd_stats, &rd_stats_y,
-                       &rd_stats_uv, mode_rate, search_state.best_rd)) {
+      if (!txfm_search(cpi, tile_data, x, bsize, mi_row, mi_col, &rd_stats,
+                       &rd_stats_y, &rd_stats_uv, mode_rate,
+                       search_state.best_rd)) {
         continue;
-      } else {
+      } else if (cpi->sf.inter_mode_rd_model_estimation == 1) {
         const int skip_ctx = av1_get_skip_context(xd);
         inter_mode_data_push(tile_data, mbmi->sb_type, rd_stats.sse,
                              rd_stats.dist,
@@ -11733,10 +12528,12 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
         search_state.best_rate_uv = rd_stats_uv.rate;
         memcpy(ctx->blk_skip, x->blk_skip,
                sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
+#if CONFIG_ONE_PASS_SVM
+        av1_copy_reg_stat(rd_cost, &rd_stats_y);
+#endif
       }
     }
   }
-#endif
 
   for (int j = 0; j < intra_mode_num; ++j) {
     const int mode_index = intra_mode_idx_ls[j];
@@ -11775,6 +12572,9 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
       search_state.best_rate_uv = intra_rd_stats_uv.rate;
       memcpy(ctx->blk_skip, x->blk_skip,
              sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
+#if CONFIG_ONE_PASS_SVM
+      av1_copy_reg_stat(rd_cost, &intra_rd_stats_y);
+#endif
     }
   }
 
@@ -11791,7 +12591,7 @@ void av1_rd_pick_inter_mode_sb(AV1_COMP *cpi, TileDataEnc *tile_data,
   }
 
   search_state.best_mbmode.skip_mode = 0;
-  if (cm->skip_mode_flag &&
+  if (cm->current_frame.skip_mode_info.skip_mode_flag &&
       !segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME) &&
       is_comp_ref_allowed(bsize)) {
     rd_pick_skip_mode(rd_cost, &search_state, cpi, x, bsize, mi_row, mi_col,
@@ -11958,7 +12758,7 @@ void av1_rd_pick_inter_mode_sb_seg_skip(const AV1_COMP *cpi,
   mbmi->interp_filters = av1_broadcast_interp_filter(best_filter);
   rate2 += av1_get_switchable_rate(cm, x, xd);
 
-  if (cm->reference_mode == REFERENCE_MODE_SELECT)
+  if (cm->current_frame.reference_mode == REFERENCE_MODE_SELECT)
     rate2 += comp_inter_cost[comp_pred];
 
   // Estimate the reference frame signaling cost and add it
@@ -12010,7 +12810,7 @@ static INLINE void calc_target_weighted_pred_above(
   int32_t *wsrc = ctxt->x->wsrc_buf + (rel_mi_col * MI_SIZE);
   int32_t *mask = ctxt->x->mask_buf + (rel_mi_col * MI_SIZE);
   const uint8_t *tmp = ctxt->tmp + rel_mi_col * MI_SIZE;
-  const int is_hbd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+  const int is_hbd = is_cur_buf_hbd(xd);
 
   if (!is_hbd) {
     for (int row = 0; row < ctxt->overlap; ++row) {
@@ -12056,7 +12856,7 @@ static INLINE void calc_target_weighted_pred_left(
   int32_t *wsrc = ctxt->x->wsrc_buf + (rel_mi_row * MI_SIZE * bw);
   int32_t *mask = ctxt->x->mask_buf + (rel_mi_row * MI_SIZE * bw);
   const uint8_t *tmp = ctxt->tmp + (rel_mi_row * MI_SIZE * ctxt->tmp_stride);
-  const int is_hbd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+  const int is_hbd = is_cur_buf_hbd(xd);
 
   if (!is_hbd) {
     for (int row = 0; row < nb_mi_height * MI_SIZE; ++row) {
@@ -12138,7 +12938,7 @@ static void calc_target_weighted_pred(const AV1_COMMON *cm, const MACROBLOCK *x,
   int32_t *mask_buf = x->mask_buf;
   int32_t *wsrc_buf = x->wsrc_buf;
 
-  const int is_hbd = (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0;
+  const int is_hbd = is_cur_buf_hbd(xd);
   const int src_scale = AOM_BLEND_A64_MAX_ALPHA * AOM_BLEND_A64_MAX_ALPHA;
 
   // plane 0 should not be subsampled
@@ -12197,3 +12997,110 @@ static void calc_target_weighted_pred(const AV1_COMMON *cm, const MACROBLOCK *x,
     }
   }
 }
+
+/* Use standard 3x3 Sobel matrix. Macro so it can be used for either high or
+   low bit-depth arrays. */
+#define SOBEL_X(src, stride, i, j)                       \
+  ((src)[((i)-1) + (stride) * ((j)-1)] -                 \
+   (src)[((i) + 1) + (stride) * ((j)-1)] +  /* NOLINT */ \
+   2 * (src)[((i)-1) + (stride) * (j)] -    /* NOLINT */ \
+   2 * (src)[((i) + 1) + (stride) * (j)] +  /* NOLINT */ \
+   (src)[((i)-1) + (stride) * ((j) + 1)] -  /* NOLINT */ \
+   (src)[((i) + 1) + (stride) * ((j) + 1)]) /* NOLINT */
+#define SOBEL_Y(src, stride, i, j)                       \
+  ((src)[((i)-1) + (stride) * ((j)-1)] +                 \
+   2 * (src)[(i) + (stride) * ((j)-1)] +    /* NOLINT */ \
+   (src)[((i) + 1) + (stride) * ((j)-1)] -  /* NOLINT */ \
+   (src)[((i)-1) + (stride) * ((j) + 1)] -  /* NOLINT */ \
+   2 * (src)[(i) + (stride) * ((j) + 1)] -  /* NOLINT */ \
+   (src)[((i) + 1) + (stride) * ((j) + 1)]) /* NOLINT */
+
+sobel_xy sobel(const uint8_t *input, int stride, int i, int j, bool high_bd) {
+  int16_t s_x;
+  int16_t s_y;
+  if (high_bd) {
+    const uint16_t *src = CONVERT_TO_SHORTPTR(input);
+    s_x = SOBEL_X(src, stride, i, j);
+    s_y = SOBEL_Y(src, stride, i, j);
+  } else {
+    s_x = SOBEL_X(input, stride, i, j);
+    s_y = SOBEL_Y(input, stride, i, j);
+  }
+  sobel_xy r = { .x = s_x, .y = s_y };
+  return r;
+}
+
+// 8-tap Gaussian convolution filter with sigma = 1.3, sums to 128,
+// all co-efficients must be even.
+DECLARE_ALIGNED(16, static const int16_t, gauss_filter[8]) = { 2,  12, 30, 40,
+                                                               30, 12, 2,  0 };
+
+void gaussian_blur(const uint8_t *src, int src_stride, int w, int h,
+                   uint8_t *dst, bool high_bd, int bd) {
+  ConvolveParams conv_params = get_conv_params(0, 0, bd);
+  InterpFilterParams filter = { .filter_ptr = gauss_filter,
+                                .taps = 8,
+                                .subpel_shifts = 0,
+                                .interp_filter = EIGHTTAP_REGULAR };
+  // Requirements from the vector-optimized implementations.
+  assert(h % 4 == 0);
+  assert(w % 8 == 0);
+  // Because we use an eight tap filter, the stride should be at least 7 + w.
+  assert(src_stride >= w + 7);
+  if (high_bd) {
+    av1_highbd_convolve_2d_sr(CONVERT_TO_SHORTPTR(src), src_stride,
+                              CONVERT_TO_SHORTPTR(dst), w, w, h, &filter,
+                              &filter, 0, 0, &conv_params, bd);
+  } else {
+    av1_convolve_2d_sr(src, src_stride, dst, w, w, h, &filter, &filter, 0, 0,
+                       &conv_params);
+  }
+}
+
+static uint16_t edge_probability(const uint8_t *input, int w, int h,
+                                 bool high_bd, int bd) {
+  // The probability of an edge in the whole image is the same as the highest
+  // probability of an edge for any individual pixel. Use Sobel as the metric
+  // for finding an edge.
+  uint16_t highest = 0;
+  // Ignore the 1 pixel border around the image for the computation.
+  for (int j = 1; j < h - 1; ++j) {
+    for (int i = 1; i < w - 1; ++i) {
+      sobel_xy g = sobel(input, w, i, j, high_bd);
+      // Scale down to 8-bit to get same output regardless of bit depth.
+      int16_t g_x = g.x >> (bd - 8);
+      int16_t g_y = g.y >> (bd - 8);
+      uint16_t magnitude = (uint16_t)sqrt(g_x * g_x + g_y * g_y);
+      highest = AOMMAX(highest, magnitude);
+    }
+  }
+  return highest;
+}
+
+/* Uses most of the Canny edge detection algorithm to find if there are any
+ * edges in the image.
+ */
+uint16_t av1_edge_exists(const uint8_t *src, int src_stride, int w, int h,
+                         bool high_bd, int bd) {
+  if (w < 3 || h < 3) {
+    return 0;
+  }
+  uint8_t *blurred;
+  if (high_bd) {
+    blurred = CONVERT_TO_BYTEPTR(aom_memalign(32, sizeof(uint16_t) * w * h));
+  } else {
+    blurred = (uint8_t *)aom_memalign(32, sizeof(uint8_t) * w * h);
+  }
+  gaussian_blur(src, src_stride, w, h, blurred, high_bd, bd);
+  // Skip the non-maximum suppression step in Canny edge detection. We just
+  // want a probability of an edge existing in the buffer, which is determined
+  // by the strongest edge in it -- we don't need to eliminate the weaker
+  // edges. Use Sobel for the edge detection.
+  uint16_t prob = edge_probability(blurred, w, h, high_bd, bd);
+  if (high_bd) {
+    aom_free(CONVERT_TO_SHORTPTR(blurred));
+  } else {
+    aom_free(blurred);
+  }
+  return prob;
+}
diff --git a/third_party/aom/av1/encoder/rdopt.h b/third_party/aom/av1/encoder/rdopt.h
index 4c11f90b8f..983d3850df 100644
--- a/third_party/aom/av1/encoder/rdopt.h
+++ b/third_party/aom/av1/encoder/rdopt.h
@@ -12,6 +12,8 @@
 #ifndef AOM_AV1_ENCODER_RDOPT_H_
 #define AOM_AV1_ENCODER_RDOPT_H_
 
+#include <stdbool.h>
+
 #include "av1/common/blockd.h"
 #include "av1/common/txb_common.h"
 
@@ -126,10 +128,31 @@ void av1_rd_pick_inter_mode_sb_seg_skip(
     struct macroblock *x, int mi_row, int mi_col, struct RD_STATS *rd_cost,
     BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx, int64_t best_rd_so_far);
 
-#if CONFIG_COLLECT_INTER_MODE_RD_STATS
+/** Returns an integer indicating the strength of the edge.
+ * 0 means no edge found, 556 is the strength of a solid black/white edge,
+ * and the number may range higher if the signal is even stronger (e.g., on a
+ * corner). high_bd is a bool indicating the source should be treated
+ * as a 16-bit array. bd is the bit depth.
+ */
+uint16_t av1_edge_exists(const uint8_t *src, int src_stride, int w, int h,
+                         bool high_bd, int bd);
+
+/** Applies a Gaussian blur with sigma = 1.3. Used by av1_edge_exists and
+ * tests.
+ */
+void gaussian_blur(const uint8_t *src, int src_stride, int w, int h,
+                   uint8_t *dst, bool high_bd, int bd);
+
+/* Applies standard 3x3 Sobel matrix. */
+typedef struct {
+  int16_t x;
+  int16_t y;
+} sobel_xy;
+
+sobel_xy sobel(const uint8_t *input, int stride, int i, int j, bool high_bd);
+
 void av1_inter_mode_data_init(struct TileDataEnc *tile_data);
 void av1_inter_mode_data_fit(TileDataEnc *tile_data, int rdmult);
-#endif
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/encoder/reconinter_enc.c b/third_party/aom/av1/encoder/reconinter_enc.c
index 23d920fc32..4b477cef62 100644
--- a/third_party/aom/av1/encoder/reconinter_enc.c
+++ b/third_party/aom/av1/encoder/reconinter_enc.c
@@ -138,26 +138,28 @@ static INLINE void build_inter_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd,
         assert(bw < 8 || bh < 8);
         ConvolveParams conv_params = get_conv_params_no_round(
             0, plane, xd->tmp_conv_dst, tmp_dst_stride, is_compound, xd->bd);
-        conv_params.use_jnt_comp_avg = 0;
+        conv_params.use_dist_wtd_comp_avg = 0;
         struct buf_2d *const dst_buf = &pd->dst;
         uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x;
 
         ref = 0;
-        const RefBuffer *ref_buf =
-            &cm->frame_refs[this_mbmi->ref_frame[ref] - LAST_FRAME];
+        const RefCntBuffer *ref_buf =
+            get_ref_frame_buf(cm, this_mbmi->ref_frame[ref]);
+        const struct scale_factors *ref_scale_factors =
+            get_ref_scale_factors_const(cm, this_mbmi->ref_frame[ref]);
 
         pd->pre[ref].buf0 =
-            (plane == 1) ? ref_buf->buf->u_buffer : ref_buf->buf->v_buffer;
+            (plane == 1) ? ref_buf->buf.u_buffer : ref_buf->buf.v_buffer;
         pd->pre[ref].buf =
             pd->pre[ref].buf0 + scaled_buffer_offset(pre_x, pre_y,
-                                                     ref_buf->buf->uv_stride,
-                                                     &ref_buf->sf);
-        pd->pre[ref].width = ref_buf->buf->uv_crop_width;
-        pd->pre[ref].height = ref_buf->buf->uv_crop_height;
-        pd->pre[ref].stride = ref_buf->buf->uv_stride;
+                                                     ref_buf->buf.uv_stride,
+                                                     ref_scale_factors);
+        pd->pre[ref].width = ref_buf->buf.uv_crop_width;
+        pd->pre[ref].height = ref_buf->buf.uv_crop_height;
+        pd->pre[ref].stride = ref_buf->buf.uv_stride;
 
         const struct scale_factors *const sf =
-            is_intrabc ? &cm->sf_identity : &ref_buf->sf;
+            is_intrabc ? &cm->sf_identity : ref_scale_factors;
         struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
 
         const MV mv = this_mbmi->mv[ref].as_mv;
@@ -194,15 +196,15 @@ static INLINE void build_inter_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd,
   {
     ConvolveParams conv_params = get_conv_params_no_round(
         0, plane, xd->tmp_conv_dst, MAX_SB_SIZE, is_compound, xd->bd);
-    av1_jnt_comp_weight_assign(cm, mi, 0, &conv_params.fwd_offset,
-                               &conv_params.bck_offset,
-                               &conv_params.use_jnt_comp_avg, is_compound);
+    av1_dist_wtd_comp_weight_assign(
+        cm, mi, 0, &conv_params.fwd_offset, &conv_params.bck_offset,
+        &conv_params.use_dist_wtd_comp_avg, is_compound);
 
     struct buf_2d *const dst_buf = &pd->dst;
     uint8_t *const dst = dst_buf->buf;
     for (ref = 0; ref < 1 + is_compound; ++ref) {
       const struct scale_factors *const sf =
-          is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf;
+          is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref];
       struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
       const MV mv = mi->mv[ref].as_mv;
 
@@ -235,46 +237,19 @@ static INLINE void build_inter_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd,
   }
 }
 
-static void build_inter_predictors_for_planes(const AV1_COMMON *cm,
-                                              MACROBLOCKD *xd, BLOCK_SIZE bsize,
-                                              int mi_row, int mi_col,
-                                              int plane_from, int plane_to) {
-  int plane;
+static void build_inter_predictors_for_plane(const AV1_COMMON *cm,
+                                             MACROBLOCKD *xd, int mi_row,
+                                             int mi_col, const BUFFER_SET *ctx,
+                                             BLOCK_SIZE bsize, int plane_idx) {
+  const struct macroblockd_plane *pd = &xd->plane[plane_idx];
+  if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                           pd->subsampling_y))
+    return;
+
   const int mi_x = mi_col * MI_SIZE;
   const int mi_y = mi_row * MI_SIZE;
-  for (plane = plane_from; plane <= plane_to; ++plane) {
-    const struct macroblockd_plane *pd = &xd->plane[plane];
-    const int bw = pd->width;
-    const int bh = pd->height;
-
-    if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
-                             pd->subsampling_y))
-      continue;
-
-    build_inter_predictors(cm, xd, plane, xd->mi[0], 0, bw, bh, mi_x, mi_y);
-  }
-}
-
-void av1_build_inter_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    int mi_row, int mi_col, BUFFER_SET *ctx,
-                                    BLOCK_SIZE bsize) {
-  av1_build_inter_predictors_sbp(cm, xd, mi_row, mi_col, ctx, bsize, 0);
-}
-
-void av1_build_inter_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                     int mi_row, int mi_col, BUFFER_SET *ctx,
-                                     BLOCK_SIZE bsize) {
-  for (int plane_idx = 1; plane_idx < MAX_MB_PLANE; plane_idx++) {
-    av1_build_inter_predictors_sbp(cm, xd, mi_row, mi_col, ctx, bsize,
-                                   plane_idx);
-  }
-}
-
-void av1_build_inter_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    int mi_row, int mi_col, BUFFER_SET *ctx,
-                                    BLOCK_SIZE bsize, int plane_idx) {
-  build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, plane_idx,
-                                    plane_idx);
+  build_inter_predictors(cm, xd, plane_idx, xd->mi[0], 0, pd->width, pd->height,
+                         mi_x, mi_y);
 
   if (is_interintra_pred(xd->mi[0])) {
     BUFFER_SET default_ctx = { { NULL, NULL, NULL }, { 0, 0, 0 } };
@@ -289,13 +264,14 @@ void av1_build_inter_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
   }
 }
 
-void av1_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                   int mi_row, int mi_col, BUFFER_SET *ctx,
-                                   BLOCK_SIZE bsize) {
-  const int num_planes = av1_num_planes(cm);
-  av1_build_inter_predictors_sby(cm, xd, mi_row, mi_col, ctx, bsize);
-  if (num_planes > 1)
-    av1_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, ctx, bsize);
+void av1_enc_build_inter_predictor(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                   int mi_row, int mi_col,
+                                   const BUFFER_SET *ctx, BLOCK_SIZE bsize,
+                                   int plane_from, int plane_to) {
+  for (int plane_idx = plane_from; plane_idx <= plane_to; ++plane_idx) {
+    build_inter_predictors_for_plane(cm, xd, mi_row, mi_col, ctx, bsize,
+                                     plane_idx);
+  }
 }
 
 // TODO(sarahparker):
@@ -308,7 +284,7 @@ void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                                InterpFilters interp_filters,
                                const WarpTypesAllowed *warp_types, int p_col,
                                int p_row, int plane, int ref,
-                               enum mv_precision precision, int x, int y,
+                               mv_precision precision, int x, int y,
                                const MACROBLOCKD *xd, int can_use_previous) {
   const int is_q4 = precision == MV_PRECISION_Q4;
   const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
@@ -338,7 +314,7 @@ static INLINE void build_prediction_by_above_pred(
   MB_MODE_INFO backup_mbmi = *above_mbmi;
 
   av1_setup_build_prediction_by_above_pred(xd, rel_mi_col, above_mi_width,
-                                           above_mbmi, ctxt, num_planes);
+                                           &backup_mbmi, ctxt, num_planes);
   mi_x = above_mi_col << MI_SIZE_LOG2;
   mi_y = ctxt->mi_row << MI_SIZE_LOG2;
 
@@ -351,9 +327,9 @@ static INLINE void build_prediction_by_above_pred(
                    block_size_high[BLOCK_64X64] >> (pd->subsampling_y + 1));
 
     if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;
-    build_inter_predictors(ctxt->cm, xd, j, above_mbmi, 1, bw, bh, mi_x, mi_y);
+    build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+                           mi_y);
   }
-  *above_mbmi = backup_mbmi;
 }
 
 void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
@@ -394,7 +370,7 @@ static INLINE void build_prediction_by_left_pred(
   MB_MODE_INFO backup_mbmi = *left_mbmi;
 
   av1_setup_build_prediction_by_left_pred(xd, rel_mi_row, left_mi_height,
-                                          left_mbmi, ctxt, num_planes);
+                                          &backup_mbmi, ctxt, num_planes);
   mi_x = ctxt->mi_col << MI_SIZE_LOG2;
   mi_y = left_mi_row << MI_SIZE_LOG2;
   const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
@@ -406,9 +382,9 @@ static INLINE void build_prediction_by_left_pred(
     int bh = (left_mi_height << MI_SIZE_LOG2) >> pd->subsampling_y;
 
     if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;
-    build_inter_predictors(ctxt->cm, xd, j, left_mbmi, 1, bw, bh, mi_x, mi_y);
+    build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+                           mi_y);
   }
-  *left_mbmi = backup_mbmi;
 }
 
 void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
@@ -451,7 +427,7 @@ void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
   int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
   int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
 
-  if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  if (is_cur_buf_hbd(xd)) {
     int len = sizeof(uint16_t);
     dst_buf1[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0]);
     dst_buf1[1] =
@@ -475,7 +451,7 @@ void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                       dst_width1, dst_height1, dst_stride1);
   av1_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
                                      dst_width2, dst_height2, dst_stride2);
-  av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, get_frame_new_buffer(cm),
+  av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, &cm->cur_frame->buf,
                        mi_row, mi_col, 0, num_planes);
   av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, dst_stride1,
                                   dst_buf2, dst_stride2);
@@ -492,7 +468,7 @@ static void build_inter_predictors_single_buf(MACROBLOCKD *xd, int plane,
   struct macroblockd_plane *const pd = &xd->plane[plane];
   const MB_MODE_INFO *mi = xd->mi[0];
 
-  const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
+  const struct scale_factors *const sf = xd->block_ref_scale_factors[ref];
   struct buf_2d *const pre_buf = &pd->pre[ref];
   uint8_t *const dst = get_buf_by_bd(xd, ext_dst) + ext_dst_stride * y + x;
   const MV mv = mi->mv[ref].as_mv;
@@ -574,37 +550,41 @@ static void build_wedge_inter_predictor_from_buf(
   uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
   mbmi->interinter_comp.seg_mask = xd->seg_mask;
   const INTERINTER_COMPOUND_DATA *comp_data = &mbmi->interinter_comp;
+  const int is_hbd = is_cur_buf_hbd(xd);
 
   if (is_compound && is_masked_compound_type(comp_data->type)) {
     if (!plane && comp_data->type == COMPOUND_DIFFWTD) {
-      if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+      if (is_hbd) {
         av1_build_compound_diffwtd_mask_highbd(
             comp_data->seg_mask, comp_data->mask_type,
             CONVERT_TO_BYTEPTR(ext_dst0), ext_dst_stride0,
             CONVERT_TO_BYTEPTR(ext_dst1), ext_dst_stride1, h, w, xd->bd);
-      else
+      } else {
         av1_build_compound_diffwtd_mask(
             comp_data->seg_mask, comp_data->mask_type, ext_dst0,
             ext_dst_stride0, ext_dst1, ext_dst_stride1, h, w);
+      }
     }
 
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+    if (is_hbd) {
       build_masked_compound_highbd(
           dst, dst_buf->stride, CONVERT_TO_BYTEPTR(ext_dst0), ext_dst_stride0,
           CONVERT_TO_BYTEPTR(ext_dst1), ext_dst_stride1, comp_data,
           mbmi->sb_type, h, w, xd->bd);
-    else
+    } else {
       build_masked_compound(dst, dst_buf->stride, ext_dst0, ext_dst_stride0,
                             ext_dst1, ext_dst_stride1, comp_data, mbmi->sb_type,
                             h, w);
+    }
   } else {
-    if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+    if (is_hbd) {
       aom_highbd_convolve_copy(CONVERT_TO_BYTEPTR(ext_dst0), ext_dst_stride0,
                                dst, dst_buf->stride, NULL, 0, NULL, 0, w, h,
                                xd->bd);
-    else
+    } else {
       aom_convolve_copy(ext_dst0, ext_dst_stride0, dst, dst_buf->stride, NULL,
                         0, NULL, 0, w, h);
+    }
   }
 }
 
diff --git a/third_party/aom/av1/encoder/reconinter_enc.h b/third_party/aom/av1/encoder/reconinter_enc.h
index 10d5e8c284..5687168315 100644
--- a/third_party/aom/av1/encoder/reconinter_enc.h
+++ b/third_party/aom/av1/encoder/reconinter_enc.h
@@ -23,21 +23,10 @@
 extern "C" {
 #endif
 
-void av1_build_inter_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    int mi_row, int mi_col, BUFFER_SET *ctx,
-                                    BLOCK_SIZE bsize);
-
-void av1_build_inter_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                     int mi_row, int mi_col, BUFFER_SET *ctx,
-                                     BLOCK_SIZE bsize);
-
-void av1_build_inter_predictors_sbp(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    int mi_row, int mi_col, BUFFER_SET *ctx,
-                                    BLOCK_SIZE bsize, int plane_idx);
-
-void av1_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                   int mi_row, int mi_col, BUFFER_SET *ctx,
-                                   BLOCK_SIZE bsize);
+void av1_enc_build_inter_predictor(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                   int mi_row, int mi_col,
+                                   const BUFFER_SET *ctx, BLOCK_SIZE bsize,
+                                   int plane_from, int plane_to);
 
 void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                                int dst_stride, const MV *src_mv,
@@ -46,7 +35,7 @@ void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                                InterpFilters interp_filters,
                                const WarpTypesAllowed *warp_types, int p_col,
                                int p_row, int plane, int ref,
-                               enum mv_precision precision, int x, int y,
+                               mv_precision precision, int x, int y,
                                const MACROBLOCKD *xd, int can_use_previous);
 
 // Detect if the block have sub-pixel level motion vectors
diff --git a/third_party/aom/av1/encoder/segmentation.c b/third_party/aom/av1/encoder/segmentation.c
index 2e9102745c..6d0c6549df 100644
--- a/third_party/aom/av1/encoder/segmentation.c
+++ b/third_party/aom/av1/encoder/segmentation.c
@@ -62,7 +62,7 @@ static void count_segs(const AV1_COMMON *cm, MACROBLOCKD *xd,
   no_pred_segcounts[segment_id]++;
 
   // Temporal prediction not allowed on key frames
-  if (cm->frame_type != KEY_FRAME) {
+  if (cm->current_frame.frame_type != KEY_FRAME) {
     const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
     // Test to see if the segment id matches the predicted value.
     const int pred_segment_id =
diff --git a/third_party/aom/av1/encoder/speed_features.c b/third_party/aom/av1/encoder/speed_features.c
index 4c35baae01..bbb68426fa 100644
--- a/third_party/aom/av1/encoder/speed_features.c
+++ b/third_party/aom/av1/encoder/speed_features.c
@@ -24,6 +24,8 @@
 #define DISABLE_TRELLISQ_SEARCH 0
 
 #define MAX_MESH_SPEED 5  // Max speed setting for mesh motion method
+// Max speed setting for tx domain evaluation
+#define MAX_TX_DOMAIN_EVAL_SPEED 5
 static MESH_PATTERN
     good_quality_mesh_patterns[MAX_MESH_SPEED + 1][MAX_MESH_STEP] = {
       { { 64, 8 }, { 28, 4 }, { 15, 1 }, { 7, 1 } },
@@ -50,6 +52,22 @@ static MESH_PATTERN intrabc_mesh_patterns[MAX_MESH_SPEED + 1][MAX_MESH_STEP] = {
 static uint8_t intrabc_max_mesh_pct[MAX_MESH_SPEED + 1] = { 100, 100, 100,
                                                             25,  25,  10 };
 
+// Threshold values to be used for pruning the txfm_domain_distortion
+// based on block MSE
+// TODO(any): Experiment the threshold logic based on variance metric
+static unsigned int tx_domain_dist_thresholds[MAX_TX_DOMAIN_EVAL_SPEED + 1] = {
+  UINT_MAX, 162754, 22026, 22026, 22026, 0
+};
+// Threshold values to be used for disabling coeff RD-optimization
+// based on block MSE
+// TODO(any): Experiment the threshold logic based on variance metric
+static unsigned int coeff_opt_dist_thresholds[5] = { UINT_MAX, 162754, 162754,
+                                                     22026, 22026 };
+// scaling values to be used for gating wedge/compound segment based on best
+// approximate rd
+static int comp_type_rd_threshold_mul[3] = { 1, 11, 12 };
+static int comp_type_rd_threshold_div[3] = { 3, 16, 16 };
+
 // Intra only frames, golden frames (except alt ref overlays) and
 // alt ref frames tend to be coded at a higher than ambient quality
 static int frame_is_boosted(const AV1_COMP *cpi) {
@@ -62,7 +80,7 @@ static int frame_is_boosted(const AV1_COMP *cpi) {
 // partly on the screen area that over which they propogate. Propogation is
 // limited by transform block size but the screen area take up by a given block
 // size will be larger for a small image format stretched to full screen.
-static BLOCK_SIZE set_partition_min_limit(AV1_COMMON *const cm) {
+static BLOCK_SIZE set_partition_min_limit(const AV1_COMMON *const cm) {
   unsigned int screen_area = (cm->width * cm->height);
 
   // Select block size based on image format size.
@@ -85,10 +103,9 @@ static int has_internal_image_edge(const AV1_COMP *cpi) {
           (cpi->twopass.this_frame_stats.inactive_zone_cols > 0));
 }
 
-static void set_good_speed_feature_framesize_dependent(AV1_COMP *cpi,
-                                                       SPEED_FEATURES *sf,
-                                                       int speed) {
-  AV1_COMMON *const cm = &cpi->common;
+static void set_good_speed_feature_framesize_dependent(
+    const AV1_COMP *const cpi, SPEED_FEATURES *const sf, int speed) {
+  const AV1_COMMON *const cm = &cpi->common;
   const int is_720p_or_larger = AOMMIN(cm->width, cm->height) >= 720;
   const int is_480p_or_larger = AOMMIN(cm->width, cm->height) >= 480;
 
@@ -126,6 +143,17 @@ static void set_good_speed_feature_framesize_dependent(AV1_COMP *cpi,
   }
 
   if (speed >= 2) {
+    if (is_720p_or_larger) {
+      sf->use_square_partition_only_threshold = BLOCK_64X64;
+    } else if (is_480p_or_larger) {
+      sf->use_square_partition_only_threshold = BLOCK_32X32;
+    } else {
+      // TODO(chiyotsai@google.com): Setting the threshold to BLOCK_16X16 incurs
+      // a large loss (about 0.584%). Try increasing the threshold on boosted
+      // frame and see if it improves the performance.
+      sf->use_square_partition_only_threshold = BLOCK_32X32;
+    }
+
     if (is_720p_or_larger) {
       sf->disable_split_mask =
           cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
@@ -172,11 +200,12 @@ static void set_good_speed_feature_framesize_dependent(AV1_COMP *cpi,
   }
 }
 
-static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
-                                                          SPEED_FEATURES *sf,
-                                                          int speed) {
-  AV1_COMMON *const cm = &cpi->common;
+static void set_good_speed_features_framesize_independent(
+    const AV1_COMP *const cpi, SPEED_FEATURES *const sf, int speed) {
+  const AV1_COMMON *const cm = &cpi->common;
   const int boosted = frame_is_boosted(cpi);
+  const int is_boosted_arf2_bwd_type =
+      boosted || cpi->refresh_bwd_ref_frame || cpi->refresh_alt2_ref_frame;
 
   // Speed 0 for all speed features that give neutral coding performance change.
   sf->reduce_inter_modes = 1;
@@ -184,54 +213,79 @@ static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
   sf->ml_prune_rect_partition = 1;
   sf->ml_prune_ab_partition = 1;
   sf->ml_prune_4_partition = 1;
+  sf->simple_motion_search_prune_rect = 1;
   sf->adaptive_txb_search_level = 1;
-  sf->jnt_comp_skip_mv_search = 1;
+  sf->use_dist_wtd_comp_flag = DIST_WTD_COMP_SKIP_MV_SEARCH;
   sf->model_based_prune_tx_search_level = 1;
   sf->model_based_post_interp_filter_breakout = 1;
+  sf->model_based_motion_mode_rd_breakout = 1;
+
+  // TODO(debargha): Test, tweak and turn on either 1 or 2
   sf->inter_mode_rd_model_estimation = 1;
-  sf->prune_ref_frame_for_rect_partitions =
-      !(boosted || cpi->refresh_bwd_ref_frame || cpi->refresh_alt2_ref_frame);
+
+  sf->prune_ref_frame_for_rect_partitions = !is_boosted_arf2_bwd_type;
+  sf->prune_ref_mode_for_partitions = sf->prune_ref_frame_for_rect_partitions;
   sf->less_rectangular_check_level = 1;
-  sf->gm_search_type = GM_REDUCED_REF_SEARCH;
+  sf->gm_search_type = GM_REDUCED_REF_SEARCH_SKIP_L2_L3;
   sf->gm_disable_recode = 1;
+  sf->use_fast_interpolation_filter_search = 1;
+  sf->intra_tx_size_search_init_depth_sqr = 1;
+  sf->intra_angle_estimation = 1;
+  sf->selective_ref_frame = 1;
+  sf->prune_wedge_pred_diff_based = 1;
+  sf->disable_wedge_search_var_thresh = 0;
+  sf->disable_wedge_search_edge_thresh = 0;
+  sf->prune_motion_mode_level = 1;
+  sf->cb_pred_filter_search = 0;
 
   if (speed >= 1) {
     sf->gm_erroradv_type = GM_ERRORADV_TR_1;
-    sf->selective_ref_frame = 1;
+    sf->selective_ref_frame = 2;
+
     sf->inter_tx_size_search_init_depth_rect = 1;
     sf->inter_tx_size_search_init_depth_sqr = 1;
     sf->intra_tx_size_search_init_depth_rect = 1;
-    sf->intra_tx_size_search_init_depth_sqr = 1;
     sf->tx_size_search_lgr_block = 1;
     if (speed >= CONFIG_2PASS_PARTITION_SEARCH_LVL) {
       sf->two_pass_partition_search = 1;
       sf->mode_pruning_based_on_two_pass_partition_search = 1;
     }
     sf->prune_ext_partition_types_search_level = 2;
-    sf->use_fast_interpolation_filter_search = 1;
     sf->skip_repeat_interpolation_filter_search = 1;
     sf->tx_type_search.skip_tx_search = 1;
     sf->tx_type_search.ml_tx_split_thresh = 40;
     sf->model_based_prune_tx_search_level = 0;
-    sf->model_based_post_interp_filter_breakout = 0;
-    // TODO(angiebird): Re-evaluate the impact of inter_mode_rd_model_estimation
-    // on speed 1
-    sf->inter_mode_rd_model_estimation = 0;
     sf->adaptive_txb_search_level = 2;
     sf->use_intra_txb_hash = 1;
     sf->optimize_b_precheck = 1;
     sf->dual_sgr_penalty_level = 1;
-    sf->use_accurate_subpel_search = 1;
+    sf->use_accurate_subpel_search = USE_4_TAPS;
     sf->reuse_inter_intra_mode = 1;
     sf->prune_comp_search_by_single_result = 1;
     sf->skip_repeated_newmv = 1;
     sf->obmc_full_pixel_search_level = 1;
+    // TODO(anyone): Following speed feature will be further explored to
+    // identify the appropriate tradeoff between encoder performance and its
+    // speed.
+    sf->prune_single_motion_modes_by_simple_trans = 1;
+
+    sf->simple_motion_search_split_only = 1;
+
+    sf->disable_wedge_search_var_thresh = 0;
+    sf->disable_wedge_search_edge_thresh = 0;
+    sf->disable_interinter_wedge_newmv_search = boosted ? 0 : 1;
+    sf->prune_comp_type_by_comp_avg = 1;
+    sf->prune_motion_mode_level = 2;
+    sf->gm_search_type = GM_REDUCED_REF_SEARCH_SKIP_L2_L3_ARF2;
+    sf->cb_pred_filter_search = 1;
+    sf->use_transform_domain_distortion = boosted ? 0 : 1;
+    sf->perform_coeff_opt = boosted ? 0 : 1;
   }
 
   if (speed >= 2) {
     sf->gm_erroradv_type = GM_ERRORADV_TR_2;
 
-    sf->selective_ref_frame = 2;
+    sf->selective_ref_frame = 3;
     sf->fast_cdef_search = 1;
 
     sf->adaptive_rd_thresh = 1;
@@ -246,10 +300,19 @@ static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
     // sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
     sf->allow_partition_search_skip = 1;
     sf->disable_wedge_search_var_thresh = 100;
+    sf->disable_wedge_search_edge_thresh = 0;
+    sf->disable_interinter_wedge_newmv_search = 1;
     sf->fast_wedge_sign_estimate = 1;
+    sf->disable_dual_filter = 1;
+    sf->use_dist_wtd_comp_flag = DIST_WTD_COMP_DISABLED;
+    sf->prune_comp_type_by_comp_avg = 2;
+    sf->cb_pred_filter_search = 0;
+    sf->adaptive_interp_filter_search = 1;
+    sf->perform_coeff_opt = boosted ? 0 : 2;
   }
 
   if (speed >= 3) {
+    sf->selective_ref_frame = 4;
     sf->tx_size_search_method = boosted ? USE_FULL_RD : USE_LARGESTALL;
     sf->less_rectangular_check_level = 2;
     sf->adaptive_pred_interp_filter = 1;
@@ -258,26 +321,33 @@ static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
     // See aomedia:1778.
     // sf->adaptive_motion_search = 1;
     sf->recode_loop = ALLOW_RECODE_KFARFGF;
-    sf->use_transform_domain_distortion = 1;
-    sf->use_accurate_subpel_search = 0;
+    sf->use_transform_domain_distortion = boosted ? 1 : 2;
+    sf->use_accurate_subpel_search = USE_2_TAPS;
     sf->adaptive_rd_thresh = 2;
     sf->tx_type_search.prune_mode = PRUNE_2D_FAST;
     sf->gm_search_type = GM_DISABLE_SEARCH;
     sf->prune_comp_search_by_single_result = 2;
+    sf->prune_motion_mode_level = boosted ? 2 : 3;
+    sf->prune_warp_using_wmtype = 1;
+    // TODO(yunqing): evaluate this speed feature for speed 1 & 2, and combine
+    // it with cpi->sf.disable_wedge_search_var_thresh.
+    sf->disable_wedge_interintra_search = 1;
+    sf->perform_coeff_opt = is_boosted_arf2_bwd_type ? 2 : 3;
   }
 
   if (speed >= 4) {
+    sf->use_intra_txb_hash = 0;
     sf->tx_type_search.fast_intra_tx_type_search = 1;
-    sf->tx_type_search.fast_inter_tx_type_search = 1;
     sf->use_square_partition_only_threshold =
         boosted ? BLOCK_128X128 : BLOCK_4X4;
-    sf->tx_size_search_method =
-        frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL;
     sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED;
     sf->adaptive_pred_interp_filter = 0;
     sf->adaptive_mode_search = 1;
     sf->cb_partition_search = !boosted;
     sf->alt_ref_search_fp = 1;
+    sf->skip_sharp_interp_filter_search = 1;
+    sf->perform_coeff_opt = is_boosted_arf2_bwd_type ? 2 : 4;
+    sf->adaptive_txb_search_level = boosted ? 2 : 3;
   }
 
   if (speed >= 5) {
@@ -294,7 +364,7 @@ static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
     sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED_MORE;
     sf->adaptive_rd_thresh = 4;
     sf->mode_search_skip_flags =
-        (cm->frame_type == KEY_FRAME)
+        (cm->current_frame.frame_type == KEY_FRAME)
             ? 0
             : FLAG_SKIP_INTRA_DIRMISMATCH | FLAG_SKIP_INTRA_BESTINTER |
                   FLAG_SKIP_COMP_BESTINTRA | FLAG_SKIP_INTRA_LOWVAR |
@@ -327,35 +397,28 @@ static void set_good_speed_features_framesize_independent(AV1_COMP *cpi,
     sf->mv.search_method = FAST_HEX;
     sf->partition_search_type = REFERENCE_PARTITION;
     sf->mode_search_skip_flags |= FLAG_SKIP_INTRA_DIRMISMATCH;
+    // TODO(any): evaluate adaptive_mode_search=1 for speed 7 & 8
+    sf->adaptive_mode_search = 2;
   }
   if (speed >= 8) {
     sf->mv.search_method = FAST_DIAMOND;
-    sf->mv.subpel_force_stop = 2;
+    sf->mv.subpel_force_stop = HALF_PEL;
     sf->lpf_pick = LPF_PICK_MINIMAL_LPF;
   }
 }
 
-void av1_set_speed_features_framesize_dependent(AV1_COMP *cpi) {
+void av1_set_speed_features_framesize_dependent(AV1_COMP *cpi, int speed) {
   SPEED_FEATURES *const sf = &cpi->sf;
   const AV1EncoderConfig *const oxcf = &cpi->oxcf;
-  RD_OPT *const rd = &cpi->rd;
-  int i;
 
   if (oxcf->mode == GOOD) {
-    set_good_speed_feature_framesize_dependent(cpi, sf, oxcf->speed);
+    set_good_speed_feature_framesize_dependent(cpi, sf, speed);
   }
 
   if (sf->disable_split_mask == DISABLE_ALL_SPLIT) {
     sf->adaptive_pred_interp_filter = 0;
   }
 
-  // Check for masked out split cases.
-  for (i = 0; i < MAX_REFS; ++i) {
-    if (sf->disable_split_mask & (1 << i)) {
-      rd->thresh_mult_sub8x8[i] = INT_MAX;
-    }
-  }
-
   // This is only used in motion vector unit test.
   if (cpi->oxcf.motion_vector_unit_test == 1)
     cpi->find_fractional_mv_step = av1_return_max_sub_pixel_mv;
@@ -363,7 +426,7 @@ void av1_set_speed_features_framesize_dependent(AV1_COMP *cpi) {
     cpi->find_fractional_mv_step = av1_return_min_sub_pixel_mv;
 }
 
-void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
+void av1_set_speed_features_framesize_independent(AV1_COMP *cpi, int speed) {
   AV1_COMMON *const cm = &cpi->common;
   SPEED_FEATURES *const sf = &cpi->sf;
   MACROBLOCK *const x = &cpi->td.mb;
@@ -376,7 +439,7 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->recode_loop = ALLOW_RECODE;
   sf->mv.subpel_search_method = SUBPEL_TREE;
   sf->mv.subpel_iters_per_step = 2;
-  sf->mv.subpel_force_stop = 0;
+  sf->mv.subpel_force_stop = EIGHTH_PEL;
 #if DISABLE_TRELLISQ_SEARCH == 2
   sf->optimize_coefficients = !is_lossless_requested(&cpi->oxcf)
                                   ? FINAL_PASS_TRELLIS_OPT
@@ -402,6 +465,7 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->tx_size_search_lgr_block = 0;
   sf->model_based_prune_tx_search_level = 0;
   sf->model_based_post_interp_filter_breakout = 0;
+  sf->model_based_motion_mode_rd_breakout = 0;
   sf->reduce_inter_modes = 0;
   sf->selective_ref_gm = 1;
   sf->adaptive_motion_search = 0;
@@ -420,6 +484,7 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->less_rectangular_check_level = 0;
   sf->use_square_partition_only_threshold = BLOCK_128X128;
   sf->prune_ref_frame_for_rect_partitions = 0;
+  sf->prune_ref_mode_for_partitions = 0;
   sf->auto_min_max_partition_size = NOT_IN_USE;
   sf->rd_auto_partition_min_limit = BLOCK_4X4;
   sf->default_max_partition_size = BLOCK_LARGEST;
@@ -429,9 +494,11 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->mode_search_skip_flags = 0;
   sf->disable_filter_search_var_thresh = 0;
   sf->allow_partition_search_skip = 0;
-  sf->use_accurate_subpel_search = 2;
+  sf->use_accurate_subpel_search = USE_8_TAPS;
+  sf->disable_wedge_search_edge_thresh = 0;
   sf->disable_wedge_search_var_thresh = 0;
   sf->fast_wedge_sign_estimate = 0;
+  sf->prune_wedge_pred_diff_based = 0;
   sf->drop_ref = 0;
   sf->skip_intra_in_interframe = 1;
   sf->txb_split_cap = 1;
@@ -439,12 +506,17 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->two_pass_partition_search = 0;
   sf->mode_pruning_based_on_two_pass_partition_search = 0;
   sf->use_intra_txb_hash = 0;
-  sf->use_inter_txb_hash = 1;
+  // TODO(any) : clean use_inter_txb_hash code
+  sf->use_inter_txb_hash = 0;
   sf->use_mb_rd_hash = 1;
   sf->optimize_b_precheck = 0;
-  sf->jnt_comp_fast_tx_search = 0;
-  sf->jnt_comp_skip_mv_search = 0;
+  sf->dist_wtd_comp_fast_tx_search = 0;
+  sf->use_dist_wtd_comp_flag = DIST_WTD_COMP_ENABLED;
   sf->reuse_inter_intra_mode = 0;
+  sf->intra_angle_estimation = 0;
+  sf->skip_obmc_in_uniform_mv_field = 0;
+  sf->skip_wm_in_uniform_mv_field = 0;
+  sf->adaptive_interp_filter_search = 0;
 
   for (i = 0; i < TX_SIZES; i++) {
     sf->intra_y_mode_mask[i] = INTRA_ALL;
@@ -466,27 +538,48 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   sf->ml_prune_ab_partition = 0;
   sf->ml_prune_4_partition = 0;
   sf->fast_cdef_search = 0;
-  for (i = 0; i < PARTITION_BLOCK_SIZES; ++i)
+  for (i = 0; i < PARTITION_BLOCK_SIZES; ++i) {
     sf->ml_partition_search_breakout_thresh[i] = -1;  // -1 means not enabled.
+  }
+  sf->simple_motion_search_split_only = 0;
+  sf->simple_motion_search_prune_rect = 0;
 
   // Set this at the appropriate speed levels
   sf->use_transform_domain_distortion = 0;
   sf->gm_search_type = GM_FULL_SEARCH;
   sf->gm_disable_recode = 0;
   sf->use_fast_interpolation_filter_search = 0;
+  sf->disable_dual_filter = 0;
   sf->skip_repeat_interpolation_filter_search = 0;
   sf->use_hash_based_trellis = 0;
   sf->prune_comp_search_by_single_result = 0;
   sf->skip_repeated_newmv = 0;
+  sf->prune_single_motion_modes_by_simple_trans = 0;
 
   // Set decoder side speed feature to use less dual sgr modes
   sf->dual_sgr_penalty_level = 0;
 
+  // TODO(angiebird, debargha): Re-evaluate the impact of
+  // inter_mode_rd_model_estimation in conjunction with
+  // model_based_motion_mode_rd_breakout
   sf->inter_mode_rd_model_estimation = 0;
+
   sf->obmc_full_pixel_search_level = 0;
+  sf->skip_sharp_interp_filter_search = 0;
+  sf->prune_comp_type_by_comp_avg = 0;
+  sf->disable_interinter_wedge_newmv_search = 0;
+  sf->prune_motion_mode_level = 0;
+  sf->prune_warp_using_wmtype = 0;
+
+  sf->disable_wedge_interintra_search = 0;
+  sf->perform_coeff_opt = 0;
 
   if (oxcf->mode == GOOD)
-    set_good_speed_features_framesize_independent(cpi, sf, oxcf->speed);
+    set_good_speed_features_framesize_independent(cpi, sf, speed);
+
+  if (!cpi->seq_params_locked) {
+    cpi->common.seq_params.enable_dual_filter &= !sf->disable_dual_filter;
+  }
 
   // sf->partition_search_breakout_dist_thr is set assuming max 64x64
   // blocks. Normalise this if the blocks are bigger.
@@ -497,28 +590,31 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   cpi->diamond_search_sad = av1_diamond_search_sad;
 
   sf->allow_exhaustive_searches = 1;
-  int speed = (oxcf->speed > MAX_MESH_SPEED) ? MAX_MESH_SPEED : oxcf->speed;
+
+  const int mesh_speed = AOMMIN(speed, MAX_MESH_SPEED);
   if (cpi->twopass.fr_content_type == FC_GRAPHICS_ANIMATION)
     sf->exhaustive_searches_thresh = (1 << 24);
   else
     sf->exhaustive_searches_thresh = (1 << 25);
-  sf->max_exaustive_pct = good_quality_max_mesh_pct[speed];
-  if (speed > 0)
+  sf->max_exaustive_pct = good_quality_max_mesh_pct[mesh_speed];
+  if (mesh_speed > 0)
     sf->exhaustive_searches_thresh = sf->exhaustive_searches_thresh << 1;
 
   for (i = 0; i < MAX_MESH_STEP; ++i) {
-    sf->mesh_patterns[i].range = good_quality_mesh_patterns[speed][i].range;
+    sf->mesh_patterns[i].range =
+        good_quality_mesh_patterns[mesh_speed][i].range;
     sf->mesh_patterns[i].interval =
-        good_quality_mesh_patterns[speed][i].interval;
+        good_quality_mesh_patterns[mesh_speed][i].interval;
   }
   if ((frame_is_intra_only(cm) && cm->allow_screen_content_tools) &&
       (cpi->twopass.fr_content_type == FC_GRAPHICS_ANIMATION ||
        cpi->oxcf.content == AOM_CONTENT_SCREEN)) {
     for (i = 0; i < MAX_MESH_STEP; ++i) {
-      sf->mesh_patterns[i].range = intrabc_mesh_patterns[speed][i].range;
-      sf->mesh_patterns[i].interval = intrabc_mesh_patterns[speed][i].interval;
+      sf->mesh_patterns[i].range = intrabc_mesh_patterns[mesh_speed][i].range;
+      sf->mesh_patterns[i].interval =
+          intrabc_mesh_patterns[mesh_speed][i].interval;
     }
-    sf->max_exaustive_pct = intrabc_max_mesh_pct[speed];
+    sf->max_exaustive_pct = intrabc_max_mesh_pct[mesh_speed];
   }
 
   // Slow quant, dct and trellis not worthwhile for first pass
@@ -544,7 +640,7 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
   cpi->optimize_speed_feature =
       oxcf->pass != 1 ? sf->optimize_coefficients : NO_TRELLIS_OPT;
   // FIXME: trellis not very efficient for quantisation matrices
-  if (cm->using_qmatrix) cpi->optimize_speed_feature = NO_TRELLIS_OPT;
+  if (oxcf->using_qm) cpi->optimize_speed_feature = NO_TRELLIS_OPT;
   if (oxcf->disable_trellis_quant) cpi->optimize_speed_feature = NO_TRELLIS_OPT;
 
   x->min_partition_size = sf->default_min_partition_size;
@@ -555,10 +651,26 @@ void av1_set_speed_features_framesize_independent(AV1_COMP *cpi) {
     cpi->find_fractional_mv_step = av1_return_max_sub_pixel_mv;
   else if (cpi->oxcf.motion_vector_unit_test == 2)
     cpi->find_fractional_mv_step = av1_return_min_sub_pixel_mv;
+  cpi->max_comp_type_rd_threshold_mul =
+      comp_type_rd_threshold_mul[sf->prune_comp_type_by_comp_avg];
+  cpi->max_comp_type_rd_threshold_div =
+      comp_type_rd_threshold_div[sf->prune_comp_type_by_comp_avg];
+  const int tx_domain_speed = AOMMIN(speed, MAX_TX_DOMAIN_EVAL_SPEED);
+  cpi->tx_domain_dist_threshold = tx_domain_dist_thresholds[tx_domain_speed];
+
+  // assert ensures that coeff_opt_dist_thresholds is accessed correctly
+  assert(cpi->sf.perform_coeff_opt >= 0 && cpi->sf.perform_coeff_opt < 5);
+  cpi->coeff_opt_dist_threshold =
+      coeff_opt_dist_thresholds[cpi->sf.perform_coeff_opt];
 
 #if CONFIG_DIST_8X8
   if (sf->use_transform_domain_distortion > 0) cpi->oxcf.using_dist_8x8 = 0;
 
   if (cpi->oxcf.using_dist_8x8) x->min_partition_size = BLOCK_8X8;
 #endif  // CONFIG_DIST_8X8
+  if (cpi->oxcf.row_mt == 1 && (cpi->oxcf.max_threads > 1)) {
+    sf->adaptive_rd_thresh = 0;
+    if (sf->inter_mode_rd_model_estimation == 1)
+      sf->inter_mode_rd_model_estimation = 0;
+  }
 }
diff --git a/third_party/aom/av1/encoder/speed_features.h b/third_party/aom/av1/encoder/speed_features.h
index 41013b2e79..2216c7be04 100644
--- a/third_party/aom/av1/encoder/speed_features.h
+++ b/third_party/aom/av1/encoder/speed_features.h
@@ -73,7 +73,7 @@ enum {
                               (1 << THR_ALTR) | (1 << THR_GOLD)
 };
 
-typedef enum {
+enum {
   TXFM_CODING_SF = 1,
   INTER_PRED_SF = 2,
   INTRA_PRED_SF = 4,
@@ -82,9 +82,9 @@ typedef enum {
   RD_SKIP_SF = 32,
   RESERVE_2_SF = 64,
   RESERVE_3_SF = 128,
-} DEV_SPEED_FEATURES;
+} UENUM1BYTE(DEV_SPEED_FEATURES);
 
-typedef enum {
+enum {
   DIAMOND = 0,
   NSTEP = 1,
   HEX = 2,
@@ -92,9 +92,9 @@ typedef enum {
   SQUARE = 4,
   FAST_HEX = 5,
   FAST_DIAMOND = 6
-} SEARCH_METHODS;
+} UENUM1BYTE(SEARCH_METHODS);
 
-typedef enum {
+enum {
   // No recode.
   DISALLOW_RECODE = 0,
   // Allow recode for KF and exceeding maximum frame bandwidth.
@@ -103,28 +103,28 @@ typedef enum {
   ALLOW_RECODE_KFARFGF = 2,
   // Allow recode for all frames based on bitrate constraints.
   ALLOW_RECODE = 3,
-} RECODE_LOOP_TYPE;
+} UENUM1BYTE(RECODE_LOOP_TYPE);
 
-typedef enum {
+enum {
   SUBPEL_TREE = 0,
   SUBPEL_TREE_PRUNED = 1,           // Prunes 1/2-pel searches
   SUBPEL_TREE_PRUNED_MORE = 2,      // Prunes 1/2-pel searches more aggressively
   SUBPEL_TREE_PRUNED_EVENMORE = 3,  // Prunes 1/2- and 1/4-pel searches
   // Other methods to come
-} SUBPEL_SEARCH_METHODS;
+} UENUM1BYTE(SUBPEL_SEARCH_METHODS);
 
-typedef enum {
+enum {
   USE_FULL_RD = 0,
   USE_FAST_RD,
   USE_LARGESTALL,
-} TX_SIZE_SEARCH_METHOD;
+} UENUM1BYTE(TX_SIZE_SEARCH_METHOD);
 
-typedef enum {
+enum {
   NOT_IN_USE = 0,
   RELAXED_NEIGHBORING_MIN_MAX = 1
-} AUTO_MIN_MAX_MODE;
+} UENUM1BYTE(AUTO_MIN_MAX_MODE);
 
-typedef enum {
+enum {
   // Try the full image with different values.
   LPF_PICK_FROM_FULL_IMAGE,
   // Try a small portion of the image with different values.
@@ -133,9 +133,9 @@ typedef enum {
   LPF_PICK_FROM_Q,
   // Pick 0 to disable LPF if LPF was enabled last frame
   LPF_PICK_MINIMAL_LPF
-} LPF_PICK_METHOD;
+} UENUM1BYTE(LPF_PICK_METHOD);
 
-typedef enum {
+enum {
   // Terminate search early based on distortion so far compared to
   // qp step, distortion in the neighborhood of the frame, etc.
   FLAG_EARLY_TERMINATE = 1 << 0,
@@ -152,9 +152,9 @@ typedef enum {
 
   // Skips intra modes other than DC_PRED if the source variance is small
   FLAG_SKIP_INTRA_LOWVAR = 1 << 5,
-} MODE_SEARCH_SKIP_LOGIC;
+} UENUM1BYTE(MODE_SEARCH_SKIP_LOGIC);
 
-typedef enum {
+enum {
   NO_PRUNE = 0,
   // eliminates one tx type in vertical and horizontal direction
   PRUNE_ONE = 1,
@@ -165,7 +165,7 @@ typedef enum {
   PRUNE_2D_ACCURATE = 3,
   // similar, but applies much more aggressive pruning to get better speed-up
   PRUNE_2D_FAST = 4,
-} TX_TYPE_PRUNE_MODE;
+} UENUM1BYTE(TX_TYPE_PRUNE_MODE);
 
 typedef struct {
   TX_TYPE_PRUNE_MODE prune_mode;
@@ -184,7 +184,7 @@ typedef struct {
   int skip_tx_search;
 } TX_TYPE_SEARCH;
 
-typedef enum {
+enum {
   // Search partitions using RD criterion
   SEARCH_PARTITION,
 
@@ -192,7 +192,14 @@ typedef enum {
   FIXED_PARTITION,
 
   REFERENCE_PARTITION
-} PARTITION_SEARCH_TYPE;
+} UENUM1BYTE(PARTITION_SEARCH_TYPE);
+
+enum {
+  EIGHTH_PEL,
+  QUARTER_PEL,
+  HALF_PEL,
+  FULL_PEL
+} UENUM1BYTE(SUBPEL_FORCE_STOP);
 
 typedef struct MV_SPEED_FEATURES {
   // Motion search method (Diamond, NSTEP, Hex, Big Diamond, Square, etc).
@@ -215,8 +222,8 @@ typedef struct MV_SPEED_FEATURES {
   // Maximum number of steps in logarithmic subpel search before giving up.
   int subpel_iters_per_step;
 
-  // Control when to stop subpel search
-  int subpel_force_stop;
+  // When to stop subpel search.
+  SUBPEL_FORCE_STOP subpel_force_stop;
 } MV_SPEED_FEATURES;
 
 #define MAX_MESH_STEP 4
@@ -226,29 +233,42 @@ typedef struct MESH_PATTERN {
   int interval;
 } MESH_PATTERN;
 
-typedef enum {
+enum {
   GM_FULL_SEARCH,
-  GM_REDUCED_REF_SEARCH,
+  GM_REDUCED_REF_SEARCH_SKIP_L2_L3,
+  GM_REDUCED_REF_SEARCH_SKIP_L2_L3_ARF2,
   GM_DISABLE_SEARCH
-} GM_SEARCH_TYPE;
+} UENUM1BYTE(GM_SEARCH_TYPE);
 
-typedef enum {
+enum {
   GM_ERRORADV_TR_0,
   GM_ERRORADV_TR_1,
   GM_ERRORADV_TR_2,
   GM_ERRORADV_TR_TYPES,
-} GM_ERRORADV_TYPE;
+} UENUM1BYTE(GM_ERRORADV_TYPE);
 
-typedef enum {
+enum {
   NO_TRELLIS_OPT,         // No trellis optimization
   FULL_TRELLIS_OPT,       // Trellis optimization in all stages
   FINAL_PASS_TRELLIS_OPT  // Trellis optimization in only the final encode pass
-} TRELLIS_OPT_TYPE;
+} UENUM1BYTE(TRELLIS_OPT_TYPE);
 
-typedef enum {
+enum {
   FULL_TXFM_RD,
   LOW_TXFM_RD,
-} TXFM_RD_MODEL;
+} UENUM1BYTE(TXFM_RD_MODEL);
+
+enum {
+  DIST_WTD_COMP_ENABLED,
+  DIST_WTD_COMP_SKIP_MV_SEARCH,
+  DIST_WTD_COMP_DISABLED,
+} UENUM1BYTE(DIST_WTD_COMP_FLAG);
+
+typedef enum {
+  FLAG_SKIP_EIGHTTAP = 1 << EIGHTTAP_REGULAR,
+  FLAG_SKIP_EIGHTTAP_SMOOTH = 1 << EIGHTTAP_SMOOTH,
+  FLAG_SKIP_EIGHTTAP_SHARP = 1 << MULTITAP_SHARP,
+} INTERP_FILTER_MASK;
 
 typedef struct SPEED_FEATURES {
   MV_SPEED_FEATURES mv;
@@ -329,11 +349,16 @@ typedef struct SPEED_FEATURES {
   // 1: use model based rd breakout
   int model_based_post_interp_filter_breakout;
 
+  // Model based breakout in motion_mode_rd
+  // 0: no breakout
+  // 1: use model based rd breakout
+  int model_based_motion_mode_rd_breakout;
+
   // Used if partition_search_type = FIXED_SIZE_PARTITION
   BLOCK_SIZE always_this_block_size;
 
   // Drop less likely to be picked reference frames in the RD search.
-  // Has three levels for now: 0, 1 and 2, where higher levels prune more
+  // Has five levels for now: 0, 1, 2, 3 and 4, where higher levels prune more
   // aggressively than lower ones. (0 means no pruning).
   int selective_ref_frame;
 
@@ -371,6 +396,9 @@ typedef struct SPEED_FEATURES {
   // Prune reference frames for rectangular partitions.
   int prune_ref_frame_for_rect_partitions;
 
+  // Prune ref/mode choices for partitions.
+  int prune_ref_mode_for_partitions;
+
   // Sets min and max partition sizes for this superblock based on the
   // same superblock in last encoded frame, and the left and above neighbor.
   AUTO_MIN_MAX_MODE auto_min_max_partition_size;
@@ -431,12 +459,19 @@ typedef struct SPEED_FEATURES {
   // Choose a very large value (UINT_MAX) to use 8-tap always
   unsigned int disable_filter_search_var_thresh;
 
-  // A source variance threshold below which wedge search is disabled
+  // Only enable wedge search if the edge strength is greater than
+  // this threshold. A value of 0 signals that this check is disabled.
+  unsigned int disable_wedge_search_edge_thresh;
+
+  // Only enable wedge search if the variance is above this threshold.
   unsigned int disable_wedge_search_var_thresh;
 
   // Whether fast wedge sign estimate is used
   int fast_wedge_sign_estimate;
 
+  // Whether to prune wedge search based on predictor difference
+  int prune_wedge_pred_diff_based;
+
   // These bit masks allow you to enable or disable intra modes for each
   // transform size separately.
   int intra_y_mode_mask[TX_SIZES];
@@ -473,7 +508,7 @@ typedef struct SPEED_FEATURES {
 
   // If true, sub-pixel search uses the exact convolve function used for final
   // encoding and decoding; otherwise, it uses bilinear interpolation.
-  int use_accurate_subpel_search;
+  SUBPEL_SEARCH_TYPE use_accurate_subpel_search;
 
   // Whether to compute distortion in the image domain (slower but
   // more accurate), or in the transform domain (faster but less acurate).
@@ -492,6 +527,9 @@ typedef struct SPEED_FEATURES {
   // usually includes EIGHTTAP_REGULAR.
   int use_fast_interpolation_filter_search;
 
+  // Disable dual filter
+  int disable_dual_filter;
+
   // Save results of interpolation_filter_search for a block
   // Check mv and ref_frames before search, if they are same with previous
   // saved results, it can be skipped.
@@ -522,18 +560,23 @@ typedef struct SPEED_FEATURES {
   // Calculate RD cost before doing optimize_b, and skip if the cost is large.
   int optimize_b_precheck;
 
-  // Use model rd instead of transform search in jnt_comp
-  int jnt_comp_fast_tx_search;
+  // Use model rd instead of transform search in dist_wtd_comp
+  int dist_wtd_comp_fast_tx_search;
 
-  // Skip mv search in jnt_comp
-  int jnt_comp_skip_mv_search;
+  // Decide when and how to use joint_comp.
+  DIST_WTD_COMP_FLAG use_dist_wtd_comp_flag;
 
   // Decoder side speed feature to add penalty for use of dual-sgr filters.
   // Takes values 0 - 10, 0 indicating no penalty and each additional level
   // adding a penalty of 1%
   int dual_sgr_penalty_level;
 
-  // Dynamically estimate final rd from prediction error and mode cost
+  // 2-pass inter mode model estimation where the preliminary pass skips
+  // transform search and uses a model to estimate rd, while the final pass
+  // computes the full transform search. two types of models are supported:
+  // 0: not used
+  // 1: used with online dynamic rd model
+  // 2: used with static rd model
   int inter_mode_rd_model_estimation;
 
   // Skip some ref frames in compound motion search by single motion search
@@ -543,6 +586,10 @@ typedef struct SPEED_FEATURES {
   // single inter mode as a group.
   int prune_comp_search_by_single_result;
 
+  // Skip certain motion modes (OBMC, warped, interintra) for single reference
+  // motion search, using the results of single ref SIMPLE_TRANSLATION
+  int prune_single_motion_modes_by_simple_trans;
+
   // Reuse the inter_intra_mode search result from NEARESTMV mode to other
   // single ref modes
   int reuse_inter_intra_mode;
@@ -554,12 +601,65 @@ typedef struct SPEED_FEATURES {
 
   // flag to skip NEWMV mode in drl if the motion search result is the same
   int skip_repeated_newmv;
+
+  // Prune intra mode candidates based on source block gradient stats.
+  int intra_angle_estimation;
+
+  // Skip obmc or warped motion mode when neighborhood motion field is
+  // identical
+  int skip_obmc_in_uniform_mv_field;
+  int skip_wm_in_uniform_mv_field;
+
+  // Enable/disable ME for interinter wedge search.
+  int disable_interinter_wedge_newmv_search;
+
+  // skip sharp_filter evaluation based on regular and smooth filter rd for
+  // dual_filter=0 case
+  int skip_sharp_interp_filter_search;
+
+  // prune wedge and compound segment approximate rd evaluation based on
+  // compound average rd/ref_best_rd
+  int prune_comp_type_by_comp_avg;
+
+  // Prune/gate motion mode evaluation based on token based rd
+  // during transform search for inter blocks
+  // Values are 0 (not used) , 1 - 3 with progressively increasing
+  // aggressiveness
+  int prune_motion_mode_level;
+
+  // Gate warp evaluation for motions of type IDENTITY,
+  // TRANSLATION and AFFINE(based on number of warp neighbors)
+  int prune_warp_using_wmtype;
+
+  // Perform simple_motion_search on each possible subblock and use it to prune
+  // PARTITION_HORZ and PARTITION_VERT.
+  int simple_motion_search_prune_rect;
+
+  // Perform simple motion search before none_partition to decide if we
+  // want to split directly without trying other partition types.
+  int simple_motion_search_split_only;
+
+  int cb_pred_filter_search;
+
+  // adaptive interp_filter search to allow skip of certain filter types.
+  int adaptive_interp_filter_search;
+
+  // mask for skip evaluation of certain interp_filter type.
+  INTERP_FILTER_MASK interp_filter_search_mask;
+
+  // Enable/disable interintra wedge search.
+  int disable_wedge_interintra_search;
+
+  // Flag used to control the extent of coeff R-D optimization
+  int perform_coeff_opt;
 } SPEED_FEATURES;
 
 struct AV1_COMP;
 
-void av1_set_speed_features_framesize_independent(struct AV1_COMP *cpi);
-void av1_set_speed_features_framesize_dependent(struct AV1_COMP *cpi);
+void av1_set_speed_features_framesize_independent(struct AV1_COMP *cpi,
+                                                  int speed);
+void av1_set_speed_features_framesize_dependent(struct AV1_COMP *cpi,
+                                                int speed);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/av1/encoder/temporal_filter.c b/third_party/aom/av1/encoder/temporal_filter.c
index 75fdf02a52..385f027932 100644
--- a/third_party/aom/av1/encoder/temporal_filter.c
+++ b/third_party/aom/av1/encoder/temporal_filter.c
@@ -34,49 +34,466 @@
 #include "aom_ports/aom_timer.h"
 #include "aom_scale/aom_scale.h"
 
+#define EDGE_THRESHOLD 50
+#define SQRT_PI_BY_2 1.25331413732
+
 static void temporal_filter_predictors_mb_c(
     MACROBLOCKD *xd, uint8_t *y_mb_ptr, uint8_t *u_mb_ptr, uint8_t *v_mb_ptr,
     int stride, int uv_block_width, int uv_block_height, int mv_row, int mv_col,
     uint8_t *pred, struct scale_factors *scale, int x, int y,
-    int can_use_previous, int num_planes) {
-  const MV mv = { mv_row, mv_col };
-  enum mv_precision mv_precision_uv;
+    int can_use_previous, int num_planes, MV *blk_mvs, int use_32x32) {
+  mv_precision mv_precision_uv;
   int uv_stride;
   // TODO(angiebird): change plane setting accordingly
   ConvolveParams conv_params = get_conv_params(0, 0, xd->bd);
-  const InterpFilters interp_filters = xd->mi[0]->interp_filters;
+  const InterpFilters interp_filters =
+      av1_make_interp_filters(MULTITAP_SHARP, MULTITAP_SHARP);
   WarpTypesAllowed warp_types;
   memset(&warp_types, 0, sizeof(WarpTypesAllowed));
 
-  if (uv_block_width == 8) {
+  const int ssx = (uv_block_width == (BW >> 1)) ? 1 : 0;
+  if (ssx) {
     uv_stride = (stride + 1) >> 1;
     mv_precision_uv = MV_PRECISION_Q4;
   } else {
     uv_stride = stride;
     mv_precision_uv = MV_PRECISION_Q3;
   }
-  av1_build_inter_predictor(y_mb_ptr, stride, &pred[0], 16, &mv, scale, 16, 16,
-                            &conv_params, interp_filters, &warp_types, x, y, 0,
-                            0, MV_PRECISION_Q3, x, y, xd, can_use_previous);
 
+  if (use_32x32) {
+    const MV mv = { mv_row, mv_col };
+
+    av1_build_inter_predictor(y_mb_ptr, stride, &pred[0], BW, &mv, scale, BW,
+                              BH, &conv_params, interp_filters, &warp_types, x,
+                              y, 0, 0, MV_PRECISION_Q3, x, y, xd,
+                              can_use_previous);
+    if (num_planes > 1) {
+      av1_build_inter_predictor(
+          u_mb_ptr, uv_stride, &pred[BLK_PELS], uv_block_width, &mv, scale,
+          uv_block_width, uv_block_height, &conv_params, interp_filters,
+          &warp_types, x, y, 1, 0, mv_precision_uv, x, y, xd, can_use_previous);
+      av1_build_inter_predictor(
+          v_mb_ptr, uv_stride, &pred[(BLK_PELS << 1)], uv_block_width, &mv,
+          scale, uv_block_width, uv_block_height, &conv_params, interp_filters,
+          &warp_types, x, y, 2, 0, mv_precision_uv, x, y, xd, can_use_previous);
+    }
+
+    return;
+  }
+
+  // While use_32x32 = 0, construct the 32x32 predictor using 4 16x16
+  // predictors.
+  int i, j, k = 0, ys = (BH >> 1), xs = (BW >> 1);
+  // Y predictor
+  for (i = 0; i < BH; i += ys) {
+    for (j = 0; j < BW; j += xs) {
+      const MV mv = blk_mvs[k];
+      const int y_offset = i * stride + j;
+      const int p_offset = i * BW + j;
+
+      av1_build_inter_predictor(y_mb_ptr + y_offset, stride, &pred[p_offset],
+                                BW, &mv, scale, xs, ys, &conv_params,
+                                interp_filters, &warp_types, x, y, 0, 0,
+                                MV_PRECISION_Q3, x, y, xd, can_use_previous);
+      k++;
+    }
+  }
+
+  // U and V predictors
   if (num_planes > 1) {
-    av1_build_inter_predictor(
-        u_mb_ptr, uv_stride, &pred[256], uv_block_width, &mv, scale,
-        uv_block_width, uv_block_height, &conv_params, interp_filters,
-        &warp_types, x, y, 1, 0, mv_precision_uv, x, y, xd, can_use_previous);
+    ys = (uv_block_height >> 1);
+    xs = (uv_block_width >> 1);
+    k = 0;
 
-    av1_build_inter_predictor(
-        v_mb_ptr, uv_stride, &pred[512], uv_block_width, &mv, scale,
-        uv_block_width, uv_block_height, &conv_params, interp_filters,
-        &warp_types, x, y, 2, 0, mv_precision_uv, x, y, xd, can_use_previous);
+    for (i = 0; i < uv_block_height; i += ys) {
+      for (j = 0; j < uv_block_width; j += xs) {
+        const MV mv = blk_mvs[k];
+        const int uv_offset = i * uv_stride + j;
+        const int p_offset = i * uv_block_width + j;
+
+        av1_build_inter_predictor(u_mb_ptr + uv_offset, uv_stride,
+                                  &pred[BLK_PELS + p_offset], uv_block_width,
+                                  &mv, scale, xs, ys, &conv_params,
+                                  interp_filters, &warp_types, x, y, 1, 0,
+                                  mv_precision_uv, x, y, xd, can_use_previous);
+        av1_build_inter_predictor(
+            v_mb_ptr + uv_offset, uv_stride, &pred[(BLK_PELS << 1) + p_offset],
+            uv_block_width, &mv, scale, xs, ys, &conv_params, interp_filters,
+            &warp_types, x, y, 2, 0, mv_precision_uv, x, y, xd,
+            can_use_previous);
+        k++;
+      }
+    }
   }
 }
 
+static void apply_temporal_filter_self(const uint8_t *pred, int buf_stride,
+                                       unsigned int block_width,
+                                       unsigned int block_height,
+                                       int filter_weight, uint32_t *accumulator,
+                                       uint16_t *count) {
+  const int modifier = filter_weight * 16;
+  unsigned int i, j, k = 0;
+  assert(filter_weight == 2);
+
+  for (i = 0; i < block_height; i++) {
+    for (j = 0; j < block_width; j++) {
+      const int pixel_value = pred[i * buf_stride + j];
+      count[k] += modifier;
+      accumulator[k] += modifier * pixel_value;
+      ++k;
+    }
+  }
+}
+
+static void highbd_apply_temporal_filter_self(
+    const uint8_t *pred8, int buf_stride, unsigned int block_width,
+    unsigned int block_height, int filter_weight, uint32_t *accumulator,
+    uint16_t *count) {
+  const int modifier = filter_weight * 16;
+  const uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  unsigned int i, j, k = 0;
+  assert(filter_weight == 2);
+
+  for (i = 0; i < block_height; i++) {
+    for (j = 0; j < block_width; j++) {
+      const int pixel_value = pred[i * buf_stride + j];
+      count[k] += modifier;
+      accumulator[k] += modifier * pixel_value;
+      ++k;
+    }
+  }
+}
+
+static INLINE int mod_index(int64_t sum_dist, int index, int rounding,
+                            int strength, int filter_weight) {
+  int mod = (int)(((sum_dist * 3) / index + rounding) >> strength);
+  mod = AOMMIN(16, mod);
+  mod = 16 - mod;
+  mod *= filter_weight;
+  return mod;
+}
+
+static INLINE void calculate_squared_errors(const uint8_t *s, int s_stride,
+                                            const uint8_t *p, int p_stride,
+                                            uint16_t *diff_sse, unsigned int w,
+                                            unsigned int h) {
+  int idx = 0;
+  unsigned int i, j;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int16_t diff = s[i * s_stride + j] - p[i * p_stride + j];
+      diff_sse[idx] = diff * diff;
+      idx++;
+    }
+  }
+}
+
+static INLINE int get_filter_weight(unsigned int i, unsigned int j,
+                                    unsigned int block_height,
+                                    unsigned int block_width, int *blk_fw,
+                                    int use_32x32) {
+  if (use_32x32)
+    // blk_fw[0] ~ blk_fw[3] are the same.
+    return blk_fw[0];
+
+  int filter_weight = 0;
+  if (i < block_height / 2) {
+    if (j < block_width / 2)
+      filter_weight = blk_fw[0];
+    else
+      filter_weight = blk_fw[1];
+  } else {
+    if (j < block_width / 2)
+      filter_weight = blk_fw[2];
+    else
+      filter_weight = blk_fw[3];
+  }
+  return filter_weight;
+}
+
+static void apply_temporal_filter(
+    const uint8_t *y_frame1, int y_stride, const uint8_t *y_pred,
+    int y_buf_stride, const uint8_t *u_frame1, const uint8_t *v_frame1,
+    int uv_stride, const uint8_t *u_pred, const uint8_t *v_pred,
+    int uv_buf_stride, unsigned int block_width, unsigned int block_height,
+    int ss_x, int ss_y, int strength, int *blk_fw, int use_32x32,
+    uint32_t *y_accumulator, uint16_t *y_count, uint32_t *u_accumulator,
+    uint16_t *u_count, uint32_t *v_accumulator, uint16_t *v_count) {
+  unsigned int i, j, k, m;
+  int modifier;
+  const int rounding = (1 << strength) >> 1;
+  const unsigned int uv_block_width = block_width >> ss_x;
+  const unsigned int uv_block_height = block_height >> ss_y;
+  DECLARE_ALIGNED(16, uint16_t, y_diff_sse[BLK_PELS]);
+  DECLARE_ALIGNED(16, uint16_t, u_diff_sse[BLK_PELS]);
+  DECLARE_ALIGNED(16, uint16_t, v_diff_sse[BLK_PELS]);
+
+  int idx = 0, idy;
+
+  memset(y_diff_sse, 0, BLK_PELS * sizeof(uint16_t));
+  memset(u_diff_sse, 0, BLK_PELS * sizeof(uint16_t));
+  memset(v_diff_sse, 0, BLK_PELS * sizeof(uint16_t));
+
+  // Calculate diff^2 for each pixel of the block.
+  // TODO(yunqing): the following code needs to be optimized.
+  calculate_squared_errors(y_frame1, y_stride, y_pred, y_buf_stride, y_diff_sse,
+                           block_width, block_height);
+  calculate_squared_errors(u_frame1, uv_stride, u_pred, uv_buf_stride,
+                           u_diff_sse, uv_block_width, uv_block_height);
+  calculate_squared_errors(v_frame1, uv_stride, v_pred, uv_buf_stride,
+                           v_diff_sse, uv_block_width, uv_block_height);
+
+  for (i = 0, k = 0, m = 0; i < block_height; i++) {
+    for (j = 0; j < block_width; j++) {
+      const int pixel_value = y_pred[i * y_buf_stride + j];
+      int filter_weight =
+          get_filter_weight(i, j, block_height, block_width, blk_fw, use_32x32);
+
+      // non-local mean approach
+      int y_index = 0;
+
+      const int uv_r = i >> ss_y;
+      const int uv_c = j >> ss_x;
+      modifier = 0;
+
+      for (idy = -1; idy <= 1; ++idy) {
+        for (idx = -1; idx <= 1; ++idx) {
+          const int row = (int)i + idy;
+          const int col = (int)j + idx;
+
+          if (row >= 0 && row < (int)block_height && col >= 0 &&
+              col < (int)block_width) {
+            modifier += y_diff_sse[row * (int)block_width + col];
+            ++y_index;
+          }
+        }
+      }
+
+      assert(y_index > 0);
+
+      modifier += u_diff_sse[uv_r * uv_block_width + uv_c];
+      modifier += v_diff_sse[uv_r * uv_block_width + uv_c];
+
+      y_index += 2;
+
+      modifier =
+          (int)mod_index(modifier, y_index, rounding, strength, filter_weight);
+
+      y_count[k] += modifier;
+      y_accumulator[k] += modifier * pixel_value;
+
+      ++k;
+
+      // Process chroma component
+      if (!(i & ss_y) && !(j & ss_x)) {
+        const int u_pixel_value = u_pred[uv_r * uv_buf_stride + uv_c];
+        const int v_pixel_value = v_pred[uv_r * uv_buf_stride + uv_c];
+
+        // non-local mean approach
+        int cr_index = 0;
+        int u_mod = 0, v_mod = 0;
+        int y_diff = 0;
+
+        for (idy = -1; idy <= 1; ++idy) {
+          for (idx = -1; idx <= 1; ++idx) {
+            const int row = uv_r + idy;
+            const int col = uv_c + idx;
+
+            if (row >= 0 && row < (int)uv_block_height && col >= 0 &&
+                col < (int)uv_block_width) {
+              u_mod += u_diff_sse[row * uv_block_width + col];
+              v_mod += v_diff_sse[row * uv_block_width + col];
+              ++cr_index;
+            }
+          }
+        }
+
+        assert(cr_index > 0);
+
+        for (idy = 0; idy < 1 + ss_y; ++idy) {
+          for (idx = 0; idx < 1 + ss_x; ++idx) {
+            const int row = (uv_r << ss_y) + idy;
+            const int col = (uv_c << ss_x) + idx;
+            y_diff += y_diff_sse[row * (int)block_width + col];
+            ++cr_index;
+          }
+        }
+
+        u_mod += y_diff;
+        v_mod += y_diff;
+
+        u_mod =
+            (int)mod_index(u_mod, cr_index, rounding, strength, filter_weight);
+        v_mod =
+            (int)mod_index(v_mod, cr_index, rounding, strength, filter_weight);
+
+        u_count[m] += u_mod;
+        u_accumulator[m] += u_mod * u_pixel_value;
+        v_count[m] += v_mod;
+        v_accumulator[m] += v_mod * v_pixel_value;
+
+        ++m;
+      }  // Complete YUV pixel
+    }
+  }
+}
+
+static INLINE void highbd_calculate_squared_errors(
+    const uint16_t *s, int s_stride, const uint16_t *p, int p_stride,
+    uint32_t *diff_sse, unsigned int w, unsigned int h) {
+  int idx = 0;
+  unsigned int i, j;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int16_t diff = s[i * s_stride + j] - p[i * p_stride + j];
+      diff_sse[idx] = diff * diff;
+      idx++;
+    }
+  }
+}
+
+static void highbd_apply_temporal_filter(
+    const uint8_t *yf, int y_stride, const uint8_t *yp, int y_buf_stride,
+    const uint8_t *uf, const uint8_t *vf, int uv_stride, const uint8_t *up,
+    const uint8_t *vp, int uv_buf_stride, unsigned int block_width,
+    unsigned int block_height, int ss_x, int ss_y, int strength, int *blk_fw,
+    int use_32x32, uint32_t *y_accumulator, uint16_t *y_count,
+    uint32_t *u_accumulator, uint16_t *u_count, uint32_t *v_accumulator,
+    uint16_t *v_count) {
+  unsigned int i, j, k, m;
+  int64_t modifier;
+  const int rounding = (1 << strength) >> 1;
+  const unsigned int uv_block_width = block_width >> ss_x;
+  const unsigned int uv_block_height = block_height >> ss_y;
+  DECLARE_ALIGNED(16, uint32_t, y_diff_sse[BLK_PELS]);
+  DECLARE_ALIGNED(16, uint32_t, u_diff_sse[BLK_PELS]);
+  DECLARE_ALIGNED(16, uint32_t, v_diff_sse[BLK_PELS]);
+
+  const uint16_t *y_frame1 = CONVERT_TO_SHORTPTR(yf);
+  const uint16_t *u_frame1 = CONVERT_TO_SHORTPTR(uf);
+  const uint16_t *v_frame1 = CONVERT_TO_SHORTPTR(vf);
+  const uint16_t *y_pred = CONVERT_TO_SHORTPTR(yp);
+  const uint16_t *u_pred = CONVERT_TO_SHORTPTR(up);
+  const uint16_t *v_pred = CONVERT_TO_SHORTPTR(vp);
+  int idx = 0, idy;
+
+  memset(y_diff_sse, 0, BLK_PELS * sizeof(uint32_t));
+  memset(u_diff_sse, 0, BLK_PELS * sizeof(uint32_t));
+  memset(v_diff_sse, 0, BLK_PELS * sizeof(uint32_t));
+
+  // Calculate diff^2 for each pixel of the block.
+  // TODO(yunqing): the following code needs to be optimized.
+  highbd_calculate_squared_errors(y_frame1, y_stride, y_pred, y_buf_stride,
+                                  y_diff_sse, block_width, block_height);
+  highbd_calculate_squared_errors(u_frame1, uv_stride, u_pred, uv_buf_stride,
+                                  u_diff_sse, uv_block_width, uv_block_height);
+  highbd_calculate_squared_errors(v_frame1, uv_stride, v_pred, uv_buf_stride,
+                                  v_diff_sse, uv_block_width, uv_block_height);
+
+  for (i = 0, k = 0, m = 0; i < block_height; i++) {
+    for (j = 0; j < block_width; j++) {
+      const int pixel_value = y_pred[i * y_buf_stride + j];
+      int filter_weight =
+          get_filter_weight(i, j, block_height, block_width, blk_fw, use_32x32);
+
+      // non-local mean approach
+      int y_index = 0;
+
+      const int uv_r = i >> ss_y;
+      const int uv_c = j >> ss_x;
+      modifier = 0;
+
+      for (idy = -1; idy <= 1; ++idy) {
+        for (idx = -1; idx <= 1; ++idx) {
+          const int row = (int)i + idy;
+          const int col = (int)j + idx;
+
+          if (row >= 0 && row < (int)block_height && col >= 0 &&
+              col < (int)block_width) {
+            modifier += y_diff_sse[row * (int)block_width + col];
+            ++y_index;
+          }
+        }
+      }
+
+      assert(y_index > 0);
+
+      modifier += u_diff_sse[uv_r * uv_block_width + uv_c];
+      modifier += v_diff_sse[uv_r * uv_block_width + uv_c];
+
+      y_index += 2;
+
+      const int final_y_mod =
+          mod_index(modifier, y_index, rounding, strength, filter_weight);
+
+      y_count[k] += final_y_mod;
+      y_accumulator[k] += final_y_mod * pixel_value;
+
+      ++k;
+
+      // Process chroma component
+      if (!(i & ss_y) && !(j & ss_x)) {
+        const int u_pixel_value = u_pred[uv_r * uv_buf_stride + uv_c];
+        const int v_pixel_value = v_pred[uv_r * uv_buf_stride + uv_c];
+
+        // non-local mean approach
+        int cr_index = 0;
+        int64_t u_mod = 0, v_mod = 0;
+        int y_diff = 0;
+
+        for (idy = -1; idy <= 1; ++idy) {
+          for (idx = -1; idx <= 1; ++idx) {
+            const int row = uv_r + idy;
+            const int col = uv_c + idx;
+
+            if (row >= 0 && row < (int)uv_block_height && col >= 0 &&
+                col < (int)uv_block_width) {
+              u_mod += u_diff_sse[row * uv_block_width + col];
+              v_mod += v_diff_sse[row * uv_block_width + col];
+              ++cr_index;
+            }
+          }
+        }
+
+        assert(cr_index > 0);
+
+        for (idy = 0; idy < 1 + ss_y; ++idy) {
+          for (idx = 0; idx < 1 + ss_x; ++idx) {
+            const int row = (uv_r << ss_y) + idy;
+            const int col = (uv_c << ss_x) + idx;
+            y_diff += y_diff_sse[row * (int)block_width + col];
+            ++cr_index;
+          }
+        }
+
+        u_mod += y_diff;
+        v_mod += y_diff;
+
+        const int final_u_mod =
+            mod_index(u_mod, cr_index, rounding, strength, filter_weight);
+        const int final_v_mod =
+            mod_index(v_mod, cr_index, rounding, strength, filter_weight);
+
+        u_count[m] += final_u_mod;
+        u_accumulator[m] += final_u_mod * u_pixel_value;
+        v_count[m] += final_v_mod;
+        v_accumulator[m] += final_v_mod * v_pixel_value;
+
+        ++m;
+      }  // Complete YUV pixel
+    }
+  }
+}
+
+// Only used in single plane case
 void av1_temporal_filter_apply_c(uint8_t *frame1, unsigned int stride,
                                  uint8_t *frame2, unsigned int block_width,
                                  unsigned int block_height, int strength,
-                                 int filter_weight, unsigned int *accumulator,
-                                 uint16_t *count) {
+                                 int *blk_fw, int use_32x32,
+                                 unsigned int *accumulator, uint16_t *count) {
   unsigned int i, j, k;
   int modifier;
   int byte = 0;
@@ -85,6 +502,8 @@ void av1_temporal_filter_apply_c(uint8_t *frame1, unsigned int stride,
   for (i = 0, k = 0; i < block_height; i++) {
     for (j = 0; j < block_width; j++, k++) {
       int pixel_value = *frame2;
+      int filter_weight =
+          get_filter_weight(i, j, block_height, block_width, blk_fw, use_32x32);
 
       // non-local mean approach
       int diff_sse[9] = { 0 };
@@ -133,10 +552,11 @@ void av1_temporal_filter_apply_c(uint8_t *frame1, unsigned int stride,
   }
 }
 
+// Only used in single plane case
 void av1_highbd_temporal_filter_apply_c(
     uint8_t *frame1_8, unsigned int stride, uint8_t *frame2_8,
     unsigned int block_width, unsigned int block_height, int strength,
-    int filter_weight, unsigned int *accumulator, uint16_t *count) {
+    int *blk_fw, int use_32x32, unsigned int *accumulator, uint16_t *count) {
   uint16_t *frame1 = CONVERT_TO_SHORTPTR(frame1_8);
   uint16_t *frame2 = CONVERT_TO_SHORTPTR(frame2_8);
   unsigned int i, j, k;
@@ -147,6 +567,8 @@ void av1_highbd_temporal_filter_apply_c(
   for (i = 0, k = 0; i < block_height; i++) {
     for (j = 0; j < block_width; j++, k++) {
       int pixel_value = *frame2;
+      int filter_weight =
+          get_filter_weight(i, j, block_height, block_width, blk_fw, use_32x32);
 
       // non-local mean approach
       int diff_sse[9] = { 0 };
@@ -198,8 +620,8 @@ void av1_highbd_temporal_filter_apply_c(
 static int temporal_filter_find_matching_mb_c(AV1_COMP *cpi,
                                               uint8_t *arf_frame_buf,
                                               uint8_t *frame_ptr_buf,
-                                              int stride, int x_pos,
-                                              int y_pos) {
+                                              int stride, int x_pos, int y_pos,
+                                              MV *blk_mvs, int *blk_bestsme) {
   MACROBLOCK *const x = &cpi->td.mb;
   MACROBLOCKD *const xd = &x->e_mbd;
   const MV_SPEED_FEATURES *const mv_sf = &cpi->sf.mv;
@@ -232,12 +654,12 @@ static int temporal_filter_find_matching_mb_c(AV1_COMP *cpi,
 
   av1_set_mv_search_range(&x->mv_limits, &best_ref_mv1);
 
-  x->mvcost = x->mv_cost_stack;
-  x->nmvjointcost = x->nmv_vec_cost;
-
-  av1_full_pixel_search(cpi, x, BLOCK_16X16, &best_ref_mv1_full, step_param,
-                        NSTEP, 1, sadpb, cond_cost_list(cpi, cost_list),
-                        &best_ref_mv1, 0, 0, x_pos, y_pos, 0);
+  // av1_full_pixel_search() parameters: best_ref_mv1_full is the start mv, and
+  // best_ref_mv1 is for mv rate calculation. The search result is stored in
+  // x->best_mv.
+  av1_full_pixel_search(cpi, x, TF_BLOCK, &best_ref_mv1_full, step_param, NSTEP,
+                        1, sadpb, cond_cost_list(cpi, cost_list), &best_ref_mv1,
+                        0, 0, x_pos, y_pos, 0);
   x->mv_limits = tmp_mv_limits;
 
   // Ignore mv costing by sending NULL pointer instead of cost array
@@ -251,19 +673,63 @@ static int temporal_filter_find_matching_mb_c(AV1_COMP *cpi,
     x->best_mv.as_mv.row *= 8;
     x->best_mv.as_mv.col *= 8;
 
-    bestsme = cpi->fn_ptr[BLOCK_16X16].vf(y + offset, y_stride, src_address,
-                                          src_stride, &sse);
-  } else {
-    bestsme = cpi->find_fractional_mv_step(
-        x, &cpi->common, 0, 0, &best_ref_mv1,
-        cpi->common.allow_high_precision_mv, x->errorperbit,
-        &cpi->fn_ptr[BLOCK_16X16], 0, mv_sf->subpel_iters_per_step,
-        cond_cost_list(cpi, cost_list), NULL, NULL, &distortion, &sse, NULL,
-        NULL, 0, 0, 0, 0, 0);
+    bestsme = cpi->fn_ptr[TF_BLOCK].vf(y + offset, y_stride, src_address,
+                                       src_stride, &sse);
+
+    x->e_mbd.mi[0]->mv[0] = x->best_mv;
+
+    // Restore input state
+    x->plane[0].src = src;
+    xd->plane[0].pre[0] = pre;
+
+    return bestsme;
   }
 
+  // find_fractional_mv_step parameters: best_ref_mv1 is for mv rate cost
+  // calculation. The start full mv and the search result are stored in
+  // x->best_mv. mi_row and mi_col are only needed for "av1_is_scaled(sf)=1"
+  // case.
+  bestsme = cpi->find_fractional_mv_step(
+      x, &cpi->common, 0, 0, &best_ref_mv1, cpi->common.allow_high_precision_mv,
+      x->errorperbit, &cpi->fn_ptr[TF_BLOCK], 0, mv_sf->subpel_iters_per_step,
+      cond_cost_list(cpi, cost_list), NULL, NULL, &distortion, &sse, NULL, NULL,
+      0, 0, BW, BH, USE_8_TAPS, 1);
+
   x->e_mbd.mi[0]->mv[0] = x->best_mv;
 
+  // DO motion search on 4 16x16 sub_blocks.
+  int i, j, k = 0;
+  best_ref_mv1.row = x->e_mbd.mi[0]->mv[0].as_mv.row;
+  best_ref_mv1.col = x->e_mbd.mi[0]->mv[0].as_mv.col;
+  best_ref_mv1_full.col = best_ref_mv1.col >> 3;
+  best_ref_mv1_full.row = best_ref_mv1.row >> 3;
+
+  for (i = 0; i < BH; i += SUB_BH) {
+    for (j = 0; j < BW; j += SUB_BW) {
+      // Setup frame pointers
+      x->plane[0].src.buf = arf_frame_buf + i * stride + j;
+      x->plane[0].src.stride = stride;
+      xd->plane[0].pre[0].buf = frame_ptr_buf + i * stride + j;
+      xd->plane[0].pre[0].stride = stride;
+
+      av1_set_mv_search_range(&x->mv_limits, &best_ref_mv1);
+      av1_full_pixel_search(
+          cpi, x, TF_SUB_BLOCK, &best_ref_mv1_full, step_param, NSTEP, 1, sadpb,
+          cond_cost_list(cpi, cost_list), &best_ref_mv1, 0, 0, x_pos, y_pos, 0);
+      x->mv_limits = tmp_mv_limits;
+
+      blk_bestsme[k] = cpi->find_fractional_mv_step(
+          x, &cpi->common, 0, 0, &best_ref_mv1,
+          cpi->common.allow_high_precision_mv, x->errorperbit,
+          &cpi->fn_ptr[TF_SUB_BLOCK], 0, mv_sf->subpel_iters_per_step,
+          cond_cost_list(cpi, cost_list), NULL, NULL, &distortion, &sse, NULL,
+          NULL, 0, 0, SUB_BW, SUB_BH, USE_8_TAPS, 1);
+
+      blk_mvs[k] = x->best_mv.as_mv;
+      k++;
+    }
+  }
+
   // Restore input state
   x->plane[0].src = src;
   xd->plane[0].pre[0] = pre;
@@ -275,37 +741,40 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
                                       YV12_BUFFER_CONFIG **frames,
                                       int frame_count, int alt_ref_index,
                                       int strength,
-                                      struct scale_factors *scale) {
+                                      struct scale_factors *ref_scale_factors) {
   const AV1_COMMON *cm = &cpi->common;
   const int num_planes = av1_num_planes(cm);
   int byte;
   int frame;
   int mb_col, mb_row;
-  unsigned int filter_weight;
-  int mb_cols = (frames[alt_ref_index]->y_crop_width + 15) >> 4;
-  int mb_rows = (frames[alt_ref_index]->y_crop_height + 15) >> 4;
+  int mb_cols = (frames[alt_ref_index]->y_crop_width + BW - 1) >> BW_LOG2;
+  int mb_rows = (frames[alt_ref_index]->y_crop_height + BH - 1) >> BH_LOG2;
   int mb_y_offset = 0;
   int mb_uv_offset = 0;
-  DECLARE_ALIGNED(16, unsigned int, accumulator[16 * 16 * 3]);
-  DECLARE_ALIGNED(16, uint16_t, count[16 * 16 * 3]);
+  DECLARE_ALIGNED(16, unsigned int, accumulator[BLK_PELS * 3]);
+  DECLARE_ALIGNED(16, uint16_t, count[BLK_PELS * 3]);
   MACROBLOCKD *mbd = &cpi->td.mb.e_mbd;
   YV12_BUFFER_CONFIG *f = frames[alt_ref_index];
   uint8_t *dst1, *dst2;
-  DECLARE_ALIGNED(32, uint16_t, predictor16[16 * 16 * 3]);
-  DECLARE_ALIGNED(32, uint8_t, predictor8[16 * 16 * 3]);
+  DECLARE_ALIGNED(32, uint16_t, predictor16[BLK_PELS * 3]);
+  DECLARE_ALIGNED(32, uint8_t, predictor8[BLK_PELS * 3]);
   uint8_t *predictor;
-  const int mb_uv_height = 16 >> mbd->plane[1].subsampling_y;
-  const int mb_uv_width = 16 >> mbd->plane[1].subsampling_x;
+  const int mb_uv_height = BH >> mbd->plane[1].subsampling_y;
+  const int mb_uv_width = BW >> mbd->plane[1].subsampling_x;
 
   // Save input state
   uint8_t *input_buffer[MAX_MB_PLANE];
   int i;
-  if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+  const int is_hbd = is_cur_buf_hbd(mbd);
+  if (is_hbd) {
     predictor = CONVERT_TO_BYTEPTR(predictor16);
   } else {
     predictor = predictor8;
   }
 
+  mbd->block_ref_scale_factors[0] = ref_scale_factors;
+  mbd->block_ref_scale_factors[1] = ref_scale_factors;
+
   for (i = 0; i < num_planes; i++) input_buffer[i] = mbd->plane[i].pre[0].buf;
 
   for (mb_row = 0; mb_row < mb_rows; mb_row++) {
@@ -321,101 +790,170 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
     // To keep the mv in play for both Y and UV planes the max that it
     //  can be on a border is therefore 16 - (2*AOM_INTERP_EXTEND+1).
     cpi->td.mb.mv_limits.row_min =
-        -((mb_row * 16) + (17 - 2 * AOM_INTERP_EXTEND));
+        -((mb_row * BH) + (17 - 2 * AOM_INTERP_EXTEND));
     cpi->td.mb.mv_limits.row_max =
-        ((mb_rows - 1 - mb_row) * 16) + (17 - 2 * AOM_INTERP_EXTEND);
+        ((mb_rows - 1 - mb_row) * BH) + (17 - 2 * AOM_INTERP_EXTEND);
 
     for (mb_col = 0; mb_col < mb_cols; mb_col++) {
       int j, k;
       int stride;
 
-      memset(accumulator, 0, 16 * 16 * 3 * sizeof(accumulator[0]));
-      memset(count, 0, 16 * 16 * 3 * sizeof(count[0]));
+      memset(accumulator, 0, BLK_PELS * 3 * sizeof(accumulator[0]));
+      memset(count, 0, BLK_PELS * 3 * sizeof(count[0]));
 
       cpi->td.mb.mv_limits.col_min =
-          -((mb_col * 16) + (17 - 2 * AOM_INTERP_EXTEND));
+          -((mb_col * BW) + (17 - 2 * AOM_INTERP_EXTEND));
       cpi->td.mb.mv_limits.col_max =
-          ((mb_cols - 1 - mb_col) * 16) + (17 - 2 * AOM_INTERP_EXTEND);
+          ((mb_cols - 1 - mb_col) * BW) + (17 - 2 * AOM_INTERP_EXTEND);
 
       for (frame = 0; frame < frame_count; frame++) {
-        const int thresh_low = 10000;
-        const int thresh_high = 20000;
+        // MVs for 4 16x16 sub blocks.
+        MV blk_mvs[4];
+        // Filter weights for 4 16x16 sub blocks.
+        int blk_fw[4] = { 0, 0, 0, 0 };
+        int use_32x32 = 0;
 
         if (frames[frame] == NULL) continue;
 
         mbd->mi[0]->mv[0].as_mv.row = 0;
         mbd->mi[0]->mv[0].as_mv.col = 0;
         mbd->mi[0]->motion_mode = SIMPLE_TRANSLATION;
+        blk_mvs[0] = kZeroMv;
+        blk_mvs[1] = kZeroMv;
+        blk_mvs[2] = kZeroMv;
+        blk_mvs[3] = kZeroMv;
 
         if (frame == alt_ref_index) {
-          filter_weight = 2;
+          blk_fw[0] = blk_fw[1] = blk_fw[2] = blk_fw[3] = 2;
+          use_32x32 = 1;
         } else {
+          int thresh_low = 10000;
+          int thresh_high = 20000;
+          int blk_bestsme[4] = { INT_MAX, INT_MAX, INT_MAX, INT_MAX };
+
           // Find best match in this frame by MC
           int err = temporal_filter_find_matching_mb_c(
               cpi, frames[alt_ref_index]->y_buffer + mb_y_offset,
               frames[frame]->y_buffer + mb_y_offset, frames[frame]->y_stride,
-              mb_col * 16, mb_row * 16);
+              mb_col * BW, mb_row * BH, blk_mvs, blk_bestsme);
 
-          // Assign higher weight to matching MB if it's error
-          // score is lower. If not applying MC default behavior
-          // is to weight all MBs equal.
-          filter_weight = err < thresh_low ? 2 : err < thresh_high ? 1 : 0;
+          int err16 =
+              blk_bestsme[0] + blk_bestsme[1] + blk_bestsme[2] + blk_bestsme[3];
+          int max_err = INT_MIN, min_err = INT_MAX;
+          for (k = 0; k < 4; k++) {
+            if (min_err > blk_bestsme[k]) min_err = blk_bestsme[k];
+            if (max_err < blk_bestsme[k]) max_err = blk_bestsme[k];
+          }
+
+          if (((err * 15 < (err16 << 4)) && max_err - min_err < 12000) ||
+              ((err * 14 < (err16 << 4)) && max_err - min_err < 6000)) {
+            use_32x32 = 1;
+            // Assign higher weight to matching MB if it's error
+            // score is lower. If not applying MC default behavior
+            // is to weight all MBs equal.
+            blk_fw[0] = err < (thresh_low << THR_SHIFT)
+                            ? 2
+                            : err < (thresh_high << THR_SHIFT) ? 1 : 0;
+            blk_fw[1] = blk_fw[2] = blk_fw[3] = blk_fw[0];
+          } else {
+            use_32x32 = 0;
+            for (k = 0; k < 4; k++)
+              blk_fw[k] = blk_bestsme[k] < thresh_low
+                              ? 2
+                              : blk_bestsme[k] < thresh_high ? 1 : 0;
+          }
         }
 
-        if (filter_weight != 0) {
+        if (blk_fw[0] || blk_fw[1] || blk_fw[2] || blk_fw[3]) {
           // Construct the predictors
           temporal_filter_predictors_mb_c(
               mbd, frames[frame]->y_buffer + mb_y_offset,
               frames[frame]->u_buffer + mb_uv_offset,
               frames[frame]->v_buffer + mb_uv_offset, frames[frame]->y_stride,
               mb_uv_width, mb_uv_height, mbd->mi[0]->mv[0].as_mv.row,
-              mbd->mi[0]->mv[0].as_mv.col, predictor, scale, mb_col * 16,
-              mb_row * 16, cm->allow_warped_motion, num_planes);
+              mbd->mi[0]->mv[0].as_mv.col, predictor, ref_scale_factors,
+              mb_col * BW, mb_row * BH, cm->allow_warped_motion, num_planes,
+              blk_mvs, use_32x32);
 
           // Apply the filter (YUV)
-          if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
-            int adj_strength = strength + 2 * (mbd->bd - 8);
-            av1_highbd_temporal_filter_apply(
-                f->y_buffer + mb_y_offset, f->y_stride, predictor, 16, 16,
-                adj_strength, filter_weight, accumulator, count);
-            if (num_planes > 1) {
-              av1_highbd_temporal_filter_apply(
-                  f->u_buffer + mb_uv_offset, f->uv_stride, predictor + 256,
-                  mb_uv_width, mb_uv_height, adj_strength, filter_weight,
-                  accumulator + 256, count + 256);
-              av1_highbd_temporal_filter_apply(
-                  f->v_buffer + mb_uv_offset, f->uv_stride, predictor + 512,
-                  mb_uv_width, mb_uv_height, adj_strength, filter_weight,
-                  accumulator + 512, count + 512);
+          if (frame == alt_ref_index) {
+            uint8_t *pred = predictor;
+            uint32_t *accum = accumulator;
+            uint16_t *cnt = count;
+            int plane;
+
+            // All 4 blk_fws are equal to 2.
+            for (plane = 0; plane < num_planes; ++plane) {
+              const int pred_stride = plane ? mb_uv_width : BW;
+              const unsigned int w = plane ? mb_uv_width : BW;
+              const unsigned int h = plane ? mb_uv_height : BH;
+
+              if (is_hbd) {
+                highbd_apply_temporal_filter_self(pred, pred_stride, w, h,
+                                                  blk_fw[0], accum, cnt);
+              } else {
+                apply_temporal_filter_self(pred, pred_stride, w, h, blk_fw[0],
+                                           accum, cnt);
+              }
+
+              pred += BLK_PELS;
+              accum += BLK_PELS;
+              cnt += BLK_PELS;
             }
           } else {
-            av1_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
-                                        predictor, 16, 16, strength,
-                                        filter_weight, accumulator, count);
-            if (num_planes > 1) {
-              av1_temporal_filter_apply_c(
-                  f->u_buffer + mb_uv_offset, f->uv_stride, predictor + 256,
-                  mb_uv_width, mb_uv_height, strength, filter_weight,
-                  accumulator + 256, count + 256);
-              av1_temporal_filter_apply_c(
-                  f->v_buffer + mb_uv_offset, f->uv_stride, predictor + 512,
-                  mb_uv_width, mb_uv_height, strength, filter_weight,
-                  accumulator + 512, count + 512);
+            if (is_hbd) {
+              const int adj_strength = strength + 2 * (mbd->bd - 8);
+
+              if (num_planes <= 1) {
+                // Single plane case
+                av1_highbd_temporal_filter_apply_c(
+                    f->y_buffer + mb_y_offset, f->y_stride, predictor, BW, BH,
+                    adj_strength, blk_fw, use_32x32, accumulator, count);
+              } else {
+                // Process 3 planes together.
+                highbd_apply_temporal_filter(
+                    f->y_buffer + mb_y_offset, f->y_stride, predictor, BW,
+                    f->u_buffer + mb_uv_offset, f->v_buffer + mb_uv_offset,
+                    f->uv_stride, predictor + BLK_PELS,
+                    predictor + (BLK_PELS << 1), mb_uv_width, BW, BH,
+                    mbd->plane[1].subsampling_x, mbd->plane[1].subsampling_y,
+                    adj_strength, blk_fw, use_32x32, accumulator, count,
+                    accumulator + BLK_PELS, count + BLK_PELS,
+                    accumulator + (BLK_PELS << 1), count + (BLK_PELS << 1));
+              }
+            } else {
+              if (num_planes <= 1) {
+                // Single plane case
+                av1_temporal_filter_apply_c(
+                    f->y_buffer + mb_y_offset, f->y_stride, predictor, BW, BH,
+                    strength, blk_fw, use_32x32, accumulator, count);
+              } else {
+                // Process 3 planes together.
+                apply_temporal_filter(
+                    f->y_buffer + mb_y_offset, f->y_stride, predictor, BW,
+                    f->u_buffer + mb_uv_offset, f->v_buffer + mb_uv_offset,
+                    f->uv_stride, predictor + BLK_PELS,
+                    predictor + (BLK_PELS << 1), mb_uv_width, BW, BH,
+                    mbd->plane[1].subsampling_x, mbd->plane[1].subsampling_y,
+                    strength, blk_fw, use_32x32, accumulator, count,
+                    accumulator + BLK_PELS, count + BLK_PELS,
+                    accumulator + (BLK_PELS << 1), count + (BLK_PELS << 1));
+              }
             }
           }
         }
       }
 
       // Normalize filter output to produce AltRef frame
-      if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+      if (is_hbd) {
         uint16_t *dst1_16;
         uint16_t *dst2_16;
         dst1 = cpi->alt_ref_buffer.y_buffer;
         dst1_16 = CONVERT_TO_SHORTPTR(dst1);
         stride = cpi->alt_ref_buffer.y_stride;
         byte = mb_y_offset;
-        for (i = 0, k = 0; i < 16; i++) {
-          for (j = 0; j < 16; j++, k++) {
+        for (i = 0, k = 0; i < BH; i++) {
+          for (j = 0; j < BW; j++, k++) {
             dst1_16[byte] =
                 (uint16_t)OD_DIVU(accumulator[k] + (count[k] >> 1), count[k]);
 
@@ -423,7 +961,7 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
             byte++;
           }
 
-          byte += stride - 16;
+          byte += stride - BW;
         }
         if (num_planes > 1) {
           dst1 = cpi->alt_ref_buffer.u_buffer;
@@ -432,9 +970,9 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
           dst2_16 = CONVERT_TO_SHORTPTR(dst2);
           stride = cpi->alt_ref_buffer.uv_stride;
           byte = mb_uv_offset;
-          for (i = 0, k = 256; i < mb_uv_height; i++) {
+          for (i = 0, k = BLK_PELS; i < mb_uv_height; i++) {
             for (j = 0; j < mb_uv_width; j++, k++) {
-              int m = k + 256;
+              int m = k + BLK_PELS;
               // U
               dst1_16[byte] =
                   (uint16_t)OD_DIVU(accumulator[k] + (count[k] >> 1), count[k]);
@@ -451,24 +989,24 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
         dst1 = cpi->alt_ref_buffer.y_buffer;
         stride = cpi->alt_ref_buffer.y_stride;
         byte = mb_y_offset;
-        for (i = 0, k = 0; i < 16; i++) {
-          for (j = 0; j < 16; j++, k++) {
+        for (i = 0, k = 0; i < BH; i++) {
+          for (j = 0; j < BW; j++, k++) {
             dst1[byte] =
                 (uint8_t)OD_DIVU(accumulator[k] + (count[k] >> 1), count[k]);
 
             // move to next pixel
             byte++;
           }
-          byte += stride - 16;
+          byte += stride - BW;
         }
         if (num_planes > 1) {
           dst1 = cpi->alt_ref_buffer.u_buffer;
           dst2 = cpi->alt_ref_buffer.v_buffer;
           stride = cpi->alt_ref_buffer.uv_stride;
           byte = mb_uv_offset;
-          for (i = 0, k = 256; i < mb_uv_height; i++) {
+          for (i = 0, k = BLK_PELS; i < mb_uv_height; i++) {
             for (j = 0; j < mb_uv_width; j++, k++) {
-              int m = k + 256;
+              int m = k + BLK_PELS;
               // U
               dst1[byte] =
                   (uint8_t)OD_DIVU(accumulator[k] + (count[k] >> 1), count[k]);
@@ -482,10 +1020,10 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
           }
         }
       }
-      mb_y_offset += 16;
+      mb_y_offset += BW;
       mb_uv_offset += mb_uv_width;
     }
-    mb_y_offset += 16 * (f->y_stride - mb_cols);
+    mb_y_offset += BH * f->y_stride - BW * mb_cols;
     mb_uv_offset += mb_uv_height * f->uv_stride - mb_uv_width * mb_cols;
   }
 
@@ -493,6 +1031,83 @@ static void temporal_filter_iterate_c(AV1_COMP *cpi,
   for (i = 0; i < num_planes; i++) mbd->plane[i].pre[0].buf = input_buffer[i];
 }
 
+// This is an adaptation of the mehtod in the following paper:
+// Shen-Chuan Tai, Shih-Ming Yang, "A fast method for image noise
+// estimation using Laplacian operator and adaptive edge detection,"
+// Proc. 3rd International Symposium on Communications, Control and
+// Signal Processing, 2008, St Julians, Malta.
+//
+// Return noise estimate, or -1.0 if there was a failure
+static double estimate_noise(const uint8_t *src, int width, int height,
+                             int stride, int edge_thresh) {
+  int64_t sum = 0;
+  int64_t num = 0;
+  for (int i = 1; i < height - 1; ++i) {
+    for (int j = 1; j < width - 1; ++j) {
+      const int k = i * stride + j;
+      // Sobel gradients
+      const int Gx = (src[k - stride - 1] - src[k - stride + 1]) +
+                     (src[k + stride - 1] - src[k + stride + 1]) +
+                     2 * (src[k - 1] - src[k + 1]);
+      const int Gy = (src[k - stride - 1] - src[k + stride - 1]) +
+                     (src[k - stride + 1] - src[k + stride + 1]) +
+                     2 * (src[k - stride] - src[k + stride]);
+      const int Ga = abs(Gx) + abs(Gy);
+      if (Ga < edge_thresh) {  // Smooth pixels
+        // Find Laplacian
+        const int v =
+            4 * src[k] -
+            2 * (src[k - 1] + src[k + 1] + src[k - stride] + src[k + stride]) +
+            (src[k - stride - 1] + src[k - stride + 1] + src[k + stride - 1] +
+             src[k + stride + 1]);
+        sum += abs(v);
+        ++num;
+      }
+    }
+  }
+  // If very few smooth pels, return -1 since the estimate is unreliable
+  if (num < 16) return -1.0;
+
+  const double sigma = (double)sum / (6 * num) * SQRT_PI_BY_2;
+  return sigma;
+}
+
+// Return noise estimate, or -1.0 if there was a failure
+static double highbd_estimate_noise(const uint8_t *src8, int width, int height,
+                                    int stride, int bd, int edge_thresh) {
+  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  int64_t sum = 0;
+  int64_t num = 0;
+  for (int i = 1; i < height - 1; ++i) {
+    for (int j = 1; j < width - 1; ++j) {
+      const int k = i * stride + j;
+      // Sobel gradients
+      const int Gx = (src[k - stride - 1] - src[k - stride + 1]) +
+                     (src[k + stride - 1] - src[k + stride + 1]) +
+                     2 * (src[k - 1] - src[k + 1]);
+      const int Gy = (src[k - stride - 1] - src[k + stride - 1]) +
+                     (src[k - stride + 1] - src[k + stride + 1]) +
+                     2 * (src[k - stride] - src[k + stride]);
+      const int Ga = ROUND_POWER_OF_TWO(abs(Gx) + abs(Gy), bd - 8);
+      if (Ga < edge_thresh) {  // Smooth pixels
+        // Find Laplacian
+        const int v =
+            4 * src[k] -
+            2 * (src[k - 1] + src[k + 1] + src[k - stride] + src[k + stride]) +
+            (src[k - stride - 1] + src[k - stride + 1] + src[k + stride - 1] +
+             src[k + stride + 1]);
+        sum += ROUND_POWER_OF_TWO(abs(v), bd - 8);
+        ++num;
+      }
+    }
+  }
+  // If very few smooth pels, return -1 since the estimate is unreliable
+  if (num < 16) return -1.0;
+
+  const double sigma = (double)sum / (6 * num) * SQRT_PI_BY_2;
+  return sigma;
+}
+
 // Apply buffer limits and context specific adjustments to arnr filter.
 static void adjust_arnr_filter(AV1_COMP *cpi, int distance, int group_boost,
                                int *arnr_frames, int *arnr_strength) {
@@ -517,16 +1132,44 @@ static void adjust_arnr_filter(AV1_COMP *cpi, int distance, int group_boost,
   frames = frames_bwd + 1 + frames_fwd;
 
   // Adjust the strength based on active max q.
-  if (cpi->common.current_video_frame > 1)
+  if (cpi->common.current_frame.frame_number > 1)
     q = ((int)av1_convert_qindex_to_q(cpi->rc.avg_frame_qindex[INTER_FRAME],
                                       cpi->common.seq_params.bit_depth));
   else
     q = ((int)av1_convert_qindex_to_q(cpi->rc.avg_frame_qindex[KEY_FRAME],
                                       cpi->common.seq_params.bit_depth));
-  if (q > 16) {
-    strength = oxcf->arnr_strength;
+  MACROBLOCKD *mbd = &cpi->td.mb.e_mbd;
+  struct lookahead_entry *buf = av1_lookahead_peek(cpi->lookahead, distance);
+  double noiselevel;
+  if (is_cur_buf_hbd(mbd)) {
+    noiselevel = highbd_estimate_noise(
+        buf->img.y_buffer, buf->img.y_crop_width, buf->img.y_crop_height,
+        buf->img.y_stride, mbd->bd, EDGE_THRESHOLD);
   } else {
-    strength = oxcf->arnr_strength - ((16 - q) / 2);
+    noiselevel = estimate_noise(buf->img.y_buffer, buf->img.y_crop_width,
+                                buf->img.y_crop_height, buf->img.y_stride,
+                                EDGE_THRESHOLD);
+  }
+  int adj_strength = oxcf->arnr_strength;
+  if (noiselevel > 0) {
+    // Get 4 integer adjustment levels in [-2, 1]
+    int noiselevel_adj;
+    if (noiselevel < 0.75)
+      noiselevel_adj = -2;
+    else if (noiselevel < 1.75)
+      noiselevel_adj = -1;
+    else if (noiselevel < 4.0)
+      noiselevel_adj = 0;
+    else
+      noiselevel_adj = 1;
+    adj_strength += noiselevel_adj;
+  }
+  // printf("[noise level: %g, strength = %d]\n", noiselevel, adj_strength);
+
+  if (q > 16) {
+    strength = adj_strength;
+  } else {
+    strength = adj_strength - ((16 - q) / 2);
     if (strength < 0) strength = 0;
   }
 
@@ -553,18 +1196,22 @@ void av1_temporal_filter(AV1_COMP *cpi, int distance) {
   int frames_to_blur_backward;
   int frames_to_blur_forward;
   struct scale_factors sf;
+
   YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS] = { NULL };
   const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
+  int rdmult = 0;
 
   // Apply context specific adjustments to the arnr filter parameters.
-  adjust_arnr_filter(cpi, distance, rc->gfu_boost, &frames_to_blur, &strength);
-  // TODO(weitinglin): Currently, we enforce the filtering strength on
-  //                   extra ARFs' to be zeros. We should investigate in which
-  //                   case it is more beneficial to use non-zero strength
-  //                   filtering.
   if (gf_group->update_type[gf_group->index] == INTNL_ARF_UPDATE) {
+    // TODO(weitinglin): Currently, we enforce the filtering strength on
+    //                   extra ARFs' to be zeros. We should investigate in which
+    //                   case it is more beneficial to use non-zero strength
+    //                   filtering.
     strength = 0;
     frames_to_blur = 1;
+  } else {
+    adjust_arnr_filter(cpi, distance, rc->gfu_boost, &frames_to_blur,
+                       &strength);
   }
 
   int which_arf = gf_group->arf_update_idx[gf_group->index];
@@ -597,6 +1244,12 @@ void av1_temporal_filter(AV1_COMP *cpi, int distance) {
         frames[0]->y_crop_width, frames[0]->y_crop_height);
   }
 
+  // Initialize errorperbit, sadperbit16 and sadperbit4.
+  rdmult = av1_compute_rd_mult_based_on_qindex(cpi, ARNR_FILT_QINDEX);
+  set_error_per_bit(&cpi->td.mb, rdmult);
+  av1_initialize_me_consts(cpi, &cpi->td.mb, ARNR_FILT_QINDEX);
+  av1_initialize_cost_tables(&cpi->common, &cpi->td.mb);
+
   temporal_filter_iterate_c(cpi, frames, frames_to_blur,
                             frames_to_blur_backward, strength, &sf);
 }
diff --git a/third_party/aom/av1/encoder/temporal_filter.h b/third_party/aom/av1/encoder/temporal_filter.h
index 2ddc68b2c9..bb26c36660 100644
--- a/third_party/aom/av1/encoder/temporal_filter.h
+++ b/third_party/aom/av1/encoder/temporal_filter.h
@@ -16,6 +16,20 @@
 extern "C" {
 #endif
 
+#define ARNR_FILT_QINDEX 128
+
+// Block size used in temporal filtering
+#define TF_BLOCK BLOCK_32X32
+#define BH 32
+#define BH_LOG2 5
+#define BW 32
+#define BW_LOG2 5
+#define BLK_PELS 1024  // Pixels in the block
+#define THR_SHIFT 2
+#define TF_SUB_BLOCK BLOCK_16X16
+#define SUB_BH 16
+#define SUB_BW 16
+
 void av1_temporal_filter(AV1_COMP *cpi, int distance);
 
 #ifdef __cplusplus
diff --git a/third_party/aom/av1/encoder/tokenize.c b/third_party/aom/av1/encoder/tokenize.c
index 16a6a9a35a..ce1a2121a4 100644
--- a/third_party/aom/av1/encoder/tokenize.c
+++ b/third_party/aom/av1/encoder/tokenize.c
@@ -29,7 +29,7 @@
 
 static int cost_and_tokenize_map(Av1ColorMapParam *param, TOKENEXTRA **t,
                                  int plane, int calc_rate, int allow_update_cdf,
-                                 FRAME_COUNTS *counts) {
+                                 FRAME_COUNTS *counts, MapCdf map_pb_cdf) {
   const uint8_t *const color_map = param->color_map;
   MapCdf map_cdf = param->map_cdf;
   ColorCost color_cost = param->color_cost;
@@ -55,7 +55,7 @@ static int cost_and_tokenize_map(Av1ColorMapParam *param, TOKENEXTRA **t,
         this_rate += (*color_cost)[palette_size_idx][color_ctx][color_new_idx];
       } else {
         (*t)->token = color_new_idx;
-        (*t)->color_map_cdf = map_cdf[palette_size_idx][color_ctx];
+        (*t)->color_map_cdf = map_pb_cdf[palette_size_idx][color_ctx];
         ++(*t);
         if (allow_update_cdf)
           update_cdf(map_cdf[palette_size_idx][color_ctx], color_new_idx, n);
@@ -107,7 +107,10 @@ int av1_cost_color_map(const MACROBLOCK *const x, int plane, BLOCK_SIZE bsize,
   assert(plane == 0 || plane == 1);
   Av1ColorMapParam color_map_params;
   get_color_map_params(x, plane, bsize, tx_size, type, &color_map_params);
-  return cost_and_tokenize_map(&color_map_params, NULL, plane, 1, 0, NULL);
+  MapCdf map_pb_cdf = plane ? x->tile_pb_ctx->palette_uv_color_index_cdf
+                            : x->tile_pb_ctx->palette_y_color_index_cdf;
+  return cost_and_tokenize_map(&color_map_params, NULL, plane, 1, 0, NULL,
+                               map_pb_cdf);
 }
 
 void av1_tokenize_color_map(const MACROBLOCK *const x, int plane,
@@ -121,13 +124,15 @@ void av1_tokenize_color_map(const MACROBLOCK *const x, int plane,
   (*t)->token = color_map_params.color_map[0];
   (*t)->color_map_cdf = NULL;
   ++(*t);
+  MapCdf map_pb_cdf = plane ? x->tile_pb_ctx->palette_uv_color_index_cdf
+                            : x->tile_pb_ctx->palette_y_color_index_cdf;
   cost_and_tokenize_map(&color_map_params, t, plane, 0, allow_update_cdf,
-                        counts);
+                        counts, map_pb_cdf);
 }
 
-void tokenize_vartx(ThreadData *td, TOKENEXTRA **t, RUN_TYPE dry_run,
-                    TX_SIZE tx_size, BLOCK_SIZE plane_bsize, int blk_row,
-                    int blk_col, int block, int plane, void *arg) {
+static void tokenize_vartx(ThreadData *td, TOKENEXTRA **t, RUN_TYPE dry_run,
+                           TX_SIZE tx_size, BLOCK_SIZE plane_bsize, int blk_row,
+                           int blk_col, int block, int plane, void *arg) {
   MACROBLOCK *const x = &td->mb;
   MACROBLOCKD *const xd = &x->e_mbd;
   MB_MODE_INFO *const mbmi = xd->mi[0];
diff --git a/third_party/aom/av1/encoder/tokenize.h b/third_party/aom/av1/encoder/tokenize.h
index 63b505f36f..c80af7bd0d 100644
--- a/third_party/aom/av1/encoder/tokenize.h
+++ b/third_party/aom/av1/encoder/tokenize.h
@@ -38,11 +38,11 @@ struct tokenize_b_args {
   uint8_t allow_update_cdf;
 };
 
-typedef enum {
+enum {
   OUTPUT_ENABLED = 0,
   DRY_RUN_NORMAL,
   DRY_RUN_COSTCOEFFS,
-} RUN_TYPE;
+} UENUM1BYTE(RUN_TYPE);
 
 // Note in all the tokenize functions rate if non NULL is incremented
 // with the coefficient token cost only if dry_run = DRY_RUN_COSTCOEFS,
diff --git a/third_party/aom/av1/encoder/tpl_model.c b/third_party/aom/av1/encoder/tpl_model.c
new file mode 100644
index 0000000000..3399116e9f
--- /dev/null
+++ b/third_party/aom/av1/encoder/tpl_model.c
@@ -0,0 +1,595 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdint.h>
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_codec.h"
+
+#include "av1/common/onyxc_int.h"
+#include "av1/common/reconintra.h"
+
+#include "av1/encoder/encoder.h"
+#include "av1/encoder/reconinter_enc.h"
+
+typedef struct GF_PICTURE {
+  YV12_BUFFER_CONFIG *frame;
+  int ref_frame[7];
+} GF_PICTURE;
+
+static void get_quantize_error(MACROBLOCK *x, int plane, tran_low_t *coeff,
+                               tran_low_t *qcoeff, tran_low_t *dqcoeff,
+                               TX_SIZE tx_size, int64_t *recon_error,
+                               int64_t *sse) {
+  const struct macroblock_plane *const p = &x->plane[plane];
+  const SCAN_ORDER *const scan_order = &av1_default_scan_orders[tx_size];
+  uint16_t eob;
+  int pix_num = 1 << num_pels_log2_lookup[txsize_to_bsize[tx_size]];
+  const int shift = tx_size == TX_32X32 ? 0 : 2;
+
+  av1_quantize_fp_32x32(coeff, pix_num, p->zbin_QTX, p->round_fp_QTX,
+                        p->quant_fp_QTX, p->quant_shift_QTX, qcoeff, dqcoeff,
+                        p->dequant_QTX, &eob, scan_order->scan,
+                        scan_order->iscan);
+
+  *recon_error = av1_block_error(coeff, dqcoeff, pix_num, sse) >> shift;
+  *recon_error = AOMMAX(*recon_error, 1);
+
+  *sse = (*sse) >> shift;
+  *sse = AOMMAX(*sse, 1);
+}
+
+static void wht_fwd_txfm(int16_t *src_diff, int bw, tran_low_t *coeff,
+                         TX_SIZE tx_size) {
+  switch (tx_size) {
+    case TX_8X8: aom_hadamard_8x8(src_diff, bw, coeff); break;
+    case TX_16X16: aom_hadamard_16x16(src_diff, bw, coeff); break;
+    case TX_32X32: aom_hadamard_32x32(src_diff, bw, coeff); break;
+    default: assert(0);
+  }
+}
+
+static uint32_t motion_compensated_prediction(AV1_COMP *cpi, ThreadData *td,
+                                              uint8_t *cur_frame_buf,
+                                              uint8_t *ref_frame_buf,
+                                              int stride, BLOCK_SIZE bsize,
+                                              int mi_row, int mi_col) {
+  AV1_COMMON *cm = &cpi->common;
+  MACROBLOCK *const x = &td->mb;
+  MACROBLOCKD *const xd = &x->e_mbd;
+  MV_SPEED_FEATURES *const mv_sf = &cpi->sf.mv;
+  const SEARCH_METHODS search_method = NSTEP;
+  int step_param;
+  int sadpb = x->sadperbit16;
+  uint32_t bestsme = UINT_MAX;
+  int distortion;
+  uint32_t sse;
+  int cost_list[5];
+  const MvLimits tmp_mv_limits = x->mv_limits;
+
+  MV best_ref_mv1 = { 0, 0 };
+  MV best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
+
+  best_ref_mv1_full.col = best_ref_mv1.col >> 3;
+  best_ref_mv1_full.row = best_ref_mv1.row >> 3;
+
+  // Setup frame pointers
+  x->plane[0].src.buf = cur_frame_buf;
+  x->plane[0].src.stride = stride;
+  xd->plane[0].pre[0].buf = ref_frame_buf;
+  xd->plane[0].pre[0].stride = stride;
+
+  step_param = mv_sf->reduce_first_step_size;
+  step_param = AOMMIN(step_param, MAX_MVSEARCH_STEPS - 2);
+
+  av1_set_mv_search_range(&x->mv_limits, &best_ref_mv1);
+
+  av1_full_pixel_search(cpi, x, bsize, &best_ref_mv1_full, step_param,
+                        search_method, 0, sadpb, cond_cost_list(cpi, cost_list),
+                        &best_ref_mv1, INT_MAX, 0, (MI_SIZE * mi_col),
+                        (MI_SIZE * mi_row), 0);
+
+  /* restore UMV window */
+  x->mv_limits = tmp_mv_limits;
+
+  const int pw = block_size_wide[bsize];
+  const int ph = block_size_high[bsize];
+  bestsme = cpi->find_fractional_mv_step(
+      x, cm, mi_row, mi_col, &best_ref_mv1, cpi->common.allow_high_precision_mv,
+      x->errorperbit, &cpi->fn_ptr[bsize], 0, mv_sf->subpel_iters_per_step,
+      cond_cost_list(cpi, cost_list), NULL, NULL, &distortion, &sse, NULL, NULL,
+      0, 0, pw, ph, 1, 1);
+
+  return bestsme;
+}
+
+static void mode_estimation(AV1_COMP *cpi, MACROBLOCK *x, MACROBLOCKD *xd,
+                            struct scale_factors *sf, GF_PICTURE *gf_picture,
+                            int frame_idx, int16_t *src_diff, tran_low_t *coeff,
+                            tran_low_t *qcoeff, tran_low_t *dqcoeff, int mi_row,
+                            int mi_col, BLOCK_SIZE bsize, TX_SIZE tx_size,
+                            YV12_BUFFER_CONFIG *ref_frame[], uint8_t *predictor,
+                            int64_t *recon_error, int64_t *sse,
+                            TplDepStats *tpl_stats) {
+  AV1_COMMON *cm = &cpi->common;
+  ThreadData *td = &cpi->td;
+
+  const int bw = 4 << mi_size_wide_log2[bsize];
+  const int bh = 4 << mi_size_high_log2[bsize];
+  const int pix_num = bw * bh;
+  int best_rf_idx = -1;
+  int_mv best_mv;
+  int64_t best_inter_cost = INT64_MAX;
+  int64_t inter_cost;
+  int rf_idx;
+  const InterpFilters kernel =
+      av1_make_interp_filters(EIGHTTAP_REGULAR, EIGHTTAP_REGULAR);
+
+  int64_t best_intra_cost = INT64_MAX;
+  int64_t intra_cost;
+  PREDICTION_MODE mode;
+  int mb_y_offset = mi_row * MI_SIZE * xd->cur_buf->y_stride + mi_col * MI_SIZE;
+  MB_MODE_INFO mi_above, mi_left;
+
+  memset(tpl_stats, 0, sizeof(*tpl_stats));
+
+  xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
+  xd->mb_to_bottom_edge = ((cm->mi_rows - 1 - mi_row) * MI_SIZE) * 8;
+  xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
+  xd->mb_to_right_edge = ((cm->mi_cols - 1 - mi_col) * MI_SIZE) * 8;
+  xd->above_mbmi = (mi_row > 0) ? &mi_above : NULL;
+  xd->left_mbmi = (mi_col > 0) ? &mi_left : NULL;
+
+  // Intra prediction search
+  for (mode = DC_PRED; mode <= PAETH_PRED; ++mode) {
+    uint8_t *src, *dst;
+    int src_stride, dst_stride;
+
+    src = xd->cur_buf->y_buffer + mb_y_offset;
+    src_stride = xd->cur_buf->y_stride;
+
+    dst = &predictor[0];
+    dst_stride = bw;
+
+    xd->mi[0]->sb_type = bsize;
+    xd->mi[0]->ref_frame[0] = INTRA_FRAME;
+
+    av1_predict_intra_block(
+        cm, xd, block_size_wide[bsize], block_size_high[bsize], tx_size, mode,
+        0, 0, FILTER_INTRA_MODES, src, src_stride, dst, dst_stride, 0, 0, 0);
+
+    if (is_cur_buf_hbd(xd)) {
+      aom_highbd_subtract_block(bh, bw, src_diff, bw, src, src_stride, dst,
+                                dst_stride, xd->bd);
+    } else {
+      aom_subtract_block(bh, bw, src_diff, bw, src, src_stride, dst,
+                         dst_stride);
+    }
+
+    wht_fwd_txfm(src_diff, bw, coeff, tx_size);
+
+    intra_cost = aom_satd(coeff, pix_num);
+
+    if (intra_cost < best_intra_cost) best_intra_cost = intra_cost;
+  }
+
+  // Motion compensated prediction
+  best_mv.as_int = 0;
+
+  (void)mb_y_offset;
+  // Motion estimation column boundary
+  x->mv_limits.col_min = -((mi_col * MI_SIZE) + (17 - 2 * AOM_INTERP_EXTEND));
+  x->mv_limits.col_max =
+      ((cm->mi_cols - 1 - mi_col) * MI_SIZE) + (17 - 2 * AOM_INTERP_EXTEND);
+
+  for (rf_idx = 0; rf_idx < 7; ++rf_idx) {
+    if (ref_frame[rf_idx] == NULL) continue;
+
+    motion_compensated_prediction(cpi, td, xd->cur_buf->y_buffer + mb_y_offset,
+                                  ref_frame[rf_idx]->y_buffer + mb_y_offset,
+                                  xd->cur_buf->y_stride, bsize, mi_row, mi_col);
+
+    // TODO(jingning): Not yet support high bit-depth in the next three
+    // steps.
+    ConvolveParams conv_params = get_conv_params(0, 0, xd->bd);
+    WarpTypesAllowed warp_types;
+    memset(&warp_types, 0, sizeof(WarpTypesAllowed));
+
+    av1_build_inter_predictor(
+        ref_frame[rf_idx]->y_buffer + mb_y_offset, ref_frame[rf_idx]->y_stride,
+        &predictor[0], bw, &x->best_mv.as_mv, sf, bw, bh, &conv_params, kernel,
+        &warp_types, mi_col * MI_SIZE, mi_row * MI_SIZE, 0, 0, MV_PRECISION_Q3,
+        mi_col * MI_SIZE, mi_row * MI_SIZE, xd, 0);
+    if (is_cur_buf_hbd(xd)) {
+      aom_highbd_subtract_block(
+          bh, bw, src_diff, bw, xd->cur_buf->y_buffer + mb_y_offset,
+          xd->cur_buf->y_stride, &predictor[0], bw, xd->bd);
+    } else {
+      aom_subtract_block(bh, bw, src_diff, bw,
+                         xd->cur_buf->y_buffer + mb_y_offset,
+                         xd->cur_buf->y_stride, &predictor[0], bw);
+    }
+    wht_fwd_txfm(src_diff, bw, coeff, tx_size);
+
+    inter_cost = aom_satd(coeff, pix_num);
+    if (inter_cost < best_inter_cost) {
+      best_rf_idx = rf_idx;
+      best_inter_cost = inter_cost;
+      best_mv.as_int = x->best_mv.as_int;
+      get_quantize_error(x, 0, coeff, qcoeff, dqcoeff, tx_size, recon_error,
+                         sse);
+    }
+  }
+  best_intra_cost = AOMMAX(best_intra_cost, 1);
+  best_inter_cost = AOMMIN(best_intra_cost, best_inter_cost);
+  tpl_stats->inter_cost = best_inter_cost << TPL_DEP_COST_SCALE_LOG2;
+  tpl_stats->intra_cost = best_intra_cost << TPL_DEP_COST_SCALE_LOG2;
+  tpl_stats->mc_dep_cost = tpl_stats->intra_cost + tpl_stats->mc_flow;
+
+  tpl_stats->ref_frame_index = gf_picture[frame_idx].ref_frame[best_rf_idx];
+  tpl_stats->mv.as_int = best_mv.as_int;
+}
+
+static int round_floor(int ref_pos, int bsize_pix) {
+  int round;
+  if (ref_pos < 0)
+    round = -(1 + (-ref_pos - 1) / bsize_pix);
+  else
+    round = ref_pos / bsize_pix;
+
+  return round;
+}
+
+static int get_overlap_area(int grid_pos_row, int grid_pos_col, int ref_pos_row,
+                            int ref_pos_col, int block, BLOCK_SIZE bsize) {
+  int width = 0, height = 0;
+  int bw = 4 << mi_size_wide_log2[bsize];
+  int bh = 4 << mi_size_high_log2[bsize];
+
+  switch (block) {
+    case 0:
+      width = grid_pos_col + bw - ref_pos_col;
+      height = grid_pos_row + bh - ref_pos_row;
+      break;
+    case 1:
+      width = ref_pos_col + bw - grid_pos_col;
+      height = grid_pos_row + bh - ref_pos_row;
+      break;
+    case 2:
+      width = grid_pos_col + bw - ref_pos_col;
+      height = ref_pos_row + bh - grid_pos_row;
+      break;
+    case 3:
+      width = ref_pos_col + bw - grid_pos_col;
+      height = ref_pos_row + bh - grid_pos_row;
+      break;
+    default: assert(0);
+  }
+
+  return width * height;
+}
+
+static void tpl_model_update_b(TplDepFrame *tpl_frame, TplDepStats *tpl_stats,
+                               int mi_row, int mi_col, const BLOCK_SIZE bsize) {
+  TplDepFrame *ref_tpl_frame = &tpl_frame[tpl_stats->ref_frame_index];
+  TplDepStats *ref_stats = ref_tpl_frame->tpl_stats_ptr;
+  MV mv = tpl_stats->mv.as_mv;
+  int mv_row = mv.row >> 3;
+  int mv_col = mv.col >> 3;
+
+  int ref_pos_row = mi_row * MI_SIZE + mv_row;
+  int ref_pos_col = mi_col * MI_SIZE + mv_col;
+
+  const int bw = 4 << mi_size_wide_log2[bsize];
+  const int bh = 4 << mi_size_high_log2[bsize];
+  const int mi_height = mi_size_high[bsize];
+  const int mi_width = mi_size_wide[bsize];
+  const int pix_num = bw * bh;
+
+  // top-left on grid block location in pixel
+  int grid_pos_row_base = round_floor(ref_pos_row, bh) * bh;
+  int grid_pos_col_base = round_floor(ref_pos_col, bw) * bw;
+  int block;
+
+  for (block = 0; block < 4; ++block) {
+    int grid_pos_row = grid_pos_row_base + bh * (block >> 1);
+    int grid_pos_col = grid_pos_col_base + bw * (block & 0x01);
+
+    if (grid_pos_row >= 0 && grid_pos_row < ref_tpl_frame->mi_rows * MI_SIZE &&
+        grid_pos_col >= 0 && grid_pos_col < ref_tpl_frame->mi_cols * MI_SIZE) {
+      int overlap_area = get_overlap_area(
+          grid_pos_row, grid_pos_col, ref_pos_row, ref_pos_col, block, bsize);
+      int ref_mi_row = round_floor(grid_pos_row, bh) * mi_height;
+      int ref_mi_col = round_floor(grid_pos_col, bw) * mi_width;
+
+      int64_t mc_flow = tpl_stats->mc_dep_cost -
+                        (tpl_stats->mc_dep_cost * tpl_stats->inter_cost) /
+                            tpl_stats->intra_cost;
+
+      int idx, idy;
+
+      for (idy = 0; idy < mi_height; ++idy) {
+        for (idx = 0; idx < mi_width; ++idx) {
+          TplDepStats *des_stats =
+              &ref_stats[(ref_mi_row + idy) * ref_tpl_frame->stride +
+                         (ref_mi_col + idx)];
+
+          des_stats->mc_flow += (mc_flow * overlap_area) / pix_num;
+          des_stats->mc_ref_cost +=
+              ((tpl_stats->intra_cost - tpl_stats->inter_cost) * overlap_area) /
+              pix_num;
+          assert(overlap_area >= 0);
+        }
+      }
+    }
+  }
+}
+
+static void tpl_model_update(TplDepFrame *tpl_frame, TplDepStats *tpl_stats,
+                             int mi_row, int mi_col, const BLOCK_SIZE bsize) {
+  int idx, idy;
+  const int mi_height = mi_size_high[bsize];
+  const int mi_width = mi_size_wide[bsize];
+
+  for (idy = 0; idy < mi_height; ++idy) {
+    for (idx = 0; idx < mi_width; ++idx) {
+      TplDepStats *tpl_ptr =
+          &tpl_stats[(mi_row + idy) * tpl_frame->stride + (mi_col + idx)];
+      tpl_model_update_b(tpl_frame, tpl_ptr, mi_row + idy, mi_col + idx,
+                         BLOCK_4X4);
+    }
+  }
+}
+
+static void tpl_model_store(TplDepStats *tpl_stats, int mi_row, int mi_col,
+                            BLOCK_SIZE bsize, int stride,
+                            const TplDepStats *src_stats) {
+  const int mi_height = mi_size_high[bsize];
+  const int mi_width = mi_size_wide[bsize];
+  int idx, idy;
+
+  int64_t intra_cost = src_stats->intra_cost / (mi_height * mi_width);
+  int64_t inter_cost = src_stats->inter_cost / (mi_height * mi_width);
+
+  TplDepStats *tpl_ptr;
+
+  intra_cost = AOMMAX(1, intra_cost);
+  inter_cost = AOMMAX(1, inter_cost);
+
+  for (idy = 0; idy < mi_height; ++idy) {
+    tpl_ptr = &tpl_stats[(mi_row + idy) * stride + mi_col];
+    for (idx = 0; idx < mi_width; ++idx) {
+      tpl_ptr->intra_cost = intra_cost;
+      tpl_ptr->inter_cost = inter_cost;
+      tpl_ptr->mc_dep_cost = tpl_ptr->intra_cost + tpl_ptr->mc_flow;
+      tpl_ptr->ref_frame_index = src_stats->ref_frame_index;
+      tpl_ptr->mv.as_int = src_stats->mv.as_int;
+      ++tpl_ptr;
+    }
+  }
+}
+
+static void mc_flow_dispenser(AV1_COMP *cpi, GF_PICTURE *gf_picture,
+                              int frame_idx) {
+  TplDepFrame *tpl_frame = &cpi->tpl_stats[frame_idx];
+  YV12_BUFFER_CONFIG *this_frame = gf_picture[frame_idx].frame;
+  YV12_BUFFER_CONFIG *ref_frame[7] = {
+    NULL, NULL, NULL, NULL, NULL, NULL, NULL
+  };
+
+  AV1_COMMON *cm = &cpi->common;
+  struct scale_factors sf;
+  int rdmult, idx;
+  ThreadData *td = &cpi->td;
+  MACROBLOCK *x = &td->mb;
+  MACROBLOCKD *xd = &x->e_mbd;
+  int mi_row, mi_col;
+
+  DECLARE_ALIGNED(16, uint16_t, predictor16[32 * 32 * 3]);
+  DECLARE_ALIGNED(16, uint8_t, predictor8[32 * 32 * 3]);
+  uint8_t *predictor;
+  DECLARE_ALIGNED(16, int16_t, src_diff[32 * 32]);
+  DECLARE_ALIGNED(16, tran_low_t, coeff[32 * 32]);
+  DECLARE_ALIGNED(16, tran_low_t, qcoeff[32 * 32]);
+  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]);
+
+  const BLOCK_SIZE bsize = BLOCK_32X32;
+  const TX_SIZE tx_size = max_txsize_lookup[bsize];
+  const int mi_height = mi_size_high[bsize];
+  const int mi_width = mi_size_wide[bsize];
+  int64_t recon_error, sse;
+
+  // Setup scaling factor
+  av1_setup_scale_factors_for_frame(
+      &sf, this_frame->y_crop_width, this_frame->y_crop_height,
+      this_frame->y_crop_width, this_frame->y_crop_height);
+
+  if (is_cur_buf_hbd(xd))
+    predictor = CONVERT_TO_BYTEPTR(predictor16);
+  else
+    predictor = predictor8;
+
+  // Prepare reference frame pointers. If any reference frame slot is
+  // unavailable, the pointer will be set to Null.
+  for (idx = 0; idx < 7; ++idx) {
+    int rf_idx = gf_picture[frame_idx].ref_frame[idx];
+    if (rf_idx != -1) ref_frame[idx] = gf_picture[rf_idx].frame;
+  }
+
+  xd->mi = cm->mi_grid_visible;
+  xd->mi[0] = cm->mi;
+  xd->cur_buf = this_frame;
+
+  // Get rd multiplier set up.
+  rdmult = (int)av1_compute_rd_mult(cpi, tpl_frame->base_qindex);
+  if (rdmult < 1) rdmult = 1;
+  set_error_per_bit(&cpi->td.mb, rdmult);
+  av1_initialize_me_consts(cpi, &cpi->td.mb, tpl_frame->base_qindex);
+
+  tpl_frame->is_valid = 1;
+
+  cm->base_qindex = tpl_frame->base_qindex;
+  av1_frame_init_quantizer(cpi);
+
+  for (mi_row = 0; mi_row < cm->mi_rows; mi_row += mi_height) {
+    // Motion estimation row boundary
+    x->mv_limits.row_min = -((mi_row * MI_SIZE) + (17 - 2 * AOM_INTERP_EXTEND));
+    x->mv_limits.row_max =
+        (cm->mi_rows - 1 - mi_row) * MI_SIZE + (17 - 2 * AOM_INTERP_EXTEND);
+    for (mi_col = 0; mi_col < cm->mi_cols; mi_col += mi_width) {
+      TplDepStats tpl_stats;
+      mode_estimation(cpi, x, xd, &sf, gf_picture, frame_idx, src_diff, coeff,
+                      qcoeff, dqcoeff, mi_row, mi_col, bsize, tx_size,
+                      ref_frame, predictor, &recon_error, &sse, &tpl_stats);
+
+      // Motion flow dependency dispenser.
+      tpl_model_store(tpl_frame->tpl_stats_ptr, mi_row, mi_col, bsize,
+                      tpl_frame->stride, &tpl_stats);
+
+      tpl_model_update(cpi->tpl_stats, tpl_frame->tpl_stats_ptr, mi_row, mi_col,
+                       bsize);
+    }
+  }
+}
+
+static void init_gop_frames(AV1_COMP *cpi, GF_PICTURE *gf_picture,
+                            const GF_GROUP *gf_group, int *tpl_group_frames,
+                            const EncodeFrameInput *const frame_input) {
+  AV1_COMMON *cm = &cpi->common;
+  const SequenceHeader *const seq_params = &cm->seq_params;
+  int frame_idx = 0;
+  int i;
+  int gld_index = -1;
+  int alt_index = -1;
+  int lst_index = -1;
+  int extend_frame_count = 0;
+  int pframe_qindex = cpi->tpl_stats[2].base_qindex;
+
+  RefCntBuffer *frame_bufs = cm->buffer_pool->frame_bufs;
+  int recon_frame_index[INTER_REFS_PER_FRAME + 1] = { -1, -1, -1, -1,
+                                                      -1, -1, -1, -1 };
+
+  // TODO(jingning): To be used later for gf frame type parsing.
+  (void)gf_group;
+
+  for (i = 0; i < FRAME_BUFFERS && frame_idx < INTER_REFS_PER_FRAME + 1; ++i) {
+    if (frame_bufs[i].ref_count == 0) {
+      alloc_frame_mvs(cm, &frame_bufs[i]);
+      if (aom_realloc_frame_buffer(
+              &frame_bufs[i].buf, cm->width, cm->height,
+              seq_params->subsampling_x, seq_params->subsampling_y,
+              seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels,
+              cm->byte_alignment, NULL, NULL, NULL))
+        aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
+                           "Failed to allocate frame buffer");
+
+      recon_frame_index[frame_idx] = i;
+      ++frame_idx;
+    }
+  }
+
+  for (i = 0; i < INTER_REFS_PER_FRAME + 1; ++i) {
+    assert(recon_frame_index[i] >= 0);
+    cpi->tpl_recon_frames[i] = &frame_bufs[recon_frame_index[i]].buf;
+  }
+
+  *tpl_group_frames = 0;
+
+  // Initialize Golden reference frame.
+  gf_picture[0].frame = NULL;
+  RefCntBuffer *ref_buf = get_ref_frame_buf(cm, GOLDEN_FRAME);
+  if (ref_buf) gf_picture[0].frame = &ref_buf->buf;
+  for (i = 0; i < 7; ++i) gf_picture[0].ref_frame[i] = -1;
+  gld_index = 0;
+  ++*tpl_group_frames;
+
+  // Initialize ARF frame
+  gf_picture[1].frame = frame_input->source;
+  gf_picture[1].ref_frame[0] = gld_index;
+  gf_picture[1].ref_frame[1] = lst_index;
+  gf_picture[1].ref_frame[2] = alt_index;
+  // TODO(yuec) Need o  figure out full AV1 reference model
+  for (i = 3; i < 7; ++i) gf_picture[1].ref_frame[i] = -1;
+  alt_index = 1;
+  ++*tpl_group_frames;
+
+  // Initialize P frames
+  for (frame_idx = 2; frame_idx < MAX_LAG_BUFFERS; ++frame_idx) {
+    struct lookahead_entry *buf =
+        av1_lookahead_peek(cpi->lookahead, frame_idx - 2);
+
+    if (buf == NULL) break;
+
+    gf_picture[frame_idx].frame = &buf->img;
+    gf_picture[frame_idx].ref_frame[0] = gld_index;
+    gf_picture[frame_idx].ref_frame[1] = lst_index;
+    gf_picture[frame_idx].ref_frame[2] = alt_index;
+    for (i = 3; i < 7; ++i) gf_picture[frame_idx].ref_frame[i] = -1;
+
+    ++*tpl_group_frames;
+    lst_index = frame_idx;
+
+    if (frame_idx == cpi->rc.baseline_gf_interval + 1) break;
+  }
+
+  gld_index = frame_idx;
+  lst_index = AOMMAX(0, frame_idx - 1);
+  alt_index = -1;
+  ++frame_idx;
+
+  // Extend two frames outside the current gf group.
+  for (; frame_idx < MAX_LAG_BUFFERS && extend_frame_count < 2; ++frame_idx) {
+    struct lookahead_entry *buf =
+        av1_lookahead_peek(cpi->lookahead, frame_idx - 2);
+
+    if (buf == NULL) break;
+
+    cpi->tpl_stats[frame_idx].base_qindex = pframe_qindex;
+
+    gf_picture[frame_idx].frame = &buf->img;
+    gf_picture[frame_idx].ref_frame[0] = gld_index;
+    gf_picture[frame_idx].ref_frame[1] = lst_index;
+    gf_picture[frame_idx].ref_frame[2] = alt_index;
+    for (i = 3; i < 7; ++i) gf_picture[frame_idx].ref_frame[i] = -1;
+    lst_index = frame_idx;
+    ++*tpl_group_frames;
+    ++extend_frame_count;
+  }
+}
+
+static void init_tpl_stats(AV1_COMP *cpi) {
+  int frame_idx;
+  for (frame_idx = 0; frame_idx < MAX_LAG_BUFFERS; ++frame_idx) {
+    TplDepFrame *tpl_frame = &cpi->tpl_stats[frame_idx];
+    memset(tpl_frame->tpl_stats_ptr, 0,
+           tpl_frame->height * tpl_frame->width *
+               sizeof(*tpl_frame->tpl_stats_ptr));
+    tpl_frame->is_valid = 0;
+  }
+}
+
+void av1_tpl_setup_stats(AV1_COMP *cpi,
+                         const EncodeFrameInput *const frame_input) {
+  GF_PICTURE gf_picture[MAX_LAG_BUFFERS];
+  const GF_GROUP *gf_group = &cpi->twopass.gf_group;
+  int tpl_group_frames = 0;
+  int frame_idx;
+
+  init_gop_frames(cpi, gf_picture, gf_group, &tpl_group_frames, frame_input);
+
+  init_tpl_stats(cpi);
+
+  // Backward propagation from tpl_group_frames to 1.
+  for (frame_idx = tpl_group_frames - 1; frame_idx > 0; --frame_idx)
+    mc_flow_dispenser(cpi, gf_picture, frame_idx);
+}
diff --git a/third_party/aom/av1/encoder/tpl_model.h b/third_party/aom/av1/encoder/tpl_model.h
new file mode 100644
index 0000000000..f6b33b07c3
--- /dev/null
+++ b/third_party/aom/av1/encoder/tpl_model.h
@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_AV1_ENCODER_TPL_MODEL_H_
+#define AOM_AV1_ENCODER_TPL_MODEL_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void av1_tpl_setup_stats(AV1_COMP *cpi,
+                         const EncodeFrameInput *const frame_input);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // AOM_AV1_ENCODER_TPL_MODEL_H_
diff --git a/third_party/aom/av1/encoder/x86/av1_fwd_txfm1d_sse4.c b/third_party/aom/av1/encoder/x86/av1_fwd_txfm1d_sse4.c
index 07615543c6..865ac31494 100644
--- a/third_party/aom/av1/encoder/x86/av1_fwd_txfm1d_sse4.c
+++ b/third_party/aom/av1/encoder/x86/av1_fwd_txfm1d_sse4.c
@@ -11,45 +11,78 @@
 
 #include "av1/encoder/x86/av1_txfm1d_sse4.h"
 
-void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
-                           int8_t cos_bit) {
+void av1_fdct32_new_sse4_1(__m128i *input, __m128i *output, int cos_bit,
+                           const int stride) {
   __m128i buf0[32];
   __m128i buf1[32];
   const int32_t *cospi;
+
+  int startidx = 0 * stride;
+  int endidx = 31 * stride;
   // stage 0
   // stage 1
-  buf1[0] = _mm_add_epi32(input[0], input[31]);
-  buf1[31] = _mm_sub_epi32(input[0], input[31]);
-  buf1[1] = _mm_add_epi32(input[1], input[30]);
-  buf1[30] = _mm_sub_epi32(input[1], input[30]);
-  buf1[2] = _mm_add_epi32(input[2], input[29]);
-  buf1[29] = _mm_sub_epi32(input[2], input[29]);
-  buf1[3] = _mm_add_epi32(input[3], input[28]);
-  buf1[28] = _mm_sub_epi32(input[3], input[28]);
-  buf1[4] = _mm_add_epi32(input[4], input[27]);
-  buf1[27] = _mm_sub_epi32(input[4], input[27]);
-  buf1[5] = _mm_add_epi32(input[5], input[26]);
-  buf1[26] = _mm_sub_epi32(input[5], input[26]);
-  buf1[6] = _mm_add_epi32(input[6], input[25]);
-  buf1[25] = _mm_sub_epi32(input[6], input[25]);
-  buf1[7] = _mm_add_epi32(input[7], input[24]);
-  buf1[24] = _mm_sub_epi32(input[7], input[24]);
-  buf1[8] = _mm_add_epi32(input[8], input[23]);
-  buf1[23] = _mm_sub_epi32(input[8], input[23]);
-  buf1[9] = _mm_add_epi32(input[9], input[22]);
-  buf1[22] = _mm_sub_epi32(input[9], input[22]);
-  buf1[10] = _mm_add_epi32(input[10], input[21]);
-  buf1[21] = _mm_sub_epi32(input[10], input[21]);
-  buf1[11] = _mm_add_epi32(input[11], input[20]);
-  buf1[20] = _mm_sub_epi32(input[11], input[20]);
-  buf1[12] = _mm_add_epi32(input[12], input[19]);
-  buf1[19] = _mm_sub_epi32(input[12], input[19]);
-  buf1[13] = _mm_add_epi32(input[13], input[18]);
-  buf1[18] = _mm_sub_epi32(input[13], input[18]);
-  buf1[14] = _mm_add_epi32(input[14], input[17]);
-  buf1[17] = _mm_sub_epi32(input[14], input[17]);
-  buf1[15] = _mm_add_epi32(input[15], input[16]);
-  buf1[16] = _mm_sub_epi32(input[15], input[16]);
+  buf1[0] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[31] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[1] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[30] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[2] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[29] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[3] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[28] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[4] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[27] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[5] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[26] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[6] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[25] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[7] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[24] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[8] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[23] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[9] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[22] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[10] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[21] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[11] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[20] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[12] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[19] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[13] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[18] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[14] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[17] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += stride;
+  endidx -= stride;
+  buf1[15] = _mm_add_epi32(input[startidx], input[endidx]);
+  buf1[16] = _mm_sub_epi32(input[startidx], input[endidx]);
 
   // stage 2
   cospi = cospi_arr(cos_bit);
@@ -296,39 +329,71 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
   btf_32_sse4_1_type1(cospi[6], cospi[58], buf1[23], buf1[24], buf0[23],
                       buf0[24], cos_bit);
 
+  startidx = 0 * stride;
+  endidx = 31 * stride;
   // stage 9
-  output[0] = buf0[0];
-  output[1] = buf0[16];
-  output[2] = buf0[8];
-  output[3] = buf0[24];
-  output[4] = buf0[4];
-  output[5] = buf0[20];
-  output[6] = buf0[12];
-  output[7] = buf0[28];
-  output[8] = buf0[2];
-  output[9] = buf0[18];
-  output[10] = buf0[10];
-  output[11] = buf0[26];
-  output[12] = buf0[6];
-  output[13] = buf0[22];
-  output[14] = buf0[14];
-  output[15] = buf0[30];
-  output[16] = buf0[1];
-  output[17] = buf0[17];
-  output[18] = buf0[9];
-  output[19] = buf0[25];
-  output[20] = buf0[5];
-  output[21] = buf0[21];
-  output[22] = buf0[13];
-  output[23] = buf0[29];
-  output[24] = buf0[3];
-  output[25] = buf0[19];
-  output[26] = buf0[11];
-  output[27] = buf0[27];
-  output[28] = buf0[7];
-  output[29] = buf0[23];
-  output[30] = buf0[15];
-  output[31] = buf0[31];
+  output[startidx] = buf0[0];
+  output[endidx] = buf0[31];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[16];
+  output[endidx] = buf0[15];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[8];
+  output[endidx] = buf0[23];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[24];
+  output[endidx] = buf0[7];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[4];
+  output[endidx] = buf0[27];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[20];
+  output[endidx] = buf0[11];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[12];
+  output[endidx] = buf0[19];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[28];
+  output[endidx] = buf0[3];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[2];
+  output[endidx] = buf0[29];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[18];
+  output[endidx] = buf0[13];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[10];
+  output[endidx] = buf0[21];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[26];
+  output[endidx] = buf0[5];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[6];
+  output[endidx] = buf0[25];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[22];
+  output[endidx] = buf0[9];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[14];
+  output[endidx] = buf0[17];
+  startidx += stride;
+  endidx -= stride;
+  output[startidx] = buf0[30];
+  output[endidx] = buf0[1];
 }
 
 void av1_fadst4_new_sse4_1(const __m128i *input, __m128i *output,
@@ -394,9 +459,8 @@ void av1_fadst4_new_sse4_1(const __m128i *input, __m128i *output,
   }
 }
 
-void av1_fdct64_new_sse4_1(const __m128i *input, __m128i *output,
-                           int8_t cos_bit, const int instride,
-                           const int outstride) {
+void av1_fdct64_new_sse4_1(__m128i *input, __m128i *output, int8_t cos_bit,
+                           const int instride, const int outstride) {
   const int32_t *cospi = cospi_arr(cos_bit);
   const __m128i __rounding = _mm_set1_epi32(1 << (cos_bit - 1));
 
@@ -479,72 +543,136 @@ void av1_fdct64_new_sse4_1(const __m128i *input, __m128i *output,
   __m128i cospi_p03 = _mm_set1_epi32(cospi[3]);
   __m128i cospi_p61 = _mm_set1_epi32(cospi[61]);
 
+  int startidx = 0 * instride;
+  int endidx = 63 * instride;
   // stage 1
   __m128i x1[64];
-  x1[0] = _mm_add_epi32(input[0 * instride], input[63 * instride]);
-  x1[63] = _mm_sub_epi32(input[0 * instride], input[63 * instride]);
-  x1[1] = _mm_add_epi32(input[1 * instride], input[62 * instride]);
-  x1[62] = _mm_sub_epi32(input[1 * instride], input[62 * instride]);
-  x1[2] = _mm_add_epi32(input[2 * instride], input[61 * instride]);
-  x1[61] = _mm_sub_epi32(input[2 * instride], input[61 * instride]);
-  x1[3] = _mm_add_epi32(input[3 * instride], input[60 * instride]);
-  x1[60] = _mm_sub_epi32(input[3 * instride], input[60 * instride]);
-  x1[4] = _mm_add_epi32(input[4 * instride], input[59 * instride]);
-  x1[59] = _mm_sub_epi32(input[4 * instride], input[59 * instride]);
-  x1[5] = _mm_add_epi32(input[5 * instride], input[58 * instride]);
-  x1[58] = _mm_sub_epi32(input[5 * instride], input[58 * instride]);
-  x1[6] = _mm_add_epi32(input[6 * instride], input[57 * instride]);
-  x1[57] = _mm_sub_epi32(input[6 * instride], input[57 * instride]);
-  x1[7] = _mm_add_epi32(input[7 * instride], input[56 * instride]);
-  x1[56] = _mm_sub_epi32(input[7 * instride], input[56 * instride]);
-  x1[8] = _mm_add_epi32(input[8 * instride], input[55 * instride]);
-  x1[55] = _mm_sub_epi32(input[8 * instride], input[55 * instride]);
-  x1[9] = _mm_add_epi32(input[9 * instride], input[54 * instride]);
-  x1[54] = _mm_sub_epi32(input[9 * instride], input[54 * instride]);
-  x1[10] = _mm_add_epi32(input[10 * instride], input[53 * instride]);
-  x1[53] = _mm_sub_epi32(input[10 * instride], input[53 * instride]);
-  x1[11] = _mm_add_epi32(input[11 * instride], input[52 * instride]);
-  x1[52] = _mm_sub_epi32(input[11 * instride], input[52 * instride]);
-  x1[12] = _mm_add_epi32(input[12 * instride], input[51 * instride]);
-  x1[51] = _mm_sub_epi32(input[12 * instride], input[51 * instride]);
-  x1[13] = _mm_add_epi32(input[13 * instride], input[50 * instride]);
-  x1[50] = _mm_sub_epi32(input[13 * instride], input[50 * instride]);
-  x1[14] = _mm_add_epi32(input[14 * instride], input[49 * instride]);
-  x1[49] = _mm_sub_epi32(input[14 * instride], input[49 * instride]);
-  x1[15] = _mm_add_epi32(input[15 * instride], input[48 * instride]);
-  x1[48] = _mm_sub_epi32(input[15 * instride], input[48 * instride]);
-  x1[16] = _mm_add_epi32(input[16 * instride], input[47 * instride]);
-  x1[47] = _mm_sub_epi32(input[16 * instride], input[47 * instride]);
-  x1[17] = _mm_add_epi32(input[17 * instride], input[46 * instride]);
-  x1[46] = _mm_sub_epi32(input[17 * instride], input[46 * instride]);
-  x1[18] = _mm_add_epi32(input[18 * instride], input[45 * instride]);
-  x1[45] = _mm_sub_epi32(input[18 * instride], input[45 * instride]);
-  x1[19] = _mm_add_epi32(input[19 * instride], input[44 * instride]);
-  x1[44] = _mm_sub_epi32(input[19 * instride], input[44 * instride]);
-  x1[20] = _mm_add_epi32(input[20 * instride], input[43 * instride]);
-  x1[43] = _mm_sub_epi32(input[20 * instride], input[43 * instride]);
-  x1[21] = _mm_add_epi32(input[21 * instride], input[42 * instride]);
-  x1[42] = _mm_sub_epi32(input[21 * instride], input[42 * instride]);
-  x1[22] = _mm_add_epi32(input[22 * instride], input[41 * instride]);
-  x1[41] = _mm_sub_epi32(input[22 * instride], input[41 * instride]);
-  x1[23] = _mm_add_epi32(input[23 * instride], input[40 * instride]);
-  x1[40] = _mm_sub_epi32(input[23 * instride], input[40 * instride]);
-  x1[24] = _mm_add_epi32(input[24 * instride], input[39 * instride]);
-  x1[39] = _mm_sub_epi32(input[24 * instride], input[39 * instride]);
-  x1[25] = _mm_add_epi32(input[25 * instride], input[38 * instride]);
-  x1[38] = _mm_sub_epi32(input[25 * instride], input[38 * instride]);
-  x1[26] = _mm_add_epi32(input[26 * instride], input[37 * instride]);
-  x1[37] = _mm_sub_epi32(input[26 * instride], input[37 * instride]);
-  x1[27] = _mm_add_epi32(input[27 * instride], input[36 * instride]);
-  x1[36] = _mm_sub_epi32(input[27 * instride], input[36 * instride]);
-  x1[28] = _mm_add_epi32(input[28 * instride], input[35 * instride]);
-  x1[35] = _mm_sub_epi32(input[28 * instride], input[35 * instride]);
-  x1[29] = _mm_add_epi32(input[29 * instride], input[34 * instride]);
-  x1[34] = _mm_sub_epi32(input[29 * instride], input[34 * instride]);
-  x1[30] = _mm_add_epi32(input[30 * instride], input[33 * instride]);
-  x1[33] = _mm_sub_epi32(input[30 * instride], input[33 * instride]);
-  x1[31] = _mm_add_epi32(input[31 * instride], input[32 * instride]);
-  x1[32] = _mm_sub_epi32(input[31 * instride], input[32 * instride]);
+  x1[0] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[63] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[1] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[62] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[2] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[61] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[3] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[60] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[4] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[59] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[5] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[58] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[6] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[57] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[7] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[56] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[8] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[55] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[9] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[54] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[10] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[53] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[11] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[52] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[12] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[51] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[13] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[50] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[14] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[49] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[15] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[48] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[16] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[47] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[17] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[46] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[18] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[45] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[19] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[44] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[20] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[43] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[21] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[42] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[22] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[41] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[23] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[40] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[24] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[39] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[25] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[38] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[26] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[37] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[27] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[36] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[28] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[35] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[29] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[34] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[30] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[33] = _mm_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[31] = _mm_add_epi32(input[startidx], input[endidx]);
+  x1[32] = _mm_sub_epi32(input[startidx], input[endidx]);
 
   // stage 2
   __m128i x2[64];
@@ -1149,69 +1277,141 @@ void av1_fdct64_new_sse4_1(const __m128i *input, __m128i *output,
   btf_32_type1_sse4_1_new(cospi_p03, cospi_p61, x9[47], x9[48], x10[47],
                           x10[48], __rounding, cos_bit);
 
+  startidx = 0 * outstride;
+  endidx = 63 * outstride;
   // stage 11
-  output[0 * outstride] = x10[0];
-  output[1 * outstride] = x10[32];
-  output[2 * outstride] = x10[16];
-  output[3 * outstride] = x10[48];
-  output[4 * outstride] = x10[8];
-  output[5 * outstride] = x10[40];
-  output[6 * outstride] = x10[24];
-  output[7 * outstride] = x10[56];
-  output[8 * outstride] = x10[4];
-  output[9 * outstride] = x10[36];
-  output[10 * outstride] = x10[20];
-  output[11 * outstride] = x10[52];
-  output[12 * outstride] = x10[12];
-  output[13 * outstride] = x10[44];
-  output[14 * outstride] = x10[28];
-  output[15 * outstride] = x10[60];
-  output[16 * outstride] = x10[2];
-  output[17 * outstride] = x10[34];
-  output[18 * outstride] = x10[18];
-  output[19 * outstride] = x10[50];
-  output[20 * outstride] = x10[10];
-  output[21 * outstride] = x10[42];
-  output[22 * outstride] = x10[26];
-  output[23 * outstride] = x10[58];
-  output[24 * outstride] = x10[6];
-  output[25 * outstride] = x10[38];
-  output[26 * outstride] = x10[22];
-  output[27 * outstride] = x10[54];
-  output[28 * outstride] = x10[14];
-  output[29 * outstride] = x10[46];
-  output[30 * outstride] = x10[30];
-  output[31 * outstride] = x10[62];
-  output[32 * outstride] = x10[1];
-  output[33 * outstride] = x10[33];
-  output[34 * outstride] = x10[17];
-  output[35 * outstride] = x10[49];
-  output[36 * outstride] = x10[9];
-  output[37 * outstride] = x10[41];
-  output[38 * outstride] = x10[25];
-  output[39 * outstride] = x10[57];
-  output[40 * outstride] = x10[5];
-  output[41 * outstride] = x10[37];
-  output[42 * outstride] = x10[21];
-  output[43 * outstride] = x10[53];
-  output[44 * outstride] = x10[13];
-  output[45 * outstride] = x10[45];
-  output[46 * outstride] = x10[29];
-  output[47 * outstride] = x10[61];
-  output[48 * outstride] = x10[3];
-  output[49 * outstride] = x10[35];
-  output[50 * outstride] = x10[19];
-  output[51 * outstride] = x10[51];
-  output[52 * outstride] = x10[11];
-  output[53 * outstride] = x10[43];
-  output[54 * outstride] = x10[27];
-  output[55 * outstride] = x10[59];
-  output[56 * outstride] = x10[7];
-  output[57 * outstride] = x10[39];
-  output[58 * outstride] = x10[23];
-  output[59 * outstride] = x10[55];
-  output[60 * outstride] = x10[15];
-  output[61 * outstride] = x10[47];
-  output[62 * outstride] = x10[31];
-  output[63 * outstride] = x10[63];
+  output[startidx] = x10[0];
+  output[endidx] = x10[63];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[32];
+  output[endidx] = x10[31];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[16];
+  output[endidx] = x10[47];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[48];
+  output[endidx] = x10[15];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[8];
+  output[endidx] = x10[55];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[40];
+  output[endidx] = x10[23];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[24];
+  output[endidx] = x10[39];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[56];
+  output[endidx] = x10[7];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[4];
+  output[endidx] = x10[59];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[36];
+  output[endidx] = x10[27];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[20];
+  output[endidx] = x10[43];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[52];
+  output[endidx] = x10[11];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[12];
+  output[endidx] = x10[51];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[44];
+  output[endidx] = x10[19];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[28];
+  output[endidx] = x10[35];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[60];
+  output[endidx] = x10[3];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[2];
+  output[endidx] = x10[61];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[34];
+  output[endidx] = x10[29];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[18];
+  output[endidx] = x10[45];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[50];
+  output[endidx] = x10[13];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[10];
+  output[endidx] = x10[53];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[42];
+  output[endidx] = x10[21];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[26];
+  output[endidx] = x10[37];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[58];
+  output[endidx] = x10[5];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[6];
+  output[endidx] = x10[57];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[38];
+  output[endidx] = x10[25];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[22];
+  output[endidx] = x10[41];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[54];
+  output[endidx] = x10[9];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[14];
+  output[endidx] = x10[49];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[46];
+  output[endidx] = x10[17];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[30];
+  output[endidx] = x10[33];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x10[62];
+  output[endidx] = x10[1];
+}
+
+void av1_idtx32_new_sse4_1(__m128i *input, __m128i *output, int cos_bit,
+                           const int col_num) {
+  (void)cos_bit;
+  for (int i = 0; i < 32; i++) {
+    output[i * col_num] = _mm_slli_epi32(input[i * col_num], 2);
+  }
 }
diff --git a/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_avx2.c b/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_avx2.c
index 592462e20d..13982ccbb9 100644
--- a/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_avx2.c
+++ b/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_avx2.c
@@ -1436,45 +1436,6 @@ static INLINE void fidentity16x32_new_avx2(const __m256i *input,
   }
 }
 
-static INLINE void av1_round_shift_array_32_avx2(__m256i *input,
-                                                 __m256i *output,
-                                                 const int size,
-                                                 const int bit) {
-  if (bit > 0) {
-    int i;
-    for (i = 0; i < size; i++) {
-      output[i] = av1_round_shift_32_avx2(input[i], bit);
-    }
-  } else {
-    int i;
-    for (i = 0; i < size; i++) {
-      output[i] = _mm256_slli_epi32(input[i], -bit);
-    }
-  }
-}
-
-static INLINE void av1_round_shift_rect_array_32_avx2(__m256i *input,
-                                                      __m256i *output,
-                                                      const int size,
-                                                      const int bit) {
-  const __m256i sqrt2 = _mm256_set1_epi32(NewSqrt2);
-  if (bit > 0) {
-    int i;
-    for (i = 0; i < size; i++) {
-      const __m256i r0 = av1_round_shift_32_avx2(input[i], bit);
-      const __m256i r1 = _mm256_mullo_epi32(sqrt2, r0);
-      output[i] = av1_round_shift_32_avx2(r1, NewSqrt2Bits);
-    }
-  } else {
-    int i;
-    for (i = 0; i < size; i++) {
-      const __m256i r0 = _mm256_slli_epi32(input[i], -bit);
-      const __m256i r1 = _mm256_mullo_epi32(sqrt2, r0);
-      output[i] = av1_round_shift_32_avx2(r1, NewSqrt2Bits);
-    }
-  }
-}
-
 static INLINE void transpose_32_8x8_avx2(int stride, const __m256i *inputA,
                                          __m256i *output) {
   __m256i temp0 = _mm256_unpacklo_epi32(inputA[0], inputA[2]);
@@ -1540,6 +1501,9 @@ static INLINE void store_rect_buffer_16bit_to_32bit_w16_avx2(
   }
 }
 
+typedef void (*transform_1d_avx2)(const __m256i *input, __m256i *output,
+                                  int8_t cos_bit);
+
 static const transform_1d_avx2 col_txfm16x32_arr[TX_TYPES] = {
   fdct16x32_new_avx2,       // DCT_DCT
   NULL,                     // ADST_DCT
@@ -1885,8 +1849,8 @@ static void lowbd_fwd_txfm2d_64x32_avx2(const int16_t *input, int32_t *output,
     }
     av1_fdct64_new_avx2(bufA, bufA, cos_bit_row);
     av1_fdct64_new_avx2(bufB, bufB, cos_bit_row);
-    av1_round_shift_rect_array_32_avx2(bufA, bufA, 32, -shift[2]);
-    av1_round_shift_rect_array_32_avx2(bufB, bufB, 32, -shift[2]);
+    av1_round_shift_rect_array_32_avx2(bufA, bufA, 32, -shift[2], NewSqrt2);
+    av1_round_shift_rect_array_32_avx2(bufB, bufB, 32, -shift[2], NewSqrt2);
 
     int32_t *output8 = output + 16 * 32 * i;
     for (int j = 0; j < 4; ++j) {
@@ -1935,8 +1899,8 @@ static void lowbd_fwd_txfm2d_32x64_avx2(const int16_t *input, int32_t *output,
     }
     av1_fdct32_new_avx2(bufA, bufA, cos_bit_row);
     av1_fdct32_new_avx2(bufB, bufB, cos_bit_row);
-    av1_round_shift_rect_array_32_avx2(bufA, bufA, 32, -shift[2]);
-    av1_round_shift_rect_array_32_avx2(bufB, bufB, 32, -shift[2]);
+    av1_round_shift_rect_array_32_avx2(bufA, bufA, 32, -shift[2], NewSqrt2);
+    av1_round_shift_rect_array_32_avx2(bufB, bufB, 32, -shift[2], NewSqrt2);
 
     int32_t *output8 = output + 16 * 32 * i;
     for (int j = 0; j < 4; ++j) {
diff --git a/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_sse4.c b/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_sse4.c
index 8ec0256eb8..193f9d10af 100644
--- a/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_sse4.c
+++ b/third_party/aom/av1/encoder/x86/av1_fwd_txfm2d_sse4.c
@@ -29,31 +29,22 @@ static INLINE void int16_array_with_stride_to_int32_array_without_stride(
   }
 }
 
-typedef void (*TxfmFuncSSE2)(const __m128i *input, __m128i *output,
+typedef void (*TxfmFuncSSE2)(__m128i *input, __m128i *output,
                              const int8_t cos_bit, const int8_t *stage_range);
 
-static void fdct32_new_sse4_1(const __m128i *input, __m128i *output,
+static void fdct32_new_sse4_1(__m128i *input, __m128i *output,
                               const int8_t cos_bit, const int8_t *stage_range) {
   const int txfm_size = 32;
   const int num_per_128 = 4;
-  __m128i buf0[32];
-  __m128i buf1[32];
   int col_num = txfm_size / num_per_128;
   int col;
   (void)stage_range;
   for (col = 0; col < col_num; col++) {
-    int j;
-    for (j = 0; j < 32; ++j) {
-      buf0[j] = input[j * col_num + col];
-    }
-    av1_fdct32_new_sse4_1(buf0, buf1, cos_bit);
-    for (j = 0; j < 32; ++j) {
-      output[j * col_num + col] = buf1[j];
-    }
+    av1_fdct32_new_sse4_1((input + col), (output + col), cos_bit, col_num);
   }
 }
 
-static void fdct64_new_sse4_1(const __m128i *input, __m128i *output,
+static void fdct64_new_sse4_1(__m128i *input, __m128i *output,
                               const int8_t cos_bit, const int8_t *stage_range) {
   const int txfm_size = 64;
   const int num_per_128 = 4;
@@ -64,11 +55,20 @@ static void fdct64_new_sse4_1(const __m128i *input, __m128i *output,
                           col_num);
   }
 }
+static void idtx32x32_sse4_1(__m128i *input, __m128i *output,
+                             const int8_t cos_bit, const int8_t *stage_range) {
+  (void)stage_range;
+
+  for (int i = 0; i < 8; i++) {
+    av1_idtx32_new_sse4_1(&input[i * 32], &output[i * 32], cos_bit, 1);
+  }
+}
 
 static INLINE TxfmFuncSSE2 fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
   switch (txfm_type) {
     case TXFM_TYPE_DCT32: return fdct32_new_sse4_1; break;
     case TXFM_TYPE_DCT64: return fdct64_new_sse4_1; break;
+    case TXFM_TYPE_IDENTITY32: return idtx32x32_sse4_1; break;
     default: assert(0);
   }
   return NULL;
@@ -142,7 +142,7 @@ static INLINE void fwd_txfm2d_64x64_sse4_1(const int16_t *input,
 
   txfm2d_size_128 = (col_num >> 1) * (txfm_size >> 1);
   av1_round_shift_array_32_sse4_1(out_128, buf_128, txfm2d_size_128, -shift[2]);
-  transpose_32x32(buf_128, out_128);
+  transpose_8nx8n(buf_128, out_128, 32, 32);
 }
 
 void av1_fwd_txfm2d_32x32_sse4_1(const int16_t *input, int32_t *output,
@@ -317,8 +317,8 @@ static void lowbd_fwd_txfm2d_32x64_sse4_1(const int16_t *input, int32_t *output,
       bufA[j] = _mm_cvtepi16_epi32(buf[j]);
       bufB[j] = _mm_cvtepi16_epi32(_mm_unpackhi_epi64(buf[j], buf[j]));
     }
-    av1_fdct32_new_sse4_1(bufA, bufA, cos_bit_row);
-    av1_fdct32_new_sse4_1(bufB, bufB, cos_bit_row);
+    av1_fdct32_new_sse4_1(bufA, bufA, cos_bit_row, 1);
+    av1_fdct32_new_sse4_1(bufB, bufB, cos_bit_row, 1);
     av1_round_shift_rect_array_32_sse4_1(bufA, bufA, 32, -shift[2], NewSqrt2);
     av1_round_shift_rect_array_32_sse4_1(bufB, bufB, 32, -shift[2], NewSqrt2);
 
diff --git a/third_party/aom/av1/encoder/x86/av1_fwd_txfm_avx2.h b/third_party/aom/av1/encoder/x86/av1_fwd_txfm_avx2.h
index 38707137c4..aaad76e5ae 100644
--- a/third_party/aom/av1/encoder/x86/av1_fwd_txfm_avx2.h
+++ b/third_party/aom/av1/encoder/x86/av1_fwd_txfm_avx2.h
@@ -13,13 +13,6 @@
 #define AOM_AV1_ENCODER_X86_AV1_FWD_TXFM_AVX2_H_
 #include <immintrin.h>
 
-static INLINE __m256i av1_round_shift_32_avx2(__m256i vec, int bit) {
-  __m256i tmp, round;
-  round = _mm256_set1_epi32(1 << (bit - 1));
-  tmp = _mm256_add_epi32(vec, round);
-  return _mm256_srai_epi32(tmp, bit);
-}
-
 // out0 = in0*w0 + in1*w1
 // out1 = -in1*w0 + in0*w1
 static INLINE void btf_32_avx2_type0(const int32_t w0, const int32_t w1,
diff --git a/third_party/aom/av1/encoder/x86/av1_txfm1d_sse4.h b/third_party/aom/av1/encoder/x86/av1_txfm1d_sse4.h
index 6df2a8bdbb..b3d5b22d79 100644
--- a/third_party/aom/av1/encoder/x86/av1_txfm1d_sse4.h
+++ b/third_party/aom/av1/encoder/x86/av1_txfm1d_sse4.h
@@ -26,12 +26,10 @@ void av1_fdct8_new_sse4_1(const __m128i *input, __m128i *output,
                           const int8_t cos_bit, const int8_t *stage_range);
 void av1_fdct16_new_sse4_1(const __m128i *input, __m128i *output,
                            const int8_t cos_bit, const int8_t *stage_range);
-void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
-                           int8_t cos_bit);
-void av1_fdct64_new_sse4_1(const __m128i *input, __m128i *output,
-                           int8_t cos_bit, const int instride,
-                           const int outstride);
-
+void av1_fdct32_new_sse4_1(__m128i *input, __m128i *output, int cos_bit,
+                           const int stride);
+void av1_fdct64_new_sse4_1(__m128i *input, __m128i *output, int8_t cos_bit,
+                           const int instride, const int outstride);
 void av1_fadst4_new_sse4_1(const __m128i *input, __m128i *output,
                            const int8_t cos_bit, const int8_t *stage_range);
 void av1_fadst8_new_sse4_1(const __m128i *input, __m128i *output,
@@ -56,6 +54,10 @@ void av1_iadst8_new_sse4_1(const __m128i *input, __m128i *output,
                            const int8_t cos_bit, const int8_t *stage_range);
 void av1_iadst16_new_sse4_1(const __m128i *input, __m128i *output,
                             const int8_t cos_bit, const int8_t *stage_range);
+
+void av1_idtx32_new_sse4_1(__m128i *input, __m128i *output, int cos_bit,
+                           const int col_num);
+
 static INLINE void transpose_32_4x4(int stride, const __m128i *input,
                                     __m128i *output) {
   __m128i temp0 = _mm_unpacklo_epi32(input[0 * stride], input[2 * stride]);
diff --git a/third_party/aom/av1/encoder/x86/corner_match_avx2.c b/third_party/aom/av1/encoder/x86/corner_match_avx2.c
new file mode 100644
index 0000000000..3351d0af19
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/corner_match_avx2.c
@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdlib.h>
+#include <memory.h>
+#include <math.h>
+#include <assert.h>
+
+#include <immintrin.h>
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/mem.h"
+#include "av1/encoder/corner_match.h"
+
+DECLARE_ALIGNED(16, static const uint8_t, byte_mask[16]) = {
+  255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0
+};
+#if MATCH_SZ != 13
+#error "Need to change byte_mask in corner_match_sse4.c if MATCH_SZ != 13"
+#endif
+
+/* Compute corr(im1, im2) * MATCH_SZ * stddev(im1), where the
+correlation/standard deviation are taken over MATCH_SZ by MATCH_SZ windows
+of each image, centered at (x1, y1) and (x2, y2) respectively.
+*/
+double compute_cross_correlation_avx2(unsigned char *im1, int stride1, int x1,
+                                      int y1, unsigned char *im2, int stride2,
+                                      int x2, int y2) {
+  int i, stride1_i = 0, stride2_i = 0;
+  __m256i temp1, sum_vec, sumsq2_vec, cross_vec, v, v1_1, v2_1;
+  const __m128i mask = _mm_load_si128((__m128i *)byte_mask);
+  const __m256i zero = _mm256_setzero_si256();
+  __m128i v1, v2;
+
+  sum_vec = zero;
+  sumsq2_vec = zero;
+  cross_vec = zero;
+
+  im1 += (y1 - MATCH_SZ_BY2) * stride1 + (x1 - MATCH_SZ_BY2);
+  im2 += (y2 - MATCH_SZ_BY2) * stride2 + (x2 - MATCH_SZ_BY2);
+
+  for (i = 0; i < MATCH_SZ; ++i) {
+    v1 = _mm_and_si128(_mm_loadu_si128((__m128i *)&im1[stride1_i]), mask);
+    v1_1 = _mm256_cvtepu8_epi16(v1);
+    v2 = _mm_and_si128(_mm_loadu_si128((__m128i *)&im2[stride2_i]), mask);
+    v2_1 = _mm256_cvtepu8_epi16(v2);
+
+    v = _mm256_insertf128_si256(_mm256_castsi128_si256(v1), v2, 1);
+    sumsq2_vec = _mm256_add_epi32(sumsq2_vec, _mm256_madd_epi16(v2_1, v2_1));
+
+    sum_vec = _mm256_add_epi16(sum_vec, _mm256_sad_epu8(v, zero));
+    cross_vec = _mm256_add_epi32(cross_vec, _mm256_madd_epi16(v1_1, v2_1));
+    stride1_i += stride1;
+    stride2_i += stride2;
+  }
+  __m256i sum_vec1 = _mm256_srli_si256(sum_vec, 8);
+  sum_vec = _mm256_add_epi32(sum_vec, sum_vec1);
+  int sum1_acc = _mm_cvtsi128_si32(_mm256_castsi256_si128(sum_vec));
+  int sum2_acc = _mm256_extract_epi32(sum_vec, 4);
+
+  __m256i unp_low = _mm256_unpacklo_epi64(sumsq2_vec, cross_vec);
+  __m256i unp_hig = _mm256_unpackhi_epi64(sumsq2_vec, cross_vec);
+  temp1 = _mm256_add_epi32(unp_low, unp_hig);
+
+  __m128i low_sumsq = _mm256_castsi256_si128(temp1);
+  low_sumsq = _mm_add_epi32(low_sumsq, _mm256_extractf128_si256(temp1, 1));
+  low_sumsq = _mm_add_epi32(low_sumsq, _mm_srli_epi64(low_sumsq, 32));
+  int sumsq2_acc = _mm_cvtsi128_si32(low_sumsq);
+  int cross_acc = _mm_extract_epi32(low_sumsq, 2);
+
+  int var2 = sumsq2_acc * MATCH_SZ_SQ - sum2_acc * sum2_acc;
+  int cov = cross_acc * MATCH_SZ_SQ - sum1_acc * sum2_acc;
+  return cov / sqrt((double)var2);
+}
diff --git a/third_party/aom/av1/encoder/x86/encodetxb_avx2.c b/third_party/aom/av1/encoder/x86/encodetxb_avx2.c
index 7642f57d18..2621301859 100644
--- a/third_party/aom/av1/encoder/x86/encodetxb_avx2.c
+++ b/third_party/aom/av1/encoder/x86/encodetxb_avx2.c
@@ -26,14 +26,6 @@ void av1_txb_init_levels_avx2(const tran_low_t *const coeff, const int width,
   const int stride = width + TX_PAD_HOR;
   const __m256i y_zeros = _mm256_setzero_si256();
 
-  const int32_t pre_len = sizeof(*levels) * TX_PAD_TOP * stride;
-  uint8_t *pre_buf = levels - TX_PAD_TOP * stride;
-  uint8_t *pre_buf_end = pre_buf + pre_len;
-  do {
-    yy_storeu_256(pre_buf, y_zeros);
-    pre_buf += 32;
-  } while (pre_buf < pre_buf_end);
-
   const int32_t bottom_len = sizeof(*levels) * (TX_PAD_BOTTOM * stride);
   uint8_t *bottom_buf_end = levels + (height + TX_PAD_BOTTOM) * stride;
   uint8_t *bottom_buf = bottom_buf_end - ((bottom_len + 31) & (~31));
diff --git a/third_party/aom/av1/encoder/x86/encodetxb_sse4.c b/third_party/aom/av1/encoder/x86/encodetxb_sse4.c
index 5e0687cd38..34c9e4f266 100644
--- a/third_party/aom/av1/encoder/x86/encodetxb_sse4.c
+++ b/third_party/aom/av1/encoder/x86/encodetxb_sse4.c
@@ -23,14 +23,6 @@ void av1_txb_init_levels_sse4_1(const tran_low_t *const coeff, const int width,
   const int stride = width + TX_PAD_HOR;
   const __m128i zeros = _mm_setzero_si128();
 
-  const int32_t pre_len = sizeof(*levels) * TX_PAD_TOP * stride;
-  uint8_t *pre_buf = levels - TX_PAD_TOP * stride;
-  uint8_t *pre_buf_end = pre_buf + pre_len;
-  do {
-    _mm_storeu_si128((__m128i *)(pre_buf), zeros);
-    pre_buf += 16;
-  } while (pre_buf < pre_buf_end);
-
   const int32_t bottom_len = sizeof(*levels) * (TX_PAD_BOTTOM * stride);
   uint8_t *bottom_buf = levels + stride * height;
   uint8_t *bottom_buf_end = bottom_buf + bottom_len;
diff --git a/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_avx2.c b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_avx2.c
new file mode 100644
index 0000000000..24c513feaf
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_avx2.c
@@ -0,0 +1,3170 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include <assert.h>
+#include <immintrin.h> /*AVX2*/
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+#include "av1/common/av1_txfm.h"
+#include "av1/encoder/av1_fwd_txfm1d_cfg.h"
+#include "aom_dsp/txfm_common.h"
+#include "aom_ports/mem.h"
+#include "aom_dsp/x86/txfm_common_sse2.h"
+#include "aom_dsp/x86/txfm_common_avx2.h"
+
+static INLINE void av1_load_buffer_8x8_avx2(const int16_t *input, __m256i *out,
+                                            int stride, int flipud, int fliplr,
+                                            int shift) {
+  __m128i out1[8];
+  if (!flipud) {
+    out1[0] = _mm_load_si128((const __m128i *)(input + 0 * stride));
+    out1[1] = _mm_load_si128((const __m128i *)(input + 1 * stride));
+    out1[2] = _mm_load_si128((const __m128i *)(input + 2 * stride));
+    out1[3] = _mm_load_si128((const __m128i *)(input + 3 * stride));
+    out1[4] = _mm_load_si128((const __m128i *)(input + 4 * stride));
+    out1[5] = _mm_load_si128((const __m128i *)(input + 5 * stride));
+    out1[6] = _mm_load_si128((const __m128i *)(input + 6 * stride));
+    out1[7] = _mm_load_si128((const __m128i *)(input + 7 * stride));
+
+  } else {
+    out1[7] = _mm_load_si128((const __m128i *)(input + 0 * stride));
+    out1[6] = _mm_load_si128((const __m128i *)(input + 1 * stride));
+    out1[5] = _mm_load_si128((const __m128i *)(input + 2 * stride));
+    out1[4] = _mm_load_si128((const __m128i *)(input + 3 * stride));
+    out1[3] = _mm_load_si128((const __m128i *)(input + 4 * stride));
+    out1[2] = _mm_load_si128((const __m128i *)(input + 5 * stride));
+    out1[1] = _mm_load_si128((const __m128i *)(input + 6 * stride));
+    out1[0] = _mm_load_si128((const __m128i *)(input + 7 * stride));
+  }
+  if (!fliplr) {
+    out[0] = _mm256_cvtepi16_epi32(out1[0]);
+    out[1] = _mm256_cvtepi16_epi32(out1[1]);
+    out[2] = _mm256_cvtepi16_epi32(out1[2]);
+    out[3] = _mm256_cvtepi16_epi32(out1[3]);
+    out[4] = _mm256_cvtepi16_epi32(out1[4]);
+    out[5] = _mm256_cvtepi16_epi32(out1[5]);
+    out[6] = _mm256_cvtepi16_epi32(out1[6]);
+    out[7] = _mm256_cvtepi16_epi32(out1[7]);
+
+  } else {
+    out[0] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[0]));
+    out[1] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[1]));
+    out[2] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[2]));
+    out[3] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[3]));
+    out[4] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[4]));
+    out[5] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[5]));
+    out[6] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[6]));
+    out[7] = _mm256_cvtepi16_epi32(mm_reverse_epi16(out1[7]));
+  }
+  out[0] = _mm256_slli_epi32(out[0], shift);
+  out[1] = _mm256_slli_epi32(out[1], shift);
+  out[2] = _mm256_slli_epi32(out[2], shift);
+  out[3] = _mm256_slli_epi32(out[3], shift);
+  out[4] = _mm256_slli_epi32(out[4], shift);
+  out[5] = _mm256_slli_epi32(out[5], shift);
+  out[6] = _mm256_slli_epi32(out[6], shift);
+  out[7] = _mm256_slli_epi32(out[7], shift);
+}
+static INLINE void col_txfm_8x8_rounding(__m256i *in, int shift) {
+  const __m256i rounding = _mm256_set1_epi32(1 << (shift - 1));
+
+  in[0] = _mm256_add_epi32(in[0], rounding);
+  in[1] = _mm256_add_epi32(in[1], rounding);
+  in[2] = _mm256_add_epi32(in[2], rounding);
+  in[3] = _mm256_add_epi32(in[3], rounding);
+  in[4] = _mm256_add_epi32(in[4], rounding);
+  in[5] = _mm256_add_epi32(in[5], rounding);
+  in[6] = _mm256_add_epi32(in[6], rounding);
+  in[7] = _mm256_add_epi32(in[7], rounding);
+
+  in[0] = _mm256_srai_epi32(in[0], shift);
+  in[1] = _mm256_srai_epi32(in[1], shift);
+  in[2] = _mm256_srai_epi32(in[2], shift);
+  in[3] = _mm256_srai_epi32(in[3], shift);
+  in[4] = _mm256_srai_epi32(in[4], shift);
+  in[5] = _mm256_srai_epi32(in[5], shift);
+  in[6] = _mm256_srai_epi32(in[6], shift);
+  in[7] = _mm256_srai_epi32(in[7], shift);
+}
+static INLINE void av1_load_buffer_8x16_avx2(const int16_t *input, __m256i *out,
+                                             int stride, int flipud, int fliplr,
+                                             int shift) {
+  const int16_t *topL = input;
+  const int16_t *botL = input + 8 * stride;
+
+  const int16_t *tmp;
+
+  if (flipud) {
+    tmp = topL;
+    topL = botL;
+    botL = tmp;
+  }
+  av1_load_buffer_8x8_avx2(topL, out, stride, flipud, fliplr, shift);
+  av1_load_buffer_8x8_avx2(botL, out + 8, stride, flipud, fliplr, shift);
+}
+static INLINE void av1_load_buffer_16xn_avx2(const int16_t *input, __m256i *out,
+                                             int stride, int height,
+                                             int outstride, int flipud,
+                                             int fliplr) {
+  __m256i out1[64];
+  if (!flipud) {
+    for (int i = 0; i < height; i++) {
+      out1[i] = _mm256_loadu_si256((const __m256i *)(input + i * stride));
+    }
+  } else {
+    for (int i = 0; i < height; i++) {
+      out1[(height - 1) - i] =
+          _mm256_loadu_si256((const __m256i *)(input + i * stride));
+    }
+  }
+  if (!fliplr) {
+    for (int i = 0; i < height; i++) {
+      out[i * outstride] =
+          _mm256_cvtepi16_epi32(_mm256_castsi256_si128(out1[i]));
+      out[i * outstride + 1] =
+          _mm256_cvtepi16_epi32(_mm256_extractf128_si256(out1[i], 1));
+    }
+  } else {
+    for (int i = 0; i < height; i++) {
+      out[i * outstride + 1] = _mm256_cvtepi16_epi32(
+          mm_reverse_epi16(_mm256_castsi256_si128(out1[i])));
+      out[i * outstride + 0] = _mm256_cvtepi16_epi32(
+          mm_reverse_epi16(_mm256_extractf128_si256(out1[i], 1)));
+    }
+  }
+}
+
+static void av1_fwd_txfm_transpose_8x8_avx2(const __m256i *in, __m256i *out,
+                                            const int instride,
+                                            const int outstride) {
+  __m256i u0, u1, u2, u3, u4, u5, u6, u7;
+  __m256i x0, x1;
+
+  u0 = _mm256_unpacklo_epi32(in[0 * instride], in[1 * instride]);
+  u1 = _mm256_unpackhi_epi32(in[0 * instride], in[1 * instride]);
+
+  u2 = _mm256_unpacklo_epi32(in[2 * instride], in[3 * instride]);
+  u3 = _mm256_unpackhi_epi32(in[2 * instride], in[3 * instride]);
+
+  u4 = _mm256_unpacklo_epi32(in[4 * instride], in[5 * instride]);
+  u5 = _mm256_unpackhi_epi32(in[4 * instride], in[5 * instride]);
+
+  u6 = _mm256_unpacklo_epi32(in[6 * instride], in[7 * instride]);
+  u7 = _mm256_unpackhi_epi32(in[6 * instride], in[7 * instride]);
+
+  x0 = _mm256_unpacklo_epi64(u0, u2);
+  x1 = _mm256_unpacklo_epi64(u4, u6);
+  out[0 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[4 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpackhi_epi64(u0, u2);
+  x1 = _mm256_unpackhi_epi64(u4, u6);
+  out[1 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[5 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpacklo_epi64(u1, u3);
+  x1 = _mm256_unpacklo_epi64(u5, u7);
+  out[2 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[6 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x31);
+
+  x0 = _mm256_unpackhi_epi64(u1, u3);
+  x1 = _mm256_unpackhi_epi64(u5, u7);
+  out[3 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x20);
+  out[7 * outstride] = _mm256_permute2f128_si256(x0, x1, 0x31);
+}
+static INLINE void av1_round_shift_32_8xn_avx2(__m256i *in, int size, int bit,
+                                               int stride) {
+  if (bit < 0) {
+    bit = -bit;
+    __m256i round = _mm256_set1_epi32(1 << (bit - 1));
+    for (int i = 0; i < size; ++i) {
+      in[stride * i] = _mm256_add_epi32(in[stride * i], round);
+      in[stride * i] = _mm256_srai_epi32(in[stride * i], bit);
+    }
+  } else if (bit > 0) {
+    for (int i = 0; i < size; ++i) {
+      in[stride * i] = _mm256_slli_epi32(in[stride * i], bit);
+    }
+  }
+}
+static INLINE void av1_store_buffer_avx2(const __m256i *const in, int32_t *out,
+                                         const int stride, const int out_size) {
+  for (int i = 0; i < out_size; ++i) {
+    _mm256_store_si256((__m256i *)(out), in[i]);
+    out += stride;
+  }
+}
+static INLINE void av1_fwd_txfm_transpose_16x16_avx2(const __m256i *in,
+                                                     __m256i *out) {
+  av1_fwd_txfm_transpose_8x8_avx2(&in[0], &out[0], 2, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(&in[1], &out[16], 2, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(&in[16], &out[1], 2, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(&in[17], &out[17], 2, 2);
+}
+
+static INLINE __m256i av1_half_btf_avx2(const __m256i *w0, const __m256i *n0,
+                                        const __m256i *w1, const __m256i *n1,
+                                        const __m256i *rounding, int bit) {
+  __m256i x, y;
+
+  x = _mm256_mullo_epi32(*w0, *n0);
+  y = _mm256_mullo_epi32(*w1, *n1);
+  x = _mm256_add_epi32(x, y);
+  x = _mm256_add_epi32(x, *rounding);
+  x = _mm256_srai_epi32(x, bit);
+  return x;
+}
+#define btf_32_avx2_type0(w0, w1, in0, in1, out0, out1, bit) \
+  do {                                                       \
+    const __m256i ww0 = _mm256_set1_epi32(w0);               \
+    const __m256i ww1 = _mm256_set1_epi32(w1);               \
+    const __m256i in0_w0 = _mm256_mullo_epi32(in0, ww0);     \
+    const __m256i in1_w1 = _mm256_mullo_epi32(in1, ww1);     \
+    out0 = _mm256_add_epi32(in0_w0, in1_w1);                 \
+    av1_round_shift_32_8xn_avx2(&out0, 1, -bit, 1);          \
+    const __m256i in0_w1 = _mm256_mullo_epi32(in0, ww1);     \
+    const __m256i in1_w0 = _mm256_mullo_epi32(in1, ww0);     \
+    out1 = _mm256_sub_epi32(in0_w1, in1_w0);                 \
+    av1_round_shift_32_8xn_avx2(&out1, 1, -bit, 1);          \
+  } while (0)
+
+#define btf_32_type0_avx2_new(ww0, ww1, in0, in1, out0, out1, r, bit) \
+  do {                                                                \
+    const __m256i in0_w0 = _mm256_mullo_epi32(in0, ww0);              \
+    const __m256i in1_w1 = _mm256_mullo_epi32(in1, ww1);              \
+    out0 = _mm256_add_epi32(in0_w0, in1_w1);                          \
+    out0 = _mm256_add_epi32(out0, r);                                 \
+    out0 = _mm256_srai_epi32(out0, bit);                              \
+    const __m256i in0_w1 = _mm256_mullo_epi32(in0, ww1);              \
+    const __m256i in1_w0 = _mm256_mullo_epi32(in1, ww0);              \
+    out1 = _mm256_sub_epi32(in0_w1, in1_w0);                          \
+    out1 = _mm256_add_epi32(out1, r);                                 \
+    out1 = _mm256_srai_epi32(out1, bit);                              \
+  } while (0)
+
+typedef void (*transform_1d_avx2)(__m256i *in, __m256i *out,
+                                  const int8_t cos_bit, int instride,
+                                  int outstride);
+static void av1_fdct8_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                           const int col_num, const int outstride) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospim32 = _mm256_set1_epi32(-cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  __m256i u[8], v[8];
+  for (int col = 0; col < col_num; ++col) {
+    u[0] = _mm256_add_epi32(in[0 * col_num + col], in[7 * col_num + col]);
+    v[7] = _mm256_sub_epi32(in[0 * col_num + col], in[7 * col_num + col]);
+    u[1] = _mm256_add_epi32(in[1 * col_num + col], in[6 * col_num + col]);
+    u[6] = _mm256_sub_epi32(in[1 * col_num + col], in[6 * col_num + col]);
+    u[2] = _mm256_add_epi32(in[2 * col_num + col], in[5 * col_num + col]);
+    u[5] = _mm256_sub_epi32(in[2 * col_num + col], in[5 * col_num + col]);
+    u[3] = _mm256_add_epi32(in[3 * col_num + col], in[4 * col_num + col]);
+    v[4] = _mm256_sub_epi32(in[3 * col_num + col], in[4 * col_num + col]);
+    v[0] = _mm256_add_epi32(u[0], u[3]);
+    v[3] = _mm256_sub_epi32(u[0], u[3]);
+    v[1] = _mm256_add_epi32(u[1], u[2]);
+    v[2] = _mm256_sub_epi32(u[1], u[2]);
+
+    v[5] = _mm256_mullo_epi32(u[5], cospim32);
+    v[6] = _mm256_mullo_epi32(u[6], cospi32);
+    v[5] = _mm256_add_epi32(v[5], v[6]);
+    v[5] = _mm256_add_epi32(v[5], rnding);
+    v[5] = _mm256_srai_epi32(v[5], bit);
+
+    u[0] = _mm256_mullo_epi32(u[5], cospi32);
+    v[6] = _mm256_mullo_epi32(u[6], cospim32);
+    v[6] = _mm256_sub_epi32(u[0], v[6]);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    // stage 3
+    // type 0
+    v[0] = _mm256_mullo_epi32(v[0], cospi32);
+    v[1] = _mm256_mullo_epi32(v[1], cospi32);
+    u[0] = _mm256_add_epi32(v[0], v[1]);
+    u[0] = _mm256_add_epi32(u[0], rnding);
+    u[0] = _mm256_srai_epi32(u[0], bit);
+
+    u[1] = _mm256_sub_epi32(v[0], v[1]);
+    u[1] = _mm256_add_epi32(u[1], rnding);
+    u[1] = _mm256_srai_epi32(u[1], bit);
+
+    // type 1
+    v[0] = _mm256_mullo_epi32(v[2], cospi48);
+    v[1] = _mm256_mullo_epi32(v[3], cospi16);
+    u[2] = _mm256_add_epi32(v[0], v[1]);
+    u[2] = _mm256_add_epi32(u[2], rnding);
+    u[2] = _mm256_srai_epi32(u[2], bit);
+
+    v[0] = _mm256_mullo_epi32(v[2], cospi16);
+    v[1] = _mm256_mullo_epi32(v[3], cospi48);
+    u[3] = _mm256_sub_epi32(v[1], v[0]);
+    u[3] = _mm256_add_epi32(u[3], rnding);
+    u[3] = _mm256_srai_epi32(u[3], bit);
+
+    u[4] = _mm256_add_epi32(v[4], v[5]);
+    u[5] = _mm256_sub_epi32(v[4], v[5]);
+    u[6] = _mm256_sub_epi32(v[7], v[6]);
+    u[7] = _mm256_add_epi32(v[7], v[6]);
+
+    // stage 4
+    // stage 5
+    v[0] = _mm256_mullo_epi32(u[4], cospi56);
+    v[1] = _mm256_mullo_epi32(u[7], cospi8);
+    v[0] = _mm256_add_epi32(v[0], v[1]);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    out[1 * outstride + col] = _mm256_srai_epi32(v[0], bit);  // buf0[4]
+
+    v[0] = _mm256_mullo_epi32(u[4], cospi8);
+    v[1] = _mm256_mullo_epi32(u[7], cospi56);
+    v[0] = _mm256_sub_epi32(v[1], v[0]);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    out[7 * outstride + col] = _mm256_srai_epi32(v[0], bit);  // buf0[7]
+
+    v[0] = _mm256_mullo_epi32(u[5], cospi24);
+    v[1] = _mm256_mullo_epi32(u[6], cospi40);
+    v[0] = _mm256_add_epi32(v[0], v[1]);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    out[5 * outstride + col] = _mm256_srai_epi32(v[0], bit);  // buf0[5]
+
+    v[0] = _mm256_mullo_epi32(u[5], cospi40);
+    v[1] = _mm256_mullo_epi32(u[6], cospi24);
+    v[0] = _mm256_sub_epi32(v[1], v[0]);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    out[3 * outstride + col] = _mm256_srai_epi32(v[0], bit);  // buf0[6]
+
+    out[0 * outstride + col] = u[0];  // buf0[0]
+    out[4 * outstride + col] = u[1];  // buf0[1]
+    out[2 * outstride + col] = u[2];  // buf0[2]
+    out[6 * outstride + col] = u[3];  // buf0[3]
+  }
+}
+static void av1_fadst8_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                            const int col_num, const int outstirde) {
+  (void)col_num;
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospim4 = _mm256_set1_epi32(-cospi[4]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospim20 = _mm256_set1_epi32(-cospi[20]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospim36 = _mm256_set1_epi32(-cospi[36]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i cospim52 = _mm256_set1_epi32(-cospi[52]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const __m256i zero = _mm256_setzero_si256();
+  __m256i u0, u1, u2, u3, u4, u5, u6, u7;
+  __m256i v0, v1, v2, v3, v4, v5, v6, v7;
+  __m256i x, y;
+  for (int col = 0; col < col_num; ++col) {
+    u0 = in[0 * col_num + col];
+    u1 = _mm256_sub_epi32(zero, in[7 * col_num + col]);
+    u2 = _mm256_sub_epi32(zero, in[3 * col_num + col]);
+    u3 = in[4 * col_num + col];
+    u4 = _mm256_sub_epi32(zero, in[1 * col_num + col]);
+    u5 = in[6 * col_num + col];
+    u6 = in[2 * col_num + col];
+    u7 = _mm256_sub_epi32(zero, in[5 * col_num + col]);
+
+    // stage 2
+    v0 = u0;
+    v1 = u1;
+
+    x = _mm256_mullo_epi32(u2, cospi32);
+    y = _mm256_mullo_epi32(u3, cospi32);
+    v2 = _mm256_add_epi32(x, y);
+    v2 = _mm256_add_epi32(v2, rnding);
+    v2 = _mm256_srai_epi32(v2, bit);
+
+    v3 = _mm256_sub_epi32(x, y);
+    v3 = _mm256_add_epi32(v3, rnding);
+    v3 = _mm256_srai_epi32(v3, bit);
+
+    v4 = u4;
+    v5 = u5;
+
+    x = _mm256_mullo_epi32(u6, cospi32);
+    y = _mm256_mullo_epi32(u7, cospi32);
+    v6 = _mm256_add_epi32(x, y);
+    v6 = _mm256_add_epi32(v6, rnding);
+    v6 = _mm256_srai_epi32(v6, bit);
+
+    v7 = _mm256_sub_epi32(x, y);
+    v7 = _mm256_add_epi32(v7, rnding);
+    v7 = _mm256_srai_epi32(v7, bit);
+
+    // stage 3
+    u0 = _mm256_add_epi32(v0, v2);
+    u1 = _mm256_add_epi32(v1, v3);
+    u2 = _mm256_sub_epi32(v0, v2);
+    u3 = _mm256_sub_epi32(v1, v3);
+    u4 = _mm256_add_epi32(v4, v6);
+    u5 = _mm256_add_epi32(v5, v7);
+    u6 = _mm256_sub_epi32(v4, v6);
+    u7 = _mm256_sub_epi32(v5, v7);
+
+    // stage 4
+    v0 = u0;
+    v1 = u1;
+    v2 = u2;
+    v3 = u3;
+
+    x = _mm256_mullo_epi32(u4, cospi16);
+    y = _mm256_mullo_epi32(u5, cospi48);
+    v4 = _mm256_add_epi32(x, y);
+    v4 = _mm256_add_epi32(v4, rnding);
+    v4 = _mm256_srai_epi32(v4, bit);
+
+    x = _mm256_mullo_epi32(u4, cospi48);
+    y = _mm256_mullo_epi32(u5, cospim16);
+    v5 = _mm256_add_epi32(x, y);
+    v5 = _mm256_add_epi32(v5, rnding);
+    v5 = _mm256_srai_epi32(v5, bit);
+
+    x = _mm256_mullo_epi32(u6, cospim48);
+    y = _mm256_mullo_epi32(u7, cospi16);
+    v6 = _mm256_add_epi32(x, y);
+    v6 = _mm256_add_epi32(v6, rnding);
+    v6 = _mm256_srai_epi32(v6, bit);
+
+    x = _mm256_mullo_epi32(u6, cospi16);
+    y = _mm256_mullo_epi32(u7, cospi48);
+    v7 = _mm256_add_epi32(x, y);
+    v7 = _mm256_add_epi32(v7, rnding);
+    v7 = _mm256_srai_epi32(v7, bit);
+
+    // stage 5
+    u0 = _mm256_add_epi32(v0, v4);
+    u1 = _mm256_add_epi32(v1, v5);
+    u2 = _mm256_add_epi32(v2, v6);
+    u3 = _mm256_add_epi32(v3, v7);
+    u4 = _mm256_sub_epi32(v0, v4);
+    u5 = _mm256_sub_epi32(v1, v5);
+    u6 = _mm256_sub_epi32(v2, v6);
+    u7 = _mm256_sub_epi32(v3, v7);
+
+    // stage 6
+    x = _mm256_mullo_epi32(u0, cospi4);
+    y = _mm256_mullo_epi32(u1, cospi60);
+    v0 = _mm256_add_epi32(x, y);
+    v0 = _mm256_add_epi32(v0, rnding);
+    v0 = _mm256_srai_epi32(v0, bit);
+
+    x = _mm256_mullo_epi32(u0, cospi60);
+    y = _mm256_mullo_epi32(u1, cospim4);
+    v1 = _mm256_add_epi32(x, y);
+    v1 = _mm256_add_epi32(v1, rnding);
+    v1 = _mm256_srai_epi32(v1, bit);
+
+    x = _mm256_mullo_epi32(u2, cospi20);
+    y = _mm256_mullo_epi32(u3, cospi44);
+    v2 = _mm256_add_epi32(x, y);
+    v2 = _mm256_add_epi32(v2, rnding);
+    v2 = _mm256_srai_epi32(v2, bit);
+
+    x = _mm256_mullo_epi32(u2, cospi44);
+    y = _mm256_mullo_epi32(u3, cospim20);
+    v3 = _mm256_add_epi32(x, y);
+    v3 = _mm256_add_epi32(v3, rnding);
+    v3 = _mm256_srai_epi32(v3, bit);
+
+    x = _mm256_mullo_epi32(u4, cospi36);
+    y = _mm256_mullo_epi32(u5, cospi28);
+    v4 = _mm256_add_epi32(x, y);
+    v4 = _mm256_add_epi32(v4, rnding);
+    v4 = _mm256_srai_epi32(v4, bit);
+
+    x = _mm256_mullo_epi32(u4, cospi28);
+    y = _mm256_mullo_epi32(u5, cospim36);
+    v5 = _mm256_add_epi32(x, y);
+    v5 = _mm256_add_epi32(v5, rnding);
+    v5 = _mm256_srai_epi32(v5, bit);
+
+    x = _mm256_mullo_epi32(u6, cospi52);
+    y = _mm256_mullo_epi32(u7, cospi12);
+    v6 = _mm256_add_epi32(x, y);
+    v6 = _mm256_add_epi32(v6, rnding);
+    v6 = _mm256_srai_epi32(v6, bit);
+
+    x = _mm256_mullo_epi32(u6, cospi12);
+    y = _mm256_mullo_epi32(u7, cospim52);
+    v7 = _mm256_add_epi32(x, y);
+    v7 = _mm256_add_epi32(v7, rnding);
+    v7 = _mm256_srai_epi32(v7, bit);
+
+    // stage 7
+    out[0 * outstirde + col] = v1;
+    out[1 * outstirde + col] = v6;
+    out[2 * outstirde + col] = v3;
+    out[3 * outstirde + col] = v4;
+    out[4 * outstirde + col] = v5;
+    out[5 * outstirde + col] = v2;
+    out[6 * outstirde + col] = v7;
+    out[7 * outstirde + col] = v0;
+  }
+}
+static void av1_idtx8_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                           int col_num, int outstride) {
+  (void)bit;
+  (void)outstride;
+  int num_iters = 8 * col_num;
+  for (int i = 0; i < num_iters; i += 8) {
+    out[i] = _mm256_add_epi32(in[i], in[i]);
+    out[i + 1] = _mm256_add_epi32(in[i + 1], in[i + 1]);
+    out[i + 2] = _mm256_add_epi32(in[i + 2], in[i + 2]);
+    out[i + 3] = _mm256_add_epi32(in[i + 3], in[i + 3]);
+    out[i + 4] = _mm256_add_epi32(in[i + 4], in[i + 4]);
+    out[i + 5] = _mm256_add_epi32(in[i + 5], in[i + 5]);
+    out[i + 6] = _mm256_add_epi32(in[i + 6], in[i + 6]);
+    out[i + 7] = _mm256_add_epi32(in[i + 7], in[i + 7]);
+  }
+}
+void av1_fwd_txfm2d_8x8_avx2(const int16_t *input, int32_t *coeff, int stride,
+                             TX_TYPE tx_type, int bd) {
+  __m256i in[8], out[8];
+  const TX_SIZE tx_size = TX_8X8;
+  const int8_t *shift = fwd_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int width = tx_size_wide[tx_size];
+  const int width_div8 = (width >> 3);
+
+  switch (tx_type) {
+    case DCT_DCT:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case ADST_DCT:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case DCT_ADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case ADST_ADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case FLIPADST_DCT:
+      av1_load_buffer_8x8_avx2(input, in, stride, 1, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case DCT_FLIPADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 1, shift[0]);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case FLIPADST_FLIPADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 1, 1, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case ADST_FLIPADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 1, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case FLIPADST_ADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 1, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case IDTX:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_idtx8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_idtx8_avx2(out, in, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case V_DCT:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_idtx8_avx2(out, in, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case H_DCT:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_idtx8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fdct8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case V_ADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_idtx8_avx2(out, in, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case H_ADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 0, shift[0]);
+      av1_idtx8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case V_FLIPADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 1, 0, shift[0]);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_idtx8_avx2(out, in, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    case H_FLIPADST:
+      av1_load_buffer_8x8_avx2(input, in, stride, 0, 1, shift[0]);
+      av1_idtx8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                     width_div8);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_fadst8_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_8x8_avx2(out, in, width_div8, width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 8);
+      break;
+    default: assert(0);
+  }
+  (void)bd;
+}
+
+static void av1_fdct16_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                            const int col_num, const int outstride) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospim32 = _mm256_set1_epi32(-cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi60 = _mm256_set1_epi32(cospi[60]);
+  const __m256i cospi4 = _mm256_set1_epi32(cospi[4]);
+  const __m256i cospi28 = _mm256_set1_epi32(cospi[28]);
+  const __m256i cospi36 = _mm256_set1_epi32(cospi[36]);
+  const __m256i cospi44 = _mm256_set1_epi32(cospi[44]);
+  const __m256i cospi20 = _mm256_set1_epi32(cospi[20]);
+  const __m256i cospi12 = _mm256_set1_epi32(cospi[12]);
+  const __m256i cospi52 = _mm256_set1_epi32(cospi[52]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  __m256i u[16], v[16], x;
+  int col;
+
+  // Calculate the column 0, 1, 2, 3
+  for (col = 0; col < col_num; ++col) {
+    // stage 0
+    // stage 1
+    u[0] = _mm256_add_epi32(in[0 * col_num + col], in[15 * col_num + col]);
+    u[15] = _mm256_sub_epi32(in[0 * col_num + col], in[15 * col_num + col]);
+    u[1] = _mm256_add_epi32(in[1 * col_num + col], in[14 * col_num + col]);
+    u[14] = _mm256_sub_epi32(in[1 * col_num + col], in[14 * col_num + col]);
+    u[2] = _mm256_add_epi32(in[2 * col_num + col], in[13 * col_num + col]);
+    u[13] = _mm256_sub_epi32(in[2 * col_num + col], in[13 * col_num + col]);
+    u[3] = _mm256_add_epi32(in[3 * col_num + col], in[12 * col_num + col]);
+    u[12] = _mm256_sub_epi32(in[3 * col_num + col], in[12 * col_num + col]);
+    u[4] = _mm256_add_epi32(in[4 * col_num + col], in[11 * col_num + col]);
+    u[11] = _mm256_sub_epi32(in[4 * col_num + col], in[11 * col_num + col]);
+    u[5] = _mm256_add_epi32(in[5 * col_num + col], in[10 * col_num + col]);
+    u[10] = _mm256_sub_epi32(in[5 * col_num + col], in[10 * col_num + col]);
+    u[6] = _mm256_add_epi32(in[6 * col_num + col], in[9 * col_num + col]);
+    u[9] = _mm256_sub_epi32(in[6 * col_num + col], in[9 * col_num + col]);
+    u[7] = _mm256_add_epi32(in[7 * col_num + col], in[8 * col_num + col]);
+    u[8] = _mm256_sub_epi32(in[7 * col_num + col], in[8 * col_num + col]);
+
+    // stage 2
+    v[0] = _mm256_add_epi32(u[0], u[7]);
+    v[7] = _mm256_sub_epi32(u[0], u[7]);
+    v[1] = _mm256_add_epi32(u[1], u[6]);
+    v[6] = _mm256_sub_epi32(u[1], u[6]);
+    v[2] = _mm256_add_epi32(u[2], u[5]);
+    v[5] = _mm256_sub_epi32(u[2], u[5]);
+    v[3] = _mm256_add_epi32(u[3], u[4]);
+    v[4] = _mm256_sub_epi32(u[3], u[4]);
+    v[8] = u[8];
+    v[9] = u[9];
+
+    v[10] = _mm256_mullo_epi32(u[10], cospim32);
+    x = _mm256_mullo_epi32(u[13], cospi32);
+    v[10] = _mm256_add_epi32(v[10], x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[13] = _mm256_mullo_epi32(u[10], cospi32);
+    x = _mm256_mullo_epi32(u[13], cospim32);
+    v[13] = _mm256_sub_epi32(v[13], x);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[11] = _mm256_mullo_epi32(u[11], cospim32);
+    x = _mm256_mullo_epi32(u[12], cospi32);
+    v[11] = _mm256_add_epi32(v[11], x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = _mm256_mullo_epi32(u[11], cospi32);
+    x = _mm256_mullo_epi32(u[12], cospim32);
+    v[12] = _mm256_sub_epi32(v[12], x);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+    v[14] = u[14];
+    v[15] = u[15];
+
+    // stage 3
+    u[0] = _mm256_add_epi32(v[0], v[3]);
+    u[3] = _mm256_sub_epi32(v[0], v[3]);
+    u[1] = _mm256_add_epi32(v[1], v[2]);
+    u[2] = _mm256_sub_epi32(v[1], v[2]);
+    u[4] = v[4];
+
+    u[5] = _mm256_mullo_epi32(v[5], cospim32);
+    x = _mm256_mullo_epi32(v[6], cospi32);
+    u[5] = _mm256_add_epi32(u[5], x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    u[6] = _mm256_mullo_epi32(v[5], cospi32);
+    x = _mm256_mullo_epi32(v[6], cospim32);
+    u[6] = _mm256_sub_epi32(u[6], x);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    u[7] = v[7];
+    u[8] = _mm256_add_epi32(v[8], v[11]);
+    u[11] = _mm256_sub_epi32(v[8], v[11]);
+    u[9] = _mm256_add_epi32(v[9], v[10]);
+    u[10] = _mm256_sub_epi32(v[9], v[10]);
+    u[12] = _mm256_sub_epi32(v[15], v[12]);
+    u[15] = _mm256_add_epi32(v[15], v[12]);
+    u[13] = _mm256_sub_epi32(v[14], v[13]);
+    u[14] = _mm256_add_epi32(v[14], v[13]);
+
+    // stage 4
+    u[0] = _mm256_mullo_epi32(u[0], cospi32);
+    u[1] = _mm256_mullo_epi32(u[1], cospi32);
+    v[0] = _mm256_add_epi32(u[0], u[1]);
+    v[0] = _mm256_add_epi32(v[0], rnding);
+    v[0] = _mm256_srai_epi32(v[0], bit);
+
+    v[1] = _mm256_sub_epi32(u[0], u[1]);
+    v[1] = _mm256_add_epi32(v[1], rnding);
+    v[1] = _mm256_srai_epi32(v[1], bit);
+
+    v[2] = _mm256_mullo_epi32(u[2], cospi48);
+    x = _mm256_mullo_epi32(u[3], cospi16);
+    v[2] = _mm256_add_epi32(v[2], x);
+    v[2] = _mm256_add_epi32(v[2], rnding);
+    v[2] = _mm256_srai_epi32(v[2], bit);
+
+    v[3] = _mm256_mullo_epi32(u[2], cospi16);
+    x = _mm256_mullo_epi32(u[3], cospi48);
+    v[3] = _mm256_sub_epi32(x, v[3]);
+    v[3] = _mm256_add_epi32(v[3], rnding);
+    v[3] = _mm256_srai_epi32(v[3], bit);
+
+    v[4] = _mm256_add_epi32(u[4], u[5]);
+    v[5] = _mm256_sub_epi32(u[4], u[5]);
+    v[6] = _mm256_sub_epi32(u[7], u[6]);
+    v[7] = _mm256_add_epi32(u[7], u[6]);
+    v[8] = u[8];
+
+    v[9] = _mm256_mullo_epi32(u[9], cospim16);
+    x = _mm256_mullo_epi32(u[14], cospi48);
+    v[9] = _mm256_add_epi32(v[9], x);
+    v[9] = _mm256_add_epi32(v[9], rnding);
+    v[9] = _mm256_srai_epi32(v[9], bit);
+
+    v[14] = _mm256_mullo_epi32(u[9], cospi48);
+    x = _mm256_mullo_epi32(u[14], cospim16);
+    v[14] = _mm256_sub_epi32(v[14], x);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[10] = _mm256_mullo_epi32(u[10], cospim48);
+    x = _mm256_mullo_epi32(u[13], cospim16);
+    v[10] = _mm256_add_epi32(v[10], x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[13] = _mm256_mullo_epi32(u[10], cospim16);
+    x = _mm256_mullo_epi32(u[13], cospim48);
+    v[13] = _mm256_sub_epi32(v[13], x);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[11] = u[11];
+    v[12] = u[12];
+    v[15] = u[15];
+
+    // stage 5
+    u[0] = v[0];
+    u[1] = v[1];
+    u[2] = v[2];
+    u[3] = v[3];
+
+    u[4] = _mm256_mullo_epi32(v[4], cospi56);
+    x = _mm256_mullo_epi32(v[7], cospi8);
+    u[4] = _mm256_add_epi32(u[4], x);
+    u[4] = _mm256_add_epi32(u[4], rnding);
+    u[4] = _mm256_srai_epi32(u[4], bit);
+
+    u[7] = _mm256_mullo_epi32(v[4], cospi8);
+    x = _mm256_mullo_epi32(v[7], cospi56);
+    u[7] = _mm256_sub_epi32(x, u[7]);
+    u[7] = _mm256_add_epi32(u[7], rnding);
+    u[7] = _mm256_srai_epi32(u[7], bit);
+
+    u[5] = _mm256_mullo_epi32(v[5], cospi24);
+    x = _mm256_mullo_epi32(v[6], cospi40);
+    u[5] = _mm256_add_epi32(u[5], x);
+    u[5] = _mm256_add_epi32(u[5], rnding);
+    u[5] = _mm256_srai_epi32(u[5], bit);
+
+    u[6] = _mm256_mullo_epi32(v[5], cospi40);
+    x = _mm256_mullo_epi32(v[6], cospi24);
+    u[6] = _mm256_sub_epi32(x, u[6]);
+    u[6] = _mm256_add_epi32(u[6], rnding);
+    u[6] = _mm256_srai_epi32(u[6], bit);
+
+    u[8] = _mm256_add_epi32(v[8], v[9]);
+    u[9] = _mm256_sub_epi32(v[8], v[9]);
+    u[10] = _mm256_sub_epi32(v[11], v[10]);
+    u[11] = _mm256_add_epi32(v[11], v[10]);
+    u[12] = _mm256_add_epi32(v[12], v[13]);
+    u[13] = _mm256_sub_epi32(v[12], v[13]);
+    u[14] = _mm256_sub_epi32(v[15], v[14]);
+    u[15] = _mm256_add_epi32(v[15], v[14]);
+
+    // stage 6
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+    v[4] = u[4];
+    v[5] = u[5];
+    v[6] = u[6];
+    v[7] = u[7];
+
+    v[8] = _mm256_mullo_epi32(u[8], cospi60);
+    x = _mm256_mullo_epi32(u[15], cospi4);
+    v[8] = _mm256_add_epi32(v[8], x);
+    v[8] = _mm256_add_epi32(v[8], rnding);
+    v[8] = _mm256_srai_epi32(v[8], bit);
+
+    v[15] = _mm256_mullo_epi32(u[8], cospi4);
+    x = _mm256_mullo_epi32(u[15], cospi60);
+    v[15] = _mm256_sub_epi32(x, v[15]);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    v[9] = _mm256_mullo_epi32(u[9], cospi28);
+    x = _mm256_mullo_epi32(u[14], cospi36);
+    v[9] = _mm256_add_epi32(v[9], x);
+    v[9] = _mm256_add_epi32(v[9], rnding);
+    v[9] = _mm256_srai_epi32(v[9], bit);
+
+    v[14] = _mm256_mullo_epi32(u[9], cospi36);
+    x = _mm256_mullo_epi32(u[14], cospi28);
+    v[14] = _mm256_sub_epi32(x, v[14]);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[10] = _mm256_mullo_epi32(u[10], cospi44);
+    x = _mm256_mullo_epi32(u[13], cospi20);
+    v[10] = _mm256_add_epi32(v[10], x);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[13] = _mm256_mullo_epi32(u[10], cospi20);
+    x = _mm256_mullo_epi32(u[13], cospi44);
+    v[13] = _mm256_sub_epi32(x, v[13]);
+    v[13] = _mm256_add_epi32(v[13], rnding);
+    v[13] = _mm256_srai_epi32(v[13], bit);
+
+    v[11] = _mm256_mullo_epi32(u[11], cospi12);
+    x = _mm256_mullo_epi32(u[12], cospi52);
+    v[11] = _mm256_add_epi32(v[11], x);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = _mm256_mullo_epi32(u[11], cospi52);
+    x = _mm256_mullo_epi32(u[12], cospi12);
+    v[12] = _mm256_sub_epi32(x, v[12]);
+    v[12] = _mm256_add_epi32(v[12], rnding);
+    v[12] = _mm256_srai_epi32(v[12], bit);
+
+    out[0 * outstride + col] = v[0];
+    out[1 * outstride + col] = v[8];
+    out[2 * outstride + col] = v[4];
+    out[3 * outstride + col] = v[12];
+    out[4 * outstride + col] = v[2];
+    out[5 * outstride + col] = v[10];
+    out[6 * outstride + col] = v[6];
+    out[7 * outstride + col] = v[14];
+    out[8 * outstride + col] = v[1];
+    out[9 * outstride + col] = v[9];
+    out[10 * outstride + col] = v[5];
+    out[11 * outstride + col] = v[13];
+    out[12 * outstride + col] = v[3];
+    out[13 * outstride + col] = v[11];
+    out[14 * outstride + col] = v[7];
+    out[15 * outstride + col] = v[15];
+  }
+}
+static void av1_fadst16_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                             const int num_cols, const int outstride) {
+  const int32_t *cospi = cospi_arr(bit);
+  const __m256i cospi32 = _mm256_set1_epi32(cospi[32]);
+  const __m256i cospi48 = _mm256_set1_epi32(cospi[48]);
+  const __m256i cospi16 = _mm256_set1_epi32(cospi[16]);
+  const __m256i cospim16 = _mm256_set1_epi32(-cospi[16]);
+  const __m256i cospim48 = _mm256_set1_epi32(-cospi[48]);
+  const __m256i cospi8 = _mm256_set1_epi32(cospi[8]);
+  const __m256i cospi56 = _mm256_set1_epi32(cospi[56]);
+  const __m256i cospim56 = _mm256_set1_epi32(-cospi[56]);
+  const __m256i cospim8 = _mm256_set1_epi32(-cospi[8]);
+  const __m256i cospi24 = _mm256_set1_epi32(cospi[24]);
+  const __m256i cospim24 = _mm256_set1_epi32(-cospi[24]);
+  const __m256i cospim40 = _mm256_set1_epi32(-cospi[40]);
+  const __m256i cospi40 = _mm256_set1_epi32(cospi[40]);
+  const __m256i cospi2 = _mm256_set1_epi32(cospi[2]);
+  const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
+  const __m256i cospim2 = _mm256_set1_epi32(-cospi[2]);
+  const __m256i cospi10 = _mm256_set1_epi32(cospi[10]);
+  const __m256i cospi54 = _mm256_set1_epi32(cospi[54]);
+  const __m256i cospim10 = _mm256_set1_epi32(-cospi[10]);
+  const __m256i cospi18 = _mm256_set1_epi32(cospi[18]);
+  const __m256i cospi46 = _mm256_set1_epi32(cospi[46]);
+  const __m256i cospim18 = _mm256_set1_epi32(-cospi[18]);
+  const __m256i cospi26 = _mm256_set1_epi32(cospi[26]);
+  const __m256i cospi38 = _mm256_set1_epi32(cospi[38]);
+  const __m256i cospim26 = _mm256_set1_epi32(-cospi[26]);
+  const __m256i cospi34 = _mm256_set1_epi32(cospi[34]);
+  const __m256i cospi30 = _mm256_set1_epi32(cospi[30]);
+  const __m256i cospim34 = _mm256_set1_epi32(-cospi[34]);
+  const __m256i cospi42 = _mm256_set1_epi32(cospi[42]);
+  const __m256i cospi22 = _mm256_set1_epi32(cospi[22]);
+  const __m256i cospim42 = _mm256_set1_epi32(-cospi[42]);
+  const __m256i cospi50 = _mm256_set1_epi32(cospi[50]);
+  const __m256i cospi14 = _mm256_set1_epi32(cospi[14]);
+  const __m256i cospim50 = _mm256_set1_epi32(-cospi[50]);
+  const __m256i cospi58 = _mm256_set1_epi32(cospi[58]);
+  const __m256i cospi6 = _mm256_set1_epi32(cospi[6]);
+  const __m256i cospim58 = _mm256_set1_epi32(-cospi[58]);
+  const __m256i rnding = _mm256_set1_epi32(1 << (bit - 1));
+  const __m256i zero = _mm256_setzero_si256();
+
+  __m256i u[16], v[16], x, y;
+  int col;
+
+  for (col = 0; col < num_cols; ++col) {
+    // stage 0
+    // stage 1
+    u[0] = in[0 * num_cols + col];
+    u[1] = _mm256_sub_epi32(zero, in[15 * num_cols + col]);
+    u[2] = _mm256_sub_epi32(zero, in[7 * num_cols + col]);
+    u[3] = in[8 * num_cols + col];
+    u[4] = _mm256_sub_epi32(zero, in[3 * num_cols + col]);
+    u[5] = in[12 * num_cols + col];
+    u[6] = in[4 * num_cols + col];
+    u[7] = _mm256_sub_epi32(zero, in[11 * num_cols + col]);
+    u[8] = _mm256_sub_epi32(zero, in[1 * num_cols + col]);
+    u[9] = in[14 * num_cols + col];
+    u[10] = in[6 * num_cols + col];
+    u[11] = _mm256_sub_epi32(zero, in[9 * num_cols + col]);
+    u[12] = in[2 * num_cols + col];
+    u[13] = _mm256_sub_epi32(zero, in[13 * num_cols + col]);
+    u[14] = _mm256_sub_epi32(zero, in[5 * num_cols + col]);
+    u[15] = in[10 * num_cols + col];
+
+    // stage 2
+    v[0] = u[0];
+    v[1] = u[1];
+
+    x = _mm256_mullo_epi32(u[2], cospi32);
+    y = _mm256_mullo_epi32(u[3], cospi32);
+    v[2] = _mm256_add_epi32(x, y);
+    v[2] = _mm256_add_epi32(v[2], rnding);
+    v[2] = _mm256_srai_epi32(v[2], bit);
+
+    v[3] = _mm256_sub_epi32(x, y);
+    v[3] = _mm256_add_epi32(v[3], rnding);
+    v[3] = _mm256_srai_epi32(v[3], bit);
+
+    v[4] = u[4];
+    v[5] = u[5];
+
+    x = _mm256_mullo_epi32(u[6], cospi32);
+    y = _mm256_mullo_epi32(u[7], cospi32);
+    v[6] = _mm256_add_epi32(x, y);
+    v[6] = _mm256_add_epi32(v[6], rnding);
+    v[6] = _mm256_srai_epi32(v[6], bit);
+
+    v[7] = _mm256_sub_epi32(x, y);
+    v[7] = _mm256_add_epi32(v[7], rnding);
+    v[7] = _mm256_srai_epi32(v[7], bit);
+
+    v[8] = u[8];
+    v[9] = u[9];
+
+    x = _mm256_mullo_epi32(u[10], cospi32);
+    y = _mm256_mullo_epi32(u[11], cospi32);
+    v[10] = _mm256_add_epi32(x, y);
+    v[10] = _mm256_add_epi32(v[10], rnding);
+    v[10] = _mm256_srai_epi32(v[10], bit);
+
+    v[11] = _mm256_sub_epi32(x, y);
+    v[11] = _mm256_add_epi32(v[11], rnding);
+    v[11] = _mm256_srai_epi32(v[11], bit);
+
+    v[12] = u[12];
+    v[13] = u[13];
+
+    x = _mm256_mullo_epi32(u[14], cospi32);
+    y = _mm256_mullo_epi32(u[15], cospi32);
+    v[14] = _mm256_add_epi32(x, y);
+    v[14] = _mm256_add_epi32(v[14], rnding);
+    v[14] = _mm256_srai_epi32(v[14], bit);
+
+    v[15] = _mm256_sub_epi32(x, y);
+    v[15] = _mm256_add_epi32(v[15], rnding);
+    v[15] = _mm256_srai_epi32(v[15], bit);
+
+    // stage 3
+    u[0] = _mm256_add_epi32(v[0], v[2]);
+    u[1] = _mm256_add_epi32(v[1], v[3]);
+    u[2] = _mm256_sub_epi32(v[0], v[2]);
+    u[3] = _mm256_sub_epi32(v[1], v[3]);
+    u[4] = _mm256_add_epi32(v[4], v[6]);
+    u[5] = _mm256_add_epi32(v[5], v[7]);
+    u[6] = _mm256_sub_epi32(v[4], v[6]);
+    u[7] = _mm256_sub_epi32(v[5], v[7]);
+    u[8] = _mm256_add_epi32(v[8], v[10]);
+    u[9] = _mm256_add_epi32(v[9], v[11]);
+    u[10] = _mm256_sub_epi32(v[8], v[10]);
+    u[11] = _mm256_sub_epi32(v[9], v[11]);
+    u[12] = _mm256_add_epi32(v[12], v[14]);
+    u[13] = _mm256_add_epi32(v[13], v[15]);
+    u[14] = _mm256_sub_epi32(v[12], v[14]);
+    u[15] = _mm256_sub_epi32(v[13], v[15]);
+
+    // stage 4
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+    v[4] = av1_half_btf_avx2(&cospi16, &u[4], &cospi48, &u[5], &rnding, bit);
+    v[5] = av1_half_btf_avx2(&cospi48, &u[4], &cospim16, &u[5], &rnding, bit);
+    v[6] = av1_half_btf_avx2(&cospim48, &u[6], &cospi16, &u[7], &rnding, bit);
+    v[7] = av1_half_btf_avx2(&cospi16, &u[6], &cospi48, &u[7], &rnding, bit);
+    v[8] = u[8];
+    v[9] = u[9];
+    v[10] = u[10];
+    v[11] = u[11];
+    v[12] = av1_half_btf_avx2(&cospi16, &u[12], &cospi48, &u[13], &rnding, bit);
+    v[13] =
+        av1_half_btf_avx2(&cospi48, &u[12], &cospim16, &u[13], &rnding, bit);
+    v[14] =
+        av1_half_btf_avx2(&cospim48, &u[14], &cospi16, &u[15], &rnding, bit);
+    v[15] = av1_half_btf_avx2(&cospi16, &u[14], &cospi48, &u[15], &rnding, bit);
+
+    // stage 5
+    u[0] = _mm256_add_epi32(v[0], v[4]);
+    u[1] = _mm256_add_epi32(v[1], v[5]);
+    u[2] = _mm256_add_epi32(v[2], v[6]);
+    u[3] = _mm256_add_epi32(v[3], v[7]);
+    u[4] = _mm256_sub_epi32(v[0], v[4]);
+    u[5] = _mm256_sub_epi32(v[1], v[5]);
+    u[6] = _mm256_sub_epi32(v[2], v[6]);
+    u[7] = _mm256_sub_epi32(v[3], v[7]);
+    u[8] = _mm256_add_epi32(v[8], v[12]);
+    u[9] = _mm256_add_epi32(v[9], v[13]);
+    u[10] = _mm256_add_epi32(v[10], v[14]);
+    u[11] = _mm256_add_epi32(v[11], v[15]);
+    u[12] = _mm256_sub_epi32(v[8], v[12]);
+    u[13] = _mm256_sub_epi32(v[9], v[13]);
+    u[14] = _mm256_sub_epi32(v[10], v[14]);
+    u[15] = _mm256_sub_epi32(v[11], v[15]);
+
+    // stage 6
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+    v[4] = u[4];
+    v[5] = u[5];
+    v[6] = u[6];
+    v[7] = u[7];
+    v[8] = av1_half_btf_avx2(&cospi8, &u[8], &cospi56, &u[9], &rnding, bit);
+    v[9] = av1_half_btf_avx2(&cospi56, &u[8], &cospim8, &u[9], &rnding, bit);
+    v[10] = av1_half_btf_avx2(&cospi40, &u[10], &cospi24, &u[11], &rnding, bit);
+    v[11] =
+        av1_half_btf_avx2(&cospi24, &u[10], &cospim40, &u[11], &rnding, bit);
+    v[12] = av1_half_btf_avx2(&cospim56, &u[12], &cospi8, &u[13], &rnding, bit);
+    v[13] = av1_half_btf_avx2(&cospi8, &u[12], &cospi56, &u[13], &rnding, bit);
+    v[14] =
+        av1_half_btf_avx2(&cospim24, &u[14], &cospi40, &u[15], &rnding, bit);
+    v[15] = av1_half_btf_avx2(&cospi40, &u[14], &cospi24, &u[15], &rnding, bit);
+
+    // stage 7
+    u[0] = _mm256_add_epi32(v[0], v[8]);
+    u[1] = _mm256_add_epi32(v[1], v[9]);
+    u[2] = _mm256_add_epi32(v[2], v[10]);
+    u[3] = _mm256_add_epi32(v[3], v[11]);
+    u[4] = _mm256_add_epi32(v[4], v[12]);
+    u[5] = _mm256_add_epi32(v[5], v[13]);
+    u[6] = _mm256_add_epi32(v[6], v[14]);
+    u[7] = _mm256_add_epi32(v[7], v[15]);
+    u[8] = _mm256_sub_epi32(v[0], v[8]);
+    u[9] = _mm256_sub_epi32(v[1], v[9]);
+    u[10] = _mm256_sub_epi32(v[2], v[10]);
+    u[11] = _mm256_sub_epi32(v[3], v[11]);
+    u[12] = _mm256_sub_epi32(v[4], v[12]);
+    u[13] = _mm256_sub_epi32(v[5], v[13]);
+    u[14] = _mm256_sub_epi32(v[6], v[14]);
+    u[15] = _mm256_sub_epi32(v[7], v[15]);
+
+    // stage 8
+    v[0] = av1_half_btf_avx2(&cospi2, &u[0], &cospi62, &u[1], &rnding, bit);
+    v[1] = av1_half_btf_avx2(&cospi62, &u[0], &cospim2, &u[1], &rnding, bit);
+    v[2] = av1_half_btf_avx2(&cospi10, &u[2], &cospi54, &u[3], &rnding, bit);
+    v[3] = av1_half_btf_avx2(&cospi54, &u[2], &cospim10, &u[3], &rnding, bit);
+    v[4] = av1_half_btf_avx2(&cospi18, &u[4], &cospi46, &u[5], &rnding, bit);
+    v[5] = av1_half_btf_avx2(&cospi46, &u[4], &cospim18, &u[5], &rnding, bit);
+    v[6] = av1_half_btf_avx2(&cospi26, &u[6], &cospi38, &u[7], &rnding, bit);
+    v[7] = av1_half_btf_avx2(&cospi38, &u[6], &cospim26, &u[7], &rnding, bit);
+    v[8] = av1_half_btf_avx2(&cospi34, &u[8], &cospi30, &u[9], &rnding, bit);
+    v[9] = av1_half_btf_avx2(&cospi30, &u[8], &cospim34, &u[9], &rnding, bit);
+    v[10] = av1_half_btf_avx2(&cospi42, &u[10], &cospi22, &u[11], &rnding, bit);
+    v[11] =
+        av1_half_btf_avx2(&cospi22, &u[10], &cospim42, &u[11], &rnding, bit);
+    v[12] = av1_half_btf_avx2(&cospi50, &u[12], &cospi14, &u[13], &rnding, bit);
+    v[13] =
+        av1_half_btf_avx2(&cospi14, &u[12], &cospim50, &u[13], &rnding, bit);
+    v[14] = av1_half_btf_avx2(&cospi58, &u[14], &cospi6, &u[15], &rnding, bit);
+    v[15] = av1_half_btf_avx2(&cospi6, &u[14], &cospim58, &u[15], &rnding, bit);
+
+    // stage 9
+    out[0 * outstride + col] = v[1];
+    out[1 * outstride + col] = v[14];
+    out[2 * outstride + col] = v[3];
+    out[3 * outstride + col] = v[12];
+    out[4 * outstride + col] = v[5];
+    out[5 * outstride + col] = v[10];
+    out[6 * outstride + col] = v[7];
+    out[7 * outstride + col] = v[8];
+    out[8 * outstride + col] = v[9];
+    out[9 * outstride + col] = v[6];
+    out[10 * outstride + col] = v[11];
+    out[11 * outstride + col] = v[4];
+    out[12 * outstride + col] = v[13];
+    out[13 * outstride + col] = v[2];
+    out[14 * outstride + col] = v[15];
+    out[15 * outstride + col] = v[0];
+  }
+}
+static void av1_idtx16_avx2(__m256i *in, __m256i *out, const int8_t bit,
+                            int col_num, const int outstride) {
+  (void)bit;
+  (void)outstride;
+  __m256i fact = _mm256_set1_epi32(2 * NewSqrt2);
+  __m256i offset = _mm256_set1_epi32(1 << (NewSqrt2Bits - 1));
+  __m256i a_low;
+
+  int num_iters = 16 * col_num;
+  for (int i = 0; i < num_iters; i++) {
+    a_low = _mm256_mullo_epi32(in[i], fact);
+    a_low = _mm256_add_epi32(a_low, offset);
+    out[i] = _mm256_srai_epi32(a_low, NewSqrt2Bits);
+  }
+}
+static const transform_1d_avx2 col_highbd_txfm8x16_arr[TX_TYPES] = {
+  av1_fdct16_avx2,   // DCT_DCT
+  av1_fadst16_avx2,  // ADST_DCT
+  av1_fdct16_avx2,   // DCT_ADST
+  av1_fadst16_avx2,  // ADST_ADST
+  av1_fadst16_avx2,  // FLIPADST_DCT
+  av1_fdct16_avx2,   // DCT_FLIPADST
+  av1_fadst16_avx2,  // FLIPADST_FLIPADST
+  av1_fadst16_avx2,  // ADST_FLIPADST
+  av1_fadst16_avx2,  // FLIPADST_ADST
+  av1_idtx16_avx2,   // IDTX
+  av1_fdct16_avx2,   // V_DCT
+  av1_idtx16_avx2,   // H_DCT
+  av1_fadst16_avx2,  // V_ADST
+  av1_idtx16_avx2,   // H_ADST
+  av1_fadst16_avx2,  // V_FLIPADST
+  av1_idtx16_avx2    // H_FLIPADST
+};
+static const transform_1d_avx2 row_highbd_txfm8x8_arr[TX_TYPES] = {
+  av1_fdct8_avx2,   // DCT_DCT
+  av1_fdct8_avx2,   // ADST_DCT
+  av1_fadst8_avx2,  // DCT_ADST
+  av1_fadst8_avx2,  // ADST_ADST
+  av1_fdct8_avx2,   // FLIPADST_DCT
+  av1_fadst8_avx2,  // DCT_FLIPADST
+  av1_fadst8_avx2,  // FLIPADST_FLIPADST
+  av1_fadst8_avx2,  // ADST_FLIPADST
+  av1_fadst8_avx2,  // FLIPADST_ADST
+  av1_idtx8_avx2,   // IDTX
+  av1_idtx8_avx2,   // V_DCT
+  av1_fdct8_avx2,   // H_DCT
+  av1_idtx8_avx2,   // V_ADST
+  av1_fadst8_avx2,  // H_ADST
+  av1_idtx8_avx2,   // V_FLIPADST
+  av1_fadst8_avx2   // H_FLIPADST
+};
+void av1_fwd_txfm2d_8x16_avx2(const int16_t *input, int32_t *coeff, int stride,
+                              TX_TYPE tx_type, int bd) {
+  __m256i in[16], out[16];
+  const int8_t *shift = fwd_txfm_shift_ls[TX_8X16];
+  const int txw_idx = get_txw_idx(TX_8X16);
+  const int txh_idx = get_txh_idx(TX_8X16);
+  const transform_1d_avx2 col_txfm = col_highbd_txfm8x16_arr[tx_type];
+  const transform_1d_avx2 row_txfm = row_highbd_txfm8x8_arr[tx_type];
+  const int8_t bit = fwd_cos_bit_col[txw_idx][txh_idx];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  av1_load_buffer_8x16_avx2(input, in, stride, ud_flip, lr_flip, shift[0]);
+  col_txfm(in, out, bit, 1, 1);
+  col_txfm_8x8_rounding(out, -shift[1]);
+  col_txfm_8x8_rounding(&out[8], -shift[1]);
+  av1_fwd_txfm_transpose_8x8_avx2(out, in, 1, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(&out[8], &in[1], 1, 2);
+  row_txfm(in, out, bit, 2, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(out, in, 2, 1);
+  av1_fwd_txfm_transpose_8x8_avx2(&out[1], &in[8], 2, 1);
+  av1_round_shift_rect_array_32_avx2(in, in, 16, -shift[2], NewSqrt2);
+  av1_store_buffer_avx2(in, coeff, 8, 16);
+  (void)bd;
+}
+static const transform_1d_avx2 col_highbd_txfm8x8_arr[TX_TYPES] = {
+  av1_fdct8_avx2,   // DCT_DCT
+  av1_fadst8_avx2,  // ADST_DCT
+  av1_fdct8_avx2,   // DCT_ADST
+  av1_fadst8_avx2,  // ADST_ADST
+  av1_fadst8_avx2,  // FLIPADST_DCT
+  av1_fdct8_avx2,   // DCT_FLIPADST
+  av1_fadst8_avx2,  // FLIPADST_FLIPADST
+  av1_fadst8_avx2,  // ADST_FLIPADST
+  av1_fadst8_avx2,  // FLIPADST_ADST
+  av1_idtx8_avx2,   // IDTX
+  av1_fdct8_avx2,   // V_DCT
+  av1_idtx8_avx2,   // H_DCT
+  av1_fadst8_avx2,  // V_ADST
+  av1_idtx8_avx2,   // H_ADST
+  av1_fadst8_avx2,  // V_FLIPADST
+  av1_idtx8_avx2    // H_FLIPADST
+};
+static const transform_1d_avx2 row_highbd_txfm8x16_arr[TX_TYPES] = {
+  av1_fdct16_avx2,   // DCT_DCT
+  av1_fdct16_avx2,   // ADST_DCT
+  av1_fadst16_avx2,  // DCT_ADST
+  av1_fadst16_avx2,  // ADST_ADST
+  av1_fdct16_avx2,   // FLIPADST_DCT
+  av1_fadst16_avx2,  // DCT_FLIPADST
+  av1_fadst16_avx2,  // FLIPADST_FLIPADST
+  av1_fadst16_avx2,  // ADST_FLIPADST
+  av1_fadst16_avx2,  // FLIPADST_ADST
+  av1_idtx16_avx2,   // IDTX
+  av1_idtx16_avx2,   // V_DCT
+  av1_fdct16_avx2,   // H_DCT
+  av1_idtx16_avx2,   // V_ADST
+  av1_fadst16_avx2,  // H_ADST
+  av1_idtx16_avx2,   // V_FLIPADST
+  av1_fadst16_avx2   // H_FLIPADST
+};
+void av1_fwd_txfm2d_16x8_avx2(const int16_t *input, int32_t *coeff, int stride,
+                              TX_TYPE tx_type, int bd) {
+  __m256i in[16], out[16];
+  const int8_t *shift = fwd_txfm_shift_ls[TX_16X8];
+  const int txw_idx = get_txw_idx(TX_16X8);
+  const int txh_idx = get_txh_idx(TX_16X8);
+  const transform_1d_avx2 col_txfm = col_highbd_txfm8x8_arr[tx_type];
+  const transform_1d_avx2 row_txfm = row_highbd_txfm8x16_arr[tx_type];
+  const int8_t bit = fwd_cos_bit_col[txw_idx][txh_idx];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  av1_load_buffer_16xn_avx2(input, in, stride, 8, 2, ud_flip, lr_flip);
+  av1_round_shift_32_8xn_avx2(in, 16, shift[0], 1);
+  col_txfm(in, out, bit, 2, 2);
+  av1_round_shift_32_8xn_avx2(out, 16, shift[1], 1);
+  av1_fwd_txfm_transpose_8x8_avx2(out, in, 2, 1);
+  av1_fwd_txfm_transpose_8x8_avx2(&out[1], &in[8], 2, 1);
+  row_txfm(in, out, bit, 1, 1);
+  av1_fwd_txfm_transpose_8x8_avx2(out, in, 1, 2);
+  av1_fwd_txfm_transpose_8x8_avx2(&out[8], &in[1], 1, 2);
+  av1_round_shift_rect_array_32_avx2(in, in, 16, -shift[2], NewSqrt2);
+  av1_store_buffer_avx2(in, coeff, 8, 16);
+  (void)bd;
+}
+void av1_fwd_txfm2d_16x16_avx2(const int16_t *input, int32_t *coeff, int stride,
+                               TX_TYPE tx_type, int bd) {
+  __m256i in[32], out[32];
+  const TX_SIZE tx_size = TX_16X16;
+  const int8_t *shift = fwd_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int width = tx_size_wide[tx_size];
+  const int height = tx_size_high[tx_size];
+  const int width_div8 = (width >> 3);
+  const int width_div16 = (width >> 4);
+  const int size = (height << 1);
+  switch (tx_type) {
+    case DCT_DCT:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case ADST_DCT:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case DCT_ADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case ADST_ADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case FLIPADST_DCT:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 1, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case DCT_FLIPADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 1);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case FLIPADST_FLIPADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 1, 1);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case ADST_FLIPADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 1);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case FLIPADST_ADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 1, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case IDTX:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_idtx16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_idtx16_avx2(out, in, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case V_DCT:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_idtx16_avx2(out, in, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case H_DCT:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_idtx16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fdct16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case V_ADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_idtx16_avx2(out, in, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case H_ADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_idtx16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case V_FLIPADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 1, 0);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_idtx16_avx2(out, in, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    case H_FLIPADST:
+      av1_load_buffer_16xn_avx2(input, in, stride, height, width_div8, 0, 1);
+      av1_round_shift_32_8xn_avx2(in, size, shift[0], width_div16);
+      av1_idtx16_avx2(in, out, fwd_cos_bit_col[txw_idx][txh_idx], width_div8,
+                      width_div8);
+      av1_round_shift_32_8xn_avx2(out, size, shift[1], width_div16);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_fadst16_avx2(in, out, fwd_cos_bit_row[txw_idx][txh_idx], width_div8,
+                       width_div8);
+      av1_fwd_txfm_transpose_16x16_avx2(out, in);
+      av1_store_buffer_avx2(in, coeff, 8, 32);
+      break;
+    default: assert(0);
+  }
+  (void)bd;
+}
+static INLINE void av1_fdct32_avx2(__m256i *input, __m256i *output,
+                                   const int8_t cos_bit, const int instride,
+                                   const int outstride) {
+  __m256i buf0[32];
+  __m256i buf1[32];
+  const int32_t *cospi;
+  int startidx = 0 * instride;
+  int endidx = 31 * instride;
+  // stage 0
+  // stage 1
+  buf1[0] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[31] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[1] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[30] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[2] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[29] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[3] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[28] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[4] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[27] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[5] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[26] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[6] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[25] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[7] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[24] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[8] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[23] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[9] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[22] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[10] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[21] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[11] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[20] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[12] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[19] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[13] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[18] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[14] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[17] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  buf1[15] = _mm256_add_epi32(input[startidx], input[endidx]);
+  buf1[16] = _mm256_sub_epi32(input[startidx], input[endidx]);
+
+  // stage 2
+  cospi = cospi_arr(cos_bit);
+  buf0[0] = _mm256_add_epi32(buf1[0], buf1[15]);
+  buf0[15] = _mm256_sub_epi32(buf1[0], buf1[15]);
+  buf0[1] = _mm256_add_epi32(buf1[1], buf1[14]);
+  buf0[14] = _mm256_sub_epi32(buf1[1], buf1[14]);
+  buf0[2] = _mm256_add_epi32(buf1[2], buf1[13]);
+  buf0[13] = _mm256_sub_epi32(buf1[2], buf1[13]);
+  buf0[3] = _mm256_add_epi32(buf1[3], buf1[12]);
+  buf0[12] = _mm256_sub_epi32(buf1[3], buf1[12]);
+  buf0[4] = _mm256_add_epi32(buf1[4], buf1[11]);
+  buf0[11] = _mm256_sub_epi32(buf1[4], buf1[11]);
+  buf0[5] = _mm256_add_epi32(buf1[5], buf1[10]);
+  buf0[10] = _mm256_sub_epi32(buf1[5], buf1[10]);
+  buf0[6] = _mm256_add_epi32(buf1[6], buf1[9]);
+  buf0[9] = _mm256_sub_epi32(buf1[6], buf1[9]);
+  buf0[7] = _mm256_add_epi32(buf1[7], buf1[8]);
+  buf0[8] = _mm256_sub_epi32(buf1[7], buf1[8]);
+  buf0[16] = buf1[16];
+  buf0[17] = buf1[17];
+  buf0[18] = buf1[18];
+  buf0[19] = buf1[19];
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf1[20], buf1[27], buf0[20],
+                    buf0[27], cos_bit);
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf1[21], buf1[26], buf0[21],
+                    buf0[26], cos_bit);
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf1[22], buf1[25], buf0[22],
+                    buf0[25], cos_bit);
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf1[23], buf1[24], buf0[23],
+                    buf0[24], cos_bit);
+  buf0[28] = buf1[28];
+  buf0[29] = buf1[29];
+  buf0[30] = buf1[30];
+  buf0[31] = buf1[31];
+
+  // stage 3
+  cospi = cospi_arr(cos_bit);
+  buf1[0] = _mm256_add_epi32(buf0[0], buf0[7]);
+  buf1[7] = _mm256_sub_epi32(buf0[0], buf0[7]);
+  buf1[1] = _mm256_add_epi32(buf0[1], buf0[6]);
+  buf1[6] = _mm256_sub_epi32(buf0[1], buf0[6]);
+  buf1[2] = _mm256_add_epi32(buf0[2], buf0[5]);
+  buf1[5] = _mm256_sub_epi32(buf0[2], buf0[5]);
+  buf1[3] = _mm256_add_epi32(buf0[3], buf0[4]);
+  buf1[4] = _mm256_sub_epi32(buf0[3], buf0[4]);
+  buf1[8] = buf0[8];
+  buf1[9] = buf0[9];
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf0[10], buf0[13], buf1[10],
+                    buf1[13], cos_bit);
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf0[11], buf0[12], buf1[11],
+                    buf1[12], cos_bit);
+  buf1[14] = buf0[14];
+  buf1[15] = buf0[15];
+  buf1[16] = _mm256_add_epi32(buf0[16], buf0[23]);
+  buf1[23] = _mm256_sub_epi32(buf0[16], buf0[23]);
+  buf1[17] = _mm256_add_epi32(buf0[17], buf0[22]);
+  buf1[22] = _mm256_sub_epi32(buf0[17], buf0[22]);
+  buf1[18] = _mm256_add_epi32(buf0[18], buf0[21]);
+  buf1[21] = _mm256_sub_epi32(buf0[18], buf0[21]);
+  buf1[19] = _mm256_add_epi32(buf0[19], buf0[20]);
+  buf1[20] = _mm256_sub_epi32(buf0[19], buf0[20]);
+  buf1[24] = _mm256_sub_epi32(buf0[31], buf0[24]);
+  buf1[31] = _mm256_add_epi32(buf0[31], buf0[24]);
+  buf1[25] = _mm256_sub_epi32(buf0[30], buf0[25]);
+  buf1[30] = _mm256_add_epi32(buf0[30], buf0[25]);
+  buf1[26] = _mm256_sub_epi32(buf0[29], buf0[26]);
+  buf1[29] = _mm256_add_epi32(buf0[29], buf0[26]);
+  buf1[27] = _mm256_sub_epi32(buf0[28], buf0[27]);
+  buf1[28] = _mm256_add_epi32(buf0[28], buf0[27]);
+
+  // stage 4
+  cospi = cospi_arr(cos_bit);
+  buf0[0] = _mm256_add_epi32(buf1[0], buf1[3]);
+  buf0[3] = _mm256_sub_epi32(buf1[0], buf1[3]);
+  buf0[1] = _mm256_add_epi32(buf1[1], buf1[2]);
+  buf0[2] = _mm256_sub_epi32(buf1[1], buf1[2]);
+  buf0[4] = buf1[4];
+  btf_32_avx2_type0(-cospi[32], cospi[32], buf1[5], buf1[6], buf0[5], buf0[6],
+                    cos_bit);
+  buf0[7] = buf1[7];
+  buf0[8] = _mm256_add_epi32(buf1[8], buf1[11]);
+  buf0[11] = _mm256_sub_epi32(buf1[8], buf1[11]);
+  buf0[9] = _mm256_add_epi32(buf1[9], buf1[10]);
+  buf0[10] = _mm256_sub_epi32(buf1[9], buf1[10]);
+  buf0[12] = _mm256_sub_epi32(buf1[15], buf1[12]);
+  buf0[15] = _mm256_add_epi32(buf1[15], buf1[12]);
+  buf0[13] = _mm256_sub_epi32(buf1[14], buf1[13]);
+  buf0[14] = _mm256_add_epi32(buf1[14], buf1[13]);
+  buf0[16] = buf1[16];
+  buf0[17] = buf1[17];
+  btf_32_avx2_type0(-cospi[16], cospi[48], buf1[18], buf1[29], buf0[18],
+                    buf0[29], cos_bit);
+  btf_32_avx2_type0(-cospi[16], cospi[48], buf1[19], buf1[28], buf0[19],
+                    buf0[28], cos_bit);
+  btf_32_avx2_type0(-cospi[48], -cospi[16], buf1[20], buf1[27], buf0[20],
+                    buf0[27], cos_bit);
+  btf_32_avx2_type0(-cospi[48], -cospi[16], buf1[21], buf1[26], buf0[21],
+                    buf0[26], cos_bit);
+  buf0[22] = buf1[22];
+  buf0[23] = buf1[23];
+  buf0[24] = buf1[24];
+  buf0[25] = buf1[25];
+  buf0[30] = buf1[30];
+  buf0[31] = buf1[31];
+
+  // stage 5
+  cospi = cospi_arr(cos_bit);
+  btf_32_avx2_type0(cospi[32], cospi[32], buf0[0], buf0[1], buf1[0], buf1[1],
+                    cos_bit);
+  btf_32_avx2_type0(cospi[16], cospi[48], buf0[3], buf0[2], buf1[2], buf1[3],
+                    cos_bit);
+  buf1[4] = _mm256_add_epi32(buf0[4], buf0[5]);
+  buf1[5] = _mm256_sub_epi32(buf0[4], buf0[5]);
+  buf1[6] = _mm256_sub_epi32(buf0[7], buf0[6]);
+  buf1[7] = _mm256_add_epi32(buf0[7], buf0[6]);
+  buf1[8] = buf0[8];
+  btf_32_avx2_type0(-cospi[16], cospi[48], buf0[9], buf0[14], buf1[9], buf1[14],
+                    cos_bit);
+  btf_32_avx2_type0(-cospi[48], -cospi[16], buf0[10], buf0[13], buf1[10],
+                    buf1[13], cos_bit);
+  buf1[11] = buf0[11];
+  buf1[12] = buf0[12];
+  buf1[15] = buf0[15];
+  buf1[16] = _mm256_add_epi32(buf0[16], buf0[19]);
+  buf1[19] = _mm256_sub_epi32(buf0[16], buf0[19]);
+  buf1[17] = _mm256_add_epi32(buf0[17], buf0[18]);
+  buf1[18] = _mm256_sub_epi32(buf0[17], buf0[18]);
+  buf1[20] = _mm256_sub_epi32(buf0[23], buf0[20]);
+  buf1[23] = _mm256_add_epi32(buf0[23], buf0[20]);
+  buf1[21] = _mm256_sub_epi32(buf0[22], buf0[21]);
+  buf1[22] = _mm256_add_epi32(buf0[22], buf0[21]);
+  buf1[24] = _mm256_add_epi32(buf0[24], buf0[27]);
+  buf1[27] = _mm256_sub_epi32(buf0[24], buf0[27]);
+  buf1[25] = _mm256_add_epi32(buf0[25], buf0[26]);
+  buf1[26] = _mm256_sub_epi32(buf0[25], buf0[26]);
+  buf1[28] = _mm256_sub_epi32(buf0[31], buf0[28]);
+  buf1[31] = _mm256_add_epi32(buf0[31], buf0[28]);
+  buf1[29] = _mm256_sub_epi32(buf0[30], buf0[29]);
+  buf1[30] = _mm256_add_epi32(buf0[30], buf0[29]);
+
+  // stage 6
+  cospi = cospi_arr(cos_bit);
+  buf0[0] = buf1[0];
+  buf0[1] = buf1[1];
+  buf0[2] = buf1[2];
+  buf0[3] = buf1[3];
+  btf_32_avx2_type0(cospi[8], cospi[56], buf1[7], buf1[4], buf0[4], buf0[7],
+                    cos_bit);
+  btf_32_avx2_type0(cospi[40], cospi[24], buf1[6], buf1[5], buf0[5], buf0[6],
+                    cos_bit);
+  buf0[8] = _mm256_add_epi32(buf1[8], buf1[9]);
+  buf0[9] = _mm256_sub_epi32(buf1[8], buf1[9]);
+  buf0[10] = _mm256_sub_epi32(buf1[11], buf1[10]);
+  buf0[11] = _mm256_add_epi32(buf1[11], buf1[10]);
+  buf0[12] = _mm256_add_epi32(buf1[12], buf1[13]);
+  buf0[13] = _mm256_sub_epi32(buf1[12], buf1[13]);
+  buf0[14] = _mm256_sub_epi32(buf1[15], buf1[14]);
+  buf0[15] = _mm256_add_epi32(buf1[15], buf1[14]);
+  buf0[16] = buf1[16];
+  btf_32_avx2_type0(-cospi[8], cospi[56], buf1[17], buf1[30], buf0[17],
+                    buf0[30], cos_bit);
+  btf_32_avx2_type0(-cospi[56], -cospi[8], buf1[18], buf1[29], buf0[18],
+                    buf0[29], cos_bit);
+  buf0[19] = buf1[19];
+  buf0[20] = buf1[20];
+  btf_32_avx2_type0(-cospi[40], cospi[24], buf1[21], buf1[26], buf0[21],
+                    buf0[26], cos_bit);
+  btf_32_avx2_type0(-cospi[24], -cospi[40], buf1[22], buf1[25], buf0[22],
+                    buf0[25], cos_bit);
+  buf0[23] = buf1[23];
+  buf0[24] = buf1[24];
+  buf0[27] = buf1[27];
+  buf0[28] = buf1[28];
+  buf0[31] = buf1[31];
+
+  // stage 7
+  cospi = cospi_arr(cos_bit);
+  buf1[0] = buf0[0];
+  buf1[1] = buf0[1];
+  buf1[2] = buf0[2];
+  buf1[3] = buf0[3];
+  buf1[4] = buf0[4];
+  buf1[5] = buf0[5];
+  buf1[6] = buf0[6];
+  buf1[7] = buf0[7];
+  btf_32_avx2_type0(cospi[4], cospi[60], buf0[15], buf0[8], buf1[8], buf1[15],
+                    cos_bit);
+  btf_32_avx2_type0(cospi[36], cospi[28], buf0[14], buf0[9], buf1[9], buf1[14],
+                    cos_bit);
+  btf_32_avx2_type0(cospi[20], cospi[44], buf0[13], buf0[10], buf1[10],
+                    buf1[13], cos_bit);
+  btf_32_avx2_type0(cospi[52], cospi[12], buf0[12], buf0[11], buf1[11],
+                    buf1[12], cos_bit);
+  buf1[16] = _mm256_add_epi32(buf0[16], buf0[17]);
+  buf1[17] = _mm256_sub_epi32(buf0[16], buf0[17]);
+  buf1[18] = _mm256_sub_epi32(buf0[19], buf0[18]);
+  buf1[19] = _mm256_add_epi32(buf0[19], buf0[18]);
+  buf1[20] = _mm256_add_epi32(buf0[20], buf0[21]);
+  buf1[21] = _mm256_sub_epi32(buf0[20], buf0[21]);
+  buf1[22] = _mm256_sub_epi32(buf0[23], buf0[22]);
+  buf1[23] = _mm256_add_epi32(buf0[23], buf0[22]);
+  buf1[24] = _mm256_add_epi32(buf0[24], buf0[25]);
+  buf1[25] = _mm256_sub_epi32(buf0[24], buf0[25]);
+  buf1[26] = _mm256_sub_epi32(buf0[27], buf0[26]);
+  buf1[27] = _mm256_add_epi32(buf0[27], buf0[26]);
+  buf1[28] = _mm256_add_epi32(buf0[28], buf0[29]);
+  buf1[29] = _mm256_sub_epi32(buf0[28], buf0[29]);
+  buf1[30] = _mm256_sub_epi32(buf0[31], buf0[30]);
+  buf1[31] = _mm256_add_epi32(buf0[31], buf0[30]);
+
+  // stage 8
+  cospi = cospi_arr(cos_bit);
+  buf0[0] = buf1[0];
+  buf0[1] = buf1[1];
+  buf0[2] = buf1[2];
+  buf0[3] = buf1[3];
+  buf0[4] = buf1[4];
+  buf0[5] = buf1[5];
+  buf0[6] = buf1[6];
+  buf0[7] = buf1[7];
+  buf0[8] = buf1[8];
+  buf0[9] = buf1[9];
+  buf0[10] = buf1[10];
+  buf0[11] = buf1[11];
+  buf0[12] = buf1[12];
+  buf0[13] = buf1[13];
+  buf0[14] = buf1[14];
+  buf0[15] = buf1[15];
+  btf_32_avx2_type0(cospi[2], cospi[62], buf1[31], buf1[16], buf0[16], buf0[31],
+                    cos_bit);
+  btf_32_avx2_type0(cospi[34], cospi[30], buf1[30], buf1[17], buf0[17],
+                    buf0[30], cos_bit);
+  btf_32_avx2_type0(cospi[18], cospi[46], buf1[29], buf1[18], buf0[18],
+                    buf0[29], cos_bit);
+  btf_32_avx2_type0(cospi[50], cospi[14], buf1[28], buf1[19], buf0[19],
+                    buf0[28], cos_bit);
+  btf_32_avx2_type0(cospi[10], cospi[54], buf1[27], buf1[20], buf0[20],
+                    buf0[27], cos_bit);
+  btf_32_avx2_type0(cospi[42], cospi[22], buf1[26], buf1[21], buf0[21],
+                    buf0[26], cos_bit);
+  btf_32_avx2_type0(cospi[26], cospi[38], buf1[25], buf1[22], buf0[22],
+                    buf0[25], cos_bit);
+  btf_32_avx2_type0(cospi[58], cospi[6], buf1[24], buf1[23], buf0[23], buf0[24],
+                    cos_bit);
+
+  startidx = 0 * outstride;
+  endidx = 31 * outstride;
+  // stage 9
+  output[startidx] = buf0[0];
+  output[endidx] = buf0[31];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[16];
+  output[endidx] = buf0[15];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[8];
+  output[endidx] = buf0[23];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[24];
+  output[endidx] = buf0[7];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[4];
+  output[endidx] = buf0[27];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[20];
+  output[endidx] = buf0[11];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[12];
+  output[endidx] = buf0[19];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[28];
+  output[endidx] = buf0[3];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[2];
+  output[endidx] = buf0[29];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[18];
+  output[endidx] = buf0[13];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[10];
+  output[endidx] = buf0[21];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[26];
+  output[endidx] = buf0[5];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[6];
+  output[endidx] = buf0[25];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[22];
+  output[endidx] = buf0[9];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[14];
+  output[endidx] = buf0[17];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = buf0[30];
+  output[endidx] = buf0[1];
+}
+static INLINE void idtx32x32_avx2(__m256i *input, __m256i *output,
+                                  const int8_t cos_bit, int instride,
+                                  int outstride) {
+  (void)cos_bit;
+  for (int i = 0; i < 32; i += 8) {
+    output[i * outstride] = _mm256_slli_epi32(input[i * instride], 2);
+    output[(i + 1) * outstride] =
+        _mm256_slli_epi32(input[(i + 1) * instride], 2);
+    output[(i + 2) * outstride] =
+        _mm256_slli_epi32(input[(i + 2) * instride], 2);
+    output[(i + 3) * outstride] =
+        _mm256_slli_epi32(input[(i + 3) * instride], 2);
+    output[(i + 4) * outstride] =
+        _mm256_slli_epi32(input[(i + 4) * instride], 2);
+    output[(i + 5) * outstride] =
+        _mm256_slli_epi32(input[(i + 5) * instride], 2);
+    output[(i + 6) * outstride] =
+        _mm256_slli_epi32(input[(i + 6) * instride], 2);
+    output[(i + 7) * outstride] =
+        _mm256_slli_epi32(input[(i + 7) * instride], 2);
+  }
+}
+static const transform_1d_avx2 col_txfm8x32_arr[TX_TYPES] = {
+  av1_fdct32_avx2,  // DCT_DCT
+  NULL,             // ADST_DCT
+  NULL,             // DCT_ADST
+  NULL,             // ADST_ADST
+  NULL,             // FLIPADST_DCT
+  NULL,             // DCT_FLIPADST
+  NULL,             // FLIPADST_FLIPADST
+  NULL,             // ADST_FLIPADST
+  NULL,             // FLIPADST_ADST
+  idtx32x32_avx2,   // IDTX
+  NULL,             // V_DCT
+  NULL,             // H_DCT
+  NULL,             // V_ADST
+  NULL,             // H_ADST
+  NULL,             // V_FLIPADST
+  NULL              // H_FLIPADST
+};
+static const transform_1d_avx2 row_txfm8x32_arr[TX_TYPES] = {
+  av1_fdct32_avx2,  // DCT_DCT
+  NULL,             // ADST_DCT
+  NULL,             // DCT_ADST
+  NULL,             // ADST_ADST
+  NULL,             // FLIPADST_DCT
+  NULL,             // DCT_FLIPADST
+  NULL,             // FLIPADST_FLIPADST
+  NULL,             // ADST_FLIPADST
+  NULL,             // FLIPADST_ADST
+  idtx32x32_avx2,   // IDTX
+  NULL,             // V_DCT
+  NULL,             // H_DCT
+  NULL,             // V_ADST
+  NULL,             // H_ADST
+  NULL,             // V_FLIPADST
+  NULL              // H_FLIPADST
+};
+void av1_fwd_txfm2d_32x32_avx2(const int16_t *input, int32_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  (void)bd;
+  __m256i buf0[128], buf1[128];
+  const int tx_size = TX_32X32;
+  const int8_t *shift = fwd_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int cos_bit_col = fwd_cos_bit_col[txw_idx][txh_idx];
+  const int cos_bit_row = fwd_cos_bit_row[txw_idx][txh_idx];
+  const int width = tx_size_wide[tx_size];
+  const int height = tx_size_high[tx_size];
+  const transform_1d_avx2 col_txfm = col_txfm8x32_arr[tx_type];
+  const transform_1d_avx2 row_txfm = row_txfm8x32_arr[tx_type];
+  int r, c;
+  const int width_div16 = (width >> 4);
+  const int width_div8 = (width >> 3);
+
+  for (int i = 0; i < width_div16; i++) {
+    av1_load_buffer_16xn_avx2(input + (i << 4), &buf0[(i << 1)], stride, height,
+                              width_div8, 0, 0);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1)], height, shift[0], width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1) + 1], height, shift[0],
+                                width_div8);
+    col_txfm(&buf0[(i << 1)], &buf0[(i << 1)], cos_bit_col, width_div8,
+             width_div8);
+    col_txfm(&buf0[(i << 1) + 1], &buf0[(i << 1) + 1], cos_bit_col, width_div8,
+             width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1)], height, shift[1], width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1) + 1], height, shift[1],
+                                width_div8);
+  }
+
+  for (r = 0; r < height; r += 8) {
+    for (c = 0; c < width_div8; c++) {
+      av1_fwd_txfm_transpose_8x8_avx2(&buf0[r * width_div8 + c],
+                                      &buf1[c * 8 * width_div8 + (r >> 3)],
+                                      width_div8, width_div8);
+    }
+  }
+
+  for (int i = 0; i < width_div16; i++) {
+    row_txfm(&buf1[(i << 1)], &buf1[(i << 1)], cos_bit_row, width_div8,
+             width_div8);
+    row_txfm(&buf1[(i << 1) + 1], &buf1[(i << 1) + 1], cos_bit_row, width_div8,
+             width_div8);
+    av1_round_shift_32_8xn_avx2(&buf1[(i << 1)], height, shift[2], width_div8);
+    av1_round_shift_32_8xn_avx2(&buf1[(i << 1) + 1], height, shift[2],
+                                width_div8);
+  }
+
+  for (r = 0; r < height; r += 8) {
+    for (c = 0; c < width_div8; c++) {
+      av1_fwd_txfm_transpose_8x8_avx2(&buf1[r * width_div8 + c],
+                                      &buf0[c * 8 * width_div8 + (r >> 3)],
+                                      width_div8, width_div8);
+    }
+  }
+
+  av1_store_buffer_avx2(buf0, output, 8, 128);
+}
+static INLINE void av1_fdct64_stage2_avx2(__m256i *x1, __m256i *x2,
+                                          __m256i *cospi_m32,
+                                          __m256i *cospi_p32,
+                                          const __m256i *__rounding,
+                                          int8_t cos_bit) {
+  x2[0] = _mm256_add_epi32(x1[0], x1[31]);
+  x2[31] = _mm256_sub_epi32(x1[0], x1[31]);
+  x2[1] = _mm256_add_epi32(x1[1], x1[30]);
+  x2[30] = _mm256_sub_epi32(x1[1], x1[30]);
+  x2[2] = _mm256_add_epi32(x1[2], x1[29]);
+  x2[29] = _mm256_sub_epi32(x1[2], x1[29]);
+  x2[3] = _mm256_add_epi32(x1[3], x1[28]);
+  x2[28] = _mm256_sub_epi32(x1[3], x1[28]);
+  x2[4] = _mm256_add_epi32(x1[4], x1[27]);
+  x2[27] = _mm256_sub_epi32(x1[4], x1[27]);
+  x2[5] = _mm256_add_epi32(x1[5], x1[26]);
+  x2[26] = _mm256_sub_epi32(x1[5], x1[26]);
+  x2[6] = _mm256_add_epi32(x1[6], x1[25]);
+  x2[25] = _mm256_sub_epi32(x1[6], x1[25]);
+  x2[7] = _mm256_add_epi32(x1[7], x1[24]);
+  x2[24] = _mm256_sub_epi32(x1[7], x1[24]);
+  x2[8] = _mm256_add_epi32(x1[8], x1[23]);
+  x2[23] = _mm256_sub_epi32(x1[8], x1[23]);
+  x2[9] = _mm256_add_epi32(x1[9], x1[22]);
+  x2[22] = _mm256_sub_epi32(x1[9], x1[22]);
+  x2[10] = _mm256_add_epi32(x1[10], x1[21]);
+  x2[21] = _mm256_sub_epi32(x1[10], x1[21]);
+  x2[11] = _mm256_add_epi32(x1[11], x1[20]);
+  x2[20] = _mm256_sub_epi32(x1[11], x1[20]);
+  x2[12] = _mm256_add_epi32(x1[12], x1[19]);
+  x2[19] = _mm256_sub_epi32(x1[12], x1[19]);
+  x2[13] = _mm256_add_epi32(x1[13], x1[18]);
+  x2[18] = _mm256_sub_epi32(x1[13], x1[18]);
+  x2[14] = _mm256_add_epi32(x1[14], x1[17]);
+  x2[17] = _mm256_sub_epi32(x1[14], x1[17]);
+  x2[15] = _mm256_add_epi32(x1[15], x1[16]);
+  x2[16] = _mm256_sub_epi32(x1[15], x1[16]);
+  x2[32] = x1[32];
+  x2[33] = x1[33];
+  x2[34] = x1[34];
+  x2[35] = x1[35];
+  x2[36] = x1[36];
+  x2[37] = x1[37];
+  x2[38] = x1[38];
+  x2[39] = x1[39];
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[40], x1[55], x2[40], x2[55],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[41], x1[54], x2[41], x2[54],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[42], x1[53], x2[42], x2[53],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[43], x1[52], x2[43], x2[52],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[44], x1[51], x2[44], x2[51],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[45], x1[50], x2[45], x2[50],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[46], x1[49], x2[46], x2[49],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x1[47], x1[48], x2[47], x2[48],
+                        *__rounding, cos_bit);
+  x2[56] = x1[56];
+  x2[57] = x1[57];
+  x2[58] = x1[58];
+  x2[59] = x1[59];
+  x2[60] = x1[60];
+  x2[61] = x1[61];
+  x2[62] = x1[62];
+  x2[63] = x1[63];
+}
+static INLINE void av1_fdct64_stage3_avx2(__m256i *x2, __m256i *x3,
+                                          __m256i *cospi_m32,
+                                          __m256i *cospi_p32,
+                                          const __m256i *__rounding,
+                                          int8_t cos_bit) {
+  x3[0] = _mm256_add_epi32(x2[0], x2[15]);
+  x3[15] = _mm256_sub_epi32(x2[0], x2[15]);
+  x3[1] = _mm256_add_epi32(x2[1], x2[14]);
+  x3[14] = _mm256_sub_epi32(x2[1], x2[14]);
+  x3[2] = _mm256_add_epi32(x2[2], x2[13]);
+  x3[13] = _mm256_sub_epi32(x2[2], x2[13]);
+  x3[3] = _mm256_add_epi32(x2[3], x2[12]);
+  x3[12] = _mm256_sub_epi32(x2[3], x2[12]);
+  x3[4] = _mm256_add_epi32(x2[4], x2[11]);
+  x3[11] = _mm256_sub_epi32(x2[4], x2[11]);
+  x3[5] = _mm256_add_epi32(x2[5], x2[10]);
+  x3[10] = _mm256_sub_epi32(x2[5], x2[10]);
+  x3[6] = _mm256_add_epi32(x2[6], x2[9]);
+  x3[9] = _mm256_sub_epi32(x2[6], x2[9]);
+  x3[7] = _mm256_add_epi32(x2[7], x2[8]);
+  x3[8] = _mm256_sub_epi32(x2[7], x2[8]);
+  x3[16] = x2[16];
+  x3[17] = x2[17];
+  x3[18] = x2[18];
+  x3[19] = x2[19];
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x2[20], x2[27], x3[20], x3[27],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x2[21], x2[26], x3[21], x3[26],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x2[22], x2[25], x3[22], x3[25],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x2[23], x2[24], x3[23], x3[24],
+                        *__rounding, cos_bit);
+  x3[28] = x2[28];
+  x3[29] = x2[29];
+  x3[30] = x2[30];
+  x3[31] = x2[31];
+  x3[32] = _mm256_add_epi32(x2[32], x2[47]);
+  x3[47] = _mm256_sub_epi32(x2[32], x2[47]);
+  x3[33] = _mm256_add_epi32(x2[33], x2[46]);
+  x3[46] = _mm256_sub_epi32(x2[33], x2[46]);
+  x3[34] = _mm256_add_epi32(x2[34], x2[45]);
+  x3[45] = _mm256_sub_epi32(x2[34], x2[45]);
+  x3[35] = _mm256_add_epi32(x2[35], x2[44]);
+  x3[44] = _mm256_sub_epi32(x2[35], x2[44]);
+  x3[36] = _mm256_add_epi32(x2[36], x2[43]);
+  x3[43] = _mm256_sub_epi32(x2[36], x2[43]);
+  x3[37] = _mm256_add_epi32(x2[37], x2[42]);
+  x3[42] = _mm256_sub_epi32(x2[37], x2[42]);
+  x3[38] = _mm256_add_epi32(x2[38], x2[41]);
+  x3[41] = _mm256_sub_epi32(x2[38], x2[41]);
+  x3[39] = _mm256_add_epi32(x2[39], x2[40]);
+  x3[40] = _mm256_sub_epi32(x2[39], x2[40]);
+  x3[48] = _mm256_sub_epi32(x2[63], x2[48]);
+  x3[63] = _mm256_add_epi32(x2[63], x2[48]);
+  x3[49] = _mm256_sub_epi32(x2[62], x2[49]);
+  x3[62] = _mm256_add_epi32(x2[62], x2[49]);
+  x3[50] = _mm256_sub_epi32(x2[61], x2[50]);
+  x3[61] = _mm256_add_epi32(x2[61], x2[50]);
+  x3[51] = _mm256_sub_epi32(x2[60], x2[51]);
+  x3[60] = _mm256_add_epi32(x2[60], x2[51]);
+  x3[52] = _mm256_sub_epi32(x2[59], x2[52]);
+  x3[59] = _mm256_add_epi32(x2[59], x2[52]);
+  x3[53] = _mm256_sub_epi32(x2[58], x2[53]);
+  x3[58] = _mm256_add_epi32(x2[58], x2[53]);
+  x3[54] = _mm256_sub_epi32(x2[57], x2[54]);
+  x3[57] = _mm256_add_epi32(x2[57], x2[54]);
+  x3[55] = _mm256_sub_epi32(x2[56], x2[55]);
+  x3[56] = _mm256_add_epi32(x2[56], x2[55]);
+}
+static INLINE void av1_fdct64_stage4_avx2(
+    __m256i *x3, __m256i *x4, __m256i *cospi_m32, __m256i *cospi_p32,
+    __m256i *cospi_m16, __m256i *cospi_p48, __m256i *cospi_m48,
+    const __m256i *__rounding, int8_t cos_bit) {
+  x4[0] = _mm256_add_epi32(x3[0], x3[7]);
+  x4[7] = _mm256_sub_epi32(x3[0], x3[7]);
+  x4[1] = _mm256_add_epi32(x3[1], x3[6]);
+  x4[6] = _mm256_sub_epi32(x3[1], x3[6]);
+  x4[2] = _mm256_add_epi32(x3[2], x3[5]);
+  x4[5] = _mm256_sub_epi32(x3[2], x3[5]);
+  x4[3] = _mm256_add_epi32(x3[3], x3[4]);
+  x4[4] = _mm256_sub_epi32(x3[3], x3[4]);
+  x4[8] = x3[8];
+  x4[9] = x3[9];
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x3[10], x3[13], x4[10], x4[13],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x3[11], x3[12], x4[11], x4[12],
+                        *__rounding, cos_bit);
+  x4[14] = x3[14];
+  x4[15] = x3[15];
+  x4[16] = _mm256_add_epi32(x3[16], x3[23]);
+  x4[23] = _mm256_sub_epi32(x3[16], x3[23]);
+  x4[17] = _mm256_add_epi32(x3[17], x3[22]);
+  x4[22] = _mm256_sub_epi32(x3[17], x3[22]);
+  x4[18] = _mm256_add_epi32(x3[18], x3[21]);
+  x4[21] = _mm256_sub_epi32(x3[18], x3[21]);
+  x4[19] = _mm256_add_epi32(x3[19], x3[20]);
+  x4[20] = _mm256_sub_epi32(x3[19], x3[20]);
+  x4[24] = _mm256_sub_epi32(x3[31], x3[24]);
+  x4[31] = _mm256_add_epi32(x3[31], x3[24]);
+  x4[25] = _mm256_sub_epi32(x3[30], x3[25]);
+  x4[30] = _mm256_add_epi32(x3[30], x3[25]);
+  x4[26] = _mm256_sub_epi32(x3[29], x3[26]);
+  x4[29] = _mm256_add_epi32(x3[29], x3[26]);
+  x4[27] = _mm256_sub_epi32(x3[28], x3[27]);
+  x4[28] = _mm256_add_epi32(x3[28], x3[27]);
+  x4[32] = x3[32];
+  x4[33] = x3[33];
+  x4[34] = x3[34];
+  x4[35] = x3[35];
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x3[36], x3[59], x4[36], x4[59],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x3[37], x3[58], x4[37], x4[58],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x3[38], x3[57], x4[38], x4[57],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x3[39], x3[56], x4[39], x4[56],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x3[40], x3[55], x4[40], x4[55],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x3[41], x3[54], x4[41], x4[54],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x3[42], x3[53], x4[42], x4[53],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x3[43], x3[52], x4[43], x4[52],
+                        *__rounding, cos_bit);
+  x4[44] = x3[44];
+  x4[45] = x3[45];
+  x4[46] = x3[46];
+  x4[47] = x3[47];
+  x4[48] = x3[48];
+  x4[49] = x3[49];
+  x4[50] = x3[50];
+  x4[51] = x3[51];
+  x4[60] = x3[60];
+  x4[61] = x3[61];
+  x4[62] = x3[62];
+  x4[63] = x3[63];
+}
+static INLINE void av1_fdct64_stage5_avx2(
+    __m256i *x4, __m256i *x5, __m256i *cospi_m32, __m256i *cospi_p32,
+    __m256i *cospi_m16, __m256i *cospi_p48, __m256i *cospi_m48,
+    const __m256i *__rounding, int8_t cos_bit) {
+  x5[0] = _mm256_add_epi32(x4[0], x4[3]);
+  x5[3] = _mm256_sub_epi32(x4[0], x4[3]);
+  x5[1] = _mm256_add_epi32(x4[1], x4[2]);
+  x5[2] = _mm256_sub_epi32(x4[1], x4[2]);
+  x5[4] = x4[4];
+  btf_32_type0_avx2_new(*cospi_m32, *cospi_p32, x4[5], x4[6], x5[5], x5[6],
+                        *__rounding, cos_bit);
+  x5[7] = x4[7];
+  x5[8] = _mm256_add_epi32(x4[8], x4[11]);
+  x5[11] = _mm256_sub_epi32(x4[8], x4[11]);
+  x5[9] = _mm256_add_epi32(x4[9], x4[10]);
+  x5[10] = _mm256_sub_epi32(x4[9], x4[10]);
+  x5[12] = _mm256_sub_epi32(x4[15], x4[12]);
+  x5[15] = _mm256_add_epi32(x4[15], x4[12]);
+  x5[13] = _mm256_sub_epi32(x4[14], x4[13]);
+  x5[14] = _mm256_add_epi32(x4[14], x4[13]);
+  x5[16] = x4[16];
+  x5[17] = x4[17];
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x4[18], x4[29], x5[18], x5[29],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x4[19], x4[28], x5[19], x5[28],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x4[20], x4[27], x5[20], x5[27],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x4[21], x4[26], x5[21], x5[26],
+                        *__rounding, cos_bit);
+  x5[22] = x4[22];
+  x5[23] = x4[23];
+  x5[24] = x4[24];
+  x5[25] = x4[25];
+  x5[30] = x4[30];
+  x5[31] = x4[31];
+  x5[32] = _mm256_add_epi32(x4[32], x4[39]);
+  x5[39] = _mm256_sub_epi32(x4[32], x4[39]);
+  x5[33] = _mm256_add_epi32(x4[33], x4[38]);
+  x5[38] = _mm256_sub_epi32(x4[33], x4[38]);
+  x5[34] = _mm256_add_epi32(x4[34], x4[37]);
+  x5[37] = _mm256_sub_epi32(x4[34], x4[37]);
+  x5[35] = _mm256_add_epi32(x4[35], x4[36]);
+  x5[36] = _mm256_sub_epi32(x4[35], x4[36]);
+  x5[40] = _mm256_sub_epi32(x4[47], x4[40]);
+  x5[47] = _mm256_add_epi32(x4[47], x4[40]);
+  x5[41] = _mm256_sub_epi32(x4[46], x4[41]);
+  x5[46] = _mm256_add_epi32(x4[46], x4[41]);
+  x5[42] = _mm256_sub_epi32(x4[45], x4[42]);
+  x5[45] = _mm256_add_epi32(x4[45], x4[42]);
+  x5[43] = _mm256_sub_epi32(x4[44], x4[43]);
+  x5[44] = _mm256_add_epi32(x4[44], x4[43]);
+  x5[48] = _mm256_add_epi32(x4[48], x4[55]);
+  x5[55] = _mm256_sub_epi32(x4[48], x4[55]);
+  x5[49] = _mm256_add_epi32(x4[49], x4[54]);
+  x5[54] = _mm256_sub_epi32(x4[49], x4[54]);
+  x5[50] = _mm256_add_epi32(x4[50], x4[53]);
+  x5[53] = _mm256_sub_epi32(x4[50], x4[53]);
+  x5[51] = _mm256_add_epi32(x4[51], x4[52]);
+  x5[52] = _mm256_sub_epi32(x4[51], x4[52]);
+  x5[56] = _mm256_sub_epi32(x4[63], x4[56]);
+  x5[63] = _mm256_add_epi32(x4[63], x4[56]);
+  x5[57] = _mm256_sub_epi32(x4[62], x4[57]);
+  x5[62] = _mm256_add_epi32(x4[62], x4[57]);
+  x5[58] = _mm256_sub_epi32(x4[61], x4[58]);
+  x5[61] = _mm256_add_epi32(x4[61], x4[58]);
+  x5[59] = _mm256_sub_epi32(x4[60], x4[59]);
+  x5[60] = _mm256_add_epi32(x4[60], x4[59]);
+}
+static INLINE void av1_fdct64_stage6_avx2(
+    __m256i *x5, __m256i *x6, __m256i *cospi_p16, __m256i *cospi_p32,
+    __m256i *cospi_m16, __m256i *cospi_p48, __m256i *cospi_m48,
+    __m256i *cospi_m08, __m256i *cospi_p56, __m256i *cospi_m56,
+    __m256i *cospi_m40, __m256i *cospi_p24, __m256i *cospi_m24,
+    const __m256i *__rounding, int8_t cos_bit) {
+  btf_32_type0_avx2_new(*cospi_p32, *cospi_p32, x5[0], x5[1], x6[0], x6[1],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_p16, *cospi_p48, x5[3], x5[2], x6[2], x6[3],
+                        *__rounding, cos_bit);
+  x6[4] = _mm256_add_epi32(x5[4], x5[5]);
+  x6[5] = _mm256_sub_epi32(x5[4], x5[5]);
+  x6[6] = _mm256_sub_epi32(x5[7], x5[6]);
+  x6[7] = _mm256_add_epi32(x5[7], x5[6]);
+  x6[8] = x5[8];
+  btf_32_type0_avx2_new(*cospi_m16, *cospi_p48, x5[9], x5[14], x6[9], x6[14],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m48, *cospi_m16, x5[10], x5[13], x6[10], x6[13],
+                        *__rounding, cos_bit);
+  x6[11] = x5[11];
+  x6[12] = x5[12];
+  x6[15] = x5[15];
+  x6[16] = _mm256_add_epi32(x5[16], x5[19]);
+  x6[19] = _mm256_sub_epi32(x5[16], x5[19]);
+  x6[17] = _mm256_add_epi32(x5[17], x5[18]);
+  x6[18] = _mm256_sub_epi32(x5[17], x5[18]);
+  x6[20] = _mm256_sub_epi32(x5[23], x5[20]);
+  x6[23] = _mm256_add_epi32(x5[23], x5[20]);
+  x6[21] = _mm256_sub_epi32(x5[22], x5[21]);
+  x6[22] = _mm256_add_epi32(x5[22], x5[21]);
+  x6[24] = _mm256_add_epi32(x5[24], x5[27]);
+  x6[27] = _mm256_sub_epi32(x5[24], x5[27]);
+  x6[25] = _mm256_add_epi32(x5[25], x5[26]);
+  x6[26] = _mm256_sub_epi32(x5[25], x5[26]);
+  x6[28] = _mm256_sub_epi32(x5[31], x5[28]);
+  x6[31] = _mm256_add_epi32(x5[31], x5[28]);
+  x6[29] = _mm256_sub_epi32(x5[30], x5[29]);
+  x6[30] = _mm256_add_epi32(x5[30], x5[29]);
+  x6[32] = x5[32];
+  x6[33] = x5[33];
+  btf_32_type0_avx2_new(*cospi_m08, *cospi_p56, x5[34], x5[61], x6[34], x6[61],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m08, *cospi_p56, x5[35], x5[60], x6[35], x6[60],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m56, *cospi_m08, x5[36], x5[59], x6[36], x6[59],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m56, *cospi_m08, x5[37], x5[58], x6[37], x6[58],
+                        *__rounding, cos_bit);
+  x6[38] = x5[38];
+  x6[39] = x5[39];
+  x6[40] = x5[40];
+  x6[41] = x5[41];
+  btf_32_type0_avx2_new(*cospi_m40, *cospi_p24, x5[42], x5[53], x6[42], x6[53],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m40, *cospi_p24, x5[43], x5[52], x6[43], x6[52],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m24, *cospi_m40, x5[44], x5[51], x6[44], x6[51],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m24, *cospi_m40, x5[45], x5[50], x6[45], x6[50],
+                        *__rounding, cos_bit);
+  x6[46] = x5[46];
+  x6[47] = x5[47];
+  x6[48] = x5[48];
+  x6[49] = x5[49];
+  x6[54] = x5[54];
+  x6[55] = x5[55];
+  x6[56] = x5[56];
+  x6[57] = x5[57];
+  x6[62] = x5[62];
+  x6[63] = x5[63];
+}
+static INLINE void av1_fdct64_stage7_avx2(
+    __m256i *x6, __m256i *x7, __m256i *cospi_p08, __m256i *cospi_p56,
+    __m256i *cospi_p40, __m256i *cospi_p24, __m256i *cospi_m08,
+    __m256i *cospi_m56, __m256i *cospi_m40, __m256i *cospi_m24,
+    const __m256i *__rounding, int8_t cos_bit) {
+  x7[0] = x6[0];
+  x7[1] = x6[1];
+  x7[2] = x6[2];
+  x7[3] = x6[3];
+  btf_32_type0_avx2_new(*cospi_p08, *cospi_p56, x6[7], x6[4], x7[4], x7[7],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_p40, *cospi_p24, x6[6], x6[5], x7[5], x7[6],
+                        *__rounding, cos_bit);
+  x7[8] = _mm256_add_epi32(x6[8], x6[9]);
+  x7[9] = _mm256_sub_epi32(x6[8], x6[9]);
+  x7[10] = _mm256_sub_epi32(x6[11], x6[10]);
+  x7[11] = _mm256_add_epi32(x6[11], x6[10]);
+  x7[12] = _mm256_add_epi32(x6[12], x6[13]);
+  x7[13] = _mm256_sub_epi32(x6[12], x6[13]);
+  x7[14] = _mm256_sub_epi32(x6[15], x6[14]);
+  x7[15] = _mm256_add_epi32(x6[15], x6[14]);
+  x7[16] = x6[16];
+  btf_32_type0_avx2_new(*cospi_m08, *cospi_p56, x6[17], x6[30], x7[17], x7[30],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m56, *cospi_m08, x6[18], x6[29], x7[18], x7[29],
+                        *__rounding, cos_bit);
+  x7[19] = x6[19];
+  x7[20] = x6[20];
+  btf_32_type0_avx2_new(*cospi_m40, *cospi_p24, x6[21], x6[26], x7[21], x7[26],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(*cospi_m24, *cospi_m40, x6[22], x6[25], x7[22], x7[25],
+                        *__rounding, cos_bit);
+  x7[23] = x6[23];
+  x7[24] = x6[24];
+  x7[27] = x6[27];
+  x7[28] = x6[28];
+  x7[31] = x6[31];
+  x7[32] = _mm256_add_epi32(x6[32], x6[35]);
+  x7[35] = _mm256_sub_epi32(x6[32], x6[35]);
+  x7[33] = _mm256_add_epi32(x6[33], x6[34]);
+  x7[34] = _mm256_sub_epi32(x6[33], x6[34]);
+  x7[36] = _mm256_sub_epi32(x6[39], x6[36]);
+  x7[39] = _mm256_add_epi32(x6[39], x6[36]);
+  x7[37] = _mm256_sub_epi32(x6[38], x6[37]);
+  x7[38] = _mm256_add_epi32(x6[38], x6[37]);
+  x7[40] = _mm256_add_epi32(x6[40], x6[43]);
+  x7[43] = _mm256_sub_epi32(x6[40], x6[43]);
+  x7[41] = _mm256_add_epi32(x6[41], x6[42]);
+  x7[42] = _mm256_sub_epi32(x6[41], x6[42]);
+  x7[44] = _mm256_sub_epi32(x6[47], x6[44]);
+  x7[47] = _mm256_add_epi32(x6[47], x6[44]);
+  x7[45] = _mm256_sub_epi32(x6[46], x6[45]);
+  x7[46] = _mm256_add_epi32(x6[46], x6[45]);
+  x7[48] = _mm256_add_epi32(x6[48], x6[51]);
+  x7[51] = _mm256_sub_epi32(x6[48], x6[51]);
+  x7[49] = _mm256_add_epi32(x6[49], x6[50]);
+  x7[50] = _mm256_sub_epi32(x6[49], x6[50]);
+  x7[52] = _mm256_sub_epi32(x6[55], x6[52]);
+  x7[55] = _mm256_add_epi32(x6[55], x6[52]);
+  x7[53] = _mm256_sub_epi32(x6[54], x6[53]);
+  x7[54] = _mm256_add_epi32(x6[54], x6[53]);
+  x7[56] = _mm256_add_epi32(x6[56], x6[59]);
+  x7[59] = _mm256_sub_epi32(x6[56], x6[59]);
+  x7[57] = _mm256_add_epi32(x6[57], x6[58]);
+  x7[58] = _mm256_sub_epi32(x6[57], x6[58]);
+  x7[60] = _mm256_sub_epi32(x6[63], x6[60]);
+  x7[63] = _mm256_add_epi32(x6[63], x6[60]);
+  x7[61] = _mm256_sub_epi32(x6[62], x6[61]);
+  x7[62] = _mm256_add_epi32(x6[62], x6[61]);
+}
+static INLINE void av1_fdct64_stage8_avx2(__m256i *x7, __m256i *x8,
+                                          const int32_t *cospi,
+                                          const __m256i *__rounding,
+                                          int8_t cos_bit) {
+  __m256i cospi_p60 = _mm256_set1_epi32(cospi[60]);
+  __m256i cospi_p04 = _mm256_set1_epi32(cospi[4]);
+  __m256i cospi_p28 = _mm256_set1_epi32(cospi[28]);
+  __m256i cospi_p36 = _mm256_set1_epi32(cospi[36]);
+  __m256i cospi_p44 = _mm256_set1_epi32(cospi[44]);
+  __m256i cospi_p20 = _mm256_set1_epi32(cospi[20]);
+  __m256i cospi_p12 = _mm256_set1_epi32(cospi[12]);
+  __m256i cospi_p52 = _mm256_set1_epi32(cospi[52]);
+  __m256i cospi_m04 = _mm256_set1_epi32(-cospi[4]);
+  __m256i cospi_m60 = _mm256_set1_epi32(-cospi[60]);
+  __m256i cospi_m36 = _mm256_set1_epi32(-cospi[36]);
+  __m256i cospi_m28 = _mm256_set1_epi32(-cospi[28]);
+  __m256i cospi_m20 = _mm256_set1_epi32(-cospi[20]);
+  __m256i cospi_m44 = _mm256_set1_epi32(-cospi[44]);
+  __m256i cospi_m52 = _mm256_set1_epi32(-cospi[52]);
+  __m256i cospi_m12 = _mm256_set1_epi32(-cospi[12]);
+
+  x8[0] = x7[0];
+  x8[1] = x7[1];
+  x8[2] = x7[2];
+  x8[3] = x7[3];
+  x8[4] = x7[4];
+  x8[5] = x7[5];
+  x8[6] = x7[6];
+  x8[7] = x7[7];
+
+  btf_32_type0_avx2_new(cospi_p04, cospi_p60, x7[15], x7[8], x8[8], x8[15],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p36, cospi_p28, x7[14], x7[9], x8[9], x8[14],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p20, cospi_p44, x7[13], x7[10], x8[10], x8[13],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p52, cospi_p12, x7[12], x7[11], x8[11], x8[12],
+                        *__rounding, cos_bit);
+  x8[16] = _mm256_add_epi32(x7[16], x7[17]);
+  x8[17] = _mm256_sub_epi32(x7[16], x7[17]);
+  x8[18] = _mm256_sub_epi32(x7[19], x7[18]);
+  x8[19] = _mm256_add_epi32(x7[19], x7[18]);
+  x8[20] = _mm256_add_epi32(x7[20], x7[21]);
+  x8[21] = _mm256_sub_epi32(x7[20], x7[21]);
+  x8[22] = _mm256_sub_epi32(x7[23], x7[22]);
+  x8[23] = _mm256_add_epi32(x7[23], x7[22]);
+  x8[24] = _mm256_add_epi32(x7[24], x7[25]);
+  x8[25] = _mm256_sub_epi32(x7[24], x7[25]);
+  x8[26] = _mm256_sub_epi32(x7[27], x7[26]);
+  x8[27] = _mm256_add_epi32(x7[27], x7[26]);
+  x8[28] = _mm256_add_epi32(x7[28], x7[29]);
+  x8[29] = _mm256_sub_epi32(x7[28], x7[29]);
+  x8[30] = _mm256_sub_epi32(x7[31], x7[30]);
+  x8[31] = _mm256_add_epi32(x7[31], x7[30]);
+  x8[32] = x7[32];
+  btf_32_type0_avx2_new(cospi_m04, cospi_p60, x7[33], x7[62], x8[33], x8[62],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_m60, cospi_m04, x7[34], x7[61], x8[34], x8[61],
+                        *__rounding, cos_bit);
+  x8[35] = x7[35];
+  x8[36] = x7[36];
+  btf_32_type0_avx2_new(cospi_m36, cospi_p28, x7[37], x7[58], x8[37], x8[58],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_m28, cospi_m36, x7[38], x7[57], x8[38], x8[57],
+                        *__rounding, cos_bit);
+  x8[39] = x7[39];
+  x8[40] = x7[40];
+  btf_32_type0_avx2_new(cospi_m20, cospi_p44, x7[41], x7[54], x8[41], x8[54],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_m44, cospi_m20, x7[42], x7[53], x8[42], x8[53],
+                        *__rounding, cos_bit);
+  x8[43] = x7[43];
+  x8[44] = x7[44];
+  btf_32_type0_avx2_new(cospi_m52, cospi_p12, x7[45], x7[50], x8[45], x8[50],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_m12, cospi_m52, x7[46], x7[49], x8[46], x8[49],
+                        *__rounding, cos_bit);
+  x8[47] = x7[47];
+  x8[48] = x7[48];
+  x8[51] = x7[51];
+  x8[52] = x7[52];
+  x8[55] = x7[55];
+  x8[56] = x7[56];
+  x8[59] = x7[59];
+  x8[60] = x7[60];
+  x8[63] = x7[63];
+}
+static INLINE void av1_fdct64_stage9_avx2(__m256i *x8, __m256i *x9,
+                                          const int32_t *cospi,
+                                          const __m256i *__rounding,
+                                          int8_t cos_bit) {
+  __m256i cospi_p62 = _mm256_set1_epi32(cospi[62]);
+  __m256i cospi_p02 = _mm256_set1_epi32(cospi[2]);
+  __m256i cospi_p30 = _mm256_set1_epi32(cospi[30]);
+  __m256i cospi_p34 = _mm256_set1_epi32(cospi[34]);
+  __m256i cospi_p46 = _mm256_set1_epi32(cospi[46]);
+  __m256i cospi_p18 = _mm256_set1_epi32(cospi[18]);
+  __m256i cospi_p14 = _mm256_set1_epi32(cospi[14]);
+  __m256i cospi_p50 = _mm256_set1_epi32(cospi[50]);
+  __m256i cospi_p54 = _mm256_set1_epi32(cospi[54]);
+  __m256i cospi_p10 = _mm256_set1_epi32(cospi[10]);
+  __m256i cospi_p22 = _mm256_set1_epi32(cospi[22]);
+  __m256i cospi_p42 = _mm256_set1_epi32(cospi[42]);
+  __m256i cospi_p38 = _mm256_set1_epi32(cospi[38]);
+  __m256i cospi_p26 = _mm256_set1_epi32(cospi[26]);
+  __m256i cospi_p06 = _mm256_set1_epi32(cospi[6]);
+  __m256i cospi_p58 = _mm256_set1_epi32(cospi[58]);
+
+  x9[0] = x8[0];
+  x9[1] = x8[1];
+  x9[2] = x8[2];
+  x9[3] = x8[3];
+  x9[4] = x8[4];
+  x9[5] = x8[5];
+  x9[6] = x8[6];
+  x9[7] = x8[7];
+  x9[8] = x8[8];
+  x9[9] = x8[9];
+  x9[10] = x8[10];
+  x9[11] = x8[11];
+  x9[12] = x8[12];
+  x9[13] = x8[13];
+  x9[14] = x8[14];
+  x9[15] = x8[15];
+  btf_32_type0_avx2_new(cospi_p02, cospi_p62, x8[31], x8[16], x9[16], x9[31],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p34, cospi_p30, x8[30], x8[17], x9[17], x9[30],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p18, cospi_p46, x8[29], x8[18], x9[18], x9[29],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p50, cospi_p14, x8[28], x8[19], x9[19], x9[28],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p10, cospi_p54, x8[27], x8[20], x9[20], x9[27],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p42, cospi_p22, x8[26], x8[21], x9[21], x9[26],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p26, cospi_p38, x8[25], x8[22], x9[22], x9[25],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p58, cospi_p06, x8[24], x8[23], x9[23], x9[24],
+                        *__rounding, cos_bit);
+  x9[32] = _mm256_add_epi32(x8[32], x8[33]);
+  x9[33] = _mm256_sub_epi32(x8[32], x8[33]);
+  x9[34] = _mm256_sub_epi32(x8[35], x8[34]);
+  x9[35] = _mm256_add_epi32(x8[35], x8[34]);
+  x9[36] = _mm256_add_epi32(x8[36], x8[37]);
+  x9[37] = _mm256_sub_epi32(x8[36], x8[37]);
+  x9[38] = _mm256_sub_epi32(x8[39], x8[38]);
+  x9[39] = _mm256_add_epi32(x8[39], x8[38]);
+  x9[40] = _mm256_add_epi32(x8[40], x8[41]);
+  x9[41] = _mm256_sub_epi32(x8[40], x8[41]);
+  x9[42] = _mm256_sub_epi32(x8[43], x8[42]);
+  x9[43] = _mm256_add_epi32(x8[43], x8[42]);
+  x9[44] = _mm256_add_epi32(x8[44], x8[45]);
+  x9[45] = _mm256_sub_epi32(x8[44], x8[45]);
+  x9[46] = _mm256_sub_epi32(x8[47], x8[46]);
+  x9[47] = _mm256_add_epi32(x8[47], x8[46]);
+  x9[48] = _mm256_add_epi32(x8[48], x8[49]);
+  x9[49] = _mm256_sub_epi32(x8[48], x8[49]);
+  x9[50] = _mm256_sub_epi32(x8[51], x8[50]);
+  x9[51] = _mm256_add_epi32(x8[51], x8[50]);
+  x9[52] = _mm256_add_epi32(x8[52], x8[53]);
+  x9[53] = _mm256_sub_epi32(x8[52], x8[53]);
+  x9[54] = _mm256_sub_epi32(x8[55], x8[54]);
+  x9[55] = _mm256_add_epi32(x8[55], x8[54]);
+  x9[56] = _mm256_add_epi32(x8[56], x8[57]);
+  x9[57] = _mm256_sub_epi32(x8[56], x8[57]);
+  x9[58] = _mm256_sub_epi32(x8[59], x8[58]);
+  x9[59] = _mm256_add_epi32(x8[59], x8[58]);
+  x9[60] = _mm256_add_epi32(x8[60], x8[61]);
+  x9[61] = _mm256_sub_epi32(x8[60], x8[61]);
+  x9[62] = _mm256_sub_epi32(x8[63], x8[62]);
+  x9[63] = _mm256_add_epi32(x8[63], x8[62]);
+}
+static INLINE void av1_fdct64_stage10_avx2(__m256i *x9, __m256i *x10,
+                                           const int32_t *cospi,
+                                           const __m256i *__rounding,
+                                           int8_t cos_bit) {
+  __m256i cospi_p63 = _mm256_set1_epi32(cospi[63]);
+  __m256i cospi_p01 = _mm256_set1_epi32(cospi[1]);
+  __m256i cospi_p31 = _mm256_set1_epi32(cospi[31]);
+  __m256i cospi_p33 = _mm256_set1_epi32(cospi[33]);
+  __m256i cospi_p47 = _mm256_set1_epi32(cospi[47]);
+  __m256i cospi_p17 = _mm256_set1_epi32(cospi[17]);
+  __m256i cospi_p15 = _mm256_set1_epi32(cospi[15]);
+  __m256i cospi_p49 = _mm256_set1_epi32(cospi[49]);
+  __m256i cospi_p55 = _mm256_set1_epi32(cospi[55]);
+  __m256i cospi_p09 = _mm256_set1_epi32(cospi[9]);
+  __m256i cospi_p23 = _mm256_set1_epi32(cospi[23]);
+  __m256i cospi_p41 = _mm256_set1_epi32(cospi[41]);
+  __m256i cospi_p39 = _mm256_set1_epi32(cospi[39]);
+  __m256i cospi_p25 = _mm256_set1_epi32(cospi[25]);
+  __m256i cospi_p07 = _mm256_set1_epi32(cospi[7]);
+  __m256i cospi_p57 = _mm256_set1_epi32(cospi[57]);
+  __m256i cospi_p59 = _mm256_set1_epi32(cospi[59]);
+  __m256i cospi_p05 = _mm256_set1_epi32(cospi[5]);
+  __m256i cospi_p27 = _mm256_set1_epi32(cospi[27]);
+  __m256i cospi_p37 = _mm256_set1_epi32(cospi[37]);
+  __m256i cospi_p43 = _mm256_set1_epi32(cospi[43]);
+  __m256i cospi_p21 = _mm256_set1_epi32(cospi[21]);
+  __m256i cospi_p11 = _mm256_set1_epi32(cospi[11]);
+  __m256i cospi_p53 = _mm256_set1_epi32(cospi[53]);
+  __m256i cospi_p51 = _mm256_set1_epi32(cospi[51]);
+  __m256i cospi_p13 = _mm256_set1_epi32(cospi[13]);
+  __m256i cospi_p19 = _mm256_set1_epi32(cospi[19]);
+  __m256i cospi_p45 = _mm256_set1_epi32(cospi[45]);
+  __m256i cospi_p35 = _mm256_set1_epi32(cospi[35]);
+  __m256i cospi_p29 = _mm256_set1_epi32(cospi[29]);
+  __m256i cospi_p03 = _mm256_set1_epi32(cospi[3]);
+  __m256i cospi_p61 = _mm256_set1_epi32(cospi[61]);
+
+  x10[0] = x9[0];
+  x10[1] = x9[1];
+  x10[2] = x9[2];
+  x10[3] = x9[3];
+  x10[4] = x9[4];
+  x10[5] = x9[5];
+  x10[6] = x9[6];
+  x10[7] = x9[7];
+  x10[8] = x9[8];
+  x10[9] = x9[9];
+  x10[10] = x9[10];
+  x10[11] = x9[11];
+  x10[12] = x9[12];
+  x10[13] = x9[13];
+  x10[14] = x9[14];
+  x10[15] = x9[15];
+  x10[16] = x9[16];
+  x10[17] = x9[17];
+  x10[18] = x9[18];
+  x10[19] = x9[19];
+  x10[20] = x9[20];
+  x10[21] = x9[21];
+  x10[22] = x9[22];
+  x10[23] = x9[23];
+  x10[24] = x9[24];
+  x10[25] = x9[25];
+  x10[26] = x9[26];
+  x10[27] = x9[27];
+  x10[28] = x9[28];
+  x10[29] = x9[29];
+  x10[30] = x9[30];
+  x10[31] = x9[31];
+  btf_32_type0_avx2_new(cospi_p01, cospi_p63, x9[63], x9[32], x10[32], x10[63],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p33, cospi_p31, x9[62], x9[33], x10[33], x10[62],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p17, cospi_p47, x9[61], x9[34], x10[34], x10[61],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p49, cospi_p15, x9[60], x9[35], x10[35], x10[60],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p09, cospi_p55, x9[59], x9[36], x10[36], x10[59],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p41, cospi_p23, x9[58], x9[37], x10[37], x10[58],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p25, cospi_p39, x9[57], x9[38], x10[38], x10[57],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p57, cospi_p07, x9[56], x9[39], x10[39], x10[56],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p05, cospi_p59, x9[55], x9[40], x10[40], x10[55],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p37, cospi_p27, x9[54], x9[41], x10[41], x10[54],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p21, cospi_p43, x9[53], x9[42], x10[42], x10[53],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p53, cospi_p11, x9[52], x9[43], x10[43], x10[52],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p13, cospi_p51, x9[51], x9[44], x10[44], x10[51],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p45, cospi_p19, x9[50], x9[45], x10[45], x10[50],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p29, cospi_p35, x9[49], x9[46], x10[46], x10[49],
+                        *__rounding, cos_bit);
+  btf_32_type0_avx2_new(cospi_p61, cospi_p03, x9[48], x9[47], x10[47], x10[48],
+                        *__rounding, cos_bit);
+}
+static void av1_fdct64_avx2(__m256i *input, __m256i *output, int8_t cos_bit,
+                            const int instride, const int outstride) {
+  const int32_t *cospi = cospi_arr(cos_bit);
+  const __m256i __rounding = _mm256_set1_epi32(1 << (cos_bit - 1));
+  __m256i cospi_m32 = _mm256_set1_epi32(-cospi[32]);
+  __m256i cospi_p32 = _mm256_set1_epi32(cospi[32]);
+  __m256i cospi_m16 = _mm256_set1_epi32(-cospi[16]);
+  __m256i cospi_p48 = _mm256_set1_epi32(cospi[48]);
+  __m256i cospi_m48 = _mm256_set1_epi32(-cospi[48]);
+  __m256i cospi_p16 = _mm256_set1_epi32(cospi[16]);
+  __m256i cospi_m08 = _mm256_set1_epi32(-cospi[8]);
+  __m256i cospi_p56 = _mm256_set1_epi32(cospi[56]);
+  __m256i cospi_m56 = _mm256_set1_epi32(-cospi[56]);
+  __m256i cospi_m40 = _mm256_set1_epi32(-cospi[40]);
+  __m256i cospi_p24 = _mm256_set1_epi32(cospi[24]);
+  __m256i cospi_m24 = _mm256_set1_epi32(-cospi[24]);
+  __m256i cospi_p08 = _mm256_set1_epi32(cospi[8]);
+  __m256i cospi_p40 = _mm256_set1_epi32(cospi[40]);
+
+  int startidx = 0 * instride;
+  int endidx = 63 * instride;
+  // stage 1
+  __m256i x1[64];
+  x1[0] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[63] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[1] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[62] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[2] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[61] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[3] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[60] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[4] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[59] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[5] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[58] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[6] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[57] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[7] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[56] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[8] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[55] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[9] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[54] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[10] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[53] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[11] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[52] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[12] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[51] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[13] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[50] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[14] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[49] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[15] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[48] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[16] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[47] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[17] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[46] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[18] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[45] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[19] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[44] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[20] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[43] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[21] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[42] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[22] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[41] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[23] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[40] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[24] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[39] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[25] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[38] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[26] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[37] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[27] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[36] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[28] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[35] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[29] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[34] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[30] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[33] = _mm256_sub_epi32(input[startidx], input[endidx]);
+  startidx += instride;
+  endidx -= instride;
+  x1[31] = _mm256_add_epi32(input[startidx], input[endidx]);
+  x1[32] = _mm256_sub_epi32(input[startidx], input[endidx]);
+
+  // stage 2
+  __m256i x2[64];
+  av1_fdct64_stage2_avx2(x1, x2, &cospi_m32, &cospi_p32, &__rounding, cos_bit);
+  // stage 3
+  av1_fdct64_stage3_avx2(x2, x1, &cospi_m32, &cospi_p32, &__rounding, cos_bit);
+  // stage 4
+  av1_fdct64_stage4_avx2(x1, x2, &cospi_m32, &cospi_p32, &cospi_m16, &cospi_p48,
+                         &cospi_m48, &__rounding, cos_bit);
+  // stage 5
+  av1_fdct64_stage5_avx2(x2, x1, &cospi_m32, &cospi_p32, &cospi_m16, &cospi_p48,
+                         &cospi_m48, &__rounding, cos_bit);
+  // stage 6
+  av1_fdct64_stage6_avx2(x1, x2, &cospi_p16, &cospi_p32, &cospi_m16, &cospi_p48,
+                         &cospi_m48, &cospi_m08, &cospi_p56, &cospi_m56,
+                         &cospi_m40, &cospi_p24, &cospi_m24, &__rounding,
+                         cos_bit);
+  // stage 7
+  av1_fdct64_stage7_avx2(x2, x1, &cospi_p08, &cospi_p56, &cospi_p40, &cospi_p24,
+                         &cospi_m08, &cospi_m56, &cospi_m40, &cospi_m24,
+                         &__rounding, cos_bit);
+  // stage 8
+  av1_fdct64_stage8_avx2(x1, x2, cospi, &__rounding, cos_bit);
+  // stage 9
+  av1_fdct64_stage9_avx2(x2, x1, cospi, &__rounding, cos_bit);
+  // stage 10
+  av1_fdct64_stage10_avx2(x1, x2, cospi, &__rounding, cos_bit);
+
+  startidx = 0 * outstride;
+  endidx = 63 * outstride;
+
+  // stage 11
+  output[startidx] = x2[0];
+  output[endidx] = x2[63];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[32];
+  output[endidx] = x2[31];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[16];
+  output[endidx] = x2[47];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[48];
+  output[endidx] = x2[15];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[8];
+  output[endidx] = x2[55];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[40];
+  output[endidx] = x2[23];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[24];
+  output[endidx] = x2[39];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[56];
+  output[endidx] = x2[7];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[4];
+  output[endidx] = x2[59];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[36];
+  output[endidx] = x2[27];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[20];
+  output[endidx] = x2[43];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[52];
+  output[endidx] = x2[11];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[12];
+  output[endidx] = x2[51];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[44];
+  output[endidx] = x2[19];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[28];
+  output[endidx] = x2[35];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[60];
+  output[endidx] = x2[3];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[2];
+  output[endidx] = x2[61];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[34];
+  output[endidx] = x2[29];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[18];
+  output[endidx] = x2[45];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[50];
+  output[endidx] = x2[13];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[10];
+  output[endidx] = x2[53];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[42];
+  output[endidx] = x2[21];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[26];
+  output[endidx] = x2[37];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[58];
+  output[endidx] = x2[5];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[6];
+  output[endidx] = x2[57];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[38];
+  output[endidx] = x2[25];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[22];
+  output[endidx] = x2[41];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[54];
+  output[endidx] = x2[9];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[14];
+  output[endidx] = x2[49];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[46];
+  output[endidx] = x2[17];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[30];
+  output[endidx] = x2[33];
+  startidx += outstride;
+  endidx -= outstride;
+  output[startidx] = x2[62];
+  output[endidx] = x2[1];
+}
+void av1_fwd_txfm2d_64x64_avx2(const int16_t *input, int32_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  (void)bd;
+  (void)tx_type;
+  assert(tx_type == DCT_DCT);
+  const TX_SIZE tx_size = TX_64X64;
+  __m256i buf0[512], buf1[512];
+  const int8_t *shift = fwd_txfm_shift_ls[tx_size];
+  const int txw_idx = get_txw_idx(tx_size);
+  const int txh_idx = get_txh_idx(tx_size);
+  const int cos_bit_col = fwd_cos_bit_col[txw_idx][txh_idx];
+  const int cos_bit_row = fwd_cos_bit_row[txw_idx][txh_idx];
+  const int width = tx_size_wide[tx_size];
+  const int height = tx_size_high[tx_size];
+  const transform_1d_avx2 col_txfm = av1_fdct64_avx2;
+  const transform_1d_avx2 row_txfm = av1_fdct64_avx2;
+  const int width_div16 = (width >> 4);
+  const int width_div8 = (width >> 3);
+  int r, c;
+  for (int i = 0; i < width_div16; i++) {
+    av1_load_buffer_16xn_avx2(input + (i << 4), &buf0[i << 1], stride, height,
+                              width_div8, 0, 0);
+    av1_round_shift_32_8xn_avx2(&buf0[i << 1], height, shift[0], width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1) + 1], height, shift[0],
+                                width_div8);
+    col_txfm(&buf0[i << 1], &buf0[i << 1], cos_bit_col, width_div8, width_div8);
+    col_txfm(&buf0[(i << 1) + 1], &buf0[(i << 1) + 1], cos_bit_col, width_div8,
+             width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[i << 1], height, shift[1], width_div8);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1) + 1], height, shift[1],
+                                width_div8);
+  }
+
+  for (r = 0; r < height; r += 8) {
+    for (c = 0; c < width_div8; c++) {
+      av1_fwd_txfm_transpose_8x8_avx2(&buf0[r * width_div8 + c],
+                                      &buf1[c * 8 * width_div8 + (r >> 3)],
+                                      width_div8, width_div8);
+    }
+  }
+
+  for (int i = 0; i < 2; i++) {
+    row_txfm(&buf1[i << 1], &buf0[i << 1], cos_bit_row, width_div8,
+             width_div16);
+    row_txfm(&buf1[(i << 1) + 1], &buf0[(i << 1) + 1], cos_bit_row, width_div8,
+             width_div16);
+    av1_round_shift_32_8xn_avx2(&buf0[i << 1], (height >> 1), shift[2],
+                                width_div16);
+    av1_round_shift_32_8xn_avx2(&buf0[(i << 1) + 1], (height >> 1), shift[2],
+                                width_div16);
+  }
+
+  for (r = 0; r < (height >> 1); r += 8) {
+    for (c = 0; c < width_div16; c++) {
+      av1_fwd_txfm_transpose_8x8_avx2(&buf0[r * width_div16 + c],
+                                      &buf1[c * 8 * width_div16 + (r >> 3)],
+                                      width_div16, width_div16);
+    }
+  }
+  av1_store_buffer_avx2(buf1, output, 8, 128);
+}
diff --git a/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
index 535485ae8e..d1059779dd 100644
--- a/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
+++ b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
@@ -59,7 +59,8 @@ static INLINE void load_buffer_4x4(const int16_t *input, __m128i *in,
 // shift[0] is used in load_buffer_4x4()
 // shift[1] is used in txfm_func_col()
 // shift[2] is used in txfm_func_row()
-static void fdct4x4_sse4_1(__m128i *in, int bit) {
+static void fdct4x4_sse4_1(__m128i *in, __m128i *out, int bit,
+                           const int num_col) {
   const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
@@ -69,10 +70,12 @@ static void fdct4x4_sse4_1(__m128i *in, int bit) {
   __m128i u0, u1, u2, u3;
   __m128i v0, v1, v2, v3;
 
-  s0 = _mm_add_epi32(in[0], in[3]);
-  s1 = _mm_add_epi32(in[1], in[2]);
-  s2 = _mm_sub_epi32(in[1], in[2]);
-  s3 = _mm_sub_epi32(in[0], in[3]);
+  int endidx = 3 * num_col;
+  s0 = _mm_add_epi32(in[0], in[endidx]);
+  s3 = _mm_sub_epi32(in[0], in[endidx]);
+  endidx -= num_col;
+  s1 = _mm_add_epi32(in[num_col], in[endidx]);
+  s2 = _mm_sub_epi32(in[num_col], in[endidx]);
 
   // btf_32_sse4_1_type0(cospi32, cospi32, s[01], u[02], bit);
   u0 = _mm_mullo_epi32(s0, cospi32);
@@ -109,10 +112,10 @@ static void fdct4x4_sse4_1(__m128i *in, int bit) {
   v2 = _mm_unpacklo_epi32(u2, u3);
   v3 = _mm_unpackhi_epi32(u2, u3);
 
-  in[0] = _mm_unpacklo_epi64(v0, v2);
-  in[1] = _mm_unpackhi_epi64(v0, v2);
-  in[2] = _mm_unpacklo_epi64(v1, v3);
-  in[3] = _mm_unpackhi_epi64(v1, v3);
+  out[0] = _mm_unpacklo_epi64(v0, v2);
+  out[1] = _mm_unpackhi_epi64(v0, v2);
+  out[2] = _mm_unpacklo_epi64(v1, v3);
+  out[3] = _mm_unpackhi_epi64(v1, v3);
 }
 
 static INLINE void write_buffer_4x4(__m128i *res, int32_t *output) {
@@ -122,7 +125,8 @@ static INLINE void write_buffer_4x4(__m128i *res, int32_t *output) {
   _mm_store_si128((__m128i *)(output + 3 * 4), res[3]);
 }
 
-static void fadst4x4_sse4_1(__m128i *in, int bit) {
+static void fadst4x4_sse4_1(__m128i *in, __m128i *out, int bit,
+                            const int num_col) {
   const int32_t *sinpi = sinpi_arr(bit);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
   const __m128i sinpi1 = _mm_set1_epi32((int)sinpi[1]);
@@ -135,15 +139,19 @@ static void fadst4x4_sse4_1(__m128i *in, int bit) {
   __m128i u0, u1, u2, u3;
   __m128i v0, v1, v2, v3;
 
-  s0 = _mm_mullo_epi32(in[0], sinpi1);
-  s1 = _mm_mullo_epi32(in[0], sinpi4);
-  s2 = _mm_mullo_epi32(in[1], sinpi2);
-  s3 = _mm_mullo_epi32(in[1], sinpi1);
-  s4 = _mm_mullo_epi32(in[2], sinpi3);
-  s5 = _mm_mullo_epi32(in[3], sinpi4);
-  s6 = _mm_mullo_epi32(in[3], sinpi2);
-  t = _mm_add_epi32(in[0], in[1]);
-  s7 = _mm_sub_epi32(t, in[3]);
+  int idx = 0 * num_col;
+  s0 = _mm_mullo_epi32(in[idx], sinpi1);
+  s1 = _mm_mullo_epi32(in[idx], sinpi4);
+  t = _mm_add_epi32(in[idx], in[idx + num_col]);
+  idx += num_col;
+  s2 = _mm_mullo_epi32(in[idx], sinpi2);
+  s3 = _mm_mullo_epi32(in[idx], sinpi1);
+  idx += num_col;
+  s4 = _mm_mullo_epi32(in[idx], sinpi3);
+  idx += num_col;
+  s5 = _mm_mullo_epi32(in[idx], sinpi4);
+  s6 = _mm_mullo_epi32(in[idx], sinpi2);
+  s7 = _mm_sub_epi32(t, in[idx]);
 
   t = _mm_add_epi32(s0, s2);
   x0 = _mm_add_epi32(t, s5);
@@ -175,12 +183,35 @@ static void fadst4x4_sse4_1(__m128i *in, int bit) {
   v2 = _mm_unpacklo_epi32(u2, u3);
   v3 = _mm_unpackhi_epi32(u2, u3);
 
-  in[0] = _mm_unpacklo_epi64(v0, v2);
-  in[1] = _mm_unpackhi_epi64(v0, v2);
-  in[2] = _mm_unpacklo_epi64(v1, v3);
-  in[3] = _mm_unpackhi_epi64(v1, v3);
+  out[0] = _mm_unpacklo_epi64(v0, v2);
+  out[1] = _mm_unpackhi_epi64(v0, v2);
+  out[2] = _mm_unpacklo_epi64(v1, v3);
+  out[3] = _mm_unpackhi_epi64(v1, v3);
 }
+static void idtx4x4_sse4_1(__m128i *in, __m128i *out, int bit, int col_num) {
+  (void)bit;
+  __m128i fact = _mm_set1_epi32(NewSqrt2);
+  __m128i offset = _mm_set1_epi32(1 << (NewSqrt2Bits - 1));
+  __m128i a_low;
+  __m128i v[4];
 
+  for (int i = 0; i < 4; i++) {
+    a_low = _mm_mullo_epi32(in[i * col_num], fact);
+    a_low = _mm_add_epi32(a_low, offset);
+    out[i] = _mm_srai_epi32(a_low, NewSqrt2Bits);
+  }
+
+  // Transpose for 4x4
+  v[0] = _mm_unpacklo_epi32(out[0], out[1]);
+  v[1] = _mm_unpackhi_epi32(out[0], out[1]);
+  v[2] = _mm_unpacklo_epi32(out[2], out[3]);
+  v[3] = _mm_unpackhi_epi32(out[2], out[3]);
+
+  out[0] = _mm_unpacklo_epi64(v[0], v[2]);
+  out[1] = _mm_unpackhi_epi64(v[0], v[2]);
+  out[2] = _mm_unpacklo_epi64(v[1], v[3]);
+  out[3] = _mm_unpackhi_epi64(v[1], v[3]);
+}
 void av1_fwd_txfm2d_4x4_sse4_1(const int16_t *input, int32_t *coeff,
                                int input_stride, TX_TYPE tx_type, int bd) {
   __m128i in[4];
@@ -191,56 +222,98 @@ void av1_fwd_txfm2d_4x4_sse4_1(const int16_t *input, int32_t *coeff,
   switch (tx_type) {
     case DCT_DCT:
       load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_DCT:
       load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case DCT_ADST:
       load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_ADST:
       load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case FLIPADST_DCT:
       load_buffer_4x4(input, in, input_stride, 1, 0, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case DCT_FLIPADST:
       load_buffer_4x4(input, in, input_stride, 0, 1, shift[0]);
-      fdct4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case FLIPADST_FLIPADST:
       load_buffer_4x4(input, in, input_stride, 1, 1, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_FLIPADST:
       load_buffer_4x4(input, in, input_stride, 0, 1, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     case FLIPADST_ADST:
       load_buffer_4x4(input, in, input_stride, 1, 0, shift[0]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_col[txw_idx][txh_idx]);
-      fadst4x4_sse4_1(in, fwd_cos_bit_row[txw_idx][txh_idx]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case IDTX:
+      load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case V_DCT:
+      load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case H_DCT:
+      load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      fdct4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case V_ADST:
+      load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case H_ADST:
+      load_buffer_4x4(input, in, input_stride, 0, 0, shift[0]);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_col[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case V_FLIPADST:
+      load_buffer_4x4(input, in, input_stride, 1, 0, shift[0]);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      write_buffer_4x4(in, coeff);
+      break;
+    case H_FLIPADST:
+      load_buffer_4x4(input, in, input_stride, 0, 1, shift[0]);
+      idtx4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
+      fadst4x4_sse4_1(in, in, fwd_cos_bit_row[txw_idx][txh_idx], 1);
       write_buffer_4x4(in, coeff);
       break;
     default: assert(0);
@@ -372,6 +445,28 @@ static INLINE void col_txfm_8x8_rounding(__m128i *in, int shift) {
   in[15] = _mm_srai_epi32(in[15], shift);
 }
 
+static INLINE void col_txfm_4x8_rounding(__m128i *in, int shift) {
+  const __m128i rounding = _mm_set1_epi32(1 << (shift - 1));
+
+  in[0] = _mm_add_epi32(in[0], rounding);
+  in[1] = _mm_add_epi32(in[1], rounding);
+  in[2] = _mm_add_epi32(in[2], rounding);
+  in[3] = _mm_add_epi32(in[3], rounding);
+  in[4] = _mm_add_epi32(in[4], rounding);
+  in[5] = _mm_add_epi32(in[5], rounding);
+  in[6] = _mm_add_epi32(in[6], rounding);
+  in[7] = _mm_add_epi32(in[7], rounding);
+
+  in[0] = _mm_srai_epi32(in[0], shift);
+  in[1] = _mm_srai_epi32(in[1], shift);
+  in[2] = _mm_srai_epi32(in[2], shift);
+  in[3] = _mm_srai_epi32(in[3], shift);
+  in[4] = _mm_srai_epi32(in[4], shift);
+  in[5] = _mm_srai_epi32(in[5], shift);
+  in[6] = _mm_srai_epi32(in[6], shift);
+  in[7] = _mm_srai_epi32(in[7], shift);
+}
+
 static INLINE void write_buffer_8x8(const __m128i *res, int32_t *output) {
   _mm_store_si128((__m128i *)(output + 0 * 4), res[0]);
   _mm_store_si128((__m128i *)(output + 1 * 4), res[1]);
@@ -417,9 +512,8 @@ static INLINE void write_buffer_16x8(const __m128i *res, int32_t *output,
   _mm_storeu_si128((__m128i *)(output + (stride * 7) + 4), res[15]);
 }
 
-static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit,
+static void fdct4x8_sse4_1(__m128i *in, __m128i *out, int bit,
                            const int col_num) {
-  (void)(col_num);
   const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospim32 = _mm_set1_epi32(-cospi[32]);
@@ -432,17 +526,25 @@ static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit,
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
   __m128i u[8], v[8];
 
+  int startidx = 0 * col_num;
+  int endidx = 7 * col_num;
   // Even 8 points 0, 2, ..., 14
   // stage 0
   // stage 1
-  u[0] = _mm_add_epi32(in[0], in[14]);
-  v[7] = _mm_sub_epi32(in[0], in[14]);  // v[7]
-  u[1] = _mm_add_epi32(in[2], in[12]);
-  u[6] = _mm_sub_epi32(in[2], in[12]);
-  u[2] = _mm_add_epi32(in[4], in[10]);
-  u[5] = _mm_sub_epi32(in[4], in[10]);
-  u[3] = _mm_add_epi32(in[6], in[8]);
-  v[4] = _mm_sub_epi32(in[6], in[8]);  // v[4]
+  u[0] = _mm_add_epi32(in[startidx], in[endidx]);
+  v[7] = _mm_sub_epi32(in[startidx], in[endidx]);  // v[7]
+  startidx += col_num;
+  endidx -= col_num;
+  u[1] = _mm_add_epi32(in[startidx], in[endidx]);
+  u[6] = _mm_sub_epi32(in[startidx], in[endidx]);
+  startidx += col_num;
+  endidx -= col_num;
+  u[2] = _mm_add_epi32(in[startidx], in[endidx]);
+  u[5] = _mm_sub_epi32(in[startidx], in[endidx]);
+  startidx += col_num;
+  endidx -= col_num;
+  u[3] = _mm_add_epi32(in[startidx], in[endidx]);
+  v[4] = _mm_sub_epi32(in[startidx], in[endidx]);  // v[4]
 
   // stage 2
   v[0] = _mm_add_epi32(u[0], u[3]);
@@ -498,126 +600,40 @@ static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit,
   v[1] = _mm_mullo_epi32(u[7], cospi8);
   v[0] = _mm_add_epi32(v[0], v[1]);
   v[0] = _mm_add_epi32(v[0], rnding);
-  out[2] = _mm_srai_epi32(v[0], bit);  // buf0[4]
+  out[1 * col_num] = _mm_srai_epi32(v[0], bit);  // buf0[4]
 
   v[0] = _mm_mullo_epi32(u[4], cospi8);
   v[1] = _mm_mullo_epi32(u[7], cospi56);
   v[0] = _mm_sub_epi32(v[1], v[0]);
   v[0] = _mm_add_epi32(v[0], rnding);
-  out[14] = _mm_srai_epi32(v[0], bit);  // buf0[7]
+  out[7 * col_num] = _mm_srai_epi32(v[0], bit);  // buf0[7]
 
   v[0] = _mm_mullo_epi32(u[5], cospi24);
   v[1] = _mm_mullo_epi32(u[6], cospi40);
   v[0] = _mm_add_epi32(v[0], v[1]);
   v[0] = _mm_add_epi32(v[0], rnding);
-  out[10] = _mm_srai_epi32(v[0], bit);  // buf0[5]
+  out[5 * col_num] = _mm_srai_epi32(v[0], bit);  // buf0[5]
 
   v[0] = _mm_mullo_epi32(u[5], cospi40);
   v[1] = _mm_mullo_epi32(u[6], cospi24);
   v[0] = _mm_sub_epi32(v[1], v[0]);
   v[0] = _mm_add_epi32(v[0], rnding);
-  out[6] = _mm_srai_epi32(v[0], bit);  // buf0[6]
+  out[3 * col_num] = _mm_srai_epi32(v[0], bit);  // buf0[6]
 
-  out[0] = u[0];   // buf0[0]
-  out[8] = u[1];   // buf0[1]
-  out[4] = u[2];   // buf0[2]
-  out[12] = u[3];  // buf0[3]
+  out[0 * col_num] = u[0];  // buf0[0]
+  out[4 * col_num] = u[1];  // buf0[1]
+  out[2 * col_num] = u[2];  // buf0[2]
+  out[6 * col_num] = u[3];  // buf0[3]
+}
 
-  // Odd 8 points: 1, 3, ..., 15
-  // stage 0
-  // stage 1
-  u[0] = _mm_add_epi32(in[1], in[15]);
-  v[7] = _mm_sub_epi32(in[1], in[15]);  // v[7]
-  u[1] = _mm_add_epi32(in[3], in[13]);
-  u[6] = _mm_sub_epi32(in[3], in[13]);
-  u[2] = _mm_add_epi32(in[5], in[11]);
-  u[5] = _mm_sub_epi32(in[5], in[11]);
-  u[3] = _mm_add_epi32(in[7], in[9]);
-  v[4] = _mm_sub_epi32(in[7], in[9]);  // v[4]
-
-  // stage 2
-  v[0] = _mm_add_epi32(u[0], u[3]);
-  v[3] = _mm_sub_epi32(u[0], u[3]);
-  v[1] = _mm_add_epi32(u[1], u[2]);
-  v[2] = _mm_sub_epi32(u[1], u[2]);
-
-  v[5] = _mm_mullo_epi32(u[5], cospim32);
-  v[6] = _mm_mullo_epi32(u[6], cospi32);
-  v[5] = _mm_add_epi32(v[5], v[6]);
-  v[5] = _mm_add_epi32(v[5], rnding);
-  v[5] = _mm_srai_epi32(v[5], bit);
-
-  u[0] = _mm_mullo_epi32(u[5], cospi32);
-  v[6] = _mm_mullo_epi32(u[6], cospim32);
-  v[6] = _mm_sub_epi32(u[0], v[6]);
-  v[6] = _mm_add_epi32(v[6], rnding);
-  v[6] = _mm_srai_epi32(v[6], bit);
-
-  // stage 3
-  // type 0
-  v[0] = _mm_mullo_epi32(v[0], cospi32);
-  v[1] = _mm_mullo_epi32(v[1], cospi32);
-  u[0] = _mm_add_epi32(v[0], v[1]);
-  u[0] = _mm_add_epi32(u[0], rnding);
-  u[0] = _mm_srai_epi32(u[0], bit);
-
-  u[1] = _mm_sub_epi32(v[0], v[1]);
-  u[1] = _mm_add_epi32(u[1], rnding);
-  u[1] = _mm_srai_epi32(u[1], bit);
-
-  // type 1
-  v[0] = _mm_mullo_epi32(v[2], cospi48);
-  v[1] = _mm_mullo_epi32(v[3], cospi16);
-  u[2] = _mm_add_epi32(v[0], v[1]);
-  u[2] = _mm_add_epi32(u[2], rnding);
-  u[2] = _mm_srai_epi32(u[2], bit);
-
-  v[0] = _mm_mullo_epi32(v[2], cospi16);
-  v[1] = _mm_mullo_epi32(v[3], cospi48);
-  u[3] = _mm_sub_epi32(v[1], v[0]);
-  u[3] = _mm_add_epi32(u[3], rnding);
-  u[3] = _mm_srai_epi32(u[3], bit);
-
-  u[4] = _mm_add_epi32(v[4], v[5]);
-  u[5] = _mm_sub_epi32(v[4], v[5]);
-  u[6] = _mm_sub_epi32(v[7], v[6]);
-  u[7] = _mm_add_epi32(v[7], v[6]);
-
-  // stage 4
-  // stage 5
-  v[0] = _mm_mullo_epi32(u[4], cospi56);
-  v[1] = _mm_mullo_epi32(u[7], cospi8);
-  v[0] = _mm_add_epi32(v[0], v[1]);
-  v[0] = _mm_add_epi32(v[0], rnding);
-  out[3] = _mm_srai_epi32(v[0], bit);  // buf0[4]
-
-  v[0] = _mm_mullo_epi32(u[4], cospi8);
-  v[1] = _mm_mullo_epi32(u[7], cospi56);
-  v[0] = _mm_sub_epi32(v[1], v[0]);
-  v[0] = _mm_add_epi32(v[0], rnding);
-  out[15] = _mm_srai_epi32(v[0], bit);  // buf0[7]
-
-  v[0] = _mm_mullo_epi32(u[5], cospi24);
-  v[1] = _mm_mullo_epi32(u[6], cospi40);
-  v[0] = _mm_add_epi32(v[0], v[1]);
-  v[0] = _mm_add_epi32(v[0], rnding);
-  out[11] = _mm_srai_epi32(v[0], bit);  // buf0[5]
-
-  v[0] = _mm_mullo_epi32(u[5], cospi40);
-  v[1] = _mm_mullo_epi32(u[6], cospi24);
-  v[0] = _mm_sub_epi32(v[1], v[0]);
-  v[0] = _mm_add_epi32(v[0], rnding);
-  out[7] = _mm_srai_epi32(v[0], bit);  // buf0[6]
-
-  out[1] = u[0];   // buf0[0]
-  out[9] = u[1];   // buf0[1]
-  out[5] = u[2];   // buf0[2]
-  out[13] = u[3];  // buf0[3]
+static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit,
+                           const int col_num) {
+  fdct4x8_sse4_1(in, out, bit, col_num);
+  fdct4x8_sse4_1(in + 1, out + 1, bit, col_num);
 }
 
 static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit,
                             const int col_num) {
-  (void)(col_num);
   const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
@@ -648,17 +664,17 @@ static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit,
   //  Odd column: 1, 3, ..., 15
   //  one even column plus one odd column constructs one row (8 coeffs)
   //  total we have 8 rows (8x8).
-  for (col = 0; col < 2; ++col) {
+  for (col = 0; col < col_num; ++col) {
     // stage 0
     // stage 1
-    u0 = in[2 * 0 + col];
-    u1 = _mm_sub_epi32(zero, in[2 * 7 + col]);
-    u2 = _mm_sub_epi32(zero, in[2 * 3 + col]);
-    u3 = in[2 * 4 + col];
-    u4 = _mm_sub_epi32(zero, in[2 * 1 + col]);
-    u5 = in[2 * 6 + col];
-    u6 = in[2 * 2 + col];
-    u7 = _mm_sub_epi32(zero, in[2 * 5 + col]);
+    u0 = in[col_num * 0 + col];
+    u1 = _mm_sub_epi32(zero, in[col_num * 7 + col]);
+    u2 = _mm_sub_epi32(zero, in[col_num * 3 + col]);
+    u3 = in[col_num * 4 + col];
+    u4 = _mm_sub_epi32(zero, in[col_num * 1 + col]);
+    u5 = in[col_num * 6 + col];
+    u6 = in[col_num * 2 + col];
+    u7 = _mm_sub_epi32(zero, in[col_num * 5 + col]);
 
     // stage 2
     v0 = u0;
@@ -787,17 +803,44 @@ static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit,
     v7 = _mm_srai_epi32(v7, bit);
 
     // stage 7
-    out[2 * 0 + col] = v1;
-    out[2 * 1 + col] = v6;
-    out[2 * 2 + col] = v3;
-    out[2 * 3 + col] = v4;
-    out[2 * 4 + col] = v5;
-    out[2 * 5 + col] = v2;
-    out[2 * 6 + col] = v7;
-    out[2 * 7 + col] = v0;
+    out[col_num * 0 + col] = v1;
+    out[col_num * 1 + col] = v6;
+    out[col_num * 2 + col] = v3;
+    out[col_num * 3 + col] = v4;
+    out[col_num * 4 + col] = v5;
+    out[col_num * 5 + col] = v2;
+    out[col_num * 6 + col] = v7;
+    out[col_num * 7 + col] = v0;
   }
 }
+static void idtx8x8_sse4_1(__m128i *in, __m128i *out, int bit, int col_num) {
+  (void)bit;
 
+  for (int i = 0; i < col_num; i += 1) {
+    out[0 + 8 * i] = _mm_add_epi32(in[0 + 8 * i], in[0 + 8 * i]);
+    out[1 + 8 * i] = _mm_add_epi32(in[1 + 8 * i], in[1 + 8 * i]);
+    out[2 + 8 * i] = _mm_add_epi32(in[2 + 8 * i], in[2 + 8 * i]);
+    out[3 + 8 * i] = _mm_add_epi32(in[3 + 8 * i], in[3 + 8 * i]);
+    out[4 + 8 * i] = _mm_add_epi32(in[4 + 8 * i], in[4 + 8 * i]);
+    out[5 + 8 * i] = _mm_add_epi32(in[5 + 8 * i], in[5 + 8 * i]);
+    out[6 + 8 * i] = _mm_add_epi32(in[6 + 8 * i], in[6 + 8 * i]);
+    out[7 + 8 * i] = _mm_add_epi32(in[7 + 8 * i], in[7 + 8 * i]);
+  }
+}
+static void idtx32x8_sse4_1(__m128i *in, __m128i *out, int bit, int col_num) {
+  (void)bit;
+  (void)col_num;
+  for (int j = 0; j < 2; j++) {
+    out[j + 8 * 0] = _mm_add_epi32(in[j + 8 * 0], in[j + 8 * 0]);
+    out[j + 8 * 1] = _mm_add_epi32(in[j + 8 * 1], in[j + 8 * 1]);
+    out[j + 8 * 2] = _mm_add_epi32(in[j + 8 * 2], in[j + 8 * 2]);
+    out[j + 8 * 3] = _mm_add_epi32(in[j + 8 * 3], in[j + 8 * 3]);
+    out[j + 8 * 4] = _mm_add_epi32(in[j + 8 * 4], in[j + 8 * 4]);
+    out[j + 8 * 5] = _mm_add_epi32(in[j + 8 * 5], in[j + 8 * 5]);
+    out[j + 8 * 6] = _mm_add_epi32(in[j + 8 * 6], in[j + 8 * 6]);
+    out[j + 8 * 7] = _mm_add_epi32(in[j + 8 * 7], in[j + 8 * 7]);
+  }
+}
 void av1_fwd_txfm2d_8x8_sse4_1(const int16_t *input, int32_t *coeff, int stride,
                                TX_TYPE tx_type, int bd) {
   __m128i in[16], out[16];
@@ -808,82 +851,145 @@ void av1_fwd_txfm2d_8x8_sse4_1(const int16_t *input, int32_t *coeff, int stride,
   switch (tx_type) {
     case DCT_DCT:
       load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_DCT:
       load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case DCT_ADST:
       load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_ADST:
       load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case FLIPADST_DCT:
       load_buffer_8x8(input, in, stride, 1, 0, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case DCT_FLIPADST:
       load_buffer_8x8(input, in, stride, 0, 1, shift[0]);
-      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case FLIPADST_FLIPADST:
       load_buffer_8x8(input, in, stride, 1, 1, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_FLIPADST:
       load_buffer_8x8(input, in, stride, 0, 1, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case FLIPADST_ADST:
       load_buffer_8x8(input, in, stride, 1, 0, shift[0]);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       col_txfm_8x8_rounding(out, -shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 0);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case IDTX:
+      load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case V_DCT:
+      load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case H_DCT:
+      load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      fdct8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case V_ADST:
+      load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case H_ADST:
+      load_buffer_8x8(input, in, stride, 0, 0, shift[0]);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case V_FLIPADST:
+      load_buffer_8x8(input, in, stride, 1, 0, shift[0]);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      transpose_8x8(out, in);
+      write_buffer_8x8(in, coeff);
+      break;
+    case H_FLIPADST:
+      load_buffer_8x8(input, in, stride, 0, 1, shift[0]);
+      idtx8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
+      col_txfm_8x8_rounding(out, -shift[1]);
+      transpose_8x8(out, in);
+      fadst8x8_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], 2);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
@@ -986,6 +1092,90 @@ static INLINE void load_buffer_8x16(const int16_t *input, __m128i *out,
   load_buffer_8x8(botL, out + 16, stride, flipud, fliplr, shift);
 }
 
+static INLINE void load_buffer_8x4(const int16_t *input, __m128i *out,
+                                   int stride, int flipud, int fliplr,
+                                   int shift) {
+  const int16_t *topL = input;
+  const int16_t *topR = input + 4;
+
+  const int16_t *tmp;
+
+  if (fliplr) {
+    tmp = topL;
+    topL = topR;
+    topR = tmp;
+  }
+
+  load_buffer_4x4(topL, out, stride, flipud, fliplr, shift);
+  load_buffer_4x4(topR, out + 4, stride, flipud, fliplr, shift);
+}
+
+static INLINE void load_buffer_16x4(const int16_t *input, __m128i *out,
+                                    int stride, int flipud, int fliplr,
+                                    int shift) {
+  const int16_t *topL = input;
+  const int16_t *topR = input + 8;
+
+  const int16_t *tmp;
+
+  if (fliplr) {
+    tmp = topL;
+    topL = topR;
+    topR = tmp;
+  }
+
+  load_buffer_8x4(topL, out, stride, flipud, fliplr, shift);
+  load_buffer_8x4(topR, out + 8, stride, flipud, fliplr, shift);
+}
+
+static INLINE void load_buffer_4x8(const int16_t *input, __m128i *out,
+                                   int stride, int flipud, int fliplr,
+                                   int shift) {
+  const int16_t *topL = input;
+  const int16_t *botL = input + 4 * stride;
+
+  const int16_t *tmp;
+
+  if (flipud) {
+    tmp = topL;
+    topL = botL;
+    botL = tmp;
+  }
+
+  load_buffer_4x4(topL, out, stride, flipud, fliplr, shift);
+  load_buffer_4x4(botL, out + 4, stride, flipud, fliplr, shift);
+}
+
+static INLINE void load_buffer_4x16(const int16_t *input, __m128i *out,
+                                    const int stride, const int flipud,
+                                    const int fliplr, const int shift) {
+  const int16_t *topL = input;
+  const int16_t *botL = input + 8 * stride;
+
+  const int16_t *tmp;
+
+  if (flipud) {
+    tmp = topL;
+    topL = botL;
+    botL = tmp;
+  }
+  load_buffer_4x8(topL, out, stride, flipud, fliplr, shift);
+  load_buffer_4x8(botL, out + 8, stride, flipud, fliplr, shift);
+}
+
+static INLINE void load_buffer_32x8n(const int16_t *input, __m128i *out,
+                                     int stride, int flipud, int fliplr,
+                                     int shift, const int height) {
+  const int16_t *in = input;
+  __m128i *output = out;
+  for (int col = 0; col < height; col++) {
+    in = input + col * stride;
+    output = out + col * 8;
+    load_buffer_4x4(in, output, 4, flipud, fliplr, shift);
+    load_buffer_4x4((in + 16), (output + 4), 4, flipud, fliplr, shift);
+  }
+}
+
 static void fdct16x16_sse4_1(__m128i *in, __m128i *out, int bit,
                              const int col_num) {
   const int32_t *cospi = cospi_arr(bit);
@@ -1544,7 +1734,19 @@ static void write_buffer_16x16(const __m128i *in, int32_t *output) {
   output += size_8x8;
   write_buffer_8x8(&in[48], output);
 }
+static void idtx16x16_sse4_1(__m128i *in, __m128i *out, int bit, int col_num) {
+  (void)bit;
+  __m128i fact = _mm_set1_epi32(2 * NewSqrt2);
+  __m128i offset = _mm_set1_epi32(1 << (NewSqrt2Bits - 1));
+  __m128i a_low;
 
+  int num_iters = 16 * col_num;
+  for (int i = 0; i < num_iters; i++) {
+    a_low = _mm_mullo_epi32(in[i], fact);
+    a_low = _mm_add_epi32(a_low, offset);
+    out[i] = _mm_srai_epi32(a_low, NewSqrt2Bits);
+  }
+}
 void av1_fwd_txfm2d_16x16_sse4_1(const int16_t *input, int32_t *coeff,
                                  int stride, TX_TYPE tx_type, int bd) {
   __m128i in[64], out[64];
@@ -1634,6 +1836,69 @@ void av1_fwd_txfm2d_16x16_sse4_1(const int16_t *input, int32_t *coeff,
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
+    case IDTX:
+      load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case V_DCT:
+      load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+      fdct16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case H_DCT:
+      load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      fdct16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case V_ADST:
+      load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+      fadst16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case H_ADST:
+      load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      fadst16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case V_FLIPADST:
+      load_buffer_16x16(input, in, stride, 1, 0, shift[0]);
+      fadst16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
+    case H_FLIPADST:
+      load_buffer_16x16(input, in, stride, 0, 1, shift[0]);
+      idtx16x16_sse4_1(in, out, fwd_cos_bit_col[txw_idx][txh_idx], col_num);
+      col_txfm_16x16_rounding(out, -shift[1]);
+      transpose_16x16(out, in);
+      fadst16x16_sse4_1(in, out, fwd_cos_bit_row[txw_idx][txh_idx], col_num);
+      transpose_16x16(out, in);
+      write_buffer_16x16(in, coeff);
+      break;
     default: assert(0);
   }
   (void)bd;
@@ -1654,13 +1919,49 @@ static const fwd_transform_1d_sse4_1 col_highbd_txfm8x8_arr[TX_TYPES] = {
   fadst8x8_sse4_1,  // FLIPADST_FLIPADST
   fadst8x8_sse4_1,  // ADST_FLIPADST
   fadst8x8_sse4_1,  // FLIPADST_ADST
-  NULL,             // IDTX
+  idtx8x8_sse4_1,   // IDTX
+  fdct8x8_sse4_1,   // V_DCT
+  idtx8x8_sse4_1,   // H_DCT
+  fadst8x8_sse4_1,  // V_ADST
+  idtx8x8_sse4_1,   // H_ADST
+  fadst8x8_sse4_1,  // V_FLIPADST
+  idtx8x8_sse4_1    // H_FLIPADST
+};
+static const fwd_transform_1d_sse4_1 row_highbd_txfm32x8_arr[TX_TYPES] = {
+  fdct8x8_sse4_1,   // DCT_DCT
+  NULL,             // ADST_DCT
+  NULL,             // DCT_ADST
+  NULL,             // ADST_ADST
+  NULL,             // FLIPADST_DCT
+  NULL,             // DCT_FLIPADST
+  NULL,             // FLIPADST_FLIPADST
+  NULL,             // ADST_FLIPADST
+  NULL,             // FLIPADST-ADST
+  idtx32x8_sse4_1,  // IDTX
   NULL,             // V_DCT
   NULL,             // H_DCT
   NULL,             // V_ADST
   NULL,             // H_ADST
   NULL,             // V_FLIPADST
-  NULL              // H_FLIPADST
+  NULL,             // H_FLIPADST
+};
+static const fwd_transform_1d_sse4_1 col_highbd_txfm4x8_arr[TX_TYPES] = {
+  fdct4x8_sse4_1,   // DCT_DCT
+  fadst8x8_sse4_1,  // ADST_DCT
+  fdct4x8_sse4_1,   // DCT_ADST
+  fadst8x8_sse4_1,  // ADST_ADST
+  fadst8x8_sse4_1,  // FLIPADST_DCT
+  fdct4x8_sse4_1,   // DCT_FLIPADST
+  fadst8x8_sse4_1,  // FLIPADST_FLIPADST
+  fadst8x8_sse4_1,  // ADST_FLIPADST
+  fadst8x8_sse4_1,  // FLIPADST_ADST
+  idtx8x8_sse4_1,   // IDTX
+  fdct4x8_sse4_1,   // V_DCT
+  idtx8x8_sse4_1,   // H_DCT
+  fadst8x8_sse4_1,  // V_ADST
+  idtx8x8_sse4_1,   // H_ADST
+  fadst8x8_sse4_1,  // V_FLIPADST
+  idtx8x8_sse4_1    // H_FLIPADST
 };
 
 static const fwd_transform_1d_sse4_1 row_highbd_txfm8x16_arr[TX_TYPES] = {
@@ -1673,13 +1974,13 @@ static const fwd_transform_1d_sse4_1 row_highbd_txfm8x16_arr[TX_TYPES] = {
   fadst16x16_sse4_1,  // FLIPADST_FLIPADST
   fadst16x16_sse4_1,  // ADST_FLIPADST
   fadst16x16_sse4_1,  // FLIPADST_ADST
-  NULL,               // IDTX
-  NULL,               // V_DCT
-  NULL,               // H_DCT
-  NULL,               // V_ADST
-  NULL,               // H_ADST
-  NULL,               // V_FLIPADST
-  NULL                // H_FLIPADST
+  idtx16x16_sse4_1,   // IDTX
+  idtx16x16_sse4_1,   // V_DCT
+  fdct16x16_sse4_1,   // H_DCT
+  idtx16x16_sse4_1,   // V_ADST
+  fadst16x16_sse4_1,  // H_ADST
+  idtx16x16_sse4_1,   // V_FLIPADST
+  fadst16x16_sse4_1   // H_FLIPADST
 };
 
 static const fwd_transform_1d_sse4_1 col_highbd_txfm8x16_arr[TX_TYPES] = {
@@ -1692,13 +1993,13 @@ static const fwd_transform_1d_sse4_1 col_highbd_txfm8x16_arr[TX_TYPES] = {
   fadst16x16_sse4_1,  // FLIPADST_FLIPADST
   fadst16x16_sse4_1,  // ADST_FLIPADST
   fadst16x16_sse4_1,  // FLIPADST_ADST
-  NULL,               // IDTX
-  NULL,               // V_DCT
-  NULL,               // H_DCT
-  NULL,               // V_ADST
-  NULL,               // H_ADST
-  NULL,               // V_FLIPADST
-  NULL                // H_FLIPADST
+  idtx16x16_sse4_1,   // IDTX
+  fdct16x16_sse4_1,   // V_DCT
+  idtx16x16_sse4_1,   // H_DCT
+  fadst16x16_sse4_1,  // V_ADST
+  idtx16x16_sse4_1,   // H_ADST
+  fadst16x16_sse4_1,  // V_FLIPADST
+  idtx16x16_sse4_1    // H_FLIPADST
 };
 static const fwd_transform_1d_sse4_1 row_highbd_txfm8x8_arr[TX_TYPES] = {
   fdct8x8_sse4_1,   // DCT_DCT
@@ -1710,13 +2011,108 @@ static const fwd_transform_1d_sse4_1 row_highbd_txfm8x8_arr[TX_TYPES] = {
   fadst8x8_sse4_1,  // FLIPADST_FLIPADST
   fadst8x8_sse4_1,  // ADST_FLIPADST
   fadst8x8_sse4_1,  // FLIPADST_ADST
-  NULL,             // IDTX
-  NULL,             // V_DCT
-  NULL,             // H_DCT
-  NULL,             // V_ADST
-  NULL,             // H_ADST
-  NULL,             // V_FLIPADST
-  NULL              // H_FLIPADST
+  idtx8x8_sse4_1,   // IDTX
+  idtx8x8_sse4_1,   // V_DCT
+  fdct8x8_sse4_1,   // H_DCT
+  idtx8x8_sse4_1,   // V_ADST
+  fadst8x8_sse4_1,  // H_ADST
+  idtx8x8_sse4_1,   // V_FLIPADST
+  fadst8x8_sse4_1   // H_FLIPADST
+};
+
+static const fwd_transform_1d_sse4_1 row_highbd_txfm4x8_arr[TX_TYPES] = {
+  fdct4x8_sse4_1,   // DCT_DCT
+  fdct4x8_sse4_1,   // ADST_DCT
+  fadst8x8_sse4_1,  // DCT_ADST
+  fadst8x8_sse4_1,  // ADST_ADST
+  fdct4x8_sse4_1,   // FLIPADST_DCT
+  fadst8x8_sse4_1,  // DCT_FLIPADST
+  fadst8x8_sse4_1,  // FLIPADST_FLIPADST
+  fadst8x8_sse4_1,  // ADST_FLIPADST
+  fadst8x8_sse4_1,  // FLIPADST_ADST
+  idtx8x8_sse4_1,   // IDTX
+  idtx8x8_sse4_1,   // V_DCT
+  fdct4x8_sse4_1,   // H_DCT
+  idtx8x8_sse4_1,   // V_ADST
+  fadst8x8_sse4_1,  // H_ADST
+  idtx8x8_sse4_1,   // V_FLIPADST
+  fadst8x8_sse4_1   // H_FLIPADST
+};
+
+static const fwd_transform_1d_sse4_1 row_highbd_txfm4x4_arr[TX_TYPES] = {
+  fdct4x4_sse4_1,   // DCT_DCT
+  fdct4x4_sse4_1,   // ADST_DCT
+  fadst4x4_sse4_1,  // DCT_ADST
+  fadst4x4_sse4_1,  // ADST_ADST
+  fdct4x4_sse4_1,   // FLIPADST_DCT
+  fadst4x4_sse4_1,  // DCT_FLIPADST
+  fadst4x4_sse4_1,  // FLIPADST_FLIPADST
+  fadst4x4_sse4_1,  // ADST_FLIPADST
+  fadst4x4_sse4_1,  // FLIPADST_ADST
+  idtx4x4_sse4_1,   // IDTX
+  idtx4x4_sse4_1,   // V_DCT
+  fdct4x4_sse4_1,   // H_DCT
+  idtx4x4_sse4_1,   // V_ADST
+  fadst4x4_sse4_1,  // H_ADST
+  idtx4x4_sse4_1,   // V_FLIPADST
+  fadst4x4_sse4_1   // H_FLIPADST
+};
+
+static const fwd_transform_1d_sse4_1 col_highbd_txfm4x4_arr[TX_TYPES] = {
+  fdct4x4_sse4_1,   // DCT_DCT
+  fadst4x4_sse4_1,  // ADST_DCT
+  fdct4x4_sse4_1,   // DCT_ADST
+  fadst4x4_sse4_1,  // ADST_ADST
+  fadst4x4_sse4_1,  // FLIPADST_DCT
+  fdct4x4_sse4_1,   // DCT_FLIPADST
+  fadst4x4_sse4_1,  // FLIPADST_FLIPADST
+  fadst4x4_sse4_1,  // ADST_FLIPADST
+  fadst4x4_sse4_1,  // FLIPADST_ADST
+  idtx4x4_sse4_1,   // IDTX
+  fdct4x4_sse4_1,   // V_DCT
+  idtx4x4_sse4_1,   // H_DCT
+  fadst4x4_sse4_1,  // V_ADST
+  idtx4x4_sse4_1,   // H_ADST
+  fadst4x4_sse4_1,  // V_FLIPADST
+  idtx4x4_sse4_1    // H_FLIPADST
+};
+
+static const fwd_transform_1d_sse4_1 col_highbd_txfm8x32_arr[TX_TYPES] = {
+  av1_fdct32_new_sse4_1,  // DCT_DCT
+  NULL,                   // ADST_DCT
+  NULL,                   // DCT_ADST
+  NULL,                   // ADST_ADST
+  NULL,                   // FLIPADST_DCT
+  NULL,                   // DCT_FLIPADST
+  NULL,                   // FLIPADST_FLIPADST
+  NULL,                   // ADST_FLIPADST
+  NULL,                   // FLIPADST_ADST
+  av1_idtx32_new_sse4_1,  // IDTX
+  NULL,                   // V_DCT
+  NULL,                   // H_DCT
+  NULL,                   // V_ADST
+  NULL,                   // H_ADST
+  NULL,                   // V_FLIPADST
+  NULL                    // H_FLIPADST
+};
+
+static const fwd_transform_1d_sse4_1 row_highbd_txfm8x32_arr[TX_TYPES] = {
+  fdct16x16_sse4_1,  // DCT_DCT
+  NULL,              // ADST_DCT
+  NULL,              // DCT_ADST
+  NULL,              // ADST_ADST
+  NULL,              // FLIPADST_DCT
+  NULL,              // DCT_FLIPADST
+  NULL,              // FLIPADST_FLIPADST
+  NULL,              // ADST_FLIPADST
+  NULL,              // FLIPADST_ADST
+  idtx16x16_sse4_1,  // IDTX
+  NULL,              // V_DCT
+  NULL,              // H_DCT
+  NULL,              // V_ADST
+  NULL,              // H_ADST
+  NULL,              // V_FLIPADST
+  NULL               // H_FLIPADST
 };
 
 void av1_fwd_txfm2d_16x8_sse4_1(const int16_t *input, int32_t *coeff,
@@ -1733,7 +2129,7 @@ void av1_fwd_txfm2d_16x8_sse4_1(const int16_t *input, int32_t *coeff,
 
   for (int i = 0; i < 2; i++) {
     load_buffer_8x8(input + i * 8, in, stride, ud_flip, 0, shift[0]);
-    col_txfm(in, in, bit, 0);
+    col_txfm(in, in, bit, 2);
     col_txfm_8x8_rounding(in, -shift[1]);
     transpose_8x8(in, out + i * 16);
   }
@@ -1773,7 +2169,7 @@ void av1_fwd_txfm2d_8x16_sse4_1(const int16_t *input, int32_t *coeff,
   transpose_8x8(in + 16, out + 16);
 
   for (int i = 0; i < 2; i++) {
-    row_txfm(out + i * 16, out, bit, 0);
+    row_txfm(out + i * 16, out, bit, 2);
     transpose_8x8(out, in);
     av1_round_shift_rect_array_32_sse4_1(in, in, 16, -shift[2], NewSqrt2);
     write_buffer_8x8(in, coeff + i * 64);
@@ -1781,3 +2177,412 @@ void av1_fwd_txfm2d_8x16_sse4_1(const int16_t *input, int32_t *coeff,
 
   (void)bd;
 }
+
+void av1_fwd_txfm2d_4x16_sse4_1(const int16_t *input, int32_t *coeff,
+                                int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[16];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_4X16];
+  const int txw_idx = get_txw_idx(TX_4X16);
+  const int txh_idx = get_txh_idx(TX_4X16);
+  const int txfm_size_col = tx_size_wide[TX_4X16];
+  const int txfm_size_row = tx_size_high[TX_4X16];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm8x16_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm4x4_arr[tx_type];
+
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+  // col transform
+  load_buffer_4x16(input, in, stride, ud_flip, lr_flip, shift[0]);
+  col_txfm(in, outcoeff128, bitcol, 1);
+  col_txfm_8x8_rounding(outcoeff128, -shift[1]);
+  transpose_8nx8n(outcoeff128, in, txfm_size_col, txfm_size_row);
+
+  // row transform
+  for (int i = 0; i < txfm_size_col; i++) {
+    row_txfm(in + i, outcoeff128 + i * txfm_size_col, bitrow, txfm_size_col);
+  }
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_16x4_sse4_1(const int16_t *input, int32_t *coeff,
+                                int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[16];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_16X4];
+  const int txw_idx = get_txw_idx(TX_16X4);
+  const int txh_idx = get_txh_idx(TX_16X4);
+  const int txfm_size_col = tx_size_wide[TX_16X4];
+  const int txfm_size_row = tx_size_high[TX_16X4];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm4x4_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm8x16_arr[tx_type];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  // col transform
+  load_buffer_16x4(input, in, stride, ud_flip, lr_flip, shift[0]);
+
+  for (int i = 0; i < txfm_size_row; i++) {
+    col_txfm(in + i * txfm_size_row, outcoeff128 + i * txfm_size_row, bitcol,
+             1);
+  }
+  col_txfm_8x8_rounding(outcoeff128, -shift[1]);
+
+  // row transform
+  row_txfm(outcoeff128, in, bitrow, 1);
+  transpose_8nx8n(in, outcoeff128, txfm_size_row, txfm_size_col);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_16x32_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[128];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_16X32];
+  const int txw_idx = get_txw_idx(TX_16X32);
+  const int txh_idx = get_txh_idx(TX_16X32);
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm8x32_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm8x32_arr[tx_type];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+
+  // column transform
+  load_buffer_16x16(input, in, stride, 0, 0, shift[0]);
+  load_buffer_16x16(input + 16 * stride, in + 64, stride, 0, 0, shift[0]);
+
+  for (int i = 0; i < 4; i++) {
+    col_txfm((in + i), (in + i), bitcol, 4);
+  }
+  col_txfm_16x16_rounding(&in[0], -shift[1]);
+  col_txfm_16x16_rounding(&in[64], -shift[1]);
+  transpose_8nx8n(in, outcoef128, 16, 32);
+
+  // row transform
+  row_txfm(outcoef128, in, bitrow, 8);
+  transpose_8nx8n(in, outcoef128, 32, 16);
+  av1_round_shift_rect_array_32_sse4_1(outcoef128, outcoef128, 128, -shift[2],
+                                       NewSqrt2);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_32x64_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  (void)tx_type;
+  __m128i in[512];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_32X64];
+  const int txw_idx = get_txw_idx(TX_32X64);
+  const int txh_idx = get_txh_idx(TX_32X64);
+  const int txfm_size_col = tx_size_wide[TX_32X64];
+  const int txfm_size_row = tx_size_high[TX_32X64];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const int num_row = txfm_size_row >> 2;
+  const int num_col = txfm_size_col >> 2;
+
+  // column transform
+  load_buffer_32x8n(input, in, stride, 0, 0, shift[0], txfm_size_row);
+  for (int i = 0; i < num_col; i++) {
+    av1_fdct64_new_sse4_1((in + i), (in + i), bitcol, num_col, num_col);
+  }
+  for (int i = 0; i < num_col; i++) {
+    col_txfm_16x16_rounding((in + i * txfm_size_row), -shift[1]);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_col, txfm_size_row);
+
+  // row transform
+  for (int i = 0; i < num_row; i++) {
+    av1_fdct32_new_sse4_1((outcoef128 + i), (in + i), bitrow, num_row);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_row, txfm_size_col);
+  av1_round_shift_rect_array_32_sse4_1(outcoef128, outcoef128, 512, -shift[2],
+                                       NewSqrt2);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_64x32_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  (void)tx_type;
+  __m128i in[512];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_64X32];
+  const int txw_idx = get_txw_idx(TX_64X32);
+  const int txh_idx = get_txh_idx(TX_64X32);
+  const int txfm_size_col = tx_size_wide[TX_64X32];
+  const int txfm_size_row = tx_size_high[TX_64X32];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const int num_row = txfm_size_row >> 2;
+  const int num_col = txfm_size_col >> 2;
+
+  // column transform
+  for (int i = 0; i < 32; i++) {
+    load_buffer_4x4(input + 0 + i * stride, in + 0 + i * 16, 4, 0, 0, shift[0]);
+    load_buffer_4x4(input + 16 + i * stride, in + 4 + i * 16, 4, 0, 0,
+                    shift[0]);
+    load_buffer_4x4(input + 32 + i * stride, in + 8 + i * 16, 4, 0, 0,
+                    shift[0]);
+    load_buffer_4x4(input + 48 + i * stride, in + 12 + i * 16, 4, 0, 0,
+                    shift[0]);
+  }
+
+  for (int i = 0; i < num_col; i++) {
+    av1_fdct32_new_sse4_1((in + i), (in + i), bitcol, num_col);
+  }
+
+  for (int i = 0; i < num_row; i++) {
+    col_txfm_16x16_rounding((in + i * txfm_size_col), -shift[1]);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_col, txfm_size_row);
+
+  // row transform
+  for (int i = 0; i < num_row; i++) {
+    av1_fdct64_new_sse4_1((outcoef128 + i), (in + i), bitrow, num_row, num_row);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_row, txfm_size_col >> 1);
+  av1_round_shift_rect_array_32_sse4_1(outcoef128, outcoef128, 512 >> 1,
+                                       -shift[2], NewSqrt2);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_32x16_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[128];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_32X16];
+  const int txw_idx = get_txw_idx(TX_32X16);
+  const int txh_idx = get_txh_idx(TX_32X16);
+  const fwd_transform_1d_sse4_1 col_txfm = row_highbd_txfm8x32_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = col_highbd_txfm8x32_arr[tx_type];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+
+  // column transform
+  load_buffer_32x8n(input, in, stride, 0, 0, shift[0], 16);
+  col_txfm(in, in, bitcol, 8);
+  col_txfm_16x16_rounding(&in[0], -shift[1]);
+  col_txfm_16x16_rounding(&in[64], -shift[1]);
+  transpose_8nx8n(in, outcoef128, 32, 16);
+
+  // row transform
+  for (int i = 0; i < 4; i++) {
+    row_txfm((outcoef128 + i), (in + i), bitrow, 4);
+  }
+  transpose_8nx8n(in, outcoef128, 16, 32);
+  av1_round_shift_rect_array_32_sse4_1(outcoef128, outcoef128, 128, -shift[2],
+                                       NewSqrt2);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_8x32_sse4_1(const int16_t *input, int32_t *coeff,
+                                int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[64];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_8X32];
+  const int txw_idx = get_txw_idx(TX_8X32);
+  const int txh_idx = get_txh_idx(TX_8X32);
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm8x32_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm32x8_arr[tx_type];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+
+  const int txfm_size_col = tx_size_wide[TX_8X32];
+  const int txfm_size_row = tx_size_high[TX_8X32];
+  const int num_col = txfm_size_col >> 2;
+
+  // column transform
+  load_buffer_8x16(input, in, stride, 0, 0, shift[0]);
+  load_buffer_8x16(input + (txfm_size_row >> 1) * stride, in + txfm_size_row,
+                   stride, 0, 0, shift[0]);
+
+  for (int i = 0; i < num_col; i++) {
+    col_txfm((in + i), (in + i), bitcol, num_col);
+  }
+  col_txfm_16x16_rounding(in, -shift[1]);
+  transpose_8nx8n(in, outcoef128, txfm_size_col, txfm_size_row);
+
+  // row transform
+  for (int i = 0; i < txfm_size_col; i += 2) {
+    row_txfm((outcoef128 + i), (in + i), bitrow, txfm_size_col);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_row, txfm_size_col);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_32x8_sse4_1(const int16_t *input, int32_t *coeff,
+                                int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[64];
+  __m128i *outcoef128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_32X8];
+  const int txw_idx = get_txw_idx(TX_32X8);
+  const int txh_idx = get_txh_idx(TX_32X8);
+  const fwd_transform_1d_sse4_1 col_txfm = row_highbd_txfm32x8_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = col_highbd_txfm8x32_arr[tx_type];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+
+  const int txfm_size_col = tx_size_wide[TX_32X8];
+  const int txfm_size_row = tx_size_high[TX_32X8];
+  const int num_col = txfm_size_row >> 2;
+
+  // column transform
+  load_buffer_32x8n(input, in, stride, 0, 0, shift[0], 8);
+  for (int i = 0; i < txfm_size_row; i += 2) {
+    col_txfm((in + i), (in + i), bitcol, txfm_size_row);
+  }
+
+  col_txfm_16x16_rounding(&in[0], -shift[1]);
+  transpose_8nx8n(in, outcoef128, txfm_size_col, txfm_size_row);
+
+  // row transform
+  for (int i = 0; i < num_col; i++) {
+    row_txfm((outcoef128 + i), (in + i), bitrow, num_col);
+  }
+  transpose_8nx8n(in, outcoef128, txfm_size_row, txfm_size_col);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_4x8_sse4_1(const int16_t *input, int32_t *coeff, int stride,
+                               TX_TYPE tx_type, int bd) {
+  __m128i in[8];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_4X8];
+  const int txw_idx = get_txw_idx(TX_4X8);
+  const int txh_idx = get_txh_idx(TX_4X8);
+  const int txfm_size_col = tx_size_wide[TX_4X8];
+  const int txfm_size_row = tx_size_high[TX_4X8];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm4x8_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm4x4_arr[tx_type];
+
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+
+  load_buffer_4x8(input, in, stride, ud_flip, lr_flip, shift[0]);
+  col_txfm(in, in, bitcol, 1);
+  col_txfm_4x8_rounding(in, -shift[1]);
+  transpose_8nx8n(in, outcoeff128, txfm_size_col, txfm_size_row);
+
+  for (int i = 0; i < 2; i++) {
+    row_txfm(outcoeff128 + i, in + i * txfm_size_col, bitrow, 2);
+  }
+  av1_round_shift_rect_array_32_sse4_1(in, outcoeff128, txfm_size_row,
+                                       -shift[2], NewSqrt2);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_8x4_sse4_1(const int16_t *input, int32_t *coeff, int stride,
+                               TX_TYPE tx_type, int bd) {
+  __m128i in[8];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_8X4];
+  const int txw_idx = get_txw_idx(TX_8X4);
+  const int txh_idx = get_txh_idx(TX_8X4);
+  const int txfm_size_col = tx_size_wide[TX_8X4];
+  const int txfm_size_row = tx_size_high[TX_8X4];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  const fwd_transform_1d_sse4_1 col_txfm = col_highbd_txfm4x4_arr[tx_type];
+  const fwd_transform_1d_sse4_1 row_txfm = row_highbd_txfm4x8_arr[tx_type];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+  // col tranform
+  load_buffer_8x4(input, in, stride, ud_flip, lr_flip, shift[0]);
+  for (int i = 0; i < 2; i++) {
+    col_txfm(in + i * txfm_size_row, in + i * txfm_size_row, bitcol, 1);
+  }
+  col_txfm_4x8_rounding(in, -shift[1]);
+
+  // row tranform
+  row_txfm(in, outcoeff128, bitrow, 1);
+  av1_round_shift_rect_array_32_sse4_1(outcoeff128, in, txfm_size_col,
+                                       -shift[2], NewSqrt2);
+  transpose_8nx8n(in, outcoeff128, txfm_size_row, txfm_size_col);
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_16x64_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[256];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_16X64];
+  const int txw_idx = get_txw_idx(TX_16X64);
+  const int txh_idx = get_txh_idx(TX_16X64);
+  const int txfm_size_col = tx_size_wide[TX_16X64];
+  const int txfm_size_row = tx_size_high[TX_16X64];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+  const int num_col = txfm_size_col >> 2;
+  // col tranform
+  for (int i = 0; i < txfm_size_row; i += num_col) {
+    load_buffer_4x4(input + (i + 0) * stride, in + (i + 0) * num_col, num_col,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + (i + 1) * stride, in + (i + 1) * num_col, num_col,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + (i + 2) * stride, in + (i + 2) * num_col, num_col,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + (i + 3) * stride, in + (i + 3) * num_col, num_col,
+                    ud_flip, lr_flip, shift[0]);
+  }
+
+  for (int i = 0; i < num_col; i++) {
+    av1_fdct64_new_sse4_1(in + i, outcoeff128 + i, bitcol, num_col, num_col);
+  }
+
+  col_txfm_16x16_rounding(outcoeff128, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 64, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 128, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 192, -shift[1]);
+
+  transpose_8nx8n(outcoeff128, in, txfm_size_col, 32);
+  fdct16x16_sse4_1(in, in, bitrow, 8);
+  transpose_8nx8n(in, outcoeff128, 32, txfm_size_col);
+  memset(coeff + txfm_size_col * 32, 0, txfm_size_col * 32 * sizeof(*coeff));
+  (void)bd;
+}
+
+void av1_fwd_txfm2d_64x16_sse4_1(const int16_t *input, int32_t *coeff,
+                                 int stride, TX_TYPE tx_type, int bd) {
+  __m128i in[256];
+  __m128i *outcoeff128 = (__m128i *)coeff;
+  const int8_t *shift = fwd_txfm_shift_ls[TX_64X16];
+  const int txw_idx = get_txw_idx(TX_64X16);
+  const int txh_idx = get_txh_idx(TX_64X16);
+  const int txfm_size_col = tx_size_wide[TX_64X16];
+  const int txfm_size_row = tx_size_high[TX_64X16];
+  int bitcol = fwd_cos_bit_col[txw_idx][txh_idx];
+  int bitrow = fwd_cos_bit_row[txw_idx][txh_idx];
+  int ud_flip, lr_flip;
+  get_flip_cfg(tx_type, &ud_flip, &lr_flip);
+  // col tranform
+  for (int i = 0; i < txfm_size_row; i++) {
+    load_buffer_4x4(input + 0 + i * stride, in + 0 + i * txfm_size_row, 4,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + 16 + i * stride, in + 4 + i * txfm_size_row, 4,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + 32 + i * stride, in + 8 + i * txfm_size_row, 4,
+                    ud_flip, lr_flip, shift[0]);
+    load_buffer_4x4(input + 48 + i * stride, in + 12 + i * txfm_size_row, 4,
+                    ud_flip, lr_flip, shift[0]);
+  }
+
+  fdct16x16_sse4_1(in, outcoeff128, bitcol, txfm_size_row);
+  col_txfm_16x16_rounding(outcoeff128, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 64, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 128, -shift[1]);
+  col_txfm_16x16_rounding(outcoeff128 + 192, -shift[1]);
+
+  transpose_8nx8n(outcoeff128, in, txfm_size_col, txfm_size_row);
+  for (int i = 0; i < 4; i++) {
+    av1_fdct64_new_sse4_1(in + i, in + i, bitrow, 4, 4);
+  }
+  transpose_8nx8n(in, outcoeff128, txfm_size_row, 32);
+  (void)bd;
+}
diff --git a/third_party/aom/av1/encoder/x86/ml_sse3.c b/third_party/aom/av1/encoder/x86/ml_sse3.c
new file mode 100644
index 0000000000..c520c3c356
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/ml_sse3.c
@@ -0,0 +1,243 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdbool.h>
+#include <assert.h>
+#include <pmmintrin.h>
+
+#include "config/av1_rtcd.h"
+#include "av1/encoder/ml.h"
+
+// In order to avoid the high-latency of swapping between FPU and SIMD
+// operations, we keep the result in a 128-bit register even though we only
+// care about a single value.
+static void nn_propagate_8to1(const float *const inputs,
+                              const float *const weights,
+                              __m128 *const output) {
+  const __m128 inputs_h = _mm_loadu_ps(&inputs[4]);
+  const __m128 inputs_l = _mm_loadu_ps(inputs);
+
+  const __m128 weights_h = _mm_loadu_ps(&weights[4]);
+  const __m128 weights_l = _mm_loadu_ps(weights);
+
+  const __m128 mul_h = _mm_mul_ps(inputs_h, weights_h);
+  const __m128 mul_l = _mm_mul_ps(inputs_l, weights_l);
+  // [7 6 5 4] [3 2 1 0] (weight and input indices)
+
+  const __m128 vadd = _mm_add_ps(mul_l, mul_h);
+  // [7+3 6+2 5+1 4+0]
+  const __m128 hadd1 = _mm_hadd_ps(vadd, vadd);
+  // [7+6+3+2 5+4+1+0 7+6+3+2 5+4+1+0]
+  const __m128 hadd2 = _mm_hadd_ps(hadd1, hadd1);
+  // [7+6+5+4+3+2+1+0 7+6+5+4+3+2+1+0 7+6+5+4+3+2+1+0 7+6+5+4+3+2+1+0]
+  *output = _mm_add_ps(*output, hadd2);
+}
+
+static void nn_propagate_4to1(const float *const inputs,
+                              const float *const weights,
+                              __m128 *const output) {
+  const __m128 inputs128 = _mm_loadu_ps(inputs);
+
+  const __m128 weights128 = _mm_loadu_ps(weights);
+
+  const __m128 mul = _mm_mul_ps(inputs128, weights128);
+  // [3 2 1 0] (weight and input indices)
+
+  const __m128 hadd1 = _mm_hadd_ps(mul, mul);
+  // [3+2 1+0 3+2 1+0]
+  const __m128 hadd2 = _mm_hadd_ps(hadd1, hadd1);
+  // [3+2+1+0 3+2+1+0 3+2+1+0 3+2+1+0]
+  *output = _mm_add_ps(*output, hadd2);
+}
+
+static void nn_propagate_4to4(const float *const inputs,
+                              const float *const weights, __m128 *const outputs,
+                              const int num_inputs) {
+  const __m128 inputs128 = _mm_loadu_ps(inputs);
+
+  __m128 hadd[2];
+  for (int i = 0; i < 2; i++) {  // For each pair of outputs
+    const __m128 weight0 = _mm_loadu_ps(&weights[2 * i * num_inputs]);
+    const __m128 mul0 = _mm_mul_ps(weight0, inputs128);
+    const __m128 weight1 = _mm_loadu_ps(&weights[(2 * i + 1) * num_inputs]);
+    const __m128 mul1 = _mm_mul_ps(weight1, inputs128);
+    hadd[i] = _mm_hadd_ps(mul0, mul1);
+  }
+  // hadd[0] = [7+6 5+4 3+2 1+0] (weight indices)
+  // hadd[1] = [15+14 13+12 11+10 9+8]
+
+  const __m128 hh = _mm_hadd_ps(hadd[0], hadd[1]);
+  // [15+14+13+12 11+10+9+8 7+6+5+4 3+2+1+0]
+
+  *outputs = _mm_add_ps(*outputs, hh);
+}
+
+static void nn_propagate_4to8(const float *const inputs,
+                              const float *const weights, __m128 *const out_h,
+                              __m128 *const out_l, const int num_inputs) {
+  const __m128 inputs128 = _mm_loadu_ps(inputs);
+
+  __m128 hadd[4];
+  for (int i = 0; i < 4; i++) {  // For each pair of outputs
+    const __m128 weight0 = _mm_loadu_ps(&weights[2 * i * num_inputs]);
+    const __m128 weight1 = _mm_loadu_ps(&weights[(2 * i + 1) * num_inputs]);
+    const __m128 mul0 = _mm_mul_ps(inputs128, weight0);
+    const __m128 mul1 = _mm_mul_ps(inputs128, weight1);
+    hadd[i] = _mm_hadd_ps(mul0, mul1);
+  }
+  // hadd[0] = [7+6 5+4 3+2 1+0] (weight indices)
+  // hadd[1] = [15+14 13+12 11+10 9+8]
+  // hadd[2] = [23+22 21+20 19+18 17+16]
+  // hadd[3] = [31+30 29+28 27+26 25+24]
+
+  const __m128 hh0 = _mm_hadd_ps(hadd[0], hadd[1]);
+  // [15+14+13+12 11+10+9+8 7+6+5+4 3+2+1+0]
+  const __m128 hh1 = _mm_hadd_ps(hadd[2], hadd[3]);
+  // [31+30+29+28 27+26+25+24 23+22+21+20 19+18+17+16]
+
+  *out_h = _mm_add_ps(*out_h, hh1);
+  *out_l = _mm_add_ps(*out_l, hh0);
+}
+
+static void nn_propagate_8to4(const float *const inputs,
+                              const float *const weights, __m128 *const outputs,
+                              const int num_inputs) {
+  const __m128 inputs_h = _mm_loadu_ps(inputs + 4);
+  const __m128 inputs_l = _mm_loadu_ps(inputs);
+  // [7 6 5 4] [3 2 1 0] (input indices)
+
+  __m128 add[4];
+  for (int i = 0; i < 4; i++) {  // For each output:
+    const __m128 weight_h = _mm_loadu_ps(&weights[i * num_inputs + 4]);
+    const __m128 weight_l = _mm_loadu_ps(&weights[i * num_inputs]);
+    const __m128 mul_h = _mm_mul_ps(inputs_h, weight_h);
+    const __m128 mul_l = _mm_mul_ps(inputs_l, weight_l);
+    add[i] = _mm_add_ps(mul_l, mul_h);
+  }
+  // add[0] = [7+3 6+2 5+1 4+0]
+  // add[1] = [15+11 14+10 13+9 12+8]
+  // add[2] = [23+19 22+18 21+17 20+16]
+  // add[3] = [31+27 30+26 29+25 28+24]
+
+  const __m128 hadd_h = _mm_hadd_ps(add[2], add[3]);
+  // [31+30+27+26 29+28+25+24 23+22+19+18 21+20+17+16]
+  const __m128 hadd_l = _mm_hadd_ps(add[0], add[1]);
+  // [15+14+11+10 13+12+9+8 7+6+3+2 5+4+1+0]
+
+  const __m128 haddhadd = _mm_hadd_ps(hadd_l, hadd_h);
+  // [31+30+29+28+27+26+25+24 23+22+21+20+19+18+17+16
+  //  15+14+13+12+11+10+9+8 7+6+5+4+3+2+1+0]
+
+  *outputs = _mm_add_ps(*outputs, haddhadd);
+}
+
+static void nn_activate8(__m128 *out_h, __m128 *out_l) {
+  const __m128 zero = _mm_setzero_ps();
+  *out_h = _mm_max_ps(*out_h, zero);
+  *out_l = _mm_max_ps(*out_l, zero);
+}
+
+static void nn_activate4(__m128 *x) { *x = _mm_max_ps(*x, _mm_setzero_ps()); }
+
+// Calculate prediction based on the given input features and neural net config.
+// Assume there are no more than NN_MAX_NODES_PER_LAYER nodes in each hidden
+// layer.
+void av1_nn_predict_sse3(const float *input_nodes,
+                         const NN_CONFIG *const nn_config,
+                         float *const output) {
+  float buf[2][NN_MAX_NODES_PER_LAYER];
+  int buf_index = 0;
+  int num_inputs = nn_config->num_inputs;
+
+  // Hidden layers, except the final iteration is the output layer.
+  for (int layer = 0; layer <= nn_config->num_hidden_layers; layer++) {
+    const float *layer_weights = nn_config->weights[layer];
+    const float *layer_bias = nn_config->bias[layer];
+    bool output_layer = (layer == nn_config->num_hidden_layers);
+    float *const output_nodes = output_layer ? output : buf[buf_index];
+    const int num_outputs = output_layer ? nn_config->num_outputs
+                                         : nn_config->num_hidden_nodes[layer];
+
+    if (num_inputs % 4 == 0 && num_outputs % 8 == 0) {
+      for (int out = 0; out < num_outputs; out += 8) {
+        __m128 out_h = _mm_loadu_ps(&layer_bias[out + 4]);
+        __m128 out_l = _mm_loadu_ps(&layer_bias[out]);
+        for (int in = 0; in < num_inputs; in += 4) {
+          nn_propagate_4to8(&input_nodes[in],
+                            &layer_weights[out * num_inputs + in], &out_h,
+                            &out_l, num_inputs);
+        }
+        if (!output_layer) nn_activate8(&out_h, &out_l);
+        _mm_storeu_ps(&output_nodes[out + 4], out_h);
+        _mm_storeu_ps(&output_nodes[out], out_l);
+      }
+    } else if (num_inputs % 8 == 0 && num_outputs % 4 == 0) {
+      for (int out = 0; out < num_outputs; out += 4) {
+        __m128 outputs = _mm_loadu_ps(&layer_bias[out]);
+        for (int in = 0; in < num_inputs; in += 8) {
+          nn_propagate_8to4(&input_nodes[in],
+                            &layer_weights[out * num_inputs + in], &outputs,
+                            num_inputs);
+        }
+        if (!output_layer) nn_activate4(&outputs);
+        _mm_storeu_ps(&output_nodes[out], outputs);
+      }
+    } else if (num_inputs % 4 == 0 && num_outputs % 4 == 0) {
+      for (int out = 0; out < num_outputs; out += 4) {
+        __m128 outputs = _mm_loadu_ps(&layer_bias[out]);
+        for (int in = 0; in < num_inputs; in += 4) {
+          nn_propagate_4to4(&input_nodes[in],
+                            &layer_weights[out * num_inputs + in], &outputs,
+                            num_inputs);
+        }
+        if (!output_layer) nn_activate4(&outputs);
+        _mm_storeu_ps(&output_nodes[out], outputs);
+      }
+    } else if (num_inputs % 8 == 0) {
+      for (int out = 0; out < num_outputs; out++) {
+        __m128 total = _mm_load1_ps(&layer_bias[out]);
+        for (int in = 0; in < num_inputs; in += 8) {
+          nn_propagate_8to1(&input_nodes[in],
+                            &layer_weights[out * num_inputs + in], &total);
+        }
+        if (!output_layer) nn_activate4(&total);
+        output_nodes[out] = _mm_cvtss_f32(total);
+      }
+    } else if (num_inputs % 4 == 0) {
+      for (int out = 0; out < num_outputs; out++) {
+        __m128 total = _mm_load1_ps(&layer_bias[out]);
+        for (int in = 0; in < num_inputs; in += 4) {
+          nn_propagate_4to1(&input_nodes[in],
+                            &layer_weights[out * num_inputs + in], &total);
+        }
+        if (!output_layer) nn_activate4(&total);
+        output_nodes[out] = _mm_cvtss_f32(total);
+      }
+    } else {
+      // Use SSE instructions for scalar operations to avoid the latency of
+      // swapping between SIMD and FPU modes.
+      for (int out = 0; out < num_outputs; out++) {
+        __m128 total = _mm_load1_ps(&layer_bias[out]);
+        for (int in_node = 0; in_node < num_inputs; in_node++) {
+          __m128 input = _mm_load1_ps(&input_nodes[in_node]);
+          __m128 weight =
+              _mm_load1_ps(&layer_weights[num_inputs * out + in_node]);
+          total = _mm_add_ps(total, _mm_mul_ps(input, weight));
+        }
+        if (!output_layer) nn_activate4(&total);
+        output_nodes[out] = _mm_cvtss_f32(total);
+      }
+    }
+    input_nodes = output_nodes;
+    num_inputs = num_outputs;
+    buf_index = 1 - buf_index;
+  }
+}
diff --git a/third_party/aom/av1/encoder/x86/pickrst_avx2.c b/third_party/aom/av1/encoder/x86/pickrst_avx2.c
index 06aaaa7eee..d00fca0594 100644
--- a/third_party/aom/av1/encoder/x86/pickrst_avx2.c
+++ b/third_party/aom/av1/encoder/x86/pickrst_avx2.c
@@ -64,14 +64,13 @@ static INLINE void acc_stat_win7_one_line_avx2(
 
 static INLINE void compute_stats_win7_opt_avx2(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start,
-    int v_end, int dgd_stride, int src_stride, double *M, double *H) {
+    int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H) {
   int i, j, k, l, m, n;
   const int wiener_win = WIENER_WIN;
   const int pixel_count = (h_end - h_start) * (v_end - v_start);
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
-      find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+  uint8_t avg = find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   int32_t M_int32[WIENER_WIN][WIENER_WIN] = { { 0 } };
   int64_t M_int64[WIENER_WIN][WIENER_WIN] = { { 0 } };
@@ -103,23 +102,281 @@ static INLINE void compute_stats_win7_opt_avx2(
     }
   }
 
-  const double avg_square_sum = avg * avg * pixel_count;
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
   for (k = 0; k < wiener_win; k++) {
     for (l = 0; l < wiener_win; l++) {
       const int32_t idx0 = l * wiener_win + k;
-      M[idx0] = M_int64[k][l] + avg_square_sum - avg * (sumX + sumY[k][l]);
-      double *H_ = H + idx0 * wiener_win2;
+      M[idx0] =
+          M_int64[k][l] + (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]));
+      int64_t *H_ = H + idx0 * wiener_win2;
       int64_t *H_int_ = &H_int64[idx0][0];
       for (m = 0; m < wiener_win; m++) {
         for (n = 0; n < wiener_win; n++) {
           H_[m * wiener_win + n] = H_int_[n * 8 + m] + avg_square_sum -
-                                   avg * (sumY[k][l] + sumY[n][m]);
+                                   (int64_t)avg * (sumY[k][l] + sumY[n][m]);
         }
       }
     }
   }
 }
 
+static INLINE void acc_stat_highbd_avx2(int64_t *dst, const uint16_t *dgd,
+                                        const __m256i *shuffle,
+                                        const __m256i *dgd_ijkl) {
+  // Load two 128-bit chunks from dgd
+  const __m256i s0 = _mm256_inserti128_si256(
+      _mm256_castsi128_si256(_mm_loadu_si128((__m128i *)dgd)),
+      _mm_loadu_si128((__m128i *)(dgd + 4)), 1);
+  // s0 = [11 10 9 8 7 6 5 4] [7 6 5 4 3 2 1 0] as u16 (values are dgd indices)
+  // The weird order is so the shuffle stays within 128-bit lanes
+
+  // Shuffle 16x u16 values within lanes according to the mask:
+  // [0 1 1 2 2 3 3 4] [0 1 1 2 2 3 3 4]
+  // (Actually we shuffle u8 values as there's no 16-bit shuffle)
+  const __m256i s1 = _mm256_shuffle_epi8(s0, *shuffle);
+  // s1 = [8 7 7 6 6 5 5 4] [4 3 3 2 2 1 1 0] as u16 (values are dgd indices)
+
+  // Multiply 16x 16-bit integers in dgd_ijkl and s1, resulting in 16x 32-bit
+  // integers then horizontally add pairs of these integers resulting in 8x
+  // 32-bit integers
+  const __m256i d0 = _mm256_madd_epi16(*dgd_ijkl, s1);
+  // d0 = [a b c d] [e f g h] as u32
+
+  // Take the lower-half of d0, extend to u64, add it on to dst (H)
+  const __m256i d0l = _mm256_cvtepu32_epi64(_mm256_extracti128_si256(d0, 0));
+  // d0l = [a b] [c d] as u64
+  const __m256i dst0 = yy_loadu_256(dst);
+  yy_storeu_256(dst, _mm256_add_epi64(d0l, dst0));
+
+  // Take the upper-half of d0, extend to u64, add it on to dst (H)
+  const __m256i d0h = _mm256_cvtepu32_epi64(_mm256_extracti128_si256(d0, 1));
+  // d0h = [e f] [g h] as u64
+  const __m256i dst1 = yy_loadu_256(dst + 4);
+  yy_storeu_256(dst + 4, _mm256_add_epi64(d0h, dst1));
+}
+
+static INLINE void acc_stat_highbd_win7_one_line_avx2(
+    const uint16_t *dgd, const uint16_t *src, int h_start, int h_end,
+    int dgd_stride, const __m256i *shuffle, int32_t *sumX,
+    int32_t sumY[WIENER_WIN][WIENER_WIN], int64_t M_int[WIENER_WIN][WIENER_WIN],
+    int64_t H_int[WIENER_WIN2][WIENER_WIN * 8]) {
+  int j, k, l;
+  const int wiener_win = WIENER_WIN;
+  for (j = h_start; j < h_end; j += 2) {
+    const uint16_t X1 = src[j];
+    const uint16_t X2 = src[j + 1];
+    *sumX += X1 + X2;
+    const uint16_t *dgd_ij = dgd + j;
+    for (k = 0; k < wiener_win; k++) {
+      const uint16_t *dgd_ijk = dgd_ij + k * dgd_stride;
+      for (l = 0; l < wiener_win; l++) {
+        int64_t *H_ = &H_int[(l * wiener_win + k)][0];
+        const uint16_t D1 = dgd_ijk[l];
+        const uint16_t D2 = dgd_ijk[l + 1];
+        sumY[k][l] += D1 + D2;
+        M_int[k][l] += D1 * X1 + D2 * X2;
+
+        // Load two u16 values from dgd_ijkl combined as a u32,
+        // then broadcast to 8x u32 slots of a 256
+        const __m256i dgd_ijkl =
+            _mm256_set1_epi32(*((uint32_t *)(dgd_ijk + l)));
+        // dgd_ijkl = [y x y x y x y x] [y x y x y x y x] where each is a u16
+
+        acc_stat_highbd_avx2(H_ + 0 * 8, dgd_ij + 0 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 1 * 8, dgd_ij + 1 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 2 * 8, dgd_ij + 2 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 3 * 8, dgd_ij + 3 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 4 * 8, dgd_ij + 4 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 5 * 8, dgd_ij + 5 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 6 * 8, dgd_ij + 6 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+      }
+    }
+  }
+}
+
+static INLINE void compute_stats_highbd_win7_opt_avx2(
+    const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end,
+    int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M,
+    int64_t *H, aom_bit_depth_t bit_depth) {
+  int i, j, k, l, m, n;
+  const int wiener_win = WIENER_WIN;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
+  const uint16_t avg =
+      find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  int64_t M_int[WIENER_WIN][WIENER_WIN] = { { 0 } };
+  int64_t H_int[WIENER_WIN2][WIENER_WIN * 8] = { { 0 } };
+  int32_t sumY[WIENER_WIN][WIENER_WIN] = { { 0 } };
+  int32_t sumX = 0;
+  const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  const __m256i shuffle = yy_loadu_256(g_shuffle_stats_highbd_data);
+  for (j = v_start; j < v_end; j += 64) {
+    const int vert_end = AOMMIN(64, v_end - j) + j;
+    for (i = j; i < vert_end; i++) {
+      acc_stat_highbd_win7_one_line_avx2(
+          dgd_win + i * dgd_stride, src + i * src_stride, h_start, h_end,
+          dgd_stride, &shuffle, &sumX, sumY, M_int, H_int);
+    }
+  }
+
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      const int32_t idx0 = l * wiener_win + k;
+      M[idx0] = (M_int[k][l] +
+                 (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) /
+                bit_depth_divider;
+      int64_t *H_ = H + idx0 * wiener_win2;
+      int64_t *H_int_ = &H_int[idx0][0];
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H_[m * wiener_win + n] =
+              (H_int_[n * 8 + m] +
+               (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) /
+              bit_depth_divider;
+        }
+      }
+    }
+  }
+}
+
+static INLINE void acc_stat_highbd_win5_one_line_avx2(
+    const uint16_t *dgd, const uint16_t *src, int h_start, int h_end,
+    int dgd_stride, const __m256i *shuffle, int32_t *sumX,
+    int32_t sumY[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA],
+    int64_t M_int[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA],
+    int64_t H_int[WIENER_WIN2_CHROMA][WIENER_WIN_CHROMA * 8]) {
+  int j, k, l;
+  const int wiener_win = WIENER_WIN_CHROMA;
+  for (j = h_start; j < h_end; j += 2) {
+    const uint16_t X1 = src[j];
+    const uint16_t X2 = src[j + 1];
+    *sumX += X1 + X2;
+    const uint16_t *dgd_ij = dgd + j;
+    for (k = 0; k < wiener_win; k++) {
+      const uint16_t *dgd_ijk = dgd_ij + k * dgd_stride;
+      for (l = 0; l < wiener_win; l++) {
+        int64_t *H_ = &H_int[(l * wiener_win + k)][0];
+        const uint16_t D1 = dgd_ijk[l];
+        const uint16_t D2 = dgd_ijk[l + 1];
+        sumY[k][l] += D1 + D2;
+        M_int[k][l] += D1 * X1 + D2 * X2;
+
+        // Load two u16 values from dgd_ijkl combined as a u32,
+        // then broadcast to 8x u32 slots of a 256
+        const __m256i dgd_ijkl =
+            _mm256_set1_epi32(*((uint32_t *)(dgd_ijk + l)));
+        // dgd_ijkl = [x y x y x y x y] [x y x y x y x y] where each is a u16
+
+        acc_stat_highbd_avx2(H_ + 0 * 8, dgd_ij + 0 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 1 * 8, dgd_ij + 1 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 2 * 8, dgd_ij + 2 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 3 * 8, dgd_ij + 3 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+        acc_stat_highbd_avx2(H_ + 4 * 8, dgd_ij + 4 * dgd_stride, shuffle,
+                             &dgd_ijkl);
+      }
+    }
+  }
+}
+
+static INLINE void compute_stats_highbd_win5_opt_avx2(
+    const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end,
+    int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M,
+    int64_t *H, aom_bit_depth_t bit_depth) {
+  int i, j, k, l, m, n;
+  const int wiener_win = WIENER_WIN_CHROMA;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
+  const uint16_t avg =
+      find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  int64_t M_int64[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
+  int64_t H_int64[WIENER_WIN2_CHROMA][WIENER_WIN_CHROMA * 8] = { { 0 } };
+  int32_t sumY[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
+  int32_t sumX = 0;
+  const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  const __m256i shuffle = yy_loadu_256(g_shuffle_stats_highbd_data);
+  for (j = v_start; j < v_end; j += 64) {
+    const int vert_end = AOMMIN(64, v_end - j) + j;
+    for (i = j; i < vert_end; i++) {
+      acc_stat_highbd_win5_one_line_avx2(
+          dgd_win + i * dgd_stride, src + i * src_stride, h_start, h_end,
+          dgd_stride, &shuffle, &sumX, sumY, M_int64, H_int64);
+    }
+  }
+
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      const int32_t idx0 = l * wiener_win + k;
+      M[idx0] = (M_int64[k][l] +
+                 (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) /
+                bit_depth_divider;
+      int64_t *H_ = H + idx0 * wiener_win2;
+      int64_t *H_int_ = &H_int64[idx0][0];
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H_[m * wiener_win + n] =
+              (H_int_[n * 8 + m] +
+               (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) /
+              bit_depth_divider;
+        }
+      }
+    }
+  }
+}
+
+void av1_compute_stats_highbd_avx2(int wiener_win, const uint8_t *dgd8,
+                                   const uint8_t *src8, int h_start, int h_end,
+                                   int v_start, int v_end, int dgd_stride,
+                                   int src_stride, int64_t *M, int64_t *H,
+                                   aom_bit_depth_t bit_depth) {
+  if (wiener_win == WIENER_WIN) {
+    compute_stats_highbd_win7_opt_avx2(dgd8, src8, h_start, h_end, v_start,
+                                       v_end, dgd_stride, src_stride, M, H,
+                                       bit_depth);
+  } else if (wiener_win == WIENER_WIN_CHROMA) {
+    compute_stats_highbd_win5_opt_avx2(dgd8, src8, h_start, h_end, v_start,
+                                       v_end, dgd_stride, src_stride, M, H,
+                                       bit_depth);
+  } else {
+    av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end, v_start,
+                               v_end, dgd_stride, src_stride, M, H, bit_depth);
+  }
+}
+
 static INLINE void acc_stat_win5_one_line_avx2(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end,
     int dgd_stride, const __m128i *shuffle, int32_t *sumX,
@@ -156,14 +413,13 @@ static INLINE void acc_stat_win5_one_line_avx2(
 
 static INLINE void compute_stats_win5_opt_avx2(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start,
-    int v_end, int dgd_stride, int src_stride, double *M, double *H) {
+    int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H) {
   int i, j, k, l, m, n;
   const int wiener_win = WIENER_WIN_CHROMA;
   const int pixel_count = (h_end - h_start) * (v_end - v_start);
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
-      find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+  uint8_t avg = find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   int32_t M_int32[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
   int64_t M_int64[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
@@ -195,17 +451,18 @@ static INLINE void compute_stats_win5_opt_avx2(
     }
   }
 
-  const double avg_square_sum = avg * avg * pixel_count;
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
   for (k = 0; k < wiener_win; k++) {
     for (l = 0; l < wiener_win; l++) {
       const int32_t idx0 = l * wiener_win + k;
-      M[idx0] = M_int64[k][l] + avg_square_sum - avg * (sumX + sumY[k][l]);
-      double *H_ = H + idx0 * wiener_win2;
+      M[idx0] =
+          M_int64[k][l] + (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]));
+      int64_t *H_ = H + idx0 * wiener_win2;
       int64_t *H_int_ = &H_int64[idx0][0];
       for (m = 0; m < wiener_win; m++) {
         for (n = 0; n < wiener_win; n++) {
           H_[m * wiener_win + n] = H_int_[n * 8 + m] + avg_square_sum -
-                                   avg * (sumY[k][l] + sumY[n][m]);
+                                   (int64_t)avg * (sumY[k][l] + sumY[n][m]);
         }
       }
     }
@@ -215,7 +472,7 @@ static INLINE void compute_stats_win5_opt_avx2(
 void av1_compute_stats_avx2(int wiener_win, const uint8_t *dgd,
                             const uint8_t *src, int h_start, int h_end,
                             int v_start, int v_end, int dgd_stride,
-                            int src_stride, double *M, double *H) {
+                            int src_stride, int64_t *M, int64_t *H) {
   if (wiener_win == WIENER_WIN) {
     compute_stats_win7_opt_avx2(dgd, src, h_start, h_end, v_start, v_end,
                                 dgd_stride, src_stride, M, H);
@@ -279,7 +536,7 @@ int64_t av1_lowbd_pixel_proj_error_avx2(
         const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
         int32_t v = xq[0] * (flt0[k] - u) + xq[1] * (flt1[k] - u);
         const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -292,22 +549,25 @@ int64_t av1_lowbd_pixel_proj_error_avx2(
       sum64 = _mm256_add_epi64(sum64, sum64_0);
       sum64 = _mm256_add_epi64(sum64, sum64_1);
     }
-  } else if (params->r[0] > 0) {
-    __m256i xq_coeff =
-        pair_set_epi16(xq[0], (-xq[0] * (1 << SGRPROJ_RST_BITS)));
+  } else if (params->r[0] > 0 || params->r[1] > 0) {
+    const int xq_active = (params->r[0] > 0) ? xq[0] : xq[1];
+    const __m256i xq_coeff =
+        pair_set_epi16(xq_active, (-xq_active * (1 << SGRPROJ_RST_BITS)));
+    const int32_t *flt = (params->r[0] > 0) ? flt0 : flt1;
+    const int flt_stride = (params->r[0] > 0) ? flt0_stride : flt1_stride;
     for (i = 0; i < height; ++i) {
       __m256i sum32 = _mm256_setzero_si256();
       for (j = 0; j <= width - 16; j += 16) {
         const __m256i d0 = _mm256_cvtepu8_epi16(xx_loadu_128(dat + j));
         const __m256i s0 = _mm256_cvtepu8_epi16(xx_loadu_128(src + j));
-        const __m256i flt0_16b = _mm256_permute4x64_epi64(
-            _mm256_packs_epi32(yy_loadu_256(flt0 + j),
-                               yy_loadu_256(flt0 + j + 8)),
+        const __m256i flt_16b = _mm256_permute4x64_epi64(
+            _mm256_packs_epi32(yy_loadu_256(flt + j),
+                               yy_loadu_256(flt + j + 8)),
             0xd8);
         const __m256i v0 =
-            _mm256_madd_epi16(xq_coeff, _mm256_unpacklo_epi16(flt0_16b, d0));
+            _mm256_madd_epi16(xq_coeff, _mm256_unpacklo_epi16(flt_16b, d0));
         const __m256i v1 =
-            _mm256_madd_epi16(xq_coeff, _mm256_unpackhi_epi16(flt0_16b, d0));
+            _mm256_madd_epi16(xq_coeff, _mm256_unpackhi_epi16(flt_16b, d0));
         const __m256i vr0 =
             _mm256_srai_epi32(_mm256_add_epi32(v0, rounding), shift);
         const __m256i vr1 =
@@ -319,53 +579,13 @@ int64_t av1_lowbd_pixel_proj_error_avx2(
       }
       for (k = j; k < width; ++k) {
         const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
-        int32_t v = xq[0] * (flt0[k] - u);
+        int32_t v = xq_active * (flt[k] - u);
         const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
-      flt0 += flt0_stride;
-      const __m256i sum64_0 =
-          _mm256_cvtepi32_epi64(_mm256_castsi256_si128(sum32));
-      const __m256i sum64_1 =
-          _mm256_cvtepi32_epi64(_mm256_extracti128_si256(sum32, 1));
-      sum64 = _mm256_add_epi64(sum64, sum64_0);
-      sum64 = _mm256_add_epi64(sum64, sum64_1);
-    }
-  } else if (params->r[1] > 0) {
-    __m256i xq_coeff = pair_set_epi16(xq[1], -(xq[1] << SGRPROJ_RST_BITS));
-    for (i = 0; i < height; ++i) {
-      __m256i sum32 = _mm256_setzero_si256();
-      for (j = 0; j <= width - 16; j += 16) {
-        const __m256i d0 = _mm256_cvtepu8_epi16(xx_loadu_128(dat + j));
-        const __m256i s0 = _mm256_cvtepu8_epi16(xx_loadu_128(src + j));
-        const __m256i flt1_16b = _mm256_permute4x64_epi64(
-            _mm256_packs_epi32(yy_loadu_256(flt1 + j),
-                               yy_loadu_256(flt1 + j + 8)),
-            0xd8);
-        const __m256i v0 =
-            _mm256_madd_epi16(xq_coeff, _mm256_unpacklo_epi16(flt1_16b, d0));
-        const __m256i v1 =
-            _mm256_madd_epi16(xq_coeff, _mm256_unpackhi_epi16(flt1_16b, d0));
-        const __m256i vr0 =
-            _mm256_srai_epi32(_mm256_add_epi32(v0, rounding), shift);
-        const __m256i vr1 =
-            _mm256_srai_epi32(_mm256_add_epi32(v1, rounding), shift);
-        const __m256i e0 = _mm256_sub_epi16(
-            _mm256_add_epi16(_mm256_packs_epi32(vr0, vr1), d0), s0);
-        const __m256i err0 = _mm256_madd_epi16(e0, e0);
-        sum32 = _mm256_add_epi32(sum32, err0);
-      }
-      for (k = j; k < width; ++k) {
-        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
-        int32_t v = xq[1] * (flt1[k] - u);
-        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
-      }
-      dat += dat_stride;
-      src += src_stride;
-      flt1 += flt1_stride;
+      flt += flt_stride;
       const __m256i sum64_0 =
           _mm256_cvtepi32_epi64(_mm256_castsi256_si128(sum32));
       const __m256i sum64_1 =
@@ -385,7 +605,7 @@ int64_t av1_lowbd_pixel_proj_error_avx2(
       }
       for (k = j; k < width; ++k) {
         const int32_t e = (int32_t)(dat[k]) - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -401,3 +621,223 @@ int64_t av1_lowbd_pixel_proj_error_avx2(
   err += sum[0] + sum[1] + sum[2] + sum[3];
   return err;
 }
+
+int64_t av1_highbd_pixel_proj_error_avx2(
+    const uint8_t *src8, int width, int height, int src_stride,
+    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
+    int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params) {
+  int i, j, k;
+  const int32_t shift = SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS;
+  const __m256i rounding = _mm256_set1_epi32(1 << (shift - 1));
+  __m256i sum64 = _mm256_setzero_si256();
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
+  int64_t err = 0;
+  if (params->r[0] > 0 && params->r[1] > 0) {  // Both filters are enabled
+    const __m256i xq0 = _mm256_set1_epi32(xq[0]);
+    const __m256i xq1 = _mm256_set1_epi32(xq[1]);
+    for (i = 0; i < height; ++i) {
+      __m256i sum32 = _mm256_setzero_si256();
+      for (j = 0; j <= width - 16; j += 16) {  // Process 16 pixels at a time
+        // Load 16 pixels each from source image and corrupted image
+        const __m256i s0 = yy_loadu_256(src + j);
+        const __m256i d0 = yy_loadu_256(dat + j);
+        // s0 = [15 14 13 12 11 10 9 8] [7 6 5 4 3 2 1 0] as u16 (indices)
+
+        // Shift-up each pixel to match filtered image scaling
+        const __m256i u0 = _mm256_slli_epi16(d0, SGRPROJ_RST_BITS);
+
+        // Split u0 into two halves and pad each from u16 to i32
+        const __m256i u0l = _mm256_cvtepu16_epi32(_mm256_castsi256_si128(u0));
+        const __m256i u0h =
+            _mm256_cvtepu16_epi32(_mm256_extracti128_si256(u0, 1));
+        // u0h, u0l = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0] as u32
+
+        // Load 16 pixels from each filtered image
+        const __m256i flt0l = yy_loadu_256(flt0 + j);
+        const __m256i flt0h = yy_loadu_256(flt0 + j + 8);
+        const __m256i flt1l = yy_loadu_256(flt1 + j);
+        const __m256i flt1h = yy_loadu_256(flt1 + j + 8);
+        // flt?l, flt?h = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0] as u32
+
+        // Subtract shifted corrupt image from each filtered image
+        const __m256i flt0l_subu = _mm256_sub_epi32(flt0l, u0l);
+        const __m256i flt0h_subu = _mm256_sub_epi32(flt0h, u0h);
+        const __m256i flt1l_subu = _mm256_sub_epi32(flt1l, u0l);
+        const __m256i flt1h_subu = _mm256_sub_epi32(flt1h, u0h);
+
+        // Multiply basis vectors by appropriate coefficients
+        const __m256i v0l = _mm256_mullo_epi32(flt0l_subu, xq0);
+        const __m256i v0h = _mm256_mullo_epi32(flt0h_subu, xq0);
+        const __m256i v1l = _mm256_mullo_epi32(flt1l_subu, xq1);
+        const __m256i v1h = _mm256_mullo_epi32(flt1h_subu, xq1);
+
+        // Add together the contributions from the two basis vectors
+        const __m256i vl = _mm256_add_epi32(v0l, v1l);
+        const __m256i vh = _mm256_add_epi32(v0h, v1h);
+
+        // Right-shift v with appropriate rounding
+        const __m256i vrl =
+            _mm256_srai_epi32(_mm256_add_epi32(vl, rounding), shift);
+        const __m256i vrh =
+            _mm256_srai_epi32(_mm256_add_epi32(vh, rounding), shift);
+        // vrh, vrl = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0]
+
+        // Saturate each i32 to an i16 then combine both halves
+        // The permute (control=[3 1 2 0]) fixes weird ordering from AVX lanes
+        const __m256i vr =
+            _mm256_permute4x64_epi64(_mm256_packs_epi32(vrl, vrh), 0xd8);
+        // intermediate = [15 14 13 12 7 6 5 4] [11 10 9 8 3 2 1 0]
+        // vr = [15 14 13 12 11 10 9 8] [7 6 5 4 3 2 1 0]
+
+        // Add twin-subspace-sgr-filter to corrupt image then subtract source
+        const __m256i e0 = _mm256_sub_epi16(_mm256_add_epi16(vr, d0), s0);
+
+        // Calculate squared error and add adjacent values
+        const __m256i err0 = _mm256_madd_epi16(e0, e0);
+
+        sum32 = _mm256_add_epi32(sum32, err0);
+      }
+
+      const __m256i sum32l =
+          _mm256_cvtepu32_epi64(_mm256_castsi256_si128(sum32));
+      sum64 = _mm256_add_epi64(sum64, sum32l);
+      const __m256i sum32h =
+          _mm256_cvtepu32_epi64(_mm256_extracti128_si256(sum32, 1));
+      sum64 = _mm256_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels in this row (modulo 16)
+      for (k = j; k < width; ++k) {
+        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
+        int32_t v = xq[0] * (flt0[k] - u) + xq[1] * (flt1[k] - u);
+        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+      flt0 += flt0_stride;
+      flt1 += flt1_stride;
+    }
+  } else if (params->r[0] > 0 || params->r[1] > 0) {  // Only one filter enabled
+    const int32_t xq_on = (params->r[0] > 0) ? xq[0] : xq[1];
+    const __m256i xq_active = _mm256_set1_epi32(xq_on);
+    const __m256i xq_inactive =
+        _mm256_set1_epi32(-xq_on * (1 << SGRPROJ_RST_BITS));
+    const int32_t *flt = (params->r[0] > 0) ? flt0 : flt1;
+    const int flt_stride = (params->r[0] > 0) ? flt0_stride : flt1_stride;
+    for (i = 0; i < height; ++i) {
+      __m256i sum32 = _mm256_setzero_si256();
+      for (j = 0; j <= width - 16; j += 16) {
+        // Load 16 pixels from source image
+        const __m256i s0 = yy_loadu_256(src + j);
+        // s0 = [15 14 13 12 11 10 9 8] [7 6 5 4 3 2 1 0] as u16
+
+        // Load 16 pixels from corrupted image and pad each u16 to i32
+        const __m256i d0 = yy_loadu_256(dat + j);
+        const __m256i d0h =
+            _mm256_cvtepu16_epi32(_mm256_extracti128_si256(d0, 1));
+        const __m256i d0l = _mm256_cvtepu16_epi32(_mm256_castsi256_si128(d0));
+        // d0 = [15 14 13 12 11 10 9 8] [7 6 5 4 3 2 1 0] as u16
+        // d0h, d0l = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0] as i32
+
+        // Load 16 pixels from the filtered image
+        const __m256i flth = yy_loadu_256(flt + j + 8);
+        const __m256i fltl = yy_loadu_256(flt + j);
+        // flth, fltl = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0] as i32
+
+        const __m256i flth_xq = _mm256_mullo_epi32(flth, xq_active);
+        const __m256i fltl_xq = _mm256_mullo_epi32(fltl, xq_active);
+        const __m256i d0h_xq = _mm256_mullo_epi32(d0h, xq_inactive);
+        const __m256i d0l_xq = _mm256_mullo_epi32(d0l, xq_inactive);
+
+        const __m256i vh = _mm256_add_epi32(flth_xq, d0h_xq);
+        const __m256i vl = _mm256_add_epi32(fltl_xq, d0l_xq);
+
+        // Shift this down with appropriate rounding
+        const __m256i vrh =
+            _mm256_srai_epi32(_mm256_add_epi32(vh, rounding), shift);
+        const __m256i vrl =
+            _mm256_srai_epi32(_mm256_add_epi32(vl, rounding), shift);
+        // vrh, vrl = [15 14 13 12] [11 10 9 8], [7 6 5 4] [3 2 1 0] as i32
+
+        // Saturate each i32 to an i16 then combine both halves
+        // The permute (control=[3 1 2 0]) fixes weird ordering from AVX lanes
+        const __m256i vr =
+            _mm256_permute4x64_epi64(_mm256_packs_epi32(vrl, vrh), 0xd8);
+        // intermediate = [15 14 13 12 7 6 5 4] [11 10 9 8 3 2 1 0] as u16
+        // vr = [15 14 13 12 11 10 9 8] [7 6 5 4 3 2 1 0] as u16
+
+        // Subtract twin-subspace-sgr filtered from source image to get error
+        const __m256i e0 = _mm256_sub_epi16(_mm256_add_epi16(vr, d0), s0);
+
+        // Calculate squared error and add adjacent values
+        const __m256i err0 = _mm256_madd_epi16(e0, e0);
+
+        sum32 = _mm256_add_epi32(sum32, err0);
+      }
+
+      const __m256i sum32l =
+          _mm256_cvtepu32_epi64(_mm256_castsi256_si128(sum32));
+      sum64 = _mm256_add_epi64(sum64, sum32l);
+      const __m256i sum32h =
+          _mm256_cvtepu32_epi64(_mm256_extracti128_si256(sum32, 1));
+      sum64 = _mm256_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels in this row (modulo 16)
+      for (k = j; k < width; ++k) {
+        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
+        int32_t v = xq_on * (flt[k] - u);
+        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+      flt += flt_stride;
+    }
+  } else {  // Neither filter is enabled
+    for (i = 0; i < height; ++i) {
+      __m256i sum32 = _mm256_setzero_si256();
+      for (j = 0; j <= width - 32; j += 32) {
+        // Load 2x16 u16 from source image
+        const __m256i s0l = yy_loadu_256(src + j);
+        const __m256i s0h = yy_loadu_256(src + j + 16);
+
+        // Load 2x16 u16 from corrupted image
+        const __m256i d0l = yy_loadu_256(dat + j);
+        const __m256i d0h = yy_loadu_256(dat + j + 16);
+
+        // Subtract corrupted image from source image
+        const __m256i diffl = _mm256_sub_epi16(d0l, s0l);
+        const __m256i diffh = _mm256_sub_epi16(d0h, s0h);
+
+        // Square error and add adjacent values
+        const __m256i err0l = _mm256_madd_epi16(diffl, diffl);
+        const __m256i err0h = _mm256_madd_epi16(diffh, diffh);
+
+        sum32 = _mm256_add_epi32(sum32, err0l);
+        sum32 = _mm256_add_epi32(sum32, err0h);
+      }
+
+      const __m256i sum32l =
+          _mm256_cvtepu32_epi64(_mm256_castsi256_si128(sum32));
+      sum64 = _mm256_add_epi64(sum64, sum32l);
+      const __m256i sum32h =
+          _mm256_cvtepu32_epi64(_mm256_extracti128_si256(sum32, 1));
+      sum64 = _mm256_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels (modulu 16)
+      for (k = j; k < width; ++k) {
+        const int32_t e = (int32_t)(dat[k]) - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+    }
+  }
+
+  // Sum 4 values from sum64l and sum64h into err
+  int64_t sum[4];
+  yy_storeu_256(sum, sum64);
+  err += sum[0] + sum[1] + sum[2] + sum[3];
+  return err;
+}
diff --git a/third_party/aom/av1/encoder/x86/pickrst_sse4.c b/third_party/aom/av1/encoder/x86/pickrst_sse4.c
index 04e4d1afc4..a94e169849 100644
--- a/third_party/aom/av1/encoder/x86/pickrst_sse4.c
+++ b/third_party/aom/av1/encoder/x86/pickrst_sse4.c
@@ -68,13 +68,13 @@ static INLINE void acc_stat_win7_one_line_sse4_1(
 
 static INLINE void compute_stats_win7_opt_sse4_1(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start,
-    int v_end, int dgd_stride, int src_stride, double *M, double *H) {
+    int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H) {
   int i, j, k, l, m, n;
   const int wiener_win = WIENER_WIN;
   const int pixel_count = (h_end - h_start) * (v_end - v_start);
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
+  const uint8_t avg =
       find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   int32_t M_int32[WIENER_WIN][WIENER_WIN] = { { 0 } };
@@ -107,23 +107,286 @@ static INLINE void compute_stats_win7_opt_sse4_1(
     }
   }
 
-  const double avg_square_sum = avg * avg * pixel_count;
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
   for (k = 0; k < wiener_win; k++) {
     for (l = 0; l < wiener_win; l++) {
       const int32_t idx0 = l * wiener_win + k;
-      M[idx0] = M_int64[k][l] + avg_square_sum - avg * (sumX + sumY[k][l]);
-      double *H_ = H + idx0 * wiener_win2;
+      M[idx0] =
+          M_int64[k][l] + (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]));
+      int64_t *H_ = H + idx0 * wiener_win2;
       int64_t *H_int_ = &H_int64[idx0][0];
       for (m = 0; m < wiener_win; m++) {
         for (n = 0; n < wiener_win; n++) {
           H_[m * wiener_win + n] = H_int_[n * 8 + m] + avg_square_sum -
-                                   avg * (sumY[k][l] + sumY[n][m]);
+                                   (int64_t)avg * (sumY[k][l] + sumY[n][m]);
         }
       }
     }
   }
 }
 
+static INLINE void acc_stat_highbd_sse41(int64_t *dst, const uint16_t *dgd,
+                                         const __m128i *shuffle,
+                                         const __m128i *dgd_ijkl) {
+  // Load 256 bits from dgd in two chunks
+  const __m128i s0l = xx_loadu_128(dgd);
+  const __m128i s0h = xx_loadu_128(dgd + 4);
+  // s0l = [7 6 5 4 3 2 1 0] as u16 values (dgd indices)
+  // s0h = [11 10 9 8 7 6 5 4] as u16 values (dgd indices)
+  // (Slightly strange order so we can apply the same shuffle to both halves)
+
+  // Shuffle the u16 values in each half (actually using 8-bit shuffle mask)
+  const __m128i s1l = _mm_shuffle_epi8(s0l, *shuffle);
+  const __m128i s1h = _mm_shuffle_epi8(s0h, *shuffle);
+  // s1l = [4 3 3 2 2 1 1 0] as u16 values (dgd indices)
+  // s1h = [8 7 7 6 6 5 5 4] as u16 values (dgd indices)
+
+  // Multiply s1 by dgd_ijkl resulting in 8x u32 values
+  // Horizontally add pairs of u32 resulting in 4x u32
+  const __m128i dl = _mm_madd_epi16(*dgd_ijkl, s1l);
+  const __m128i dh = _mm_madd_epi16(*dgd_ijkl, s1h);
+  // dl = [d c b a] as u32 values
+  // dh = [h g f e] as u32 values
+
+  // Add these 8x u32 results on to dst in four parts
+  const __m128i dll = _mm_cvtepu32_epi64(dl);
+  const __m128i dlh = _mm_cvtepu32_epi64(_mm_srli_si128(dl, 8));
+  const __m128i dhl = _mm_cvtepu32_epi64(dh);
+  const __m128i dhh = _mm_cvtepu32_epi64(_mm_srli_si128(dh, 8));
+  // dll = [b a] as u64 values, etc.
+
+  const __m128i rll = _mm_add_epi64(xx_loadu_128(dst), dll);
+  xx_storeu_128(dst, rll);
+  const __m128i rlh = _mm_add_epi64(xx_loadu_128(dst + 2), dlh);
+  xx_storeu_128(dst + 2, rlh);
+  const __m128i rhl = _mm_add_epi64(xx_loadu_128(dst + 4), dhl);
+  xx_storeu_128(dst + 4, rhl);
+  const __m128i rhh = _mm_add_epi64(xx_loadu_128(dst + 6), dhh);
+  xx_storeu_128(dst + 6, rhh);
+}
+
+static INLINE void acc_stat_highbd_win7_one_line_sse4_1(
+    const uint16_t *dgd, const uint16_t *src, int h_start, int h_end,
+    int dgd_stride, const __m128i *shuffle, int32_t *sumX,
+    int32_t sumY[WIENER_WIN][WIENER_WIN], int64_t M_int[WIENER_WIN][WIENER_WIN],
+    int64_t H_int[WIENER_WIN2][WIENER_WIN * 8]) {
+  int j, k, l;
+  const int wiener_win = WIENER_WIN;
+  for (j = h_start; j < h_end; j += 2) {
+    const uint16_t X1 = src[j];
+    const uint16_t X2 = src[j + 1];
+    *sumX += X1 + X2;
+    const uint16_t *dgd_ij = dgd + j;
+    for (k = 0; k < wiener_win; k++) {
+      const uint16_t *dgd_ijk = dgd_ij + k * dgd_stride;
+      for (l = 0; l < wiener_win; l++) {
+        int64_t *H_ = &H_int[(l * wiener_win + k)][0];
+        const uint16_t D1 = dgd_ijk[l];
+        const uint16_t D2 = dgd_ijk[l + 1];
+        sumY[k][l] += D1 + D2;
+        M_int[k][l] += D1 * X1 + D2 * X2;
+
+        // Load two u16 values from dgd as a single u32
+        // Then broadcast to 4x u32 slots of a 128
+        const __m128i dgd_ijkl = _mm_set1_epi32(*((uint32_t *)(dgd_ijk + l)));
+        // dgd_ijkl = [y x y x y x y x] as u16
+
+        acc_stat_highbd_sse41(H_ + 0 * 8, dgd_ij + 0 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 1 * 8, dgd_ij + 1 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 2 * 8, dgd_ij + 2 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 3 * 8, dgd_ij + 3 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 4 * 8, dgd_ij + 4 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 5 * 8, dgd_ij + 5 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 6 * 8, dgd_ij + 6 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+      }
+    }
+  }
+}
+
+static INLINE void compute_stats_highbd_win7_opt_sse4_1(
+    const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end,
+    int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M,
+    int64_t *H, aom_bit_depth_t bit_depth) {
+  int i, j, k, l, m, n;
+  const int wiener_win = WIENER_WIN;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
+  const uint16_t avg =
+      find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  int64_t M_int[WIENER_WIN][WIENER_WIN] = { { 0 } };
+  int64_t H_int[WIENER_WIN2][WIENER_WIN * 8] = { { 0 } };
+  int32_t sumY[WIENER_WIN][WIENER_WIN] = { { 0 } };
+  int32_t sumX = 0;
+  const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  // Load just half of the 256-bit shuffle control used for the AVX2 version
+  const __m128i shuffle = xx_loadu_128(g_shuffle_stats_highbd_data);
+  for (j = v_start; j < v_end; j += 64) {
+    const int vert_end = AOMMIN(64, v_end - j) + j;
+    for (i = j; i < vert_end; i++) {
+      acc_stat_highbd_win7_one_line_sse4_1(
+          dgd_win + i * dgd_stride, src + i * src_stride, h_start, h_end,
+          dgd_stride, &shuffle, &sumX, sumY, M_int, H_int);
+    }
+  }
+
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      const int32_t idx0 = l * wiener_win + k;
+      M[idx0] = (M_int[k][l] +
+                 (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) /
+                bit_depth_divider;
+      int64_t *H_ = H + idx0 * wiener_win2;
+      int64_t *H_int_ = &H_int[idx0][0];
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H_[m * wiener_win + n] =
+              (H_int_[n * 8 + m] +
+               (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) /
+              bit_depth_divider;
+        }
+      }
+    }
+  }
+}
+
+static INLINE void acc_stat_highbd_win5_one_line_sse4_1(
+    const uint16_t *dgd, const uint16_t *src, int h_start, int h_end,
+    int dgd_stride, const __m128i *shuffle, int32_t *sumX,
+    int32_t sumY[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA],
+    int64_t M_int[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA],
+    int64_t H_int[WIENER_WIN2_CHROMA][WIENER_WIN_CHROMA * 8]) {
+  int j, k, l;
+  const int wiener_win = WIENER_WIN_CHROMA;
+  for (j = h_start; j < h_end; j += 2) {
+    const uint16_t X1 = src[j];
+    const uint16_t X2 = src[j + 1];
+    *sumX += X1 + X2;
+    const uint16_t *dgd_ij = dgd + j;
+    for (k = 0; k < wiener_win; k++) {
+      const uint16_t *dgd_ijk = dgd_ij + k * dgd_stride;
+      for (l = 0; l < wiener_win; l++) {
+        int64_t *H_ = &H_int[(l * wiener_win + k)][0];
+        const uint16_t D1 = dgd_ijk[l];
+        const uint16_t D2 = dgd_ijk[l + 1];
+        sumY[k][l] += D1 + D2;
+        M_int[k][l] += D1 * X1 + D2 * X2;
+
+        // Load two u16 values from dgd as a single u32
+        // then broadcast to 4x u32 slots of a 128
+        const __m128i dgd_ijkl = _mm_set1_epi32(*((uint32_t *)(dgd_ijk + l)));
+        // dgd_ijkl = [y x y x y x y x] as u16
+
+        acc_stat_highbd_sse41(H_ + 0 * 8, dgd_ij + 0 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 1 * 8, dgd_ij + 1 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 2 * 8, dgd_ij + 2 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 3 * 8, dgd_ij + 3 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+        acc_stat_highbd_sse41(H_ + 4 * 8, dgd_ij + 4 * dgd_stride, shuffle,
+                              &dgd_ijkl);
+      }
+    }
+  }
+}
+
+static INLINE void compute_stats_highbd_win5_opt_sse4_1(
+    const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end,
+    int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M,
+    int64_t *H, aom_bit_depth_t bit_depth) {
+  int i, j, k, l, m, n;
+  const int wiener_win = WIENER_WIN_CHROMA;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
+  const uint16_t avg =
+      find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  int64_t M_int[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
+  int64_t H_int[WIENER_WIN2_CHROMA][WIENER_WIN_CHROMA * 8] = { { 0 } };
+  int32_t sumY[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
+  int32_t sumX = 0;
+  const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  // Load just half of the 256-bit shuffle control used for the AVX2 version
+  const __m128i shuffle = xx_loadu_128(g_shuffle_stats_highbd_data);
+  for (j = v_start; j < v_end; j += 64) {
+    const int vert_end = AOMMIN(64, v_end - j) + j;
+    for (i = j; i < vert_end; i++) {
+      acc_stat_highbd_win5_one_line_sse4_1(
+          dgd_win + i * dgd_stride, src + i * src_stride, h_start, h_end,
+          dgd_stride, &shuffle, &sumX, sumY, M_int, H_int);
+    }
+  }
+
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      const int32_t idx0 = l * wiener_win + k;
+      M[idx0] = (M_int[k][l] +
+                 (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) /
+                bit_depth_divider;
+      int64_t *H_ = H + idx0 * wiener_win2;
+      int64_t *H_int_ = &H_int[idx0][0];
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H_[m * wiener_win + n] =
+              (H_int_[n * 8 + m] +
+               (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) /
+              bit_depth_divider;
+        }
+      }
+    }
+  }
+}
+
+void av1_compute_stats_highbd_sse4_1(int wiener_win, const uint8_t *dgd8,
+                                     const uint8_t *src8, int h_start,
+                                     int h_end, int v_start, int v_end,
+                                     int dgd_stride, int src_stride, int64_t *M,
+                                     int64_t *H, aom_bit_depth_t bit_depth) {
+  if (wiener_win == WIENER_WIN) {
+    compute_stats_highbd_win7_opt_sse4_1(dgd8, src8, h_start, h_end, v_start,
+                                         v_end, dgd_stride, src_stride, M, H,
+                                         bit_depth);
+  } else if (wiener_win == WIENER_WIN_CHROMA) {
+    compute_stats_highbd_win5_opt_sse4_1(dgd8, src8, h_start, h_end, v_start,
+                                         v_end, dgd_stride, src_stride, M, H,
+                                         bit_depth);
+  } else {
+    av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end, v_start,
+                               v_end, dgd_stride, src_stride, M, H, bit_depth);
+  }
+}
+
 static INLINE void acc_stat_win5_one_line_sse4_1(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end,
     int dgd_stride, const __m128i *shuffle, int32_t *sumX,
@@ -160,13 +423,13 @@ static INLINE void acc_stat_win5_one_line_sse4_1(
 
 static INLINE void compute_stats_win5_opt_sse4_1(
     const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start,
-    int v_end, int dgd_stride, int src_stride, double *M, double *H) {
+    int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H) {
   int i, j, k, l, m, n;
   const int wiener_win = WIENER_WIN_CHROMA;
   const int pixel_count = (h_end - h_start) * (v_end - v_start);
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
+  const uint8_t avg =
       find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   int32_t M_int32[WIENER_WIN_CHROMA][WIENER_WIN_CHROMA] = { { 0 } };
@@ -199,17 +462,18 @@ static INLINE void compute_stats_win5_opt_sse4_1(
     }
   }
 
-  const double avg_square_sum = avg * avg * pixel_count;
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
   for (k = 0; k < wiener_win; k++) {
     for (l = 0; l < wiener_win; l++) {
       const int32_t idx0 = l * wiener_win + k;
-      M[idx0] = M_int64[k][l] + avg_square_sum - avg * (sumX + sumY[k][l]);
-      double *H_ = H + idx0 * wiener_win2;
+      M[idx0] =
+          M_int64[k][l] + (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]));
+      int64_t *H_ = H + idx0 * wiener_win2;
       int64_t *H_int_ = &H_int64[idx0][0];
       for (m = 0; m < wiener_win; m++) {
         for (n = 0; n < wiener_win; n++) {
           H_[m * wiener_win + n] = H_int_[n * 8 + m] + avg_square_sum -
-                                   avg * (sumY[k][l] + sumY[n][m]);
+                                   (int64_t)avg * (sumY[k][l] + sumY[n][m]);
         }
       }
     }
@@ -218,7 +482,7 @@ static INLINE void compute_stats_win5_opt_sse4_1(
 void av1_compute_stats_sse4_1(int wiener_win, const uint8_t *dgd,
                               const uint8_t *src, int h_start, int h_end,
                               int v_start, int v_end, int dgd_stride,
-                              int src_stride, double *M, double *H) {
+                              int src_stride, int64_t *M, int64_t *H) {
   if (wiener_win == WIENER_WIN) {
     compute_stats_win7_opt_sse4_1(dgd, src, h_start, h_end, v_start, v_end,
                                   dgd_stride, src_stride, M, H);
@@ -250,7 +514,7 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
     __m128i xq_coeff = pair_set_epi16(xq[0], xq[1]);
     for (i = 0; i < height; ++i) {
       __m128i sum32 = _mm_setzero_si128();
-      for (j = 0; j < width - 8; j += 8) {
+      for (j = 0; j <= width - 8; j += 8) {
         const __m128i d0 = _mm_cvtepu8_epi16(xx_loadl_64(dat + j));
         const __m128i s0 = _mm_cvtepu8_epi16(xx_loadl_64(src + j));
         const __m128i flt0_16b =
@@ -275,7 +539,7 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
         const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
         int32_t v = xq[0] * (flt0[k] - u) + xq[1] * (flt1[k] - u);
         const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -286,19 +550,23 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
       sum64 = _mm_add_epi64(sum64, sum64_0);
       sum64 = _mm_add_epi64(sum64, sum64_1);
     }
-  } else if (params->r[0] > 0) {
-    __m128i xq_coeff = pair_set_epi16(xq[0], -(xq[0] << SGRPROJ_RST_BITS));
+  } else if (params->r[0] > 0 || params->r[1] > 0) {
+    const int xq_active = (params->r[0] > 0) ? xq[0] : xq[1];
+    const __m128i xq_coeff =
+        pair_set_epi16(xq_active, -(xq_active << SGRPROJ_RST_BITS));
+    const int32_t *flt = (params->r[0] > 0) ? flt0 : flt1;
+    const int flt_stride = (params->r[0] > 0) ? flt0_stride : flt1_stride;
     for (i = 0; i < height; ++i) {
       __m128i sum32 = _mm_setzero_si128();
-      for (j = 0; j < width - 8; j += 8) {
+      for (j = 0; j <= width - 8; j += 8) {
         const __m128i d0 = _mm_cvtepu8_epi16(xx_loadl_64(dat + j));
         const __m128i s0 = _mm_cvtepu8_epi16(xx_loadl_64(src + j));
-        const __m128i flt0_16b =
-            _mm_packs_epi32(xx_loadu_128(flt0 + j), xx_loadu_128(flt0 + j + 4));
+        const __m128i flt_16b =
+            _mm_packs_epi32(xx_loadu_128(flt + j), xx_loadu_128(flt + j + 4));
         const __m128i v0 =
-            _mm_madd_epi16(xq_coeff, _mm_unpacklo_epi16(flt0_16b, d0));
+            _mm_madd_epi16(xq_coeff, _mm_unpacklo_epi16(flt_16b, d0));
         const __m128i v1 =
-            _mm_madd_epi16(xq_coeff, _mm_unpackhi_epi16(flt0_16b, d0));
+            _mm_madd_epi16(xq_coeff, _mm_unpackhi_epi16(flt_16b, d0));
         const __m128i vr0 = _mm_srai_epi32(_mm_add_epi32(v0, rounding), shift);
         const __m128i vr1 = _mm_srai_epi32(_mm_add_epi32(v1, rounding), shift);
         const __m128i e0 =
@@ -308,47 +576,13 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
       }
       for (k = j; k < width; ++k) {
         const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
-        int32_t v = xq[0] * (flt0[k] - u);
+        int32_t v = xq_active * (flt[k] - u);
         const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
-      flt0 += flt0_stride;
-      const __m128i sum64_0 = _mm_cvtepi32_epi64(sum32);
-      const __m128i sum64_1 = _mm_cvtepi32_epi64(_mm_srli_si128(sum32, 8));
-      sum64 = _mm_add_epi64(sum64, sum64_0);
-      sum64 = _mm_add_epi64(sum64, sum64_1);
-    }
-  } else if (params->r[1] > 0) {
-    __m128i xq_coeff = pair_set_epi16(xq[1], -(xq[1] << SGRPROJ_RST_BITS));
-    for (i = 0; i < height; ++i) {
-      __m128i sum32 = _mm_setzero_si128();
-      for (j = 0; j < width - 8; j += 8) {
-        const __m128i d0 = _mm_cvtepu8_epi16(xx_loadl_64(dat + j));
-        const __m128i s0 = _mm_cvtepu8_epi16(xx_loadl_64(src + j));
-        const __m128i flt1_16b =
-            _mm_packs_epi32(xx_loadu_128(flt1 + j), xx_loadu_128(flt1 + j + 4));
-        const __m128i v0 =
-            _mm_madd_epi16(xq_coeff, _mm_unpacklo_epi16(flt1_16b, d0));
-        const __m128i v1 =
-            _mm_madd_epi16(xq_coeff, _mm_unpackhi_epi16(flt1_16b, d0));
-        const __m128i vr0 = _mm_srai_epi32(_mm_add_epi32(v0, rounding), shift);
-        const __m128i vr1 = _mm_srai_epi32(_mm_add_epi32(v1, rounding), shift);
-        const __m128i e0 =
-            _mm_sub_epi16(_mm_add_epi16(_mm_packs_epi32(vr0, vr1), d0), s0);
-        const __m128i err0 = _mm_madd_epi16(e0, e0);
-        sum32 = _mm_add_epi32(sum32, err0);
-      }
-      for (k = j; k < width; ++k) {
-        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
-        int32_t v = xq[1] * (flt1[k] - u);
-        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
-        err += e * e;
-      }
-      dat += dat_stride;
-      src += src_stride;
-      flt1 += flt1_stride;
+      flt += flt_stride;
       const __m128i sum64_0 = _mm_cvtepi32_epi64(sum32);
       const __m128i sum64_1 = _mm_cvtepi32_epi64(_mm_srli_si128(sum32, 8));
       sum64 = _mm_add_epi64(sum64, sum64_0);
@@ -357,7 +591,7 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
   } else {
     __m128i sum32 = _mm_setzero_si128();
     for (i = 0; i < height; ++i) {
-      for (j = 0; j < width - 16; j += 16) {
+      for (j = 0; j <= width - 16; j += 16) {
         const __m128i d = xx_loadu_128(dat + j);
         const __m128i s = xx_loadu_128(src + j);
         const __m128i d0 = _mm_cvtepu8_epi16(d);
@@ -373,7 +607,7 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
       }
       for (k = j; k < width; ++k) {
         const int32_t e = (int32_t)(dat[k]) - src[k];
-        err += e * e;
+        err += ((int64_t)e * e);
       }
       dat += dat_stride;
       src += src_stride;
@@ -387,3 +621,209 @@ int64_t av1_lowbd_pixel_proj_error_sse4_1(
   err += sum[0] + sum[1];
   return err;
 }
+
+int64_t av1_highbd_pixel_proj_error_sse4_1(
+    const uint8_t *src8, int width, int height, int src_stride,
+    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
+    int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params) {
+  int i, j, k;
+  const int32_t shift = SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS;
+  const __m128i rounding = _mm_set1_epi32(1 << (shift - 1));
+  __m128i sum64 = _mm_setzero_si128();
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
+  int64_t err = 0;
+  if (params->r[0] > 0 && params->r[1] > 0) {  // Both filters are enabled
+    const __m128i xq0 = _mm_set1_epi32(xq[0]);
+    const __m128i xq1 = _mm_set1_epi32(xq[1]);
+
+    for (i = 0; i < height; ++i) {
+      __m128i sum32 = _mm_setzero_si128();
+      for (j = 0; j <= width - 8; j += 8) {
+        // Load 8x pixels from source image
+        const __m128i s0 = xx_loadu_128(src + j);
+        // s0 = [7 6 5 4 3 2 1 0] as i16 (indices of src[])
+
+        // Load 8x pixels from corrupted image
+        const __m128i d0 = xx_loadu_128(dat + j);
+        // d0 = [7 6 5 4 3 2 1 0] as i16 (indices of dat[])
+
+        // Shift each pixel value up by SGRPROJ_RST_BITS
+        const __m128i u0 = _mm_slli_epi16(d0, SGRPROJ_RST_BITS);
+
+        // Split u0 into two halves and pad each from u16 to i32
+        const __m128i u0l = _mm_cvtepu16_epi32(u0);
+        const __m128i u0h = _mm_cvtepu16_epi32(_mm_srli_si128(u0, 8));
+        // u0h = [7 6 5 4] as i32, u0l = [3 2 1 0] as i32, all dat[] indices
+
+        // Load 8 pixels from first and second filtered images
+        const __m128i flt0l = xx_loadu_128(flt0 + j);
+        const __m128i flt0h = xx_loadu_128(flt0 + j + 4);
+        const __m128i flt1l = xx_loadu_128(flt1 + j);
+        const __m128i flt1h = xx_loadu_128(flt1 + j + 4);
+        // flt0 = [7 6 5 4] [3 2 1 0] as i32 (indices of flt0+j)
+        // flt1 = [7 6 5 4] [3 2 1 0] as i32 (indices of flt1+j)
+
+        // Subtract shifted corrupt image from each filtered image
+        // This gives our two basis vectors for the projection
+        const __m128i flt0l_subu = _mm_sub_epi32(flt0l, u0l);
+        const __m128i flt0h_subu = _mm_sub_epi32(flt0h, u0h);
+        const __m128i flt1l_subu = _mm_sub_epi32(flt1l, u0l);
+        const __m128i flt1h_subu = _mm_sub_epi32(flt1h, u0h);
+        // flt?h_subu = [ f[7]-u[7] f[6]-u[6] f[5]-u[5] f[4]-u[4] ] as i32
+        // flt?l_subu = [ f[3]-u[3] f[2]-u[2] f[1]-u[1] f[0]-u[0] ] as i32
+
+        // Multiply each basis vector by the corresponding coefficient
+        const __m128i v0l = _mm_mullo_epi32(flt0l_subu, xq0);
+        const __m128i v0h = _mm_mullo_epi32(flt0h_subu, xq0);
+        const __m128i v1l = _mm_mullo_epi32(flt1l_subu, xq1);
+        const __m128i v1h = _mm_mullo_epi32(flt1h_subu, xq1);
+
+        // Add together the contribution from each scaled basis vector
+        const __m128i vl = _mm_add_epi32(v0l, v1l);
+        const __m128i vh = _mm_add_epi32(v0h, v1h);
+
+        // Right-shift v with appropriate rounding
+        const __m128i vrl = _mm_srai_epi32(_mm_add_epi32(vl, rounding), shift);
+        const __m128i vrh = _mm_srai_epi32(_mm_add_epi32(vh, rounding), shift);
+
+        // Saturate each i32 value to i16 and combine lower and upper halves
+        const __m128i vr = _mm_packs_epi32(vrl, vrh);
+
+        // Add twin-subspace-sgr-filter to corrupt image then subtract source
+        const __m128i e0 = _mm_sub_epi16(_mm_add_epi16(vr, d0), s0);
+
+        // Calculate squared error and add adjacent values
+        const __m128i err0 = _mm_madd_epi16(e0, e0);
+
+        sum32 = _mm_add_epi32(sum32, err0);
+      }
+
+      const __m128i sum32l = _mm_cvtepu32_epi64(sum32);
+      sum64 = _mm_add_epi64(sum64, sum32l);
+      const __m128i sum32h = _mm_cvtepu32_epi64(_mm_srli_si128(sum32, 8));
+      sum64 = _mm_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels in this row (modulo 8)
+      for (k = j; k < width; ++k) {
+        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
+        int32_t v = xq[0] * (flt0[k] - u) + xq[1] * (flt1[k] - u);
+        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+      flt0 += flt0_stride;
+      flt1 += flt1_stride;
+    }
+  } else if (params->r[0] > 0 || params->r[1] > 0) {  // Only one filter enabled
+    const int32_t xq_on = (params->r[0] > 0) ? xq[0] : xq[1];
+    const __m128i xq_active = _mm_set1_epi32(xq_on);
+    const __m128i xq_inactive =
+        _mm_set1_epi32(-xq_on * (1 << SGRPROJ_RST_BITS));
+    const int32_t *flt = (params->r[0] > 0) ? flt0 : flt1;
+    const int flt_stride = (params->r[0] > 0) ? flt0_stride : flt1_stride;
+    for (i = 0; i < height; ++i) {
+      __m128i sum32 = _mm_setzero_si128();
+      for (j = 0; j <= width - 8; j += 8) {
+        // Load 8x pixels from source image
+        const __m128i s0 = xx_loadu_128(src + j);
+        // s0 = [7 6 5 4 3 2 1 0] as u16 (indices of src[])
+
+        // Load 8x pixels from corrupted image and pad each u16 to i32
+        const __m128i d0 = xx_loadu_128(dat + j);
+        const __m128i d0h = _mm_cvtepu16_epi32(_mm_srli_si128(d0, 8));
+        const __m128i d0l = _mm_cvtepu16_epi32(d0);
+        // d0h, d0l = [7 6 5 4], [3 2 1 0] as u32 (indices of dat[])
+
+        // Load 8 pixels from the filtered image
+        const __m128i flth = xx_loadu_128(flt + j + 4);
+        const __m128i fltl = xx_loadu_128(flt + j);
+        // flth, fltl = [7 6 5 4], [3 2 1 0] as i32 (indices of flt+j)
+
+        const __m128i flth_xq = _mm_mullo_epi32(flth, xq_active);
+        const __m128i fltl_xq = _mm_mullo_epi32(fltl, xq_active);
+        const __m128i d0h_xq = _mm_mullo_epi32(d0h, xq_inactive);
+        const __m128i d0l_xq = _mm_mullo_epi32(d0l, xq_inactive);
+
+        const __m128i vh = _mm_add_epi32(flth_xq, d0h_xq);
+        const __m128i vl = _mm_add_epi32(fltl_xq, d0l_xq);
+        // vh = [ xq0(f[7]-d[7]) xq0(f[6]-d[6]) xq0(f[5]-d[5]) xq0(f[4]-d[4]) ]
+        // vl = [ xq0(f[3]-d[3]) xq0(f[2]-d[2]) xq0(f[1]-d[1]) xq0(f[0]-d[0]) ]
+
+        // Shift this down with appropriate rounding
+        const __m128i vrh = _mm_srai_epi32(_mm_add_epi32(vh, rounding), shift);
+        const __m128i vrl = _mm_srai_epi32(_mm_add_epi32(vl, rounding), shift);
+
+        // Saturate vr0 and vr1 from i32 to i16 then pack together
+        const __m128i vr = _mm_packs_epi32(vrl, vrh);
+
+        // Subtract twin-subspace-sgr filtered from source image to get error
+        const __m128i e0 = _mm_sub_epi16(_mm_add_epi16(vr, d0), s0);
+
+        // Calculate squared error and add adjacent values
+        const __m128i err0 = _mm_madd_epi16(e0, e0);
+
+        sum32 = _mm_add_epi32(sum32, err0);
+      }
+
+      const __m128i sum32l = _mm_cvtepu32_epi64(sum32);
+      sum64 = _mm_add_epi64(sum64, sum32l);
+      const __m128i sum32h = _mm_cvtepu32_epi64(_mm_srli_si128(sum32, 8));
+      sum64 = _mm_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels in this row (modulo 8)
+      for (k = j; k < width; ++k) {
+        const int32_t u = (int32_t)(dat[k] << SGRPROJ_RST_BITS);
+        int32_t v = xq_on * (flt[k] - u);
+        const int32_t e = ROUND_POWER_OF_TWO(v, shift) + dat[k] - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+      flt += flt_stride;
+    }
+  } else {  // Neither filter is enabled
+    for (i = 0; i < height; ++i) {
+      __m128i sum32 = _mm_setzero_si128();
+      for (j = 0; j <= width - 16; j += 16) {
+        // Load 2x8 u16 from source image
+        const __m128i s0 = xx_loadu_128(src + j);
+        const __m128i s1 = xx_loadu_128(src + j + 8);
+        // Load 2x8 u16 from corrupted image
+        const __m128i d0 = xx_loadu_128(dat + j);
+        const __m128i d1 = xx_loadu_128(dat + j + 8);
+
+        // Subtract corrupted image from source image
+        const __m128i diff0 = _mm_sub_epi16(d0, s0);
+        const __m128i diff1 = _mm_sub_epi16(d1, s1);
+
+        // Square error and add adjacent values
+        const __m128i err0 = _mm_madd_epi16(diff0, diff0);
+        const __m128i err1 = _mm_madd_epi16(diff1, diff1);
+
+        sum32 = _mm_add_epi32(sum32, err0);
+        sum32 = _mm_add_epi32(sum32, err1);
+      }
+
+      const __m128i sum32l = _mm_cvtepu32_epi64(sum32);
+      sum64 = _mm_add_epi64(sum64, sum32l);
+      const __m128i sum32h = _mm_cvtepu32_epi64(_mm_srli_si128(sum32, 8));
+      sum64 = _mm_add_epi64(sum64, sum32h);
+
+      // Process remaining pixels (modulu 8)
+      for (k = j; k < width; ++k) {
+        const int32_t e = (int32_t)(dat[k]) - src[k];
+        err += ((int64_t)e * e);
+      }
+      dat += dat_stride;
+      src += src_stride;
+    }
+  }
+
+  // Sum 4 values from sum64l and sum64h into err
+  int64_t sum[2];
+  xx_storeu_128(sum, sum64);
+  err += sum[0] + sum[1];
+  return err;
+}
diff --git a/third_party/aom/av1/encoder/x86/rdopt_avx2.c b/third_party/aom/av1/encoder/x86/rdopt_avx2.c
new file mode 100644
index 0000000000..f588badc7c
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/rdopt_avx2.c
@@ -0,0 +1,256 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+#include <immintrin.h>
+#include "aom_dsp/x86/synonyms_avx2.h"
+#include "aom_ports/system_state.h"
+
+#include "config/av1_rtcd.h"
+#include "av1/encoder/rdopt.h"
+
+// Process horizontal and vertical correlations in a 4x4 block of pixels.
+// We actually use the 4x4 pixels to calculate correlations corresponding to
+// the top-left 3x3 pixels, so this function must be called with 1x1 overlap,
+// moving the window along/down by 3 pixels at a time.
+INLINE static void horver_correlation_4x4(const int16_t *diff, int stride,
+                                          __m256i *xy_sum_32,
+                                          __m256i *xz_sum_32, __m256i *x_sum_32,
+                                          __m256i *x2_sum_32) {
+  // Pixels in this 4x4   [ a b c d ]
+  // are referred to as:  [ e f g h ]
+  //                      [ i j k l ]
+  //                      [ m n o p ]
+
+  const __m256i pixels = _mm256_set_epi64x(
+      *(uint64_t *)&diff[0 * stride], *(uint64_t *)&diff[1 * stride],
+      *(uint64_t *)&diff[2 * stride], *(uint64_t *)&diff[3 * stride]);
+  // pixels = [d c b a h g f e] [l k j i p o n m] as i16
+
+  const __m256i slli = _mm256_slli_epi64(pixels, 16);
+  // slli = [c b a 0 g f e 0] [k j i 0 o n m 0] as i16
+
+  const __m256i madd_xy = _mm256_madd_epi16(pixels, slli);
+  // madd_xy = [bc+cd ab fg+gh ef] [jk+kl ij no+op mn] as i32
+  *xy_sum_32 = _mm256_add_epi32(*xy_sum_32, madd_xy);
+
+  // Permute control [3 2] [1 0] => [2 1] [0 0], 0b10010000 = 0x90
+  const __m256i perm = _mm256_permute4x64_epi64(slli, 0x90);
+  // perm = [g f e 0 k j i 0] [o n m 0 o n m 0] as i16
+
+  const __m256i madd_xz = _mm256_madd_epi16(slli, perm);
+  // madd_xz = [cg+bf ae gk+fj ei] [ko+jn im oo+nn mm] as i32
+  *xz_sum_32 = _mm256_add_epi32(*xz_sum_32, madd_xz);
+
+  // Sum every element in slli (and then also their squares)
+  const __m256i madd1_slli = _mm256_madd_epi16(slli, _mm256_set1_epi16(1));
+  // madd1_slli = [c+b a g+f e] [k+j i o+n m] as i32
+  *x_sum_32 = _mm256_add_epi32(*x_sum_32, madd1_slli);
+
+  const __m256i madd_slli = _mm256_madd_epi16(slli, slli);
+  // madd_slli = [cc+bb aa gg+ff ee] [kk+jj ii oo+nn mm] as i32
+  *x2_sum_32 = _mm256_add_epi32(*x2_sum_32, madd_slli);
+}
+
+void av1_get_horver_correlation_full_avx2(const int16_t *diff, int stride,
+                                          int width, int height, float *hcorr,
+                                          float *vcorr) {
+  // The following notation is used:
+  // x - current pixel
+  // y - right neighbour pixel
+  // z - below neighbour pixel
+  // w - down-right neighbour pixel
+  int64_t xy_sum = 0, xz_sum = 0;
+  int64_t x_sum = 0, x2_sum = 0;
+
+  // Process horizontal and vertical correlations through the body in 4x4
+  // blocks.  This excludes the final row and column and possibly one extra
+  // column depending how 3 divides into width and height
+  int32_t xy_xz_tmp[8] = { 0 }, x_x2_tmp[8] = { 0 };
+  __m256i xy_sum_32 = _mm256_setzero_si256();
+  __m256i xz_sum_32 = _mm256_setzero_si256();
+  __m256i x_sum_32 = _mm256_setzero_si256();
+  __m256i x2_sum_32 = _mm256_setzero_si256();
+  for (int i = 0; i <= height - 4; i += 3) {
+    for (int j = 0; j <= width - 4; j += 3) {
+      horver_correlation_4x4(&diff[i * stride + j], stride, &xy_sum_32,
+                             &xz_sum_32, &x_sum_32, &x2_sum_32);
+    }
+    const __m256i hadd_xy_xz = _mm256_hadd_epi32(xy_sum_32, xz_sum_32);
+    // hadd_xy_xz = [ae+bf+cg ei+fj+gk ab+bc+cd ef+fg+gh]
+    //              [im+jn+ko mm+nn+oo ij+jk+kl mn+no+op] as i32
+    yy_storeu_256(xy_xz_tmp, hadd_xy_xz);
+    xy_sum += (int64_t)xy_xz_tmp[5] + xy_xz_tmp[4] + xy_xz_tmp[1];
+    xz_sum += (int64_t)xy_xz_tmp[7] + xy_xz_tmp[6] + xy_xz_tmp[3];
+
+    const __m256i hadd_x_x2 = _mm256_hadd_epi32(x_sum_32, x2_sum_32);
+    // hadd_x_x2 = [aa+bb+cc ee+ff+gg a+b+c e+f+g]
+    //             [ii+jj+kk mm+nn+oo i+j+k m+n+o] as i32
+    yy_storeu_256(x_x2_tmp, hadd_x_x2);
+    x_sum += (int64_t)x_x2_tmp[5] + x_x2_tmp[4] + x_x2_tmp[1];
+    x2_sum += (int64_t)x_x2_tmp[7] + x_x2_tmp[6] + x_x2_tmp[3];
+
+    xy_sum_32 = _mm256_setzero_si256();
+    xz_sum_32 = _mm256_setzero_si256();
+    x_sum_32 = _mm256_setzero_si256();
+    x2_sum_32 = _mm256_setzero_si256();
+  }
+
+  // x_sum now covers every pixel except the final 1-2 rows and 1-2 cols
+  int64_t x_finalrow = 0, x_finalcol = 0, x2_finalrow = 0, x2_finalcol = 0;
+
+  // Do we have 2 rows remaining or just the one?  Note that width and height
+  // are powers of 2, so each modulo 3 must be 1 or 2.
+  if (height % 3 == 1) {  // Just horiz corrs on the final row
+    const int16_t x0 = diff[(height - 1) * stride];
+    x_sum += x0;
+    x_finalrow += x0;
+    x2_sum += x0 * x0;
+    x2_finalrow += x0 * x0;
+    for (int j = 0; j < width - 1; ++j) {
+      const int16_t x = diff[(height - 1) * stride + j];
+      const int16_t y = diff[(height - 1) * stride + j + 1];
+      xy_sum += x * y;
+      x_sum += y;
+      x2_sum += y * y;
+      x_finalrow += y;
+      x2_finalrow += y * y;
+    }
+  } else {  // Two rows remaining to do
+    const int16_t x0 = diff[(height - 2) * stride];
+    const int16_t z0 = diff[(height - 1) * stride];
+    x_sum += x0 + z0;
+    x2_sum += x0 * x0 + z0 * z0;
+    x_finalrow += z0;
+    x2_finalrow += z0 * z0;
+    for (int j = 0; j < width - 1; ++j) {
+      const int16_t x = diff[(height - 2) * stride + j];
+      const int16_t y = diff[(height - 2) * stride + j + 1];
+      const int16_t z = diff[(height - 1) * stride + j];
+      const int16_t w = diff[(height - 1) * stride + j + 1];
+
+      // Horizontal and vertical correlations for the penultimate row:
+      xy_sum += x * y;
+      xz_sum += x * z;
+
+      // Now just horizontal correlations for the final row:
+      xy_sum += z * w;
+
+      x_sum += y + w;
+      x2_sum += y * y + w * w;
+      x_finalrow += w;
+      x2_finalrow += w * w;
+    }
+  }
+
+  // Do we have 2 columns remaining or just the one?
+  if (width % 3 == 1) {  // Just vert corrs on the final col
+    const int16_t x0 = diff[width - 1];
+    x_sum += x0;
+    x_finalcol += x0;
+    x2_sum += x0 * x0;
+    x2_finalcol += x0 * x0;
+    for (int i = 0; i < height - 1; ++i) {
+      const int16_t x = diff[i * stride + width - 1];
+      const int16_t z = diff[(i + 1) * stride + width - 1];
+      xz_sum += x * z;
+      x_finalcol += z;
+      x2_finalcol += z * z;
+      // So the bottom-right elements don't get counted twice:
+      if (i < height - (height % 3 == 1 ? 2 : 3)) {
+        x_sum += z;
+        x2_sum += z * z;
+      }
+    }
+  } else {  // Two cols remaining
+    const int16_t x0 = diff[width - 2];
+    const int16_t y0 = diff[width - 1];
+    x_sum += x0 + y0;
+    x2_sum += x0 * x0 + y0 * y0;
+    x_finalcol += y0;
+    x2_finalcol += y0 * y0;
+    for (int i = 0; i < height - 1; ++i) {
+      const int16_t x = diff[i * stride + width - 2];
+      const int16_t y = diff[i * stride + width - 1];
+      const int16_t z = diff[(i + 1) * stride + width - 2];
+      const int16_t w = diff[(i + 1) * stride + width - 1];
+
+      // Horizontal and vertical correlations for the penultimate col:
+      // Skip these on the last iteration of this loop if we also had two
+      // rows remaining, otherwise the final horizontal and vertical correlation
+      // get erroneously processed twice
+      if (i < height - 2 || height % 3 == 1) {
+        xy_sum += x * y;
+        xz_sum += x * z;
+      }
+
+      x_finalcol += w;
+      x2_finalcol += w * w;
+      // So the bottom-right elements don't get counted twice:
+      if (i < height - (height % 3 == 1 ? 2 : 3)) {
+        x_sum += z + w;
+        x2_sum += z * z + w * w;
+      }
+
+      // Now just vertical correlations for the final column:
+      xz_sum += y * w;
+    }
+  }
+
+  // Calculate the simple sums and squared-sums
+  int64_t x_firstrow = 0, x_firstcol = 0;
+  int64_t x2_firstrow = 0, x2_firstcol = 0;
+
+  for (int j = 0; j < width; ++j) {
+    x_firstrow += diff[j];
+    x2_firstrow += diff[j] * diff[j];
+  }
+  for (int i = 0; i < height; ++i) {
+    x_firstcol += diff[i * stride];
+    x2_firstcol += diff[i * stride] * diff[i * stride];
+  }
+
+  int64_t xhor_sum = x_sum - x_finalcol;
+  int64_t xver_sum = x_sum - x_finalrow;
+  int64_t y_sum = x_sum - x_firstcol;
+  int64_t z_sum = x_sum - x_firstrow;
+  int64_t x2hor_sum = x2_sum - x2_finalcol;
+  int64_t x2ver_sum = x2_sum - x2_finalrow;
+  int64_t y2_sum = x2_sum - x2_firstcol;
+  int64_t z2_sum = x2_sum - x2_firstrow;
+
+  aom_clear_system_state();
+
+  const float num_hor = (float)(height * (width - 1));
+  const float num_ver = (float)((height - 1) * width);
+
+  const float xhor_var_n = x2hor_sum - (xhor_sum * xhor_sum) / num_hor;
+  const float xver_var_n = x2ver_sum - (xver_sum * xver_sum) / num_ver;
+
+  const float y_var_n = y2_sum - (y_sum * y_sum) / num_hor;
+  const float z_var_n = z2_sum - (z_sum * z_sum) / num_ver;
+
+  const float xy_var_n = xy_sum - (xhor_sum * y_sum) / num_hor;
+  const float xz_var_n = xz_sum - (xver_sum * z_sum) / num_ver;
+
+  if (xhor_var_n > 0 && y_var_n > 0) {
+    *hcorr = xy_var_n / sqrtf(xhor_var_n * y_var_n);
+    *hcorr = *hcorr < 0 ? 0 : *hcorr;
+  } else {
+    *hcorr = 1.0;
+  }
+  if (xver_var_n > 0 && z_var_n > 0) {
+    *vcorr = xz_var_n / sqrtf(xver_var_n * z_var_n);
+    *vcorr = *vcorr < 0 ? 0 : *vcorr;
+  } else {
+    *vcorr = 1.0;
+  }
+}
diff --git a/third_party/aom/av1/encoder/x86/rdopt_sse4.c b/third_party/aom/av1/encoder/x86/rdopt_sse4.c
new file mode 100644
index 0000000000..67d94b4ca8
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/rdopt_sse4.c
@@ -0,0 +1,275 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+#include <emmintrin.h>
+#include "aom_dsp/x86/synonyms.h"
+#include "aom_ports/system_state.h"
+
+#include "config/av1_rtcd.h"
+#include "av1/encoder/rdopt.h"
+
+// Process horizontal and vertical correlations in a 4x4 block of pixels.
+// We actually use the 4x4 pixels to calculate correlations corresponding to
+// the top-left 3x3 pixels, so this function must be called with 1x1 overlap,
+// moving the window along/down by 3 pixels at a time.
+INLINE static void horver_correlation_4x4(const int16_t *diff, int stride,
+                                          __m128i *xy_sum_32,
+                                          __m128i *xz_sum_32, __m128i *x_sum_32,
+                                          __m128i *x2_sum_32) {
+  // Pixels in this 4x4   [ a b c d ]
+  // are referred to as:  [ e f g h ]
+  //                      [ i j k l ]
+  //                      [ m n o p ]
+
+  const __m128i pixelsa = _mm_set_epi64x(*(uint64_t *)&diff[0 * stride],
+                                         *(uint64_t *)&diff[2 * stride]);
+  const __m128i pixelsb = _mm_set_epi64x(*(uint64_t *)&diff[1 * stride],
+                                         *(uint64_t *)&diff[3 * stride]);
+  // pixelsa = [d c b a l k j i] as i16
+  // pixelsb = [h g f e p o n m] as i16
+
+  const __m128i slli_a = _mm_slli_epi64(pixelsa, 16);
+  const __m128i slli_b = _mm_slli_epi64(pixelsb, 16);
+  // slli_a = [c b a 0 k j i 0] as i16
+  // slli_b = [g f e 0 o n m 0] as i16
+
+  const __m128i xy_madd_a = _mm_madd_epi16(pixelsa, slli_a);
+  const __m128i xy_madd_b = _mm_madd_epi16(pixelsb, slli_b);
+  // xy_madd_a = [bc+cd ab jk+kl ij] as i32
+  // xy_madd_b = [fg+gh ef no+op mn] as i32
+
+  const __m128i xy32 = _mm_hadd_epi32(xy_madd_b, xy_madd_a);
+  // xy32 = [ab+bc+cd ij+jk+kl ef+fg+gh mn+no+op] as i32
+  *xy_sum_32 = _mm_add_epi32(*xy_sum_32, xy32);
+
+  const __m128i xz_madd_a = _mm_madd_epi16(slli_a, slli_b);
+  // xz_madd_a = [bf+cg ae jn+ko im] i32
+
+  const __m128i swap_b = _mm_srli_si128(slli_b, 8);
+  // swap_b = [0 0 0 0 g f e 0] as i16
+  const __m128i xz_madd_b = _mm_madd_epi16(slli_a, swap_b);
+  // xz_madd_b = [0 0 gk+fj ei] i32
+
+  const __m128i xz32 = _mm_hadd_epi32(xz_madd_b, xz_madd_a);
+  // xz32 = [ae+bf+cg im+jn+ko 0 ei+fj+gk] i32
+  *xz_sum_32 = _mm_add_epi32(*xz_sum_32, xz32);
+
+  // Now calculate the straight sums, x_sum += a+b+c+e+f+g+i+j+k
+  // (sum up every element in slli_a and swap_b)
+  const __m128i sum_slli_a = _mm_hadd_epi16(slli_a, slli_a);
+  const __m128i sum_slli_a32 = _mm_cvtepi16_epi32(sum_slli_a);
+  // sum_slli_a32 = [c+b a k+j i] as i32
+  const __m128i swap_b32 = _mm_cvtepi16_epi32(swap_b);
+  // swap_b32 = [g f e 0] as i32
+  *x_sum_32 = _mm_add_epi32(*x_sum_32, sum_slli_a32);
+  *x_sum_32 = _mm_add_epi32(*x_sum_32, swap_b32);
+  // sum = [c+b+g a+f k+j+e i] as i32
+
+  // Also sum their squares
+  const __m128i slli_a_2 = _mm_madd_epi16(slli_a, slli_a);
+  const __m128i swap_b_2 = _mm_madd_epi16(swap_b, swap_b);
+  // slli_a_2 = [c2+b2 a2 k2+j2 i2]
+  // swap_b_2 = [0 0 g2+f2 e2]
+  const __m128i sum2 = _mm_hadd_epi32(slli_a_2, swap_b_2);
+  // sum2 = [0 g2+f2+e2 c2+b2+a2 k2+j2+i2]
+  *x2_sum_32 = _mm_add_epi32(*x2_sum_32, sum2);
+}
+
+void av1_get_horver_correlation_full_sse4_1(const int16_t *diff, int stride,
+                                            int width, int height, float *hcorr,
+                                            float *vcorr) {
+  // The following notation is used:
+  // x - current pixel
+  // y - right neighbour pixel
+  // z - below neighbour pixel
+  // w - down-right neighbour pixel
+  int64_t xy_sum = 0, xz_sum = 0;
+  int64_t x_sum = 0, x2_sum = 0;
+
+  // Process horizontal and vertical correlations through the body in 4x4
+  // blocks.  This excludes the final row and column and possibly one extra
+  // column depending how 3 divides into width and height
+  int32_t xy_tmp[4] = { 0 }, xz_tmp[4] = { 0 };
+  int32_t x_tmp[4] = { 0 }, x2_tmp[4] = { 0 };
+  __m128i xy_sum_32 = _mm_setzero_si128();
+  __m128i xz_sum_32 = _mm_setzero_si128();
+  __m128i x_sum_32 = _mm_setzero_si128();
+  __m128i x2_sum_32 = _mm_setzero_si128();
+  for (int i = 0; i <= height - 4; i += 3) {
+    for (int j = 0; j <= width - 4; j += 3) {
+      horver_correlation_4x4(&diff[i * stride + j], stride, &xy_sum_32,
+                             &xz_sum_32, &x_sum_32, &x2_sum_32);
+    }
+    xx_storeu_128(xy_tmp, xy_sum_32);
+    xx_storeu_128(xz_tmp, xz_sum_32);
+    xx_storeu_128(x_tmp, x_sum_32);
+    xx_storeu_128(x2_tmp, x2_sum_32);
+    xy_sum += (int64_t)xy_tmp[3] + xy_tmp[2] + xy_tmp[1];
+    xz_sum += (int64_t)xz_tmp[3] + xz_tmp[2] + xz_tmp[0];
+    x_sum += (int64_t)x_tmp[3] + x_tmp[2] + x_tmp[1] + x_tmp[0];
+    x2_sum += (int64_t)x2_tmp[2] + x2_tmp[1] + x2_tmp[0];
+    xy_sum_32 = _mm_setzero_si128();
+    xz_sum_32 = _mm_setzero_si128();
+    x_sum_32 = _mm_setzero_si128();
+    x2_sum_32 = _mm_setzero_si128();
+  }
+
+  // x_sum now covers every pixel except the final 1-2 rows and 1-2 cols
+  int64_t x_finalrow = 0, x_finalcol = 0, x2_finalrow = 0, x2_finalcol = 0;
+
+  // Do we have 2 rows remaining or just the one?  Note that width and height
+  // are powers of 2, so each modulo 3 must be 1 or 2.
+  if (height % 3 == 1) {  // Just horiz corrs on the final row
+    const int16_t x0 = diff[(height - 1) * stride];
+    x_sum += x0;
+    x_finalrow += x0;
+    x2_sum += x0 * x0;
+    x2_finalrow += x0 * x0;
+    for (int j = 0; j < width - 1; ++j) {
+      const int16_t x = diff[(height - 1) * stride + j];
+      const int16_t y = diff[(height - 1) * stride + j + 1];
+      xy_sum += x * y;
+      x_sum += y;
+      x2_sum += y * y;
+      x_finalrow += y;
+      x2_finalrow += y * y;
+    }
+  } else {  // Two rows remaining to do
+    const int16_t x0 = diff[(height - 2) * stride];
+    const int16_t z0 = diff[(height - 1) * stride];
+    x_sum += x0 + z0;
+    x2_sum += x0 * x0 + z0 * z0;
+    x_finalrow += z0;
+    x2_finalrow += z0 * z0;
+    for (int j = 0; j < width - 1; ++j) {
+      const int16_t x = diff[(height - 2) * stride + j];
+      const int16_t y = diff[(height - 2) * stride + j + 1];
+      const int16_t z = diff[(height - 1) * stride + j];
+      const int16_t w = diff[(height - 1) * stride + j + 1];
+
+      // Horizontal and vertical correlations for the penultimate row:
+      xy_sum += x * y;
+      xz_sum += x * z;
+
+      // Now just horizontal correlations for the final row:
+      xy_sum += z * w;
+
+      x_sum += y + w;
+      x2_sum += y * y + w * w;
+      x_finalrow += w;
+      x2_finalrow += w * w;
+    }
+  }
+
+  // Do we have 2 columns remaining or just the one?
+  if (width % 3 == 1) {  // Just vert corrs on the final col
+    const int16_t x0 = diff[width - 1];
+    x_sum += x0;
+    x_finalcol += x0;
+    x2_sum += x0 * x0;
+    x2_finalcol += x0 * x0;
+    for (int i = 0; i < height - 1; ++i) {
+      const int16_t x = diff[i * stride + width - 1];
+      const int16_t z = diff[(i + 1) * stride + width - 1];
+      xz_sum += x * z;
+      x_finalcol += z;
+      x2_finalcol += z * z;
+      // So the bottom-right elements don't get counted twice:
+      if (i < height - (height % 3 == 1 ? 2 : 3)) {
+        x_sum += z;
+        x2_sum += z * z;
+      }
+    }
+  } else {  // Two cols remaining
+    const int16_t x0 = diff[width - 2];
+    const int16_t y0 = diff[width - 1];
+    x_sum += x0 + y0;
+    x2_sum += x0 * x0 + y0 * y0;
+    x_finalcol += y0;
+    x2_finalcol += y0 * y0;
+    for (int i = 0; i < height - 1; ++i) {
+      const int16_t x = diff[i * stride + width - 2];
+      const int16_t y = diff[i * stride + width - 1];
+      const int16_t z = diff[(i + 1) * stride + width - 2];
+      const int16_t w = diff[(i + 1) * stride + width - 1];
+
+      // Horizontal and vertical correlations for the penultimate col:
+      // Skip these on the last iteration of this loop if we also had two
+      // rows remaining, otherwise the final horizontal and vertical correlation
+      // get erroneously processed twice
+      if (i < height - 2 || height % 3 == 1) {
+        xy_sum += x * y;
+        xz_sum += x * z;
+      }
+
+      x_finalcol += w;
+      x2_finalcol += w * w;
+      // So the bottom-right elements don't get counted twice:
+      if (i < height - (height % 3 == 1 ? 2 : 3)) {
+        x_sum += z + w;
+        x2_sum += z * z + w * w;
+      }
+
+      // Now just vertical correlations for the final column:
+      xz_sum += y * w;
+    }
+  }
+
+  // Calculate the simple sums and squared-sums
+  int64_t x_firstrow = 0, x_firstcol = 0;
+  int64_t x2_firstrow = 0, x2_firstcol = 0;
+
+  for (int j = 0; j < width; ++j) {
+    x_firstrow += diff[j];
+    x2_firstrow += diff[j] * diff[j];
+  }
+  for (int i = 0; i < height; ++i) {
+    x_firstcol += diff[i * stride];
+    x2_firstcol += diff[i * stride] * diff[i * stride];
+  }
+
+  int64_t xhor_sum = x_sum - x_finalcol;
+  int64_t xver_sum = x_sum - x_finalrow;
+  int64_t y_sum = x_sum - x_firstcol;
+  int64_t z_sum = x_sum - x_firstrow;
+  int64_t x2hor_sum = x2_sum - x2_finalcol;
+  int64_t x2ver_sum = x2_sum - x2_finalrow;
+  int64_t y2_sum = x2_sum - x2_firstcol;
+  int64_t z2_sum = x2_sum - x2_firstrow;
+
+  aom_clear_system_state();
+
+  const float num_hor = (float)(height * (width - 1));
+  const float num_ver = (float)((height - 1) * width);
+
+  const float xhor_var_n = x2hor_sum - (xhor_sum * xhor_sum) / num_hor;
+  const float xver_var_n = x2ver_sum - (xver_sum * xver_sum) / num_ver;
+
+  const float y_var_n = y2_sum - (y_sum * y_sum) / num_hor;
+  const float z_var_n = z2_sum - (z_sum * z_sum) / num_ver;
+
+  const float xy_var_n = xy_sum - (xhor_sum * y_sum) / num_hor;
+  const float xz_var_n = xz_sum - (xver_sum * z_sum) / num_ver;
+
+  if (xhor_var_n > 0 && y_var_n > 0) {
+    *hcorr = xy_var_n / sqrtf(xhor_var_n * y_var_n);
+    *hcorr = *hcorr < 0 ? 0 : *hcorr;
+  } else {
+    *hcorr = 1.0;
+  }
+  if (xver_var_n > 0 && z_var_n > 0) {
+    *vcorr = xz_var_n / sqrtf(xver_var_n * z_var_n);
+    *vcorr = *vcorr < 0 ? 0 : *vcorr;
+  } else {
+    *vcorr = 1.0;
+  }
+}
diff --git a/third_party/aom/av1/encoder/x86/temporal_filter_apply_sse2.asm b/third_party/aom/av1/encoder/x86/temporal_filter_apply_sse2.asm
index 30983d1c10..0cae3a748a 100644
--- a/third_party/aom/av1/encoder/x86/temporal_filter_apply_sse2.asm
+++ b/third_party/aom/av1/encoder/x86/temporal_filter_apply_sse2.asm
@@ -16,6 +16,7 @@
 
 SECTION .text
 
+; TODO(yunqing) The following optimization is not used since the c code changes.
 ; void av1_temporal_filter_apply_sse2 | arg
 ;  (unsigned char  *frame1,           |  0
 ;   unsigned int    stride,           |  1
diff --git a/third_party/aom/av1/exports_ident b/third_party/aom/av1/exports_ident
new file mode 100644
index 0000000000..b523a679d5
--- /dev/null
+++ b/third_party/aom/av1/exports_ident
@@ -0,0 +1,2 @@
+text ifd_init
+text ifd_inspect
diff --git a/third_party/aom/build/cmake/aom_config_defaults.cmake b/third_party/aom/build/cmake/aom_config_defaults.cmake
index a07438cfe3..47aba55af0 100644
--- a/third_party/aom/build/cmake/aom_config_defaults.cmake
+++ b/third_party/aom/build/cmake/aom_config_defaults.cmake
@@ -101,32 +101,22 @@ set_aom_config_var(CONFIG_DENOISE 1 NUMBER
                    "Denoise/noise modeling support in encoder.")
 set_aom_config_var(CONFIG_FILEOPTIONS 1 NUMBER
                    "Enables encoder config file support.")
-set_aom_config_var(CONFIG_FIX_GF_LENGTH 1 NUMBER
-                   "Fix the GF length if possible")
 set_aom_config_var(CONFIG_INSPECTION 0 NUMBER "Enables bitstream inspection.")
 set_aom_config_var(CONFIG_INTERNAL_STATS 0 NUMBER
                    "Enables internal encoder stats.")
-set_aom_config_var(CONFIG_LOWBITDEPTH 0 NUMBER
+set_aom_config_var(CONFIG_LOWBITDEPTH 1 NUMBER
                    "Enables 8-bit optimized pipeline.")
 set_aom_config_var(CONFIG_MAX_DECODE_PROFILE 2 NUMBER
                    "Max profile to support decoding.")
 set_aom_config_var(CONFIG_NORMAL_TILE_MODE 0 NUMBER
                    "Only enables normal tile mode.")
-set_aom_config_var(
-  CONFIG_REDUCED_ENCODER_BORDER 0 NUMBER
-  "Enable reduced border extention for encoder. \
-                    Disables superres and resize support."
-  )
 set_aom_config_var(CONFIG_SIZE_LIMIT 0 NUMBER "Limit max decode width/height.")
 set_aom_config_var(CONFIG_SPATIAL_RESAMPLING 1 NUMBER "Spatial resampling.")
 set_aom_config_var(DECODE_HEIGHT_LIMIT 0 NUMBER "Set limit for decode height.")
 set_aom_config_var(DECODE_WIDTH_LIMIT 0 NUMBER "Set limit for decode width.")
-set_aom_config_var(CONFIG_GLOBAL_MOTION_SEARCH 1 NUMBER
-                   "Global motion search flag.")
 
 # AV1 experiment flags.
-set_aom_config_var(CONFIG_COLLECT_INTER_MODE_RD_STATS 1 NUMBER
-                   "AV1 experiment flag.")
+set_aom_config_var(CONFIG_SPEED_STATS 0 NUMBER "AV1 experiment flag.")
 set_aom_config_var(CONFIG_COLLECT_RD_STATS 0 NUMBER "AV1 experiment flag.")
 set_aom_config_var(CONFIG_DIST_8X8 0 NUMBER "AV1 experiment flag.")
 set_aom_config_var(CONFIG_ENTROPY_STATS 0 NUMBER "AV1 experiment flag.")
@@ -135,8 +125,12 @@ set_aom_config_var(CONFIG_INTER_STATS_ONLY 0 NUMBER "AV1 experiment flag.")
 set_aom_config_var(CONFIG_RD_DEBUG 0 NUMBER "AV1 experiment flag.")
 set_aom_config_var(CONFIG_2PASS_PARTITION_SEARCH_LVL 1 NUMBER
                    "AV1 experiment flag.")
-set_aom_config_var(CONFIG_SHARP_SETTINGS 0 NUMBER
-                   "Use sharper encoding settings")
+set_aom_config_var(CONFIG_SHARP_SETTINGS 0 NUMBER "AV1 experiment flag.")
+set_aom_config_var(CONFIG_ONE_PASS_SVM 0 NUMBER "AV1 experiment flag.")
+set_aom_config_var(CONFIG_DISABLE_FULL_PIXEL_SPLIT_8X8 1 NUMBER
+                   "Disable full_pixel_motion_search_based_split on BLOCK_8X8.")
+set_aom_config_var(CONFIG_COLLECT_PARTITION_STATS 0 NUMBER
+                   "Collect stats on partition decisions.")
 
 #
 # Variables in this section control optional features of the build system.
diff --git a/third_party/aom/build/cmake/aom_configure.cmake b/third_party/aom/build/cmake/aom_configure.cmake
index c0c7381e87..cb64713e58 100644
--- a/third_party/aom/build/cmake/aom_configure.cmake
+++ b/third_party/aom/build/cmake/aom_configure.cmake
@@ -17,9 +17,6 @@ include(FindGit)
 include(FindPerl)
 include(FindThreads)
 
-set(AOM_SUPPORTED_CPU_TARGETS
-    "arm64 armv7 armv7s generic mips32 mips64 ppc x86 x86_64")
-
 include("${AOM_ROOT}/build/cmake/aom_config_defaults.cmake")
 include("${AOM_ROOT}/build/cmake/aom_experiment_deps.cmake")
 include("${AOM_ROOT}/build/cmake/aom_optimization.cmake")
@@ -70,8 +67,12 @@ if(NOT AOM_TARGET_CPU)
 endif()
 
 if(CMAKE_TOOLCHAIN_FILE) # Add toolchain file to config string.
-  file(RELATIVE_PATH toolchain_path "${AOM_CONFIG_DIR}"
-                     "${CMAKE_TOOLCHAIN_FILE}")
+  if(IS_ABSOLUTE "${CMAKE_TOOLCHAIN_FILE}")
+    file(RELATIVE_PATH toolchain_path "${AOM_CONFIG_DIR}"
+                       "${CMAKE_TOOLCHAIN_FILE}")
+  else()
+    set(toolchain_path "${CMAKE_TOOLCHAIN_FILE}")
+  endif()
   set(toolchain_string "-DCMAKE_TOOLCHAIN_FILE=\\\"${toolchain_path}\\\"")
   set(AOM_CMAKE_CONFIG "${toolchain_string} ${AOM_CMAKE_CONFIG}")
 else()
@@ -104,7 +105,7 @@ if(NOT MSVC)
     # work.
     set(CMAKE_POSITION_INDEPENDENT_CODE ON)
     if("${AOM_TARGET_SYSTEM}" STREQUAL "Linux" AND "${AOM_TARGET_CPU}" MATCHES
-       "^armv7")
+       "^armv[78]")
       set(AOM_AS_FLAGS ${AOM_AS_FLAGS} --defsym PIC=1)
     else()
       set(AOM_AS_FLAGS ${AOM_AS_FLAGS} -DPIC)
@@ -112,13 +113,6 @@ if(NOT MSVC)
   endif()
 endif()
 
-if(NOT "${AOM_SUPPORTED_CPU_TARGETS}" MATCHES "${AOM_TARGET_CPU}")
-  message(FATAL_ERROR
-            "No RTCD support for ${AOM_TARGET_CPU}. Create it, or "
-            "add -DAOM_TARGET_CPU=generic to your cmake command line for a "
-            "generic build of libaom and tools.")
-endif()
-
 if("${AOM_TARGET_CPU}" STREQUAL "x86" OR "${AOM_TARGET_CPU}" STREQUAL "x86_64")
   find_program(AS_EXECUTABLE yasm $ENV{YASM_PATH})
   if(NOT AS_EXECUTABLE OR ENABLE_NASM)
@@ -184,6 +178,12 @@ if("${AOM_TARGET_SYSTEM}" MATCHES "Darwin\|Linux\|Windows")
   set(CONFIG_OS_SUPPORT 1)
 endif()
 
+# The default _WIN32_WINNT value in MinGW is 0x0502 (Windows XP with SP2). Set
+# it to 0x0601 (Windows 7).
+if("${AOM_TARGET_SYSTEM}" STREQUAL "Windows")
+  add_compiler_flag_if_supported("-D_WIN32_WINNT=0x0601")
+endif()
+
 #
 # Fix CONFIG_* dependencies. This must be done before including cpu.cmake to
 # ensure RTCD_CONFIG_* are properly set.
@@ -252,6 +252,7 @@ if(MSVC)
   endif()
 else()
   require_c_flag("-std=c99" YES)
+  require_cxx_flag_nomsvc("-std=c++11" YES)
   add_compiler_flag_if_supported("-Wall")
   add_compiler_flag_if_supported("-Wdisabled-optimization")
   add_compiler_flag_if_supported("-Wextra")
@@ -266,8 +267,17 @@ else()
   add_compiler_flag_if_supported("-Wunused")
   add_compiler_flag_if_supported("-Wvla")
 
-  add_c_flag_if_supported("-Wstack-usage=100000")
-  add_cxx_flag_if_supported("-Wstack-usage=360000")
+  if(CMAKE_C_COMPILER_ID MATCHES "GNU" AND "${SANITIZE}" MATCHES
+     "address|undefined")
+
+    # This combination has more stack overhead, so we account for it by
+    # providing higher stack limit than usual.
+    add_c_flag_if_supported("-Wstack-usage=170000")
+    add_cxx_flag_if_supported("-Wstack-usage=270000")
+  else()
+    add_c_flag_if_supported("-Wstack-usage=100000")
+    add_cxx_flag_if_supported("-Wstack-usage=240000")
+  endif()
 
   # TODO(jzern): this could be added as a cxx flags for test/*.cc only, avoiding
   # third_party.
@@ -365,13 +375,3 @@ execute_process(COMMAND ${CMAKE_COMMAND} -DAOM_CONFIG_DIR=${AOM_CONFIG_DIR}
                         -DGIT_EXECUTABLE=${GIT_EXECUTABLE}
                         -DPERL_EXECUTABLE=${PERL_EXECUTABLE} -P
                         "${AOM_ROOT}/build/cmake/version.cmake")
-
-if(NOT MSVC) # Generate aom.pc (pkg-config file).
-  execute_process(COMMAND ${CMAKE_COMMAND} -DAOM_CONFIG_DIR=${AOM_CONFIG_DIR}
-                          -DAOM_ROOT=${AOM_ROOT}
-                          -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}
-                          -DCMAKE_PROJECT_NAME=${CMAKE_PROJECT_NAME}
-                          -DCONFIG_MULTITHREAD=${CONFIG_MULTITHREAD}
-                          -DHAVE_PTHREAD_H=${HAVE_PTHREAD_H} -P
-                          "${AOM_ROOT}/build/cmake/pkg_config.cmake")
-endif()
diff --git a/third_party/aom/build/cmake/aom_experiment_deps.cmake b/third_party/aom/build/cmake/aom_experiment_deps.cmake
index 0688704e5f..2e3615791e 100644
--- a/third_party/aom/build/cmake/aom_experiment_deps.cmake
+++ b/third_party/aom/build/cmake/aom_experiment_deps.cmake
@@ -21,10 +21,6 @@ macro(fix_experiment_configs)
     change_config_and_warn(CONFIG_INSPECTION 1 CONFIG_ANALYZER)
   endif()
 
-  if(CONFIG_RD_DEBUG)
-    change_config_and_warn(CONFIG_RD_DEBUG 0 CONFIG_JNT_COMP)
-  endif()
-
   if(CONFIG_DIST_8X8 AND CONFIG_MULTITHREAD)
     change_config_and_warn(CONFIG_DIST_8X8 0 CONFIG_MULTITHREAD)
   endif()
diff --git a/third_party/aom/build/cmake/aom_install.cmake b/third_party/aom/build/cmake/aom_install.cmake
new file mode 100644
index 0000000000..47206d8bfb
--- /dev/null
+++ b/third_party/aom/build/cmake/aom_install.cmake
@@ -0,0 +1,90 @@
+#
+# Copyright (c) 2018, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and the
+# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was
+# not distributed with this source code in the LICENSE file, you can obtain it
+# at www.aomedia.org/license/software. If the Alliance for Open Media Patent
+# License 1.0 was not distributed with this source code in the PATENTS file, you
+# can obtain it at www.aomedia.org/license/patent.
+#
+list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aom.h"
+            "${AOM_ROOT}/aom/aom_codec.h" "${AOM_ROOT}/aom/aom_frame_buffer.h"
+            "${AOM_ROOT}/aom/aom_image.h" "${AOM_ROOT}/aom/aom_integer.h"
+            "${AOM_ROOT}/aom/aom.h")
+
+if(CONFIG_AV1_DECODER)
+  list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aom_decoder.h"
+              "${AOM_ROOT}/aom/aomdx.h")
+endif()
+
+if(CONFIG_AV1_ENCODER)
+  list(APPEND AOM_INSTALL_INCS "${AOM_ROOT}/aom/aomcx.h"
+              "${AOM_ROOT}/aom/aom_encoder.h")
+endif()
+
+# Generate aom.pc and setup dependencies to ensure it is created when necessary.
+# Note: aom.pc generation uses GNUInstallDirs:
+# https://cmake.org/cmake/help/latest/module/GNUInstallDirs.html
+macro(setup_aom_install_targets)
+  if(NOT (MSVC OR XCODE))
+    include("GNUInstallDirs")
+    set(AOM_PKG_CONFIG_FILE "${AOM_CONFIG_DIR}/aom.pc")
+
+    # Create a dummy library target for creating aom.pc.
+    create_dummy_source_file(aom_pc c AOM_PKG_CONFIG_SOURCES)
+    add_library(aom_pc ${AOM_PKG_CONFIG_SOURCES})
+
+    # Setup a rule to generate aom.pc.
+    add_custom_command(OUTPUT "${AOM_PKG_CONFIG_FILE}"
+                       COMMAND
+                         ${CMAKE_COMMAND} ARGS
+                         -DAOM_CONFIG_DIR=${AOM_CONFIG_DIR}
+                         -DAOM_ROOT=${AOM_ROOT}
+                         -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}
+                         -DCMAKE_INSTALL_BINDIR=${CMAKE_INSTALL_BINDIR}
+                         -DCMAKE_INSTALL_INCLUDEDIR=${CMAKE_INSTALL_INCLUDEDIR}
+                         -DCMAKE_INSTALL_LIBDIR=${CMAKE_INSTALL_LIBDIR}
+                         -DCMAKE_PROJECT_NAME=${CMAKE_PROJECT_NAME}
+                         -DCONFIG_MULTITHREAD=${CONFIG_MULTITHREAD}
+                         -DHAVE_PTHREAD_H=${HAVE_PTHREAD_H} -P
+                         "${AOM_ROOT}/build/cmake/pkg_config.cmake"
+                       COMMENT "Writing aom.pc" VERBATIM)
+
+    # Explicitly add a dependency on the pkg-config file to ensure it's built.
+    get_property(aom_pc_sources TARGET aom_pc PROPERTY SOURCES)
+    set_source_files_properties(${aom_pc_sources} OBJECT_DEPENDS
+                                "${AOM_PKG_CONFIG_FILE}")
+
+    # Our pkg-config file carries version information: add a dependency on the
+    # version rule.
+    add_dependencies(aom_pc aom_version)
+
+    if(CONFIG_AV1_DECODER)
+      if(ENABLE_EXAMPLES)
+        list(APPEND AOM_INSTALL_BINS aomdec)
+      endif()
+    endif()
+
+    if(CONFIG_AV1_ENCODER)
+      if(ENABLE_EXAMPLES)
+        list(APPEND AOM_INSTALL_BINS aomenc)
+      endif()
+    endif()
+
+    set(AOM_INSTALL_LIBS aom)
+
+    # Setup the install rules.
+    install(FILES ${AOM_INSTALL_INCS} DESTINATION
+                  "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_INCLUDEDIR}/aom")
+    install(FILES "${AOM_PKG_CONFIG_FILE}" DESTINATION
+                  "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}/pkgconfig")
+    install(TARGETS ${AOM_INSTALL_LIBS} DESTINATION
+                    "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}")
+
+    if(ENABLE_EXAMPLES)
+      install(TARGETS ${AOM_INSTALL_BINS} DESTINATION
+                      "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}")
+    endif()
+  endif()
+endmacro()
diff --git a/third_party/aom/build/cmake/cpu.cmake b/third_party/aom/build/cmake/cpu.cmake
index 6e8089e635..ef2d7552bb 100644
--- a/third_party/aom/build/cmake/cpu.cmake
+++ b/third_party/aom/build/cmake/cpu.cmake
@@ -9,18 +9,7 @@
 # can obtain it at www.aomedia.org/license/patent.
 #
 
-if("${AOM_TARGET_CPU}" STREQUAL "arm64")
-  set(ARCH_ARM 1)
-  set(RTCD_ARCH_ARM "yes")
-
-  if(ENABLE_NEON)
-    set(HAVE_NEON 1)
-    set(RTCD_HAVE_NEON "yes")
-  else()
-    set(HAVE_NEON 0)
-    set(AOM_RTCD_FLAGS ${AOM_RTCD_FLAGS} --disable-neon)
-  endif()
-elseif("${AOM_TARGET_CPU}" MATCHES "^armv7")
+if("${AOM_TARGET_CPU}" MATCHES "^arm")
   set(ARCH_ARM 1)
   set(RTCD_ARCH_ARM "yes")
 
diff --git a/third_party/aom/build/cmake/exports.cmake b/third_party/aom/build/cmake/exports.cmake
index e0813dc0f2..b6e14d917b 100644
--- a/third_party/aom/build/cmake/exports.cmake
+++ b/third_party/aom/build/cmake/exports.cmake
@@ -35,6 +35,7 @@ function(setup_exports_target)
                             -DAOM_XCODE=${XCODE} -DCONFIG_NAME=$<CONFIG>
                             -DCONFIG_AV1_DECODER=${CONFIG_AV1_DECODER}
                             -DCONFIG_AV1_ENCODER=${CONFIG_AV1_ENCODER}
+                            -DCONFIG_INSPECTION=${CONFIG_INSPECTION}
                             -DENABLE_TESTS=${ENABLE_TESTS} -P
                             "${AOM_ROOT}/build/cmake/generate_exports.cmake"
                     SOURCES ${AOM_EXPORTS_SOURCES}
diff --git a/third_party/aom/build/cmake/exports_sources.cmake b/third_party/aom/build/cmake/exports_sources.cmake
index 576920e36c..46bf001d86 100644
--- a/third_party/aom/build/cmake/exports_sources.cmake
+++ b/third_party/aom/build/cmake/exports_sources.cmake
@@ -19,6 +19,9 @@ list(APPEND AOM_EXPORTS_SOURCES "${AOM_ROOT}/aom/exports_com"
 if(CONFIG_AV1_DECODER)
   list(APPEND AOM_EXPORTS_SOURCES "${AOM_ROOT}/aom/exports_dec"
               "${AOM_ROOT}/av1/exports_dec")
+  if(CONFIG_INSPECTION)
+    list(APPEND AOM_EXPORTS_SOURCES "${AOM_ROOT}/av1/exports_ident")
+  endif()
 endif()
 
 if(CONFIG_AV1_ENCODER)
diff --git a/third_party/aom/build/cmake/pkg_config.cmake b/third_party/aom/build/cmake/pkg_config.cmake
index 64e20214e9..358c1a20dd 100644
--- a/third_party/aom/build/cmake/pkg_config.cmake
+++ b/third_party/aom/build/cmake/pkg_config.cmake
@@ -11,6 +11,7 @@
 cmake_minimum_required(VERSION 3.5)
 
 set(REQUIRED_ARGS "AOM_ROOT" "AOM_CONFIG_DIR" "CMAKE_INSTALL_PREFIX"
+    "CMAKE_INSTALL_BINDIR" "CMAKE_INSTALL_INCLUDEDIR" "CMAKE_INSTALL_LIBDIR"
     "CMAKE_PROJECT_NAME" "CONFIG_MULTITHREAD" "HAVE_PTHREAD_H")
 
 foreach(arg ${REQUIRED_ARGS})
@@ -34,25 +35,28 @@ endif()
 
 # Write pkg-config info.
 set(prefix "${CMAKE_INSTALL_PREFIX}")
+set(bindir "${CMAKE_INSTALL_BINDIR}")
+set(includedir "${CMAKE_INSTALL_INCLUDEDIR}")
+set(libdir "${CMAKE_INSTALL_LIBDIR}")
 set(pkgconfig_file "${AOM_CONFIG_DIR}/aom.pc")
 string(TOLOWER ${CMAKE_PROJECT_NAME} pkg_name)
 file(WRITE "${pkgconfig_file}" "# libaom pkg-config.\n")
 file(APPEND "${pkgconfig_file}" "prefix=${prefix}\n")
-file(APPEND "${pkgconfig_file}" "exec_prefix=\${prefix}/bin\n")
-file(APPEND "${pkgconfig_file}" "libdir=\${prefix}/lib\n")
-file(APPEND "${pkgconfig_file}" "includedir=\${prefix}/include\n\n")
+file(APPEND "${pkgconfig_file}" "exec_prefix=\${prefix}\n")
+file(APPEND "${pkgconfig_file}" "includedir=\${prefix}/${includedir}\n")
+file(APPEND "${pkgconfig_file}" "libdir=\${exec_prefix}/${libdir}\n\n")
 file(APPEND "${pkgconfig_file}" "Name: ${pkg_name}\n")
-file(APPEND "${pkgconfig_file}"
-            "Description: AV1 codec library v${aom_version}.\n")
+file(
+  APPEND
+    "${pkgconfig_file}"
+    "Description: Alliance for Open Media AV1 codec library v${aom_version}.\n")
 file(APPEND "${pkgconfig_file}" "Version: ${package_version}\n")
 file(APPEND "${pkgconfig_file}" "Requires:\n")
 file(APPEND "${pkgconfig_file}" "Conflicts:\n")
+file(APPEND "${pkgconfig_file}" "Libs: -L\${libdir} -l${pkg_name}\n")
 if(CONFIG_MULTITHREAD AND HAVE_PTHREAD_H)
-  file(APPEND "${pkgconfig_file}"
-              "Libs: -L\${prefix}/lib -l${pkg_name} -lm -lpthread\n")
   file(APPEND "${pkgconfig_file}" "Libs.private: -lm -lpthread\n")
 else()
-  file(APPEND "${pkgconfig_file}" "Libs: -L\${prefix}/lib -l${pkg_name} -lm\n")
   file(APPEND "${pkgconfig_file}" "Libs.private: -lm\n")
 endif()
-file(APPEND "${pkgconfig_file}" "Cflags: -I\${prefix}/include\n")
+file(APPEND "${pkgconfig_file}" "Cflags: -I\${includedir}\n")
diff --git a/third_party/aom/build/cmake/rtcd.pl b/third_party/aom/build/cmake/rtcd.pl
index 46e06907cc..dafccdca9c 100755
--- a/third_party/aom/build/cmake/rtcd.pl
+++ b/third_party/aom/build/cmake/rtcd.pl
@@ -426,10 +426,10 @@ if ($opts{arch} eq 'x86') {
     @ALL_ARCHS = filter("$opts{arch}", qw/msa/);
   }
   mips;
-} elsif ($opts{arch} =~ /armv7\w?/) {
+} elsif ($opts{arch} =~ /armv[78]\w?/) {
   @ALL_ARCHS = filter(qw/neon/);
   arm;
-} elsif ($opts{arch} eq 'armv8' || $opts{arch} eq 'arm64' ) {
+} elsif ($opts{arch} eq 'arm64' ) {
   @ALL_ARCHS = filter(qw/neon/);
   &require("neon");
   arm;
diff --git a/third_party/aom/build/cmake/toolchains/arm64-mingw-gcc.cmake b/third_party/aom/build/cmake/toolchains/arm64-mingw-gcc.cmake
index b5b2ff1cd3..bfeac928b7 100644
--- a/third_party/aom/build/cmake/toolchains/arm64-mingw-gcc.cmake
+++ b/third_party/aom/build/cmake/toolchains/arm64-mingw-gcc.cmake
@@ -27,6 +27,3 @@ set(CMAKE_RANLIB ${CROSS}ranlib CACHE FILEPATH Indexer)
 
 # No runtime cpu detect for arm64-mingw-gcc.
 set(CONFIG_RUNTIME_CPU_DETECT 0 CACHE NUMBER "")
-
-# Disable the use of the gtest's CMake support.
-set(AOM_DISABLE_GTEST_CMAKE 1)
diff --git a/third_party/aom/build/cmake/toolchains/armv7-mingw-gcc.cmake b/third_party/aom/build/cmake/toolchains/armv7-mingw-gcc.cmake
index cf06a11b38..eb488ecc1b 100644
--- a/third_party/aom/build/cmake/toolchains/armv7-mingw-gcc.cmake
+++ b/third_party/aom/build/cmake/toolchains/armv7-mingw-gcc.cmake
@@ -27,6 +27,3 @@ set(CMAKE_RANLIB ${CROSS}ranlib CACHE FILEPATH Indexer)
 
 # No runtime cpu detect for armv7-mingw-gcc.
 set(CONFIG_RUNTIME_CPU_DETECT 0 CACHE NUMBER "")
-
-# Disable the use of the gtest's CMake support.
-set(AOM_DISABLE_GTEST_CMAKE 1)
diff --git a/third_party/aom/build/cmake/toolchains/x86-mingw-gcc.cmake b/third_party/aom/build/cmake/toolchains/x86-mingw-gcc.cmake
index c986c4ee36..4839c9d455 100644
--- a/third_party/aom/build/cmake/toolchains/x86-mingw-gcc.cmake
+++ b/third_party/aom/build/cmake/toolchains/x86-mingw-gcc.cmake
@@ -26,6 +26,3 @@ set(CMAKE_C_COMPILER ${CROSS}gcc)
 set(CMAKE_CXX_COMPILER ${CROSS}g++)
 set(CMAKE_AR ${CROSS}ar CACHE FILEPATH Archiver)
 set(CMAKE_RANLIB ${CROSS}ranlib CACHE FILEPATH Indexer)
-
-# Disable the use of the gtest's CMake support.
-set(AOM_DISABLE_GTEST_CMAKE 1)
diff --git a/third_party/aom/build/cmake/toolchains/x86_64-mingw-gcc.cmake b/third_party/aom/build/cmake/toolchains/x86_64-mingw-gcc.cmake
index 00d94d5f1f..4b2d28debc 100644
--- a/third_party/aom/build/cmake/toolchains/x86_64-mingw-gcc.cmake
+++ b/third_party/aom/build/cmake/toolchains/x86_64-mingw-gcc.cmake
@@ -24,6 +24,3 @@ set(CMAKE_C_COMPILER ${CROSS}gcc)
 set(CMAKE_CXX_COMPILER ${CROSS}g++)
 set(CMAKE_AR ${CROSS}ar CACHE FILEPATH Archiver)
 set(CMAKE_RANLIB ${CROSS}ranlib CACHE FILEPATH Indexer)
-
-# Disable the use of the gtest's CMake support.
-set(AOM_DISABLE_GTEST_CMAKE 1)
diff --git a/third_party/aom/build/cmake/util.cmake b/third_party/aom/build/cmake/util.cmake
index b70ec40137..53379410f1 100644
--- a/third_party/aom/build/cmake/util.cmake
+++ b/third_party/aom/build/cmake/util.cmake
@@ -16,10 +16,10 @@ set(AOM_BUILD_CMAKE_UTIL_CMAKE_ 1)
 # Directory where generated sources will be written.
 set(AOM_GEN_SRC_DIR "${AOM_CONFIG_DIR}/gen_src")
 
-# Creates dummy source file in $AOM_CONFIG_DIR named $basename.$extension and
-# returns the full path to the dummy source file via the $out_file_path
-# parameter.
-macro(create_dummy_source_file basename extension out_file_path)
+# Creates dummy source file in $AOM_GEN_SRC_DIR named $basename.$extension and
+# returns the full path to the dummy source file via appending it to the list
+# variable referred to by $out_file_list_var parameter.
+macro(create_dummy_source_file basename extension out_file_list_var)
   set(dummy_source_file "${AOM_GEN_SRC_DIR}/${basename}_dummy.${extension}")
   file(
     WRITE
@@ -27,13 +27,15 @@ macro(create_dummy_source_file basename extension out_file_path)
       "// ${target_name} needs a ${extension} file to force link language, \n"
       "// or to silence a harmless CMake warning: Ignore me.\n"
       "void ${target_name}_dummy_function(void) {}\n")
+  list(APPEND "${out_file_list_var}" "${dummy_source_file}")
 endmacro()
 
 # Convenience function for adding a dummy source file to $target_name using
 # $extension as the file extension. Wraps create_dummy_source_file().
 function(add_dummy_source_file_to_target target_name extension)
-  create_dummy_source_file("${target_name}" "${extension}" "dummy_source_file")
-  target_sources(${target_name} PRIVATE ${dummy_source_file})
+  create_dummy_source_file("${target_name}" "${extension}"
+                           "dummy_source_file_list")
+  target_sources(${target_name} PRIVATE ${dummy_source_file_list})
 endfunction()
 
 # Sets the value of the variable referenced by $feature to $value, and reports
diff --git a/third_party/aom/common/av1_config.c b/third_party/aom/common/av1_config.c
index e8decf76fe..90955fb5b3 100644
--- a/third_party/aom/common/av1_config.c
+++ b/third_party/aom/common/av1_config.c
@@ -322,7 +322,7 @@ static int parse_sequence_header(const uint8_t *const buffer, size_t length,
   AV1C_READ_BITS_OR_RETURN_ERROR(max_frame_height_minus_1,
                                  frame_height_bits_minus_1 + 1);
 
-  int frame_id_numbers_present = 0;
+  uint8_t frame_id_numbers_present = 0;
   if (!reduced_still_picture_header) {
     AV1C_READ_BIT_OR_RETURN_ERROR(frame_id_numbers_present_flag);
     frame_id_numbers_present = frame_id_numbers_present_flag;
@@ -345,7 +345,7 @@ static int parse_sequence_header(const uint8_t *const buffer, size_t length,
 
     AV1C_READ_BIT_OR_RETURN_ERROR(enable_order_hint);
     if (enable_order_hint) {
-      AV1C_READ_BIT_OR_RETURN_ERROR(enable_jnt_comp);
+      AV1C_READ_BIT_OR_RETURN_ERROR(enable_dist_wtd_comp);
       AV1C_READ_BIT_OR_RETURN_ERROR(enable_ref_frame_mvs);
     }
 
diff --git a/third_party/aom/common/obudec.c b/third_party/aom/common/obudec.c
index acbd12e0c8..bd9f98d3ae 100644
--- a/third_party/aom/common/obudec.c
+++ b/third_party/aom/common/obudec.c
@@ -12,9 +12,11 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <assert.h>
 
 #include "common/obudec.h"
 
+#include "aom_dsp/aom_dsp_common.h"
 #include "aom_ports/mem_ops.h"
 #include "av1/common/common.h"
 #include "av1/common/obu_util.h"
@@ -24,8 +26,12 @@
 #define OBU_HEADER_SIZE 1
 #define OBU_EXTENSION_SIZE 1
 #define OBU_MAX_LENGTH_FIELD_SIZE 8
+
+#define OBU_MAX_HEADER_SIZE \
+  (OBU_HEADER_SIZE + OBU_EXTENSION_SIZE + 2 * OBU_MAX_LENGTH_FIELD_SIZE)
+
 #define OBU_DETECTION_SIZE \
-  (OBU_HEADER_SIZE + OBU_EXTENSION_SIZE + 3 * OBU_MAX_LENGTH_FIELD_SIZE)
+  (OBU_HEADER_SIZE + OBU_EXTENSION_SIZE + 4 * OBU_MAX_LENGTH_FIELD_SIZE)
 
 // Reads unsigned LEB128 integer and returns 0 upon successful read and decode.
 // Stores raw bytes in 'value_buffer', length of the number in 'value_length',
@@ -117,20 +123,20 @@ static int obudec_read_obu_header_and_size(FILE *f, size_t buffer_capacity,
                                            size_t *bytes_read,
                                            size_t *payload_length,
                                            ObuHeader *obu_header) {
-  const size_t kMinimumBufferSize =
-      (OBU_HEADER_SIZE + OBU_EXTENSION_SIZE + OBU_MAX_LENGTH_FIELD_SIZE);
+  const size_t kMinimumBufferSize = OBU_MAX_HEADER_SIZE;
   if (!f || !buffer || !bytes_read || !payload_length || !obu_header ||
       buffer_capacity < kMinimumBufferSize) {
     return -1;
   }
 
-  size_t leb128_length = 0;
+  size_t leb128_length_obu = 0;
+  size_t leb128_length_payload = 0;
   uint64_t obu_size = 0;
   if (is_annexb) {
-    if (obudec_read_leb128(f, &buffer[0], &leb128_length, &obu_size) != 0) {
+    if (obudec_read_leb128(f, &buffer[0], &leb128_length_obu, &obu_size) != 0) {
       fprintf(stderr, "obudec: Failure reading OBU size length.\n");
       return -1;
-    } else if (leb128_length == 0) {
+    } else if (leb128_length_obu == 0) {
       *payload_length = 0;
       return 0;
     }
@@ -141,8 +147,8 @@ static int obudec_read_obu_header_and_size(FILE *f, size_t buffer_capacity,
   }
 
   size_t header_size = 0;
-  if (obudec_read_obu_header(f, buffer_capacity - leb128_length, is_annexb,
-                             buffer + leb128_length, obu_header,
+  if (obudec_read_obu_header(f, buffer_capacity - leb128_length_obu, is_annexb,
+                             buffer + leb128_length_obu, obu_header,
                              &header_size) != 0) {
     return -1;
   } else if (header_size == 0) {
@@ -150,7 +156,8 @@ static int obudec_read_obu_header_and_size(FILE *f, size_t buffer_capacity,
     return 0;
   }
 
-  if (is_annexb) {
+  if (!obu_header->has_size_field) {
+    assert(is_annexb);
     if (obu_size < header_size) {
       fprintf(stderr, "obudec: OBU size is too small.\n");
       return -1;
@@ -158,8 +165,8 @@ static int obudec_read_obu_header_and_size(FILE *f, size_t buffer_capacity,
     *payload_length = (size_t)obu_size - header_size;
   } else {
     uint64_t u64_payload_length = 0;
-    if (obudec_read_leb128(f, &buffer[header_size], &leb128_length,
-                           &u64_payload_length) != 0) {
+    if (obudec_read_leb128(f, &buffer[leb128_length_obu + header_size],
+                           &leb128_length_payload, &u64_payload_length) != 0) {
       fprintf(stderr, "obudec: Failure reading OBU payload length.\n");
       return -1;
     }
@@ -171,7 +178,39 @@ static int obudec_read_obu_header_and_size(FILE *f, size_t buffer_capacity,
     *payload_length = (size_t)u64_payload_length;
   }
 
-  *bytes_read = leb128_length + header_size;
+  *bytes_read = leb128_length_obu + header_size + leb128_length_payload;
+  return 0;
+}
+
+static int obudec_grow_buffer(size_t growth_amount, uint8_t **obu_buffer,
+                              size_t *obu_buffer_capacity) {
+  if (!*obu_buffer || !obu_buffer_capacity || growth_amount == 0) {
+    return -1;
+  }
+
+  const size_t capacity = *obu_buffer_capacity;
+  if (SIZE_MAX - growth_amount < capacity) {
+    fprintf(stderr, "obudec: cannot grow buffer, capacity will roll over.\n");
+    return -1;
+  }
+
+  const size_t new_capacity = capacity + growth_amount;
+
+#if defined AOM_MAX_ALLOCABLE_MEMORY
+  if (new_capacity > AOM_MAX_ALLOCABLE_MEMORY) {
+    fprintf(stderr, "obudec: OBU size exceeds max alloc size.\n");
+    return -1;
+  }
+#endif
+
+  uint8_t *new_buffer = (uint8_t *)realloc(*obu_buffer, new_capacity);
+  if (!new_buffer) {
+    fprintf(stderr, "obudec: Failed to allocate compressed data buffer.\n");
+    return -1;
+  }
+
+  *obu_buffer = new_buffer;
+  *obu_buffer_capacity = new_capacity;
   return 0;
 }
 
@@ -179,12 +218,25 @@ static int obudec_read_one_obu(FILE *f, uint8_t **obu_buffer,
                                size_t obu_bytes_buffered,
                                size_t *obu_buffer_capacity, size_t *obu_length,
                                ObuHeader *obu_header, int is_annexb) {
-  size_t available_buffer_capacity = *obu_buffer_capacity - obu_bytes_buffered;
-
-  if (!(*obu_buffer)) return -1;
+  if (!f || !(*obu_buffer) || !obu_buffer_capacity || !obu_length ||
+      !obu_header) {
+    return -1;
+  }
 
   size_t bytes_read = 0;
   size_t obu_payload_length = 0;
+  size_t available_buffer_capacity = *obu_buffer_capacity - obu_bytes_buffered;
+
+  if (available_buffer_capacity < OBU_MAX_HEADER_SIZE) {
+    if (obudec_grow_buffer(AOMMAX(*obu_buffer_capacity, OBU_MAX_HEADER_SIZE),
+                           obu_buffer, obu_buffer_capacity) != 0) {
+      *obu_length = bytes_read;
+      return -1;
+    }
+    available_buffer_capacity +=
+        AOMMAX(*obu_buffer_capacity, OBU_MAX_HEADER_SIZE);
+  }
+
   const int status = obudec_read_obu_header_and_size(
       f, available_buffer_capacity, is_annexb, *obu_buffer + obu_bytes_buffered,
       &bytes_read, &obu_payload_length, obu_header);
@@ -199,28 +251,11 @@ static int obudec_read_one_obu(FILE *f, uint8_t **obu_buffer,
     return -1;
   }
 
-  if (bytes_read + obu_payload_length > available_buffer_capacity) {
-    // TODO(tomfinegan): Add overflow check.
-    const size_t new_capacity =
-        obu_bytes_buffered + bytes_read + 2 * obu_payload_length;
-
-#if defined AOM_MAX_ALLOCABLE_MEMORY
-    if (new_capacity > AOM_MAX_ALLOCABLE_MEMORY) {
-      fprintf(stderr, "obudec: OBU size exceeds max alloc size.\n");
-      return -1;
-    }
-#endif
-
-    uint8_t *new_buffer = (uint8_t *)realloc(*obu_buffer, new_capacity);
-
-    if (new_buffer) {
-      *obu_buffer = new_buffer;
-      *obu_buffer_capacity = new_capacity;
-    } else {
-      fprintf(stderr, "obudec: Failed to allocate compressed data buffer\n");
-      *obu_length = bytes_read + obu_payload_length;
-      return -1;
-    }
+  if (bytes_read + obu_payload_length > available_buffer_capacity &&
+      obudec_grow_buffer(AOMMAX(*obu_buffer_capacity, obu_payload_length),
+                         obu_buffer, obu_buffer_capacity) != 0) {
+    *obu_length = bytes_read + obu_payload_length;
+    return -1;
   }
 
   if (obu_payload_length > 0 &&
diff --git a/third_party/aom/common/rawenc.c b/third_party/aom/common/rawenc.c
index 5a2731d3ac..b72132c2e9 100644
--- a/third_party/aom/common/rawenc.c
+++ b/third_party/aom/common/rawenc.c
@@ -9,36 +9,88 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <stdbool.h>
 #include "common/rawenc.h"
 
-void raw_write_image_file(const aom_image_t *img, const int *planes,
-                          const int num_planes, FILE *file) {
-  const int bytes_per_sample = ((img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1);
+#define BATCH_SIZE 8
+// When writing greyscale color, batch 8 writes for low bit-depth, 4 writes
+// for high bit-depth.
+static const uint8_t batched[BATCH_SIZE] = { 128, 128, 128, 128,
+                                             128, 128, 128, 128 };
+static const uint8_t batched_hbd[BATCH_SIZE] = {
+  0, 128, 0, 128, 0, 128, 0, 128
+};
+
+// Interface to writing to either a file or MD5Context. Takes a pointer to
+// either the file or MD5Context, the buffer, the size of each element, and
+// number of elements to write. Note that size and nmemb (last two args) must
+// be unsigned int, as the interface to MD5Update requires that.
+typedef void (*WRITER)(void *, const uint8_t *, unsigned int, unsigned int);
+
+static void write_file(void *fp, const uint8_t *buffer, unsigned int size,
+                       unsigned int nmemb) {
+  fwrite(buffer, size, nmemb, (FILE *)fp);
+}
+
+static void write_md5(void *md5, const uint8_t *buffer, unsigned int size,
+                      unsigned int nmemb) {
+  MD5Update((MD5Context *)md5, buffer, size * nmemb);
+}
+
+// Writes out n greyscale values.
+static void write_greyscale(const bool high_bitdepth, int n, WRITER writer_func,
+                            void *file_or_md5) {
+  const uint8_t *b = batched;
+  if (high_bitdepth) {
+    b = batched_hbd;
+  }
+  const int num_batched_writes =
+      high_bitdepth ? n / (BATCH_SIZE / 2) : n / BATCH_SIZE;
+  for (int i = 0; i < num_batched_writes; ++i) {
+    writer_func(file_or_md5, b, sizeof(uint8_t), BATCH_SIZE);
+  }
+  const int remaining = high_bitdepth ? n % (BATCH_SIZE / 2) : n % BATCH_SIZE;
+  for (int i = 0; i < remaining; ++i) {
+    if (high_bitdepth) {
+      writer_func(file_or_md5, batched_hbd, sizeof(uint8_t), 2);
+    } else {
+      writer_func(file_or_md5, batched, sizeof(uint8_t), 1);
+    }
+  }
+}
+
+// Encapsulates the logic for writing raw data to either an image file or
+// to an MD5 context.
+static void raw_write_image_file_or_md5(const aom_image_t *img,
+                                        const int *planes, const int num_planes,
+                                        void *file_or_md5, WRITER writer_func) {
+  const bool high_bitdepth = img->fmt & AOM_IMG_FMT_HIGHBITDEPTH;
+  const int bytes_per_sample = high_bitdepth ? 2 : 1;
   for (int i = 0; i < num_planes; ++i) {
     const int plane = planes[i];
-    const unsigned char *buf = img->planes[plane];
-    const int stride = img->stride[plane];
     const int w = aom_img_plane_width(img, plane);
     const int h = aom_img_plane_height(img, plane);
+    // If we're on a color plane and the output is monochrome, write a greyscale
+    // value. Since there are only YUV planes, compare against Y.
+    if (img->monochrome && plane != AOM_PLANE_Y) {
+      write_greyscale(high_bitdepth, w * h, writer_func, file_or_md5);
+      continue;
+    }
+    const unsigned char *buf = img->planes[plane];
+    const int stride = img->stride[plane];
     for (int y = 0; y < h; ++y) {
-      fwrite(buf, bytes_per_sample, w, file);
+      writer_func(file_or_md5, buf, bytes_per_sample, w);
       buf += stride;
     }
   }
 }
 
+void raw_write_image_file(const aom_image_t *img, const int *planes,
+                          const int num_planes, FILE *file) {
+  raw_write_image_file_or_md5(img, planes, num_planes, file, write_file);
+}
+
 void raw_update_image_md5(const aom_image_t *img, const int *planes,
                           const int num_planes, MD5Context *md5) {
-  for (int i = 0; i < num_planes; ++i) {
-    const int plane = planes[i];
-    const unsigned char *buf = img->planes[plane];
-    const int stride = img->stride[plane];
-    const int w = aom_img_plane_width(img, plane) *
-                  ((img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1);
-    const int h = aom_img_plane_height(img, plane);
-    for (int y = 0; y < h; ++y) {
-      MD5Update(md5, buf, w);
-      buf += stride;
-    }
-  }
+  raw_write_image_file_or_md5(img, planes, num_planes, md5, write_md5);
 }
diff --git a/third_party/aom/common/tools_common.c b/third_party/aom/common/tools_common.c
index 21cd800262..51c1c52a1a 100644
--- a/third_party/aom/common/tools_common.c
+++ b/third_party/aom/common/tools_common.c
@@ -149,6 +149,11 @@ const AvxInterface *get_aom_encoder_by_name(const char *name) {
 
   return NULL;
 }
+
+// large scale tile encoding
+static const AvxInterface aom_lst_encoder = { "av1", LST_FOURCC,
+                                              &aom_codec_av1_cx };
+const AvxInterface *get_aom_lst_encoder(void) { return &aom_lst_encoder; }
 #endif  // CONFIG_AV1_ENCODER
 
 #if CONFIG_AV1_DECODER
@@ -283,6 +288,7 @@ static void lowbd_img_upshift(aom_image_t *dst, const aom_image_t *src,
     fatal("Unsupported image conversion");
   }
   switch (src->fmt) {
+    case AOM_IMG_FMT_YV12:
     case AOM_IMG_FMT_I420:
     case AOM_IMG_FMT_I422:
     case AOM_IMG_FMT_I444: break;
@@ -423,3 +429,80 @@ void aom_img_downshift(aom_image_t *dst, const aom_image_t *src,
     lowbd_img_downshift(dst, src, down_shift);
   }
 }
+
+static int img_shifted_realloc_required(const aom_image_t *img,
+                                        const aom_image_t *shifted,
+                                        aom_img_fmt_t required_fmt) {
+  return img->d_w != shifted->d_w || img->d_h != shifted->d_h ||
+         required_fmt != shifted->fmt;
+}
+
+void aom_shift_img(unsigned int output_bit_depth, aom_image_t **img_ptr,
+                   aom_image_t **img_shifted_ptr) {
+  aom_image_t *img = *img_ptr;
+  aom_image_t *img_shifted = *img_shifted_ptr;
+
+  const aom_img_fmt_t shifted_fmt = output_bit_depth == 8
+                                        ? img->fmt & ~AOM_IMG_FMT_HIGHBITDEPTH
+                                        : img->fmt | AOM_IMG_FMT_HIGHBITDEPTH;
+
+  if (shifted_fmt != img->fmt || output_bit_depth != img->bit_depth) {
+    if (img_shifted &&
+        img_shifted_realloc_required(img, img_shifted, shifted_fmt)) {
+      aom_img_free(img_shifted);
+      img_shifted = NULL;
+    }
+    if (img_shifted) {
+      img_shifted->monochrome = img->monochrome;
+    }
+    if (!img_shifted) {
+      img_shifted = aom_img_alloc(NULL, shifted_fmt, img->d_w, img->d_h, 16);
+      img_shifted->bit_depth = output_bit_depth;
+      img_shifted->monochrome = img->monochrome;
+      img_shifted->csp = img->csp;
+    }
+    if (output_bit_depth > img->bit_depth) {
+      aom_img_upshift(img_shifted, img, output_bit_depth - img->bit_depth);
+    } else {
+      aom_img_downshift(img_shifted, img, img->bit_depth - output_bit_depth);
+    }
+    *img_shifted_ptr = img_shifted;
+    *img_ptr = img_shifted;
+  }
+}
+
+// Related to I420, NV12 format has one luma "luminance" plane Y and one plane
+// with U and V values interleaved.
+void aom_img_write_nv12(const aom_image_t *img, FILE *file) {
+  // Y plane
+  const unsigned char *buf = img->planes[0];
+  int stride = img->stride[0];
+  int w = aom_img_plane_width(img, 0) *
+          ((img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1);
+  int h = aom_img_plane_height(img, 0);
+  int x, y;
+
+  for (y = 0; y < h; ++y) {
+    fwrite(buf, 1, w, file);
+    buf += stride;
+  }
+
+  // Interleaved U and V plane
+  const unsigned char *ubuf = img->planes[1];
+  const unsigned char *vbuf = img->planes[2];
+  const size_t size = (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1;
+  stride = img->stride[1];
+  w = aom_img_plane_width(img, 1);
+  h = aom_img_plane_height(img, 1);
+
+  for (y = 0; y < h; ++y) {
+    for (x = 0; x < w; ++x) {
+      fwrite(ubuf, size, 1, file);
+      fwrite(vbuf, size, 1, file);
+      ubuf += size;
+      vbuf += size;
+    }
+    ubuf += (stride - w * size);
+    vbuf += (stride - w * size);
+  }
+}
diff --git a/third_party/aom/common/tools_common.h b/third_party/aom/common/tools_common.h
index 4e1d12f4a0..d9a68f01bf 100644
--- a/third_party/aom/common/tools_common.h
+++ b/third_party/aom/common/tools_common.h
@@ -18,6 +18,7 @@
 #include "aom/aom_codec.h"
 #include "aom/aom_image.h"
 #include "aom/aom_integer.h"
+#include "aom_ports/mem.h"
 #include "aom_ports/msvc.h"
 
 #if CONFIG_AV1_ENCODER
@@ -77,6 +78,16 @@ enum VideoFileType {
   FILE_TYPE_WEBM
 };
 
+// Used in lightfield example.
+enum {
+  YUV1D,  // 1D tile output for conformance test.
+  YUV,    // Tile output in YUV format.
+  NV12,   // Tile output in NV12 format.
+} UENUM1BYTE(OUTPUT_FORMAT);
+
+// The fourcc for large_scale_tile encoding is "LSTC".
+#define LST_FOURCC 0x4354534c
+
 struct FileTypeDetectionBuffer {
   char buf[4];
   size_t buf_read;
@@ -142,6 +153,7 @@ typedef struct AvxInterface {
 int get_aom_encoder_count(void);
 const AvxInterface *get_aom_encoder_by_index(int i);
 const AvxInterface *get_aom_encoder_by_name(const char *name);
+const AvxInterface *get_aom_lst_encoder(void);
 
 int get_aom_decoder_count(void);
 const AvxInterface *get_aom_decoder_by_index(int i);
@@ -155,8 +167,13 @@ double sse_to_psnr(double samples, double peak, double mse);
 void aom_img_upshift(aom_image_t *dst, const aom_image_t *src, int input_shift);
 void aom_img_downshift(aom_image_t *dst, const aom_image_t *src,
                        int down_shift);
+void aom_shift_img(unsigned int output_bit_depth, aom_image_t **img_ptr,
+                   aom_image_t **img_shifted_ptr);
 void aom_img_truncate_16_to_8(aom_image_t *dst, const aom_image_t *src);
 
+// Output in NV12 format.
+void aom_img_write_nv12(const aom_image_t *img, FILE *file);
+
 #ifdef __cplusplus
 } /* extern "C" */
 #endif
diff --git a/third_party/aom/common/video_reader.c b/third_party/aom/common/video_reader.c
index 47ad6e1891..7b021bc40f 100644
--- a/third_party/aom/common/video_reader.c
+++ b/third_party/aom/common/video_reader.c
@@ -121,3 +121,7 @@ FILE *aom_video_reader_get_file(AvxVideoReader *reader) {
 const AvxVideoInfo *aom_video_reader_get_info(AvxVideoReader *reader) {
   return &reader->info;
 }
+
+void aom_video_reader_set_fourcc(AvxVideoReader *reader, uint32_t fourcc) {
+  reader->info.codec_fourcc = fourcc;
+}
diff --git a/third_party/aom/common/video_reader.h b/third_party/aom/common/video_reader.h
index 903deae849..9ab439e8af 100644
--- a/third_party/aom/common/video_reader.h
+++ b/third_party/aom/common/video_reader.h
@@ -50,6 +50,9 @@ FILE *aom_video_reader_get_file(AvxVideoReader *reader);
 // Fills AvxVideoInfo with information from opened video file.
 const AvxVideoInfo *aom_video_reader_get_info(AvxVideoReader *reader);
 
+// Set fourcc.
+void aom_video_reader_set_fourcc(AvxVideoReader *reader, uint32_t fourcc);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/third_party/aom/common/video_writer.c b/third_party/aom/common/video_writer.c
index a7ec309fc7..2b42e365cc 100644
--- a/third_party/aom/common/video_writer.c
+++ b/third_party/aom/common/video_writer.c
@@ -75,3 +75,7 @@ int aom_video_writer_write_frame(AvxVideoWriter *writer, const uint8_t *buffer,
 
   return 1;
 }
+
+void aom_video_writer_set_fourcc(AvxVideoWriter *writer, uint32_t fourcc) {
+  writer->info.codec_fourcc = fourcc;
+}
diff --git a/third_party/aom/common/video_writer.h b/third_party/aom/common/video_writer.h
index 3e2b6554b8..8712d47a58 100644
--- a/third_party/aom/common/video_writer.h
+++ b/third_party/aom/common/video_writer.h
@@ -14,7 +14,7 @@
 
 #include "common/video_common.h"
 
-typedef enum { kContainerIVF } AvxContainer;
+enum { kContainerIVF } UENUM1BYTE(AvxContainer);
 
 struct AvxVideoWriterStruct;
 typedef struct AvxVideoWriterStruct AvxVideoWriter;
@@ -37,6 +37,8 @@ void aom_video_writer_close(AvxVideoWriter *writer);
 // Writes frame bytes to the file.
 int aom_video_writer_write_frame(AvxVideoWriter *writer, const uint8_t *buffer,
                                  size_t size, int64_t pts);
+// Set fourcc.
+void aom_video_writer_set_fourcc(AvxVideoWriter *writer, uint32_t fourcc);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/common/webmenc.cc b/third_party/aom/common/webmenc.cc
index 58ab336707..6ae7df646f 100644
--- a/third_party/aom/common/webmenc.cc
+++ b/third_party/aom/common/webmenc.cc
@@ -11,8 +11,11 @@
 
 #include "common/webmenc.h"
 
+#include <stdio.h>
+
 #include <string>
 
+#include "common/av1_config.h"
 #include "third_party/libwebm/mkvmuxer/mkvmuxer.h"
 #include "third_party/libwebm/mkvmuxer/mkvmuxerutil.h"
 #include "third_party/libwebm/mkvmuxer/mkvwriter.h"
@@ -22,17 +25,33 @@ const uint64_t kDebugTrackUid = 0xDEADBEEF;
 const int kVideoTrackNumber = 1;
 }  // namespace
 
-void write_webm_file_header(struct WebmOutputContext *webm_ctx,
-                            const aom_codec_enc_cfg_t *cfg,
-                            stereo_format_t stereo_fmt, unsigned int fourcc,
-                            const struct AvxRational *par) {
+int write_webm_file_header(struct WebmOutputContext *webm_ctx,
+                           aom_codec_ctx_t *encoder_ctx,
+                           const aom_codec_enc_cfg_t *cfg,
+                           stereo_format_t stereo_fmt, unsigned int fourcc,
+                           const struct AvxRational *par) {
   mkvmuxer::MkvWriter *const writer = new mkvmuxer::MkvWriter(webm_ctx->stream);
   mkvmuxer::Segment *const segment = new mkvmuxer::Segment();
-  segment->Init(writer);
+  if (!writer || !segment) {
+    fprintf(stderr, "webmenc> mkvmuxer objects alloc failed, out of memory?\n");
+    return -1;
+  }
+
+  bool ok = segment->Init(writer);
+  if (!ok) {
+    fprintf(stderr, "webmenc> mkvmuxer Init failed.\n");
+    return -1;
+  }
+
   segment->set_mode(mkvmuxer::Segment::kFile);
   segment->OutputCues(true);
 
   mkvmuxer::SegmentInfo *const info = segment->GetSegmentInfo();
+  if (!info) {
+    fprintf(stderr, "webmenc> Cannot retrieve Segment Info.\n");
+    return -1;
+  }
+
   const uint64_t kTimecodeScale = 1000000;
   info->set_timecode_scale(kTimecodeScale);
   std::string version = "aomenc";
@@ -46,13 +65,48 @@ void write_webm_file_header(struct WebmOutputContext *webm_ctx,
                              static_cast<int>(cfg->g_h), kVideoTrackNumber);
   mkvmuxer::VideoTrack *const video_track = static_cast<mkvmuxer::VideoTrack *>(
       segment->GetTrackByNumber(video_track_id));
-  video_track->SetStereoMode(stereo_fmt);
-  const char *codec_id;
-  switch (fourcc) {
-    case AV1_FOURCC: codec_id = "V_AV1"; break;
-    default: codec_id = "V_AV1"; break;
+
+  if (!video_track) {
+    fprintf(stderr, "webmenc> Video track creation failed.\n");
+    return -1;
   }
-  video_track->set_codec_id(codec_id);
+
+  ok = false;
+  aom_fixed_buf_t *obu_sequence_header =
+      aom_codec_get_global_headers(encoder_ctx);
+  if (obu_sequence_header) {
+    Av1Config av1_config;
+    if (get_av1config_from_obu(
+            reinterpret_cast<const uint8_t *>(obu_sequence_header->buf),
+            obu_sequence_header->sz, false, &av1_config) == 0) {
+      uint8_t av1_config_buffer[4] = { 0 };
+      size_t bytes_written = 0;
+      if (write_av1config(&av1_config, sizeof(av1_config_buffer),
+                          &bytes_written, av1_config_buffer) == 0) {
+        ok = video_track->SetCodecPrivate(av1_config_buffer,
+                                          sizeof(av1_config_buffer));
+      }
+    }
+    free(obu_sequence_header->buf);
+    free(obu_sequence_header);
+  }
+  if (!ok) {
+    fprintf(stderr, "webmenc> Unable to set AV1 config.\n");
+    return -1;
+  }
+
+  ok = video_track->SetStereoMode(stereo_fmt);
+  if (!ok) {
+    fprintf(stderr, "webmenc> Unable to set stereo mode.\n");
+    return -1;
+  }
+
+  if (fourcc != AV1_FOURCC) {
+    fprintf(stderr, "webmenc> Unsupported codec (unknown 4 CC).\n");
+    return -1;
+  }
+  video_track->set_codec_id("V_AV1");
+
   if (par->numerator > 1 || par->denominator > 1) {
     // TODO(fgalligan): Add support of DisplayUnit, Display Aspect Ratio type
     // to WebM format.
@@ -61,16 +115,24 @@ void write_webm_file_header(struct WebmOutputContext *webm_ctx,
     video_track->set_display_width(display_width);
     video_track->set_display_height(cfg->g_h);
   }
+
   if (webm_ctx->debug) {
     video_track->set_uid(kDebugTrackUid);
   }
+
   webm_ctx->writer = writer;
   webm_ctx->segment = segment;
+
+  return 0;
 }
 
-void write_webm_block(struct WebmOutputContext *webm_ctx,
-                      const aom_codec_enc_cfg_t *cfg,
-                      const aom_codec_cx_pkt_t *pkt) {
+int write_webm_block(struct WebmOutputContext *webm_ctx,
+                     const aom_codec_enc_cfg_t *cfg,
+                     const aom_codec_cx_pkt_t *pkt) {
+  if (!webm_ctx->segment) {
+    fprintf(stderr, "webmenc> segment is NULL.\n");
+    return -1;
+  }
   mkvmuxer::Segment *const segment =
       reinterpret_cast<mkvmuxer::Segment *>(webm_ctx->segment);
   int64_t pts_ns = pkt->data.frame.pts * 1000000000ll * cfg->g_timebase.num /
@@ -78,19 +140,34 @@ void write_webm_block(struct WebmOutputContext *webm_ctx,
   if (pts_ns <= webm_ctx->last_pts_ns) pts_ns = webm_ctx->last_pts_ns + 1000000;
   webm_ctx->last_pts_ns = pts_ns;
 
-  segment->AddFrame(static_cast<uint8_t *>(pkt->data.frame.buf),
-                    pkt->data.frame.sz, kVideoTrackNumber, pts_ns,
-                    pkt->data.frame.flags & AOM_FRAME_IS_KEY);
+  if (!segment->AddFrame(static_cast<uint8_t *>(pkt->data.frame.buf),
+                         pkt->data.frame.sz, kVideoTrackNumber, pts_ns,
+                         pkt->data.frame.flags & AOM_FRAME_IS_KEY)) {
+    fprintf(stderr, "webmenc> AddFrame failed.\n");
+    return -1;
+  }
+  return 0;
 }
 
-void write_webm_file_footer(struct WebmOutputContext *webm_ctx) {
+int write_webm_file_footer(struct WebmOutputContext *webm_ctx) {
+  if (!webm_ctx->writer || !webm_ctx->segment) {
+    fprintf(stderr, "webmenc> segment or writer NULL.\n");
+    return -1;
+  }
   mkvmuxer::MkvWriter *const writer =
       reinterpret_cast<mkvmuxer::MkvWriter *>(webm_ctx->writer);
   mkvmuxer::Segment *const segment =
       reinterpret_cast<mkvmuxer::Segment *>(webm_ctx->segment);
-  segment->Finalize();
+  const bool ok = segment->Finalize();
   delete segment;
   delete writer;
   webm_ctx->writer = NULL;
   webm_ctx->segment = NULL;
+
+  if (!ok) {
+    fprintf(stderr, "webmenc> Segment::Finalize failed.\n");
+    return -1;
+  }
+
+  return 0;
 }
diff --git a/third_party/aom/common/webmenc.h b/third_party/aom/common/webmenc.h
index aa9832fbae..a4aa992b02 100644
--- a/third_party/aom/common/webmenc.h
+++ b/third_party/aom/common/webmenc.h
@@ -30,24 +30,28 @@ struct WebmOutputContext {
 };
 
 /* Stereo 3D packed frame format */
-typedef enum stereo_format {
+enum {
   STEREO_FORMAT_MONO = 0,
   STEREO_FORMAT_LEFT_RIGHT = 1,
   STEREO_FORMAT_BOTTOM_TOP = 2,
   STEREO_FORMAT_TOP_BOTTOM = 3,
   STEREO_FORMAT_RIGHT_LEFT = 11
-} stereo_format_t;
+} UENUM1BYTE(stereo_format_t);
 
-void write_webm_file_header(struct WebmOutputContext *webm_ctx,
-                            const aom_codec_enc_cfg_t *cfg,
-                            stereo_format_t stereo_fmt, unsigned int fourcc,
-                            const struct AvxRational *par);
+// The following functions wrap libwebm's mkvmuxer. All functions return 0 upon
+// success, or -1 upon failure.
 
-void write_webm_block(struct WebmOutputContext *webm_ctx,
-                      const aom_codec_enc_cfg_t *cfg,
-                      const aom_codec_cx_pkt_t *pkt);
+int write_webm_file_header(struct WebmOutputContext *webm_ctx,
+                           aom_codec_ctx_t *encoder_ctx,
+                           const aom_codec_enc_cfg_t *cfg,
+                           stereo_format_t stereo_fmt, unsigned int fourcc,
+                           const struct AvxRational *par);
 
-void write_webm_file_footer(struct WebmOutputContext *webm_ctx);
+int write_webm_block(struct WebmOutputContext *webm_ctx,
+                     const aom_codec_enc_cfg_t *cfg,
+                     const aom_codec_cx_pkt_t *pkt);
+
+int write_webm_file_footer(struct WebmOutputContext *webm_ctx);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/third_party/aom/common/y4menc.c b/third_party/aom/common/y4menc.c
index 585d221976..e3f5d5b387 100644
--- a/third_party/aom/common/y4menc.c
+++ b/third_party/aom/common/y4menc.c
@@ -15,7 +15,7 @@
 #include "common/y4menc.h"
 
 // Returns the Y4M name associated with the monochrome colorspace.
-const char *monochrome_colorspace(unsigned int bit_depth) {
+static const char *monochrome_colorspace(unsigned int bit_depth) {
   switch (bit_depth) {
     case 8: return "Cmono";
     case 9: return "Cmono9";
@@ -30,12 +30,15 @@ const char *monochrome_colorspace(unsigned int bit_depth) {
 // image format.
 const char *colorspace8(aom_chroma_sample_position_t csp, aom_img_fmt_t fmt) {
   switch (fmt) {
-    case AOM_IMG_FMT_444A: return "C444alpha";
     case AOM_IMG_FMT_I444: return "C444";
     case AOM_IMG_FMT_I422: return "C422";
     default:
       if (csp == AOM_CSP_VERTICAL) {
         return "C420mpeg2 XYSCSS=420MPEG2";
+      } else if (csp == AOM_CSP_COLOCATED) {
+        // Note that Y4M does not have a dedicated header for colocated chroma,
+        // and that FFMPEG interprets C420 as C420jpeg.
+        return "C420";
       } else {
         return "C420jpeg";
       }
@@ -43,8 +46,9 @@ const char *colorspace8(aom_chroma_sample_position_t csp, aom_img_fmt_t fmt) {
 }
 
 // Return the Y4M name of the colorspace, given the bit depth and image format.
-const char *colorspace(unsigned int bit_depth, aom_chroma_sample_position_t csp,
-                       aom_img_fmt_t fmt) {
+static const char *colorspace(unsigned int bit_depth,
+                              aom_chroma_sample_position_t csp,
+                              aom_img_fmt_t fmt) {
   switch (bit_depth) {
     case 8: return colorspace8(csp, fmt);
     case 9:
diff --git a/third_party/aom/common/y4minput.c b/third_party/aom/common/y4minput.c
index eca8b1bbac..f3dfaafc68 100644
--- a/third_party/aom/common/y4minput.c
+++ b/third_party/aom/common/y4minput.c
@@ -16,6 +16,7 @@
 #include <string.h>
 
 #include "aom/aom_integer.h"
+#include "aom_ports/msvc.h"
 #include "y4minput.h"
 
 // Reads 'size' bytes from 'file' into 'buf' with some fault tolerance.
@@ -109,7 +110,8 @@ static int y4m_parse_tags(y4m_input *_y4m, char *_tags) {
   if (!got_par) _y4m->par_n = _y4m->par_d = 0;
   /*Chroma-type is not specified in older files, e.g., those generated by
      mplayer.*/
-  if (!got_chroma) strcpy(_y4m->chroma_type, "420");
+  if (!got_chroma)
+    snprintf(_y4m->chroma_type, sizeof(_y4m->chroma_type), "420");
   return 0;
 }
 
@@ -778,7 +780,7 @@ static void y4m_convert_null(y4m_input *_y4m, unsigned char *_dst,
 }
 
 int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
-                   int only_420) {
+                   aom_chroma_sample_position_t csp, int only_420) {
   char buffer[80] = { 0 };
   int ret;
   int i;
@@ -821,6 +823,19 @@ int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
             "Only progressive scan handled.\n");
     return -1;
   }
+  /* Only support vertical chroma sample position if the input format is
+   * already 420mpeg2. Colocated is not supported in Y4M.
+   */
+  if (csp == AOM_CSP_VERTICAL && strcmp(_y4m->chroma_type, "420mpeg2") != 0) {
+    fprintf(stderr,
+            "Vertical chroma sample position only supported "
+            "for 420mpeg2 input\n");
+    return -1;
+  }
+  if (csp == AOM_CSP_COLOCATED) {
+    fprintf(stderr, "Colocated chroma sample position not supported in Y4M\n");
+    return -1;
+  }
   _y4m->aom_fmt = AOM_IMG_FMT_I420;
   _y4m->bps = 12;
   _y4m->bit_depth = 8;
@@ -877,7 +892,11 @@ int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
     /*Chroma filter required: read into the aux buf first.*/
     _y4m->aux_buf_sz = _y4m->aux_buf_read_sz =
         2 * ((_y4m->pic_w + 1) / 2) * ((_y4m->pic_h + 1) / 2);
-    _y4m->convert = y4m_convert_42xmpeg2_42xjpeg;
+    _y4m->convert = y4m_convert_null;
+    if (csp != AOM_CSP_VERTICAL) {
+      _y4m->convert = y4m_convert_42xmpeg2_42xjpeg;
+      snprintf(_y4m->chroma_type, sizeof(_y4m->chroma_type), "420");
+    }
   } else if (strcmp(_y4m->chroma_type, "420paldv") == 0) {
     _y4m->src_c_dec_h = _y4m->dst_c_dec_h = _y4m->src_c_dec_v =
         _y4m->dst_c_dec_v = 2;
@@ -1037,14 +1056,8 @@ int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
       _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 3 * _y4m->pic_w * _y4m->pic_h;
       _y4m->convert = y4m_convert_444_420jpeg;
     } else {
-      _y4m->aom_fmt = AOM_IMG_FMT_444A;
-      _y4m->bps = 32;
-      _y4m->dst_c_dec_h = _y4m->src_c_dec_h;
-      _y4m->dst_c_dec_v = _y4m->src_c_dec_v;
-      _y4m->dst_buf_read_sz = 4 * _y4m->pic_w * _y4m->pic_h;
-      /*Natively supported: no conversion required.*/
-      _y4m->aux_buf_sz = _y4m->aux_buf_read_sz = 0;
-      _y4m->convert = y4m_convert_null;
+      fprintf(stderr, "Unsupported format: 444A\n");
+      return -1;
     }
   } else if (strcmp(_y4m->chroma_type, "mono") == 0) {
     _y4m->src_c_dec_h = _y4m->src_c_dec_v = 0;
@@ -1131,12 +1144,10 @@ int y4m_input_fetch_frame(y4m_input *_y4m, FILE *_fin, aom_image_t *_img) {
   c_w *= bytes_per_sample;
   c_h = (_y4m->pic_h + _y4m->dst_c_dec_v - 1) / _y4m->dst_c_dec_v;
   c_sz = c_w * c_h;
-  _img->stride[AOM_PLANE_Y] = _img->stride[AOM_PLANE_ALPHA] =
-      _y4m->pic_w * bytes_per_sample;
+  _img->stride[AOM_PLANE_Y] = _y4m->pic_w * bytes_per_sample;
   _img->stride[AOM_PLANE_U] = _img->stride[AOM_PLANE_V] = c_w;
   _img->planes[AOM_PLANE_Y] = _y4m->dst_buf;
   _img->planes[AOM_PLANE_U] = _y4m->dst_buf + pic_sz;
   _img->planes[AOM_PLANE_V] = _y4m->dst_buf + pic_sz + c_sz;
-  _img->planes[AOM_PLANE_ALPHA] = _y4m->dst_buf + pic_sz + 2 * c_sz;
   return 1;
 }
diff --git a/third_party/aom/common/y4minput.h b/third_party/aom/common/y4minput.h
index 01b9ce9726..f6c5a3d3ab 100644
--- a/third_party/aom/common/y4minput.h
+++ b/third_party/aom/common/y4minput.h
@@ -57,8 +57,14 @@ struct y4m_input {
   unsigned int bit_depth;
 };
 
+/**
+ * Open the input file, treating it as Y4M. y4m_input is filled in after
+ * reading it. Note that chroma-sample-position should only be set for 420
+ * input, and the input chroma is shifted if necessary. The code does not
+ * support the conversion from co-located to vertical.
+ */
 int y4m_input_open(y4m_input *_y4m, FILE *_fin, char *_skip, int _nskip,
-                   int only_420);
+                   aom_chroma_sample_position_t csp, int only_420);
 void y4m_input_close(y4m_input *_y4m);
 int y4m_input_fetch_frame(y4m_input *_y4m, FILE *_fin, aom_image_t *img);
 
diff --git a/third_party/aom/examples/analyzer.cc b/third_party/aom/examples/analyzer.cc
index 6a42eca248..261d085697 100644
--- a/third_party/aom/examples/analyzer.cc
+++ b/third_party/aom/examples/analyzer.cc
@@ -162,7 +162,7 @@ bool AV1Decoder::setInspectionCallback() {
 
 void AV1Decoder::inspect(void *pbi, void *data) {
   AV1Decoder *decoder = (AV1Decoder *)data;
-  ifd_inspect(&decoder->frame_data, pbi);
+  ifd_inspect(&decoder->frame_data, pbi, 0);
 }
 
 #define MIN_ZOOM (1)
diff --git a/third_party/aom/examples/aom_cx_set_ref.c b/third_party/aom/examples/aom_cx_set_ref.c
index 8e3d216fe6..5a78af9f78 100644
--- a/third_party/aom/examples/aom_cx_set_ref.c
+++ b/third_party/aom/examples/aom_cx_set_ref.c
@@ -141,9 +141,7 @@ static int encode_frame(aom_codec_ctx_t *ecodec, aom_image_t *img,
     if (pkt->kind == AOM_CODEC_CX_FRAME_PKT) {
       const int keyframe = (pkt->data.frame.flags & AOM_FRAME_IS_KEY) != 0;
 
-      if (!(pkt->data.frame.flags & AOM_FRAME_IS_FRAGMENT)) {
-        *frame_out += 1;
-      }
+      ++*frame_out;
 
       if (!aom_video_writer_write_frame(writer, pkt->data.frame.buf,
                                         pkt->data.frame.sz,
diff --git a/third_party/aom/examples/av1_dec_fuzzer.cc b/third_party/aom/examples/av1_dec_fuzzer.cc
new file mode 100644
index 0000000000..96d16a8564
--- /dev/null
+++ b/third_party/aom/examples/av1_dec_fuzzer.cc
@@ -0,0 +1,72 @@
+/*
+ * Copyright (c) 2019, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+/*
+ * See build_av1_dec_fuzzer.sh for building instructions.
+ */
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <memory>
+
+#include "config/aom_config.h"
+#include "aom/aom_decoder.h"
+#include "aom/aomdx.h"
+#include "aom_ports/mem_ops.h"
+#include "common/ivfdec.h"
+
+static void close_file(FILE *file) { fclose(file); }
+
+extern "C" void usage_exit(void) { exit(EXIT_FAILURE); }
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+  std::unique_ptr<FILE, decltype(&close_file)> file(
+      fmemopen((void *)data, size, "rb"), &close_file);
+  if (file == nullptr) {
+    return 0;
+  }
+
+  char header[32];
+  if (fread(header, 1, 32, file.get()) != 32) {
+    return 0;
+  }
+  const AvxInterface *decoder = get_aom_decoder_by_name("av1");
+  if (decoder == nullptr) {
+    return 0;
+  }
+
+  aom_codec_ctx_t codec;
+  // Set thread count in the range [1, 64].
+  const unsigned int threads = (header[0] & 0x3f) + 1;
+  aom_codec_dec_cfg_t cfg = { threads, 0, 0, CONFIG_LOWBITDEPTH };
+  if (aom_codec_dec_init(&codec, decoder->codec_interface(), &cfg, 0)) {
+    return 0;
+  }
+
+  uint8_t *buffer = nullptr;
+  size_t buffer_size = 0;
+  size_t frame_size = 0;
+  while (!ivf_read_frame(file.get(), &buffer, &frame_size, &buffer_size,
+                         nullptr)) {
+    const aom_codec_err_t err =
+        aom_codec_decode(&codec, buffer, frame_size, nullptr);
+    static_cast<void>(err);
+    aom_codec_iter_t iter = nullptr;
+    aom_image_t *img = nullptr;
+    while ((img = aom_codec_get_frame(&codec, &iter)) != nullptr) {
+    }
+  }
+  aom_codec_destroy(&codec);
+  free(buffer);
+  return 0;
+}
diff --git a/third_party/aom/examples/build_av1_dec_fuzzer.sh b/third_party/aom/examples/build_av1_dec_fuzzer.sh
new file mode 100644
index 0000000000..86992a0990
--- /dev/null
+++ b/third_party/aom/examples/build_av1_dec_fuzzer.sh
@@ -0,0 +1,78 @@
+#!/bin/bash
+#
+# Copyright (c) 2019, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and
+# the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+# was not distributed with this source code in the LICENSE file, you can
+# obtain it at www.aomedia.org/license/software. If the Alliance for Open
+# Media Patent License 1.0 was not distributed with this source code in the
+# PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+#
+###############################################################################
+# Fuzzer for libaom decoder.
+# ==========================
+# Requirements
+# ---------------------
+# Clang6.0 or above (must support -fsanitize=fuzzer)
+#
+# References:
+# ---------------------
+# http://llvm.org/docs/LibFuzzer.html
+# https://github.com/google/oss-fuzz
+#
+# Steps to build / run
+# ---------------------
+
+set -eu
+
+# Have a copy of AOM and a build directory ready.
+if [[ $# -ne 2 ]]; then
+  echo "Pass in the AOM source tree as first argument, and a build directory "
+  echo "as the second argument. The AOM source tree can be obtained via: "
+  echo "  git clone https://aomedia.googlesource.com/aom"
+  exit 2
+fi
+if [[ -z "$CC" ]]; then
+  echo "Set the CC environment variable to point to your C compiler."
+  exit 2
+fi
+if [[ -z "$CXX" ]]; then
+  echo "Set the CXX environment variable to point to your C++ compiler."
+  exit 2
+fi
+
+AOM_DIR=$1
+BUILD_DIR=$2
+# Run CMake with address sanitizer enabled and build the codec.
+# Enable DO_RANGE_CHECK_CLAMP to suppress the noise of integer overflows
+# in the transform functions. Also set memory limits.
+EXTRA_C_FLAGS='-DDO_RANGE_CHECK_CLAMP=1 -DAOM_MAX_ALLOCABLE_MEMORY=1073741824'
+cd "${BUILD_DIR}"
+cmake "${AOM_DIR}" -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCONFIG_PIC=1 \
+  -DCONFIG_SCALABILITY=0 -DCONFIG_LOWBITDEPTH=1 -DCONFIG_AV1_ENCODER=0 \
+  -DENABLE_EXAMPLES=0 -DENABLE_DOCS=0 -DENABLE_TESTS=0 -DCONFIG_SIZE_LIMIT=1 \
+  -DDECODE_HEIGHT_LIMIT=12288 -DDECODE_WIDTH_LIMIT=12288 \
+  -DAOM_EXTRA_C_FLAGS="${EXTRA_C_FLAGS}" \
+  -DAOM_EXTRA_CXX_FLAGS="${EXTRA_C_FLAGS}" -DSANITIZE=address
+
+# Build the codec.
+make -j$(nproc)
+
+# Build some libaom utils that are not part of the core lib.
+$CC -std=c99 -c -I${AOM_DIR} -I${BUILD_DIR} \
+  ${AOM_DIR}/common/ivfdec.c -o ${BUILD_DIR}/ivfdec.o
+
+$CC -std=c99 -c -I${AOM_DIR} -I${BUILD_DIR} \
+  ${AOM_DIR}/common/tools_common.c -o ${BUILD_DIR}/tools_common.o
+
+# Build the av1 fuzzer
+$CXX -std=c++11 -DDECODER=av1 -I${AOM_DIR} -I${BUILD_DIR} \
+    -fsanitize=fuzzer -Wl,--start-group \
+    ${AOM_DIR}/examples/av1_dec_fuzzer.cc -o ${BUILD_DIR}/av1_dec_fuzzer \
+    ${BUILD_DIR}/libaom.a ${BUILD_DIR}/ivfdec.o ${BUILD_DIR}/tools_common.o \
+    -Wl,--end-group
+
+echo "Fuzzer built at ${BUILD_DIR}/av1_dec_fuzzer."
+echo "Create a corpus directory, copy IVF files in there, and run:"
+echo "  av1_dec_fuzzer CORPUS_DIR"
diff --git a/third_party/aom/examples/inspect.c b/third_party/aom/examples/inspect.c
index 9d5f0dcfc4..050e3da3da 100644
--- a/third_party/aom/examples/inspect.c
+++ b/third_party/aom/examples/inspect.c
@@ -60,7 +60,9 @@ typedef enum {
   DUAL_FILTER_LAYER = 1 << 12,
   Q_INDEX_LAYER = 1 << 13,
   SEGMENT_ID_LAYER = 1 << 14,
-  ALL_LAYERS = (1 << 15) - 1
+  MOTION_MODE_LAYER = 1 << 15,
+  COMPOUND_TYPE_LAYER = 1 << 16,
+  ALL_LAYERS = (1 << 17) - 1
 } LayerType;
 
 static LayerType layers = 0;
@@ -84,6 +86,10 @@ static const arg_def_t dump_transform_size_arg =
 static const arg_def_t dump_transform_type_arg =
     ARG_DEF("tt", "transformType", 0, "Dump Transform Type");
 static const arg_def_t dump_mode_arg = ARG_DEF("m", "mode", 0, "Dump Mode");
+static const arg_def_t dump_motion_mode_arg =
+    ARG_DEF("mm", "motion_mode", 0, "Dump Motion Modes");
+static const arg_def_t dump_compound_type_arg =
+    ARG_DEF("ct", "compound_type", 0, "Dump Compound Types");
 static const arg_def_t dump_uv_mode_arg =
     ARG_DEF("uvm", "uv_mode", 0, "Dump UV Intra Prediction Modes");
 static const arg_def_t dump_skip_arg = ARG_DEF("s", "skip", 0, "Dump Skip");
@@ -101,6 +107,13 @@ static const arg_def_t dump_delta_q_arg =
 static const arg_def_t dump_seg_id_arg =
     ARG_DEF("si", "seg_id", 0, "Dump Segment ID");
 static const arg_def_t usage_arg = ARG_DEF("h", "help", 0, "Help");
+static const arg_def_t skip_non_transform_arg = ARG_DEF(
+    "snt", "skip_non_transform", 1, "Skip is counted as a non transform.");
+static const arg_def_t combined_arg =
+    ARG_DEF("comb", "combined", 1, "combinining parameters into one output.");
+
+int combined_parm_list[5];
+int combined_parm_count = 0;
 
 static const arg_def_t *main_args[] = { &limit_arg,
                                         &dump_all_arg,
@@ -113,6 +126,8 @@ static const arg_def_t *main_args[] = { &limit_arg,
                                         &dump_transform_type_arg,
                                         &dump_mode_arg,
                                         &dump_uv_mode_arg,
+                                        &dump_motion_mode_arg,
+                                        &dump_compound_type_arg,
                                         &dump_skip_arg,
                                         &dump_filter_arg,
                                         &dump_cdef_arg,
@@ -123,6 +138,8 @@ static const arg_def_t *main_args[] = { &limit_arg,
                                         &dump_delta_q_arg,
                                         &dump_seg_id_arg,
                                         &usage_arg,
+                                        &skip_non_transform_arg,
+                                        &combined_arg,
                                         NULL };
 #define ENUM(name) \
   { #name, name }
@@ -150,12 +167,14 @@ const map_entry block_size_map[] = {
   ENUM(BLOCK_64X16),   LAST_ENUM
 };
 
+#define TX_SKIP -1
+
 const map_entry tx_size_map[] = {
   ENUM(TX_4X4),   ENUM(TX_8X8),   ENUM(TX_16X16), ENUM(TX_32X32),
   ENUM(TX_64X64), ENUM(TX_4X8),   ENUM(TX_8X4),   ENUM(TX_8X16),
   ENUM(TX_16X8),  ENUM(TX_16X32), ENUM(TX_32X16), ENUM(TX_32X64),
   ENUM(TX_64X32), ENUM(TX_4X16),  ENUM(TX_16X4),  ENUM(TX_8X32),
-  ENUM(TX_32X8),  LAST_ENUM
+  ENUM(TX_32X8),  ENUM(TX_16X64), ENUM(TX_64X16), LAST_ENUM
 };
 
 const map_entry tx_type_map[] = { ENUM(DCT_DCT),
@@ -193,6 +212,15 @@ const map_entry prediction_mode_map[] = {
   ENUM(NEW_NEWMV),   ENUM(INTRA_INVALID), LAST_ENUM
 };
 
+const map_entry motion_mode_map[] = { ENUM(SIMPLE_TRANSLATION),
+                                      ENUM(OBMC_CAUSAL),    // 2-sided OBMC
+                                      ENUM(WARPED_CAUSAL),  // 2-sided WARPED
+                                      LAST_ENUM };
+
+const map_entry compound_type_map[] = { ENUM(COMPOUND_AVERAGE),
+                                        ENUM(COMPOUND_WEDGE),
+                                        ENUM(COMPOUND_DIFFWTD), LAST_ENUM };
+
 const map_entry uv_prediction_mode_map[] = {
   ENUM(UV_DC_PRED),       ENUM(UV_V_PRED),
   ENUM(UV_H_PRED),        ENUM(UV_D45_PRED),
@@ -212,6 +240,43 @@ const map_entry config_map[] = { ENUM(MI_SIZE), LAST_ENUM };
 
 static const char *exec_name;
 
+struct parm_offset {
+  char parm[60];
+  char offset;
+};
+struct parm_offset parm_offsets[] = {
+  { "blockSize", offsetof(insp_mi_data, sb_type) },
+  { "transformSize", offsetof(insp_mi_data, tx_size) },
+  { "transformType", offsetof(insp_mi_data, tx_type) },
+  { "dualFilterType", offsetof(insp_mi_data, dual_filter_type) },
+  { "mode", offsetof(insp_mi_data, mode) },
+  { "uv_mode", offsetof(insp_mi_data, uv_mode) },
+  { "motion_mode", offsetof(insp_mi_data, motion_mode) },
+  { "compound_type", offsetof(insp_mi_data, compound_type) },
+  { "referenceFrame", offsetof(insp_mi_data, ref_frame) },
+  { "skip", offsetof(insp_mi_data, skip) },
+};
+int parm_count = sizeof(parm_offsets) / sizeof(parm_offsets[0]);
+
+int convert_to_indices(char *str, int *indices, int maxCount, int *count) {
+  *count = 0;
+  do {
+    char *comma = strchr(str, ',');
+    int length = (comma ? (int)(comma - str) : (int)strlen(str));
+    int i;
+    for (i = 0; i < parm_count; ++i) {
+      if (!strncmp(str, parm_offsets[i].parm, length)) {
+        break;
+      }
+    }
+    if (i == parm_count) return 0;
+    indices[(*count)++] = i;
+    if (*count > maxCount) return 0;
+    str += length + 1;
+  } while (strlen(str) > 0);
+  return 1;
+}
+
 insp_frame_data frame_data;
 int frame_count = 0;
 int decoded_frame_count = 0;
@@ -382,6 +447,38 @@ int put_motion_vectors(char *buffer) {
   return (int)(buf - buffer);
 }
 
+int put_combined(char *buffer) {
+  const int mi_rows = frame_data.mi_rows;
+  const int mi_cols = frame_data.mi_cols;
+  char *buf = buffer;
+  int r, c, p;
+  buf += put_str(buf, "  \"");
+  for (p = 0; p < combined_parm_count; ++p) {
+    if (p) buf += put_str(buf, "&");
+    buf += put_str(buf, parm_offsets[combined_parm_list[p]].parm);
+  }
+  buf += put_str(buf, "\": [");
+  for (r = 0; r < mi_rows; ++r) {
+    *(buf++) = '[';
+    for (c = 0; c < mi_cols; ++c) {
+      insp_mi_data *mi = &frame_data.mi_grid[r * mi_cols + c];
+      *(buf++) = '[';
+      for (p = 0; p < combined_parm_count; ++p) {
+        if (p) *(buf++) = ',';
+        int16_t *v = (int16_t *)(((int8_t *)mi) +
+                                 parm_offsets[combined_parm_list[p]].offset);
+        buf += put_num(buf, 0, v[0], 0);
+      }
+      *(buf++) = ']';
+      if (c < mi_cols - 1) *(buf++) = ',';
+    }
+    *(buf++) = ']';
+    if (r < mi_rows - 1) *(buf++) = ',';
+  }
+  buf += put_str(buf, "],\n");
+  return (int)(buf - buffer);
+}
+
 int put_block_info(char *buffer, const map_entry *map, const char *name,
                    size_t offset, int len) {
   const int mi_rows = frame_data.mi_rows;
@@ -490,9 +587,16 @@ int put_accounting(char *buffer) {
 }
 #endif
 
+int skip_non_transform = 0;
+
 void inspect(void *pbi, void *data) {
   /* Fetch frame data. */
-  ifd_inspect(&frame_data, pbi);
+  ifd_inspect(&frame_data, pbi, skip_non_transform);
+
+  // Show existing frames just show a reference buffer we've already decoded.
+  // There's no information to show.
+  if (frame_data.show_existing_frame) return;
+
   (void)data;
   // We allocate enough space and hope we don't write out of bounds. Totally
   // unsafe but this speeds things up, especially when compiled to Javascript.
@@ -523,6 +627,14 @@ void inspect(void *pbi, void *data) {
     buf += put_block_info(buf, uv_prediction_mode_map, "uv_mode",
                           offsetof(insp_mi_data, uv_mode), 0);
   }
+  if (layers & MOTION_MODE_LAYER) {
+    buf += put_block_info(buf, motion_mode_map, "motion_mode",
+                          offsetof(insp_mi_data, motion_mode), 0);
+  }
+  if (layers & COMPOUND_TYPE_LAYER) {
+    buf += put_block_info(buf, compound_type_map, "compound_type",
+                          offsetof(insp_mi_data, compound_type), 0);
+  }
   if (layers & SKIP_LAYER) {
     buf +=
         put_block_info(buf, skip_map, "skip", offsetof(insp_mi_data, skip), 0);
@@ -554,6 +666,7 @@ void inspect(void *pbi, void *data) {
   if (layers & MOTION_VECTORS_LAYER) {
     buf += put_motion_vectors(buf);
   }
+  if (combined_parm_count > 0) buf += put_combined(buf);
   if (layers & REFERENCE_FRAME_LAYER) {
     buf += put_block_info(buf, refs_map, "referenceFrame",
                           offsetof(insp_mi_data, ref_frame), 2);
@@ -563,7 +676,8 @@ void inspect(void *pbi, void *data) {
     buf += put_accounting(buf);
   }
 #endif
-  buf += snprintf(buf, MAX_BUFFER, "  \"frame\": %d,\n", decoded_frame_count);
+  buf +=
+      snprintf(buf, MAX_BUFFER, "  \"frame\": %d,\n", frame_data.frame_number);
   buf += snprintf(buf, MAX_BUFFER, "  \"showFrame\": %d,\n",
                   frame_data.show_frame);
   buf += snprintf(buf, MAX_BUFFER, "  \"frameType\": %d,\n",
@@ -618,21 +732,51 @@ int open_file(char *file) {
   return EXIT_SUCCESS;
 }
 
+Av1DecodeReturn adr;
+int have_frame = 0;
+const unsigned char *frame;
+const unsigned char *end_frame;
+size_t frame_size = 0;
+
 EMSCRIPTEN_KEEPALIVE
 int read_frame() {
-  if (!aom_video_reader_read_frame(reader)) return EXIT_FAILURE;
   img = NULL;
-  aom_codec_iter_t iter = NULL;
-  size_t frame_size = 0;
-  const unsigned char *frame = aom_video_reader_get_frame(reader, &frame_size);
-  if (aom_codec_decode(&codec, frame, (unsigned int)frame_size, NULL) !=
-      AOM_CODEC_OK) {
-    die_codec(&codec, "Failed to decode frame.");
-  }
+
+  // This loop skips over any frames that are show_existing_frames,  as
+  // there is nothing to analyze.
+  do {
+    if (!have_frame) {
+      if (!aom_video_reader_read_frame(reader)) return EXIT_FAILURE;
+      frame = aom_video_reader_get_frame(reader, &frame_size);
+
+      have_frame = 1;
+      end_frame = frame + frame_size;
+    }
+
+    if (aom_codec_decode(&codec, frame, (unsigned int)frame_size, &adr) !=
+        AOM_CODEC_OK) {
+      die_codec(&codec, "Failed to decode frame.");
+    }
+
+    frame = adr.buf;
+    if (frame == end_frame) have_frame = 0;
+  } while (adr.show_existing);
+
   int got_any_frames = 0;
   aom_image_t *frame_img;
-  while ((frame_img = aom_codec_get_frame(&codec, &iter))) {
-    img = frame_img;
+  struct av1_ref_frame ref_dec;
+  ref_dec.idx = adr.idx;
+
+  // ref_dec.idx is the index to the reference buffer idx to AV1_GET_REFERENCE
+  // if its -1 the decoder didn't update any reference buffer and the only
+  // way to see the frame is aom_codec_get_frame.
+  if (ref_dec.idx == -1) {
+    aom_codec_iter_t iter = NULL;
+    img = frame_img = aom_codec_get_frame(&codec, &iter);
+    ++frame_count;
+    got_any_frames = 1;
+  } else if (!aom_codec_control(&codec, AV1_GET_REFERENCE, &ref_dec)) {
+    img = frame_img = &ref_dec.img;
     ++frame_count;
     got_any_frames = 1;
   }
@@ -692,6 +836,10 @@ static void parse_args(char **argv) {
       layers |= MODE_LAYER;
     else if (arg_match(&arg, &dump_uv_mode_arg, argi))
       layers |= UV_MODE_LAYER;
+    else if (arg_match(&arg, &dump_motion_mode_arg, argi))
+      layers |= MOTION_MODE_LAYER;
+    else if (arg_match(&arg, &dump_compound_type_arg, argi))
+      layers |= COMPOUND_TYPE_LAYER;
     else if (arg_match(&arg, &dump_skip_arg, argi))
       layers |= SKIP_LAYER;
     else if (arg_match(&arg, &dump_filter_arg, argi))
@@ -718,6 +866,13 @@ static void parse_args(char **argv) {
       usage_exit();
     else if (arg_match(&arg, &limit_arg, argi))
       stop_after = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &skip_non_transform_arg, argi))
+      skip_non_transform = arg_parse_uint(&arg);
+    else if (arg_match(&arg, &combined_arg, argi))
+      convert_to_indices(
+          (char *)arg.val, combined_parm_list,
+          sizeof(combined_parm_list) / sizeof(combined_parm_list[0]),
+          &combined_parm_count);
     else
       argj++;
   }
diff --git a/third_party/aom/examples/lightfield_bitstream_parsing.c b/third_party/aom/examples/lightfield_bitstream_parsing.c
index 159f1617a3..afacf44043 100644
--- a/third_party/aom/examples/lightfield_bitstream_parsing.c
+++ b/third_party/aom/examples/lightfield_bitstream_parsing.c
@@ -13,13 +13,28 @@
 // ============================
 //
 // This is a lightfield bitstream parsing example. It takes an input file
-// containing the whole compressed lightfield bitstream(ivf file), and parses it
-// and constructs and outputs a new bitstream that can be decoded by an AV1
-// decoder. The output bitstream contains reference frames(i.e. anchor frames),
-// camera frame header, and tile list OBUs. num_references is the number of
-// anchor frames coded at the beginning of the light field file.
-// After running the lightfield encoder, run lightfield bitstream parsing:
+// containing the whole compressed lightfield bitstream(ivf file) and a text
+// file containing a stream of tiles to decode and then constructs and outputs
+// a new bitstream that can be decoded by an AV1 decoder. The output bitstream
+// contains reference frames(i.e. anchor frames), camera frame header, and
+// tile list OBUs. num_references is the number of anchor frames coded at the
+// beginning of the light field file.  After running the lightfield encoder,
+// run lightfield bitstream parsing:
 // examples/lightfield_bitstream_parsing vase10x10.ivf vase_tile_list.ivf 4
+//   tile_list.txt
+//
+// The tile_list.txt is expected to be of the form:
+// Frame <frame_index0>
+// <image_index0> <anchor_index0> <tile_col0> <tile_row0>
+// <image_index1> <anchor_index1> <tile_col1> <tile_row1>
+// ...
+// Frame <frame_index1)
+// ...
+//
+// The "Frame" markers indicate a new render frame and thus a new tile list
+// will be started and the old one flushed.  The image_indexN, anchor_indexN,
+// tile_colN, and tile_rowN identify an individual tile to be decoded and
+// to use anchor_indexN anchor image for MCP.
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -39,7 +54,7 @@
 static const char *exec_name;
 
 void usage_exit(void) {
-  fprintf(stderr, "Usage: %s <infile> <outfile> <num_references> \n",
+  fprintf(stderr, "Usage: %s <infile> <outfile> <num_references> <tile_list>\n",
           exec_name);
   exit(EXIT_FAILURE);
 }
@@ -47,9 +62,8 @@ void usage_exit(void) {
 #define ALIGN_POWER_OF_TWO(value, n) \
   (((value) + ((1 << (n)) - 1)) & ~((1 << (n)) - 1))
 
-// SB size: 64x64
-const uint8_t output_frame_width_in_tiles_minus_1 = 512 / 64 - 1;
-const uint8_t output_frame_height_in_tiles_minus_1 = 512 / 64 - 1;
+const int output_frame_width = 512;
+const int output_frame_height = 512;
 
 // Spec:
 // typedef struct {
@@ -68,32 +82,6 @@ typedef struct {
   int tile_row;
 } TILE_LIST_INFO;
 
-// M references: 0 - M-1; N images(including references): 0 - N-1;
-// Note: order the image index incrementally, so that we only go through the
-// bitstream once to construct the tile list.
-const int num_tile_lists = 2;
-const uint16_t tile_count_minus_1 = 9 - 1;
-const TILE_LIST_INFO tile_list[2][9] = {
-  { { 16, 0, 4, 5 },
-    { 83, 3, 13, 2 },
-    { 57, 2, 2, 6 },
-    { 31, 1, 11, 5 },
-    { 2, 0, 7, 4 },
-    { 77, 3, 9, 9 },
-    { 49, 1, 0, 1 },
-    { 6, 0, 3, 10 },
-    { 63, 2, 5, 8 } },
-  { { 65, 2, 11, 1 },
-    { 42, 1, 3, 7 },
-    { 88, 3, 8, 4 },
-    { 76, 3, 1, 15 },
-    { 1, 0, 2, 2 },
-    { 19, 0, 5, 6 },
-    { 60, 2, 4, 0 },
-    { 25, 1, 11, 15 },
-    { 50, 2, 5, 4 } },
-};
-
 static int get_image_bps(aom_img_fmt_t fmt) {
   switch (fmt) {
     case AOM_IMG_FMT_I420: return 12;
@@ -107,6 +95,102 @@ static int get_image_bps(aom_img_fmt_t fmt) {
   return 0;
 }
 
+void process_tile_list(const TILE_LIST_INFO *tiles, int num_tiles,
+                       aom_codec_pts_t tl_pts, unsigned char **frames,
+                       const size_t *frame_sizes, aom_codec_ctx_t *codec,
+                       unsigned char *tl_buf, AvxVideoWriter *writer,
+                       uint8_t output_frame_width_in_tiles_minus_1,
+                       uint8_t output_frame_height_in_tiles_minus_1) {
+  unsigned char *tl = tl_buf;
+  struct aom_write_bit_buffer wb = { tl, 0 };
+  unsigned char *saved_obu_size_loc = NULL;
+  uint32_t tile_list_obu_header_size = 0;
+  uint32_t tile_list_obu_size = 0;
+  int num_tiles_minus_1 = num_tiles - 1;
+  int i;
+
+  // Write the tile list OBU header that is 1 byte long.
+  aom_wb_write_literal(&wb, 0, 1);  // forbidden bit.
+  aom_wb_write_literal(&wb, 8, 4);  // tile list OBU: "1000"
+  aom_wb_write_literal(&wb, 0, 1);  // obu_extension = 0
+  aom_wb_write_literal(&wb, 1, 1);  // obu_has_size_field
+  aom_wb_write_literal(&wb, 0, 1);  // reserved
+  tl++;
+  tile_list_obu_header_size++;
+
+  // Write the OBU size using a fixed length_field_size of 4 bytes.
+  saved_obu_size_loc = tl;
+  // aom_wb_write_unsigned_literal(&wb, data, bits) requires that bits <= 32.
+  aom_wb_write_unsigned_literal(&wb, 0, 32);
+  tl += 4;
+  tile_list_obu_header_size += 4;
+
+  // write_tile_list_obu()
+  aom_wb_write_literal(&wb, output_frame_width_in_tiles_minus_1, 8);
+  aom_wb_write_literal(&wb, output_frame_height_in_tiles_minus_1, 8);
+  aom_wb_write_literal(&wb, num_tiles_minus_1, 16);
+  tl += 4;
+  tile_list_obu_size += 4;
+
+  // Write each tile's data
+  for (i = 0; i <= num_tiles_minus_1; i++) {
+    aom_tile_data tile_data = { 0, NULL, 0 };
+
+    int image_idx = tiles[i].image_idx;
+    int ref_idx = tiles[i].reference_idx;
+    int tc = tiles[i].tile_col;
+    int tr = tiles[i].tile_row;
+
+    // Reset bit writer to the right location.
+    wb.bit_buffer = tl;
+    wb.bit_offset = 0;
+
+    size_t frame_size = frame_sizes[image_idx];
+    const unsigned char *frame = frames[image_idx];
+
+    aom_codec_control_(codec, AV1_SET_DECODE_TILE_ROW, tr);
+    aom_codec_control_(codec, AV1_SET_DECODE_TILE_COL, tc);
+
+    aom_codec_err_t aom_status =
+        aom_codec_decode(codec, frame, frame_size, NULL);
+    if (aom_status) die_codec(codec, "Failed to decode tile.");
+
+    aom_codec_control_(codec, AV1D_GET_TILE_DATA, &tile_data);
+
+    // Copy over tile info.
+    //  uint8_t anchor_frame_idx;
+    //  uint8_t tile_row;
+    //  uint8_t tile_col;
+    //  uint16_t coded_tile_data_size_minus_1;
+    //  uint8_t *coded_tile_data;
+    uint32_t tile_info_bytes = 5;
+    aom_wb_write_literal(&wb, ref_idx, 8);
+    aom_wb_write_literal(&wb, tr, 8);
+    aom_wb_write_literal(&wb, tc, 8);
+    aom_wb_write_literal(&wb, (int)tile_data.coded_tile_data_size - 1, 16);
+    tl += tile_info_bytes;
+
+    memcpy(tl, (uint8_t *)tile_data.coded_tile_data,
+           tile_data.coded_tile_data_size);
+    tl += tile_data.coded_tile_data_size;
+
+    tile_list_obu_size +=
+        tile_info_bytes + (uint32_t)tile_data.coded_tile_data_size;
+  }
+
+  // Write tile list OBU size.
+  size_t bytes_written = 0;
+  if (aom_uleb_encode_fixed_size(tile_list_obu_size, 4, 4, saved_obu_size_loc,
+                                 &bytes_written))
+    die_codec(codec, "Failed to encode the tile list obu size.");
+
+  // Copy the tile list.
+  if (!aom_video_writer_write_frame(
+          writer, tl_buf, tile_list_obu_header_size + tile_list_obu_size,
+          tl_pts))
+    die_codec(codec, "Failed to copy compressed tile list.");
+}
+
 int main(int argc, char **argv) {
   aom_codec_ctx_t codec;
   AvxVideoReader *reader = NULL;
@@ -114,11 +198,12 @@ int main(int argc, char **argv) {
   const AvxInterface *decoder = NULL;
   const AvxVideoInfo *info = NULL;
   int num_references;
-  int n, i;
+  int i;
   aom_codec_pts_t pts;
+  const char *tile_list_file = NULL;
 
   exec_name = argv[0];
-  if (argc != 4) die("Invalid number of arguments.");
+  if (argc != 5) die("Invalid number of arguments.");
 
   reader = aom_video_reader_open(argv[1]);
   if (!reader) die("Failed to open %s for reading.", argv[1]);
@@ -126,11 +211,15 @@ int main(int argc, char **argv) {
   num_references = (int)strtol(argv[3], NULL, 0);
   info = aom_video_reader_get_info(reader);
 
+  aom_video_reader_set_fourcc(reader, AV1_FOURCC);
+
   // The writer to write out ivf file in tile list OBU, which can be decoded by
   // AV1 decoder.
   writer = aom_video_writer_open(argv[2], kContainerIVF, info);
   if (!writer) die("Failed to open %s for writing", argv[2]);
 
+  tile_list_file = argv[4];
+
   decoder = get_aom_decoder_by_fourcc(info->codec_fourcc);
   if (!decoder) die("Unknown input codec.");
   printf("Using %s\n", aom_codec_iface_name(decoder->codec_interface()));
@@ -141,6 +230,7 @@ int main(int argc, char **argv) {
   // Decode anchor frames.
   aom_codec_control_(&codec, AV1_SET_TILE_MODE, 0);
 
+  printf("Reading %d reference images.\n", num_references);
   for (i = 0; i < num_references; ++i) {
     aom_video_reader_read_frame(reader);
 
@@ -165,16 +255,38 @@ int main(int argc, char **argv) {
   // Record the offset of the first camera image.
   const FileOffset camera_frame_pos = ftello(infile);
 
-  // Read out the first camera frame.
-  aom_video_reader_read_frame(reader);
+  printf("Loading compressed frames into memory.\n");
+
+  // Count the frames in the lightfield.
+  int num_frames = 0;
+  while (aom_video_reader_read_frame(reader)) {
+    ++num_frames;
+  }
+  if (num_frames < 1) die("Input light field has no frames.");
+
+  // Read all of the lightfield frames into memory.
+  unsigned char **frames =
+      (unsigned char **)malloc(num_frames * sizeof(unsigned char *));
+  size_t *frame_sizes = (size_t *)malloc(num_frames * sizeof(size_t));
+  // Seek to the first camera image.
+  fseeko(infile, camera_frame_pos, SEEK_SET);
+  for (int f = 0; f < num_frames; ++f) {
+    aom_video_reader_read_frame(reader);
+    size_t frame_size = 0;
+    const unsigned char *frame =
+        aom_video_reader_get_frame(reader, &frame_size);
+    frames[f] = (unsigned char *)malloc(frame_size * sizeof(unsigned char));
+    memcpy(frames[f], frame, frame_size);
+    frame_sizes[f] = frame_size;
+  }
+  printf("Read %d frames.\n", num_frames);
 
   // Copy first camera frame for getting camera frame header. This is done
   // only once.
   {
-    size_t frame_size = 0;
-    const unsigned char *frame =
-        aom_video_reader_get_frame(reader, &frame_size);
-    pts = (aom_codec_pts_t)aom_video_reader_get_frame_pts(reader);
+    size_t frame_size = frame_sizes[0];
+    const unsigned char *frame = frames[0];
+    pts = num_references;
     aom_tile_data frame_header_info = { 0, NULL, 0 };
 
     // Need to decode frame header to get camera frame header info. So, here
@@ -231,115 +343,68 @@ int main(int argc, char **argv) {
   // Allocate a buffer to store tile list bitstream.
   const size_t data_sz = MAX_TILES * ALIGN_POWER_OF_TWO(tile_width, 5) *
                          ALIGN_POWER_OF_TWO(tile_height, 5) * bps / 8;
+
   unsigned char *tl_buf = (unsigned char *)malloc(data_sz);
   if (tl_buf == NULL) die_codec(&codec, "Failed to allocate tile list buffer.");
 
-  aom_codec_pts_t tl_pts = pts;
+  aom_codec_pts_t tl_pts = num_references;
+  const uint8_t output_frame_width_in_tiles_minus_1 =
+      output_frame_width / tile_width - 1;
+  const uint8_t output_frame_height_in_tiles_minus_1 =
+      output_frame_height / tile_height - 1;
 
-  // Process 1 tile list.
-  for (n = 0; n < num_tile_lists; n++) {
-    unsigned char *tl = tl_buf;
-    struct aom_write_bit_buffer wb = { tl, 0 };
-    unsigned char *saved_obu_size_loc = NULL;
-    uint32_t tile_list_obu_header_size = 0;
-    uint32_t tile_list_obu_size = 0;
-
-    // Write the tile list OBU header that is 1 byte long.
-    aom_wb_write_literal(&wb, 0, 1);  // forbidden bit.
-    aom_wb_write_literal(&wb, 8, 4);  // tile list OBU: "1000"
-    aom_wb_write_literal(&wb, 0, 1);  // obu_extension = 0
-    aom_wb_write_literal(&wb, 1, 1);  // obu_has_size_field
-    aom_wb_write_literal(&wb, 0, 1);  // reserved
-    tl++;
-    tile_list_obu_header_size++;
-
-    // Write the OBU size using a fixed length_field_size of 4 bytes.
-    saved_obu_size_loc = tl;
-    // aom_wb_write_unsigned_literal(&wb, data, bits) requires that bits <= 32.
-    aom_wb_write_unsigned_literal(&wb, 0, 32);
-    tl += 4;
-    tile_list_obu_header_size += 4;
-
-    // write_tile_list_obu()
-    aom_wb_write_literal(&wb, output_frame_width_in_tiles_minus_1, 8);
-    aom_wb_write_literal(&wb, output_frame_height_in_tiles_minus_1, 8);
-    aom_wb_write_literal(&wb, tile_count_minus_1, 16);
-    tl += 4;
-    tile_list_obu_size += 4;
-
-    // Write each tile's data
-    for (i = 0; i <= tile_count_minus_1; i++) {
-      aom_tile_data tile_data = { 0, NULL, 0 };
-
-      int image_idx = tile_list[n][i].image_idx;
-      int ref_idx = tile_list[n][i].reference_idx;
-      int tc = tile_list[n][i].tile_col;
-      int tr = tile_list[n][i].tile_row;
-      int frame_cnt = -1;
-
-      // Reset bit writer to the right location.
-      wb.bit_buffer = tl;
-      wb.bit_offset = 0;
-
-      // Seek to the first camera image.
-      fseeko(infile, camera_frame_pos, SEEK_SET);
-
-      // Read out the camera image
-      while (frame_cnt != image_idx) {
-        aom_video_reader_read_frame(reader);
-        frame_cnt++;
+  printf("Reading tile list from file.\n");
+  char line[1024];
+  FILE *tile_list_fptr = fopen(tile_list_file, "r");
+  if (!tile_list_fptr) die_codec(&codec, "Failed to open tile list file.");
+  int num_tiles = 0;
+  TILE_LIST_INFO tiles[MAX_TILES];
+  while ((fgets(line, 1024, tile_list_fptr)) != NULL) {
+    if (line[0] == 'F' || num_tiles >= MAX_TILES) {
+      // Flush existing tile list and start another, either because we hit a
+      // new render frame or because we've hit our max number of tiles per list.
+      if (num_tiles > 0) {
+        process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec,
+                          tl_buf, writer, output_frame_width_in_tiles_minus_1,
+                          output_frame_height_in_tiles_minus_1);
+        ++tl_pts;
       }
-
-      size_t frame_size = 0;
-      const unsigned char *frame =
-          aom_video_reader_get_frame(reader, &frame_size);
-
-      aom_codec_control_(&codec, AV1_SET_DECODE_TILE_ROW, tr);
-      aom_codec_control_(&codec, AV1_SET_DECODE_TILE_COL, tc);
-
-      aom_codec_err_t aom_status =
-          aom_codec_decode(&codec, frame, frame_size, NULL);
-      if (aom_status) die_codec(&codec, "Failed to decode tile.");
-
-      aom_codec_control_(&codec, AV1D_GET_TILE_DATA, &tile_data);
-
-      // Copy over tile info.
-      //  uint8_t anchor_frame_idx;
-      //  uint8_t tile_row;
-      //  uint8_t tile_col;
-      //  uint16_t coded_tile_data_size_minus_1;
-      //  uint8_t *coded_tile_data;
-      uint32_t tile_info_bytes = 5;
-      aom_wb_write_literal(&wb, ref_idx, 8);
-      aom_wb_write_literal(&wb, tr, 8);
-      aom_wb_write_literal(&wb, tc, 8);
-      aom_wb_write_literal(&wb, (int)tile_data.coded_tile_data_size - 1, 16);
-      tl += tile_info_bytes;
-
-      memcpy(tl, (uint8_t *)tile_data.coded_tile_data,
-             tile_data.coded_tile_data_size);
-      tl += tile_data.coded_tile_data_size;
-
-      tile_list_obu_size +=
-          tile_info_bytes + (uint32_t)tile_data.coded_tile_data_size;
+      num_tiles = 0;
     }
-
-    // Write tile list OBU size.
-    size_t bytes_written = 0;
-    if (aom_uleb_encode_fixed_size(tile_list_obu_size, 4, 4, saved_obu_size_loc,
-                                   &bytes_written))
-      die_codec(&codec, "Failed to encode the tile list obu size.");
-
-    // Copy the tile list.
-    if (!aom_video_writer_write_frame(
-            writer, tl_buf, tile_list_obu_header_size + tile_list_obu_size,
-            tl_pts))
-      die_codec(&codec, "Failed to copy compressed tile list.");
-
-    tl_pts++;
+    if (line[0] == 'F') {
+      continue;
+    }
+    if (sscanf(line, "%d %d %d %d", &tiles[num_tiles].image_idx,
+               &tiles[num_tiles].reference_idx, &tiles[num_tiles].tile_col,
+               &tiles[num_tiles].tile_row) == 4) {
+      if (tiles[num_tiles].image_idx >= num_frames) {
+        die("Tile list image_idx out of bounds: %d >= %d.",
+            tiles[num_tiles].image_idx, num_frames);
+      }
+      if (tiles[num_tiles].reference_idx >= num_references) {
+        die("Tile list reference_idx out of bounds: %d >= %d.",
+            tiles[num_tiles].reference_idx, num_references);
+      }
+      ++num_tiles;
+    }
+  }
+  if (num_tiles > 0) {
+    // Flush out the last tile list.
+    process_tile_list(tiles, num_tiles, tl_pts, frames, frame_sizes, &codec,
+                      tl_buf, writer, output_frame_width_in_tiles_minus_1,
+                      output_frame_height_in_tiles_minus_1);
+    ++tl_pts;
   }
 
+  const int num_tile_lists = (int)(tl_pts - pts);
+  printf("Finished processing tile lists.  Num tile lists: %d.\n",
+         num_tile_lists);
   free(tl_buf);
+  for (int f = 0; f < num_frames; ++f) {
+    free(frames[f]);
+  }
+  free(frame_sizes);
+  free(frames);
   if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
   aom_video_writer_close(writer);
   aom_video_reader_close(reader);
diff --git a/third_party/aom/examples/lightfield_decoder.c b/third_party/aom/examples/lightfield_decoder.c
index f5e54db7f1..7a445f003f 100644
--- a/third_party/aom/examples/lightfield_decoder.c
+++ b/third_party/aom/examples/lightfield_decoder.c
@@ -14,10 +14,26 @@
 //
 // This is an example of a simple lightfield decoder. It builds upon the
 // simple_decoder.c example.  It takes an input file containing the compressed
-// data (in ivf format), treating it as a lightfield instead of a video.
+// data (in ivf format), treating it as a lightfield instead of a video; and a
+// text file with a list of tiles to decode. There is an optional parameter
+// allowing to choose the output format, and the supported formats are
+// YUV1D(default), YUV, and NV12.
 // After running the lightfield encoder, run lightfield decoder to decode a
 // batch of tiles:
-// examples/lightfield_decoder vase10x10.ivf vase_reference.yuv 4
+// examples/lightfield_decoder vase10x10.ivf vase_reference.yuv 4 tile_list.txt
+// 0(optional)
+// The tile_list.txt is expected to be of the form:
+// Frame <frame_index0>
+// <image_index0> <anchor_index0> <tile_col0> <tile_row0>
+// <image_index1> <anchor_index1> <tile_col1> <tile_row1>
+// ...
+// Frame <frame_index1)
+// ...
+//
+// The "Frame" markers indicate a new render frame and thus a new tile list
+// will be started and the old one flushed.  The image_indexN, anchor_indexN,
+// tile_colN, and tile_rowN identify an individual tile to be decoded and
+// to use anchor_indexN anchor image for MCP.
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -33,43 +49,109 @@
 static const char *exec_name;
 
 void usage_exit(void) {
-  fprintf(stderr, "Usage: %s <infile> <outfile> <num_references>\n", exec_name);
+  fprintf(stderr,
+          "Usage: %s <infile> <outfile> <num_references> <tile_list> <output "
+          "format(optional)>\n",
+          exec_name);
   exit(EXIT_FAILURE);
 }
 
-// Tile list entry provided by the application
-typedef struct {
-  int image_idx;
-  int reference_idx;
-  int tile_col;
-  int tile_row;
-} TILE_LIST_INFO;
+// Output frame size
+const int output_frame_width = 512;
+const int output_frame_height = 512;
 
-// M references: 0 - M-1; N images(including references): 0 - N-1;
-// Note: order the image index incrementally, so that we only go through the
-// bitstream once to construct the tile list.
-const int num_tile_lists = 2;
-const uint16_t tile_count_minus_1 = 9 - 1;
-const TILE_LIST_INFO tile_list[2][9] = {
-  { { 16, 0, 4, 5 },
-    { 83, 3, 13, 2 },
-    { 57, 2, 2, 6 },
-    { 31, 1, 11, 5 },
-    { 2, 0, 7, 4 },
-    { 77, 3, 9, 9 },
-    { 49, 1, 0, 1 },
-    { 6, 0, 3, 10 },
-    { 63, 2, 5, 8 } },
-  { { 65, 2, 11, 1 },
-    { 42, 1, 3, 7 },
-    { 88, 3, 8, 4 },
-    { 76, 3, 1, 15 },
-    { 1, 0, 2, 2 },
-    { 19, 0, 5, 6 },
-    { 60, 2, 4, 0 },
-    { 25, 1, 11, 15 },
-    { 50, 2, 5, 4 } },
-};
+static void aom_img_copy_tile(const aom_image_t *src, const aom_image_t *dst,
+                              int dst_row_offset, int dst_col_offset) {
+  const int shift = (src->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 1 : 0;
+  int plane;
+
+  for (plane = 0; plane < 3; ++plane) {
+    const unsigned char *src_buf = src->planes[plane];
+    const int src_stride = src->stride[plane];
+    unsigned char *dst_buf = dst->planes[plane];
+    const int dst_stride = dst->stride[plane];
+    const int roffset =
+        (plane > 0) ? dst_row_offset >> dst->y_chroma_shift : dst_row_offset;
+    const int coffset =
+        (plane > 0) ? dst_col_offset >> dst->x_chroma_shift : dst_col_offset;
+
+    // col offset needs to be adjusted for HBD.
+    dst_buf += roffset * dst_stride + (coffset << shift);
+
+    const int w = (aom_img_plane_width(src, plane) << shift);
+    const int h = aom_img_plane_height(src, plane);
+    int y;
+
+    for (y = 0; y < h; ++y) {
+      memcpy(dst_buf, src_buf, w);
+      src_buf += src_stride;
+      dst_buf += dst_stride;
+    }
+  }
+}
+
+void decode_tile(aom_codec_ctx_t *codec, const unsigned char *frame,
+                 size_t frame_size, int tr, int tc, int ref_idx,
+                 aom_image_t *reference_images, aom_image_t *output,
+                 int *tile_idx, unsigned int *output_bit_depth,
+                 aom_image_t **img_ptr, int output_format) {
+  aom_codec_control_(codec, AV1_SET_TILE_MODE, 1);
+  aom_codec_control_(codec, AV1D_EXT_TILE_DEBUG, 1);
+  aom_codec_control_(codec, AV1_SET_DECODE_TILE_ROW, tr);
+  aom_codec_control_(codec, AV1_SET_DECODE_TILE_COL, tc);
+
+  av1_ref_frame_t ref;
+  ref.idx = 0;
+  ref.use_external_ref = 1;
+  ref.img = reference_images[ref_idx];
+  if (aom_codec_control(codec, AV1_SET_REFERENCE, &ref)) {
+    die_codec(codec, "Failed to set reference frame.");
+  }
+
+  aom_codec_err_t aom_status = aom_codec_decode(codec, frame, frame_size, NULL);
+  if (aom_status) die_codec(codec, "Failed to decode tile.");
+
+  aom_codec_iter_t iter = NULL;
+  aom_image_t *img = aom_codec_get_frame(codec, &iter);
+  if (!img) die_codec(codec, "Failed to get frame.");
+  *img_ptr = img;
+
+  // aom_img_alloc() sets bit_depth as follows:
+  // output->bit_depth = (fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 16 : 8;
+  // Use img->bit_depth(read from bitstream), so that aom_shift_img()
+  // works as expected.
+  output->bit_depth = img->bit_depth;
+  *output_bit_depth = img->bit_depth;
+
+  if (output_format != YUV1D) {
+    // read out the tile size.
+    unsigned int tile_size = 0;
+    if (aom_codec_control(codec, AV1D_GET_TILE_SIZE, &tile_size))
+      die_codec(codec, "Failed to get the tile size");
+    const unsigned int tile_width = tile_size >> 16;
+    const unsigned int tile_height = tile_size & 65535;
+    const uint8_t output_frame_width_in_tiles = output_frame_width / tile_width;
+
+    // Copy the tile to the output frame.
+    const int row_offset =
+        (*tile_idx / output_frame_width_in_tiles) * tile_height;
+    const int col_offset =
+        (*tile_idx % output_frame_width_in_tiles) * tile_width;
+
+    aom_img_copy_tile(img, output, row_offset, col_offset);
+    (*tile_idx)++;
+  }
+}
+
+static void img_write_to_file(const aom_image_t *img, FILE *file,
+                              int output_format) {
+  if (output_format == YUV)
+    aom_img_write(img, file);
+  else if (output_format == NV12)
+    aom_img_write_nv12(img, file);
+  else
+    die("Invalid output format");
+}
 
 int main(int argc, char **argv) {
   FILE *outfile = NULL;
@@ -78,13 +160,18 @@ int main(int argc, char **argv) {
   const AvxInterface *decoder = NULL;
   const AvxVideoInfo *info = NULL;
   int num_references;
+  aom_img_fmt_t ref_fmt = 0;
   aom_image_t reference_images[MAX_EXTERNAL_REFERENCES];
+  aom_image_t output;
+  aom_image_t *output_shifted = NULL;
   size_t frame_size = 0;
   const unsigned char *frame = NULL;
-  int n, i, j;
+  int i, j;
+  const char *tile_list_file = NULL;
+  int output_format = YUV1D;
   exec_name = argv[0];
 
-  if (argc != 4) die("Invalid number of arguments.");
+  if (argc < 5) die("Invalid number of arguments.");
 
   reader = aom_video_reader_open(argv[1]);
   if (!reader) die("Failed to open %s for reading.", argv[1]);
@@ -93,11 +180,18 @@ int main(int argc, char **argv) {
     die("Failed to open %s for writing.", argv[2]);
 
   num_references = (int)strtol(argv[3], NULL, 0);
+  tile_list_file = argv[4];
+
+  if (argc > 5) output_format = (int)strtol(argv[5], NULL, 0);
+  if (output_format < YUV1D || output_format > NV12)
+    die("Output format out of range [0, 2]");
 
   info = aom_video_reader_get_info(reader);
 
-  decoder = get_aom_decoder_by_fourcc(info->codec_fourcc);
-  if (!decoder) die("Unknown input codec.");
+  if (info->codec_fourcc == LST_FOURCC)
+    decoder = get_aom_decoder_by_fourcc(AV1_FOURCC);
+  else
+    die("Unknown input codec.");
   printf("Using %s\n", aom_codec_iface_name(decoder->codec_interface()));
 
   if (aom_codec_dec_init(&codec, decoder->codec_interface(), NULL, 0))
@@ -116,7 +210,6 @@ int main(int argc, char **argv) {
       die_codec(&codec, "Failed to decode frame.");
 
     if (i == 0) {
-      aom_img_fmt_t ref_fmt = 0;
       if (aom_codec_control(&codec, AV1D_GET_IMG_FORMAT, &ref_fmt))
         die_codec(&codec, "Failed to get the image format");
 
@@ -127,7 +220,7 @@ int main(int argc, char **argv) {
       // Allocate memory to store decoded references. Allocate memory with the
       // border so that it can be used as a reference.
       for (j = 0; j < num_references; j++) {
-        unsigned int border = AOM_BORDER_IN_PIXELS;
+        unsigned int border = AOM_DEC_BORDER_IN_PIXELS;
         if (!aom_img_alloc_with_border(&reference_images[j], ref_fmt,
                                        frame_res[0], frame_res[1], 32, 8,
                                        border)) {
@@ -156,50 +249,110 @@ int main(int argc, char **argv) {
   // Record the offset of the first camera image.
   const FileOffset camera_frame_pos = ftello(infile);
 
-  // Process 1 tile.
-  for (n = 0; n < num_tile_lists; n++) {
-    for (i = 0; i <= tile_count_minus_1; i++) {
-      int image_idx = tile_list[n][i].image_idx;
-      int ref_idx = tile_list[n][i].reference_idx;
-      int tc = tile_list[n][i].tile_col;
-      int tr = tile_list[n][i].tile_row;
-      int frame_cnt = -1;
+  printf("Loading compressed frames into memory.\n");
 
-      // Seek to the first camera image.
-      fseeko(infile, camera_frame_pos, SEEK_SET);
+  // Count the frames in the lightfield.
+  int num_frames = 0;
+  while (aom_video_reader_read_frame(reader)) {
+    ++num_frames;
+  }
+  if (num_frames < 1) die("Input light field has no frames.");
 
-      // Read out the camera image
-      while (frame_cnt != image_idx) {
-        aom_video_reader_read_frame(reader);
-        frame_cnt++;
+  // Read all of the lightfield frames into memory.
+  unsigned char **frames =
+      (unsigned char **)malloc(num_frames * sizeof(unsigned char *));
+  size_t *frame_sizes = (size_t *)malloc(num_frames * sizeof(size_t));
+  // Seek to the first camera image.
+  fseeko(infile, camera_frame_pos, SEEK_SET);
+  for (int f = 0; f < num_frames; ++f) {
+    aom_video_reader_read_frame(reader);
+    frame = aom_video_reader_get_frame(reader, &frame_size);
+    frames[f] = (unsigned char *)malloc(frame_size * sizeof(unsigned char));
+    memcpy(frames[f], frame, frame_size);
+    frame_sizes[f] = frame_size;
+  }
+  printf("Read %d frames.\n", num_frames);
+
+  if (output_format != YUV1D) {
+    // Allocate the output frame.
+    aom_img_fmt_t out_fmt = ref_fmt;
+    if (!CONFIG_LOWBITDEPTH) out_fmt |= AOM_IMG_FMT_HIGHBITDEPTH;
+    if (!aom_img_alloc(&output, out_fmt, output_frame_width,
+                       output_frame_height, 32))
+      die("Failed to allocate output image.");
+  }
+
+  printf("Decoding tile list from file.\n");
+  char line[1024];
+  FILE *tile_list_fptr = fopen(tile_list_file, "r");
+  int tile_list_cnt = 0;
+  int tile_list_writes = 0;
+  int tile_idx = 0;
+  aom_image_t *out = NULL;
+  unsigned int output_bit_depth = 0;
+
+  while ((fgets(line, 1024, tile_list_fptr)) != NULL) {
+    if (line[0] == 'F') {
+      if (output_format != YUV1D) {
+        // Write out the tile list.
+        if (tile_list_cnt) {
+          out = &output;
+          if (output_bit_depth != 0)
+            aom_shift_img(output_bit_depth, &out, &output_shifted);
+          img_write_to_file(out, outfile, output_format);
+          tile_list_writes++;
+        }
+
+        tile_list_cnt++;
+        tile_idx = 0;
+        // Then memset the frame.
+        memset(output.img_data, 0, output.sz);
       }
+      continue;
+    }
 
-      frame = aom_video_reader_get_frame(reader, &frame_size);
+    int image_idx, ref_idx, tc, tr;
+    sscanf(line, "%d %d %d %d", &image_idx, &ref_idx, &tc, &tr);
+    if (image_idx >= num_frames) {
+      die("Tile list image_idx out of bounds: %d >= %d.", image_idx,
+          num_frames);
+    }
+    if (ref_idx >= num_references) {
+      die("Tile list ref_idx out of bounds: %d >= %d.", ref_idx,
+          num_references);
+    }
+    frame = frames[image_idx];
+    frame_size = frame_sizes[image_idx];
 
-      aom_codec_control_(&codec, AV1_SET_TILE_MODE, 1);
-      aom_codec_control_(&codec, AV1D_EXT_TILE_DEBUG, 1);
-      aom_codec_control_(&codec, AV1_SET_DECODE_TILE_ROW, tr);
-      aom_codec_control_(&codec, AV1_SET_DECODE_TILE_COL, tc);
-
-      av1_ref_frame_t ref;
-      ref.idx = 0;
-      ref.use_external_ref = 1;
-      ref.img = reference_images[ref_idx];
-      if (aom_codec_control(&codec, AV1_SET_REFERENCE, &ref)) {
-        die_codec(&codec, "Failed to set reference frame.");
-      }
-
-      aom_codec_err_t aom_status =
-          aom_codec_decode(&codec, frame, frame_size, NULL);
-      if (aom_status) die_codec(&codec, "Failed to decode tile.");
-
-      aom_codec_iter_t iter = NULL;
-      aom_image_t *img = aom_codec_get_frame(&codec, &iter);
-      aom_img_write(img, outfile);
+    aom_image_t *img = NULL;
+    decode_tile(&codec, frame, frame_size, tr, tc, ref_idx, reference_images,
+                &output, &tile_idx, &output_bit_depth, &img, output_format);
+    if (output_format == YUV1D) {
+      out = img;
+      if (output_bit_depth != 0)
+        aom_shift_img(output_bit_depth, &out, &output_shifted);
+      aom_img_write(out, outfile);
     }
   }
 
+  if (output_format != YUV1D) {
+    // Write out the last tile list.
+    if (tile_list_writes < tile_list_cnt) {
+      out = &output;
+      if (output_bit_depth != 0)
+        aom_shift_img(output_bit_depth, &out, &output_shifted);
+      img_write_to_file(out, outfile, output_format);
+    }
+  }
+
+  if (output_shifted) aom_img_free(output_shifted);
+  if (output_format != YUV1D) aom_img_free(&output);
   for (i = 0; i < num_references; i++) aom_img_free(&reference_images[i]);
+  for (int f = 0; f < num_frames; ++f) {
+    free(frames[f]);
+  }
+  free(frame_sizes);
+  free(frames);
   if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
   aom_video_reader_close(reader);
   fclose(outfile);
diff --git a/third_party/aom/examples/lightfield_encoder.c b/third_party/aom/examples/lightfield_encoder.c
index e55cd5ce3c..4dd71cac83 100644
--- a/third_party/aom/examples/lightfield_encoder.c
+++ b/third_party/aom/examples/lightfield_encoder.c
@@ -275,9 +275,13 @@ static void pass1(aom_image_t *raw, FILE *infile, const char *outfile_name,
   aom_img_fmt_t ref_fmt = AOM_IMG_FMT_I420;
   if (!CONFIG_LOWBITDEPTH) ref_fmt |= AOM_IMG_FMT_HIGHBITDEPTH;
   // Allocate memory with the border so that it can be used as a reference.
+  int border_in_pixels =
+      (codec.config.enc->rc_resize_mode || codec.config.enc->rc_superres_mode)
+          ? AOM_BORDER_IN_PIXELS
+          : AOM_ENC_NO_SCALE_BORDER;
   for (i = 0; i < reference_image_num; i++) {
     if (!aom_img_alloc_with_border(&reference_images[i], ref_fmt, cfg->g_w,
-                                   cfg->g_h, 32, 8, AOM_BORDER_IN_PIXELS)) {
+                                   cfg->g_h, 32, 8, border_in_pixels)) {
       die("Failed to allocate image.");
     }
   }
@@ -393,6 +397,10 @@ static void pass1(aom_image_t *raw, FILE *infile, const char *outfile_name,
   for (i = 0; i < reference_image_num; i++) aom_img_free(&reference_images[i]);
 
   if (aom_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec.");
+
+  // Modify large_scale_file fourcc.
+  if (cfg->large_scale_tile == 1)
+    aom_video_writer_set_fourcc(writer, LST_FOURCC);
   aom_video_writer_close(writer);
 
   printf("\nSecond pass complete. Processed %d frames.\n", frame_count);
diff --git a/third_party/aom/examples/lightfield_tile_list_decoder.c b/third_party/aom/examples/lightfield_tile_list_decoder.c
index 5556bf0e7d..87a8b436db 100644
--- a/third_party/aom/examples/lightfield_tile_list_decoder.c
+++ b/third_party/aom/examples/lightfield_tile_list_decoder.c
@@ -18,10 +18,12 @@
 // compressed tile data. This input file is reconstructed from the encoded
 // lightfield ivf file, and is decodable by AV1 decoder. num_references is
 // the number of anchor frames coded at the beginning of the light field file.
-// num_tile_lists is the number of tile lists need to be decoded.
+// num_tile_lists is the number of tile lists need to be decoded. There is an
+// optional parameter allowing to choose the output format, and the supported
+// formats are YUV1D(default), YUV, and NV12.
 // Run lightfield tile list decoder to decode an AV1 tile list file:
 // examples/lightfield_tile_list_decoder vase_tile_list.ivf vase_tile_list.yuv
-// 4 2
+// 4 2 0(optional)
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -39,11 +41,61 @@ static const char *exec_name;
 
 void usage_exit(void) {
   fprintf(stderr,
-          "Usage: %s <infile> <outfile> <num_references> <num_tile_lists>\n",
+          "Usage: %s <infile> <outfile> <num_references> <num_tile_lists> "
+          "<output format(optional)>\n",
           exec_name);
   exit(EXIT_FAILURE);
 }
 
+static void write_tile_yuv1d(aom_codec_ctx_t *codec, const aom_image_t *img,
+                             FILE *file) {
+  // read out the tile size.
+  unsigned int tile_size = 0;
+  if (aom_codec_control(codec, AV1D_GET_TILE_SIZE, &tile_size))
+    die_codec(codec, "Failed to get the tile size");
+  const unsigned int tile_width = tile_size >> 16;
+  const unsigned int tile_height = tile_size & 65535;
+  const uint8_t output_frame_width_in_tiles = img->d_w / tile_width;
+
+  unsigned int tile_count = 0;
+  if (aom_codec_control(codec, AV1D_GET_TILE_COUNT, &tile_count))
+    die_codec(codec, "Failed to get the tile size");
+
+  // Write tile to file.
+  const int shift = (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 1 : 0;
+  unsigned int tile_idx;
+
+  for (tile_idx = 0; tile_idx < tile_count; ++tile_idx) {
+    const int row_offset =
+        (tile_idx / output_frame_width_in_tiles) * tile_height;
+    const int col_offset =
+        (tile_idx % output_frame_width_in_tiles) * tile_width;
+    int plane;
+
+    for (plane = 0; plane < 3; ++plane) {
+      const unsigned char *buf = img->planes[plane];
+      const int stride = img->stride[plane];
+      const int roffset =
+          (plane > 0) ? row_offset >> img->y_chroma_shift : row_offset;
+      const int coffset =
+          (plane > 0) ? col_offset >> img->x_chroma_shift : col_offset;
+      const int w = (plane > 0) ? ((tile_width >> img->x_chroma_shift) << shift)
+                                : (tile_width << shift);
+      const int h =
+          (plane > 0) ? (tile_height >> img->y_chroma_shift) : tile_height;
+      int y;
+
+      // col offset needs to be adjusted for HBD.
+      buf += roffset * stride + (coffset << shift);
+
+      for (y = 0; y < h; ++y) {
+        fwrite(buf, 1, w, file);
+        buf += stride;
+      }
+    }
+  }
+}
+
 int main(int argc, char **argv) {
   FILE *outfile = NULL;
   aom_codec_ctx_t codec;
@@ -55,11 +107,12 @@ int main(int argc, char **argv) {
   aom_image_t reference_images[MAX_EXTERNAL_REFERENCES];
   size_t frame_size = 0;
   const unsigned char *frame = NULL;
+  int output_format = YUV1D;
   int i, j, n;
 
   exec_name = argv[0];
 
-  if (argc != 5) die("Invalid number of arguments.");
+  if (argc < 5) die("Invalid number of arguments.");
 
   reader = aom_video_reader_open(argv[1]);
   if (!reader) die("Failed to open %s for reading.", argv[1]);
@@ -70,6 +123,10 @@ int main(int argc, char **argv) {
   num_references = (int)strtol(argv[3], NULL, 0);
   num_tile_lists = (int)strtol(argv[4], NULL, 0);
 
+  if (argc > 5) output_format = (int)strtol(argv[5], NULL, 0);
+  if (output_format < YUV1D || output_format > NV12)
+    die("Output format out of range [0, 2]");
+
   info = aom_video_reader_get_info(reader);
 
   decoder = get_aom_decoder_by_fourcc(info->codec_fourcc);
@@ -103,7 +160,7 @@ int main(int argc, char **argv) {
       // Allocate memory to store decoded references. Allocate memory with the
       // border so that it can be used as a reference.
       for (j = 0; j < num_references; j++) {
-        unsigned int border = AOM_BORDER_IN_PIXELS;
+        unsigned int border = AOM_DEC_BORDER_IN_PIXELS;
         if (!aom_img_alloc_with_border(&reference_images[j], ref_fmt,
                                        frame_res[0], frame_res[1], 32, 8,
                                        border)) {
@@ -147,9 +204,17 @@ int main(int argc, char **argv) {
     if (aom_codec_decode(&codec, frame, frame_size, NULL))
       die_codec(&codec, "Failed to decode the tile list.");
     aom_codec_iter_t iter = NULL;
-    aom_image_t *img;
-    while ((img = aom_codec_get_frame(&codec, &iter)))
-      fwrite(img->img_data, 1, img->sz, outfile);
+    aom_image_t *img = aom_codec_get_frame(&codec, &iter);
+    if (!img) die_codec(&codec, "Failed to get frame.");
+
+    if (output_format == YUV1D)
+      // write the tile to the output file in 1D format.
+      write_tile_yuv1d(&codec, img, outfile);
+    else if (output_format == YUV)
+      aom_img_write(img, outfile);
+    else
+      // NV12 output format
+      aom_img_write_nv12(img, outfile);
   }
 
   for (i = 0; i < num_references; i++) aom_img_free(&reference_images[i]);
diff --git a/third_party/aom/examples/noise_model.c b/third_party/aom/examples/noise_model.c
index 5cc6003b67..45f1b4dbaf 100644
--- a/third_party/aom/examples/noise_model.c
+++ b/third_party/aom/examples/noise_model.c
@@ -114,7 +114,7 @@ typedef struct {
   const char *debug_file;
 } noise_model_args_t;
 
-void parse_args(noise_model_args_t *noise_args, int *argc, char **argv) {
+static void parse_args(noise_model_args_t *noise_args, int *argc, char **argv) {
   struct arg arg;
   static const arg_def_t *main_args[] = { &help,
                                           &input_arg,
diff --git a/third_party/aom/test/acm_random.h b/third_party/aom/test/acm_random.h
index 0a8317fd57..a14b671ffe 100644
--- a/third_party/aom/test/acm_random.h
+++ b/third_party/aom/test/acm_random.h
@@ -26,6 +26,7 @@ class ACMRandom {
 
   void Reset(int seed) { random_.Reseed(seed); }
 
+  // Generates a random 31-bit unsigned integer from [0, 2^31).
   uint32_t Rand31(void) {
     return random_.Generate(testing::internal::Random::kMaxRange);
   }
diff --git a/third_party/aom/test/active_map_test.cc b/third_party/aom/test/active_map_test.cc
index a2b0546edb..0f8a7329e0 100644
--- a/third_party/aom/test/active_map_test.cc
+++ b/third_party/aom/test/active_map_test.cc
@@ -37,7 +37,7 @@ class ActiveMapTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, cpu_used_);
     } else if (video->frame() == 3) {
       aom_active_map_t map = aom_active_map_t();
diff --git a/third_party/aom/test/aq_segment_test.cc b/third_party/aom/test/aq_segment_test.cc
index bbb5027d4d..51557a5cd9 100644
--- a/third_party/aom/test/aq_segment_test.cc
+++ b/third_party/aom/test/aq_segment_test.cc
@@ -35,7 +35,7 @@ class AqSegmentTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
       encoder->Control(AV1E_SET_AQ_MODE, aq_mode_);
       encoder->Control(AV1E_SET_DELTAQ_MODE, deltaq_mode_);
diff --git a/third_party/aom/test/arf_freq_test.cc b/third_party/aom/test/arf_freq_test.cc
index 083f4022f4..50b478ba65 100644
--- a/third_party/aom/test/arf_freq_test.cc
+++ b/third_party/aom/test/arf_freq_test.cc
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
+
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
 #include "test/codec_factory.h"
@@ -185,7 +187,7 @@ TEST_P(ArfFreqTestLarge, MinArfFreqTest) {
   init_flags_ = AOM_CODEC_USE_PSNR;
   if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
 
-  testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+  std::unique_ptr<libaom_test::VideoSource> video;
   if (is_extension_y4m(test_video_param_.filename)) {
     video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
                                                 kFrames));
diff --git a/third_party/aom/test/av1_convolve_2d_test.cc b/third_party/aom/test/av1_convolve_2d_test.cc
index 03286260e8..b0cef81532 100644
--- a/third_party/aom/test/av1_convolve_2d_test.cc
+++ b/third_party/aom/test/av1_convolve_2d_test.cc
@@ -12,13 +12,14 @@
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 #include "test/av1_convolve_2d_test_util.h"
 
-using ::testing::make_tuple;
-using ::testing::tuple;
 using libaom_test::ACMRandom;
 using libaom_test::AV1Convolve2D::AV1Convolve2DSrTest;
 using libaom_test::AV1Convolve2D::AV1JntConvolve2DTest;
 using libaom_test::AV1HighbdConvolve2D::AV1HighbdConvolve2DSrTest;
 using libaom_test::AV1HighbdConvolve2D::AV1HighbdJntConvolve2DTest;
+using ::testing::make_tuple;
+using ::testing::tuple;
+
 namespace {
 
 TEST_P(AV1Convolve2DSrTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
@@ -89,68 +90,72 @@ INSTANTIATE_TEST_CASE_P(NEON_COPY, AV1Convolve2DSrTest,
 TEST_P(AV1JntConvolve2DTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
 TEST_P(AV1JntConvolve2DTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
 
-INSTANTIATE_TEST_CASE_P(
-    C_COPY, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_copy_c, 0, 0));
+INSTANTIATE_TEST_CASE_P(C_COPY, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_2d_copy_c, 0, 0));
 
 INSTANTIATE_TEST_CASE_P(
     C_X, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_c, 1, 0));
+    libaom_test::AV1Convolve2D::BuildParams(av1_dist_wtd_convolve_x_c, 1, 0));
 
 INSTANTIATE_TEST_CASE_P(
     C_Y, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_c, 0, 1));
+    libaom_test::AV1Convolve2D::BuildParams(av1_dist_wtd_convolve_y_c, 0, 1));
 
 #if HAVE_SSE2
 INSTANTIATE_TEST_CASE_P(SSE2_COPY, AV1JntConvolve2DTest,
                         libaom_test::AV1Convolve2D::BuildParams(
-                            av1_jnt_convolve_2d_copy_sse2, 0, 0));
-#if HAVE_SSE4_1
-INSTANTIATE_TEST_CASE_P(
-    SSE2_X, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_sse2, 1, 0));
+                            av1_dist_wtd_convolve_2d_copy_sse2, 0, 0));
+INSTANTIATE_TEST_CASE_P(SSE2, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_2d_sse2, 1, 1));
 
-INSTANTIATE_TEST_CASE_P(
-    SSE2_Y, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_sse2, 0, 1));
+INSTANTIATE_TEST_CASE_P(SSE2_X, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_x_sse2, 1, 0));
 
-INSTANTIATE_TEST_CASE_P(
-    SSSE3, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_ssse3, 1, 1));
+INSTANTIATE_TEST_CASE_P(SSE2_Y, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_y_sse2, 0, 1));
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_2d_ssse3, 1, 1));
 
 #if HAVE_AVX2
 INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1JntConvolve2DTest,
                         libaom_test::AV1Convolve2D::BuildParams(
-                            av1_jnt_convolve_2d_copy_avx2, 0, 0));
-INSTANTIATE_TEST_CASE_P(
-    AVX2_X, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_avx2, 1, 0));
+                            av1_dist_wtd_convolve_2d_copy_avx2, 0, 0));
+INSTANTIATE_TEST_CASE_P(AVX2_X, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_x_avx2, 1, 0));
 
-INSTANTIATE_TEST_CASE_P(
-    AVX2_Y, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_avx2, 0, 1));
+INSTANTIATE_TEST_CASE_P(AVX2_Y, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_y_avx2, 0, 1));
 
-INSTANTIATE_TEST_CASE_P(
-    AVX2, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_avx2, 1, 1));
+INSTANTIATE_TEST_CASE_P(AVX2, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_2d_avx2, 1, 1));
 #endif  // HAVE_AVX2
-#endif  // HAVE_SSE4_1
+#endif  // HAVE_SSSE3
 #endif  // HAVE_SSE2
 #if HAVE_NEON
 INSTANTIATE_TEST_CASE_P(NEON_COPY, AV1JntConvolve2DTest,
                         libaom_test::AV1Convolve2D::BuildParams(
-                            av1_jnt_convolve_2d_copy_neon, 0, 0));
+                            av1_dist_wtd_convolve_2d_copy_neon, 0, 0));
 
-INSTANTIATE_TEST_CASE_P(
-    NEON, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_neon, 1, 1));
-INSTANTIATE_TEST_CASE_P(
-    NEON_X, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_neon, 1, 0));
+INSTANTIATE_TEST_CASE_P(NEON, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_2d_neon, 1, 1));
+INSTANTIATE_TEST_CASE_P(NEON_X, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_x_neon, 1, 0));
 
-INSTANTIATE_TEST_CASE_P(
-    NEON_Y, AV1JntConvolve2DTest,
-    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_neon, 0, 1));
+INSTANTIATE_TEST_CASE_P(NEON_Y, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_dist_wtd_convolve_y_neon, 0, 1));
 #endif  // HAVE_NEON
 
 TEST_P(AV1HighbdConvolve2DSrTest, CheckOutput) { RunCheckOutput(GET_PARAM(1)); }
@@ -209,41 +214,41 @@ TEST_P(AV1HighbdJntConvolve2DTest, DISABLED_Speed) {
 
 INSTANTIATE_TEST_CASE_P(C_X, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_x_c, 1, 0));
+                            av1_highbd_dist_wtd_convolve_x_c, 1, 0));
 
 INSTANTIATE_TEST_CASE_P(C_Y, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_y_c, 0, 1));
+                            av1_highbd_dist_wtd_convolve_y_c, 0, 1));
 
 INSTANTIATE_TEST_CASE_P(C_COPY, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_2d_copy_c, 0, 0));
+                            av1_highbd_dist_wtd_convolve_2d_copy_c, 0, 0));
 #if HAVE_SSE4_1
 INSTANTIATE_TEST_CASE_P(SSE4_1_COPY, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_2d_copy_sse4_1, 0, 0));
+                            av1_highbd_dist_wtd_convolve_2d_copy_sse4_1, 0, 0));
 INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_2d_sse4_1, 1, 1));
+                            av1_highbd_dist_wtd_convolve_2d_sse4_1, 1, 1));
 INSTANTIATE_TEST_CASE_P(SSE4_1_X, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_x_sse4_1, 1, 0));
+                            av1_highbd_dist_wtd_convolve_x_sse4_1, 1, 0));
 INSTANTIATE_TEST_CASE_P(SSE4_1_Y, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_y_sse4_1, 0, 1));
+                            av1_highbd_dist_wtd_convolve_y_sse4_1, 0, 1));
 #if HAVE_AVX2
 INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_2d_copy_avx2, 0, 0));
+                            av1_highbd_dist_wtd_convolve_2d_copy_avx2, 0, 0));
 INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_2d_avx2, 1, 1));
+                            av1_highbd_dist_wtd_convolve_2d_avx2, 1, 1));
 INSTANTIATE_TEST_CASE_P(AVX2_X, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_x_avx2, 1, 0));
+                            av1_highbd_dist_wtd_convolve_x_avx2, 1, 0));
 INSTANTIATE_TEST_CASE_P(AVX2_Y, AV1HighbdJntConvolve2DTest,
                         libaom_test::AV1HighbdConvolve2D::BuildParams(
-                            av1_highbd_jnt_convolve_y_avx2, 0, 1));
+                            av1_highbd_dist_wtd_convolve_y_avx2, 0, 1));
 #endif  // HAVE_AVX2
 #endif  // HAVE_SSE4_1
 }  // namespace
diff --git a/third_party/aom/test/av1_convolve_2d_test_util.cc b/third_party/aom/test/av1_convolve_2d_test_util.cc
index 409fd23e10..9cfe3e6b33 100644
--- a/third_party/aom/test/av1_convolve_2d_test_util.cc
+++ b/third_party/aom/test/av1_convolve_2d_test_util.cc
@@ -200,9 +200,9 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
         ConvolveParams conv_params2 =
             get_conv_params_no_round(do_average, 0, output2, MAX_SB_SIZE, 1, 8);
 
-        // Test special case where jnt_comp_avg is not used
-        conv_params1.use_jnt_comp_avg = 0;
-        conv_params2.use_jnt_comp_avg = 0;
+        // Test special case where dist_wtd_comp_avg is not used
+        conv_params1.use_dist_wtd_comp_avg = 0;
+        conv_params2.use_dist_wtd_comp_avg = 0;
 
         const int subx_range = has_subx ? 16 : 1;
         const int suby_range = has_suby ? 16 : 1;
@@ -211,9 +211,10 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
             // Choose random locations within the source block
             const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
             const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
-            av1_jnt_convolve_2d_c(input + offset_r * w + offset_c, w, output8_1,
-                                  MAX_SB_SIZE, out_w, out_h, filter_params_x,
-                                  filter_params_y, subx, suby, &conv_params1);
+            av1_dist_wtd_convolve_2d_c(input + offset_r * w + offset_c, w,
+                                       output8_1, MAX_SB_SIZE, out_w, out_h,
+                                       filter_params_x, filter_params_y, subx,
+                                       suby, &conv_params1);
             test_impl(input + offset_r * w + offset_c, w, output8_2,
                       MAX_SB_SIZE, out_w, out_h, filter_params_x,
                       filter_params_y, subx, suby, &conv_params2);
@@ -222,7 +223,7 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
               for (int j = 0; j < out_w; ++j) {
                 int idx = i * MAX_SB_SIZE + j;
                 ASSERT_EQ(output1[idx], output2[idx])
-                    << "Mismatch at unit tests for av1_jnt_convolve_2d\n"
+                    << "Mismatch at unit tests for av1_dist_wtd_convolve_2d\n"
                     << out_w << "x" << out_h << " Pixel mismatch at index "
                     << idx << " = (" << i << ", " << j
                     << "), sub pixel offset = (" << suby << ", " << subx << ")";
@@ -247,8 +248,8 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
         // Test different combination of fwd and bck offset weights
         for (int k = 0; k < 2; ++k) {
           for (int l = 0; l < 4; ++l) {
-            conv_params1.use_jnt_comp_avg = 1;
-            conv_params2.use_jnt_comp_avg = 1;
+            conv_params1.use_dist_wtd_comp_avg = 1;
+            conv_params2.use_dist_wtd_comp_avg = 1;
             conv_params1.fwd_offset = quant_dist_lookup_table[k][l][0];
             conv_params1.bck_offset = quant_dist_lookup_table[k][l][1];
             conv_params2.fwd_offset = quant_dist_lookup_table[k][l][0];
@@ -259,10 +260,10 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
                 // Choose random locations within the source block
                 const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
                 const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
-                av1_jnt_convolve_2d_c(input + offset_r * w + offset_c, w,
-                                      output8_1, MAX_SB_SIZE, out_w, out_h,
-                                      filter_params_x, filter_params_y, subx,
-                                      suby, &conv_params1);
+                av1_dist_wtd_convolve_2d_c(input + offset_r * w + offset_c, w,
+                                           output8_1, MAX_SB_SIZE, out_w, out_h,
+                                           filter_params_x, filter_params_y,
+                                           subx, suby, &conv_params1);
                 test_impl(input + offset_r * w + offset_c, w, output8_2,
                           MAX_SB_SIZE, out_w, out_h, filter_params_x,
                           filter_params_y, subx, suby, &conv_params2);
@@ -272,7 +273,7 @@ void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
                     int idx = i * MAX_SB_SIZE + j;
                     ASSERT_EQ(output1[idx], output2[idx])
                         << "Mismatch at unit tests for "
-                           "av1_jnt_convolve_2d\n"
+                           "av1_dist_wtd_convolve_2d\n"
                         << out_w << "x" << out_h << " Pixel mismatch at index "
                         << idx << " = (" << i << ", " << j
                         << "), sub pixel offset = (" << suby << ", " << subx
@@ -333,7 +334,7 @@ void AV1JntConvolve2DTest::RunSpeedTest(convolve_2d_func test_impl) {
   ConvolveParams conv_params =
       get_conv_params_no_round(do_average, 0, output, MAX_SB_SIZE, 1, 8);
 
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
 
   // Choose random locations within the source block
   const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
@@ -540,8 +541,8 @@ void AV1HighbdJntConvolve2DTest::RunSpeedTest(
   ConvolveParams conv_params =
       get_conv_params_no_round(do_average, 0, output, MAX_SB_SIZE, 1, bd);
 
-  // Test special case where jnt_comp_avg is not used
-  conv_params.use_jnt_comp_avg = 0;
+  // Test special case where dist_wtd_comp_avg is not used
+  conv_params.use_dist_wtd_comp_avg = 0;
 
   subx = 0;
   suby = 0;
@@ -601,9 +602,9 @@ void AV1HighbdJntConvolve2DTest::RunCheckOutput(
         ConvolveParams conv_params2 = get_conv_params_no_round(
             do_average, 0, output2, MAX_SB_SIZE, 1, bd);
 
-        // Test special case where jnt_comp_avg is not used
-        conv_params1.use_jnt_comp_avg = 0;
-        conv_params2.use_jnt_comp_avg = 0;
+        // Test special case where dist_wtd_comp_avg is not used
+        conv_params1.use_dist_wtd_comp_avg = 0;
+        conv_params2.use_dist_wtd_comp_avg = 0;
 
         const int subx_range = has_subx ? 16 : 1;
         const int suby_range = has_suby ? 16 : 1;
@@ -612,10 +613,10 @@ void AV1HighbdJntConvolve2DTest::RunCheckOutput(
             // Choose random locations within the source block
             const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
             const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
-            av1_highbd_jnt_convolve_2d_c(input + offset_r * w + offset_c, w,
-                                         output16_1, MAX_SB_SIZE, out_w, out_h,
-                                         filter_params_x, filter_params_y, subx,
-                                         suby, &conv_params1, bd);
+            av1_highbd_dist_wtd_convolve_2d_c(
+                input + offset_r * w + offset_c, w, output16_1, MAX_SB_SIZE,
+                out_w, out_h, filter_params_x, filter_params_y, subx, suby,
+                &conv_params1, bd);
             test_impl(input + offset_r * w + offset_c, w, output16_2,
                       MAX_SB_SIZE, out_w, out_h, filter_params_x,
                       filter_params_y, subx, suby, &conv_params2, bd);
@@ -648,8 +649,8 @@ void AV1HighbdJntConvolve2DTest::RunCheckOutput(
         // Test different combination of fwd and bck offset weights
         for (int k = 0; k < 2; ++k) {
           for (int l = 0; l < 4; ++l) {
-            conv_params1.use_jnt_comp_avg = 1;
-            conv_params2.use_jnt_comp_avg = 1;
+            conv_params1.use_dist_wtd_comp_avg = 1;
+            conv_params2.use_dist_wtd_comp_avg = 1;
             conv_params1.fwd_offset = quant_dist_lookup_table[k][l][0];
             conv_params1.bck_offset = quant_dist_lookup_table[k][l][1];
             conv_params2.fwd_offset = quant_dist_lookup_table[k][l][0];
@@ -662,7 +663,7 @@ void AV1HighbdJntConvolve2DTest::RunCheckOutput(
                 // Choose random locations within the source block
                 const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
                 const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
-                av1_highbd_jnt_convolve_2d_c(
+                av1_highbd_dist_wtd_convolve_2d_c(
                     input + offset_r * w + offset_c, w, output16_1, MAX_SB_SIZE,
                     out_w, out_h, filter_params_x, filter_params_y, subx, suby,
                     &conv_params1, bd);
diff --git a/third_party/aom/test/av1_convolve_scale_test.cc b/third_party/aom/test/av1_convolve_scale_test.cc
index 3b1698eeb4..a933fc9238 100644
--- a/third_party/aom/test/av1_convolve_scale_test.cc
+++ b/third_party/aom/test/av1_convolve_scale_test.cc
@@ -32,9 +32,9 @@ const int kHPad = 32;
 const int kXStepQn = 16;
 const int kYStepQn = 20;
 
+using libaom_test::ACMRandom;
 using ::testing::make_tuple;
 using ::testing::tuple;
-using libaom_test::ACMRandom;
 
 enum NTaps { EIGHT_TAP, TEN_TAP, TWELVE_TAP };
 int NTapsToInt(NTaps ntaps) { return 8 + static_cast<int>(ntaps) * 2; }
@@ -286,13 +286,13 @@ class ConvolveScaleTestBase : public ::testing::Test {
   }
 
   void SetConvParamOffset(int i, int j, int is_compound, int do_average,
-                          int use_jnt_comp_avg) {
+                          int use_dist_wtd_comp_avg) {
     if (i == -1 && j == -1) {
-      convolve_params_.use_jnt_comp_avg = use_jnt_comp_avg;
+      convolve_params_.use_dist_wtd_comp_avg = use_dist_wtd_comp_avg;
       convolve_params_.is_compound = is_compound;
       convolve_params_.do_average = do_average;
     } else {
-      convolve_params_.use_jnt_comp_avg = use_jnt_comp_avg;
+      convolve_params_.use_dist_wtd_comp_avg = use_dist_wtd_comp_avg;
       convolve_params_.fwd_offset = quant_dist_lookup_table[i][j][0];
       convolve_params_.bck_offset = quant_dist_lookup_table[i][j][1];
       convolve_params_.is_compound = is_compound;
@@ -312,12 +312,12 @@ class ConvolveScaleTestBase : public ::testing::Test {
 
       is_compound = 1;
       for (int do_average = 0; do_average < 2; do_average++) {
-        for (int use_jnt_comp_avg = 0; use_jnt_comp_avg < 2;
-             use_jnt_comp_avg++) {
+        for (int use_dist_wtd_comp_avg = 0; use_dist_wtd_comp_avg < 2;
+             use_dist_wtd_comp_avg++) {
           for (int j = 0; j < 2; ++j) {
             for (int k = 0; k < 4; ++k) {
               SetConvParamOffset(j, k, is_compound, do_average,
-                                 use_jnt_comp_avg);
+                                 use_dist_wtd_comp_avg);
               Prep(&rnd);
               RunOne(true);
               RunOne(false);
diff --git a/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc b/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc
index e8470e5d50..76b82f58f7 100644
--- a/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc
+++ b/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
+
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
 #include "test/codec_factory.h"
@@ -95,7 +97,7 @@ class AVxEncoderParmsGetToDecoder
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_COLOR_PRIMARIES, encode_parms.color_primaries);
       encoder->Control(AV1E_SET_TRANSFER_CHARACTERISTICS,
                        encode_parms.transfer_characteristics);
@@ -146,7 +148,7 @@ class AVxEncoderParmsGetToDecoder
 TEST_P(AVxEncoderParmsGetToDecoder, BitstreamParms) {
   init_flags_ = AOM_CODEC_USE_PSNR;
 
-  testing::internal::scoped_ptr<libaom_test::VideoSource> video(
+  std::unique_ptr<libaom_test::VideoSource> video(
       new libaom_test::Y4mVideoSource(test_video_.name, 0, test_video_.frames));
   ASSERT_TRUE(video.get() != NULL);
 
diff --git a/third_party/aom/test/av1_fwd_txfm1d_test.cc b/third_party/aom/test/av1_fwd_txfm1d_test.cc
index 49a6668793..863cb396ae 100644
--- a/third_party/aom/test/av1_fwd_txfm1d_test.cc
+++ b/third_party/aom/test/av1_fwd_txfm1d_test.cc
@@ -13,12 +13,12 @@
 #include "test/av1_txfm_test.h"
 
 using libaom_test::ACMRandom;
+using libaom_test::input_base;
+using libaom_test::reference_hybrid_1d;
 using libaom_test::TYPE_ADST;
 using libaom_test::TYPE_DCT;
 using libaom_test::TYPE_IDTX;
 using libaom_test::TYPE_TXFM;
-using libaom_test::input_base;
-using libaom_test::reference_hybrid_1d;
 
 namespace {
 const int txfm_type_num = 3;
@@ -56,7 +56,7 @@ TEST(av1_fwd_txfm1d, av1_cospi_arr_data) {
   for (int i = 0; i < 7; i++) {
     for (int j = 0; j < 64; j++) {
       EXPECT_EQ(av1_cospi_arr_data[i][j],
-                (int32_t)round(cos(M_PI * j / 128) * (1 << (cos_bit_min + i))));
+                (int32_t)round(cos(PI * j / 128) * (1 << (cos_bit_min + i))));
     }
   }
 }
diff --git a/third_party/aom/test/av1_fwd_txfm2d_test.cc b/third_party/aom/test/av1_fwd_txfm2d_test.cc
index 75f20536b2..d55085f069 100644
--- a/third_party/aom/test/av1_fwd_txfm2d_test.cc
+++ b/third_party/aom/test/av1_fwd_txfm2d_test.cc
@@ -23,10 +23,10 @@
 #include "av1/encoder/hybrid_fwd_txfm.h"
 
 using libaom_test::ACMRandom;
-using libaom_test::TYPE_TXFM;
 using libaom_test::bd;
 using libaom_test::compute_avg_abs_error;
 using libaom_test::input_base;
+using libaom_test::TYPE_TXFM;
 
 using std::vector;
 
@@ -507,5 +507,12 @@ INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdFwdTxfm2dTest,
                         Combine(ValuesIn(Highbd_fwd_txfm_for_sse4_1),
                                 Values(av1_highbd_fwd_txfm)));
 #endif  // HAVE_SSE4_1
+#if HAVE_AVX2
+static TX_SIZE Highbd_fwd_txfm_for_avx2[] = { TX_8X8,   TX_16X16, TX_32X32,
+                                              TX_64X64, TX_8X16,  TX_16X8 };
 
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdFwdTxfm2dTest,
+                        Combine(ValuesIn(Highbd_fwd_txfm_for_avx2),
+                                Values(av1_highbd_fwd_txfm)));
+#endif  // HAVE_AVX2
 }  // namespace
diff --git a/third_party/aom/test/av1_highbd_iht_test.cc b/third_party/aom/test/av1_highbd_iht_test.cc
index 2d6490c2ad..7f077b657c 100644
--- a/third_party/aom/test/av1_highbd_iht_test.cc
+++ b/third_party/aom/test/av1_highbd_iht_test.cc
@@ -25,8 +25,8 @@
 
 namespace {
 
-using ::testing::tuple;
 using libaom_test::ACMRandom;
+using ::testing::tuple;
 
 typedef void (*HbdHtFunc)(const int16_t *input, int32_t *output, int stride,
                           TX_TYPE tx_type, int bd);
diff --git a/third_party/aom/test/av1_horz_only_frame_superres_test.cc b/third_party/aom/test/av1_horz_only_frame_superres_test.cc
index fd77ef35d7..ffc9136762 100644
--- a/third_party/aom/test/av1_horz_only_frame_superres_test.cc
+++ b/third_party/aom/test/av1_horz_only_frame_superres_test.cc
@@ -30,9 +30,9 @@ const int kPerfIters = 1000;
 const int kVPad = 32;
 const int kHPad = 32;
 
+using libaom_test::ACMRandom;
 using ::testing::make_tuple;
 using ::testing::tuple;
-using libaom_test::ACMRandom;
 
 template <typename Pixel>
 class TestImage {
diff --git a/third_party/aom/test/av1_inv_txfm2d_test.cc b/third_party/aom/test/av1_inv_txfm2d_test.cc
index 11e231ba64..5432130d76 100644
--- a/third_party/aom/test/av1_inv_txfm2d_test.cc
+++ b/third_party/aom/test/av1_inv_txfm2d_test.cc
@@ -24,11 +24,11 @@
 #include "test/util.h"
 
 using libaom_test::ACMRandom;
-using libaom_test::InvTxfm2dFunc;
-using libaom_test::LbdInvTxfm2dFunc;
 using libaom_test::bd;
 using libaom_test::compute_avg_abs_error;
 using libaom_test::input_base;
+using libaom_test::InvTxfm2dFunc;
+using libaom_test::LbdInvTxfm2dFunc;
 
 using ::testing::Combine;
 using ::testing::Range;
diff --git a/third_party/aom/test/av1_nn_predict_test.cc b/third_party/aom/test/av1_nn_predict_test.cc
new file mode 100644
index 0000000000..0574a15ad3
--- /dev/null
+++ b/third_party/aom/test/av1_nn_predict_test.cc
@@ -0,0 +1,215 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "aom/aom_integer.h"
+#include "aom_ports/aom_timer.h"
+#include "av1/encoder/ml.h"
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+#include "test/util.h"
+#include "test/register_state_check.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+
+namespace {
+typedef void (*NnPredict_Func)(const float *const input_nodes,
+                               const NN_CONFIG *const nn_config,
+                               float *const output);
+
+typedef ::testing::tuple<const NnPredict_Func> NnPredictTestParam;
+
+const float epsilon = 1e-3f;  // Error threshold for functional equivalence
+
+class NnPredictTest : public ::testing::TestWithParam<NnPredictTestParam> {
+ public:
+  virtual void SetUp() {
+    const int MAX_NODES2 = NN_MAX_NODES_PER_LAYER * NN_MAX_NODES_PER_LAYER;
+    // Allocate two massive buffers on the heap for edge weights and node bias
+    // Then set-up the double-dimension arrays pointing into the big buffers
+    weights_buf = (float *)aom_malloc(MAX_NODES2 * (NN_MAX_HIDDEN_LAYERS + 1) *
+                                      sizeof(*weights_buf));
+    bias_buf =
+        (float *)aom_malloc(NN_MAX_NODES_PER_LAYER *
+                            (NN_MAX_HIDDEN_LAYERS + 1) * sizeof(*bias_buf));
+    ASSERT_NE(weights_buf, nullptr);
+    ASSERT_NE(bias_buf, nullptr);
+    for (int i = 0; i < NN_MAX_HIDDEN_LAYERS + 1; i++) {
+      weights[i] = &weights_buf[i * MAX_NODES2];
+      bias[i] = &bias_buf[i * NN_MAX_NODES_PER_LAYER];
+    }
+    target_func_ = GET_PARAM(0);
+  }
+  virtual void TearDown() {
+    aom_free(weights_buf);
+    aom_free(bias_buf);
+  }
+  void RunNnPredictTest(const NN_CONFIG *const shape);
+  void RunNnPredictSpeedTest(const NN_CONFIG *const shape, const int run_times);
+  void RunNnPredictTest_all(const NN_CONFIG *const shapes,
+                            const int num_shapes);
+  void RunNnPredictSpeedTest_all(const NN_CONFIG *const shapes,
+                                 const int num_shapes, const int run_times);
+
+ private:
+  NnPredict_Func target_func_;
+  libaom_test::ACMRandom rng_;
+  float *weights[NN_MAX_HIDDEN_LAYERS + 1] = { 0 };
+  float *bias[NN_MAX_HIDDEN_LAYERS + 1] = { 0 };
+  float *weights_buf = nullptr, *bias_buf = nullptr;
+};
+
+void NnPredictTest::RunNnPredictTest(const NN_CONFIG *const shape) {
+  libaom_test::ClearSystemState();
+  float inputs[NN_MAX_NODES_PER_LAYER] = { 0 };
+  float outputs_test[NN_MAX_NODES_PER_LAYER] = { 0 };
+  float outputs_ref[NN_MAX_NODES_PER_LAYER] = { 0 };
+
+  NN_CONFIG nn_config;
+  memcpy(&nn_config, shape, sizeof(nn_config));
+
+  char shape_str[32] = { 0 };
+  snprintf(shape_str, sizeof(shape_str), "%d", shape->num_inputs);
+  for (int layer = 0; layer < shape->num_hidden_layers; layer++)
+    snprintf(&shape_str[strlen(shape_str)],
+             sizeof(shape_str) - strlen(shape_str), "x%d",
+             shape->num_hidden_nodes[layer]);
+  snprintf(&shape_str[strlen(shape_str)], sizeof(shape_str) - strlen(shape_str),
+           "x%d", shape->num_outputs);
+
+  for (int i = 0; i < NN_MAX_HIDDEN_LAYERS + 1; i++) {
+    nn_config.weights[i] = weights[i];
+    nn_config.bias[i] = bias[i];
+  }
+
+  for (int iter = 0; iter < 10000 && !HasFatalFailure(); ++iter) {
+    for (int node = 0; node < shape->num_inputs; node++) {
+      inputs[node] = ((float)rng_.Rand31() - (1 << 30)) / (1u << 31);
+    }
+    for (int layer = 0; layer < shape->num_hidden_layers; layer++) {
+      for (int node = 0; node < NN_MAX_NODES_PER_LAYER; node++) {
+        bias[layer][node] = ((float)rng_.Rand31() - (1 << 30)) / (1u << 31);
+      }
+      for (int node = 0; node < NN_MAX_NODES_PER_LAYER * NN_MAX_NODES_PER_LAYER;
+           node++) {
+        weights[layer][node] = ((float)rng_.Rand31() - (1 << 30)) / (1u << 31);
+      }
+    }
+    // Now the outputs:
+    int layer = shape->num_hidden_layers;
+    for (int node = 0; node < NN_MAX_NODES_PER_LAYER; node++) {
+      bias[layer][node] = ((float)rng_.Rand31() - (1 << 30)) / (1u << 31);
+    }
+    for (int node = 0; node < NN_MAX_NODES_PER_LAYER * NN_MAX_NODES_PER_LAYER;
+         node++) {
+      weights[layer][node] = ((float)rng_.Rand31() - (1 << 30)) / (1u << 31);
+    }
+
+    av1_nn_predict_c(inputs, &nn_config, outputs_ref);
+    target_func_(inputs, &nn_config, outputs_test);
+    libaom_test::ClearSystemState();
+
+    for (int node = 0; node < shape->num_outputs; node++) {
+      if (outputs_ref[node] < epsilon) {
+        ASSERT_LE(outputs_test[node], epsilon)
+            << "Reference output was near-zero, test output was not ("
+            << shape_str << ")";
+      } else {
+        const float error = outputs_ref[node] - outputs_test[node];
+        const float relative_error = fabsf(error / outputs_ref[node]);
+        ASSERT_LE(relative_error, epsilon)
+            << "Excessive relative error between reference and test ("
+            << shape_str << ")";
+      }
+    }
+  }
+}
+
+void NnPredictTest::RunNnPredictSpeedTest(const NN_CONFIG *const shape,
+                                          const int run_times) {
+  libaom_test::ClearSystemState();
+  float inputs[NN_MAX_NODES_PER_LAYER] = { 0 };
+  float outputs_test[NN_MAX_NODES_PER_LAYER] = { 0 };
+  float outputs_ref[NN_MAX_NODES_PER_LAYER] = { 0 };
+
+  NN_CONFIG nn_config;
+  memcpy(&nn_config, shape, sizeof(nn_config));
+
+  for (int i = 0; i < NN_MAX_HIDDEN_LAYERS; i++) {
+    nn_config.weights[i] = weights[i];
+    nn_config.bias[i] = bias[i];
+  }
+  // Don't bother actually changing the values for inputs/weights/bias: it
+  // shouldn't make any difference for a speed test.
+
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    av1_nn_predict_c(inputs, &nn_config, outputs_ref);
+  }
+  aom_usec_timer_mark(&timer);
+  const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    target_func_(inputs, &nn_config, outputs_test);
+  }
+  aom_usec_timer_mark(&timer);
+  libaom_test::ClearSystemState();
+  const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+
+  printf("%d", shape->num_inputs);
+  for (int layer = 0; layer < shape->num_hidden_layers; layer++)
+    printf("x%d", shape->num_hidden_nodes[layer]);
+  printf("x%d: ", shape->num_outputs);
+  printf("%7.2f/%7.2fns (%3.2f)\n", time1, time2, time1 / time2);
+}
+
+// This is all the neural network shapes observed executed in a few different
+// runs of the encoder.  It also conveniently covers all the kernels
+// implemented.
+static const NN_CONFIG shapes[] = {
+  { 10, 16, 1, { 64 }, { 0 }, { 0 } }, { 12, 1, 1, { 12 }, { 0 }, { 0 } },
+  { 12, 1, 1, { 24 }, { 0 }, { 0 } },  { 12, 1, 1, { 32 }, { 0 }, { 0 } },
+  { 18, 4, 1, { 24 }, { 0 }, { 0 } },  { 18, 4, 1, { 32 }, { 0 }, { 0 } },
+  { 4, 1, 1, { 16 }, { 0 }, { 0 } },   { 8, 1, 1, { 16 }, { 0 }, { 0 } },
+  { 8, 4, 1, { 16 }, { 0 }, { 0 } },   { 8, 1, 1, { 24 }, { 0 }, { 0 } },
+  { 8, 1, 1, { 32 }, { 0 }, { 0 } },   { 8, 1, 1, { 64 }, { 0 }, { 0 } },
+  { 9, 3, 1, { 32 }, { 0 }, { 0 } },   { 4, 4, 1, { 8 }, { 0 }, { 0 } },
+};
+
+void NnPredictTest::RunNnPredictTest_all(const NN_CONFIG *const shapes,
+                                         const int num_shapes) {
+  for (int i = 0; i < num_shapes; i++) RunNnPredictTest(&shapes[i]);
+}
+
+void NnPredictTest::RunNnPredictSpeedTest_all(const NN_CONFIG *const shapes,
+                                              const int num_shapes,
+                                              const int run_times) {
+  for (int i = 0; i < num_shapes; i++)
+    NnPredictTest::RunNnPredictSpeedTest(&shapes[i], run_times);
+}
+
+TEST_P(NnPredictTest, RandomValues) {
+  RunNnPredictTest_all(shapes, sizeof(shapes) / sizeof(*shapes));
+}
+
+TEST_P(NnPredictTest, DISABLED_Speed) {
+  RunNnPredictSpeedTest_all(shapes, sizeof(shapes) / sizeof(*shapes), 10000000);
+}
+
+#if HAVE_SSE3
+INSTANTIATE_TEST_CASE_P(SSE3, NnPredictTest,
+                        ::testing::Values(av1_nn_predict_sse3));
+#endif
+
+}  // namespace
diff --git a/third_party/aom/test/av1_round_shift_array_test.cc b/third_party/aom/test/av1_round_shift_array_test.cc
index 181a394604..61dbed51f4 100644
--- a/third_party/aom/test/av1_round_shift_array_test.cc
+++ b/third_party/aom/test/av1_round_shift_array_test.cc
@@ -13,7 +13,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
 
 #include "aom_mem/aom_mem.h"
 #include "aom_ports/aom_timer.h"
diff --git a/third_party/aom/test/av1_txfm_test.cc b/third_party/aom/test/av1_txfm_test.cc
index d5b0ce3255..abbc475918 100644
--- a/third_party/aom/test/av1_txfm_test.cc
+++ b/third_party/aom/test/av1_txfm_test.cc
@@ -94,7 +94,7 @@ double Sqrt2 = pow(2, 0.5);
 double invSqrt2 = 1 / pow(2, 0.5);
 
 double dct_matrix(double n, double k, int size) {
-  return cos(M_PI * (2 * n + 1) * k / (2 * size));
+  return cos(PI * (2 * n + 1) * k / (2 * size));
 }
 
 void reference_dct_1d(const double *in, double *out, int size) {
@@ -179,7 +179,7 @@ void reference_adst_1d(const double *in, double *out, int size) {
   for (int k = 0; k < size; ++k) {
     out[k] = 0;
     for (int n = 0; n < size; ++n) {
-      out[k] += in[n] * sin(M_PI * (2 * n + 1) * (2 * k + 1) / (4 * size));
+      out[k] += in[n] * sin(PI * (2 * n + 1) * (2 * k + 1) / (4 * size));
     }
   }
 }
diff --git a/third_party/aom/test/av1_txfm_test.h b/third_party/aom/test/av1_txfm_test.h
index a181647411..5a56d28f1a 100644
--- a/third_party/aom/test/av1_txfm_test.h
+++ b/third_party/aom/test/av1_txfm_test.h
@@ -29,14 +29,14 @@
 #include "av1/common/enums.h"
 
 namespace libaom_test {
-typedef enum {
+enum {
   TYPE_DCT = 0,
   TYPE_ADST,
   TYPE_IDTX,
   TYPE_IDCT,
   TYPE_IADST,
   TYPE_LAST
-} TYPE_TXFM;
+} UENUM1BYTE(TYPE_TXFM);
 
 int get_txfm1d_size(TX_SIZE tx_size);
 
diff --git a/third_party/aom/test/blend_a64_mask_test.cc b/third_party/aom/test/blend_a64_mask_test.cc
index 66ca6fc5f3..75925337f5 100644
--- a/third_party/aom/test/blend_a64_mask_test.cc
+++ b/third_party/aom/test/blend_a64_mask_test.cc
@@ -86,6 +86,7 @@ class BlendA64MaskTest : public FunctionEquivalenceTest<BlendA64Func> {
     w_ = block_size_wide[block_size];
     h_ = block_size_high[block_size];
     run_times = run_times > 1 ? run_times / w_ : 1;
+    ASSERT_GT(run_times, 0);
     subx_ = subx;
     suby_ = suby;
 
@@ -248,13 +249,13 @@ TEST_P(BlendA64MaskTest8B, DISABLED_Speed) {
 INSTANTIATE_TEST_CASE_P(SSE4_1, BlendA64MaskTest8B,
                         ::testing::Values(TestFuncs(
                             aom_blend_a64_mask_c, aom_blend_a64_mask_sse4_1)));
-#endif  // HAVE_AVX2
+#endif  // HAVE_SSE4_1
 
 #if HAVE_AVX2
 INSTANTIATE_TEST_CASE_P(AVX2, BlendA64MaskTest8B,
                         ::testing::Values(TestFuncs(aom_blend_a64_mask_sse4_1,
                                                     aom_blend_a64_mask_avx2)));
-#endif  // HAVE_SSE4_1
+#endif  // HAVE_AVX2
 
 //////////////////////////////////////////////////////////////////////////////
 // 8 bit _d16 version
@@ -482,6 +483,7 @@ class BlendA64MaskTestHBD_d16
   static const int kSrcMaxBitsMaskHBD = (1 << 16) - 1;
 
   void Execute(const uint16_t *p_src0, const uint16_t *p_src1, int run_times) {
+    ASSERT_GT(run_times, 0) << "Cannot run 0 iterations of the test.";
     ConvolveParams conv_params;
     conv_params.round_0 = (bit_depth_ == 12) ? ROUND0_BITS + 2 : ROUND0_BITS;
     conv_params.round_1 = COMPOUND_ROUND1_BITS;
@@ -566,11 +568,45 @@ TEST_P(BlendA64MaskTestHBD_d16, DISABLED_SaturatedValues) {
     }
   }
 }
+TEST_P(BlendA64MaskTestHBD_d16, DISABLED_Speed) {
+  const int kRunTimes = 10000000;
+  for (int bsize = 0; bsize < BLOCK_SIZES_ALL; ++bsize) {
+    for (bit_depth_ = 8; bit_depth_ <= 12; bit_depth_ += 2) {
+      for (int i = 0; i < kBufSize; ++i) {
+        dst_ref_[i] = rng_.Rand12() % (1 << bit_depth_);
+        dst_tst_[i] = rng_.Rand12() % (1 << bit_depth_);
+
+        src0_[i] = rng_.Rand16();
+        src1_[i] = rng_.Rand16();
+      }
+
+      for (int i = 0; i < kMaxMaskSize; ++i)
+        mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+      RunOneTest(bsize, 1, 1, kRunTimes);
+      RunOneTest(bsize, 0, 0, kRunTimes);
+    }
+  }
+}
 
 INSTANTIATE_TEST_CASE_P(
     C, BlendA64MaskTestHBD_d16,
     ::testing::Values(TestFuncsHBD_d16(aom_highbd_blend_a64_d16_mask_c, NULL)));
 
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64MaskTestHBD_d16,
+    ::testing::Values(TestFuncsHBD_d16(aom_highbd_blend_a64_d16_mask_c,
+                                       aom_highbd_blend_a64_d16_mask_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, BlendA64MaskTestHBD_d16,
+    ::testing::Values(TestFuncsHBD_d16(aom_highbd_blend_a64_d16_mask_c,
+                                       aom_highbd_blend_a64_d16_mask_avx2)));
+#endif  // HAVE_AVX2
+
 // TODO(slavarnway): Enable the following in the avx2 commit. (56501)
 #if 0
 #if HAVE_AVX2
diff --git a/third_party/aom/test/borders_test.cc b/third_party/aom/test/borders_test.cc
index 893237ef30..31eacab12e 100644
--- a/third_party/aom/test/borders_test.cc
+++ b/third_party/aom/test/borders_test.cc
@@ -33,7 +33,7 @@ class BordersTestLarge
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, 1);
       encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
       encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
diff --git a/third_party/aom/test/cdef_test.cc b/third_party/aom/test/cdef_test.cc
index becc072918..8201818795 100644
--- a/third_party/aom/test/cdef_test.cc
+++ b/third_party/aom/test/cdef_test.cc
@@ -116,7 +116,7 @@ void test_cdef(int bsize, int iterations, cdef_filter_block_func cdef,
                   ref_cdef(depth == 8 ? (uint8_t *)ref_d : 0, ref_d, size,
                            s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
                            pristrength, secstrength, dir, pridamping,
-                           secdamping, bsize, (1 << depth) - 1, depth - 8);
+                           secdamping, bsize, depth - 8);
                   // If cdef and ref_cdef are the same, we're just testing
                   // speed
                   if (cdef != ref_cdef)
@@ -124,7 +124,7 @@ void test_cdef(int bsize, int iterations, cdef_filter_block_func cdef,
                         cdef(depth == 8 ? (uint8_t *)d : 0, d, size,
                              s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
                              pristrength, secstrength, dir, pridamping,
-                             secdamping, bsize, (1 << depth) - 1, depth - 8));
+                             secdamping, bsize, depth - 8));
                   if (ref_cdef != cdef) {
                     for (pos = 0; pos < max_pos && !error; pos++) {
                       error = ref_d[pos] != d[pos];
diff --git a/third_party/aom/test/cfl_test.cc b/third_party/aom/test/cfl_test.cc
index e4d438d6ab..f087dd9fbd 100644
--- a/third_party/aom/test/cfl_test.cc
+++ b/third_party/aom/test/cfl_test.cc
@@ -84,7 +84,7 @@ class CFLTest {
     tx_size = tx;
     width = tx_size_wide[tx_size];
     height = tx_size_high[tx_size];
-    rnd(ACMRandom::DeterministicSeed());
+    rnd.Reset(ACMRandom::DeterministicSeed());
   }
 
  protected:
diff --git a/third_party/aom/test/comp_avg_pred_test.cc b/third_party/aom/test/comp_avg_pred_test.cc
index 9ad8973f08..3e5632e7eb 100644
--- a/third_party/aom/test/comp_avg_pred_test.cc
+++ b/third_party/aom/test/comp_avg_pred_test.cc
@@ -11,62 +11,66 @@
 
 #include "test/comp_avg_pred_test.h"
 
+using libaom_test::ACMRandom;
+using libaom_test::AV1DISTWTDCOMPAVG::AV1DISTWTDCOMPAVGTest;
+using libaom_test::AV1DISTWTDCOMPAVG::AV1DISTWTDCOMPAVGUPSAMPLEDTest;
+using libaom_test::AV1DISTWTDCOMPAVG::AV1HighBDDISTWTDCOMPAVGTest;
+using libaom_test::AV1DISTWTDCOMPAVG::AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest;
 using ::testing::make_tuple;
 using ::testing::tuple;
-using libaom_test::ACMRandom;
-using libaom_test::AV1JNTCOMPAVG::AV1HighBDJNTCOMPAVGTest;
-using libaom_test::AV1JNTCOMPAVG::AV1HighBDJNTCOMPAVGUPSAMPLEDTest;
-using libaom_test::AV1JNTCOMPAVG::AV1JNTCOMPAVGTest;
-using libaom_test::AV1JNTCOMPAVG::AV1JNTCOMPAVGUPSAMPLEDTest;
 
 namespace {
 
-TEST_P(AV1JNTCOMPAVGTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
+TEST_P(AV1DISTWTDCOMPAVGTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
 
-TEST_P(AV1JNTCOMPAVGTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
+TEST_P(AV1DISTWTDCOMPAVGTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
 
 #if HAVE_SSSE3
-INSTANTIATE_TEST_CASE_P(
-    SSSE3, AV1JNTCOMPAVGTest,
-    libaom_test::AV1JNTCOMPAVG::BuildParams(aom_jnt_comp_avg_pred_ssse3));
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1DISTWTDCOMPAVGTest,
+                        libaom_test::AV1DISTWTDCOMPAVG::BuildParams(
+                            aom_dist_wtd_comp_avg_pred_ssse3));
 #endif
 
-TEST_P(AV1JNTCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
+TEST_P(AV1DISTWTDCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
   RunSpeedTest(GET_PARAM(0));
 }
 
-TEST_P(AV1JNTCOMPAVGUPSAMPLEDTest, CheckOutput) {
+TEST_P(AV1DISTWTDCOMPAVGUPSAMPLEDTest, CheckOutput) {
   RunCheckOutput(GET_PARAM(0));
 }
 
 #if HAVE_SSSE3
-INSTANTIATE_TEST_CASE_P(SSSE3, AV1JNTCOMPAVGUPSAMPLEDTest,
-                        libaom_test::AV1JNTCOMPAVG::BuildParams(
-                            aom_jnt_comp_avg_upsampled_pred_ssse3));
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1DISTWTDCOMPAVGUPSAMPLEDTest,
+                        libaom_test::AV1DISTWTDCOMPAVG::BuildParams(
+                            aom_dist_wtd_comp_avg_upsampled_pred_ssse3));
 #endif
 
-TEST_P(AV1HighBDJNTCOMPAVGTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(1)); }
-
-TEST_P(AV1HighBDJNTCOMPAVGTest, CheckOutput) { RunCheckOutput(GET_PARAM(1)); }
-
-#if HAVE_SSE2
-INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDJNTCOMPAVGTest,
-                        libaom_test::AV1JNTCOMPAVG::BuildParams(
-                            aom_highbd_jnt_comp_avg_pred_sse2, 1));
-#endif
-
-TEST_P(AV1HighBDJNTCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
+TEST_P(AV1HighBDDISTWTDCOMPAVGTest, DISABLED_Speed) {
   RunSpeedTest(GET_PARAM(1));
 }
 
-TEST_P(AV1HighBDJNTCOMPAVGUPSAMPLEDTest, CheckOutput) {
+TEST_P(AV1HighBDDISTWTDCOMPAVGTest, CheckOutput) {
   RunCheckOutput(GET_PARAM(1));
 }
 
 #if HAVE_SSE2
-INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDJNTCOMPAVGUPSAMPLEDTest,
-                        libaom_test::AV1JNTCOMPAVG::BuildParams(
-                            aom_highbd_jnt_comp_avg_upsampled_pred_sse2));
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDDISTWTDCOMPAVGTest,
+                        libaom_test::AV1DISTWTDCOMPAVG::BuildParams(
+                            aom_highbd_dist_wtd_comp_avg_pred_sse2, 1));
+#endif
+
+TEST_P(AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(1));
+}
+
+TEST_P(AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(1));
+}
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest,
+                        libaom_test::AV1DISTWTDCOMPAVG::BuildParams(
+                            aom_highbd_dist_wtd_comp_avg_upsampled_pred_sse2));
 #endif
 
 }  // namespace
diff --git a/third_party/aom/test/comp_avg_pred_test.h b/third_party/aom/test/comp_avg_pred_test.h
index 9661dd9f54..01ea35d59b 100644
--- a/third_party/aom/test/comp_avg_pred_test.h
+++ b/third_party/aom/test/comp_avg_pred_test.h
@@ -25,72 +25,73 @@
 namespace libaom_test {
 const int kMaxSize = 128 + 32;  // padding
 
-namespace AV1JNTCOMPAVG {
+namespace AV1DISTWTDCOMPAVG {
 
-typedef void (*jntcompavg_func)(uint8_t *comp_pred, const uint8_t *pred,
-                                int width, int height, const uint8_t *ref,
-                                int ref_stride,
-                                const JNT_COMP_PARAMS *jcp_param);
+typedef void (*distwtdcompavg_func)(uint8_t *comp_pred, const uint8_t *pred,
+                                    int width, int height, const uint8_t *ref,
+                                    int ref_stride,
+                                    const DIST_WTD_COMP_PARAMS *jcp_param);
 
-typedef void (*jntcompavgupsampled_func)(
+typedef void (*distwtdcompavgupsampled_func)(
     MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
-    int ref_stride, const JNT_COMP_PARAMS *jcp_param, int subpel_search);
+    int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search);
 
-typedef void (*highbdjntcompavgupsampled_func)(
+typedef void (*highbddistwtdcompavgupsampled_func)(
     MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
     const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
     int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
-    int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param,
+    int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param,
     int subpel_search);
 
-typedef ::testing::tuple<jntcompavg_func, BLOCK_SIZE> JNTCOMPAVGParam;
+typedef ::testing::tuple<distwtdcompavg_func, BLOCK_SIZE> DISTWTDCOMPAVGParam;
 
-typedef ::testing::tuple<jntcompavgupsampled_func, BLOCK_SIZE>
-    JNTCOMPAVGUPSAMPLEDParam;
+typedef ::testing::tuple<distwtdcompavgupsampled_func, BLOCK_SIZE>
+    DISTWTDCOMPAVGUPSAMPLEDParam;
 
-typedef ::testing::tuple<int, jntcompavg_func, BLOCK_SIZE>
-    HighbdJNTCOMPAVGParam;
+typedef ::testing::tuple<int, distwtdcompavg_func, BLOCK_SIZE>
+    HighbdDISTWTDCOMPAVGParam;
 
-typedef ::testing::tuple<int, highbdjntcompavgupsampled_func, BLOCK_SIZE>
-    HighbdJNTCOMPAVGUPSAMPLEDParam;
+typedef ::testing::tuple<int, highbddistwtdcompavgupsampled_func, BLOCK_SIZE>
+    HighbdDISTWTDCOMPAVGUPSAMPLEDParam;
 
-::testing::internal::ParamGenerator<JNTCOMPAVGParam> BuildParams(
-    jntcompavg_func filter) {
+::testing::internal::ParamGenerator<DISTWTDCOMPAVGParam> BuildParams(
+    distwtdcompavg_func filter) {
   return ::testing::Combine(::testing::Values(filter),
                             ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
 }
 
-::testing::internal::ParamGenerator<JNTCOMPAVGUPSAMPLEDParam> BuildParams(
-    jntcompavgupsampled_func filter) {
+::testing::internal::ParamGenerator<DISTWTDCOMPAVGUPSAMPLEDParam> BuildParams(
+    distwtdcompavgupsampled_func filter) {
   return ::testing::Combine(::testing::Values(filter),
                             ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
 }
 
-::testing::internal::ParamGenerator<HighbdJNTCOMPAVGParam> BuildParams(
-    jntcompavg_func filter, int is_hbd) {
+::testing::internal::ParamGenerator<HighbdDISTWTDCOMPAVGParam> BuildParams(
+    distwtdcompavg_func filter, int is_hbd) {
   (void)is_hbd;
   return ::testing::Combine(::testing::Range(8, 13, 2),
                             ::testing::Values(filter),
                             ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
 }
 
-::testing::internal::ParamGenerator<HighbdJNTCOMPAVGUPSAMPLEDParam> BuildParams(
-    highbdjntcompavgupsampled_func filter) {
+::testing::internal::ParamGenerator<HighbdDISTWTDCOMPAVGUPSAMPLEDParam>
+BuildParams(highbddistwtdcompavgupsampled_func filter) {
   return ::testing::Combine(::testing::Range(8, 13, 2),
                             ::testing::Values(filter),
                             ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
 }
 
-class AV1JNTCOMPAVGTest : public ::testing::TestWithParam<JNTCOMPAVGParam> {
+class AV1DISTWTDCOMPAVGTest
+    : public ::testing::TestWithParam<DISTWTDCOMPAVGParam> {
  public:
-  ~AV1JNTCOMPAVGTest() {}
+  ~AV1DISTWTDCOMPAVGTest() {}
   void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
   void TearDown() { libaom_test::ClearSystemState(); }
 
  protected:
-  void RunCheckOutput(jntcompavg_func test_impl) {
+  void RunCheckOutput(distwtdcompavg_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(1);
 
@@ -107,27 +108,27 @@ class AV1JNTCOMPAVGTest : public ::testing::TestWithParam<JNTCOMPAVGParam> {
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
     for (int ii = 0; ii < 2; ii++) {
       for (int jj = 0; jj < 4; jj++) {
-        jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
-        jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+        dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+        dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
 
         const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
         const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
-        aom_jnt_comp_avg_pred_c(output, pred8 + offset_r * w + offset_c, in_w,
-                                in_h, ref8 + offset_r * w + offset_c, in_w,
-                                &jnt_comp_params);
+        aom_dist_wtd_comp_avg_pred_c(output, pred8 + offset_r * w + offset_c,
+                                     in_w, in_h, ref8 + offset_r * w + offset_c,
+                                     in_w, &dist_wtd_comp_params);
         test_impl(output2, pred8 + offset_r * w + offset_c, in_w, in_h,
-                  ref8 + offset_r * w + offset_c, in_w, &jnt_comp_params);
+                  ref8 + offset_r * w + offset_c, in_w, &dist_wtd_comp_params);
 
         for (int i = 0; i < in_h; ++i) {
           for (int j = 0; j < in_w; ++j) {
             int idx = i * in_w + j;
             ASSERT_EQ(output[idx], output2[idx])
-                << "Mismatch at unit tests for AV1JNTCOMPAVGTest\n"
+                << "Mismatch at unit tests for AV1DISTWTDCOMPAVGTest\n"
                 << in_w << "x" << in_h << " Pixel mismatch at index " << idx
                 << " = (" << i << ", " << j << ")";
           }
@@ -135,7 +136,7 @@ class AV1JNTCOMPAVGTest : public ::testing::TestWithParam<JNTCOMPAVGParam> {
       }
     }
   }
-  void RunSpeedTest(jntcompavg_func test_impl) {
+  void RunSpeedTest(distwtdcompavg_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(1);
 
@@ -152,49 +153,49 @@ class AV1JNTCOMPAVGTest : public ::testing::TestWithParam<JNTCOMPAVGParam> {
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
-    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
-    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+    dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
 
     const int num_loops = 1000000000 / (in_w + in_h);
     aom_usec_timer timer;
     aom_usec_timer_start(&timer);
 
     for (int i = 0; i < num_loops; ++i)
-      aom_jnt_comp_avg_pred_c(output, pred8, in_w, in_h, ref8, in_w,
-                              &jnt_comp_params);
+      aom_dist_wtd_comp_avg_pred_c(output, pred8, in_w, in_h, ref8, in_w,
+                                   &dist_wtd_comp_params);
 
     aom_usec_timer_mark(&timer);
     const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
-    printf("jntcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("distwtdcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time / num_loops);
 
     aom_usec_timer timer1;
     aom_usec_timer_start(&timer1);
 
     for (int i = 0; i < num_loops; ++i)
-      test_impl(output2, pred8, in_w, in_h, ref8, in_w, &jnt_comp_params);
+      test_impl(output2, pred8, in_w, in_h, ref8, in_w, &dist_wtd_comp_params);
 
     aom_usec_timer_mark(&timer1);
     const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
-    printf("jntcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("distwtdcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time1 / num_loops);
   }
 
   libaom_test::ACMRandom rnd_;
-};  // class AV1JNTCOMPAVGTest
+};  // class AV1DISTWTDCOMPAVGTest
 
-class AV1JNTCOMPAVGUPSAMPLEDTest
-    : public ::testing::TestWithParam<JNTCOMPAVGUPSAMPLEDParam> {
+class AV1DISTWTDCOMPAVGUPSAMPLEDTest
+    : public ::testing::TestWithParam<DISTWTDCOMPAVGUPSAMPLEDParam> {
  public:
-  ~AV1JNTCOMPAVGUPSAMPLEDTest() {}
+  ~AV1DISTWTDCOMPAVGUPSAMPLEDTest() {}
   void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
   void TearDown() { libaom_test::ClearSystemState(); }
 
  protected:
-  void RunCheckOutput(jntcompavgupsampled_func test_impl) {
+  void RunCheckOutput(distwtdcompavgupsampled_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(1);
 
@@ -211,37 +212,40 @@ class AV1JNTCOMPAVGUPSAMPLEDTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
     int sub_x_q3, sub_y_q3;
     int subpel_search;
-    for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+    for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+         ++subpel_search) {
       for (sub_x_q3 = 0; sub_x_q3 < 8; ++sub_x_q3) {
         for (sub_y_q3 = 0; sub_y_q3 < 8; ++sub_y_q3) {
           for (int ii = 0; ii < 2; ii++) {
             for (int jj = 0; jj < 4; jj++) {
-              jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
-              jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              dist_wtd_comp_params.fwd_offset =
+                  quant_dist_lookup_table[ii][jj][0];
+              dist_wtd_comp_params.bck_offset =
+                  quant_dist_lookup_table[ii][jj][1];
 
               const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
               const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
 
-              aom_jnt_comp_avg_upsampled_pred_c(
+              aom_dist_wtd_comp_avg_upsampled_pred_c(
                   NULL, NULL, 0, 0, NULL, output,
                   pred8 + offset_r * w + offset_c, in_w, in_h, sub_x_q3,
                   sub_y_q3, ref8 + offset_r * w + offset_c, in_w,
-                  &jnt_comp_params, subpel_search);
+                  &dist_wtd_comp_params, subpel_search);
               test_impl(NULL, NULL, 0, 0, NULL, output2,
                         pred8 + offset_r * w + offset_c, in_w, in_h, sub_x_q3,
                         sub_y_q3, ref8 + offset_r * w + offset_c, in_w,
-                        &jnt_comp_params, subpel_search);
+                        &dist_wtd_comp_params, subpel_search);
 
               for (int i = 0; i < in_h; ++i) {
                 for (int j = 0; j < in_w; ++j) {
                   int idx = i * in_w + j;
                   ASSERT_EQ(output[idx], output2[idx])
                       << "Mismatch at unit tests for "
-                         "AV1JNTCOMPAVGUPSAMPLEDTest\n"
+                         "AV1DISTWTDCOMPAVGUPSAMPLEDTest\n"
                       << in_w << "x" << in_h << " Pixel mismatch at index "
                       << idx << " = (" << i << ", " << j
                       << "), sub pixel offset = (" << sub_y_q3 << ", "
@@ -254,7 +258,7 @@ class AV1JNTCOMPAVGUPSAMPLEDTest
       }
     }
   }
-  void RunSpeedTest(jntcompavgupsampled_func test_impl) {
+  void RunSpeedTest(distwtdcompavgupsampled_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(1);
 
@@ -271,11 +275,11 @@ class AV1JNTCOMPAVGUPSAMPLEDTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
-    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
-    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+    dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
 
     int sub_x_q3 = 0;
     int sub_y_q3 = 0;
@@ -283,16 +287,16 @@ class AV1JNTCOMPAVGUPSAMPLEDTest
     const int num_loops = 1000000000 / (in_w + in_h);
     aom_usec_timer timer;
     aom_usec_timer_start(&timer);
-    int subpel_search = 2;  // set to 1 to test 4-tap filter.
+    int subpel_search = USE_8_TAPS;  // set to USE_4_TAPS to test 4-tap filter.
 
     for (int i = 0; i < num_loops; ++i)
-      aom_jnt_comp_avg_upsampled_pred_c(NULL, NULL, 0, 0, NULL, output, pred8,
-                                        in_w, in_h, sub_x_q3, sub_y_q3, ref8,
-                                        in_w, &jnt_comp_params, subpel_search);
+      aom_dist_wtd_comp_avg_upsampled_pred_c(
+          NULL, NULL, 0, 0, NULL, output, pred8, in_w, in_h, sub_x_q3, sub_y_q3,
+          ref8, in_w, &dist_wtd_comp_params, subpel_search);
 
     aom_usec_timer_mark(&timer);
     const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
-    printf("jntcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("distwtdcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time / num_loops);
 
     aom_usec_timer timer1;
@@ -300,27 +304,27 @@ class AV1JNTCOMPAVGUPSAMPLEDTest
 
     for (int i = 0; i < num_loops; ++i)
       test_impl(NULL, NULL, 0, 0, NULL, output2, pred8, in_w, in_h, sub_x_q3,
-                sub_y_q3, ref8, in_w, &jnt_comp_params, subpel_search);
+                sub_y_q3, ref8, in_w, &dist_wtd_comp_params, subpel_search);
 
     aom_usec_timer_mark(&timer1);
     const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
-    printf("jntcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("distwtdcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time1 / num_loops);
   }
 
   libaom_test::ACMRandom rnd_;
-};  // class AV1JNTCOMPAVGUPSAMPLEDTest
+};  // class AV1DISTWTDCOMPAVGUPSAMPLEDTest
 
-class AV1HighBDJNTCOMPAVGTest
-    : public ::testing::TestWithParam<HighbdJNTCOMPAVGParam> {
+class AV1HighBDDISTWTDCOMPAVGTest
+    : public ::testing::TestWithParam<HighbdDISTWTDCOMPAVGParam> {
  public:
-  ~AV1HighBDJNTCOMPAVGTest() {}
+  ~AV1HighBDDISTWTDCOMPAVGTest() {}
   void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
 
   void TearDown() { libaom_test::ClearSystemState(); }
 
  protected:
-  void RunCheckOutput(jntcompavg_func test_impl) {
+  void RunCheckOutput(distwtdcompavg_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(2);
     const int bd = GET_PARAM(0);
@@ -337,31 +341,31 @@ class AV1HighBDJNTCOMPAVGTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
     for (int ii = 0; ii < 2; ii++) {
       for (int jj = 0; jj < 4; jj++) {
-        jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
-        jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+        dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+        dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
 
         const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
         const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
-        aom_highbd_jnt_comp_avg_pred_c(
+        aom_highbd_dist_wtd_comp_avg_pred_c(
             CONVERT_TO_BYTEPTR(output),
             CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w, in_h,
             CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c, in_w,
-            &jnt_comp_params);
+            &dist_wtd_comp_params);
         test_impl(CONVERT_TO_BYTEPTR(output2),
                   CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w,
                   in_h, CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c,
-                  in_w, &jnt_comp_params);
+                  in_w, &dist_wtd_comp_params);
 
         for (int i = 0; i < in_h; ++i) {
           for (int j = 0; j < in_w; ++j) {
             int idx = i * in_w + j;
             ASSERT_EQ(output[idx], output2[idx])
-                << "Mismatch at unit tests for AV1HighBDJNTCOMPAVGTest\n"
+                << "Mismatch at unit tests for AV1HighBDDISTWTDCOMPAVGTest\n"
                 << in_w << "x" << in_h << " Pixel mismatch at index " << idx
                 << " = (" << i << ", " << j << ")";
           }
@@ -369,7 +373,7 @@ class AV1HighBDJNTCOMPAVGTest
       }
     }
   }
-  void RunSpeedTest(jntcompavg_func test_impl) {
+  void RunSpeedTest(distwtdcompavg_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(2);
     const int bd = GET_PARAM(0);
@@ -386,24 +390,24 @@ class AV1HighBDJNTCOMPAVGTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
-    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
-    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+    dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
 
     const int num_loops = 1000000000 / (in_w + in_h);
     aom_usec_timer timer;
     aom_usec_timer_start(&timer);
 
     for (int i = 0; i < num_loops; ++i)
-      aom_highbd_jnt_comp_avg_pred_c(
+      aom_highbd_dist_wtd_comp_avg_pred_c(
           CONVERT_TO_BYTEPTR(output), CONVERT_TO_BYTEPTR(pred8), in_w, in_h,
-          CONVERT_TO_BYTEPTR(ref8), in_w, &jnt_comp_params);
+          CONVERT_TO_BYTEPTR(ref8), in_w, &dist_wtd_comp_params);
 
     aom_usec_timer_mark(&timer);
     const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
-    printf("highbdjntcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("highbddistwtdcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time / num_loops);
 
     aom_usec_timer timer1;
@@ -411,33 +415,33 @@ class AV1HighBDJNTCOMPAVGTest
 
     for (int i = 0; i < num_loops; ++i)
       test_impl(CONVERT_TO_BYTEPTR(output2), CONVERT_TO_BYTEPTR(pred8), in_w,
-                in_h, CONVERT_TO_BYTEPTR(ref8), in_w, &jnt_comp_params);
+                in_h, CONVERT_TO_BYTEPTR(ref8), in_w, &dist_wtd_comp_params);
 
     aom_usec_timer_mark(&timer1);
     const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
-    printf("highbdjntcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+    printf("highbddistwtdcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
            1000.0 * elapsed_time1 / num_loops);
   }
 
   libaom_test::ACMRandom rnd_;
-};  // class AV1HighBDJNTCOMPAVGTest
+};  // class AV1HighBDDISTWTDCOMPAVGTest
 
-class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
-    : public ::testing::TestWithParam<HighbdJNTCOMPAVGUPSAMPLEDParam> {
+class AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest
+    : public ::testing::TestWithParam<HighbdDISTWTDCOMPAVGUPSAMPLEDParam> {
  public:
-  ~AV1HighBDJNTCOMPAVGUPSAMPLEDTest() {}
+  ~AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest() {}
   void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
   void TearDown() { libaom_test::ClearSystemState(); }
 
  protected:
-  void RunCheckOutput(highbdjntcompavgupsampled_func test_impl) {
+  void RunCheckOutput(highbddistwtdcompavgupsampled_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(2);
     const int bd = GET_PARAM(0);
     uint16_t pred8[kMaxSize * kMaxSize];
     uint16_t ref8[kMaxSize * kMaxSize];
-    uint16_t output[kMaxSize * kMaxSize];
-    uint16_t output2[kMaxSize * kMaxSize];
+    DECLARE_ALIGNED(16, uint16_t, output[kMaxSize * kMaxSize]);
+    DECLARE_ALIGNED(16, uint16_t, output2[kMaxSize * kMaxSize]);
 
     for (int i = 0; i < h; ++i)
       for (int j = 0; j < w; ++j) {
@@ -447,39 +451,42 @@ class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
     int sub_x_q3, sub_y_q3;
     int subpel_search;
-    for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+    for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+         ++subpel_search) {
       for (sub_x_q3 = 0; sub_x_q3 < 8; ++sub_x_q3) {
         for (sub_y_q3 = 0; sub_y_q3 < 8; ++sub_y_q3) {
           for (int ii = 0; ii < 2; ii++) {
             for (int jj = 0; jj < 4; jj++) {
-              jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
-              jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              dist_wtd_comp_params.fwd_offset =
+                  quant_dist_lookup_table[ii][jj][0];
+              dist_wtd_comp_params.bck_offset =
+                  quant_dist_lookup_table[ii][jj][1];
 
               const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
               const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
 
-              aom_highbd_jnt_comp_avg_upsampled_pred_c(
+              aom_highbd_dist_wtd_comp_avg_upsampled_pred_c(
                   NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output),
                   CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w,
                   in_h, sub_x_q3, sub_y_q3,
                   CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c, in_w, bd,
-                  &jnt_comp_params, subpel_search);
+                  &dist_wtd_comp_params, subpel_search);
               test_impl(NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output2),
                         CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c,
                         in_w, in_h, sub_x_q3, sub_y_q3,
                         CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c,
-                        in_w, bd, &jnt_comp_params, subpel_search);
+                        in_w, bd, &dist_wtd_comp_params, subpel_search);
 
               for (int i = 0; i < in_h; ++i) {
                 for (int j = 0; j < in_w; ++j) {
                   int idx = i * in_w + j;
                   ASSERT_EQ(output[idx], output2[idx])
                       << "Mismatch at unit tests for "
-                         "AV1HighBDJNTCOMPAVGUPSAMPLEDTest\n"
+                         "AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest\n"
                       << in_w << "x" << in_h << " Pixel mismatch at index "
                       << idx << " = (" << i << ", " << j
                       << "), sub pixel offset = (" << sub_y_q3 << ", "
@@ -492,14 +499,14 @@ class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
       }
     }
   }
-  void RunSpeedTest(highbdjntcompavgupsampled_func test_impl) {
+  void RunSpeedTest(highbddistwtdcompavgupsampled_func test_impl) {
     const int w = kMaxSize, h = kMaxSize;
     const int block_idx = GET_PARAM(2);
     const int bd = GET_PARAM(0);
     uint16_t pred8[kMaxSize * kMaxSize];
     uint16_t ref8[kMaxSize * kMaxSize];
-    uint16_t output[kMaxSize * kMaxSize];
-    uint16_t output2[kMaxSize * kMaxSize];
+    DECLARE_ALIGNED(16, uint16_t, output[kMaxSize * kMaxSize]);
+    DECLARE_ALIGNED(16, uint16_t, output2[kMaxSize * kMaxSize]);
 
     for (int i = 0; i < h; ++i)
       for (int j = 0; j < w; ++j) {
@@ -509,27 +516,28 @@ class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
     const int in_w = block_size_wide[block_idx];
     const int in_h = block_size_high[block_idx];
 
-    JNT_COMP_PARAMS jnt_comp_params;
-    jnt_comp_params.use_jnt_comp_avg = 1;
+    DIST_WTD_COMP_PARAMS dist_wtd_comp_params;
+    dist_wtd_comp_params.use_dist_wtd_comp_avg = 1;
 
-    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
-    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+    dist_wtd_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    dist_wtd_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
     int sub_x_q3 = 0;
     int sub_y_q3 = 0;
     const int num_loops = 1000000000 / (in_w + in_h);
     aom_usec_timer timer;
     aom_usec_timer_start(&timer);
-    int subpel_search = 2;  // set to 1 to test 4-tap filter.
+    int subpel_search = USE_8_TAPS;  // set to USE_4_TAPS to test 4-tap filter.
     for (int i = 0; i < num_loops; ++i)
-      aom_highbd_jnt_comp_avg_upsampled_pred_c(
+      aom_highbd_dist_wtd_comp_avg_upsampled_pred_c(
           NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output),
           CONVERT_TO_BYTEPTR(pred8), in_w, in_h, sub_x_q3, sub_y_q3,
-          CONVERT_TO_BYTEPTR(ref8), in_w, bd, &jnt_comp_params, subpel_search);
+          CONVERT_TO_BYTEPTR(ref8), in_w, bd, &dist_wtd_comp_params,
+          subpel_search);
 
     aom_usec_timer_mark(&timer);
     const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
-    printf("highbdjntcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
-           1000.0 * elapsed_time / num_loops);
+    printf("highbddistwtdcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w,
+           in_h, 1000.0 * elapsed_time / num_loops);
 
     aom_usec_timer timer1;
     aom_usec_timer_start(&timer1);
@@ -537,19 +545,19 @@ class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
     for (int i = 0; i < num_loops; ++i)
       test_impl(NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output2),
                 CONVERT_TO_BYTEPTR(pred8), in_w, in_h, sub_x_q3, sub_y_q3,
-                CONVERT_TO_BYTEPTR(ref8), in_w, bd, &jnt_comp_params,
+                CONVERT_TO_BYTEPTR(ref8), in_w, bd, &dist_wtd_comp_params,
                 subpel_search);
 
     aom_usec_timer_mark(&timer1);
     const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
-    printf("highbdjntcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w,
+    printf("highbddistwtdcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w,
            in_h, 1000.0 * elapsed_time1 / num_loops);
   }
 
   libaom_test::ACMRandom rnd_;
-};  // class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
+};  // class AV1HighBDDISTWTDCOMPAVGUPSAMPLEDTest
 
-}  // namespace AV1JNTCOMPAVG
+}  // namespace AV1DISTWTDCOMPAVG
 }  // namespace libaom_test
 
 #endif  // AOM_TEST_COMP_AVG_PRED_TEST_H_
diff --git a/third_party/aom/test/comp_mask_variance_test.cc b/third_party/aom/test/comp_mask_variance_test.cc
index 34be2aa6db..53ba56c819 100644
--- a/third_party/aom/test/comp_mask_variance_test.cc
+++ b/third_party/aom/test/comp_mask_variance_test.cc
@@ -34,7 +34,7 @@ typedef void (*comp_mask_pred_func)(uint8_t *comp_pred, const uint8_t *pred,
                                     int ref_stride, const uint8_t *mask,
                                     int mask_stride, int invert_mask);
 
-#if HAVE_SSSE3 || HAVE_AV2
+#if HAVE_SSSE3 || HAVE_SSE2 || HAVE_AV2
 const BLOCK_SIZE kValidBlockSize[] = {
   BLOCK_8X8,   BLOCK_8X16, BLOCK_8X32,  BLOCK_16X8,  BLOCK_16X16,
   BLOCK_16X32, BLOCK_32X8, BLOCK_32X16, BLOCK_32X32,
@@ -191,7 +191,8 @@ void AV1CompMaskUpVarianceTest::RunCheckOutput(comp_mask_pred_func test_impl,
   const int h = block_size_high[bsize];
   int wedge_types = (1 << get_wedge_bits_lookup(bsize));
   int subpel_search;
-  for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+  for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+       ++subpel_search) {
     // loop through subx and suby
     for (int sub = 0; sub < 8 * 8; ++sub) {
       int subx = sub & 0x7;
@@ -231,7 +232,7 @@ void AV1CompMaskUpVarianceTest::RunSpeedTest(comp_mask_pred_func test_impl,
   const int num_loops = 1000000000 / (w + h);
   comp_mask_pred_func funcs[2] = { &aom_comp_mask_pred_c, test_impl };
   double elapsed_time[2] = { 0 };
-  int subpel_search = 2;  // set to 1 to test 4-tap filter.
+  int subpel_search = USE_8_TAPS;  // set to USE_4_TAPS to test 4-tap filter.
   for (int i = 0; i < 2; ++i) {
     aom_usec_timer timer;
     aom_usec_timer_start(&timer);
@@ -458,6 +459,7 @@ AV1HighbdCompMaskUpVarianceTest::~AV1HighbdCompMaskUpVarianceTest() { ; }
 
 void AV1HighbdCompMaskUpVarianceTest::RunCheckOutput(
     highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int inv) {
+  (void)test_impl;
   int bd_ = GET_PARAM(2);
   const int w = block_size_wide[bsize];
   const int h = block_size_high[bsize];
@@ -480,19 +482,24 @@ void AV1HighbdCompMaskUpVarianceTest::RunCheckOutput(
         const uint8_t *mask =
             av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
 
-        aom_highbd_comp_mask_pred = aom_highbd_comp_mask_pred_c;  // ref
-        aom_highbd_comp_mask_upsampled_pred(
-            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred1_),
-            CONVERT_TO_BYTEPTR(pred_), w, h, subx, suby,
-            CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv, bd_,
-            subpel_search);
+        // ref
+        aom_highbd_upsampled_pred_c(
+            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred1_), w, h, subx,
+            suby, CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, bd_, subpel_search);
+
+        aom_highbd_comp_mask_pred_c(
+            CONVERT_TO_BYTEPTR(comp_pred1_), CONVERT_TO_BYTEPTR(pred_), w, h,
+            CONVERT_TO_BYTEPTR(comp_pred1_), w, mask, w, inv);
+
+        // test
+        aom_highbd_upsampled_pred(
+            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred2_), w, h, subx,
+            suby, CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, bd_, subpel_search);
+
+        aom_highbd_comp_mask_pred(
+            CONVERT_TO_BYTEPTR(comp_pred2_), CONVERT_TO_BYTEPTR(pred_), w, h,
+            CONVERT_TO_BYTEPTR(comp_pred2_), w, mask, w, inv);
 
-        aom_highbd_comp_mask_pred = test_impl;  // test
-        aom_highbd_comp_mask_upsampled_pred(
-            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred2_),
-            CONVERT_TO_BYTEPTR(pred_), w, h, subx, suby,
-            CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv, bd_,
-            subpel_search);
         ASSERT_EQ(CheckResult(w, h), true)
             << " wedge " << wedge_index << " inv " << inv << "sub (" << subx
             << "," << suby << ")";
diff --git a/third_party/aom/test/convolve_test.cc b/third_party/aom/test/convolve_test.cc
index de3f47628a..760d2c579d 100644
--- a/third_party/aom/test/convolve_test.cc
+++ b/third_party/aom/test/convolve_test.cc
@@ -273,6 +273,8 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
     input_ = reinterpret_cast<uint8_t *>(
                  aom_memalign(kDataAlignment, kInputBufferSize + 1)) +
              1;
+    ref8_ = reinterpret_cast<uint8_t *>(
+        aom_memalign(kDataAlignment, kOutputStride * kMaxDimension));
     output_ = reinterpret_cast<uint8_t *>(
         aom_memalign(kDataAlignment, kOutputBufferSize));
     output_ref_ = reinterpret_cast<uint8_t *>(
@@ -280,6 +282,8 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
     input16_ = reinterpret_cast<uint16_t *>(aom_memalign(
                    kDataAlignment, (kInputBufferSize + 1) * sizeof(uint16_t))) +
                1;
+    ref16_ = reinterpret_cast<uint16_t *>(aom_memalign(
+        kDataAlignment, kOutputStride * kMaxDimension * sizeof(uint16_t)));
     output16_ = reinterpret_cast<uint16_t *>(
         aom_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
     output16_ref_ = reinterpret_cast<uint16_t *>(
@@ -291,12 +295,16 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
   static void TearDownTestCase() {
     aom_free(input_ - 1);
     input_ = NULL;
+    aom_free(ref8_);
+    ref8_ = NULL;
     aom_free(output_);
     output_ = NULL;
     aom_free(output_ref_);
     output_ref_ = NULL;
     aom_free(input16_ - 1);
     input16_ = NULL;
+    aom_free(ref16_);
+    ref16_ = NULL;
     aom_free(output16_);
     output16_ = NULL;
     aom_free(output16_ref_);
@@ -449,18 +457,22 @@ class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
 
   const ConvolveFunctions *UUT_;
   static uint8_t *input_;
+  static uint8_t *ref8_;
   static uint8_t *output_;
   static uint8_t *output_ref_;
   static uint16_t *input16_;
+  static uint16_t *ref16_;
   static uint16_t *output16_;
   static uint16_t *output16_ref_;
   int mask_;
 };
 
 uint8_t *ConvolveTest::input_ = NULL;
+uint8_t *ConvolveTest::ref8_ = NULL;
 uint8_t *ConvolveTest::output_ = NULL;
 uint8_t *ConvolveTest::output_ref_ = NULL;
 uint16_t *ConvolveTest::input16_ = NULL;
+uint16_t *ConvolveTest::ref16_ = NULL;
 uint16_t *ConvolveTest::output16_ = NULL;
 uint16_t *ConvolveTest::output16_ref_ = NULL;
 
@@ -486,26 +498,28 @@ const int kNumFilterBanks = SWITCHABLE_FILTERS;
 const int kNumFilters = 16;
 
 TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
-  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
-    const InterpFilter filter = (InterpFilter)filter_bank;
-    const InterpKernel *filters =
-        (const InterpKernel *)av1_get_interp_filter_kernel(filter);
-    const InterpFilterParams *filter_params =
-        av1_get_interp_filter_params_with_block_size(filter, 8);
-    if (filter_params->taps != SUBPEL_TAPS) continue;
-    for (int i = 0; i < kNumFilters; i++) {
-      const int p0 = filters[i][0] + filters[i][1];
-      const int p1 = filters[i][2] + filters[i][3];
-      const int p2 = filters[i][4] + filters[i][5];
-      const int p3 = filters[i][6] + filters[i][7];
-      EXPECT_LE(p0, 128);
-      EXPECT_LE(p1, 128);
-      EXPECT_LE(p2, 128);
-      EXPECT_LE(p3, 128);
-      EXPECT_LE(p0 + p3, 128);
-      EXPECT_LE(p0 + p3 + p1, 128);
-      EXPECT_LE(p0 + p3 + p1 + p2, 128);
-      EXPECT_EQ(p0 + p1 + p2 + p3, 128);
+  int subpel_search;
+  for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+       ++subpel_search) {
+    for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+      const InterpFilter filter = (InterpFilter)filter_bank;
+      const InterpKernel *filters =
+          (const InterpKernel *)av1_get_interp_filter_kernel(filter,
+                                                             subpel_search);
+      for (int i = 0; i < kNumFilters; i++) {
+        const int p0 = filters[i][0] + filters[i][1];
+        const int p1 = filters[i][2] + filters[i][3];
+        const int p2 = filters[i][4] + filters[i][5];
+        const int p3 = filters[i][6] + filters[i][7];
+        EXPECT_LE(p0, 128);
+        EXPECT_LE(p1, 128);
+        EXPECT_LE(p2, 128);
+        EXPECT_LE(p3, 128);
+        EXPECT_LE(p0 + p3, 128);
+        EXPECT_LE(p0 + p3 + p1, 128);
+        EXPECT_LE(p0 + p3 + p1 + p2, 128);
+        EXPECT_EQ(p0 + p1 + p2 + p3, 128);
+      }
     }
   }
 }
@@ -515,53 +529,51 @@ const int16_t kInvalidFilter[8] = { 0 };
 TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
   uint8_t *const in = input();
   uint8_t *const out = output();
-  uint8_t ref8[kOutputStride * kMaxDimension];
-  uint16_t ref16[kOutputStride * kMaxDimension];
   uint8_t *ref;
   if (UUT_->use_highbd_ == 0) {
-    ref = ref8;
+    ref = ref8_;
   } else {
-    ref = CONVERT_TO_BYTEPTR(ref16);
+    ref = CONVERT_TO_BYTEPTR(ref16_);
   }
+  int subpel_search;
+  for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+       ++subpel_search) {
+    for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+      const InterpFilter filter = (InterpFilter)filter_bank;
+      const InterpKernel *filters =
+          (const InterpKernel *)av1_get_interp_filter_kernel(filter,
+                                                             subpel_search);
+      for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+        for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+          wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
+                                     filters[filter_y], ref, kOutputStride,
+                                     Width(), Height());
 
-  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
-    const InterpFilter filter = (InterpFilter)filter_bank;
-    const InterpKernel *filters =
-        (const InterpKernel *)av1_get_interp_filter_kernel(filter);
-    const InterpFilterParams *filter_params =
-        av1_get_interp_filter_params_with_block_size(filter, 8);
-    if (filter_params->taps != SUBPEL_TAPS) continue;
+          if (filter_x && filter_y)
+            continue;
+          else if (filter_y)
+            ASM_REGISTER_STATE_CHECK(
+                UUT_->v8_(in, kInputStride, out, kOutputStride, kInvalidFilter,
+                          16, filters[filter_y], 16, Width(), Height()));
+          else if (filter_x)
+            ASM_REGISTER_STATE_CHECK(UUT_->h8_(
+                in, kInputStride, out, kOutputStride, filters[filter_x], 16,
+                kInvalidFilter, 16, Width(), Height()));
+          else
+            ASM_REGISTER_STATE_CHECK(UUT_->copy_(
+                in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
+                kInvalidFilter, 0, Width(), Height()));
 
-    for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
-      for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
-        wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
-                                   filters[filter_y], ref, kOutputStride,
-                                   Width(), Height());
+          CheckGuardBlocks();
 
-        if (filter_x && filter_y)
-          continue;
-        else if (filter_y)
-          ASM_REGISTER_STATE_CHECK(
-              UUT_->v8_(in, kInputStride, out, kOutputStride, kInvalidFilter,
-                        16, filters[filter_y], 16, Width(), Height()));
-        else if (filter_x)
-          ASM_REGISTER_STATE_CHECK(
-              UUT_->h8_(in, kInputStride, out, kOutputStride, filters[filter_x],
-                        16, kInvalidFilter, 16, Width(), Height()));
-        else
-          ASM_REGISTER_STATE_CHECK(
-              UUT_->copy_(in, kInputStride, out, kOutputStride, kInvalidFilter,
-                          0, kInvalidFilter, 0, Width(), Height()));
-
-        CheckGuardBlocks();
-
-        for (int y = 0; y < Height(); ++y)
-          for (int x = 0; x < Width(); ++x)
-            ASSERT_EQ(lookup(ref, y * kOutputStride + x),
-                      lookup(out, y * kOutputStride + x))
-                << "mismatch at (" << x << "," << y << "), "
-                << "filters (" << filter_bank << "," << filter_x << ","
-                << filter_y << ")";
+          for (int y = 0; y < Height(); ++y)
+            for (int x = 0; x < Width(); ++x)
+              ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                        lookup(out, y * kOutputStride + x))
+                  << "mismatch at (" << x << "," << y << "), "
+                  << "filters (" << filter_bank << "," << filter_x << ","
+                  << filter_y << ")";
+        }
       }
     }
   }
@@ -570,13 +582,11 @@ TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
 TEST_P(ConvolveTest, FilterExtremes) {
   uint8_t *const in = input();
   uint8_t *const out = output();
-  uint8_t ref8[kOutputStride * kMaxDimension];
-  uint16_t ref16[kOutputStride * kMaxDimension];
   uint8_t *ref;
   if (UUT_->use_highbd_ == 0) {
-    ref = ref8;
+    ref = ref8_;
   } else {
-    ref = CONVERT_TO_BYTEPTR(ref16);
+    ref = CONVERT_TO_BYTEPTR(ref16_);
   }
 
   // Populate ref and out with some random data
@@ -609,41 +619,43 @@ TEST_P(ConvolveTest, FilterExtremes) {
           seed_val++;
       }
       if (axis) seed_val += 8;
+      int subpel_search;
+      for (subpel_search = USE_4_TAPS; subpel_search <= USE_8_TAPS;
+           ++subpel_search) {
+        for (int filter_bank = 0; filter_bank < kNumFilterBanks;
+             ++filter_bank) {
+          const InterpFilter filter = (InterpFilter)filter_bank;
+          const InterpKernel *filters =
+              (const InterpKernel *)av1_get_interp_filter_kernel(filter,
+                                                                 subpel_search);
+          for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+            for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+              wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
+                                         filters[filter_y], ref, kOutputStride,
+                                         Width(), Height());
+              if (filter_x && filter_y)
+                continue;
+              else if (filter_y)
+                ASM_REGISTER_STATE_CHECK(UUT_->v8_(
+                    in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
+                    filters[filter_y], 16, Width(), Height()));
+              else if (filter_x)
+                ASM_REGISTER_STATE_CHECK(UUT_->h8_(
+                    in, kInputStride, out, kOutputStride, filters[filter_x], 16,
+                    kInvalidFilter, 16, Width(), Height()));
+              else
+                ASM_REGISTER_STATE_CHECK(UUT_->copy_(
+                    in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
+                    kInvalidFilter, 0, Width(), Height()));
 
-      for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
-        const InterpFilter filter = (InterpFilter)filter_bank;
-        const InterpKernel *filters =
-            (const InterpKernel *)av1_get_interp_filter_kernel(filter);
-        const InterpFilterParams *filter_params =
-            av1_get_interp_filter_params_with_block_size(filter, 8);
-        if (filter_params->taps != SUBPEL_TAPS) continue;
-        for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
-          for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
-            wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
-                                       filters[filter_y], ref, kOutputStride,
-                                       Width(), Height());
-            if (filter_x && filter_y)
-              continue;
-            else if (filter_y)
-              ASM_REGISTER_STATE_CHECK(UUT_->v8_(
-                  in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
-                  filters[filter_y], 16, Width(), Height()));
-            else if (filter_x)
-              ASM_REGISTER_STATE_CHECK(UUT_->h8_(
-                  in, kInputStride, out, kOutputStride, filters[filter_x], 16,
-                  kInvalidFilter, 16, Width(), Height()));
-            else
-              ASM_REGISTER_STATE_CHECK(UUT_->copy_(
-                  in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
-                  kInvalidFilter, 0, Width(), Height()));
-
-            for (int y = 0; y < Height(); ++y)
-              for (int x = 0; x < Width(); ++x)
-                ASSERT_EQ(lookup(ref, y * kOutputStride + x),
-                          lookup(out, y * kOutputStride + x))
-                    << "mismatch at (" << x << "," << y << "), "
-                    << "filters (" << filter_bank << "," << filter_x << ","
-                    << filter_y << ")";
+              for (int y = 0; y < Height(); ++y)
+                for (int x = 0; x < Width(); ++x)
+                  ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                            lookup(out, y * kOutputStride + x))
+                      << "mismatch at (" << x << "," << y << "), "
+                      << "filters (" << filter_bank << "," << filter_x << ","
+                      << filter_y << ")";
+            }
           }
         }
       }
@@ -674,13 +686,11 @@ TEST_P(ConvolveTest, DISABLED_Copy_Speed) {
 TEST_P(ConvolveTest, DISABLED_Speed) {
   uint8_t *const in = input();
   uint8_t *const out = output();
-  uint8_t ref8[kOutputStride * kMaxDimension];
-  uint16_t ref16[kOutputStride * kMaxDimension];
   uint8_t *ref;
   if (UUT_->use_highbd_ == 0) {
-    ref = ref8;
+    ref = ref8_;
   } else {
-    ref = CONVERT_TO_BYTEPTR(ref16);
+    ref = CONVERT_TO_BYTEPTR(ref16_);
   }
 
   // Populate ref and out with some random data
@@ -700,7 +710,7 @@ TEST_P(ConvolveTest, DISABLED_Speed) {
 
   const InterpFilter filter = (InterpFilter)1;
   const InterpKernel *filters =
-      (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+      (const InterpKernel *)av1_get_interp_filter_kernel(filter, USE_8_TAPS);
   wrapper_filter_average_block2d_8_c(in, kInputStride, filters[1], filters[1],
                                      out, kOutputStride, Width(), Height());
 
@@ -712,11 +722,8 @@ TEST_P(ConvolveTest, DISABLED_Speed) {
     for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
       const InterpFilter filter = (InterpFilter)filter_bank;
       const InterpKernel *filters =
-          (const InterpKernel *)av1_get_interp_filter_kernel(filter);
-      const InterpFilterParams *filter_params =
-          av1_get_interp_filter_params_with_block_size(filter, 8);
-      if (filter_params->taps != SUBPEL_TAPS) continue;
-
+          (const InterpKernel *)av1_get_interp_filter_kernel(filter,
+                                                             USE_8_TAPS);
       for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
         for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
           if (filter_x && filter_y) continue;
@@ -805,18 +812,22 @@ const ConvolveParam kArrayConvolve_c[] = {
 INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::ValuesIn(kArrayConvolve_c));
 
 #if HAVE_SSE2 && ARCH_X86_64
-const ConvolveFunctions convolve8_sse2(wrap_convolve_copy_sse2_8,
-                                       wrap_convolve8_horiz_sse2_8,
-                                       wrap_convolve8_vert_sse2_8, 8);
-const ConvolveFunctions convolve10_sse2(wrap_convolve_copy_sse2_10,
-                                        wrap_convolve8_horiz_sse2_10,
-                                        wrap_convolve8_vert_sse2_10, 10);
-const ConvolveFunctions convolve12_sse2(wrap_convolve_copy_sse2_12,
-                                        wrap_convolve8_horiz_sse2_12,
-                                        wrap_convolve8_vert_sse2_12, 12);
+const ConvolveFunctions convolve8_sse2(aom_convolve_copy_c,
+                                       aom_convolve8_horiz_sse2,
+                                       aom_convolve8_vert_sse2, 0);
+const ConvolveFunctions wrap_convolve8_sse2(wrap_convolve_copy_sse2_8,
+                                            wrap_convolve8_horiz_sse2_8,
+                                            wrap_convolve8_vert_sse2_8, 8);
+const ConvolveFunctions wrap_convolve10_sse2(wrap_convolve_copy_sse2_10,
+                                             wrap_convolve8_horiz_sse2_10,
+                                             wrap_convolve8_vert_sse2_10, 10);
+const ConvolveFunctions wrap_convolve12_sse2(wrap_convolve_copy_sse2_12,
+                                             wrap_convolve8_horiz_sse2_12,
+                                             wrap_convolve8_vert_sse2_12, 12);
 const ConvolveParam kArrayConvolve_sse2[] = { ALL_SIZES(convolve8_sse2),
-                                              ALL_SIZES(convolve10_sse2),
-                                              ALL_SIZES(convolve12_sse2) };
+                                              ALL_SIZES(wrap_convolve8_sse2),
+                                              ALL_SIZES(wrap_convolve10_sse2),
+                                              ALL_SIZES(wrap_convolve12_sse2) };
 INSTANTIATE_TEST_CASE_P(SSE2, ConvolveTest,
                         ::testing::ValuesIn(kArrayConvolve_sse2));
 #endif
diff --git a/third_party/aom/test/corner_match_test.cc b/third_party/aom/test/corner_match_test.cc
index 58e3139c5f..af2baa7b14 100644
--- a/third_party/aom/test/corner_match_test.cc
+++ b/third_party/aom/test/corner_match_test.cc
@@ -24,9 +24,13 @@ namespace AV1CornerMatch {
 
 using libaom_test::ACMRandom;
 
+typedef double (*ComputeCrossCorrFunc)(unsigned char *im1, int stride1, int x1,
+                                       int y1, unsigned char *im2, int stride2,
+                                       int x2, int y2);
+
 using ::testing::make_tuple;
 using ::testing::tuple;
-typedef tuple<int> CornerMatchParam;
+typedef tuple<int, ComputeCrossCorrFunc> CornerMatchParam;
 
 class AV1CornerMatchTest : public ::testing::TestWithParam<CornerMatchParam> {
  public:
@@ -36,19 +40,24 @@ class AV1CornerMatchTest : public ::testing::TestWithParam<CornerMatchParam> {
   virtual void TearDown();
 
  protected:
-  void RunCheckOutput();
+  void RunCheckOutput(int run_times);
+  ComputeCrossCorrFunc target_func;
 
   libaom_test::ACMRandom rnd_;
 };
 
 AV1CornerMatchTest::~AV1CornerMatchTest() {}
-void AV1CornerMatchTest::SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+void AV1CornerMatchTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+  target_func = GET_PARAM(1);
+}
 void AV1CornerMatchTest::TearDown() { libaom_test::ClearSystemState(); }
 
-void AV1CornerMatchTest::RunCheckOutput() {
+void AV1CornerMatchTest::RunCheckOutput(int run_times) {
   const int w = 128, h = 128;
   const int num_iters = 10000;
   int i, j;
+  aom_usec_timer ref_timer, test_timer;
 
   uint8_t *input1 = new uint8_t[w * h];
   uint8_t *input2 = new uint8_t[w * h];
@@ -80,21 +89,54 @@ void AV1CornerMatchTest::RunCheckOutput() {
 
     double res_c =
         compute_cross_correlation_c(input1, w, x1, y1, input2, w, x2, y2);
-    double res_sse4 =
-        compute_cross_correlation_sse4_1(input1, w, x1, y1, input2, w, x2, y2);
+    double res_simd = target_func(input1, w, x1, y1, input2, w, x2, y2);
 
-    ASSERT_EQ(res_sse4, res_c);
+    if (run_times > 1) {
+      aom_usec_timer_start(&ref_timer);
+      for (j = 0; j < run_times; j++) {
+        compute_cross_correlation_c(input1, w, x1, y1, input2, w, x2, y2);
+      }
+      aom_usec_timer_mark(&ref_timer);
+      const int elapsed_time_c =
+          static_cast<int>(aom_usec_timer_elapsed(&ref_timer));
+
+      aom_usec_timer_start(&test_timer);
+      for (j = 0; j < run_times; j++) {
+        target_func(input1, w, x1, y1, input2, w, x2, y2);
+      }
+      aom_usec_timer_mark(&test_timer);
+      const int elapsed_time_simd =
+          static_cast<int>(aom_usec_timer_elapsed(&test_timer));
+
+      printf(
+          "c_time=%d \t simd_time=%d \t "
+          "gain=%d\n",
+          elapsed_time_c, elapsed_time_simd,
+          (elapsed_time_c / elapsed_time_simd));
+    } else {
+      ASSERT_EQ(res_simd, res_c);
+    }
   }
-
   delete[] input1;
   delete[] input2;
 }
 
-TEST_P(AV1CornerMatchTest, CheckOutput) { RunCheckOutput(); }
+TEST_P(AV1CornerMatchTest, CheckOutput) { RunCheckOutput(1); }
+TEST_P(AV1CornerMatchTest, DISABLED_Speed) { RunCheckOutput(100000); }
 
-INSTANTIATE_TEST_CASE_P(SSE4_1, AV1CornerMatchTest,
-                        ::testing::Values(make_tuple(0), make_tuple(1)));
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AV1CornerMatchTest,
+    ::testing::Values(make_tuple(0, compute_cross_correlation_sse4_1),
+                      make_tuple(1, compute_cross_correlation_sse4_1)));
+#endif
 
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1CornerMatchTest,
+    ::testing::Values(make_tuple(0, compute_cross_correlation_avx2),
+                      make_tuple(1, compute_cross_correlation_avx2)));
+#endif
 }  // namespace AV1CornerMatch
 
 }  // namespace test_libaom
diff --git a/third_party/aom/test/cpu_speed_test.cc b/third_party/aom/test/cpu_speed_test.cc
index 8ea3e69650..2a164974b0 100644
--- a/third_party/aom/test/cpu_speed_test.cc
+++ b/third_party/aom/test/cpu_speed_test.cc
@@ -46,7 +46,7 @@ class CpuSpeedTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
       encoder->Control(AV1E_SET_TUNE_CONTENT, tune_content_);
       if (encoding_mode_ != ::libaom_test::kRealTime) {
diff --git a/third_party/aom/test/decode_multithreaded_test.cc b/third_party/aom/test/decode_multithreaded_test.cc
index cea1d144f1..92253ede81 100644
--- a/third_party/aom/test/decode_multithreaded_test.cc
+++ b/third_party/aom/test/decode_multithreaded_test.cc
@@ -76,7 +76,7 @@ class AV1DecodeMultiThreadedTest
 
   virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
                                   libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
       encoder->Control(AV1E_SET_TILE_ROWS, n_tile_rows_);
       encoder->Control(AV1E_SET_NUM_TG, n_tile_groups_);
diff --git a/third_party/aom/test/decode_perf_test.cc b/third_party/aom/test/decode_perf_test.cc
index bb7b00032e..2f67342570 100644
--- a/third_party/aom/test/decode_perf_test.cc
+++ b/third_party/aom/test/decode_perf_test.cc
@@ -122,7 +122,7 @@ class AV1NewEncodeDecodePerfTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, speed_);
       encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
       encoder->Control(AV1E_SET_TILE_COLUMNS, 2);
diff --git a/third_party/aom/test/dr_prediction_test.cc b/third_party/aom/test/dr_prediction_test.cc
index ff2c1de4e9..603694f3ae 100644
--- a/third_party/aom/test/dr_prediction_test.cc
+++ b/third_party/aom/test/dr_prediction_test.cc
@@ -1,4 +1,4 @@
-/*
+/*
  * Copyright (c) 2018, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
@@ -135,7 +135,7 @@ struct DrPredFunc {
 template <typename Pixel, typename FuncType>
 class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
  protected:
-  static const int kMaxNumTests = 100000;
+  static const int kMaxNumTests = 10000;
   static const int kIterations = 10;
   static const int kDstStride = 64;
   static const int kDstSize = kDstStride * kDstStride;
@@ -162,6 +162,7 @@ class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
 
     for (int i = 0; i < kDstSize; ++i) {
       dst_ref_[i] = 0;
+      dst_tst_[i] = 0;
     }
   }
 
@@ -170,7 +171,7 @@ class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
   void Predict(bool speedtest, int tx) {
     const int kNumTests = speedtest ? kMaxNumTests : 1;
     aom_usec_timer timer;
-
+    int tst_time = 0;
     aom_usec_timer_start(&timer);
     for (int k = 0; k < kNumTests; ++k) {
       params_.ref_fn(dst_ref_, dst_stride_, bw_, bh_, above_, left_,
@@ -179,32 +180,39 @@ class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
     aom_usec_timer_mark(&timer);
     const int ref_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
 
-    aom_usec_timer_start(&timer);
     if (params_.tst_fn) {
+      aom_usec_timer_start(&timer);
       for (int k = 0; k < kNumTests; ++k) {
         ASM_REGISTER_STATE_CHECK(params_.tst_fn(dst_tst_, dst_stride_, bw_, bh_,
                                                 above_, left_, upsample_above_,
                                                 upsample_left_, dx_, dy_, bd_));
       }
+      aom_usec_timer_mark(&timer);
+      tst_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    } else {
+      for (int i = 0; i < kDstSize; ++i) {
+        dst_ref_[i] = dst_tst_[i];
+      }
     }
-    aom_usec_timer_mark(&timer);
-    const int tst_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
 
     OutputTimes(kNumTests, ref_time, tst_time, tx);
   }
 
-  void RunTest(bool speedtest, int p_angle) {
-    for (int i = 0; i < kBufSize; ++i) {
-      above_data_[i] = left_data_[i] = (1 << bd_) - 1;
+  void RunTest(bool speedtest, bool needsaturation, int p_angle) {
+    if (needsaturation) {
+      for (int i = 0; i < kBufSize; ++i) {
+        above_data_[i] = left_data_[i] = (1 << bd_) - 1;
+      }
     }
-
     for (int tx = 0; tx < TX_SIZES_ALL; ++tx) {
       if (params_.tst_fn == NULL) {
         for (int i = 0; i < kDstSize; ++i) {
           dst_tst_[i] = (1 << bd_) - 1;
+          dst_ref_[i] = (1 << bd_) - 1;
         }
       } else {
         for (int i = 0; i < kDstSize; ++i) {
+          dst_ref_[i] = 0;
           dst_tst_[i] = 0;
         }
       }
@@ -282,26 +290,11 @@ class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
 class LowbdDrPredTest : public DrPredTest<uint8_t, DrPred> {};
 
 TEST_P(LowbdDrPredTest, SaturatedValues) {
-  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
-    enable_upsample_ = iter & 1;
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
     for (int angle = start_angle_; angle < stop_angle_; ++angle) {
       dx_ = av1_get_dx(angle);
       dy_ = av1_get_dy(angle);
-      if (dx_ && dy_) RunTest(false, angle);
-    }
-  }
-}
-
-TEST_P(LowbdDrPredTest, DISABLED_Speed) {
-  const int angles[] = { 3, 45, 87 };
-  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
-    for (int i = 0; i < 3; ++i) {
-      const int angle = angles[i] + start_angle_;
-      dx_ = av1_get_dx(angle);
-      dy_ = av1_get_dy(angle);
-      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
-             enable_upsample_, angle);
-      if (dx_ && dy_) RunTest(true, angle);
+      if (dx_ && dy_) RunTest(false, true, angle);
     }
   }
 }
@@ -320,26 +313,11 @@ INSTANTIATE_TEST_CASE_P(
 class HighbdDrPredTest : public DrPredTest<uint16_t, DrPred_Hbd> {};
 
 TEST_P(HighbdDrPredTest, SaturatedValues) {
-  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
-    enable_upsample_ = iter & 1;
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
     for (int angle = start_angle_; angle < stop_angle_; ++angle) {
       dx_ = av1_get_dx(angle);
       dy_ = av1_get_dy(angle);
-      if (dx_ && dy_) RunTest(false, angle);
-    }
-  }
-}
-
-TEST_P(HighbdDrPredTest, DISABLED_Speed) {
-  const int angles[] = { 3, 45, 87 };
-  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
-    for (int i = 0; i < 3; ++i) {
-      const int angle = angles[i] + start_angle_;
-      dx_ = av1_get_dx(angle);
-      dy_ = av1_get_dy(angle);
-      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
-             enable_upsample_, angle);
-      if (dx_ && dy_) RunTest(true, angle);
+      if (dx_ && dy_) RunTest(false, true, angle);
     }
   }
 }
@@ -366,4 +344,112 @@ INSTANTIATE_TEST_CASE_P(
         DrPredFunc<DrPred_Hbd>(&z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
                                NULL, AOM_BITS_12, kZ3Start)));
 
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, LowbdDrPredTest,
+    ::testing::Values(DrPredFunc<DrPred>(&z1_wrapper<av1_dr_prediction_z1_c>,
+                                         &z1_wrapper<av1_dr_prediction_z1_avx2>,
+                                         AOM_BITS_8, kZ1Start),
+                      DrPredFunc<DrPred>(&z2_wrapper<av1_dr_prediction_z2_c>,
+                                         &z2_wrapper<av1_dr_prediction_z2_avx2>,
+                                         AOM_BITS_8, kZ2Start),
+                      DrPredFunc<DrPred>(&z3_wrapper<av1_dr_prediction_z3_c>,
+                                         &z3_wrapper<av1_dr_prediction_z3_avx2>,
+                                         AOM_BITS_8, kZ3Start)));
+
+TEST_P(LowbdDrPredTest, DISABLED_Speed) {
+  const int angles[] = { 3, 45, 87 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int i = 0; i < 3; ++i) {
+      const int angle = angles[i] + start_angle_;
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
+             enable_upsample_, angle);
+      if (dx_ && dy_) RunTest(true, false, angle);
+    }
+  }
+}
+
+TEST_P(LowbdDrPredTest, OperationCheck) {
+  if (params_.tst_fn == NULL) return;
+  // const int angles[] = { 3, 45, 81, 87, 93, 100, 145, 187, 199, 260 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int angle = start_angle_; angle < stop_angle_; ++angle) {
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      if (dx_ && dy_) RunTest(false, false, angle);
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, HighbdDrPredTest,
+    ::testing::Values(DrPredFunc<DrPred_Hbd>(
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_avx2>,
+                          AOM_BITS_8, kZ1Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_avx2>,
+                          AOM_BITS_10, kZ1Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                          &z1_wrapper_hbd<av1_highbd_dr_prediction_z1_avx2>,
+                          AOM_BITS_12, kZ1Start),
+                      /* TODO(https://crbug.com/aomedia/2288)
+                      DrPredFunc<DrPred_Hbd>(
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_avx2>,
+                          AOM_BITS_8, kZ2Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_avx2>,
+                          AOM_BITS_10, kZ2Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                          &z2_wrapper_hbd<av1_highbd_dr_prediction_z2_avx2>,
+                          AOM_BITS_12, kZ2Start),
+                          */
+                      DrPredFunc<DrPred_Hbd>(
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_avx2>,
+                          AOM_BITS_8, kZ3Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_avx2>,
+                          AOM_BITS_10, kZ3Start),
+                      DrPredFunc<DrPred_Hbd>(
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                          &z3_wrapper_hbd<av1_highbd_dr_prediction_z3_avx2>,
+                          AOM_BITS_12, kZ3Start)));
+
+TEST_P(HighbdDrPredTest, DISABLED_Speed) {
+  const int angles[] = { 3, 45, 87 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int i = 0; i < 3; ++i) {
+      int angle = angles[i] + start_angle_;
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
+             enable_upsample_, angle);
+      if (dx_ && dy_) RunTest(true, false, angle);
+    }
+  }
+}
+
+TEST_P(HighbdDrPredTest, OperationCheck) {
+  if (params_.tst_fn == NULL) return;
+  // const int angles[] = { 3, 45, 81, 87, 93, 100, 145, 187, 199, 260 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int angle = start_angle_; angle < stop_angle_; angle++) {
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      if (dx_ && dy_) RunTest(false, false, angle);
+    }
+  }
+}
+
+#endif  // HAVE_AVX2
+
 }  // namespace
diff --git a/third_party/aom/test/ec_test.cc b/third_party/aom/test/ec_test.cc
index e6a5ea63bf..853abcbc5a 100644
--- a/third_party/aom/test/ec_test.cc
+++ b/third_party/aom/test/ec_test.cc
@@ -22,7 +22,6 @@ TEST(EC_TEST, random_ec_test) {
   int sz;
   int i;
   int ret;
-  unsigned int sym;
   unsigned int seed;
   unsigned char *ptr;
   uint32_t ptr_sz;
@@ -90,6 +89,8 @@ TEST(EC_TEST, random_ec_test) {
         << " (Random seed: " << seed << ").\n";
     for (j = 0; j < sz; j++) {
       int dec_method;
+      unsigned int sym = data[j] + 1;  // Initialize sym to an invalid value.
+
       if (CDF_SHIFT == 0) {
         dec_method = 3 + (rand() & 1);
       } else {
diff --git a/third_party/aom/test/edge_detect_test.cc b/third_party/aom/test/edge_detect_test.cc
new file mode 100644
index 0000000000..47466cb872
--- /dev/null
+++ b/third_party/aom/test/edge_detect_test.cc
@@ -0,0 +1,340 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdbool.h>
+#include "aom_mem/aom_mem.h"
+#include "av1/encoder/rdopt.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+using ::testing::get;
+using ::testing::tuple;
+
+static int get_pix(uint8_t *buf, int i, bool high_bd) {
+  if (high_bd) {
+    return *CONVERT_TO_SHORTPTR(buf + i);
+  } else {
+    return buf[i];
+  }
+}
+
+/** Get the (i, j) value from the input; if i or j is outside of the width
+ * or height, the nearest pixel value is returned.
+ */
+static int get_nearest_pix(const int *buf, int w, int h, int i, int j) {
+  int offset = AOMMAX(AOMMIN(i, w - 1), 0) + w * AOMMAX(AOMMIN(j, h - 1), 0);
+  return buf[offset];
+}
+
+/** Given the image data, creates a new image with padded values, so an
+ * 8-tap filter can be convolved. The padded value is the same as the closest
+ * value in the image. Returns a pointer to the start of the image in the
+ * padded data. Must be freed with free_pad_8tap. The output will be either
+ * 8-bit or 16-bit, depending on the high bit-depth (high_bd) field.
+ */
+static uint8_t *pad_8tap_convolve(const int *data, int w, int h, bool high_bd) {
+  // SIMD optimizations require the width to be a multiple of 8 and the height
+  // to be multiples of 4.
+  assert(w % 8 == 0);
+  assert(h % 4 == 0);
+  // For an 8-tap filter, we need to pad with 3 lines on top and on the left,
+  // and 4 lines on the right and bottom, for 7 extra lines.
+  const int pad_w = w + 7;
+  const int pad_h = h + 7;
+
+  uint8_t *dst;
+  if (high_bd) {
+    dst =
+        CONVERT_TO_BYTEPTR(aom_memalign(32, sizeof(uint16_t) * pad_w * pad_h));
+  } else {
+    dst = (uint8_t *)aom_memalign(32, sizeof(uint8_t) * pad_w * pad_h);
+  }
+  for (int j = 0; j < pad_h; ++j) {
+    for (int i = 0; i < pad_w; ++i) {
+      const int v = get_nearest_pix(data, w, h, i - 3, j - 3);
+      if (high_bd) {
+        *CONVERT_TO_SHORTPTR(dst + i + j * pad_w) = v;
+      } else {
+        dst[i + j * pad_w] = v;
+      }
+    }
+  }
+  return dst + (w + 7) * 3 + 3;
+}
+
+static int stride_8tap(int width) { return width + 7; }
+
+static void free_pad_8tap(uint8_t *padded, int width, bool high_bd) {
+  if (high_bd) {
+    aom_free(CONVERT_TO_SHORTPTR(padded - (width + 7) * 3 - 3));
+  } else {
+    aom_free(padded - (width + 7) * 3 - 3);
+  }
+}
+
+static uint8_t *malloc_bd(int num_entries, bool high_bd) {
+  const int bytes_per_entry = high_bd ? sizeof(uint16_t) : sizeof(uint8_t);
+
+  uint8_t *buf = (uint8_t *)aom_memalign(32, bytes_per_entry * num_entries);
+  if (high_bd) {
+    return CONVERT_TO_BYTEPTR(buf);
+  } else {
+    return buf;
+  }
+}
+
+static void free_bd(uint8_t *p, bool high_bd) {
+  if (high_bd) {
+    aom_free(CONVERT_TO_SHORTPTR(p));
+  } else {
+    aom_free(p);
+  }
+}
+
+class EdgeDetectBrightnessTest :
+    // Parameters are (brightness, width, height, high bit depth representation,
+    // bit depth).
+    public ::testing::TestWithParam<tuple<int, int, int, bool, int> > {
+ protected:
+  void SetUp() override {
+    // Allocate a (width by height) array of luma values in orig_.
+    // padded_ will be filled by the pad() call, which adds a border around
+    // the orig_. The output_ array has enough space for the computation.
+    const int brightness = GET_PARAM(0);
+    const int width = GET_PARAM(1);
+    const int height = GET_PARAM(2);
+    const bool high_bd = GET_PARAM(3);
+
+    // Create the padded image of uniform brightness.
+    int *orig = (int *)malloc(width * height * sizeof(int));
+    for (int i = 0; i < width * height; ++i) {
+      orig[i] = brightness;
+    }
+    input_ = pad_8tap_convolve(orig, width, height, high_bd);
+    free(orig);
+    output_ = malloc_bd(width * height, high_bd);
+  }
+
+  void TearDown() override {
+    const int width = GET_PARAM(1);
+    const bool high_bd = GET_PARAM(3);
+    free_pad_8tap(input_, width, high_bd);
+    free_bd(output_, high_bd);
+  }
+
+  // Skip the tests where brightness exceeds the bit-depth; we run into this
+  // issue because of gtest's limitation on valid combinations of test
+  // parameters. Also skip the tests where bit depth is greater than 8, but
+  // high bit depth representation is not set.
+  bool should_skip() const {
+    const int brightness = GET_PARAM(0);
+    const int bd = GET_PARAM(4);
+    if (brightness >= (1 << bd)) {
+      return true;
+    }
+    const bool high_bd = GET_PARAM(3);
+    if (bd > 8 && !high_bd) {
+      return true;
+    }
+    return false;
+  }
+
+  uint8_t *input_;
+  uint8_t *output_;
+};
+
+TEST_P(EdgeDetectBrightnessTest, BlurUniformBrightness) {
+  // Some combination of parameters are non-sensical, due to limitations
+  // of the testing framework. Ignore these.
+  if (should_skip()) {
+    return;
+  }
+
+  // For varying levels of brightness, the algorithm should
+  // produce the same output.
+  const int brightness = GET_PARAM(0);
+  const int width = GET_PARAM(1);
+  const int height = GET_PARAM(2);
+  const bool high_bd = GET_PARAM(3);
+  const int bd = GET_PARAM(4);
+
+  gaussian_blur(input_, stride_8tap(width), width, height, output_, high_bd,
+                bd);
+  for (int i = 0; i < width * height; ++i) {
+    ASSERT_EQ(brightness, get_pix(output_, i, high_bd));
+  }
+}
+
+// No edges on a uniformly bright image.
+TEST_P(EdgeDetectBrightnessTest, DetectUniformBrightness) {
+  if (should_skip()) {
+    return;
+  }
+  const int width = GET_PARAM(1);
+  const int height = GET_PARAM(2);
+  const bool high_bd = GET_PARAM(3);
+  const int bd = GET_PARAM(4);
+
+  ASSERT_EQ(0, av1_edge_exists(input_, stride_8tap(width), width, height,
+                               high_bd, bd));
+}
+
+INSTANTIATE_TEST_CASE_P(ImageBrightnessTests, EdgeDetectBrightnessTest,
+                        ::testing::Combine(
+                            // Brightness
+                            ::testing::Values(0, 1, 2, 127, 128, 129, 254, 255,
+                                              256, 511, 512, 1023, 1024, 2048,
+                                              4095),
+                            // Width
+                            ::testing::Values(8, 16, 32),
+                            // Height
+                            ::testing::Values(4, 8, 12, 32),
+                            // High bit depth representation
+                            ::testing::Bool(),
+                            // Bit depth
+                            ::testing::Values(8, 10, 12)));
+
+class EdgeDetectImageTest :
+    // Parameters are (width, height, high bit depth representation, bit depth).
+    public ::testing::TestWithParam<tuple<int, int, bool, int> > {
+ protected:
+  // Skip the tests where bit depth is greater than 8, but high bit depth
+  // representation is not set (limitation of testing framework).
+  bool should_skip() const {
+    const bool high_bd = GET_PARAM(2);
+    const int bd = GET_PARAM(3);
+    return bd > 8 && !high_bd;
+  }
+};
+
+// Generate images with black on one side and white on the other.
+TEST_P(EdgeDetectImageTest, BlackWhite) {
+  // Some combination of parameters are non-sensical, due to limitations
+  // of the testing framework. Ignore these.
+  if (should_skip()) {
+    return;
+  }
+
+  const int width = GET_PARAM(0);
+  const int height = GET_PARAM(1);
+  const bool high_bd = GET_PARAM(2);
+  const int bd = GET_PARAM(3);
+
+  const int white = (1 << bd) - 1;
+  int *orig = (int *)malloc(width * height * sizeof(int));
+  for (int j = 0; j < height; ++j) {
+    for (int i = 0; i < width; ++i) {
+      if (i < width / 2) {
+        orig[i + j * width] = 0;
+      } else {
+        orig[i + j * width] = white;
+      }
+    }
+  }
+  uint8_t *padded = pad_8tap_convolve(orig, width, height, high_bd);
+  free(orig);
+  // Value should be between 556 and 560.
+  ASSERT_LE(556, av1_edge_exists(padded, stride_8tap(width), width, height,
+                                 high_bd, bd));
+  ASSERT_GE(560, av1_edge_exists(padded, stride_8tap(width), width, height,
+                                 high_bd, bd));
+
+  free_pad_8tap(padded, width, high_bd);
+}
+
+// Hardcoded blur tests.
+static const int luma[32] = { 241, 147, 7,   90,  184, 103, 28,  186,
+                              2,   248, 49,  242, 114, 146, 127, 22,
+                              121, 228, 167, 108, 158, 174, 41,  168,
+                              214, 99,  184, 109, 114, 247, 117, 119 };
+static const uint8_t expected[] = { 161, 138, 119, 118, 123, 118, 113, 122,
+                                    143, 140, 134, 133, 134, 126, 116, 114,
+                                    147, 149, 145, 142, 143, 138, 126, 118,
+                                    164, 156, 148, 144, 148, 148, 138, 126 };
+
+static void hardcoded_blur_test_aux(const bool high_bd) {
+  const int w = 8;
+  const int h = 4;
+  for (int bd = 8; bd <= 12; bd += 2) {
+    // Skip the tests where bit depth is greater than 8, but high bit depth
+    // representation is not set.
+    if (bd > 8 && !high_bd) {
+      break;
+    }
+    uint8_t *output = malloc_bd(w * h, high_bd);
+    uint8_t *padded = pad_8tap_convolve(luma, w, h, high_bd);
+    gaussian_blur(padded, stride_8tap(w), w, h, output, high_bd, bd);
+    for (int i = 0; i < w * h; ++i) {
+      ASSERT_EQ(expected[i], get_pix(output, i, high_bd));
+    }
+    free_pad_8tap(padded, w, high_bd);
+    free_bd(output, high_bd);
+
+    // If we multiply the inputs by a constant factor, the output should not
+    // vary more than 0.5 * factor.
+    for (int c = 2; c < (1 << (bd - 8)); ++c) {
+      int scaled_luma[32];
+      for (int i = 0; i < 32; ++i) {
+        scaled_luma[i] = luma[i] * c;
+      }
+      uint8_t *output = malloc_bd(w * h, high_bd);
+      uint8_t *padded = pad_8tap_convolve(scaled_luma, w, h, high_bd);
+      gaussian_blur(padded, stride_8tap(w), w, h, output, high_bd, bd);
+      for (int i = 0; i < w * h; ++i) {
+        ASSERT_GE(c / 2, abs(expected[i] * c - get_pix(output, i, high_bd)));
+      }
+      free_pad_8tap(padded, w, high_bd);
+      free_bd(output, high_bd);
+    }
+  }
+}
+
+TEST(EdgeDetectImageTest, HardcodedBlurTest) {
+  hardcoded_blur_test_aux(false);
+  hardcoded_blur_test_aux(true);
+}
+
+TEST(EdgeDetectImageTest, SobelTest) {
+  // Randomly generated 3x3. Compute Sobel for middle value.
+  const uint8_t buf[9] = { 241, 147, 7, 90, 184, 103, 28, 186, 2 };
+  const int stride = 3;
+  bool high_bd = false;
+  sobel_xy result = sobel(buf, stride, 1, 1, high_bd);
+  ASSERT_EQ(234, result.x);
+  ASSERT_EQ(140, result.y);
+
+  // Verify it works for 8-bit values in a high bit-depth buffer.
+  const uint16_t buf8_16[9] = { 241, 147, 7, 90, 184, 103, 28, 186, 2 };
+  high_bd = true;
+  result = sobel(CONVERT_TO_BYTEPTR(buf8_16), stride, 1, 1, high_bd);
+  ASSERT_EQ(234, result.x);
+  ASSERT_EQ(140, result.y);
+
+  // Verify it works for high bit-depth values as well.
+  const uint16_t buf16[9] = { 241, 147, 7, 90, 184, 2003, 1028, 186, 2 };
+  result = sobel(CONVERT_TO_BYTEPTR(buf16), stride, 1, 1, high_bd);
+  ASSERT_EQ(-2566, result.x);
+  ASSERT_EQ(-860, result.y);
+}
+
+INSTANTIATE_TEST_CASE_P(EdgeDetectImages, EdgeDetectImageTest,
+                        ::testing::Combine(
+                            // Width
+                            ::testing::Values(8, 16, 32),
+                            // Height
+                            ::testing::Values(4, 8, 12, 32),
+                            // High bit depth representation
+                            ::testing::Bool(),
+                            // Bit depth
+                            ::testing::Values(8, 10, 12)));
+}  // namespace
diff --git a/third_party/aom/test/encode_test_driver.cc b/third_party/aom/test/encode_test_driver.cc
index f3d61dc36a..d06168fec7 100644
--- a/third_party/aom/test/encode_test_driver.cc
+++ b/third_party/aom/test/encode_test_driver.cc
@@ -9,6 +9,7 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
 #include <string>
 
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
@@ -192,7 +193,7 @@ void EncoderTest::RunLoop(VideoSource *video) {
       cfg_.g_pass = AOM_RC_LAST_PASS;
 
     BeginPassHook(pass);
-    testing::internal::scoped_ptr<Encoder> encoder(
+    std::unique_ptr<Encoder> encoder(
         codec_->CreateEncoder(cfg_, init_flags_, &stats_));
     ASSERT_TRUE(encoder.get() != NULL);
 
@@ -205,7 +206,7 @@ void EncoderTest::RunLoop(VideoSource *video) {
 
     ASSERT_FALSE(::testing::Test::HasFatalFailure());
 
-    testing::internal::scoped_ptr<Decoder> decoder(
+    std::unique_ptr<Decoder> decoder(
         codec_->CreateDecoder(dec_cfg, 0 /* flags */));
 #if CONFIG_AV1_DECODER
     if (decoder->IsAV1()) {
diff --git a/third_party/aom/test/end_to_end_test.cc b/third_party/aom/test/end_to_end_test.cc
index 1ac0ae9312..6ea09a6d9d 100644
--- a/third_party/aom/test/end_to_end_test.cc
+++ b/third_party/aom/test/end_to_end_test.cc
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
+
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
 #include "test/codec_factory.h"
@@ -51,6 +53,13 @@ typedef struct {
   unsigned int profile;
 } TestVideoParam;
 
+std::ostream &operator<<(std::ostream &os, const TestVideoParam &test_arg) {
+  return os << "TestVideoParam { filename:" << test_arg.filename
+            << " input_bit_depth:" << test_arg.input_bit_depth
+            << " fmt:" << test_arg.fmt << " bit_depth:" << test_arg.bit_depth
+            << " profile:" << test_arg.profile << "}";
+}
+
 const TestVideoParam kTestVectors[] = {
   { "park_joy_90p_8_420.y4m", 8, AOM_IMG_FMT_I420, AOM_BITS_8, 0 },
   { "park_joy_90p_8_422.y4m", 8, AOM_IMG_FMT_I422, AOM_BITS_8, 2 },
@@ -120,7 +129,7 @@ class EndToEndTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
       encoder->Control(AV1E_SET_TILE_COLUMNS, 4);
       encoder->Control(AOME_SET_CPUUSED, cpu_used_);
@@ -155,7 +164,7 @@ class EndToEndTest
     init_flags_ = AOM_CODEC_USE_PSNR;
     if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
 
-    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    std::unique_ptr<libaom_test::VideoSource> video;
     if (is_extension_y4m(test_video_param_.filename)) {
       video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
                                                   kFrames));
diff --git a/third_party/aom/test/error_resilience_test.cc b/third_party/aom/test/error_resilience_test.cc
index 13ac0bf93d..10bcd6734c 100644
--- a/third_party/aom/test/error_resilience_test.cc
+++ b/third_party/aom/test/error_resilience_test.cc
@@ -145,6 +145,23 @@ class ErrorResilienceTestLarge
     }
   }
 
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    // Check that the encode frame flags are correctly reflected
+    // in the output frame flags.
+    const int encode_flags = pkt->data.frame.flags >> 16;
+    if ((encode_flags & (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
+                         AOM_EFLAG_NO_UPD_ARF)) ==
+        (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF)) {
+      ASSERT_TRUE(!!(pkt->data.frame.flags & AOM_FRAME_IS_DROPPABLE));
+    }
+    if (encode_flags & AOM_EFLAG_SET_S_FRAME) {
+      ASSERT_TRUE(!!(pkt->data.frame.flags & AOM_FRAME_IS_SWITCH));
+    }
+    if (encode_flags & AOM_EFLAG_ERROR_RESILIENT) {
+      ASSERT_TRUE(!!(pkt->data.frame.flags & AOM_FRAME_IS_ERROR_RESILIENT));
+    }
+  }
+
   double GetAveragePsnr() const {
     if (nframes_) return psnr_ / nframes_;
     return 0.0;
@@ -342,7 +359,7 @@ TEST_P(ErrorResilienceTestLarge, DropFramesWithoutRecovery) {
 
   // Set an arbitrary set of error frames same as droppable frames.
   unsigned int num_droppable_frames = 3;
-  unsigned int droppable_frame_list[] = { 5, 10, 13 };
+  unsigned int droppable_frame_list[] = { 5, 11, 13 };
   SetDroppableFrames(num_droppable_frames, droppable_frame_list);
   SetErrorFrames(num_droppable_frames, droppable_frame_list);
   ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
diff --git a/third_party/aom/test/ethread_test.cc b/third_party/aom/test/ethread_test.cc
index d9ac782828..306cc2f3a8 100644
--- a/third_party/aom/test/ethread_test.cc
+++ b/third_party/aom/test/ethread_test.cc
@@ -20,14 +20,15 @@
 
 namespace {
 class AVxEncoderThreadTest
-    : public ::libaom_test::CodecTestWith4Params<libaom_test::TestMode, int,
-                                                 int, int>,
+    : public ::libaom_test::CodecTestWith5Params<libaom_test::TestMode, int,
+                                                 int, int, int>,
       public ::libaom_test::EncoderTest {
  protected:
   AVxEncoderThreadTest()
       : EncoderTest(GET_PARAM(0)), encoder_initialized_(false),
         encoding_mode_(GET_PARAM(1)), set_cpu_used_(GET_PARAM(2)),
-        tile_cols_(GET_PARAM(3)), tile_rows_(GET_PARAM(4)) {
+        tile_cols_(GET_PARAM(3)), tile_rows_(GET_PARAM(4)),
+        row_mt_(GET_PARAM(5)) {
     init_flags_ = AOM_CODEC_USE_PSNR;
     aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
     cfg.w = 1280;
@@ -119,87 +120,87 @@ class AVxEncoderThreadTest
         "niklas_640_480_30.yuv", AOM_IMG_FMT_I420, 640, 480, 30, 1, 15, 21);
     cfg_.rc_target_bitrate = 1000;
 
-    // Encode using single thread.
-    row_mt_ = 0;
-    cfg_.g_threads = 1;
-    init_flags_ = AOM_CODEC_USE_PSNR;
-    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    std::vector<size_t> single_thr_size_enc;
-    std::vector<std::string> single_thr_md5_enc;
-    std::vector<std::string> single_thr_md5_dec;
-    single_thr_size_enc = size_enc_;
-    single_thr_md5_enc = md5_enc_;
-    single_thr_md5_dec = md5_dec_;
-    size_enc_.clear();
-    md5_enc_.clear();
-    md5_dec_.clear();
+    if (row_mt_ == 0) {
+      // Encode using single thread.
+      cfg_.g_threads = 1;
+      init_flags_ = AOM_CODEC_USE_PSNR;
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      std::vector<size_t> single_thr_size_enc;
+      std::vector<std::string> single_thr_md5_enc;
+      std::vector<std::string> single_thr_md5_dec;
+      single_thr_size_enc = size_enc_;
+      single_thr_md5_enc = md5_enc_;
+      single_thr_md5_dec = md5_dec_;
+      size_enc_.clear();
+      md5_enc_.clear();
+      md5_dec_.clear();
 
-    // Encode using multiple threads.
-    cfg_.g_threads = 4;
-    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    std::vector<size_t> multi_thr_size_enc;
-    std::vector<std::string> multi_thr_md5_enc;
-    std::vector<std::string> multi_thr_md5_dec;
-    multi_thr_size_enc = size_enc_;
-    multi_thr_md5_enc = md5_enc_;
-    multi_thr_md5_dec = md5_dec_;
-    size_enc_.clear();
-    md5_enc_.clear();
-    md5_dec_.clear();
+      // Encode using multiple threads.
+      cfg_.g_threads = 4;
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      std::vector<size_t> multi_thr_size_enc;
+      std::vector<std::string> multi_thr_md5_enc;
+      std::vector<std::string> multi_thr_md5_dec;
+      multi_thr_size_enc = size_enc_;
+      multi_thr_md5_enc = md5_enc_;
+      multi_thr_md5_dec = md5_dec_;
+      size_enc_.clear();
+      md5_enc_.clear();
+      md5_dec_.clear();
 
-    // Check that the vectors are equal.
-    ASSERT_EQ(single_thr_size_enc, multi_thr_size_enc);
-    ASSERT_EQ(single_thr_md5_enc, multi_thr_md5_enc);
-    ASSERT_EQ(single_thr_md5_dec, multi_thr_md5_dec);
+      // Check that the vectors are equal.
+      ASSERT_EQ(single_thr_size_enc, multi_thr_size_enc);
+      ASSERT_EQ(single_thr_md5_enc, multi_thr_md5_enc);
+      ASSERT_EQ(single_thr_md5_dec, multi_thr_md5_dec);
+    } else if (row_mt_ == 1) {
+      // Encode using multiple threads row-mt enabled.
+      cfg_.g_threads = 2;
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      std::vector<size_t> multi_thr2_row_mt_size_enc;
+      std::vector<std::string> multi_thr2_row_mt_md5_enc;
+      std::vector<std::string> multi_thr2_row_mt_md5_dec;
+      multi_thr2_row_mt_size_enc = size_enc_;
+      multi_thr2_row_mt_md5_enc = md5_enc_;
+      multi_thr2_row_mt_md5_dec = md5_dec_;
+      size_enc_.clear();
+      md5_enc_.clear();
+      md5_dec_.clear();
 
-    // Encode using multiple threads row-mt enabled.
-    row_mt_ = 1;
-    cfg_.g_threads = 2;
-    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    std::vector<size_t> multi_thr2_row_mt_size_enc;
-    std::vector<std::string> multi_thr2_row_mt_md5_enc;
-    std::vector<std::string> multi_thr2_row_mt_md5_dec;
-    multi_thr2_row_mt_size_enc = size_enc_;
-    multi_thr2_row_mt_md5_enc = md5_enc_;
-    multi_thr2_row_mt_md5_dec = md5_dec_;
-    size_enc_.clear();
-    md5_enc_.clear();
-    md5_dec_.clear();
+      // Disable threads=3 test for now to reduce the time so that the nightly
+      // test would not time out.
+      // cfg_.g_threads = 3;
+      // ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      // std::vector<size_t> multi_thr3_row_mt_size_enc;
+      // std::vector<std::string> multi_thr3_row_mt_md5_enc;
+      // std::vector<std::string> multi_thr3_row_mt_md5_dec;
+      // multi_thr3_row_mt_size_enc = size_enc_;
+      // multi_thr3_row_mt_md5_enc = md5_enc_;
+      // multi_thr3_row_mt_md5_dec = md5_dec_;
+      // size_enc_.clear();
+      // md5_enc_.clear();
+      // md5_dec_.clear();
+      // Check that the vectors are equal.
+      // ASSERT_EQ(multi_thr3_row_mt_size_enc, multi_thr2_row_mt_size_enc);
+      // ASSERT_EQ(multi_thr3_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
+      // ASSERT_EQ(multi_thr3_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
 
-    // Disable threads=3 test for now to reduce the time so that the nightly
-    // test would not time out.
-    // cfg_.g_threads = 3;
-    // ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    // std::vector<size_t> multi_thr3_row_mt_size_enc;
-    // std::vector<std::string> multi_thr3_row_mt_md5_enc;
-    // std::vector<std::string> multi_thr3_row_mt_md5_dec;
-    // multi_thr3_row_mt_size_enc = size_enc_;
-    // multi_thr3_row_mt_md5_enc = md5_enc_;
-    // multi_thr3_row_mt_md5_dec = md5_dec_;
-    // size_enc_.clear();
-    // md5_enc_.clear();
-    // md5_dec_.clear();
-    // Check that the vectors are equal.
-    // ASSERT_EQ(multi_thr3_row_mt_size_enc, multi_thr2_row_mt_size_enc);
-    // ASSERT_EQ(multi_thr3_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
-    // ASSERT_EQ(multi_thr3_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
+      cfg_.g_threads = 4;
+      ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+      std::vector<size_t> multi_thr4_row_mt_size_enc;
+      std::vector<std::string> multi_thr4_row_mt_md5_enc;
+      std::vector<std::string> multi_thr4_row_mt_md5_dec;
+      multi_thr4_row_mt_size_enc = size_enc_;
+      multi_thr4_row_mt_md5_enc = md5_enc_;
+      multi_thr4_row_mt_md5_dec = md5_dec_;
+      size_enc_.clear();
+      md5_enc_.clear();
+      md5_dec_.clear();
 
-    cfg_.g_threads = 4;
-    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
-    std::vector<size_t> multi_thr4_row_mt_size_enc;
-    std::vector<std::string> multi_thr4_row_mt_md5_enc;
-    std::vector<std::string> multi_thr4_row_mt_md5_dec;
-    multi_thr4_row_mt_size_enc = size_enc_;
-    multi_thr4_row_mt_md5_enc = md5_enc_;
-    multi_thr4_row_mt_md5_dec = md5_dec_;
-    size_enc_.clear();
-    md5_enc_.clear();
-    md5_dec_.clear();
-
-    // Check that the vectors are equal.
-    ASSERT_EQ(multi_thr4_row_mt_size_enc, multi_thr2_row_mt_size_enc);
-    ASSERT_EQ(multi_thr4_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
-    ASSERT_EQ(multi_thr4_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
+      // Check that the vectors are equal.
+      ASSERT_EQ(multi_thr4_row_mt_size_enc, multi_thr2_row_mt_size_enc);
+      ASSERT_EQ(multi_thr4_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
+      ASSERT_EQ(multi_thr4_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
+    }
   }
 
   bool encoder_initialized_;
@@ -233,14 +234,15 @@ TEST_P(AVxEncoderThreadTestLarge, EncoderResultTest) {
 AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadTest,
                           ::testing::Values(::libaom_test::kTwoPassGood),
                           ::testing::Range(2, 4), ::testing::Values(0, 2),
-                          ::testing::Values(0, 1));
+                          ::testing::Values(0, 1), ::testing::Values(0, 1));
 
 // Test cpu_used 0 and 1.
 AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadTestLarge,
                           ::testing::Values(::libaom_test::kTwoPassGood,
                                             ::libaom_test::kOnePassGood),
                           ::testing::Range(0, 2), ::testing::Values(0, 1, 2, 6),
-                          ::testing::Values(0, 1, 2, 6));
+                          ::testing::Values(0, 1, 2, 6),
+                          ::testing::Values(0, 1));
 
 class AVxEncoderThreadLSTest : public AVxEncoderThreadTest {
   virtual void SetTileSize(libaom_test::Encoder *encoder) {
@@ -269,5 +271,5 @@ AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadLSTestLarge,
                           ::testing::Values(::libaom_test::kTwoPassGood,
                                             ::libaom_test::kOnePassGood),
                           ::testing::Range(0, 4), ::testing::Values(0, 6),
-                          ::testing::Values(0, 6));
+                          ::testing::Values(0, 6), ::testing::Values(0, 1));
 }  // namespace
diff --git a/third_party/aom/test/external_frame_buffer_test.cc b/third_party/aom/test/external_frame_buffer_test.cc
index c2af059a4e..4938a64bbb 100644
--- a/third_party/aom/test/external_frame_buffer_test.cc
+++ b/third_party/aom/test/external_frame_buffer_test.cc
@@ -8,8 +8,9 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include <memory>
 #include <string>
-
+#include "common/tools_common.h"
 #include "config/aom_config.h"
 #include "test/codec_factory.h"
 #include "test/decode_test_driver.h"
@@ -57,7 +58,7 @@ class ExternalFrameBufferList {
 
   // Searches the frame buffer list for a free frame buffer. Makes sure
   // that the frame buffer is at least |min_size| in bytes. Marks that the
-  // frame buffer is in use by libvpx. Finally sets |fb| to point to the
+  // frame buffer is in use by libaom. Finally sets |fb| to point to the
   // external frame buffer. Returns < 0 on an error.
   int GetFreeFrameBuffer(size_t min_size, aom_codec_frame_buffer_t *fb) {
     EXPECT_TRUE(fb != NULL);
@@ -113,9 +114,9 @@ class ExternalFrameBufferList {
     return 0;
   }
 
-  // Checks that the ximage data is contained within the external frame buffer
-  // private data passed back in the ximage.
-  void CheckXImageFrameBuffer(const aom_image_t *img) {
+  // Checks that the aom_image_t data is contained within the external frame
+  // buffer private data passed back in the aom_image_t.
+  void CheckImageFrameBuffer(const aom_image_t *img) {
     if (img->fb_priv != NULL) {
       const struct ExternalFrameBuffer *const ext_fb =
           reinterpret_cast<ExternalFrameBuffer *>(img->fb_priv);
@@ -157,7 +158,7 @@ class ExternalFrameBufferList {
 
 #if CONFIG_WEBM_IO
 
-// Callback used by libvpx to request the application to return a frame
+// Callback used by libaom to request the application to return a frame
 // buffer of at least |min_size| in bytes.
 int get_aom_frame_buffer(void *user_priv, size_t min_size,
                          aom_codec_frame_buffer_t *fb) {
@@ -166,7 +167,7 @@ int get_aom_frame_buffer(void *user_priv, size_t min_size,
   return fb_list->GetFreeFrameBuffer(min_size, fb);
 }
 
-// Callback used by libvpx to tell the application that |fb| is not needed
+// Callback used by libaom to tell the application that |fb| is not needed
 // anymore.
 int release_aom_frame_buffer(void *user_priv, aom_codec_frame_buffer_t *fb) {
   ExternalFrameBufferList *const fb_list =
@@ -217,7 +218,7 @@ class ExternalFrameBufferMD5Test
       const libaom_test::CompressedVideoSource &video,
       libaom_test::Decoder *decoder) {
     if (num_buffers_ > 0 && video.frame_number() == 0) {
-      // Have libvpx use frame buffers we create.
+      // Have libaom use frame buffers we create.
       ASSERT_TRUE(fb_list_.CreateBufferList(num_buffers_));
       ASSERT_EQ(AOM_CODEC_OK,
                 decoder->SetFrameBufferFunctions(GetAV1FrameBuffer,
@@ -243,7 +244,23 @@ class ExternalFrameBufferMD5Test
     expected_md5[32] = '\0';
 
     ::libaom_test::MD5 md5_res;
-    md5_res.Add(&img);
+#if !CONFIG_LOWBITDEPTH
+    const aom_img_fmt_t shifted_fmt =
+        (aom_img_fmt)(img.fmt & ~AOM_IMG_FMT_HIGHBITDEPTH);
+    if (img.bit_depth == 8 && shifted_fmt != img.fmt) {
+      aom_image_t *img_shifted =
+          aom_img_alloc(NULL, shifted_fmt, img.d_w, img.d_h, 16);
+      img_shifted->bit_depth = img.bit_depth;
+      img_shifted->monochrome = img.monochrome;
+      aom_img_downshift(img_shifted, &img, 0);
+      md5_res.Add(img_shifted);
+      aom_img_free(img_shifted);
+    } else {
+#endif
+      md5_res.Add(&img);
+#if !CONFIG_LOWBITDEPTH
+    }
+#endif
     const char *const actual_md5 = md5_res.Get();
 
     // Check md5 match.
@@ -279,16 +296,16 @@ class ExternalFrameBufferMD5Test
 };
 
 #if CONFIG_WEBM_IO
-const char kAV1TestFile[] = "av1-1-b8-01-size-226x226.ivf";
+const char kAV1TestFile[] = "av1-1-b8-03-sizeup.mkv";
 const char kAV1NonRefTestFile[] = "av1-1-b8-01-size-226x226.ivf";
 
-// Class for testing passing in external frame buffers to libvpx.
+// Class for testing passing in external frame buffers to libaom.
 class ExternalFrameBufferTest : public ::testing::Test {
  protected:
   ExternalFrameBufferTest() : video_(NULL), decoder_(NULL), num_buffers_(0) {}
 
   virtual void SetUp() {
-    video_ = new libaom_test::IVFVideoSource(kAV1TestFile);
+    video_ = new libaom_test::WebMVideoSource(kAV1TestFile);
     ASSERT_TRUE(video_ != NULL);
     video_->Init();
     video_->Begin();
@@ -305,7 +322,7 @@ class ExternalFrameBufferTest : public ::testing::Test {
     video_ = NULL;
   }
 
-  // Passes the external frame buffer information to libvpx.
+  // Passes the external frame buffer information to libaom.
   aom_codec_err_t SetFrameBufferFunctions(
       int num_buffers, aom_get_frame_buffer_cb_fn_t cb_get,
       aom_release_frame_buffer_cb_fn_t cb_release) {
@@ -342,11 +359,11 @@ class ExternalFrameBufferTest : public ::testing::Test {
 
     // Get decompressed data
     while ((img = dec_iter.Next()) != NULL) {
-      fb_list_.CheckXImageFrameBuffer(img);
+      fb_list_.CheckImageFrameBuffer(img);
     }
   }
 
-  libaom_test::IVFVideoSource *video_;
+  libaom_test::CompressedVideoSource *video_;
   libaom_test::AV1Decoder *decoder_;
   int num_buffers_;
   ExternalFrameBufferList fb_list_;
@@ -373,12 +390,13 @@ class ExternalFrameBufferNonRefTest : public ExternalFrameBufferTest {
 #endif  // CONFIG_WEBM_IO
 
 // This test runs through the set of test vectors, and decodes them.
-// Libvpx will call into the application to allocate a frame buffer when
+// Libaom will call into the application to allocate a frame buffer when
 // needed. The md5 checksums are computed for each frame in the video file.
 // If md5 checksums match the correct md5 data, then the test is passed.
 // Otherwise, the test failed.
-TEST_P(ExternalFrameBufferMD5Test, DISABLED_ExtFBMD5Match) {
+TEST_P(ExternalFrameBufferMD5Test, ExtFBMD5Match) {
   const std::string filename = GET_PARAM(kVideoNameParam);
+  aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
 
   // Number of buffers equals #AOM_MAXIMUM_REF_BUFFERS +
   // #AOM_MAXIMUM_WORK_BUFFERS + four jitter buffers.
@@ -388,7 +406,7 @@ TEST_P(ExternalFrameBufferMD5Test, DISABLED_ExtFBMD5Match) {
   set_num_buffers(num_buffers);
 
   // Open compressed video file.
-  testing::internal::scoped_ptr<libaom_test::CompressedVideoSource> video;
+  std::unique_ptr<libaom_test::CompressedVideoSource> video;
   if (filename.substr(filename.length() - 3, 3) == "ivf") {
     video.reset(new libaom_test::IVFVideoSource(filename));
   } else {
@@ -407,8 +425,12 @@ TEST_P(ExternalFrameBufferMD5Test, DISABLED_ExtFBMD5Match) {
   const std::string md5_filename = filename + ".md5";
   OpenMD5File(md5_filename);
 
+  // Set decode config.
+  cfg.allow_lowbitdepth = CONFIG_LOWBITDEPTH;
+  set_cfg(cfg);
+
   // Decode frame, and check the md5 matching.
-  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get(), cfg));
 }
 
 #if CONFIG_WEBM_IO
@@ -434,7 +456,7 @@ TEST_F(ExternalFrameBufferTest, EightJitterBuffers) {
   ASSERT_EQ(AOM_CODEC_OK, DecodeRemainingFrames());
 }
 
-TEST_F(ExternalFrameBufferTest, DISABLED_NotEnoughBuffers) {
+TEST_F(ExternalFrameBufferTest, NotEnoughBuffers) {
   // Minimum number of external frame buffers for AV1 is
   // #AOM_MAXIMUM_REF_BUFFERS + #AOM_MAXIMUM_WORK_BUFFERS. Most files will
   // only use 5 frame buffers at one time.
@@ -448,7 +470,7 @@ TEST_F(ExternalFrameBufferTest, DISABLED_NotEnoughBuffers) {
   ASSERT_EQ(AOM_CODEC_MEM_ERROR, DecodeRemainingFrames());
 }
 
-TEST_F(ExternalFrameBufferTest, DISABLED_NoRelease) {
+TEST_F(ExternalFrameBufferTest, NoRelease) {
   const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
   ASSERT_EQ(AOM_CODEC_OK,
             SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
diff --git a/third_party/aom/test/filterintra_test.cc b/third_party/aom/test/filterintra_test.cc
index 5971349406..93e26ae46b 100644
--- a/third_party/aom/test/filterintra_test.cc
+++ b/third_party/aom/test/filterintra_test.cc
@@ -21,8 +21,8 @@
 
 namespace {
 
-using ::testing::tuple;
 using libaom_test::ACMRandom;
+using ::testing::tuple;
 
 typedef void (*Predictor)(uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size,
                           const uint8_t *above, const uint8_t *left, int mode);
diff --git a/third_party/aom/test/frame_size_tests.cc b/third_party/aom/test/frame_size_tests.cc
index eaf0b8370a..1546012a30 100644
--- a/third_party/aom/test/frame_size_tests.cc
+++ b/third_party/aom/test/frame_size_tests.cc
@@ -35,7 +35,7 @@ class AV1FrameSizeTests : public ::testing::Test,
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, 7);
       encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
       encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
diff --git a/third_party/aom/test/hiprec_convolve_test.cc b/third_party/aom/test/hiprec_convolve_test.cc
index f94a0730c1..dcf85230ed 100644
--- a/third_party/aom/test/hiprec_convolve_test.cc
+++ b/third_party/aom/test/hiprec_convolve_test.cc
@@ -12,11 +12,11 @@
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 #include "test/hiprec_convolve_test_util.h"
 
-using ::testing::make_tuple;
-using ::testing::tuple;
 using libaom_test::ACMRandom;
 using libaom_test::AV1HighbdHiprecConvolve::AV1HighbdHiprecConvolveTest;
 using libaom_test::AV1HiprecConvolve::AV1HiprecConvolveTest;
+using ::testing::make_tuple;
+using ::testing::tuple;
 
 namespace {
 
diff --git a/third_party/aom/test/hiprec_convolve_test_util.cc b/third_party/aom/test/hiprec_convolve_test_util.cc
index 2672bcec3d..f5bf56e907 100644
--- a/third_party/aom/test/hiprec_convolve_test_util.cc
+++ b/third_party/aom/test/hiprec_convolve_test_util.cc
@@ -31,7 +31,7 @@ static void generate_kernels(ACMRandom *rnd, InterpKernel hkernel,
   hkernel[2] = hkernel[4] =
       WIENER_FILT_TAP2_MINV +
       rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 1 - WIENER_FILT_TAP2_MINV);
-  hkernel[3] = -(hkernel[0] + hkernel[1] + hkernel[2]);
+  hkernel[3] = -2 * (hkernel[0] + hkernel[1] + hkernel[2]);
   hkernel[7] = 0;
 
   vkernel[0] = vkernel[6] =
@@ -43,7 +43,7 @@ static void generate_kernels(ACMRandom *rnd, InterpKernel hkernel,
   vkernel[2] = vkernel[4] =
       WIENER_FILT_TAP2_MINV +
       rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 1 - WIENER_FILT_TAP2_MINV);
-  vkernel[3] = -(vkernel[0] + vkernel[1] + vkernel[2]);
+  vkernel[3] = -2 * (vkernel[0] + vkernel[1] + vkernel[2]);
   vkernel[7] = 0;
 }
 
diff --git a/third_party/aom/test/horver_correlation_test.cc b/third_party/aom/test/horver_correlation_test.cc
new file mode 100644
index 0000000000..72fd46ac23
--- /dev/null
+++ b/third_party/aom/test/horver_correlation_test.cc
@@ -0,0 +1,146 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+typedef void (*HorverFunc)(const int16_t *diff, int stride, int w, int h,
+                           float *hcorr, float *vcorr);
+
+typedef ::testing::tuple<const HorverFunc> HorverTestParam;
+
+class HorverTest : public ::testing::TestWithParam<HorverTestParam> {
+ public:
+  virtual void SetUp() {
+    data_buf_ = (int16_t *)aom_malloc(MAX_SB_SQUARE * sizeof(int16_t));
+    ASSERT_NE(data_buf_, nullptr);
+    target_func_ = GET_PARAM(0);
+  }
+  virtual void TearDown() { aom_free(data_buf_); }
+  void RunHorverTest(void);
+  void RunHorverTest_ExtremeValues(void);
+  void RunHorverSpeedTest(int run_times);
+
+ private:
+  HorverFunc target_func_;
+  ACMRandom rng_;
+  int16_t *data_buf_;
+};
+
+void HorverTest::RunHorverTest(void) {
+  for (int block_size = 0; block_size < BLOCK_SIZES_ALL; block_size++) {
+    const int w = block_size_wide[block_size];
+    const int h = block_size_high[block_size];
+    for (int iter = 0; iter < 1000 && !HasFatalFailure(); ++iter) {
+      float hcorr_ref = 0.0, vcorr_ref = 0.0;
+      float hcorr_test = 0.0, vcorr_test = 0.0;
+
+      for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+        data_buf_[i] = (rng_.Rand16() % (1 << 12)) - (1 << 11);
+      }
+
+      av1_get_horver_correlation_full_c(data_buf_, MAX_SB_SIZE, w, h,
+                                        &hcorr_ref, &vcorr_ref);
+
+      target_func_(data_buf_, MAX_SB_SIZE, w, h, &hcorr_test, &vcorr_test);
+
+      ASSERT_LE(fabs(hcorr_ref - hcorr_test), 1e-6)
+          << "hcorr incorrect (" << w << "x" << h << ")";
+      ASSERT_LE(fabs(vcorr_ref - vcorr_test), 1e-6)
+          << "vcorr incorrect (" << w << "x" << h << ")";
+    }
+    //    printf("(%3dx%-3d) passed\n", w, h);
+  }
+}
+
+void HorverTest::RunHorverSpeedTest(int run_times) {
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    data_buf_[i] = rng_.Rand16() % (1 << 12);
+  }
+
+  for (int block_size = 0; block_size < BLOCK_SIZES_ALL; block_size++) {
+    const int w = block_size_wide[block_size];
+    const int h = block_size_high[block_size];
+    float hcorr_ref = 0.0, vcorr_ref = 0.0;
+    float hcorr_test = 0.0, vcorr_test = 0.0;
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      av1_get_horver_correlation_full_c(data_buf_, MAX_SB_SIZE, w, h,
+                                        &hcorr_ref, &vcorr_ref);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      target_func_(data_buf_, MAX_SB_SIZE, w, h, &hcorr_test, &vcorr_test);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+
+    printf("%3dx%-3d:%7.2f/%7.2fns (%3.2f)\n", w, h, time1, time2,
+           time1 / time2);
+  }
+}
+
+void HorverTest::RunHorverTest_ExtremeValues(void) {
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    // Most of get_horver_test is squaring and summing, so simply saturating
+    // the whole buffer is mostly likely to cause an overflow.
+    data_buf_[i] = (1 << 12) - 1;
+  }
+
+  for (int block_size = 0; block_size < BLOCK_SIZES_ALL; block_size++) {
+    const int w = block_size_wide[block_size];
+    const int h = block_size_high[block_size];
+    float hcorr_ref = 0.0, vcorr_ref = 0.0;
+    float hcorr_test = 0.0, vcorr_test = 0.0;
+
+    av1_get_horver_correlation_full_c(data_buf_, MAX_SB_SIZE, w, h, &hcorr_ref,
+                                      &vcorr_ref);
+    target_func_(data_buf_, MAX_SB_SIZE, w, h, &hcorr_test, &vcorr_test);
+
+    ASSERT_LE(fabs(hcorr_ref - hcorr_test), 1e-6) << "hcorr incorrect";
+    ASSERT_LE(fabs(vcorr_ref - vcorr_test), 1e-6) << "vcorr incorrect";
+  }
+}
+
+TEST_P(HorverTest, RandomValues) { RunHorverTest(); }
+
+TEST_P(HorverTest, ExtremeValues) { RunHorverTest_ExtremeValues(); }
+
+TEST_P(HorverTest, DISABLED_Speed) { RunHorverSpeedTest(100000); }
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, HorverTest,
+    ::testing::Values(av1_get_horver_correlation_full_sse4_1));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, HorverTest, ::testing::Values(av1_get_horver_correlation_full_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/horz_superres_test.cc b/third_party/aom/test/horz_superres_test.cc
index 973f55b668..1627684b9e 100644
--- a/third_party/aom/test/horz_superres_test.cc
+++ b/third_party/aom/test/horz_superres_test.cc
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
+
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
 #include "av1/encoder/encoder.h"
@@ -125,7 +127,7 @@ class HorzSuperresEndToEndTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
       encoder->Control(AV1E_SET_TILE_COLUMNS, 4);
 
@@ -152,7 +154,7 @@ class HorzSuperresEndToEndTest
   double GetPsnrThreshold() { return kPSNRThresholds[test_video_idx_]; }
 
   void DoTest() {
-    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    std::unique_ptr<libaom_test::VideoSource> video;
     video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
                                                 test_video_param_.limit));
     ASSERT_TRUE(video.get() != NULL);
@@ -248,7 +250,7 @@ class HorzSuperresQThreshEndToEndTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
       encoder->Control(AV1E_SET_TILE_COLUMNS, 0);
 
@@ -275,7 +277,7 @@ class HorzSuperresQThreshEndToEndTest
   double GetPsnrThreshold() { return kPSNRThresholds[test_video_idx_]; }
 
   void DoTest() {
-    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    std::unique_ptr<libaom_test::VideoSource> video;
     video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
                                                 test_video_param_.limit));
     ASSERT_TRUE(video.get() != NULL);
diff --git a/third_party/aom/test/intrapred_test.cc b/third_party/aom/test/intrapred_test.cc
index 1a1c0fc421..43ab773b61 100644
--- a/third_party/aom/test/intrapred_test.cc
+++ b/third_party/aom/test/intrapred_test.cc
@@ -197,8 +197,8 @@ TEST_P(LowbdIntraPredTest, Bitexact) {
       highbd_entry(type, 32, 16, opt, bd), highbd_entry(type, 32, 32, opt, bd)
 #endif
 
-  // ---------------------------------------------------------------------------
-  // Low Bit Depth Tests
+// ---------------------------------------------------------------------------
+// Low Bit Depth Tests
 
 #define lowbd_entry(type, width, height, opt)                                  \
   IntraPredFunc<IntraPred>(&aom_##type##_predictor_##width##x##height##_##opt, \
diff --git a/third_party/aom/test/invalid_file_test.cc b/third_party/aom/test/invalid_file_test.cc
index 5b4f5a6c3d..0478c7e5e1 100644
--- a/third_party/aom/test/invalid_file_test.cc
+++ b/third_party/aom/test/invalid_file_test.cc
@@ -44,6 +44,11 @@ class InvalidFileTest : public ::libaom_test::DecoderTest,
         << "Result file open failed. Filename: " << res_file_name;
   }
 
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     const unsigned int /*frame_number*/) {
+    EXPECT_NE(img.fb_priv, nullptr);
+  }
+
   virtual bool HandleDecodeResult(
       const aom_codec_err_t res_dec,
       const libaom_test::CompressedVideoSource &video,
@@ -108,12 +113,22 @@ TEST_P(InvalidFileTest, ReturnCode) { RunTest(); }
 
 const DecodeParam kAV1InvalidFileTests[] = {
   { 1, "invalid-bug-1814.ivf" },
+  { 1, "invalid-chromium-906381.ivf" },
+  { 1, "invalid-oss-fuzz-9288.ivf" },
   { 4, "invalid-oss-fuzz-9463.ivf" },
   { 1, "invalid-oss-fuzz-9482.ivf" },
   { 1, "invalid-oss-fuzz-9720.ivf" },
   { 1, "invalid-oss-fuzz-10061.ivf" },
   { 1, "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf" },
   { 1, "invalid-oss-fuzz-10227.ivf" },
+  { 1, "invalid-oss-fuzz-10389.ivf" },
+  { 4, "invalid-oss-fuzz-10555.ivf" },
+  { 1, "invalid-oss-fuzz-10705.ivf" },
+  { 1, "invalid-oss-fuzz-10723.ivf" },
+  { 1, "invalid-oss-fuzz-10779.ivf" },
+  { 1, "invalid-oss-fuzz-11477.ivf" },
+  { 1, "invalid-oss-fuzz-11479.ivf" },
+  { 1, "invalid-oss-fuzz-11523.ivf" },
 };
 
 AV1_INSTANTIATE_TEST_CASE(InvalidFileTest,
diff --git a/third_party/aom/test/lightfield_test.sh b/third_party/aom/test/lightfield_test.sh
index b957a6b794..19b6934a3b 100755
--- a/third_party/aom/test/lightfield_test.sh
+++ b/third_party/aom/test/lightfield_test.sh
@@ -50,13 +50,14 @@ lightfield_test() {
   # be decoded by an AV1 decoder.
   local bs_decoder="${LIBAOM_BIN_PATH}/lightfield_bitstream_parsing${AOM_TEST_EXE_SUFFIX}"
   local tl_file="${AOM_TEST_OUTPUT_DIR}/vase_tile_list.ivf"
+  local tl_text_file="${LIBAOM_TEST_DATA_PATH}/vase10x10_tiles.txt"
   if [ ! -x "${bs_decoder}" ]; then
     elog "${bs_decoder} does not exist or is not executable."
     return 1
   fi
 
   eval "${AOM_TEST_PREFIX}" "${bs_decoder}" "${lf_file}" "${tl_file}" \
-      "${num_references}" ${devnull}
+      "${num_references}" "${tl_text_file}" ${devnull}
 
   [ -e "${tl_file}" ] || return 1
 
@@ -82,7 +83,7 @@ lightfield_test() {
   fi
 
   eval "${AOM_TEST_PREFIX}" "${ref_decoder}" "${lf_file}" "${tl_reffile}" \
-      "${num_references}" ${devnull}
+      "${num_references}" "${tl_text_file}" ${devnull}
 
   [ -e "${tl_reffile}" ] || return 1
 
diff --git a/third_party/aom/test/lossless_test.cc b/third_party/aom/test/lossless_test.cc
index 3f8e89c815..71ae5e72be 100644
--- a/third_party/aom/test/lossless_test.cc
+++ b/third_party/aom/test/lossless_test.cc
@@ -40,7 +40,7 @@ class LosslessTestLarge
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       // Only call Control if quantizer > 0 to verify that using quantizer
       // alone will activate lossless
       if (cfg_.rc_max_quantizer > 0 || cfg_.rc_min_quantizer > 0) {
diff --git a/third_party/aom/test/motion_vector_test.cc b/third_party/aom/test/motion_vector_test.cc
index 27eb938930..2636c39aa2 100644
--- a/third_party/aom/test/motion_vector_test.cc
+++ b/third_party/aom/test/motion_vector_test.cc
@@ -9,6 +9,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
+#include <memory>
+
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
 #include "test/codec_factory.h"
@@ -60,7 +62,7 @@ class MotionVectorTestLarge
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, cpu_used_);
       encoder->Control(AV1E_ENABLE_MOTION_VECTOR_UNIT_TEST, mv_test_mode_);
       if (encoding_mode_ != ::libaom_test::kRealTime) {
@@ -90,7 +92,7 @@ TEST_P(MotionVectorTestLarge, OverallTest) {
   cfg_.g_profile = 0;
   init_flags_ = AOM_CODEC_USE_PSNR;
 
-  testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+  std::unique_ptr<libaom_test::VideoSource> video;
   video.reset(new libaom_test::YUVVideoSource(
       "niklas_640_480_30.yuv", AOM_IMG_FMT_I420, width, height, 30, 1, 0, 3));
 
diff --git a/third_party/aom/test/pickrst_test.cc b/third_party/aom/test/pickrst_test.cc
index 040e8e8b76..0aa49b63f9 100644
--- a/third_party/aom/test/pickrst_test.cc
+++ b/third_party/aom/test/pickrst_test.cc
@@ -11,19 +11,20 @@
 
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
-#include "test/function_equivalence_test.h"
 #include "test/register_state_check.h"
+#include "test/acm_random.h"
+#include "test/util.h"
 
 #include "config/aom_config.h"
 #include "config/aom_dsp_rtcd.h"
 
 #include "aom/aom_integer.h"
+#include "aom_ports/aom_timer.h"
 #include "av1/encoder/pickrst.h"
-using libaom_test::FunctionEquivalenceTest;
 
 #define MAX_DATA_BLOCK 384
 
-namespace {
+namespace pickrst_test_lowbd {
 static const int kIterations = 100;
 
 typedef int64_t (*lowbd_pixel_proj_error_func)(
@@ -31,8 +32,6 @@ typedef int64_t (*lowbd_pixel_proj_error_func)(
     const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
     int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params);
 
-typedef libaom_test::FuncParam<lowbd_pixel_proj_error_func> TestFuncs;
-
 ////////////////////////////////////////////////////////////////////////////////
 // 8 bit
 ////////////////////////////////////////////////////////////////////////////////
@@ -46,13 +45,17 @@ class PixelProjErrorTest
   virtual void SetUp() {
     target_func_ = GET_PARAM(0);
     src_ = (uint8_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
-                                  sizeof(uint8_t)));
+                                  sizeof(*src_)));
+    ASSERT_NE(src_, nullptr);
     dgd_ = (uint8_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
-                                  sizeof(uint8_t)));
+                                  sizeof(*dgd_)));
+    ASSERT_NE(dgd_, nullptr);
     flt0_ = (int32_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
-                                   sizeof(int32_t)));
+                                   sizeof(*flt0_)));
+    ASSERT_NE(flt0_, nullptr);
     flt1_ = (int32_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
-                                   sizeof(int32_t)));
+                                   sizeof(*flt1_)));
+    ASSERT_NE(flt1_, nullptr);
   }
   virtual void TearDown() {
     aom_free(src_);
@@ -60,19 +63,19 @@ class PixelProjErrorTest
     aom_free(flt0_);
     aom_free(flt1_);
   }
-  void runPixelProjErrorTest(int32_t run_times);
-  void runPixelProjErrorTest_ExtremeValues();
+  void RunPixelProjErrorTest(int32_t run_times);
+  void RunPixelProjErrorTest_ExtremeValues();
 
  private:
   lowbd_pixel_proj_error_func target_func_;
-  ACMRandom rng_;
+  libaom_test::ACMRandom rng_;
   uint8_t *src_;
   uint8_t *dgd_;
   int32_t *flt0_;
   int32_t *flt1_;
 };
 
-void PixelProjErrorTest::runPixelProjErrorTest(int32_t run_times) {
+void PixelProjErrorTest::RunPixelProjErrorTest(int32_t run_times) {
   int h_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
   int v_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
   const int dgd_stride = MAX_DATA_BLOCK;
@@ -124,7 +127,7 @@ void PixelProjErrorTest::runPixelProjErrorTest(int32_t run_times) {
   }
 }
 
-void PixelProjErrorTest::runPixelProjErrorTest_ExtremeValues() {
+void PixelProjErrorTest::RunPixelProjErrorTest_ExtremeValues() {
   const int h_start = 0;
   int h_end = 192;
   const int v_start = 0;
@@ -165,13 +168,13 @@ void PixelProjErrorTest::runPixelProjErrorTest_ExtremeValues() {
   }
 }
 
-TEST_P(PixelProjErrorTest, RandomValues) { runPixelProjErrorTest(1); }
+TEST_P(PixelProjErrorTest, RandomValues) { RunPixelProjErrorTest(1); }
 
 TEST_P(PixelProjErrorTest, ExtremeValues) {
-  runPixelProjErrorTest_ExtremeValues();
+  RunPixelProjErrorTest_ExtremeValues();
 }
 
-TEST_P(PixelProjErrorTest, DISABLED_Speed) { runPixelProjErrorTest(200000); }
+TEST_P(PixelProjErrorTest, DISABLED_Speed) { RunPixelProjErrorTest(200000); }
 
 #if HAVE_SSE4_1
 INSTANTIATE_TEST_CASE_P(SSE4_1, PixelProjErrorTest,
@@ -184,4 +187,171 @@ INSTANTIATE_TEST_CASE_P(AVX2, PixelProjErrorTest,
                         ::testing::Values(av1_lowbd_pixel_proj_error_avx2));
 #endif  // HAVE_AVX2
 
-}  // namespace
+}  // namespace pickrst_test_lowbd
+
+namespace pickrst_test_highbd {
+static const int kIterations = 100;
+
+typedef int64_t (*highbd_pixel_proj_error_func)(
+    const uint8_t *src8, int width, int height, int src_stride,
+    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
+    int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params);
+
+////////////////////////////////////////////////////////////////////////////////
+// High bit-depth
+////////////////////////////////////////////////////////////////////////////////
+
+typedef ::testing::tuple<const highbd_pixel_proj_error_func>
+    PixelProjErrorTestParam;
+
+class PixelProjHighbdErrorTest
+    : public ::testing::TestWithParam<PixelProjErrorTestParam> {
+ public:
+  virtual void SetUp() {
+    target_func_ = GET_PARAM(0);
+    src_ =
+        (uint16_t *)aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*src_));
+    ASSERT_NE(src_, nullptr);
+    dgd_ =
+        (uint16_t *)aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*dgd_));
+    ASSERT_NE(dgd_, nullptr);
+    flt0_ =
+        (int32_t *)aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*flt0_));
+    ASSERT_NE(flt0_, nullptr);
+    flt1_ =
+        (int32_t *)aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*flt1_));
+    ASSERT_NE(flt1_, nullptr);
+  }
+  virtual void TearDown() {
+    aom_free(src_);
+    aom_free(dgd_);
+    aom_free(flt0_);
+    aom_free(flt1_);
+  }
+  void RunPixelProjErrorTest(int32_t run_times);
+  void RunPixelProjErrorTest_ExtremeValues();
+
+ private:
+  highbd_pixel_proj_error_func target_func_;
+  libaom_test::ACMRandom rng_;
+  uint16_t *src_;
+  uint16_t *dgd_;
+  int32_t *flt0_;
+  int32_t *flt1_;
+};
+
+void PixelProjHighbdErrorTest::RunPixelProjErrorTest(int32_t run_times) {
+  int h_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
+  int v_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
+  const int dgd_stride = MAX_DATA_BLOCK;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int flt0_stride = MAX_DATA_BLOCK;
+  const int flt1_stride = MAX_DATA_BLOCK;
+  sgr_params_type params;
+  int xq[2];
+  const int iters = run_times == 1 ? kIterations : 4;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    int64_t err_ref = 0, err_test = 1;
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_[i] = rng_.Rand16() % (1 << 12);
+      src_[i] = rng_.Rand16() % (1 << 12);
+      flt0_[i] = rng_.Rand15Signed();
+      flt1_[i] = rng_.Rand15Signed();
+    }
+    xq[0] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    xq[1] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    params.r[0] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter % 2);
+    params.r[1] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter / 2);
+    params.s[0] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter % 2);
+    params.s[1] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter / 2);
+    uint8_t *dgd8 = CONVERT_TO_BYTEPTR(dgd_);
+    uint8_t *src8 = CONVERT_TO_BYTEPTR(src_);
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      err_ref = av1_highbd_pixel_proj_error_c(
+          src8, h_end, v_end, src_stride, dgd8, dgd_stride, flt0_, flt0_stride,
+          flt1_, flt1_stride, xq, &params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      err_test =
+          target_func_(src8, h_end, v_end, src_stride, dgd8, dgd_stride, flt0_,
+                       flt0_stride, flt1_, flt1_stride, xq, &params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 10) {
+      printf("r0 %d r1 %d %3dx%-3d:%7.2f/%7.2fns (%3.2f)\n", params.r[0],
+             params.r[1], h_end, v_end, time1, time2, time1 / time2);
+    }
+    ASSERT_EQ(err_ref, err_test);
+  }
+}
+
+void PixelProjHighbdErrorTest::RunPixelProjErrorTest_ExtremeValues() {
+  const int h_start = 0;
+  int h_end = 192;
+  const int v_start = 0;
+  int v_end = 192;
+  const int dgd_stride = MAX_DATA_BLOCK;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int flt0_stride = MAX_DATA_BLOCK;
+  const int flt1_stride = MAX_DATA_BLOCK;
+  sgr_params_type params;
+  int xq[2];
+  const int iters = kIterations;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    int64_t err_ref = 0, err_test = 1;
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_[i] = 0;
+      src_[i] = (1 << 12) - 1;
+      flt0_[i] = rng_.Rand15Signed();
+      flt1_[i] = rng_.Rand15Signed();
+    }
+    xq[0] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    xq[1] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    params.r[0] = rng_.Rand8() % MAX_RADIUS;
+    params.r[1] = rng_.Rand8() % MAX_RADIUS;
+    params.s[0] = rng_.Rand8() % MAX_RADIUS;
+    params.s[1] = rng_.Rand8() % MAX_RADIUS;
+    uint8_t *dgd8 = CONVERT_TO_BYTEPTR(dgd_);
+    uint8_t *src8 = CONVERT_TO_BYTEPTR(src_);
+
+    err_ref = av1_highbd_pixel_proj_error_c(
+        src8, h_end - h_start, v_end - v_start, src_stride, dgd8, dgd_stride,
+        flt0_, flt0_stride, flt1_, flt1_stride, xq, &params);
+
+    err_test = target_func_(src8, h_end - h_start, v_end - v_start, src_stride,
+                            dgd8, dgd_stride, flt0_, flt0_stride, flt1_,
+                            flt1_stride, xq, &params);
+
+    ASSERT_EQ(err_ref, err_test);
+  }
+}
+
+TEST_P(PixelProjHighbdErrorTest, RandomValues) { RunPixelProjErrorTest(1); }
+
+TEST_P(PixelProjHighbdErrorTest, ExtremeValues) {
+  RunPixelProjErrorTest_ExtremeValues();
+}
+
+TEST_P(PixelProjHighbdErrorTest, DISABLED_Speed) {
+  RunPixelProjErrorTest(200000);
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, PixelProjHighbdErrorTest,
+                        ::testing::Values(av1_highbd_pixel_proj_error_sse4_1));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+
+INSTANTIATE_TEST_CASE_P(AVX2, PixelProjHighbdErrorTest,
+                        ::testing::Values(av1_highbd_pixel_proj_error_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace pickrst_test_highbd
diff --git a/third_party/aom/test/qm_test.cc b/third_party/aom/test/qm_test.cc
index c87506b41a..d1dfbb849b 100644
--- a/third_party/aom/test/qm_test.cc
+++ b/third_party/aom/test/qm_test.cc
@@ -33,7 +33,7 @@ class QMTest
 
   virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
                                   ::libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
       encoder->Control(AV1E_SET_ENABLE_QM, 1);
       encoder->Control(AV1E_SET_QM_MIN, qm_min_);
diff --git a/third_party/aom/test/quantize_func_test.cc b/third_party/aom/test/quantize_func_test.cc
index 554d0c7214..2975c1dff5 100644
--- a/third_party/aom/test/quantize_func_test.cc
+++ b/third_party/aom/test/quantize_func_test.cc
@@ -63,7 +63,7 @@ void highbd_quan64x64_wrapper(QUAN_PARAM_LIST) {
   HBD_QUAN_FUNC;
 }
 
-typedef enum { TYPE_B, TYPE_DC, TYPE_FP } QuantType;
+enum { TYPE_B, TYPE_DC, TYPE_FP } UENUM1BYTE(QuantType);
 
 using ::testing::tuple;
 typedef tuple<QuantizeFunc, QuantizeFunc, TX_SIZE, QuantType, aom_bit_depth_t>
@@ -286,19 +286,29 @@ TEST_P(QuantizeTest, DISABLED_Speed) {
   const int16_t *quant_shift = qtab_->quant.y_quant_shift[q];
   const int16_t *dequant = qtab_->dequant.y_dequant_QTX[q];
   const int kNumTests = 5000000;
-  aom_usec_timer timer;
+  aom_usec_timer timer, simd_timer;
 
   FillCoeffRandom();
 
   aom_usec_timer_start(&timer);
   for (int n = 0; n < kNumTests; ++n) {
-    quant_(coeff_ptr, n_coeffs, zbin, round_fp, quant_fp, quant_shift, qcoeff,
-           dqcoeff, dequant, eob, sc->scan, sc->iscan);
+    quant_ref_(coeff_ptr, n_coeffs, zbin, round_fp, quant_fp, quant_shift,
+               qcoeff, dqcoeff, dequant, eob, sc->scan, sc->iscan);
   }
   aom_usec_timer_mark(&timer);
 
+  aom_usec_timer_start(&simd_timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    quant_(coeff_ptr, n_coeffs, zbin, round_fp, quant_fp, quant_shift, qcoeff,
+           dqcoeff, dequant, eob, sc->scan, sc->iscan);
+  }
+  aom_usec_timer_mark(&simd_timer);
+
   const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
-  printf("Elapsed time: %d us\n", elapsed_time);
+  const int simd_elapsed_time =
+      static_cast<int>(aom_usec_timer_elapsed(&simd_timer));
+  printf("c_time = %d \t simd_time = %d \t Gain = %d \n", elapsed_time,
+         simd_elapsed_time, (elapsed_time / simd_elapsed_time));
 }
 
 using ::testing::make_tuple;
@@ -398,14 +408,12 @@ INSTANTIATE_TEST_CASE_P(SSE2, QuantizeTest,
 INSTANTIATE_TEST_CASE_P(
     SSSE3, QuantizeTest,
     ::testing::Values(make_tuple(&aom_quantize_b_c, &aom_quantize_b_ssse3,
-                                 TX_16X16, TYPE_B, AOM_BITS_8)));
-
-// Like libvpx, the ssse3 and avx quantize tests do not pass.
-// https://bugs.chromium.org/p/webm/issues/detail?id=1448
-INSTANTIATE_TEST_CASE_P(
-    DISABLED_SSSE3_32x32, QuantizeTest,
-    ::testing::Values(make_tuple(&aom_quantize_b_32x32_c,
-                                 &aom_quantize_b_32x32_ssse3, TX_16X16, TYPE_B,
+                                 TX_16X16, TYPE_B, AOM_BITS_8),
+                      make_tuple(&aom_quantize_b_32x32_c,
+                                 &aom_quantize_b_32x32_ssse3, TX_32X32, TYPE_B,
+                                 AOM_BITS_8),
+                      make_tuple(&aom_quantize_b_64x64_c,
+                                 &aom_quantize_b_64x64_ssse3, TX_64X64, TYPE_B,
                                  AOM_BITS_8)));
 
 #endif  // HAVE_SSSE3 && ARCH_X86_64
@@ -413,13 +421,11 @@ INSTANTIATE_TEST_CASE_P(
 #if HAVE_AVX && ARCH_X86_64
 INSTANTIATE_TEST_CASE_P(
     AVX, QuantizeTest,
-    ::testing::Values(
-        make_tuple(&aom_quantize_b_c, &aom_quantize_b_avx, TX_16X16, TYPE_B,
-                   AOM_BITS_8),
-        // Although these tests will not pass against _c, test them against each
-        // other so there is some minor checking.
-        make_tuple(&aom_quantize_b_32x32_ssse3, &aom_quantize_b_32x32_avx,
-                   TX_32X32, TYPE_B, AOM_BITS_8)));
+    ::testing::Values(make_tuple(&aom_quantize_b_c, &aom_quantize_b_avx,
+                                 TX_16X16, TYPE_B, AOM_BITS_8),
+                      make_tuple(&aom_quantize_b_32x32_c,
+                                 &aom_quantize_b_32x32_avx, TX_32X32, TYPE_B,
+                                 AOM_BITS_8)));
 
 #endif  // HAVE_AVX && ARCH_X86_64
 }  // namespace
diff --git a/third_party/aom/test/resize_test.cc b/third_party/aom/test/resize_test.cc
index b270b8362f..34964f0a30 100644
--- a/third_party/aom/test/resize_test.cc
+++ b/third_party/aom/test/resize_test.cc
@@ -297,7 +297,7 @@ class ResizeInternalTestLarge : public ResizeTest {
 
   virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
     if (frame0_psnr_ == 0.) frame0_psnr_ = pkt->data.psnr.psnr[0];
-    EXPECT_NEAR(pkt->data.psnr.psnr[0], frame0_psnr_, 2.5);
+    EXPECT_NEAR(pkt->data.psnr.psnr[0], frame0_psnr_, 3.0);
   }
 
 #if WRITE_COMPRESSED_STREAM
diff --git a/third_party/aom/test/sad_test.cc b/third_party/aom/test/sad_test.cc
index 845fe79da1..87dbb33afe 100644
--- a/third_party/aom/test/sad_test.cc
+++ b/third_party/aom/test/sad_test.cc
@@ -35,22 +35,25 @@ typedef uint32_t (*SadMxNAvgFunc)(const uint8_t *src_ptr, int src_stride,
                                   const uint8_t *second_pred);
 typedef ::testing::tuple<int, int, SadMxNAvgFunc, int> SadMxNAvgParam;
 
-typedef void (*JntCompAvgFunc)(uint8_t *comp_pred, const uint8_t *pred,
-                               int width, int height, const uint8_t *ref,
-                               int ref_stride,
-                               const JNT_COMP_PARAMS *jcp_param);
-typedef ::testing::tuple<int, int, JntCompAvgFunc, int> JntCompAvgParam;
+typedef void (*DistWtdCompAvgFunc)(uint8_t *comp_pred, const uint8_t *pred,
+                                   int width, int height, const uint8_t *ref,
+                                   int ref_stride,
+                                   const DIST_WTD_COMP_PARAMS *jcp_param);
+typedef ::testing::tuple<int, int, DistWtdCompAvgFunc, int> DistWtdCompAvgParam;
 
-typedef unsigned int (*JntSadMxhFunc)(const uint8_t *src_ptr, int src_stride,
-                                      const uint8_t *ref_ptr, int ref_stride,
-                                      int width, int height);
-typedef ::testing::tuple<int, int, JntSadMxhFunc, int> JntSadMxhParam;
+typedef unsigned int (*DistWtdSadMxhFunc)(const uint8_t *src_ptr,
+                                          int src_stride,
+                                          const uint8_t *ref_ptr,
+                                          int ref_stride, int width,
+                                          int height);
+typedef ::testing::tuple<int, int, DistWtdSadMxhFunc, int> DistWtdSadMxhParam;
 
-typedef uint32_t (*JntSadMxNAvgFunc)(const uint8_t *src_ptr, int src_stride,
-                                     const uint8_t *ref_ptr, int ref_stride,
-                                     const uint8_t *second_pred,
-                                     const JNT_COMP_PARAMS *jcp_param);
-typedef ::testing::tuple<int, int, JntSadMxNAvgFunc, int> JntSadMxNAvgParam;
+typedef uint32_t (*DistWtdSadMxNAvgFunc)(const uint8_t *src_ptr, int src_stride,
+                                         const uint8_t *ref_ptr, int ref_stride,
+                                         const uint8_t *second_pred,
+                                         const DIST_WTD_COMP_PARAMS *jcp_param);
+typedef ::testing::tuple<int, int, DistWtdSadMxNAvgFunc, int>
+    DistWtdSadMxNAvgParam;
 
 typedef void (*SadMxNx4Func)(const uint8_t *src_ptr, int src_stride,
                              const uint8_t *const ref_ptr[], int ref_stride,
@@ -203,7 +206,7 @@ class SADTestBase : public ::testing::Test {
     return sad;
   }
 
-  void ReferenceJntCompAvg(int block_idx) {
+  void ReferenceDistWtdCompAvg(int block_idx) {
     const uint8_t *const reference8 = GetReference(block_idx);
     const uint8_t *const second_pred8 = second_pred_;
     uint8_t *const comp_pred8 = comp_pred_;
@@ -228,7 +231,7 @@ class SADTestBase : public ::testing::Test {
     }
   }
 
-  unsigned int ReferenceJntSADavg(int block_idx) {
+  unsigned int ReferenceDistWtdSADavg(int block_idx) {
     unsigned int sad = 0;
     const uint8_t *const reference8 = GetReference(block_idx);
     const uint8_t *const source8 = source_data_;
@@ -305,7 +308,7 @@ class SADTestBase : public ::testing::Test {
   static uint8_t *comp_pred_test_;
   static uint8_t *comp_pred8_test_;
   static uint16_t *comp_pred16_test_;
-  JNT_COMP_PARAMS jcp_param_;
+  DIST_WTD_COMP_PARAMS jcp_param_;
 
   ACMRandom rnd_;
 };
@@ -391,13 +394,15 @@ class SADavgTest : public ::testing::WithParamInterface<SadMxNAvgParam>,
   }
 };
 
-class JntCompAvgTest : public ::testing::WithParamInterface<JntCompAvgParam>,
-                       public SADTestBase {
+class DistWtdCompAvgTest
+    : public ::testing::WithParamInterface<DistWtdCompAvgParam>,
+      public SADTestBase {
  public:
-  JntCompAvgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+  DistWtdCompAvgTest()
+      : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
 
  protected:
-  void jnt_comp_avg(int block_idx) {
+  void dist_wtd_comp_avg(int block_idx) {
     const uint8_t *const reference = GetReference(block_idx);
 
     ASM_REGISTER_STATE_CHECK(GET_PARAM(2)(comp_pred_test_, second_pred_, width_,
@@ -411,8 +416,8 @@ class JntCompAvgTest : public ::testing::WithParamInterface<JntCompAvgParam>,
         jcp_param_.fwd_offset = quant_dist_lookup_table[j][i][0];
         jcp_param_.bck_offset = quant_dist_lookup_table[j][i][1];
 
-        ReferenceJntCompAvg(0);
-        jnt_comp_avg(0);
+        ReferenceDistWtdCompAvg(0);
+        dist_wtd_comp_avg(0);
 
         for (int y = 0; y < height_; ++y)
           for (int x = 0; x < width_; ++x)
@@ -423,10 +428,10 @@ class JntCompAvgTest : public ::testing::WithParamInterface<JntCompAvgParam>,
   }
 };
 
-class JntSADTest : public ::testing::WithParamInterface<JntSadMxhParam>,
-                   public SADTestBase {
+class DistWtdSADTest : public ::testing::WithParamInterface<DistWtdSadMxhParam>,
+                       public SADTestBase {
  public:
-  JntSADTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+  DistWtdSADTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
 
  protected:
   unsigned int SAD(int block_idx) {
@@ -455,13 +460,14 @@ class JntSADTest : public ::testing::WithParamInterface<JntSadMxhParam>,
   }
 };
 
-class JntSADavgTest : public ::testing::WithParamInterface<JntSadMxNAvgParam>,
-                      public SADTestBase {
+class DistWtdSADavgTest
+    : public ::testing::WithParamInterface<DistWtdSadMxNAvgParam>,
+      public SADTestBase {
  public:
-  JntSADavgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+  DistWtdSADavgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
 
  protected:
-  unsigned int jnt_SAD_avg(int block_idx) {
+  unsigned int dist_wtd_SAD_avg(int block_idx) {
     unsigned int ret;
     const uint8_t *const reference = GetReference(block_idx);
 
@@ -477,8 +483,8 @@ class JntSADavgTest : public ::testing::WithParamInterface<JntSadMxNAvgParam>,
         jcp_param_.fwd_offset = quant_dist_lookup_table[j][i][0];
         jcp_param_.bck_offset = quant_dist_lookup_table[j][i][1];
 
-        const unsigned int reference_sad = ReferenceJntSADavg(0);
-        const unsigned int exp_sad = jnt_SAD_avg(0);
+        const unsigned int reference_sad = ReferenceDistWtdSADavg(0);
+        const unsigned int exp_sad = dist_wtd_SAD_avg(0);
 
         ASSERT_EQ(reference_sad, exp_sad);
       }
@@ -608,19 +614,19 @@ TEST_P(SADavgTest, ShortSrc) {
   source_stride_ = tmp_stride;
 }
 
-TEST_P(JntCompAvgTest, MaxRef) {
+TEST_P(DistWtdCompAvgTest, MaxRef) {
   FillConstant(reference_data_, reference_stride_, mask_);
   FillConstant(second_pred_, width_, 0);
   CheckCompAvg();
 }
 
-TEST_P(JntCompAvgTest, MaxSecondPred) {
+TEST_P(DistWtdCompAvgTest, MaxSecondPred) {
   FillConstant(reference_data_, reference_stride_, 0);
   FillConstant(second_pred_, width_, mask_);
   CheckCompAvg();
 }
 
-TEST_P(JntCompAvgTest, ShortRef) {
+TEST_P(DistWtdCompAvgTest, ShortRef) {
   const int tmp_stride = reference_stride_;
   reference_stride_ >>= 1;
   FillRandom(reference_data_, reference_stride_);
@@ -629,7 +635,7 @@ TEST_P(JntCompAvgTest, ShortRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntCompAvgTest, UnalignedRef) {
+TEST_P(DistWtdCompAvgTest, UnalignedRef) {
   // The reference frame, but not the source frame, may be unaligned for
   // certain types of searches.
   const int tmp_stride = reference_stride_;
@@ -640,19 +646,19 @@ TEST_P(JntCompAvgTest, UnalignedRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADTest, MaxRef) {
+TEST_P(DistWtdSADTest, MaxRef) {
   FillConstant(source_data_, source_stride_, 0);
   FillConstant(reference_data_, reference_stride_, mask_);
   CheckSAD();
 }
 
-TEST_P(JntSADTest, MaxSrc) {
+TEST_P(DistWtdSADTest, MaxSrc) {
   FillConstant(source_data_, source_stride_, mask_);
   FillConstant(reference_data_, reference_stride_, 0);
   CheckSAD();
 }
 
-TEST_P(JntSADTest, ShortRef) {
+TEST_P(DistWtdSADTest, ShortRef) {
   const int tmp_stride = reference_stride_;
   reference_stride_ >>= 1;
   FillRandom(source_data_, source_stride_);
@@ -661,7 +667,7 @@ TEST_P(JntSADTest, ShortRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADTest, UnalignedRef) {
+TEST_P(DistWtdSADTest, UnalignedRef) {
   // The reference frame, but not the source frame, may be unaligned for
   // certain types of searches.
   const int tmp_stride = reference_stride_;
@@ -672,7 +678,7 @@ TEST_P(JntSADTest, UnalignedRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADTest, ShortSrc) {
+TEST_P(DistWtdSADTest, ShortSrc) {
   const int tmp_stride = source_stride_;
   source_stride_ >>= 1;
   int test_count = 2000;
@@ -685,20 +691,20 @@ TEST_P(JntSADTest, ShortSrc) {
   source_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADavgTest, MaxRef) {
+TEST_P(DistWtdSADavgTest, MaxRef) {
   FillConstant(source_data_, source_stride_, 0);
   FillConstant(reference_data_, reference_stride_, mask_);
   FillConstant(second_pred_, width_, 0);
   CheckSAD();
 }
-TEST_P(JntSADavgTest, MaxSrc) {
+TEST_P(DistWtdSADavgTest, MaxSrc) {
   FillConstant(source_data_, source_stride_, mask_);
   FillConstant(reference_data_, reference_stride_, 0);
   FillConstant(second_pred_, width_, 0);
   CheckSAD();
 }
 
-TEST_P(JntSADavgTest, ShortRef) {
+TEST_P(DistWtdSADavgTest, ShortRef) {
   const int tmp_stride = reference_stride_;
   reference_stride_ >>= 1;
   FillRandom(source_data_, source_stride_);
@@ -708,7 +714,7 @@ TEST_P(JntSADavgTest, ShortRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADavgTest, UnalignedRef) {
+TEST_P(DistWtdSADavgTest, UnalignedRef) {
   // The reference frame, but not the source frame, may be unaligned for
   // certain types of searches.
   const int tmp_stride = reference_stride_;
@@ -720,7 +726,7 @@ TEST_P(JntSADavgTest, UnalignedRef) {
   reference_stride_ = tmp_stride;
 }
 
-TEST_P(JntSADavgTest, ShortSrc) {
+TEST_P(DistWtdSADavgTest, ShortSrc) {
   const int tmp_stride = source_stride_;
   source_stride_ >>= 1;
   int test_count = 2000;
@@ -947,47 +953,48 @@ const SadMxNAvgParam avg_c_tests[] = {
 INSTANTIATE_TEST_CASE_P(C, SADavgTest, ::testing::ValuesIn(avg_c_tests));
 
 // TODO(chengchen): add highbd tests
-const JntCompAvgParam jnt_comp_avg_c_tests[] = {
-  make_tuple(128, 128, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(128, 64, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(64, 128, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(64, 64, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(64, 32, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(32, 64, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(32, 32, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(32, 16, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(16, 32, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(16, 16, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(16, 8, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(8, 16, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(8, 8, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(8, 4, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(4, 8, &aom_jnt_comp_avg_pred_c, -1),
-  make_tuple(4, 4, &aom_jnt_comp_avg_pred_c, -1),
+const DistWtdCompAvgParam dist_wtd_comp_avg_c_tests[] = {
+  make_tuple(128, 128, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(128, 64, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(64, 128, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(64, 64, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(64, 32, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(32, 64, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(32, 32, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(32, 16, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(16, 32, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(16, 16, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(16, 8, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(8, 16, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(8, 8, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(8, 4, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(4, 8, &aom_dist_wtd_comp_avg_pred_c, -1),
+  make_tuple(4, 4, &aom_dist_wtd_comp_avg_pred_c, -1),
 };
 
-INSTANTIATE_TEST_CASE_P(C, JntCompAvgTest,
-                        ::testing::ValuesIn(jnt_comp_avg_c_tests));
+INSTANTIATE_TEST_CASE_P(C, DistWtdCompAvgTest,
+                        ::testing::ValuesIn(dist_wtd_comp_avg_c_tests));
 
-const JntSadMxNAvgParam jnt_avg_c_tests[] = {
-  make_tuple(128, 128, &aom_jnt_sad128x128_avg_c, -1),
-  make_tuple(128, 64, &aom_jnt_sad128x64_avg_c, -1),
-  make_tuple(64, 128, &aom_jnt_sad64x128_avg_c, -1),
-  make_tuple(64, 64, &aom_jnt_sad64x64_avg_c, -1),
-  make_tuple(64, 32, &aom_jnt_sad64x32_avg_c, -1),
-  make_tuple(32, 64, &aom_jnt_sad32x64_avg_c, -1),
-  make_tuple(32, 32, &aom_jnt_sad32x32_avg_c, -1),
-  make_tuple(32, 16, &aom_jnt_sad32x16_avg_c, -1),
-  make_tuple(16, 32, &aom_jnt_sad16x32_avg_c, -1),
-  make_tuple(16, 16, &aom_jnt_sad16x16_avg_c, -1),
-  make_tuple(16, 8, &aom_jnt_sad16x8_avg_c, -1),
-  make_tuple(8, 16, &aom_jnt_sad8x16_avg_c, -1),
-  make_tuple(8, 8, &aom_jnt_sad8x8_avg_c, -1),
-  make_tuple(8, 4, &aom_jnt_sad8x4_avg_c, -1),
-  make_tuple(4, 8, &aom_jnt_sad4x8_avg_c, -1),
-  make_tuple(4, 4, &aom_jnt_sad4x4_avg_c, -1),
+const DistWtdSadMxNAvgParam dist_wtd_avg_c_tests[] = {
+  make_tuple(128, 128, &aom_dist_wtd_sad128x128_avg_c, -1),
+  make_tuple(128, 64, &aom_dist_wtd_sad128x64_avg_c, -1),
+  make_tuple(64, 128, &aom_dist_wtd_sad64x128_avg_c, -1),
+  make_tuple(64, 64, &aom_dist_wtd_sad64x64_avg_c, -1),
+  make_tuple(64, 32, &aom_dist_wtd_sad64x32_avg_c, -1),
+  make_tuple(32, 64, &aom_dist_wtd_sad32x64_avg_c, -1),
+  make_tuple(32, 32, &aom_dist_wtd_sad32x32_avg_c, -1),
+  make_tuple(32, 16, &aom_dist_wtd_sad32x16_avg_c, -1),
+  make_tuple(16, 32, &aom_dist_wtd_sad16x32_avg_c, -1),
+  make_tuple(16, 16, &aom_dist_wtd_sad16x16_avg_c, -1),
+  make_tuple(16, 8, &aom_dist_wtd_sad16x8_avg_c, -1),
+  make_tuple(8, 16, &aom_dist_wtd_sad8x16_avg_c, -1),
+  make_tuple(8, 8, &aom_dist_wtd_sad8x8_avg_c, -1),
+  make_tuple(8, 4, &aom_dist_wtd_sad8x4_avg_c, -1),
+  make_tuple(4, 8, &aom_dist_wtd_sad4x8_avg_c, -1),
+  make_tuple(4, 4, &aom_dist_wtd_sad4x4_avg_c, -1),
 };
-INSTANTIATE_TEST_CASE_P(C, JntSADavgTest, ::testing::ValuesIn(jnt_avg_c_tests));
+INSTANTIATE_TEST_CASE_P(C, DistWtdSADavgTest,
+                        ::testing::ValuesIn(dist_wtd_avg_c_tests));
 
 const SadMxNx4Param x4d_c_tests[] = {
   make_tuple(128, 128, &aom_sad128x128x4d_c, -1),
@@ -1251,7 +1258,7 @@ INSTANTIATE_TEST_CASE_P(SSE2, SADx4Test, ::testing::ValuesIn(x4d_sse2_tests));
 #if HAVE_SSSE3
 // Note: These are named sse2, but part of ssse3 file and only built and linked
 // when ssse3 is enabled.
-const JntSadMxhParam jnt_sad_sse2_tests[] = {
+const DistWtdSadMxhParam dist_wtd_sad_sse2_tests[] = {
   make_tuple(4, 4, &aom_sad4xh_sse2, -1),
   make_tuple(4, 8, &aom_sad4xh_sse2, -1),
   make_tuple(8, 4, &aom_sad8xh_sse2, -1),
@@ -1275,8 +1282,8 @@ const JntSadMxhParam jnt_sad_sse2_tests[] = {
   make_tuple(16, 64, &aom_sad16xh_sse2, -1),
   make_tuple(64, 16, &aom_sad64xh_sse2, -1),
 };
-INSTANTIATE_TEST_CASE_P(SSE2, JntSADTest,
-                        ::testing::ValuesIn(jnt_sad_sse2_tests));
+INSTANTIATE_TEST_CASE_P(SSE2, DistWtdSADTest,
+                        ::testing::ValuesIn(dist_wtd_sad_sse2_tests));
 
 #endif  // HAVE_SSSE3
 
@@ -1285,49 +1292,49 @@ INSTANTIATE_TEST_CASE_P(SSE2, JntSADTest,
 #endif  // HAVE_SSE3
 
 #if HAVE_SSSE3
-const JntCompAvgParam jnt_comp_avg_ssse3_tests[] = {
-  make_tuple(128, 128, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(128, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(64, 128, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(64, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(64, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(32, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(32, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(32, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(16, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(16, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(16, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(8, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(8, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(8, 4, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(4, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(4, 4, &aom_jnt_comp_avg_pred_ssse3, -1),
-  make_tuple(16, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
+const DistWtdCompAvgParam dist_wtd_comp_avg_ssse3_tests[] = {
+  make_tuple(128, 128, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(128, 64, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 128, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 64, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 32, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 64, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 32, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 16, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 32, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 16, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 8, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 16, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 8, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 4, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(4, 8, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(4, 4, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 16, &aom_dist_wtd_comp_avg_pred_ssse3, -1),
 };
 
-INSTANTIATE_TEST_CASE_P(SSSE3, JntCompAvgTest,
-                        ::testing::ValuesIn(jnt_comp_avg_ssse3_tests));
+INSTANTIATE_TEST_CASE_P(SSSE3, DistWtdCompAvgTest,
+                        ::testing::ValuesIn(dist_wtd_comp_avg_ssse3_tests));
 
-const JntSadMxNAvgParam jnt_avg_ssse3_tests[] = {
-  make_tuple(128, 128, &aom_jnt_sad128x128_avg_ssse3, -1),
-  make_tuple(128, 64, &aom_jnt_sad128x64_avg_ssse3, -1),
-  make_tuple(64, 128, &aom_jnt_sad64x128_avg_ssse3, -1),
-  make_tuple(64, 64, &aom_jnt_sad64x64_avg_ssse3, -1),
-  make_tuple(64, 32, &aom_jnt_sad64x32_avg_ssse3, -1),
-  make_tuple(32, 64, &aom_jnt_sad32x64_avg_ssse3, -1),
-  make_tuple(32, 32, &aom_jnt_sad32x32_avg_ssse3, -1),
-  make_tuple(32, 16, &aom_jnt_sad32x16_avg_ssse3, -1),
-  make_tuple(16, 32, &aom_jnt_sad16x32_avg_ssse3, -1),
-  make_tuple(16, 16, &aom_jnt_sad16x16_avg_ssse3, -1),
-  make_tuple(16, 8, &aom_jnt_sad16x8_avg_ssse3, -1),
-  make_tuple(8, 16, &aom_jnt_sad8x16_avg_ssse3, -1),
-  make_tuple(8, 8, &aom_jnt_sad8x8_avg_ssse3, -1),
-  make_tuple(8, 4, &aom_jnt_sad8x4_avg_ssse3, -1),
-  make_tuple(4, 8, &aom_jnt_sad4x8_avg_ssse3, -1),
-  make_tuple(4, 4, &aom_jnt_sad4x4_avg_ssse3, -1),
+const DistWtdSadMxNAvgParam dist_wtd_avg_ssse3_tests[] = {
+  make_tuple(128, 128, &aom_dist_wtd_sad128x128_avg_ssse3, -1),
+  make_tuple(128, 64, &aom_dist_wtd_sad128x64_avg_ssse3, -1),
+  make_tuple(64, 128, &aom_dist_wtd_sad64x128_avg_ssse3, -1),
+  make_tuple(64, 64, &aom_dist_wtd_sad64x64_avg_ssse3, -1),
+  make_tuple(64, 32, &aom_dist_wtd_sad64x32_avg_ssse3, -1),
+  make_tuple(32, 64, &aom_dist_wtd_sad32x64_avg_ssse3, -1),
+  make_tuple(32, 32, &aom_dist_wtd_sad32x32_avg_ssse3, -1),
+  make_tuple(32, 16, &aom_dist_wtd_sad32x16_avg_ssse3, -1),
+  make_tuple(16, 32, &aom_dist_wtd_sad16x32_avg_ssse3, -1),
+  make_tuple(16, 16, &aom_dist_wtd_sad16x16_avg_ssse3, -1),
+  make_tuple(16, 8, &aom_dist_wtd_sad16x8_avg_ssse3, -1),
+  make_tuple(8, 16, &aom_dist_wtd_sad8x16_avg_ssse3, -1),
+  make_tuple(8, 8, &aom_dist_wtd_sad8x8_avg_ssse3, -1),
+  make_tuple(8, 4, &aom_dist_wtd_sad8x4_avg_ssse3, -1),
+  make_tuple(4, 8, &aom_dist_wtd_sad4x8_avg_ssse3, -1),
+  make_tuple(4, 4, &aom_dist_wtd_sad4x4_avg_ssse3, -1),
 };
-INSTANTIATE_TEST_CASE_P(SSSE3, JntSADavgTest,
-                        ::testing::ValuesIn(jnt_avg_ssse3_tests));
+INSTANTIATE_TEST_CASE_P(SSSE3, DistWtdSADavgTest,
+                        ::testing::ValuesIn(dist_wtd_avg_ssse3_tests));
 #endif  // HAVE_SSSE3
 
 #if HAVE_SSE4_1
diff --git a/third_party/aom/test/selfguided_filter_test.cc b/third_party/aom/test/selfguided_filter_test.cc
index d2d5c6105d..464a58f3fd 100644
--- a/third_party/aom/test/selfguided_filter_test.cc
+++ b/third_party/aom/test/selfguided_filter_test.cc
@@ -26,9 +26,9 @@
 
 namespace {
 
+using libaom_test::ACMRandom;
 using ::testing::make_tuple;
 using ::testing::tuple;
-using libaom_test::ACMRandom;
 
 typedef void (*SgrFunc)(const uint8_t *dat8, int width, int height, int stride,
                         int eps, const int *xqd, uint8_t *dst8, int dst_stride,
diff --git a/third_party/aom/test/simd_cmp_impl.h b/third_party/aom/test/simd_cmp_impl.h
index b98af9aade..2aa02c864f 100644
--- a/third_party/aom/test/simd_cmp_impl.h
+++ b/third_party/aom/test/simd_cmp_impl.h
@@ -471,7 +471,7 @@ typedef struct {
 
 #define MAP(name)                                \
   {                                              \
-    #name, reinterpret_cast < fptr > (c_##name), \
+#name, reinterpret_cast < fptr > (c_##name), \
         reinterpret_cast < fptr > (name)         \
   }
 
diff --git a/third_party/aom/test/sum_squares_test.cc b/third_party/aom/test/sum_squares_test.cc
index f109984986..f26a64654d 100644
--- a/third_party/aom/test/sum_squares_test.cc
+++ b/third_party/aom/test/sum_squares_test.cc
@@ -27,6 +27,10 @@
 
 using libaom_test::ACMRandom;
 using libaom_test::FunctionEquivalenceTest;
+using ::testing::Combine;
+using ::testing::Range;
+using ::testing::Values;
+using ::testing::ValuesIn;
 
 namespace {
 const int kNumIterations = 10000;
@@ -225,4 +229,170 @@ INSTANTIATE_TEST_CASE_P(SSE2, SumSquares1DTest,
                             aom_sum_squares_i16_c, aom_sum_squares_i16_sse2)));
 
 #endif  // HAVE_SSE2
+
+typedef int64_t (*sse_func)(const uint8_t *a, int a_stride, const uint8_t *b,
+                            int b_stride, int width, int height);
+typedef libaom_test::FuncParam<sse_func> TestSSEFuncs;
+
+typedef ::testing::tuple<TestSSEFuncs, int> SSETestParam;
+
+class SSETest : public ::testing::TestWithParam<SSETestParam> {
+ public:
+  virtual ~SSETest() {}
+  virtual void SetUp() {
+    params_ = GET_PARAM(0);
+    width_ = GET_PARAM(1);
+    isHbd_ = params_.ref_func == aom_highbd_sse_c;
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    src_ = reinterpret_cast<uint8_t *>(aom_memalign(32, 256 * 256 * 2));
+    ref_ = reinterpret_cast<uint8_t *>(aom_memalign(32, 256 * 256 * 2));
+    ASSERT_TRUE(src_ != NULL);
+    ASSERT_TRUE(ref_ != NULL);
+  }
+
+  virtual void TearDown() {
+    libaom_test::ClearSystemState();
+    aom_free(src_);
+    aom_free(ref_);
+  }
+  void RunTest(int isRandom, int width, int height, int run_times);
+
+  void GenRandomData(int width, int height, int stride) {
+    uint16_t *pSrc = (uint16_t *)src_;
+    uint16_t *pRef = (uint16_t *)ref_;
+    const int msb = 11;  // Up to 12 bit input
+    const int limit = 1 << (msb + 1);
+    for (int ii = 0; ii < height; ii++) {
+      for (int jj = 0; jj < width; jj++) {
+        if (!isHbd_) {
+          src_[ii * stride + jj] = rnd_.Rand8();
+          ref_[ii * stride + jj] = rnd_.Rand8();
+        } else {
+          pSrc[ii * stride + jj] = rnd_(limit);
+          pRef[ii * stride + jj] = rnd_(limit);
+        }
+      }
+    }
+  }
+
+  void GenExtremeData(int width, int height, int stride, uint8_t *data,
+                      int16_t val) {
+    uint16_t *pData = (uint16_t *)data;
+    for (int ii = 0; ii < height; ii++) {
+      for (int jj = 0; jj < width; jj++) {
+        if (!isHbd_) {
+          data[ii * stride + jj] = (uint8_t)val;
+        } else {
+          pData[ii * stride + jj] = val;
+        }
+      }
+    }
+  }
+
+ protected:
+  int isHbd_;
+  int width_;
+  TestSSEFuncs params_;
+  uint8_t *src_;
+  uint8_t *ref_;
+  ACMRandom rnd_;
+};
+
+void SSETest::RunTest(int isRandom, int width, int height, int run_times) {
+  int failed = 0;
+  aom_usec_timer ref_timer, test_timer;
+  for (int k = 0; k < 3; k++) {
+    int stride = 4 << rnd_(7);  // Up to 256 stride
+    while (stride < width) {    // Make sure it's valid
+      stride = 4 << rnd_(7);
+    }
+    if (isRandom) {
+      GenRandomData(width, height, stride);
+    } else {
+      const int msb = isHbd_ ? 12 : 8;  // Up to 12 bit input
+      const int limit = (1 << msb) - 1;
+      if (k == 0) {
+        GenExtremeData(width, height, stride, src_, 0);
+        GenExtremeData(width, height, stride, ref_, limit);
+      } else {
+        GenExtremeData(width, height, stride, src_, limit);
+        GenExtremeData(width, height, stride, ref_, 0);
+      }
+    }
+    int64_t res_ref, res_tst;
+    uint8_t *pSrc = src_;
+    uint8_t *pRef = ref_;
+    if (isHbd_) {
+      pSrc = CONVERT_TO_BYTEPTR(src_);
+      pRef = CONVERT_TO_BYTEPTR(ref_);
+    }
+    res_ref = params_.ref_func(pSrc, stride, pRef, stride, width, height);
+    res_tst = params_.tst_func(pSrc, stride, pRef, stride, width, height);
+    if (run_times > 1) {
+      aom_usec_timer_start(&ref_timer);
+      for (int j = 0; j < run_times; j++) {
+        params_.ref_func(pSrc, stride, pRef, stride, width, height);
+      }
+      aom_usec_timer_mark(&ref_timer);
+      const int elapsed_time_c =
+          static_cast<int>(aom_usec_timer_elapsed(&ref_timer));
+
+      aom_usec_timer_start(&test_timer);
+      for (int j = 0; j < run_times; j++) {
+        params_.tst_func(pSrc, stride, pRef, stride, width, height);
+      }
+      aom_usec_timer_mark(&test_timer);
+      const int elapsed_time_simd =
+          static_cast<int>(aom_usec_timer_elapsed(&test_timer));
+
+      printf(
+          "c_time=%d \t simd_time=%d \t "
+          "gain=%d\n",
+          elapsed_time_c, elapsed_time_simd,
+          (elapsed_time_c / elapsed_time_simd));
+    } else {
+      if (!failed) {
+        failed = res_ref != res_tst;
+        EXPECT_EQ(res_ref, res_tst)
+            << "Error:" << (isHbd_ ? "hbd " : " ") << k << " SSE Test ["
+            << width << "x" << height
+            << "] C output does not match optimized output.";
+      }
+    }
+  }
+}
+
+TEST_P(SSETest, OperationCheck) {
+  for (int height = 4; height <= 128; height += 4) {
+    RunTest(1, width_, height, 1);  // GenRandomData
+  }
+}
+
+TEST_P(SSETest, ExtremeValues) {
+  for (int height = 4; height <= 128; height += 4) {
+    RunTest(0, width_, height, 1);
+  }
+}
+
+TEST_P(SSETest, DISABLED_Speed) {
+  for (int height = 4; height <= 128; height += 4) {
+    RunTest(1, width_, height, 100);
+  }
+}
+#if HAVE_SSE4_1
+TestSSEFuncs sse_sse4[] = { TestSSEFuncs(&aom_sse_c, &aom_sse_sse4_1),
+                            TestSSEFuncs(&aom_highbd_sse_c,
+                                         &aom_highbd_sse_sse4_1) };
+INSTANTIATE_TEST_CASE_P(SSE4_1, SSETest,
+                        Combine(ValuesIn(sse_sse4), Range(4, 129, 4)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+
+TestSSEFuncs sse_avx2[] = { TestSSEFuncs(&aom_sse_c, &aom_sse_avx2),
+                            TestSSEFuncs(&aom_highbd_sse_c,
+                                         &aom_highbd_sse_avx2) };
+INSTANTIATE_TEST_CASE_P(AVX2, SSETest,
+                        Combine(ValuesIn(sse_avx2), Range(4, 129, 4)));
+#endif  // HAVE_AVX2
 }  // namespace
diff --git a/third_party/aom/test/superframe_test.cc b/third_party/aom/test/superframe_test.cc
index 7be18f72a3..2cec95aa52 100644
--- a/third_party/aom/test/superframe_test.cc
+++ b/third_party/aom/test/superframe_test.cc
@@ -45,7 +45,7 @@ class SuperframeTest
 
   virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
                                   libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
       encoder->Control(AOME_SET_CPUUSED, 2);
       encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
diff --git a/third_party/aom/test/test-data.sha1 b/third_party/aom/test/test-data.sha1
index b6ee34701a..95342a8c38 100644
--- a/third_party/aom/test/test-data.sha1
+++ b/third_party/aom/test/test-data.sha1
@@ -2,12 +2,32 @@ d5dfb0151c9051f8c85999255645d7a23916d3c0 *hantro_collage_w352h288.yuv
 b87815bf86020c592ccc7a846ba2e28ec8043902 *hantro_odd.yuv
 26b7f64399b84db4b4c9c915d743ec5c2619d4b9 *invalid-bug-1814.ivf
 d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-bug-1814.ivf.res
+09aa07e5325b3bb5462182eb30b8ecc914630740 *invalid-chromium-906381.ivf
+09d2af8dd22201dd8d48e5dcfcaed281ff9422c7 *invalid-chromium-906381.ivf.res
 fa06784f23751d8c37be94160fb821e855199af4 *invalid-oss-fuzz-10061.ivf
 b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-10061.ivf.res
 c9e06c4c7fb7d69fd635a1f606a5e478d60e99cf *invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf
 88e18e61bd2b7457b4c71ebefbdff0029c41cc04 *invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf.res
 91a5bedeb4832c1c2900736cc0f644bb63971bbc *invalid-oss-fuzz-10227.ivf
 b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-10227.ivf.res
+b2d0a29a65879436bf483d04865faca7d11cc2ee *invalid-oss-fuzz-10389.ivf
+9655e6275888547ecd1f14e20e08ce4891372e76 *invalid-oss-fuzz-10389.ivf.res
+11df8e9a068669c678097d460b63609d3da73828 *invalid-oss-fuzz-10555.ivf
+b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-10555.ivf.res
+cf5945085fe85456a1f74bf4cc7998b88b3f4b62 *invalid-oss-fuzz-10705.ivf
+758671858368ffd2a2c0727898de5661f7cf7d68 *invalid-oss-fuzz-10705.ivf.res
+88e29851122cca3f336824f7fa4d9f757f91110c *invalid-oss-fuzz-10723.ivf
+1af486cd2cc83ebeddc76ca7a1c512cc0ec568d5 *invalid-oss-fuzz-10723.ivf.res
+0784acc8931090ec24eba752d6c27e359e68fe7d *invalid-oss-fuzz-10779.ivf
+5d9474c0309b7ca09a182d888f73b37a8fe1362c *invalid-oss-fuzz-10779.ivf.res
+7d37be9357f89a100ced694aee1ca5a6fad35ba9 *invalid-oss-fuzz-11477.ivf
+15932651aacfc4622f0910f728f3f95e08e1753d *invalid-oss-fuzz-11477.ivf.res
+1674787c38ddf82a2e5c804203f04f56a304e8e0 *invalid-oss-fuzz-11479.ivf
+1af486cd2cc83ebeddc76ca7a1c512cc0ec568d5 *invalid-oss-fuzz-11479.ivf.res
+b1a45514f0c59be03c9991cd04882426b9b930fa *invalid-oss-fuzz-11523.ivf
+7c44ac1723c14d98bcb888fbf118c959511519ba *invalid-oss-fuzz-11523.ivf.res
+ccbe4081557eb44820a0e6337c4a094421826b9a *invalid-oss-fuzz-9288.ivf
+67c54283fe1a26ccf02cc991e4f9a1eea3ac5e78 *invalid-oss-fuzz-9288.ivf.res
 c0960f032484579f967881cc025b71cfd7a79ee1 *invalid-oss-fuzz-9463.ivf
 d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-oss-fuzz-9463.ivf.res
 f448caf378e250b7eea4fa2d1c3cd7ef4a3211ce *invalid-oss-fuzz-9482.ivf
@@ -505,3 +525,10 @@ e69e41fee40b408b6eebcc79f266a95f2ee24f9e *av1-1-b8-03-sizedown.mkv
 8de81b170635d456602dc8923a8b39c534d01fa8 *av1-1-b8-03-sizeup.mkv.md5
 d3ed7de0aa8c155fe35e0f5f4203240710d31383 *park_joy_90p_8_420_monochrome.y4m
 5b3f0907407b809aa66b62cb080feda8c92454ca *park_joy_90p_8_420_vertical_csp.y4m
+caf8b6a5f1a5bcb38afae8a54a08c4f4459aafa3 *vase10x10_tiles.txt
+e14825f50ff845b8a6932c64cb254007a0b5e3a1 *av1-1-b8-22-svc-L2T1.ivf
+0f75f2ac44e61fc83be70c955410fa378e433237 *av1-1-b8-22-svc-L2T1.ivf.md5
+e94687eb0e90179b3800b6d5e11eb7e9bfb34eec *av1-1-b8-22-svc-L1T2.ivf
+2bc12b16385ea14323bc79607fb8dfbd7edaf8ef *av1-1-b8-22-svc-L1T2.ivf.md5
+32ef2f14ee9cb11a24a22934f4c065e926e5d236 *av1-1-b8-22-svc-L2T2.ivf
+f476a10ff06d750129f8229755d51e17ff141b2a *av1-1-b8-22-svc-L2T2.ivf.md5
diff --git a/third_party/aom/test/test.cmake b/third_party/aom/test/test.cmake
index b16ae14c3d..1b11373c76 100644
--- a/third_party/aom/test/test.cmake
+++ b/third_party/aom/test/test.cmake
@@ -170,6 +170,7 @@ if(NOT BUILD_SHARED_LIBS)
               "${AOM_ROOT}/test/av1_fwd_txfm2d_test.cc"
               "${AOM_ROOT}/test/av1_inv_txfm1d_test.cc"
               "${AOM_ROOT}/test/av1_inv_txfm2d_test.cc"
+              "${AOM_ROOT}/test/av1_nn_predict_test.cc"
               "${AOM_ROOT}/test/av1_round_shift_array_test.cc"
               "${AOM_ROOT}/test/av1_txfm_test.cc"
               "${AOM_ROOT}/test/av1_txfm_test.h"
@@ -179,10 +180,12 @@ if(NOT BUILD_SHARED_LIBS)
               "${AOM_ROOT}/test/comp_avg_pred_test.cc"
               "${AOM_ROOT}/test/comp_avg_pred_test.h"
               "${AOM_ROOT}/test/comp_mask_variance_test.cc"
+              "${AOM_ROOT}/test/edge_detect_test.cc"
               "${AOM_ROOT}/test/encodetxb_test.cc"
               "${AOM_ROOT}/test/error_block_test.cc"
               "${AOM_ROOT}/test/fft_test.cc"
               "${AOM_ROOT}/test/fwht4x4_test.cc"
+              "${AOM_ROOT}/test/horver_correlation_test.cc"
               "${AOM_ROOT}/test/masked_sad_test.cc"
               "${AOM_ROOT}/test/masked_variance_test.cc"
               "${AOM_ROOT}/test/motion_vector_test.cc"
@@ -230,13 +233,6 @@ if(ENABLE_TESTS)
               "make sure it's in your PATH.")
   endif()
 
-  if(MSVC) # Force static run time to avoid collisions with googletest.
-    include("${AOM_ROOT}/build/cmake/msvc_runtime.cmake")
-    if(BUILD_SHARED_LIBS)
-      set(AOM_DISABLE_GTEST_CMAKE 1)
-    endif()
-  endif()
-
   if(BUILD_SHARED_LIBS AND APPLE) # Silence an RPATH warning.
     set(CMAKE_MACOSX_RPATH 1)
   endif()
@@ -244,15 +240,16 @@ if(ENABLE_TESTS)
   include_directories(
     "${AOM_ROOT}/third_party/googletest/src/googletest/include")
 
-  if(AOM_DISABLE_GTEST_CMAKE)
-    include_directories("${AOM_ROOT}/third_party/googletest/src/googletest")
-    add_library(
-      gtest
-      STATIC
-      "${AOM_ROOT}/third_party/googletest/src/googletest/src/gtest-all.cc")
+  include_directories("${AOM_ROOT}/third_party/googletest/src/googletest")
+  add_library(
+    aom_gtest
+    STATIC "${AOM_ROOT}/third_party/googletest/src/googletest/src/gtest-all.cc")
+  if(MSVC OR WIN32)
+    target_compile_definitions(aom_gtest PRIVATE GTEST_OS_WINDOWS=1)
+  elseif(CONFIG_MULTITHREAD AND CMAKE_USE_PTHREADS_INIT)
+    target_compile_definitions(aom_gtest PRIVATE GTEST_HAS_PTHREAD=1)
   else()
-    add_subdirectory("${AOM_ROOT}/third_party/googletest/src/googletest"
-                     EXCLUDE_FROM_ALL)
+    target_compile_definitions(aom_gtest PRIVATE GTEST_HAS_PTHREAD=0)
   endif()
 endif()
 
@@ -304,12 +301,12 @@ function(setup_aom_test_targets)
       add_executable(test_intra_pred_speed ${AOM_TEST_INTRA_PRED_SPEED_SOURCES}
                      $<TARGET_OBJECTS:aom_common_app_util>)
       target_link_libraries(test_intra_pred_speed ${AOM_LIB_LINK_TYPE} aom
-                            gtest)
+                            aom_gtest)
       list(APPEND AOM_APP_TARGETS test_intra_pred_speed)
     endif()
   endif()
 
-  target_link_libraries(test_libaom ${AOM_LIB_LINK_TYPE} aom gtest)
+  target_link_libraries(test_libaom ${AOM_LIB_LINK_TYPE} aom aom_gtest)
 
   if(CONFIG_LIBYUV)
     target_sources(test_libaom PRIVATE $<TARGET_OBJECTS:yuv>)
diff --git a/third_party/aom/test/test_data_util.cmake b/third_party/aom/test/test_data_util.cmake
index 45c951478a..6d684cbc27 100644
--- a/third_party/aom/test/test_data_util.cmake
+++ b/third_party/aom/test/test_data_util.cmake
@@ -12,20 +12,6 @@
 list(APPEND AOM_TEST_DATA_FILE_NAMES
             "hantro_collage_w352h288.yuv"
             "hantro_odd.yuv"
-            "invalid-bug-1814.ivf"
-            "invalid-bug-1814.ivf.res"
-            "invalid-oss-fuzz-10061.ivf"
-            "invalid-oss-fuzz-10061.ivf.res"
-            "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf"
-            "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf.res"
-            "invalid-oss-fuzz-10227.ivf"
-            "invalid-oss-fuzz-10227.ivf.res"
-            "invalid-oss-fuzz-9463.ivf"
-            "invalid-oss-fuzz-9463.ivf.res"
-            "invalid-oss-fuzz-9482.ivf"
-            "invalid-oss-fuzz-9482.ivf.res"
-            "invalid-oss-fuzz-9720.ivf"
-            "invalid-oss-fuzz-9720.ivf.res"
             "park_joy_90p_10_420.y4m"
             "park_joy_90p_10_422.y4m"
             "park_joy_90p_10_444.y4m"
@@ -43,7 +29,8 @@ list(APPEND AOM_TEST_DATA_FILE_NAMES
             "rush_hour_444.y4m"
             "screendata.y4m"
             "niklas_640_480_30.yuv"
-            "vase10x10.yuv")
+            "vase10x10.yuv"
+            "vase10x10_tiles.txt")
 
 if(ENABLE_DECODE_PERF_TESTS AND CONFIG_AV1_ENCODER)
   list(APPEND AOM_TEST_DATA_FILE_NAMES "niklas_1280_720_30.yuv")
@@ -512,7 +499,47 @@ if(CONFIG_AV1_DECODER)
               "av1-1-b8-03-sizeup.mkv"
               "av1-1-b8-03-sizeup.mkv.md5"
               "av1-1-b8-03-sizedown.mkv"
-              "av1-1-b8-03-sizedown.mkv.md5")
+              "av1-1-b8-03-sizedown.mkv.md5"
+              "av1-1-b8-22-svc-L2T1.ivf"
+              "av1-1-b8-22-svc-L2T1.ivf.md5"
+              "av1-1-b8-22-svc-L1T2.ivf"
+              "av1-1-b8-22-svc-L1T2.ivf.md5"
+              "av1-1-b8-22-svc-L2T2.ivf"
+              "av1-1-b8-22-svc-L2T2.ivf.md5"
+              "invalid-bug-1814.ivf"
+              "invalid-bug-1814.ivf.res"
+              "invalid-chromium-906381.ivf"
+              "invalid-chromium-906381.ivf.res"
+              "invalid-oss-fuzz-10061.ivf"
+              "invalid-oss-fuzz-10061.ivf.res"
+              "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf"
+              "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf.res"
+              "invalid-oss-fuzz-10227.ivf"
+              "invalid-oss-fuzz-10227.ivf.res"
+              "invalid-oss-fuzz-10389.ivf"
+              "invalid-oss-fuzz-10389.ivf.res"
+              "invalid-oss-fuzz-10555.ivf"
+              "invalid-oss-fuzz-10555.ivf.res"
+              "invalid-oss-fuzz-10705.ivf"
+              "invalid-oss-fuzz-10705.ivf.res"
+              "invalid-oss-fuzz-10723.ivf"
+              "invalid-oss-fuzz-10723.ivf.res"
+              "invalid-oss-fuzz-10779.ivf"
+              "invalid-oss-fuzz-10779.ivf.res"
+              "invalid-oss-fuzz-11477.ivf"
+              "invalid-oss-fuzz-11477.ivf.res"
+              "invalid-oss-fuzz-11479.ivf"
+              "invalid-oss-fuzz-11479.ivf.res"
+              "invalid-oss-fuzz-11523.ivf"
+              "invalid-oss-fuzz-11523.ivf.res"
+              "invalid-oss-fuzz-9288.ivf"
+              "invalid-oss-fuzz-9288.ivf.res"
+              "invalid-oss-fuzz-9463.ivf"
+              "invalid-oss-fuzz-9463.ivf.res"
+              "invalid-oss-fuzz-9482.ivf"
+              "invalid-oss-fuzz-9482.ivf.res"
+              "invalid-oss-fuzz-9720.ivf"
+              "invalid-oss-fuzz-9720.ivf.res")
 endif()
 
 if(ENABLE_ENCODE_PERF_TESTS AND CONFIG_AV1_ENCODER)
diff --git a/third_party/aom/test/test_vector_test.cc b/third_party/aom/test/test_vector_test.cc
index 286988b170..9f070b9b24 100644
--- a/third_party/aom/test/test_vector_test.cc
+++ b/third_party/aom/test/test_vector_test.cc
@@ -11,6 +11,7 @@
 
 #include <cstdio>
 #include <cstdlib>
+#include <memory>
 #include <set>
 #include <string>
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
@@ -116,7 +117,7 @@ TEST_P(TestVectorTest, MD5Match) {
   SCOPED_TRACE(str);
 
   // Open compressed video file.
-  testing::internal::scoped_ptr<libaom_test::CompressedVideoSource> video;
+  std::unique_ptr<libaom_test::CompressedVideoSource> video;
   if (filename.substr(filename.length() - 3, 3) == "ivf") {
     video.reset(new libaom_test::IVFVideoSource(filename));
   } else if (filename.substr(filename.length() - 4, 4) == "webm" ||
diff --git a/third_party/aom/test/test_vectors.cc b/third_party/aom/test/test_vectors.cc
index 71e431e18b..d2f333f74b 100644
--- a/third_party/aom/test/test_vectors.cc
+++ b/third_party/aom/test/test_vectors.cc
@@ -132,7 +132,8 @@ const char *const kAV1TestVectors[] = {
   "av1-1-b8-01-size-66x32.ivf",    "av1-1-b8-01-size-66x34.ivf",
   "av1-1-b8-01-size-66x64.ivf",    "av1-1-b8-01-size-66x66.ivf",
   "av1-1-b8-02-allintra.ivf",      "av1-1-b8-03-sizedown.mkv",
-  "av1-1-b8-03-sizeup.mkv"
+  "av1-1-b8-03-sizeup.mkv",        "av1-1-b8-22-svc-L1T2.ivf",
+  "av1-1-b8-22-svc-L2T1.ivf",      "av1-1-b8-22-svc-L2T2.ivf"
 };
 const int kNumAV1TestVectors = NELEMENTS(kAV1TestVectors);
 #endif  // CONFIG_AV1_DECODER
diff --git a/third_party/aom/test/tile_independence_test.cc b/third_party/aom/test/tile_independence_test.cc
index cf534c0c59..4f7c4a475e 100644
--- a/third_party/aom/test/tile_independence_test.cc
+++ b/third_party/aom/test/tile_independence_test.cc
@@ -59,7 +59,7 @@ class TileIndependenceTest
 
   virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
                                   libaom_test::Encoder *encoder) {
-    if (video->frame() == 1) {
+    if (video->frame() == 0) {
       encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
       encoder->Control(AV1E_SET_TILE_ROWS, n_tile_rows_);
       SetCpuUsed(encoder);
diff --git a/third_party/aom/test/variance_test.cc b/third_party/aom/test/variance_test.cc
index 0df314b0fd..d57fef99a1 100644
--- a/third_party/aom/test/variance_test.cc
+++ b/third_party/aom/test/variance_test.cc
@@ -43,10 +43,10 @@ typedef unsigned int (*SubpixAvgVarMxNFunc)(const uint8_t *a, int a_stride,
 typedef unsigned int (*Get4x4SseFunc)(const uint8_t *a, int a_stride,
                                       const uint8_t *b, int b_stride);
 typedef unsigned int (*SumOfSquaresFunction)(const int16_t *src);
-typedef unsigned int (*JntSubpixAvgVarMxNFunc)(
+typedef unsigned int (*DistWtdSubpixAvgVarMxNFunc)(
     const uint8_t *a, int a_stride, int xoffset, int yoffset, const uint8_t *b,
     int b_stride, uint32_t *sse, const uint8_t *second_pred,
-    const JNT_COMP_PARAMS *jcp_param);
+    const DIST_WTD_COMP_PARAMS *jcp_param);
 typedef uint32_t (*ObmcSubpelVarFunc)(const uint8_t *pre, int pre_stride,
                                       int xoffset, int yoffset,
                                       const int32_t *wsrc, const int32_t *mask,
@@ -216,10 +216,10 @@ static uint32_t subpel_avg_variance_ref(const uint8_t *ref, const uint8_t *src,
   return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
 }
 
-static uint32_t jnt_subpel_avg_variance_ref(
+static uint32_t dist_wtd_subpel_avg_variance_ref(
     const uint8_t *ref, const uint8_t *src, const uint8_t *second_pred, int l2w,
     int l2h, int xoff, int yoff, uint32_t *sse_ptr, bool use_high_bit_depth,
-    aom_bit_depth_t bit_depth, JNT_COMP_PARAMS *jcp_param) {
+    aom_bit_depth_t bit_depth, DIST_WTD_COMP_PARAMS *jcp_param) {
   int64_t se = 0;
   uint64_t sse = 0;
   const int w = 1 << l2w;
@@ -709,7 +709,7 @@ class SubpelVarianceTest
   uint8_t *ref_;
   uint8_t *sec_;
   TestParams<FunctionType> params_;
-  JNT_COMP_PARAMS jcp_param_;
+  DIST_WTD_COMP_PARAMS jcp_param_;
 
   // some relay helpers
   bool use_high_bit_depth() const { return params_.use_high_bit_depth; }
@@ -820,7 +820,7 @@ void SubpelVarianceTest<SubpixAvgVarMxNFunc>::RefTest() {
 }
 
 template <>
-void SubpelVarianceTest<JntSubpixAvgVarMxNFunc>::RefTest() {
+void SubpelVarianceTest<DistWtdSubpixAvgVarMxNFunc>::RefTest() {
   for (int x = 0; x < 8; ++x) {
     for (int y = 0; y < 8; ++y) {
       if (!use_high_bit_depth()) {
@@ -849,7 +849,7 @@ void SubpelVarianceTest<JntSubpixAvgVarMxNFunc>::RefTest() {
           ASM_REGISTER_STATE_CHECK(var1 = params_.func(ref_, width() + 0, x, y,
                                                        src_, width(), &sse1,
                                                        sec_, &jcp_param_));
-          var2 = jnt_subpel_avg_variance_ref(
+          var2 = dist_wtd_subpel_avg_variance_ref(
               ref_, src_, sec_, params_.log2width, params_.log2height, x, y,
               &sse2, use_high_bit_depth(), params_.bit_depth, &jcp_param_);
           EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
@@ -1022,7 +1022,8 @@ typedef MainTestClass<VarianceMxNFunc> AvxMseTest;
 typedef MainTestClass<VarianceMxNFunc> AvxVarianceTest;
 typedef SubpelVarianceTest<SubpixVarMxNFunc> AvxSubpelVarianceTest;
 typedef SubpelVarianceTest<SubpixAvgVarMxNFunc> AvxSubpelAvgVarianceTest;
-typedef SubpelVarianceTest<JntSubpixAvgVarMxNFunc> AvxJntSubpelAvgVarianceTest;
+typedef SubpelVarianceTest<DistWtdSubpixAvgVarMxNFunc>
+    AvxDistWtdSubpelAvgVarianceTest;
 typedef ObmcVarianceTest<ObmcSubpelVarFunc> AvxObmcSubpelVarianceTest;
 
 TEST_P(AvxSseTest, RefSse) { RefTestSse(); }
@@ -1039,7 +1040,7 @@ TEST_P(SumOfSquaresTest, Ref) { RefTest(); }
 TEST_P(AvxSubpelVarianceTest, Ref) { RefTest(); }
 TEST_P(AvxSubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
 TEST_P(AvxSubpelAvgVarianceTest, Ref) { RefTest(); }
-TEST_P(AvxJntSubpelAvgVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxDistWtdSubpelAvgVarianceTest, Ref) { RefTest(); }
 TEST_P(AvxObmcSubpelVarianceTest, Ref) { RefTest(); }
 TEST_P(AvxObmcSubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
 TEST_P(AvxObmcSubpelVarianceTest, DISABLED_Speed) { SpeedTest(); }
@@ -1121,36 +1122,35 @@ INSTANTIATE_TEST_CASE_P(
         SubpelAvgVarianceParams(2, 3, &aom_sub_pixel_avg_variance4x8_c, 0),
         SubpelAvgVarianceParams(2, 2, &aom_sub_pixel_avg_variance4x4_c, 0)));
 
-typedef TestParams<JntSubpixAvgVarMxNFunc> JntSubpelAvgVarianceParams;
+typedef TestParams<DistWtdSubpixAvgVarMxNFunc> DistWtdSubpelAvgVarianceParams;
 INSTANTIATE_TEST_CASE_P(
-    C, AvxJntSubpelAvgVarianceTest,
-    ::testing::Values(
-        JntSubpelAvgVarianceParams(6, 6, &aom_jnt_sub_pixel_avg_variance64x64_c,
-                                   0),
-        JntSubpelAvgVarianceParams(6, 5, &aom_jnt_sub_pixel_avg_variance64x32_c,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 6, &aom_jnt_sub_pixel_avg_variance32x64_c,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 5, &aom_jnt_sub_pixel_avg_variance32x32_c,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 4, &aom_jnt_sub_pixel_avg_variance32x16_c,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 5, &aom_jnt_sub_pixel_avg_variance16x32_c,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 4, &aom_jnt_sub_pixel_avg_variance16x16_c,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 3, &aom_jnt_sub_pixel_avg_variance16x8_c,
-                                   0),
-        JntSubpelAvgVarianceParams(3, 4, &aom_jnt_sub_pixel_avg_variance8x16_c,
-                                   0),
-        JntSubpelAvgVarianceParams(3, 3, &aom_jnt_sub_pixel_avg_variance8x8_c,
-                                   0),
-        JntSubpelAvgVarianceParams(3, 2, &aom_jnt_sub_pixel_avg_variance8x4_c,
-                                   0),
-        JntSubpelAvgVarianceParams(2, 3, &aom_jnt_sub_pixel_avg_variance4x8_c,
-                                   0),
-        JntSubpelAvgVarianceParams(2, 2, &aom_jnt_sub_pixel_avg_variance4x4_c,
-                                   0)));
+    C, AvxDistWtdSubpelAvgVarianceTest,
+    ::testing::Values(DistWtdSubpelAvgVarianceParams(
+                          6, 6, &aom_dist_wtd_sub_pixel_avg_variance64x64_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          6, 5, &aom_dist_wtd_sub_pixel_avg_variance64x32_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          5, 6, &aom_dist_wtd_sub_pixel_avg_variance32x64_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          5, 5, &aom_dist_wtd_sub_pixel_avg_variance32x32_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          5, 4, &aom_dist_wtd_sub_pixel_avg_variance32x16_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          4, 5, &aom_dist_wtd_sub_pixel_avg_variance16x32_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          4, 4, &aom_dist_wtd_sub_pixel_avg_variance16x16_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          4, 3, &aom_dist_wtd_sub_pixel_avg_variance16x8_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          3, 4, &aom_dist_wtd_sub_pixel_avg_variance8x16_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          3, 3, &aom_dist_wtd_sub_pixel_avg_variance8x8_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          3, 2, &aom_dist_wtd_sub_pixel_avg_variance8x4_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          2, 3, &aom_dist_wtd_sub_pixel_avg_variance4x8_c, 0),
+                      DistWtdSubpelAvgVarianceParams(
+                          2, 2, &aom_dist_wtd_sub_pixel_avg_variance4x4_c, 0)));
 
 INSTANTIATE_TEST_CASE_P(
     C, AvxObmcSubpelVarianceTest,
@@ -1840,44 +1840,34 @@ INSTANTIATE_TEST_CASE_P(
                                 0)));
 
 INSTANTIATE_TEST_CASE_P(
-    SSSE3, AvxJntSubpelAvgVarianceTest,
+    SSSE3, AvxDistWtdSubpelAvgVarianceTest,
     ::testing::Values(
-        JntSubpelAvgVarianceParams(6, 6,
-                                   &aom_jnt_sub_pixel_avg_variance64x64_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(6, 5,
-                                   &aom_jnt_sub_pixel_avg_variance64x32_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 6,
-                                   &aom_jnt_sub_pixel_avg_variance32x64_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 5,
-                                   &aom_jnt_sub_pixel_avg_variance32x32_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(5, 4,
-                                   &aom_jnt_sub_pixel_avg_variance32x16_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 5,
-                                   &aom_jnt_sub_pixel_avg_variance16x32_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 4,
-                                   &aom_jnt_sub_pixel_avg_variance16x16_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(4, 3,
-                                   &aom_jnt_sub_pixel_avg_variance16x8_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(3, 4,
-                                   &aom_jnt_sub_pixel_avg_variance8x16_ssse3,
-                                   0),
-        JntSubpelAvgVarianceParams(3, 3,
-                                   &aom_jnt_sub_pixel_avg_variance8x8_ssse3, 0),
-        JntSubpelAvgVarianceParams(3, 2,
-                                   &aom_jnt_sub_pixel_avg_variance8x4_ssse3, 0),
-        JntSubpelAvgVarianceParams(2, 3,
-                                   &aom_jnt_sub_pixel_avg_variance4x8_ssse3, 0),
-        JntSubpelAvgVarianceParams(2, 2,
-                                   &aom_jnt_sub_pixel_avg_variance4x4_ssse3,
-                                   0)));
+        DistWtdSubpelAvgVarianceParams(
+            6, 6, &aom_dist_wtd_sub_pixel_avg_variance64x64_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            6, 5, &aom_dist_wtd_sub_pixel_avg_variance64x32_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            5, 6, &aom_dist_wtd_sub_pixel_avg_variance32x64_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            5, 5, &aom_dist_wtd_sub_pixel_avg_variance32x32_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            5, 4, &aom_dist_wtd_sub_pixel_avg_variance32x16_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            4, 5, &aom_dist_wtd_sub_pixel_avg_variance16x32_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            4, 4, &aom_dist_wtd_sub_pixel_avg_variance16x16_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            4, 3, &aom_dist_wtd_sub_pixel_avg_variance16x8_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            3, 4, &aom_dist_wtd_sub_pixel_avg_variance8x16_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            3, 3, &aom_dist_wtd_sub_pixel_avg_variance8x8_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            3, 2, &aom_dist_wtd_sub_pixel_avg_variance8x4_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            2, 3, &aom_dist_wtd_sub_pixel_avg_variance4x8_ssse3, 0),
+        DistWtdSubpelAvgVarianceParams(
+            2, 2, &aom_dist_wtd_sub_pixel_avg_variance4x4_ssse3, 0)));
 #endif  // HAVE_SSSE3
 
 #if HAVE_SSE4_1
diff --git a/third_party/aom/test/warp_filter_test.cc b/third_party/aom/test/warp_filter_test.cc
index 19a4e8b6a1..d7b3ec9fb2 100644
--- a/third_party/aom/test/warp_filter_test.cc
+++ b/third_party/aom/test/warp_filter_test.cc
@@ -10,11 +10,11 @@
  */
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 #include "test/warp_filter_test_util.h"
-using ::testing::make_tuple;
-using ::testing::tuple;
 using libaom_test::ACMRandom;
 using libaom_test::AV1HighbdWarpFilter::AV1HighbdWarpFilterTest;
 using libaom_test::AV1WarpFilter::AV1WarpFilterTest;
+using ::testing::make_tuple;
+using ::testing::tuple;
 
 namespace {
 
diff --git a/third_party/aom/test/warp_filter_test_util.cc b/third_party/aom/test/warp_filter_test_util.cc
index 69b2ed4afe..9208af84e3 100644
--- a/third_party/aom/test/warp_filter_test_util.cc
+++ b/third_party/aom/test/warp_filter_test_util.cc
@@ -149,7 +149,7 @@ void AV1WarpFilterTest::RunSpeedTest(warp_affine_func test_impl) {
   int do_average = 0;
 
   conv_params = get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
 
   const int num_loops = 1000000000 / (out_w + out_h);
   aom_usec_timer timer;
@@ -222,9 +222,9 @@ void AV1WarpFilterTest::RunCheckOutput(warp_affine_func test_impl) {
                 conv_params = get_conv_params(0, 0, bd);
               }
               if (jj >= 4) {
-                conv_params.use_jnt_comp_avg = 0;
+                conv_params.use_dist_wtd_comp_avg = 0;
               } else {
-                conv_params.use_jnt_comp_avg = 1;
+                conv_params.use_dist_wtd_comp_avg = 1;
                 conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
                 conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
               }
@@ -236,9 +236,9 @@ void AV1WarpFilterTest::RunCheckOutput(warp_affine_func test_impl) {
                     get_conv_params_no_round(do_average, 0, dstb, out_w, 1, bd);
               }
               if (jj >= 4) {
-                conv_params.use_jnt_comp_avg = 0;
+                conv_params.use_dist_wtd_comp_avg = 0;
               } else {
-                conv_params.use_jnt_comp_avg = 1;
+                conv_params.use_dist_wtd_comp_avg = 1;
                 conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
                 conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
               }
@@ -342,7 +342,7 @@ void AV1HighbdWarpFilterTest::RunSpeedTest(highbd_warp_affine_func test_impl) {
   sub_x = 0;
   sub_y = 0;
   int do_average = 0;
-  conv_params.use_jnt_comp_avg = 0;
+  conv_params.use_dist_wtd_comp_avg = 0;
   conv_params = get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
 
   const int num_loops = 1000000000 / (out_w + out_h);
@@ -419,9 +419,9 @@ void AV1HighbdWarpFilterTest::RunCheckOutput(
                 conv_params = get_conv_params(0, 0, bd);
               }
               if (jj >= 4) {
-                conv_params.use_jnt_comp_avg = 0;
+                conv_params.use_dist_wtd_comp_avg = 0;
               } else {
-                conv_params.use_jnt_comp_avg = 1;
+                conv_params.use_dist_wtd_comp_avg = 1;
                 conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
                 conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
               }
@@ -436,9 +436,9 @@ void AV1HighbdWarpFilterTest::RunCheckOutput(
                     get_conv_params_no_round(do_average, 0, dstb, out_w, 1, bd);
               }
               if (jj >= 4) {
-                conv_params.use_jnt_comp_avg = 0;
+                conv_params.use_dist_wtd_comp_avg = 0;
               } else {
-                conv_params.use_jnt_comp_avg = 1;
+                conv_params.use_dist_wtd_comp_avg = 1;
                 conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
                 conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
               }
diff --git a/third_party/aom/test/wiener_test.cc b/third_party/aom/test/wiener_test.cc
index dfec09119d..8f49af6213 100644
--- a/third_party/aom/test/wiener_test.cc
+++ b/third_party/aom/test/wiener_test.cc
@@ -13,32 +13,33 @@
 
 #include "third_party/googletest/src/googletest/include/gtest/gtest.h"
 
-#include "test/function_equivalence_test.h"
 #include "test/register_state_check.h"
+#include "test/acm_random.h"
+#include "test/util.h"
 
 #include "config/aom_config.h"
 #include "config/aom_dsp_rtcd.h"
 
 #include "aom/aom_integer.h"
+#include "aom_ports/aom_timer.h"
 #include "av1/encoder/pickrst.h"
 
 #define MAX_WIENER_BLOCK 384
 #define MAX_DATA_BLOCK (MAX_WIENER_BLOCK + WIENER_WIN)
-using libaom_test::FunctionEquivalenceTest;
 
-namespace {
+// 8-bit-depth tests
+namespace wiener_lowbd {
 
 static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd,
                                     const uint8_t *src, int h_start, int h_end,
                                     int v_start, int v_end, int dgd_stride,
-                                    int src_stride, double *M, double *H) {
+                                    int src_stride, int64_t *M, int64_t *H) {
   ASSERT_TRUE(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA);
   int i, j, k, l, m, n;
   const int pixel_count = (h_end - h_start) * (v_end - v_start);
   const int wiener_win2 = wiener_win * wiener_win;
   const int wiener_halfwin = (wiener_win >> 1);
-  const double avg =
-      find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+  uint8_t avg = find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
 
   std::vector<std::vector<int64_t> > M_int(wiener_win,
                                            std::vector<int64_t>(wiener_win, 0));
@@ -75,16 +76,16 @@ static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd,
     }
   }
 
-  const double avg_square_sum = avg * avg * pixel_count;
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
   for (k = 0; k < wiener_win; k++) {
     for (l = 0; l < wiener_win; l++) {
       M[l * wiener_win + k] =
-          M_int[l][k] + avg_square_sum - avg * (sumX + sumY[k][l]);
+          M_int[l][k] + avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]);
       for (m = 0; m < wiener_win; m++) {
         for (n = 0; n < wiener_win; n++) {
           H[(l * wiener_win + k) * wiener_win2 + m * wiener_win + n] =
               H_int[(l * wiener_win + k)][n * 8 + m] + avg_square_sum -
-              avg * (sumY[k][l] + sumY[n][m]);
+              (int64_t)avg * (sumY[k][l] + sumY[n][m]);
         }
       }
     }
@@ -93,7 +94,8 @@ static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd,
 
 void compute_stats_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src,
                          int h_start, int h_end, int v_start, int v_end,
-                         int dgd_stride, int src_stride, double *M, double *H) {
+                         int dgd_stride, int src_stride, int64_t *M,
+                         int64_t *H) {
   if (wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA) {
     compute_stats_win_opt_c(wiener_win, dgd, src, h_start, h_end, v_start,
                             v_end, dgd_stride, src_stride, M, H);
@@ -104,13 +106,10 @@ void compute_stats_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src,
 }
 
 static const int kIterations = 100;
-static const double min_error = (double)(0.01);
 typedef void (*compute_stats_Func)(int wiener_win, const uint8_t *dgd,
                                    const uint8_t *src, int h_start, int h_end,
                                    int v_start, int v_end, int dgd_stride,
-                                   int src_stride, double *M, double *H);
-
-typedef libaom_test::FuncParam<compute_stats_Func> TestFuncs;
+                                   int src_stride, int64_t *M, int64_t *H);
 
 ////////////////////////////////////////////////////////////////////////////////
 // 8 bit
@@ -120,24 +119,34 @@ typedef ::testing::tuple<const compute_stats_Func> WienerTestParam;
 
 class WienerTest : public ::testing::TestWithParam<WienerTestParam> {
  public:
-  virtual void SetUp() { target_func_ = GET_PARAM(0); }
-  void runWienerTest(const int32_t wiener_win, int32_t run_times);
-  void runWienerTest_ExtremeValues(const int32_t wiener_win);
+  virtual void SetUp() {
+    src_buf = (uint8_t *)aom_memalign(
+        32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*src_buf));
+    dgd_buf = (uint8_t *)aom_memalign(
+        32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*dgd_buf));
+    target_func_ = GET_PARAM(0);
+  }
+  virtual void TearDown() {
+    aom_free(src_buf);
+    aom_free(dgd_buf);
+  }
+  void RunWienerTest(const int32_t wiener_win, int32_t run_times);
+  void RunWienerTest_ExtremeValues(const int32_t wiener_win);
 
  private:
   compute_stats_Func target_func_;
-  ACMRandom rng_;
+  libaom_test::ACMRandom rng_;
+  uint8_t *src_buf;
+  uint8_t *dgd_buf;
 };
 
-void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) {
+void WienerTest::RunWienerTest(const int32_t wiener_win, int32_t run_times) {
   const int32_t wiener_halfwin = wiener_win >> 1;
   const int32_t wiener_win2 = wiener_win * wiener_win;
-  DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
-  DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
-  DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]);
   const int h_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7));
   int h_end =
       run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8;
@@ -177,19 +186,18 @@ void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) {
     }
     int failed = 0;
     for (int i = 0; i < wiener_win2; ++i) {
-      if (fabs(M_ref[i] - M_test[i]) > min_error) {
+      if (M_ref[i] != M_test[i]) {
         failed = 1;
-        printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
-               iter, i, M_ref[i], M_test[i]);
+        printf("win %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n",
+               wiener_win, iter, i, M_ref[i], M_test[i]);
         break;
       }
     }
-    // ASSERT_EQ(failed, 0);
     for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
-      if (fabs(H_ref[i] - H_test[i]) > min_error) {
+      if (H_ref[i] != H_test[i]) {
         failed = 1;
-        printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
-               iter, i, H_ref[i], H_test[i]);
+        printf("win %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n",
+               wiener_win, iter, i, H_ref[i], H_test[i]);
         break;
       }
     }
@@ -197,15 +205,13 @@ void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) {
   }
 }
 
-void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) {
+void WienerTest::RunWienerTest_ExtremeValues(const int32_t wiener_win) {
   const int32_t wiener_halfwin = wiener_win >> 1;
   const int32_t wiener_win2 = wiener_win * wiener_win;
-  DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
-  DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
-  DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]);
-  DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]);
   const int h_start = 16;
   const int h_end = MAX_WIENER_BLOCK;
   const int v_start = 16;
@@ -229,19 +235,18 @@ void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) {
 
     int failed = 0;
     for (int i = 0; i < wiener_win2; ++i) {
-      if (fabs(M_ref[i] - M_test[i]) > min_error) {
+      if (M_ref[i] != M_test[i]) {
         failed = 1;
-        printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
-               iter, i, M_ref[i], M_test[i]);
+        printf("win %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n",
+               wiener_win, iter, i, M_ref[i], M_test[i]);
         break;
       }
     }
-    // ASSERT_EQ(failed, 0);
     for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
-      if (fabs(H_ref[i] - H_test[i]) > min_error) {
+      if (H_ref[i] != H_test[i]) {
         failed = 1;
-        printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
-               iter, i, H_ref[i], H_test[i]);
+        printf("win %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n",
+               wiener_win, iter, i, H_ref[i], H_test[i]);
         break;
       }
     }
@@ -250,18 +255,18 @@ void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) {
 }
 
 TEST_P(WienerTest, RandomValues) {
-  runWienerTest(WIENER_WIN, 1);
-  runWienerTest(WIENER_WIN_CHROMA, 1);
+  RunWienerTest(WIENER_WIN, 1);
+  RunWienerTest(WIENER_WIN_CHROMA, 1);
 }
 
 TEST_P(WienerTest, ExtremeValues) {
-  runWienerTest_ExtremeValues(WIENER_WIN);
-  runWienerTest_ExtremeValues(WIENER_WIN_CHROMA);
+  RunWienerTest_ExtremeValues(WIENER_WIN);
+  RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA);
 }
 
 TEST_P(WienerTest, DISABLED_Speed) {
-  runWienerTest(WIENER_WIN, 200);
-  runWienerTest(WIENER_WIN_CHROMA, 200);
+  RunWienerTest(WIENER_WIN, 200);
+  RunWienerTest(WIENER_WIN_CHROMA, 200);
 }
 
 INSTANTIATE_TEST_CASE_P(C, WienerTest, ::testing::Values(compute_stats_opt_c));
@@ -277,4 +282,303 @@ INSTANTIATE_TEST_CASE_P(AVX2, WienerTest,
                         ::testing::Values(av1_compute_stats_avx2));
 #endif  // HAVE_AVX2
 
-}  // namespace
+}  // namespace wiener_lowbd
+
+// High bit-depth tests:
+namespace wiener_highbd {
+
+static void compute_stats_highbd_win_opt_c(int wiener_win, const uint8_t *dgd8,
+                                           const uint8_t *src8, int h_start,
+                                           int h_end, int v_start, int v_end,
+                                           int dgd_stride, int src_stride,
+                                           int64_t *M, int64_t *H,
+                                           aom_bit_depth_t bit_depth) {
+  ASSERT_TRUE(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA);
+  int i, j, k, l, m, n;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
+  const uint16_t avg =
+      find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  std::vector<std::vector<int64_t> > M_int(wiener_win,
+                                           std::vector<int64_t>(wiener_win, 0));
+  std::vector<std::vector<int64_t> > H_int(
+      wiener_win * wiener_win, std::vector<int64_t>(wiener_win * 8, 0));
+  std::vector<std::vector<int32_t> > sumY(wiener_win,
+                                          std::vector<int32_t>(wiener_win, 0));
+
+  memset(M, 0, sizeof(*M) * wiener_win2);
+  memset(H, 0, sizeof(*H) * wiener_win2 * wiener_win2);
+
+  int64_t sumX = 0;
+  const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  for (i = v_start; i < v_end; i++) {
+    for (j = h_start; j < h_end; j += 2) {
+      const uint16_t X1 = src[i * src_stride + j];
+      const uint16_t X2 = src[i * src_stride + j + 1];
+      sumX += X1 + X2;
+
+      const uint16_t *dgd_ij = dgd_win + i * dgd_stride + j;
+      for (k = 0; k < wiener_win; k++) {
+        for (l = 0; l < wiener_win; l++) {
+          const uint16_t *dgd_ijkl = dgd_ij + k * dgd_stride + l;
+          int64_t *H_int_temp = &H_int[(l * wiener_win + k)][0];
+          const uint16_t D1 = dgd_ijkl[0];
+          const uint16_t D2 = dgd_ijkl[1];
+          sumY[k][l] += D1 + D2;
+          M_int[l][k] += D1 * X1 + D2 * X2;
+          for (m = 0; m < wiener_win; m++) {
+            for (n = 0; n < wiener_win; n++) {
+              H_int_temp[m * 8 + n] += D1 * dgd_ij[n + dgd_stride * m] +
+                                       D2 * dgd_ij[n + dgd_stride * m + 1];
+            }
+          }
+        }
+      }
+    }
+  }
+
+  uint8_t bit_depth_divider = 1;
+  if (bit_depth == AOM_BITS_12)
+    bit_depth_divider = 16;
+  else if (bit_depth == AOM_BITS_10)
+    bit_depth_divider = 4;
+
+  const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      M[l * wiener_win + k] =
+          (M_int[l][k] +
+           (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) /
+          bit_depth_divider;
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H[(l * wiener_win + k) * wiener_win2 + m * wiener_win + n] =
+              (H_int[(l * wiener_win + k)][n * 8 + m] +
+               (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) /
+              bit_depth_divider;
+        }
+      }
+    }
+  }
+}
+
+void compute_stats_highbd_opt_c(int wiener_win, const uint8_t *dgd,
+                                const uint8_t *src, int h_start, int h_end,
+                                int v_start, int v_end, int dgd_stride,
+                                int src_stride, int64_t *M, int64_t *H,
+                                aom_bit_depth_t bit_depth) {
+  if (wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA) {
+    compute_stats_highbd_win_opt_c(wiener_win, dgd, src, h_start, h_end,
+                                   v_start, v_end, dgd_stride, src_stride, M, H,
+                                   bit_depth);
+  } else {
+    av1_compute_stats_highbd_c(wiener_win, dgd, src, h_start, h_end, v_start,
+                               v_end, dgd_stride, src_stride, M, H, bit_depth);
+  }
+}
+
+static const int kIterations = 100;
+typedef void (*compute_stats_Func)(int wiener_win, const uint8_t *dgd,
+                                   const uint8_t *src, int h_start, int h_end,
+                                   int v_start, int v_end, int dgd_stride,
+                                   int src_stride, int64_t *M, int64_t *H,
+                                   aom_bit_depth_t bit_depth);
+
+typedef ::testing::tuple<const compute_stats_Func> WienerTestParam;
+
+class WienerTestHighbd : public ::testing::TestWithParam<WienerTestParam> {
+ public:
+  virtual void SetUp() {
+    src_buf = (uint16_t *)aom_memalign(
+        32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*src_buf));
+    dgd_buf = (uint16_t *)aom_memalign(
+        32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*dgd_buf));
+    target_func_ = GET_PARAM(0);
+  }
+  virtual void TearDown() {
+    aom_free(src_buf);
+    aom_free(dgd_buf);
+  }
+  void RunWienerTest(const int32_t wiener_win, int32_t run_times,
+                     aom_bit_depth_t bit_depth);
+  void RunWienerTest_ExtremeValues(const int32_t wiener_win,
+                                   aom_bit_depth_t bit_depth);
+
+ private:
+  compute_stats_Func target_func_;
+  libaom_test::ACMRandom rng_;
+  uint16_t *src_buf;
+  uint16_t *dgd_buf;
+};
+
+void WienerTestHighbd::RunWienerTest(const int32_t wiener_win,
+                                     int32_t run_times,
+                                     aom_bit_depth_t bit_depth) {
+  const int32_t wiener_halfwin = wiener_win >> 1;
+  const int32_t wiener_win2 = wiener_win * wiener_win;
+  DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  const int h_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7));
+  const int h_end =
+      run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8;
+  const int v_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7));
+  const int v_end =
+      run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8;
+  const int dgd_stride = h_end;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int iters = run_times == 1 ? kIterations : 2;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_buf[i] = rng_.Rand16() % (1 << bit_depth);
+      src_buf[i] = rng_.Rand16() % (1 << bit_depth);
+    }
+    const uint8_t *dgd8 = CONVERT_TO_BYTEPTR(
+        dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin);
+    const uint8_t *src8 = CONVERT_TO_BYTEPTR(src_buf);
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end,
+                                 v_start, v_end, dgd_stride, src_stride, M_ref,
+                                 H_ref, bit_depth);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      target_func_(wiener_win, dgd8, src8, h_start, h_end, v_start, v_end,
+                   dgd_stride, src_stride, M_test, H_test, bit_depth);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 10) {
+      printf("win %d bd %d %3dx%-3d:%7.2f/%7.2fns", wiener_win, bit_depth,
+             h_end, v_end, time1, time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+    int failed = 0;
+    for (int i = 0; i < wiener_win2; ++i) {
+      if (M_ref[i] != M_test[i]) {
+        failed = 1;
+        printf("win %d bd %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64
+               " \n",
+               wiener_win, bit_depth, iter, i, M_ref[i], M_test[i]);
+        break;
+      }
+    }
+    for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
+      if (H_ref[i] != H_test[i]) {
+        failed = 1;
+        printf("win %d bd %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64
+               " \n",
+               wiener_win, bit_depth, iter, i, H_ref[i], H_test[i]);
+        break;
+      }
+    }
+    ASSERT_EQ(failed, 0);
+  }
+}
+
+void WienerTestHighbd::RunWienerTest_ExtremeValues(const int32_t wiener_win,
+                                                   aom_bit_depth_t bit_depth) {
+  const int32_t wiener_halfwin = wiener_win >> 1;
+  const int32_t wiener_win2 = wiener_win * wiener_win;
+  DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  const int h_start = 16;
+  const int h_end = MAX_WIENER_BLOCK;
+  const int v_start = 16;
+  const int v_end = MAX_WIENER_BLOCK;
+  const int dgd_stride = h_end;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int iters = 1;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_buf[i] = ((uint16_t)1 << bit_depth) - 1;
+      src_buf[i] = ((uint16_t)1 << bit_depth) - 1;
+    }
+    const uint8_t *dgd8 = CONVERT_TO_BYTEPTR(
+        dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin);
+    const uint8_t *src8 = CONVERT_TO_BYTEPTR(src_buf);
+
+    av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end, v_start,
+                               v_end, dgd_stride, src_stride, M_ref, H_ref,
+                               bit_depth);
+
+    target_func_(wiener_win, dgd8, src8, h_start, h_end, v_start, v_end,
+                 dgd_stride, src_stride, M_test, H_test, bit_depth);
+
+    int failed = 0;
+    for (int i = 0; i < wiener_win2; ++i) {
+      if (M_ref[i] != M_test[i]) {
+        failed = 1;
+        printf("win %d bd %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64
+               " \n",
+               wiener_win, bit_depth, iter, i, M_ref[i], M_test[i]);
+        break;
+      }
+    }
+    for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
+      if (H_ref[i] != H_test[i]) {
+        failed = 1;
+        printf("win %d bd %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64
+               " \n",
+               wiener_win, bit_depth, iter, i, H_ref[i], H_test[i]);
+        break;
+      }
+    }
+    ASSERT_EQ(failed, 0);
+  }
+}
+
+TEST_P(WienerTestHighbd, RandomValues) {
+  RunWienerTest(WIENER_WIN, 1, AOM_BITS_8);
+  RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_8);
+  RunWienerTest(WIENER_WIN, 1, AOM_BITS_10);
+  RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_10);
+  RunWienerTest(WIENER_WIN, 1, AOM_BITS_12);
+  RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_12);
+}
+
+TEST_P(WienerTestHighbd, ExtremeValues) {
+  RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_8);
+  RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_8);
+  RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_10);
+  RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_10);
+  RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_12);
+  RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_12);
+}
+
+TEST_P(WienerTestHighbd, DISABLED_Speed) {
+  RunWienerTest(WIENER_WIN, 200, AOM_BITS_8);
+  RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_8);
+  RunWienerTest(WIENER_WIN, 200, AOM_BITS_10);
+  RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_10);
+  RunWienerTest(WIENER_WIN, 200, AOM_BITS_12);
+  RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_12);
+}
+
+INSTANTIATE_TEST_CASE_P(C, WienerTestHighbd,
+                        ::testing::Values(compute_stats_highbd_opt_c));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, WienerTestHighbd,
+                        ::testing::Values(av1_compute_stats_highbd_sse4_1));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, WienerTestHighbd,
+                        ::testing::Values(av1_compute_stats_highbd_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace wiener_highbd
diff --git a/third_party/aom/test/y4m_video_source.h b/third_party/aom/test/y4m_video_source.h
index 3dea901e6d..63f74f567f 100644
--- a/third_party/aom/test/y4m_video_source.h
+++ b/third_party/aom/test/y4m_video_source.h
@@ -11,6 +11,7 @@
 #ifndef AOM_TEST_Y4M_VIDEO_SOURCE_H_
 #define AOM_TEST_Y4M_VIDEO_SOURCE_H_
 #include <algorithm>
+#include <memory>
 #include <string>
 
 #include "common/y4minput.h"
@@ -41,7 +42,8 @@ class Y4mVideoSource : public VideoSource {
 
   virtual void ReadSourceToStart() {
     ASSERT_TRUE(input_file_ != NULL);
-    ASSERT_FALSE(y4m_input_open(&y4m_, input_file_, NULL, 0, 0));
+    ASSERT_FALSE(
+        y4m_input_open(&y4m_, input_file_, NULL, 0, AOM_CSP_UNKNOWN, 0));
     framerate_numerator_ = y4m_.fps_n;
     framerate_denominator_ = y4m_.fps_d;
     frame_ = 0;
@@ -109,7 +111,7 @@ class Y4mVideoSource : public VideoSource {
 
   std::string file_name_;
   FILE *input_file_;
-  testing::internal::scoped_ptr<aom_image_t> img_;
+  std::unique_ptr<aom_image_t> img_;
   unsigned int start_;
   unsigned int limit_;
   unsigned int frame_;
diff --git a/third_party/aom/third_party/libwebm/README.libaom b/third_party/aom/third_party/libwebm/README.libaom
index bd288d201d..17b2f47152 100644
--- a/third_party/aom/third_party/libwebm/README.libaom
+++ b/third_party/aom/third_party/libwebm/README.libaom
@@ -1,5 +1,5 @@
 URL: https://chromium.googlesource.com/webm/libwebm
-Version: af81f26025b7435fa9a14ad07c58b44cf9280430
+Version: 9f23fbc50e7a76c815b1d3f0309abe1066301331
 License: BSD
 License File: LICENSE.txt
 
@@ -7,8 +7,6 @@ Description:
 libwebm is used to handle WebM container I/O.
 
 Local Changes:
-Add av1 codec as an eligible codec for webm:
- https://aomedia-review.googlesource.com/c/aom/+/15103
 Only keep:
  - Android.mk
  - AUTHORS.TXT
diff --git a/third_party/aom/third_party/libwebm/common/file_util.cc b/third_party/aom/third_party/libwebm/common/file_util.cc
index 618ffc087f..e6109d5a0e 100644
--- a/third_party/aom/third_party/libwebm/common/file_util.cc
+++ b/third_party/aom/third_party/libwebm/common/file_util.cc
@@ -46,7 +46,7 @@ std::string GetTempFileName() {
   errno_t err = tmpnam_s(tmp_file_name);
 #else
   char* fname_pointer = tmpnam(tmp_file_name);
-  errno_t err = (fname_pointer == &tmp_file_name[0]) ? 0 : -1;
+  int err = (fname_pointer == &tmp_file_name[0]) ? 0 : -1;
 #endif
   if (err == 0) {
     return std::string(tmp_file_name);
diff --git a/third_party/aom/third_party/libwebm/common/webmids.h b/third_party/aom/third_party/libwebm/common/webmids.h
index 89d722a71b..fc0c208140 100644
--- a/third_party/aom/third_party/libwebm/common/webmids.h
+++ b/third_party/aom/third_party/libwebm/common/webmids.h
@@ -93,6 +93,7 @@ enum MkvId {
   kMkvDisplayHeight = 0x54BA,
   kMkvDisplayUnit = 0x54B2,
   kMkvAspectRatioType = 0x54B3,
+  kMkvColourSpace = 0x2EB524,
   kMkvFrameRate = 0x2383E3,
   // end video
   // colour
diff --git a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.cc b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.cc
index bae2c99b87..5120312119 100644
--- a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.cc
+++ b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.cc
@@ -773,6 +773,14 @@ bool Track::Write(IMkvWriter* writer) const {
   if (!type_ || !codec_id_)
     return false;
 
+  // AV1 tracks require a CodecPrivate. See
+  // https://github.com/Matroska-Org/matroska-specification/blob/av1-mappin/codec/av1.md
+  // TODO(tomfinegan): Update the above link to the AV1 Matroska mappings to
+  // point to a stable version once it is finalized, or our own WebM mappings
+  // page on webmproject.org should we decide to release them.
+  if (!strcmp(codec_id_, Tracks::kAv1CodecId) && !codec_private_)
+    return false;
+
   // |size| may be bigger than what is written out in this function because
   // derived classes may write out more data in the Track element.
   const uint64_t payload_size = PayloadSize();
@@ -1027,19 +1035,16 @@ bool MasteringMetadata::Write(IMkvWriter* writer) const {
       !WriteEbmlElement(writer, libwebm::kMkvLuminanceMin, luminance_min_)) {
     return false;
   }
-  if (r_ &&
-      !r_->Write(writer, libwebm::kMkvPrimaryRChromaticityX,
-                 libwebm::kMkvPrimaryRChromaticityY)) {
+  if (r_ && !r_->Write(writer, libwebm::kMkvPrimaryRChromaticityX,
+                       libwebm::kMkvPrimaryRChromaticityY)) {
     return false;
   }
-  if (g_ &&
-      !g_->Write(writer, libwebm::kMkvPrimaryGChromaticityX,
-                 libwebm::kMkvPrimaryGChromaticityY)) {
+  if (g_ && !g_->Write(writer, libwebm::kMkvPrimaryGChromaticityX,
+                       libwebm::kMkvPrimaryGChromaticityY)) {
     return false;
   }
-  if (b_ &&
-      !b_->Write(writer, libwebm::kMkvPrimaryBChromaticityX,
-                 libwebm::kMkvPrimaryBChromaticityY)) {
+  if (b_ && !b_->Write(writer, libwebm::kMkvPrimaryBChromaticityX,
+                       libwebm::kMkvPrimaryBChromaticityY)) {
     return false;
   }
   if (white_point_ &&
@@ -1421,6 +1426,7 @@ VideoTrack::VideoTrack(unsigned int* seed)
       stereo_mode_(0),
       alpha_mode_(0),
       width_(0),
+      colour_space_(NULL),
       colour_(NULL),
       projection_(NULL) {}
 
@@ -1518,6 +1524,10 @@ bool VideoTrack::Write(IMkvWriter* writer) const {
                           static_cast<uint64>(alpha_mode_)))
       return false;
   }
+  if (colour_space_) {
+    if (!WriteEbmlElement(writer, libwebm::kMkvColourSpace, colour_space_))
+      return false;
+  }
   if (frame_rate_ > 0.0) {
     if (!WriteEbmlElement(writer, libwebm::kMkvFrameRate,
                           static_cast<float>(frame_rate_))) {
@@ -1542,6 +1552,22 @@ bool VideoTrack::Write(IMkvWriter* writer) const {
   return true;
 }
 
+void VideoTrack::set_colour_space(const char* colour_space) {
+  if (colour_space) {
+    delete[] colour_space_;
+
+    const size_t length = strlen(colour_space) + 1;
+    colour_space_ = new (std::nothrow) char[length];  // NOLINT
+    if (colour_space_) {
+#ifdef _MSC_VER
+      strcpy_s(colour_space_, length, colour_space);
+#else
+      strcpy(colour_space_, colour_space);
+#endif
+    }
+  }
+}
+
 bool VideoTrack::SetColour(const Colour& colour) {
   std::unique_ptr<Colour> colour_ptr(new Colour());
   if (!colour_ptr.get())
@@ -1625,6 +1651,8 @@ uint64_t VideoTrack::VideoPayloadSize() const {
   if (frame_rate_ > 0.0)
     size += EbmlElementSize(libwebm::kMkvFrameRate,
                             static_cast<float>(frame_rate_));
+  if (colour_space_)
+    size += EbmlElementSize(libwebm::kMkvColourSpace, colour_space_);
   if (colour_)
     size += colour_->ColourSize();
   if (projection_)
@@ -1702,10 +1730,9 @@ bool AudioTrack::Write(IMkvWriter* writer) const {
 
 const char Tracks::kOpusCodecId[] = "A_OPUS";
 const char Tracks::kVorbisCodecId[] = "A_VORBIS";
+const char Tracks::kAv1CodecId[] = "V_AV1";
 const char Tracks::kVp8CodecId[] = "V_VP8";
 const char Tracks::kVp9CodecId[] = "V_VP9";
-const char Tracks::kVp10CodecId[] = "V_VP10";
-const char Tracks::kAV1CodecId[] = "V_AV1";
 const char Tracks::kWebVttCaptionsId[] = "D_WEBVTT/CAPTIONS";
 const char Tracks::kWebVttDescriptionsId[] = "D_WEBVTT/DESCRIPTIONS";
 const char Tracks::kWebVttMetadataId[] = "D_WEBVTT/METADATA";
@@ -4161,15 +4188,15 @@ bool Segment::WriteFramesLessThan(uint64_t timestamp) {
 }
 
 bool Segment::DocTypeIsWebm() const {
-  const int kNumCodecIds = 10;
+  const int kNumCodecIds = 9;
 
   // TODO(vigneshv): Tweak .clang-format.
   const char* kWebmCodecIds[kNumCodecIds] = {
       Tracks::kOpusCodecId,          Tracks::kVorbisCodecId,
-      Tracks::kVp8CodecId,           Tracks::kVp9CodecId,
-      Tracks::kVp10CodecId,          Tracks::kAV1CodecId,
-      Tracks::kWebVttCaptionsId,     Tracks::kWebVttDescriptionsId,
-      Tracks::kWebVttMetadataId,     Tracks::kWebVttSubtitlesId};
+      Tracks::kAv1CodecId,           Tracks::kVp8CodecId,
+      Tracks::kVp9CodecId,           Tracks::kWebVttCaptionsId,
+      Tracks::kWebVttDescriptionsId, Tracks::kWebVttMetadataId,
+      Tracks::kWebVttSubtitlesId};
 
   const int num_tracks = static_cast<int>(tracks_.track_entries_size());
   for (int track_index = 0; track_index < num_tracks; ++track_index) {
diff --git a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.h b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.h
index 9e817bced4..f2db377145 100644
--- a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.h
+++ b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxer.h
@@ -795,6 +795,8 @@ class VideoTrack : public Track {
   uint64_t alpha_mode() { return alpha_mode_; }
   void set_width(uint64_t width) { width_ = width; }
   uint64_t width() const { return width_; }
+  void set_colour_space(const char* colour_space);
+  const char* colour_space() const { return colour_space_; }
 
   Colour* colour() { return colour_; }
 
@@ -824,6 +826,7 @@ class VideoTrack : public Track {
   uint64_t stereo_mode_;
   uint64_t alpha_mode_;
   uint64_t width_;
+  char* colour_space_;
 
   Colour* colour_;
   Projection* projection_;
@@ -871,10 +874,9 @@ class Tracks {
 
   static const char kOpusCodecId[];
   static const char kVorbisCodecId[];
+  static const char kAv1CodecId[];
   static const char kVp8CodecId[];
   static const char kVp9CodecId[];
-  static const char kVp10CodecId[];
-  static const char kAV1CodecId[];
   static const char kWebVttCaptionsId[];
   static const char kWebVttDescriptionsId[];
   static const char kWebVttMetadataId[];
diff --git a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxerutil.cc b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxerutil.cc
index 355d4e22b3..3bff7cd51a 100644
--- a/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxerutil.cc
+++ b/third_party/aom/third_party/libwebm/mkvmuxer/mkvmuxerutil.cc
@@ -136,9 +136,8 @@ uint64 WriteBlock(IMkvWriter* writer, const Frame* const frame, int64 timecode,
     return false;
   }
 
-  if (!frame->is_key() &&
-      !WriteEbmlElement(writer, libwebm::kMkvReferenceBlock,
-                        reference_block_timestamp)) {
+  if (!frame->is_key() && !WriteEbmlElement(writer, libwebm::kMkvReferenceBlock,
+                                            reference_block_timestamp)) {
     return false;
   }
 
diff --git a/third_party/aom/third_party/libwebm/mkvmuxer/mkvwriter.cc b/third_party/aom/third_party/libwebm/mkvmuxer/mkvwriter.cc
index 84655d802a..d668384d85 100644
--- a/third_party/aom/third_party/libwebm/mkvmuxer/mkvwriter.cc
+++ b/third_party/aom/third_party/libwebm/mkvmuxer/mkvwriter.cc
@@ -78,6 +78,8 @@ int32 MkvWriter::Position(int64 position) {
 
 #ifdef _MSC_VER
   return _fseeki64(file_, position, SEEK_SET);
+#elif defined(_WIN32)
+  return fseeko64(file_, static_cast<off_t>(position), SEEK_SET);
 #else
   return fseeko(file_, static_cast<off_t>(position), SEEK_SET);
 #endif
diff --git a/third_party/aom/third_party/libwebm/mkvparser/mkvparser.cc b/third_party/aom/third_party/libwebm/mkvparser/mkvparser.cc
index e7b76f7da1..9c78ead2b0 100644
--- a/third_party/aom/third_party/libwebm/mkvparser/mkvparser.cc
+++ b/third_party/aom/third_party/libwebm/mkvparser/mkvparser.cc
@@ -36,8 +36,6 @@ inline bool isnan(double val) { return std::isnan(val); }
 inline bool isinf(double val) { return std::isinf(val); }
 #endif  // MSC_COMPAT
 
-IMkvReader::~IMkvReader() {}
-
 template <typename Type>
 Type* SafeArrayAlloc(unsigned long long num_elements,
                      unsigned long long element_size) {
@@ -5274,6 +5272,7 @@ bool Projection::Parse(IMkvReader* reader, long long start, long long size,
 VideoTrack::VideoTrack(Segment* pSegment, long long element_start,
                        long long element_size)
     : Track(pSegment, element_start, element_size),
+      m_colour_space(NULL),
       m_colour(NULL),
       m_projection(NULL) {}
 
@@ -5299,6 +5298,7 @@ long VideoTrack::Parse(Segment* pSegment, const Info& info,
   long long stereo_mode = 0;
 
   double rate = 0.0;
+  char* colour_space = NULL;
 
   IMkvReader* const pReader = pSegment->m_pReader;
 
@@ -5312,7 +5312,7 @@ long VideoTrack::Parse(Segment* pSegment, const Info& info,
   const long long stop = pos + s.size;
 
   Colour* colour = NULL;
-  Projection* projection = NULL;
+  std::unique_ptr<Projection> projection_ptr;
 
   while (pos < stop) {
     long long id, size;
@@ -5364,8 +5364,16 @@ long VideoTrack::Parse(Segment* pSegment, const Info& info,
       if (!Colour::Parse(pReader, pos, size, &colour))
         return E_FILE_FORMAT_INVALID;
     } else if (id == libwebm::kMkvProjection) {
-      if (!Projection::Parse(pReader, pos, size, &projection))
+      Projection* projection = NULL;
+      if (!Projection::Parse(pReader, pos, size, &projection)) {
         return E_FILE_FORMAT_INVALID;
+      } else {
+        projection_ptr.reset(projection);
+      }
+    } else if (id == libwebm::kMkvColourSpace) {
+      const long status = UnserializeString(pReader, pos, size, colour_space);
+      if (status < 0)
+        return status;
     }
 
     pos += size;  // consume payload
@@ -5397,7 +5405,8 @@ long VideoTrack::Parse(Segment* pSegment, const Info& info,
   pTrack->m_stereo_mode = stereo_mode;
   pTrack->m_rate = rate;
   pTrack->m_colour = colour;
-  pTrack->m_projection = projection;
+  pTrack->m_colour_space = colour_space;
+  pTrack->m_projection = projection_ptr.release();
 
   pResult = pTrack;
   return 0;  // success
diff --git a/third_party/aom/third_party/libwebm/mkvparser/mkvparser.h b/third_party/aom/third_party/libwebm/mkvparser/mkvparser.h
index 26c2b7e5eb..848d01f03e 100644
--- a/third_party/aom/third_party/libwebm/mkvparser/mkvparser.h
+++ b/third_party/aom/third_party/libwebm/mkvparser/mkvparser.h
@@ -22,7 +22,7 @@ class IMkvReader {
   virtual int Length(long long* total, long long* available) = 0;
 
  protected:
-  virtual ~IMkvReader();
+  virtual ~IMkvReader() {}
 };
 
 template <typename Type>
@@ -527,6 +527,8 @@ class VideoTrack : public Track {
 
   Projection* GetProjection() const;
 
+  const char* GetColourSpace() const { return m_colour_space; }
+
  private:
   long long m_width;
   long long m_height;
@@ -534,7 +536,7 @@ class VideoTrack : public Track {
   long long m_display_height;
   long long m_display_unit;
   long long m_stereo_mode;
-
+  char* m_colour_space;
   double m_rate;
 
   Colour* m_colour;
diff --git a/third_party/aom/third_party/libwebm/mkvparser/mkvreader.cc b/third_party/aom/third_party/libwebm/mkvparser/mkvreader.cc
index 23d68f5089..9d19c1be56 100644
--- a/third_party/aom/third_party/libwebm/mkvparser/mkvreader.cc
+++ b/third_party/aom/third_party/libwebm/mkvparser/mkvreader.cc
@@ -118,6 +118,8 @@ int MkvReader::Read(long long offset, long len, unsigned char* buffer) {
 
   if (status)
     return -1;  // error
+#elif defined(_WIN32)
+  fseeko64(m_file, static_cast<off_t>(offset), SEEK_SET);
 #else
   fseeko(m_file, static_cast<off_t>(offset), SEEK_SET);
 #endif
diff --git a/third_party/aom/tools/txfm_analyzer/txfm_graph.h b/third_party/aom/tools/txfm_analyzer/txfm_graph.h
index 2e3c9551e2..8dc36146dd 100644
--- a/third_party/aom/tools/txfm_analyzer/txfm_graph.h
+++ b/third_party/aom/tools/txfm_analyzer/txfm_graph.h
@@ -23,7 +23,6 @@ struct Node {
   int visited;
 };
 
-#define PI (3.141592653589793238462643383279502884)
 #define STAGENUM (10)
 #define NODENUM (32)
 #define COS_MOD (128)