Update aom to v1.0.0

Update aom to commit id d14c5bb4f336ef1842046089849dee4a301fbbf0.

third_party/aom/av1/common/alloccommon.c

@@ -10,7 +10,8 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom_mem/aom_mem.h"
 
 #include "av1/common/alloccommon.h"
@@ -25,16 +26,43 @@ int av1_get_MBs(int width, int height) {
   const int mi_cols = aligned_width >> MI_SIZE_LOG2;
   const int mi_rows = aligned_height >> MI_SIZE_LOG2;
 
-#if CONFIG_CB4X4
   const int mb_cols = (mi_cols + 2) >> 2;
   const int mb_rows = (mi_rows + 2) >> 2;
-#else
-  const int mb_cols = (mi_cols + 1) >> 1;
-  const int mb_rows = (mi_rows + 1) >> 1;
-#endif
   return mb_rows * mb_cols;
 }
 
+#if LOOP_FILTER_BITMASK
+static int alloc_loop_filter_mask(AV1_COMMON *cm) {
+  aom_free(cm->lf.lfm);
+  cm->lf.lfm = NULL;
+
+  // Each lfm holds bit masks for all the 4x4 blocks in a max
+  // 64x64 (128x128 for ext_partitions) region.  The stride
+  // and rows are rounded up / truncated to a multiple of 16
+  // (32 for ext_partition).
+  cm->lf.lfm_stride = (cm->mi_cols + (MI_SIZE_64X64 - 1)) >> MIN_MIB_SIZE_LOG2;
+  cm->lf.lfm_num = ((cm->mi_rows + (MI_SIZE_64X64 - 1)) >> MIN_MIB_SIZE_LOG2) *
+                   cm->lf.lfm_stride;
+  cm->lf.lfm =
+      (LoopFilterMask *)aom_calloc(cm->lf.lfm_num, sizeof(*cm->lf.lfm));
+  if (!cm->lf.lfm) return 1;
+
+  unsigned int i;
+  for (i = 0; i < cm->lf.lfm_num; ++i) av1_zero(cm->lf.lfm[i]);
+
+  return 0;
+}
+
+static void free_loop_filter_mask(AV1_COMMON *cm) {
+  if (cm->lf.lfm == NULL) return;
+
+  aom_free(cm->lf.lfm);
+  cm->lf.lfm = NULL;
+  cm->lf.lfm_num = 0;
+  cm->lf.lfm_stride = 0;
+}
+#endif
+
 void av1_set_mb_mi(AV1_COMMON *cm, int width, int height) {
   // Ensure that the decoded width and height are both multiples of
   // 8 luma pixels (note: this may only be a multiple of 4 chroma pixels if
@@ -48,79 +76,13 @@ void av1_set_mb_mi(AV1_COMMON *cm, int width, int height) {
   cm->mi_rows = aligned_height >> MI_SIZE_LOG2;
   cm->mi_stride = calc_mi_size(cm->mi_cols);
 
-#if CONFIG_CB4X4
   cm->mb_cols = (cm->mi_cols + 2) >> 2;
   cm->mb_rows = (cm->mi_rows + 2) >> 2;
-#else
-  cm->mb_cols = (cm->mi_cols + 1) >> 1;
-  cm->mb_rows = (cm->mi_rows + 1) >> 1;
-#endif
   cm->MBs = cm->mb_rows * cm->mb_cols;
-}
 
-static int alloc_seg_map(AV1_COMMON *cm, int seg_map_size) {
-  int i;
-
-  for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
-    cm->seg_map_array[i] = (uint8_t *)aom_calloc(seg_map_size, 1);
-    if (cm->seg_map_array[i] == NULL) return 1;
-  }
-  cm->seg_map_alloc_size = seg_map_size;
-
-  // Init the index.
-  cm->seg_map_idx = 0;
-  cm->prev_seg_map_idx = 1;
-
-  cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
-  if (!cm->frame_parallel_decode)
-    cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
-
-  return 0;
-}
-
-static void free_seg_map(AV1_COMMON *cm) {
-  int i;
-
-  for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
-    aom_free(cm->seg_map_array[i]);
-    cm->seg_map_array[i] = NULL;
-  }
-
-  cm->current_frame_seg_map = NULL;
-
-  if (!cm->frame_parallel_decode) {
-    cm->last_frame_seg_map = NULL;
-  }
-  cm->seg_map_alloc_size = 0;
-}
-
-static void free_scratch_buffers(AV1_COMMON *cm) {
-  (void)cm;
-#if CONFIG_NCOBMC && CONFIG_NCOBMC_ADAPT_WEIGHT
-  for (int i = 0; i < 4; ++i) {
-    if (cm->ncobmcaw_buf[i]) {
-      aom_free(cm->ncobmcaw_buf[i]);
-      cm->ncobmcaw_buf[i] = NULL;
-    }
-  }
-#endif  // CONFIG_NCOBMC && CONFIG_NCOBMC_ADAPT_WEIGHT
-}
-
-static int alloc_scratch_buffers(AV1_COMMON *cm) {
-  (void)cm;
-#if CONFIG_NCOBMC && CONFIG_NCOBMC_ADAPT_WEIGHT
-  // If not allocated already, allocate
-  if (!cm->ncobmcaw_buf[0] && !cm->ncobmcaw_buf[1] && !cm->ncobmcaw_buf[2] &&
-      !cm->ncobmcaw_buf[3]) {
-    for (int i = 0; i < 4; ++i) {
-      CHECK_MEM_ERROR(
-          cm, cm->ncobmcaw_buf[i],
-          (uint8_t *)aom_memalign(
-              16, (1 + CONFIG_HIGHBITDEPTH) * MAX_MB_PLANE * MAX_SB_SQUARE));
-    }
-  }
-#endif  // CONFIG_NCOBMC && CONFIG_NCOBMC_ADAPT_WEIGHT
-  return 0;
+#if LOOP_FILTER_BITMASK
+  alloc_loop_filter_mask(cm);
+#endif
 }
 
 void av1_free_ref_frame_buffers(BufferPool *pool) {
@@ -134,97 +96,179 @@ void av1_free_ref_frame_buffers(BufferPool *pool) {
     }
     aom_free(pool->frame_bufs[i].mvs);
     pool->frame_bufs[i].mvs = NULL;
-#if CONFIG_MFMV
-    aom_free(pool->frame_bufs[i].tpl_mvs);
-    pool->frame_bufs[i].tpl_mvs = NULL;
-#endif
+    aom_free(pool->frame_bufs[i].seg_map);
+    pool->frame_bufs[i].seg_map = NULL;
     aom_free_frame_buffer(&pool->frame_bufs[i].buf);
-#if CONFIG_HASH_ME
-    av1_hash_table_destroy(&pool->frame_bufs[i].hash_table);
-#endif
   }
 }
 
-#if CONFIG_LOOP_RESTORATION
-// Assumes cm->rst_info[p].restoration_tilesize is already initialized
+// Assumes cm->rst_info[p].restoration_unit_size is already initialized
 void av1_alloc_restoration_buffers(AV1_COMMON *cm) {
-  int p;
-#if CONFIG_FRAME_SUPERRES
-  int width = cm->superres_upscaled_width;
-  int height = cm->superres_upscaled_height;
-#else
-  int width = cm->width;
-  int height = cm->height;
-#endif  // CONFIG_FRAME_SUPERRES
-  av1_alloc_restoration_struct(cm, &cm->rst_info[0], width, height);
-  for (p = 1; p < MAX_MB_PLANE; ++p)
-    av1_alloc_restoration_struct(cm, &cm->rst_info[p],
-                                 ROUND_POWER_OF_TWO(width, cm->subsampling_x),
-                                 ROUND_POWER_OF_TWO(height, cm->subsampling_y));
-  aom_free(cm->rst_internal.tmpbuf);
-  CHECK_MEM_ERROR(cm, cm->rst_internal.tmpbuf,
-                  (int32_t *)aom_memalign(16, RESTORATION_TMPBUF_SIZE));
+  const int num_planes = av1_num_planes(cm);
+  for (int p = 0; p < num_planes; ++p)
+    av1_alloc_restoration_struct(cm, &cm->rst_info[p], p > 0);
 
-#if CONFIG_STRIPED_LOOP_RESTORATION
-  // Allocate internal storage for the loop restoration stripe boundary lines
-  for (p = 0; p < MAX_MB_PLANE; ++p) {
-    int w = p == 0 ? width : ROUND_POWER_OF_TWO(width, cm->subsampling_x);
-    int align_bits = 5;  // align for efficiency
-    int stride = ALIGN_POWER_OF_TWO(w, align_bits);
-    int num_stripes = (height + 63) / 64;
-    // for each processing stripe: 2 lines above, 2 below
-    int buf_size = num_stripes * 2 * stride;
-    uint8_t *above_buf, *below_buf;
-
-    aom_free(cm->rst_internal.stripe_boundary_above[p]);
-    aom_free(cm->rst_internal.stripe_boundary_below[p]);
-
-#if CONFIG_HIGHBITDEPTH
-    if (cm->use_highbitdepth) buf_size = buf_size * 2;
-#endif
-    CHECK_MEM_ERROR(cm, above_buf,
-                    (uint8_t *)aom_memalign(1 << align_bits, buf_size));
-    CHECK_MEM_ERROR(cm, below_buf,
-                    (uint8_t *)aom_memalign(1 << align_bits, buf_size));
-    cm->rst_internal.stripe_boundary_above[p] = above_buf;
-    cm->rst_internal.stripe_boundary_below[p] = below_buf;
-    cm->rst_internal.stripe_boundary_stride[p] = stride;
+  if (cm->rst_tmpbuf == NULL) {
+    CHECK_MEM_ERROR(cm, cm->rst_tmpbuf,
+                    (int32_t *)aom_memalign(16, RESTORATION_TMPBUF_SIZE));
+  }
+
+  if (cm->rlbs == NULL) {
+    CHECK_MEM_ERROR(cm, cm->rlbs, aom_malloc(sizeof(RestorationLineBuffers)));
+  }
+
+  // For striped loop restoration, we divide each row of tiles into "stripes",
+  // of height 64 luma pixels but with an offset by RESTORATION_UNIT_OFFSET
+  // luma pixels to match the output from CDEF. We will need to store 2 *
+  // RESTORATION_CTX_VERT lines of data for each stripe, and also need to be
+  // able to quickly answer the question "Where is the <n>'th stripe for tile
+  // row <m>?" To make that efficient, we generate the rst_last_stripe array.
+  int num_stripes = 0;
+  for (int i = 0; i < cm->tile_rows; ++i) {
+    TileInfo tile_info;
+    av1_tile_set_row(&tile_info, cm, i);
+    const int mi_h = tile_info.mi_row_end - tile_info.mi_row_start;
+    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->use_highbitdepth ? 1 : 0;
+
+  for (int p = 0; p < num_planes; ++p) {
+    const int is_uv = p > 0;
+    const int ss_x = is_uv && cm->subsampling_x;
+    const int plane_w = ((frame_w + ss_x) >> ss_x) + 2 * RESTORATION_EXTRA_HORZ;
+    const int stride = ALIGN_POWER_OF_TWO(plane_w, 5);
+    const int buf_size = num_stripes * stride * RESTORATION_CTX_VERT
+                         << use_highbd;
+    RestorationStripeBoundaries *boundaries = &cm->rst_info[p].boundaries;
+
+    if (buf_size != boundaries->stripe_boundary_size ||
+        boundaries->stripe_boundary_above == NULL ||
+        boundaries->stripe_boundary_below == NULL) {
+      aom_free(boundaries->stripe_boundary_above);
+      aom_free(boundaries->stripe_boundary_below);
+
+      CHECK_MEM_ERROR(cm, boundaries->stripe_boundary_above,
+                      (uint8_t *)aom_memalign(32, buf_size));
+      CHECK_MEM_ERROR(cm, boundaries->stripe_boundary_below,
+                      (uint8_t *)aom_memalign(32, buf_size));
+
+      boundaries->stripe_boundary_size = buf_size;
+    }
+    boundaries->stripe_boundary_stride = stride;
   }
-#endif  // CONFIG_STRIPED_LOOP_RESTORATION
 }
 
 void av1_free_restoration_buffers(AV1_COMMON *cm) {
   int p;
   for (p = 0; p < MAX_MB_PLANE; ++p)
     av1_free_restoration_struct(&cm->rst_info[p]);
-  aom_free(cm->rst_internal.tmpbuf);
-  cm->rst_internal.tmpbuf = NULL;
-}
-#endif  // CONFIG_LOOP_RESTORATION
+  aom_free(cm->rst_tmpbuf);
+  cm->rst_tmpbuf = NULL;
+  aom_free(cm->rlbs);
+  cm->rlbs = NULL;
+  for (p = 0; p < MAX_MB_PLANE; ++p) {
+    RestorationStripeBoundaries *boundaries = &cm->rst_info[p].boundaries;
+    aom_free(boundaries->stripe_boundary_above);
+    aom_free(boundaries->stripe_boundary_below);
+    boundaries->stripe_boundary_above = NULL;
+    boundaries->stripe_boundary_below = NULL;
+  }
 
-void av1_free_context_buffers(AV1_COMMON *cm) {
+  aom_free_frame_buffer(&cm->rst_frame);
+}
+
+void av1_free_above_context_buffers(AV1_COMMON *cm,
+                                    int num_free_above_contexts) {
   int i;
-  cm->free_mi(cm);
-  free_seg_map(cm);
-  free_scratch_buffers(cm);
-  for (i = 0; i < MAX_MB_PLANE; i++) {
+  const int num_planes = cm->num_allocated_above_context_planes;
+
+  for (int tile_row = 0; tile_row < num_free_above_contexts; tile_row++) {
+    for (i = 0; i < num_planes; i++) {
+      aom_free(cm->above_context[i][tile_row]);
+      cm->above_context[i][tile_row] = NULL;
+    }
+    aom_free(cm->above_seg_context[tile_row]);
+    cm->above_seg_context[tile_row] = NULL;
+
+    aom_free(cm->above_txfm_context[tile_row]);
+    cm->above_txfm_context[tile_row] = NULL;
+  }
+  for (i = 0; i < num_planes; i++) {
     aom_free(cm->above_context[i]);
     cm->above_context[i] = NULL;
   }
   aom_free(cm->above_seg_context);
   cm->above_seg_context = NULL;
-  cm->above_context_alloc_cols = 0;
-#if CONFIG_VAR_TX
+
   aom_free(cm->above_txfm_context);
   cm->above_txfm_context = NULL;
 
-  for (i = 0; i < MAX_MB_PLANE; ++i) {
-    aom_free(cm->top_txfm_context[i]);
-    cm->top_txfm_context[i] = NULL;
-  }
+  cm->num_allocated_above_contexts = 0;
+  cm->num_allocated_above_context_mi_col = 0;
+  cm->num_allocated_above_context_planes = 0;
+}
+
+void av1_free_context_buffers(AV1_COMMON *cm) {
+  cm->free_mi(cm);
+
+  av1_free_above_context_buffers(cm, cm->num_allocated_above_contexts);
+
+#if LOOP_FILTER_BITMASK
+  free_loop_filter_mask(cm);
 #endif
 }
 
+int av1_alloc_above_context_buffers(AV1_COMMON *cm,
+                                    int num_alloc_above_contexts) {
+  const int num_planes = av1_num_planes(cm);
+  int plane_idx;
+  const int aligned_mi_cols =
+      ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
+
+  // Allocate above context buffers
+  cm->num_allocated_above_contexts = num_alloc_above_contexts;
+  cm->num_allocated_above_context_mi_col = aligned_mi_cols;
+  cm->num_allocated_above_context_planes = num_planes;
+  for (plane_idx = 0; plane_idx < num_planes; plane_idx++) {
+    cm->above_context[plane_idx] = (ENTROPY_CONTEXT **)aom_calloc(
+        num_alloc_above_contexts, sizeof(cm->above_context[0]));
+    if (!cm->above_context[plane_idx]) return 1;
+  }
+
+  cm->above_seg_context = (PARTITION_CONTEXT **)aom_calloc(
+      num_alloc_above_contexts, sizeof(cm->above_seg_context));
+  if (!cm->above_seg_context) return 1;
+
+  cm->above_txfm_context = (TXFM_CONTEXT **)aom_calloc(
+      num_alloc_above_contexts, sizeof(cm->above_txfm_context));
+  if (!cm->above_txfm_context) return 1;
+
+  for (int tile_row = 0; tile_row < num_alloc_above_contexts; tile_row++) {
+    for (plane_idx = 0; plane_idx < num_planes; plane_idx++) {
+      cm->above_context[plane_idx][tile_row] = (ENTROPY_CONTEXT *)aom_calloc(
+          aligned_mi_cols, sizeof(*cm->above_context[0][tile_row]));
+      if (!cm->above_context[plane_idx][tile_row]) return 1;
+    }
+
+    cm->above_seg_context[tile_row] = (PARTITION_CONTEXT *)aom_calloc(
+        aligned_mi_cols, sizeof(*cm->above_seg_context[tile_row]));
+    if (!cm->above_seg_context[tile_row]) return 1;
+
+    cm->above_txfm_context[tile_row] = (TXFM_CONTEXT *)aom_calloc(
+        aligned_mi_cols, sizeof(*cm->above_txfm_context[tile_row]));
+    if (!cm->above_txfm_context[tile_row]) return 1;
+  }
+
+  return 0;
+}
+
 int av1_alloc_context_buffers(AV1_COMMON *cm, int width, int height) {
   int new_mi_size;
 
@@ -235,52 +279,6 @@ int av1_alloc_context_buffers(AV1_COMMON *cm, int width, int height) {
     if (cm->alloc_mi(cm, new_mi_size)) goto fail;
   }
 
-  if (cm->seg_map_alloc_size < cm->mi_rows * cm->mi_cols) {
-    // Create the segmentation map structure and set to 0.
-    free_seg_map(cm);
-    if (alloc_seg_map(cm, cm->mi_rows * cm->mi_cols)) goto fail;
-  }
-  if (alloc_scratch_buffers(cm)) goto fail;
-
-  if (cm->above_context_alloc_cols < cm->mi_cols) {
-    // TODO(geza.lore): These are bigger than they need to be.
-    // cm->tile_width would be enough but it complicates indexing a
-    // little elsewhere.
-    const int aligned_mi_cols =
-        ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
-    int i;
-
-    for (i = 0; i < MAX_MB_PLANE; i++) {
-      aom_free(cm->above_context[i]);
-      cm->above_context[i] = (ENTROPY_CONTEXT *)aom_calloc(
-          aligned_mi_cols << (MI_SIZE_LOG2 - tx_size_wide_log2[0]),
-          sizeof(*cm->above_context[0]));
-      if (!cm->above_context[i]) goto fail;
-    }
-
-    aom_free(cm->above_seg_context);
-    cm->above_seg_context = (PARTITION_CONTEXT *)aom_calloc(
-        aligned_mi_cols, sizeof(*cm->above_seg_context));
-    if (!cm->above_seg_context) goto fail;
-
-#if CONFIG_VAR_TX
-    aom_free(cm->above_txfm_context);
-    cm->above_txfm_context = (TXFM_CONTEXT *)aom_calloc(
-        aligned_mi_cols << TX_UNIT_WIDE_LOG2, sizeof(*cm->above_txfm_context));
-    if (!cm->above_txfm_context) goto fail;
-
-    for (i = 0; i < MAX_MB_PLANE; ++i) {
-      aom_free(cm->top_txfm_context[i]);
-      cm->top_txfm_context[i] =
-          (TXFM_CONTEXT *)aom_calloc(aligned_mi_cols << TX_UNIT_WIDE_LOG2,
-                                     sizeof(*cm->top_txfm_context[0]));
-      if (!cm->top_txfm_context[i]) goto fail;
-    }
-#endif
-
-    cm->above_context_alloc_cols = aligned_mi_cols;
-  }
-
   return 0;
 
 fail:
@@ -299,18 +297,4 @@ void av1_remove_common(AV1_COMMON *cm) {
   cm->frame_contexts = NULL;
 }
 
-void av1_init_context_buffers(AV1_COMMON *cm) {
-  cm->setup_mi(cm);
-  if (cm->last_frame_seg_map && !cm->frame_parallel_decode)
-    memset(cm->last_frame_seg_map, 0, cm->mi_rows * cm->mi_cols);
-}
-
-void av1_swap_current_and_last_seg_map(AV1_COMMON *cm) {
-  // Swap indices.
-  const int tmp = cm->seg_map_idx;
-  cm->seg_map_idx = cm->prev_seg_map_idx;
-  cm->prev_seg_map_idx = tmp;
-
-  cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
-  cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
-}
+void av1_init_context_buffers(AV1_COMMON *cm) { cm->setup_mi(cm); }

third_party/aom/av1/common/alloccommon.h

@@ -23,15 +23,17 @@ struct BufferPool;
 
 void av1_remove_common(struct AV1Common *cm);
 
+int av1_alloc_above_context_buffers(struct AV1Common *cm,
+                                    int num_alloc_above_contexts);
+void av1_free_above_context_buffers(struct AV1Common *cm,
+                                    int num_free_above_contexts);
 int av1_alloc_context_buffers(struct AV1Common *cm, int width, int height);
 void av1_init_context_buffers(struct AV1Common *cm);
 void av1_free_context_buffers(struct AV1Common *cm);
 
 void av1_free_ref_frame_buffers(struct BufferPool *pool);
-#if CONFIG_LOOP_RESTORATION
 void av1_alloc_restoration_buffers(struct AV1Common *cm);
 void av1_free_restoration_buffers(struct AV1Common *cm);
-#endif  // CONFIG_LOOP_RESTORATION
 
 int av1_alloc_state_buffers(struct AV1Common *cm, int width, int height);
 void av1_free_state_buffers(struct AV1Common *cm);
@@ -39,8 +41,6 @@ void av1_free_state_buffers(struct AV1Common *cm);
 void av1_set_mb_mi(struct AV1Common *cm, int width, int height);
 int av1_get_MBs(int width, int height);
 
-void av1_swap_current_and_last_seg_map(struct AV1Common *cm);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/arm/av1_txfm_neon.c

@@ -0,0 +1,28 @@
+/*
+ *
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "aom_ports/mem.h"
+#include "av1/common/arm/mem_neon.h"
+
+void av1_round_shift_array_neon(int32_t *arr, int size, int bit) {
+  assert(!(size % 4));
+  if (!bit) return;
+  const int32x4_t dup_bits_n_32x4 = vdupq_n_s32((int32_t)(-bit));
+  for (int i = 0; i < size; i += 4) {
+    int32x4_t tmp_q_s32 = vld1q_s32(arr);
+    tmp_q_s32 = vrshlq_s32(tmp_q_s32, dup_bits_n_32x4);
+    vst1q_s32(arr, tmp_q_s32);
+    arr += 4;
+  }
+}

third_party/aom/av1/common/arm/blend_a64_hmask_neon.c

@@ -0,0 +1,134 @@
+/*
+ *
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "aom/aom_integer.h"
+#include "aom_dsp/blend.h"
+#include "aom_ports/mem.h"
+#include "av1/common/arm/mem_neon.h"
+#include "aom_dsp/aom_dsp_common.h"
+#include "config/aom_dsp_rtcd.h"
+
+void aom_blend_a64_hmask_neon(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) {
+  assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
+  assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
+
+  assert(h >= 2);
+  assert(w >= 2);
+  assert(IS_POWER_OF_TWO(h));
+  assert(IS_POWER_OF_TWO(w));
+  uint8x8_t tmp0, tmp1;
+  uint8x16_t res_q;
+  uint16x8_t res, res_low, res_high;
+  uint32x2_t tmp0_32, tmp1_32;
+  uint16x4_t tmp0_16, tmp1_16;
+  const uint8x8_t vdup_64 = vdup_n_u8((uint8_t)64);
+
+  if (w >= 16) {
+    const uint8x16_t vdup_64_q = vdupq_n_u8((uint8_t)64);
+    for (int i = 0; i < h; ++i) {
+      for (int j = 0; j < w; j += 16) {
+        __builtin_prefetch(src0);
+        __builtin_prefetch(src1);
+        const uint8x16_t tmp0_q = vld1q_u8(src0);
+        const uint8x16_t tmp1_q = vld1q_u8(src1);
+        const uint8x16_t m_q = vld1q_u8(mask);
+        const uint8x16_t max_minus_m_q = vsubq_u8(vdup_64_q, m_q);
+        res_low = vmull_u8(vget_low_u8(m_q), vget_low_u8(tmp0_q));
+        res_low =
+            vmlal_u8(res_low, vget_low_u8(max_minus_m_q), vget_low_u8(tmp1_q));
+        res_high = vmull_u8(vget_high_u8(m_q), vget_high_u8(tmp0_q));
+        res_high = vmlal_u8(res_high, vget_high_u8(max_minus_m_q),
+                            vget_high_u8(tmp1_q));
+        res_q = vcombine_u8(vrshrn_n_u16(res_low, AOM_BLEND_A64_ROUND_BITS),
+                            vrshrn_n_u16(res_high, AOM_BLEND_A64_ROUND_BITS));
+        vst1q_u8(dst, res_q);
+        src0 += 16;
+        src1 += 16;
+        dst += 16;
+        mask += 16;
+      }
+      src0 += src0_stride - w;
+      src1 += src1_stride - w;
+      dst += dst_stride - w;
+      mask -= w;
+    }
+  } else if (w == 8) {
+    const uint8x8_t m = vld1_u8(mask);
+    const uint8x8_t max_minus_m = vsub_u8(vdup_64, m);
+    for (int i = 0; i < h; ++i) {
+      __builtin_prefetch(src0);
+      __builtin_prefetch(src1);
+      tmp0 = vld1_u8(src0);
+      tmp1 = vld1_u8(src1);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_u8(dst, vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS));
+      src0 += src0_stride;
+      src1 += src1_stride;
+      dst += dst_stride;
+    }
+  } else if (w == 4) {
+    const uint8x8_t m = vreinterpret_u8_u32(vld1_dup_u32((uint32_t *)mask));
+    const uint8x8_t max_minus_m = vsub_u8(vdup_64, m);
+    for (int i = 0; i < h; i += 2) {
+      __builtin_prefetch(src0 + 0 * src0_stride);
+      __builtin_prefetch(src0 + 1 * src0_stride);
+      __builtin_prefetch(src1 + 0 * src1_stride);
+      __builtin_prefetch(src1 + 1 * src1_stride);
+      load_unaligned_u8_4x2(src0, src0_stride, &tmp0_32);
+      tmp0 = vreinterpret_u8_u32(tmp0_32);
+      load_unaligned_u8_4x2(src1, src1_stride, &tmp1_32);
+      tmp1 = vreinterpret_u8_u32(tmp1_32);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_lane_u32(
+          (uint32_t *)(dst + (0 * dst_stride)),
+          vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
+      vst1_lane_u32(
+          (uint32_t *)(dst + (1 * dst_stride)),
+          vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
+      src0 += (2 * src0_stride);
+      src1 += (2 * src1_stride);
+      dst += (2 * dst_stride);
+    }
+  } else if (w == 2) {
+    const uint8x8_t m = vreinterpret_u8_u16(vld1_dup_u16((uint16_t *)mask));
+    const uint8x8_t max_minus_m = vsub_u8(vdup_64, m);
+    for (int i = 0; i < h; i += 2) {
+      __builtin_prefetch(src0 + 0 * src0_stride);
+      __builtin_prefetch(src0 + 1 * src0_stride);
+      __builtin_prefetch(src1 + 0 * src1_stride);
+      __builtin_prefetch(src1 + 1 * src1_stride);
+      load_unaligned_u8_2x2(src0, src0_stride, &tmp0_16);
+      tmp0 = vreinterpret_u8_u16(tmp0_16);
+      load_unaligned_u8_2x2(src1, src1_stride, &tmp1_16);
+      tmp1 = vreinterpret_u8_u16(tmp1_16);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_lane_u16(
+          (uint16_t *)(dst + (0 * dst_stride)),
+          vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
+      vst1_lane_u16(
+          (uint16_t *)(dst + (1 * dst_stride)),
+          vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
+      src0 += (2 * src0_stride);
+      src1 += (2 * src1_stride);
+      dst += (2 * dst_stride);
+    }
+  }
+}

third_party/aom/av1/common/arm/blend_a64_vmask_neon.c

@@ -0,0 +1,141 @@
+/*
+ *
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "aom/aom_integer.h"
+#include "aom_dsp/blend.h"
+#include "aom_ports/mem.h"
+#include "av1/common/arm/mem_neon.h"
+#include "aom_dsp/aom_dsp_common.h"
+#include "config/aom_dsp_rtcd.h"
+
+void aom_blend_a64_vmask_neon(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) {
+  uint8x8_t tmp0, tmp1;
+  uint8x16_t tmp0_q, tmp1_q, res_q;
+  uint16x8_t res, res_low, res_high;
+  uint32x2_t tmp0_32, tmp1_32;
+  uint16x4_t tmp0_16, tmp1_16;
+  assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
+  assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
+
+  assert(h >= 2);
+  assert(w >= 2);
+  assert(IS_POWER_OF_TWO(h));
+  assert(IS_POWER_OF_TWO(w));
+
+  if (w >= 16) {
+    for (int i = 0; i < h; ++i) {
+      const uint8x8_t m = vdup_n_u8((uint8_t)mask[i]);
+      const uint8x8_t max_minus_m = vdup_n_u8(64 - (uint8_t)mask[i]);
+      for (int j = 0; j < w; j += 16) {
+        __builtin_prefetch(src0);
+        __builtin_prefetch(src1);
+        tmp0_q = vld1q_u8(src0);
+        tmp1_q = vld1q_u8(src1);
+        res_low = vmull_u8(m, vget_low_u8(tmp0_q));
+        res_low = vmlal_u8(res_low, max_minus_m, vget_low_u8(tmp1_q));
+        res_high = vmull_u8(m, vget_high_u8(tmp0_q));
+        res_high = vmlal_u8(res_high, max_minus_m, vget_high_u8(tmp1_q));
+        res_q = vcombine_u8(vrshrn_n_u16(res_low, AOM_BLEND_A64_ROUND_BITS),
+                            vrshrn_n_u16(res_high, AOM_BLEND_A64_ROUND_BITS));
+        vst1q_u8(dst, res_q);
+        src0 += 16;
+        src1 += 16;
+        dst += 16;
+      }
+      src0 += src0_stride - w;
+      src1 += src1_stride - w;
+      dst += dst_stride - w;
+    }
+  } else if (w == 8) {
+    for (int i = 0; i < h; ++i) {
+      __builtin_prefetch(src0);
+      __builtin_prefetch(src1);
+      const uint8x8_t m = vdup_n_u8((uint8_t)mask[i]);
+      const uint8x8_t max_minus_m = vdup_n_u8(64 - (uint8_t)mask[i]);
+      tmp0 = vld1_u8(src0);
+      tmp1 = vld1_u8(src1);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_u8(dst, vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS));
+      src0 += src0_stride;
+      src1 += src1_stride;
+      dst += dst_stride;
+    }
+  } else if (w == 4) {
+    for (int i = 0; i < h; i += 2) {
+      __builtin_prefetch(src0 + 0 * src0_stride);
+      __builtin_prefetch(src0 + 1 * src0_stride);
+      __builtin_prefetch(src1 + 0 * src1_stride);
+      __builtin_prefetch(src1 + 1 * src1_stride);
+      const uint16x4_t m1 = vdup_n_u16((uint16_t)mask[i]);
+      const uint16x4_t m2 = vdup_n_u16((uint16_t)mask[i + 1]);
+      const uint8x8_t m = vmovn_u16(vcombine_u16(m1, m2));
+      const uint16x4_t max_minus_m1 = vdup_n_u16(64 - (uint16_t)mask[i]);
+      const uint16x4_t max_minus_m2 = vdup_n_u16(64 - (uint16_t)mask[i + 1]);
+      const uint8x8_t max_minus_m =
+          vmovn_u16(vcombine_u16(max_minus_m1, max_minus_m2));
+      load_unaligned_u8_4x2(src0, src0_stride, &tmp0_32);
+      tmp0 = vreinterpret_u8_u32(tmp0_32);
+      load_unaligned_u8_4x2(src1, src1_stride, &tmp1_32);
+      tmp1 = vreinterpret_u8_u32(tmp1_32);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_lane_u32(
+          (uint32_t *)(dst + (0 * dst_stride)),
+          vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
+      vst1_lane_u32(
+          (uint32_t *)(dst + (1 * dst_stride)),
+          vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
+      src0 += (2 * src0_stride);
+      src1 += (2 * src1_stride);
+      dst += (2 * dst_stride);
+    }
+  } else if (w == 2) {
+    for (int i = 0; i < h; i += 2) {
+      __builtin_prefetch(src0 + 0 * src0_stride);
+      __builtin_prefetch(src0 + 1 * src0_stride);
+      __builtin_prefetch(src1 + 0 * src1_stride);
+      __builtin_prefetch(src1 + 1 * src1_stride);
+      const uint8x8_t m1 = vdup_n_u8(mask[i]);
+      const uint8x8_t m2 = vdup_n_u8(mask[i + 1]);
+      const uint16x4x2_t m_trn =
+          vtrn_u16(vreinterpret_u16_u8(m1), vreinterpret_u16_u8(m2));
+      const uint8x8_t m = vreinterpret_u8_u16(m_trn.val[0]);
+      const uint8x8_t max_minus_m1 = vdup_n_u8(64 - mask[i]);
+      const uint8x8_t max_minus_m2 = vdup_n_u8(64 - mask[i + 1]);
+      const uint16x4x2_t max_minus_m_trn = vtrn_u16(
+          vreinterpret_u16_u8(max_minus_m1), vreinterpret_u16_u8(max_minus_m2));
+      const uint8x8_t max_minus_m = vreinterpret_u8_u16(max_minus_m_trn.val[0]);
+      load_unaligned_u8_2x2(src0, src0_stride, &tmp0_16);
+      tmp0 = vreinterpret_u8_u16(tmp0_16);
+      load_unaligned_u8_2x2(src1, src1_stride, &tmp1_16);
+      tmp1 = vreinterpret_u8_u16(tmp1_16);
+      res = vmull_u8(m, tmp0);
+      res = vmlal_u8(res, max_minus_m, tmp1);
+      vst1_lane_u16(
+          (uint16_t *)(dst + (0 * dst_stride)),
+          vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
+      vst1_lane_u16(
+          (uint16_t *)(dst + (1 * dst_stride)),
+          vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
+      src0 += (2 * src0_stride);
+      src1 += (2 * src1_stride);
+      dst += (2 * dst_stride);
+    }
+  }
+}

third_party/aom/av1/common/arm/cfl_neon.c

@@ -0,0 +1,584 @@
+/*
+ * Copyright (c) 2017, 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 <arm_neon.h>
+
+#include "config/av1_rtcd.h"
+
+#include "av1/common/cfl.h"
+
+static INLINE void vldsubstq_s16(int16_t *dst, const uint16_t *src, int offset,
+                                 int16x8_t sub) {
+  vst1q_s16(dst + offset,
+            vsubq_s16(vreinterpretq_s16_u16(vld1q_u16(src + offset)), sub));
+}
+
+static INLINE uint16x8_t vldaddq_u16(const uint16_t *buf, size_t offset) {
+  return vaddq_u16(vld1q_u16(buf), vld1q_u16(buf + offset));
+}
+
+// Load half of a vector and duplicated in other half
+static INLINE uint8x8_t vldh_dup_u8(const uint8_t *ptr) {
+  return vreinterpret_u8_u32(vld1_dup_u32((const uint32_t *)ptr));
+}
+
+// Store half of a vector.
+static INLINE void vsth_u16(uint16_t *ptr, uint16x4_t val) {
+  *((uint32_t *)ptr) = vreinterpret_u32_u16(val)[0];
+}
+
+// Store half of a vector.
+static INLINE void vsth_u8(uint8_t *ptr, uint8x8_t val) {
+  *((uint32_t *)ptr) = vreinterpret_u32_u8(val)[0];
+}
+
+static void cfl_luma_subsampling_420_lbd_neon(const uint8_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + (height >> 1) * CFL_BUF_LINE;
+  const int luma_stride = input_stride << 1;
+  do {
+    if (width == 4) {
+      const uint16x4_t top = vpaddl_u8(vldh_dup_u8(input));
+      const uint16x4_t sum = vpadal_u8(top, vldh_dup_u8(input + input_stride));
+      vsth_u16(pred_buf_q3, vshl_n_u16(sum, 1));
+    } else if (width == 8) {
+      const uint16x4_t top = vpaddl_u8(vld1_u8(input));
+      const uint16x4_t sum = vpadal_u8(top, vld1_u8(input + input_stride));
+      vst1_u16(pred_buf_q3, vshl_n_u16(sum, 1));
+    } else if (width == 16) {
+      const uint16x8_t top = vpaddlq_u8(vld1q_u8(input));
+      const uint16x8_t sum = vpadalq_u8(top, vld1q_u8(input + input_stride));
+      vst1q_u16(pred_buf_q3, vshlq_n_u16(sum, 1));
+    } else {
+      const uint8x8x4_t top = vld4_u8(input);
+      const uint8x8x4_t bot = vld4_u8(input + input_stride);
+      // equivalent to a vpaddlq_u8 (because vld4q interleaves)
+      const uint16x8_t top_0 = vaddl_u8(top.val[0], top.val[1]);
+      // equivalent to a vpaddlq_u8 (because vld4q interleaves)
+      const uint16x8_t bot_0 = vaddl_u8(bot.val[0], bot.val[1]);
+      // equivalent to a vpaddlq_u8 (because vld4q interleaves)
+      const uint16x8_t top_1 = vaddl_u8(top.val[2], top.val[3]);
+      // equivalent to a vpaddlq_u8 (because vld4q interleaves)
+      const uint16x8_t bot_1 = vaddl_u8(bot.val[2], bot.val[3]);
+      uint16x8x2_t sum;
+      sum.val[0] = vshlq_n_u16(vaddq_u16(top_0, bot_0), 1);
+      sum.val[1] = vshlq_n_u16(vaddq_u16(top_1, bot_1), 1);
+      vst2q_u16(pred_buf_q3, sum);
+    }
+    input += luma_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+static void cfl_luma_subsampling_422_lbd_neon(const uint8_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + height * CFL_BUF_LINE;
+  do {
+    if (width == 4) {
+      const uint16x4_t top = vpaddl_u8(vldh_dup_u8(input));
+      vsth_u16(pred_buf_q3, vshl_n_u16(top, 2));
+    } else if (width == 8) {
+      const uint16x4_t top = vpaddl_u8(vld1_u8(input));
+      vst1_u16(pred_buf_q3, vshl_n_u16(top, 2));
+    } else if (width == 16) {
+      const uint16x8_t top = vpaddlq_u8(vld1q_u8(input));
+      vst1q_u16(pred_buf_q3, vshlq_n_u16(top, 2));
+    } else {
+      const uint8x8x4_t top = vld4_u8(input);
+      uint16x8x2_t sum;
+      // vaddl_u8 is equivalent to a vpaddlq_u8 (because vld4q interleaves)
+      sum.val[0] = vshlq_n_u16(vaddl_u8(top.val[0], top.val[1]), 2);
+      sum.val[1] = vshlq_n_u16(vaddl_u8(top.val[2], top.val[3]), 2);
+      vst2q_u16(pred_buf_q3, sum);
+    }
+    input += input_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+static void cfl_luma_subsampling_444_lbd_neon(const uint8_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + height * CFL_BUF_LINE;
+  do {
+    if (width == 4) {
+      const uint16x8_t top = vshll_n_u8(vldh_dup_u8(input), 3);
+      vst1_u16(pred_buf_q3, vget_low_u16(top));
+    } else if (width == 8) {
+      const uint16x8_t top = vshll_n_u8(vld1_u8(input), 3);
+      vst1q_u16(pred_buf_q3, top);
+    } else {
+      const uint8x16_t top = vld1q_u8(input);
+      vst1q_u16(pred_buf_q3, vshll_n_u8(vget_low_u8(top), 3));
+      vst1q_u16(pred_buf_q3 + 8, vshll_n_u8(vget_high_u8(top), 3));
+      if (width == 32) {
+        const uint8x16_t next_top = vld1q_u8(input + 16);
+        vst1q_u16(pred_buf_q3 + 16, vshll_n_u8(vget_low_u8(next_top), 3));
+        vst1q_u16(pred_buf_q3 + 24, vshll_n_u8(vget_high_u8(next_top), 3));
+      }
+    }
+    input += input_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+#if __ARM_ARCH <= 7
+uint16x8_t vpaddq_u16(uint16x8_t a, uint16x8_t b) {
+  return vcombine_u16(vpadd_u16(vget_low_u16(a), vget_high_u16(a)),
+                      vpadd_u16(vget_low_u16(b), vget_high_u16(b)));
+}
+#endif
+
+static void cfl_luma_subsampling_420_hbd_neon(const uint16_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + (height >> 1) * CFL_BUF_LINE;
+  const int luma_stride = input_stride << 1;
+  do {
+    if (width == 4) {
+      const uint16x4_t top = vld1_u16(input);
+      const uint16x4_t bot = vld1_u16(input + input_stride);
+      const uint16x4_t sum = vadd_u16(top, bot);
+      const uint16x4_t hsum = vpadd_u16(sum, sum);
+      vsth_u16(pred_buf_q3, vshl_n_u16(hsum, 1));
+    } else if (width < 32) {
+      const uint16x8_t top = vld1q_u16(input);
+      const uint16x8_t bot = vld1q_u16(input + input_stride);
+      const uint16x8_t sum = vaddq_u16(top, bot);
+      if (width == 8) {
+        const uint16x4_t hsum = vget_low_u16(vpaddq_u16(sum, sum));
+        vst1_u16(pred_buf_q3, vshl_n_u16(hsum, 1));
+      } else {
+        const uint16x8_t top_1 = vld1q_u16(input + 8);
+        const uint16x8_t bot_1 = vld1q_u16(input + 8 + input_stride);
+        const uint16x8_t sum_1 = vaddq_u16(top_1, bot_1);
+        const uint16x8_t hsum = vpaddq_u16(sum, sum_1);
+        vst1q_u16(pred_buf_q3, vshlq_n_u16(hsum, 1));
+      }
+    } else {
+      const uint16x8x4_t top = vld4q_u16(input);
+      const uint16x8x4_t bot = vld4q_u16(input + input_stride);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t top_0 = vaddq_u16(top.val[0], top.val[1]);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t bot_0 = vaddq_u16(bot.val[0], bot.val[1]);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t top_1 = vaddq_u16(top.val[2], top.val[3]);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t bot_1 = vaddq_u16(bot.val[2], bot.val[3]);
+      uint16x8x2_t sum;
+      sum.val[0] = vshlq_n_u16(vaddq_u16(top_0, bot_0), 1);
+      sum.val[1] = vshlq_n_u16(vaddq_u16(top_1, bot_1), 1);
+      vst2q_u16(pred_buf_q3, sum);
+    }
+    input += luma_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+static void cfl_luma_subsampling_422_hbd_neon(const uint16_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + height * CFL_BUF_LINE;
+  do {
+    if (width == 4) {
+      const uint16x4_t top = vld1_u16(input);
+      const uint16x4_t hsum = vpadd_u16(top, top);
+      vsth_u16(pred_buf_q3, vshl_n_u16(hsum, 2));
+    } else if (width == 8) {
+      const uint16x4x2_t top = vld2_u16(input);
+      // equivalent to a vpadd_u16 (because vld2 interleaves)
+      const uint16x4_t hsum = vadd_u16(top.val[0], top.val[1]);
+      vst1_u16(pred_buf_q3, vshl_n_u16(hsum, 2));
+    } else if (width == 16) {
+      const uint16x8x2_t top = vld2q_u16(input);
+      // equivalent to a vpaddq_u16 (because vld2q interleaves)
+      const uint16x8_t hsum = vaddq_u16(top.val[0], top.val[1]);
+      vst1q_u16(pred_buf_q3, vshlq_n_u16(hsum, 2));
+    } else {
+      const uint16x8x4_t top = vld4q_u16(input);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t hsum_0 = vaddq_u16(top.val[0], top.val[1]);
+      // equivalent to a vpaddq_u16 (because vld4q interleaves)
+      const uint16x8_t hsum_1 = vaddq_u16(top.val[2], top.val[3]);
+      uint16x8x2_t result = { { vshlq_n_u16(hsum_0, 2),
+                                vshlq_n_u16(hsum_1, 2) } };
+      vst2q_u16(pred_buf_q3, result);
+    }
+    input += input_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+static void cfl_luma_subsampling_444_hbd_neon(const uint16_t *input,
+                                              int input_stride,
+                                              uint16_t *pred_buf_q3, int width,
+                                              int height) {
+  const uint16_t *end = pred_buf_q3 + height * CFL_BUF_LINE;
+  do {
+    if (width == 4) {
+      const uint16x4_t top = vld1_u16(input);
+      vst1_u16(pred_buf_q3, vshl_n_u16(top, 3));
+    } else if (width == 8) {
+      const uint16x8_t top = vld1q_u16(input);
+      vst1q_u16(pred_buf_q3, vshlq_n_u16(top, 3));
+    } else if (width == 16) {
+      uint16x8x2_t top = vld2q_u16(input);
+      top.val[0] = vshlq_n_u16(top.val[0], 3);
+      top.val[1] = vshlq_n_u16(top.val[1], 3);
+      vst2q_u16(pred_buf_q3, top);
+    } else {
+      uint16x8x4_t top = vld4q_u16(input);
+      top.val[0] = vshlq_n_u16(top.val[0], 3);
+      top.val[1] = vshlq_n_u16(top.val[1], 3);
+      top.val[2] = vshlq_n_u16(top.val[2], 3);
+      top.val[3] = vshlq_n_u16(top.val[3], 3);
+      vst4q_u16(pred_buf_q3, top);
+    }
+    input += input_stride;
+  } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+}
+
+CFL_GET_SUBSAMPLE_FUNCTION(neon)
+
+static INLINE void subtract_average_neon(const uint16_t *src, int16_t *dst,
+                                         int width, int height,
+                                         int round_offset,
+                                         const int num_pel_log2) {
+  const uint16_t *const end = src + height * CFL_BUF_LINE;
+
+  // Round offset is not needed, because NEON will handle the rounding.
+  (void)round_offset;
+
+  // To optimize the use of the CPU pipeline, we process 4 rows per iteration
+  const int step = 4 * CFL_BUF_LINE;
+
+  // At this stage, the prediction buffer contains scaled reconstructed luma
+  // pixels, which are positive integer and only require 15 bits. By using
+  // unsigned integer for the sum, we can do one addition operation inside 16
+  // bits (8 lanes) before having to convert to 32 bits (4 lanes).
+  const uint16_t *sum_buf = src;
+  uint32x4_t sum_32x4 = { 0, 0, 0, 0 };
+  do {
+    // For all widths, we load, add and combine the data so it fits in 4 lanes.
+    if (width == 4) {
+      const uint16x4_t a0 =
+          vadd_u16(vld1_u16(sum_buf), vld1_u16(sum_buf + CFL_BUF_LINE));
+      const uint16x4_t a1 = vadd_u16(vld1_u16(sum_buf + 2 * CFL_BUF_LINE),
+                                     vld1_u16(sum_buf + 3 * CFL_BUF_LINE));
+      sum_32x4 = vaddq_u32(sum_32x4, vaddl_u16(a0, a1));
+    } else if (width == 8) {
+      const uint16x8_t a0 = vldaddq_u16(sum_buf, CFL_BUF_LINE);
+      const uint16x8_t a1 =
+          vldaddq_u16(sum_buf + 2 * CFL_BUF_LINE, CFL_BUF_LINE);
+      sum_32x4 = vpadalq_u16(sum_32x4, a0);
+      sum_32x4 = vpadalq_u16(sum_32x4, a1);
+    } else {
+      const uint16x8_t row0 = vldaddq_u16(sum_buf, 8);
+      const uint16x8_t row1 = vldaddq_u16(sum_buf + CFL_BUF_LINE, 8);
+      const uint16x8_t row2 = vldaddq_u16(sum_buf + 2 * CFL_BUF_LINE, 8);
+      const uint16x8_t row3 = vldaddq_u16(sum_buf + 3 * CFL_BUF_LINE, 8);
+      sum_32x4 = vpadalq_u16(sum_32x4, row0);
+      sum_32x4 = vpadalq_u16(sum_32x4, row1);
+      sum_32x4 = vpadalq_u16(sum_32x4, row2);
+      sum_32x4 = vpadalq_u16(sum_32x4, row3);
+
+      if (width == 32) {
+        const uint16x8_t row0_1 = vldaddq_u16(sum_buf + 16, 8);
+        const uint16x8_t row1_1 = vldaddq_u16(sum_buf + CFL_BUF_LINE + 16, 8);
+        const uint16x8_t row2_1 =
+            vldaddq_u16(sum_buf + 2 * CFL_BUF_LINE + 16, 8);
+        const uint16x8_t row3_1 =
+            vldaddq_u16(sum_buf + 3 * CFL_BUF_LINE + 16, 8);
+
+        sum_32x4 = vpadalq_u16(sum_32x4, row0_1);
+        sum_32x4 = vpadalq_u16(sum_32x4, row1_1);
+        sum_32x4 = vpadalq_u16(sum_32x4, row2_1);
+        sum_32x4 = vpadalq_u16(sum_32x4, row3_1);
+      }
+    }
+    sum_buf += step;
+  } while (sum_buf < end);
+
+  // Permute and add in such a way that each lane contains the block sum.
+  // [A+C+B+D, B+D+A+C, C+A+D+B, D+B+C+A]
+#if __ARM_ARCH >= 8
+  sum_32x4 = vpaddq_u32(sum_32x4, sum_32x4);
+  sum_32x4 = vpaddq_u32(sum_32x4, sum_32x4);
+#else
+  uint32x4_t flip =
+      vcombine_u32(vget_high_u32(sum_32x4), vget_low_u32(sum_32x4));
+  sum_32x4 = vaddq_u32(sum_32x4, flip);
+  sum_32x4 = vaddq_u32(sum_32x4, vrev64q_u32(sum_32x4));
+#endif
+
+  // Computing the average could be done using scalars, but getting off the NEON
+  // engine introduces latency, so we use vqrshrn.
+  int16x4_t avg_16x4;
+  // Constant propagation makes for some ugly code.
+  switch (num_pel_log2) {
+    case 4: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 4)); break;
+    case 5: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 5)); break;
+    case 6: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 6)); break;
+    case 7: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 7)); break;
+    case 8: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 8)); break;
+    case 9: avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 9)); break;
+    case 10:
+      avg_16x4 = vreinterpret_s16_u16(vqrshrn_n_u32(sum_32x4, 10));
+      break;
+    default: assert(0);
+  }
+
+  if (width == 4) {
+    do {
+      vst1_s16(dst, vsub_s16(vreinterpret_s16_u16(vld1_u16(src)), avg_16x4));
+      src += CFL_BUF_LINE;
+      dst += CFL_BUF_LINE;
+    } while (src < end);
+  } else {
+    const int16x8_t avg_16x8 = vcombine_s16(avg_16x4, avg_16x4);
+    do {
+      vldsubstq_s16(dst, src, 0, avg_16x8);
+      vldsubstq_s16(dst, src, CFL_BUF_LINE, avg_16x8);
+      vldsubstq_s16(dst, src, 2 * CFL_BUF_LINE, avg_16x8);
+      vldsubstq_s16(dst, src, 3 * CFL_BUF_LINE, avg_16x8);
+
+      if (width > 8) {
+        vldsubstq_s16(dst, src, 8, avg_16x8);
+        vldsubstq_s16(dst, src, 8 + CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 8 + 2 * CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 8 + 3 * CFL_BUF_LINE, avg_16x8);
+      }
+      if (width == 32) {
+        vldsubstq_s16(dst, src, 16, avg_16x8);
+        vldsubstq_s16(dst, src, 16 + CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 16 + 2 * CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 16 + 3 * CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 24, avg_16x8);
+        vldsubstq_s16(dst, src, 24 + CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 24 + 2 * CFL_BUF_LINE, avg_16x8);
+        vldsubstq_s16(dst, src, 24 + 3 * CFL_BUF_LINE, avg_16x8);
+      }
+      src += step;
+      dst += step;
+    } while (src < end);
+  }
+}
+
+CFL_SUB_AVG_FN(neon)
+
+// Saturating negate 16-bit integers in a when the corresponding signed 16-bit
+// integer in b is negative.
+// Notes:
+//   * Negating INT16_MIN results in INT16_MIN. However, this cannot occur in
+//   practice, as scaled_luma is the multiplication of two absolute values.
+//   * In the Intel equivalent, elements in a are zeroed out when the
+//   corresponding elements in b are zero. Because vsign is used twice in a
+//   row, with b in the first call becoming a in the second call, there's no
+//   impact from not zeroing out.
+static int16x4_t vsign_s16(int16x4_t a, int16x4_t b) {
+  const int16x4_t mask = vshr_n_s16(b, 15);
+  return veor_s16(vadd_s16(a, mask), mask);
+}
+
+// Saturating negate 16-bit integers in a when the corresponding signed 16-bit
+// integer in b is negative.
+// Notes:
+//   * Negating INT16_MIN results in INT16_MIN. However, this cannot occur in
+//   practice, as scaled_luma is the multiplication of two absolute values.
+//   * In the Intel equivalent, elements in a are zeroed out when the
+//   corresponding elements in b are zero. Because vsignq is used twice in a
+//   row, with b in the first call becoming a in the second call, there's no
+//   impact from not zeroing out.
+static int16x8_t vsignq_s16(int16x8_t a, int16x8_t b) {
+  const int16x8_t mask = vshrq_n_s16(b, 15);
+  return veorq_s16(vaddq_s16(a, mask), mask);
+}
+
+static INLINE int16x4_t predict_w4(const int16_t *pred_buf_q3,
+                                   int16x4_t alpha_sign, int abs_alpha_q12,
+                                   int16x4_t dc) {
+  const int16x4_t ac_q3 = vld1_s16(pred_buf_q3);
+  const int16x4_t ac_sign = veor_s16(alpha_sign, ac_q3);
+  int16x4_t scaled_luma = vqrdmulh_n_s16(vabs_s16(ac_q3), abs_alpha_q12);
+  return vadd_s16(vsign_s16(scaled_luma, ac_sign), dc);
+}
+
+static INLINE int16x8_t predict_w8(const int16_t *pred_buf_q3,
+                                   int16x8_t alpha_sign, int abs_alpha_q12,
+                                   int16x8_t dc) {
+  const int16x8_t ac_q3 = vld1q_s16(pred_buf_q3);
+  const int16x8_t ac_sign = veorq_s16(alpha_sign, ac_q3);
+  int16x8_t scaled_luma = vqrdmulhq_n_s16(vabsq_s16(ac_q3), abs_alpha_q12);
+  return vaddq_s16(vsignq_s16(scaled_luma, ac_sign), dc);
+}
+
+static INLINE int16x8x2_t predict_w16(const int16_t *pred_buf_q3,
+                                      int16x8_t alpha_sign, int abs_alpha_q12,
+                                      int16x8_t dc) {
+  // vld2q_s16 interleaves, which is not useful for prediction. vst1q_s16_x2
+  // does not interleave, but is not currently available in the compilier used
+  // by the AOM build system.
+  const int16x8x2_t ac_q3 = vld2q_s16(pred_buf_q3);
+  const int16x8_t ac_sign_0 = veorq_s16(alpha_sign, ac_q3.val[0]);
+  const int16x8_t ac_sign_1 = veorq_s16(alpha_sign, ac_q3.val[1]);
+  const int16x8_t scaled_luma_0 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[0]), abs_alpha_q12);
+  const int16x8_t scaled_luma_1 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[1]), abs_alpha_q12);
+  int16x8x2_t result;
+  result.val[0] = vaddq_s16(vsignq_s16(scaled_luma_0, ac_sign_0), dc);
+  result.val[1] = vaddq_s16(vsignq_s16(scaled_luma_1, ac_sign_1), dc);
+  return result;
+}
+
+static INLINE int16x8x4_t predict_w32(const int16_t *pred_buf_q3,
+                                      int16x8_t alpha_sign, int abs_alpha_q12,
+                                      int16x8_t dc) {
+  // vld4q_s16 interleaves, which is not useful for prediction. vst1q_s16_x4
+  // does not interleave, but is not currently available in the compilier used
+  // by the AOM build system.
+  const int16x8x4_t ac_q3 = vld4q_s16(pred_buf_q3);
+  const int16x8_t ac_sign_0 = veorq_s16(alpha_sign, ac_q3.val[0]);
+  const int16x8_t ac_sign_1 = veorq_s16(alpha_sign, ac_q3.val[1]);
+  const int16x8_t ac_sign_2 = veorq_s16(alpha_sign, ac_q3.val[2]);
+  const int16x8_t ac_sign_3 = veorq_s16(alpha_sign, ac_q3.val[3]);
+  const int16x8_t scaled_luma_0 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[0]), abs_alpha_q12);
+  const int16x8_t scaled_luma_1 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[1]), abs_alpha_q12);
+  const int16x8_t scaled_luma_2 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[2]), abs_alpha_q12);
+  const int16x8_t scaled_luma_3 =
+      vqrdmulhq_n_s16(vabsq_s16(ac_q3.val[3]), abs_alpha_q12);
+  int16x8x4_t result;
+  result.val[0] = vaddq_s16(vsignq_s16(scaled_luma_0, ac_sign_0), dc);
+  result.val[1] = vaddq_s16(vsignq_s16(scaled_luma_1, ac_sign_1), dc);
+  result.val[2] = vaddq_s16(vsignq_s16(scaled_luma_2, ac_sign_2), dc);
+  result.val[3] = vaddq_s16(vsignq_s16(scaled_luma_3, ac_sign_3), dc);
+  return result;
+}
+
+static INLINE void cfl_predict_lbd_neon(const int16_t *pred_buf_q3,
+                                        uint8_t *dst, int dst_stride,
+                                        int alpha_q3, int width, int height) {
+  const int16_t abs_alpha_q12 = abs(alpha_q3) << 9;
+  const int16_t *const end = pred_buf_q3 + height * CFL_BUF_LINE;
+  if (width == 4) {
+    const int16x4_t alpha_sign = vdup_n_s16(alpha_q3);
+    const int16x4_t dc = vdup_n_s16(*dst);
+    do {
+      const int16x4_t pred =
+          predict_w4(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+      vsth_u8(dst, vqmovun_s16(vcombine_s16(pred, pred)));
+      dst += dst_stride;
+    } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+  } else {
+    const int16x8_t alpha_sign = vdupq_n_s16(alpha_q3);
+    const int16x8_t dc = vdupq_n_s16(*dst);
+    do {
+      if (width == 8) {
+        vst1_u8(dst, vqmovun_s16(predict_w8(pred_buf_q3, alpha_sign,
+                                            abs_alpha_q12, dc)));
+      } else if (width == 16) {
+        const int16x8x2_t pred =
+            predict_w16(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+        const uint8x8x2_t predun = { { vqmovun_s16(pred.val[0]),
+                                       vqmovun_s16(pred.val[1]) } };
+        vst2_u8(dst, predun);
+      } else {
+        const int16x8x4_t pred =
+            predict_w32(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+        const uint8x8x4_t predun = {
+          { vqmovun_s16(pred.val[0]), vqmovun_s16(pred.val[1]),
+            vqmovun_s16(pred.val[2]), vqmovun_s16(pred.val[3]) }
+        };
+        vst4_u8(dst, predun);
+      }
+      dst += dst_stride;
+    } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+  }
+}
+
+CFL_PREDICT_FN(neon, lbd)
+
+static INLINE uint16x4_t clamp_s16(int16x4_t a, int16x4_t max) {
+  return vreinterpret_u16_s16(vmax_s16(vmin_s16(a, max), vdup_n_s16(0)));
+}
+
+static INLINE uint16x8_t clampq_s16(int16x8_t a, int16x8_t max) {
+  return vreinterpretq_u16_s16(vmaxq_s16(vminq_s16(a, max), vdupq_n_s16(0)));
+}
+
+static INLINE uint16x8x2_t clamp2q_s16(int16x8x2_t a, int16x8_t max) {
+  uint16x8x2_t result;
+  result.val[0] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[0], max), vdupq_n_s16(0)));
+  result.val[1] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[1], max), vdupq_n_s16(0)));
+  return result;
+}
+
+static INLINE uint16x8x4_t clamp4q_s16(int16x8x4_t a, int16x8_t max) {
+  uint16x8x4_t result;
+  result.val[0] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[0], max), vdupq_n_s16(0)));
+  result.val[1] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[1], max), vdupq_n_s16(0)));
+  result.val[2] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[2], max), vdupq_n_s16(0)));
+  result.val[3] = vreinterpretq_u16_s16(
+      vmaxq_s16(vminq_s16(a.val[3], max), vdupq_n_s16(0)));
+  return result;
+}
+
+static INLINE void cfl_predict_hbd_neon(const int16_t *pred_buf_q3,
+                                        uint16_t *dst, int dst_stride,
+                                        int alpha_q3, int bd, int width,
+                                        int height) {
+  const int max = (1 << bd) - 1;
+  const int16_t abs_alpha_q12 = abs(alpha_q3) << 9;
+  const int16_t *const end = pred_buf_q3 + height * CFL_BUF_LINE;
+  if (width == 4) {
+    const int16x4_t alpha_sign = vdup_n_s16(alpha_q3);
+    const int16x4_t dc = vdup_n_s16(*dst);
+    const int16x4_t max_16x4 = vdup_n_s16(max);
+    do {
+      const int16x4_t scaled_luma =
+          predict_w4(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+      vst1_u16(dst, clamp_s16(scaled_luma, max_16x4));
+      dst += dst_stride;
+    } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+  } else {
+    const int16x8_t alpha_sign = vdupq_n_s16(alpha_q3);
+    const int16x8_t dc = vdupq_n_s16(*dst);
+    const int16x8_t max_16x8 = vdupq_n_s16(max);
+    do {
+      if (width == 8) {
+        const int16x8_t pred =
+            predict_w8(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+        vst1q_u16(dst, clampq_s16(pred, max_16x8));
+      } else if (width == 16) {
+        const int16x8x2_t pred =
+            predict_w16(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+        vst2q_u16(dst, clamp2q_s16(pred, max_16x8));
+      } else {
+        const int16x8x4_t pred =
+            predict_w32(pred_buf_q3, alpha_sign, abs_alpha_q12, dc);
+        vst4q_u16(dst, clamp4q_s16(pred, max_16x8));
+      }
+      dst += dst_stride;
+    } while ((pred_buf_q3 += CFL_BUF_LINE) < end);
+  }
+}
+
+CFL_PREDICT_FN(neon, hbd)

third_party/aom/av1/common/arm/convolve_neon.h

@@ -0,0 +1,228 @@
+/*
+ *  Copyright (c) 2018, Alliance for Open Media. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef AV1_COMMON_ARM_CONVOLVE_NEON_H_
+#define AV1_COMMON_ARM_CONVOLVE_NEON_H_
+
+#include <arm_neon.h>
+
+#define HORIZ_EXTRA_ROWS ((SUBPEL_TAPS + 7) & ~0x07)
+
+static INLINE uint8x8_t wiener_convolve8_vert_4x8(
+    const int16x8_t s0, const int16x8_t s1, const int16x8_t s2,
+    const int16x8_t s3, const int16x8_t s4, const int16x8_t s5,
+    const int16x8_t s6, int16_t *filter_y, const int bd,
+    const int round1_bits) {
+  int16x8_t ss0, ss1, ss2;
+  int32x4_t sum0, sum1;
+  uint16x4_t tmp0, tmp1;
+  uint16x8_t tmp;
+  uint8x8_t res;
+
+  const int32_t round_const = (1 << (bd + round1_bits - 1));
+  const int32x4_t round_bits = vdupq_n_s32(-round1_bits);
+  const int32x4_t zero = vdupq_n_s32(0);
+  const int32x4_t round_vec = vdupq_n_s32(round_const);
+
+  ss0 = vaddq_s16(s0, s6);
+  ss1 = vaddq_s16(s1, s5);
+  ss2 = vaddq_s16(s2, s4);
+
+  sum0 = vmull_n_s16(vget_low_s16(ss0), filter_y[0]);
+  sum0 = vmlal_n_s16(sum0, vget_low_s16(ss1), filter_y[1]);
+  sum0 = vmlal_n_s16(sum0, vget_low_s16(ss2), filter_y[2]);
+  sum0 = vmlal_n_s16(sum0, vget_low_s16(s3), filter_y[3]);
+
+  sum1 = vmull_n_s16(vget_high_s16(ss0), filter_y[0]);
+  sum1 = vmlal_n_s16(sum1, vget_high_s16(ss1), filter_y[1]);
+  sum1 = vmlal_n_s16(sum1, vget_high_s16(ss2), filter_y[2]);
+  sum1 = vmlal_n_s16(sum1, vget_high_s16(s3), filter_y[3]);
+
+  sum0 = vsubq_s32(sum0, round_vec);
+  sum1 = vsubq_s32(sum1, round_vec);
+
+  /* right shift & rounding */
+  sum0 = vrshlq_s32(sum0, round_bits);
+  sum1 = vrshlq_s32(sum1, round_bits);
+
+  sum0 = vmaxq_s32(sum0, zero);
+  sum1 = vmaxq_s32(sum1, zero);
+
+  /* from int32x4_t to uint8x8_t */
+  tmp0 = vqmovn_u32(vreinterpretq_u32_s32(sum0));
+  tmp1 = vqmovn_u32(vreinterpretq_u32_s32(sum1));
+  tmp = vcombine_u16(tmp0, tmp1);
+  res = vqmovn_u16(tmp);
+
+  return res;
+}
+
+static INLINE uint16x8_t wiener_convolve8_horiz_8x8(
+    const int16x8_t s0, const int16x8_t s1, const int16x8_t s2,
+    const int16x8_t s3, int16_t *filter_x, const int bd,
+    const int round0_bits) {
+  int16x8_t sum;
+  uint16x8_t res;
+  int32x4_t sum_0, sum_1;
+  int32x4_t s3_0, s3_1;
+  const int32_t round_const_0 = (1 << (bd + FILTER_BITS - 1));
+  const int32_t round_const_1 = (1 << ((bd) + 1 + FILTER_BITS - round0_bits));
+
+  /* for the purpose of right shift by { conv_params->round_0 } */
+  const int32x4_t round_bits = vdupq_n_s32(-round0_bits);
+
+  const int32x4_t round_vec_0 = vdupq_n_s32(round_const_0);
+  const int32x4_t round_vec_1 = vdupq_n_s32(round_const_1);
+
+  sum = vmulq_n_s16(s0, filter_x[0]);
+  sum = vmlaq_n_s16(sum, s1, filter_x[1]);
+  sum = vmlaq_n_s16(sum, s2, filter_x[2]);
+
+  /* sum from 16x8 to 2 32x4 registers */
+  sum_0 = vmovl_s16(vget_low_s16(sum));
+  sum_1 = vmovl_s16(vget_high_s16(sum));
+
+  /* s[3]*128 -- and filter coef max can be 128
+   *  then max value possible = 128*128*255 exceeding 16 bit
+   */
+
+  s3_0 = vmull_n_s16(vget_low_s16(s3), filter_x[3]);
+  s3_1 = vmull_n_s16(vget_high_s16(s3), filter_x[3]);
+  sum_0 = vaddq_s32(sum_0, s3_0);
+  sum_1 = vaddq_s32(sum_1, s3_1);
+
+  /* Add the constant value */
+  sum_0 = vaddq_s32(sum_0, round_vec_0);
+  sum_1 = vaddq_s32(sum_1, round_vec_0);
+
+  /* right shift & rounding & saturating */
+  sum_0 = vqrshlq_s32(sum_0, round_bits);
+  sum_1 = vqrshlq_s32(sum_1, round_bits);
+
+  /* Clipping to max value */
+  sum_0 = vminq_s32(sum_0, round_vec_1);
+  sum_1 = vminq_s32(sum_1, round_vec_1);
+
+  res = vcombine_u16(vqmovun_s32(sum_0), vqmovun_s32(sum_1));
+  return res;
+}
+
+static INLINE uint16x4_t wiener_convolve8_horiz_4x8(
+    const int16x4_t s0, const int16x4_t s1, const int16x4_t s2,
+    const int16x4_t s3, const int16x4_t s4, const int16x4_t s5,
+    const int16x4_t s6, int16_t *filter_x, const int bd,
+    const int round0_bits) {
+  uint16x4_t res;
+  int32x4_t sum_0, s3_0;
+  int16x4_t sum, temp0, temp1, temp2;
+
+  const int32_t round_const_0 = (1 << (bd + FILTER_BITS - 1));
+  const int32_t round_const_1 = (1 << ((bd) + 1 + FILTER_BITS - round0_bits));
+  const int32x4_t round_bits = vdupq_n_s32(-round0_bits);
+  const int32x4_t zero = vdupq_n_s32(0);
+  const int32x4_t round_vec_0 = vdupq_n_s32(round_const_0);
+  const int32x4_t round_vec_1 = vdupq_n_s32(round_const_1);
+
+  temp0 = vadd_s16(s0, s6);
+  temp1 = vadd_s16(s1, s5);
+  temp2 = vadd_s16(s2, s4);
+
+  sum = vmul_n_s16(temp0, filter_x[0]);
+  sum = vmla_n_s16(sum, temp1, filter_x[1]);
+  sum = vmla_n_s16(sum, temp2, filter_x[2]);
+  sum_0 = vmovl_s16(sum);
+
+  /* s[3]*128 -- and filter coff max can be 128.
+   * then max value possible = 128*128*255 Therefore, 32 bits are required to
+   * hold the result.
+   */
+  s3_0 = vmull_n_s16(s3, filter_x[3]);
+  sum_0 = vaddq_s32(sum_0, s3_0);
+
+  sum_0 = vaddq_s32(sum_0, round_vec_0);
+  sum_0 = vrshlq_s32(sum_0, round_bits);
+
+  sum_0 = vmaxq_s32(sum_0, zero);
+  sum_0 = vminq_s32(sum_0, round_vec_1);
+  res = vqmovun_s32(sum_0);
+  return res;
+}
+
+static INLINE int16x8_t
+convolve8_8x8_s16(const int16x8_t s0, const int16x8_t s1, const int16x8_t s2,
+                  const int16x8_t s3, const int16x8_t s4, const int16x8_t s5,
+                  const int16x8_t s6, const int16x8_t s7, const int16_t *filter,
+                  const int16x8_t horiz_const, const int16x8_t shift_round_0) {
+  int16x8_t sum;
+  int16x8_t res;
+
+  sum = horiz_const;
+  sum = vmlaq_n_s16(sum, s0, filter[0]);
+  sum = vmlaq_n_s16(sum, s1, filter[1]);
+  sum = vmlaq_n_s16(sum, s2, filter[2]);
+  sum = vmlaq_n_s16(sum, s3, filter[3]);
+  sum = vmlaq_n_s16(sum, s4, filter[4]);
+  sum = vmlaq_n_s16(sum, s5, filter[5]);
+  sum = vmlaq_n_s16(sum, s6, filter[6]);
+  sum = vmlaq_n_s16(sum, s7, filter[7]);
+
+  res = vqrshlq_s16(sum, shift_round_0);
+
+  return res;
+}
+
+static INLINE int16x4_t
+convolve8_4x4_s16(const int16x4_t s0, const int16x4_t s1, const int16x4_t s2,
+                  const int16x4_t s3, const int16x4_t s4, const int16x4_t s5,
+                  const int16x4_t s6, const int16x4_t s7, const int16_t *filter,
+                  const int16x4_t horiz_const, const int16x4_t shift_round_0) {
+  int16x4_t sum;
+  sum = horiz_const;
+  sum = vmla_n_s16(sum, s0, filter[0]);
+  sum = vmla_n_s16(sum, s1, filter[1]);
+  sum = vmla_n_s16(sum, s2, filter[2]);
+  sum = vmla_n_s16(sum, s3, filter[3]);
+  sum = vmla_n_s16(sum, s4, filter[4]);
+  sum = vmla_n_s16(sum, s5, filter[5]);
+  sum = vmla_n_s16(sum, s6, filter[6]);
+  sum = vmla_n_s16(sum, s7, filter[7]);
+
+  sum = vqrshl_s16(sum, shift_round_0);
+
+  return sum;
+}
+
+static INLINE uint16x4_t convolve8_4x4_s32(
+    const int16x4_t s0, const int16x4_t s1, const int16x4_t s2,
+    const int16x4_t s3, const int16x4_t s4, const int16x4_t s5,
+    const int16x4_t s6, const int16x4_t s7, const int16_t *y_filter,
+    const int32x4_t round_shift_vec, const int32x4_t offset_const) {
+  int32x4_t sum0;
+  uint16x4_t res;
+  const int32x4_t zero = vdupq_n_s32(0);
+
+  sum0 = vmull_n_s16(s0, y_filter[0]);
+  sum0 = vmlal_n_s16(sum0, s1, y_filter[1]);
+  sum0 = vmlal_n_s16(sum0, s2, y_filter[2]);
+  sum0 = vmlal_n_s16(sum0, s3, y_filter[3]);
+  sum0 = vmlal_n_s16(sum0, s4, y_filter[4]);
+  sum0 = vmlal_n_s16(sum0, s5, y_filter[5]);
+  sum0 = vmlal_n_s16(sum0, s6, y_filter[6]);
+  sum0 = vmlal_n_s16(sum0, s7, y_filter[7]);
+
+  sum0 = vaddq_s32(sum0, offset_const);
+  sum0 = vqrshlq_s32(sum0, round_shift_vec);
+  sum0 = vmaxq_s32(sum0, zero);
+  res = vmovn_u32(vreinterpretq_u32_s32(sum0));
+
+  return res;
+}
+
+#endif  // AV1_COMMON_ARM_CONVOLVE_NEON_H_

third_party/aom/av1/common/arm/intrapred_neon.c

@@ -0,0 +1,79 @@
+/*
+ *
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/mem.h"
+#include "av1/common/arm/mem_neon.h"
+#include "config/aom_dsp_rtcd.h"
+
+static INLINE void highbd_dc_predictor_neon(uint16_t *dst, ptrdiff_t stride,
+                                            int bw, const uint16_t *above,
+                                            const uint16_t *left) {
+  assert(bw >= 4);
+  assert(IS_POWER_OF_TWO(bw));
+  int expected_dc, sum = 0;
+  const int count = bw * 2;
+  uint32x4_t sum_q = vdupq_n_u32(0);
+  uint32x2_t sum_d;
+  uint16_t *dst_1;
+  if (bw >= 8) {
+    for (int i = 0; i < bw; i += 8) {
+      sum_q = vpadalq_u16(sum_q, vld1q_u16(above));
+      sum_q = vpadalq_u16(sum_q, vld1q_u16(left));
+      above += 8;
+      left += 8;
+    }
+    sum_d = vadd_u32(vget_low_u32(sum_q), vget_high_u32(sum_q));
+    sum = vget_lane_s32(vreinterpret_s32_u64(vpaddl_u32(sum_d)), 0);
+    expected_dc = (sum + (count >> 1)) / count;
+    const uint16x8_t dc = vdupq_n_u16((uint16_t)expected_dc);
+    for (int r = 0; r < bw; r++) {
+      dst_1 = dst;
+      for (int i = 0; i < bw; i += 8) {
+        vst1q_u16(dst_1, dc);
+        dst_1 += 8;
+      }
+      dst += stride;
+    }
+  } else {  // 4x4
+    sum_q = vaddl_u16(vld1_u16(above), vld1_u16(left));
+    sum_d = vadd_u32(vget_low_u32(sum_q), vget_high_u32(sum_q));
+    sum = vget_lane_s32(vreinterpret_s32_u64(vpaddl_u32(sum_d)), 0);
+    expected_dc = (sum + (count >> 1)) / count;
+    const uint16x4_t dc = vdup_n_u16((uint16_t)expected_dc);
+    for (int r = 0; r < bw; r++) {
+      vst1_u16(dst, dc);
+      dst += stride;
+    }
+  }
+}
+
+#define intra_pred_highbd_sized(type, width)                         \
+  void aom_highbd_##type##_predictor_##width##x##width##_neon(       \
+      uint16_t *dst, ptrdiff_t stride, const uint16_t *above,        \
+      const uint16_t *left, int bd) {                                \
+    (void)bd;                                                        \
+    highbd_##type##_predictor_neon(dst, stride, width, above, left); \
+  }
+
+#define intra_pred_square(type)      \
+  intra_pred_highbd_sized(type, 4);  \
+  intra_pred_highbd_sized(type, 8);  \
+  intra_pred_highbd_sized(type, 16); \
+  intra_pred_highbd_sized(type, 32); \
+  intra_pred_highbd_sized(type, 64);
+
+intra_pred_square(dc);
+
+#undef intra_pred_square

third_party/aom/av1/common/arm/mem_neon.h

@@ -0,0 +1,401 @@
+/*
+ *  Copyright (c) 2018, Alliance for Open Media. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef AV1_COMMON_ARM_MEM_NEON_H_
+#define AV1_COMMON_ARM_MEM_NEON_H_
+
+#include <arm_neon.h>
+#include <string.h>
+
+static INLINE void store_row2_u8_8x8(uint8_t *s, int p, const uint8x8_t s0,
+                                     const uint8x8_t s1) {
+  vst1_u8(s, s0);
+  s += p;
+  vst1_u8(s, s1);
+  s += p;
+}
+
+static INLINE void load_u8_8x8(const uint8_t *s, ptrdiff_t p,
+                               uint8x8_t *const s0, uint8x8_t *const s1,
+                               uint8x8_t *const s2, uint8x8_t *const s3,
+                               uint8x8_t *const s4, uint8x8_t *const s5,
+                               uint8x8_t *const s6, uint8x8_t *const s7) {
+  *s0 = vld1_u8(s);
+  s += p;
+  *s1 = vld1_u8(s);
+  s += p;
+  *s2 = vld1_u8(s);
+  s += p;
+  *s3 = vld1_u8(s);
+  s += p;
+  *s4 = vld1_u8(s);
+  s += p;
+  *s5 = vld1_u8(s);
+  s += p;
+  *s6 = vld1_u8(s);
+  s += p;
+  *s7 = vld1_u8(s);
+}
+
+static INLINE void load_u8_8x16(const uint8_t *s, ptrdiff_t p,
+                                uint8x16_t *const s0, uint8x16_t *const s1,
+                                uint8x16_t *const s2, uint8x16_t *const s3) {
+  *s0 = vld1q_u8(s);
+  s += p;
+  *s1 = vld1q_u8(s);
+  s += p;
+  *s2 = vld1q_u8(s);
+  s += p;
+  *s3 = vld1q_u8(s);
+}
+
+static INLINE void load_u8_8x4(const uint8_t *s, const ptrdiff_t p,
+                               uint8x8_t *const s0, uint8x8_t *const s1,
+                               uint8x8_t *const s2, uint8x8_t *const s3) {
+  *s0 = vld1_u8(s);
+  s += p;
+  *s1 = vld1_u8(s);
+  s += p;
+  *s2 = vld1_u8(s);
+  s += p;
+  *s3 = vld1_u8(s);
+}
+
+static INLINE void load_u16_4x4(const uint16_t *s, const ptrdiff_t p,
+                                uint16x4_t *const s0, uint16x4_t *const s1,
+                                uint16x4_t *const s2, uint16x4_t *const s3) {
+  *s0 = vld1_u16(s);
+  s += p;
+  *s1 = vld1_u16(s);
+  s += p;
+  *s2 = vld1_u16(s);
+  s += p;
+  *s3 = vld1_u16(s);
+  s += p;
+}
+
+static INLINE void load_u16_8x4(const uint16_t *s, const ptrdiff_t p,
+                                uint16x8_t *const s0, uint16x8_t *const s1,
+                                uint16x8_t *const s2, uint16x8_t *const s3) {
+  *s0 = vld1q_u16(s);
+  s += p;
+  *s1 = vld1q_u16(s);
+  s += p;
+  *s2 = vld1q_u16(s);
+  s += p;
+  *s3 = vld1q_u16(s);
+  s += p;
+}
+
+static INLINE void load_s16_4x8(const int16_t *s, ptrdiff_t p,
+                                int16x4_t *const s0, int16x4_t *const s1,
+                                int16x4_t *const s2, int16x4_t *const s3,
+                                int16x4_t *const s4, int16x4_t *const s5,
+                                int16x4_t *const s6, int16x4_t *const s7) {
+  *s0 = vld1_s16(s);
+  s += p;
+  *s1 = vld1_s16(s);
+  s += p;
+  *s2 = vld1_s16(s);
+  s += p;
+  *s3 = vld1_s16(s);
+  s += p;
+  *s4 = vld1_s16(s);
+  s += p;
+  *s5 = vld1_s16(s);
+  s += p;
+  *s6 = vld1_s16(s);
+  s += p;
+  *s7 = vld1_s16(s);
+}
+
+static INLINE void load_s16_4x4(const int16_t *s, ptrdiff_t p,
+                                int16x4_t *const s0, int16x4_t *const s1,
+                                int16x4_t *const s2, int16x4_t *const s3) {
+  *s0 = vld1_s16(s);
+  s += p;
+  *s1 = vld1_s16(s);
+  s += p;
+  *s2 = vld1_s16(s);
+  s += p;
+  *s3 = vld1_s16(s);
+}
+
+static INLINE void store_u8_8x8(uint8_t *s, ptrdiff_t p, const uint8x8_t s0,
+                                const uint8x8_t s1, const uint8x8_t s2,
+                                const uint8x8_t s3, const uint8x8_t s4,
+                                const uint8x8_t s5, const uint8x8_t s6,
+                                const uint8x8_t s7) {
+  vst1_u8(s, s0);
+  s += p;
+  vst1_u8(s, s1);
+  s += p;
+  vst1_u8(s, s2);
+  s += p;
+  vst1_u8(s, s3);
+  s += p;
+  vst1_u8(s, s4);
+  s += p;
+  vst1_u8(s, s5);
+  s += p;
+  vst1_u8(s, s6);
+  s += p;
+  vst1_u8(s, s7);
+}
+
+static INLINE void store_u8_8x4(uint8_t *s, ptrdiff_t p, const uint8x8_t s0,
+                                const uint8x8_t s1, const uint8x8_t s2,
+                                const uint8x8_t s3) {
+  vst1_u8(s, s0);
+  s += p;
+  vst1_u8(s, s1);
+  s += p;
+  vst1_u8(s, s2);
+  s += p;
+  vst1_u8(s, s3);
+}
+
+static INLINE void store_u8_8x16(uint8_t *s, ptrdiff_t p, const uint8x16_t s0,
+                                 const uint8x16_t s1, const uint8x16_t s2,
+                                 const uint8x16_t s3) {
+  vst1q_u8(s, s0);
+  s += p;
+  vst1q_u8(s, s1);
+  s += p;
+  vst1q_u8(s, s2);
+  s += p;
+  vst1q_u8(s, s3);
+}
+
+static INLINE void store_u16_8x8(uint16_t *s, ptrdiff_t dst_stride,
+                                 const uint16x8_t s0, const uint16x8_t s1,
+                                 const uint16x8_t s2, const uint16x8_t s3,
+                                 const uint16x8_t s4, const uint16x8_t s5,
+                                 const uint16x8_t s6, const uint16x8_t s7) {
+  vst1q_u16(s, s0);
+  s += dst_stride;
+  vst1q_u16(s, s1);
+  s += dst_stride;
+  vst1q_u16(s, s2);
+  s += dst_stride;
+  vst1q_u16(s, s3);
+  s += dst_stride;
+  vst1q_u16(s, s4);
+  s += dst_stride;
+  vst1q_u16(s, s5);
+  s += dst_stride;
+  vst1q_u16(s, s6);
+  s += dst_stride;
+  vst1q_u16(s, s7);
+}
+
+static INLINE void store_u16_4x4(uint16_t *s, ptrdiff_t dst_stride,
+                                 const uint16x4_t s0, const uint16x4_t s1,
+                                 const uint16x4_t s2, const uint16x4_t s3) {
+  vst1_u16(s, s0);
+  s += dst_stride;
+  vst1_u16(s, s1);
+  s += dst_stride;
+  vst1_u16(s, s2);
+  s += dst_stride;
+  vst1_u16(s, s3);
+}
+
+static INLINE void store_u16_8x4(uint16_t *s, ptrdiff_t dst_stride,
+                                 const uint16x8_t s0, const uint16x8_t s1,
+                                 const uint16x8_t s2, const uint16x8_t s3) {
+  vst1q_u16(s, s0);
+  s += dst_stride;
+  vst1q_u16(s, s1);
+  s += dst_stride;
+  vst1q_u16(s, s2);
+  s += dst_stride;
+  vst1q_u16(s, s3);
+}
+
+static INLINE void store_s16_8x8(int16_t *s, ptrdiff_t dst_stride,
+                                 const int16x8_t s0, const int16x8_t s1,
+                                 const int16x8_t s2, const int16x8_t s3,
+                                 const int16x8_t s4, const int16x8_t s5,
+                                 const int16x8_t s6, const int16x8_t s7) {
+  vst1q_s16(s, s0);
+  s += dst_stride;
+  vst1q_s16(s, s1);
+  s += dst_stride;
+  vst1q_s16(s, s2);
+  s += dst_stride;
+  vst1q_s16(s, s3);
+  s += dst_stride;
+  vst1q_s16(s, s4);
+  s += dst_stride;
+  vst1q_s16(s, s5);
+  s += dst_stride;
+  vst1q_s16(s, s6);
+  s += dst_stride;
+  vst1q_s16(s, s7);
+}
+
+static INLINE void load_s16_8x8(const int16_t *s, ptrdiff_t p,
+                                int16x8_t *const s0, int16x8_t *const s1,
+                                int16x8_t *const s2, int16x8_t *const s3,
+                                int16x8_t *const s4, int16x8_t *const s5,
+                                int16x8_t *const s6, int16x8_t *const s7) {
+  *s0 = vld1q_s16(s);
+  s += p;
+  *s1 = vld1q_s16(s);
+  s += p;
+  *s2 = vld1q_s16(s);
+  s += p;
+  *s3 = vld1q_s16(s);
+  s += p;
+  *s4 = vld1q_s16(s);
+  s += p;
+  *s5 = vld1q_s16(s);
+  s += p;
+  *s6 = vld1q_s16(s);
+  s += p;
+  *s7 = vld1q_s16(s);
+}
+
+static INLINE void load_s16_8x4(const int16_t *s, ptrdiff_t p,
+                                int16x8_t *const s0, int16x8_t *const s1,
+                                int16x8_t *const s2, int16x8_t *const s3) {
+  *s0 = vld1q_s16(s);
+  s += p;
+  *s1 = vld1q_s16(s);
+  s += p;
+  *s2 = vld1q_s16(s);
+  s += p;
+  *s3 = vld1q_s16(s);
+}
+
+static INLINE void load_unaligned_u8_4x8(const uint8_t *buf, int stride,
+                                         uint32x2_t *tu0, uint32x2_t *tu1,
+                                         uint32x2_t *tu2, uint32x2_t *tu3) {
+  uint32_t a;
+
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 0);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 1);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu1 = vset_lane_u32(a, *tu1, 0);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu1 = vset_lane_u32(a, *tu1, 1);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu2 = vset_lane_u32(a, *tu2, 0);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu2 = vset_lane_u32(a, *tu2, 1);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu3 = vset_lane_u32(a, *tu3, 0);
+  memcpy(&a, buf, 4);
+  *tu3 = vset_lane_u32(a, *tu3, 1);
+}
+
+static INLINE void load_unaligned_u8_4x4(const uint8_t *buf, int stride,
+                                         uint32x2_t *tu0, uint32x2_t *tu1) {
+  uint32_t a;
+
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 0);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 1);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu1 = vset_lane_u32(a, *tu1, 0);
+  memcpy(&a, buf, 4);
+  *tu1 = vset_lane_u32(a, *tu1, 1);
+}
+
+static INLINE void load_unaligned_u8_4x2(const uint8_t *buf, int stride,
+                                         uint32x2_t *tu0) {
+  uint32_t a;
+
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 0);
+  memcpy(&a, buf, 4);
+  buf += stride;
+  *tu0 = vset_lane_u32(a, *tu0, 1);
+}
+
+static INLINE void load_unaligned_u8_2x2(const uint8_t *buf, int stride,
+                                         uint16x4_t *tu0) {
+  uint16_t a;
+
+  memcpy(&a, buf, 2);
+  buf += stride;
+  *tu0 = vset_lane_u16(a, *tu0, 0);
+  memcpy(&a, buf, 2);
+  buf += stride;
+  *tu0 = vset_lane_u16(a, *tu0, 1);
+}
+
+static INLINE void load_u8_16x8(const uint8_t *s, ptrdiff_t p,
+                                uint8x16_t *const s0, uint8x16_t *const s1,
+                                uint8x16_t *const s2, uint8x16_t *const s3,
+                                uint8x16_t *const s4, uint8x16_t *const s5,
+                                uint8x16_t *const s6, uint8x16_t *const s7) {
+  *s0 = vld1q_u8(s);
+  s += p;
+  *s1 = vld1q_u8(s);
+  s += p;
+  *s2 = vld1q_u8(s);
+  s += p;
+  *s3 = vld1q_u8(s);
+  s += p;
+  *s4 = vld1q_u8(s);
+  s += p;
+  *s5 = vld1q_u8(s);
+  s += p;
+  *s6 = vld1q_u8(s);
+  s += p;
+  *s7 = vld1q_u8(s);
+}
+
+static INLINE void load_u8_16x4(const uint8_t *s, ptrdiff_t p,
+                                uint8x16_t *const s0, uint8x16_t *const s1,
+                                uint8x16_t *const s2, uint8x16_t *const s3) {
+  *s0 = vld1q_u8(s);
+  s += p;
+  *s1 = vld1q_u8(s);
+  s += p;
+  *s2 = vld1q_u8(s);
+  s += p;
+  *s3 = vld1q_u8(s);
+}
+
+static INLINE void load_unaligned_u16_4x4(const uint16_t *buf, uint32_t stride,
+                                          uint64x2_t *tu0, uint64x2_t *tu1) {
+  uint64_t a;
+
+  memcpy(&a, buf, 8);
+  buf += stride;
+  *tu0 = vsetq_lane_u64(a, *tu0, 0);
+  memcpy(&a, buf, 8);
+  buf += stride;
+  *tu0 = vsetq_lane_u64(a, *tu0, 1);
+  memcpy(&a, buf, 8);
+  buf += stride;
+  *tu1 = vsetq_lane_u64(a, *tu1, 0);
+  memcpy(&a, buf, 8);
+  *tu1 = vsetq_lane_u64(a, *tu1, 1);
+}
+
+#endif  // AV1_COMMON_ARM_MEM_NEON_H_

third_party/aom/av1/common/arm/neon/iht4x4_add_neon.c

@@ -1,228 +0,0 @@
-/*
- * 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 <arm_neon.h>
-#include <assert.h>
-
-#include "./aom_config.h"
-#include "./av1_rtcd.h"
-#include "aom_dsp/txfm_common.h"
-#include "av1/common/common.h"
-
-static INLINE void TRANSPOSE4X4(int16x8_t *q8s16, int16x8_t *q9s16) {
-  int32x4_t q8s32, q9s32;
-  int16x4x2_t d0x2s16, d1x2s16;
-  int32x4x2_t q0x2s32;
-
-  d0x2s16 = vtrn_s16(vget_low_s16(*q8s16), vget_high_s16(*q8s16));
-  d1x2s16 = vtrn_s16(vget_low_s16(*q9s16), vget_high_s16(*q9s16));
-
-  q8s32 = vreinterpretq_s32_s16(vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]));
-  q9s32 = vreinterpretq_s32_s16(vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]));
-  q0x2s32 = vtrnq_s32(q8s32, q9s32);
-
-  *q8s16 = vreinterpretq_s16_s32(q0x2s32.val[0]);
-  *q9s16 = vreinterpretq_s16_s32(q0x2s32.val[1]);
-  return;
-}
-
-static INLINE void GENERATE_COSINE_CONSTANTS(int16x4_t *d0s16, int16x4_t *d1s16,
-                                             int16x4_t *d2s16) {
-  *d0s16 = vdup_n_s16((int16_t)cospi_8_64);
-  *d1s16 = vdup_n_s16((int16_t)cospi_16_64);
-  *d2s16 = vdup_n_s16((int16_t)cospi_24_64);
-  return;
-}
-
-static INLINE void GENERATE_SINE_CONSTANTS(int16x4_t *d3s16, int16x4_t *d4s16,
-                                           int16x4_t *d5s16, int16x8_t *q3s16) {
-  *d3s16 = vdup_n_s16((int16_t)sinpi_1_9);
-  *d4s16 = vdup_n_s16((int16_t)sinpi_2_9);
-  *q3s16 = vdupq_n_s16((int16_t)sinpi_3_9);
-  *d5s16 = vdup_n_s16((int16_t)sinpi_4_9);
-  return;
-}
-
-static INLINE void IDCT4x4_1D(int16x4_t *d0s16, int16x4_t *d1s16,
-                              int16x4_t *d2s16, int16x8_t *q8s16,
-                              int16x8_t *q9s16) {
-  int16x4_t d16s16, d17s16, d18s16, d19s16, d23s16, d24s16;
-  int16x4_t d26s16, d27s16, d28s16, d29s16;
-  int32x4_t q10s32, q13s32, q14s32, q15s32;
-  int16x8_t q13s16, q14s16;
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-
-  d23s16 = vadd_s16(d16s16, d18s16);
-  d24s16 = vsub_s16(d16s16, d18s16);
-
-  q15s32 = vmull_s16(d17s16, *d2s16);
-  q10s32 = vmull_s16(d17s16, *d0s16);
-  q13s32 = vmull_s16(d23s16, *d1s16);
-  q14s32 = vmull_s16(d24s16, *d1s16);
-  q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16);
-  q10s32 = vmlal_s16(q10s32, d19s16, *d2s16);
-
-  d26s16 = vqrshrn_n_s32(q13s32, 14);
-  d27s16 = vqrshrn_n_s32(q14s32, 14);
-  d29s16 = vqrshrn_n_s32(q15s32, 14);
-  d28s16 = vqrshrn_n_s32(q10s32, 14);
-
-  q13s16 = vcombine_s16(d26s16, d27s16);
-  q14s16 = vcombine_s16(d28s16, d29s16);
-  *q8s16 = vaddq_s16(q13s16, q14s16);
-  *q9s16 = vsubq_s16(q13s16, q14s16);
-  *q9s16 = vcombine_s16(vget_high_s16(*q9s16), vget_low_s16(*q9s16));  // vswp
-  return;
-}
-
-static INLINE void IADST4x4_1D(int16x4_t *d3s16, int16x4_t *d4s16,
-                               int16x4_t *d5s16, int16x8_t *q3s16,
-                               int16x8_t *q8s16, int16x8_t *q9s16) {
-  int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16;
-  int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
-
-  d6s16 = vget_low_s16(*q3s16);
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-
-  q10s32 = vmull_s16(*d3s16, d16s16);
-  q11s32 = vmull_s16(*d4s16, d16s16);
-  q12s32 = vmull_s16(d6s16, d17s16);
-  q13s32 = vmull_s16(*d5s16, d18s16);
-  q14s32 = vmull_s16(*d3s16, d18s16);
-  q15s32 = vmovl_s16(d16s16);
-  q15s32 = vaddw_s16(q15s32, d19s16);
-  q8s32 = vmull_s16(*d4s16, d19s16);
-  q15s32 = vsubw_s16(q15s32, d18s16);
-  q9s32 = vmull_s16(*d5s16, d19s16);
-
-  q10s32 = vaddq_s32(q10s32, q13s32);
-  q10s32 = vaddq_s32(q10s32, q8s32);
-  q11s32 = vsubq_s32(q11s32, q14s32);
-  q8s32 = vdupq_n_s32((int32_t)sinpi_3_9);
-  q11s32 = vsubq_s32(q11s32, q9s32);
-  q15s32 = vmulq_s32(q15s32, q8s32);
-
-  q13s32 = vaddq_s32(q10s32, q12s32);
-  q10s32 = vaddq_s32(q10s32, q11s32);
-  q14s32 = vaddq_s32(q11s32, q12s32);
-  q10s32 = vsubq_s32(q10s32, q12s32);
-
-  d16s16 = vqrshrn_n_s32(q13s32, 14);
-  d17s16 = vqrshrn_n_s32(q14s32, 14);
-  d18s16 = vqrshrn_n_s32(q15s32, 14);
-  d19s16 = vqrshrn_n_s32(q10s32, 14);
-
-  *q8s16 = vcombine_s16(d16s16, d17s16);
-  *q9s16 = vcombine_s16(d18s16, d19s16);
-  return;
-}
-
-void av1_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest,
-                            int dest_stride, const TxfmParam *txfm_param) {
-  uint8x8_t d26u8, d27u8;
-  int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16;
-  uint32x2_t d26u32, d27u32;
-  int16x8_t q3s16, q8s16, q9s16;
-  uint16x8_t q8u16, q9u16;
-
-  d26u32 = d27u32 = vdup_n_u32(0);
-
-  q8s16 = vld1q_s16(input);
-  q9s16 = vld1q_s16(input + 8);
-
-  TRANSPOSE4X4(&q8s16, &q9s16);
-
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  switch (tx_type) {
-    case DCT_DCT:  // idct_idct is not supported. Fall back to C
-      av1_iht4x4_16_add_c(input, dest, dest_stride, txfm_param);
-      return;
-      break;
-    case ADST_DCT:  // iadst_idct
-      // generate constants
-      GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
-      GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
-
-      // first transform rows
-      IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
-
-      // transpose the matrix
-      TRANSPOSE4X4(&q8s16, &q9s16);
-
-      // then transform columns
-      IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
-      break;
-    case DCT_ADST:  // idct_iadst
-      // generate constantsyy
-      GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
-      GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
-
-      // first transform rows
-      IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
-
-      // transpose the matrix
-      TRANSPOSE4X4(&q8s16, &q9s16);
-
-      // then transform columns
-      IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
-      break;
-    case ADST_ADST:  // iadst_iadst
-      // generate constants
-      GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
-
-      // first transform rows
-      IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
-
-      // transpose the matrix
-      TRANSPOSE4X4(&q8s16, &q9s16);
-
-      // then transform columns
-      IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
-      break;
-    default:  // iadst_idct
-      assert(0);
-      break;
-  }
-
-  q8s16 = vrshrq_n_s16(q8s16, 4);
-  q9s16 = vrshrq_n_s16(q9s16, 4);
-
-  d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 0);
-  dest += dest_stride;
-  d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 1);
-  dest += dest_stride;
-  d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 0);
-  dest += dest_stride;
-  d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 1);
-
-  q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u32(d26u32));
-  q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u32(d27u32));
-
-  d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
-  d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
-
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 1);
-  dest -= dest_stride;
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 0);
-  dest -= dest_stride;
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 1);
-  dest -= dest_stride;
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 0);
-  return;
-}

third_party/aom/av1/common/arm/neon/iht8x8_add_neon.c

@@ -1,594 +0,0 @@
-/*
- * 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 <arm_neon.h>
-#include <assert.h>
-
-#include "./aom_config.h"
-#include "./av1_rtcd.h"
-#include "aom_dsp/txfm_common.h"
-#include "av1/common/common.h"
-
-static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
-                                int16x8_t *q10s16, int16x8_t *q11s16,
-                                int16x8_t *q12s16, int16x8_t *q13s16,
-                                int16x8_t *q14s16, int16x8_t *q15s16) {
-  int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
-  int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
-  int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
-  int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-  d20s16 = vget_low_s16(*q10s16);
-  d21s16 = vget_high_s16(*q10s16);
-  d22s16 = vget_low_s16(*q11s16);
-  d23s16 = vget_high_s16(*q11s16);
-  d24s16 = vget_low_s16(*q12s16);
-  d25s16 = vget_high_s16(*q12s16);
-  d26s16 = vget_low_s16(*q13s16);
-  d27s16 = vget_high_s16(*q13s16);
-  d28s16 = vget_low_s16(*q14s16);
-  d29s16 = vget_high_s16(*q14s16);
-  d30s16 = vget_low_s16(*q15s16);
-  d31s16 = vget_high_s16(*q15s16);
-
-  *q8s16 = vcombine_s16(d16s16, d24s16);   // vswp d17, d24
-  *q9s16 = vcombine_s16(d18s16, d26s16);   // vswp d19, d26
-  *q10s16 = vcombine_s16(d20s16, d28s16);  // vswp d21, d28
-  *q11s16 = vcombine_s16(d22s16, d30s16);  // vswp d23, d30
-  *q12s16 = vcombine_s16(d17s16, d25s16);
-  *q13s16 = vcombine_s16(d19s16, d27s16);
-  *q14s16 = vcombine_s16(d21s16, d29s16);
-  *q15s16 = vcombine_s16(d23s16, d31s16);
-
-  q0x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
-  q1x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
-  q2x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
-  q3x2s32 =
-      vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
-
-  q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]),   // q8
-                      vreinterpretq_s16_s32(q1x2s32.val[0]));  // q9
-  q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]),   // q10
-                      vreinterpretq_s16_s32(q1x2s32.val[1]));  // q11
-  q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]),   // q12
-                      vreinterpretq_s16_s32(q3x2s32.val[0]));  // q13
-  q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]),   // q14
-                      vreinterpretq_s16_s32(q3x2s32.val[1]));  // q15
-
-  *q8s16 = q0x2s16.val[0];
-  *q9s16 = q0x2s16.val[1];
-  *q10s16 = q1x2s16.val[0];
-  *q11s16 = q1x2s16.val[1];
-  *q12s16 = q2x2s16.val[0];
-  *q13s16 = q2x2s16.val[1];
-  *q14s16 = q3x2s16.val[0];
-  *q15s16 = q3x2s16.val[1];
-  return;
-}
-
-static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
-                              int16x8_t *q10s16, int16x8_t *q11s16,
-                              int16x8_t *q12s16, int16x8_t *q13s16,
-                              int16x8_t *q14s16, int16x8_t *q15s16) {
-  int16x4_t d0s16, d1s16, d2s16, d3s16;
-  int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
-  int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
-  int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
-  int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16;
-  int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32;
-  int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32;
-
-  d0s16 = vdup_n_s16((int16_t)cospi_28_64);
-  d1s16 = vdup_n_s16((int16_t)cospi_4_64);
-  d2s16 = vdup_n_s16((int16_t)cospi_12_64);
-  d3s16 = vdup_n_s16((int16_t)cospi_20_64);
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-  d20s16 = vget_low_s16(*q10s16);
-  d21s16 = vget_high_s16(*q10s16);
-  d22s16 = vget_low_s16(*q11s16);
-  d23s16 = vget_high_s16(*q11s16);
-  d24s16 = vget_low_s16(*q12s16);
-  d25s16 = vget_high_s16(*q12s16);
-  d26s16 = vget_low_s16(*q13s16);
-  d27s16 = vget_high_s16(*q13s16);
-  d28s16 = vget_low_s16(*q14s16);
-  d29s16 = vget_high_s16(*q14s16);
-  d30s16 = vget_low_s16(*q15s16);
-  d31s16 = vget_high_s16(*q15s16);
-
-  q2s32 = vmull_s16(d18s16, d0s16);
-  q3s32 = vmull_s16(d19s16, d0s16);
-  q5s32 = vmull_s16(d26s16, d2s16);
-  q6s32 = vmull_s16(d27s16, d2s16);
-
-  q2s32 = vmlsl_s16(q2s32, d30s16, d1s16);
-  q3s32 = vmlsl_s16(q3s32, d31s16, d1s16);
-  q5s32 = vmlsl_s16(q5s32, d22s16, d3s16);
-  q6s32 = vmlsl_s16(q6s32, d23s16, d3s16);
-
-  d8s16 = vqrshrn_n_s32(q2s32, 14);
-  d9s16 = vqrshrn_n_s32(q3s32, 14);
-  d10s16 = vqrshrn_n_s32(q5s32, 14);
-  d11s16 = vqrshrn_n_s32(q6s32, 14);
-  q4s16 = vcombine_s16(d8s16, d9s16);
-  q5s16 = vcombine_s16(d10s16, d11s16);
-
-  q2s32 = vmull_s16(d18s16, d1s16);
-  q3s32 = vmull_s16(d19s16, d1s16);
-  q9s32 = vmull_s16(d26s16, d3s16);
-  q13s32 = vmull_s16(d27s16, d3s16);
-
-  q2s32 = vmlal_s16(q2s32, d30s16, d0s16);
-  q3s32 = vmlal_s16(q3s32, d31s16, d0s16);
-  q9s32 = vmlal_s16(q9s32, d22s16, d2s16);
-  q13s32 = vmlal_s16(q13s32, d23s16, d2s16);
-
-  d14s16 = vqrshrn_n_s32(q2s32, 14);
-  d15s16 = vqrshrn_n_s32(q3s32, 14);
-  d12s16 = vqrshrn_n_s32(q9s32, 14);
-  d13s16 = vqrshrn_n_s32(q13s32, 14);
-  q6s16 = vcombine_s16(d12s16, d13s16);
-  q7s16 = vcombine_s16(d14s16, d15s16);
-
-  d0s16 = vdup_n_s16((int16_t)cospi_16_64);
-
-  q2s32 = vmull_s16(d16s16, d0s16);
-  q3s32 = vmull_s16(d17s16, d0s16);
-  q13s32 = vmull_s16(d16s16, d0s16);
-  q15s32 = vmull_s16(d17s16, d0s16);
-
-  q2s32 = vmlal_s16(q2s32, d24s16, d0s16);
-  q3s32 = vmlal_s16(q3s32, d25s16, d0s16);
-  q13s32 = vmlsl_s16(q13s32, d24s16, d0s16);
-  q15s32 = vmlsl_s16(q15s32, d25s16, d0s16);
-
-  d0s16 = vdup_n_s16((int16_t)cospi_24_64);
-  d1s16 = vdup_n_s16((int16_t)cospi_8_64);
-
-  d18s16 = vqrshrn_n_s32(q2s32, 14);
-  d19s16 = vqrshrn_n_s32(q3s32, 14);
-  d22s16 = vqrshrn_n_s32(q13s32, 14);
-  d23s16 = vqrshrn_n_s32(q15s32, 14);
-  *q9s16 = vcombine_s16(d18s16, d19s16);
-  *q11s16 = vcombine_s16(d22s16, d23s16);
-
-  q2s32 = vmull_s16(d20s16, d0s16);
-  q3s32 = vmull_s16(d21s16, d0s16);
-  q8s32 = vmull_s16(d20s16, d1s16);
-  q12s32 = vmull_s16(d21s16, d1s16);
-
-  q2s32 = vmlsl_s16(q2s32, d28s16, d1s16);
-  q3s32 = vmlsl_s16(q3s32, d29s16, d1s16);
-  q8s32 = vmlal_s16(q8s32, d28s16, d0s16);
-  q12s32 = vmlal_s16(q12s32, d29s16, d0s16);
-
-  d26s16 = vqrshrn_n_s32(q2s32, 14);
-  d27s16 = vqrshrn_n_s32(q3s32, 14);
-  d30s16 = vqrshrn_n_s32(q8s32, 14);
-  d31s16 = vqrshrn_n_s32(q12s32, 14);
-  *q13s16 = vcombine_s16(d26s16, d27s16);
-  *q15s16 = vcombine_s16(d30s16, d31s16);
-
-  q0s16 = vaddq_s16(*q9s16, *q15s16);
-  q1s16 = vaddq_s16(*q11s16, *q13s16);
-  q2s16 = vsubq_s16(*q11s16, *q13s16);
-  q3s16 = vsubq_s16(*q9s16, *q15s16);
-
-  *q13s16 = vsubq_s16(q4s16, q5s16);
-  q4s16 = vaddq_s16(q4s16, q5s16);
-  *q14s16 = vsubq_s16(q7s16, q6s16);
-  q7s16 = vaddq_s16(q7s16, q6s16);
-  d26s16 = vget_low_s16(*q13s16);
-  d27s16 = vget_high_s16(*q13s16);
-  d28s16 = vget_low_s16(*q14s16);
-  d29s16 = vget_high_s16(*q14s16);
-
-  d16s16 = vdup_n_s16((int16_t)cospi_16_64);
-
-  q9s32 = vmull_s16(d28s16, d16s16);
-  q10s32 = vmull_s16(d29s16, d16s16);
-  q11s32 = vmull_s16(d28s16, d16s16);
-  q12s32 = vmull_s16(d29s16, d16s16);
-
-  q9s32 = vmlsl_s16(q9s32, d26s16, d16s16);
-  q10s32 = vmlsl_s16(q10s32, d27s16, d16s16);
-  q11s32 = vmlal_s16(q11s32, d26s16, d16s16);
-  q12s32 = vmlal_s16(q12s32, d27s16, d16s16);
-
-  d10s16 = vqrshrn_n_s32(q9s32, 14);
-  d11s16 = vqrshrn_n_s32(q10s32, 14);
-  d12s16 = vqrshrn_n_s32(q11s32, 14);
-  d13s16 = vqrshrn_n_s32(q12s32, 14);
-  q5s16 = vcombine_s16(d10s16, d11s16);
-  q6s16 = vcombine_s16(d12s16, d13s16);
-
-  *q8s16 = vaddq_s16(q0s16, q7s16);
-  *q9s16 = vaddq_s16(q1s16, q6s16);
-  *q10s16 = vaddq_s16(q2s16, q5s16);
-  *q11s16 = vaddq_s16(q3s16, q4s16);
-  *q12s16 = vsubq_s16(q3s16, q4s16);
-  *q13s16 = vsubq_s16(q2s16, q5s16);
-  *q14s16 = vsubq_s16(q1s16, q6s16);
-  *q15s16 = vsubq_s16(q0s16, q7s16);
-  return;
-}
-
-static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
-                               int16x8_t *q10s16, int16x8_t *q11s16,
-                               int16x8_t *q12s16, int16x8_t *q13s16,
-                               int16x8_t *q14s16, int16x8_t *q15s16) {
-  int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
-  int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
-  int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
-  int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
-  int16x8_t q2s16, q4s16, q5s16, q6s16;
-  int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q7s32, q8s32;
-  int32x4_t q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
-
-  d16s16 = vget_low_s16(*q8s16);
-  d17s16 = vget_high_s16(*q8s16);
-  d18s16 = vget_low_s16(*q9s16);
-  d19s16 = vget_high_s16(*q9s16);
-  d20s16 = vget_low_s16(*q10s16);
-  d21s16 = vget_high_s16(*q10s16);
-  d22s16 = vget_low_s16(*q11s16);
-  d23s16 = vget_high_s16(*q11s16);
-  d24s16 = vget_low_s16(*q12s16);
-  d25s16 = vget_high_s16(*q12s16);
-  d26s16 = vget_low_s16(*q13s16);
-  d27s16 = vget_high_s16(*q13s16);
-  d28s16 = vget_low_s16(*q14s16);
-  d29s16 = vget_high_s16(*q14s16);
-  d30s16 = vget_low_s16(*q15s16);
-  d31s16 = vget_high_s16(*q15s16);
-
-  d14s16 = vdup_n_s16((int16_t)cospi_2_64);
-  d15s16 = vdup_n_s16((int16_t)cospi_30_64);
-
-  q1s32 = vmull_s16(d30s16, d14s16);
-  q2s32 = vmull_s16(d31s16, d14s16);
-  q3s32 = vmull_s16(d30s16, d15s16);
-  q4s32 = vmull_s16(d31s16, d15s16);
-
-  d30s16 = vdup_n_s16((int16_t)cospi_18_64);
-  d31s16 = vdup_n_s16((int16_t)cospi_14_64);
-
-  q1s32 = vmlal_s16(q1s32, d16s16, d15s16);
-  q2s32 = vmlal_s16(q2s32, d17s16, d15s16);
-  q3s32 = vmlsl_s16(q3s32, d16s16, d14s16);
-  q4s32 = vmlsl_s16(q4s32, d17s16, d14s16);
-
-  q5s32 = vmull_s16(d22s16, d30s16);
-  q6s32 = vmull_s16(d23s16, d30s16);
-  q7s32 = vmull_s16(d22s16, d31s16);
-  q8s32 = vmull_s16(d23s16, d31s16);
-
-  q5s32 = vmlal_s16(q5s32, d24s16, d31s16);
-  q6s32 = vmlal_s16(q6s32, d25s16, d31s16);
-  q7s32 = vmlsl_s16(q7s32, d24s16, d30s16);
-  q8s32 = vmlsl_s16(q8s32, d25s16, d30s16);
-
-  q11s32 = vaddq_s32(q1s32, q5s32);
-  q12s32 = vaddq_s32(q2s32, q6s32);
-  q1s32 = vsubq_s32(q1s32, q5s32);
-  q2s32 = vsubq_s32(q2s32, q6s32);
-
-  d22s16 = vqrshrn_n_s32(q11s32, 14);
-  d23s16 = vqrshrn_n_s32(q12s32, 14);
-  *q11s16 = vcombine_s16(d22s16, d23s16);
-
-  q12s32 = vaddq_s32(q3s32, q7s32);
-  q15s32 = vaddq_s32(q4s32, q8s32);
-  q3s32 = vsubq_s32(q3s32, q7s32);
-  q4s32 = vsubq_s32(q4s32, q8s32);
-
-  d2s16 = vqrshrn_n_s32(q1s32, 14);
-  d3s16 = vqrshrn_n_s32(q2s32, 14);
-  d24s16 = vqrshrn_n_s32(q12s32, 14);
-  d25s16 = vqrshrn_n_s32(q15s32, 14);
-  d6s16 = vqrshrn_n_s32(q3s32, 14);
-  d7s16 = vqrshrn_n_s32(q4s32, 14);
-  *q12s16 = vcombine_s16(d24s16, d25s16);
-
-  d0s16 = vdup_n_s16((int16_t)cospi_10_64);
-  d1s16 = vdup_n_s16((int16_t)cospi_22_64);
-  q4s32 = vmull_s16(d26s16, d0s16);
-  q5s32 = vmull_s16(d27s16, d0s16);
-  q2s32 = vmull_s16(d26s16, d1s16);
-  q6s32 = vmull_s16(d27s16, d1s16);
-
-  d30s16 = vdup_n_s16((int16_t)cospi_26_64);
-  d31s16 = vdup_n_s16((int16_t)cospi_6_64);
-
-  q4s32 = vmlal_s16(q4s32, d20s16, d1s16);
-  q5s32 = vmlal_s16(q5s32, d21s16, d1s16);
-  q2s32 = vmlsl_s16(q2s32, d20s16, d0s16);
-  q6s32 = vmlsl_s16(q6s32, d21s16, d0s16);
-
-  q0s32 = vmull_s16(d18s16, d30s16);
-  q13s32 = vmull_s16(d19s16, d30s16);
-
-  q0s32 = vmlal_s16(q0s32, d28s16, d31s16);
-  q13s32 = vmlal_s16(q13s32, d29s16, d31s16);
-
-  q10s32 = vmull_s16(d18s16, d31s16);
-  q9s32 = vmull_s16(d19s16, d31s16);
-
-  q10s32 = vmlsl_s16(q10s32, d28s16, d30s16);
-  q9s32 = vmlsl_s16(q9s32, d29s16, d30s16);
-
-  q14s32 = vaddq_s32(q2s32, q10s32);
-  q15s32 = vaddq_s32(q6s32, q9s32);
-  q2s32 = vsubq_s32(q2s32, q10s32);
-  q6s32 = vsubq_s32(q6s32, q9s32);
-
-  d28s16 = vqrshrn_n_s32(q14s32, 14);
-  d29s16 = vqrshrn_n_s32(q15s32, 14);
-  d4s16 = vqrshrn_n_s32(q2s32, 14);
-  d5s16 = vqrshrn_n_s32(q6s32, 14);
-  *q14s16 = vcombine_s16(d28s16, d29s16);
-
-  q9s32 = vaddq_s32(q4s32, q0s32);
-  q10s32 = vaddq_s32(q5s32, q13s32);
-  q4s32 = vsubq_s32(q4s32, q0s32);
-  q5s32 = vsubq_s32(q5s32, q13s32);
-
-  d30s16 = vdup_n_s16((int16_t)cospi_8_64);
-  d31s16 = vdup_n_s16((int16_t)cospi_24_64);
-
-  d18s16 = vqrshrn_n_s32(q9s32, 14);
-  d19s16 = vqrshrn_n_s32(q10s32, 14);
-  d8s16 = vqrshrn_n_s32(q4s32, 14);
-  d9s16 = vqrshrn_n_s32(q5s32, 14);
-  *q9s16 = vcombine_s16(d18s16, d19s16);
-
-  q5s32 = vmull_s16(d2s16, d30s16);
-  q6s32 = vmull_s16(d3s16, d30s16);
-  q7s32 = vmull_s16(d2s16, d31s16);
-  q0s32 = vmull_s16(d3s16, d31s16);
-
-  q5s32 = vmlal_s16(q5s32, d6s16, d31s16);
-  q6s32 = vmlal_s16(q6s32, d7s16, d31s16);
-  q7s32 = vmlsl_s16(q7s32, d6s16, d30s16);
-  q0s32 = vmlsl_s16(q0s32, d7s16, d30s16);
-
-  q1s32 = vmull_s16(d4s16, d30s16);
-  q3s32 = vmull_s16(d5s16, d30s16);
-  q10s32 = vmull_s16(d4s16, d31s16);
-  q2s32 = vmull_s16(d5s16, d31s16);
-
-  q1s32 = vmlsl_s16(q1s32, d8s16, d31s16);
-  q3s32 = vmlsl_s16(q3s32, d9s16, d31s16);
-  q10s32 = vmlal_s16(q10s32, d8s16, d30s16);
-  q2s32 = vmlal_s16(q2s32, d9s16, d30s16);
-
-  *q8s16 = vaddq_s16(*q11s16, *q9s16);
-  *q11s16 = vsubq_s16(*q11s16, *q9s16);
-  q4s16 = vaddq_s16(*q12s16, *q14s16);
-  *q12s16 = vsubq_s16(*q12s16, *q14s16);
-
-  q14s32 = vaddq_s32(q5s32, q1s32);
-  q15s32 = vaddq_s32(q6s32, q3s32);
-  q5s32 = vsubq_s32(q5s32, q1s32);
-  q6s32 = vsubq_s32(q6s32, q3s32);
-
-  d18s16 = vqrshrn_n_s32(q14s32, 14);
-  d19s16 = vqrshrn_n_s32(q15s32, 14);
-  d10s16 = vqrshrn_n_s32(q5s32, 14);
-  d11s16 = vqrshrn_n_s32(q6s32, 14);
-  *q9s16 = vcombine_s16(d18s16, d19s16);
-
-  q1s32 = vaddq_s32(q7s32, q10s32);
-  q3s32 = vaddq_s32(q0s32, q2s32);
-  q7s32 = vsubq_s32(q7s32, q10s32);
-  q0s32 = vsubq_s32(q0s32, q2s32);
-
-  d28s16 = vqrshrn_n_s32(q1s32, 14);
-  d29s16 = vqrshrn_n_s32(q3s32, 14);
-  d14s16 = vqrshrn_n_s32(q7s32, 14);
-  d15s16 = vqrshrn_n_s32(q0s32, 14);
-  *q14s16 = vcombine_s16(d28s16, d29s16);
-
-  d30s16 = vdup_n_s16((int16_t)cospi_16_64);
-
-  d22s16 = vget_low_s16(*q11s16);
-  d23s16 = vget_high_s16(*q11s16);
-  q2s32 = vmull_s16(d22s16, d30s16);
-  q3s32 = vmull_s16(d23s16, d30s16);
-  q13s32 = vmull_s16(d22s16, d30s16);
-  q1s32 = vmull_s16(d23s16, d30s16);
-
-  d24s16 = vget_low_s16(*q12s16);
-  d25s16 = vget_high_s16(*q12s16);
-  q2s32 = vmlal_s16(q2s32, d24s16, d30s16);
-  q3s32 = vmlal_s16(q3s32, d25s16, d30s16);
-  q13s32 = vmlsl_s16(q13s32, d24s16, d30s16);
-  q1s32 = vmlsl_s16(q1s32, d25s16, d30s16);
-
-  d4s16 = vqrshrn_n_s32(q2s32, 14);
-  d5s16 = vqrshrn_n_s32(q3s32, 14);
-  d24s16 = vqrshrn_n_s32(q13s32, 14);
-  d25s16 = vqrshrn_n_s32(q1s32, 14);
-  q2s16 = vcombine_s16(d4s16, d5s16);
-  *q12s16 = vcombine_s16(d24s16, d25s16);
-
-  q13s32 = vmull_s16(d10s16, d30s16);
-  q1s32 = vmull_s16(d11s16, d30s16);
-  q11s32 = vmull_s16(d10s16, d30s16);
-  q0s32 = vmull_s16(d11s16, d30s16);
-
-  q13s32 = vmlal_s16(q13s32, d14s16, d30s16);
-  q1s32 = vmlal_s16(q1s32, d15s16, d30s16);
-  q11s32 = vmlsl_s16(q11s32, d14s16, d30s16);
-  q0s32 = vmlsl_s16(q0s32, d15s16, d30s16);
-
-  d20s16 = vqrshrn_n_s32(q13s32, 14);
-  d21s16 = vqrshrn_n_s32(q1s32, 14);
-  d12s16 = vqrshrn_n_s32(q11s32, 14);
-  d13s16 = vqrshrn_n_s32(q0s32, 14);
-  *q10s16 = vcombine_s16(d20s16, d21s16);
-  q6s16 = vcombine_s16(d12s16, d13s16);
-
-  q5s16 = vdupq_n_s16(0);
-
-  *q9s16 = vsubq_s16(q5s16, *q9s16);
-  *q11s16 = vsubq_s16(q5s16, q2s16);
-  *q13s16 = vsubq_s16(q5s16, q6s16);
-  *q15s16 = vsubq_s16(q5s16, q4s16);
-  return;
-}
-
-void av1_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
-                            int dest_stride, const TxfmParam *txfm_param) {
-  int i;
-  uint8_t *d1, *d2;
-  uint8x8_t d0u8, d1u8, d2u8, d3u8;
-  uint64x1_t d0u64, d1u64, d2u64, d3u64;
-  int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16;
-  uint16x8_t q8u16, q9u16, q10u16, q11u16;
-
-  q8s16 = vld1q_s16(input);
-  q9s16 = vld1q_s16(input + 8);
-  q10s16 = vld1q_s16(input + 8 * 2);
-  q11s16 = vld1q_s16(input + 8 * 3);
-  q12s16 = vld1q_s16(input + 8 * 4);
-  q13s16 = vld1q_s16(input + 8 * 5);
-  q14s16 = vld1q_s16(input + 8 * 6);
-  q15s16 = vld1q_s16(input + 8 * 7);
-
-  TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-               &q15s16);
-
-  const TX_TYPE tx_type = txfm_param->tx_type;
-  switch (tx_type) {
-    case DCT_DCT:  // idct_idct is not supported. Fall back to C
-      av1_iht8x8_64_add_c(input, dest, dest_stride, txfm_param);
-      return;
-      break;
-    case ADST_DCT:  // iadst_idct
-      // generate IDCT constants
-      // GENERATE_IDCT_CONSTANTS
-
-      // first transform rows
-      IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                 &q15s16);
-
-      // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
-
-      // generate IADST constants
-      // GENERATE_IADST_CONSTANTS
-
-      // then transform columns
-      IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                  &q15s16);
-      break;
-    case DCT_ADST:  // idct_iadst
-      // generate IADST constants
-      // GENERATE_IADST_CONSTANTS
-
-      // first transform rows
-      IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                  &q15s16);
-
-      // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
-
-      // generate IDCT constants
-      // GENERATE_IDCT_CONSTANTS
-
-      // then transform columns
-      IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                 &q15s16);
-      break;
-    case ADST_ADST:  // iadst_iadst
-      // generate IADST constants
-      // GENERATE_IADST_CONSTANTS
-
-      // first transform rows
-      IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                  &q15s16);
-
-      // transpose the matrix
-      TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                   &q15s16);
-
-      // then transform columns
-      IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
-                  &q15s16);
-      break;
-    default:  // iadst_idct
-      assert(0);
-      break;
-  }
-
-  q8s16 = vrshrq_n_s16(q8s16, 5);
-  q9s16 = vrshrq_n_s16(q9s16, 5);
-  q10s16 = vrshrq_n_s16(q10s16, 5);
-  q11s16 = vrshrq_n_s16(q11s16, 5);
-  q12s16 = vrshrq_n_s16(q12s16, 5);
-  q13s16 = vrshrq_n_s16(q13s16, 5);
-  q14s16 = vrshrq_n_s16(q14s16, 5);
-  q15s16 = vrshrq_n_s16(q15s16, 5);
-
-  for (d1 = d2 = dest, i = 0; i < 2; i++) {
-    if (i != 0) {
-      q8s16 = q12s16;
-      q9s16 = q13s16;
-      q10s16 = q14s16;
-      q11s16 = q15s16;
-    }
-
-    d0u64 = vld1_u64((uint64_t *)d1);
-    d1 += dest_stride;
-    d1u64 = vld1_u64((uint64_t *)d1);
-    d1 += dest_stride;
-    d2u64 = vld1_u64((uint64_t *)d1);
-    d1 += dest_stride;
-    d3u64 = vld1_u64((uint64_t *)d1);
-    d1 += dest_stride;
-
-    q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u64(d0u64));
-    q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u64(d1u64));
-    q10u16 =
-        vaddw_u8(vreinterpretq_u16_s16(q10s16), vreinterpret_u8_u64(d2u64));
-    q11u16 =
-        vaddw_u8(vreinterpretq_u16_s16(q11s16), vreinterpret_u8_u64(d3u64));
-
-    d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
-    d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
-    d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16));
-    d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16));
-
-    vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8));
-    d2 += dest_stride;
-    vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8));
-    d2 += dest_stride;
-    vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8));
-    d2 += dest_stride;
-    vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8));
-    d2 += dest_stride;
-  }
-  return;
-}

third_party/aom/av1/common/arm/reconinter_neon.c

@@ -0,0 +1,86 @@
+/*
+ *
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "aom/aom_integer.h"
+#include "aom_dsp/blend.h"
+#include "aom_ports/mem.h"
+#include "av1/common/arm/mem_neon.h"
+#include "av1/common/blockd.h"
+#include "config/av1_rtcd.h"
+
+void av1_build_compound_diffwtd_mask_d16_neon(
+    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) {
+  assert(h >= 4);
+  assert(w >= 4);
+  assert((mask_type == DIFFWTD_38_INV) || (mask_type == DIFFWTD_38));
+  const int round =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1 + (bd - 8);
+  uint16x8_t diff_q, tmp0, tmp1;
+  uint8x8_t diff_d, diff_select;
+  const CONV_BUF_TYPE *src0_1, *src1_1;
+  const int16x8_t dup_round = vdupq_n_s16((int16_t)(-round));
+  const uint8x8_t dup_38 = vdup_n_u8(38);
+  const uint8x8_t dup_64 = vdup_n_u8(AOM_BLEND_A64_MAX_ALPHA);
+  if (mask_type == DIFFWTD_38) {
+    diff_select = vdup_n_u8(255);
+  } else {
+    diff_select = vdup_n_u8(0);
+  }
+  if (w >= 8) {
+    for (int i = 0; i < h; ++i) {
+      src0_1 = src0;
+      src1_1 = src1;
+      for (int j = 0; j < w; j += 8) {
+        __builtin_prefetch(src0_1);
+        __builtin_prefetch(src1_1);
+        diff_q = vabdq_u16(vld1q_u16(src0_1), vld1q_u16(src1_1));
+        diff_q = vrshlq_u16(diff_q, dup_round);
+        diff_d = vshrn_n_u16(diff_q, DIFF_FACTOR_LOG2);
+        diff_d = vmin_u8(vadd_u8(diff_d, dup_38), dup_64);
+        diff_d = vbsl_u8(diff_select, diff_d, vsub_u8(dup_64, diff_d));
+        vst1_u8(mask, diff_d);
+        src0_1 += 8;
+        src1_1 += 8;
+        mask += 8;
+      }
+      src0 += src0_stride;
+      src1 += src1_stride;
+    }
+  } else if (w == 4) {
+    for (int i = 0; i < h; i += 2) {
+      src0_1 = src0;
+      src1_1 = src1;
+      __builtin_prefetch(src0_1 + 0 * src0_stride);
+      __builtin_prefetch(src0_1 + 1 * src0_stride);
+      __builtin_prefetch(src1_1 + 0 * src1_stride);
+      __builtin_prefetch(src1_1 + 1 * src1_stride);
+      tmp0 = vcombine_u16(vld1_u16(src0_1 + (0 * src0_stride)),
+                          vld1_u16(src0_1 + (1 * src0_stride)));
+      tmp1 = vcombine_u16(vld1_u16(src1_1 + (0 * src1_stride)),
+                          vld1_u16(src1_1 + (1 * src1_stride)));
+      diff_q = vabdq_u16(tmp0, tmp1);
+      diff_q = vrshlq_u16(diff_q, dup_round);
+      diff_d = vshrn_n_u16(diff_q, DIFF_FACTOR_LOG2);
+      diff_d = vmin_u8(vadd_u8(diff_d, dup_38), dup_64);
+      diff_d = vbsl_u8(diff_select, diff_d, vsub_u8(dup_64, diff_d));
+      vst1_u8(mask, diff_d);
+      src0 += src0_stride * 2;
+      src1 += src1_stride * 2;
+      mask += w * 2;
+    }
+  }
+}

third_party/aom/av1/common/arm/transpose_neon.h

@@ -0,0 +1,422 @@
+/*
+ *  Copyright (c) 2018, Alliance for Open Media. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef AV1_COMMON_ARM_TRANSPOSE_NEON_H_
+#define AV1_COMMON_ARM_TRANSPOSE_NEON_H_
+
+#include <arm_neon.h>
+
+static INLINE void transpose_u8_8x8(uint8x8_t *a0, uint8x8_t *a1, uint8x8_t *a2,
+                                    uint8x8_t *a3, uint8x8_t *a4, uint8x8_t *a5,
+                                    uint8x8_t *a6, uint8x8_t *a7) {
+  // Swap 8 bit elements. Goes from:
+  // a0: 00 01 02 03 04 05 06 07
+  // a1: 10 11 12 13 14 15 16 17
+  // a2: 20 21 22 23 24 25 26 27
+  // a3: 30 31 32 33 34 35 36 37
+  // a4: 40 41 42 43 44 45 46 47
+  // a5: 50 51 52 53 54 55 56 57
+  // a6: 60 61 62 63 64 65 66 67
+  // a7: 70 71 72 73 74 75 76 77
+  // to:
+  // b0.val[0]: 00 10 02 12 04 14 06 16  40 50 42 52 44 54 46 56
+  // b0.val[1]: 01 11 03 13 05 15 07 17  41 51 43 53 45 55 47 57
+  // b1.val[0]: 20 30 22 32 24 34 26 36  60 70 62 72 64 74 66 76
+  // b1.val[1]: 21 31 23 33 25 35 27 37  61 71 63 73 65 75 67 77
+
+  const uint8x16x2_t b0 =
+      vtrnq_u8(vcombine_u8(*a0, *a4), vcombine_u8(*a1, *a5));
+  const uint8x16x2_t b1 =
+      vtrnq_u8(vcombine_u8(*a2, *a6), vcombine_u8(*a3, *a7));
+
+  // Swap 16 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30 04 14 24 34  40 50 60 70 44 54 64 74
+  // c0.val[1]: 02 12 22 32 06 16 26 36  42 52 62 72 46 56 66 76
+  // c1.val[0]: 01 11 21 31 05 15 25 35  41 51 61 71 45 55 65 75
+  // c1.val[1]: 03 13 23 33 07 17 27 37  43 53 63 73 47 57 67 77
+
+  const uint16x8x2_t c0 = vtrnq_u16(vreinterpretq_u16_u8(b0.val[0]),
+                                    vreinterpretq_u16_u8(b1.val[0]));
+  const uint16x8x2_t c1 = vtrnq_u16(vreinterpretq_u16_u8(b0.val[1]),
+                                    vreinterpretq_u16_u8(b1.val[1]));
+
+  // Unzip 32 bit elements resulting in:
+  // d0.val[0]: 00 10 20 30 40 50 60 70  01 11 21 31 41 51 61 71
+  // d0.val[1]: 04 14 24 34 44 54 64 74  05 15 25 35 45 55 65 75
+  // d1.val[0]: 02 12 22 32 42 52 62 72  03 13 23 33 43 53 63 73
+  // d1.val[1]: 06 16 26 36 46 56 66 76  07 17 27 37 47 57 67 77
+  const uint32x4x2_t d0 = vuzpq_u32(vreinterpretq_u32_u16(c0.val[0]),
+                                    vreinterpretq_u32_u16(c1.val[0]));
+  const uint32x4x2_t d1 = vuzpq_u32(vreinterpretq_u32_u16(c0.val[1]),
+                                    vreinterpretq_u32_u16(c1.val[1]));
+
+  *a0 = vreinterpret_u8_u32(vget_low_u32(d0.val[0]));
+  *a1 = vreinterpret_u8_u32(vget_high_u32(d0.val[0]));
+  *a2 = vreinterpret_u8_u32(vget_low_u32(d1.val[0]));
+  *a3 = vreinterpret_u8_u32(vget_high_u32(d1.val[0]));
+  *a4 = vreinterpret_u8_u32(vget_low_u32(d0.val[1]));
+  *a5 = vreinterpret_u8_u32(vget_high_u32(d0.val[1]));
+  *a6 = vreinterpret_u8_u32(vget_low_u32(d1.val[1]));
+  *a7 = vreinterpret_u8_u32(vget_high_u32(d1.val[1]));
+}
+
+static INLINE void transpose_u8_8x4(uint8x8_t *a0, uint8x8_t *a1, uint8x8_t *a2,
+                                    uint8x8_t *a3) {
+  // Swap 8 bit elements. Goes from:
+  // a0: 00 01 02 03 04 05 06 07
+  // a1: 10 11 12 13 14 15 16 17
+  // a2: 20 21 22 23 24 25 26 27
+  // a3: 30 31 32 33 34 35 36 37
+  // to:
+  // b0.val[0]: 00 10 02 12 04 14 06 16
+  // b0.val[1]: 01 11 03 13 05 15 07 17
+  // b1.val[0]: 20 30 22 32 24 34 26 36
+  // b1.val[1]: 21 31 23 33 25 35 27 37
+
+  const uint8x8x2_t b0 = vtrn_u8(*a0, *a1);
+  const uint8x8x2_t b1 = vtrn_u8(*a2, *a3);
+
+  // Swap 16 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30 04 14 24 34
+  // c0.val[1]: 02 12 22 32 06 16 26 36
+  // c1.val[0]: 01 11 21 31 05 15 25 35
+  // c1.val[1]: 03 13 23 33 07 17 27 37
+
+  const uint16x4x2_t c0 =
+      vtrn_u16(vreinterpret_u16_u8(b0.val[0]), vreinterpret_u16_u8(b1.val[0]));
+  const uint16x4x2_t c1 =
+      vtrn_u16(vreinterpret_u16_u8(b0.val[1]), vreinterpret_u16_u8(b1.val[1]));
+
+  *a0 = vreinterpret_u8_u16(c0.val[0]);
+  *a1 = vreinterpret_u8_u16(c1.val[0]);
+  *a2 = vreinterpret_u8_u16(c0.val[1]);
+  *a3 = vreinterpret_u8_u16(c1.val[1]);
+}
+
+static INLINE void transpose_u8_4x4(uint8x8_t *a0, uint8x8_t *a1) {
+  // Swap 16 bit elements. Goes from:
+  // a0: 00 01 02 03  10 11 12 13
+  // a1: 20 21 22 23  30 31 32 33
+  // to:
+  // b0.val[0]: 00 01 20 21  10 11 30 31
+  // b0.val[1]: 02 03 22 23  12 13 32 33
+
+  const uint16x4x2_t b0 =
+      vtrn_u16(vreinterpret_u16_u8(*a0), vreinterpret_u16_u8(*a1));
+
+  // Swap 32 bit elements resulting in:
+  // c0.val[0]: 00 01 20 21  02 03 22 23
+  // c0.val[1]: 10 11 30 31  12 13 32 33
+
+  const uint32x2x2_t c0 = vtrn_u32(vreinterpret_u32_u16(b0.val[0]),
+                                   vreinterpret_u32_u16(b0.val[1]));
+
+  // Swap 8 bit elements resulting in:
+  // d0.val[0]: 00 10 20 30  02 12 22 32
+  // d0.val[1]: 01 11 21 31  03 13 23 33
+
+  const uint8x8x2_t d0 =
+      vtrn_u8(vreinterpret_u8_u32(c0.val[0]), vreinterpret_u8_u32(c0.val[1]));
+
+  *a0 = d0.val[0];
+  *a1 = d0.val[1];
+}
+
+static INLINE void transpose_u8_4x8(uint8x8_t *a0, uint8x8_t *a1, uint8x8_t *a2,
+                                    uint8x8_t *a3, const uint8x8_t a4,
+                                    const uint8x8_t a5, const uint8x8_t a6,
+                                    const uint8x8_t a7) {
+  // Swap 32 bit elements. Goes from:
+  // a0: 00 01 02 03 XX XX XX XX
+  // a1: 10 11 12 13 XX XX XX XX
+  // a2: 20 21 22 23 XX XX XX XX
+  // a3; 30 31 32 33 XX XX XX XX
+  // a4: 40 41 42 43 XX XX XX XX
+  // a5: 50 51 52 53 XX XX XX XX
+  // a6: 60 61 62 63 XX XX XX XX
+  // a7: 70 71 72 73 XX XX XX XX
+  // to:
+  // b0.val[0]: 00 01 02 03 40 41 42 43
+  // b1.val[0]: 10 11 12 13 50 51 52 53
+  // b2.val[0]: 20 21 22 23 60 61 62 63
+  // b3.val[0]: 30 31 32 33 70 71 72 73
+
+  const uint32x2x2_t b0 =
+      vtrn_u32(vreinterpret_u32_u8(*a0), vreinterpret_u32_u8(a4));
+  const uint32x2x2_t b1 =
+      vtrn_u32(vreinterpret_u32_u8(*a1), vreinterpret_u32_u8(a5));
+  const uint32x2x2_t b2 =
+      vtrn_u32(vreinterpret_u32_u8(*a2), vreinterpret_u32_u8(a6));
+  const uint32x2x2_t b3 =
+      vtrn_u32(vreinterpret_u32_u8(*a3), vreinterpret_u32_u8(a7));
+
+  // Swap 16 bit elements resulting in:
+  // c0.val[0]: 00 01 20 21 40 41 60 61
+  // c0.val[1]: 02 03 22 23 42 43 62 63
+  // c1.val[0]: 10 11 30 31 50 51 70 71
+  // c1.val[1]: 12 13 32 33 52 53 72 73
+
+  const uint16x4x2_t c0 = vtrn_u16(vreinterpret_u16_u32(b0.val[0]),
+                                   vreinterpret_u16_u32(b2.val[0]));
+  const uint16x4x2_t c1 = vtrn_u16(vreinterpret_u16_u32(b1.val[0]),
+                                   vreinterpret_u16_u32(b3.val[0]));
+
+  // Swap 8 bit elements resulting in:
+  // d0.val[0]: 00 10 20 30 40 50 60 70
+  // d0.val[1]: 01 11 21 31 41 51 61 71
+  // d1.val[0]: 02 12 22 32 42 52 62 72
+  // d1.val[1]: 03 13 23 33 43 53 63 73
+
+  const uint8x8x2_t d0 =
+      vtrn_u8(vreinterpret_u8_u16(c0.val[0]), vreinterpret_u8_u16(c1.val[0]));
+  const uint8x8x2_t d1 =
+      vtrn_u8(vreinterpret_u8_u16(c0.val[1]), vreinterpret_u8_u16(c1.val[1]));
+
+  *a0 = d0.val[0];
+  *a1 = d0.val[1];
+  *a2 = d1.val[0];
+  *a3 = d1.val[1];
+}
+
+static INLINE void transpose_u16_4x8(uint16x4_t *a0, uint16x4_t *a1,
+                                     uint16x4_t *a2, uint16x4_t *a3,
+                                     uint16x4_t *a4, uint16x4_t *a5,
+                                     uint16x4_t *a6, uint16x4_t *a7,
+                                     uint16x8_t *o0, uint16x8_t *o1,
+                                     uint16x8_t *o2, uint16x8_t *o3) {
+  // Swap 16 bit elements. Goes from:
+  // a0: 00 01 02 03
+  // a1: 10 11 12 13
+  // a2: 20 21 22 23
+  // a3: 30 31 32 33
+  // a4: 40 41 42 43
+  // a5: 50 51 52 53
+  // a6: 60 61 62 63
+  // a7: 70 71 72 73
+  // to:
+  // b0.val[0]: 00 10 02 12
+  // b0.val[1]: 01 11 03 13
+  // b1.val[0]: 20 30 22 32
+  // b1.val[1]: 21 31 23 33
+  // b2.val[0]: 40 50 42 52
+  // b2.val[1]: 41 51 43 53
+  // b3.val[0]: 60 70 62 72
+  // b3.val[1]: 61 71 63 73
+
+  uint16x4x2_t b0 = vtrn_u16(*a0, *a1);
+  uint16x4x2_t b1 = vtrn_u16(*a2, *a3);
+  uint16x4x2_t b2 = vtrn_u16(*a4, *a5);
+  uint16x4x2_t b3 = vtrn_u16(*a6, *a7);
+
+  // Swap 32 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30
+  // c0.val[1]: 02 12 22 32
+  // c1.val[0]: 01 11 21 31
+  // c1.val[1]: 03 13 23 33
+  // c2.val[0]: 40 50 60 70
+  // c2.val[1]: 42 52 62 72
+  // c3.val[0]: 41 51 61 71
+  // c3.val[1]: 43 53 63 73
+
+  uint32x2x2_t c0 = vtrn_u32(vreinterpret_u32_u16(b0.val[0]),
+                             vreinterpret_u32_u16(b1.val[0]));
+  uint32x2x2_t c1 = vtrn_u32(vreinterpret_u32_u16(b0.val[1]),
+                             vreinterpret_u32_u16(b1.val[1]));
+  uint32x2x2_t c2 = vtrn_u32(vreinterpret_u32_u16(b2.val[0]),
+                             vreinterpret_u32_u16(b3.val[0]));
+  uint32x2x2_t c3 = vtrn_u32(vreinterpret_u32_u16(b2.val[1]),
+                             vreinterpret_u32_u16(b3.val[1]));
+
+  // Swap 64 bit elements resulting in:
+  // o0: 00 10 20 30 40 50 60 70
+  // o1: 01 11 21 31 41 51 61 71
+  // o2: 02 12 22 32 42 52 62 72
+  // o3: 03 13 23 33 43 53 63 73
+
+  *o0 = vcombine_u16(vreinterpret_u16_u32(c0.val[0]),
+                     vreinterpret_u16_u32(c2.val[0]));
+  *o1 = vcombine_u16(vreinterpret_u16_u32(c1.val[0]),
+                     vreinterpret_u16_u32(c3.val[0]));
+  *o2 = vcombine_u16(vreinterpret_u16_u32(c0.val[1]),
+                     vreinterpret_u16_u32(c2.val[1]));
+  *o3 = vcombine_u16(vreinterpret_u16_u32(c1.val[1]),
+                     vreinterpret_u16_u32(c3.val[1]));
+}
+
+static INLINE void transpose_u16_8x8(uint16x8_t *a0, uint16x8_t *a1,
+                                     uint16x8_t *a2, uint16x8_t *a3,
+                                     uint16x8_t *a4, uint16x8_t *a5,
+                                     uint16x8_t *a6, uint16x8_t *a7) {
+  // Swap 16 bit elements. Goes from:
+  // a0: 00 01 02 03 04 05 06 07
+  // a1: 10 11 12 13 14 15 16 17
+  // a2: 20 21 22 23 24 25 26 27
+  // a3: 30 31 32 33 34 35 36 37
+  // a4: 40 41 42 43 44 45 46 47
+  // a5: 50 51 52 53 54 55 56 57
+  // a6: 60 61 62 63 64 65 66 67
+  // a7: 70 71 72 73 74 75 76 77
+  // to:
+  // b0.val[0]: 00 10 02 12 04 14 06 16
+  // b0.val[1]: 01 11 03 13 05 15 07 17
+  // b1.val[0]: 20 30 22 32 24 34 26 36
+  // b1.val[1]: 21 31 23 33 25 35 27 37
+  // b2.val[0]: 40 50 42 52 44 54 46 56
+  // b2.val[1]: 41 51 43 53 45 55 47 57
+  // b3.val[0]: 60 70 62 72 64 74 66 76
+  // b3.val[1]: 61 71 63 73 65 75 67 77
+
+  const uint16x8x2_t b0 = vtrnq_u16(*a0, *a1);
+  const uint16x8x2_t b1 = vtrnq_u16(*a2, *a3);
+  const uint16x8x2_t b2 = vtrnq_u16(*a4, *a5);
+  const uint16x8x2_t b3 = vtrnq_u16(*a6, *a7);
+
+  // Swap 32 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30 04 14 24 34
+  // c0.val[1]: 02 12 22 32 06 16 26 36
+  // c1.val[0]: 01 11 21 31 05 15 25 35
+  // c1.val[1]: 03 13 23 33 07 17 27 37
+  // c2.val[0]: 40 50 60 70 44 54 64 74
+  // c2.val[1]: 42 52 62 72 46 56 66 76
+  // c3.val[0]: 41 51 61 71 45 55 65 75
+  // c3.val[1]: 43 53 63 73 47 57 67 77
+
+  const uint32x4x2_t c0 = vtrnq_u32(vreinterpretq_u32_u16(b0.val[0]),
+                                    vreinterpretq_u32_u16(b1.val[0]));
+  const uint32x4x2_t c1 = vtrnq_u32(vreinterpretq_u32_u16(b0.val[1]),
+                                    vreinterpretq_u32_u16(b1.val[1]));
+  const uint32x4x2_t c2 = vtrnq_u32(vreinterpretq_u32_u16(b2.val[0]),
+                                    vreinterpretq_u32_u16(b3.val[0]));
+  const uint32x4x2_t c3 = vtrnq_u32(vreinterpretq_u32_u16(b2.val[1]),
+                                    vreinterpretq_u32_u16(b3.val[1]));
+
+  *a0 = vcombine_u16(vget_low_u16(vreinterpretq_u16_u32(c0.val[0])),
+                     vget_low_u16(vreinterpretq_u16_u32(c2.val[0])));
+  *a4 = vcombine_u16(vget_high_u16(vreinterpretq_u16_u32(c0.val[0])),
+                     vget_high_u16(vreinterpretq_u16_u32(c2.val[0])));
+
+  *a2 = vcombine_u16(vget_low_u16(vreinterpretq_u16_u32(c0.val[1])),
+                     vget_low_u16(vreinterpretq_u16_u32(c2.val[1])));
+  *a6 = vcombine_u16(vget_high_u16(vreinterpretq_u16_u32(c0.val[1])),
+                     vget_high_u16(vreinterpretq_u16_u32(c2.val[1])));
+
+  *a1 = vcombine_u16(vget_low_u16(vreinterpretq_u16_u32(c1.val[0])),
+                     vget_low_u16(vreinterpretq_u16_u32(c3.val[0])));
+  *a5 = vcombine_u16(vget_high_u16(vreinterpretq_u16_u32(c1.val[0])),
+                     vget_high_u16(vreinterpretq_u16_u32(c3.val[0])));
+
+  *a3 = vcombine_u16(vget_low_u16(vreinterpretq_u16_u32(c1.val[1])),
+                     vget_low_u16(vreinterpretq_u16_u32(c3.val[1])));
+  *a7 = vcombine_u16(vget_high_u16(vreinterpretq_u16_u32(c1.val[1])),
+                     vget_high_u16(vreinterpretq_u16_u32(c3.val[1])));
+}
+
+static INLINE void transpose_s16_8x8(int16x8_t *a0, int16x8_t *a1,
+                                     int16x8_t *a2, int16x8_t *a3,
+                                     int16x8_t *a4, int16x8_t *a5,
+                                     int16x8_t *a6, int16x8_t *a7) {
+  // Swap 16 bit elements. Goes from:
+  // a0: 00 01 02 03 04 05 06 07
+  // a1: 10 11 12 13 14 15 16 17
+  // a2: 20 21 22 23 24 25 26 27
+  // a3: 30 31 32 33 34 35 36 37
+  // a4: 40 41 42 43 44 45 46 47
+  // a5: 50 51 52 53 54 55 56 57
+  // a6: 60 61 62 63 64 65 66 67
+  // a7: 70 71 72 73 74 75 76 77
+  // to:
+  // b0.val[0]: 00 10 02 12 04 14 06 16
+  // b0.val[1]: 01 11 03 13 05 15 07 17
+  // b1.val[0]: 20 30 22 32 24 34 26 36
+  // b1.val[1]: 21 31 23 33 25 35 27 37
+  // b2.val[0]: 40 50 42 52 44 54 46 56
+  // b2.val[1]: 41 51 43 53 45 55 47 57
+  // b3.val[0]: 60 70 62 72 64 74 66 76
+  // b3.val[1]: 61 71 63 73 65 75 67 77
+
+  const int16x8x2_t b0 = vtrnq_s16(*a0, *a1);
+  const int16x8x2_t b1 = vtrnq_s16(*a2, *a3);
+  const int16x8x2_t b2 = vtrnq_s16(*a4, *a5);
+  const int16x8x2_t b3 = vtrnq_s16(*a6, *a7);
+
+  // Swap 32 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30 04 14 24 34
+  // c0.val[1]: 02 12 22 32 06 16 26 36
+  // c1.val[0]: 01 11 21 31 05 15 25 35
+  // c1.val[1]: 03 13 23 33 07 17 27 37
+  // c2.val[0]: 40 50 60 70 44 54 64 74
+  // c2.val[1]: 42 52 62 72 46 56 66 76
+  // c3.val[0]: 41 51 61 71 45 55 65 75
+  // c3.val[1]: 43 53 63 73 47 57 67 77
+
+  const int32x4x2_t c0 = vtrnq_s32(vreinterpretq_s32_s16(b0.val[0]),
+                                   vreinterpretq_s32_s16(b1.val[0]));
+  const int32x4x2_t c1 = vtrnq_s32(vreinterpretq_s32_s16(b0.val[1]),
+                                   vreinterpretq_s32_s16(b1.val[1]));
+  const int32x4x2_t c2 = vtrnq_s32(vreinterpretq_s32_s16(b2.val[0]),
+                                   vreinterpretq_s32_s16(b3.val[0]));
+  const int32x4x2_t c3 = vtrnq_s32(vreinterpretq_s32_s16(b2.val[1]),
+                                   vreinterpretq_s32_s16(b3.val[1]));
+
+  *a0 = vcombine_s16(vget_low_s16(vreinterpretq_s16_s32(c0.val[0])),
+                     vget_low_s16(vreinterpretq_s16_s32(c2.val[0])));
+  *a4 = vcombine_s16(vget_high_s16(vreinterpretq_s16_s32(c0.val[0])),
+                     vget_high_s16(vreinterpretq_s16_s32(c2.val[0])));
+
+  *a2 = vcombine_s16(vget_low_s16(vreinterpretq_s16_s32(c0.val[1])),
+                     vget_low_s16(vreinterpretq_s16_s32(c2.val[1])));
+  *a6 = vcombine_s16(vget_high_s16(vreinterpretq_s16_s32(c0.val[1])),
+                     vget_high_s16(vreinterpretq_s16_s32(c2.val[1])));
+
+  *a1 = vcombine_s16(vget_low_s16(vreinterpretq_s16_s32(c1.val[0])),
+                     vget_low_s16(vreinterpretq_s16_s32(c3.val[0])));
+  *a5 = vcombine_s16(vget_high_s16(vreinterpretq_s16_s32(c1.val[0])),
+                     vget_high_s16(vreinterpretq_s16_s32(c3.val[0])));
+
+  *a3 = vcombine_s16(vget_low_s16(vreinterpretq_s16_s32(c1.val[1])),
+                     vget_low_s16(vreinterpretq_s16_s32(c3.val[1])));
+  *a7 = vcombine_s16(vget_high_s16(vreinterpretq_s16_s32(c1.val[1])),
+                     vget_high_s16(vreinterpretq_s16_s32(c3.val[1])));
+}
+
+static INLINE void transpose_s16_4x4d(int16x4_t *a0, int16x4_t *a1,
+                                      int16x4_t *a2, int16x4_t *a3) {
+  // Swap 16 bit elements. Goes from:
+  // a0: 00 01 02 03
+  // a1: 10 11 12 13
+  // a2: 20 21 22 23
+  // a3: 30 31 32 33
+  // to:
+  // b0.val[0]: 00 10 02 12
+  // b0.val[1]: 01 11 03 13
+  // b1.val[0]: 20 30 22 32
+  // b1.val[1]: 21 31 23 33
+
+  const int16x4x2_t b0 = vtrn_s16(*a0, *a1);
+  const int16x4x2_t b1 = vtrn_s16(*a2, *a3);
+
+  // Swap 32 bit elements resulting in:
+  // c0.val[0]: 00 10 20 30
+  // c0.val[1]: 02 12 22 32
+  // c1.val[0]: 01 11 21 31
+  // c1.val[1]: 03 13 23 33
+
+  const int32x2x2_t c0 = vtrn_s32(vreinterpret_s32_s16(b0.val[0]),
+                                  vreinterpret_s32_s16(b1.val[0]));
+  const int32x2x2_t c1 = vtrn_s32(vreinterpret_s32_s16(b0.val[1]),
+                                  vreinterpret_s32_s16(b1.val[1]));
+
+  *a0 = vreinterpret_s16_s32(c0.val[0]);
+  *a1 = vreinterpret_s16_s32(c1.val[0]);
+  *a2 = vreinterpret_s16_s32(c0.val[1]);
+  *a3 = vreinterpret_s16_s32(c1.val[1]);
+}
+
+#endif  // AV1_COMMON_ARM_TRANSPOSE_NEON_H_

third_party/aom/av1/common/arm/wiener_convolve_neon.c

@@ -0,0 +1,401 @@
+/*
+ * 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 <arm_neon.h>
+#include <assert.h>
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_dsp/txfm_common.h"
+#include "aom_ports/mem.h"
+#include "av1/common/common.h"
+#include "av1/common/arm/convolve_neon.h"
+#include "av1/common/arm/mem_neon.h"
+#include "av1/common/arm/transpose_neon.h"
+
+/* Wiener filter 2D
+   Apply horizontal filter and store in a temporary buffer. When applying
+   vertical filter, overwrite the original pixel values.
+ */
+
+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) {
+  uint16_t *d_tmp;
+  uint8_t *d;
+  const uint8_t *src_ptr, *s_tmp;
+  uint16_t *dst_ptr;
+  (void)x_step_q4;
+  (void)y_step_q4;
+
+  int width, height;
+  const int bd = 8;
+  const int intermediate_height = h + SUBPEL_TAPS - 1;
+  const int center_tap = ((SUBPEL_TAPS - 1) / 2);
+  int16_t filter_x_tmp[7], filter_y_tmp[7];
+
+  DECLARE_ALIGNED(16, uint16_t,
+                  temp[(MAX_SB_SIZE + HORIZ_EXTRA_ROWS) * MAX_SB_SIZE]);
+
+  assert(x_step_q4 == 16 && y_step_q4 == 16);
+  assert(!(w % 8));
+
+  assert(w <= MAX_SB_SIZE);
+  assert(h <= MAX_SB_SIZE);
+
+  assert(filter_x[7] == 0);
+  assert(filter_y[7] == 0);
+
+  /* assumption of horizontal filtering output will not exceed 15 bit.
+     ((bd) + 1 + FILTER_BITS - conv_params->round_0) <= 15
+     16 - conv_params->round_0 <= 15 -- (conv_params->round_0) >= 1
+   */
+  assert((conv_params->round_0) >= 1);
+
+  memcpy(&filter_x_tmp[0], filter_x, sizeof(*filter_x) * FILTER_BITS);
+  memcpy(&filter_y_tmp[0], filter_y, sizeof(*filter_y) * FILTER_BITS);
+
+  filter_x_tmp[3] += (1 << FILTER_BITS);
+  filter_y_tmp[3] += (1 << FILTER_BITS);
+
+  s_tmp = src - center_tap * src_stride - center_tap;
+  dst_ptr = temp;
+  src_ptr = s_tmp;
+  height = intermediate_height;
+
+  /* if height is a multiple of 8 */
+  if (!(h & 7)) {
+    int16x8_t res0, res1, res2, res3;
+    uint16x8_t res4, res5, res6, res7, res8, res9, res10, res11;
+    uint8x8_t t0, t1, t2, t3, t4, t5, t6, t7;
+    uint8x8_t t8, t9, t10, t11, t12, t13, t14;
+
+    do {
+      const uint8_t *s;
+
+      __builtin_prefetch(src_ptr + 0 * src_stride);
+      __builtin_prefetch(src_ptr + 1 * src_stride);
+      __builtin_prefetch(src_ptr + 2 * src_stride);
+      __builtin_prefetch(src_ptr + 3 * src_stride);
+      __builtin_prefetch(src_ptr + 4 * src_stride);
+      __builtin_prefetch(src_ptr + 5 * src_stride);
+      __builtin_prefetch(src_ptr + 6 * src_stride);
+      __builtin_prefetch(src_ptr + 7 * src_stride);
+
+      load_u8_8x8(src_ptr, src_stride, &t0, &t1, &t2, &t3, &t4, &t5, &t6, &t7);
+      transpose_u8_8x8(&t0, &t1, &t2, &t3, &t4, &t5, &t6, &t7);
+
+      s = src_ptr + 7;
+      d_tmp = dst_ptr;
+      width = w;
+
+      __builtin_prefetch(dst_ptr + 0 * dst_stride);
+      __builtin_prefetch(dst_ptr + 1 * dst_stride);
+      __builtin_prefetch(dst_ptr + 2 * dst_stride);
+      __builtin_prefetch(dst_ptr + 3 * dst_stride);
+      __builtin_prefetch(dst_ptr + 4 * dst_stride);
+      __builtin_prefetch(dst_ptr + 5 * dst_stride);
+      __builtin_prefetch(dst_ptr + 6 * dst_stride);
+      __builtin_prefetch(dst_ptr + 7 * dst_stride);
+
+      do {
+        load_u8_8x8(s, src_stride, &t7, &t8, &t9, &t10, &t11, &t12, &t13, &t14);
+        transpose_u8_8x8(&t7, &t8, &t9, &t10, &t11, &t12, &t13, &t14);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t0, t6));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t1, t5));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t2, t4));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t3));
+        res4 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t1, t7));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t2, t6));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t3, t5));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t4));
+        res5 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t2, t8));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t3, t7));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t4, t6));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t5));
+        res6 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t3, t9));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t4, t8));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t5, t7));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t6));
+        res7 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t4, t10));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t5, t9));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t6, t8));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t7));
+        res8 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t5, t11));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t6, t10));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t7, t9));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t8));
+        res9 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                          bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t6, t12));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t7, t11));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t8, t10));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t9));
+        res10 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                           bd, conv_params->round_0);
+
+        res0 = vreinterpretq_s16_u16(vaddl_u8(t7, t13));
+        res1 = vreinterpretq_s16_u16(vaddl_u8(t8, t12));
+        res2 = vreinterpretq_s16_u16(vaddl_u8(t9, t11));
+        res3 = vreinterpretq_s16_u16(vmovl_u8(t10));
+        res11 = wiener_convolve8_horiz_8x8(res0, res1, res2, res3, filter_x_tmp,
+                                           bd, conv_params->round_0);
+
+        transpose_u16_8x8(&res4, &res5, &res6, &res7, &res8, &res9, &res10,
+                          &res11);
+        store_u16_8x8(d_tmp, MAX_SB_SIZE, res4, res5, res6, res7, res8, res9,
+                      res10, res11);
+
+        t0 = t8;
+        t1 = t9;
+        t2 = t10;
+        t3 = t11;
+        t4 = t12;
+        t5 = t13;
+        t6 = t14;
+        s += 8;
+        d_tmp += 8;
+        width -= 8;
+      } while (width > 0);
+      src_ptr += 8 * src_stride;
+      dst_ptr += 8 * MAX_SB_SIZE;
+      height -= 8;
+    } while (height > 0);
+  } else {
+    /*if height is a multiple of 4*/
+    int16x8_t tt0, tt1, tt2, tt3;
+    const uint8_t *s;
+    uint16x4_t res0, res1, res2, res3, res4, res5, res6, res7;
+    uint16x8_t d0, d1, d2, d3;
+    int16x4_t s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10;
+    int16x4_t s11, s12, s13, s14;
+    uint8x8_t t0, t1, t2, t3;
+
+    do {
+      __builtin_prefetch(src_ptr + 0 * src_stride);
+      __builtin_prefetch(src_ptr + 1 * src_stride);
+      __builtin_prefetch(src_ptr + 2 * src_stride);
+      __builtin_prefetch(src_ptr + 3 * src_stride);
+
+      load_u8_8x4(src_ptr, src_stride, &t0, &t1, &t2, &t3); /*8x4*/
+      transpose_u8_8x4(&t0, &t1, &t2,
+                       &t3); /*first 8 pixels of 4 rows transposed-- 4x8*/
+
+      tt0 = vreinterpretq_s16_u16(vmovl_u8(t0));
+      tt1 = vreinterpretq_s16_u16(vmovl_u8(t1));
+      tt2 = vreinterpretq_s16_u16(vmovl_u8(t2));
+      tt3 = vreinterpretq_s16_u16(vmovl_u8(t3));
+
+      s0 = vget_low_s16(tt0);  /*pa0 pb0 pc0 pd0 -- pixel_a0*/
+      s1 = vget_low_s16(tt1);  /*pa1 pb1 pc1 pd1 */
+      s2 = vget_low_s16(tt2);  /*pa2 pb2 pc2 pd2 */
+      s3 = vget_low_s16(tt3);  /*pa3 pb3 pc3 pd3 */
+      s4 = vget_high_s16(tt0); /*pa4 pb4 pc4 pd4 */
+      s5 = vget_high_s16(tt1); /*pa5 pb5 pc5 pd5 */
+      s6 = vget_high_s16(tt2); /*pa6 pb6 pc6 pd6 */
+
+      __builtin_prefetch(dst_ptr + 0 * dst_stride);
+      __builtin_prefetch(dst_ptr + 1 * dst_stride);
+      __builtin_prefetch(dst_ptr + 2 * dst_stride);
+      __builtin_prefetch(dst_ptr + 3 * dst_stride);
+
+      s = src_ptr + 7;
+      d_tmp = dst_ptr;
+      width = w;
+
+      do {
+        load_u8_8x4(s, src_stride, &t0, &t1, &t2, &t3); /*8x4*/
+        transpose_u8_8x4(&t0, &t1, &t2, &t3);
+
+        tt0 = vreinterpretq_s16_u16(vmovl_u8(t0));
+        tt1 = vreinterpretq_s16_u16(vmovl_u8(t1));
+        tt2 = vreinterpretq_s16_u16(vmovl_u8(t2));
+        tt3 = vreinterpretq_s16_u16(vmovl_u8(t3));
+
+        s7 = vget_low_s16(tt0); /*pa7  pb7  pc7  pd7  */ /*4x8*/
+        s8 = vget_low_s16(tt1);   /*pa8  pb8  pc8  pd8  */
+        s9 = vget_low_s16(tt2);   /*pa9  pb9  pc9  pd9  */
+        s10 = vget_low_s16(tt3);  /*pa10 pb10 pc10 pd10 */
+        s11 = vget_high_s16(tt0); /*pa11 pb11 pc11 pd11 */
+        s12 = vget_high_s16(tt1); /*pa12 pb12 pc12 pd12 */
+        s13 = vget_high_s16(tt2); /*pa13 pb13 pc13 pd13 */
+        s14 = vget_high_s16(tt3); /*pa14 pb14 pc14 pd14 */
+
+        res0 = wiener_convolve8_horiz_4x8(
+            s0, s1, s2, s3, s4, s5, s6, filter_x_tmp, bd, conv_params->round_0);
+        res1 = wiener_convolve8_horiz_4x8(
+            s1, s2, s3, s4, s5, s6, s7, filter_x_tmp, bd, conv_params->round_0);
+        res2 = wiener_convolve8_horiz_4x8(
+            s2, s3, s4, s5, s6, s7, s8, filter_x_tmp, bd, conv_params->round_0);
+        res3 = wiener_convolve8_horiz_4x8(
+            s3, s4, s5, s6, s7, s8, s9, filter_x_tmp, bd, conv_params->round_0);
+        res4 =
+            wiener_convolve8_horiz_4x8(s4, s5, s6, s7, s8, s9, s10,
+                                       filter_x_tmp, bd, conv_params->round_0);
+        res5 =
+            wiener_convolve8_horiz_4x8(s5, s6, s7, s8, s9, s10, s11,
+                                       filter_x_tmp, bd, conv_params->round_0);
+        res6 =
+            wiener_convolve8_horiz_4x8(s6, s7, s8, s9, s10, s11, s12,
+                                       filter_x_tmp, bd, conv_params->round_0);
+        res7 =
+            wiener_convolve8_horiz_4x8(s7, s8, s9, s10, s11, s12, s13,
+                                       filter_x_tmp, bd, conv_params->round_0);
+
+        transpose_u16_4x8(&res0, &res1, &res2, &res3, &res4, &res5, &res6,
+                          &res7, &d0, &d1, &d2, &d3);
+
+        store_u16_8x4(d_tmp, MAX_SB_SIZE, d0, d1, d2, d3);
+
+        s0 = s8;
+        s1 = s9;
+        s2 = s10;
+        s3 = s11;
+        s4 = s12;
+        s5 = s13;
+        s6 = s14;
+        s += 8;
+        d_tmp += 8;
+        width -= 8;
+      } while (width > 0);
+
+      src_ptr += 4 * src_stride;
+      dst_ptr += 4 * MAX_SB_SIZE;
+      height -= 4;
+    } while (height > 0);
+  }
+
+  {
+    int16x8_t s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10;
+    uint8x8_t t0, t1, t2, t3;
+    int16_t *src_tmp_ptr, *s;
+    uint8_t *dst_tmp_ptr;
+    height = h;
+    width = w;
+    src_tmp_ptr = (int16_t *)temp;
+    dst_tmp_ptr = dst;
+    src_stride = MAX_SB_SIZE;
+
+    do {
+      s = src_tmp_ptr;
+      s0 = vld1q_s16(s);
+      s += src_stride;
+      s1 = vld1q_s16(s);
+      s += src_stride;
+      s2 = vld1q_s16(s);
+      s += src_stride;
+      s3 = vld1q_s16(s);
+      s += src_stride;
+      s4 = vld1q_s16(s);
+      s += src_stride;
+      s5 = vld1q_s16(s);
+      s += src_stride;
+      s6 = vld1q_s16(s);
+      s += src_stride;
+      d = dst_tmp_ptr;
+      height = h;
+
+      do {
+        __builtin_prefetch(dst_tmp_ptr + 0 * dst_stride);
+        __builtin_prefetch(dst_tmp_ptr + 1 * dst_stride);
+        __builtin_prefetch(dst_tmp_ptr + 2 * dst_stride);
+        __builtin_prefetch(dst_tmp_ptr + 3 * dst_stride);
+
+        s7 = vld1q_s16(s);
+        s += src_stride;
+        s8 = vld1q_s16(s);
+        s += src_stride;
+        s9 = vld1q_s16(s);
+        s += src_stride;
+        s10 = vld1q_s16(s);
+        s += src_stride;
+
+        t0 = wiener_convolve8_vert_4x8(s0, s1, s2, s3, s4, s5, s6, filter_y_tmp,
+                                       bd, conv_params->round_1);
+        t1 = wiener_convolve8_vert_4x8(s1, s2, s3, s4, s5, s6, s7, filter_y_tmp,
+                                       bd, conv_params->round_1);
+        t2 = wiener_convolve8_vert_4x8(s2, s3, s4, s5, s6, s7, s8, filter_y_tmp,
+                                       bd, conv_params->round_1);
+        t3 = wiener_convolve8_vert_4x8(s3, s4, s5, s6, s7, s8, s9, filter_y_tmp,
+                                       bd, conv_params->round_1);
+
+        vst1_u8(d, t0);
+        d += dst_stride;
+        vst1_u8(d, t1);
+        d += dst_stride;
+        vst1_u8(d, t2);
+        d += dst_stride;
+        vst1_u8(d, t3);
+        d += dst_stride;
+
+        s0 = s4;
+        s1 = s5;
+        s2 = s6;
+        s3 = s7;
+        s4 = s8;
+        s5 = s9;
+        s6 = s10;
+        height -= 4;
+      } while (height > 3);
+
+      if (height != 0) {
+        __builtin_prefetch(dst_tmp_ptr + 0 * dst_stride);
+        __builtin_prefetch(dst_tmp_ptr + 1 * dst_stride);
+
+        do {
+          s7 = vld1q_s16(s);
+          s += src_stride;
+
+          t0 =
+              wiener_convolve8_vert_4x8(s0, s1, s2, s3, s4, s5, s6,
+                                        filter_y_tmp, bd, conv_params->round_1);
+          vst1_u8(d, t0);
+          d += dst_stride;
+
+          s0 = s1;
+          s1 = s2;
+          s2 = s3;
+          s3 = s4;
+          s4 = s5;
+          s5 = s6;
+          s6 = s7;
+          height -= 1;
+        } while (height > 0);
+      }
+
+      src_tmp_ptr += 8;
+      dst_tmp_ptr += 8;
+
+      w -= 8;
+    } while (w > 0);
+  }
+}

third_party/aom/av1/common/av1_fwd_txfm1d.h

@@ -1,61 +0,0 @@
-/*
- * 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.
- */
-
-#ifndef AV1_FWD_TXFM1D_H_
-#define AV1_FWD_TXFM1D_H_
-
-#include "av1/common/av1_txfm.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-void av1_fdct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
-                   const int8_t *stage_range);
-void av1_fdct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
-                   const int8_t *stage_range);
-void av1_fdct16_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fdct32_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_TX64X64
-void av1_fdct64_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-#endif  // CONFIG_TX64X64
-
-void av1_fadst4_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fadst8_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fadst16_new(const int32_t *input, int32_t *output,
-                     const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fadst32_new(const int32_t *input, int32_t *output,
-                     const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_EXT_TX
-void av1_fidentity4_c(const int32_t *input, int32_t *output,
-                      const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fidentity8_c(const int32_t *input, int32_t *output,
-                      const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fidentity16_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-void av1_fidentity32_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_TX64X64
-void av1_fidentity64_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_EXT_TX
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif  // AV1_FWD_TXFM1D_H_

third_party/aom/av1/common/av1_fwd_txfm1d_cfg.h

@@ -1,363 +0,0 @@
-/*
- * 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.
- */
-
-#ifndef AV1_FWD_TXFM2D_CFG_H_
-#define AV1_FWD_TXFM2D_CFG_H_
-#include "av1/common/enums.h"
-#include "av1/common/av1_fwd_txfm1d.h"
-
-//  ---------------- 4x4 1D constants -----------------------
-// shift
-static const int8_t fwd_shift_4[3] = { 2, 0, 0 };
-
-// stage range
-static const int8_t fwd_stage_range_col_dct_4[4] = { 0, 1, 2, 2 };
-static const int8_t fwd_stage_range_row_dct_4[4] = { 2, 3, 3, 3 };
-static const int8_t fwd_stage_range_col_adst_4[6] = { 0, 0, 1, 2, 2, 2 };
-static const int8_t fwd_stage_range_row_adst_4[6] = { 2, 2, 2, 3, 3, 3 };
-static const int8_t fwd_stage_range_idx_4[1] = { 0 };
-
-// cos bit
-static const int8_t fwd_cos_bit_col_dct_4[4] = { 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_dct_4[4] = { 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_col_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
-
-//  ---------------- 8x8 1D constants -----------------------
-// shift
-static const int8_t fwd_shift_8[3] = { 2, -1, 0 };
-
-// stage range
-static const int8_t fwd_stage_range_col_dct_8[6] = { 0, 1, 2, 3, 3, 3 };
-static const int8_t fwd_stage_range_row_dct_8[6] = { 3, 4, 5, 5, 5, 5 };
-static const int8_t fwd_stage_range_col_adst_8[8] = { 0, 0, 1, 2, 2, 3, 3, 3 };
-static const int8_t fwd_stage_range_row_adst_8[8] = { 3, 3, 3, 4, 4, 5, 5, 5 };
-static const int8_t fwd_stage_range_idx_8[1] = { 0 };
-
-// cos bit
-static const int8_t fwd_cos_bit_col_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_col_adst_8[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
-};
-static const int8_t fwd_cos_bit_row_adst_8[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
-};
-
-//  ---------------- 16x16 1D constants -----------------------
-// shift
-static const int8_t fwd_shift_16[3] = { 2, -2, 0 };
-
-// stage range
-static const int8_t fwd_stage_range_col_dct_16[8] = { 0, 1, 2, 3, 4, 4, 4, 4 };
-static const int8_t fwd_stage_range_row_dct_16[8] = { 4, 5, 6, 7, 7, 7, 7, 7 };
-static const int8_t fwd_stage_range_col_adst_16[10] = { 0, 0, 1, 2, 2,
-                                                        3, 3, 4, 4, 4 };
-static const int8_t fwd_stage_range_row_adst_16[10] = {
-  4, 4, 4, 5, 5, 6, 6, 7, 7, 7,
-};
-static const int8_t fwd_stage_range_idx_16[1] = { 0 };
-
-// cos bit
-static const int8_t fwd_cos_bit_col_dct_16[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
-};
-static const int8_t fwd_cos_bit_row_dct_16[8] = {
-  12, 12, 12, 12, 12, 12, 12, 12
-};
-static const int8_t fwd_cos_bit_col_adst_16[10] = { 13, 13, 13, 13, 13,
-                                                    13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_adst_16[10] = { 12, 12, 12, 12, 12,
-                                                    12, 12, 12, 12, 12 };
-
-//  ---------------- 32x32 1D constants -----------------------
-// shift
-static const int8_t fwd_shift_32[3] = { 2, -4, 0 };
-
-// stage range
-static const int8_t fwd_stage_range_col_dct_32[10] = { 0, 1, 2, 3, 4,
-                                                       5, 5, 5, 5, 5 };
-static const int8_t fwd_stage_range_row_dct_32[10] = { 5, 6, 7, 8, 9,
-                                                       9, 9, 9, 9, 9 };
-static const int8_t fwd_stage_range_col_adst_32[12] = { 0, 0, 1, 2, 2, 3,
-                                                        3, 4, 4, 5, 5, 5 };
-static const int8_t fwd_stage_range_row_adst_32[12] = { 5, 5, 5, 6, 6, 7,
-                                                        7, 8, 8, 9, 9, 9 };
-static const int8_t fwd_stage_range_idx_32[1] = { 0 };
-
-// cos bit
-static const int8_t fwd_cos_bit_col_dct_32[10] = { 12, 12, 12, 12, 12,
-                                                   12, 12, 12, 12, 12 };
-static const int8_t fwd_cos_bit_row_dct_32[10] = { 12, 12, 12, 12, 12,
-                                                   12, 12, 12, 12, 12 };
-static const int8_t fwd_cos_bit_col_adst_32[12] = { 12, 12, 12, 12, 12, 12,
-                                                    12, 12, 12, 12, 12, 12 };
-static const int8_t fwd_cos_bit_row_adst_32[12] = { 12, 12, 12, 12, 12, 12,
-                                                    12, 12, 12, 12, 12, 12 };
-
-//  ---------------- 64x64 1D constants -----------------------
-// shift
-static const int8_t fwd_shift_64[3] = { 0, -2, -2 };
-
-// stage range
-static const int8_t fwd_stage_range_col_dct_64[12] = { 0, 1, 2, 3, 4, 5,
-                                                       6, 6, 6, 6, 6, 6 };
-static const int8_t fwd_stage_range_row_dct_64[12] = { 6,  7,  8,  9,  10, 11,
-                                                       11, 11, 11, 11, 11, 11 };
-static const int8_t fwd_stage_range_idx_64[1] = { 0 };
-
-// cos bit
-static const int8_t fwd_cos_bit_col_dct_64[12] = { 15, 15, 15, 15, 15, 14,
-                                                   13, 13, 13, 13, 13, 13 };
-static const int8_t fwd_cos_bit_row_dct_64[12] = { 15, 14, 13, 12, 11, 10,
-                                                   10, 10, 10, 10, 10, 10 };
-
-//  ---------------- row config fwd_dct_4 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_dct_4 = {
-  4,  // .txfm_size
-  4,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_4,                // .shift
-  fwd_stage_range_row_dct_4,  // .stage_range
-  fwd_cos_bit_row_dct_4,      // .cos_bit
-  TXFM_TYPE_DCT4              // .txfm_type
-};
-
-//  ---------------- row config fwd_dct_8 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_dct_8 = {
-  8,  // .txfm_size
-  6,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_8,                // .shift
-  fwd_stage_range_row_dct_8,  // .stage_range
-  fwd_cos_bit_row_dct_8,      // .cos_bit_
-  TXFM_TYPE_DCT8              // .txfm_type
-};
-//  ---------------- row config fwd_dct_16 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_dct_16 = {
-  16,  // .txfm_size
-  8,   // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_16,                // .shift
-  fwd_stage_range_row_dct_16,  // .stage_range
-  fwd_cos_bit_row_dct_16,      // .cos_bit
-  TXFM_TYPE_DCT16              // .txfm_type
-};
-
-//  ---------------- row config fwd_dct_32 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_dct_32 = {
-  32,  // .txfm_size
-  10,  // .stage_num
-  // 1,  // .log_scale
-  fwd_shift_32,                // .shift
-  fwd_stage_range_row_dct_32,  // .stage_range
-  fwd_cos_bit_row_dct_32,      // .cos_bit_row
-  TXFM_TYPE_DCT32              // .txfm_type
-};
-
-//  ---------------- row config fwd_dct_64 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_dct_64 = {
-  64,                          // .txfm_size
-  12,                          // .stage_num
-  fwd_shift_64,                // .shift
-  fwd_stage_range_row_dct_64,  // .stage_range
-  fwd_cos_bit_row_dct_64,      // .cos_bit
-  TXFM_TYPE_DCT64,             // .txfm_type_col
-};
-
-//  ---------------- row config fwd_adst_4 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_adst_4 = {
-  4,  // .txfm_size
-  6,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_4,                 // .shift
-  fwd_stage_range_row_adst_4,  // .stage_range
-  fwd_cos_bit_row_adst_4,      // .cos_bit
-  TXFM_TYPE_ADST4,             // .txfm_type
-};
-
-//  ---------------- row config fwd_adst_8 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_adst_8 = {
-  8,  // .txfm_size
-  8,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_8,                 // .shift
-  fwd_stage_range_row_adst_8,  // .stage_range
-  fwd_cos_bit_row_adst_8,      // .cos_bit
-  TXFM_TYPE_ADST8,             // .txfm_type_col
-};
-
-//  ---------------- row config fwd_adst_16 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_adst_16 = {
-  16,  // .txfm_size
-  10,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_16,                 // .shift
-  fwd_stage_range_row_adst_16,  // .stage_range
-  fwd_cos_bit_row_adst_16,      // .cos_bit
-  TXFM_TYPE_ADST16,             // .txfm_type
-};
-
-//  ---------------- row config fwd_adst_32 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_row_cfg_adst_32 = {
-  32,  // .txfm_size
-  12,  // .stage_num
-  // 1,  // .log_scale
-  fwd_shift_32,                 // .shift
-  fwd_stage_range_row_adst_32,  // .stage_range
-  fwd_cos_bit_row_adst_32,      // .cos_bit
-  TXFM_TYPE_ADST32,             // .txfm_type
-};
-
-//  ---------------- col config fwd_dct_4 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_dct_4 = {
-  4,  // .txfm_size
-  4,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_4,                // .shift
-  fwd_stage_range_col_dct_4,  // .stage_range
-  fwd_cos_bit_col_dct_4,      // .cos_bit
-  TXFM_TYPE_DCT4              // .txfm_type
-};
-
-//  ---------------- col config fwd_dct_8 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_dct_8 = {
-  8,  // .txfm_size
-  6,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_8,                // .shift
-  fwd_stage_range_col_dct_8,  // .stage_range
-  fwd_cos_bit_col_dct_8,      // .cos_bit_
-  TXFM_TYPE_DCT8              // .txfm_type
-};
-//  ---------------- col config fwd_dct_16 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_dct_16 = {
-  16,  // .txfm_size
-  8,   // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_16,                // .shift
-  fwd_stage_range_col_dct_16,  // .stage_range
-  fwd_cos_bit_col_dct_16,      // .cos_bit
-  TXFM_TYPE_DCT16              // .txfm_type
-};
-
-//  ---------------- col config fwd_dct_32 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_dct_32 = {
-  32,  // .txfm_size
-  10,  // .stage_num
-  // 1,  // .log_scale
-  fwd_shift_32,                // .shift
-  fwd_stage_range_col_dct_32,  // .stage_range
-  fwd_cos_bit_col_dct_32,      // .cos_bit_col
-  TXFM_TYPE_DCT32              // .txfm_type
-};
-
-//  ---------------- col config fwd_dct_64 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_dct_64 = {
-  64,                          // .txfm_size
-  12,                          // .stage_num
-  fwd_shift_64,                // .shift
-  fwd_stage_range_col_dct_64,  // .stage_range
-  fwd_cos_bit_col_dct_64,      // .cos_bit
-  TXFM_TYPE_DCT64,             // .txfm_type_col
-};
-
-//  ---------------- col config fwd_adst_4 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_adst_4 = {
-  4,  // .txfm_size
-  6,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_4,                 // .shift
-  fwd_stage_range_col_adst_4,  // .stage_range
-  fwd_cos_bit_col_adst_4,      // .cos_bit
-  TXFM_TYPE_ADST4,             // .txfm_type
-};
-
-//  ---------------- col config fwd_adst_8 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_adst_8 = {
-  8,  // .txfm_size
-  8,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_8,                 // .shift
-  fwd_stage_range_col_adst_8,  // .stage_range
-  fwd_cos_bit_col_adst_8,      // .cos_bit
-  TXFM_TYPE_ADST8,             // .txfm_type_col
-};
-
-//  ---------------- col config fwd_adst_16 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_adst_16 = {
-  16,  // .txfm_size
-  10,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_16,                 // .shift
-  fwd_stage_range_col_adst_16,  // .stage_range
-  fwd_cos_bit_col_adst_16,      // .cos_bit
-  TXFM_TYPE_ADST16,             // .txfm_type
-};
-
-//  ---------------- col config fwd_adst_32 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_col_cfg_adst_32 = {
-  32,  // .txfm_size
-  12,  // .stage_num
-  // 1,  // .log_scale
-  fwd_shift_32,                 // .shift
-  fwd_stage_range_col_adst_32,  // .stage_range
-  fwd_cos_bit_col_adst_32,      // .cos_bit
-  TXFM_TYPE_ADST32,             // .txfm_type
-};
-
-#if CONFIG_EXT_TX
-// identity does not need to differentiate between row and col
-//  ---------------- row/col config fwd_identity_4 ----------
-static const TXFM_1D_CFG fwd_txfm_1d_cfg_identity_4 = {
-  4,  // .txfm_size
-  1,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_4,            // .shift
-  fwd_stage_range_idx_4,  // .stage_range
-  NULL,                   // .cos_bit
-  TXFM_TYPE_IDENTITY4,    // .txfm_type
-};
-
-//  ---------------- row/col config fwd_identity_8 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_cfg_identity_8 = {
-  8,  // .txfm_size
-  1,  // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_8,            // .shift
-  fwd_stage_range_idx_8,  // .stage_range
-  NULL,                   // .cos_bit
-  TXFM_TYPE_IDENTITY8,    // .txfm_type
-};
-
-//  ---------------- row/col config fwd_identity_16 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_cfg_identity_16 = {
-  16,  // .txfm_size
-  1,   // .stage_num
-  // 0,  // .log_scale
-  fwd_shift_16,            // .shift
-  fwd_stage_range_idx_16,  // .stage_range
-  NULL,                    // .cos_bit
-  TXFM_TYPE_IDENTITY16,    // .txfm_type
-};
-
-//  ---------------- row/col config fwd_identity_32 ----------------
-static const TXFM_1D_CFG fwd_txfm_1d_cfg_identity_32 = {
-  32,  // .txfm_size
-  1,   // .stage_num
-  // 1,  // .log_scale
-  fwd_shift_32,            // .shift
-  fwd_stage_range_idx_32,  // .stage_range
-  NULL,                    // .cos_bit
-  TXFM_TYPE_IDENTITY32,    // .txfm_type
-};
-#endif  // CONFIG_EXT_TX
-#endif  // AV1_FWD_TXFM2D_CFG_H_

third_party/aom/av1/common/av1_fwd_txfm2d.c

@@ -1,413 +0,0 @@
-/*
- * 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 <assert.h>
-
-#include "./av1_rtcd.h"
-#include "aom_dsp/txfm_common.h"
-#include "av1/common/enums.h"
-#include "av1/common/av1_fwd_txfm1d.h"
-#include "av1/common/av1_fwd_txfm1d_cfg.h"
-#include "av1/common/av1_txfm.h"
-
-static INLINE TxfmFunc fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
-  switch (txfm_type) {
-    case TXFM_TYPE_DCT4: return av1_fdct4_new;
-    case TXFM_TYPE_DCT8: return av1_fdct8_new;
-    case TXFM_TYPE_DCT16: return av1_fdct16_new;
-    case TXFM_TYPE_DCT32: return av1_fdct32_new;
-#if CONFIG_TX64X64
-    case TXFM_TYPE_DCT64: return av1_fdct64_new;
-#endif  // CONFIG_TX64X64
-    case TXFM_TYPE_ADST4: return av1_fadst4_new;
-    case TXFM_TYPE_ADST8: return av1_fadst8_new;
-    case TXFM_TYPE_ADST16: return av1_fadst16_new;
-    case TXFM_TYPE_ADST32: return av1_fadst32_new;
-#if CONFIG_EXT_TX
-    case TXFM_TYPE_IDENTITY4: return av1_fidentity4_c;
-    case TXFM_TYPE_IDENTITY8: return av1_fidentity8_c;
-    case TXFM_TYPE_IDENTITY16: return av1_fidentity16_c;
-    case TXFM_TYPE_IDENTITY32: return av1_fidentity32_c;
-#if CONFIG_TX64X64
-    case TXFM_TYPE_IDENTITY64: return av1_fidentity64_c;
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_EXT_TX
-    default: assert(0); return NULL;
-  }
-}
-
-void av1_gen_fwd_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
-                             const TXFM_2D_FLIP_CFG *cfg, int bd) {
-  // Note when assigning txfm_size_col, we use the txfm_size from the
-  // row configuration and vice versa. This is intentionally done to
-  // accurately perform rectangular transforms. When the transform is
-  // rectangular, the number of columns will be the same as the
-  // txfm_size stored in the row cfg struct. It will make no difference
-  // for square transforms.
-  const int txfm_size_col = cfg->row_cfg->txfm_size;
-  const int txfm_size_row = cfg->col_cfg->txfm_size;
-  // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
-  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->col_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_col[i] = cfg->col_cfg->stage_range[i] + shift[0] + bd + 1;
-  }
-
-  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->row_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_row[i] =
-        cfg->row_cfg->stage_range[i] + shift[0] + shift[1] + bd + 1;
-  }
-}
-
-static INLINE void fwd_txfm2d_c(const int16_t *input, int32_t *output,
-                                const int stride, const TXFM_2D_FLIP_CFG *cfg,
-                                int32_t *buf, int bd) {
-  int c, r;
-  // Note when assigning txfm_size_col, we use the txfm_size from the
-  // row configuration and vice versa. This is intentionally done to
-  // accurately perform rectangular transforms. When the transform is
-  // rectangular, the number of columns will be the same as the
-  // txfm_size stored in the row cfg struct. It will make no difference
-  // for square transforms.
-  const int txfm_size_col = cfg->row_cfg->txfm_size;
-  const int txfm_size_row = cfg->col_cfg->txfm_size;
-  // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
-  int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
-  int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
-  assert(cfg->col_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
-  assert(cfg->row_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
-  av1_gen_fwd_stage_range(stage_range_col, stage_range_row, cfg, bd);
-
-  const int8_t *cos_bit_col = cfg->col_cfg->cos_bit;
-  const int8_t *cos_bit_row = cfg->row_cfg->cos_bit;
-  const TxfmFunc txfm_func_col = fwd_txfm_type_to_func(cfg->col_cfg->txfm_type);
-  const TxfmFunc txfm_func_row = fwd_txfm_type_to_func(cfg->row_cfg->txfm_type);
-
-  // use output buffer as temp buffer
-  int32_t *temp_in = output;
-  int32_t *temp_out = output + txfm_size_row;
-
-  // Columns
-  for (c = 0; c < txfm_size_col; ++c) {
-    if (cfg->ud_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r) temp_in[r] = input[r * stride + c];
-    } else {
-      for (r = 0; r < txfm_size_row; ++r)
-        // flip upside down
-        temp_in[r] = input[(txfm_size_row - r - 1) * stride + c];
-    }
-    round_shift_array(temp_in, txfm_size_row, -shift[0]);
-    // Multiply everything by Sqrt2 on the larger dimension if the
-    // transform is rectangular
-    if (txfm_size_col > txfm_size_row) {
-      for (r = 0; r < txfm_size_row; ++r)
-        temp_in[r] = (int32_t)fdct_round_shift(temp_in[r] * Sqrt2);
-    }
-    txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
-    round_shift_array(temp_out, txfm_size_row, -shift[1]);
-    if (cfg->lr_flip == 0) {
-      for (r = 0; r < txfm_size_row; ++r)
-        buf[r * txfm_size_col + c] = temp_out[r];
-    } else {
-      for (r = 0; r < txfm_size_row; ++r)
-        // flip from left to right
-        buf[r * txfm_size_col + (txfm_size_col - c - 1)] = temp_out[r];
-    }
-  }
-
-  // Rows
-  for (r = 0; r < txfm_size_row; ++r) {
-    // Multiply everything by Sqrt2 on the larger dimension if the
-    // transform is rectangular
-    if (txfm_size_row > txfm_size_col) {
-      for (c = 0; c < txfm_size_col; ++c)
-        buf[r * txfm_size_col + c] =
-            (int32_t)fdct_round_shift(buf[r * txfm_size_col + c] * Sqrt2);
-    }
-    txfm_func_row(buf + r * txfm_size_col, output + r * txfm_size_col,
-                  cos_bit_row, stage_range_row);
-    round_shift_array(output + r * txfm_size_col, txfm_size_col, -shift[2]);
-  }
-}
-
-void av1_fwd_txfm2d_4x8_c(const int16_t *input, int32_t *output, int stride,
-                          TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int32_t txfm_buf[4 * 8];
-  int16_t rinput[4 * 8];
-  TX_SIZE tx_size = TX_4X8;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(rtx_type, rtx_size);
-  fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
-  transpose_int32(output, w, txfm_buf, rw, rw, rh);
-#else
-  int32_t txfm_buf[4 * 8];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_4X8);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-#endif
-}
-
-void av1_fwd_txfm2d_8x4_c(const int16_t *input, int32_t *output, int stride,
-                          TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[8 * 4];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_8X4);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_8x16_c(const int16_t *input, int32_t *output, int stride,
-                           TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int32_t txfm_buf[8 * 16];
-  int16_t rinput[8 * 16];
-  TX_SIZE tx_size = TX_8X16;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(rtx_type, rtx_size);
-  fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
-  transpose_int32(output, w, txfm_buf, rw, rw, rh);
-#else
-  int32_t txfm_buf[8 * 16];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_8X16);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-#endif
-}
-
-void av1_fwd_txfm2d_16x8_c(const int16_t *input, int32_t *output, int stride,
-                           TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[16 * 8];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_16X8);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_16x32_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int32_t txfm_buf[16 * 32];
-  int16_t rinput[16 * 32];
-  TX_SIZE tx_size = TX_16X32;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int16(rinput, rw, input, stride, w, h);
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(rtx_type, rtx_size);
-  fwd_txfm2d_c(rinput, txfm_buf, rw, &cfg, output, bd);
-  transpose_int32(output, w, txfm_buf, rw, rw, rh);
-#else
-  int32_t txfm_buf[16 * 32];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_16X32);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-#endif
-}
-
-void av1_fwd_txfm2d_32x16_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[32 * 16];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_32X16);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_4x4_c(const int16_t *input, int32_t *output, int stride,
-                          TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[4 * 4];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_4X4);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_8x8_c(const int16_t *input, int32_t *output, int stride,
-                          TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[8 * 8];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_8X8);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_16x16_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[16 * 16];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_16X16);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_32x32_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[32 * 32];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_32X32);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-#if CONFIG_TX64X64
-void av1_fwd_txfm2d_64x64_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[64 * 64];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_64x64_cfg(tx_type);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_32x64_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[32 * 64];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_32x64_cfg(tx_type);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-
-void av1_fwd_txfm2d_64x32_c(const int16_t *input, int32_t *output, int stride,
-                            TX_TYPE tx_type, int bd) {
-  int32_t txfm_buf[64 * 32];
-  TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_64x32_cfg(tx_type);
-  fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf, bd);
-}
-#endif  // CONFIG_TX64X64
-
-static const TXFM_1D_CFG *fwd_txfm_col_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
-  // DCT
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_col_cfg_dct_4, &fwd_txfm_1d_col_cfg_dct_8,
-      &fwd_txfm_1d_col_cfg_dct_16, &fwd_txfm_1d_col_cfg_dct_32 },
-  // ADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_col_cfg_adst_4, &fwd_txfm_1d_col_cfg_adst_8,
-      &fwd_txfm_1d_col_cfg_adst_16, &fwd_txfm_1d_col_cfg_adst_32 },
-#if CONFIG_EXT_TX
-  // FLIPADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_col_cfg_adst_4, &fwd_txfm_1d_col_cfg_adst_8,
-      &fwd_txfm_1d_col_cfg_adst_16, &fwd_txfm_1d_col_cfg_adst_32 },
-  // IDENTITY
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_cfg_identity_4, &fwd_txfm_1d_cfg_identity_8,
-      &fwd_txfm_1d_cfg_identity_16, &fwd_txfm_1d_cfg_identity_32 },
-#endif  // CONFIG_EXT_TX
-};
-
-static const TXFM_1D_CFG *fwd_txfm_row_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
-  // DCT
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_row_cfg_dct_4, &fwd_txfm_1d_row_cfg_dct_8,
-      &fwd_txfm_1d_row_cfg_dct_16, &fwd_txfm_1d_row_cfg_dct_32 },
-  // ADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_row_cfg_adst_4, &fwd_txfm_1d_row_cfg_adst_8,
-      &fwd_txfm_1d_row_cfg_adst_16, &fwd_txfm_1d_row_cfg_adst_32 },
-#if CONFIG_EXT_TX
-  // FLIPADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_row_cfg_adst_4, &fwd_txfm_1d_row_cfg_adst_8,
-      &fwd_txfm_1d_row_cfg_adst_16, &fwd_txfm_1d_row_cfg_adst_32 },
-  // IDENTITY
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &fwd_txfm_1d_cfg_identity_4, &fwd_txfm_1d_cfg_identity_8,
-      &fwd_txfm_1d_cfg_identity_16, &fwd_txfm_1d_cfg_identity_32 },
-#endif  // CONFIG_EXT_TX
-};
-
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size) {
-  TXFM_2D_FLIP_CFG cfg;
-  set_flip_cfg(tx_type, &cfg);
-  const TX_TYPE_1D tx_type_col = vtx_tab[tx_type];
-  const TX_TYPE_1D tx_type_row = htx_tab[tx_type];
-  const TX_SIZE tx_size_col = txsize_vert_map[tx_size];
-  const TX_SIZE tx_size_row = txsize_horz_map[tx_size];
-  cfg.col_cfg = fwd_txfm_col_cfg_ls[tx_type_col][tx_size_col];
-  cfg.row_cfg = fwd_txfm_row_cfg_ls[tx_type_row][tx_size_row];
-  return cfg;
-}
-
-#if CONFIG_TX64X64
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_32x64_cfg(TX_TYPE tx_type) {
-  TXFM_2D_FLIP_CFG cfg;
-  const TX_TYPE_1D tx_type_row = htx_tab[tx_type];
-  const TX_SIZE tx_size_row = txsize_horz_map[TX_32X64];
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = &fwd_txfm_1d_col_cfg_dct_64;
-      cfg.row_cfg = fwd_txfm_row_cfg_ls[tx_type_row][tx_size_row];
-      cfg.ud_flip = 0;
-      cfg.lr_flip = 0;
-      break;
-    default: assert(0);
-  }
-  return cfg;
-}
-
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x32_cfg(TX_TYPE tx_type) {
-  TXFM_2D_FLIP_CFG cfg;
-  const TX_TYPE_1D tx_type_col = vtx_tab[tx_type];
-  const TX_SIZE tx_size_col = txsize_vert_map[TX_64X32];
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = fwd_txfm_col_cfg_ls[tx_type_col][tx_size_col];
-      cfg.row_cfg = &fwd_txfm_1d_row_cfg_dct_64;
-      cfg.ud_flip = 0;
-      cfg.lr_flip = 0;
-      break;
-    default: assert(0);
-  }
-  return cfg;
-}
-
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(TX_TYPE tx_type) {
-  TXFM_2D_FLIP_CFG cfg;
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = &fwd_txfm_1d_col_cfg_dct_64;
-      cfg.row_cfg = &fwd_txfm_1d_row_cfg_dct_64;
-      cfg.ud_flip = 0;
-      cfg.lr_flip = 0;
-      break;
-    default:
-      cfg.ud_flip = 0;
-      cfg.lr_flip = 0;
-      assert(0);
-  }
-  return cfg;
-}
-#endif  // CONFIG_TX64X64

third_party/aom/av1/common/av1_inv_txfm1d.h

@@ -18,41 +18,41 @@
 extern "C" {
 #endif
 
-void av1_idct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
-                   const int8_t *stage_range);
-void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
-                   const int8_t *stage_range);
-void av1_idct16_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_idct32_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_TX64X64
-void av1_idct64_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-#endif  // CONFIG_TX64X64
+static INLINE int32_t clamp_value(int32_t value, int8_t bit) {
+  if (bit <= 0) return value;  // Do nothing for invalid clamp bit.
+  const int64_t max_value = (1LL << (bit - 1)) - 1;
+  const int64_t min_value = -(1LL << (bit - 1));
+  return (int32_t)clamp64(value, min_value, max_value);
+}
 
-void av1_iadst4_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iadst8_new(const int32_t *input, int32_t *output,
-                    const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iadst16_new(const int32_t *input, int32_t *output,
-                     const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iadst32_new(const int32_t *input, int32_t *output,
-                     const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_EXT_TX
-void av1_iidentity4_c(const int32_t *input, int32_t *output,
-                      const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iidentity8_c(const int32_t *input, int32_t *output,
-                      const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iidentity16_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-void av1_iidentity32_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-#if CONFIG_TX64X64
-void av1_iidentity64_c(const int32_t *input, int32_t *output,
-                       const int8_t *cos_bit, const int8_t *stage_range);
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_EXT_TX
+static INLINE void clamp_buf(int32_t *buf, int32_t size, int8_t bit) {
+  for (int i = 0; i < size; ++i) buf[i] = clamp_value(buf[i], bit);
+}
+
+void av1_idct4_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                   const int8_t *stage_range);
+void av1_idct8_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                   const int8_t *stage_range);
+void av1_idct16_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                    const int8_t *stage_range);
+void av1_idct32_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                    const int8_t *stage_range);
+void av1_idct64_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                    const int8_t *stage_range);
+void av1_iadst4_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                    const int8_t *stage_range);
+void av1_iadst8_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                    const int8_t *stage_range);
+void av1_iadst16_new(const int32_t *input, int32_t *output, int8_t cos_bit,
+                     const int8_t *stage_range);
+void av1_iidentity4_c(const int32_t *input, int32_t *output, int8_t cos_bit,
+                      const int8_t *stage_range);
+void av1_iidentity8_c(const int32_t *input, int32_t *output, int8_t cos_bit,
+                      const int8_t *stage_range);
+void av1_iidentity16_c(const int32_t *input, int32_t *output, int8_t cos_bit,
+                       const int8_t *stage_range);
+void av1_iidentity32_c(const int32_t *input, int32_t *output, int8_t cos_bit,
+                       const int8_t *stage_range);
 
 #ifdef __cplusplus
 }

third_party/aom/av1/common/av1_inv_txfm1d_cfg.h

@@ -14,358 +14,34 @@
 #include "av1/common/av1_inv_txfm1d.h"
 
 // sum of fwd_shift_##
-#if CONFIG_CHROMA_2X2
-#if CONFIG_TX64X64
-static const int8_t fwd_shift_sum[TX_SIZES] = { 3, 2, 1, 0, -2, -4 };
-#else   // CONFIG_TX64X64
-static const int8_t fwd_shift_sum[TX_SIZES] = { 3, 2, 1, 0, -2 };
-#endif  // CONFIG_TX64X64
-#else   // CONFIG_CHROMA_2X2
-#if CONFIG_TX64X64
-static const int8_t fwd_shift_sum[TX_SIZES] = { 2, 1, 0, -2, -4 };
-#else  // CONFIG_TX64X64
-static const int8_t fwd_shift_sum[TX_SIZES] = { 2, 1, 0, -2 };
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_CHROMA_2X2
-
-//  ---------------- 4x4 1D config -----------------------
-// shift
-static const int8_t inv_shift_4[2] = { 0, -4 };
-
-// stage range
-static const int8_t inv_stage_range_col_dct_4[4] = { 3, 3, 2, 2 };
-static const int8_t inv_stage_range_row_dct_4[4] = { 3, 3, 3, 3 };
-static const int8_t inv_stage_range_col_adst_4[6] = { 3, 3, 3, 3, 2, 2 };
-static const int8_t inv_stage_range_row_adst_4[6] = { 3, 3, 3, 3, 3, 3 };
-static const int8_t inv_stage_range_idx_4[1] = { 0 };
-
-// cos bit
-static const int8_t inv_cos_bit_col_dct_4[4] = { 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_dct_4[4] = { 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_col_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
-
-//  ---------------- 8x8 1D constants -----------------------
-// shift
-static const int8_t inv_shift_8[2] = { 0, -5 };
-
-// stage range
-static const int8_t inv_stage_range_col_dct_8[6] = { 5, 5, 5, 5, 4, 4 };
-static const int8_t inv_stage_range_row_dct_8[6] = { 5, 5, 5, 5, 5, 5 };
-static const int8_t inv_stage_range_col_adst_8[8] = { 5, 5, 5, 5, 5, 5, 4, 4 };
-static const int8_t inv_stage_range_row_adst_8[8] = { 5, 5, 5, 5, 5, 5, 5, 5 };
-static const int8_t inv_stage_range_idx_8[1] = { 0 };
-
-// cos bit
-static const int8_t inv_cos_bit_col_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_col_adst_8[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
-};
-static const int8_t inv_cos_bit_row_adst_8[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
+static const int8_t inv_start_range[TX_SIZES_ALL] = {
+  5,  // 4x4 transform
+  6,  // 8x8 transform
+  7,  // 16x16 transform
+  7,  // 32x32 transform
+  7,  // 64x64 transform
+  5,  // 4x8 transform
+  5,  // 8x4 transform
+  6,  // 8x16 transform
+  6,  // 16x8 transform
+  6,  // 16x32 transform
+  6,  // 32x16 transform
+  6,  // 32x64 transform
+  6,  // 64x32 transform
+  6,  // 4x16 transform
+  6,  // 16x4 transform
+  7,  // 8x32 transform
+  7,  // 32x8 transform
+  7,  // 16x64 transform
+  7,  // 64x16 transform
 };
 
-//  ---------------- 16x16 1D constants -----------------------
-// shift
-static const int8_t inv_shift_16[2] = { -1, -5 };
+extern const int8_t *inv_txfm_shift_ls[TX_SIZES_ALL];
 
-// stage range
-static const int8_t inv_stage_range_col_dct_16[8] = { 7, 7, 7, 7, 7, 7, 6, 6 };
-static const int8_t inv_stage_range_row_dct_16[8] = { 7, 7, 7, 7, 7, 7, 7, 7 };
-static const int8_t inv_stage_range_col_adst_16[10] = { 7, 7, 7, 7, 7,
-                                                        7, 7, 7, 6, 6 };
-static const int8_t inv_stage_range_row_adst_16[10] = { 7, 7, 7, 7, 7,
-                                                        7, 7, 7, 7, 7 };
-static const int8_t inv_stage_range_idx_16[1] = { 0 };
+// Values in both inv_cos_bit_col and inv_cos_bit_row are always 12
+// for each valid row and col combination
+#define INV_COS_BIT 12
+extern const int8_t inv_cos_bit_col[5 /*row*/][5 /*col*/];
+extern const int8_t inv_cos_bit_row[5 /*row*/][5 /*col*/];
 
-// cos bit
-static const int8_t inv_cos_bit_col_dct_16[8] = {
-  13, 13, 13, 13, 13, 13, 13, 13
-};
-static const int8_t inv_cos_bit_row_dct_16[8] = {
-  12, 12, 12, 12, 12, 12, 12, 12
-};
-static const int8_t inv_cos_bit_col_adst_16[10] = { 13, 13, 13, 13, 13,
-                                                    13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_adst_16[10] = { 12, 12, 12, 12, 12,
-                                                    12, 12, 12, 12, 12 };
-
-//  ---------------- 32x32 1D constants -----------------------
-// shift
-static const int8_t inv_shift_32[2] = { -1, -5 };
-
-// stage range
-static const int8_t inv_stage_range_col_dct_32[10] = { 9, 9, 9, 9, 9,
-                                                       9, 9, 9, 8, 8 };
-static const int8_t inv_stage_range_row_dct_32[10] = { 9, 9, 9, 9, 9,
-                                                       9, 9, 9, 9, 9 };
-static const int8_t inv_stage_range_col_adst_32[12] = { 9, 9, 9, 9, 9, 9,
-                                                        9, 9, 9, 9, 8, 8 };
-static const int8_t inv_stage_range_row_adst_32[12] = { 9, 9, 9, 9, 9, 9,
-                                                        9, 9, 9, 9, 9, 9 };
-static const int8_t inv_stage_range_idx_32[1] = { 0 };
-
-// cos bit
-static const int8_t inv_cos_bit_col_dct_32[10] = { 13, 13, 13, 13, 13,
-                                                   13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_dct_32[10] = { 12, 12, 12, 12, 12,
-                                                   12, 12, 12, 12, 12 };
-static const int8_t inv_cos_bit_col_adst_32[12] = { 13, 13, 13, 13, 13, 13,
-                                                    13, 13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_adst_32[12] = { 12, 12, 12, 12, 12, 12,
-                                                    12, 12, 12, 12, 12, 12 };
-
-//  ---------------- 64x64 1D constants -----------------------
-// shift
-static const int8_t inv_shift_64[2] = { -1, -5 };
-
-// stage range
-static const int8_t inv_stage_range_col_dct_64[12] = { 11, 11, 11, 11, 11, 11,
-                                                       11, 11, 11, 11, 10, 10 };
-static const int8_t inv_stage_range_row_dct_64[12] = { 11, 11, 11, 11, 11, 11,
-                                                       11, 11, 11, 11, 11, 11 };
-
-static const int8_t inv_stage_range_idx_64[1] = { 0 };
-
-// cos bit
-static const int8_t inv_cos_bit_col_dct_64[12] = { 13, 13, 13, 13, 13, 13,
-                                                   13, 13, 13, 13, 13, 13 };
-static const int8_t inv_cos_bit_row_dct_64[12] = { 12, 12, 12, 12, 12, 12,
-                                                   12, 12, 12, 12, 12, 12 };
-
-//  ---------------- row config inv_dct_4 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_4 = {
-  4,                          // .txfm_size
-  4,                          // .stage_num
-  inv_shift_4,                // .shift
-  inv_stage_range_row_dct_4,  // .stage_range
-  inv_cos_bit_row_dct_4,      // .cos_bit
-  TXFM_TYPE_DCT4              // .txfm_type
-};
-
-//  ---------------- row config inv_dct_8 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_8 = {
-  8,                          // .txfm_size
-  6,                          // .stage_num
-  inv_shift_8,                // .shift
-  inv_stage_range_row_dct_8,  // .stage_range
-  inv_cos_bit_row_dct_8,      // .cos_bit_
-  TXFM_TYPE_DCT8              // .txfm_type
-};
-//  ---------------- row config inv_dct_16 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_16 = {
-  16,                          // .txfm_size
-  8,                           // .stage_num
-  inv_shift_16,                // .shift
-  inv_stage_range_row_dct_16,  // .stage_range
-  inv_cos_bit_row_dct_16,      // .cos_bit
-  TXFM_TYPE_DCT16              // .txfm_type
-};
-
-//  ---------------- row config inv_dct_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_32 = {
-  32,                          // .txfm_size
-  10,                          // .stage_num
-  inv_shift_32,                // .shift
-  inv_stage_range_row_dct_32,  // .stage_range
-  inv_cos_bit_row_dct_32,      // .cos_bit_row
-  TXFM_TYPE_DCT32              // .txfm_type
-};
-
-#if CONFIG_TX64X64
-//  ---------------- row config inv_dct_64 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_dct_64 = {
-  64,                          // .txfm_size
-  12,                          // .stage_num
-  inv_shift_64,                // .shift
-  inv_stage_range_row_dct_64,  // .stage_range
-  inv_cos_bit_row_dct_64,      // .cos_bit
-  TXFM_TYPE_DCT64,             // .txfm_type_col
-};
-#endif  // CONFIG_TX64X64
-
-//  ---------------- row config inv_adst_4 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_4 = {
-  4,                           // .txfm_size
-  6,                           // .stage_num
-  inv_shift_4,                 // .shift
-  inv_stage_range_row_adst_4,  // .stage_range
-  inv_cos_bit_row_adst_4,      // .cos_bit
-  TXFM_TYPE_ADST4,             // .txfm_type
-};
-
-//  ---------------- row config inv_adst_8 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_8 = {
-  8,                           // .txfm_size
-  8,                           // .stage_num
-  inv_shift_8,                 // .shift
-  inv_stage_range_row_adst_8,  // .stage_range
-  inv_cos_bit_row_adst_8,      // .cos_bit
-  TXFM_TYPE_ADST8,             // .txfm_type_col
-};
-
-//  ---------------- row config inv_adst_16 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_16 = {
-  16,                           // .txfm_size
-  10,                           // .stage_num
-  inv_shift_16,                 // .shift
-  inv_stage_range_row_adst_16,  // .stage_range
-  inv_cos_bit_row_adst_16,      // .cos_bit
-  TXFM_TYPE_ADST16,             // .txfm_type
-};
-
-//  ---------------- row config inv_adst_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_32 = {
-  32,                           // .txfm_size
-  12,                           // .stage_num
-  inv_shift_32,                 // .shift
-  inv_stage_range_row_adst_32,  // .stage_range
-  inv_cos_bit_row_adst_32,      // .cos_bit
-  TXFM_TYPE_ADST32,             // .txfm_type
-};
-
-//  ---------------- col config inv_dct_4 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_4 = {
-  4,                          // .txfm_size
-  4,                          // .stage_num
-  inv_shift_4,                // .shift
-  inv_stage_range_col_dct_4,  // .stage_range
-  inv_cos_bit_col_dct_4,      // .cos_bit
-  TXFM_TYPE_DCT4              // .txfm_type
-};
-
-//  ---------------- col config inv_dct_8 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_8 = {
-  8,                          // .txfm_size
-  6,                          // .stage_num
-  inv_shift_8,                // .shift
-  inv_stage_range_col_dct_8,  // .stage_range
-  inv_cos_bit_col_dct_8,      // .cos_bit_
-  TXFM_TYPE_DCT8              // .txfm_type
-};
-//  ---------------- col config inv_dct_16 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_16 = {
-  16,                          // .txfm_size
-  8,                           // .stage_num
-  inv_shift_16,                // .shift
-  inv_stage_range_col_dct_16,  // .stage_range
-  inv_cos_bit_col_dct_16,      // .cos_bit
-  TXFM_TYPE_DCT16              // .txfm_type
-};
-
-//  ---------------- col config inv_dct_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_32 = {
-  32,                          // .txfm_size
-  10,                          // .stage_num
-  inv_shift_32,                // .shift
-  inv_stage_range_col_dct_32,  // .stage_range
-  inv_cos_bit_col_dct_32,      // .cos_bit_col
-  TXFM_TYPE_DCT32              // .txfm_type
-};
-
-//  ---------------- col config inv_dct_64 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_dct_64 = {
-  64,                          // .txfm_size
-  12,                          // .stage_num
-  inv_shift_64,                // .shift
-  inv_stage_range_col_dct_64,  // .stage_range
-  inv_cos_bit_col_dct_64,      // .cos_bit
-  TXFM_TYPE_DCT64,             // .txfm_type_col
-};
-
-//  ---------------- col config inv_adst_4 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_4 = {
-  4,                           // .txfm_size
-  6,                           // .stage_num
-  inv_shift_4,                 // .shift
-  inv_stage_range_col_adst_4,  // .stage_range
-  inv_cos_bit_col_adst_4,      // .cos_bit
-  TXFM_TYPE_ADST4,             // .txfm_type
-};
-
-//  ---------------- col config inv_adst_8 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_8 = {
-  8,                           // .txfm_size
-  8,                           // .stage_num
-  inv_shift_8,                 // .shift
-  inv_stage_range_col_adst_8,  // .stage_range
-  inv_cos_bit_col_adst_8,      // .cos_bit
-  TXFM_TYPE_ADST8,             // .txfm_type_col
-};
-
-//  ---------------- col config inv_adst_16 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_16 = {
-  16,                           // .txfm_size
-  10,                           // .stage_num
-  inv_shift_16,                 // .shift
-  inv_stage_range_col_adst_16,  // .stage_range
-  inv_cos_bit_col_adst_16,      // .cos_bit
-  TXFM_TYPE_ADST16,             // .txfm_type
-};
-
-//  ---------------- col config inv_adst_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_col_cfg_adst_32 = {
-  32,                           // .txfm_size
-  12,                           // .stage_num
-  inv_shift_32,                 // .shift
-  inv_stage_range_col_adst_32,  // .stage_range
-  inv_cos_bit_col_adst_32,      // .cos_bit
-  TXFM_TYPE_ADST32,             // .txfm_type
-};
-
-#if CONFIG_EXT_TX
-// identity does not need to differentiate between row and col
-//  ---------------- row/col config inv_identity_4 ----------
-static const TXFM_1D_CFG inv_txfm_1d_cfg_identity_4 = {
-  4,                      // .txfm_size
-  1,                      // .stage_num
-  inv_shift_4,            // .shift
-  inv_stage_range_idx_4,  // .stage_range
-  NULL,                   // .cos_bit
-  TXFM_TYPE_IDENTITY4,    // .txfm_type
-};
-
-//  ---------------- row/col config inv_identity_8 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_cfg_identity_8 = {
-  8,                      // .txfm_size
-  1,                      // .stage_num
-  inv_shift_8,            // .shift
-  inv_stage_range_idx_8,  // .stage_range
-  NULL,                   // .cos_bit
-  TXFM_TYPE_IDENTITY8,    // .txfm_type
-};
-
-//  ---------------- row/col config inv_identity_16 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_cfg_identity_16 = {
-  16,                      // .txfm_size
-  1,                       // .stage_num
-  inv_shift_16,            // .shift
-  inv_stage_range_idx_16,  // .stage_range
-  NULL,                    // .cos_bit
-  TXFM_TYPE_IDENTITY16,    // .txfm_type
-};
-
-//  ---------------- row/col config inv_identity_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_cfg_identity_32 = {
-  32,                      // .txfm_size
-  1,                       // .stage_num
-  inv_shift_32,            // .shift
-  inv_stage_range_idx_32,  // .stage_range
-  NULL,                    // .cos_bit
-  TXFM_TYPE_IDENTITY32,    // .txfm_type
-};
-
-#if CONFIG_TX64X64
-//  ---------------- row/col config inv_identity_32 ----------------
-static const TXFM_1D_CFG inv_txfm_1d_cfg_identity_64 = {
-  64,                      // .txfm_size
-  1,                       // .stage_num
-  inv_shift_64,            // .shift
-  inv_stage_range_idx_64,  // .stage_range
-  NULL,                    // .cos_bit
-  TXFM_TYPE_IDENTITY64,    // .txfm_type
-};
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_EXT_TX
 #endif  // AV1_INV_TXFM2D_CFG_H_

third_party/aom/av1/common/av1_inv_txfm2d.c

@@ -9,218 +9,252 @@
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */
 
-#include "./av1_rtcd.h"
-#include "aom_dsp/inv_txfm.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
 #include "av1/common/enums.h"
 #include "av1/common/av1_txfm.h"
 #include "av1/common/av1_inv_txfm1d.h"
 #include "av1/common/av1_inv_txfm1d_cfg.h"
 
+void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
+                                 int stride, int bd) {
+  /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
+     0.5 shifts per pixel. */
+  int i;
+  tran_low_t output[16];
+  tran_low_t a1, b1, c1, d1, e1;
+  const tran_low_t *ip = input;
+  tran_low_t *op = output;
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+
+  for (i = 0; i < 4; i++) {
+    a1 = ip[0] >> UNIT_QUANT_SHIFT;
+    c1 = ip[1] >> UNIT_QUANT_SHIFT;
+    d1 = ip[2] >> UNIT_QUANT_SHIFT;
+    b1 = ip[3] >> UNIT_QUANT_SHIFT;
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+
+    op[0] = a1;
+    op[1] = b1;
+    op[2] = c1;
+    op[3] = d1;
+    ip += 4;
+    op += 4;
+  }
+
+  ip = output;
+  for (i = 0; i < 4; i++) {
+    a1 = ip[4 * 0];
+    c1 = ip[4 * 1];
+    d1 = ip[4 * 2];
+    b1 = ip[4 * 3];
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
+
+    range_check_value(a1, bd + 1);
+    range_check_value(b1, bd + 1);
+    range_check_value(c1, bd + 1);
+    range_check_value(d1, bd + 1);
+
+    dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
+    dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
+    dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
+    dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
+
+    ip++;
+    dest++;
+  }
+}
+
+void av1_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
+                                int dest_stride, int bd) {
+  int i;
+  tran_low_t a1, e1;
+  tran_low_t tmp[4];
+  const tran_low_t *ip = in;
+  tran_low_t *op = tmp;
+  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
+  (void)bd;
+
+  a1 = ip[0] >> UNIT_QUANT_SHIFT;
+  e1 = a1 >> 1;
+  a1 -= e1;
+  op[0] = a1;
+  op[1] = op[2] = op[3] = e1;
+
+  ip = tmp;
+  for (i = 0; i < 4; i++) {
+    e1 = ip[0] >> 1;
+    a1 = ip[0] - e1;
+    dest[dest_stride * 0] =
+        highbd_clip_pixel_add(dest[dest_stride * 0], a1, bd);
+    dest[dest_stride * 1] =
+        highbd_clip_pixel_add(dest[dest_stride * 1], e1, bd);
+    dest[dest_stride * 2] =
+        highbd_clip_pixel_add(dest[dest_stride * 2], e1, bd);
+    dest[dest_stride * 3] =
+        highbd_clip_pixel_add(dest[dest_stride * 3], e1, bd);
+    ip++;
+    dest++;
+  }
+}
+
 static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
   switch (txfm_type) {
     case TXFM_TYPE_DCT4: return av1_idct4_new;
     case TXFM_TYPE_DCT8: return av1_idct8_new;
     case TXFM_TYPE_DCT16: return av1_idct16_new;
     case TXFM_TYPE_DCT32: return av1_idct32_new;
-#if CONFIG_TX64X64
     case TXFM_TYPE_DCT64: return av1_idct64_new;
-#endif  // CONFIG_TX64X64
     case TXFM_TYPE_ADST4: return av1_iadst4_new;
     case TXFM_TYPE_ADST8: return av1_iadst8_new;
     case TXFM_TYPE_ADST16: return av1_iadst16_new;
-    case TXFM_TYPE_ADST32: return av1_iadst32_new;
-#if CONFIG_EXT_TX
     case TXFM_TYPE_IDENTITY4: return av1_iidentity4_c;
     case TXFM_TYPE_IDENTITY8: return av1_iidentity8_c;
     case TXFM_TYPE_IDENTITY16: return av1_iidentity16_c;
     case TXFM_TYPE_IDENTITY32: return av1_iidentity32_c;
-#if CONFIG_TX64X64
-    case TXFM_TYPE_IDENTITY64: return av1_iidentity64_c;
-#endif  // CONFIG_TX64X64
-#endif  // CONFIG_EXT_TX
     default: assert(0); return NULL;
   }
 }
 
-static const TXFM_1D_CFG *inv_txfm_col_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
-  // DCT
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_col_cfg_dct_4, &inv_txfm_1d_col_cfg_dct_8,
-      &inv_txfm_1d_col_cfg_dct_16, &inv_txfm_1d_col_cfg_dct_32,
-#if CONFIG_TX64X64
-      &inv_txfm_1d_col_cfg_dct_64
-#endif  // CONFIG_TX64X64
-  },
-  // ADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_col_cfg_adst_4, &inv_txfm_1d_col_cfg_adst_8,
-      &inv_txfm_1d_col_cfg_adst_16, &inv_txfm_1d_col_cfg_adst_32,
-#if CONFIG_TX64X64
-      NULL
-#endif  // CONFIG_TX64X64
-  },
-#if CONFIG_EXT_TX
-  // FLIPADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_col_cfg_adst_4, &inv_txfm_1d_col_cfg_adst_8,
-      &inv_txfm_1d_col_cfg_adst_16, &inv_txfm_1d_col_cfg_adst_32,
-#if CONFIG_TX64X64
-      NULL
-#endif  // CONFIG_TX64X64
-  },
-  // IDENTITY
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_cfg_identity_4, &inv_txfm_1d_cfg_identity_8,
-      &inv_txfm_1d_cfg_identity_16, &inv_txfm_1d_cfg_identity_32,
-#if CONFIG_TX64X64
-      &inv_txfm_1d_cfg_identity_64
-#endif  // CONFIG_TX64X64
-  },
-#endif  // CONFIG_EXT_TX
+static const int8_t inv_shift_4x4[2] = { 0, -4 };
+static const int8_t inv_shift_8x8[2] = { -1, -4 };
+static const int8_t inv_shift_16x16[2] = { -2, -4 };
+static const int8_t inv_shift_32x32[2] = { -2, -4 };
+static const int8_t inv_shift_64x64[2] = { -2, -4 };
+static const int8_t inv_shift_4x8[2] = { 0, -4 };
+static const int8_t inv_shift_8x4[2] = { 0, -4 };
+static const int8_t inv_shift_8x16[2] = { -1, -4 };
+static const int8_t inv_shift_16x8[2] = { -1, -4 };
+static const int8_t inv_shift_16x32[2] = { -1, -4 };
+static const int8_t inv_shift_32x16[2] = { -1, -4 };
+static const int8_t inv_shift_32x64[2] = { -1, -4 };
+static const int8_t inv_shift_64x32[2] = { -1, -4 };
+static const int8_t inv_shift_4x16[2] = { -1, -4 };
+static const int8_t inv_shift_16x4[2] = { -1, -4 };
+static const int8_t inv_shift_8x32[2] = { -2, -4 };
+static const int8_t inv_shift_32x8[2] = { -2, -4 };
+static const int8_t inv_shift_16x64[2] = { -2, -4 };
+static const int8_t inv_shift_64x16[2] = { -2, -4 };
+
+const int8_t *inv_txfm_shift_ls[TX_SIZES_ALL] = {
+  inv_shift_4x4,   inv_shift_8x8,   inv_shift_16x16, inv_shift_32x32,
+  inv_shift_64x64, inv_shift_4x8,   inv_shift_8x4,   inv_shift_8x16,
+  inv_shift_16x8,  inv_shift_16x32, inv_shift_32x16, inv_shift_32x64,
+  inv_shift_64x32, inv_shift_4x16,  inv_shift_16x4,  inv_shift_8x32,
+  inv_shift_32x8,  inv_shift_16x64, inv_shift_64x16,
 };
 
-static const TXFM_1D_CFG *inv_txfm_row_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
-  // DCT
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_row_cfg_dct_4, &inv_txfm_1d_row_cfg_dct_8,
-      &inv_txfm_1d_row_cfg_dct_16, &inv_txfm_1d_row_cfg_dct_32,
-#if CONFIG_TX64X64
-      &inv_txfm_1d_row_cfg_dct_64,
-#endif  // CONFIG_TX64X64
-  },
-  // ADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_row_cfg_adst_4, &inv_txfm_1d_row_cfg_adst_8,
-      &inv_txfm_1d_row_cfg_adst_16, &inv_txfm_1d_row_cfg_adst_32,
-#if CONFIG_TX64X64
-      NULL
-#endif  // CONFIG_TX64X64
-  },
-#if CONFIG_EXT_TX
-  // FLIPADST
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_row_cfg_adst_4, &inv_txfm_1d_row_cfg_adst_8,
-      &inv_txfm_1d_row_cfg_adst_16, &inv_txfm_1d_row_cfg_adst_32,
-#if CONFIG_TX64X64
-      NULL
-#endif  // CONFIG_TX64X64
-  },
-  // IDENTITY
-  {
-#if CONFIG_CHROMA_2X2
-      NULL,
-#endif
-      &inv_txfm_1d_cfg_identity_4, &inv_txfm_1d_cfg_identity_8,
-      &inv_txfm_1d_cfg_identity_16, &inv_txfm_1d_cfg_identity_32,
-#if CONFIG_TX64X64
-      &inv_txfm_1d_cfg_identity_64
-#endif  // CONFIG_TX64X64
-  },
-#endif  // CONFIG_EXT_TX
-};
+/* clang-format off */
+const int8_t inv_cos_bit_col[MAX_TXWH_IDX]      // txw_idx
+                            [MAX_TXWH_IDX] = {  // txh_idx
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT,           0,           0 },
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT,           0 },
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
+    {           0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
+    {           0,           0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT }
+  };
 
-TXFM_2D_FLIP_CFG av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size) {
-  TXFM_2D_FLIP_CFG cfg;
-  set_flip_cfg(tx_type, &cfg);
-  const TX_TYPE_1D tx_type_col = vtx_tab[tx_type];
-  const TX_TYPE_1D tx_type_row = htx_tab[tx_type];
-  const TX_SIZE tx_size_col = txsize_vert_map[tx_size];
-  const TX_SIZE tx_size_row = txsize_horz_map[tx_size];
-  cfg.col_cfg = inv_txfm_col_cfg_ls[tx_type_col][tx_size_col];
-  cfg.row_cfg = inv_txfm_row_cfg_ls[tx_type_row][tx_size_row];
-  return cfg;
-}
+const int8_t inv_cos_bit_row[MAX_TXWH_IDX]      // txw_idx
+                            [MAX_TXWH_IDX] = {  // txh_idx
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT,           0,           0 },
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT,           0 },
+    { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
+    {           0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
+    {           0,           0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT }
+  };
+/* clang-format on */
 
-#if CONFIG_TX64X64
-TXFM_2D_FLIP_CFG av1_get_inv_txfm_64x64_cfg(TX_TYPE tx_type) {
-  TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL, NULL };
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = &inv_txfm_1d_col_cfg_dct_64;
-      cfg.row_cfg = &inv_txfm_1d_row_cfg_dct_64;
-      set_flip_cfg(tx_type, &cfg);
-      break;
-    default: assert(0);
+const int8_t iadst4_range[7] = { 0, 1, 0, 0, 0, 0, 0 };
+
+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);
+  const TX_TYPE_1D tx_type_1d_col = vtx_tab[tx_type];
+  const TX_TYPE_1D tx_type_1d_row = htx_tab[tx_type];
+  cfg->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);
+  cfg->cos_bit_col = inv_cos_bit_col[txw_idx][txh_idx];
+  cfg->cos_bit_row = inv_cos_bit_row[txw_idx][txh_idx];
+  cfg->txfm_type_col = av1_txfm_type_ls[txh_idx][tx_type_1d_col];
+  if (cfg->txfm_type_col == TXFM_TYPE_ADST4) {
+    memcpy(cfg->stage_range_col, iadst4_range, sizeof(iadst4_range));
   }
-  return cfg;
-}
-
-TXFM_2D_FLIP_CFG av1_get_inv_txfm_32x64_cfg(int tx_type) {
-  TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL, NULL };
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = &inv_txfm_1d_col_cfg_dct_64;
-      cfg.row_cfg = &inv_txfm_1d_row_cfg_dct_32;
-      set_flip_cfg(tx_type, &cfg);
-      break;
-    default: assert(0);
+  cfg->txfm_type_row = av1_txfm_type_ls[txw_idx][tx_type_1d_row];
+  if (cfg->txfm_type_row == TXFM_TYPE_ADST4) {
+    memcpy(cfg->stage_range_row, iadst4_range, sizeof(iadst4_range));
   }
-  return cfg;
+  cfg->stage_num_col = av1_txfm_stage_num_list[cfg->txfm_type_col];
+  cfg->stage_num_row = av1_txfm_stage_num_list[cfg->txfm_type_row];
 }
 
-TXFM_2D_FLIP_CFG av1_get_inv_txfm_64x32_cfg(int tx_type) {
-  TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL, NULL };
-  switch (tx_type) {
-    case DCT_DCT:
-      cfg.col_cfg = &inv_txfm_1d_col_cfg_dct_32;
-      cfg.row_cfg = &inv_txfm_1d_row_cfg_dct_64;
-      set_flip_cfg(tx_type, &cfg);
-      break;
-    default: assert(0);
-  }
-  return cfg;
-}
-#endif  // CONFIG_TX64X64
-
 void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
-                             const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift,
+                             const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
                              int bd) {
-  // Note when assigning txfm_size_col, we use the txfm_size from the
-  // row configuration and vice versa. This is intentionally done to
-  // accurately perform rectangular transforms. When the transform is
-  // rectangular, the number of columns will be the same as the
-  // txfm_size stored in the row cfg struct. It will make no difference
-  // for square transforms.
-  const int txfm_size_col = cfg->row_cfg->txfm_size;
-  const int txfm_size_row = cfg->col_cfg->txfm_size;
-  // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
-  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->row_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_row[i] = cfg->row_cfg->stage_range[i] + fwd_shift + bd + 1;
+  const int fwd_shift = inv_start_range[tx_size];
+  const int8_t *shift = cfg->shift;
+  int8_t opt_range_row, opt_range_col;
+  if (bd == 8) {
+    opt_range_row = 16;
+    opt_range_col = 16;
+  } else if (bd == 10) {
+    opt_range_row = 18;
+    opt_range_col = 16;
+  } else {
+    assert(bd == 12);
+    opt_range_row = 20;
+    opt_range_col = 18;
   }
   // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
-  for (int i = 0; i < cfg->col_cfg->stage_num && i < MAX_TXFM_STAGE_NUM; ++i) {
-    stage_range_col[i] =
-        cfg->col_cfg->stage_range[i] + fwd_shift + shift[0] + bd + 1;
+  for (int i = 0; i < cfg->stage_num_row && i < MAX_TXFM_STAGE_NUM; ++i) {
+    int real_range_row = cfg->stage_range_row[i] + fwd_shift + bd + 1;
+    (void)real_range_row;
+    if (cfg->txfm_type_row == TXFM_TYPE_ADST4 && i == 1) {
+      // the adst4 may use 1 extra bit on top of opt_range_row at stage 1
+      // so opt_range_col >= real_range_col will not hold
+      stage_range_row[i] = opt_range_row;
+    } else {
+      assert(opt_range_row >= real_range_row);
+      stage_range_row[i] = opt_range_row;
+    }
+  }
+  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
+  for (int i = 0; i < cfg->stage_num_col && i < MAX_TXFM_STAGE_NUM; ++i) {
+    int real_range_col =
+        cfg->stage_range_col[i] + fwd_shift + shift[0] + bd + 1;
+    (void)real_range_col;
+    if (cfg->txfm_type_col == TXFM_TYPE_ADST4 && i == 1) {
+      // the adst4 may use 1 extra bit on top of opt_range_row at stage 1
+      // so opt_range_col >= real_range_col will not hold
+      stage_range_col[i] = opt_range_col;
+    } else {
+      assert(opt_range_col >= real_range_col);
+      stage_range_col[i] = opt_range_col;
+    }
   }
 }
 
 static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
                                     int stride, TXFM_2D_FLIP_CFG *cfg,
-                                    int32_t *txfm_buf, int8_t fwd_shift,
+                                    int32_t *txfm_buf, TX_SIZE tx_size,
                                     int bd) {
   // Note when assigning txfm_size_col, we use the txfm_size from the
   // row configuration and vice versa. This is intentionally done to
@@ -228,39 +262,48 @@ static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
   // rectangular, the number of columns will be the same as the
   // txfm_size stored in the row cfg struct. It will make no difference
   // for square transforms.
-  const int txfm_size_col = cfg->row_cfg->txfm_size;
-  const int txfm_size_row = cfg->col_cfg->txfm_size;
+  const int txfm_size_col = tx_size_wide[cfg->tx_size];
+  const int txfm_size_row = tx_size_high[cfg->tx_size];
   // Take the shift from the larger dimension in the rectangular case.
-  const int8_t *shift = (txfm_size_col > txfm_size_row) ? cfg->row_cfg->shift
-                                                        : cfg->col_cfg->shift;
+  const int8_t *shift = cfg->shift;
+  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
   int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
   int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
-  assert(cfg->row_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
-  assert(cfg->col_cfg->stage_num <= MAX_TXFM_STAGE_NUM);
-  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, fwd_shift, bd);
+  assert(cfg->stage_num_row <= MAX_TXFM_STAGE_NUM);
+  assert(cfg->stage_num_col <= MAX_TXFM_STAGE_NUM);
+  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, tx_size, bd);
 
-  const int8_t *cos_bit_col = cfg->col_cfg->cos_bit;
-  const int8_t *cos_bit_row = cfg->row_cfg->cos_bit;
-  const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->col_cfg->txfm_type);
-  const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->row_cfg->txfm_type);
+  const int8_t cos_bit_col = cfg->cos_bit_col;
+  const int8_t cos_bit_row = cfg->cos_bit_row;
+  const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->txfm_type_col);
+  const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->txfm_type_row);
 
-  // txfm_buf's length is  txfm_size_row * txfm_size_col + 2 * txfm_size_row
+  // txfm_buf's length is  txfm_size_row * txfm_size_col + 2 *
+  // AOMMAX(txfm_size_row, txfm_size_col)
   // it is used for intermediate data buffering
+  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
   int32_t *temp_in = txfm_buf;
-  int32_t *temp_out = temp_in + txfm_size_row;
-  int32_t *buf = temp_out + txfm_size_row;
+  int32_t *temp_out = temp_in + buf_offset;
+  int32_t *buf = temp_out + buf_offset;
   int32_t *buf_ptr = buf;
   int c, r;
 
   // Rows
   for (r = 0; r < txfm_size_row; ++r) {
-    txfm_func_row(input, buf_ptr, cos_bit_row, stage_range_row);
-    round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
-    // Multiply everything by Sqrt2 if the transform is rectangular
-    if (txfm_size_row != txfm_size_col) {
-      for (c = 0; c < txfm_size_col; ++c)
-        buf_ptr[c] = (int32_t)dct_const_round_shift(buf_ptr[c] * Sqrt2);
+    if (abs(rect_type) == 1) {
+      for (c = 0; c < txfm_size_col; ++c) {
+        temp_in[c] = round_shift((int64_t)input[c] * NewInvSqrt2, NewSqrt2Bits);
+      }
+      clamp_buf(temp_in, txfm_size_col, bd + 8);
+      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
+    } else {
+      for (c = 0; c < txfm_size_col; ++c) {
+        temp_in[c] = input[c];
+      }
+      clamp_buf(temp_in, txfm_size_col, bd + 8);
+      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
     }
+    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
     input += txfm_size_col;
     buf_ptr += txfm_size_col;
   }
@@ -275,8 +318,9 @@ static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
       for (r = 0; r < txfm_size_row; ++r)
         temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
     }
+    clamp_buf(temp_in, txfm_size_row, AOMMAX(bd + 6, 16));
     txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
-    round_shift_array(temp_out, txfm_size_row, -shift[1]);
+    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
     if (cfg->ud_flip == 0) {
       for (r = 0; r < txfm_size_row; ++r) {
         output[r * stride + c] =
@@ -296,156 +340,166 @@ static INLINE void inv_txfm2d_add_facade(const int32_t *input, uint16_t *output,
                                          int stride, int32_t *txfm_buf,
                                          TX_TYPE tx_type, TX_SIZE tx_size,
                                          int bd) {
-  TXFM_2D_FLIP_CFG cfg = av1_get_inv_txfm_cfg(tx_type, tx_size);
-  TX_SIZE tx_size_sqr = txsize_sqr_map[tx_size];
-  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf,
-                   fwd_shift_sum[tx_size_sqr], bd);
+  TXFM_2D_FLIP_CFG cfg;
+  av1_get_inv_txfm_cfg(tx_type, tx_size, &cfg);
+  // Forward shift sum uses larger square size, to be consistent with what
+  // av1_gen_inv_stage_range() does for inverse shifts.
+  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf, tx_size, bd);
 }
 
 void av1_inv_txfm2d_add_4x8_c(const int32_t *input, uint16_t *output,
                               int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[4 * 8 + 8 + 8];
+  DECLARE_ALIGNED(32, int, txfm_buf[4 * 8 + 8 + 8]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X8, bd);
 }
 
 void av1_inv_txfm2d_add_8x4_c(const int32_t *input, uint16_t *output,
                               int stride, TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int txfm_buf[8 * 4 + 8 + 8];
-  int32_t rinput[8 * 4];
-  uint16_t routput[8 * 4];
-  TX_SIZE tx_size = TX_8X4;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int32(rinput, rw, input, w, w, h);
-  transpose_uint16(routput, rw, output, stride, w, h);
-  inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
-  transpose_uint16(output, stride, routput, rw, rw, rh);
-#else
-  int txfm_buf[8 * 4 + 4 + 4];
+  DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X4, bd);
-#endif
 }
 
 void av1_inv_txfm2d_add_8x16_c(const int32_t *input, uint16_t *output,
                                int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[8 * 16 + 16 + 16];
+  DECLARE_ALIGNED(32, int, txfm_buf[8 * 16 + 16 + 16]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X16, bd);
 }
 
 void av1_inv_txfm2d_add_16x8_c(const int32_t *input, uint16_t *output,
                                int stride, TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int txfm_buf[16 * 8 + 16 + 16];
-  int32_t rinput[16 * 8];
-  uint16_t routput[16 * 8];
-  TX_SIZE tx_size = TX_16X8;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int32(rinput, rw, input, w, w, h);
-  transpose_uint16(routput, rw, output, stride, w, h);
-  inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
-  transpose_uint16(output, stride, routput, rw, rw, rh);
-#else
-  int txfm_buf[16 * 8 + 8 + 8];
+  DECLARE_ALIGNED(32, int, txfm_buf[16 * 8 + 16 + 16]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X8, bd);
-#endif
 }
 
 void av1_inv_txfm2d_add_16x32_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[16 * 32 + 32 + 32];
+  DECLARE_ALIGNED(32, int, txfm_buf[16 * 32 + 32 + 32]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X32, bd);
 }
 
 void av1_inv_txfm2d_add_32x16_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int txfm_buf[32 * 16 + 32 + 32];
-  int32_t rinput[32 * 16];
-  uint16_t routput[32 * 16];
-  TX_SIZE tx_size = TX_32X16;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int32(rinput, rw, input, w, w, h);
-  transpose_uint16(routput, rw, output, stride, w, h);
-  inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
-  transpose_uint16(output, stride, routput, rw, rw, rh);
-#else
-  int txfm_buf[32 * 16 + 16 + 16];
+  DECLARE_ALIGNED(32, int, txfm_buf[32 * 16 + 32 + 32]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X16, bd);
-#endif
 }
 
 void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
                               int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[4 * 4 + 4 + 4];
+  DECLARE_ALIGNED(32, int, txfm_buf[4 * 4 + 4 + 4]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X4, bd);
 }
 
 void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output,
                               int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[8 * 8 + 8 + 8];
+  DECLARE_ALIGNED(32, int, txfm_buf[8 * 8 + 8 + 8]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X8, bd);
 }
 
 void av1_inv_txfm2d_add_16x16_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[16 * 16 + 16 + 16];
+  DECLARE_ALIGNED(32, int, txfm_buf[16 * 16 + 16 + 16]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X16, bd);
 }
 
 void av1_inv_txfm2d_add_32x32_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[32 * 32 + 32 + 32];
+  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
   inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X32, bd);
 }
 
-#if CONFIG_TX64X64
 void av1_inv_txfm2d_add_64x64_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[64 * 64 + 64 + 64];
-  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_64X64, bd);
+  // TODO(urvang): Can the same array be reused, instead of using a new array?
+  // Remap 32x32 input into a modified 64x64 by:
+  // - Copying over these values in top-left 32x32 locations.
+  // - Setting the rest of the locations to 0.
+  int32_t mod_input[64 * 64];
+  for (int 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));
+  }
+  memset(mod_input + 32 * 64, 0, 32 * 64 * sizeof(*mod_input));
+  DECLARE_ALIGNED(32, int, txfm_buf[64 * 64 + 64 + 64]);
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X64,
+                        bd);
 }
 
 void av1_inv_txfm2d_add_64x32_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-#if CONFIG_TXMG
-  int txfm_buf[64 * 32 + 64 + 64];
-  int32_t rinput[64 * 32];
-  uint16_t routput[64 * 32];
-  TX_SIZE tx_size = TX_64X32;
-  TX_SIZE rtx_size = av1_rotate_tx_size(tx_size);
-  TX_TYPE rtx_type = av1_rotate_tx_type(tx_type);
-  int w = tx_size_wide[tx_size];
-  int h = tx_size_high[tx_size];
-  int rw = h;
-  int rh = w;
-  transpose_int32(rinput, rw, input, w, w, h);
-  transpose_uint16(routput, rw, output, stride, w, h);
-  inv_txfm2d_add_facade(rinput, routput, rw, txfm_buf, rtx_type, rtx_size, bd);
-  transpose_uint16(output, stride, routput, rw, rw, rh);
-#else
-  int txfm_buf[64 * 32 + 64 + 64];
-  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_64X32, bd);
-#endif
+  // Remap 32x32 input into a modified 64x32 by:
+  // - Copying over these values in top-left 32x32 locations.
+  // - Setting the rest of the locations to 0.
+  int32_t mod_input[64 * 32];
+  for (int 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));
+  }
+  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X32,
+                        bd);
 }
 
 void av1_inv_txfm2d_add_32x64_c(const int32_t *input, uint16_t *output,
                                 int stride, TX_TYPE tx_type, int bd) {
-  int txfm_buf[64 * 32 + 64 + 64];
-  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X64, bd);
+  // Remap 32x32 input into a modified 32x64 input by:
+  // - Copying over these values in top-left 32x32 locations.
+  // - Setting the rest of the locations to 0.
+  int32_t mod_input[32 * 64];
+  memcpy(mod_input, input, 32 * 32 * sizeof(*mod_input));
+  memset(mod_input + 32 * 32, 0, 32 * 32 * sizeof(*mod_input));
+  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_32X64,
+                        bd);
+}
+
+void av1_inv_txfm2d_add_16x64_c(const int32_t *input, uint16_t *output,
+                                int stride, TX_TYPE tx_type, int bd) {
+  // Remap 16x32 input into a modified 16x64 input by:
+  // - Copying over these values in top-left 16x32 locations.
+  // - Setting the rest of the locations to 0.
+  int32_t mod_input[16 * 64];
+  memcpy(mod_input, input, 16 * 32 * sizeof(*mod_input));
+  memset(mod_input + 16 * 32, 0, 16 * 32 * sizeof(*mod_input));
+  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_16X64,
+                        bd);
+}
+
+void av1_inv_txfm2d_add_64x16_c(const int32_t *input, uint16_t *output,
+                                int stride, TX_TYPE tx_type, int bd) {
+  // Remap 32x16 input into a modified 64x16 by:
+  // - Copying over these values in top-left 32x16 locations.
+  // - Setting the rest of the locations to 0.
+  int32_t mod_input[64 * 16];
+  for (int 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));
+  }
+  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
+  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X16,
+                        bd);
+}
+
+void av1_inv_txfm2d_add_4x16_c(const int32_t *input, uint16_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X16, bd);
+}
+
+void av1_inv_txfm2d_add_16x4_c(const int32_t *input, uint16_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X4, bd);
+}
+
+void av1_inv_txfm2d_add_8x32_c(const int32_t *input, uint16_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X32, bd);
+}
+
+void av1_inv_txfm2d_add_32x8_c(const int32_t *input, uint16_t *output,
+                               int stride, TX_TYPE tx_type, int bd) {
+  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
+  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X8, bd);
 }
-#endif  // CONFIG_TX64X64

third_party/aom/av1/common/av1_loopfilter.h

@@ -12,9 +12,9 @@
 #ifndef AV1_COMMON_LOOPFILTER_H_
 #define AV1_COMMON_LOOPFILTER_H_
 
-#include "aom_ports/mem.h"
-#include "./aom_config.h"
+#include "config/aom_config.h"
 
+#include "aom_ports/mem.h"
 #include "av1/common/blockd.h"
 #include "av1/common/seg_common.h"
 
@@ -27,37 +27,111 @@ extern "C" {
 
 #define SIMD_WIDTH 16
 
-#define MAX_MODE_LF_DELTAS 2
-
 enum lf_path {
   LF_PATH_420,
   LF_PATH_444,
   LF_PATH_SLOW,
 };
 
+#if LOOP_FILTER_BITMASK
+typedef struct {
+  uint64_t bits[4];
+} FilterMask;
+
+// 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
+// uint64_t; For U, V plane, for 420 format, plane size is 32x32, thus we use
+// a uint64_t to represent bitmask.
+// Left_ entries refer to whether we apply a filter on the border to the
+// left of the block.   Above_ entries refer to whether or not to apply a
+// filter on the above border.
+// Since each transform is accompanied by a potentially different type of
+// loop filter there is a different entry in the array for each transform size.
+typedef struct {
+  FilterMask left_y[TX_SIZES];
+  FilterMask above_y[TX_SIZES];
+  FilterMask left_u[TX_SIZES];
+  FilterMask above_u[TX_SIZES];
+  FilterMask left_v[TX_SIZES];
+  FilterMask above_v[TX_SIZES];
+
+  // Y plane vertical edge and horizontal edge filter level
+  uint8_t lfl_y_hor[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_y_ver[MI_SIZE_64X64][MI_SIZE_64X64];
+
+  // U plane vertical edge and horizontal edge filter level
+  uint8_t lfl_u_hor[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_u_ver[MI_SIZE_64X64][MI_SIZE_64X64];
+
+  // V plane vertical edge and horizontal edge filter level
+  uint8_t lfl_v_hor[MI_SIZE_64X64][MI_SIZE_64X64];
+  uint8_t lfl_v_ver[MI_SIZE_64X64][MI_SIZE_64X64];
+} LoopFilterMask;
+
+// To determine whether to apply loop filtering at one transform block edge,
+// we need information of the neighboring transform block. Specifically,
+// in determining a vertical edge, we need the information of the tx block
+// to its left. For a horizontal edge, we need info of the tx block above it.
+// Thus, we need to record info of right column and bottom row of tx blocks.
+// We record the information of the neighboring superblock, when bitmask
+// building for a superblock is finished. And it will be used for next
+// superblock bitmask building.
+// Information includes:
+// ------------------------------------------------------------
+//                    MI_SIZE_64X64
+// Y  tx_size above |--------------|
+// Y  tx_size left  |--------------|
+// UV tx_size above |--------------|
+// UV tx_size left  |--------------|
+// Y level above    |--------------|
+// Y level left     |--------------|
+// U level above    |--------------|
+// U level left     |--------------|
+// V level above    |--------------|
+// V level left     |--------------|
+// skip             |--------------|
+// ------------------------------------------------------------
+typedef struct {
+  TX_SIZE tx_size_y_above[MI_SIZE_64X64];
+  TX_SIZE tx_size_y_left[MI_SIZE_64X64];
+  TX_SIZE tx_size_uv_above[MI_SIZE_64X64];
+  TX_SIZE tx_size_uv_left[MI_SIZE_64X64];
+  uint8_t y_level_above[MI_SIZE_64X64];
+  uint8_t y_level_left[MI_SIZE_64X64];
+  uint8_t u_level_above[MI_SIZE_64X64];
+  uint8_t u_level_left[MI_SIZE_64X64];
+  uint8_t v_level_above[MI_SIZE_64X64];
+  uint8_t v_level_left[MI_SIZE_64X64];
+  uint8_t skip[MI_SIZE_64X64];
+} LpfSuperblockInfo;
+#endif  // LOOP_FILTER_BITMASK
+
 struct loopfilter {
-#if CONFIG_LOOPFILTER_LEVEL
   int filter_level[2];
   int filter_level_u;
   int filter_level_v;
-#else
-  int filter_level;
-#endif
 
   int sharpness_level;
-  int last_sharpness_level;
 
   uint8_t mode_ref_delta_enabled;
   uint8_t mode_ref_delta_update;
 
-  // 0 = Intra, Last, Last2+Last3(CONFIG_EXT_REFS),
-  // GF, BRF(CONFIG_EXT_REFS), ARF2(CONFIG_EXT_REFS), ARF
-  int8_t ref_deltas[TOTAL_REFS_PER_FRAME];
-  int8_t last_ref_deltas[TOTAL_REFS_PER_FRAME];
+  // 0 = Intra, Last, Last2+Last3,
+  // GF, BRF, ARF2, ARF
+  int8_t ref_deltas[REF_FRAMES];
 
   // 0 = ZERO_MV, MV
   int8_t mode_deltas[MAX_MODE_LF_DELTAS];
-  int8_t last_mode_deltas[MAX_MODE_LF_DELTAS];
+
+  int combine_vert_horz_lf;
+
+#if LOOP_FILTER_BITMASK
+  LoopFilterMask *lfm;
+  size_t lfm_num;
+  int lfm_stride;
+  LpfSuperblockInfo neighbor_sb_lpf_info;
+#endif  // LOOP_FILTER_BITMASK
 };
 
 // Need to align this structure so when it is declared and
@@ -70,127 +144,56 @@ typedef struct {
 
 typedef struct {
   loop_filter_thresh lfthr[MAX_LOOP_FILTER + 1];
-#if CONFIG_LOOPFILTER_LEVEL
-  uint8_t lvl[MAX_SEGMENTS][2][TOTAL_REFS_PER_FRAME][MAX_MODE_LF_DELTAS];
-#else
-  uint8_t lvl[MAX_SEGMENTS][TOTAL_REFS_PER_FRAME][MAX_MODE_LF_DELTAS];
-#endif
+  uint8_t lvl[MAX_MB_PLANE][MAX_SEGMENTS][2][REF_FRAMES][MAX_MODE_LF_DELTAS];
 } loop_filter_info_n;
 
-// This structure holds bit masks for all 8x8 blocks in a 64x64 region.
-// Each 1 bit represents a position in which we want to apply the loop filter.
-// Left_ entries refer to whether we apply a filter on the border to the
-// left of the block.   Above_ entries refer to whether or not to apply a
-// filter on the above border.   Int_ entries refer to whether or not to
-// apply borders on the 4x4 edges within the 8x8 block that each bit
-// represents.
-// Since each transform is accompanied by a potentially different type of
-// loop filter there is a different entry in the array for each transform size.
-typedef struct {
-  uint64_t left_y[TX_SIZES];
-  uint64_t above_y[TX_SIZES];
-  uint64_t int_4x4_y;
-  uint16_t left_uv[TX_SIZES];
-  uint16_t above_uv[TX_SIZES];
-  uint16_t left_int_4x4_uv;
-  uint16_t above_int_4x4_uv;
-  uint8_t lfl_y[MAX_MIB_SIZE][MAX_MIB_SIZE];
-  uint8_t lfl_uv[MAX_MIB_SIZE / 2][MAX_MIB_SIZE / 2];
-} LOOP_FILTER_MASK;
-
 /* assorted loopfilter functions which get used elsewhere */
 struct AV1Common;
 struct macroblockd;
 struct AV1LfSyncData;
 
-// This function sets up the bit masks for the entire 64x64 region represented
-// by mi_row, mi_col.
-void av1_setup_mask(struct AV1Common *const cm, const int mi_row,
-                    const int mi_col, MODE_INFO **mi_8x8,
-                    const int mode_info_stride, LOOP_FILTER_MASK *lfm);
-
-void av1_filter_block_plane_ss00_ver(struct AV1Common *const cm,
-                                     struct macroblockd_plane *const plane,
-                                     int mi_row, LOOP_FILTER_MASK *lfm);
-void av1_filter_block_plane_ss00_hor(struct AV1Common *const cm,
-                                     struct macroblockd_plane *const plane,
-                                     int mi_row, LOOP_FILTER_MASK *lfm);
-void av1_filter_block_plane_ss11_ver(struct AV1Common *const cm,
-                                     struct macroblockd_plane *const plane,
-                                     int mi_row, LOOP_FILTER_MASK *lfm);
-void av1_filter_block_plane_ss11_hor(struct AV1Common *const cm,
-                                     struct macroblockd_plane *const plane,
-                                     int mi_row, LOOP_FILTER_MASK *lfm);
-
-void av1_filter_block_plane_non420_ver(struct AV1Common *const cm,
-                                       struct macroblockd_plane *plane,
-                                       MODE_INFO **mi_8x8, int mi_row,
-                                       int mi_col, int pl);
-void av1_filter_block_plane_non420_hor(struct AV1Common *const cm,
-                                       struct macroblockd_plane *plane,
-                                       MODE_INFO **mi_8x8, int mi_row,
-                                       int mi_col, int pl);
-
 void av1_loop_filter_init(struct AV1Common *cm);
 
-// Update the loop filter for the current frame.
-// This should be called before av1_loop_filter_rows(),
-// av1_loop_filter_frame()
-// calls this function directly.
-void av1_loop_filter_frame_init(struct AV1Common *cm, int default_filt_lvl,
-                                int default_filt_lvl_r
-#if CONFIG_LOOPFILTER_LEVEL
-                                ,
-                                int plane
-#endif
-                                );
+void av1_loop_filter_frame_init(struct AV1Common *cm, int plane_start,
+                                int plane_end);
 
-#if CONFIG_LPF_SB
 void av1_loop_filter_frame(YV12_BUFFER_CONFIG *frame, struct AV1Common *cm,
-                           struct macroblockd *mbd, int filter_level,
-                           int y_only, int partial_frame, int mi_row,
-                           int mi_col);
+                           struct macroblockd *mbd, int plane_start,
+                           int plane_end, int partial_frame);
 
-// Apply the loop filter to [start, stop) macro block rows in frame_buffer.
-void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
-                          struct AV1Common *cm,
-                          struct macroblockd_plane *planes, int start, int stop,
-                          int col_start, int col_end, int y_only);
+void av1_filter_block_plane_vert(const struct AV1Common *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);
 
-void av1_loop_filter_sb_level_init(struct AV1Common *cm, int mi_row, int mi_col,
-                                   int lvl);
-#else
-void av1_loop_filter_frame(YV12_BUFFER_CONFIG *frame, struct AV1Common *cm,
-                           struct macroblockd *mbd, int filter_level,
-#if CONFIG_LOOPFILTER_LEVEL
-                           int filter_level_r,
-#endif
-                           int y_only, int partial_frame);
-
-// Apply the loop filter to [start, stop) macro block rows in frame_buffer.
-void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
-                          struct AV1Common *cm,
-                          struct macroblockd_plane *planes, int start, int stop,
-                          int y_only);
-#endif  // CONFIG_LPF_SB
+void av1_filter_block_plane_horz(const struct AV1Common *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);
 
 typedef struct LoopFilterWorkerData {
   YV12_BUFFER_CONFIG *frame_buffer;
   struct AV1Common *cm;
   struct macroblockd_plane planes[MAX_MB_PLANE];
-
-  int start;
-  int stop;
-  int y_only;
+  // TODO(Ranjit): When the filter functions are modified to use xd->lossless
+  // add lossless as a member here.
+  MACROBLOCKD *xd;
 } LFWorkerData;
 
-void av1_loop_filter_data_reset(LFWorkerData *lf_data,
-                                YV12_BUFFER_CONFIG *frame_buffer,
-                                struct AV1Common *cm,
-                                const struct macroblockd_plane *planes);
+#if LOOP_FILTER_BITMASK
+void av1_setup_bitmask(struct AV1Common *const cm, int mi_row, int mi_col,
+                       int plane, int subsampling_x, int subsampling_y,
+                       int row_end, int col_end);
+
+void av1_filter_block_plane_ver(struct AV1Common *const cm,
+                                struct macroblockd_plane *const plane_ptr,
+                                int pl, int mi_row, int mi_col);
+
+void av1_filter_block_plane_hor(struct AV1Common *const cm,
+                                struct macroblockd_plane *const plane, int pl,
+                                int mi_row, int mi_col);
+#endif
 
-// Operates on the rows described by 'lf_data'.
-int av1_loop_filter_worker(LFWorkerData *const lf_data, void *unused);
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/av1_rtcd.c

@@ -8,9 +8,11 @@
  * 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_config.h"
+#include "config/aom_config.h"
+
 #define RTCD_C
-#include "./av1_rtcd.h"
+#include "config/av1_rtcd.h"
+
 #include "aom_ports/aom_once.h"
 
 void av1_rtcd() {

third_party/aom/av1/common/av1_rtcd_defs.pl

@@ -1,3 +1,13 @@
+##
+## Copyright (c) 2017, 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.
+##
 sub av1_common_forward_decls() {
 print <<EOF
 /*
@@ -13,6 +23,7 @@ print <<EOF
 #include "av1/common/convolve.h"
 #include "av1/common/av1_txfm.h"
 #include "av1/common/odintrin.h"
+#include "av1/common/restoration.h"
 
 struct macroblockd;
 
@@ -21,9 +32,22 @@ struct macroblock;
 struct txfm_param;
 struct aom_variance_vtable;
 struct search_site_config;
-struct mv;
-union int_mv;
 struct yv12_buffer_config;
+
+/* Function pointers return by CfL functions */
+typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
+                                     uint16_t *output_q3);
+
+typedef void (*cfl_subsample_hbd_fn)(const uint16_t *input, int input_stride,
+                                     uint16_t *output_q3);
+
+typedef void (*cfl_subtract_average_fn)(const uint16_t *src, int16_t *dst);
+
+typedef void (*cfl_predict_lbd_fn)(const int16_t *src, uint8_t *dst,
+                                   int dst_stride, int alpha_q3);
+
+typedef void (*cfl_predict_hbd_fn)(const int16_t *src, uint16_t *dst,
+                                   int dst_stride, int alpha_q3, int bd);
 EOF
 }
 forward_decls qw/av1_common_forward_decls/;
@@ -38,233 +62,55 @@ if ($opts{arch} eq "x86_64") {
   $avx2_x86_64 = 'avx2';
 }
 
-#
-# 10/12-tap convolution filters
-#
-add_proto qw/void av1_lowbd_convolve_init/, "void";
-specialize qw/av1_lowbd_convolve_init ssse3/;
+add_proto qw/void av1_convolve_horiz_rs/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn";
+specialize qw/av1_convolve_horiz_rs sse4_1/;
 
-add_proto qw/void av1_convolve_horiz/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, ConvolveParams *conv_params";
-specialize qw/av1_convolve_horiz ssse3/;
+add_proto qw/void av1_highbd_convolve_horiz_rs/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn, int bd";
+specialize qw/av1_highbd_convolve_horiz_rs sse4_1/;
 
-add_proto qw/void av1_convolve_vert/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, ConvolveParams *conv_params";
-specialize qw/av1_convolve_vert ssse3/;
+add_proto qw/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";
 
-if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-  add_proto qw/void av1_highbd_convolve_init/, "void";
-  specialize qw/av1_highbd_convolve_init sse4_1/;
-  add_proto qw/void av1_highbd_convolve_horiz/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
-  specialize qw/av1_highbd_convolve_horiz sse4_1/;
-  add_proto qw/void av1_highbd_convolve_vert/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
-  specialize qw/av1_highbd_convolve_vert sse4_1/;
-}
+add_proto qw/void av1_highbd_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, int bps";
 
-#
-# Inverse dct
-#
-if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-  add_proto qw/void av1_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht4x4_16_add sse2/;
+specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/;
+specialize qw/av1_highbd_wiener_convolve_add_src ssse3/;
+specialize qw/av1_highbd_wiener_convolve_add_src avx2/;
 
-  add_proto qw/void av1_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht4x8_32_add sse2/;
+# 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";
+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";
+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";
 
-  add_proto qw/void av1_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht8x4_32_add sse2/;
-
-  add_proto qw/void av1_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht8x16_128_add sse2/;
-
-  add_proto qw/void av1_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht16x8_128_add sse2/;
-
-  add_proto qw/void av1_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht16x32_512_add sse2/;
-
-  add_proto qw/void av1_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht32x16_512_add sse2/;
-
-  add_proto qw/void av1_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht8x8_64_add sse2/;
-
-  add_proto qw/void av1_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-  specialize qw/av1_iht16x16_256_add sse2 avx2/;
-
-  add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-} else {
-  add_proto qw/void av1_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-    specialize qw/av1_iht4x4_16_add sse2 neon/;
-  }
-
-  add_proto qw/void av1_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht4x8_32_add sse2/;
-
-  add_proto qw/void av1_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht8x4_32_add sse2/;
-
-  add_proto qw/void av1_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht8x16_128_add sse2/;
-
-  add_proto qw/void av1_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht16x8_128_add sse2/;
-
-  add_proto qw/void av1_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht16x32_512_add sse2/;
-
-  add_proto qw/void av1_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  specialize qw/av1_iht32x16_512_add sse2/;
-
-  add_proto qw/void av1_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-    specialize qw/av1_iht8x8_64_add sse2 neon/;
-  }
-
-  add_proto qw/void av1_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-    specialize qw/av1_iht16x16_256_add sse2 avx2/;
-  }
-
-  add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-
-  if (aom_config("CONFIG_EXT_TX") ne "yes") {
-    if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-      specialize qw/av1_iht4x4_16_add msa/;
-    }
-    if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-      specialize qw/av1_iht8x8_64_add msa/;
-    }
-    if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-      specialize qw/av1_iht16x16_256_add msa/;
-    }
-  }
-}
-
-add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-
-if (aom_config("CONFIG_TX64X64") eq "yes") {
-  add_proto qw/void av1_iht64x64_4096_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-  add_proto qw/void av1_iht32x64_2048_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-  add_proto qw/void av1_iht64x32_2048_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-}
-
-if (aom_config("CONFIG_NEW_QUANT") eq "yes") {
-  add_proto qw/void quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-  add_proto qw/void quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-  add_proto qw/void quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-  add_proto qw/void quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-  if (aom_config("CONFIG_TX64X64") eq "yes") {
-    add_proto qw/void quantize_64x64_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-    add_proto qw/void quantize_64x64_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-  }
-}
 
 # FILTER_INTRA predictor functions
-if (aom_config("CONFIG_FILTER_INTRA") eq "yes") {
-  add_proto qw/void av1_dc_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_v_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_h_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d45_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d135_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d117_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d153_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d207_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_d63_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  add_proto qw/void av1_tm_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left";
-  # High bitdepth functions
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/void av1_highbd_dc_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_v_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_h_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d45_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d135_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d117_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d153_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d207_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_d63_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-    add_proto qw/void av1_highbd_tm_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint16_t *above, const uint16_t *left, int bd";
-  }
-}
+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";
+specialize qw/av1_filter_intra_predictor sse4_1/;
 
 # High bitdepth functions
-if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-  #
-  # Sub Pixel Filters
-  #
-  add_proto qw/void av1_highbd_convolve_copy/, "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";
 
-  add_proto qw/void av1_highbd_convolve_avg/, "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";
+#
+# Sub Pixel Filters
+#
+add_proto qw/void av1_highbd_convolve_copy/, "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";
 
-  add_proto qw/void av1_highbd_convolve8/, "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/av1_highbd_convolve8/, "$sse2_x86_64";
+add_proto qw/void av1_highbd_convolve_avg/, "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";
 
-  add_proto qw/void av1_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/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
+add_proto qw/void av1_highbd_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+specialize qw/av1_highbd_convolve8/, "$sse2_x86_64";
 
-  add_proto qw/void av1_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/av1_highbd_convolve8_vert/, "$sse2_x86_64";
+add_proto qw/void av1_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
 
-  add_proto qw/void av1_highbd_convolve8_avg/, "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/av1_highbd_convolve8_avg/, "$sse2_x86_64";
-
-  add_proto qw/void av1_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
-  specialize qw/av1_highbd_convolve8_avg_horiz/, "$sse2_x86_64";
-
-  add_proto qw/void av1_highbd_convolve8_avg_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/av1_highbd_convolve8_avg_vert/, "$sse2_x86_64";
-
-  #
-  # dct
-  #
-  add_proto qw/void av1_highbd_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
-
-  add_proto qw/void av1_highbd_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, const struct txfm_param *param";
-}
+add_proto qw/void av1_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/av1_highbd_convolve8_vert/, "$sse2_x86_64";
 
 #inv txfm
+add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_inv_txfm_add ssse3 avx2/;
+
+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";
+
 add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
@@ -272,26 +118,41 @@ add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *out
 add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-  specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
-}
+specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
 add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-  specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
-}
+specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
 add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-  specialize qw/av1_inv_txfm2d_add_16x16 sse4_1/;
-}
+specialize qw/av1_inv_txfm2d_add_16x16 sse4_1/;
 add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-if (aom_config("CONFIG_DAALA_DCT32") ne "yes") {
-  specialize qw/av1_inv_txfm2d_add_32x32 avx2/;
-}
-if (aom_config("CONFIG_TX64X64") eq "yes") {
-  add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-  add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-  add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
-}
+specialize qw/av1_inv_txfm2d_add_32x32 avx2/;
+
+add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+
+specialize qw/av1_inv_txfm2d_add_64x64 sse4_1/;
+
+add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
+
+# 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";
+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";
+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";
+
+# 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";
+specialize qw/av1_build_compound_diffwtd_mask sse4_1/;
+
+add_proto qw/void av1_build_compound_diffwtd_mask_highbd/, "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, int bd";
+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 neon/;
 
 #
 # Encoder functions below this point.
@@ -300,199 +161,85 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 
   # ENCODEMB INVOKE
 
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    # the transform coefficients are held in 32-bit
-    # values, so the assembler code for  av1_block_error can no longer be used.
-    add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
-    specialize qw/av1_block_error avx2/;
+  # the transform coefficients are held in 32-bit
+  # values, so the assembler code for  av1_block_error can no longer be used.
+  add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
+  specialize qw/av1_block_error avx2/;
 
-    add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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/av1_quantize_fp sse2 avx2/;
+  add_proto qw/void av1_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";
+  specialize qw/av1_quantize_fp sse2 avx2/;
 
-    add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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/av1_quantize_fp_32x32 avx2/;
+  add_proto qw/void av1_quantize_fp_32x32/, "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/av1_quantize_fp_32x32 avx2/;
 
-    if (aom_config("CONFIG_TX64X64") eq "yes") {
-      add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
-    }
-  } else {
-    add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
-    specialize qw/av1_block_error sse2 avx2 msa/;
-
-    add_proto qw/int64_t av1_block_error_fp/, "const int16_t *coeff, const int16_t *dqcoeff, int block_size";
-    specialize qw/av1_block_error_fp neon sse2/;
-
-    add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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/av1_quantize_fp neon sse2 avx2/, "$ssse3_x86_64";
-
-    add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-    specialize qw/av1_quantize_fp_32x32 avx2/, "$ssse3_x86_64";
-
-    if (aom_config("CONFIG_TX64X64") eq "yes") {
-      add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
-    }
-
-  }
+  add_proto qw/void av1_quantize_fp_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/av1_quantize_fp_64x64 avx2/;
 
   # fdct functions
 
-  add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-    specialize qw/av1_fht4x4 sse2/;
-  }
-
   add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
 
-  add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-    specialize qw/av1_fht8x8 sse2/;
-  }
-
-  add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-    specialize qw/av1_fht16x16 sse2 avx2/;
-  }
-
-  add_proto qw/void av1_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  if (aom_config("CONFIG_DAALA_DCT32") ne "yes") {
-    specialize qw/av1_fht32x32 sse2 avx2/;
-  }
-
-  if (aom_config("CONFIG_TX64X64") eq "yes") {
-    add_proto qw/void av1_fht64x64/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    add_proto qw/void av1_fht64x32/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    add_proto qw/void av1_fht32x64/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  }
-
-  add_proto qw/void av1_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht4x8 sse2/;
-
-  add_proto qw/void av1_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht8x4 sse2/;
-
-  add_proto qw/void av1_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht8x16 sse2/;
-
-  add_proto qw/void av1_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht16x8 sse2/;
-
-  add_proto qw/void av1_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht16x32 sse2/;
-
-  add_proto qw/void av1_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  specialize qw/av1_fht32x16 sse2/;
-
-  add_proto qw/void av1_fht4x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-
-  add_proto qw/void av1_fht16x4/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-
-  add_proto qw/void av1_fht8x32/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-
-  add_proto qw/void av1_fht32x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-
-  if (aom_config("CONFIG_HIGHBITDEPTH") ne "yes") {
-    if (aom_config("CONFIG_EXT_TX") ne "yes") {
-      if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-        specialize qw/av1_fht4x4 msa/;
-      }
-      if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-        specialize qw/av1_fht8x8 msa/;
-      }
-      if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-        specialize qw/av1_fht16x16 msa/;
-      }
-    }
-  }
-
-  add_proto qw/void av1_fwd_idtx/, "const int16_t *src_diff, tran_low_t *coeff, int stride, int bsx, int bsy, TX_TYPE tx_type";
-
   #fwd txfm
+  add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param";
+  specialize qw/av1_lowbd_fwd_txfm sse2 sse4_1/;
+
   add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
   add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  if (aom_config("CONFIG_DAALA_DCT4") ne "yes") {
-    specialize qw/av1_fwd_txfm2d_4x4 sse4_1/;
-  }
+  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";
-  if (aom_config("CONFIG_DAALA_DCT8") ne "yes") {
-    specialize qw/av1_fwd_txfm2d_8x8 sse4_1/;
-  }
+  specialize qw/av1_fwd_txfm2d_8x8 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  if (aom_config("CONFIG_DAALA_DCT16") ne "yes") {
-    specialize qw/av1_fwd_txfm2d_16x16 sse4_1/;
-  }
+  specialize qw/av1_fwd_txfm2d_16x16 sse4_1/;
   add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  if (aom_config("CONFIG_DAALA_DCT32") ne "yes") {
-    specialize qw/av1_fwd_txfm2d_32x32 sse4_1/;
-  }
+  specialize qw/av1_fwd_txfm2d_32x32 sse4_1/;
+
+  add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
+  add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 
-  if (aom_config("CONFIG_TX64X64") eq "yes") {
-    add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-    add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-    add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
-  }
   #
   # Motion search
   #
-  add_proto qw/int av1_full_search_sad/, "const struct macroblock *x, const struct mv *ref_mv, int sad_per_bit, int distance, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv, struct mv *best_mv";
-  specialize qw/av1_full_search_sad sse3 sse4_1/;
-  $av1_full_search_sad_sse3=av1_full_search_sadx3;
-  $av1_full_search_sad_sse4_1=av1_full_search_sadx8;
+  add_proto qw/int av1_diamond_search_sad/, "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/int av1_diamond_search_sad/, "struct macroblock *x, const struct search_site_config *cfg,  struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv";
-
-  add_proto qw/int av1_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv";
+  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/;
 
-  if (aom_config("CONFIG_AOM_QM") eq "yes") {
-    add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
-  } else {
-    add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
-  }
+  add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
 
-  if (aom_config("CONFIG_LGT_FROM_PRED") eq "yes") {
-    add_proto qw/void flgt2d_from_pred/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-  }
+  # ENCODEMB INVOKE
 
-  if (aom_config("CONFIG_HIGHBITDEPTH") 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/;
 
-    # ENCODEMB INVOKE
-    if (aom_config("CONFIG_NEW_QUANT") eq "yes") {
-      add_proto qw/void highbd_quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
+  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 highbd_quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-      add_proto qw/void highbd_quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-      add_proto qw/void highbd_quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-      if (aom_config("CONFIG_TX64X64") eq "yes") {
-        add_proto qw/void highbd_quantize_64x64_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-
-        add_proto qw/void highbd_quantize_64x64_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
-      }
-    }
-
-    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_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, 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";
+  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/;
 
   add_proto qw/void av1_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
 
   # End av1_high encoder functions
 
+  # txb
+  add_proto qw/void av1_get_nz_map_contexts/, "const uint8_t *const levels, const int16_t *const scan, const uint16_t eob, const TX_SIZE tx_size, const TX_CLASS tx_class, int8_t *const coeff_contexts";
+  specialize qw/av1_get_nz_map_contexts sse2/;
+  add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels";
+  specialize qw/av1_txb_init_levels sse4_1/;
+
   add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
   specialize qw/av1_wedge_sse_from_residuals sse2/;
   add_proto qw/int av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
@@ -500,179 +247,132 @@ if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
   add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
   specialize qw/av1_wedge_compute_delta_squares sse2/;
 
+  # hash
+  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/;
+
 }
 # end encoder functions
 
-# If PVQ is enabled, fwd transforms are required by decoder
-if (aom_config("CONFIG_PVQ") eq "yes") {
-  # fdct functions
-
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht4x4 sse2/;
-
-    add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht8x8 sse2/;
-
-    add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht16x16 sse2/;
-
-    add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/av1_fwht4x4 sse2/;
-  } else {
-    add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht4x4 sse2 msa/;
-
-    add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht8x8 sse2 msa/;
-
-    add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, struct txfm_param *param";
-    specialize qw/av1_fht16x16 sse2 msa/;
-
-    add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
-    specialize qw/av1_fwht4x4 msa sse2/;
-  }
-
-}
-
 # Deringing Functions
 
-if (aom_config("CONFIG_CDEF") eq "yes") {
-  add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift";
-  if (aom_config("CONFIG_CDEF_SINGLEPASS") ne "yes") {
-    add_proto qw/void aom_clpf_block_hbd/, "uint16_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
-    add_proto qw/void aom_clpf_hblock_hbd/, "uint16_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
-    add_proto qw/void aom_clpf_block/, "uint8_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
-    add_proto qw/void aom_clpf_hblock/, "uint8_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
-    add_proto qw/void cdef_direction_4x4/, "uint16_t *y, int ystride, const uint16_t *in, int threshold, int dir, int damping";
-    add_proto qw/void cdef_direction_8x8/, "uint16_t *y, int ystride, const uint16_t *in, int threshold, int dir, int damping";
-    add_proto qw/void copy_8x8_16bit_to_8bit/, "uint8_t *dst, int dstride, const uint16_t *src, int sstride";
-    add_proto qw/void copy_4x4_16bit_to_8bit/, "uint8_t *dst, int dstride, const uint16_t *src, int sstride";
-    add_proto qw/void copy_8x8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride";
-    add_proto qw/void copy_4x4_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride";
-  } else {
-    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";
-  }
+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 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";
+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";
 
-  # VS compiling for 32 bit targets does not support vector types in
-  # structs as arguments, which makes the v256 type of the intrinsics
-  # hard to support, so optimizations for this target are disabled.
-  if ($opts{config} !~ /libs-x86-win32-vs.*/) {
-    if (aom_config("CONFIG_CDEF_SINGLEPASS") eq "yes") {
-      specialize qw/cdef_find_dir sse2 ssse3 sse4_1 avx2 neon/;
-      specialize qw/cdef_filter_block sse2 ssse3 sse4_1 avx2 neon/;
-      specialize qw/copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
-      specialize qw/copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
-    } else {
-      specialize qw/cdef_find_dir sse2 ssse3 sse4_1 neon/;
-      specialize qw/aom_clpf_block_hbd sse2 ssse3 sse4_1 neon/;
-      specialize qw/aom_clpf_hblock_hbd sse2 ssse3 sse4_1 neon/;
-      specialize qw/aom_clpf_block sse2 ssse3 sse4_1 neon/;
-      specialize qw/aom_clpf_hblock sse2 ssse3 sse4_1 neon/;
-      specialize qw/cdef_find_dir sse2 ssse3 sse4_1 neon/;
-      specialize qw/cdef_direction_4x4 sse2 ssse3 sse4_1 neon/;
-      specialize qw/cdef_direction_8x8 sse2 ssse3 sse4_1 neon/;
-
-      specialize qw/copy_8x8_16bit_to_8bit sse2 ssse3 sse4_1 neon/;
-      specialize qw/copy_4x4_16bit_to_8bit sse2 ssse3 sse4_1 neon/;
-      specialize qw/copy_8x8_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
-      specialize qw/copy_4x4_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
-      specialize qw/copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 neon/;
-      specialize qw/copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
-    }
-  }
-}
-
-# PVQ Functions
-
-if (aom_config("CONFIG_PVQ") eq "yes") {
-  add_proto qw/double pvq_search_rdo_double/, "const od_val16 *xcoeff, int n, int k, int *ypulse, double g2, double pvq_norm_lambda, int prev_k";
-  specialize qw/pvq_search_rdo_double sse4_1/;
+# VS compiling for 32 bit targets does not support vector types in
+# structs as arguments, which makes the v256 type of the intrinsics
+# hard to support, so optimizations for this target are disabled.
+if ($opts{config} !~ /libs-x86-win32-vs.*/) {
+  specialize qw/cdef_find_dir sse2 ssse3 sse4_1 avx2 neon/;
+  specialize qw/cdef_filter_block sse2 ssse3 sse4_1 avx2 neon/;
+  specialize qw/copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
+  specialize qw/copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
 }
 
 # WARPED_MOTION / GLOBAL_MOTION functions
 
-if ((aom_config("CONFIG_WARPED_MOTION") eq "yes") ||
-    (aom_config("CONFIG_GLOBAL_MOTION") eq "yes")) {
-  add_proto qw/void av1_warp_affine/, "const int32_t *mat, const uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
-  specialize qw/av1_warp_affine sse2 ssse3/;
+add_proto qw/void av1_warp_affine/, "const int32_t *mat, const uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
+specialize qw/av1_warp_affine sse4_1/;
 
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/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";
-    specialize qw/av1_highbd_warp_affine ssse3/;
-  }
-}
+add_proto qw/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";
+specialize qw/av1_highbd_warp_affine sse4_1/;
 
-if (aom_config("CONFIG_GLOBAL_MOTION") eq "yes" &&
-    aom_config("CONFIG_AV1_ENCODER") eq "yes") {
+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/;
 }
 
 # LOOP_RESTORATION functions
 
-if (aom_config("CONFIG_LOOP_RESTORATION") eq "yes") {
-  add_proto qw/void apply_selfguided_restoration/, "uint8_t *dat, int width, int height, int stride, int eps, int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf";
-  specialize qw/apply_selfguided_restoration sse4_1/;
+add_proto qw/void apply_selfguided_restoration/, "const uint8_t *dat, int width, int height, int stride, int eps, const int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf, int bit_depth, int highbd";
+specialize qw/apply_selfguided_restoration sse4_1 avx2/;
 
-  add_proto qw/void av1_selfguided_restoration/, "uint8_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps";
-  specialize qw/av1_selfguided_restoration sse4_1/;
-
-  add_proto qw/void av1_highpass_filter/, "uint8_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps";
-  specialize qw/av1_highpass_filter sse4_1/;
-
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/void apply_selfguided_restoration_highbd/, "uint16_t *dat, int width, int height, int stride, int bit_depth, int eps, int *xqd, uint16_t *dst, int dst_stride, int32_t *tmpbuf";
-    specialize qw/apply_selfguided_restoration_highbd sse4_1/;
-
-    add_proto qw/void av1_selfguided_restoration_highbd/, "uint16_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int bit_depth, int r, int eps";
-    specialize qw/av1_selfguided_restoration_highbd sse4_1/;
-
-    add_proto qw/void av1_highpass_filter_highbd/, "uint16_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps";
-    specialize qw/av1_highpass_filter_highbd sse4_1/;
-  }
-}
+add_proto qw/void av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height,
+                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
+                                  int sgr_params_idx, int bit_depth, int highbd";
+specialize qw/av1_selfguided_restoration sse4_1 avx2/;
 
 # CONVOLVE_ROUND/COMPOUND_ROUND functions
 
-if (aom_config("CONFIG_CONVOLVE_ROUND") eq "yes") {
-  add_proto qw/void av1_convolve_2d/, "const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
-  specialize qw/av1_convolve_2d sse2/;
-  add_proto qw/void av1_convolve_rounding/, "const int32_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, int bits";
-  specialize qw/av1_convolve_rounding avx2/;
+add_proto qw/void av1_convolve_2d_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params";
+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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, 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";
-  if (aom_config("CONFIG_COMPOUND_ROUND") ne "yes") {
-    specialize qw/av1_convolve_2d_scale sse4_1/;
-  }
+  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, InterpFilterParams *filter_params_x, 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, InterpFilterParams *filter_params_x, 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";
 
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/void av1_highbd_convolve_2d/, "const uint16_t *src, int src_stride, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, InterpFilterParams *filter_params_y, const int subpel_x_q4, const int subpel_y_q4, ConvolveParams *conv_params, int bd";
-    specialize qw/av1_highbd_convolve_2d ssse3/;
-    add_proto qw/void av1_highbd_convolve_rounding/, "const int32_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, int bits, int bd";
-    specialize qw/av1_highbd_convolve_rounding avx2/;
-
-    add_proto qw/void av1_highbd_convolve_2d_scale/, "const uint16_t *src, int src_stride, CONV_BUF_TYPE *dst, int dst_stride, int w, int h, InterpFilterParams *filter_params_x, 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";
-    if (aom_config("CONFIG_COMPOUND_ROUND") ne "yes") {
-        specialize qw/av1_highbd_convolve_2d_scale sse4_1/;
-    }
-  }
-}
+  specialize qw/av1_convolve_2d_sr sse2 avx2 neon/;
+  specialize qw/av1_convolve_2d_copy_sr sse2 avx2 neon/;
+  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_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/;
 
 # INTRA_EDGE functions
-if (aom_config("CONFIG_INTRA_EDGE") eq "yes") {
-  add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
-  specialize qw/av1_filter_intra_edge sse4_1/;
-  add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz";
-  specialize qw/av1_upsample_intra_edge sse4_1/;
-  if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
-    add_proto qw/void av1_filter_intra_edge_high/, "uint16_t *p, int sz, int strength";
-    specialize qw/av1_filter_intra_edge_high sse4_1/;
-    add_proto qw/void av1_upsample_intra_edge_high/, "uint16_t *p, int sz, int bd";
-    specialize qw/av1_upsample_intra_edge_high sse4_1/;
-  }
-}
+add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
+specialize qw/av1_filter_intra_edge sse4_1/;
+add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz";
+specialize qw/av1_upsample_intra_edge sse4_1/;
 
+add_proto qw/void av1_filter_intra_edge_high/, "uint16_t *p, int sz, int strength";
+specialize qw/av1_filter_intra_edge_high sse4_1/;
+add_proto qw/void av1_upsample_intra_edge_high/, "uint16_t *p, int sz, int bd";
+specialize qw/av1_upsample_intra_edge_high sse4_1/;
+
+# CFL
+add_proto qw/cfl_subtract_average_fn get_subtract_average_fn/, "TX_SIZE tx_size";
+specialize qw/get_subtract_average_fn sse2 avx2 neon vsx/;
+
+add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size";
+specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/;
+
+add_proto qw/cfl_predict_lbd_fn get_predict_lbd_fn/, "TX_SIZE tx_size";
+specialize qw/get_predict_lbd_fn ssse3 avx2 neon/;
+
+add_proto qw/cfl_predict_hbd_fn get_predict_hbd_fn/, "TX_SIZE tx_size";
+specialize qw/get_predict_hbd_fn ssse3 avx2 neon/;
+
+1;

third_party/aom/av1/common/av1_txfm.c

@@ -0,0 +1,110 @@
+/*
+ * Copyright (c) 2017, 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 "config/aom_dsp_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)));
+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,
+    822,  807,  792,  775,  759,  742,  724,  706,  688,  669, 650, 630, 610,
+    590,  569,  548,  526,  505,  483,  460,  438,  415,  392, 369, 345, 321,
+    297,  273,  249,  224,  200,  175,  150,  125,  100,  75,  50,  25 },
+  { 2048, 2047, 2046, 2042, 2038, 2033, 2026, 2018, 2009, 1998, 1987,
+    1974, 1960, 1945, 1928, 1911, 1892, 1872, 1851, 1829, 1806, 1782,
+    1757, 1730, 1703, 1674, 1645, 1615, 1583, 1551, 1517, 1483, 1448,
+    1412, 1375, 1338, 1299, 1260, 1220, 1179, 1138, 1096, 1053, 1009,
+    965,  921,  876,  830,  784,  737,  690,  642,  595,  546,  498,
+    449,  400,  350,  301,  251,  201,  151,  100,  50 },
+  { 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3996, 3973,
+    3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3612, 3564,
+    3513, 3461, 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896,
+    2824, 2751, 2675, 2598, 2520, 2440, 2359, 2276, 2191, 2106, 2019,
+    1931, 1842, 1751, 1660, 1567, 1474, 1380, 1285, 1189, 1092, 995,
+    897,  799,  700,  601,  501,  401,  301,  201,  101 },
+  { 8192, 8190, 8182, 8170, 8153, 8130, 8103, 8071, 8035, 7993, 7946,
+    7895, 7839, 7779, 7713, 7643, 7568, 7489, 7405, 7317, 7225, 7128,
+    7027, 6921, 6811, 6698, 6580, 6458, 6333, 6203, 6070, 5933, 5793,
+    5649, 5501, 5351, 5197, 5040, 4880, 4717, 4551, 4383, 4212, 4038,
+    3862, 3683, 3503, 3320, 3135, 2948, 2760, 2570, 2378, 2185, 1990,
+    1795, 1598, 1401, 1202, 1003, 803,  603,  402,  201 },
+  { 16384, 16379, 16364, 16340, 16305, 16261, 16207, 16143, 16069, 15986, 15893,
+    15791, 15679, 15557, 15426, 15286, 15137, 14978, 14811, 14635, 14449, 14256,
+    14053, 13842, 13623, 13395, 13160, 12916, 12665, 12406, 12140, 11866, 11585,
+    11297, 11003, 10702, 10394, 10080, 9760,  9434,  9102,  8765,  8423,  8076,
+    7723,  7366,  7005,  6639,  6270,  5897,  5520,  5139,  4756,  4370,  3981,
+    3590,  3196,  2801,  2404,  2006,  1606,  1205,  804,   402 },
+  { 32768, 32758, 32729, 32679, 32610, 32522, 32413, 32286, 32138, 31972, 31786,
+    31581, 31357, 31114, 30853, 30572, 30274, 29957, 29622, 29269, 28899, 28511,
+    28106, 27684, 27246, 26791, 26320, 25833, 25330, 24812, 24279, 23732, 23170,
+    22595, 22006, 21403, 20788, 20160, 19520, 18868, 18205, 17531, 16846, 16151,
+    15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512,  8740,  7962,
+    7180,  6393,  5602,  4808,  4011,  3212,  2411,  1608,  804 },
+  { 65536, 65516, 65457, 65358, 65220, 65043, 64827, 64571, 64277, 63944, 63572,
+    63162, 62714, 62228, 61705, 61145, 60547, 59914, 59244, 58538, 57798, 57022,
+    56212, 55368, 54491, 53581, 52639, 51665, 50660, 49624, 48559, 47464, 46341,
+    45190, 44011, 42806, 41576, 40320, 39040, 37736, 36410, 35062, 33692, 32303,
+    30893, 29466, 28020, 26558, 25080, 23586, 22078, 20557, 19024, 17479, 15924,
+    14359, 12785, 11204, 9616,  8022,  6424,  4821,  3216,  1608 }
+};
+
+// av1_sinpi_arr_data[i][j] = (int)round((sqrt(2) * sin(j*Pi/9) * 2 / 3) * (1
+// << (cos_bit_min + i))) modified so that elements j=1,2 sum to element j=4.
+const int32_t av1_sinpi_arr_data[7][5] = {
+  { 0, 330, 621, 836, 951 },        { 0, 660, 1241, 1672, 1901 },
+  { 0, 1321, 2482, 3344, 3803 },    { 0, 2642, 4964, 6689, 7606 },
+  { 0, 5283, 9929, 13377, 15212 },  { 0, 10566, 19858, 26755, 30424 },
+  { 0, 21133, 39716, 53510, 60849 }
+};
+
+void av1_round_shift_array_c(int32_t *arr, int size, int bit) {
+  int i;
+  if (bit == 0) {
+    return;
+  } else {
+    if (bit > 0) {
+      for (i = 0; i < size; i++) {
+        arr[i] = round_shift(arr[i], bit);
+      }
+    } else {
+      for (i = 0; i < size; i++) {
+        arr[i] = (int32_t)clamp64(((int64_t)1 << (-bit)) * arr[i], INT32_MIN,
+                                  INT32_MAX);
+      }
+    }
+  }
+}
+
+const TXFM_TYPE av1_txfm_type_ls[5][TX_TYPES_1D] = {
+  { TXFM_TYPE_DCT4, TXFM_TYPE_ADST4, TXFM_TYPE_ADST4, TXFM_TYPE_IDENTITY4 },
+  { TXFM_TYPE_DCT8, TXFM_TYPE_ADST8, TXFM_TYPE_ADST8, TXFM_TYPE_IDENTITY8 },
+  { TXFM_TYPE_DCT16, TXFM_TYPE_ADST16, TXFM_TYPE_ADST16, TXFM_TYPE_IDENTITY16 },
+  { TXFM_TYPE_DCT32, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID,
+    TXFM_TYPE_IDENTITY32 },
+  { TXFM_TYPE_DCT64, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID }
+};
+
+const int8_t av1_txfm_stage_num_list[TXFM_TYPES] = {
+  4,   // TXFM_TYPE_DCT4
+  6,   // TXFM_TYPE_DCT8
+  8,   // TXFM_TYPE_DCT16
+  10,  // TXFM_TYPE_DCT32
+  12,  // TXFM_TYPE_DCT64
+  7,   // TXFM_TYPE_ADST4
+  8,   // TXFM_TYPE_ADST8
+  10,  // TXFM_TYPE_ADST16
+  1,   // TXFM_TYPE_IDENTITY4
+  1,   // TXFM_TYPE_IDENTITY8
+  1,   // TXFM_TYPE_IDENTITY16
+  1,   // TXFM_TYPE_IDENTITY32
+};

third_party/aom/av1/common/av1_txfm.h

@@ -16,6 +16,8 @@
 #include <math.h>
 #include <stdio.h>
 
+#include "config/aom_config.h"
+
 #include "av1/common/enums.h"
 #include "av1/common/blockd.h"
 #include "aom/aom_integer.h"
@@ -25,100 +27,73 @@
 extern "C" {
 #endif
 
+#if !defined(DO_RANGE_CHECK_CLAMP)
+#define DO_RANGE_CHECK_CLAMP 0
+#endif
+
+extern const int32_t av1_cospi_arr_data[7][64];
+extern const int32_t av1_sinpi_arr_data[7][5];
+
 #define MAX_TXFM_STAGE_NUM 12
 
 static const int cos_bit_min = 10;
 static const int cos_bit_max = 16;
 
-// cospi_arr[i][j] = (int)round(cos(M_PI*j/128) * (1<<(cos_bit_min+i)));
-static const int32_t 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,
-    822,  807,  792,  775,  759,  742,  724,  706,  688,  669, 650, 630, 610,
-    590,  569,  548,  526,  505,  483,  460,  438,  415,  392, 369, 345, 321,
-    297,  273,  249,  224,  200,  175,  150,  125,  100,  75,  50,  25 },
-  { 2048, 2047, 2046, 2042, 2038, 2033, 2026, 2018, 2009, 1998, 1987,
-    1974, 1960, 1945, 1928, 1911, 1892, 1872, 1851, 1829, 1806, 1782,
-    1757, 1730, 1703, 1674, 1645, 1615, 1583, 1551, 1517, 1483, 1448,
-    1412, 1375, 1338, 1299, 1260, 1220, 1179, 1138, 1096, 1053, 1009,
-    965,  921,  876,  830,  784,  737,  690,  642,  595,  546,  498,
-    449,  400,  350,  301,  251,  201,  151,  100,  50 },
-  { 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3996, 3973,
-    3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3612, 3564,
-    3513, 3461, 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896,
-    2824, 2751, 2675, 2598, 2520, 2440, 2359, 2276, 2191, 2106, 2019,
-    1931, 1842, 1751, 1660, 1567, 1474, 1380, 1285, 1189, 1092, 995,
-    897,  799,  700,  601,  501,  401,  301,  201,  101 },
-  { 8192, 8190, 8182, 8170, 8153, 8130, 8103, 8071, 8035, 7993, 7946,
-    7895, 7839, 7779, 7713, 7643, 7568, 7489, 7405, 7317, 7225, 7128,
-    7027, 6921, 6811, 6698, 6580, 6458, 6333, 6203, 6070, 5933, 5793,
-    5649, 5501, 5351, 5197, 5040, 4880, 4717, 4551, 4383, 4212, 4038,
-    3862, 3683, 3503, 3320, 3135, 2948, 2760, 2570, 2378, 2185, 1990,
-    1795, 1598, 1401, 1202, 1003, 803,  603,  402,  201 },
-  { 16384, 16379, 16364, 16340, 16305, 16261, 16207, 16143, 16069, 15986, 15893,
-    15791, 15679, 15557, 15426, 15286, 15137, 14978, 14811, 14635, 14449, 14256,
-    14053, 13842, 13623, 13395, 13160, 12916, 12665, 12406, 12140, 11866, 11585,
-    11297, 11003, 10702, 10394, 10080, 9760,  9434,  9102,  8765,  8423,  8076,
-    7723,  7366,  7005,  6639,  6270,  5897,  5520,  5139,  4756,  4370,  3981,
-    3590,  3196,  2801,  2404,  2006,  1606,  1205,  804,   402 },
-  { 32768, 32758, 32729, 32679, 32610, 32522, 32413, 32286, 32138, 31972, 31786,
-    31581, 31357, 31114, 30853, 30572, 30274, 29957, 29622, 29269, 28899, 28511,
-    28106, 27684, 27246, 26791, 26320, 25833, 25330, 24812, 24279, 23732, 23170,
-    22595, 22006, 21403, 20788, 20160, 19520, 18868, 18205, 17531, 16846, 16151,
-    15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512,  8740,  7962,
-    7180,  6393,  5602,  4808,  4011,  3212,  2411,  1608,  804 },
-  { 65536, 65516, 65457, 65358, 65220, 65043, 64827, 64571, 64277, 63944, 63572,
-    63162, 62714, 62228, 61705, 61145, 60547, 59914, 59244, 58538, 57798, 57022,
-    56212, 55368, 54491, 53581, 52639, 51665, 50660, 49624, 48559, 47464, 46341,
-    45190, 44011, 42806, 41576, 40320, 39040, 37736, 36410, 35062, 33692, 32303,
-    30893, 29466, 28020, 26558, 25080, 23586, 22078, 20557, 19024, 17479, 15924,
-    14359, 12785, 11204, 9616,  8022,  6424,  4821,  3216,  1608 }
-};
+static const int NewSqrt2Bits = 12;
+// 2^12 * sqrt(2)
+static const int32_t NewSqrt2 = 5793;
+// 2^12 / sqrt(2)
+static const int32_t NewInvSqrt2 = 2896;
 
 static INLINE const int32_t *cospi_arr(int n) {
-  return cospi_arr_data[n - cos_bit_min];
+  return av1_cospi_arr_data[n - cos_bit_min];
 }
 
-static INLINE int32_t round_shift(int32_t value, int bit) {
-  assert(bit >= 1);
-  return (value + (1 << (bit - 1))) >> bit;
+static INLINE const int32_t *sinpi_arr(int n) {
+  return av1_sinpi_arr_data[n - cos_bit_min];
 }
 
-static INLINE void round_shift_array(int32_t *arr, int size, int bit) {
-  int i;
-  if (bit == 0) {
-    return;
-  } else {
-    if (bit > 0) {
-      for (i = 0; i < size; i++) {
-        arr[i] = round_shift(arr[i], bit);
-      }
-    } else {
-      for (i = 0; i < size; i++) {
-        arr[i] = arr[i] * (1 << (-bit));
-      }
-    }
+static INLINE int32_t range_check_value(int32_t value, int8_t bit) {
+#if CONFIG_COEFFICIENT_RANGE_CHECKING
+  const int64_t max_value = (1LL << (bit - 1)) - 1;
+  const int64_t min_value = -(1LL << (bit - 1));
+  if (value < min_value || value > max_value) {
+    fprintf(stderr, "coeff out of bit range, value: %d bit %d\n", value, bit);
+    assert(0);
   }
+#endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
+#if DO_RANGE_CHECK_CLAMP
+  bit = AOMMIN(bit, 31);
+  return clamp(value, (1 << (bit - 1)) - 1, -(1 << (bit - 1)));
+#endif  // DO_RANGE_CHECK_CLAMP
+  (void)bit;
+  return value;
+}
+
+static INLINE int32_t round_shift(int64_t value, int bit) {
+  assert(bit >= 1);
+  return (int32_t)((value + (1ll << (bit - 1))) >> bit);
 }
 
 static INLINE int32_t half_btf(int32_t w0, int32_t in0, int32_t w1, int32_t in1,
                                int bit) {
-  int32_t result_32 = w0 * in0 + w1 * in1;
+  int64_t result_64 = (int64_t)(w0 * in0) + (int64_t)(w1 * in1);
 #if CONFIG_COEFFICIENT_RANGE_CHECKING
-  int64_t result_64 = (int64_t)w0 * (int64_t)in0 + (int64_t)w1 * (int64_t)in1;
-  if (result_64 < INT32_MIN || result_64 > INT32_MAX) {
-    printf("%s %d overflow result_32: %d result_64: %" PRId64
-           " w0: %d in0: %d w1: %d in1: "
-           "%d\n",
-           __FILE__, __LINE__, result_32, result_64, w0, in0, w1, in1);
-    assert(0 && "half_btf overflow");
-  }
+  assert(result_64 >= INT32_MIN && result_64 <= INT32_MAX);
 #endif
-  return round_shift(result_32, bit);
+  return round_shift(result_64, bit);
 }
 
-typedef void (*TxfmFunc)(const int32_t *input, int32_t *output,
-                         const int8_t *cos_bit, const int8_t *stage_range);
+static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans,
+                                             int bd) {
+  return clip_pixel_highbd(dest + (int)trans, bd);
+}
+
+typedef void (*TxfmFunc)(const int32_t *input, int32_t *output, int8_t cos_bit,
+                         const int8_t *stage_range);
+
+typedef void (*FwdTxfm2dFunc)(const int16_t *input, int32_t *output, int stride,
+                              TX_TYPE tx_type, int bd);
 
 typedef enum TXFM_TYPE {
   TXFM_TYPE_DCT4,
@@ -129,88 +104,82 @@ typedef enum TXFM_TYPE {
   TXFM_TYPE_ADST4,
   TXFM_TYPE_ADST8,
   TXFM_TYPE_ADST16,
-  TXFM_TYPE_ADST32,
   TXFM_TYPE_IDENTITY4,
   TXFM_TYPE_IDENTITY8,
   TXFM_TYPE_IDENTITY16,
   TXFM_TYPE_IDENTITY32,
-  TXFM_TYPE_IDENTITY64,
+  TXFM_TYPES,
+  TXFM_TYPE_INVALID,
 } TXFM_TYPE;
 
-typedef struct TXFM_1D_CFG {
-  const int txfm_size;
-  const int stage_num;
-
-  const int8_t *shift;
-  const int8_t *stage_range;
-  const int8_t *cos_bit;
-  const TXFM_TYPE txfm_type;
-} TXFM_1D_CFG;
-
 typedef struct TXFM_2D_FLIP_CFG {
+  TX_SIZE tx_size;
   int ud_flip;  // flip upside down
   int lr_flip;  // flip left to right
-  const TXFM_1D_CFG *col_cfg;
-  const TXFM_1D_CFG *row_cfg;
+  const int8_t *shift;
+  int8_t cos_bit_col;
+  int8_t cos_bit_row;
+  int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
+  int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
+  TXFM_TYPE txfm_type_col;
+  TXFM_TYPE txfm_type_row;
+  int stage_num_col;
+  int stage_num_row;
 } TXFM_2D_FLIP_CFG;
 
-static INLINE void set_flip_cfg(TX_TYPE tx_type, TXFM_2D_FLIP_CFG *cfg) {
+static INLINE void get_flip_cfg(TX_TYPE tx_type, int *ud_flip, int *lr_flip) {
   switch (tx_type) {
     case DCT_DCT:
     case ADST_DCT:
     case DCT_ADST:
     case ADST_ADST:
-      cfg->ud_flip = 0;
-      cfg->lr_flip = 0;
+      *ud_flip = 0;
+      *lr_flip = 0;
       break;
-#if CONFIG_EXT_TX
     case IDTX:
     case V_DCT:
     case H_DCT:
     case V_ADST:
     case H_ADST:
-      cfg->ud_flip = 0;
-      cfg->lr_flip = 0;
+      *ud_flip = 0;
+      *lr_flip = 0;
       break;
     case FLIPADST_DCT:
     case FLIPADST_ADST:
     case V_FLIPADST:
-      cfg->ud_flip = 1;
-      cfg->lr_flip = 0;
+      *ud_flip = 1;
+      *lr_flip = 0;
       break;
     case DCT_FLIPADST:
     case ADST_FLIPADST:
     case H_FLIPADST:
-      cfg->ud_flip = 0;
-      cfg->lr_flip = 1;
+      *ud_flip = 0;
+      *lr_flip = 1;
       break;
     case FLIPADST_FLIPADST:
-      cfg->ud_flip = 1;
-      cfg->lr_flip = 1;
+      *ud_flip = 1;
+      *lr_flip = 1;
       break;
-#endif  // CONFIG_EXT_TX
     default:
-      cfg->ud_flip = 0;
-      cfg->lr_flip = 0;
+      *ud_flip = 0;
+      *lr_flip = 0;
       assert(0);
   }
 }
 
-#if CONFIG_TXMG
+static INLINE void set_flip_cfg(TX_TYPE tx_type, TXFM_2D_FLIP_CFG *cfg) {
+  get_flip_cfg(tx_type, &cfg->ud_flip, &cfg->lr_flip);
+}
+
 static INLINE TX_SIZE av1_rotate_tx_size(TX_SIZE tx_size) {
   switch (tx_size) {
-#if CONFIG_CHROMA_2X2
-    case TX_2X2: return TX_2X2;
-#endif
     case TX_4X4: return TX_4X4;
     case TX_8X8: return TX_8X8;
     case TX_16X16: return TX_16X16;
     case TX_32X32: return TX_32X32;
-#if CONFIG_TX64X64
     case TX_64X64: return TX_64X64;
     case TX_32X64: return TX_64X32;
     case TX_64X32: return TX_32X64;
-#endif
     case TX_4X8: return TX_8X4;
     case TX_8X4: return TX_4X8;
     case TX_8X16: return TX_16X8;
@@ -221,6 +190,8 @@ static INLINE TX_SIZE av1_rotate_tx_size(TX_SIZE tx_size) {
     case TX_16X4: return TX_4X16;
     case TX_8X32: return TX_32X8;
     case TX_32X8: return TX_8X32;
+    case TX_16X64: return TX_64X16;
+    case TX_64X16: return TX_16X64;
     default: assert(0); return TX_INVALID;
   }
 }
@@ -231,7 +202,6 @@ static INLINE TX_TYPE av1_rotate_tx_type(TX_TYPE tx_type) {
     case ADST_DCT: return DCT_ADST;
     case DCT_ADST: return ADST_DCT;
     case ADST_ADST: return ADST_ADST;
-#if CONFIG_EXT_TX
     case FLIPADST_DCT: return DCT_FLIPADST;
     case DCT_FLIPADST: return FLIPADST_DCT;
     case FLIPADST_FLIPADST: return FLIPADST_FLIPADST;
@@ -244,123 +214,46 @@ static INLINE TX_TYPE av1_rotate_tx_type(TX_TYPE tx_type) {
     case H_ADST: return V_ADST;
     case V_FLIPADST: return H_FLIPADST;
     case H_FLIPADST: return V_FLIPADST;
-#endif  // CONFIG_EXT_TX
-#if CONFIG_MRC_TX
-    case MRC_DCT: return MRC_DCT;
-#endif  // CONFIG_MRC_TX
     default: assert(0); return TX_TYPES;
   }
 }
-#endif  // CONFIG_TXMG
 
-#if CONFIG_MRC_TX
-static INLINE int get_mrc_diff_mask_inter(const int16_t *diff, int diff_stride,
-                                          uint8_t *mask, int mask_stride,
-                                          int width, int height) {
-  // placeholder mask generation function
-  assert(SIGNAL_MRC_MASK_INTER);
-  int n_masked_vals = 0;
-  for (int i = 0; i < height; ++i) {
-    for (int j = 0; j < width; ++j) {
-      mask[i * mask_stride + j] = diff[i * diff_stride + j] > 100 ? 1 : 0;
-      n_masked_vals += mask[i * mask_stride + j];
-    }
-  }
-  return n_masked_vals;
-}
-
-static INLINE int get_mrc_pred_mask_inter(const uint8_t *pred, int pred_stride,
-                                          uint8_t *mask, int mask_stride,
-                                          int width, int height) {
-  // placeholder mask generation function
-  int n_masked_vals = 0;
-  for (int i = 0; i < height; ++i) {
-    for (int j = 0; j < width; ++j) {
-      mask[i * mask_stride + j] = pred[i * pred_stride + j] > 100 ? 1 : 0;
-      n_masked_vals += mask[i * mask_stride + j];
-    }
-  }
-  return n_masked_vals;
-}
-
-static INLINE int get_mrc_diff_mask_intra(const int16_t *diff, int diff_stride,
-                                          uint8_t *mask, int mask_stride,
-                                          int width, int height) {
-  // placeholder mask generation function
-  assert(SIGNAL_MRC_MASK_INTRA);
-  int n_masked_vals = 0;
-  for (int i = 0; i < height; ++i) {
-    for (int j = 0; j < width; ++j) {
-      mask[i * mask_stride + j] = diff[i * diff_stride + j] > 100 ? 1 : 0;
-      n_masked_vals += mask[i * mask_stride + j];
-    }
-  }
-  return n_masked_vals;
-}
-
-static INLINE int get_mrc_pred_mask_intra(const uint8_t *pred, int pred_stride,
-                                          uint8_t *mask, int mask_stride,
-                                          int width, int height) {
-  // placeholder mask generation function
-  int n_masked_vals = 0;
-  for (int i = 0; i < height; ++i) {
-    for (int j = 0; j < width; ++j) {
-      mask[i * mask_stride + j] = pred[i * pred_stride + j] > 100 ? 1 : 0;
-      n_masked_vals += mask[i * mask_stride + j];
-    }
-  }
-  return n_masked_vals;
-}
-
-static INLINE int get_mrc_diff_mask(const int16_t *diff, int diff_stride,
-                                    uint8_t *mask, int mask_stride, int width,
-                                    int height, int is_inter) {
-  if (is_inter) {
-    assert(USE_MRC_INTER && "MRC invalid for inter blocks");
-    assert(SIGNAL_MRC_MASK_INTER);
-    return get_mrc_diff_mask_inter(diff, diff_stride, mask, mask_stride, width,
-                                   height);
+// Utility function that returns the log of the ratio of the col and row
+// sizes.
+static INLINE int get_rect_tx_log_ratio(int col, int row) {
+  if (col == row) return 0;
+  if (col > row) {
+    if (col == row * 2) return 1;
+    if (col == row * 4) return 2;
+    assert(0 && "Unsupported transform size");
   } else {
-    assert(USE_MRC_INTRA && "MRC invalid for intra blocks");
-    assert(SIGNAL_MRC_MASK_INTRA);
-    return get_mrc_diff_mask_intra(diff, diff_stride, mask, mask_stride, width,
-                                   height);
+    if (row == col * 2) return -1;
+    if (row == col * 4) return -2;
+    assert(0 && "Unsupported transform size");
   }
+  return 0;  // Invalid
 }
 
-static INLINE int get_mrc_pred_mask(const uint8_t *pred, int pred_stride,
-                                    uint8_t *mask, int mask_stride, int width,
-                                    int height, int is_inter) {
-  if (is_inter) {
-    assert(USE_MRC_INTER && "MRC invalid for inter blocks");
-    return get_mrc_pred_mask_inter(pred, pred_stride, mask, mask_stride, width,
-                                   height);
-  } else {
-    assert(USE_MRC_INTRA && "MRC invalid for intra blocks");
-    return get_mrc_pred_mask_intra(pred, pred_stride, mask, mask_stride, width,
-                                   height);
-  }
-}
-
-static INLINE int is_valid_mrc_mask(int n_masked_vals, int width, int height) {
-  return !(n_masked_vals == 0 || n_masked_vals == (width * height));
-}
-#endif  // CONFIG_MRC_TX
-
 void av1_gen_fwd_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
                              const TXFM_2D_FLIP_CFG *cfg, int bd);
 
 void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
-                             const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift,
+                             const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
                              int bd);
 
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size);
-#if CONFIG_TX64X64
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(TX_TYPE tx_type);
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x32_cfg(TX_TYPE tx_type);
-TXFM_2D_FLIP_CFG av1_get_fwd_txfm_32x64_cfg(TX_TYPE tx_type);
-#endif  // CONFIG_TX64X64
-TXFM_2D_FLIP_CFG av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size);
+void av1_get_fwd_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
+                          TXFM_2D_FLIP_CFG *cfg);
+void av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
+                          TXFM_2D_FLIP_CFG *cfg);
+extern const TXFM_TYPE av1_txfm_type_ls[5][TX_TYPES_1D];
+extern const int8_t av1_txfm_stage_num_list[TXFM_TYPES];
+static INLINE int get_txw_idx(TX_SIZE tx_size) {
+  return tx_size_wide_log2[tx_size] - tx_size_wide_log2[0];
+}
+static INLINE int get_txh_idx(TX_SIZE tx_size) {
+  return tx_size_high_log2[tx_size] - tx_size_high_log2[0];
+}
+#define MAX_TXWH_IDX 5
 #ifdef __cplusplus
 }
 #endif  // __cplusplus

third_party/aom/av1/common/blockd.c

@@ -16,110 +16,18 @@
 #include "av1/common/blockd.h"
 #include "av1/common/onyxc_int.h"
 
-PREDICTION_MODE av1_left_block_mode(const MODE_INFO *cur_mi,
-                                    const MODE_INFO *left_mi, int b) {
-  if (b == 0 || b == 2) {
-    if (!left_mi || is_inter_block(&left_mi->mbmi)) return DC_PRED;
-
-    return get_y_mode(left_mi, b + 1);
-  } else {
-    assert(b == 1 || b == 3);
-    return cur_mi->bmi[b - 1].as_mode;
-  }
+PREDICTION_MODE av1_left_block_mode(const MB_MODE_INFO *left_mi) {
+  if (!left_mi) return DC_PRED;
+  assert(!is_inter_block(left_mi) || is_intrabc_block(left_mi));
+  return left_mi->mode;
 }
 
-PREDICTION_MODE av1_above_block_mode(const MODE_INFO *cur_mi,
-                                     const MODE_INFO *above_mi, int b) {
-  if (b == 0 || b == 1) {
-    if (!above_mi || is_inter_block(&above_mi->mbmi)) return DC_PRED;
-
-    return get_y_mode(above_mi, b + 2);
-  } else {
-    assert(b == 2 || b == 3);
-    return cur_mi->bmi[b - 2].as_mode;
-  }
+PREDICTION_MODE av1_above_block_mode(const MB_MODE_INFO *above_mi) {
+  if (!above_mi) return DC_PRED;
+  assert(!is_inter_block(above_mi) || is_intrabc_block(above_mi));
+  return above_mi->mode;
 }
 
-#if CONFIG_COEF_INTERLEAVE
-void av1_foreach_transformed_block_interleave(
-    const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
-    foreach_transformed_block_visitor visit, void *arg) {
-  const struct macroblockd_plane *const pd_y = &xd->plane[0];
-  const struct macroblockd_plane *const pd_c = &xd->plane[1];
-  const MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
-
-  const TX_SIZE tx_log2_y = mbmi->tx_size;
-  const TX_SIZE tx_log2_c = av1_get_uv_tx_size(mbmi, pd_c);
-  const int tx_sz_y = (1 << tx_log2_y);
-  const int tx_sz_c = (1 << tx_log2_c);
-
-  const BLOCK_SIZE plane_bsize_y = get_plane_block_size(bsize, pd_y);
-  const BLOCK_SIZE plane_bsize_c = get_plane_block_size(bsize, pd_c);
-
-  const int num_4x4_w_y = num_4x4_blocks_wide_lookup[plane_bsize_y];
-  const int num_4x4_w_c = num_4x4_blocks_wide_lookup[plane_bsize_c];
-  const int num_4x4_h_y = num_4x4_blocks_high_lookup[plane_bsize_y];
-  const int num_4x4_h_c = num_4x4_blocks_high_lookup[plane_bsize_c];
-
-  const int step_y = 1 << (tx_log2_y << 1);
-  const int step_c = 1 << (tx_log2_c << 1);
-
-  const int max_4x4_w_y =
-      get_max_4x4_size(num_4x4_w_y, xd->mb_to_right_edge, pd_y->subsampling_x);
-  const int max_4x4_h_y =
-      get_max_4x4_size(num_4x4_h_y, xd->mb_to_bottom_edge, pd_y->subsampling_y);
-
-  const int extra_step_y = ((num_4x4_w_y - max_4x4_w_y) >> tx_log2_y) * step_y;
-
-  const int max_4x4_w_c =
-      get_max_4x4_size(num_4x4_w_c, xd->mb_to_right_edge, pd_c->subsampling_x);
-  const int max_4x4_h_c =
-      get_max_4x4_size(num_4x4_h_c, xd->mb_to_bottom_edge, pd_c->subsampling_y);
-
-  const int extra_step_c = ((num_4x4_w_c - max_4x4_w_c) >> tx_log2_c) * step_c;
-
-  // The max_4x4_w/h may be smaller than tx_sz under some corner cases,
-  // i.e. when the SB is splitted by tile boundaries.
-  const int tu_num_w_y = (max_4x4_w_y + tx_sz_y - 1) / tx_sz_y;
-  const int tu_num_h_y = (max_4x4_h_y + tx_sz_y - 1) / tx_sz_y;
-  const int tu_num_w_c = (max_4x4_w_c + tx_sz_c - 1) / tx_sz_c;
-  const int tu_num_h_c = (max_4x4_h_c + tx_sz_c - 1) / tx_sz_c;
-  const int tu_num_c = tu_num_w_c * tu_num_h_c;
-
-  int tu_idx_c = 0;
-  int offset_y, row_y, col_y;
-  int offset_c, row_c, col_c;
-
-  for (row_y = 0; row_y < tu_num_h_y; row_y++) {
-    for (col_y = 0; col_y < tu_num_w_y; col_y++) {
-      // luma
-      offset_y = (row_y * tu_num_w_y + col_y) * step_y + row_y * extra_step_y;
-      visit(0, offset_y, row_y * tx_sz_y, col_y * tx_sz_y, plane_bsize_y,
-            tx_log2_y, arg);
-      // chroma
-      if (tu_idx_c < tu_num_c) {
-        row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-        col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-        offset_c = tu_idx_c * step_c + (tu_idx_c / tu_num_w_c) * extra_step_c;
-        visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
-        visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
-        tu_idx_c++;
-      }
-    }
-  }
-
-  // In 422 case, it's possible that Chroma has more TUs than Luma
-  while (tu_idx_c < tu_num_c) {
-    row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
-    col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
-    offset_c = tu_idx_c * step_c + row_c * extra_step_c;
-    visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
-    visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
-    tu_idx_c++;
-  }
-}
-#endif
-
 void av1_foreach_transformed_block_in_plane(
     const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
     foreach_transformed_block_visitor visit, void *arg) {
@@ -128,12 +36,8 @@ void av1_foreach_transformed_block_in_plane(
   // 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
   // transform size varies per plane, look it up in a common way.
   const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
-#if CONFIG_CHROMA_SUB8X8
   const BLOCK_SIZE plane_bsize =
-      AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#else
-  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#endif
+      get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
   const uint8_t txw_unit = tx_size_wide_unit[tx_size];
   const uint8_t txh_unit = tx_size_high_unit[tx_size];
   const int step = txw_unit * txh_unit;
@@ -147,7 +51,8 @@ void av1_foreach_transformed_block_in_plane(
 
   int blk_row, blk_col;
 
-  const BLOCK_SIZE max_unit_bsize = get_plane_block_size(BLOCK_64X64, pd);
+  const BLOCK_SIZE max_unit_bsize =
+      get_plane_block_size(BLOCK_64X64, pd->subsampling_x, pd->subsampling_y);
   int mu_blocks_wide = block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
   int mu_blocks_high = block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
   mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
@@ -170,95 +75,60 @@ void av1_foreach_transformed_block_in_plane(
   }
 }
 
-#if CONFIG_LV_MAP
 void av1_foreach_transformed_block(const MACROBLOCKD *const xd,
                                    BLOCK_SIZE bsize, int mi_row, int mi_col,
                                    foreach_transformed_block_visitor visit,
-                                   void *arg) {
-  int plane;
-
-  for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-#if CONFIG_CB4X4
+                                   void *arg, const int num_planes) {
+  for (int plane = 0; plane < num_planes; ++plane) {
     if (!is_chroma_reference(mi_row, mi_col, bsize,
                              xd->plane[plane].subsampling_x,
                              xd->plane[plane].subsampling_y))
       continue;
-#else
-    (void)mi_row;
-    (void)mi_col;
-#endif
     av1_foreach_transformed_block_in_plane(xd, bsize, plane, visit, arg);
   }
 }
-#endif
 
-#if !CONFIG_PVQ || CONFIG_VAR_TX
 void av1_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
-                      int plane, TX_SIZE tx_size, int has_eob, int aoff,
-                      int loff) {
+                      int plane, BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
+                      int has_eob, int aoff, int loff) {
   ENTROPY_CONTEXT *const a = pd->above_context + aoff;
   ENTROPY_CONTEXT *const l = pd->left_context + loff;
   const int txs_wide = tx_size_wide_unit[tx_size];
   const int txs_high = tx_size_high_unit[tx_size];
-#if CONFIG_CB4X4
-  const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
-#else
-  const BLOCK_SIZE bsize = AOMMAX(xd->mi[0]->mbmi.sb_type, BLOCK_8X8);
-#endif
-  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
 
   // above
   if (has_eob && xd->mb_to_right_edge < 0) {
-    int i;
     const int blocks_wide = max_block_wide(xd, plane_bsize, plane);
-    int above_contexts = txs_wide;
-    if (above_contexts + aoff > blocks_wide)
-      above_contexts = blocks_wide - aoff;
-
-    for (i = 0; i < above_contexts; ++i) a[i] = has_eob;
-    for (i = above_contexts; i < txs_wide; ++i) a[i] = 0;
+    const int above_contexts = AOMMIN(txs_wide, blocks_wide - aoff);
+    memset(a, has_eob, sizeof(*a) * above_contexts);
+    memset(a + above_contexts, 0, sizeof(*a) * (txs_wide - above_contexts));
   } else {
-    memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * txs_wide);
+    memset(a, has_eob, sizeof(*a) * txs_wide);
   }
 
   // left
   if (has_eob && xd->mb_to_bottom_edge < 0) {
-    int i;
     const int blocks_high = max_block_high(xd, plane_bsize, plane);
-    int left_contexts = txs_high;
-    if (left_contexts + loff > blocks_high) left_contexts = blocks_high - loff;
-
-    for (i = 0; i < left_contexts; ++i) l[i] = has_eob;
-    for (i = left_contexts; i < txs_high; ++i) l[i] = 0;
+    const int left_contexts = AOMMIN(txs_high, blocks_high - loff);
+    memset(l, has_eob, sizeof(*l) * left_contexts);
+    memset(l + left_contexts, 0, sizeof(*l) * (txs_high - left_contexts));
   } else {
-    memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * txs_high);
+    memset(l, has_eob, sizeof(*l) * txs_high);
   }
 }
-#endif
-
 void av1_reset_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col,
-                            BLOCK_SIZE bsize) {
+                            BLOCK_SIZE bsize, const int num_planes) {
   int i;
   int nplanes;
-#if CONFIG_CB4X4
   int chroma_ref;
   chroma_ref =
       is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x,
                           xd->plane[1].subsampling_y);
-  nplanes = 1 + (MAX_MB_PLANE - 1) * chroma_ref;
-#else
-  (void)mi_row;
-  (void)mi_col;
-  nplanes = MAX_MB_PLANE;
-#endif
+  nplanes = 1 + (num_planes - 1) * chroma_ref;
   for (i = 0; i < nplanes; i++) {
     struct macroblockd_plane *const pd = &xd->plane[i];
-#if CONFIG_CHROMA_SUB8X8
     const BLOCK_SIZE plane_bsize =
-        AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
-#else
-    const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
-#endif
+        get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y);
     const int txs_wide = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
     const int txs_high = block_size_high[plane_bsize] >> tx_size_high_log2[0];
     memset(pd->above_context, 0, sizeof(ENTROPY_CONTEXT) * txs_wide);
@@ -266,38 +136,61 @@ void av1_reset_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col,
   }
 }
 
-void av1_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y) {
+void av1_reset_loop_filter_delta(MACROBLOCKD *xd, int num_planes) {
+  xd->delta_lf_from_base = 0;
+  const int frame_lf_count =
+      num_planes > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2;
+  for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) xd->delta_lf[lf_id] = 0;
+}
+
+void av1_reset_loop_restoration(MACROBLOCKD *xd, const int num_planes) {
+  for (int p = 0; p < num_planes; ++p) {
+    set_default_wiener(xd->wiener_info + p);
+    set_default_sgrproj(xd->sgrproj_info + p);
+  }
+}
+
+void av1_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y,
+                            const int num_planes) {
   int i;
 
-  for (i = 0; i < MAX_MB_PLANE; i++) {
+  for (i = 0; i < num_planes; i++) {
     xd->plane[i].plane_type = get_plane_type(i);
     xd->plane[i].subsampling_x = i ? ss_x : 0;
     xd->plane[i].subsampling_y = i ? ss_y : 0;
   }
 }
 
-#if CONFIG_EXT_INTRA
 const int16_t dr_intra_derivative[90] = {
-  1,    14666, 7330, 4884, 3660, 2926, 2435, 2084, 1821, 1616, 1451, 1317, 1204,
-  1108, 1026,  955,  892,  837,  787,  743,  703,  666,  633,  603,  574,  548,
-  524,  502,   481,  461,  443,  426,  409,  394,  379,  365,  352,  339,  327,
-  316,  305,   294,  284,  274,  265,  256,  247,  238,  230,  222,  214,  207,
-  200,  192,   185,  179,  172,  166,  159,  153,  147,  141,  136,  130,  124,
-  119,  113,   108,  103,  98,   93,   88,   83,   78,   73,   68,   63,   59,
-  54,   49,    45,   40,   35,   31,   26,   22,   17,   13,   8,    4,
+  // More evenly spread out angles and limited to 10-bit
+  // Values that are 0 will never be used
+  //                    Approx angle
+  0,    0, 0,        //
+  1023, 0, 0,        // 3, ...
+  547,  0, 0,        // 6, ...
+  372,  0, 0, 0, 0,  // 9, ...
+  273,  0, 0,        // 14, ...
+  215,  0, 0,        // 17, ...
+  178,  0, 0,        // 20, ...
+  151,  0, 0,        // 23, ... (113 & 203 are base angles)
+  132,  0, 0,        // 26, ...
+  116,  0, 0,        // 29, ...
+  102,  0, 0, 0,     // 32, ...
+  90,   0, 0,        // 36, ...
+  80,   0, 0,        // 39, ...
+  71,   0, 0,        // 42, ...
+  64,   0, 0,        // 45, ... (45 & 135 are base angles)
+  57,   0, 0,        // 48, ...
+  51,   0, 0,        // 51, ...
+  45,   0, 0, 0,     // 54, ...
+  40,   0, 0,        // 58, ...
+  35,   0, 0,        // 61, ...
+  31,   0, 0,        // 64, ...
+  27,   0, 0,        // 67, ... (67 & 157 are base angles)
+  23,   0, 0,        // 70, ...
+  19,   0, 0,        // 73, ...
+  15,   0, 0, 0, 0,  // 76, ...
+  11,   0, 0,        // 81, ...
+  7,    0, 0,        // 84, ...
+  3,    0, 0,        // 87, ...
 };
-
-#if CONFIG_INTRA_INTERP
-int av1_is_intra_filter_switchable(int angle) {
-  assert(angle > 0 && angle < 270);
-  if (angle % 45 == 0) return 0;
-  if (angle > 90 && angle < 180) {
-    return 1;
-  } else {
-    return ((angle < 90 ? dr_intra_derivative[angle]
-                        : dr_intra_derivative[270 - angle]) &
-            0xFF) > 0;
-  }
-}
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA

third_party/aom/av1/common/cdef.c

@@ -13,7 +13,8 @@
 #include <math.h>
 #include <string.h>
 
-#include "./aom_scale_rtcd.h"
+#include "config/aom_scale_rtcd.h"
+
 #include "aom/aom_integer.h"
 #include "av1/common/cdef.h"
 #include "av1/common/cdef_block.h"
@@ -21,7 +22,6 @@
 #include "av1/common/reconinter.h"
 
 int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col) {
-  int r, c;
   int maxc, maxr;
   int skip = 1;
   maxc = cm->mi_cols - mi_col;
@@ -30,38 +30,40 @@ int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col) {
   maxr = AOMMIN(maxr, MI_SIZE_64X64);
   maxc = AOMMIN(maxc, MI_SIZE_64X64);
 
-  for (r = 0; r < maxr; r++) {
-    for (c = 0; c < maxc; c++) {
-      skip = skip &&
-             cm->mi_grid_visible[(mi_row + r) * cm->mi_stride + mi_col + c]
-                 ->mbmi.skip;
+  for (int r = 0; r < maxr; r++) {
+    for (int c = 0; c < maxc; c++) {
+      skip =
+          skip &&
+          cm->mi_grid_visible[(mi_row + r) * cm->mi_stride + mi_col + c]->skip;
     }
   }
   return skip;
 }
 
-static int is_8x8_block_skip(MODE_INFO **grid, int mi_row, int mi_col,
+static int is_8x8_block_skip(MB_MODE_INFO **grid, int mi_row, int mi_col,
                              int mi_stride) {
   int is_skip = 1;
   for (int r = 0; r < mi_size_high[BLOCK_8X8]; ++r)
     for (int c = 0; c < mi_size_wide[BLOCK_8X8]; ++c)
-      is_skip &= grid[(mi_row + r) * mi_stride + (mi_col + c)]->mbmi.skip;
+      is_skip &= grid[(mi_row + r) * mi_stride + (mi_col + c)]->skip;
 
   return is_skip;
 }
 
 int sb_compute_cdef_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
-                         cdef_list *dlist, int filter_skip) {
-  int r, c;
-  int maxc, maxr;
-  MODE_INFO **grid;
-  int count = 0;
-  grid = cm->mi_grid_visible;
-  maxc = cm->mi_cols - mi_col;
-  maxr = cm->mi_rows - mi_row;
+                         cdef_list *dlist, BLOCK_SIZE bs) {
+  MB_MODE_INFO **grid = cm->mi_grid_visible;
+  int maxc = cm->mi_cols - mi_col;
+  int maxr = cm->mi_rows - mi_row;
 
-  maxr = AOMMIN(maxr, MI_SIZE_64X64);
-  maxc = AOMMIN(maxc, MI_SIZE_64X64);
+  if (bs == BLOCK_128X128 || bs == BLOCK_128X64)
+    maxc = AOMMIN(maxc, MI_SIZE_128X128);
+  else
+    maxc = AOMMIN(maxc, MI_SIZE_64X64);
+  if (bs == BLOCK_128X128 || bs == BLOCK_64X128)
+    maxr = AOMMIN(maxr, MI_SIZE_128X128);
+  else
+    maxr = AOMMIN(maxr, MI_SIZE_64X64);
 
   const int r_step = mi_size_high[BLOCK_8X8];
   const int c_step = mi_size_wide[BLOCK_8X8];
@@ -71,36 +73,25 @@ int sb_compute_cdef_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
   assert(r_step == 1 || r_step == 2);
   assert(c_step == 1 || c_step == 2);
 
-  if (filter_skip) {
-    for (r = 0; r < maxr; r += r_step) {
-      for (c = 0; c < maxc; c += c_step) {
+  int count = 0;
+
+  for (int r = 0; r < maxr; r += r_step) {
+    for (int c = 0; c < maxc; c += c_step) {
+      if (!is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride)) {
         dlist[count].by = r >> r_shift;
         dlist[count].bx = c >> c_shift;
-        dlist[count].skip =
-            is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride);
+        dlist[count].skip = 0;
         count++;
       }
     }
-  } else {
-    for (r = 0; r < maxr; r += r_step) {
-      for (c = 0; c < maxc; c += c_step) {
-        if (!is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride)) {
-          dlist[count].by = r >> r_shift;
-          dlist[count].bx = c >> c_shift;
-          dlist[count].skip = 0;
-          count++;
-        }
-      }
-    }
   }
   return count;
 }
 
 void copy_rect8_8bit_to_16bit_c(uint16_t *dst, int dstride, const uint8_t *src,
                                 int sstride, int v, int h) {
-  int i, j;
-  for (i = 0; i < v; i++) {
-    for (j = 0; j < h; j++) {
+  for (int i = 0; i < v; i++) {
+    for (int j = 0; j < h; j++) {
       dst[i * dstride + j] = src[i * sstride + j];
     }
   }
@@ -109,36 +100,30 @@ void copy_rect8_8bit_to_16bit_c(uint16_t *dst, int dstride, const uint8_t *src,
 void copy_rect8_16bit_to_16bit_c(uint16_t *dst, int dstride,
                                  const uint16_t *src, int sstride, int v,
                                  int h) {
-  int i, j;
-  for (i = 0; i < v; i++) {
-    for (j = 0; j < h; j++) {
+  for (int i = 0; i < v; i++) {
+    for (int j = 0; j < h; j++) {
       dst[i * dstride + j] = src[i * sstride + j];
     }
   }
 }
 
-static void copy_sb8_16(UNUSED AV1_COMMON *cm, uint16_t *dst, int dstride,
+static void copy_sb8_16(AOM_UNUSED 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 CONFIG_HIGHBITDEPTH
   if (cm->use_highbitdepth) {
     const uint16_t *base =
         &CONVERT_TO_SHORTPTR(src)[src_voffset * sstride + src_hoffset];
     copy_rect8_16bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
   } else {
-#endif
     const uint8_t *base = &src[src_voffset * sstride + src_hoffset];
     copy_rect8_8bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
-#if CONFIG_HIGHBITDEPTH
   }
-#endif
 }
 
 static INLINE void fill_rect(uint16_t *dst, int dstride, int v, int h,
                              uint16_t x) {
-  int i, j;
-  for (i = 0; i < v; i++) {
-    for (j = 0; j < h; j++) {
+  for (int i = 0; i < v; i++) {
+    for (int j = 0; j < h; j++) {
       dst[i * dstride + j] = x;
     }
   }
@@ -146,9 +131,8 @@ static INLINE void fill_rect(uint16_t *dst, int dstride, int v, int h,
 
 static INLINE void copy_rect(uint16_t *dst, int dstride, const uint16_t *src,
                              int sstride, int v, int h) {
-  int i, j;
-  for (i = 0; i < v; i++) {
-    for (j = 0; j < h; j++) {
+  for (int i = 0; i < v; i++) {
+    for (int j = 0; j < h; j++) {
       dst[i * dstride + j] = src[i * sstride + j];
     }
   }
@@ -156,9 +140,8 @@ 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) {
-  int fbr, fbc;
-  int nhfb, nvfb;
-  uint16_t src[CDEF_INBUF_SIZE];
+  const int num_planes = av1_num_planes(cm);
+  DECLARE_ALIGNED(16, uint16_t, src[CDEF_INBUF_SIZE]);
   uint16_t *linebuf[3];
   uint16_t *colbuf[3];
   cdef_list dlist[MI_SIZE_64X64 * MI_SIZE_64X64];
@@ -166,48 +149,42 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
   int cdef_count;
   int dir[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
   int var[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
-  int stride;
   int mi_wide_l2[3];
   int mi_high_l2[3];
   int xdec[3];
   int ydec[3];
-  int pli;
-  int cdef_left;
   int coeff_shift = AOMMAX(cm->bit_depth - 8, 0);
-  int nplanes = MAX_MB_PLANE;
-  int chroma_cdef = xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
-                    xd->plane[2].subsampling_x == xd->plane[2].subsampling_y;
-  nvfb = (cm->mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
-  nhfb = (cm->mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
-  av1_setup_dst_planes(xd->plane, cm->sb_size, frame, 0, 0);
+  const int nvfb = (cm->mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
+  const int nhfb = (cm->mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
+  av1_setup_dst_planes(xd->plane, cm->seq_params.sb_size, frame, 0, 0, 0,
+                       num_planes);
   row_cdef = aom_malloc(sizeof(*row_cdef) * (nhfb + 2) * 2);
   memset(row_cdef, 1, sizeof(*row_cdef) * (nhfb + 2) * 2);
   prev_row_cdef = row_cdef + 1;
   curr_row_cdef = prev_row_cdef + nhfb + 2;
-  for (pli = 0; pli < nplanes; pli++) {
+  for (int pli = 0; pli < num_planes; pli++) {
     xdec[pli] = xd->plane[pli].subsampling_x;
     ydec[pli] = xd->plane[pli].subsampling_y;
     mi_wide_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_x;
     mi_high_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_y;
-    if (xdec[pli] != ydec[pli]) nplanes = 1;
   }
-  stride = (cm->mi_cols << MI_SIZE_LOG2) + 2 * CDEF_HBORDER;
-  for (pli = 0; pli < nplanes; pli++) {
+  const int stride = (cm->mi_cols << MI_SIZE_LOG2) + 2 * CDEF_HBORDER;
+  for (int pli = 0; pli < num_planes; pli++) {
     linebuf[pli] = aom_malloc(sizeof(*linebuf) * CDEF_VBORDER * stride);
     colbuf[pli] =
         aom_malloc(sizeof(*colbuf) *
                    ((CDEF_BLOCKSIZE << mi_high_l2[pli]) + 2 * CDEF_VBORDER) *
                    CDEF_HBORDER);
   }
-  for (fbr = 0; fbr < nvfb; fbr++) {
-    for (pli = 0; pli < nplanes; pli++) {
+  for (int fbr = 0; fbr < nvfb; fbr++) {
+    for (int pli = 0; pli < num_planes; pli++) {
       const int block_height =
           (MI_SIZE_64X64 << mi_high_l2[pli]) + 2 * CDEF_VBORDER;
       fill_rect(colbuf[pli], CDEF_HBORDER, block_height, CDEF_HBORDER,
                 CDEF_VERY_LARGE);
     }
-    cdef_left = 1;
-    for (fbc = 0; fbc < nhfb; fbc++) {
+    int cdef_left = 1;
+    for (int fbc = 0; fbc < nhfb; fbc++) {
       int level, sec_strength;
       int uv_level, uv_sec_strength;
       int nhb, nvb;
@@ -217,38 +194,43 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
                               MI_SIZE_64X64 * fbc] == NULL ||
           cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
                               MI_SIZE_64X64 * fbc]
-                  ->mbmi.cdef_strength == -1) {
+                  ->cdef_strength == -1) {
         cdef_left = 0;
         continue;
       }
       if (!cdef_left) cstart = -CDEF_HBORDER;
       nhb = AOMMIN(MI_SIZE_64X64, cm->mi_cols - MI_SIZE_64X64 * fbc);
       nvb = AOMMIN(MI_SIZE_64X64, cm->mi_rows - MI_SIZE_64X64 * fbr);
-      int tile_top, tile_left, tile_bottom, tile_right;
-      int mi_idx = MI_SIZE_64X64 * fbr * cm->mi_stride + MI_SIZE_64X64 * fbc;
-      MODE_INFO *const mi_tl = cm->mi + mi_idx;
-      BOUNDARY_TYPE boundary_tl = mi_tl->mbmi.boundary_info;
-      tile_top = boundary_tl & TILE_ABOVE_BOUNDARY;
-      tile_left = boundary_tl & TILE_LEFT_BOUNDARY;
+      int frame_top, frame_left, frame_bottom, frame_right;
 
-      if (fbr != nvfb - 1 &&
-          (&cm->mi[mi_idx + (MI_SIZE_64X64 - 1) * cm->mi_stride]))
-        tile_bottom = cm->mi[mi_idx + (MI_SIZE_64X64 - 1) * cm->mi_stride]
-                          .mbmi.boundary_info &
-                      TILE_BOTTOM_BOUNDARY;
-      else
-        tile_bottom = 1;
+      int mi_row = MI_SIZE_64X64 * fbr;
+      int mi_col = MI_SIZE_64X64 * fbc;
+      // for the current filter block, it's top left corner mi structure (mi_tl)
+      // is first accessed to check whether the top and left boundaries are
+      // frame boundaries. Then bottom-left and top-right mi structures are
+      // accessed to check whether the bottom and right boundaries
+      // (respectively) are frame boundaries.
+      //
+      // Note that we can't just check the bottom-right mi structure - eg. if
+      // we're at the right-hand edge of the frame but not the bottom, then
+      // the bottom-right mi is NULL but the bottom-left is not.
+      frame_top = (mi_row == 0) ? 1 : 0;
+      frame_left = (mi_col == 0) ? 1 : 0;
 
-      if (fbc != nhfb - 1 && (&cm->mi[mi_idx + MI_SIZE_64X64 - 1]))
-        tile_right = cm->mi[mi_idx + MI_SIZE_64X64 - 1].mbmi.boundary_info &
-                     TILE_RIGHT_BOUNDARY;
+      if (fbr != nvfb - 1)
+        frame_bottom = (mi_row + MI_SIZE_64X64 == cm->mi_rows) ? 1 : 0;
       else
-        tile_right = 1;
+        frame_bottom = 1;
+
+      if (fbc != nhfb - 1)
+        frame_right = (mi_col + MI_SIZE_64X64 == cm->mi_cols) ? 1 : 0;
+      else
+        frame_right = 1;
 
       const int mbmi_cdef_strength =
           cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
                               MI_SIZE_64X64 * fbc]
-              ->mbmi.cdef_strength;
+              ->cdef_strength;
       level = cm->cdef_strengths[mbmi_cdef_strength] / CDEF_SEC_STRENGTHS;
       sec_strength =
           cm->cdef_strengths[mbmi_cdef_strength] % CDEF_SEC_STRENGTHS;
@@ -259,23 +241,15 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
       uv_sec_strength += uv_sec_strength == 3;
       if ((level == 0 && sec_strength == 0 && uv_level == 0 &&
            uv_sec_strength == 0) ||
-          (cdef_count = sb_compute_cdef_list(
-               cm, fbr * MI_SIZE_64X64, fbc * MI_SIZE_64X64, dlist,
-#if CONFIG_CDEF_SINGLEPASS
-               (level & 1) || (uv_level & 1))) == 0)
-#else
-                 get_filter_skip(level) || get_filter_skip(uv_level))) == 0)
-#endif
-      {
+          (cdef_count = sb_compute_cdef_list(cm, fbr * MI_SIZE_64X64,
+                                             fbc * MI_SIZE_64X64, dlist,
+                                             BLOCK_64X64)) == 0) {
         cdef_left = 0;
         continue;
       }
 
       curr_row_cdef[fbc] = 1;
-      for (pli = 0; pli < nplanes; pli++) {
-#if !CONFIG_CDEF_SINGLEPASS
-        uint16_t dst[CDEF_BLOCKSIZE * CDEF_BLOCKSIZE];
-#endif
+      for (int pli = 0; pli < num_planes; pli++) {
         int coffset;
         int rend, cend;
         int pri_damping = cm->cdef_pri_damping;
@@ -284,10 +258,7 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
         int vsize = nvb << mi_high_l2[pli];
 
         if (pli) {
-          if (chroma_cdef)
-            level = uv_level;
-          else
-            level = 0;
+          level = uv_level;
           sec_strength = uv_sec_strength;
         }
 
@@ -375,81 +346,57 @@ void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
             (MI_SIZE_64X64 << mi_high_l2[pli]) * (fbr + 1) - CDEF_VBORDER,
             coffset, xd->plane[pli].dst.stride, CDEF_VBORDER, hsize);
 
-        if (tile_top) {
+        if (frame_top) {
           fill_rect(src, CDEF_BSTRIDE, CDEF_VBORDER, hsize + 2 * CDEF_HBORDER,
                     CDEF_VERY_LARGE);
         }
-        if (tile_left) {
+        if (frame_left) {
           fill_rect(src, CDEF_BSTRIDE, vsize + 2 * CDEF_VBORDER, CDEF_HBORDER,
                     CDEF_VERY_LARGE);
         }
-        if (tile_bottom) {
+        if (frame_bottom) {
           fill_rect(&src[(vsize + CDEF_VBORDER) * CDEF_BSTRIDE], CDEF_BSTRIDE,
                     CDEF_VBORDER, hsize + 2 * CDEF_HBORDER, CDEF_VERY_LARGE);
         }
-        if (tile_right) {
+        if (frame_right) {
           fill_rect(&src[hsize + CDEF_HBORDER], CDEF_BSTRIDE,
                     vsize + 2 * CDEF_VBORDER, CDEF_HBORDER, CDEF_VERY_LARGE);
         }
-#if CONFIG_HIGHBITDEPTH
+
         if (cm->use_highbitdepth) {
           cdef_filter_fb(
-#if CONFIG_CDEF_SINGLEPASS
               NULL,
-              &CONVERT_TO_SHORTPTR(xd->plane[pli].dst.buf)
-#else
-              (uint8_t *)&CONVERT_TO_SHORTPTR(xd->plane[pli].dst.buf)
-#endif
-                  [xd->plane[pli].dst.stride *
-                       (MI_SIZE_64X64 * fbr << mi_high_l2[pli]) +
-                   (fbc * MI_SIZE_64X64 << mi_wide_l2[pli])],
-#if CONFIG_CDEF_SINGLEPASS
+              &CONVERT_TO_SHORTPTR(
+                  xd->plane[pli]
+                      .dst.buf)[xd->plane[pli].dst.stride *
+                                    (MI_SIZE_64X64 * fbr << mi_high_l2[pli]) +
+                                (fbc * MI_SIZE_64X64 << mi_wide_l2[pli])],
               xd->plane[pli].dst.stride,
-#else
-              xd->plane[pli].dst.stride, dst,
-#endif
               &src[CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER], xdec[pli],
               ydec[pli], dir, NULL, var, pli, dlist, cdef_count, level,
-#if CONFIG_CDEF_SINGLEPASS
               sec_strength, pri_damping, sec_damping, coeff_shift);
-#else
-              sec_strength, sec_damping, pri_damping, coeff_shift, 0, 1);
-#endif
         } else {
-#endif
           cdef_filter_fb(
               &xd->plane[pli]
                    .dst.buf[xd->plane[pli].dst.stride *
                                 (MI_SIZE_64X64 * fbr << mi_high_l2[pli]) +
                             (fbc * MI_SIZE_64X64 << mi_wide_l2[pli])],
-#if CONFIG_CDEF_SINGLEPASS
               NULL, xd->plane[pli].dst.stride,
-#else
-              xd->plane[pli].dst.stride, dst,
-#endif
               &src[CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER], xdec[pli],
               ydec[pli], dir, NULL, var, pli, dlist, cdef_count, level,
-#if CONFIG_CDEF_SINGLEPASS
               sec_strength, pri_damping, sec_damping, coeff_shift);
-#else
-              sec_strength, sec_damping, pri_damping, coeff_shift, 0, 0);
-#endif
-
-#if CONFIG_HIGHBITDEPTH
         }
-#endif
       }
       cdef_left = 1;
     }
     {
-      unsigned char *tmp;
-      tmp = prev_row_cdef;
+      unsigned char *tmp = prev_row_cdef;
       prev_row_cdef = curr_row_cdef;
       curr_row_cdef = tmp;
     }
   }
   aom_free(row_cdef);
-  for (pli = 0; pli < nplanes; pli++) {
+  for (int pli = 0; pli < num_planes; pli++) {
     aom_free(linebuf[pli]);
     aom_free(colbuf[pli]);
   }

third_party/aom/av1/common/cdef.h

@@ -11,12 +11,13 @@
 #ifndef AV1_COMMON_CDEF_H_
 #define AV1_COMMON_CDEF_H_
 
-#define CDEF_STRENGTH_BITS 7
+#define CDEF_STRENGTH_BITS 6
 
-#define CDEF_PRI_STRENGTHS 32
+#define CDEF_PRI_STRENGTHS 16
 #define CDEF_SEC_STRENGTHS 4
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom/aom_integer.h"
 #include "aom_ports/mem.h"
 #include "av1/common/cdef_block.h"
@@ -38,7 +39,7 @@ extern "C" {
 
 int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col);
 int sb_compute_cdef_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
-                         cdef_list *dlist, int filter_skip);
+                         cdef_list *dlist, BLOCK_SIZE bsize);
 void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, MACROBLOCKD *xd);
 
 void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,

third_party/aom/av1/common/cdef_block.c

@@ -12,28 +12,13 @@
 #include <math.h>
 #include <stdlib.h>
 
-#ifdef HAVE_CONFIG_H
-#include "./config.h"
-#endif
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
 
-#include "./aom_dsp_rtcd.h"
-#include "./av1_rtcd.h"
-#include "./cdef.h"
+#include "av1/common/cdef.h"
 
 /* Generated from gen_filter_tables.c. */
-#if !CONFIG_CDEF_SINGLEPASS || CDEF_FULL
-const int cdef_directions[8][3] = {
-  { -1 * CDEF_BSTRIDE + 1, -2 * CDEF_BSTRIDE + 2, -3 * CDEF_BSTRIDE + 3 },
-  { 0 * CDEF_BSTRIDE + 1, -1 * CDEF_BSTRIDE + 2, -1 * CDEF_BSTRIDE + 3 },
-  { 0 * CDEF_BSTRIDE + 1, 0 * CDEF_BSTRIDE + 2, 0 * CDEF_BSTRIDE + 3 },
-  { 0 * CDEF_BSTRIDE + 1, 1 * CDEF_BSTRIDE + 2, 1 * CDEF_BSTRIDE + 3 },
-  { 1 * CDEF_BSTRIDE + 1, 2 * CDEF_BSTRIDE + 2, 3 * CDEF_BSTRIDE + 3 },
-  { 1 * CDEF_BSTRIDE + 0, 2 * CDEF_BSTRIDE + 1, 3 * CDEF_BSTRIDE + 1 },
-  { 1 * CDEF_BSTRIDE + 0, 2 * CDEF_BSTRIDE + 0, 3 * CDEF_BSTRIDE + 0 },
-  { 1 * CDEF_BSTRIDE + 0, 2 * CDEF_BSTRIDE - 1, 3 * CDEF_BSTRIDE - 1 }
-};
-#else
-const int cdef_directions[8][2] = {
+DECLARE_ALIGNED(16, const int, cdef_directions[8][2]) = {
   { -1 * CDEF_BSTRIDE + 1, -2 * CDEF_BSTRIDE + 2 },
   { 0 * CDEF_BSTRIDE + 1, -1 * CDEF_BSTRIDE + 2 },
   { 0 * CDEF_BSTRIDE + 1, 0 * CDEF_BSTRIDE + 2 },
@@ -43,7 +28,6 @@ const int cdef_directions[8][2] = {
   { 1 * CDEF_BSTRIDE + 0, 2 * CDEF_BSTRIDE + 0 },
   { 1 * CDEF_BSTRIDE + 0, 2 * CDEF_BSTRIDE - 1 }
 };
-#endif
 
 /* Detect direction. 0 means 45-degree up-right, 2 is horizontal, and so on.
    The search minimizes the weighted variance along all the lines in a
@@ -123,65 +107,38 @@ int cdef_find_dir_c(const uint16_t *img, int stride, int32_t *var,
   return best_dir;
 }
 
-#if CONFIG_CDEF_SINGLEPASS
-#if CDEF_FULL
-const int cdef_pri_taps[2][3] = { { 3, 2, 1 }, { 2, 2, 2 } };
-const int cdef_sec_taps[2][2] = { { 3, 1 }, { 3, 1 } };
-#else
 const int cdef_pri_taps[2][2] = { { 4, 2 }, { 3, 3 } };
 const int cdef_sec_taps[2][2] = { { 2, 1 }, { 2, 1 } };
-#endif
 
 /* Smooth in the direction detected. */
-#if CDEF_CAP
 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,
-                         UNUSED int max_unused)
-#else
-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)
-#endif
-{
+                         AOM_UNUSED int max_unused, int coeff_shift) {
   int i, j, k;
   const int s = CDEF_BSTRIDE;
-  const int *pri_taps = cdef_pri_taps[pri_strength & 1];
-  const int *sec_taps = cdef_sec_taps[pri_strength & 1];
-  for (i = 0; i < 4 << (bsize == BLOCK_8X8); i++) {
-    for (j = 0; j < 4 << (bsize == BLOCK_8X8); j++) {
+  const int *pri_taps = cdef_pri_taps[(pri_strength >> coeff_shift) & 1];
+  const int *sec_taps = cdef_sec_taps[(pri_strength >> coeff_shift) & 1];
+  for (i = 0; i < 4 << (bsize == BLOCK_8X8 || bsize == BLOCK_4X8); i++) {
+    for (j = 0; j < 4 << (bsize == BLOCK_8X8 || bsize == BLOCK_8X4); j++) {
       int16_t sum = 0;
       int16_t y;
       int16_t x = in[i * s + j];
-#if CDEF_CAP
       int max = x;
       int min = x;
-#endif
-#if CDEF_FULL
-      for (k = 0; k < 3; k++)
-#else
-      for (k = 0; k < 2; k++)
-#endif
-      {
+      for (k = 0; k < 2; k++) {
         int16_t p0 = in[i * s + j + cdef_directions[dir][k]];
         int16_t p1 = in[i * s + j - cdef_directions[dir][k]];
         sum += pri_taps[k] * constrain(p0 - x, pri_strength, pri_damping);
         sum += pri_taps[k] * constrain(p1 - x, pri_strength, pri_damping);
-#if CDEF_CAP
         if (p0 != CDEF_VERY_LARGE) max = AOMMAX(p0, max);
         if (p1 != CDEF_VERY_LARGE) max = AOMMAX(p1, max);
         min = AOMMIN(p0, min);
         min = AOMMIN(p1, min);
-#endif
-#if CDEF_FULL
-        if (k == 2) continue;
-#endif
         int16_t s0 = in[i * s + j + cdef_directions[(dir + 2) & 7][k]];
         int16_t s1 = in[i * s + j - cdef_directions[(dir + 2) & 7][k]];
         int16_t s2 = in[i * s + j + cdef_directions[(dir + 6) & 7][k]];
         int16_t s3 = in[i * s + j - cdef_directions[(dir + 6) & 7][k]];
-#if CDEF_CAP
         if (s0 != CDEF_VERY_LARGE) max = AOMMAX(s0, max);
         if (s1 != CDEF_VERY_LARGE) max = AOMMAX(s1, max);
         if (s2 != CDEF_VERY_LARGE) max = AOMMAX(s2, max);
@@ -190,17 +147,12 @@ void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride,
         min = AOMMIN(s1, min);
         min = AOMMIN(s2, min);
         min = AOMMIN(s3, min);
-#endif
         sum += sec_taps[k] * constrain(s0 - x, sec_strength, sec_damping);
         sum += sec_taps[k] * constrain(s1 - x, sec_strength, sec_damping);
         sum += sec_taps[k] * constrain(s2 - x, sec_strength, sec_damping);
         sum += sec_taps[k] * constrain(s3 - x, sec_strength, sec_damping);
       }
-#if CDEF_CAP
       y = clamp((int16_t)x + ((8 + sum - (sum < 0)) >> 4), min, max);
-#else
-      y = clamp((int16_t)x + ((8 + sum - (sum < 0)) >> 4), 0, max);
-#endif
       if (dst8)
         dst8[i * dstride + j] = (uint8_t)y;
       else
@@ -209,67 +161,6 @@ void cdef_filter_block_c(uint8_t *dst8, uint16_t *dst16, int dstride,
   }
 }
 
-#else
-
-/* Smooth in the direction detected. */
-void cdef_direction_8x8_c(uint16_t *y, int ystride, const uint16_t *in,
-                          int threshold, int dir, int damping) {
-  int i;
-  int j;
-  int k;
-  static const int taps[3] = { 3, 2, 1 };
-  for (i = 0; i < 8; i++) {
-    for (j = 0; j < 8; j++) {
-      int16_t sum;
-      int16_t xx;
-      int16_t yy;
-      xx = in[i * CDEF_BSTRIDE + j];
-      sum = 0;
-      for (k = 0; k < 3; k++) {
-        int16_t p0;
-        int16_t p1;
-        p0 = in[i * CDEF_BSTRIDE + j + cdef_directions[dir][k]] - xx;
-        p1 = in[i * CDEF_BSTRIDE + j - cdef_directions[dir][k]] - xx;
-        sum += taps[k] * constrain(p0, threshold, damping);
-        sum += taps[k] * constrain(p1, threshold, damping);
-      }
-      sum = (sum + 8) >> 4;
-      yy = xx + sum;
-      y[i * ystride + j] = yy;
-    }
-  }
-}
-
-/* Smooth in the direction detected. */
-void cdef_direction_4x4_c(uint16_t *y, int ystride, const uint16_t *in,
-                          int threshold, int dir, int damping) {
-  int i;
-  int j;
-  int k;
-  static const int taps[2] = { 4, 1 };
-  for (i = 0; i < 4; i++) {
-    for (j = 0; j < 4; j++) {
-      int16_t sum;
-      int16_t xx;
-      int16_t yy;
-      xx = in[i * CDEF_BSTRIDE + j];
-      sum = 0;
-      for (k = 0; k < 2; k++) {
-        int16_t p0;
-        int16_t p1;
-        p0 = in[i * CDEF_BSTRIDE + j + cdef_directions[dir][k]] - xx;
-        p1 = in[i * CDEF_BSTRIDE + j - cdef_directions[dir][k]] - xx;
-        sum += taps[k] * constrain(p0, threshold, damping);
-        sum += taps[k] * constrain(p1, threshold, damping);
-      }
-      sum = (sum + 8) >> 4;
-      yy = xx + sum;
-      y[i * ystride + j] = yy;
-    }
-  }
-}
-#endif
-
 /* Compute the primary filter strength for an 8x8 block based on the
    directional variance difference. A high variance difference means
    that we have a highly directional pattern (e.g. a high contrast
@@ -282,172 +173,26 @@ static INLINE int adjust_strength(int strength, int32_t var) {
   return var ? (strength * (4 + i) + 8) >> 4 : 0;
 }
 
-#if !CONFIG_CDEF_SINGLEPASS
-void copy_8x8_16bit_to_16bit_c(uint16_t *dst, int dstride, const uint16_t *src,
-                               int sstride) {
-  int i, j;
-  for (i = 0; i < 8; i++)
-    for (j = 0; j < 8; j++) dst[i * dstride + j] = src[i * sstride + j];
-}
-
-void copy_4x4_16bit_to_16bit_c(uint16_t *dst, int dstride, const uint16_t *src,
-                               int sstride) {
-  int i, j;
-  for (i = 0; i < 4; i++)
-    for (j = 0; j < 4; j++) dst[i * dstride + j] = src[i * sstride + j];
-}
-
-static void copy_block_16bit_to_16bit(uint16_t *dst, int dstride, uint16_t *src,
-                                      cdef_list *dlist, int cdef_count,
-                                      int bsize) {
-  int bi, bx, by;
-
-  if (bsize == BLOCK_8X8) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_8x8_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
-                              &src[bi << (3 + 3)], 8);
-    }
-  } else if (bsize == BLOCK_4X8) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 2)], dstride,
-                              &src[bi << (3 + 2)], 4);
-      copy_4x4_16bit_to_16bit(&dst[((by << 3) + 4) * dstride + (bx << 2)],
-                              dstride, &src[(bi << (3 + 2)) + 4 * 4], 4);
-    }
-  } else if (bsize == BLOCK_8X4) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 3)], dstride,
-                              &src[bi << (2 + 3)], 8);
-      copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 3) + 4],
-                              dstride, &src[(bi << (2 + 3)) + 4], 8);
-    }
-  } else {
-    assert(bsize == BLOCK_4X4);
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
-                              &src[bi << (2 + 2)], 4);
-    }
-  }
-}
-
-void copy_8x8_16bit_to_8bit_c(uint8_t *dst, int dstride, const uint16_t *src,
-                              int sstride) {
-  int i, j;
-  for (i = 0; i < 8; i++)
-    for (j = 0; j < 8; j++)
-      dst[i * dstride + j] = (uint8_t)src[i * sstride + j];
-}
-
-void copy_4x4_16bit_to_8bit_c(uint8_t *dst, int dstride, const uint16_t *src,
-                              int sstride) {
-  int i, j;
-  for (i = 0; i < 4; i++)
-    for (j = 0; j < 4; j++)
-      dst[i * dstride + j] = (uint8_t)src[i * sstride + j];
-}
-
-static void copy_block_16bit_to_8bit(uint8_t *dst, int dstride,
-                                     const uint16_t *src, cdef_list *dlist,
-                                     int cdef_count, int bsize) {
-  int bi, bx, by;
-  if (bsize == BLOCK_8X8) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_8x8_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
-                             &src[bi << (3 + 3)], 8);
-    }
-  } else if (bsize == BLOCK_4X8) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 2)], dstride,
-                             &src[bi << (3 + 2)], 4);
-      copy_4x4_16bit_to_8bit(&dst[((by << 3) + 4) * dstride + (bx << 2)],
-                             dstride, &src[(bi << (3 + 2)) + 4 * 4], 4);
-    }
-  } else if (bsize == BLOCK_8X4) {
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 3)], dstride,
-                             &src[bi << (2 + 3)], 8);
-      copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 3) + 4], dstride,
-                             &src[(bi << (2 + 3)) + 4], 8);
-    }
-  } else {
-    assert(bsize == BLOCK_4X4);
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
-                             &src[bi << (2 * 2)], 4);
-    }
-  }
-}
-
-int get_filter_skip(int level) {
-  int filter_skip = level & 1;
-  if (level == 1) filter_skip = 0;
-  return filter_skip;
-}
-
-void cdef_filter_fb(uint8_t *dst, int dstride, uint16_t *y, uint16_t *in,
-                    int xdec, int ydec, int dir[CDEF_NBLOCKS][CDEF_NBLOCKS],
-                    int *dirinit, int var[CDEF_NBLOCKS][CDEF_NBLOCKS], int pli,
-                    cdef_list *dlist, int cdef_count, int level,
-                    int sec_strength, int sec_damping, int pri_damping,
-                    int coeff_shift, int skip_dering, int hbd) {
-#else
-
 void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
                     int xdec, int ydec, int dir[CDEF_NBLOCKS][CDEF_NBLOCKS],
                     int *dirinit, int var[CDEF_NBLOCKS][CDEF_NBLOCKS], int pli,
                     cdef_list *dlist, int cdef_count, int level,
                     int sec_strength, int pri_damping, int sec_damping,
                     int coeff_shift) {
-#endif
   int bi;
   int bx;
   int by;
   int bsize, bsizex, bsizey;
 
-#if CONFIG_CDEF_SINGLEPASS
-  int pri_strength = (level >> 1) << coeff_shift;
-  int filter_skip = level & 1;
-  if (!pri_strength && !sec_strength && filter_skip) {
-    pri_strength = 19 << coeff_shift;
-    sec_strength = 7 << coeff_shift;
-  }
-#else
-  int threshold = (level >> 1) << coeff_shift;
-  int filter_skip = get_filter_skip(level);
-  if (level == 1) threshold = 31 << coeff_shift;
-
-  cdef_direction_func cdef_direction[] = { cdef_direction_4x4,
-                                           cdef_direction_8x8 };
-#endif
+  int pri_strength = level << coeff_shift;
+  sec_strength <<= coeff_shift;
   sec_damping += coeff_shift - (pli != AOM_PLANE_Y);
   pri_damping += coeff_shift - (pli != AOM_PLANE_Y);
   bsize =
       ydec ? (xdec ? BLOCK_4X4 : BLOCK_8X4) : (xdec ? BLOCK_4X8 : BLOCK_8X8);
   bsizex = 3 - xdec;
   bsizey = 3 - ydec;
-#if CONFIG_CDEF_SINGLEPASS
-  if (dirinit && pri_strength == 0 && sec_strength == 0)
-#else
-  if (!skip_dering)
-#endif
-  {
-#if CONFIG_CDEF_SINGLEPASS
+  if (dirinit && pri_strength == 0 && sec_strength == 0) {
     // If we're here, both primary and secondary strengths are 0, and
     // we still haven't written anything to y[] yet, so we just copy
     // the input to y[]. This is necessary only for av1_cdef_search()
@@ -455,97 +200,16 @@ void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
     for (bi = 0; bi < cdef_count; bi++) {
       by = dlist[bi].by;
       bx = dlist[bi].bx;
-#else
-    if (pli == 0) {
-      if (!dirinit || !*dirinit) {
-        for (bi = 0; bi < cdef_count; bi++) {
-          by = dlist[bi].by;
-          bx = dlist[bi].bx;
-          dir[by][bx] = cdef_find_dir(&in[8 * by * CDEF_BSTRIDE + 8 * bx],
-                                      CDEF_BSTRIDE, &var[by][bx], coeff_shift);
-        }
-        if (dirinit) *dirinit = 1;
-      }
-    }
-    // Only run dering for non-zero threshold (which is always the case for
-    // 4:2:2 or 4:4:0). If we don't dering, we still need to eventually write
-    // something out in y[] later.
-    if (threshold != 0) {
-      assert(bsize == BLOCK_8X8 || bsize == BLOCK_4X4);
-      for (bi = 0; bi < cdef_count; bi++) {
-        int t = !filter_skip && dlist[bi].skip ? 0 : threshold;
-        by = dlist[bi].by;
-        bx = dlist[bi].bx;
-        (cdef_direction[bsize == BLOCK_8X8])(
-            &y[bi << (bsizex + bsizey)], 1 << bsizex,
-            &in[(by * CDEF_BSTRIDE << bsizey) + (bx << bsizex)],
-            pli ? t : adjust_strength(t, var[by][bx]), dir[by][bx],
-            pri_damping);
-      }
-    }
-  }
-
-  if (sec_strength) {
-    if (threshold && !skip_dering)
-      copy_block_16bit_to_16bit(in, CDEF_BSTRIDE, y, dlist, cdef_count, bsize);
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-      int py = by << bsizey;
-      int px = bx << bsizex;
-
-      if (!filter_skip && dlist[bi].skip) continue;
-      if (!dst || hbd) {
-        // 16 bit destination if high bitdepth or 8 bit destination not given
-        (!threshold || (dir[by][bx] < 4 && dir[by][bx]) ? aom_clpf_block_hbd
-                                                        : aom_clpf_hblock_hbd)(
-            dst ? (uint16_t *)dst + py * dstride + px
-                : &y[bi << (bsizex + bsizey)],
-            in + py * CDEF_BSTRIDE + px, dst && hbd ? dstride : 1 << bsizex,
-            CDEF_BSTRIDE, 1 << bsizex, 1 << bsizey, sec_strength << coeff_shift,
-            sec_damping);
-      } else {
-        // Do clpf and write the result to an 8 bit destination
-        (!threshold || (dir[by][bx] < 4 && dir[by][bx]) ? aom_clpf_block
-                                                        : aom_clpf_hblock)(
-            dst + py * dstride + px, in + py * CDEF_BSTRIDE + px, dstride,
-            CDEF_BSTRIDE, 1 << bsizex, 1 << bsizey, sec_strength << coeff_shift,
-            sec_damping);
-      }
-    }
-  } else if (threshold != 0) {
-    // No clpf, so copy instead
-    if (hbd) {
-      copy_block_16bit_to_16bit((uint16_t *)dst, dstride, y, dlist, cdef_count,
-                                bsize);
-    } else {
-      copy_block_16bit_to_8bit(dst, dstride, y, dlist, cdef_count, bsize);
-    }
-  } else if (dirinit) {
-    // If we're here, both dering and clpf are off, and we still haven't written
-    // anything to y[] yet, so we just copy the input to y[]. This is necessary
-    // only for av1_cdef_search() and only av1_cdef_search() sets dirinit.
-    for (bi = 0; bi < cdef_count; bi++) {
-      by = dlist[bi].by;
-      bx = dlist[bi].bx;
-#endif
       int iy, ix;
       // TODO(stemidts/jmvalin): SIMD optimisations
       for (iy = 0; iy < 1 << bsizey; iy++)
         for (ix = 0; ix < 1 << bsizex; ix++)
-#if CONFIG_CDEF_SINGLEPASS
           dst16[(bi << (bsizex + bsizey)) + (iy << bsizex) + ix] =
-#else
-          y[(bi << (bsizex + bsizey)) + (iy << bsizex) + ix] =
-#endif
               in[((by << bsizey) + iy) * CDEF_BSTRIDE + (bx << bsizex) + ix];
     }
-#if CONFIG_CDEF_SINGLEPASS
     return;
-#endif
   }
 
-#if CONFIG_CDEF_SINGLEPASS
   if (pli == 0) {
     if (!dirinit || !*dirinit) {
       for (bi = 0; bi < cdef_count; bi++) {
@@ -557,19 +221,28 @@ void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
       if (dirinit) *dirinit = 1;
     }
   }
+  if (pli == 1 && xdec != ydec) {
+    for (bi = 0; bi < cdef_count; bi++) {
+      static const int conv422[8] = { 7, 0, 2, 4, 5, 6, 6, 6 };
+      static const int conv440[8] = { 1, 2, 2, 2, 3, 4, 6, 0 };
+      by = dlist[bi].by;
+      bx = dlist[bi].bx;
+      dir[by][bx] = (xdec ? conv422 : conv440)[dir[by][bx]];
+    }
+  }
 
-  assert(bsize == BLOCK_8X8 || bsize == BLOCK_4X4);
   for (bi = 0; bi < cdef_count; bi++) {
-    int t = !filter_skip && dlist[bi].skip ? 0 : pri_strength;
-    int s = !filter_skip && dlist[bi].skip ? 0 : sec_strength;
+    int t = dlist[bi].skip ? 0 : pri_strength;
+    int s = dlist[bi].skip ? 0 : sec_strength;
     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);
+      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);
     else
       cdef_filter_block(
           NULL,
@@ -578,7 +251,7 @@ 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);
+          pri_damping, sec_damping, bsize, (256 << coeff_shift) - 1,
+          coeff_shift);
   }
-#endif
 }

third_party/aom/av1/common/cdef_block.h

@@ -12,43 +12,28 @@
 #if !defined(_CDEF_BLOCK_H)
 #define _CDEF_BLOCK_H (1)
 
-#include "./odintrin.h"
+#include "av1/common/odintrin.h"
 
 #define CDEF_BLOCKSIZE 64
 #define CDEF_BLOCKSIZE_LOG2 6
-#define CDEF_NBLOCKS (CDEF_BLOCKSIZE / 8)
-#if CONFIG_CDEF_SINGLEPASS
+#define CDEF_NBLOCKS ((1 << MAX_SB_SIZE_LOG2) / 8)
 #define CDEF_SB_SHIFT (MAX_SB_SIZE_LOG2 - CDEF_BLOCKSIZE_LOG2)
-#endif
 
 /* We need to buffer three vertical lines. */
 #define CDEF_VBORDER (3)
 /* We only need to buffer three horizontal pixels too, but let's align to
    16 bytes (8 x 16 bits) to make vectorization easier. */
 #define CDEF_HBORDER (8)
-#define CDEF_BSTRIDE ALIGN_POWER_OF_TWO(CDEF_BLOCKSIZE + 2 * CDEF_HBORDER, 3)
+#define CDEF_BSTRIDE \
+  ALIGN_POWER_OF_TWO((1 << MAX_SB_SIZE_LOG2) + 2 * CDEF_HBORDER, 3)
 
 #define CDEF_VERY_LARGE (30000)
-#define CDEF_INBUF_SIZE (CDEF_BSTRIDE * (CDEF_BLOCKSIZE + 2 * CDEF_VBORDER))
+#define CDEF_INBUF_SIZE \
+  (CDEF_BSTRIDE * ((1 << MAX_SB_SIZE_LOG2) + 2 * CDEF_VBORDER))
 
-#if CONFIG_CDEF_SINGLEPASS
-// Filter configuration
-#define CDEF_CAP 1   // 1 = Cap change to largest diff
-#define CDEF_FULL 0  // 1 = 7x7 filter, 0 = 5x5 filter
-
-#if CDEF_FULL
-extern const int cdef_pri_taps[2][3];
-extern const int cdef_sec_taps[2][2];
-extern const int cdef_directions[8][3];
-#else
 extern const int cdef_pri_taps[2][2];
 extern const int cdef_sec_taps[2][2];
-extern const int cdef_directions[8][2];
-#endif
-
-#else  // CONFIG_CDEF_SINGLEPASS
-extern const int cdef_directions[8][3];
-#endif
+DECLARE_ALIGNED(16, extern const int, cdef_directions[8][2]);
 
 typedef struct {
   uint8_t by;
@@ -56,35 +41,19 @@ typedef struct {
   uint8_t skip;
 } cdef_list;
 
-#if CONFIG_CDEF_SINGLEPASS
 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 max,
+                                       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);
-#else
-typedef void (*cdef_direction_func)(uint16_t *y, int ystride,
-                                    const uint16_t *in, int threshold, int dir,
-                                    int damping);
 
-int get_filter_skip(int level);
-#endif
-
-#if CONFIG_CDEF_SINGLEPASS
 void cdef_filter_fb(uint8_t *dst8, uint16_t *dst16, int dstride, uint16_t *in,
                     int xdec, int ydec, int dir[CDEF_NBLOCKS][CDEF_NBLOCKS],
                     int *dirinit, int var[CDEF_NBLOCKS][CDEF_NBLOCKS], int pli,
                     cdef_list *dlist, int cdef_count, int level,
                     int sec_strength, int pri_damping, int sec_damping,
                     int coeff_shift);
-#else
-void cdef_filter_fb(uint8_t *dst, int dstride, uint16_t *y, uint16_t *in,
-                    int xdec, int ydec, int dir[CDEF_NBLOCKS][CDEF_NBLOCKS],
-                    int *dirinit, int var[CDEF_NBLOCKS][CDEF_NBLOCKS], int pli,
-                    cdef_list *dlist, int cdef_count, int level,
-                    int sec_strength, int sec_damping, int pri_damping,
-                    int coeff_shift, int skip_dering, int hbd);
-#endif
 #endif

third_party/aom/av1/common/cdef_block_avx2.c

@@ -11,4 +11,4 @@
 
 #include "aom_dsp/aom_simd.h"
 #define SIMD_FUNC(name) name##_avx2
-#include "./cdef_block_simd.h"
+#include "av1/common/cdef_block_simd.h"

third_party/aom/av1/common/cdef_block_neon.c

@@ -11,4 +11,4 @@
 
 #include "aom_dsp/aom_simd.h"
 #define SIMD_FUNC(name) name##_neon
-#include "./cdef_block_simd.h"
+#include "av1/common/cdef_block_simd.h"

third_party/aom/av1/common/cdef_block_sse2.c

@@ -11,4 +11,4 @@
 
 #include "aom_dsp/aom_simd.h"
 #define SIMD_FUNC(name) name##_sse2
-#include "./cdef_block_simd.h"
+#include "av1/common/cdef_block_simd.h"

third_party/aom/av1/common/cdef_block_sse4.c

@@ -11,4 +11,4 @@
 
 #include "aom_dsp/aom_simd.h"
 #define SIMD_FUNC(name) name##_sse4_1
-#include "./cdef_block_simd.h"
+#include "av1/common/cdef_block_simd.h"

third_party/aom/av1/common/cdef_block_ssse3.c

@@ -11,4 +11,4 @@
 
 #include "aom_dsp/aom_simd.h"
 #define SIMD_FUNC(name) name##_ssse3
-#include "./cdef_block_simd.h"
+#include "av1/common/cdef_block_simd.h"

third_party/aom/av1/common/cfl.c

@@ -13,20 +13,77 @@
 #include "av1/common/common_data.h"
 #include "av1/common/onyxc_int.h"
 
+#include "config/av1_rtcd.h"
+
 void cfl_init(CFL_CTX *cfl, AV1_COMMON *cm) {
-  if (!((cm->subsampling_x == 0 && cm->subsampling_y == 0) ||
-        (cm->subsampling_x == 1 && cm->subsampling_y == 1))) {
+  assert(block_size_wide[CFL_MAX_BLOCK_SIZE] == CFL_BUF_LINE);
+  assert(block_size_high[CFL_MAX_BLOCK_SIZE] == CFL_BUF_LINE);
+  if (!(cm->subsampling_x == 0 && cm->subsampling_y == 0) &&
+      !(cm->subsampling_x == 1 && cm->subsampling_y == 1) &&
+      !(cm->subsampling_x == 1 && cm->subsampling_y == 0)) {
     aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
-                       "Only 4:4:4 and 4:2:0 are currently supported by CfL");
+                       "Only 4:4:4, 4:2:2 and 4:2:0 are currently supported by "
+                       "CfL, %d %d subsampling is not supported.\n",
+                       cm->subsampling_x, cm->subsampling_y);
   }
-  memset(&cfl->pred_buf_q3, 0, sizeof(cfl->pred_buf_q3));
+  memset(&cfl->recon_buf_q3, 0, sizeof(cfl->recon_buf_q3));
+  memset(&cfl->ac_buf_q3, 0, sizeof(cfl->ac_buf_q3));
   cfl->subsampling_x = cm->subsampling_x;
   cfl->subsampling_y = cm->subsampling_y;
   cfl->are_parameters_computed = 0;
   cfl->store_y = 0;
-#if CONFIG_CHROMA_SUB8X8 && CONFIG_DEBUG
-  cfl_clear_sub8x8_val(cfl);
-#endif  // CONFIG_CHROMA_SUB8X8 && CONFIG_DEBUG
+  // The DC_PRED cache is disabled by default and is only enabled in
+  // cfl_rd_pick_alpha
+  cfl->use_dc_pred_cache = 0;
+  cfl->dc_pred_is_cached[CFL_PRED_U] = 0;
+  cfl->dc_pred_is_cached[CFL_PRED_V] = 0;
+}
+
+void cfl_store_dc_pred(MACROBLOCKD *const xd, const uint8_t *input,
+                       CFL_PRED_TYPE pred_plane, int width) {
+  assert(pred_plane < CFL_PRED_PLANES);
+  assert(width <= CFL_BUF_LINE);
+
+  if (get_bitdepth_data_path_index(xd)) {
+    uint16_t *const input_16 = CONVERT_TO_SHORTPTR(input);
+    memcpy(xd->cfl.dc_pred_cache[pred_plane], input_16, width << 1);
+    return;
+  }
+
+  memcpy(xd->cfl.dc_pred_cache[pred_plane], input, width);
+}
+
+static void cfl_load_dc_pred_lbd(const int16_t *dc_pred_cache, uint8_t *dst,
+                                 int dst_stride, int width, int height) {
+  for (int j = 0; j < height; j++) {
+    memcpy(dst, dc_pred_cache, width);
+    dst += dst_stride;
+  }
+}
+
+static void cfl_load_dc_pred_hbd(const int16_t *dc_pred_cache, uint16_t *dst,
+                                 int dst_stride, int width, int height) {
+  const size_t num_bytes = width << 1;
+  for (int j = 0; j < height; j++) {
+    memcpy(dst, dc_pred_cache, num_bytes);
+    dst += dst_stride;
+  }
+}
+void cfl_load_dc_pred(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
+                      TX_SIZE tx_size, CFL_PRED_TYPE pred_plane) {
+  const int width = tx_size_wide[tx_size];
+  const int height = tx_size_high[tx_size];
+  assert(pred_plane < CFL_PRED_PLANES);
+  assert(width <= CFL_BUF_LINE);
+  assert(height <= CFL_BUF_LINE);
+  if (get_bitdepth_data_path_index(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);
+    return;
+  }
+  cfl_load_dc_pred_lbd(xd->cfl.dc_pred_cache[pred_plane], dst, dst_stride,
+                       width, height);
 }
 
 // Due to frame boundary issues, it is possible that the total area covered by
@@ -38,217 +95,54 @@ static INLINE void cfl_pad(CFL_CTX *cfl, int width, int height) {
 
   if (diff_width > 0) {
     const int min_height = height - diff_height;
-    int16_t *pred_buf_q3 = cfl->pred_buf_q3 + (width - diff_width);
+    uint16_t *recon_buf_q3 = cfl->recon_buf_q3 + (width - diff_width);
     for (int j = 0; j < min_height; j++) {
-      const int last_pixel = pred_buf_q3[-1];
+      const uint16_t last_pixel = recon_buf_q3[-1];
+      assert(recon_buf_q3 + diff_width <= cfl->recon_buf_q3 + CFL_BUF_SQUARE);
       for (int i = 0; i < diff_width; i++) {
-        pred_buf_q3[i] = last_pixel;
+        recon_buf_q3[i] = last_pixel;
       }
-      pred_buf_q3 += MAX_SB_SIZE;
+      recon_buf_q3 += CFL_BUF_LINE;
     }
     cfl->buf_width = width;
   }
   if (diff_height > 0) {
-    int16_t *pred_buf_q3 =
-        cfl->pred_buf_q3 + ((height - diff_height) * MAX_SB_SIZE);
+    uint16_t *recon_buf_q3 =
+        cfl->recon_buf_q3 + ((height - diff_height) * CFL_BUF_LINE);
     for (int j = 0; j < diff_height; j++) {
-      const int16_t *last_row_q3 = pred_buf_q3 - MAX_SB_SIZE;
+      const uint16_t *last_row_q3 = recon_buf_q3 - CFL_BUF_LINE;
+      assert(recon_buf_q3 + width <= cfl->recon_buf_q3 + CFL_BUF_SQUARE);
       for (int i = 0; i < width; i++) {
-        pred_buf_q3[i] = last_row_q3[i];
+        recon_buf_q3[i] = last_row_q3[i];
       }
-      pred_buf_q3 += MAX_SB_SIZE;
+      recon_buf_q3 += CFL_BUF_LINE;
     }
     cfl->buf_height = height;
   }
 }
 
-static void sum_above_row_lbd(const uint8_t *above_u, const uint8_t *above_v,
-                              int width, int *out_sum_u, int *out_sum_v) {
-  int sum_u = 0;
-  int sum_v = 0;
-  for (int i = 0; i < width; i++) {
-    sum_u += above_u[i];
-    sum_v += above_v[i];
-  }
-  *out_sum_u += sum_u;
-  *out_sum_v += sum_v;
-}
-#if CONFIG_HIGHBITDEPTH
-static void sum_above_row_hbd(const uint16_t *above_u, const uint16_t *above_v,
-                              int width, int *out_sum_u, int *out_sum_v) {
-  int sum_u = 0;
-  int sum_v = 0;
-  for (int i = 0; i < width; i++) {
-    sum_u += above_u[i];
-    sum_v += above_v[i];
-  }
-  *out_sum_u += sum_u;
-  *out_sum_v += sum_v;
-}
-#endif  // CONFIG_HIGHBITDEPTH
-
-static void sum_above_row(const MACROBLOCKD *xd, int width, int *out_sum_u,
-                          int *out_sum_v) {
-  const struct macroblockd_plane *const pd_u = &xd->plane[AOM_PLANE_U];
-  const struct macroblockd_plane *const pd_v = &xd->plane[AOM_PLANE_V];
-#if CONFIG_HIGHBITDEPTH
-  if (get_bitdepth_data_path_index(xd)) {
-    const uint16_t *above_u_16 =
-        CONVERT_TO_SHORTPTR(pd_u->dst.buf) - pd_u->dst.stride;
-    const uint16_t *above_v_16 =
-        CONVERT_TO_SHORTPTR(pd_v->dst.buf) - pd_v->dst.stride;
-    sum_above_row_hbd(above_u_16, above_v_16, width, out_sum_u, out_sum_v);
-    return;
-  }
-#endif  // CONFIG_HIGHBITDEPTH
-  const uint8_t *above_u = pd_u->dst.buf - pd_u->dst.stride;
-  const uint8_t *above_v = pd_v->dst.buf - pd_v->dst.stride;
-  sum_above_row_lbd(above_u, above_v, width, out_sum_u, out_sum_v);
-}
-
-static void sum_left_col_lbd(const uint8_t *left_u, int u_stride,
-                             const uint8_t *left_v, int v_stride, int height,
-                             int *out_sum_u, int *out_sum_v) {
-  int sum_u = 0;
-  int sum_v = 0;
-  for (int i = 0; i < height; i++) {
-    sum_u += left_u[i * u_stride];
-    sum_v += left_v[i * v_stride];
-  }
-  *out_sum_u += sum_u;
-  *out_sum_v += sum_v;
-}
-#if CONFIG_HIGHBITDEPTH
-static void sum_left_col_hbd(const uint16_t *left_u, int u_stride,
-                             const uint16_t *left_v, int v_stride, int height,
-                             int *out_sum_u, int *out_sum_v) {
-  int sum_u = 0;
-  int sum_v = 0;
-  for (int i = 0; i < height; i++) {
-    sum_u += left_u[i * u_stride];
-    sum_v += left_v[i * v_stride];
-  }
-  *out_sum_u += sum_u;
-  *out_sum_v += sum_v;
-}
-#endif  // CONFIG_HIGHBITDEPTH
-static void sum_left_col(const MACROBLOCKD *xd, int height, int *out_sum_u,
-                         int *out_sum_v) {
-  const struct macroblockd_plane *const pd_u = &xd->plane[AOM_PLANE_U];
-  const struct macroblockd_plane *const pd_v = &xd->plane[AOM_PLANE_V];
-
-#if CONFIG_HIGHBITDEPTH
-  if (get_bitdepth_data_path_index(xd)) {
-    const uint16_t *left_u_16 = CONVERT_TO_SHORTPTR(pd_u->dst.buf) - 1;
-    const uint16_t *left_v_16 = CONVERT_TO_SHORTPTR(pd_v->dst.buf) - 1;
-    sum_left_col_hbd(left_u_16, pd_u->dst.stride, left_v_16, pd_v->dst.stride,
-                     height, out_sum_u, out_sum_v);
-    return;
-  }
-#endif  // CONFIG_HIGHBITDEPTH
-  const uint8_t *left_u = pd_u->dst.buf - 1;
-  const uint8_t *left_v = pd_v->dst.buf - 1;
-  sum_left_col_lbd(left_u, pd_u->dst.stride, left_v, pd_v->dst.stride, height,
-                   out_sum_u, out_sum_v);
-}
-
-// CfL computes its own block-level DC_PRED. This is required to compute both
-// alpha_cb and alpha_cr before the prediction are computed.
-static void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize) {
-  CFL_CTX *const cfl = xd->cfl;
-
-  // Compute DC_PRED until block boundary. We can't assume the neighbor will use
-  // the same transform size.
-  const int width = max_block_wide(xd, plane_bsize, AOM_PLANE_U)
-                    << tx_size_wide_log2[0];
-  const int height = max_block_high(xd, plane_bsize, AOM_PLANE_U)
-                     << tx_size_high_log2[0];
-  // Number of pixel on the top and left borders.
-  const int num_pel = width + height;
-
-  int sum_u = 0;
-  int sum_v = 0;
-
-// Match behavior of build_intra_predictors_high (reconintra.c) at superblock
-// boundaries:
-// base-1 base-1 base-1 .. base-1 base-1 base-1 base-1 base-1 base-1
-// base+1   A      B  ..     Y      Z
-// base+1   C      D  ..     W      X
-// base+1   E      F  ..     U      V
-// base+1   G      H  ..     S      T      T      T      T      T
-// ..
-
-#if CONFIG_CHROMA_SUB8X8
-  if (xd->chroma_up_available && xd->mb_to_right_edge >= 0) {
-#else
-  if (xd->up_available && xd->mb_to_right_edge >= 0) {
-#endif
-    sum_above_row(xd, width, &sum_u, &sum_v);
-  } else {
-    const int base = 128 << (xd->bd - 8);
-    sum_u = width * (base - 1);
-    sum_v = width * (base - 1);
-  }
-
-#if CONFIG_CHROMA_SUB8X8
-  if (xd->chroma_left_available && xd->mb_to_bottom_edge >= 0) {
-#else
-  if (xd->left_available && xd->mb_to_bottom_edge >= 0) {
-#endif
-    sum_left_col(xd, height, &sum_u, &sum_v);
-  } else {
-    const int base = 128 << (xd->bd - 8);
-    sum_u += height * (base + 1);
-    sum_v += height * (base + 1);
-  }
-
-  // TODO(ltrudeau) Because of max_block_wide and max_block_high, num_pel will
-  // not be a power of two. So these divisions will have to use a lookup table.
-  cfl->dc_pred[CFL_PRED_U] = (sum_u + (num_pel >> 1)) / num_pel;
-  cfl->dc_pred[CFL_PRED_V] = (sum_v + (num_pel >> 1)) / num_pel;
-}
-
-static void cfl_subtract_averages(CFL_CTX *cfl, TX_SIZE tx_size) {
-  const int width = cfl->uv_width;
-  const int height = cfl->uv_height;
-  const int tx_height = tx_size_high[tx_size];
-  const int tx_width = tx_size_wide[tx_size];
-  const int block_row_stride = MAX_SB_SIZE << tx_size_high_log2[tx_size];
-  const int num_pel_log2 =
-      (tx_size_high_log2[tx_size] + tx_size_wide_log2[tx_size]);
-
-  int16_t *pred_buf_q3 = cfl->pred_buf_q3;
-
-  cfl_pad(cfl, width, height);
-
-  for (int b_j = 0; b_j < height; b_j += tx_height) {
-    for (int b_i = 0; b_i < width; b_i += tx_width) {
-      int sum_q3 = 0;
-      int16_t *tx_pred_buf_q3 = pred_buf_q3;
-      for (int t_j = 0; t_j < tx_height; t_j++) {
-        for (int t_i = b_i; t_i < b_i + tx_width; t_i++) {
-          sum_q3 += tx_pred_buf_q3[t_i];
-        }
-        tx_pred_buf_q3 += MAX_SB_SIZE;
-      }
-      int avg_q3 = (sum_q3 + (1 << (num_pel_log2 - 1))) >> num_pel_log2;
-      // Loss is never more than 1/2 (in Q3)
-      assert(fabs((double)avg_q3 - (sum_q3 / ((double)(1 << num_pel_log2)))) <=
-             0.5);
-
-      tx_pred_buf_q3 = pred_buf_q3;
-      for (int t_j = 0; t_j < tx_height; t_j++) {
-        for (int t_i = b_i; t_i < b_i + tx_width; t_i++) {
-          tx_pred_buf_q3[t_i] -= avg_q3;
-        }
-
-        tx_pred_buf_q3 += MAX_SB_SIZE;
-      }
+static void subtract_average_c(const uint16_t *src, int16_t *dst, int width,
+                               int height, int round_offset, int num_pel_log2) {
+  int sum = round_offset;
+  const uint16_t *recon = src;
+  for (int j = 0; j < height; j++) {
+    for (int i = 0; i < width; i++) {
+      sum += recon[i];
     }
-    pred_buf_q3 += block_row_stride;
+    recon += CFL_BUF_LINE;
+  }
+  const int avg = sum >> num_pel_log2;
+  for (int j = 0; j < height; j++) {
+    for (int i = 0; i < width; i++) {
+      dst[i] = src[i] - avg;
+    }
+    src += CFL_BUF_LINE;
+    dst += CFL_BUF_LINE;
   }
 }
 
+CFL_SUB_AVG_FN(c)
+
 static INLINE int cfl_idx_to_alpha(int alpha_idx, int joint_sign,
                                    CFL_PRED_TYPE pred_type) {
   const int alpha_sign = (pred_type == CFL_PRED_U) ? CFL_SIGN_U(joint_sign)
@@ -259,159 +153,218 @@ static INLINE int cfl_idx_to_alpha(int alpha_idx, int joint_sign,
   return (alpha_sign == CFL_SIGN_POS) ? abs_alpha_q3 + 1 : -abs_alpha_q3 - 1;
 }
 
-static void cfl_build_prediction_lbd(const int16_t *pred_buf_q3, uint8_t *dst,
-                                     int dst_stride, int width, int height,
-                                     int alpha_q3, int dc_pred) {
+static INLINE void cfl_predict_lbd_c(const int16_t *ac_buf_q3, uint8_t *dst,
+                                     int dst_stride, int alpha_q3, int width,
+                                     int height) {
   for (int j = 0; j < height; j++) {
     for (int i = 0; i < width; i++) {
-      dst[i] =
-          clip_pixel(get_scaled_luma_q0(alpha_q3, pred_buf_q3[i]) + dc_pred);
+      dst[i] = clip_pixel(get_scaled_luma_q0(alpha_q3, ac_buf_q3[i]) + dst[i]);
     }
     dst += dst_stride;
-    pred_buf_q3 += MAX_SB_SIZE;
+    ac_buf_q3 += CFL_BUF_LINE;
   }
 }
 
-#if CONFIG_HIGHBITDEPTH
-static void cfl_build_prediction_hbd(const int16_t *pred_buf_q3, uint16_t *dst,
-                                     int dst_stride, int width, int height,
-                                     int alpha_q3, int dc_pred, int bit_depth) {
+// 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,
+                       int alpha_q3, int bit_depth, int width, int height) {
   for (int j = 0; j < height; j++) {
     for (int i = 0; i < width; i++) {
       dst[i] = clip_pixel_highbd(
-          get_scaled_luma_q0(alpha_q3, pred_buf_q3[i]) + dc_pred, bit_depth);
+          get_scaled_luma_q0(alpha_q3, ac_buf_q3[i]) + dst[i], bit_depth);
     }
     dst += dst_stride;
-    pred_buf_q3 += MAX_SB_SIZE;
+    ac_buf_q3 += CFL_BUF_LINE;
   }
 }
-#endif  // CONFIG_HIGHBITDEPTH
 
-static void cfl_build_prediction(const int16_t *pred_buf_q3, uint8_t *dst,
-                                 int dst_stride, int width, int height,
-                                 int alpha_q3, int dc_pred, int use_hbd,
-                                 int bit_depth) {
-#if CONFIG_HIGHBITDEPTH
-  if (use_hbd) {
-    uint16_t *dst_16 = CONVERT_TO_SHORTPTR(dst);
-    cfl_build_prediction_hbd(pred_buf_q3, dst_16, dst_stride, width, height,
-                             alpha_q3, dc_pred, bit_depth);
-    return;
-  }
-#endif  // CONFIG_HIGHBITDEPTH
-  (void)use_hbd;
-  (void)bit_depth;
-  cfl_build_prediction_lbd(pred_buf_q3, dst, dst_stride, width, height,
-                           alpha_q3, dc_pred);
+// 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) {
+  CFL_CTX *const cfl = &xd->cfl;
+  // Do not call cfl_compute_parameters multiple time on the same values.
+  assert(cfl->are_parameters_computed == 0);
+
+  cfl_pad(cfl, tx_size_wide[tx_size], tx_size_high[tx_size]);
+  get_subtract_average_fn(tx_size)(cfl->recon_buf_q3, cfl->ac_buf_q3);
+  cfl->are_parameters_computed = 1;
 }
 
 void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
-                       int row, int col, TX_SIZE tx_size, int plane) {
-  CFL_CTX *const cfl = xd->cfl;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+                       TX_SIZE tx_size, int plane) {
+  CFL_CTX *const cfl = &xd->cfl;
+  MB_MODE_INFO *mbmi = xd->mi[0];
+  assert(is_cfl_allowed(xd));
 
-  // CfL parameters must be computed before prediction can be done.
-  assert(cfl->are_parameters_computed == 1);
+  if (!cfl->are_parameters_computed) cfl_compute_parameters(xd, tx_size);
 
-  const int16_t *pred_buf_q3 =
-      cfl->pred_buf_q3 + ((row * MAX_SB_SIZE + col) << tx_size_wide_log2[0]);
   const int alpha_q3 =
       cfl_idx_to_alpha(mbmi->cfl_alpha_idx, mbmi->cfl_alpha_signs, plane - 1);
-
-  cfl_build_prediction(pred_buf_q3, dst, dst_stride, tx_size_wide[tx_size],
-                       tx_size_high[tx_size], alpha_q3, cfl->dc_pred[plane - 1],
-                       get_bitdepth_data_path_index(xd), xd->bd);
+  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)) {
+    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);
+    return;
+  }
+  get_predict_lbd_fn(tx_size)(cfl->ac_buf_q3, dst, dst_stride, alpha_q3);
 }
 
-static void cfl_luma_subsampling_420_lbd(const uint8_t *input, int input_stride,
-                                         int16_t *output_q3, int width,
-                                         int height) {
-  for (int j = 0; j < height; j++) {
-    for (int i = 0; i < width; i++) {
-      int top = i << 1;
-      int bot = top + input_stride;
-      output_q3[i] = (input[top] + input[top + 1] + input[bot] + input[bot + 1])
-                     << 1;
+// 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,
+                                           int height) {
+  for (int j = 0; j < height; j += 2) {
+    for (int i = 0; i < width; i += 2) {
+      const int bot = i + input_stride;
+      output_q3[i >> 1] =
+          (input[i] + input[i + 1] + input[bot] + input[bot + 1]) << 1;
     }
     input += input_stride << 1;
-    output_q3 += MAX_SB_SIZE;
+    output_q3 += CFL_BUF_LINE;
   }
 }
 
-static void cfl_luma_subsampling_444_lbd(const uint8_t *input, int input_stride,
-                                         int16_t *output_q3, int width,
-                                         int height) {
+static void cfl_luma_subsampling_422_lbd_c(const uint8_t *input,
+                                           int input_stride,
+                                           uint16_t *output_q3, int width,
+                                           int height) {
+  assert((height - 1) * CFL_BUF_LINE + width <= CFL_BUF_SQUARE);
+  for (int j = 0; j < height; j++) {
+    for (int i = 0; i < width; i += 2) {
+      output_q3[i >> 1] = (input[i] + input[i + 1]) << 2;
+    }
+    input += input_stride;
+    output_q3 += CFL_BUF_LINE;
+  }
+}
+
+static void cfl_luma_subsampling_444_lbd_c(const uint8_t *input,
+                                           int input_stride,
+                                           uint16_t *output_q3, int width,
+                                           int height) {
+  assert((height - 1) * CFL_BUF_LINE + width <= CFL_BUF_SQUARE);
   for (int j = 0; j < height; j++) {
     for (int i = 0; i < width; i++) {
       output_q3[i] = input[i] << 3;
     }
     input += input_stride;
-    output_q3 += MAX_SB_SIZE;
+    output_q3 += CFL_BUF_LINE;
   }
 }
 
-#if CONFIG_HIGHBITDEPTH
-static void cfl_luma_subsampling_420_hbd(const uint16_t *input,
-                                         int input_stride, int16_t *output_q3,
-                                         int width, int height) {
-  for (int j = 0; j < height; j++) {
-    for (int i = 0; i < width; i++) {
-      int top = i << 1;
-      int bot = top + input_stride;
-      output_q3[i] = (input[top] + input[top + 1] + input[bot] + input[bot + 1])
-                     << 1;
+static void cfl_luma_subsampling_420_hbd_c(const uint16_t *input,
+                                           int input_stride,
+                                           uint16_t *output_q3, int width,
+                                           int height) {
+  for (int j = 0; j < height; j += 2) {
+    for (int i = 0; i < width; i += 2) {
+      const int bot = i + input_stride;
+      output_q3[i >> 1] =
+          (input[i] + input[i + 1] + input[bot] + input[bot + 1]) << 1;
     }
     input += input_stride << 1;
-    output_q3 += MAX_SB_SIZE;
+    output_q3 += CFL_BUF_LINE;
   }
 }
 
-static void cfl_luma_subsampling_444_hbd(const uint16_t *input,
-                                         int input_stride, int16_t *output_q3,
-                                         int width, int height) {
+static void cfl_luma_subsampling_422_hbd_c(const uint16_t *input,
+                                           int input_stride,
+                                           uint16_t *output_q3, int width,
+                                           int height) {
+  assert((height - 1) * CFL_BUF_LINE + width <= CFL_BUF_SQUARE);
+  for (int j = 0; j < height; j++) {
+    for (int i = 0; i < width; i += 2) {
+      output_q3[i >> 1] = (input[i] + input[i + 1]) << 2;
+    }
+    input += input_stride;
+    output_q3 += CFL_BUF_LINE;
+  }
+}
+
+static void cfl_luma_subsampling_444_hbd_c(const uint16_t *input,
+                                           int input_stride,
+                                           uint16_t *output_q3, int width,
+                                           int height) {
+  assert((height - 1) * CFL_BUF_LINE + width <= CFL_BUF_SQUARE);
   for (int j = 0; j < height; j++) {
     for (int i = 0; i < width; i++) {
       output_q3[i] = input[i] << 3;
     }
     input += input_stride;
-    output_q3 += MAX_SB_SIZE;
+    output_q3 += CFL_BUF_LINE;
   }
 }
-#endif  // CONFIG_HIGHBITDEPTH
 
-static void cfl_luma_subsampling_420(const uint8_t *input, int input_stride,
-                                     int16_t *output_q3, int width, int height,
-                                     int use_hbd) {
-#if CONFIG_HIGHBITDEPTH
-  if (use_hbd) {
-    const uint16_t *input_16 = CONVERT_TO_SHORTPTR(input);
-    cfl_luma_subsampling_420_hbd(input_16, input_stride, output_q3, width,
-                                 height);
-    return;
+CFL_GET_SUBSAMPLE_FUNCTION(c)
+
+static INLINE cfl_subsample_hbd_fn cfl_subsampling_hbd(TX_SIZE tx_size,
+                                                       int sub_x, int sub_y) {
+  if (sub_x == 1) {
+    if (sub_y == 1) {
+      return cfl_get_luma_subsampling_420_hbd(tx_size);
+    }
+    return cfl_get_luma_subsampling_422_hbd(tx_size);
   }
-#endif  // CONFIG_HIGHBITDEPTH
-  (void)use_hbd;
-  cfl_luma_subsampling_420_lbd(input, input_stride, output_q3, width, height);
+  return cfl_get_luma_subsampling_444_hbd(tx_size);
 }
 
-static void cfl_luma_subsampling_444(const uint8_t *input, int input_stride,
-                                     int16_t *output_q3, int width, int height,
-                                     int use_hbd) {
-#if CONFIG_HIGHBITDEPTH
-  if (use_hbd) {
-    uint16_t *input_16 = CONVERT_TO_SHORTPTR(input);
-    cfl_luma_subsampling_444_hbd(input_16, input_stride, output_q3, width,
-                                 height);
-    return;
+static INLINE cfl_subsample_lbd_fn cfl_subsampling_lbd(TX_SIZE tx_size,
+                                                       int sub_x, int sub_y) {
+  if (sub_x == 1) {
+    if (sub_y == 1) {
+      return cfl_get_luma_subsampling_420_lbd(tx_size);
+    }
+    return cfl_get_luma_subsampling_422_lbd(tx_size);
   }
-#endif  // CONFIG_HIGHBITDEPTH
-  (void)use_hbd;
-  cfl_luma_subsampling_444_lbd(input, input_stride, output_q3, width, height);
+  return cfl_get_luma_subsampling_444_lbd(tx_size);
 }
 
-static INLINE void cfl_store(CFL_CTX *cfl, const uint8_t *input,
-                             int input_stride, int row, int col, int width,
-                             int height, int use_hbd) {
+static void cfl_store(CFL_CTX *cfl, const uint8_t *input, int input_stride,
+                      int row, int col, TX_SIZE tx_size, int use_hbd) {
+  const int width = tx_size_wide[tx_size];
+  const int height = tx_size_high[tx_size];
   const int tx_off_log2 = tx_size_wide_log2[0];
   const int sub_x = cfl->subsampling_x;
   const int sub_y = cfl->subsampling_y;
@@ -435,26 +388,22 @@ static INLINE void cfl_store(CFL_CTX *cfl, const uint8_t *input,
   }
 
   // Check that we will remain inside the pixel buffer.
-  assert(store_row + store_height <= MAX_SB_SIZE);
-  assert(store_col + store_width <= MAX_SB_SIZE);
+  assert(store_row + store_height <= CFL_BUF_LINE);
+  assert(store_col + store_width <= CFL_BUF_LINE);
 
   // Store the input into the CfL pixel buffer
-  int16_t *pred_buf_q3 =
-      cfl->pred_buf_q3 + (store_row * MAX_SB_SIZE + store_col);
+  uint16_t *recon_buf_q3 =
+      cfl->recon_buf_q3 + (store_row * CFL_BUF_LINE + store_col);
 
-  if (sub_y == 0 && sub_x == 0) {
-    cfl_luma_subsampling_444(input, input_stride, pred_buf_q3, store_width,
-                             store_height, use_hbd);
-  } else if (sub_y == 1 && sub_x == 1) {
-    cfl_luma_subsampling_420(input, input_stride, pred_buf_q3, store_width,
-                             store_height, use_hbd);
+  if (use_hbd) {
+    cfl_subsampling_hbd(tx_size, sub_x, sub_y)(CONVERT_TO_SHORTPTR(input),
+                                               input_stride, recon_buf_q3);
   } else {
-    // TODO(ltrudeau) add support for 4:2:2
-    assert(0);  // Unsupported chroma subsampling
+    cfl_subsampling_lbd(tx_size, sub_x, sub_y)(input, input_stride,
+                                               recon_buf_q3);
   }
 }
 
-#if CONFIG_CHROMA_SUB8X8
 // Adjust the row and column of blocks smaller than 8X8, as chroma-referenced
 // and non-chroma-referenced blocks are stored together in the CfL buffer.
 static INLINE void sub8x8_adjust_offset(const CFL_CTX *cfl, int *row_out,
@@ -471,99 +420,36 @@ static INLINE void sub8x8_adjust_offset(const CFL_CTX *cfl, int *row_out,
     (*col_out)++;
   }
 }
-#if CONFIG_DEBUG
-static INLINE void sub8x8_set_val(CFL_CTX *cfl, int row, int col, int val_high,
-                                  int val_wide) {
-  for (int val_r = 0; val_r < val_high; val_r++) {
-    assert(row + val_r < CFL_SUB8X8_VAL_MI_SIZE);
-    int row_off = (row + val_r) * CFL_SUB8X8_VAL_MI_SIZE;
-    for (int val_c = 0; val_c < val_wide; val_c++) {
-      assert(col + val_c < CFL_SUB8X8_VAL_MI_SIZE);
-      assert(cfl->sub8x8_val[row_off + col + val_c] == 0);
-      cfl->sub8x8_val[row_off + col + val_c]++;
-    }
-  }
-}
-#endif  // CONFIG_DEBUG
-#endif  // CONFIG_CHROMA_SUB8X8
 
 void cfl_store_tx(MACROBLOCKD *const xd, int row, int col, TX_SIZE tx_size,
                   BLOCK_SIZE bsize) {
-  CFL_CTX *const cfl = xd->cfl;
+  CFL_CTX *const cfl = &xd->cfl;
   struct macroblockd_plane *const pd = &xd->plane[AOM_PLANE_Y];
   uint8_t *dst =
       &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
-  (void)bsize;
-#if CONFIG_CHROMA_SUB8X8
 
   if (block_size_high[bsize] == 4 || block_size_wide[bsize] == 4) {
     // Only dimensions of size 4 can have an odd offset.
     assert(!((col & 1) && tx_size_wide[tx_size] != 4));
     assert(!((row & 1) && tx_size_high[tx_size] != 4));
     sub8x8_adjust_offset(cfl, &row, &col);
-#if CONFIG_DEBUG
-    sub8x8_set_val(cfl, row, col, tx_size_high_unit[tx_size],
-                   tx_size_wide_unit[tx_size]);
-#endif  // CONFIG_DEBUG
   }
-#endif
-  cfl_store(cfl, dst, pd->dst.stride, row, col, tx_size_wide[tx_size],
-            tx_size_high[tx_size], get_bitdepth_data_path_index(xd));
-}
-
-void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size) {
-  CFL_CTX *const cfl = xd->cfl;
-  struct macroblockd_plane *const pd = &xd->plane[AOM_PLANE_Y];
-  int row = 0;
-  int col = 0;
-#if CONFIG_CHROMA_SUB8X8
-  bsize = AOMMAX(BLOCK_4X4, bsize);
-  if (block_size_high[bsize] == 4 || block_size_wide[bsize] == 4) {
-    sub8x8_adjust_offset(cfl, &row, &col);
-#if CONFIG_DEBUG
-    sub8x8_set_val(cfl, row, col, mi_size_high[bsize], mi_size_wide[bsize]);
-#endif  // CONFIG_DEBUG
-  }
-#endif  // CONFIG_CHROMA_SUB8X8
-  const int width = max_intra_block_width(xd, bsize, AOM_PLANE_Y, tx_size);
-  const int height = max_intra_block_height(xd, bsize, AOM_PLANE_Y, tx_size);
-  cfl_store(cfl, pd->dst.buf, pd->dst.stride, row, col, width, height,
+  cfl_store(cfl, dst, pd->dst.stride, row, col, tx_size,
             get_bitdepth_data_path_index(xd));
 }
 
-void cfl_compute_parameters(MACROBLOCKD *const xd, TX_SIZE tx_size) {
-  CFL_CTX *const cfl = xd->cfl;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size) {
+  CFL_CTX *const cfl = &xd->cfl;
+  struct macroblockd_plane *const pd = &xd->plane[AOM_PLANE_Y];
+  int row = 0;
+  int col = 0;
 
-  // Do not call cfl_compute_parameters multiple time on the same values.
-  assert(cfl->are_parameters_computed == 0);
-
-#if CONFIG_CHROMA_SUB8X8
-  const BLOCK_SIZE plane_bsize = AOMMAX(
-      BLOCK_4X4, get_plane_block_size(mbmi->sb_type, &xd->plane[AOM_PLANE_U]));
-#if CONFIG_DEBUG
-  if (mbmi->sb_type < BLOCK_8X8) {
-    for (int val_r = 0; val_r < mi_size_high[mbmi->sb_type]; val_r++) {
-      for (int val_c = 0; val_c < mi_size_wide[mbmi->sb_type]; val_c++) {
-        assert(cfl->sub8x8_val[val_r * CFL_SUB8X8_VAL_MI_SIZE + val_c] == 1);
-      }
-    }
-    cfl_clear_sub8x8_val(cfl);
+  if (block_size_high[bsize] == 4 || block_size_wide[bsize] == 4) {
+    sub8x8_adjust_offset(cfl, &row, &col);
   }
-#endif  // CONFIG_DEBUG
-#else
-  const BLOCK_SIZE plane_bsize =
-      get_plane_block_size(mbmi->sb_type, &xd->plane[AOM_PLANE_U]);
-#endif
-  // AOM_PLANE_U is used, but both planes will have the same sizes.
-  cfl->uv_width = max_intra_block_width(xd, plane_bsize, AOM_PLANE_U, tx_size);
-  cfl->uv_height =
-      max_intra_block_height(xd, plane_bsize, AOM_PLANE_U, tx_size);
-
-  assert(cfl->buf_width <= cfl->uv_width);
-  assert(cfl->buf_height <= cfl->uv_height);
-
-  cfl_dc_pred(xd, plane_bsize);
-  cfl_subtract_averages(cfl, tx_size);
-  cfl->are_parameters_computed = 1;
+  const int width = max_intra_block_width(xd, bsize, AOM_PLANE_Y, 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));
 }

third_party/aom/av1/common/cfl.h

@@ -13,20 +13,290 @@
 #define AV1_COMMON_CFL_H_
 
 #include "av1/common/blockd.h"
+#include "av1/common/onyxc_int.h"
+
+// Can we use CfL for the current block?
+static INLINE CFL_ALLOWED_TYPE is_cfl_allowed(const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *mbmi = xd->mi[0];
+  const BLOCK_SIZE bsize = mbmi->sb_type;
+  assert(bsize < BLOCK_SIZES_ALL);
+  if (xd->lossless[mbmi->segment_id]) {
+    // In lossless, CfL is available when the partition size is equal to the
+    // transform size.
+    const int ssx = xd->plane[AOM_PLANE_U].subsampling_x;
+    const int ssy = xd->plane[AOM_PLANE_U].subsampling_y;
+    const int plane_bsize = get_plane_block_size(bsize, ssx, ssy);
+    return (CFL_ALLOWED_TYPE)(plane_bsize == BLOCK_4X4);
+  }
+  // Spec: CfL is available to luma partitions lesser than or equal to 32x32
+  return (CFL_ALLOWED_TYPE)(block_size_wide[bsize] <= 32 &&
+                            block_size_high[bsize] <= 32);
+}
+
+// Do we need to save the luma pixels from the current block,
+// for a possible future CfL prediction?
+static INLINE CFL_ALLOWED_TYPE store_cfl_required(const AV1_COMMON *cm,
+                                                  const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *mbmi = xd->mi[0];
+
+  if (cm->seq_params.monochrome) return CFL_DISALLOWED;
+
+  if (!xd->cfl.is_chroma_reference) {
+    // For non-chroma-reference blocks, we should always store the luma pixels,
+    // in case the corresponding chroma-reference block uses CfL.
+    // Note that this can only happen for block sizes which are <8 on
+    // their shortest side, as otherwise they would be chroma reference
+    // blocks.
+    return CFL_ALLOWED;
+  }
+
+  // If this block has chroma information, we know whether we're
+  // actually going to perform a CfL prediction
+  return (CFL_ALLOWED_TYPE)(!is_inter_block(mbmi) &&
+                            mbmi->uv_mode == UV_CFL_PRED);
+}
 
 static INLINE int get_scaled_luma_q0(int alpha_q3, int16_t pred_buf_q3) {
   int scaled_luma_q6 = alpha_q3 * pred_buf_q3;
   return ROUND_POWER_OF_TWO_SIGNED(scaled_luma_q6, 6);
 }
 
+static INLINE CFL_PRED_TYPE get_cfl_pred_type(PLANE_TYPE plane) {
+  assert(plane > 0);
+  return (CFL_PRED_TYPE)(plane - 1);
+}
+
 void cfl_predict_block(MACROBLOCKD *const xd, uint8_t *dst, int dst_stride,
-                       int row, int col, TX_SIZE tx_size, int plane);
+                       TX_SIZE tx_size, int plane);
 
 void cfl_store_block(MACROBLOCKD *const xd, BLOCK_SIZE bsize, TX_SIZE tx_size);
 
 void cfl_store_tx(MACROBLOCKD *const xd, int row, int col, TX_SIZE tx_size,
                   BLOCK_SIZE bsize);
 
-void cfl_compute_parameters(MACROBLOCKD *const xd, TX_SIZE tx_size);
+void cfl_store_dc_pred(MACROBLOCKD *const xd, const uint8_t *input,
+                       CFL_PRED_TYPE pred_plane, int width);
+
+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
+
+// 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
+// goodness.
+#define CFL_SUBSAMPLE(arch, sub, bd, width, height)                       \
+  void subsample_##bd##_##sub##_##width##x##height##_##arch(              \
+      const CFL_##bd##_TYPE, int input_stride, uint16_t *output_q3) {     \
+    cfl_luma_subsampling_##sub##_##bd##_##arch(cfl_type, input_stride,    \
+                                               output_q3, width, height); \
+  }
+
+// Declare size-specific wrappers for all valid CfL sizes.
+#define CFL_SUBSAMPLE_FUNCTIONS(arch, sub, bd)                            \
+  CFL_SUBSAMPLE(arch, sub, bd, 4, 4)                                      \
+  CFL_SUBSAMPLE(arch, sub, bd, 8, 8)                                      \
+  CFL_SUBSAMPLE(arch, sub, bd, 16, 16)                                    \
+  CFL_SUBSAMPLE(arch, sub, bd, 32, 32)                                    \
+  CFL_SUBSAMPLE(arch, sub, bd, 4, 8)                                      \
+  CFL_SUBSAMPLE(arch, sub, bd, 8, 4)                                      \
+  CFL_SUBSAMPLE(arch, sub, bd, 8, 16)                                     \
+  CFL_SUBSAMPLE(arch, sub, bd, 16, 8)                                     \
+  CFL_SUBSAMPLE(arch, sub, bd, 16, 32)                                    \
+  CFL_SUBSAMPLE(arch, sub, bd, 32, 16)                                    \
+  CFL_SUBSAMPLE(arch, sub, bd, 4, 16)                                     \
+  CFL_SUBSAMPLE(arch, sub, bd, 16, 4)                                     \
+  CFL_SUBSAMPLE(arch, sub, bd, 8, 32)                                     \
+  CFL_SUBSAMPLE(arch, sub, bd, 32, 8)                                     \
+  cfl_subsample_##bd##_fn cfl_get_luma_subsampling_##sub##_##bd##_##arch( \
+      TX_SIZE tx_size) {                                                  \
+    CFL_SUBSAMPLE_FUNCTION_ARRAY(arch, sub, bd)                           \
+    return subfn_##sub[tx_size];                                          \
+  }
+
+// Declare an architecture-specific array of function pointers for size-specific
+// wrappers.
+#define CFL_SUBSAMPLE_FUNCTION_ARRAY(arch, sub, bd)                       \
+  static const cfl_subsample_##bd##_fn subfn_##sub[TX_SIZES_ALL] = {      \
+    subsample_##bd##_##sub##_4x4_##arch,   /* 4x4 */                      \
+    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) */ \
+    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) */ \
+    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) */ \
+  };
+
+// The RTCD script does not support passing in an array, so we wrap it in this
+// function.
+#define CFL_GET_SUBSAMPLE_FUNCTION(arch)  \
+  CFL_SUBSAMPLE_FUNCTIONS(arch, 420, lbd) \
+  CFL_SUBSAMPLE_FUNCTIONS(arch, 422, lbd) \
+  CFL_SUBSAMPLE_FUNCTIONS(arch, 444, lbd) \
+  CFL_SUBSAMPLE_FUNCTIONS(arch, 420, hbd) \
+  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
+// goodness.
+#define CFL_SUB_AVG_X(arch, width, height, round_offset, num_pel_log2)   \
+  void subtract_average_##width##x##height##_##arch(const uint16_t *src, \
+                                                    int16_t *dst) {      \
+    subtract_average_##arch(src, dst, width, height, round_offset,       \
+                            num_pel_log2);                               \
+  }
+
+// Declare size-specific wrappers for all valid CfL sizes.
+#define CFL_SUB_AVG_FN(arch)                                                \
+  CFL_SUB_AVG_X(arch, 4, 4, 8, 4)                                           \
+  CFL_SUB_AVG_X(arch, 4, 8, 16, 5)                                          \
+  CFL_SUB_AVG_X(arch, 4, 16, 32, 6)                                         \
+  CFL_SUB_AVG_X(arch, 8, 4, 16, 5)                                          \
+  CFL_SUB_AVG_X(arch, 8, 8, 32, 6)                                          \
+  CFL_SUB_AVG_X(arch, 8, 16, 64, 7)                                         \
+  CFL_SUB_AVG_X(arch, 8, 32, 128, 8)                                        \
+  CFL_SUB_AVG_X(arch, 16, 4, 32, 6)                                         \
+  CFL_SUB_AVG_X(arch, 16, 8, 64, 7)                                         \
+  CFL_SUB_AVG_X(arch, 16, 16, 128, 8)                                       \
+  CFL_SUB_AVG_X(arch, 16, 32, 256, 9)                                       \
+  CFL_SUB_AVG_X(arch, 32, 8, 128, 8)                                        \
+  CFL_SUB_AVG_X(arch, 32, 16, 256, 9)                                       \
+  CFL_SUB_AVG_X(arch, 32, 32, 512, 10)                                      \
+  cfl_subtract_average_fn get_subtract_average_fn_##arch(TX_SIZE tx_size) { \
+    static const cfl_subtract_average_fn sub_avg[TX_SIZES_ALL] = {          \
+      subtract_average_4x4_##arch,   /* 4x4 */                              \
+      subtract_average_8x8_##arch,   /* 8x8 */                              \
+      subtract_average_16x16_##arch, /* 16x16 */                            \
+      subtract_average_32x32_##arch, /* 32x32 */                            \
+      cfl_subtract_average_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) */         \
+      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) */         \
+    };                                                                      \
+    /* Modulo TX_SIZES_ALL to ensure that an attacker won't be able to */   \
+    /* index the function pointer array out of bounds. */                   \
+    return sub_avg[tx_size % TX_SIZES_ALL];                                 \
+  }
+
+// For VSX SIMD optimization, the C versions of width == 4 subtract are
+// faster than the VSX. As such, the VSX code calls the C versions.
+void subtract_average_4x4_c(const uint16_t *src, int16_t *dst);
+void subtract_average_4x8_c(const uint16_t *src, int16_t *dst);
+void subtract_average_4x16_c(const uint16_t *src, int16_t *dst);
+
+#define CFL_PREDICT_lbd(arch, width, height)                                 \
+  void predict_lbd_##width##x##height##_##arch(const int16_t *pred_buf_q3,   \
+                                               uint8_t *dst, int dst_stride, \
+                                               int alpha_q3) {               \
+    cfl_predict_lbd_##arch(pred_buf_q3, dst, dst_stride, alpha_q3, width,    \
+                           height);                                          \
+  }
+
+#define CFL_PREDICT_hbd(arch, width, height)                                  \
+  void predict_hbd_##width##x##height##_##arch(const int16_t *pred_buf_q3,    \
+                                               uint16_t *dst, int dst_stride, \
+                                               int alpha_q3, int bd) {        \
+    cfl_predict_hbd_##arch(pred_buf_q3, dst, dst_stride, alpha_q3, bd, width, \
+                           height);                                           \
+  }
+
+// This wrapper exists because clang format does not like calling macros with
+// lowercase letters.
+#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)                                           \
+  CFL_PREDICT_X(arch, 4, 16, bd)                                          \
+  CFL_PREDICT_X(arch, 8, 4, bd)                                           \
+  CFL_PREDICT_X(arch, 8, 8, bd)                                           \
+  CFL_PREDICT_X(arch, 8, 16, bd)                                          \
+  CFL_PREDICT_X(arch, 8, 32, bd)                                          \
+  CFL_PREDICT_X(arch, 16, 4, bd)                                          \
+  CFL_PREDICT_X(arch, 16, 8, bd)                                          \
+  CFL_PREDICT_X(arch, 16, 16, bd)                                         \
+  CFL_PREDICT_X(arch, 16, 32, bd)                                         \
+  CFL_PREDICT_X(arch, 32, 8, bd)                                          \
+  CFL_PREDICT_X(arch, 32, 16, bd)                                         \
+  CFL_PREDICT_X(arch, 32, 32, bd)                                         \
+  cfl_predict_##bd##_fn get_predict_##bd##_fn_##arch(TX_SIZE tx_size) {   \
+    static const cfl_predict_##bd##_fn pred[TX_SIZES_ALL] = {             \
+      predict_##bd##_4x4_##arch,   /* 4x4 */                              \
+      predict_##bd##_8x8_##arch,   /* 8x8 */                              \
+      predict_##bd##_16x16_##arch, /* 16x16 */                            \
+      predict_##bd##_32x32_##arch, /* 32x32 */                            \
+      cfl_predict_##bd##_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) */         \
+      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) */         \
+    };                                                                    \
+    /* Modulo TX_SIZES_ALL to ensure that an attacker won't be able to */ \
+    /* index the function pointer array out of bounds. */                 \
+    return pred[tx_size % TX_SIZES_ALL];                                  \
+  }
 
 #endif  // AV1_COMMON_CFL_H_

third_party/aom/av1/common/clpf.c

@@ -1,115 +0,0 @@
-/*
- * 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 "./av1_rtcd.h"
-#include "./cdef.h"
-#include "aom/aom_image.h"
-#include "aom_dsp/aom_dsp_common.h"
-
-static int clpf_sample(int X, int A, int B, int C, int D, int E, int F, int G,
-                       int H, int s, unsigned int dmp) {
-  int delta = 1 * constrain(A - X, s, dmp) + 3 * constrain(B - X, s, dmp) +
-              1 * constrain(C - X, s, dmp) + 3 * constrain(D - X, s, dmp) +
-              3 * constrain(E - X, s, dmp) + 1 * constrain(F - X, s, dmp) +
-              3 * constrain(G - X, s, dmp) + 1 * constrain(H - X, s, dmp);
-  return (8 + delta - (delta < 0)) >> 4;
-}
-
-static int clpf_hsample(int X, int A, int B, int C, int D, int s,
-                        unsigned int dmp) {
-  int delta = 1 * constrain(A - X, s, dmp) + 3 * constrain(B - X, s, dmp) +
-              3 * constrain(C - X, s, dmp) + 1 * constrain(D - X, s, dmp);
-  return (4 + delta - (delta < 0)) >> 3;
-}
-
-void aom_clpf_block_c(uint8_t *dst, const uint16_t *src, int dstride,
-                      int sstride, int sizex, int sizey, unsigned int strength,
-                      unsigned int damping) {
-  int x, y;
-
-  for (y = 0; y < sizey; y++) {
-    for (x = 0; x < sizex; x++) {
-      const int X = src[y * sstride + x];
-      const int A = src[(y - 2) * sstride + x];
-      const int B = src[(y - 1) * sstride + x];
-      const int C = src[y * sstride + x - 2];
-      const int D = src[y * sstride + x - 1];
-      const int E = src[y * sstride + x + 1];
-      const int F = src[y * sstride + x + 2];
-      const int G = src[(y + 1) * sstride + x];
-      const int H = src[(y + 2) * sstride + x];
-      const int delta =
-          clpf_sample(X, A, B, C, D, E, F, G, H, strength, damping);
-      dst[y * dstride + x] = X + delta;
-    }
-  }
-}
-
-// Identical to aom_clpf_block_c() apart from "dst".
-void aom_clpf_block_hbd_c(uint16_t *dst, const uint16_t *src, int dstride,
-                          int sstride, int sizex, int sizey,
-                          unsigned int strength, unsigned int damping) {
-  int x, y;
-
-  for (y = 0; y < sizey; y++) {
-    for (x = 0; x < sizex; x++) {
-      const int X = src[y * sstride + x];
-      const int A = src[(y - 2) * sstride + x];
-      const int B = src[(y - 1) * sstride + x];
-      const int C = src[y * sstride + x - 2];
-      const int D = src[y * sstride + x - 1];
-      const int E = src[y * sstride + x + 1];
-      const int F = src[y * sstride + x + 2];
-      const int G = src[(y + 1) * sstride + x];
-      const int H = src[(y + 2) * sstride + x];
-      const int delta =
-          clpf_sample(X, A, B, C, D, E, F, G, H, strength, damping);
-      dst[y * dstride + x] = X + delta;
-    }
-  }
-}
-
-// Vertically restricted filter
-void aom_clpf_hblock_c(uint8_t *dst, const uint16_t *src, int dstride,
-                       int sstride, int sizex, int sizey, unsigned int strength,
-                       unsigned int damping) {
-  int x, y;
-
-  for (y = 0; y < sizey; y++) {
-    for (x = 0; x < sizex; x++) {
-      const int X = src[y * sstride + x];
-      const int A = src[y * sstride + x - 2];
-      const int B = src[y * sstride + x - 1];
-      const int C = src[y * sstride + x + 1];
-      const int D = src[y * sstride + x + 2];
-      const int delta = clpf_hsample(X, A, B, C, D, strength, damping);
-      dst[y * dstride + x] = X + delta;
-    }
-  }
-}
-
-void aom_clpf_hblock_hbd_c(uint16_t *dst, const uint16_t *src, int dstride,
-                           int sstride, int sizex, int sizey,
-                           unsigned int strength, unsigned int damping) {
-  int x, y;
-
-  for (y = 0; y < sizey; y++) {
-    for (x = 0; x < sizex; x++) {
-      const int X = src[y * sstride + x];
-      const int A = src[y * sstride + x - 2];
-      const int B = src[y * sstride + x - 1];
-      const int C = src[y * sstride + x + 1];
-      const int D = src[y * sstride + x + 2];
-      const int delta = clpf_hsample(X, A, B, C, D, strength, damping);
-      dst[y * dstride + x] = X + delta;
-    }
-  }
-}

third_party/aom/av1/common/clpf_neon.c

@@ -1,14 +0,0 @@
-/*
- * 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_dsp/aom_simd.h"
-#define SIMD_FUNC(name) name##_neon
-#include "./clpf_simd.h"

third_party/aom/av1/common/clpf_simd.h

@@ -1,456 +0,0 @@
-/*
- * 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 "./av1_rtcd.h"
-#include "aom_ports/bitops.h"
-#include "aom_ports/mem.h"
-
-// sign(a-b) * min(abs(a-b), max(0, threshold - (abs(a-b) >> adjdamp)))
-SIMD_INLINE v128 constrain16(v128 a, v128 b, unsigned int threshold,
-                             unsigned int adjdamp) {
-  v128 diff = v128_sub_16(a, b);
-  const v128 sign = v128_shr_n_s16(diff, 15);
-  diff = v128_abs_s16(diff);
-  const v128 s =
-      v128_ssub_u16(v128_dup_16(threshold), v128_shr_u16(diff, adjdamp));
-  return v128_xor(v128_add_16(sign, v128_min_s16(diff, s)), sign);
-}
-
-// sign(a - b) * min(abs(a - b), max(0, strength - (abs(a - b) >> adjdamp)))
-SIMD_INLINE v128 constrain(v256 a, v256 b, unsigned int strength,
-                           unsigned int adjdamp) {
-  const v256 diff16 = v256_sub_16(a, b);
-  v128 diff = v128_pack_s16_s8(v256_high_v128(diff16), v256_low_v128(diff16));
-  const v128 sign = v128_cmplt_s8(diff, v128_zero());
-  diff = v128_abs_s8(diff);
-  return v128_xor(
-      v128_add_8(sign,
-                 v128_min_u8(diff, v128_ssub_u8(v128_dup_8(strength),
-                                                v128_shr_u8(diff, adjdamp)))),
-      sign);
-}
-
-// delta = 1/16 * constrain(a, x, s, d) + 3/16 * constrain(b, x, s, d) +
-//         1/16 * constrain(c, x, s, d) + 3/16 * constrain(d, x, s, d) +
-//         3/16 * constrain(e, x, s, d) + 1/16 * constrain(f, x, s, d) +
-//         3/16 * constrain(g, x, s, d) + 1/16 * constrain(h, x, s, d)
-SIMD_INLINE v128 calc_delta(v256 x, v256 a, v256 b, v256 c, v256 d, v256 e,
-                            v256 f, v256 g, v256 h, unsigned int s,
-                            unsigned int dmp) {
-  const v128 bdeg =
-      v128_add_8(v128_add_8(constrain(b, x, s, dmp), constrain(d, x, s, dmp)),
-                 v128_add_8(constrain(e, x, s, dmp), constrain(g, x, s, dmp)));
-  const v128 delta = v128_add_8(
-      v128_add_8(v128_add_8(constrain(a, x, s, dmp), constrain(c, x, s, dmp)),
-                 v128_add_8(constrain(f, x, s, dmp), constrain(h, x, s, dmp))),
-      v128_add_8(v128_add_8(bdeg, bdeg), bdeg));
-  return v128_add_8(
-      v128_pack_s16_u8(v256_high_v128(x), v256_low_v128(x)),
-      v128_shr_s8(
-          v128_add_8(v128_dup_8(8),
-                     v128_add_8(delta, v128_cmplt_s8(delta, v128_zero()))),
-          4));
-}
-
-// delta = 1/8 * constrain(a, x, s, d) + 3/8 * constrain(b, x, s, d) +
-//         3/8 * constrain(c, x, s, d) + 1/8 * constrain(d, x, s, d) +
-SIMD_INLINE v128 calc_hdelta(v256 x, v256 a, v256 b, v256 c, v256 d,
-                             unsigned int s, unsigned int dmp) {
-  const v128 bc = v128_add_8(constrain(b, x, s, dmp), constrain(c, x, s, dmp));
-  const v128 delta =
-      v128_add_8(v128_add_8(constrain(a, x, s, dmp), constrain(d, x, s, dmp)),
-                 v128_add_8(v128_add_8(bc, bc), bc));
-  return v128_add_8(
-      v128_pack_s16_u8(v256_high_v128(x), v256_low_v128(x)),
-      v128_shr_s8(
-          v128_add_8(v128_dup_8(4),
-                     v128_add_8(delta, v128_cmplt_s8(delta, v128_zero()))),
-          3));
-}
-
-// Process blocks of width 8, two lines at a time, 8 bit.
-static void SIMD_FUNC(clpf_block8)(uint8_t *dst, const uint16_t *src,
-                                   int dstride, int sstride, int sizey,
-                                   unsigned int strength,
-                                   unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 2) {
-    const v128 l1 = v128_load_aligned(src);
-    const v128 l2 = v128_load_aligned(src + sstride);
-    const v128 l3 = v128_load_aligned(src - sstride);
-    const v128 l4 = v128_load_aligned(src + 2 * sstride);
-    const v256 a = v256_from_v128(v128_load_aligned(src - 2 * sstride), l3);
-    const v256 b = v256_from_v128(l3, l1);
-    const v256 g = v256_from_v128(l2, l4);
-    const v256 h = v256_from_v128(l4, v128_load_aligned(src + 3 * sstride));
-    const v256 c = v256_from_v128(v128_load_unaligned(src - 2),
-                                  v128_load_unaligned(src - 2 + sstride));
-    const v256 d = v256_from_v128(v128_load_unaligned(src - 1),
-                                  v128_load_unaligned(src - 1 + sstride));
-    const v256 e = v256_from_v128(v128_load_unaligned(src + 1),
-                                  v128_load_unaligned(src + 1 + sstride));
-    const v256 f = v256_from_v128(v128_load_unaligned(src + 2),
-                                  v128_load_unaligned(src + 2 + sstride));
-    const v128 o = calc_delta(v256_from_v128(l1, l2), a, b, c, d, e, f, g, h,
-                              strength, adjdamp);
-
-    v64_store_aligned(dst, v128_high_v64(o));
-    v64_store_aligned(dst + dstride, v128_low_v64(o));
-    src += sstride * 2;
-    dst += dstride * 2;
-  }
-}
-
-// Process blocks of width 4, four lines at a time, 8 bit.
-static void SIMD_FUNC(clpf_block4)(uint8_t *dst, const uint16_t *src,
-                                   int dstride, int sstride, int sizey,
-                                   unsigned int strength,
-                                   unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 4) {
-    const v64 l0 = v64_load_aligned(src - 2 * sstride);
-    const v64 l1 = v64_load_aligned(src - sstride);
-    const v64 l2 = v64_load_aligned(src);
-    const v64 l3 = v64_load_aligned(src + sstride);
-    const v64 l4 = v64_load_aligned(src + 2 * sstride);
-    const v64 l5 = v64_load_aligned(src + 3 * sstride);
-    const v64 l6 = v64_load_aligned(src + 4 * sstride);
-    const v64 l7 = v64_load_aligned(src + 5 * sstride);
-    const v128 o =
-        calc_delta(v256_from_v64(l2, l3, l4, l5), v256_from_v64(l0, l1, l2, l3),
-                   v256_from_v64(l1, l2, l3, l4),
-                   v256_from_v64(v64_load_unaligned(src - 2),
-                                 v64_load_unaligned(src + sstride - 2),
-                                 v64_load_unaligned(src + 2 * sstride - 2),
-                                 v64_load_unaligned(src + 3 * sstride - 2)),
-                   v256_from_v64(v64_load_unaligned(src - 1),
-                                 v64_load_unaligned(src + sstride - 1),
-                                 v64_load_unaligned(src + 2 * sstride - 1),
-                                 v64_load_unaligned(src + 3 * sstride - 1)),
-                   v256_from_v64(v64_load_unaligned(src + 1),
-                                 v64_load_unaligned(src + sstride + 1),
-                                 v64_load_unaligned(src + 2 * sstride + 1),
-                                 v64_load_unaligned(src + 3 * sstride + 1)),
-                   v256_from_v64(v64_load_unaligned(src + 2),
-                                 v64_load_unaligned(src + sstride + 2),
-                                 v64_load_unaligned(src + 2 * sstride + 2),
-                                 v64_load_unaligned(src + 3 * sstride + 2)),
-                   v256_from_v64(l3, l4, l5, l6), v256_from_v64(l4, l5, l6, l7),
-                   strength, adjdamp);
-
-    u32_store_aligned(dst, v128_low_u32(v128_shr_n_byte(o, 12)));
-    u32_store_aligned(dst + dstride, v128_low_u32(v128_shr_n_byte(o, 8)));
-    u32_store_aligned(dst + 2 * dstride, v128_low_u32(v128_shr_n_byte(o, 4)));
-    u32_store_aligned(dst + 3 * dstride, v128_low_u32(o));
-
-    dst += 4 * dstride;
-    src += 4 * sstride;
-  }
-}
-
-static void SIMD_FUNC(clpf_hblock8)(uint8_t *dst, const uint16_t *src,
-                                    int dstride, int sstride, int sizey,
-                                    unsigned int strength,
-                                    unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 2) {
-    const v256 x = v256_from_v128(v128_load_aligned(src),
-                                  v128_load_aligned(src + sstride));
-    const v256 a = v256_from_v128(v128_load_unaligned(src - 2),
-                                  v128_load_unaligned(src - 2 + sstride));
-    const v256 b = v256_from_v128(v128_load_unaligned(src - 1),
-                                  v128_load_unaligned(src - 1 + sstride));
-    const v256 c = v256_from_v128(v128_load_unaligned(src + 1),
-                                  v128_load_unaligned(src + 1 + sstride));
-    const v256 d = v256_from_v128(v128_load_unaligned(src + 2),
-                                  v128_load_unaligned(src + 2 + sstride));
-    const v128 o = calc_hdelta(x, a, b, c, d, strength, adjdamp);
-
-    v64_store_aligned(dst, v128_high_v64(o));
-    v64_store_aligned(dst + dstride, v128_low_v64(o));
-    src += sstride * 2;
-    dst += dstride * 2;
-  }
-}
-
-// Process blocks of width 4, four lines at a time, 8 bit.
-static void SIMD_FUNC(clpf_hblock4)(uint8_t *dst, const uint16_t *src,
-                                    int dstride, int sstride, int sizey,
-                                    unsigned int strength,
-                                    unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 4) {
-    const v256 a = v256_from_v64(v64_load_unaligned(src - 2),
-                                 v64_load_unaligned(src + sstride - 2),
-                                 v64_load_unaligned(src + 2 * sstride - 2),
-                                 v64_load_unaligned(src + 3 * sstride - 2));
-    const v256 b = v256_from_v64(v64_load_unaligned(src - 1),
-                                 v64_load_unaligned(src + sstride - 1),
-                                 v64_load_unaligned(src + 2 * sstride - 1),
-                                 v64_load_unaligned(src + 3 * sstride - 1));
-    const v256 c = v256_from_v64(v64_load_unaligned(src + 1),
-                                 v64_load_unaligned(src + sstride + 1),
-                                 v64_load_unaligned(src + 2 * sstride + 1),
-                                 v64_load_unaligned(src + 3 * sstride + 1));
-    const v256 d = v256_from_v64(v64_load_unaligned(src + 2),
-                                 v64_load_unaligned(src + sstride + 2),
-                                 v64_load_unaligned(src + 2 * sstride + 2),
-                                 v64_load_unaligned(src + 3 * sstride + 2));
-
-    const v128 o = calc_hdelta(
-        v256_from_v64(v64_load_aligned(src), v64_load_aligned(src + sstride),
-                      v64_load_aligned(src + 2 * sstride),
-                      v64_load_aligned(src + 3 * sstride)),
-        a, b, c, d, strength, adjdamp);
-
-    u32_store_aligned(dst, v128_low_u32(v128_shr_n_byte(o, 12)));
-    u32_store_aligned(dst + dstride, v128_low_u32(v128_shr_n_byte(o, 8)));
-    u32_store_aligned(dst + 2 * dstride, v128_low_u32(v128_shr_n_byte(o, 4)));
-    u32_store_aligned(dst + 3 * dstride, v128_low_u32(o));
-
-    dst += 4 * dstride;
-    src += 4 * sstride;
-  }
-}
-
-void SIMD_FUNC(aom_clpf_block)(uint8_t *dst, const uint16_t *src, int dstride,
-                               int sstride, int sizex, int sizey,
-                               unsigned int strength, unsigned int dmp) {
-  if ((sizex != 4 && sizex != 8) || ((sizey & 3) && sizex == 4)) {
-    // Fallback to C for odd sizes:
-    // * block widths not 4 or 8
-    // * block heights not a multiple of 4 if the block width is 4
-    aom_clpf_block_c(dst, src, dstride, sstride, sizex, sizey, strength, dmp);
-  } else {
-    (sizex == 4 ? SIMD_FUNC(clpf_block4) : SIMD_FUNC(clpf_block8))(
-        dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
-  }
-}
-
-void SIMD_FUNC(aom_clpf_hblock)(uint8_t *dst, const uint16_t *src, int dstride,
-                                int sstride, int sizex, int sizey,
-                                unsigned int strength, unsigned int dmp) {
-  if ((sizex != 4 && sizex != 8) || ((sizey & 3) && sizex == 4)) {
-    // Fallback to C for odd sizes:
-    // * block widths not 4 or 8
-    // * block heights not a multiple of 4 if the block width is 4
-    aom_clpf_hblock_c(dst, src, dstride, sstride, sizex, sizey, strength, dmp);
-  } else {
-    (sizex == 4 ? SIMD_FUNC(clpf_hblock4) : SIMD_FUNC(clpf_hblock8))(
-        dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
-  }
-}
-
-// delta = 1/16 * constrain(a, x, s, d) + 3/16 * constrain(b, x, s, d) +
-//         1/16 * constrain(c, x, s, d) + 3/16 * constrain(d, x, s, d) +
-//         3/16 * constrain(e, x, s, d) + 1/16 * constrain(f, x, s, d) +
-//         3/16 * constrain(g, x, s, d) + 1/16 * constrain(h, x, s, d)
-SIMD_INLINE v128 calc_delta_hbd(v128 x, v128 a, v128 b, v128 c, v128 d, v128 e,
-                                v128 f, v128 g, v128 h, unsigned int s,
-                                unsigned int dmp) {
-  const v128 bdeg = v128_add_16(
-      v128_add_16(constrain16(b, x, s, dmp), constrain16(d, x, s, dmp)),
-      v128_add_16(constrain16(e, x, s, dmp), constrain16(g, x, s, dmp)));
-  const v128 delta = v128_add_16(
-      v128_add_16(
-          v128_add_16(constrain16(a, x, s, dmp), constrain16(c, x, s, dmp)),
-          v128_add_16(constrain16(f, x, s, dmp), constrain16(h, x, s, dmp))),
-      v128_add_16(v128_add_16(bdeg, bdeg), bdeg));
-  return v128_add_16(
-      x,
-      v128_shr_s16(
-          v128_add_16(v128_dup_16(8),
-                      v128_add_16(delta, v128_cmplt_s16(delta, v128_zero()))),
-          4));
-}
-
-static void calc_delta_hbd4(v128 o, v128 a, v128 b, v128 c, v128 d, v128 e,
-                            v128 f, v128 g, v128 h, uint16_t *dst,
-                            unsigned int s, unsigned int dmp, int dstride) {
-  o = calc_delta_hbd(o, a, b, c, d, e, f, g, h, s, dmp);
-  v64_store_aligned(dst, v128_high_v64(o));
-  v64_store_aligned(dst + dstride, v128_low_v64(o));
-}
-
-static void calc_delta_hbd8(v128 o, v128 a, v128 b, v128 c, v128 d, v128 e,
-                            v128 f, v128 g, v128 h, uint16_t *dst,
-                            unsigned int s, unsigned int adjdamp) {
-  v128_store_aligned(dst,
-                     calc_delta_hbd(o, a, b, c, d, e, f, g, h, s, adjdamp));
-}
-
-// delta = 1/16 * constrain(a, x, s, dmp) + 3/16 * constrain(b, x, s, dmp) +
-//         3/16 * constrain(c, x, s, dmp) + 1/16 * constrain(d, x, s, dmp)
-SIMD_INLINE v128 calc_hdelta_hbd(v128 x, v128 a, v128 b, v128 c, v128 d,
-                                 unsigned int s, unsigned int dmp) {
-  const v128 bc =
-      v128_add_16(constrain16(b, x, s, dmp), constrain16(c, x, s, dmp));
-  const v128 delta = v128_add_16(
-      v128_add_16(constrain16(a, x, s, dmp), constrain16(d, x, s, dmp)),
-      v128_add_16(v128_add_16(bc, bc), bc));
-  return v128_add_16(
-      x,
-      v128_shr_s16(
-          v128_add_16(v128_dup_16(4),
-                      v128_add_16(delta, v128_cmplt_s16(delta, v128_zero()))),
-          3));
-}
-
-static void calc_hdelta_hbd4(v128 o, v128 a, v128 b, v128 c, v128 d,
-                             uint16_t *dst, unsigned int s,
-                             unsigned int adjdamp, int dstride) {
-  o = calc_hdelta_hbd(o, a, b, c, d, s, adjdamp);
-  v64_store_aligned(dst, v128_high_v64(o));
-  v64_store_aligned(dst + dstride, v128_low_v64(o));
-}
-
-static void calc_hdelta_hbd8(v128 o, v128 a, v128 b, v128 c, v128 d,
-                             uint16_t *dst, unsigned int s,
-                             unsigned int adjdamp) {
-  v128_store_aligned(dst, calc_hdelta_hbd(o, a, b, c, d, s, adjdamp));
-}
-
-// Process blocks of width 4, two lines at time.
-static void SIMD_FUNC(clpf_block_hbd4)(uint16_t *dst, const uint16_t *src,
-                                       int dstride, int sstride, int sizey,
-                                       unsigned int strength,
-                                       unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 2) {
-    const v64 l1 = v64_load_aligned(src);
-    const v64 l2 = v64_load_aligned(src + sstride);
-    const v64 l3 = v64_load_aligned(src - sstride);
-    const v64 l4 = v64_load_aligned(src + 2 * sstride);
-    const v128 a = v128_from_v64(v64_load_aligned(src - 2 * sstride), l3);
-    const v128 b = v128_from_v64(l3, l1);
-    const v128 g = v128_from_v64(l2, l4);
-    const v128 h = v128_from_v64(l4, v64_load_aligned(src + 3 * sstride));
-    const v128 c = v128_from_v64(v64_load_unaligned(src - 2),
-                                 v64_load_unaligned(src - 2 + sstride));
-    const v128 d = v128_from_v64(v64_load_unaligned(src - 1),
-                                 v64_load_unaligned(src - 1 + sstride));
-    const v128 e = v128_from_v64(v64_load_unaligned(src + 1),
-                                 v64_load_unaligned(src + 1 + sstride));
-    const v128 f = v128_from_v64(v64_load_unaligned(src + 2),
-                                 v64_load_unaligned(src + 2 + sstride));
-
-    calc_delta_hbd4(v128_from_v64(l1, l2), a, b, c, d, e, f, g, h, dst,
-                    strength, adjdamp, dstride);
-    src += sstride * 2;
-    dst += dstride * 2;
-  }
-}
-
-// The most simple case.  Start here if you need to understand the functions.
-static void SIMD_FUNC(clpf_block_hbd)(uint16_t *dst, const uint16_t *src,
-                                      int dstride, int sstride, int sizey,
-                                      unsigned int strength,
-                                      unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y++) {
-    const v128 o = v128_load_aligned(src);
-    const v128 a = v128_load_aligned(src - 2 * sstride);
-    const v128 b = v128_load_aligned(src - 1 * sstride);
-    const v128 g = v128_load_aligned(src + sstride);
-    const v128 h = v128_load_aligned(src + 2 * sstride);
-    const v128 c = v128_load_unaligned(src - 2);
-    const v128 d = v128_load_unaligned(src - 1);
-    const v128 e = v128_load_unaligned(src + 1);
-    const v128 f = v128_load_unaligned(src + 2);
-
-    calc_delta_hbd8(o, a, b, c, d, e, f, g, h, dst, strength, adjdamp);
-    src += sstride;
-    dst += dstride;
-  }
-}
-
-// Process blocks of width 4, horizontal filter, two lines at time.
-static void SIMD_FUNC(clpf_hblock_hbd4)(uint16_t *dst, const uint16_t *src,
-                                        int dstride, int sstride, int sizey,
-                                        unsigned int strength,
-                                        unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y += 2) {
-    const v128 a = v128_from_v64(v64_load_unaligned(src - 2),
-                                 v64_load_unaligned(src - 2 + sstride));
-    const v128 b = v128_from_v64(v64_load_unaligned(src - 1),
-                                 v64_load_unaligned(src - 1 + sstride));
-    const v128 c = v128_from_v64(v64_load_unaligned(src + 1),
-                                 v64_load_unaligned(src + 1 + sstride));
-    const v128 d = v128_from_v64(v64_load_unaligned(src + 2),
-                                 v64_load_unaligned(src + 2 + sstride));
-
-    calc_hdelta_hbd4(v128_from_v64(v64_load_unaligned(src),
-                                   v64_load_unaligned(src + sstride)),
-                     a, b, c, d, dst, strength, adjdamp, dstride);
-    src += sstride * 2;
-    dst += dstride * 2;
-  }
-}
-
-// Process blocks of width 8, horizontal filter, two lines at time.
-static void SIMD_FUNC(clpf_hblock_hbd)(uint16_t *dst, const uint16_t *src,
-                                       int dstride, int sstride, int sizey,
-                                       unsigned int strength,
-                                       unsigned int adjdamp) {
-  int y;
-
-  for (y = 0; y < sizey; y++) {
-    const v128 o = v128_load_aligned(src);
-    const v128 a = v128_load_unaligned(src - 2);
-    const v128 b = v128_load_unaligned(src - 1);
-    const v128 c = v128_load_unaligned(src + 1);
-    const v128 d = v128_load_unaligned(src + 2);
-
-    calc_hdelta_hbd8(o, a, b, c, d, dst, strength, adjdamp);
-    src += sstride;
-    dst += dstride;
-  }
-}
-
-void SIMD_FUNC(aom_clpf_block_hbd)(uint16_t *dst, const uint16_t *src,
-                                   int dstride, int sstride, int sizex,
-                                   int sizey, unsigned int strength,
-                                   unsigned int dmp) {
-  if ((sizex != 4 && sizex != 8) || ((sizey & 1) && sizex == 4)) {
-    // Fallback to C for odd sizes:
-    // * block width not 4 or 8
-    // * block heights not a multiple of 2 if the block width is 4
-    aom_clpf_block_hbd_c(dst, src, dstride, sstride, sizex, sizey, strength,
-                         dmp);
-  } else {
-    (sizex == 4 ? SIMD_FUNC(clpf_block_hbd4) : SIMD_FUNC(clpf_block_hbd))(
-        dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
-  }
-}
-
-void SIMD_FUNC(aom_clpf_hblock_hbd)(uint16_t *dst, const uint16_t *src,
-                                    int dstride, int sstride, int sizex,
-                                    int sizey, unsigned int strength,
-                                    unsigned int dmp) {
-  if ((sizex != 4 && sizex != 8) || ((sizey & 1) && sizex == 4)) {
-    // Fallback to C for odd sizes:
-    // * block width not 4 or 8
-    // * block heights not a multiple of 2 if the block width is 4
-    aom_clpf_hblock_hbd_c(dst, src, dstride, sstride, sizex, sizey, strength,
-                          dmp);
-  } else {
-    (sizex == 4 ? SIMD_FUNC(clpf_hblock_hbd4) : SIMD_FUNC(clpf_hblock_hbd))(
-        dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
-  }
-}

third_party/aom/av1/common/clpf_sse2.c

@@ -1,14 +0,0 @@
-/*
- * 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_dsp/aom_simd.h"
-#define SIMD_FUNC(name) name##_sse2
-#include "./clpf_simd.h"

third_party/aom/av1/common/clpf_sse4.c

@@ -1,14 +0,0 @@
-/*
- * 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_dsp/aom_simd.h"
-#define SIMD_FUNC(name) name##_sse4_1
-#include "./clpf_simd.h"

third_party/aom/av1/common/clpf_ssse3.c

@@ -1,14 +0,0 @@
-/*
- * 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_dsp/aom_simd.h"
-#define SIMD_FUNC(name) name##_ssse3
-#include "./clpf_simd.h"

third_party/aom/av1/common/common.h

@@ -20,6 +20,7 @@
 #include "aom_mem/aom_mem.h"
 #include "aom/aom_integer.h"
 #include "aom_ports/bitops.h"
+#include "config/aom_config.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -53,6 +54,8 @@ static INLINE int get_unsigned_bits(unsigned int num_values) {
 
 #define AOM_FRAME_MARKER 0x2
 
+#define AV1_MIN_TILE_SIZE_BYTES 1
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/convolve.h

@@ -17,140 +17,119 @@
 extern "C" {
 #endif
 
-typedef enum CONVOLVE_OPT {
-  // indicate the results in dst buf is rounded by FILTER_BITS or not
-  CONVOLVE_OPT_ROUND,
-  CONVOLVE_OPT_NO_ROUND,
-} CONVOLVE_OPT;
-
-typedef int32_t CONV_BUF_TYPE;
-
+typedef uint16_t CONV_BUF_TYPE;
 typedef struct ConvolveParams {
   int ref;
   int do_average;
-  CONVOLVE_OPT round;
   CONV_BUF_TYPE *dst;
   int dst_stride;
   int round_0;
   int round_1;
   int plane;
-  int do_post_rounding;
+  int is_compound;
+  int use_jnt_comp_avg;
+  int fwd_offset;
+  int bck_offset;
 } ConvolveParams;
 
-static INLINE ConvolveParams get_conv_params(int ref, int do_average,
-                                             int plane) {
-  ConvolveParams conv_params;
-  conv_params.ref = ref;
-  conv_params.do_average = do_average;
-  conv_params.round = CONVOLVE_OPT_ROUND;
-  conv_params.plane = plane;
-  conv_params.do_post_rounding = 0;
-  return conv_params;
-}
+#define ROUND0_BITS 3
+#define COMPOUND_ROUND1_BITS 7
+#define WIENER_ROUND0_BITS 3
 
-#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
-static INLINE void av1_convolve_filter_params_fixup_1212(
-    const InterpFilterParams *params_x, InterpFilterParams *params_y) {
-  if (params_x->interp_filter == MULTITAP_SHARP &&
-      params_y->interp_filter == MULTITAP_SHARP) {
-    // Avoid two directions both using 12-tap filter.
-    // This will reduce hardware implementation cost.
-    *params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
-  }
-}
-#endif
+#define WIENER_CLAMP_LIMIT(r0, bd) (1 << ((bd) + 1 + FILTER_BITS - r0))
 
-static INLINE void av1_get_convolve_filter_params(
-    InterpFilters interp_filters, int avoid_1212, InterpFilterParams *params_x,
-    InterpFilterParams *params_y) {
-#if CONFIG_DUAL_FILTER
+typedef void (*aom_convolve_fn_t)(const uint8_t *src, int src_stride,
+                                  uint8_t *dst, int dst_stride, int w, int h,
+                                  InterpFilterParams *filter_params_x,
+                                  InterpFilterParams *filter_params_y,
+                                  const int subpel_x_q4, const int subpel_y_q4,
+                                  ConvolveParams *conv_params);
+
+typedef void (*aom_highbd_convolve_fn_t)(
+    const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w,
+    int h, InterpFilterParams *filter_params_x,
+    InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+static INLINE void av1_get_convolve_filter_params(InterpFilters interp_filters,
+                                                  InterpFilterParams *params_x,
+                                                  InterpFilterParams *params_y,
+                                                  int w, int h) {
   InterpFilter filter_x = av1_extract_interp_filter(interp_filters, 1);
   InterpFilter filter_y = av1_extract_interp_filter(interp_filters, 0);
-#else
-  InterpFilter filter_x = av1_extract_interp_filter(interp_filters, 0);
-  InterpFilter filter_y = av1_extract_interp_filter(interp_filters, 0);
-#endif
-
-  *params_x = av1_get_interp_filter_params(filter_x);
-  *params_y = av1_get_interp_filter_params(filter_y);
-
-  if (avoid_1212) {
-#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
-    convolve_filter_params_fixup_1212(params_x, params_y);
-#endif
-  }
+  *params_x = av1_get_interp_filter_params_with_block_size(filter_x, w);
+  *params_y = av1_get_interp_filter_params_with_block_size(filter_y, h);
 }
 
 struct AV1Common;
-void av1_convolve_init(struct AV1Common *cm);
+struct scale_factors;
 
-#if CONFIG_CONVOLVE_ROUND
 void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
                             int dst_stride, int w, int h,
                             InterpFilters interp_filters, const int subpel_x_q4,
                             int x_step_q4, const int subpel_y_q4, int y_step_q4,
-                            int scaled, ConvolveParams *conv_params);
+                            int scaled, ConvolveParams *conv_params,
+                            const struct scale_factors *sf);
 
 static INLINE ConvolveParams get_conv_params_no_round(int ref, int do_average,
-                                                      int plane, int32_t *dst,
-                                                      int dst_stride) {
+                                                      int plane,
+                                                      CONV_BUF_TYPE *dst,
+                                                      int dst_stride,
+                                                      int is_compound, int bd) {
   ConvolveParams conv_params;
   conv_params.ref = ref;
   conv_params.do_average = do_average;
-  conv_params.round = CONVOLVE_OPT_NO_ROUND;
-#if CONFIG_COMPOUND_ROUND
-  conv_params.round_0 = FILTER_BITS;
-#else
-  conv_params.round_0 = 5;
-#endif
-  conv_params.round_1 = 0;
+  assert(IMPLIES(do_average, is_compound));
+  conv_params.is_compound = is_compound;
+  conv_params.round_0 = ROUND0_BITS;
+  conv_params.round_1 = is_compound ? COMPOUND_ROUND1_BITS
+                                    : 2 * FILTER_BITS - conv_params.round_0;
+  const int intbufrange = bd + FILTER_BITS - conv_params.round_0 + 2;
+  assert(IMPLIES(bd < 12, intbufrange <= 16));
+  if (intbufrange > 16) {
+    conv_params.round_0 += intbufrange - 16;
+    if (!is_compound) conv_params.round_1 -= intbufrange - 16;
+  }
+  // TODO(yunqing): The following dst should only be valid while
+  // is_compound = 1;
   conv_params.dst = dst;
   conv_params.dst_stride = dst_stride;
   conv_params.plane = plane;
-  conv_params.do_post_rounding = 0;
   return conv_params;
 }
 
-#if CONFIG_HIGHBITDEPTH
+static INLINE ConvolveParams get_conv_params(int ref, int do_average, int plane,
+                                             int bd) {
+  return get_conv_params_no_round(ref, do_average, plane, NULL, 0, 0, bd);
+}
+
+static INLINE ConvolveParams get_conv_params_wiener(int bd) {
+  ConvolveParams conv_params;
+  (void)bd;
+  conv_params.ref = 0;
+  conv_params.do_average = 0;
+  conv_params.is_compound = 0;
+  conv_params.round_0 = WIENER_ROUND0_BITS;
+  conv_params.round_1 = 2 * FILTER_BITS - conv_params.round_0;
+  const int intbufrange = bd + FILTER_BITS - conv_params.round_0 + 2;
+  assert(IMPLIES(bd < 12, intbufrange <= 16));
+  if (intbufrange > 16) {
+    conv_params.round_0 += intbufrange - 16;
+    conv_params.round_1 -= intbufrange - 16;
+  }
+  conv_params.dst = NULL;
+  conv_params.dst_stride = 0;
+  conv_params.plane = 0;
+  return conv_params;
+}
+
 void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
                                    uint8_t *dst, int dst_stride, int w, int h,
                                    InterpFilters interp_filters,
                                    const int subpel_x_q4, int x_step_q4,
                                    const int subpel_y_q4, int y_step_q4,
                                    int scaled, ConvolveParams *conv_params,
-                                   int bd);
-#endif
-#endif  // CONFIG_CONVOLVE_ROUND
-
-void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
-                  int dst_stride, int w, int h, InterpFilters interp_filters,
-                  const int subpel_x, int xstep, const int subpel_y, int ystep,
-                  ConvolveParams *conv_params);
-
-void av1_convolve_c(const uint8_t *src, int src_stride, uint8_t *dst,
-                    int dst_stride, int w, int h, InterpFilters interp_filters,
-                    const int subpel_x, int xstep, const int subpel_y,
-                    int ystep, ConvolveParams *conv_params);
-
-void av1_convolve_scale(const uint8_t *src, int src_stride, uint8_t *dst,
-                        int dst_stride, int w, int h,
-                        InterpFilters interp_filters, const int subpel_x,
-                        int xstep, const int subpel_y, int ystep,
-                        ConvolveParams *conv_params);
-
-#if CONFIG_HIGHBITDEPTH
-void av1_highbd_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
-                         int dst_stride, int w, int h,
-                         InterpFilters interp_filters, const int subpel_x,
-                         int xstep, const int subpel_y, int ystep, int avg,
-                         int bd);
-
-void av1_highbd_convolve_scale(const uint8_t *src, int src_stride, uint8_t *dst,
-                               int dst_stride, int w, int h,
-                               InterpFilters interp_filters, const int subpel_x,
-                               int xstep, const int subpel_y, int ystep,
-                               int avg, int bd);
-#endif  // CONFIG_HIGHBITDEPTH
+                                   const struct scale_factors *sf, int bd);
 
 #ifdef __cplusplus
 }  // extern "C"

third_party/aom/av1/common/daala_tx.h

@@ -1,53 +0,0 @@
-#ifndef AOM_DSP_DAALA_TX_H_
-#define AOM_DSP_DAALA_TX_H_
-
-#include "aom_dsp/aom_dsp_common.h"
-#include "av1/common/odintrin.h"
-
-void daala_fdct4(const tran_low_t *input, tran_low_t *output);
-void daala_idct4(const tran_low_t *input, tran_low_t *output);
-void daala_fdst4(const tran_low_t *input, tran_low_t *output);
-void daala_idst4(const tran_low_t *input, tran_low_t *output);
-void daala_idtx4(const tran_low_t *input, tran_low_t *output);
-void daala_fdct8(const tran_low_t *input, tran_low_t *output);
-void daala_idct8(const tran_low_t *input, tran_low_t *output);
-void daala_fdst8(const tran_low_t *input, tran_low_t *output);
-void daala_idst8(const tran_low_t *input, tran_low_t *output);
-void daala_idtx8(const tran_low_t *input, tran_low_t *output);
-void daala_fdct16(const tran_low_t *input, tran_low_t *output);
-void daala_idct16(const tran_low_t *input, tran_low_t *output);
-void daala_fdst16(const tran_low_t *input, tran_low_t *output);
-void daala_idst16(const tran_low_t *input, tran_low_t *output);
-void daala_idtx16(const tran_low_t *input, tran_low_t *output);
-void daala_fdct32(const tran_low_t *input, tran_low_t *output);
-void daala_idct32(const tran_low_t *input, tran_low_t *output);
-void daala_fdst32(const tran_low_t *input, tran_low_t *output);
-void daala_idst32(const tran_low_t *input, tran_low_t *output);
-void daala_idtx32(const tran_low_t *input, tran_low_t *output);
-#if CONFIG_TX64X64
-void daala_fdct64(const tran_low_t *input, tran_low_t *output);
-void daala_idct64(const tran_low_t *input, tran_low_t *output);
-void daala_fdst64(const tran_low_t *input, tran_low_t *output);
-void daala_idst64(const tran_low_t *input, tran_low_t *output);
-void daala_idtx64(const tran_low_t *input, tran_low_t *output);
-#endif
-
-void od_bin_fdct4(od_coeff y[4], const od_coeff *x, int xstride);
-void od_bin_idct4(od_coeff *x, int xstride, const od_coeff y[4]);
-void od_bin_fdst4(od_coeff y[4], const od_coeff *x, int xstride);
-void od_bin_idst4(od_coeff *x, int xstride, const od_coeff y[4]);
-void od_bin_fdct8(od_coeff y[8], const od_coeff *x, int xstride);
-void od_bin_idct8(od_coeff *x, int xstride, const od_coeff y[8]);
-void od_bin_fdst8(od_coeff y[8], const od_coeff *x, int xstride);
-void od_bin_idst8(od_coeff *x, int xstride, const od_coeff y[8]);
-void od_bin_fdct16(od_coeff y[16], const od_coeff *x, int xstride);
-void od_bin_idct16(od_coeff *x, int xstride, const od_coeff y[16]);
-void od_bin_fdst16(od_coeff y[16], const od_coeff *x, int xstride);
-void od_bin_idst16(od_coeff *x, int xstride, const od_coeff y[16]);
-void od_bin_fdct32(od_coeff y[32], const od_coeff *x, int xstride);
-void od_bin_idct32(od_coeff *x, int xstride, const od_coeff y[32]);
-#if CONFIG_TX64X64
-void od_bin_fdct64(od_coeff y[64], const od_coeff *x, int xstride);
-void od_bin_idct64(od_coeff *x, int xstride, const od_coeff y[64]);
-#endif
-#endif

third_party/aom/av1/common/debugmodes.c

@@ -27,7 +27,7 @@ static void log_frame_info(AV1_COMMON *cm, const char *str, FILE *f) {
 static void print_mi_data(AV1_COMMON *cm, FILE *file, const char *descriptor,
                           size_t member_offset) {
   int mi_row, mi_col;
-  MODE_INFO **mi = cm->mi_grid_visible;
+  MB_MODE_INFO **mi = cm->mi_grid_visible;
   int rows = cm->mi_rows;
   int cols = cm->mi_cols;
   char prefix = descriptor[0];
@@ -36,8 +36,7 @@ static void print_mi_data(AV1_COMMON *cm, FILE *file, const char *descriptor,
   for (mi_row = 0; mi_row < rows; mi_row++) {
     fprintf(file, "%c ", prefix);
     for (mi_col = 0; mi_col < cols; mi_col++) {
-      fprintf(file, "%2d ",
-              *((char *)((char *)(&mi[0]->mbmi) + member_offset)));
+      fprintf(file, "%2d ", *((char *)((char *)(mi[0]) + member_offset)));
       mi++;
     }
     fprintf(file, "\n");
@@ -50,7 +49,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
   int mi_row;
   int mi_col;
   FILE *mvs = fopen(file, "a");
-  MODE_INFO **mi = cm->mi_grid_visible;
+  MB_MODE_INFO **mi = cm->mi_grid_visible;
   int rows = cm->mi_rows;
   int cols = cm->mi_cols;
 
@@ -65,7 +64,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
   for (mi_row = 0; mi_row < rows; mi_row++) {
     fprintf(mvs, "S ");
     for (mi_col = 0; mi_col < cols; mi_col++) {
-      fprintf(mvs, "%2d ", mi[0]->mbmi.skip);
+      fprintf(mvs, "%2d ", mi[0]->skip);
       mi++;
     }
     fprintf(mvs, "\n");
@@ -79,8 +78,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
   for (mi_row = 0; mi_row < rows; mi_row++) {
     fprintf(mvs, "V ");
     for (mi_col = 0; mi_col < cols; mi_col++) {
-      fprintf(mvs, "%4d:%4d ", mi[0]->mbmi.mv[0].as_mv.row,
-              mi[0]->mbmi.mv[0].as_mv.col);
+      fprintf(mvs, "%4d:%4d ", mi[0]->mv[0].as_mv.row, mi[0]->mv[0].as_mv.col);
       mi++;
     }
     fprintf(mvs, "\n");
@@ -90,3 +88,20 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
 
   fclose(mvs);
 }
+
+void av1_print_uncompressed_frame_header(const uint8_t *data, int size,
+                                         const char *filename) {
+  FILE *hdrFile = fopen(filename, "w");
+  fwrite(data, size, sizeof(uint8_t), hdrFile);
+  fclose(hdrFile);
+}
+
+void av1_print_frame_contexts(const FRAME_CONTEXT *fc, const char *filename) {
+  FILE *fcFile = fopen(filename, "w");
+  const uint16_t *fcp = (uint16_t *)fc;
+  const unsigned int n_contexts = sizeof(FRAME_CONTEXT) / sizeof(uint16_t);
+  unsigned int i;
+
+  for (i = 0; i < n_contexts; ++i) fprintf(fcFile, "%d ", *fcp++);
+  fclose(fcFile);
+}

third_party/aom/av1/common/entropy.h

@@ -12,7 +12,8 @@
 #ifndef AV1_COMMON_ENTROPY_H_
 #define AV1_COMMON_ENTROPY_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom/aom_integer.h"
 #include "aom_dsp/prob.h"
 
@@ -24,82 +25,35 @@
 extern "C" {
 #endif
 
-#define DIFF_UPDATE_PROB 252
-#define GROUP_DIFF_UPDATE_PROB 252
-
-#if CONFIG_Q_ADAPT_PROBS
 #define TOKEN_CDF_Q_CTXS 4
-#endif  // CONFIG_Q_ADAPT_PROBS
 
-// Coefficient token alphabet
-#define ZERO_TOKEN 0        // 0     Extra Bits 0+0
-#define ONE_TOKEN 1         // 1     Extra Bits 0+1
-#define TWO_TOKEN 2         // 2     Extra Bits 0+1
-#define THREE_TOKEN 3       // 3     Extra Bits 0+1
-#define FOUR_TOKEN 4        // 4     Extra Bits 0+1
-#define CATEGORY1_TOKEN 5   // 5-6   Extra Bits 1+1
-#define CATEGORY2_TOKEN 6   // 7-10  Extra Bits 2+1
-#define CATEGORY3_TOKEN 7   // 11-18 Extra Bits 3+1
-#define CATEGORY4_TOKEN 8   // 19-34 Extra Bits 4+1
-#define CATEGORY5_TOKEN 9   // 35-66 Extra Bits 5+1
-#define CATEGORY6_TOKEN 10  // 67+   Extra Bits 14+1
-#define EOB_TOKEN 11        // EOB   Extra Bits 0+0
-#define NO_EOB 0            // Not an end-of-block
-#define EARLY_EOB 1         // End of block before the last position
-#define LAST_EOB 2          // End of block in the last position (implicit)
-#define BLOCK_Z_TOKEN 255   // block zero
-#define HEAD_TOKENS 5
-#define TAIL_TOKENS 9
-#define ONE_TOKEN_EOB 1
-#define ONE_TOKEN_NEOB 2
-#define TWO_TOKEN_PLUS_EOB 3
-#define TWO_TOKEN_PLUS_NEOB 4
-#define ENTROPY_TOKENS 12
-
-#define ENTROPY_NODES 11
-
-#if CONFIG_LV_MAP
 #define TXB_SKIP_CONTEXTS 13
 
-#if CONFIG_CTX1D
-#define EOB_COEF_CONTEXTS_2D 25
-#define EOB_COEF_CONTEXTS_1D 25
-#define EOB_COEF_CONTEXTS \
-  (EOB_COEF_CONTEXTS_2D + EOB_COEF_CONTEXTS_1D + EOB_COEF_CONTEXTS_1D)
-#else  // CONFIG_CTX1D
-#define EOB_COEF_CONTEXTS 25
-#endif  // CONFIG_CTX1D
+#define EOB_COEF_CONTEXTS 9
 
-#if CONFIG_EXT_TX
-#define SIG_COEF_CONTEXTS_2D 16
+#define SIG_COEF_CONTEXTS_2D 26
 #define SIG_COEF_CONTEXTS_1D 16
-#define SIG_COEF_CONTEXTS \
-  (SIG_COEF_CONTEXTS_2D + SIG_COEF_CONTEXTS_1D + SIG_COEF_CONTEXTS_1D)
-#else  // CONFIG_EXT_TX
-#define SIG_COEF_CONTEXTS_2D 16
-#define SIG_COEF_CONTEXTS 16
-#endif  // CONFIG_EXT_TX
-#define COEFF_BASE_CONTEXTS 42
+#define SIG_COEF_CONTEXTS_EOB 4
+#define SIG_COEF_CONTEXTS (SIG_COEF_CONTEXTS_2D + SIG_COEF_CONTEXTS_1D)
+
+#define COEFF_BASE_CONTEXTS (SIG_COEF_CONTEXTS)
 #define DC_SIGN_CONTEXTS 3
 
 #define BR_TMP_OFFSET 12
 #define BR_REF_CAT 4
-#define LEVEL_CONTEXTS (BR_TMP_OFFSET * BR_REF_CAT)
+#define LEVEL_CONTEXTS 21
 
 #define NUM_BASE_LEVELS 2
-#define COEFF_BASE_RANGE (16 - NUM_BASE_LEVELS)
-#define BASE_RANGE_SETS 3
+
+#define BR_CDF_SIZE (4)
+#define COEFF_BASE_RANGE (4 * (BR_CDF_SIZE - 1))
 
 #define COEFF_CONTEXT_BITS 6
 #define COEFF_CONTEXT_MASK ((1 << COEFF_CONTEXT_BITS) - 1)
+#define MAX_BASE_BR_RANGE (COEFF_BASE_RANGE + NUM_BASE_LEVELS + 1)
 
 #define BASE_CONTEXT_POSITION_NUM 12
 
-#if CONFIG_CTX1D
-#define EMPTY_LINE_CONTEXTS 5
-#define HV_EOB_CONTEXTS 24
-#endif  // CONFIG_CTX1D
-
 typedef enum TX_CLASS {
   TX_CLASS_2D = 0,
   TX_CLASS_HORIZ = 1,
@@ -107,161 +61,21 @@ typedef enum TX_CLASS {
   TX_CLASSES = 3,
 } TX_CLASS;
 
-#endif
-
-DECLARE_ALIGNED(16, extern const uint8_t, av1_pt_energy_class[ENTROPY_TOKENS]);
-
-#define CAT1_MIN_VAL 5
-#define CAT2_MIN_VAL 7
-#define CAT3_MIN_VAL 11
-#define CAT4_MIN_VAL 19
-#define CAT5_MIN_VAL 35
-#define CAT6_MIN_VAL 67
-
-// Extra bit probabilities.
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat1_prob[1]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat2_prob[2]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat3_prob[3]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat4_prob[4]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat5_prob[5]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_cat6_prob[18]);
-#if CONFIG_NEW_MULTISYMBOL
-extern const aom_cdf_prob *av1_cat1_cdf[];
-extern const aom_cdf_prob *av1_cat2_cdf[];
-extern const aom_cdf_prob *av1_cat3_cdf[];
-extern const aom_cdf_prob *av1_cat4_cdf[];
-extern const aom_cdf_prob *av1_cat5_cdf[];
-extern const aom_cdf_prob *av1_cat6_cdf[];
-#endif
-
-#define EOB_MODEL_TOKEN 3
-
-typedef struct {
-#if CONFIG_NEW_MULTISYMBOL
-  const aom_cdf_prob **cdf;
-#else
-  const aom_prob *prob;
-#endif
-  int len;
-  int base_val;
-  const int16_t *cost;
-} av1_extra_bit;
-
-// indexed by token value
-extern const av1_extra_bit av1_extra_bits[ENTROPY_TOKENS];
-
-static INLINE int av1_get_cat6_extrabits_size(TX_SIZE tx_size,
-                                              aom_bit_depth_t bit_depth) {
-  tx_size = txsize_sqr_up_map[tx_size];
-#if CONFIG_TX64X64
-  // TODO(debargha): Does TX_64X64 require an additional extrabit?
-  if (tx_size > TX_32X32) tx_size = TX_32X32;
-#endif
-#if CONFIG_CHROMA_2X2
-  int tx_offset = (tx_size < TX_4X4) ? 0 : (int)(tx_size - TX_4X4);
-#else
-  int tx_offset = (int)(tx_size - TX_4X4);
-#endif
-  int bits = (int)bit_depth + 3 + tx_offset;
-#if CONFIG_NEW_MULTISYMBOL
-  // Round up
-  bits = AOMMIN((int)sizeof(av1_cat6_prob), ((bits + 3) & ~3));
-#endif
-  assert(bits <= (int)sizeof(av1_cat6_prob));
-  return bits;
-}
-
 #define DCT_MAX_VALUE 16384
-#if CONFIG_HIGHBITDEPTH
 #define DCT_MAX_VALUE_HIGH10 65536
 #define DCT_MAX_VALUE_HIGH12 262144
-#endif  // CONFIG_HIGHBITDEPTH
-
-/* Coefficients are predicted via a 3-dimensional probability table. */
 
+/* Coefficients are predicted via a 3-dimensional probability table indexed on
+ * REF_TYPES, COEF_BANDS and COEF_CONTEXTS. */
 #define REF_TYPES 2  // intra=0, inter=1
 
-/* Middle dimension reflects the coefficient position within the transform. */
-#define COEF_BANDS 6
-
-/* Inside dimension is measure of nearby complexity, that reflects the energy
-   of nearby coefficients are nonzero.  For the first coefficient (DC, unless
-   block type is 0), we look at the (already encoded) blocks above and to the
-   left of the current block.  The context index is then the number (0,1,or 2)
-   of these blocks having nonzero coefficients.
-   After decoding a coefficient, the measure is determined by the size of the
-   most recently decoded coefficient.
-   Note that the intuitive meaning of this measure changes as coefficients
-   are decoded, e.g., prior to the first token, a zero means that my neighbors
-   are empty while, after the first token, because of the use of end-of-block,
-   a zero means we just decoded a zero and hence guarantees that a non-zero
-   coefficient will appear later in this block.  However, this shift
-   in meaning is perfectly OK because our context depends also on the
-   coefficient band (and since zigzag positions 0, 1, and 2 are in
-   distinct bands). */
-
-#define COEFF_CONTEXTS 6
-#define COEFF_CONTEXTS0 3  // for band 0
-#define BAND_COEFF_CONTEXTS(band) \
-  ((band) == 0 ? COEFF_CONTEXTS0 : COEFF_CONTEXTS)
-
-#define SUBEXP_PARAM 4   /* Subexponential code parameter */
-#define MODULUS_PARAM 13 /* Modulus parameter */
-
 struct AV1Common;
 struct frame_contexts;
+void av1_reset_cdf_symbol_counters(struct frame_contexts *fc);
 void av1_default_coef_probs(struct AV1Common *cm);
-#if CONFIG_LV_MAP
-void av1_adapt_coef_probs(struct AV1Common *cm);
-#endif  // CONFIG_LV_MAP
 
-// This is the index in the scan order beyond which all coefficients for
-// 8x8 transform and above are in the top band.
-// This macro is currently unused but may be used by certain implementations
-#define MAXBAND_INDEX 21
-
-DECLARE_ALIGNED(16, extern const uint8_t,
-                av1_coefband_trans_8x8plus[MAX_TX_SQUARE]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_coefband_trans_4x8_8x4[32]);
-DECLARE_ALIGNED(16, extern const uint8_t, av1_coefband_trans_4x4[16]);
-
-DECLARE_ALIGNED(16, extern const uint16_t, band_count_table[TX_SIZES_ALL][8]);
-DECLARE_ALIGNED(16, extern const uint16_t,
-                band_cum_count_table[TX_SIZES_ALL][8]);
-
-static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) {
-  switch (tx_size) {
-    case TX_4X4: return av1_coefband_trans_4x4;
-    case TX_8X4:
-    case TX_4X8: return av1_coefband_trans_4x8_8x4;
-    default: return av1_coefband_trans_8x8plus;
-  }
-}
-
-// 128 lists of probabilities are stored for the following ONE node probs:
-// 1, 3, 5, 7, ..., 253, 255
-// In between probabilities are interpolated linearly
-
-#define COEFF_PROB_MODELS 255
-
-#define UNCONSTRAINED_NODES 3
-
-#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES)
-#define TAIL_NODES (MODEL_NODES + 1)
-extern const aom_tree_index av1_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)];
-extern const aom_prob av1_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES];
-
-typedef aom_cdf_prob coeff_cdf_model[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
-                                    [CDF_SIZE(ENTROPY_TOKENS)];
-extern const aom_cdf_prob av1_pareto8_token_probs[COEFF_PROB_MODELS]
-                                                 [ENTROPY_TOKENS - 2];
-extern const aom_cdf_prob av1_pareto8_tail_probs[COEFF_PROB_MODELS]
-                                                [ENTROPY_TOKENS - 3];
 struct frame_contexts;
 
-void av1_coef_head_cdfs(struct frame_contexts *fc);
-void av1_coef_pareto_cdfs(struct frame_contexts *fc);
-
 typedef char ENTROPY_CONTEXT;
 
 static INLINE int combine_entropy_contexts(ENTROPY_CONTEXT a,
@@ -273,93 +87,6 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
                                       const ENTROPY_CONTEXT *l) {
   ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
 
-#if CONFIG_CHROMA_2X2
-  switch (tx_size) {
-    case TX_2X2:
-      above_ec = a[0] != 0;
-      left_ec = l[0] != 0;
-      break;
-    case TX_4X4:
-      above_ec = !!*(const uint16_t *)a;
-      left_ec = !!*(const uint16_t *)l;
-      break;
-    case TX_4X8:
-      above_ec = !!*(const uint16_t *)a;
-      left_ec = !!*(const uint32_t *)l;
-      break;
-    case TX_8X4:
-      above_ec = !!*(const uint32_t *)a;
-      left_ec = !!*(const uint16_t *)l;
-      break;
-    case TX_8X8:
-      above_ec = !!*(const uint32_t *)a;
-      left_ec = !!*(const uint32_t *)l;
-      break;
-    case TX_8X16:
-      above_ec = !!*(const uint32_t *)a;
-      left_ec = !!*(const uint64_t *)l;
-      break;
-    case TX_16X8:
-      above_ec = !!*(const uint64_t *)a;
-      left_ec = !!*(const uint32_t *)l;
-      break;
-    case TX_16X16:
-      above_ec = !!*(const uint64_t *)a;
-      left_ec = !!*(const uint64_t *)l;
-      break;
-    case TX_16X32:
-      above_ec = !!*(const uint64_t *)a;
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
-      break;
-    case TX_32X16:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
-      left_ec = !!*(const uint64_t *)l;
-      break;
-    case TX_32X32:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
-      break;
-#if CONFIG_TX64X64
-    case TX_64X64:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8) |
-                    *(const uint64_t *)(a + 16) | *(const uint64_t *)(a + 24));
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8) |
-                   *(const uint64_t *)(l + 16) | *(const uint64_t *)(l + 24));
-      break;
-    case TX_32X64:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8) |
-                   *(const uint64_t *)(l + 16) | *(const uint64_t *)(l + 24));
-      break;
-    case TX_64X32:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8) |
-                    *(const uint64_t *)(a + 16) | *(const uint64_t *)(a + 24));
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
-      break;
-#endif  // CONFIG_TX64X64
-#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-    case TX_4X16:
-      above_ec = !!*(const uint16_t *)a;
-      left_ec = !!*(const uint64_t *)l;
-      break;
-    case TX_16X4:
-      above_ec = !!*(const uint64_t *)a;
-      left_ec = !!*(const uint16_t *)l;
-      break;
-    case TX_8X32:
-      above_ec = !!*(const uint32_t *)a;
-      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
-      break;
-    case TX_32X8:
-      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
-      left_ec = !!*(const uint32_t *)l;
-      break;
-#endif
-    default: assert(0 && "Invalid transform size."); break;
-  }
-  return combine_entropy_contexts(above_ec, left_ec);
-#endif  // CONFIG_CHROMA_2X2
-
   switch (tx_size) {
     case TX_4X4:
       above_ec = a[0] != 0;
@@ -401,7 +128,6 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
       above_ec = !!*(const uint64_t *)a;
       left_ec = !!*(const uint64_t *)l;
       break;
-#if CONFIG_TX64X64
     case TX_64X64:
       above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
       left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
@@ -414,8 +140,6 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
       above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
       left_ec = !!*(const uint64_t *)l;
       break;
-#endif  // CONFIG_TX64X64
-#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
     case TX_4X16:
       above_ec = a[0] != 0;
       left_ec = !!*(const uint32_t *)l;
@@ -432,55 +156,24 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
       above_ec = !!*(const uint64_t *)a;
       left_ec = !!*(const uint16_t *)l;
       break;
-#endif
+    case TX_16X64:
+      above_ec = !!*(const uint32_t *)a;
+      left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
+      break;
+    case TX_64X16:
+      above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
+      left_ec = !!*(const uint32_t *)l;
+      break;
     default: assert(0 && "Invalid transform size."); break;
   }
   return combine_entropy_contexts(above_ec, left_ec);
 }
 
-#define COEF_COUNT_SAT 24
-#define COEF_MAX_UPDATE_FACTOR 112
-#define COEF_COUNT_SAT_AFTER_KEY 24
-#define COEF_MAX_UPDATE_FACTOR_AFTER_KEY 128
-
-#if CONFIG_ADAPT_SCAN
-#define ADAPT_SCAN_PROB_PRECISION 10
-// 1/8 update rate
-#define ADAPT_SCAN_UPDATE_LOG_RATE 3
-#define ADAPT_SCAN_UPDATE_RATE \
-  (1 << (ADAPT_SCAN_PROB_PRECISION - ADAPT_SCAN_UPDATE_LOG_RATE))
-#endif
-
-static INLINE aom_prob av1_merge_probs(aom_prob pre_prob,
-                                       const unsigned int ct[2],
-                                       unsigned int count_sat,
-                                       unsigned int max_update_factor) {
-  return merge_probs(pre_prob, ct, count_sat, max_update_factor);
+static INLINE TX_SIZE get_txsize_entropy_ctx(TX_SIZE txsize) {
+  return (TX_SIZE)((txsize_sqr_map[txsize] + txsize_sqr_up_map[txsize] + 1) >>
+                   1);
 }
 
-static INLINE aom_prob av1_mode_mv_merge_probs(aom_prob pre_prob,
-                                               const unsigned int ct[2]) {
-  return mode_mv_merge_probs(pre_prob, ct);
-}
-
-void av1_average_tile_coef_cdfs(struct frame_contexts *fc,
-                                struct frame_contexts *ec_ctxs[],
-                                aom_cdf_prob *cdf_ptrs[], int num_tiles);
-void av1_average_tile_mv_cdfs(struct frame_contexts *fc,
-                              struct frame_contexts *ec_ctxs[],
-                              aom_cdf_prob *cdf_ptrs[], int num_tiles);
-void av1_average_tile_intra_cdfs(struct frame_contexts *fc,
-                                 struct frame_contexts *ec_ctxs[],
-                                 aom_cdf_prob *cdf_ptrs[], int num_tiles);
-void av1_average_tile_inter_cdfs(struct AV1Common *cm,
-                                 struct frame_contexts *fc,
-                                 struct frame_contexts *ec_ctxs[],
-                                 aom_cdf_prob *cdf_ptrs[], int num_tiles);
-#if CONFIG_PVQ
-void av1_default_pvq_probs(struct AV1Common *cm);
-void av1_average_tile_pvq_cdfs(struct frame_contexts *fc,
-                               struct frame_contexts *ec_ctxs[], int num_tiles);
-#endif  // CONFIG_PVQ
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/entropymode.h

@@ -18,25 +18,16 @@
 #include "av1/common/seg_common.h"
 #include "aom_dsp/aom_filter.h"
 
-#if CONFIG_PVQ
-#include "av1/common/pvq.h"
-#include "av1/common/pvq_state.h"
-#include "av1/common/generic_code.h"
-#endif  // CONFIG_PVQ
-
 #ifdef __cplusplus
 extern "C" {
 #endif
 
 #define BLOCK_SIZE_GROUPS 4
 
-#define TX_SIZE_CONTEXTS 2
+#define TX_SIZE_CONTEXTS 3
 
 #define INTER_OFFSET(mode) ((mode)-NEARESTMV)
-#if CONFIG_COMPOUND_SINGLEREF
-#define INTER_SINGLEREF_COMP_OFFSET(mode) ((mode)-SR_NEAREST_NEARMV)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-#define INTER_COMPOUND_OFFSET(mode) ((mode)-NEAREST_NEARESTMV)
+#define INTER_COMPOUND_OFFSET(mode) (uint8_t)((mode)-NEAREST_NEARESTMV)
 
 // Number of possible contexts for a color index.
 // As can be seen from av1_get_palette_color_index_context(), the possible
@@ -44,14 +35,6 @@ extern "C" {
 // a value from 0 to 4 using 'palette_color_index_context_lookup' table.
 #define PALETTE_COLOR_INDEX_CONTEXTS 5
 
-// Maximum number of colors in a palette.
-#define PALETTE_MAX_SIZE 8
-// Minimum number of colors in a palette.
-#define PALETTE_MIN_SIZE 2
-
-// Palette mode is available for block sizes >= 8x8.
-#define PALETTE_BLOCK_SIZES (BLOCK_LARGEST - BLOCK_8X8 + 1)
-
 // Palette Y mode context for a block is determined by number of neighboring
 // blocks (top and/or left) using a palette for Y plane. So, possible Y mode'
 // context values are:
@@ -66,11 +49,14 @@ extern "C" {
 // 1 if this block uses palette for Y plane (i.e. Y palette size > 0).
 #define PALETTE_UV_MODE_CONTEXTS 2
 
-#define PALETTE_MAX_BLOCK_SIZE (64 * 64)
+// Map the number of pixels in a block size to a context
+//   64(BLOCK_8X8, BLOCK_4x16, BLOCK_16X4)  -> 0
+//  128(BLOCK_8X16, BLOCK_16x8)             -> 1
+//   ...
+// 4096(BLOCK_64X64)                        -> 6
+#define PALATTE_BSIZE_CTXS 7
 
-#if CONFIG_KF_CTX
 #define KF_MODE_CONTEXTS 5
-#endif
 
 struct AV1Common;
 
@@ -80,643 +66,128 @@ typedef struct {
   const int16_t *neighbors;
 } SCAN_ORDER;
 
-struct seg_counts {
-  unsigned int tree_total[MAX_SEGMENTS];
-  unsigned int tree_mispred[MAX_SEGMENTS];
-  unsigned int pred[PREDICTION_PROBS][2];
-};
-
 typedef struct frame_contexts {
-  aom_prob y_mode_prob[BLOCK_SIZE_GROUPS][INTRA_MODES - 1];
-  aom_prob uv_mode_prob[INTRA_MODES][UV_INTRA_MODES - 1];
-#if CONFIG_EXT_PARTITION_TYPES
-  aom_prob partition_prob[PARTITION_CONTEXTS][EXT_PARTITION_TYPES - 1];
-#else
-  aom_prob partition_prob[PARTITION_CONTEXTS][PARTITION_TYPES - 1];
-#endif
-  coeff_cdf_model coef_tail_cdfs[TX_SIZES][PLANE_TYPES];
-  coeff_cdf_model coef_head_cdfs[TX_SIZES][PLANE_TYPES];
-#if CONFIG_ADAPT_SCAN
-// TODO(angiebird): try aom_prob
-#if CONFIG_CHROMA_2X2
-  uint32_t non_zero_prob_2x2[TX_TYPES][4];
-#endif
-  uint32_t non_zero_prob_4X4[TX_TYPES][16];
-  uint32_t non_zero_prob_8X8[TX_TYPES][64];
-  uint32_t non_zero_prob_16X16[TX_TYPES][256];
-  uint32_t non_zero_prob_32X32[TX_TYPES][1024];
-
-  uint32_t non_zero_prob_4X8[TX_TYPES][32];
-  uint32_t non_zero_prob_8X4[TX_TYPES][32];
-  uint32_t non_zero_prob_16X8[TX_TYPES][128];
-  uint32_t non_zero_prob_8X16[TX_TYPES][128];
-  uint32_t non_zero_prob_32X16[TX_TYPES][512];
-  uint32_t non_zero_prob_16X32[TX_TYPES][512];
-
-#if CONFIG_CHROMA_2X2
-  DECLARE_ALIGNED(16, int16_t, scan_2x2[TX_TYPES][4]);
-#endif
-  DECLARE_ALIGNED(16, int16_t, scan_4X4[TX_TYPES][16]);
-  DECLARE_ALIGNED(16, int16_t, scan_8X8[TX_TYPES][64]);
-  DECLARE_ALIGNED(16, int16_t, scan_16X16[TX_TYPES][256]);
-  DECLARE_ALIGNED(16, int16_t, scan_32X32[TX_TYPES][1024]);
-
-  DECLARE_ALIGNED(16, int16_t, scan_4X8[TX_TYPES][32]);
-  DECLARE_ALIGNED(16, int16_t, scan_8X4[TX_TYPES][32]);
-  DECLARE_ALIGNED(16, int16_t, scan_8X16[TX_TYPES][128]);
-  DECLARE_ALIGNED(16, int16_t, scan_16X8[TX_TYPES][128]);
-  DECLARE_ALIGNED(16, int16_t, scan_16X32[TX_TYPES][512]);
-  DECLARE_ALIGNED(16, int16_t, scan_32X16[TX_TYPES][512]);
-
-#if CONFIG_CHROMA_2X2
-  DECLARE_ALIGNED(16, int16_t, iscan_2x2[TX_TYPES][4]);
-#endif
-  DECLARE_ALIGNED(16, int16_t, iscan_4X4[TX_TYPES][16]);
-  DECLARE_ALIGNED(16, int16_t, iscan_8X8[TX_TYPES][64]);
-  DECLARE_ALIGNED(16, int16_t, iscan_16X16[TX_TYPES][256]);
-  DECLARE_ALIGNED(16, int16_t, iscan_32X32[TX_TYPES][1024]);
-
-  DECLARE_ALIGNED(16, int16_t, iscan_4X8[TX_TYPES][32]);
-  DECLARE_ALIGNED(16, int16_t, iscan_8X4[TX_TYPES][32]);
-  DECLARE_ALIGNED(16, int16_t, iscan_8X16[TX_TYPES][128]);
-  DECLARE_ALIGNED(16, int16_t, iscan_16X8[TX_TYPES][128]);
-  DECLARE_ALIGNED(16, int16_t, iscan_16X32[TX_TYPES][512]);
-  DECLARE_ALIGNED(16, int16_t, iscan_32X16[TX_TYPES][512]);
-
-#if CONFIG_CHROMA_2X2
-  int16_t nb_2x2[TX_TYPES][(4 + 1) * 2];
-#endif
-  int16_t nb_4X4[TX_TYPES][(16 + 1) * 2];
-  int16_t nb_8X8[TX_TYPES][(64 + 1) * 2];
-  int16_t nb_16X16[TX_TYPES][(256 + 1) * 2];
-  int16_t nb_32X32[TX_TYPES][(1024 + 1) * 2];
-
-  int16_t nb_4X8[TX_TYPES][(32 + 1) * 2];
-  int16_t nb_8X4[TX_TYPES][(32 + 1) * 2];
-  int16_t nb_8X16[TX_TYPES][(128 + 1) * 2];
-  int16_t nb_16X8[TX_TYPES][(128 + 1) * 2];
-  int16_t nb_16X32[TX_TYPES][(512 + 1) * 2];
-  int16_t nb_32X16[TX_TYPES][(512 + 1) * 2];
-
-  SCAN_ORDER sc[TX_SIZES_ALL][TX_TYPES];
-
-  int16_t eob_threshold[TX_SIZES_ALL][TX_TYPES][EOB_THRESHOLD_NUM];
-#endif  // CONFIG_ADAPT_SCAN
-
-#if CONFIG_LV_MAP
-  aom_prob txb_skip[TX_SIZES][TXB_SKIP_CONTEXTS];
-  aom_prob nz_map[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS];
-  aom_prob eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS];
-  aom_prob dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS];
-  aom_prob coeff_base[TX_SIZES][PLANE_TYPES][NUM_BASE_LEVELS]
-                     [COEFF_BASE_CONTEXTS];
-  aom_prob coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS];
-#if BR_NODE
-  aom_prob coeff_br[TX_SIZES][PLANE_TYPES][BASE_RANGE_SETS][LEVEL_CONTEXTS];
-#endif
-#if CONFIG_CTX1D
-  aom_prob eob_mode[TX_SIZES][PLANE_TYPES][TX_CLASSES];
-  aom_prob empty_line[TX_SIZES][PLANE_TYPES][TX_CLASSES][EMPTY_LINE_CONTEXTS];
-  aom_prob hv_eob[TX_SIZES][PLANE_TYPES][TX_CLASSES][HV_EOB_CONTEXTS];
-#endif  // CONFIG_CTX1D
-
-#if LV_MAP_PROB
   aom_cdf_prob txb_skip_cdf[TX_SIZES][TXB_SKIP_CONTEXTS][CDF_SIZE(2)];
-  aom_cdf_prob nz_map_cdf[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS]
-                         [CDF_SIZE(2)];
-  aom_cdf_prob eob_flag_cdf[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS]
-                           [CDF_SIZE(2)];
-  aom_cdf_prob dc_sign_cdf[PLANE_TYPES][DC_SIGN_CONTEXTS][CDF_SIZE(2)];
-  aom_cdf_prob coeff_base_cdf[TX_SIZES][PLANE_TYPES][NUM_BASE_LEVELS]
-                             [COEFF_BASE_CONTEXTS][CDF_SIZE(2)];
-  aom_cdf_prob coeff_lps_cdf[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS]
+  aom_cdf_prob eob_extra_cdf[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS]
                             [CDF_SIZE(2)];
-#if BR_NODE
-  aom_cdf_prob coeff_br_cdf[TX_SIZES][PLANE_TYPES][BASE_RANGE_SETS]
-                           [LEVEL_CONTEXTS][CDF_SIZE(2)];
-#endif
-#if CONFIG_CTX1D
-  aom_cdf_prob eob_mode_cdf[TX_SIZES][PLANE_TYPES][TX_CLASSES][CDF_SIZE(2)];
-  aom_cdf_prob empty_line_cdf[TX_SIZES][PLANE_TYPES][TX_CLASSES]
-                             [EMPTY_LINE_CONTEXTS][CDF_SIZE(2)];
-  aom_cdf_prob hv_eob_cdf[TX_SIZES][PLANE_TYPES][TX_CLASSES][HV_EOB_CONTEXTS]
-                         [CDF_SIZE(2)];
-#endif  // CONFIG_CTX1D
-#endif  // LV_MAP_PROB
-#endif
+  aom_cdf_prob dc_sign_cdf[PLANE_TYPES][DC_SIGN_CONTEXTS][CDF_SIZE(2)];
+  aom_cdf_prob eob_flag_cdf16[PLANE_TYPES][2][CDF_SIZE(5)];
+  aom_cdf_prob eob_flag_cdf32[PLANE_TYPES][2][CDF_SIZE(6)];
+  aom_cdf_prob eob_flag_cdf64[PLANE_TYPES][2][CDF_SIZE(7)];
+  aom_cdf_prob eob_flag_cdf128[PLANE_TYPES][2][CDF_SIZE(8)];
+  aom_cdf_prob eob_flag_cdf256[PLANE_TYPES][2][CDF_SIZE(9)];
+  aom_cdf_prob eob_flag_cdf512[PLANE_TYPES][2][CDF_SIZE(10)];
+  aom_cdf_prob eob_flag_cdf1024[PLANE_TYPES][2][CDF_SIZE(11)];
+  aom_cdf_prob coeff_base_eob_cdf[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS_EOB]
+                                 [CDF_SIZE(3)];
+  aom_cdf_prob coeff_base_cdf[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS]
+                             [CDF_SIZE(4)];
+  aom_cdf_prob coeff_br_cdf[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS]
+                           [CDF_SIZE(BR_CDF_SIZE)];
 
-  aom_prob newmv_prob[NEWMV_MODE_CONTEXTS];
-  aom_prob zeromv_prob[ZEROMV_MODE_CONTEXTS];
-  aom_prob refmv_prob[REFMV_MODE_CONTEXTS];
-  aom_prob drl_prob[DRL_MODE_CONTEXTS];
-#if CONFIG_NEW_MULTISYMBOL
   aom_cdf_prob newmv_cdf[NEWMV_MODE_CONTEXTS][CDF_SIZE(2)];
-  aom_cdf_prob zeromv_cdf[ZEROMV_MODE_CONTEXTS][CDF_SIZE(2)];
+  aom_cdf_prob zeromv_cdf[GLOBALMV_MODE_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob refmv_cdf[REFMV_MODE_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob drl_cdf[DRL_MODE_CONTEXTS][CDF_SIZE(2)];
-#endif
 
-  aom_prob inter_compound_mode_probs[INTER_MODE_CONTEXTS]
-                                    [INTER_COMPOUND_MODES - 1];
   aom_cdf_prob inter_compound_mode_cdf[INTER_MODE_CONTEXTS]
                                       [CDF_SIZE(INTER_COMPOUND_MODES)];
-#if CONFIG_COMPOUND_SINGLEREF
-  aom_prob inter_singleref_comp_mode_probs[INTER_MODE_CONTEXTS]
-                                          [INTER_SINGLEREF_COMP_MODES - 1];
-  aom_cdf_prob inter_singleref_comp_mode_cdf[INTER_MODE_CONTEXTS][CDF_SIZE(
-      INTER_SINGLEREF_COMP_MODES)];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  aom_prob compound_type_prob[BLOCK_SIZES_ALL][COMPOUND_TYPES - 1];
-#if CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
-  aom_cdf_prob compound_type_cdf[BLOCK_SIZES_ALL][CDF_SIZE(COMPOUND_TYPES)];
-#endif  // CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
-#if CONFIG_INTERINTRA
-  aom_prob interintra_prob[BLOCK_SIZE_GROUPS];
-  aom_prob wedge_interintra_prob[BLOCK_SIZES_ALL];
-  aom_prob interintra_mode_prob[BLOCK_SIZE_GROUPS][INTERINTRA_MODES - 1];
-#if CONFIG_NEW_MULTISYMBOL
+  aom_cdf_prob compound_type_cdf[BLOCK_SIZES_ALL][CDF_SIZE(COMPOUND_TYPES - 1)];
+  aom_cdf_prob wedge_idx_cdf[BLOCK_SIZES_ALL][CDF_SIZE(16)];
   aom_cdf_prob interintra_cdf[BLOCK_SIZE_GROUPS][CDF_SIZE(2)];
   aom_cdf_prob wedge_interintra_cdf[BLOCK_SIZES_ALL][CDF_SIZE(2)];
-#endif
   aom_cdf_prob interintra_mode_cdf[BLOCK_SIZE_GROUPS]
                                   [CDF_SIZE(INTERINTRA_MODES)];
-#endif  // CONFIG_INTERINTRA
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-  aom_prob motion_mode_prob[BLOCK_SIZES_ALL][MOTION_MODES - 1];
   aom_cdf_prob motion_mode_cdf[BLOCK_SIZES_ALL][CDF_SIZE(MOTION_MODES)];
-#if CONFIG_NCOBMC_ADAPT_WEIGHT && CONFIG_MOTION_VAR
-  aom_prob ncobmc_mode_prob[ADAPT_OVERLAP_BLOCKS][MAX_NCOBMC_MODES - 1];
-  aom_cdf_prob ncobmc_mode_cdf[ADAPT_OVERLAP_BLOCKS]
-                              [CDF_SIZE(MAX_NCOBMC_MODES)];
-#endif
-#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-  aom_prob ncobmc_prob[BLOCK_SIZES_ALL][OBMC_FAMILY_MODES - 1];
-  aom_cdf_prob ncobmc_cdf[BLOCK_SIZES_ALL][CDF_SIZE(OBMC_FAMILY_MODES)];
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-  aom_prob obmc_prob[BLOCK_SIZES_ALL];
-#if CONFIG_NEW_MULTISYMBOL || CONFIG_NCOBMC_ADAPT_WEIGHT
   aom_cdf_prob obmc_cdf[BLOCK_SIZES_ALL][CDF_SIZE(2)];
-#endif  // CONFIG_NEW_MULTISYMBOL
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-  aom_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
-  aom_prob comp_inter_prob[COMP_INTER_CONTEXTS];
-  aom_cdf_prob palette_y_size_cdf[PALETTE_BLOCK_SIZES][CDF_SIZE(PALETTE_SIZES)];
-  aom_cdf_prob palette_uv_size_cdf[PALETTE_BLOCK_SIZES]
-                                  [CDF_SIZE(PALETTE_SIZES)];
+  aom_cdf_prob palette_y_size_cdf[PALATTE_BSIZE_CTXS][CDF_SIZE(PALETTE_SIZES)];
+  aom_cdf_prob palette_uv_size_cdf[PALATTE_BSIZE_CTXS][CDF_SIZE(PALETTE_SIZES)];
   aom_cdf_prob palette_y_color_index_cdf[PALETTE_SIZES]
                                         [PALETTE_COLOR_INDEX_CONTEXTS]
                                         [CDF_SIZE(PALETTE_COLORS)];
   aom_cdf_prob palette_uv_color_index_cdf[PALETTE_SIZES]
                                          [PALETTE_COLOR_INDEX_CONTEXTS]
                                          [CDF_SIZE(PALETTE_COLORS)];
-#if CONFIG_MRC_TX
-  aom_cdf_prob mrc_mask_inter_cdf[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS]
-                                 [CDF_SIZE(PALETTE_COLORS)];
-  aom_cdf_prob mrc_mask_intra_cdf[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS]
-                                 [CDF_SIZE(PALETTE_COLORS)];
-#endif  // CONFIG_MRC_TX
-#if CONFIG_NEW_MULTISYMBOL
-  aom_cdf_prob palette_y_mode_cdf[PALETTE_BLOCK_SIZES][PALETTE_Y_MODE_CONTEXTS]
+  aom_cdf_prob palette_y_mode_cdf[PALATTE_BSIZE_CTXS][PALETTE_Y_MODE_CONTEXTS]
                                  [CDF_SIZE(2)];
   aom_cdf_prob palette_uv_mode_cdf[PALETTE_UV_MODE_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob comp_inter_cdf[COMP_INTER_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob single_ref_cdf[REF_CONTEXTS][SINGLE_REFS - 1][CDF_SIZE(2)];
-#endif
-#if CONFIG_EXT_COMP_REFS
-  aom_prob comp_ref_type_prob[COMP_REF_TYPE_CONTEXTS];
-  aom_prob uni_comp_ref_prob[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1];
-#if CONFIG_NEW_MULTISYMBOL
   aom_cdf_prob comp_ref_type_cdf[COMP_REF_TYPE_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob uni_comp_ref_cdf[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1]
                                [CDF_SIZE(2)];
-#endif  // CONFIG_NEW_MULTISYMBOL
-#endif  // CONFIG_EXT_COMP_REFS
-  aom_prob single_ref_prob[REF_CONTEXTS][SINGLE_REFS - 1];
-#if CONFIG_EXT_REFS
-  aom_prob comp_ref_prob[REF_CONTEXTS][FWD_REFS - 1];
-  aom_prob comp_bwdref_prob[REF_CONTEXTS][BWD_REFS - 1];
-#else
-  aom_prob comp_ref_prob[REF_CONTEXTS][COMP_REFS - 1];
-#endif  // CONFIG_EXT_REFS
-#if CONFIG_NEW_MULTISYMBOL
-#if CONFIG_EXT_REFS
   aom_cdf_prob comp_ref_cdf[REF_CONTEXTS][FWD_REFS - 1][CDF_SIZE(2)];
   aom_cdf_prob comp_bwdref_cdf[REF_CONTEXTS][BWD_REFS - 1][CDF_SIZE(2)];
-#else
-  aom_cdf_prob comp_ref_cdf[REF_CONTEXTS][COMP_REFS - 1][CDF_SIZE(2)];
-#endif  // CONFIG_EXT_REFS
-#endif
-#if CONFIG_COMPOUND_SINGLEREF
-  aom_prob comp_inter_mode_prob[COMP_INTER_MODE_CONTEXTS];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-  aom_prob quarter_tx_size_prob;
-#if CONFIG_NEW_MULTISYMBOL
-  aom_cdf_prob quarter_tx_size_cdf[CDF_SIZE(2)];
-#endif
-#endif
-#if CONFIG_VAR_TX
-  aom_prob txfm_partition_prob[TXFM_PARTITION_CONTEXTS];
-#if CONFIG_NEW_MULTISYMBOL
   aom_cdf_prob txfm_partition_cdf[TXFM_PARTITION_CONTEXTS][CDF_SIZE(2)];
-#endif
-#endif  // CONFIG_VAR_TX
-  aom_prob skip_probs[SKIP_CONTEXTS];
-#if CONFIG_NEW_MULTISYMBOL
+  aom_cdf_prob compound_index_cdf[COMP_INDEX_CONTEXTS][CDF_SIZE(2)];
+  aom_cdf_prob comp_group_idx_cdf[COMP_GROUP_IDX_CONTEXTS][CDF_SIZE(2)];
+  aom_cdf_prob skip_mode_cdfs[SKIP_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob skip_cdfs[SKIP_CONTEXTS][CDF_SIZE(2)];
   aom_cdf_prob intra_inter_cdf[INTRA_INTER_CONTEXTS][CDF_SIZE(2)];
-#endif
-  nmv_context nmvc[NMV_CONTEXTS];
-#if CONFIG_INTRABC
+  nmv_context nmvc;
   nmv_context ndvc;
   aom_cdf_prob intrabc_cdf[CDF_SIZE(2)];
-#endif
-  int initialized;
-#if CONFIG_SUPERTX
-  aom_prob supertx_prob[PARTITION_SUPERTX_CONTEXTS][TX_SIZES];
-#endif  // CONFIG_SUPERTX
   struct segmentation_probs seg;
-#if CONFIG_EXT_INTRA
-#if CONFIG_INTRA_INTERP
-  aom_prob intra_filter_probs[INTRA_FILTERS + 1][INTRA_FILTERS - 1];
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
-#if CONFIG_FILTER_INTRA
-  aom_prob filter_intra_probs[PLANE_TYPES];
-#endif  // CONFIG_FILTER_INTRA
-#if CONFIG_LOOP_RESTORATION
-  aom_prob switchable_restore_prob[RESTORE_SWITCHABLE_TYPES - 1];
-#endif  // CONFIG_LOOP_RESTORATION
+  aom_cdf_prob filter_intra_cdfs[BLOCK_SIZES_ALL][CDF_SIZE(2)];
+  aom_cdf_prob filter_intra_mode_cdf[CDF_SIZE(FILTER_INTRA_MODES)];
+  aom_cdf_prob switchable_restore_cdf[CDF_SIZE(RESTORE_SWITCHABLE_TYPES)];
+  aom_cdf_prob wiener_restore_cdf[CDF_SIZE(2)];
+  aom_cdf_prob sgrproj_restore_cdf[CDF_SIZE(2)];
   aom_cdf_prob y_mode_cdf[BLOCK_SIZE_GROUPS][CDF_SIZE(INTRA_MODES)];
-  aom_cdf_prob uv_mode_cdf[INTRA_MODES][CDF_SIZE(UV_INTRA_MODES)];
-#if CONFIG_EXT_PARTITION_TYPES
+  aom_cdf_prob uv_mode_cdf[CFL_ALLOWED_TYPES][INTRA_MODES]
+                          [CDF_SIZE(UV_INTRA_MODES)];
   aom_cdf_prob partition_cdf[PARTITION_CONTEXTS][CDF_SIZE(EXT_PARTITION_TYPES)];
-#else
-  aom_cdf_prob partition_cdf[PARTITION_CONTEXTS][CDF_SIZE(PARTITION_TYPES)];
-#endif
   aom_cdf_prob switchable_interp_cdf[SWITCHABLE_FILTER_CONTEXTS]
                                     [CDF_SIZE(SWITCHABLE_FILTERS)];
-/* kf_y_cdf is discarded after use, so does not require persistent storage.
-   However, we keep it with the other CDFs in this struct since it needs to
-   be copied to each tile to support parallelism just like the others.
-*/
-#if CONFIG_KF_CTX
+  /* kf_y_cdf is discarded after use, so does not require persistent storage.
+     However, we keep it with the other CDFs in this struct since it needs to
+     be copied to each tile to support parallelism just like the others.
+  */
   aom_cdf_prob kf_y_cdf[KF_MODE_CONTEXTS][KF_MODE_CONTEXTS]
                        [CDF_SIZE(INTRA_MODES)];
-#else
-  aom_cdf_prob kf_y_cdf[INTRA_MODES][INTRA_MODES][CDF_SIZE(INTRA_MODES)];
-#endif
-  aom_cdf_prob tx_size_cdf[MAX_TX_DEPTH][TX_SIZE_CONTEXTS]
+
+  aom_cdf_prob angle_delta_cdf[DIRECTIONAL_MODES]
+                              [CDF_SIZE(2 * MAX_ANGLE_DELTA + 1)];
+
+  aom_cdf_prob tx_size_cdf[MAX_TX_CATS][TX_SIZE_CONTEXTS]
                           [CDF_SIZE(MAX_TX_DEPTH + 1)];
   aom_cdf_prob delta_q_cdf[CDF_SIZE(DELTA_Q_PROBS + 1)];
-#if CONFIG_EXT_DELTA_Q
-#if CONFIG_LOOPFILTER_LEVEL
   aom_cdf_prob delta_lf_multi_cdf[FRAME_LF_COUNT][CDF_SIZE(DELTA_LF_PROBS + 1)];
-#endif  // CONFIG_LOOPFILTER_LEVEL
   aom_cdf_prob delta_lf_cdf[CDF_SIZE(DELTA_LF_PROBS + 1)];
-#endif
-#if CONFIG_EXT_TX
   aom_cdf_prob intra_ext_tx_cdf[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
                                [CDF_SIZE(TX_TYPES)];
   aom_cdf_prob inter_ext_tx_cdf[EXT_TX_SETS_INTER][EXT_TX_SIZES]
                                [CDF_SIZE(TX_TYPES)];
-#else
-  aom_cdf_prob intra_ext_tx_cdf[EXT_TX_SIZES][TX_TYPES][CDF_SIZE(TX_TYPES)];
-  aom_cdf_prob inter_ext_tx_cdf[EXT_TX_SIZES][CDF_SIZE(TX_TYPES)];
-#endif  // CONFIG_EXT_TX
-#if CONFIG_LGT_FROM_PRED
-  aom_prob intra_lgt_prob[LGT_SIZES][INTRA_MODES];
-  aom_prob inter_lgt_prob[LGT_SIZES];
-#endif  // CONFIG_LGT_FROM_PRED
-#if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
-  aom_cdf_prob intra_filter_cdf[INTRA_FILTERS + 1][CDF_SIZE(INTRA_FILTERS)];
-#endif  // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
-  aom_prob delta_q_prob[DELTA_Q_PROBS];
-#if CONFIG_EXT_DELTA_Q
-#if CONFIG_LOOPFILTER_LEVEL
-  aom_prob delta_lf_multi_prob[FRAME_LF_COUNT][DELTA_LF_PROBS];
-#endif  // CONFIG_LOOPFILTER_LEVEL
-  aom_prob delta_lf_prob[DELTA_LF_PROBS];
-#endif
-#if CONFIG_PVQ
-  // TODO(any): If PVQ is enabled, most of coefficient related cdf,
-  // such as coef_cdfs[], coef_tail_cdfs[], and coef_heaf_cdfs[] can be removed.
-  od_adapt_ctx pvq_context;
-#endif  // CONFIG_PVQ
-#if CONFIG_CFL
   aom_cdf_prob cfl_sign_cdf[CDF_SIZE(CFL_JOINT_SIGNS)];
   aom_cdf_prob cfl_alpha_cdf[CFL_ALPHA_CONTEXTS][CDF_SIZE(CFL_ALPHABET_SIZE)];
-#endif
-#if CONFIG_LPF_SB
-  aom_cdf_prob lpf_reuse_cdf[LPF_REUSE_CONTEXT][CDF_SIZE(2)];
-  aom_cdf_prob lpf_delta_cdf[LPF_DELTA_CONTEXT][CDF_SIZE(DELTA_RANGE)];
-  aom_cdf_prob lpf_sign_cdf[LPF_REUSE_CONTEXT][LPF_SIGN_CONTEXT][CDF_SIZE(2)];
-#endif  // CONFIG_LPF_SB
+  int initialized;
 } FRAME_CONTEXT;
 
-typedef struct FRAME_COUNTS {
-// Note: This structure should only contain 'unsigned int' fields, or
-// aggregates built solely from 'unsigned int' fields/elements
-#if CONFIG_ENTROPY_STATS
-  unsigned int kf_y_mode[INTRA_MODES][INTRA_MODES][INTRA_MODES];
-  unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
-  unsigned int uv_mode[INTRA_MODES][UV_INTRA_MODES];
-#endif  // CONFIG_ENTROPY_STATS
-#if CONFIG_EXT_PARTITION_TYPES
-  unsigned int partition[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
-#else
-  unsigned int partition[PARTITION_CONTEXTS][PARTITION_TYPES];
-#endif
-  unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
-                                [SWITCHABLE_FILTERS];
-#if CONFIG_ADAPT_SCAN
-#if CONFIG_CHROMA_2X2
-  unsigned int non_zero_count_2x2[TX_TYPES][4];
-#endif  // CONFIG_CHROMA_2X2
-  unsigned int non_zero_count_4X4[TX_TYPES][16];
-  unsigned int non_zero_count_8X8[TX_TYPES][64];
-  unsigned int non_zero_count_16X16[TX_TYPES][256];
-  unsigned int non_zero_count_32X32[TX_TYPES][1024];
-
-  unsigned int non_zero_count_4x8[TX_TYPES][32];
-  unsigned int non_zero_count_8x4[TX_TYPES][32];
-  unsigned int non_zero_count_8x16[TX_TYPES][128];
-  unsigned int non_zero_count_16x8[TX_TYPES][128];
-  unsigned int non_zero_count_16x32[TX_TYPES][512];
-  unsigned int non_zero_count_32x16[TX_TYPES][512];
-
-  unsigned int txb_count[TX_SIZES_ALL][TX_TYPES];
-#endif  // CONFIG_ADAPT_SCAN
-
-#if CONFIG_LV_MAP
-  unsigned int txb_skip[TX_SIZES][TXB_SKIP_CONTEXTS][2];
-  unsigned int nz_map[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS][2];
-  unsigned int eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS][2];
-  unsigned int dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS][2];
-  unsigned int coeff_base[TX_SIZES][PLANE_TYPES][NUM_BASE_LEVELS]
-                         [COEFF_BASE_CONTEXTS][2];
-  unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS][2];
-  unsigned int coeff_br[TX_SIZES][PLANE_TYPES][BASE_RANGE_SETS][LEVEL_CONTEXTS]
-                       [2];
-#if CONFIG_CTX1D
-  unsigned int eob_mode[TX_SIZES][PLANE_TYPES][TX_CLASSES][2];
-  unsigned int empty_line[TX_SIZES][PLANE_TYPES][TX_CLASSES]
-                         [EMPTY_LINE_CONTEXTS][2];
-  unsigned int hv_eob[TX_SIZES][PLANE_TYPES][TX_CLASSES][HV_EOB_CONTEXTS][2];
-#endif  // CONFIG_CTX1D
-#endif  // CONFIG_LV_MAP
-
-#if CONFIG_SYMBOLRATE
-  unsigned int coeff_num[2];   // 0: zero coeff 1: non-zero coeff
-  unsigned int symbol_num[2];  // 0: entropy symbol 1: non-entropy symbol
-#endif
-
-  unsigned int newmv_mode[NEWMV_MODE_CONTEXTS][2];
-  unsigned int zeromv_mode[ZEROMV_MODE_CONTEXTS][2];
-  unsigned int refmv_mode[REFMV_MODE_CONTEXTS][2];
-  unsigned int drl_mode[DRL_MODE_CONTEXTS][2];
-
-  unsigned int inter_compound_mode[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES];
-#if CONFIG_COMPOUND_SINGLEREF
-  unsigned int inter_singleref_comp_mode[INTER_MODE_CONTEXTS]
-                                        [INTER_SINGLEREF_COMP_MODES];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-#if CONFIG_INTERINTRA
-  unsigned int interintra[BLOCK_SIZE_GROUPS][2];
-  unsigned int interintra_mode[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
-  unsigned int wedge_interintra[BLOCK_SIZES_ALL][2];
-#endif  // CONFIG_INTERINTRA
-  unsigned int compound_interinter[BLOCK_SIZES_ALL][COMPOUND_TYPES];
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-  unsigned int motion_mode[BLOCK_SIZES_ALL][MOTION_MODES];
-#if CONFIG_NCOBMC_ADAPT_WEIGHT && CONFIG_MOTION_VAR
-  unsigned int ncobmc_mode[ADAPT_OVERLAP_BLOCKS][MAX_NCOBMC_MODES];
-#endif
-#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-  unsigned int ncobmc[BLOCK_SIZES_ALL][OBMC_FAMILY_MODES];
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-  unsigned int obmc[BLOCK_SIZES_ALL][2];
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-  unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
-  unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
-#if CONFIG_EXT_COMP_REFS
-  unsigned int comp_ref_type[COMP_REF_TYPE_CONTEXTS][2];
-  unsigned int uni_comp_ref[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1][2];
-#endif  // CONFIG_EXT_COMP_REFS
-  unsigned int single_ref[REF_CONTEXTS][SINGLE_REFS - 1][2];
-#if CONFIG_EXT_REFS
-  unsigned int comp_ref[REF_CONTEXTS][FWD_REFS - 1][2];
-  unsigned int comp_bwdref[REF_CONTEXTS][BWD_REFS - 1][2];
-#else
-  unsigned int comp_ref[REF_CONTEXTS][COMP_REFS - 1][2];
-#endif  // CONFIG_EXT_REFS
-#if CONFIG_COMPOUND_SINGLEREF
-  unsigned int comp_inter_mode[COMP_INTER_MODE_CONTEXTS][2];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  // TODO(urvang): Only needed for !CONFIG_VAR_TX case. So can be removed when
-  // CONFIG_VAR_TX flag is removed.
-  unsigned int tx_size[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][MAX_TX_DEPTH + 1];
-#if CONFIG_RECT_TX_EXT && (CONFIG_EXT_TX || CONFIG_VAR_TX)
-  unsigned int quarter_tx_size[2];
-#endif
-#if CONFIG_VAR_TX
-  unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
-#endif
-  unsigned int skip[SKIP_CONTEXTS][2];
-  nmv_context_counts mv[NMV_CONTEXTS];
-#if CONFIG_INTRABC
-  unsigned int intrabc[2];
-  nmv_context_counts dv;
-#endif
-#if CONFIG_LGT_FROM_PRED
-  unsigned int intra_lgt[LGT_SIZES][INTRA_MODES][2];
-  unsigned int inter_lgt[LGT_SIZES][2];
-#endif  // CONFIG_LGT_FROM_PRED
-  unsigned int delta_q[DELTA_Q_PROBS][2];
-#if CONFIG_EXT_DELTA_Q
-#if CONFIG_LOOPFILTER_LEVEL
-  unsigned int delta_lf_multi[FRAME_LF_COUNT][DELTA_LF_PROBS][2];
-#endif  // CONFIG_LOOPFILTER_LEVEL
-  unsigned int delta_lf[DELTA_LF_PROBS][2];
-#endif
-#if CONFIG_EXT_TX && CONFIG_RECT_TX
-  unsigned int tx_size_implied[TX_SIZES][TX_SIZES];
-#endif  // CONFIG_EXT_TX && CONFIG_RECT_TX
-#if CONFIG_ENTROPY_STATS
-#if CONFIG_EXT_TX
-  unsigned int inter_ext_tx[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
-  unsigned int intra_ext_tx[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
-                           [TX_TYPES];
-#else
-  unsigned int intra_ext_tx[EXT_TX_SIZES][TX_TYPES][TX_TYPES];
-  unsigned int inter_ext_tx[EXT_TX_SIZES][TX_TYPES];
-#endif  // CONFIG_EXT_TX
-#endif  // CONFIG_ENTROPY_STATS
-#if CONFIG_SUPERTX
-  unsigned int supertx[PARTITION_SUPERTX_CONTEXTS][TX_SIZES][2];
-  unsigned int supertx_size[TX_SIZES];
-#endif  // CONFIG_SUPERTX
-  struct seg_counts seg;
-#if CONFIG_EXT_INTRA
-#if CONFIG_INTRA_INTERP
-  unsigned int intra_filter[INTRA_FILTERS + 1][INTRA_FILTERS];
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
-#if CONFIG_FILTER_INTRA
-  unsigned int filter_intra[PLANE_TYPES][2];
-#endif  // CONFIG_FILTER_INTRA
-#if CONFIG_LPF_SB
-  unsigned int lpf_reuse[LPF_REUSE_CONTEXT][2];
-  unsigned int lpf_delta[LPF_DELTA_CONTEXT][DELTA_RANGE];
-  unsigned int lpf_sign[LPF_SIGN_CONTEXT][2];
-#endif  // CONFIG_LPF_SB
-} FRAME_COUNTS;
-
-#if CONFIG_KF_CTX
-extern const aom_cdf_prob default_kf_y_mode_cdf[KF_MODE_CONTEXTS]
-                                               [KF_MODE_CONTEXTS]
-                                               [CDF_SIZE(INTRA_MODES)];
-#else
-extern const aom_cdf_prob default_kf_y_mode_cdf[INTRA_MODES][INTRA_MODES]
-                                               [CDF_SIZE(INTRA_MODES)];
-#endif
-
-extern const aom_prob av1_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES]
-                                                     [PALETTE_Y_MODE_CONTEXTS];
-extern const aom_prob
-    av1_default_palette_uv_mode_prob[PALETTE_UV_MODE_CONTEXTS];
-
-#if CONFIG_EXT_TX
 static const int av1_ext_tx_ind[EXT_TX_SET_TYPES][TX_TYPES] = {
-  {
-      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-#if CONFIG_MRC_TX
-  {
-      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
-  },
-  {
-      1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2,
-  },
-#endif  // CONFIG_MRC_TX
-  {
-      1, 3, 4, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      1, 5, 6, 4, 0, 0, 0, 0, 0, 0, 2, 3, 0, 0, 0, 0,
-  },
-  {
-      3, 4, 5, 8, 6, 7, 9, 10, 11, 0, 1, 2, 0, 0, 0, 0,
-  },
-  {
-      7, 8, 9, 12, 10, 11, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6,
-  },
+  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 1, 3, 4, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 1, 5, 6, 4, 0, 0, 0, 0, 0, 0, 2, 3, 0, 0, 0, 0 },
+  { 3, 4, 5, 8, 6, 7, 9, 10, 11, 0, 1, 2, 0, 0, 0, 0 },
+  { 7, 8, 9, 12, 10, 11, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6 },
 };
 
 static const int av1_ext_tx_inv[EXT_TX_SET_TYPES][TX_TYPES] = {
-  {
-      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-#if CONFIG_MRC_TX
-  {
-      0, 16, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      9, 0, 16, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-#endif  // CONFIG_MRC_TX
-  {
-      9, 0, 3, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      9, 0, 10, 11, 3, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-  },
-  {
-      9, 10, 11, 0, 1, 2, 4, 5, 3, 6, 7, 8, 0, 0, 0, 0,
-  },
-  {
-      9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 4, 5, 3, 6, 7, 8,
-  },
+  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 9, 0, 3, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 9, 0, 10, 11, 3, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
+  { 9, 10, 11, 0, 1, 2, 4, 5, 3, 6, 7, 8, 0, 0, 0, 0 },
+  { 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 4, 5, 3, 6, 7, 8 },
 };
-#else
-#if CONFIG_MRC_TX
-static const int av1_ext_tx_ind[TX_TYPES] = {
-  0, 3, 4, 2, 1,
-};
-static const int av1_ext_tx_inv[TX_TYPES] = {
-  0, 4, 3, 1, 2,
-};
-#else
-static const int av1_ext_tx_ind[TX_TYPES] = {
-  0, 2, 3, 1,
-};
-static const int av1_ext_tx_inv[TX_TYPES] = {
-  0, 3, 1, 2,
-};
-#endif  // CONFIG_MRC_TX
-#endif  // CONFIG_EXT_TX
-
-#if CONFIG_INTERINTRA
-extern const aom_tree_index
-    av1_interintra_mode_tree[TREE_SIZE(INTERINTRA_MODES)];
-#endif
-extern const aom_tree_index
-    av1_inter_compound_mode_tree[TREE_SIZE(INTER_COMPOUND_MODES)];
-#if CONFIG_COMPOUND_SINGLEREF
-extern const aom_tree_index
-    av1_inter_singleref_comp_mode_tree[TREE_SIZE(INTER_SINGLEREF_COMP_MODES)];
-#endif  // CONFIG_COMPOUND_SINGLEREF
-extern const aom_tree_index av1_compound_type_tree[TREE_SIZE(COMPOUND_TYPES)];
-extern const aom_tree_index av1_partition_tree[TREE_SIZE(PARTITION_TYPES)];
-#if CONFIG_EXT_PARTITION_TYPES
-extern const aom_tree_index
-    av1_ext_partition_tree[TREE_SIZE(EXT_PARTITION_TYPES)];
-#endif
-extern const aom_tree_index
-    av1_palette_color_index_tree[PALETTE_SIZES][TREE_SIZE(PALETTE_COLORS)];
-#if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
-extern const aom_tree_index av1_intra_filter_tree[TREE_SIZE(INTRA_FILTERS)];
-#endif  // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
-#if CONFIG_EXT_TX
-extern const aom_tree_index av1_ext_tx_tree[EXT_TX_SET_TYPES]
-                                           [TREE_SIZE(TX_TYPES)];
-#else
-extern const aom_tree_index av1_ext_tx_tree[TREE_SIZE(TX_TYPES)];
-#endif  // CONFIG_EXT_TX
-#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-extern const aom_tree_index av1_motion_mode_tree[TREE_SIZE(MOTION_MODES)];
-#endif  // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-extern const aom_tree_index av1_ncobmc_mode_tree[TREE_SIZE(MAX_NCOBMC_MODES)];
-#if CONFIG_WARPED_MOTION
-extern const aom_tree_index av1_ncobmc_tree[TREE_SIZE(OBMC_FAMILY_MODES)];
-#endif  // CONFIG_WARPED_MOTION
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-#if CONFIG_LOOP_RESTORATION
-#define RESTORE_NONE_SGRPROJ_PROB 64
-#define RESTORE_NONE_BILATERAL_PROB 16
-#define RESTORE_NONE_WIENER_PROB 64
-#define RESTORE_NONE_DOMAINTXFMRF_PROB 64
-extern const aom_tree_index
-    av1_switchable_restore_tree[TREE_SIZE(RESTORE_SWITCHABLE_TYPES)];
-#endif  // CONFIG_LOOP_RESTORATION
 
+void av1_set_default_ref_deltas(int8_t *ref_deltas);
+void av1_set_default_mode_deltas(int8_t *mode_deltas);
+void av1_setup_frame_contexts(struct AV1Common *cm);
 void av1_setup_past_independence(struct AV1Common *cm);
 
-void av1_adapt_intra_frame_probs(struct AV1Common *cm);
-void av1_adapt_inter_frame_probs(struct AV1Common *cm);
-
 static INLINE int av1_ceil_log2(int n) {
+  if (n < 2) return 0;
   int i = 1, p = 2;
   while (p < n) {
     i++;

third_party/aom/av1/common/entropymv.c

@@ -12,100 +12,51 @@
 #include "av1/common/onyxc_int.h"
 #include "av1/common/entropymv.h"
 
-// Integer pel reference mv threshold for use of high-precision 1/8 mv
-#define COMPANDED_MVREF_THRESH 8
-
-const aom_tree_index av1_mv_joint_tree[TREE_SIZE(MV_JOINTS)] = {
-  -MV_JOINT_ZERO, 2, -MV_JOINT_HNZVZ, 4, -MV_JOINT_HZVNZ, -MV_JOINT_HNZVNZ
-};
-
-/* clang-format off */
-const aom_tree_index av1_mv_class_tree[TREE_SIZE(MV_CLASSES)] = {
-  -MV_CLASS_0, 2,
-  -MV_CLASS_1, 4,
-  6, 8,
-  -MV_CLASS_2, -MV_CLASS_3,
-  10, 12,
-  -MV_CLASS_4, -MV_CLASS_5,
-  -MV_CLASS_6, 14,
-  16, 18,
-  -MV_CLASS_7, -MV_CLASS_8,
-  -MV_CLASS_9, -MV_CLASS_10,
-};
-/* clang-format on */
-
-const aom_tree_index av1_mv_class0_tree[TREE_SIZE(CLASS0_SIZE)] = {
-  -0, -1,
-};
-
-const aom_tree_index av1_mv_fp_tree[TREE_SIZE(MV_FP_SIZE)] = { -0, 2,  -1,
-                                                               4,  -2, -3 };
-
 static const nmv_context default_nmv_context = {
-  { 32, 64, 96 },  // joints
-  { AOM_ICDF(4096), AOM_ICDF(11264), AOM_ICDF(19328), AOM_ICDF(32768),
-    0 },  // joint_cdf
+  { AOM_CDF4(4096, 11264, 19328) },  // joints_cdf
   { {
         // Vertical component
-        128,                                                   // sign
-        { 224, 144, 192, 168, 192, 176, 192, 198, 198, 245 },  // class
-        { AOM_ICDF(28672), AOM_ICDF(30976), AOM_ICDF(31858), AOM_ICDF(32320),
-          AOM_ICDF(32551), AOM_ICDF(32656), AOM_ICDF(32740), AOM_ICDF(32757),
-          AOM_ICDF(32762), AOM_ICDF(32767), AOM_ICDF(32768), 0 },  // class_cdf
-        { 216 },                                                   // class0
-        { 136, 140, 148, 160, 176, 192, 224, 234, 234, 240 },      // bits
-        { { 128, 128, 64 }, { 96, 112, 64 } },                     // class0_fp
-        { 64, 96, 64 },                                            // fp
-        { { AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(26624), AOM_ICDF(32768),
-            0 },
-          { AOM_ICDF(12288), AOM_ICDF(21248), AOM_ICDF(24128), AOM_ICDF(32768),
-            0 } },  // class0_fp_cdf
-        { AOM_ICDF(8192), AOM_ICDF(17408), AOM_ICDF(21248), AOM_ICDF(32768),
-          0 },  // fp_cdf
-        160,    // class0_hp bit
-        128,    // hp
-#if CONFIG_NEW_MULTISYMBOL
-        { AOM_ICDF(160 * 128), AOM_ICDF(32768), 0 },
-        { AOM_ICDF(128 * 128), AOM_ICDF(32768), 0 },
-        { AOM_ICDF(216 * 128), AOM_ICDF(32768), 0 },
-        { { AOM_ICDF(128 * 196), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 198), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 208), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 224), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 245), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 240), AOM_ICDF(32768), 0 } },  // bits_cdf
-#endif
+        { AOM_CDF11(28672, 30976, 31858, 32320, 32551, 32656, 32740, 32757,
+                    32762, 32767) },  // class_cdf // fp
+        { { AOM_CDF4(16384, 24576, 26624) },
+          { AOM_CDF4(12288, 21248, 24128) } },  // class0_fp_cdf
+        { AOM_CDF4(8192, 17408, 21248) },       // fp_cdf
+        { AOM_CDF2(128 * 128) },                // sign_cdf
+        { AOM_CDF2(160 * 128) },                // class0_hp_cdf
+        { AOM_CDF2(128 * 128) },                // hp_cdf
+        { AOM_CDF2(216 * 128) },                // class0_cdf
+        { { AOM_CDF2(128 * 136) },
+          { AOM_CDF2(128 * 140) },
+          { AOM_CDF2(128 * 148) },
+          { AOM_CDF2(128 * 160) },
+          { AOM_CDF2(128 * 176) },
+          { AOM_CDF2(128 * 192) },
+          { AOM_CDF2(128 * 224) },
+          { AOM_CDF2(128 * 234) },
+          { AOM_CDF2(128 * 234) },
+          { AOM_CDF2(128 * 240) } },  // bits_cdf
     },
     {
         // Horizontal component
-        128,                                                   // sign
-        { 216, 128, 176, 160, 176, 176, 192, 198, 198, 208 },  // class
-        { AOM_ICDF(28672), AOM_ICDF(30976), AOM_ICDF(31858), AOM_ICDF(32320),
-          AOM_ICDF(32551), AOM_ICDF(32656), AOM_ICDF(32740), AOM_ICDF(32757),
-          AOM_ICDF(32762), AOM_ICDF(32767), AOM_ICDF(32768), 0 },  // class_cdf
-        { 208 },                                                   // class0
-        { 136, 140, 148, 160, 176, 192, 224, 234, 234, 240 },      // bits
-        { { 128, 128, 64 }, { 96, 112, 64 } },                     // class0_fp
-        { 64, 96, 64 },                                            // fp
-        { { AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(26624), AOM_ICDF(32768),
-            0 },
-          { AOM_ICDF(12288), AOM_ICDF(21248), AOM_ICDF(24128), AOM_ICDF(32768),
-            0 } },  // class0_fp_cdf
-        { AOM_ICDF(8192), AOM_ICDF(17408), AOM_ICDF(21248), AOM_ICDF(32768),
-          0 },  // fp_cdf
-        160,    // class0_hp bit
-        128,    // hp
-#if CONFIG_NEW_MULTISYMBOL
-        { AOM_ICDF(160 * 128), AOM_ICDF(32768), 0 },
-        { AOM_ICDF(128 * 128), AOM_ICDF(32768), 0 },
-        { AOM_ICDF(216 * 128), AOM_ICDF(32768), 0 },
-        { { AOM_ICDF(128 * 196), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 198), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 208), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 224), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 245), AOM_ICDF(32768), 0 },
-          { AOM_ICDF(128 * 240), AOM_ICDF(32768), 0 } },  // bits_cdf
-#endif
+        { AOM_CDF11(28672, 30976, 31858, 32320, 32551, 32656, 32740, 32757,
+                    32762, 32767) },  // class_cdf // fp
+        { { AOM_CDF4(16384, 24576, 26624) },
+          { AOM_CDF4(12288, 21248, 24128) } },  // class0_fp_cdf
+        { AOM_CDF4(8192, 17408, 21248) },       // fp_cdf
+        { AOM_CDF2(128 * 128) },                // sign_cdf
+        { AOM_CDF2(160 * 128) },                // class0_hp_cdf
+        { AOM_CDF2(128 * 128) },                // hp_cdf
+        { AOM_CDF2(216 * 128) },                // class0_cdf
+        { { AOM_CDF2(128 * 136) },
+          { AOM_CDF2(128 * 140) },
+          { AOM_CDF2(128 * 148) },
+          { AOM_CDF2(128 * 160) },
+          { AOM_CDF2(128 * 176) },
+          { AOM_CDF2(128 * 192) },
+          { AOM_CDF2(128 * 224) },
+          { AOM_CDF2(128 * 234) },
+          { AOM_CDF2(128 * 234) },
+          { AOM_CDF2(128 * 240) } },  // bits_cdf
     } },
 };
 
@@ -164,104 +115,8 @@ MV_CLASS_TYPE av1_get_mv_class(int z, int *offset) {
   return c;
 }
 
-static void inc_mv_component(int v, nmv_component_counts *comp_counts, int incr,
-                             MvSubpelPrecision precision) {
-  int s, z, c, o, d, e, f;
-  assert(v != 0); /* should not be zero */
-  s = v < 0;
-  comp_counts->sign[s] += incr;
-  z = (s ? -v : v) - 1; /* magnitude - 1 */
-
-  c = av1_get_mv_class(z, &o);
-  comp_counts->classes[c] += incr;
-
-  d = (o >> 3);     /* int mv data */
-  f = (o >> 1) & 3; /* fractional pel mv data */
-  e = (o & 1);      /* high precision mv data */
-
-  if (c == MV_CLASS_0) {
-    comp_counts->class0[d] += incr;
-#if CONFIG_INTRABC || CONFIG_AMVR
-    if (precision > MV_SUBPEL_NONE)
-#endif
-      comp_counts->class0_fp[d][f] += incr;
-    if (precision > MV_SUBPEL_LOW_PRECISION) comp_counts->class0_hp[e] += incr;
-  } else {
-    int i;
-    int b = c + CLASS0_BITS - 1;  // number of bits
-    for (i = 0; i < b; ++i) comp_counts->bits[i][((d >> i) & 1)] += incr;
-#if CONFIG_INTRABC || CONFIG_AMVR
-    if (precision > MV_SUBPEL_NONE)
-#endif
-      comp_counts->fp[f] += incr;
-    if (precision > MV_SUBPEL_LOW_PRECISION) comp_counts->hp[e] += incr;
-  }
-}
-
-void av1_inc_mv(const MV *mv, nmv_context_counts *counts,
-                MvSubpelPrecision precision) {
-  if (counts != NULL) {
-    const MV_JOINT_TYPE j = av1_get_mv_joint(mv);
-    ++counts->joints[j];
-
-    if (mv_joint_vertical(j))
-      inc_mv_component(mv->row, &counts->comps[0], 1, precision);
-
-    if (mv_joint_horizontal(j))
-      inc_mv_component(mv->col, &counts->comps[1], 1, precision);
-  }
-}
-
-void av1_adapt_mv_probs(AV1_COMMON *cm, int allow_hp) {
-  int i, j;
-  int idx;
-  for (idx = 0; idx < NMV_CONTEXTS; ++idx) {
-    nmv_context *nmvc = &cm->fc->nmvc[idx];
-    const nmv_context *pre_nmvc = &cm->pre_fc->nmvc[idx];
-    const nmv_context_counts *counts = &cm->counts.mv[idx];
-    aom_tree_merge_probs(av1_mv_joint_tree, pre_nmvc->joints, counts->joints,
-                         nmvc->joints);
-    for (i = 0; i < 2; ++i) {
-      nmv_component *comp = &nmvc->comps[i];
-      const nmv_component *pre_comp = &pre_nmvc->comps[i];
-      const nmv_component_counts *c = &counts->comps[i];
-
-      comp->sign = av1_mode_mv_merge_probs(pre_comp->sign, c->sign);
-      aom_tree_merge_probs(av1_mv_class_tree, pre_comp->classes, c->classes,
-                           comp->classes);
-      aom_tree_merge_probs(av1_mv_class0_tree, pre_comp->class0, c->class0,
-                           comp->class0);
-
-      for (j = 0; j < MV_OFFSET_BITS; ++j)
-        comp->bits[j] = av1_mode_mv_merge_probs(pre_comp->bits[j], c->bits[j]);
-#if CONFIG_AMVR
-      if (cm->cur_frame_mv_precision_level == 0) {
-#endif
-        for (j = 0; j < CLASS0_SIZE; ++j)
-          aom_tree_merge_probs(av1_mv_fp_tree, pre_comp->class0_fp[j],
-                               c->class0_fp[j], comp->class0_fp[j]);
-
-        aom_tree_merge_probs(av1_mv_fp_tree, pre_comp->fp, c->fp, comp->fp);
-
-        if (allow_hp) {
-          comp->class0_hp =
-              av1_mode_mv_merge_probs(pre_comp->class0_hp, c->class0_hp);
-          comp->hp = av1_mode_mv_merge_probs(pre_comp->hp, c->hp);
-        }
-#if CONFIG_AMVR
-      }
-#endif
-    }
-  }
-}
-
 void av1_init_mv_probs(AV1_COMMON *cm) {
-  int i;
-  for (i = 0; i < NMV_CONTEXTS; ++i) {
-    // NB: this sets CDFs too
-    cm->fc->nmvc[i] = default_nmv_context;
-  }
-#if CONFIG_INTRABC
+  // NB: this sets CDFs too
+  cm->fc->nmvc = default_nmv_context;
   cm->fc->ndvc = default_nmv_context;
-#endif  // CONFIG_INTRABC
 }

third_party/aom/av1/common/entropymv.h

@@ -12,7 +12,7 @@
 #ifndef AV1_COMMON_ENTROPYMV_H_
 #define AV1_COMMON_ENTROPYMV_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
 
 #include "aom_dsp/prob.h"
 
@@ -26,8 +26,6 @@ struct AV1Common;
 
 void av1_init_mv_probs(struct AV1Common *cm);
 
-void av1_adapt_mv_probs(struct AV1Common *cm, int usehp);
-
 #define MV_UPDATE_PROB 252
 
 /* Symbols for coding which components are zero jointly */
@@ -66,9 +64,7 @@ typedef enum {
 #define CLASS0_BITS 1 /* bits at integer precision for class 0 */
 #define CLASS0_SIZE (1 << CLASS0_BITS)
 #define MV_OFFSET_BITS (MV_CLASSES + CLASS0_BITS - 2)
-#if CONFIG_NEW_MULTISYMBOL
 #define MV_BITS_CONTEXTS 6
-#endif
 #define MV_FP_SIZE 4
 
 #define MV_MAX_BITS (MV_CLASSES + CLASS0_BITS + 2)
@@ -76,37 +72,22 @@ typedef enum {
 #define MV_VALS ((MV_MAX << 1) + 1)
 
 #define MV_IN_USE_BITS 14
-#define MV_UPP ((1 << MV_IN_USE_BITS) - 1)
+#define MV_UPP (1 << MV_IN_USE_BITS)
 #define MV_LOW (-(1 << MV_IN_USE_BITS))
 
-extern const aom_tree_index av1_mv_joint_tree[];
-extern const aom_tree_index av1_mv_class_tree[];
-extern const aom_tree_index av1_mv_class0_tree[];
-extern const aom_tree_index av1_mv_fp_tree[];
-
 typedef struct {
-  aom_prob sign;
-  aom_prob classes[MV_CLASSES - 1];
-  aom_cdf_prob class_cdf[CDF_SIZE(MV_CLASSES)];
-  aom_prob class0[CLASS0_SIZE - 1];
-  aom_prob bits[MV_OFFSET_BITS];
-  aom_prob class0_fp[CLASS0_SIZE][MV_FP_SIZE - 1];
-  aom_prob fp[MV_FP_SIZE - 1];
+  aom_cdf_prob classes_cdf[CDF_SIZE(MV_CLASSES)];
   aom_cdf_prob class0_fp_cdf[CLASS0_SIZE][CDF_SIZE(MV_FP_SIZE)];
   aom_cdf_prob fp_cdf[CDF_SIZE(MV_FP_SIZE)];
-  aom_prob class0_hp;
-  aom_prob hp;
-#if CONFIG_NEW_MULTISYMBOL
+  aom_cdf_prob sign_cdf[CDF_SIZE(2)];
   aom_cdf_prob class0_hp_cdf[CDF_SIZE(2)];
   aom_cdf_prob hp_cdf[CDF_SIZE(2)];
   aom_cdf_prob class0_cdf[CDF_SIZE(CLASS0_SIZE)];
-  aom_cdf_prob bits_cdf[MV_BITS_CONTEXTS][CDF_SIZE(2)];
-#endif
+  aom_cdf_prob bits_cdf[MV_OFFSET_BITS][CDF_SIZE(2)];
 } nmv_component;
 
 typedef struct {
-  aom_prob joints[MV_JOINTS - 1];
-  aom_cdf_prob joint_cdf[CDF_SIZE(MV_JOINTS)];
+  aom_cdf_prob joints_cdf[CDF_SIZE(MV_JOINTS)];
   nmv_component comps[2];
 } nmv_context;
 
@@ -120,33 +101,12 @@ static INLINE MV_JOINT_TYPE av1_get_mv_joint(const MV *mv) {
 
 MV_CLASS_TYPE av1_get_mv_class(int z, int *offset);
 
-typedef struct {
-  unsigned int sign[2];
-  unsigned int classes[MV_CLASSES];
-  unsigned int class0[CLASS0_SIZE];
-  unsigned int bits[MV_OFFSET_BITS][2];
-  unsigned int class0_fp[CLASS0_SIZE][MV_FP_SIZE];
-  unsigned int fp[MV_FP_SIZE];
-  unsigned int class0_hp[2];
-  unsigned int hp[2];
-} nmv_component_counts;
-
-typedef struct {
-  unsigned int joints[MV_JOINTS];
-  nmv_component_counts comps[2];
-} nmv_context_counts;
-
 typedef enum {
-#if CONFIG_INTRABC || CONFIG_AMVR
   MV_SUBPEL_NONE = -1,
-#endif
   MV_SUBPEL_LOW_PRECISION = 0,
   MV_SUBPEL_HIGH_PRECISION,
 } MvSubpelPrecision;
 
-void av1_inc_mv(const MV *mv, nmv_context_counts *mvctx,
-                MvSubpelPrecision precision);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/enums.h

@@ -12,7 +12,8 @@
 #ifndef AV1_COMMON_ENUMS_H_
 #define AV1_COMMON_ENUMS_H_
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom/aom_codec.h"
 #include "aom/aom_integer.h"
 
@@ -22,22 +23,8 @@ extern "C" {
 
 #undef MAX_SB_SIZE
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-#define TWO_MODE
-#endif
-
-#if CONFIG_NCOBMC || CONFIG_NCOBMC_ADAPT_WEIGHT
-#define NC_MODE_INFO 1
-#else
-#define NC_MODE_INFO 0
-#endif
-
 // Max superblock size
-#if CONFIG_EXT_PARTITION
 #define MAX_SB_SIZE_LOG2 7
-#else
-#define MAX_SB_SIZE_LOG2 6
-#endif  // CONFIG_EXT_PARTITION
 #define MAX_SB_SIZE (1 << MAX_SB_SIZE_LOG2)
 #define MAX_SB_SQUARE (MAX_SB_SIZE * MAX_SB_SIZE)
 
@@ -45,11 +32,7 @@ extern "C" {
 #define MIN_SB_SIZE_LOG2 6
 
 // Pixels per Mode Info (MI) unit
-#if CONFIG_CB4X4
 #define MI_SIZE_LOG2 2
-#else
-#define MI_SIZE_LOG2 3
-#endif
 #define MI_SIZE (1 << MI_SIZE_LOG2)
 
 // MI-units per max superblock (MI Block - MIB)
@@ -63,73 +46,78 @@ extern "C" {
 #define MAX_MIB_MASK (MAX_MIB_SIZE - 1)
 
 // Maximum number of tile rows and tile columns
-#if CONFIG_EXT_TILE
-#define MAX_TILE_ROWS 1024
-#define MAX_TILE_COLS 1024
-#else
-#if CONFIG_MAX_TILE
 #define MAX_TILE_ROWS 64
 #define MAX_TILE_COLS 64
-#else
-#define MAX_TILE_ROWS 4
-#define MAX_TILE_COLS 64
-#endif
-#endif  // CONFIG_EXT_TILE
 
-#if CONFIG_VAR_TX
 #define MAX_VARTX_DEPTH 2
-#define SQR_VARTX_DEPTH_INIT 0
-#define RECT_VARTX_DEPTH_INIT 0
-#endif
 
 #define MI_SIZE_64X64 (64 >> MI_SIZE_LOG2)
+#define MI_SIZE_128X128 (128 >> MI_SIZE_LOG2)
+
+#define MAX_PALETTE_SQUARE (64 * 64)
+// Maximum number of colors in a palette.
+#define PALETTE_MAX_SIZE 8
+// Minimum number of colors in a palette.
+#define PALETTE_MIN_SIZE 2
+
+#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)
 
-#if CONFIG_LOOPFILTER_LEVEL
 // 4 frame filter levels: y plane vertical, y plane horizontal,
 // u plane, and v plane
 #define FRAME_LF_COUNT 4
 #define DEFAULT_DELTA_LF_MULTI 0
-#endif  // CONFIG_LOOPFILTER_LEVEL
+#define MAX_MODE_LF_DELTAS 2
 
-#if CONFIG_LPF_SB
-#define LPF_DELTA_BITS 3
-#define LPF_STEP 2
-#define DELTA_RANGE (1 << LPF_DELTA_BITS)
-#define MAX_LPF_OFFSET (LPF_STEP * ((1 << LPF_DELTA_BITS) - 1))
+#define DIST_PRECISION_BITS 4
+#define DIST_PRECISION (1 << DIST_PRECISION_BITS)  // 16
 
-#define LPF_REUSE_CONTEXT 2
-#define LPF_DELTA_CONTEXT DELTA_RANGE
-#define LPF_SIGN_CONTEXT 2
+// TODO(chengchen): Temporal flag serve as experimental flag for WIP
+// bitmask construction.
+// Shall be removed when bitmask code is completely checkedin
+#define LOOP_FILTER_BITMASK 0
 
-// Half of maximum loop filter length (15-tap)
-#define FILT_BOUNDARY_OFFSET 8
-#define FILT_BOUNDARY_MI_OFFSET (FILT_BOUNDARY_OFFSET >> MI_SIZE_LOG2)
-#endif  // CONFIG_LPF_SB
-
-// Bitstream profiles indicated by 2-3 bits in the uncompressed header.
-// 00: Profile 0.  8-bit 4:2:0 only.
-// 10: Profile 1.  8-bit 4:4:4, 4:2:2, and 4:4:0.
-// 01: Profile 2.  10-bit and 12-bit color only, with 4:2:0 sampling.
-// 110: Profile 3. 10-bit and 12-bit color only, with 4:2:2/4:4:4/4:4:0
-//                 sampling.
-// 111: Undefined profile.
+#define PROFILE_BITS 3
+// The following three profiles are currently defined.
+// Profile 0.  8-bit and 10-bit 4:2:0 and 4:0:0 only.
+// Profile 1.  8-bit and 10-bit 4:4:4
+// 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 {
   PROFILE_0,
   PROFILE_1,
   PROFILE_2,
-  PROFILE_3,
-  MAX_PROFILES
+  MAX_PROFILES,
 } BITSTREAM_PROFILE;
 
+#define LEVEL_MAJOR_BITS 3
+#define LEVEL_MINOR_BITS 2
+#define LEVEL_BITS (LEVEL_MAJOR_BITS + LEVEL_MINOR_BITS)
+
+#define LEVEL_MAJOR_MIN 2
+#define LEVEL_MAJOR_MAX ((1 << LEVEL_MAJOR_BITS) - 1 + LEVEL_MAJOR_MIN)
+#define LEVEL_MINOR_MIN 0
+#define LEVEL_MINOR_MAX ((1 << LEVEL_MINOR_BITS) - 1)
+
+#define OP_POINTS_CNT_MINUS_1_BITS 5
+#define OP_POINTS_IDC_BITS 12
+
 // Note: Some enums use the attribute 'packed' to use smallest possible integer
 // type, so that we can save memory when they are used in structs/arrays.
 
 typedef enum ATTRIBUTE_PACKED {
-#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-  BLOCK_2X2,
-  BLOCK_2X4,
-  BLOCK_4X2,
-#endif
   BLOCK_4X4,
   BLOCK_4X8,
   BLOCK_8X4,
@@ -143,33 +131,29 @@ typedef enum ATTRIBUTE_PACKED {
   BLOCK_32X64,
   BLOCK_64X32,
   BLOCK_64X64,
-#if CONFIG_EXT_PARTITION
   BLOCK_64X128,
   BLOCK_128X64,
   BLOCK_128X128,
-#endif  // CONFIG_EXT_PARTITION
   BLOCK_4X16,
   BLOCK_16X4,
   BLOCK_8X32,
   BLOCK_32X8,
   BLOCK_16X64,
   BLOCK_64X16,
-#if CONFIG_EXT_PARTITION
-  BLOCK_32X128,
-  BLOCK_128X32,
-#endif  // CONFIG_EXT_PARTITION
   BLOCK_SIZES_ALL,
   BLOCK_SIZES = BLOCK_4X16,
   BLOCK_INVALID = 255,
   BLOCK_LARGEST = (BLOCK_SIZES - 1)
 } BLOCK_SIZE;
 
-typedef enum {
+// 4X4, 8X8, 16X16, 32X32, 64X64, 128X128
+#define SQR_BLOCK_SIZES 6
+
+typedef enum ATTRIBUTE_PACKED {
   PARTITION_NONE,
   PARTITION_HORZ,
   PARTITION_VERT,
   PARTITION_SPLIT,
-#if CONFIG_EXT_PARTITION_TYPES
   PARTITION_HORZ_A,  // HORZ split and the top partition is split again
   PARTITION_HORZ_B,  // HORZ split and the bottom partition is split again
   PARTITION_VERT_A,  // VERT split and the left partition is split again
@@ -177,134 +161,104 @@ typedef enum {
   PARTITION_HORZ_4,  // 4:1 horizontal partition
   PARTITION_VERT_4,  // 4:1 vertical partition
   EXT_PARTITION_TYPES,
-#endif  // CONFIG_EXT_PARTITION_TYPES
   PARTITION_TYPES = PARTITION_SPLIT + 1,
   PARTITION_INVALID = 255
 } PARTITION_TYPE;
 
 typedef char PARTITION_CONTEXT;
 #define PARTITION_PLOFFSET 4  // number of probability models per block size
-#define PARTITION_BLOCK_SIZES (4 + CONFIG_EXT_PARTITION)
-#define PARTITION_CONTEXTS_PRIMARY (PARTITION_BLOCK_SIZES * PARTITION_PLOFFSET)
-#if CONFIG_UNPOISON_PARTITION_CTX
-#define INVALID_PARTITION_CTX (-1)
-#define PARTITION_CONTEXTS \
-  (PARTITION_CONTEXTS_PRIMARY + 2 * PARTITION_BLOCK_SIZES)
-#else
-#define PARTITION_CONTEXTS PARTITION_CONTEXTS_PRIMARY
-#endif
+#define PARTITION_BLOCK_SIZES 5
+#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 {
-#if CONFIG_CHROMA_2X2
-  TX_2X2,  // 2x2 transform
 #endif
-  TX_4X4,    // 4x4 transform
-  TX_8X8,    // 8x8 transform
-  TX_16X16,  // 16x16 transform
-  TX_32X32,  // 32x32 transform
-#if CONFIG_TX64X64
-  TX_64X64,  // 64x64 transform
-#endif       // CONFIG_TX64X64
-  TX_4X8,    // 4x8 transform
-  TX_8X4,    // 8x4 transform
-  TX_8X16,   // 8x16 transform
-  TX_16X8,   // 16x8 transform
-  TX_16X32,  // 16x32 transform
-  TX_32X16,  // 32x16 transform
-#if CONFIG_TX64X64
+  TX_4X4,             // 4x4 transform
+  TX_8X8,             // 8x8 transform
+  TX_16X16,           // 16x16 transform
+  TX_32X32,           // 32x32 transform
+  TX_64X64,           // 64x64 transform
+  TX_4X8,             // 4x8 transform
+  TX_8X4,             // 8x4 transform
+  TX_8X16,            // 8x16 transform
+  TX_16X8,            // 16x8 transform
+  TX_16X32,           // 16x32 transform
+  TX_32X16,           // 32x16 transform
   TX_32X64,           // 32x64 transform
   TX_64X32,           // 64x32 transform
-#endif                // CONFIG_TX64X64
   TX_4X16,            // 4x16 transform
   TX_16X4,            // 16x4 transform
   TX_8X32,            // 8x32 transform
   TX_32X8,            // 32x8 transform
+  TX_16X64,           // 16x64 transform
+  TX_64X16,           // 64x16 transform
   TX_SIZES_ALL,       // Includes rectangular transforms
   TX_SIZES = TX_4X8,  // Does NOT include rectangular transforms
-  TX_INVALID = 255    // Invalid transform size
+  TX_SIZES_LARGEST = TX_64X64,
+  TX_INVALID = 255  // Invalid transform size
+#if defined(_MSC_VER)
+};
+#else
 } TX_SIZE;
+#endif
 
 #define TX_SIZE_LUMA_MIN (TX_4X4)
 /* We don't need to code a transform size unless the allowed size is at least
    one more than the minimum. */
 #define TX_SIZE_CTX_MIN (TX_SIZE_LUMA_MIN + 1)
 
-#define MAX_TX_DEPTH (TX_SIZES - TX_SIZE_CTX_MIN)
+// Maximum tx_size categories
+#define MAX_TX_CATS (TX_SIZES - TX_SIZE_CTX_MIN)
+#define MAX_TX_DEPTH 2
 
-#if CONFIG_CTX1D
-#define MAX_HVTX_SIZE (1 << 5)
-#endif  // CONFIG_CTX1D
-
-#define MAX_TX_SIZE_LOG2 (5 + CONFIG_TX64X64)
+#define MAX_TX_SIZE_LOG2 (6)
 #define MAX_TX_SIZE (1 << MAX_TX_SIZE_LOG2)
 #define MIN_TX_SIZE_LOG2 2
 #define MIN_TX_SIZE (1 << MIN_TX_SIZE_LOG2)
 #define MAX_TX_SQUARE (MAX_TX_SIZE * MAX_TX_SIZE)
 
+// Pad 4 extra columns to remove horizontal availability check.
+#define TX_PAD_HOR_LOG2 2
+#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_BOTTOM 4
+#define TX_PAD_VER (TX_PAD_TOP + TX_PAD_BOTTOM)
+// Pad 16 extra bytes to avoid reading overflow in SIMD optimization.
+#define TX_PAD_END 16
+#define TX_PAD_2D ((32 + TX_PAD_HOR) * (32 + TX_PAD_VER) + TX_PAD_END)
+
 // Number of maxium size transform blocks in the maximum size superblock
 #define MAX_TX_BLOCKS_IN_MAX_SB_LOG2 ((MAX_SB_SIZE_LOG2 - MAX_TX_SIZE_LOG2) * 2)
 #define MAX_TX_BLOCKS_IN_MAX_SB (1 << MAX_TX_BLOCKS_IN_MAX_SB_LOG2)
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-typedef enum ATTRIBUTE_PACKED {
-  NCOBMC_MODE_0,
-  NCOBMC_MODE_1,
-  NCOBMC_MODE_2,
-  NCOBMC_MODE_3,
-  NCOBMC_MODE_4,
-  NCOBMC_MODE_5,
-  NCOBMC_MODE_6,
-  NCOBMC_MODE_7,
-  ALL_NCOBMC_MODES,
-#ifdef TWO_MODE
-  MAX_NCOBMC_MODES = NCOBMC_MODE_1 + 1,
-#else
-  MAX_NCOBMC_MODES = ALL_NCOBMC_MODES,
-#endif
-  NO_OVERLAP = MAX_NCOBMC_MODES + 1
-} NCOBMC_MODE;
-
-typedef enum {
-  ADAPT_OVERLAP_BLOCK_8X8,
-  ADAPT_OVERLAP_BLOCK_16X16,
-  ADAPT_OVERLAP_BLOCK_32X32,
-  ADAPT_OVERLAP_BLOCK_64X64,
-  ADAPT_OVERLAP_BLOCKS,
-  ADAPT_OVERLAP_BLOCK_INVALID = 255
-} ADAPT_OVERLAP_BLOCK;
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-
 // frame transform mode
-typedef enum {
-  ONLY_4X4,     // only 4x4 transform used
-  ALLOW_8X8,    // allow block transform size up to 8x8
-  ALLOW_16X16,  // allow block transform size up to 16x16
-  ALLOW_32X32,  // allow block transform size up to 32x32
-#if CONFIG_TX64X64
-  ALLOW_64X64,  // allow block transform size up to 64x64
-#endif
-  TX_MODE_SELECT,  // transform specified for each block
+typedef enum ATTRIBUTE_PACKED {
+  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;
 
 // 1D tx types
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   DCT_1D,
   ADST_1D,
   FLIPADST_1D,
   IDTX_1D,
-  // TODO(sarahparker) need to eventually put something here for the
-  // mrc experiment to make this work with the ext-tx pruning functions
   TX_TYPES_1D,
 } TX_TYPE_1D;
 
-typedef enum {
+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
-#if CONFIG_EXT_TX
   FLIPADST_DCT,
   DCT_FLIPADST,
   FLIPADST_FLIPADST,
@@ -317,25 +271,26 @@ typedef enum {
   H_ADST,
   V_FLIPADST,
   H_FLIPADST,
-#endif  // CONFIG_EXT_TX
-#if CONFIG_MRC_TX
-  MRC_DCT,  // DCT in both directions with mrc based bitmask
-#endif      // CONFIG_MRC_TX
   TX_TYPES,
 } TX_TYPE;
 
-#if CONFIG_EXT_TX
 typedef enum {
+  REG_REG,
+  REG_SMOOTH,
+  REG_SHARP,
+  SMOOTH_REG,
+  SMOOTH_SMOOTH,
+  SMOOTH_SHARP,
+  SHARP_REG,
+  SHARP_SMOOTH,
+  SHARP_SHARP,
+} DUAL_FILTER_TYPE;
+
+typedef enum ATTRIBUTE_PACKED {
   // DCT only
   EXT_TX_SET_DCTONLY,
   // DCT + Identity only
   EXT_TX_SET_DCT_IDTX,
-#if CONFIG_MRC_TX
-  // DCT + MRC_DCT
-  EXT_TX_SET_MRC_DCT,
-  // DCT + MRC_DCT + IDTX
-  EXT_TX_SET_MRC_DCT_IDTX,
-#endif  // CONFIG_MRC_TX
   // Discrete Trig transforms w/o flip (4) + Identity (1)
   EXT_TX_SET_DTT4_IDTX,
   // Discrete Trig transforms w/o flip (4) + Identity (1) + 1D Hor/vert DCT (2)
@@ -348,45 +303,13 @@ typedef enum {
 } TxSetType;
 
 #define IS_2D_TRANSFORM(tx_type) (tx_type < IDTX)
-#else
-#define IS_2D_TRANSFORM(tx_type) 1
-#endif
 
-typedef enum {
-  TILE_LEFT_BOUNDARY = 1,
-  TILE_RIGHT_BOUNDARY = 2,
-  TILE_ABOVE_BOUNDARY = 4,
-  TILE_BOTTOM_BOUNDARY = 8,
-  FRAME_LEFT_BOUNDARY = 16,
-  FRAME_RIGHT_BOUNDARY = 32,
-  FRAME_ABOVE_BOUNDARY = 64,
-  FRAME_BOTTOM_BOUNDARY = 128,
-} BOUNDARY_TYPE;
-
-#if CONFIG_EXT_TX
-#if CONFIG_CHROMA_2X2
-#define EXT_TX_SIZES 5  // number of sizes that use extended transforms
-#else
-#define EXT_TX_SIZES 4  // number of sizes that use extended transforms
-#endif                  // CONFIG_CHROMA_2X2
-#if CONFIG_MRC_TX
-#define EXT_TX_SETS_INTER 5  // Sets of transform selections for INTER
-#define EXT_TX_SETS_INTRA 4  // Sets of transform selections for INTRA
-#else                        // CONFIG_MRC_TX
+#define EXT_TX_SIZES 4       // number of sizes that use extended transforms
 #define EXT_TX_SETS_INTER 4  // Sets of transform selections for INTER
 #define EXT_TX_SETS_INTRA 3  // Sets of transform selections for INTRA
-#endif                       // CONFIG_MRC_TX
-#else
-#if CONFIG_CHROMA_2X2
-#define EXT_TX_SIZES 4  // number of sizes that use extended transforms
-#else
-#define EXT_TX_SIZES 3  // number of sizes that use extended transforms
-#endif
-#endif  // CONFIG_EXT_TX
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   AOM_LAST_FLAG = 1 << 0,
-#if CONFIG_EXT_REFS
   AOM_LAST2_FLAG = 1 << 1,
   AOM_LAST3_FLAG = 1 << 2,
   AOM_GOLD_FLAG = 1 << 3,
@@ -394,43 +317,45 @@ typedef enum {
   AOM_ALT2_FLAG = 1 << 5,
   AOM_ALT_FLAG = 1 << 6,
   AOM_REFFRAME_ALL = (1 << 7) - 1
-#else   // !CONFIG_EXT_REFS
-  AOM_GOLD_FLAG = 1 << 1,
-  AOM_ALT_FLAG = 1 << 2,
-  AOM_REFFRAME_ALL = (1 << 3) - 1
-#endif  // CONFIG_EXT_REFS
 } AOM_REFFRAME;
 
-#if CONFIG_EXT_COMP_REFS
-#define USE_UNI_COMP_REFS 1
-
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   UNIDIR_COMP_REFERENCE,
   BIDIR_COMP_REFERENCE,
   COMP_REFERENCE_TYPES,
 } COMP_REFERENCE_TYPE;
-#else  // !CONFIG_EXT_COMP_REFS
-#define USE_UNI_COMP_REFS 0
-#endif  // CONFIG_EXT_COMP_REFS
 
-typedef enum { PLANE_TYPE_Y, PLANE_TYPE_UV, PLANE_TYPES } PLANE_TYPE;
+typedef enum ATTRIBUTE_PACKED {
+  PLANE_TYPE_Y,
+  PLANE_TYPE_UV,
+  PLANE_TYPES
+} PLANE_TYPE;
 
-#if CONFIG_CFL
 #define CFL_ALPHABET_SIZE_LOG2 4
 #define CFL_ALPHABET_SIZE (1 << CFL_ALPHABET_SIZE_LOG2)
 #define CFL_MAGS_SIZE ((2 << CFL_ALPHABET_SIZE_LOG2) + 1)
 #define CFL_IDX_U(idx) (idx >> CFL_ALPHABET_SIZE_LOG2)
 #define CFL_IDX_V(idx) (idx & (CFL_ALPHABET_SIZE - 1))
 
-typedef enum { CFL_PRED_U, CFL_PRED_V, CFL_PRED_PLANES } CFL_PRED_TYPE;
+typedef enum ATTRIBUTE_PACKED {
+  CFL_PRED_U,
+  CFL_PRED_V,
+  CFL_PRED_PLANES
+} CFL_PRED_TYPE;
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   CFL_SIGN_ZERO,
   CFL_SIGN_NEG,
   CFL_SIGN_POS,
   CFL_SIGNS
 } CFL_SIGN_TYPE;
 
+typedef enum ATTRIBUTE_PACKED {
+  CFL_DISALLOWED,
+  CFL_ALLOWED,
+  CFL_ALLOWED_TYPES
+} CFL_ALLOWED_TYPE;
+
 // CFL_SIGN_ZERO,CFL_SIGN_ZERO is invalid
 #define CFL_JOINT_SIGNS (CFL_SIGNS * CFL_SIGNS - 1)
 // CFL_SIGN_U is equivalent to (js + 1) / 3 for js in 0 to 8
@@ -445,17 +370,13 @@ typedef enum {
 // Also, the contexts are symmetric under swapping the planes.
 #define CFL_CONTEXT_V(js) \
   (CFL_SIGN_V(js) * CFL_SIGNS + CFL_SIGN_U(js) - CFL_SIGNS)
-#endif
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   PALETTE_MAP,
-#if CONFIG_MRC_TX
-  MRC_MAP,
-#endif  // CONFIG_MRC_TX
   COLOR_MAP_TYPES,
 } COLOR_MAP_TYPE;
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   TWO_COLORS,
   THREE_COLORS,
   FOUR_COLORS,
@@ -466,7 +387,7 @@ typedef enum {
   PALETTE_SIZES
 } PALETTE_SIZE;
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   PALETTE_COLOR_ONE,
   PALETTE_COLOR_TWO,
   PALETTE_COLOR_THREE,
@@ -478,36 +399,26 @@ typedef enum {
   PALETTE_COLORS
 } PALETTE_COLOR;
 
-// Note: All directional predictors must be between V_PRED and D63_PRED (both
+// Note: All directional predictors must be between V_PRED and D67_PRED (both
 // inclusive).
 typedef enum ATTRIBUTE_PACKED {
-  DC_PRED,      // Average of above and left pixels
-  V_PRED,       // Vertical
-  H_PRED,       // Horizontal
-  D45_PRED,     // Directional 45  deg = round(arctan(1/1) * 180/pi)
-  D135_PRED,    // Directional 135 deg = 180 - 45
-  D117_PRED,    // Directional 117 deg = 180 - 63
-  D153_PRED,    // Directional 153 deg = 180 - 27
-  D207_PRED,    // Directional 207 deg = 180 + 27
-  D63_PRED,     // Directional 63  deg = round(arctan(2/1) * 180/pi)
-  SMOOTH_PRED,  // Combination of horizontal and vertical interpolation
-#if CONFIG_SMOOTH_HV
+  DC_PRED,        // Average of above and left pixels
+  V_PRED,         // Vertical
+  H_PRED,         // Horizontal
+  D45_PRED,       // Directional 45  degree
+  D135_PRED,      // Directional 135 degree
+  D113_PRED,      // Directional 113 degree
+  D157_PRED,      // Directional 157 degree
+  D203_PRED,      // Directional 203 degree
+  D67_PRED,       // Directional 67  degree
+  SMOOTH_PRED,    // Combination of horizontal and vertical interpolation
   SMOOTH_V_PRED,  // Vertical interpolation
   SMOOTH_H_PRED,  // Horizontal interpolation
-#endif            // CONFIG_SMOOTH_HV
-  TM_PRED,        // True-motion
+  PAETH_PRED,     // Predict from the direction of smallest gradient
   NEARESTMV,
   NEARMV,
-  ZEROMV,
+  GLOBALMV,
   NEWMV,
-#if CONFIG_COMPOUND_SINGLEREF
-  // Single ref compound modes
-  SR_NEAREST_NEARMV,
-  // SR_NEAREST_NEWMV,
-  SR_NEAR_NEWMV,
-  SR_ZERO_NEWMV,
-  SR_NEW_NEWMV,
-#endif  // CONFIG_COMPOUND_SINGLEREF
   // Compound ref compound modes
   NEAREST_NEARESTMV,
   NEAR_NEARMV,
@@ -515,175 +426,131 @@ typedef enum ATTRIBUTE_PACKED {
   NEW_NEARESTMV,
   NEAR_NEWMV,
   NEW_NEARMV,
-  ZERO_ZEROMV,
+  GLOBAL_GLOBALMV,
   NEW_NEWMV,
   MB_MODE_COUNT,
-  INTRA_MODES = TM_PRED + 1,     // TM_PRED has to be the last intra mode.
+  INTRA_MODE_START = DC_PRED,
+  INTRA_MODE_END = NEARESTMV,
+  INTRA_MODE_NUM = INTRA_MODE_END - INTRA_MODE_START,
+  SINGLE_INTER_MODE_START = NEARESTMV,
+  SINGLE_INTER_MODE_END = NEAREST_NEARESTMV,
+  SINGLE_INTER_MODE_NUM = SINGLE_INTER_MODE_END - SINGLE_INTER_MODE_START,
+  COMP_INTER_MODE_START = NEAREST_NEARESTMV,
+  COMP_INTER_MODE_END = MB_MODE_COUNT,
+  COMP_INTER_MODE_NUM = COMP_INTER_MODE_END - COMP_INTER_MODE_START,
+  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;
 
-#if CONFIG_CFL
 // TODO(ltrudeau) Do we really want to pack this?
 // TODO(ltrudeau) Do we match with PREDICTION_MODE?
 typedef enum ATTRIBUTE_PACKED {
-  UV_DC_PRED,      // Average of above and left pixels
-  UV_V_PRED,       // Vertical
-  UV_H_PRED,       // Horizontal
-  UV_D45_PRED,     // Directional 45  deg = round(arctan(1/1) * 180/pi)
-  UV_D135_PRED,    // Directional 135 deg = 180 - 45
-  UV_D117_PRED,    // Directional 117 deg = 180 - 63
-  UV_D153_PRED,    // Directional 153 deg = 180 - 27
-  UV_D207_PRED,    // Directional 207 deg = 180 + 27
-  UV_D63_PRED,     // Directional 63  deg = round(arctan(2/1) * 180/pi)
-  UV_SMOOTH_PRED,  // Combination of horizontal and vertical interpolation
-#if CONFIG_SMOOTH_HV
+  UV_DC_PRED,        // Average of above and left pixels
+  UV_V_PRED,         // Vertical
+  UV_H_PRED,         // Horizontal
+  UV_D45_PRED,       // Directional 45  degree
+  UV_D135_PRED,      // Directional 135 degree
+  UV_D113_PRED,      // Directional 113 degree
+  UV_D157_PRED,      // Directional 157 degree
+  UV_D203_PRED,      // Directional 203 degree
+  UV_D67_PRED,       // Directional 67  degree
+  UV_SMOOTH_PRED,    // Combination of horizontal and vertical interpolation
   UV_SMOOTH_V_PRED,  // Vertical interpolation
   UV_SMOOTH_H_PRED,  // Horizontal interpolation
-#endif               // CONFIG_SMOOTH_HV
-  UV_TM_PRED,        // True-motion
+  UV_PAETH_PRED,     // Predict from the direction of smallest gradient
   UV_CFL_PRED,       // Chroma-from-Luma
   UV_INTRA_MODES,
   UV_MODE_INVALID,  // For uv_mode in inter blocks
 } UV_PREDICTION_MODE;
-#else
-#define UV_INTRA_MODES (INTRA_MODES)
-#define UV_PREDICTION_MODE PREDICTION_MODE
-#define UV_DC_PRED (DC_PRED)
-#define UV_MODE_INVALID (INTRA_INVALID)
-#endif  // CONFIG_CFL
 
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   SIMPLE_TRANSLATION,
-#if CONFIG_MOTION_VAR
-  OBMC_CAUSAL,  // 2-sided OBMC
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-  NCOBMC_ADAPT_WEIGHT,
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
-#endif  // CONFIG_MOTION_VAR
-#if CONFIG_WARPED_MOTION
+  OBMC_CAUSAL,    // 2-sided OBMC
   WARPED_CAUSAL,  // 2-sided WARPED
-#endif            // CONFIG_WARPED_MOTION
   MOTION_MODES
-#if CONFIG_NCOBMC_ADAPT_WEIGHT && CONFIG_WARPED_MOTION
-  ,
-  OBMC_FAMILY_MODES = NCOBMC_ADAPT_WEIGHT + 1
-#endif
 } MOTION_MODE;
 
-#if CONFIG_INTERINTRA
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   II_DC_PRED,
   II_V_PRED,
   II_H_PRED,
   II_SMOOTH_PRED,
   INTERINTRA_MODES
 } INTERINTRA_MODE;
-#endif
 
 typedef enum {
   COMPOUND_AVERAGE,
-#if CONFIG_WEDGE
   COMPOUND_WEDGE,
-#endif  // CONFIG_WEDGE
-#if CONFIG_COMPOUND_SEGMENT
-  COMPOUND_SEG,
-#endif  // CONFIG_COMPOUND_SEGMENT
+  COMPOUND_DIFFWTD,
   COMPOUND_TYPES,
 } COMPOUND_TYPE;
 
-// TODO(huisu): Consider adding FILTER_SMOOTH_PRED to "FILTER_INTRA_MODE".
-#if CONFIG_FILTER_INTRA
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   FILTER_DC_PRED,
   FILTER_V_PRED,
   FILTER_H_PRED,
-  FILTER_D45_PRED,
-  FILTER_D135_PRED,
-  FILTER_D117_PRED,
-  FILTER_D153_PRED,
-  FILTER_D207_PRED,
-  FILTER_D63_PRED,
-  FILTER_TM_PRED,
+  FILTER_D157_PRED,
+  FILTER_PAETH_PRED,
   FILTER_INTRA_MODES,
 } FILTER_INTRA_MODE;
-#endif  // CONFIG_FILTER_INTRA
 
-#if CONFIG_EXT_INTRA
 #define DIRECTIONAL_MODES 8
-#endif  // CONFIG_EXT_INTRA
+#define MAX_ANGLE_DELTA 3
+#define ANGLE_STEP 3
 
 #define INTER_MODES (1 + NEWMV - NEARESTMV)
 
-#if CONFIG_COMPOUND_SINGLEREF
-#define INTER_SINGLEREF_COMP_MODES (1 + SR_NEW_NEWMV - SR_NEAREST_NEARMV)
-#endif  // CONFIG_COMPOUND_SINGLEREF
-
 #define INTER_COMPOUND_MODES (1 + NEW_NEWMV - NEAREST_NEARESTMV)
 
 #define SKIP_CONTEXTS 3
+#define SKIP_MODE_CONTEXTS 3
+
+#define COMP_INDEX_CONTEXTS 6
+#define COMP_GROUP_IDX_CONTEXTS 6
 
 #define NMV_CONTEXTS 3
 
-#define NEWMV_MODE_CONTEXTS 7
-#define ZEROMV_MODE_CONTEXTS 2
-#define REFMV_MODE_CONTEXTS 9
-#define DRL_MODE_CONTEXTS 5
+#define NEWMV_MODE_CONTEXTS 6
+#define GLOBALMV_MODE_CONTEXTS 2
+#define REFMV_MODE_CONTEXTS 6
+#define DRL_MODE_CONTEXTS 3
 
-#define ZEROMV_OFFSET 3
+#define GLOBALMV_OFFSET 3
 #define REFMV_OFFSET 4
 
-#define NEWMV_CTX_MASK ((1 << ZEROMV_OFFSET) - 1)
-#define ZEROMV_CTX_MASK ((1 << (REFMV_OFFSET - ZEROMV_OFFSET)) - 1)
+#define NEWMV_CTX_MASK ((1 << GLOBALMV_OFFSET) - 1)
+#define GLOBALMV_CTX_MASK ((1 << (REFMV_OFFSET - GLOBALMV_OFFSET)) - 1)
 #define REFMV_CTX_MASK ((1 << (8 - REFMV_OFFSET)) - 1)
 
-#define ALL_ZERO_FLAG_OFFSET 8
-#define SKIP_NEARESTMV_OFFSET 9
-#define SKIP_NEARMV_OFFSET 10
-#define SKIP_NEARESTMV_SUB8X8_OFFSET 11
+#define COMP_NEWMV_CTXS 5
+#define INTER_MODE_CONTEXTS 8
 
-#define INTER_MODE_CONTEXTS 7
 #define DELTA_Q_SMALL 3
 #define DELTA_Q_PROBS (DELTA_Q_SMALL)
 #define DEFAULT_DELTA_Q_RES 4
-#if CONFIG_EXT_DELTA_Q
 #define DELTA_LF_SMALL 3
 #define DELTA_LF_PROBS (DELTA_LF_SMALL)
 #define DEFAULT_DELTA_LF_RES 2
-#endif
 
 /* Segment Feature Masks */
 #define MAX_MV_REF_CANDIDATES 2
 
-#define MAX_REF_MV_STACK_SIZE 16
-#if CONFIG_EXT_PARTITION
+#define MAX_REF_MV_STACK_SIZE 8
 #define REF_CAT_LEVEL 640
-#else
-#define REF_CAT_LEVEL 255
-#endif  // CONFIG_EXT_PARTITION
 
 #define INTRA_INTER_CONTEXTS 4
 #define COMP_INTER_CONTEXTS 5
-#define REF_CONTEXTS 5
+#define REF_CONTEXTS 3
 
-#if CONFIG_EXT_COMP_REFS
 #define COMP_REF_TYPE_CONTEXTS 5
 #define UNI_COMP_REF_CONTEXTS 3
-#endif  // CONFIG_EXT_COMP_REFS
 
-#if CONFIG_COMPOUND_SINGLEREF
-#define COMP_INTER_MODE_CONTEXTS 4
-#endif  // CONFIG_COMPOUND_SINGLEREF
-
-#if CONFIG_VAR_TX
-#define TXFM_PARTITION_CONTEXTS ((TX_SIZES - TX_8X8) * 6 - 2)
+#define TXFM_PARTITION_CONTEXTS ((TX_SIZES - TX_8X8) * 6 - 3)
 typedef uint8_t TXFM_CONTEXT;
-#endif
 
 #define NONE_FRAME -1
 #define INTRA_FRAME 0
 #define LAST_FRAME 1
-
-#if CONFIG_EXT_REFS
 #define LAST2_FRAME 2
 #define LAST3_FRAME 3
 #define GOLDEN_FRAME 4
@@ -691,94 +558,55 @@ typedef uint8_t TXFM_CONTEXT;
 #define ALTREF2_FRAME 6
 #define ALTREF_FRAME 7
 #define LAST_REF_FRAMES (LAST3_FRAME - LAST_FRAME + 1)
-#else  // !CONFIG_EXT_REFS
-#define GOLDEN_FRAME 2
-#define ALTREF_FRAME 3
-#endif  // CONFIG_EXT_REFS
 
 #define INTER_REFS_PER_FRAME (ALTREF_FRAME - LAST_FRAME + 1)
-#define TOTAL_REFS_PER_FRAME (ALTREF_FRAME - INTRA_FRAME + 1)
 
 #define FWD_REFS (GOLDEN_FRAME - LAST_FRAME + 1)
 #define FWD_RF_OFFSET(ref) (ref - LAST_FRAME)
-#if CONFIG_EXT_REFS
 #define BWD_REFS (ALTREF_FRAME - BWDREF_FRAME + 1)
 #define BWD_RF_OFFSET(ref) (ref - BWDREF_FRAME)
-#else
-#define BWD_REFS 1
-#define BWD_RF_OFFSET(ref) (ref - ALTREF_FRAME)
-#endif  // CONFIG_EXT_REFS
 
 #define SINGLE_REFS (FWD_REFS + BWD_REFS)
-#if CONFIG_EXT_COMP_REFS
-typedef enum {
-  LAST_LAST2_FRAMES,     // { LAST_FRAME, LAST2_FRAME }
-  LAST_LAST3_FRAMES,     // { LAST_FRAME, LAST3_FRAME }
-  LAST_GOLDEN_FRAMES,    // { LAST_FRAME, GOLDEN_FRAME }
-  BWDREF_ALTREF_FRAMES,  // { BWDREF_FRAME, ALTREF_FRAME }
-  UNIDIR_COMP_REFS
+
+typedef enum ATTRIBUTE_PACKED {
+  LAST_LAST2_FRAMES,      // { LAST_FRAME, LAST2_FRAME }
+  LAST_LAST3_FRAMES,      // { LAST_FRAME, LAST3_FRAME }
+  LAST_GOLDEN_FRAMES,     // { LAST_FRAME, GOLDEN_FRAME }
+  BWDREF_ALTREF_FRAMES,   // { BWDREF_FRAME, ALTREF_FRAME }
+  LAST2_LAST3_FRAMES,     // { LAST2_FRAME, LAST3_FRAME }
+  LAST2_GOLDEN_FRAMES,    // { LAST2_FRAME, GOLDEN_FRAME }
+  LAST3_GOLDEN_FRAMES,    // { LAST3_FRAME, GOLDEN_FRAME }
+  BWDREF_ALTREF2_FRAMES,  // { BWDREF_FRAME, ALTREF2_FRAME }
+  ALTREF2_ALTREF_FRAMES,  // { ALTREF2_FRAME, ALTREF_FRAME }
+  TOTAL_UNIDIR_COMP_REFS,
+  // 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;
+
+#define TOTAL_COMP_REFS (FWD_REFS * BWD_REFS + TOTAL_UNIDIR_COMP_REFS)
+
 #define COMP_REFS (FWD_REFS * BWD_REFS + UNIDIR_COMP_REFS)
-#else  // !CONFIG_EXT_COMP_REFS
-#define COMP_REFS (FWD_REFS * BWD_REFS)
-#endif  // CONFIG_EXT_COMP_REFS
 
-#define MODE_CTX_REF_FRAMES (TOTAL_REFS_PER_FRAME + COMP_REFS)
+// NOTE: A limited number of unidirectional reference pairs can be signalled for
+//       compound prediction. The use of skip mode, on the other hand, makes it
+//       possible to have a reference pair not listed for explicit signaling.
+#define MODE_CTX_REF_FRAMES (REF_FRAMES + TOTAL_COMP_REFS)
 
-#if CONFIG_SUPERTX
-#define PARTITION_SUPERTX_CONTEXTS 2
-#define MAX_SUPERTX_BLOCK_SIZE BLOCK_32X32
-#endif  // CONFIG_SUPERTX
-
-#if CONFIG_LOOP_RESTORATION
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   RESTORE_NONE,
   RESTORE_WIENER,
   RESTORE_SGRPROJ,
   RESTORE_SWITCHABLE,
   RESTORE_SWITCHABLE_TYPES = RESTORE_SWITCHABLE,
-  RESTORE_TYPES,
+  RESTORE_TYPES = 4,
 } RestorationType;
-#endif  // CONFIG_LOOP_RESTORATION
 
-#if CONFIG_FRAME_SUPERRES
 #define SUPERRES_SCALE_BITS 3
-#define SUPERRES_SCALE_DENOMINATOR_MIN 8
-#endif  // CONFIG_FRAME_SUPERRES
+#define SUPERRES_SCALE_DENOMINATOR_MIN (SCALE_NUMERATOR + 1)
 
-#if CONFIG_LPF_DIRECT
-typedef enum {
-  VERT_HORZ,
-  DEGREE_30,
-  DEGREE_45,
-  DEGREE_60,
-  DEGREE_120,
-  DEGREE_135,
-  DEGREE_150,
-  FILTER_DEGREES,
-} FILTER_DEGREE;
-#endif  // CONFIG_LPF_DIRECT
-
-#if CONFIG_OBU
-// R19
-typedef enum {
-  OBU_SEQUENCE_HEADER = 1,
-  OBU_TD = 2,
-  OBU_FRAME_HEADER = 3,
-  OBU_TILE_GROUP = 4,
-  OBU_METADATA = 5,
-  OBU_PADDING = 15,
-} OBU_TYPE;
-#endif
-
-#if CONFIG_LGT_FROM_PRED
-#define LGT_SIZES 2
-// Note: at least one of LGT_FROM_PRED_INTRA and LGT_FROM_PRED_INTER must be 1
-#define LGT_FROM_PRED_INTRA 1
-#define LGT_FROM_PRED_INTER 1
-// LGT_SL_INTRA: LGTs with a mode-dependent first self-loop and a break point
-#define LGT_SL_INTRA 0
-#endif  // CONFIG_LGT_FROM_PRED
+// In large_scale_tile coding, external references are used.
+#define MAX_EXTERNAL_REFERENCES 128
 
 #ifdef __cplusplus
 }  // extern "C"

third_party/aom/av1/common/filter.c

@@ -25,153 +25,6 @@ DECLARE_ALIGNED(256, static const InterpKernel,
   { 0, 0, 0, 16, 112, 0, 0, 0 }, { 0, 0, 0, 8, 120, 0, 0, 0 }
 };
 
-#if USE_TEMPORALFILTER_12TAP
-DECLARE_ALIGNED(16, static const int16_t,
-                sub_pel_filters_temporalfilter_12[SUBPEL_SHIFTS][12]) = {
-  // intfilt 0.8
-  { 0, 0, 0, 0, 0, 128, 0, 0, 0, 0, 0, 0 },
-  { 0, 1, -1, 3, -7, 127, 8, -4, 2, -1, 0, 0 },
-  { 0, 1, -3, 5, -12, 124, 18, -8, 4, -2, 1, 0 },
-  { -1, 2, -4, 8, -17, 120, 28, -11, 6, -3, 1, -1 },
-  { -1, 2, -4, 10, -21, 114, 38, -15, 8, -4, 2, -1 },
-  { -1, 3, -5, 11, -23, 107, 49, -18, 9, -5, 2, -1 },
-  { -1, 3, -6, 12, -25, 99, 60, -21, 11, -6, 3, -1 },
-  { -1, 3, -6, 12, -25, 90, 70, -23, 12, -6, 3, -1 },
-  { -1, 3, -6, 12, -24, 80, 80, -24, 12, -6, 3, -1 },
-  { -1, 3, -6, 12, -23, 70, 90, -25, 12, -6, 3, -1 },
-  { -1, 3, -6, 11, -21, 60, 99, -25, 12, -6, 3, -1 },
-  { -1, 2, -5, 9, -18, 49, 107, -23, 11, -5, 3, -1 },
-  { -1, 2, -4, 8, -15, 38, 114, -21, 10, -4, 2, -1 },
-  { -1, 1, -3, 6, -11, 28, 120, -17, 8, -4, 2, -1 },
-  { 0, 1, -2, 4, -8, 18, 124, -12, 5, -3, 1, 0 },
-  { 0, 0, -1, 2, -4, 8, 127, -7, 3, -1, 1, 0 },
-};
-#endif  // USE_TEMPORALFILTER_12TAP
-
-#if USE_EXTRA_FILTER
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_8[SUBPEL_SHIFTS]) = {
-  { 0, 0, 0, 128, 0, 0, 0, 0 },      { 0, 2, -6, 126, 8, -2, 0, 0 },
-  { 0, 2, -10, 122, 18, -4, 0, 0 },  { 0, 2, -12, 116, 28, -8, 2, 0 },
-  { 0, 2, -14, 110, 38, -10, 2, 0 }, { 0, 2, -14, 102, 48, -12, 2, 0 },
-  { 0, 2, -16, 94, 58, -12, 2, 0 },  { 0, 2, -14, 84, 66, -12, 2, 0 },
-  { 0, 2, -14, 76, 76, -14, 2, 0 },  { 0, 2, -12, 66, 84, -14, 2, 0 },
-  { 0, 2, -12, 58, 94, -16, 2, 0 },  { 0, 2, -12, 48, 102, -14, 2, 0 },
-  { 0, 2, -10, 38, 110, -14, 2, 0 }, { 0, 2, -8, 28, 116, -12, 2, 0 },
-  { 0, 0, -4, 18, 122, -10, 2, 0 },  { 0, 0, -2, 8, 126, -6, 2, 0 }
-};
-
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_regular_uv[SUBPEL_SHIFTS]) = {
-  { 0, 0, 0, 128, 0, 0, 0, 0 },      { 0, 2, -6, 126, 8, -2, 0, 0 },
-  { 0, 2, -10, 122, 18, -4, 0, 0 },  { 0, 2, -12, 116, 28, -8, 2, 0 },
-  { 0, 2, -14, 110, 38, -10, 2, 0 }, { 0, 2, -14, 102, 48, -12, 2, 0 },
-  { 0, 2, -16, 94, 58, -12, 2, 0 },  { 0, 2, -14, 84, 66, -12, 2, 0 },
-  { 0, 2, -14, 76, 76, -14, 2, 0 },  { 0, 2, -12, 66, 84, -14, 2, 0 },
-  { 0, 2, -12, 58, 94, -16, 2, 0 },  { 0, 2, -12, 48, 102, -14, 2, 0 },
-  { 0, 2, -10, 38, 110, -14, 2, 0 }, { 0, 2, -8, 28, 116, -12, 2, 0 },
-  { 0, 0, -4, 18, 122, -10, 2, 0 },  { 0, 0, -2, 8, 126, -6, 2, 0 }
-};
-
-#if USE_12TAP_FILTER
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
-  // intfilt 0.8
-  { 0, 0, 0, 128, 0, 0, 0, 0 },         { -1, 2, -6, 127, 9, -4, 2, -1 },
-  { -2, 5, -12, 124, 18, -7, 4, -2 },   { -2, 7, -16, 119, 28, -11, 5, -2 },
-  { -3, 8, -19, 114, 38, -14, 7, -3 },  { -3, 9, -22, 107, 49, -17, 8, -3 },
-  { -4, 10, -23, 99, 60, -20, 10, -4 }, { -4, 11, -23, 90, 70, -22, 10, -4 },
-  { -4, 11, -23, 80, 80, -23, 11, -4 }, { -4, 10, -22, 70, 90, -23, 11, -4 },
-  { -4, 10, -20, 60, 99, -23, 10, -4 }, { -3, 8, -17, 49, 107, -22, 9, -3 },
-  { -3, 7, -14, 38, 114, -19, 8, -3 },  { -2, 5, -11, 28, 119, -16, 7, -2 },
-  { -2, 4, -7, 18, 124, -12, 5, -2 },   { -1, 2, -4, 9, 127, -6, 2, -1 },
-};
-
-DECLARE_ALIGNED(256, static const int16_t,
-                sub_pel_filters_10sharp[SUBPEL_SHIFTS][12]) = {
-  // intfilt 0.85
-  { 0, 0, 0, 0, 0, 128, 0, 0, 0, 0, 0, 0 },
-  { 0, 1, -2, 3, -7, 127, 8, -4, 2, -1, 1, 0 },
-  { 0, 1, -3, 6, -13, 124, 18, -8, 4, -2, 1, 0 },
-  { 0, 2, -4, 8, -18, 120, 28, -12, 6, -4, 2, 0 },
-  { 0, 2, -5, 10, -21, 114, 38, -15, 8, -5, 2, 0 },
-  { 0, 3, -6, 11, -24, 107, 49, -19, 10, -6, 3, 0 },
-  { 0, 3, -7, 12, -25, 99, 59, -21, 11, -6, 3, 0 },
-  { 0, 3, -7, 12, -25, 90, 70, -23, 12, -7, 3, 0 },
-  { 0, 3, -7, 12, -25, 81, 81, -25, 12, -7, 3, 0 },
-  { 0, 3, -7, 12, -23, 70, 90, -25, 12, -7, 3, 0 },
-  { 0, 3, -6, 11, -21, 59, 99, -25, 12, -7, 3, 0 },
-  { 0, 3, -6, 10, -19, 49, 107, -24, 11, -6, 3, 0 },
-  { 0, 2, -5, 8, -15, 38, 114, -21, 10, -5, 2, 0 },
-  { 0, 2, -4, 6, -12, 28, 120, -18, 8, -4, 2, 0 },
-  { 0, 1, -2, 4, -8, 18, 124, -13, 6, -3, 1, 0 },
-  { 0, 1, -1, 2, -4, 8, 127, -7, 3, -2, 1, 0 },
-};
-#else
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
-  { 0, 0, 0, 128, 0, 0, 0, 0 },         { -2, 2, -6, 126, 8, -2, 2, 0 },
-  { -2, 6, -12, 124, 16, -6, 4, -2 },   { -2, 8, -18, 120, 26, -10, 6, -2 },
-  { -4, 10, -22, 116, 38, -14, 6, -2 }, { -4, 10, -22, 108, 48, -18, 8, -2 },
-  { -4, 10, -24, 100, 60, -20, 8, -2 }, { -4, 10, -24, 90, 70, -22, 10, -2 },
-  { -4, 12, -24, 80, 80, -24, 12, -4 }, { -2, 10, -22, 70, 90, -24, 10, -4 },
-  { -2, 8, -20, 60, 100, -24, 10, -4 }, { -2, 8, -18, 48, 108, -22, 10, -4 },
-  { -2, 6, -14, 38, 116, -22, 10, -4 }, { -2, 6, -10, 26, 120, -18, 8, -2 },
-  { -2, 4, -6, 16, 124, -12, 6, -2 },   { 0, 2, -2, 8, 126, -6, 2, -2 }
-};
-#endif
-
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_8smooth2[SUBPEL_SHIFTS]) = {
-  // freqmultiplier = 0.2
-  { 0, 0, 0, 128, 0, 0, 0, 0 },   { 0, 9, 30, 44, 32, 11, 2, 0 },
-  { 0, 8, 28, 44, 34, 12, 2, 0 }, { 0, 7, 27, 44, 35, 13, 2, 0 },
-  { 0, 6, 26, 43, 37, 14, 2, 0 }, { 0, 5, 24, 43, 38, 16, 2, 0 },
-  { 0, 5, 23, 42, 38, 17, 3, 0 }, { 0, 4, 21, 41, 40, 19, 3, 0 },
-  { 0, 4, 20, 40, 40, 20, 4, 0 }, { 0, 3, 19, 40, 41, 21, 4, 0 },
-  { 0, 3, 17, 38, 42, 23, 5, 0 }, { 0, 2, 16, 38, 43, 24, 5, 0 },
-  { 0, 2, 14, 37, 43, 26, 6, 0 }, { 0, 2, 13, 35, 44, 27, 7, 0 },
-  { 0, 2, 12, 34, 44, 28, 8, 0 }, { 0, 2, 11, 32, 44, 30, 9, 0 },
-};
-
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_smooth2_uv[SUBPEL_SHIFTS]) = {
-  // freqmultiplier = 0.2
-  { 0, 0, 0, 128, 0, 0, 0, 0 },   { 0, 9, 30, 44, 32, 11, 2, 0 },
-  { 0, 8, 28, 44, 34, 12, 2, 0 }, { 0, 7, 27, 44, 35, 13, 2, 0 },
-  { 0, 6, 26, 43, 37, 14, 2, 0 }, { 0, 5, 24, 43, 38, 16, 2, 0 },
-  { 0, 5, 23, 42, 38, 17, 3, 0 }, { 0, 4, 21, 41, 40, 19, 3, 0 },
-  { 0, 4, 20, 40, 40, 20, 4, 0 }, { 0, 3, 19, 40, 41, 21, 4, 0 },
-  { 0, 3, 17, 38, 42, 23, 5, 0 }, { 0, 2, 16, 38, 43, 24, 5, 0 },
-  { 0, 2, 14, 37, 43, 26, 6, 0 }, { 0, 2, 13, 35, 44, 27, 7, 0 },
-  { 0, 2, 12, 34, 44, 28, 8, 0 }, { 0, 2, 11, 32, 44, 30, 9, 0 },
-};
-
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_8smooth[SUBPEL_SHIFTS]) = {
-  { 0, 0, 0, 128, 0, 0, 0, 0 },     { 0, 2, 28, 62, 34, 2, 0, 0 },
-  { 0, 0, 26, 62, 36, 4, 0, 0 },    { 0, 0, 22, 62, 40, 4, 0, 0 },
-  { 0, 0, 20, 60, 42, 6, 0, 0 },    { 0, 0, 18, 58, 44, 8, 0, 0 },
-  { 0, 0, 16, 56, 46, 10, 0, 0 },   { 0, -2, 16, 54, 48, 12, 0, 0 },
-  { 0, -2, 14, 52, 52, 14, -2, 0 }, { 0, 0, 12, 48, 54, 16, -2, 0 },
-  { 0, 0, 10, 46, 56, 16, 0, 0 },   { 0, 0, 8, 44, 58, 18, 0, 0 },
-  { 0, 0, 6, 42, 60, 20, 0, 0 },    { 0, 0, 4, 40, 62, 22, 0, 0 },
-  { 0, 0, 4, 36, 62, 26, 0, 0 },    { 0, 0, 2, 34, 62, 28, 2, 0 }
-};
-
-DECLARE_ALIGNED(256, static const InterpKernel,
-                sub_pel_filters_smooth_uv[SUBPEL_SHIFTS]) = {
-  { 0, 0, 0, 128, 0, 0, 0, 0 },     { 0, 2, 28, 62, 34, 2, 0, 0 },
-  { 0, 0, 26, 62, 36, 4, 0, 0 },    { 0, 0, 22, 62, 40, 4, 0, 0 },
-  { 0, 0, 20, 60, 42, 6, 0, 0 },    { 0, 0, 18, 58, 44, 8, 0, 0 },
-  { 0, 0, 16, 56, 46, 10, 0, 0 },   { 0, -2, 16, 54, 48, 12, 0, 0 },
-  { 0, -2, 14, 52, 52, 14, -2, 0 }, { 0, 0, 12, 48, 54, 16, -2, 0 },
-  { 0, 0, 10, 46, 56, 16, 0, 0 },   { 0, 0, 8, 44, 58, 18, 0, 0 },
-  { 0, 0, 6, 42, 60, 20, 0, 0 },    { 0, 0, 4, 40, 62, 22, 0, 0 },
-  { 0, 0, 4, 36, 62, 26, 0, 0 },    { 0, 0, 2, 34, 62, 28, 2, 0 }
-};
-#else   // USE_EXTRA_FILTER
-
 DECLARE_ALIGNED(256, static const InterpKernel,
                 sub_pel_filters_8[SUBPEL_SHIFTS]) = {
   { 0, 0, 0, 128, 0, 0, 0, 0 },      { 0, 2, -6, 126, 8, -2, 0, 0 },
@@ -207,49 +60,7 @@ DECLARE_ALIGNED(256, static const InterpKernel,
   { 0, 0, 6, 42, 60, 20, 0, 0 },    { 0, 0, 4, 40, 62, 22, 0, 0 },
   { 0, 0, 4, 36, 62, 26, 0, 0 },    { 0, 0, 2, 34, 62, 28, 2, 0 }
 };
-#endif  // USE_EXTRA_FILTER
 
-#if CONFIG_EXT_INTRA
-#if CONFIG_INTRA_INTERP
-const InterpKernel *av1_intra_filter_kernels[INTRA_FILTERS] = {
-  bilinear_filters,         // INTRA_FILTER_LINEAR
-  sub_pel_filters_8,        // INTRA_FILTER_8TAP
-  sub_pel_filters_8sharp,   // INTRA_FILTER_8TAP_SHARP
-  sub_pel_filters_8smooth,  // INTRA_FILTER_8TAP_SMOOTH
-};
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
-
-#if USE_EXTRA_FILTER
-static const InterpFilterParams
-    av1_interp_filter_params_list[SWITCHABLE_FILTERS + EXTRA_FILTERS] = {
-      { (const int16_t *)sub_pel_filters_8, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        EIGHTTAP_REGULAR },
-      { (const int16_t *)sub_pel_filters_8smooth, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        EIGHTTAP_SMOOTH },
-#if USE_12TAP_FILTER
-      { (const int16_t *)sub_pel_filters_10sharp, 12, SUBPEL_SHIFTS,
-        MULTITAP_SHARP },
-#else
-      { (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        EIGHTTAP_SHARP },
-#endif
-      { (const int16_t *)sub_pel_filters_8smooth2, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        EIGHTTAP_SMOOTH2 },
-      { (const int16_t *)bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        BILINEAR },
-      { (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        EIGHTTAP_SHARP },
-      { (const int16_t *)sub_pel_filters_regular_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        FILTER_REGULAR_UV },
-      { (const int16_t *)sub_pel_filters_smooth_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        FILTER_SMOOTH_UV },
-      { (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        FILTER_SHARP_UV },
-      { (const int16_t *)sub_pel_filters_smooth2_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
-        FILTER_SMOOTH2_UV },
-    };
-#else
 static const InterpFilterParams
     av1_interp_filter_params_list[SWITCHABLE_FILTERS + 1] = {
       { (const int16_t *)sub_pel_filters_8, SUBPEL_TAPS, SUBPEL_SHIFTS,
@@ -261,62 +72,49 @@ static const InterpFilterParams
       { (const int16_t *)bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
         BILINEAR }
     };
-#endif  // USE_EXTRA_FILTER
 
-#if USE_TEMPORALFILTER_12TAP
-static const InterpFilterParams av1_interp_temporalfilter_12tap = {
-  (const int16_t *)sub_pel_filters_temporalfilter_12, 12, SUBPEL_SHIFTS,
-  TEMPORALFILTER_12TAP
+DECLARE_ALIGNED(256, static const InterpKernel,
+                sub_pel_filters_4[SUBPEL_SHIFTS]) = {
+  { 0, 0, 0, 128, 0, 0, 0, 0 },     { 0, 0, -4, 126, 8, -2, 0, 0 },
+  { 0, 0, -8, 122, 18, -4, 0, 0 },  { 0, 0, -10, 116, 28, -6, 0, 0 },
+  { 0, 0, -12, 110, 38, -8, 0, 0 }, { 0, 0, -12, 102, 48, -10, 0, 0 },
+  { 0, 0, -14, 94, 58, -10, 0, 0 }, { 0, 0, -12, 84, 66, -10, 0, 0 },
+  { 0, 0, -12, 76, 76, -12, 0, 0 }, { 0, 0, -10, 66, 84, -12, 0, 0 },
+  { 0, 0, -10, 58, 94, -14, 0, 0 }, { 0, 0, -10, 48, 102, -12, 0, 0 },
+  { 0, 0, -8, 38, 110, -12, 0, 0 }, { 0, 0, -6, 28, 116, -10, 0, 0 },
+  { 0, 0, -4, 18, 122, -8, 0, 0 },  { 0, 0, -2, 8, 126, -4, 0, 0 }
+};
+DECLARE_ALIGNED(256, static const InterpKernel,
+                sub_pel_filters_4smooth[SUBPEL_SHIFTS]) = {
+  { 0, 0, 0, 128, 0, 0, 0, 0 },   { 0, 0, 30, 62, 34, 2, 0, 0 },
+  { 0, 0, 26, 62, 36, 4, 0, 0 },  { 0, 0, 22, 62, 40, 4, 0, 0 },
+  { 0, 0, 20, 60, 42, 6, 0, 0 },  { 0, 0, 18, 58, 44, 8, 0, 0 },
+  { 0, 0, 16, 56, 46, 10, 0, 0 }, { 0, 0, 14, 54, 48, 12, 0, 0 },
+  { 0, 0, 12, 52, 52, 12, 0, 0 }, { 0, 0, 12, 48, 54, 14, 0, 0 },
+  { 0, 0, 10, 46, 56, 16, 0, 0 }, { 0, 0, 8, 44, 58, 18, 0, 0 },
+  { 0, 0, 6, 42, 60, 20, 0, 0 },  { 0, 0, 4, 40, 62, 22, 0, 0 },
+  { 0, 0, 4, 36, 62, 26, 0, 0 },  { 0, 0, 2, 34, 62, 30, 0, 0 }
 };
-#endif  // USE_TEMPORALFILTER_12TAP
 
-InterpFilterParams av1_get_interp_filter_params(
-    const InterpFilter interp_filter) {
-#if USE_TEMPORALFILTER_12TAP
-  if (interp_filter == TEMPORALFILTER_12TAP)
-    return av1_interp_temporalfilter_12tap;
-#endif  // USE_TEMPORALFILTER_12TAP
+static const InterpFilterParams av1_interp_4tap[2] = {
+  { (const int16_t *)sub_pel_filters_4, SUBPEL_TAPS, SUBPEL_SHIFTS,
+    EIGHTTAP_REGULAR },
+  { (const int16_t *)sub_pel_filters_4smooth, SUBPEL_TAPS, SUBPEL_SHIFTS,
+    EIGHTTAP_SMOOTH },
+};
+
+InterpFilterParams av1_get_interp_filter_params_with_block_size(
+    const InterpFilter interp_filter, const int w) {
+  if (w <= 4 &&
+      (interp_filter == MULTITAP_SHARP || interp_filter == EIGHTTAP_REGULAR))
+    return av1_interp_4tap[0];
+  else if (w <= 4 && interp_filter == EIGHTTAP_SMOOTH)
+    return av1_interp_4tap[1];
+
   return av1_interp_filter_params_list[interp_filter];
 }
 
 const int16_t *av1_get_interp_filter_kernel(const InterpFilter interp_filter) {
-#if USE_TEMPORALFILTER_12TAP
-  if (interp_filter == TEMPORALFILTER_12TAP)
-    return av1_interp_temporalfilter_12tap.filter_ptr;
-#endif  // USE_TEMPORALFILTER_12TAP
   return (const int16_t *)av1_interp_filter_params_list[interp_filter]
       .filter_ptr;
 }
-
-#if CONFIG_DUAL_FILTER
-InterpFilter av1_get_plane_interp_filter(InterpFilter interp_filter,
-                                         int plane) {
-#if USE_TEMPORALFILTER_12TAP
-#if USE_EXTRA_FILTER
-  assert(interp_filter <= EIGHTTAP_SHARP ||
-         interp_filter == TEMPORALFILTER_12TAP);
-#else   // USE_EXTRA_FILTER
-  assert(interp_filter <= SWITCHABLE_FILTERS ||
-         interp_filter == TEMPORALFILTER_12TAP);
-#endif  // USE_EXTRA_FILTER
-#else
-  assert(interp_filter <= EIGHTTAP_SHARP);
-#endif
-#if USE_EXTRA_FILTER
-  if (plane == 0) {
-    return interp_filter;
-  } else {
-    switch (interp_filter) {
-      case EIGHTTAP_REGULAR: return FILTER_REGULAR_UV;
-      case EIGHTTAP_SMOOTH: return FILTER_SMOOTH_UV;
-      case MULTITAP_SHARP: return FILTER_SHARP_UV;
-      case EIGHTTAP_SMOOTH2: return FILTER_SMOOTH2_UV;
-      default: return interp_filter;
-    }
-  }
-#else   // USE_EXTRA_FILTER
-  (void)plane;
-  return interp_filter;
-#endif  // USE_EXTRA_FILTER
-}
-#endif

third_party/aom/av1/common/filter.h

@@ -14,7 +14,8 @@
 
 #include <assert.h>
 
-#include "./aom_config.h"
+#include "config/aom_config.h"
+
 #include "aom/aom_integer.h"
 #include "aom_dsp/aom_filter.h"
 #include "aom_ports/mem.h"
@@ -23,34 +24,17 @@
 extern "C" {
 #endif
 
-#define USE_TEMPORALFILTER_12TAP 1
-#define MAX_FILTER_TAP 12
+#define MAX_FILTER_TAP 8
 
-#define USE_12TAP_FILTER 0
-#define USE_EXTRA_FILTER 0
-
-typedef enum {
+typedef enum ATTRIBUTE_PACKED {
   EIGHTTAP_REGULAR,
   EIGHTTAP_SMOOTH,
-#if USE_EXTRA_FILTER
-  EIGHTTAP_SMOOTH2,
-#endif  // USE_EXTRA_FILTER
   MULTITAP_SHARP,
   BILINEAR,
-#if USE_EXTRA_FILTER
-  EIGHTTAP_SHARP,
-  FILTER_REGULAR_UV,
-  FILTER_SMOOTH_UV,
-  FILTER_SHARP_UV,
-  FILTER_SMOOTH2_UV,
-#endif  // USE_EXTRA_FILTER
   INTERP_FILTERS_ALL,
   SWITCHABLE_FILTERS = BILINEAR,
   SWITCHABLE = SWITCHABLE_FILTERS + 1, /* the last switchable one */
   EXTRA_FILTERS = INTERP_FILTERS_ALL - SWITCHABLE_FILTERS,
-#if USE_TEMPORALFILTER_12TAP
-  TEMPORALFILTER_12TAP = SWITCHABLE_FILTERS + EXTRA_FILTERS,
-#endif
 } InterpFilter;
 
 // With CONFIG_DUAL_FILTER, pack two InterpFilter's into a uint32_t: since
@@ -59,73 +43,34 @@ typedef enum {
 // setting a (pair of) filters.
 //
 // Without CONFIG_DUAL_FILTER,
-#if CONFIG_DUAL_FILTER
 typedef uint32_t InterpFilters;
 static INLINE InterpFilter av1_extract_interp_filter(InterpFilters filters,
                                                      int x_filter) {
-  return (InterpFilter)((filters >> (x_filter ? 16 : 0)) & 0xffff);
+  return (InterpFilter)((filters >> (x_filter ? 16 : 0)) & 0xf);
 }
 
 static INLINE InterpFilters av1_make_interp_filters(InterpFilter y_filter,
                                                     InterpFilter x_filter) {
-  uint16_t y16 = y_filter & 0xffff;
-  uint16_t x16 = x_filter & 0xffff;
+  uint16_t y16 = y_filter & 0xf;
+  uint16_t x16 = x_filter & 0xf;
   return y16 | ((uint32_t)x16 << 16);
 }
 
 static INLINE InterpFilters av1_broadcast_interp_filter(InterpFilter filter) {
   return av1_make_interp_filters(filter, filter);
 }
-#else
-typedef InterpFilter InterpFilters;
-static INLINE InterpFilter av1_extract_interp_filter(InterpFilters filters,
-                                                     int x_filter) {
-#ifdef NDEBUG
-  (void)x_filter;
-#endif
-  assert(!x_filter);
-  return filters;
-}
-
-static INLINE InterpFilters av1_broadcast_interp_filter(InterpFilter filter) {
-  return filter;
-}
-#endif
 
 static INLINE InterpFilter av1_unswitchable_filter(InterpFilter filter) {
   return filter == SWITCHABLE ? EIGHTTAP_REGULAR : filter;
 }
 
-#if USE_EXTRA_FILTER
-#define LOG_SWITCHABLE_FILTERS \
-  3 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
-#else
 #define LOG_SWITCHABLE_FILTERS \
   2 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
-#endif
 
-#if CONFIG_DUAL_FILTER
 #define MAX_SUBPEL_TAPS 12
 #define SWITCHABLE_FILTER_CONTEXTS ((SWITCHABLE_FILTERS + 1) * 4)
 #define INTER_FILTER_COMP_OFFSET (SWITCHABLE_FILTERS + 1)
 #define INTER_FILTER_DIR_OFFSET ((SWITCHABLE_FILTERS + 1) * 2)
-#else  // CONFIG_DUAL_FILTER
-#define SWITCHABLE_FILTER_CONTEXTS (SWITCHABLE_FILTERS + 1)
-#endif  // CONFIG_DUAL_FILTER
-
-#if CONFIG_EXT_INTRA
-#if CONFIG_INTRA_INTERP
-typedef enum {
-  INTRA_FILTER_LINEAR,
-  INTRA_FILTER_8TAP,
-  INTRA_FILTER_8TAP_SHARP,
-  INTRA_FILTER_8TAP_SMOOTH,
-  INTRA_FILTERS,
-} INTRA_FILTER;
-
-extern const InterpKernel *av1_intra_filter_kernels[INTRA_FILTERS];
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
 
 typedef struct InterpFilterParams {
   const int16_t *filter_ptr;
@@ -134,26 +79,16 @@ typedef struct InterpFilterParams {
   InterpFilter interp_filter;
 } InterpFilterParams;
 
-InterpFilterParams av1_get_interp_filter_params(
-    const InterpFilter interp_filter);
-
 const int16_t *av1_get_interp_filter_kernel(const InterpFilter interp_filter);
 
+InterpFilterParams av1_get_interp_filter_params_with_block_size(
+    const InterpFilter interp_filter, const int w);
+
 static INLINE const int16_t *av1_get_interp_filter_subpel_kernel(
     const InterpFilterParams filter_params, const int subpel) {
   return filter_params.filter_ptr + filter_params.taps * subpel;
 }
 
-static INLINE int av1_is_interpolating_filter(
-    const InterpFilter interp_filter) {
-  const InterpFilterParams ip = av1_get_interp_filter_params(interp_filter);
-  return (ip.filter_ptr[ip.taps / 2 - 1] == 128);
-}
-
-#if CONFIG_DUAL_FILTER
-InterpFilter av1_get_plane_interp_filter(InterpFilter interp_filter, int plane);
-#endif
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/frame_buffers.c

@@ -75,5 +75,6 @@ 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;
 }

third_party/aom/av1/common/generic_code.c

@@ -1,112 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include "generic_code.h"
-
-void aom_cdf_init_q15_1D(uint16_t *cdf, int nsyms, int cdf_size) {
-  int i;
-  for (i = 0; i < nsyms; i++)
-    cdf[i] = AOM_ICDF((i + 1)*CDF_PROB_TOP/nsyms);
-
-  cdf[cdf_size - 1] = 0;
-}
-
-/** Adapts a Q15 cdf after encoding/decoding a symbol. */
-void aom_cdf_adapt_q15(int val, uint16_t *cdf, int n, int *count, int rate) {
-  int i;
-  *count = OD_MINI(*count + 1, 1 << rate);
-  OD_ASSERT(AOM_ICDF(cdf[n - 1]) == 32768);
-  if (*count >= 1 << rate) {
-    /* Steady-state adaptation based on a simple IIR with dyadic rate. */
-    for (i = 0; i < n; i++) {
-      int tmp;
-      /* When (i < val), we want the adjustment ((cdf[i] - tmp) >> rate) to be
-         positive so long as (cdf[i] > i + 1), and 0 when (cdf[i] == i + 1),
-         to ensure we don't drive any probabilities to 0. Replacing cdf[i] with
-         (i + 2) and solving ((i + 2 - tmp) >> rate == 1) for tmp produces
-         tmp == i + 2 - (1 << rate). Using this value of tmp with
-         cdf[i] == i + 1 instead gives an adjustment of 0 as desired.
-
-         When (i >= val), we want ((cdf[i] - tmp) >> rate) to be negative so
-         long as cdf[i] < 32768 - (n - 1 - i), and 0 when
-         cdf[i] == 32768 - (n - 1 - i), again to ensure we don't drive any
-         probabilities to 0. Since right-shifting any negative value is still
-         negative, we can solve (32768 - (n - 1 - i) - tmp == 0) for tmp,
-         producing tmp = 32769 - n + i. Using this value of tmp with smaller
-         values of cdf[i] instead gives negative adjustments, as desired.
-
-         Combining the two cases gives the expression below. These could be
-         stored in a lookup table indexed by n and rate to avoid the
-         arithmetic. */
-      tmp = 2 - (1<<rate) + i + (32767 + (1<<rate) - n)*(i >= val);
-      cdf[i] = AOM_ICDF(AOM_ICDF(cdf[i]) - ((AOM_ICDF(cdf[i]) - tmp) >> rate));
-    }
-  }
-  else {
-    int alpha;
-    /* Initial adaptation for the first symbols. The adaptation rate is
-       computed to be equivalent to what od_{en,de}code_cdf_adapt() does
-       when the initial cdf is set to increment/4. */
-    alpha = 4*32768/(n + 4**count);
-    for (i = 0; i < n; i++) {
-      int tmp;
-      tmp = (32768 - n)*(i >= val) + i + 1;
-      cdf[i] = AOM_ICDF(AOM_ICDF(cdf[i])
-          - (((AOM_ICDF(cdf[i]) - tmp)*alpha) >> 15));
-    }
-  }
-  OD_ASSERT(AOM_ICDF(cdf[n - 1]) == 32768);
-}
-
-/** Takes the base-2 log of E(x) in Q1.
- *
- * @param [in] ExQ16 expectation of x in Q16
- *
- * @retval 2*log2(ExQ16/2^16)
- */
-int log_ex(int ex_q16) {
-  int lg;
-  int lg_q1;
-  int odd;
-  lg = OD_ILOG(ex_q16);
-  if (lg < 15) {
-    odd = ex_q16*ex_q16 > 2 << 2*lg;
-  }
-  else {
-    int tmp;
-    tmp = ex_q16 >> (lg - 8);
-    odd = tmp*tmp > (1 << 15);
-  }
-  lg_q1 = OD_MAXI(0, 2*lg - 33 + odd);
-  return lg_q1;
-}
-
-/** Updates the probability model based on the encoded/decoded value
- *
- * @param [in,out] model generic prob model
- * @param [in,out] ExQ16 expectation of x
- * @param [in]     x     variable encoded/decoded (used for ExQ16)
- * @param [in]     xs    variable x after shift (used for the model)
- * @param [in]     id    id of the icdf to adapt
- * @param [in]     integration integration period of ExQ16 (leaky average over
- * 1<<integration samples)
- */
-void generic_model_update(int *ex_q16, int x, int integration) {
-  /* We could have saturated ExQ16 directly, but this is safe and simpler */
-  x = OD_MINI(x, 32767);
-  OD_IIR_DIADIC(*ex_q16, x << 16, integration);
-}

third_party/aom/av1/common/generic_code.h

@@ -1,81 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#if !defined(_generic_code_H)
-# define _generic_code_H
-
-# include "aom_dsp/bitreader.h"
-# include "aom_dsp/bitwriter.h"
-
-# define GENERIC_TABLES 12
-
-#define generic_decode(r, model, ex_q16, integration, ACCT_STR_NAME) \
-  generic_decode_(r, model, ex_q16, integration ACCT_STR_ARG(ACCT_STR_NAME))
-#define aom_decode_cdf_adapt_q15(r, cdf, n, count, rate, ACCT_STR_NAME) \
-  aom_decode_cdf_adapt_q15_(r, cdf, n, count, rate ACCT_STR_ARG(ACCT_STR_NAME))
-#define aom_decode_cdf_adapt(r, cdf, n, increment, ACCT_STR_NAME) \
-  aom_decode_cdf_adapt_(r, cdf, n, increment ACCT_STR_ARG(ACCT_STR_NAME))
-
-typedef struct {
-  /** cdf for multiple expectations of x */
-  uint16_t cdf[GENERIC_TABLES][CDF_SIZE(16)];
-} generic_encoder;
-
-#define OD_IIR_DIADIC(y, x, shift) ((y) += ((x) - (y)) >> (shift))
-
-void generic_model_init(generic_encoder *model);
-
-/* Initialize a CDF for use by aom_write_symbol_pvq()/aom_read_symbol_pvq().
-   This is used for CDFs whose size might not match the declared array size.
-   The only real requirement is that the first value of every CDF be zero.
-   Then aom_cdf_init_q15_1D() will be called with the real size the first time
-   the CDF is used. */
-#define OD_CDFS_INIT_DYNAMIC(cdf) (memset(cdf, 0, sizeof(cdf)))
-
-// WARNING: DO NOT USE this init function,
-// if the size of cdf is different from what is declared by code.
-#define OD_CDFS_INIT_Q15(cdfs) \
-  { int n_cdfs = sizeof(cdfs)/sizeof(cdfs[0]); \
-    int cdf_size = sizeof(cdfs[0])/sizeof(cdfs[0][0]); \
-    int nsyms = cdf_size - 1; \
-    int i_; \
-    for (i_ = 0; i_ < n_cdfs; i_++) \
-      aom_cdf_init_q15_1D(cdfs[i_], nsyms, cdf_size); \
-  }
-
-void aom_cdf_init(uint16_t *cdf, int ncdfs, int nsyms, int val, int first);
-
-void aom_cdf_init_q15_1D(uint16_t *cdf, int nsyms, int cdf_size);
-
-void aom_cdf_adapt_q15(int val, uint16_t *cdf, int n, int *count, int rate);
-
-void aom_encode_cdf_adapt_q15(aom_writer *w, int val, uint16_t *cdf, int n,
- int *count, int rate);
-
-void generic_encode(aom_writer *w, generic_encoder *model, int x,
- int *ex_q16, int integration);
-double generic_encode_cost(generic_encoder *model, int x, int *ex_q16);
-
-double od_encode_cdf_cost(int val, uint16_t *cdf, int n);
-
-int aom_decode_cdf_adapt_q15_(aom_reader *r, uint16_t *cdf, int n,
- int *count, int rate ACCT_STR_PARAM);
-
-int generic_decode_(aom_reader *r, generic_encoder *model,
- int *ex_q16, int integration ACCT_STR_PARAM);
-
-int log_ex(int ex_q16);
-
-void generic_model_update(int *ex_q16, int x, int integration);
-
-#endif

third_party/aom/av1/common/idct.h

@@ -12,15 +12,12 @@
 #ifndef AV1_COMMON_IDCT_H_
 #define AV1_COMMON_IDCT_H_
 
-#include <assert.h>
+#include "config/aom_config.h"
 
-#include "./aom_config.h"
 #include "av1/common/blockd.h"
 #include "av1/common/common.h"
 #include "av1/common/enums.h"
-#include "aom_dsp/inv_txfm.h"
 #include "aom_dsp/txfm_common.h"
-#include "aom_ports/mem.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -32,64 +29,16 @@ typedef struct {
   transform_1d cols, rows;  // vertical and horizontal
 } transform_2d;
 
-#if CONFIG_LGT
-int get_lgt4(const TxfmParam *txfm_param, int is_col,
-             const tran_high_t **lgtmtx);
-int get_lgt8(const TxfmParam *txfm_param, int is_col,
-             const tran_high_t **lgtmtx);
-#endif  // CONFIG_LGT
-
-#if CONFIG_LGT_FROM_PRED
-void get_lgt4_from_pred(const TxfmParam *txfm_param, int is_col,
-                        const tran_high_t **lgtmtx, int ntx);
-void get_lgt8_from_pred(const TxfmParam *txfm_param, int is_col,
-                        const tran_high_t **lgtmtx, int ntx);
-void get_lgt16up_from_pred(const TxfmParam *txfm_param, int is_col,
-                           const tran_high_t **lgtmtx, int ntx);
-#endif  // CONFIG_LGT_FROM_PRED
-
-#if CONFIG_HIGHBITDEPTH
-typedef void (*highbd_transform_1d)(const tran_low_t *, tran_low_t *, int bd);
-
-typedef struct {
-  highbd_transform_1d cols, rows;  // vertical and horizontal
-} highbd_transform_2d;
-#endif  // CONFIG_HIGHBITDEPTH
-
 #define MAX_TX_SCALE 1
 int av1_get_tx_scale(const TX_SIZE tx_size);
 
-void av1_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
-                     const TxfmParam *txfm_param);
-void av1_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
-                     const TxfmParam *txfm_param);
-
-void av1_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
-                      TxfmParam *txfm_param);
 void av1_inverse_transform_block(const MACROBLOCKD *xd,
-                                 const tran_low_t *dqcoeff,
-#if CONFIG_LGT_FROM_PRED
-                                 PREDICTION_MODE mode,
-#endif
-#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
-                                 uint8_t *mrc_mask,
-#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
+                                 const tran_low_t *dqcoeff, int plane,
                                  TX_TYPE tx_type, TX_SIZE tx_size, uint8_t *dst,
-                                 int stride, int eob);
-void av1_inverse_transform_block_facade(MACROBLOCKD *xd, int plane, int block,
-                                        int blk_row, int blk_col, int eob);
+                                 int stride, int eob, int reduced_tx_set);
 
-void av1_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
-                            int eob, int bd);
 void av1_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
                                  int stride, const TxfmParam *param);
-void av1_highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest,
-                                 int stride, const TxfmParam *param);
-void av1_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
-                                 int stride, const TxfmParam *param);
-void av1_highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
-                             TxfmParam *txfm_param);
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/laplace_tables.c

@@ -1,657 +0,0 @@
-/* This file is auto-generated using "gen_laplace_tables 128 7" */
-
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include "aom_dsp/prob.h"
-#include "pvq.h"
-
-const uint16_t EXP_CDF_TABLE[128][16] = {
-  {AOM_ICDF(32753), AOM_ICDF(32754), AOM_ICDF(32755), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(32499), AOM_ICDF(32753), AOM_ICDF(32755), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(32243), AOM_ICDF(32747), AOM_ICDF(32755), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(31987), AOM_ICDF(32737), AOM_ICDF(32755), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(31732), AOM_ICDF(32724), AOM_ICDF(32755), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(31476), AOM_ICDF(32706), AOM_ICDF(32754), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(31220), AOM_ICDF(32684), AOM_ICDF(32753), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(30964), AOM_ICDF(32658), AOM_ICDF(32751), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(30708), AOM_ICDF(32628), AOM_ICDF(32748), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(30452), AOM_ICDF(32594), AOM_ICDF(32745), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(30198), AOM_ICDF(32558), AOM_ICDF(32742), AOM_ICDF(32756),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(29941), AOM_ICDF(32515), AOM_ICDF(32736), AOM_ICDF(32755),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(29686), AOM_ICDF(32470), AOM_ICDF(32731), AOM_ICDF(32755),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(29429), AOM_ICDF(32419), AOM_ICDF(32723), AOM_ICDF(32754),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(29174), AOM_ICDF(32366), AOM_ICDF(32715), AOM_ICDF(32753),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(28918), AOM_ICDF(32308), AOM_ICDF(32705), AOM_ICDF(32752),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(28662), AOM_ICDF(32246), AOM_ICDF(32694), AOM_ICDF(32750),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(28406), AOM_ICDF(32180), AOM_ICDF(32681), AOM_ICDF(32748),
-    AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(28150), AOM_ICDF(32110), AOM_ICDF(32667), AOM_ICDF(32745),
-    AOM_ICDF(32756), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(27894), AOM_ICDF(32036), AOM_ICDF(32651), AOM_ICDF(32742),
-    AOM_ICDF(32756), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(27639), AOM_ICDF(31959), AOM_ICDF(32634), AOM_ICDF(32739),
-    AOM_ICDF(32755), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(27383), AOM_ICDF(31877), AOM_ICDF(32614), AOM_ICDF(32735),
-    AOM_ICDF(32755), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(27126), AOM_ICDF(31790), AOM_ICDF(32592), AOM_ICDF(32730),
-    AOM_ICDF(32754), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(26871), AOM_ICDF(31701), AOM_ICDF(32569), AOM_ICDF(32725),
-    AOM_ICDF(32753), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(26615), AOM_ICDF(31607), AOM_ICDF(32543), AOM_ICDF(32719),
-    AOM_ICDF(32752), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(26361), AOM_ICDF(31511), AOM_ICDF(32517), AOM_ICDF(32713),
-    AOM_ICDF(32751), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(26104), AOM_ICDF(31408), AOM_ICDF(32485), AOM_ICDF(32704),
-    AOM_ICDF(32748), AOM_ICDF(32757), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(25848), AOM_ICDF(31302), AOM_ICDF(32452), AOM_ICDF(32695),
-    AOM_ICDF(32746), AOM_ICDF(32757), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(25591), AOM_ICDF(31191), AOM_ICDF(32416), AOM_ICDF(32684),
-    AOM_ICDF(32743), AOM_ICDF(32756), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(25336), AOM_ICDF(31078), AOM_ICDF(32379), AOM_ICDF(32674),
-    AOM_ICDF(32741), AOM_ICDF(32756), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(25080), AOM_ICDF(30960), AOM_ICDF(32338), AOM_ICDF(32661),
-    AOM_ICDF(32737), AOM_ICDF(32755), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(24824), AOM_ICDF(30838), AOM_ICDF(32295), AOM_ICDF(32648),
-    AOM_ICDF(32733), AOM_ICDF(32754), AOM_ICDF(32759), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(24568), AOM_ICDF(30712), AOM_ICDF(32248), AOM_ICDF(32632),
-    AOM_ICDF(32728), AOM_ICDF(32752), AOM_ICDF(32758), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(24313), AOM_ICDF(30583), AOM_ICDF(32199), AOM_ICDF(32616),
-    AOM_ICDF(32723), AOM_ICDF(32751), AOM_ICDF(32758), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(24057), AOM_ICDF(30449), AOM_ICDF(32147), AOM_ICDF(32598),
-    AOM_ICDF(32718), AOM_ICDF(32750), AOM_ICDF(32758), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(23801), AOM_ICDF(30311), AOM_ICDF(32091), AOM_ICDF(32578),
-    AOM_ICDF(32711), AOM_ICDF(32747), AOM_ICDF(32757), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(23546), AOM_ICDF(30170), AOM_ICDF(32033), AOM_ICDF(32557),
-    AOM_ICDF(32704), AOM_ICDF(32745), AOM_ICDF(32757), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(23288), AOM_ICDF(30022), AOM_ICDF(31969), AOM_ICDF(32532),
-    AOM_ICDF(32695), AOM_ICDF(32742), AOM_ICDF(32756), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(23033), AOM_ICDF(29873), AOM_ICDF(31904), AOM_ICDF(32507),
-    AOM_ICDF(32686), AOM_ICDF(32739), AOM_ICDF(32755), AOM_ICDF(32760),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(22778), AOM_ICDF(29720), AOM_ICDF(31835), AOM_ICDF(32479),
-    AOM_ICDF(32675), AOM_ICDF(32735), AOM_ICDF(32753), AOM_ICDF(32759),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(22521), AOM_ICDF(29561), AOM_ICDF(31761), AOM_ICDF(32449),
-    AOM_ICDF(32664), AOM_ICDF(32731), AOM_ICDF(32752), AOM_ICDF(32759),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(22267), AOM_ICDF(29401), AOM_ICDF(31686), AOM_ICDF(32418),
-    AOM_ICDF(32652), AOM_ICDF(32727), AOM_ICDF(32751), AOM_ICDF(32759),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(22011), AOM_ICDF(29235), AOM_ICDF(31605), AOM_ICDF(32383),
-    AOM_ICDF(32638), AOM_ICDF(32722), AOM_ICDF(32749), AOM_ICDF(32758),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(21754), AOM_ICDF(29064), AOM_ICDF(31520), AOM_ICDF(32345),
-    AOM_ICDF(32622), AOM_ICDF(32715), AOM_ICDF(32746), AOM_ICDF(32757),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(21501), AOM_ICDF(28893), AOM_ICDF(31434), AOM_ICDF(32307),
-    AOM_ICDF(32607), AOM_ICDF(32710), AOM_ICDF(32745), AOM_ICDF(32757),
-    AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(21243), AOM_ICDF(28713), AOM_ICDF(31339), AOM_ICDF(32262),
-    AOM_ICDF(32587), AOM_ICDF(32701), AOM_ICDF(32741), AOM_ICDF(32755),
-    AOM_ICDF(32760), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(20988), AOM_ICDF(28532), AOM_ICDF(31243), AOM_ICDF(32217),
-    AOM_ICDF(32567), AOM_ICDF(32693), AOM_ICDF(32738), AOM_ICDF(32754),
-    AOM_ICDF(32760), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(20730), AOM_ICDF(28344), AOM_ICDF(31140), AOM_ICDF(32167),
-    AOM_ICDF(32544), AOM_ICDF(32682), AOM_ICDF(32733), AOM_ICDF(32752),
-    AOM_ICDF(32759), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(20476), AOM_ICDF(28156), AOM_ICDF(31036), AOM_ICDF(32116),
-    AOM_ICDF(32521), AOM_ICDF(32673), AOM_ICDF(32730), AOM_ICDF(32751),
-    AOM_ICDF(32759), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(20220), AOM_ICDF(27962), AOM_ICDF(30926), AOM_ICDF(32061),
-    AOM_ICDF(32495), AOM_ICDF(32661), AOM_ICDF(32725), AOM_ICDF(32749),
-    AOM_ICDF(32758), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(19963), AOM_ICDF(27763), AOM_ICDF(30810), AOM_ICDF(32000),
-    AOM_ICDF(32465), AOM_ICDF(32647), AOM_ICDF(32718), AOM_ICDF(32746),
-    AOM_ICDF(32757), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(19708), AOM_ICDF(27562), AOM_ICDF(30691), AOM_ICDF(31938),
-    AOM_ICDF(32435), AOM_ICDF(32633), AOM_ICDF(32712), AOM_ICDF(32743),
-    AOM_ICDF(32756), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(19454), AOM_ICDF(27358), AOM_ICDF(30569), AOM_ICDF(31873),
-    AOM_ICDF(32403), AOM_ICDF(32618), AOM_ICDF(32705), AOM_ICDF(32741),
-    AOM_ICDF(32755), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(19196), AOM_ICDF(27146), AOM_ICDF(30438), AOM_ICDF(31801),
-    AOM_ICDF(32365), AOM_ICDF(32599), AOM_ICDF(32696), AOM_ICDF(32736),
-    AOM_ICDF(32753), AOM_ICDF(32760), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(18942), AOM_ICDF(26934), AOM_ICDF(30306), AOM_ICDF(31728),
-    AOM_ICDF(32328), AOM_ICDF(32581), AOM_ICDF(32688), AOM_ICDF(32733),
-    AOM_ICDF(32752), AOM_ICDF(32760), AOM_ICDF(32763), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(18684), AOM_ICDF(26714), AOM_ICDF(30164), AOM_ICDF(31647),
-    AOM_ICDF(32284), AOM_ICDF(32558), AOM_ICDF(32676), AOM_ICDF(32727),
-    AOM_ICDF(32749), AOM_ICDF(32758), AOM_ICDF(32762), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(18429), AOM_ICDF(26493), AOM_ICDF(30021), AOM_ICDF(31565),
-    AOM_ICDF(32240), AOM_ICDF(32535), AOM_ICDF(32664), AOM_ICDF(32721),
-    AOM_ICDF(32746), AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(18174), AOM_ICDF(26268), AOM_ICDF(29872), AOM_ICDF(31477),
-    AOM_ICDF(32192), AOM_ICDF(32510), AOM_ICDF(32652), AOM_ICDF(32715),
-    AOM_ICDF(32743), AOM_ICDF(32756), AOM_ICDF(32762), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(17920), AOM_ICDF(26040), AOM_ICDF(29719), AOM_ICDF(31386),
-    AOM_ICDF(32141), AOM_ICDF(32483), AOM_ICDF(32638), AOM_ICDF(32708),
-    AOM_ICDF(32740), AOM_ICDF(32754), AOM_ICDF(32761), AOM_ICDF(32764),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(17661), AOM_ICDF(25803), AOM_ICDF(29556), AOM_ICDF(31286),
-    AOM_ICDF(32083), AOM_ICDF(32451), AOM_ICDF(32620), AOM_ICDF(32698),
-    AOM_ICDF(32734), AOM_ICDF(32751), AOM_ICDF(32759), AOM_ICDF(32763),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(17406), AOM_ICDF(25566), AOM_ICDF(29391), AOM_ICDF(31184),
-    AOM_ICDF(32024), AOM_ICDF(32418), AOM_ICDF(32603), AOM_ICDF(32690),
-    AOM_ICDF(32731), AOM_ICDF(32750), AOM_ICDF(32759), AOM_ICDF(32763),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(17151), AOM_ICDF(25325), AOM_ICDF(29220), AOM_ICDF(31076),
-    AOM_ICDF(31961), AOM_ICDF(32383), AOM_ICDF(32584), AOM_ICDF(32680),
-    AOM_ICDF(32726), AOM_ICDF(32748), AOM_ICDF(32758), AOM_ICDF(32763),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(16896), AOM_ICDF(25080), AOM_ICDF(29044), AOM_ICDF(30964),
-    AOM_ICDF(31894), AOM_ICDF(32344), AOM_ICDF(32562), AOM_ICDF(32668),
-    AOM_ICDF(32719), AOM_ICDF(32744), AOM_ICDF(32756), AOM_ICDF(32762),
-    AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(16639), AOM_ICDF(24829), AOM_ICDF(28860), AOM_ICDF(30844),
-    AOM_ICDF(31821), AOM_ICDF(32302), AOM_ICDF(32539), AOM_ICDF(32655),
-    AOM_ICDF(32712), AOM_ICDF(32740), AOM_ICDF(32754), AOM_ICDF(32761),
-    AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(28672), AOM_ICDF(30720),
-    AOM_ICDF(31744), AOM_ICDF(32256), AOM_ICDF(32512), AOM_ICDF(32640),
-    AOM_ICDF(32704), AOM_ICDF(32736), AOM_ICDF(32752), AOM_ICDF(32760),
-    AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(16130), AOM_ICDF(24320), AOM_ICDF(28479), AOM_ICDF(30591),
-    AOM_ICDF(31663), AOM_ICDF(32208), AOM_ICDF(32485), AOM_ICDF(32625),
-    AOM_ICDF(32696), AOM_ICDF(32732), AOM_ICDF(32750), AOM_ICDF(32759),
-    AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
-  {AOM_ICDF(15872), AOM_ICDF(24056), AOM_ICDF(28276), AOM_ICDF(30452),
-    AOM_ICDF(31574), AOM_ICDF(32152), AOM_ICDF(32450), AOM_ICDF(32604),
-    AOM_ICDF(32683), AOM_ICDF(32724), AOM_ICDF(32745), AOM_ICDF(32756),
-    AOM_ICDF(32762), AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32768)},
-  {AOM_ICDF(15615), AOM_ICDF(23789), AOM_ICDF(28068), AOM_ICDF(30308),
-    AOM_ICDF(31480), AOM_ICDF(32094), AOM_ICDF(32415), AOM_ICDF(32583),
-    AOM_ICDF(32671), AOM_ICDF(32717), AOM_ICDF(32741), AOM_ICDF(32754),
-    AOM_ICDF(32761), AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32768)},
-  {AOM_ICDF(15361), AOM_ICDF(23521), AOM_ICDF(27856), AOM_ICDF(30159),
-    AOM_ICDF(31382), AOM_ICDF(32032), AOM_ICDF(32377), AOM_ICDF(32560),
-    AOM_ICDF(32657), AOM_ICDF(32709), AOM_ICDF(32737), AOM_ICDF(32752),
-    AOM_ICDF(32760), AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32768)},
-  {AOM_ICDF(15103), AOM_ICDF(23245), AOM_ICDF(27634), AOM_ICDF(30000),
-    AOM_ICDF(31275), AOM_ICDF(31963), AOM_ICDF(32334), AOM_ICDF(32534),
-    AOM_ICDF(32642), AOM_ICDF(32700), AOM_ICDF(32731), AOM_ICDF(32748),
-    AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32765), AOM_ICDF(32768)},
-  {AOM_ICDF(14848), AOM_ICDF(22968), AOM_ICDF(27409), AOM_ICDF(29837),
-    AOM_ICDF(31165), AOM_ICDF(31891), AOM_ICDF(32288), AOM_ICDF(32505),
-    AOM_ICDF(32624), AOM_ICDF(32689), AOM_ICDF(32725), AOM_ICDF(32744),
-    AOM_ICDF(32755), AOM_ICDF(32761), AOM_ICDF(32764), AOM_ICDF(32768)},
-  {AOM_ICDF(14592), AOM_ICDF(22686), AOM_ICDF(27176), AOM_ICDF(29666),
-    AOM_ICDF(31047), AOM_ICDF(31813), AOM_ICDF(32238), AOM_ICDF(32474),
-    AOM_ICDF(32605), AOM_ICDF(32678), AOM_ICDF(32718), AOM_ICDF(32740),
-    AOM_ICDF(32752), AOM_ICDF(32759), AOM_ICDF(32763), AOM_ICDF(32768)},
-  {AOM_ICDF(14336), AOM_ICDF(22400), AOM_ICDF(26936), AOM_ICDF(29488),
-    AOM_ICDF(30923), AOM_ICDF(31730), AOM_ICDF(32184), AOM_ICDF(32439),
-    AOM_ICDF(32583), AOM_ICDF(32664), AOM_ICDF(32709), AOM_ICDF(32735),
-    AOM_ICDF(32749), AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32768)},
-  {AOM_ICDF(14079), AOM_ICDF(22109), AOM_ICDF(26689), AOM_ICDF(29301),
-    AOM_ICDF(30791), AOM_ICDF(31641), AOM_ICDF(32125), AOM_ICDF(32401),
-    AOM_ICDF(32559), AOM_ICDF(32649), AOM_ICDF(32700), AOM_ICDF(32729),
-    AOM_ICDF(32746), AOM_ICDF(32756), AOM_ICDF(32761), AOM_ICDF(32768)},
-  {AOM_ICDF(13825), AOM_ICDF(21817), AOM_ICDF(26437), AOM_ICDF(29108),
-    AOM_ICDF(30652), AOM_ICDF(31545), AOM_ICDF(32061), AOM_ICDF(32359),
-    AOM_ICDF(32532), AOM_ICDF(32632), AOM_ICDF(32690), AOM_ICDF(32723),
-    AOM_ICDF(32742), AOM_ICDF(32753), AOM_ICDF(32759), AOM_ICDF(32768)},
-  {AOM_ICDF(13568), AOM_ICDF(21518), AOM_ICDF(26176), AOM_ICDF(28905),
-    AOM_ICDF(30504), AOM_ICDF(31441), AOM_ICDF(31990), AOM_ICDF(32312),
-    AOM_ICDF(32501), AOM_ICDF(32611), AOM_ICDF(32676), AOM_ICDF(32714),
-    AOM_ICDF(32736), AOM_ICDF(32749), AOM_ICDF(32757), AOM_ICDF(32768)},
-  {AOM_ICDF(13314), AOM_ICDF(21218), AOM_ICDF(25911), AOM_ICDF(28697),
-    AOM_ICDF(30351), AOM_ICDF(31333), AOM_ICDF(31916), AOM_ICDF(32262),
-    AOM_ICDF(32468), AOM_ICDF(32590), AOM_ICDF(32662), AOM_ICDF(32705),
-    AOM_ICDF(32731), AOM_ICDF(32746), AOM_ICDF(32755), AOM_ICDF(32768)},
-  {AOM_ICDF(13054), AOM_ICDF(20908), AOM_ICDF(25633), AOM_ICDF(28475),
-    AOM_ICDF(30185), AOM_ICDF(31214), AOM_ICDF(31833), AOM_ICDF(32205),
-    AOM_ICDF(32429), AOM_ICDF(32564), AOM_ICDF(32645), AOM_ICDF(32694),
-    AOM_ICDF(32723), AOM_ICDF(32741), AOM_ICDF(32752), AOM_ICDF(32768)},
-  {AOM_ICDF(12803), AOM_ICDF(20603), AOM_ICDF(25356), AOM_ICDF(28252),
-    AOM_ICDF(30017), AOM_ICDF(31093), AOM_ICDF(31748), AOM_ICDF(32147),
-    AOM_ICDF(32390), AOM_ICDF(32538), AOM_ICDF(32628), AOM_ICDF(32683),
-    AOM_ICDF(32717), AOM_ICDF(32737), AOM_ICDF(32749), AOM_ICDF(32768)},
-  {AOM_ICDF(12544), AOM_ICDF(20286), AOM_ICDF(25064), AOM_ICDF(28013),
-    AOM_ICDF(29833), AOM_ICDF(30956), AOM_ICDF(31649), AOM_ICDF(32077),
-    AOM_ICDF(32341), AOM_ICDF(32504), AOM_ICDF(32605), AOM_ICDF(32667),
-    AOM_ICDF(32705), AOM_ICDF(32729), AOM_ICDF(32744), AOM_ICDF(32768)},
-  {AOM_ICDF(12288), AOM_ICDF(19968), AOM_ICDF(24768), AOM_ICDF(27768),
-    AOM_ICDF(29643), AOM_ICDF(30815), AOM_ICDF(31547), AOM_ICDF(32005),
-    AOM_ICDF(32291), AOM_ICDF(32470), AOM_ICDF(32582), AOM_ICDF(32652),
-    AOM_ICDF(32696), AOM_ICDF(32723), AOM_ICDF(32740), AOM_ICDF(32768)},
-  {AOM_ICDF(12033), AOM_ICDF(19647), AOM_ICDF(24465), AOM_ICDF(27514),
-    AOM_ICDF(29443), AOM_ICDF(30664), AOM_ICDF(31437), AOM_ICDF(31926),
-    AOM_ICDF(32235), AOM_ICDF(32431), AOM_ICDF(32555), AOM_ICDF(32633),
-    AOM_ICDF(32683), AOM_ICDF(32714), AOM_ICDF(32734), AOM_ICDF(32768)},
-  {AOM_ICDF(11777), AOM_ICDF(19321), AOM_ICDF(24154), AOM_ICDF(27250),
-    AOM_ICDF(29233), AOM_ICDF(30504), AOM_ICDF(31318), AOM_ICDF(31839),
-    AOM_ICDF(32173), AOM_ICDF(32387), AOM_ICDF(32524), AOM_ICDF(32612),
-    AOM_ICDF(32668), AOM_ICDF(32704), AOM_ICDF(32727), AOM_ICDF(32768)},
-  {AOM_ICDF(11521), AOM_ICDF(18991), AOM_ICDF(23835), AOM_ICDF(26976),
-    AOM_ICDF(29013), AOM_ICDF(30334), AOM_ICDF(31190), AOM_ICDF(31745),
-    AOM_ICDF(32105), AOM_ICDF(32338), AOM_ICDF(32489), AOM_ICDF(32587),
-    AOM_ICDF(32651), AOM_ICDF(32692), AOM_ICDF(32719), AOM_ICDF(32768)},
-  {AOM_ICDF(11265), AOM_ICDF(18657), AOM_ICDF(23508), AOM_ICDF(26691),
-    AOM_ICDF(28780), AOM_ICDF(30151), AOM_ICDF(31051), AOM_ICDF(31641),
-    AOM_ICDF(32028), AOM_ICDF(32282), AOM_ICDF(32449), AOM_ICDF(32559),
-    AOM_ICDF(32631), AOM_ICDF(32678), AOM_ICDF(32709), AOM_ICDF(32768)},
-  {AOM_ICDF(11006), AOM_ICDF(18316), AOM_ICDF(23170), AOM_ICDF(26394),
-    AOM_ICDF(28535), AOM_ICDF(29957), AOM_ICDF(30901), AOM_ICDF(31528),
-    AOM_ICDF(31944), AOM_ICDF(32220), AOM_ICDF(32404), AOM_ICDF(32526),
-    AOM_ICDF(32607), AOM_ICDF(32661), AOM_ICDF(32697), AOM_ICDF(32768)},
-  {AOM_ICDF(10752), AOM_ICDF(17976), AOM_ICDF(22830), AOM_ICDF(26091),
-    AOM_ICDF(28282), AOM_ICDF(29754), AOM_ICDF(30743), AOM_ICDF(31408),
-    AOM_ICDF(31854), AOM_ICDF(32154), AOM_ICDF(32356), AOM_ICDF(32491),
-    AOM_ICDF(32582), AOM_ICDF(32643), AOM_ICDF(32684), AOM_ICDF(32768)},
-  {AOM_ICDF(10496), AOM_ICDF(17630), AOM_ICDF(22479), AOM_ICDF(25775),
-    AOM_ICDF(28015), AOM_ICDF(29538), AOM_ICDF(30573), AOM_ICDF(31276),
-    AOM_ICDF(31754), AOM_ICDF(32079), AOM_ICDF(32300), AOM_ICDF(32450),
-    AOM_ICDF(32552), AOM_ICDF(32621), AOM_ICDF(32668), AOM_ICDF(32768)},
-  {AOM_ICDF(10240), AOM_ICDF(17280), AOM_ICDF(22120), AOM_ICDF(25448),
-    AOM_ICDF(27736), AOM_ICDF(29309), AOM_ICDF(30390), AOM_ICDF(31133),
-    AOM_ICDF(31644), AOM_ICDF(31995), AOM_ICDF(32237), AOM_ICDF(32403),
-    AOM_ICDF(32517), AOM_ICDF(32595), AOM_ICDF(32649), AOM_ICDF(32768)},
-  { AOM_ICDF(9984), AOM_ICDF(16926), AOM_ICDF(21753), AOM_ICDF(25109),
-    AOM_ICDF(27443), AOM_ICDF(29066), AOM_ICDF(30194), AOM_ICDF(30978),
-    AOM_ICDF(31523), AOM_ICDF(31902), AOM_ICDF(32166), AOM_ICDF(32349),
-    AOM_ICDF(32476), AOM_ICDF(32565), AOM_ICDF(32627), AOM_ICDF(32768)},
-  { AOM_ICDF(9728), AOM_ICDF(16568), AOM_ICDF(21377), AOM_ICDF(24759),
-    AOM_ICDF(27137), AOM_ICDF(28809), AOM_ICDF(29984), AOM_ICDF(30811),
-    AOM_ICDF(31392), AOM_ICDF(31801), AOM_ICDF(32088), AOM_ICDF(32290),
-    AOM_ICDF(32432), AOM_ICDF(32532), AOM_ICDF(32602), AOM_ICDF(32768)},
-  { AOM_ICDF(9474), AOM_ICDF(16208), AOM_ICDF(20995), AOM_ICDF(24399),
-    AOM_ICDF(26819), AOM_ICDF(28539), AOM_ICDF(29762), AOM_ICDF(30631),
-    AOM_ICDF(31249), AOM_ICDF(31688), AOM_ICDF(32000), AOM_ICDF(32222),
-    AOM_ICDF(32380), AOM_ICDF(32492), AOM_ICDF(32572), AOM_ICDF(32768)},
-  { AOM_ICDF(9216), AOM_ICDF(15840), AOM_ICDF(20601), AOM_ICDF(24023),
-    AOM_ICDF(26483), AOM_ICDF(28251), AOM_ICDF(29522), AOM_ICDF(30435),
-    AOM_ICDF(31091), AOM_ICDF(31563), AOM_ICDF(31902), AOM_ICDF(32146),
-    AOM_ICDF(32321), AOM_ICDF(32447), AOM_ICDF(32537), AOM_ICDF(32768)},
-  { AOM_ICDF(8959), AOM_ICDF(15469), AOM_ICDF(20199), AOM_ICDF(23636),
-    AOM_ICDF(26133), AOM_ICDF(27947), AOM_ICDF(29265), AOM_ICDF(30223),
-    AOM_ICDF(30919), AOM_ICDF(31425), AOM_ICDF(31792), AOM_ICDF(32059),
-    AOM_ICDF(32253), AOM_ICDF(32394), AOM_ICDF(32496), AOM_ICDF(32768)},
-  { AOM_ICDF(8705), AOM_ICDF(15097), AOM_ICDF(19791), AOM_ICDF(23238),
-    AOM_ICDF(25770), AOM_ICDF(27629), AOM_ICDF(28994), AOM_ICDF(29997),
-    AOM_ICDF(30733), AOM_ICDF(31274), AOM_ICDF(31671), AOM_ICDF(31963),
-    AOM_ICDF(32177), AOM_ICDF(32334), AOM_ICDF(32449), AOM_ICDF(32768)},
-  { AOM_ICDF(8449), AOM_ICDF(14719), AOM_ICDF(19373), AOM_ICDF(22827),
-    AOM_ICDF(25390), AOM_ICDF(27292), AOM_ICDF(28704), AOM_ICDF(29752),
-    AOM_ICDF(30530), AOM_ICDF(31107), AOM_ICDF(31535), AOM_ICDF(31853),
-    AOM_ICDF(32089), AOM_ICDF(32264), AOM_ICDF(32394), AOM_ICDF(32768)},
-  { AOM_ICDF(8192), AOM_ICDF(14336), AOM_ICDF(18944), AOM_ICDF(22400),
-    AOM_ICDF(24992), AOM_ICDF(26936), AOM_ICDF(28394), AOM_ICDF(29488),
-    AOM_ICDF(30308), AOM_ICDF(30923), AOM_ICDF(31384), AOM_ICDF(31730),
-    AOM_ICDF(31989), AOM_ICDF(32184), AOM_ICDF(32330), AOM_ICDF(32768)},
-  { AOM_ICDF(7936), AOM_ICDF(13950), AOM_ICDF(18507), AOM_ICDF(21961),
-    AOM_ICDF(24578), AOM_ICDF(26561), AOM_ICDF(28064), AOM_ICDF(29203),
-    AOM_ICDF(30066), AOM_ICDF(30720), AOM_ICDF(31216), AOM_ICDF(31592),
-    AOM_ICDF(31877), AOM_ICDF(32093), AOM_ICDF(32256), AOM_ICDF(32768)},
-  { AOM_ICDF(7678), AOM_ICDF(13558), AOM_ICDF(18060), AOM_ICDF(21507),
-    AOM_ICDF(24146), AOM_ICDF(26166), AOM_ICDF(27713), AOM_ICDF(28897),
-    AOM_ICDF(29804), AOM_ICDF(30498), AOM_ICDF(31030), AOM_ICDF(31437),
-    AOM_ICDF(31749), AOM_ICDF(31988), AOM_ICDF(32171), AOM_ICDF(32768)},
-  { AOM_ICDF(7423), AOM_ICDF(13165), AOM_ICDF(17606), AOM_ICDF(21041),
-    AOM_ICDF(23698), AOM_ICDF(25753), AOM_ICDF(27342), AOM_ICDF(28571),
-    AOM_ICDF(29522), AOM_ICDF(30257), AOM_ICDF(30826), AOM_ICDF(31266),
-    AOM_ICDF(31606), AOM_ICDF(31869), AOM_ICDF(32073), AOM_ICDF(32768)},
-  { AOM_ICDF(7168), AOM_ICDF(12768), AOM_ICDF(17143), AOM_ICDF(20561),
-    AOM_ICDF(23231), AOM_ICDF(25317), AOM_ICDF(26947), AOM_ICDF(28220),
-    AOM_ICDF(29215), AOM_ICDF(29992), AOM_ICDF(30599), AOM_ICDF(31073),
-    AOM_ICDF(31444), AOM_ICDF(31734), AOM_ICDF(31960), AOM_ICDF(32768)},
-  { AOM_ICDF(6911), AOM_ICDF(12365), AOM_ICDF(16669), AOM_ICDF(20065),
-    AOM_ICDF(22744), AOM_ICDF(24858), AOM_ICDF(26526), AOM_ICDF(27842),
-    AOM_ICDF(28881), AOM_ICDF(29701), AOM_ICDF(30348), AOM_ICDF(30858),
-    AOM_ICDF(31261), AOM_ICDF(31579), AOM_ICDF(31830), AOM_ICDF(32768)},
-  { AOM_ICDF(6657), AOM_ICDF(11961), AOM_ICDF(16188), AOM_ICDF(19556),
-    AOM_ICDF(22240), AOM_ICDF(24379), AOM_ICDF(26083), AOM_ICDF(27441),
-    AOM_ICDF(28523), AOM_ICDF(29385), AOM_ICDF(30072), AOM_ICDF(30620),
-    AOM_ICDF(31056), AOM_ICDF(31404), AOM_ICDF(31681), AOM_ICDF(32768)},
-  { AOM_ICDF(6400), AOM_ICDF(11550), AOM_ICDF(15694), AOM_ICDF(19029),
-    AOM_ICDF(21712), AOM_ICDF(23871), AOM_ICDF(25609), AOM_ICDF(27007),
-    AOM_ICDF(28132), AOM_ICDF(29037), AOM_ICDF(29766), AOM_ICDF(30352),
-    AOM_ICDF(30824), AOM_ICDF(31204), AOM_ICDF(31509), AOM_ICDF(32768)},
-  { AOM_ICDF(6142), AOM_ICDF(11134), AOM_ICDF(15190), AOM_ICDF(18486),
-    AOM_ICDF(21164), AOM_ICDF(23340), AOM_ICDF(25108), AOM_ICDF(26544),
-    AOM_ICDF(27711), AOM_ICDF(28659), AOM_ICDF(29429), AOM_ICDF(30055),
-    AOM_ICDF(30564), AOM_ICDF(30977), AOM_ICDF(31313), AOM_ICDF(32768)},
-  { AOM_ICDF(5890), AOM_ICDF(10720), AOM_ICDF(14682), AOM_ICDF(17932),
-    AOM_ICDF(20598), AOM_ICDF(22785), AOM_ICDF(24579), AOM_ICDF(26051),
-    AOM_ICDF(27258), AOM_ICDF(28248), AOM_ICDF(29060), AOM_ICDF(29726),
-    AOM_ICDF(30273), AOM_ICDF(30721), AOM_ICDF(31089), AOM_ICDF(32768)},
-  { AOM_ICDF(5631), AOM_ICDF(10295), AOM_ICDF(14157), AOM_ICDF(17356),
-    AOM_ICDF(20005), AOM_ICDF(22199), AOM_ICDF(24016), AOM_ICDF(25520),
-    AOM_ICDF(26766), AOM_ICDF(27798), AOM_ICDF(28652), AOM_ICDF(29359),
-    AOM_ICDF(29945), AOM_ICDF(30430), AOM_ICDF(30832), AOM_ICDF(32768)},
-  { AOM_ICDF(5377), AOM_ICDF(9871), AOM_ICDF(13628), AOM_ICDF(16768),
-    AOM_ICDF(19393), AOM_ICDF(21587), AOM_ICDF(23421), AOM_ICDF(24954),
-    AOM_ICDF(26236), AOM_ICDF(27308), AOM_ICDF(28204), AOM_ICDF(28953),
-    AOM_ICDF(29579), AOM_ICDF(30102), AOM_ICDF(30539), AOM_ICDF(32768)},
-  { AOM_ICDF(5121), AOM_ICDF(9441), AOM_ICDF(13086), AOM_ICDF(16161),
-    AOM_ICDF(18756), AOM_ICDF(20945), AOM_ICDF(22792), AOM_ICDF(24351),
-    AOM_ICDF(25666), AOM_ICDF(26776), AOM_ICDF(27712), AOM_ICDF(28502),
-    AOM_ICDF(29169), AOM_ICDF(29731), AOM_ICDF(30206), AOM_ICDF(32768)},
-  { AOM_ICDF(4865), AOM_ICDF(9007), AOM_ICDF(12534), AOM_ICDF(15538),
-    AOM_ICDF(18096), AOM_ICDF(20274), AOM_ICDF(22129), AOM_ICDF(23708),
-    AOM_ICDF(25053), AOM_ICDF(26198), AOM_ICDF(27173), AOM_ICDF(28004),
-    AOM_ICDF(28711), AOM_ICDF(29313), AOM_ICDF(29826), AOM_ICDF(32768)},
-  { AOM_ICDF(4608), AOM_ICDF(8568), AOM_ICDF(11971), AOM_ICDF(14896),
-    AOM_ICDF(17409), AOM_ICDF(19569), AOM_ICDF(21425), AOM_ICDF(23020),
-    AOM_ICDF(24391), AOM_ICDF(25569), AOM_ICDF(26581), AOM_ICDF(27451),
-    AOM_ICDF(28199), AOM_ICDF(28842), AOM_ICDF(29394), AOM_ICDF(32768)},
-  { AOM_ICDF(4351), AOM_ICDF(8125), AOM_ICDF(11398), AOM_ICDF(14236),
-    AOM_ICDF(16697), AOM_ICDF(18831), AOM_ICDF(20682), AOM_ICDF(22287),
-    AOM_ICDF(23679), AOM_ICDF(24886), AOM_ICDF(25933), AOM_ICDF(26841),
-    AOM_ICDF(27628), AOM_ICDF(28311), AOM_ICDF(28903), AOM_ICDF(32768)},
-  { AOM_ICDF(4096), AOM_ICDF(7680), AOM_ICDF(10816), AOM_ICDF(13560),
-    AOM_ICDF(15961), AOM_ICDF(18062), AOM_ICDF(19900), AOM_ICDF(21508),
-    AOM_ICDF(22915), AOM_ICDF(24146), AOM_ICDF(25224), AOM_ICDF(26167),
-    AOM_ICDF(26992), AOM_ICDF(27714), AOM_ICDF(28346), AOM_ICDF(32768)},
-  { AOM_ICDF(3840), AOM_ICDF(7230), AOM_ICDF(10223), AOM_ICDF(12865),
-    AOM_ICDF(15197), AOM_ICDF(17256), AOM_ICDF(19074), AOM_ICDF(20679),
-    AOM_ICDF(22096), AOM_ICDF(23347), AOM_ICDF(24451), AOM_ICDF(25426),
-    AOM_ICDF(26287), AOM_ICDF(27047), AOM_ICDF(27718), AOM_ICDF(32768)},
-  { AOM_ICDF(3584), AOM_ICDF(6776), AOM_ICDF(9619), AOM_ICDF(12151),
-    AOM_ICDF(14406), AOM_ICDF(16414), AOM_ICDF(18203), AOM_ICDF(19796),
-    AOM_ICDF(21215), AOM_ICDF(22479), AOM_ICDF(23604), AOM_ICDF(24606),
-    AOM_ICDF(25499), AOM_ICDF(26294), AOM_ICDF(27002), AOM_ICDF(32768)},
-  { AOM_ICDF(3328), AOM_ICDF(6318), AOM_ICDF(9004), AOM_ICDF(11417),
-    AOM_ICDF(13585), AOM_ICDF(15533), AOM_ICDF(17283), AOM_ICDF(18856),
-    AOM_ICDF(20269), AOM_ICDF(21538), AOM_ICDF(22678), AOM_ICDF(23703),
-    AOM_ICDF(24624), AOM_ICDF(25451), AOM_ICDF(26194), AOM_ICDF(32768)},
-  { AOM_ICDF(3072), AOM_ICDF(5856), AOM_ICDF(8379), AOM_ICDF(10665),
-    AOM_ICDF(12737), AOM_ICDF(14615), AOM_ICDF(16317), AOM_ICDF(17859),
-    AOM_ICDF(19257), AOM_ICDF(20524), AOM_ICDF(21672), AOM_ICDF(22712),
-    AOM_ICDF(23655), AOM_ICDF(24509), AOM_ICDF(25283), AOM_ICDF(32768)},
-  { AOM_ICDF(2816), AOM_ICDF(5390), AOM_ICDF(7743), AOM_ICDF(9894),
-    AOM_ICDF(11860), AOM_ICDF(13657), AOM_ICDF(15299), AOM_ICDF(16800),
-    AOM_ICDF(18172), AOM_ICDF(19426), AOM_ICDF(20573), AOM_ICDF(21621),
-    AOM_ICDF(22579), AOM_ICDF(23455), AOM_ICDF(24255), AOM_ICDF(32768)},
-  { AOM_ICDF(2560), AOM_ICDF(4920), AOM_ICDF(7096), AOM_ICDF(9102),
-    AOM_ICDF(10951), AOM_ICDF(12656), AOM_ICDF(14227), AOM_ICDF(15676),
-    AOM_ICDF(17011), AOM_ICDF(18242), AOM_ICDF(19377), AOM_ICDF(20423),
-    AOM_ICDF(21388), AOM_ICDF(22277), AOM_ICDF(23097), AOM_ICDF(32768)},
-  { AOM_ICDF(2304), AOM_ICDF(4446), AOM_ICDF(6437), AOM_ICDF(8288),
-    AOM_ICDF(10009), AOM_ICDF(11609), AOM_ICDF(13097), AOM_ICDF(14480),
-    AOM_ICDF(15766), AOM_ICDF(16961), AOM_ICDF(18072), AOM_ICDF(19105),
-    AOM_ICDF(20066), AOM_ICDF(20959), AOM_ICDF(21789), AOM_ICDF(32768)},
-  { AOM_ICDF(2048), AOM_ICDF(3968), AOM_ICDF(5768), AOM_ICDF(7456),
-    AOM_ICDF(9038), AOM_ICDF(10521), AOM_ICDF(11911), AOM_ICDF(13215),
-    AOM_ICDF(14437), AOM_ICDF(15583), AOM_ICDF(16657), AOM_ICDF(17664),
-    AOM_ICDF(18608), AOM_ICDF(19493), AOM_ICDF(20323), AOM_ICDF(32768)},
-  { AOM_ICDF(1792), AOM_ICDF(3486), AOM_ICDF(5087), AOM_ICDF(6601),
-    AOM_ICDF(8032), AOM_ICDF(9385), AOM_ICDF(10664), AOM_ICDF(11873),
-    AOM_ICDF(13016), AOM_ICDF(14096), AOM_ICDF(15117), AOM_ICDF(16082),
-    AOM_ICDF(16995), AOM_ICDF(17858), AOM_ICDF(18673), AOM_ICDF(32768)},
-  { AOM_ICDF(1536), AOM_ICDF(3000), AOM_ICDF(4395), AOM_ICDF(5725),
-    AOM_ICDF(6993), AOM_ICDF(8201), AOM_ICDF(9353), AOM_ICDF(10451),
-    AOM_ICDF(11497), AOM_ICDF(12494), AOM_ICDF(13444), AOM_ICDF(14350),
-    AOM_ICDF(15213), AOM_ICDF(16036), AOM_ICDF(16820), AOM_ICDF(32768)},
-  { AOM_ICDF(1280), AOM_ICDF(2510), AOM_ICDF(3692), AOM_ICDF(4828),
-    AOM_ICDF(5919), AOM_ICDF(6968), AOM_ICDF(7976), AOM_ICDF(8944),
-    AOM_ICDF(9875), AOM_ICDF(10769), AOM_ICDF(11628), AOM_ICDF(12454),
-    AOM_ICDF(13248), AOM_ICDF(14011), AOM_ICDF(14744), AOM_ICDF(32768)},
-  { AOM_ICDF(1024), AOM_ICDF(2016), AOM_ICDF(2977), AOM_ICDF(3908),
-    AOM_ICDF(4810), AOM_ICDF(5684), AOM_ICDF(6530), AOM_ICDF(7350),
-    AOM_ICDF(8144), AOM_ICDF(8913), AOM_ICDF(9658), AOM_ICDF(10380),
-    AOM_ICDF(11080), AOM_ICDF(11758), AOM_ICDF(12415), AOM_ICDF(32768)},
-  {  AOM_ICDF(768), AOM_ICDF(1518), AOM_ICDF(2250), AOM_ICDF(2965),
-    AOM_ICDF(3663), AOM_ICDF(4345), AOM_ICDF(5011), AOM_ICDF(5662),
-    AOM_ICDF(6297), AOM_ICDF(6917), AOM_ICDF(7523), AOM_ICDF(8115),
-    AOM_ICDF(8693), AOM_ICDF(9257), AOM_ICDF(9808), AOM_ICDF(32768)},
-  {  AOM_ICDF(512), AOM_ICDF(1016), AOM_ICDF(1512), AOM_ICDF(2000),
-    AOM_ICDF(2481), AOM_ICDF(2954), AOM_ICDF(3420), AOM_ICDF(3879),
-    AOM_ICDF(4330), AOM_ICDF(4774), AOM_ICDF(5211), AOM_ICDF(5642),
-    AOM_ICDF(6066), AOM_ICDF(6483), AOM_ICDF(6894), AOM_ICDF(32768)},
-  {  AOM_ICDF(256),  AOM_ICDF(510),  AOM_ICDF(762), AOM_ICDF(1012),
-    AOM_ICDF(1260), AOM_ICDF(1506), AOM_ICDF(1750), AOM_ICDF(1992),
-    AOM_ICDF(2232), AOM_ICDF(2471), AOM_ICDF(2708), AOM_ICDF(2943),
-    AOM_ICDF(3176), AOM_ICDF(3407), AOM_ICDF(3636), AOM_ICDF(32768)},
-};
-
-
-const uint16_t LAPLACE_OFFSET[128] = {
-  0,
-  29871,
-  28672,
-  27751,
-  26975,
-  26291,
-  25673,
-  25105,
-  24576,
-  24079,
-  23609,
-  23162,
-  22734,
-  22325,
-  21931,
-  21550,
-  21182,
-  20826,
-  20480,
-  20143,
-  19815,
-  19495,
-  19183,
-  18877,
-  18579,
-  18286,
-  17999,
-  17718,
-  17442,
-  17170,
-  16904,
-  16642,
-  16384,
-  16129,
-  15879,
-  15633,
-  15390,
-  15150,
-  14913,
-  14680,
-  14450,
-  14222,
-  13997,
-  13775,
-  13556,
-  13338,
-  13124,
-  12911,
-  12701,
-  12493,
-  12288,
-  12084,
-  11882,
-  11682,
-  11484,
-  11288,
-  11094,
-  10901,
-  10710,
-  10521,
-  10333,
-  10147,
-  9962,
-  9779,
-  9597,
-  9417,
-  9238,
-  9060,
-  8884,
-  8709,
-  8535,
-  8363,
-  8192,
-  8021,
-  7853,
-  7685,
-  7518,
-  7352,
-  7188,
-  7025,
-  6862,
-  6701,
-  6540,
-  6381,
-  6222,
-  6065,
-  5908,
-  5753,
-  5598,
-  5444,
-  5291,
-  5138,
-  4987,
-  4837,
-  4687,
-  4538,
-  4390,
-  4242,
-  4096,
-  3950,
-  3804,
-  3660,
-  3516,
-  3373,
-  3231,
-  3089,
-  2948,
-  2808,
-  2668,
-  2529,
-  2391,
-  2253,
-  2116,
-  1979,
-  1843,
-  1708,
-  1573,
-  1439,
-  1306,
-  1172,
-  1040,
-  908,
-  777,
-  646,
-  516,
-  386,
-  257,
-  128,
-};

third_party/aom/av1/common/mips/msa/av1_idct16x16_msa.c

@@ -1,81 +0,0 @@
-/*
- * 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 <assert.h>
-
-#include "av1/common/enums.h"
-#include "aom_dsp/mips/inv_txfm_msa.h"
-
-void av1_iht16x16_256_add_msa(const int16_t *input, uint8_t *dst,
-                              int32_t dst_stride, TxfmParam *txfm_param) {
-  int32_t i;
-  DECLARE_ALIGNED(32, int16_t, out[16 * 16]);
-  int16_t *out_ptr = &out[0];
-  const TX_TYPE tx_type = txfm_param->tx_type;
-
-  switch (tx_type) {
-    case DCT_DCT:
-      /* transform rows */
-      for (i = 0; i < 2; ++i) {
-        /* process 16 * 8 block */
-        aom_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
-      }
-
-      /* transform columns */
-      for (i = 0; i < 2; ++i) {
-        /* process 8 * 16 block */
-        aom_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
-                                         dst_stride);
-      }
-      break;
-    case ADST_DCT:
-      /* transform rows */
-      for (i = 0; i < 2; ++i) {
-        /* process 16 * 8 block */
-        aom_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
-      }
-
-      /* transform columns */
-      for (i = 0; i < 2; ++i) {
-        aom_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
-                                          (dst + (i << 3)), dst_stride);
-      }
-      break;
-    case DCT_ADST:
-      /* transform rows */
-      for (i = 0; i < 2; ++i) {
-        /* process 16 * 8 block */
-        aom_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
-      }
-
-      /* transform columns */
-      for (i = 0; i < 2; ++i) {
-        /* process 8 * 16 block */
-        aom_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
-                                         dst_stride);
-      }
-      break;
-    case ADST_ADST:
-      /* transform rows */
-      for (i = 0; i < 2; ++i) {
-        /* process 16 * 8 block */
-        aom_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
-      }
-
-      /* transform columns */
-      for (i = 0; i < 2; ++i) {
-        aom_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
-                                          (dst + (i << 3)), dst_stride);
-      }
-      break;
-    default: assert(0); break;
-  }
-}

third_party/aom/av1/common/mips/msa/av1_idct4x4_msa.c

@@ -1,62 +0,0 @@
-/*
- * 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 <assert.h>
-
-#include "av1/common/enums.h"
-#include "aom_dsp/mips/inv_txfm_msa.h"
-
-void av1_iht4x4_16_add_msa(const int16_t *input, uint8_t *dst,
-                           int32_t dst_stride, TxfmParam *txfm_param) {
-  v8i16 in0, in1, in2, in3;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-
-  /* load vector elements of 4x4 block */
-  LD4x4_SH(input, in0, in1, in2, in3);
-  TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
-
-  switch (tx_type) {
-    case DCT_DCT:
-      /* DCT in horizontal */
-      AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      /* DCT in vertical */
-      TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
-      AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      break;
-    case ADST_DCT:
-      /* DCT in horizontal */
-      AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      /* ADST in vertical */
-      TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
-      AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      break;
-    case DCT_ADST:
-      /* ADST in horizontal */
-      AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      /* DCT in vertical */
-      TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
-      AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      break;
-    case ADST_ADST:
-      /* ADST in horizontal */
-      AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      /* ADST in vertical */
-      TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
-      AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
-      break;
-    default: assert(0); break;
-  }
-
-  /* final rounding (add 2^3, divide by 2^4) and shift */
-  SRARI_H4_SH(in0, in1, in2, in3, 4);
-  /* add block and store 4x4 */
-  ADDBLK_ST4x4_UB(in0, in1, in2, in3, dst, dst_stride);
-}

third_party/aom/av1/common/mips/msa/av1_idct8x8_msa.c

@@ -1,80 +0,0 @@
-/*
- * 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 <assert.h>
-
-#include "av1/common/enums.h"
-#include "aom_dsp/mips/inv_txfm_msa.h"
-
-void av1_iht8x8_64_add_msa(const int16_t *input, uint8_t *dst,
-                           int32_t dst_stride, TxfmParam *txfm_param) {
-  v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
-  const TX_TYPE tx_type = txfm_param->tx_type;
-
-  /* load vector elements of 8x8 block */
-  LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
-
-  TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                     in4, in5, in6, in7);
-
-  switch (tx_type) {
-    case DCT_DCT:
-      /* DCT in horizontal */
-      AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                     in4, in5, in6, in7);
-      /* DCT in vertical */
-      TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
-                         in3, in4, in5, in6, in7);
-      AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                     in4, in5, in6, in7);
-      break;
-    case ADST_DCT:
-      /* DCT in horizontal */
-      AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                     in4, in5, in6, in7);
-      /* ADST in vertical */
-      TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
-                         in3, in4, in5, in6, in7);
-      AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
-                in5, in6, in7);
-      break;
-    case DCT_ADST:
-      /* ADST in horizontal */
-      AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
-                in5, in6, in7);
-      /* DCT in vertical */
-      TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
-                         in3, in4, in5, in6, in7);
-      AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                     in4, in5, in6, in7);
-      break;
-    case ADST_ADST:
-      /* ADST in horizontal */
-      AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
-                in5, in6, in7);
-      /* ADST in vertical */
-      TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
-                         in3, in4, in5, in6, in7);
-      AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
-                in5, in6, in7);
-      break;
-    default: assert(0); break;
-  }
-
-  /* final rounding (add 2^4, divide by 2^5) and shift */
-  SRARI_H4_SH(in0, in1, in2, in3, 5);
-  SRARI_H4_SH(in4, in5, in6, in7, 5);
-
-  /* add block and store 8x8 */
-  AOM_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
-  dst += (4 * dst_stride);
-  AOM_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
-}

third_party/aom/av1/common/mv.h

@@ -27,6 +27,8 @@ typedef struct mv {
   int16_t col;
 } MV;
 
+static const MV kZeroMv = { 0, 0 };
+
 typedef union int_mv {
   uint32_t as_int;
   MV as_mv;
@@ -37,11 +39,6 @@ typedef struct mv32 {
   int32_t col;
 } MV32;
 
-#if CONFIG_WARPED_MOTION
-#define WARPED_MOTION_SORT_SAMPLES 1
-#endif  // CONFIG_WARPED_MOTION
-
-#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
 // Bits of precision used for the model
 #define WARPEDMODEL_PREC_BITS 16
 #define WARPEDMODEL_ROW3HOMO_PREC_BITS 16
@@ -54,19 +51,8 @@ typedef struct mv32 {
 #define WARPEDPIXEL_PREC_BITS 6
 #define WARPEDPIXEL_PREC_SHIFTS (1 << WARPEDPIXEL_PREC_BITS)
 
-// Taps for ntap filter
-#define WARPEDPIXEL_FILTER_TAPS 6
-
-// Precision of filter taps
-#define WARPEDPIXEL_FILTER_BITS 7
-
 #define WARP_PARAM_REDUCE_BITS 6
 
-// Precision bits reduction after horizontal shear
-#define HORSHEAR_REDUCE_PREC_BITS 5
-#define VERSHEAR_REDUCE_PREC_BITS \
-  (2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)
-
 #define WARPEDDIFF_PREC_BITS (WARPEDMODEL_PREC_BITS - WARPEDPIXEL_PREC_BITS)
 
 /* clang-format off */
@@ -75,10 +61,7 @@ typedef enum {
   TRANSLATION = 1,   // translational motion 2-parameter
   ROTZOOM = 2,       // simplified affine with rotation + zoom only, 4-parameter
   AFFINE = 3,        // affine, 6-parameter
-  HORTRAPEZOID = 4,  // constrained homography, hor trapezoid, 6-parameter
-  VERTRAPEZOID = 5,  // constrained homography, ver trapezoid, 6-parameter
-  HOMOGRAPHY = 6,    // homography, 8-parameter
-  TRANS_TYPES = 7,
+  TRANS_TYPES,
 } TransformationType;
 /* clang-format on */
 
@@ -90,24 +73,13 @@ typedef enum {
 // GLOBAL_TRANS_TYPES 7 - up to full homography
 #define GLOBAL_TRANS_TYPES 4
 
-#if GLOBAL_TRANS_TYPES > 4
-// First bit indicates whether using identity or not
-// GLOBAL_TYPE_BITS=ceiling(log2(GLOBAL_TRANS_TYPES-1)) is the
-// number of bits needed to cover the remaining possibilities
-#define GLOBAL_TYPE_BITS (get_msb(2 * GLOBAL_TRANS_TYPES - 3))
-#endif  // GLOBAL_TRANS_TYPES > 4
-
 typedef struct {
-#if CONFIG_GLOBAL_MOTION
   int global_warp_allowed;
-#endif  // CONFIG_GLOBAL_MOTION
-#if CONFIG_WARPED_MOTION
   int local_warp_allowed;
-#endif  // CONFIG_WARPED_MOTION
 } WarpTypesAllowed;
 
 // number of parameters used by each transformation in TransformationTypes
-static const int trans_model_params[TRANS_TYPES] = { 0, 2, 4, 6, 6, 6, 8 };
+static const int trans_model_params[TRANS_TYPES] = { 0, 2, 4, 6 };
 
 // The order of values in the wmmat matrix below is best described
 // by the homography:
@@ -118,6 +90,7 @@ typedef struct {
   TransformationType wmtype;
   int32_t wmmat[8];
   int16_t alpha, beta, gamma, delta;
+  int8_t invalid;
 } WarpedMotionParams;
 
 /* clang-format off */
@@ -125,12 +98,11 @@ 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
+  0, 0, 0, 0,
+  0,
 };
 /* clang-format on */
-#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
 
-#if CONFIG_GLOBAL_MOTION
 // The following constants describe the various precisions
 // of different parameters in the global motion experiment.
 //
@@ -187,9 +159,6 @@ static const WarpedMotionParams default_warp_params = {
 #define GM_ALPHA_MIN -GM_ALPHA_MAX
 #define GM_ROW3HOMO_MIN -GM_ROW3HOMO_MAX
 
-// Use global motion parameters for sub8x8 blocks
-#define GLOBAL_SUB8X8_USED 0
-
 static INLINE int block_center_x(int mi_col, BLOCK_SIZE bs) {
   const int bw = block_size_wide[bs];
   return mi_col * MI_SIZE + bw / 2 - 1;
@@ -206,7 +175,6 @@ static INLINE int convert_to_trans_prec(int allow_hp, int coor) {
   else
     return ROUND_POWER_OF_TWO_SIGNED(coor, WARPEDMODEL_PREC_BITS - 2) * 2;
 }
-#if CONFIG_AMVR
 static INLINE void integer_mv_precision(MV *mv) {
   int mod = (mv->row % 8);
   if (mod != 0) {
@@ -232,7 +200,6 @@ static INLINE void integer_mv_precision(MV *mv) {
     }
   }
 }
-#endif
 // Convert a global motion vector into a motion vector at the centre of the
 // given block.
 //
@@ -242,14 +209,15 @@ static INLINE void integer_mv_precision(MV *mv) {
 // represents an integer)
 static INLINE int_mv gm_get_motion_vector(const WarpedMotionParams *gm,
                                           int allow_hp, BLOCK_SIZE bsize,
-                                          int mi_col, int mi_row, int block_idx
-#if CONFIG_AMVR
-                                          ,
-                                          int is_integer
-#endif
-                                          ) {
-  const int unify_bsize = CONFIG_CB4X4;
+                                          int mi_col, int mi_row,
+                                          int is_integer) {
   int_mv res;
+
+  if (gm->wmtype == IDENTITY) {
+    res.as_int = 0;
+    return res;
+  }
+
   const int32_t *mat = gm->wmmat;
   int x, y, tx, ty;
 
@@ -265,65 +233,37 @@ static INLINE int_mv gm_get_motion_vector(const WarpedMotionParams *gm,
     res.as_mv.row = gm->wmmat[0] >> GM_TRANS_ONLY_PREC_DIFF;
     res.as_mv.col = gm->wmmat[1] >> GM_TRANS_ONLY_PREC_DIFF;
     assert(IMPLIES(1 & (res.as_mv.row | res.as_mv.col), allow_hp));
-#if CONFIG_AMVR
     if (is_integer) {
       integer_mv_precision(&res.as_mv);
     }
-#endif
     return res;
   }
 
-  if (bsize >= BLOCK_8X8 || unify_bsize) {
-    x = block_center_x(mi_col, bsize);
-    y = block_center_y(mi_row, bsize);
-  } else {
-    x = block_center_x(mi_col, bsize);
-    y = block_center_y(mi_row, bsize);
-    x += (block_idx & 1) * MI_SIZE / 2;
-    y += (block_idx & 2) * MI_SIZE / 4;
-  }
+  x = block_center_x(mi_col, bsize);
+  y = block_center_y(mi_row, bsize);
 
   if (gm->wmtype == ROTZOOM) {
     assert(gm->wmmat[5] == gm->wmmat[2]);
     assert(gm->wmmat[4] == -gm->wmmat[3]);
   }
-  if (gm->wmtype > AFFINE) {
-    int xc = (int)((int64_t)mat[2] * x + (int64_t)mat[3] * y + mat[0]);
-    int yc = (int)((int64_t)mat[4] * x + (int64_t)mat[5] * y + mat[1]);
-    const int Z = (int)((int64_t)mat[6] * x + (int64_t)mat[7] * y +
-                        (1 << WARPEDMODEL_ROW3HOMO_PREC_BITS));
-    xc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
-    yc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
-    xc = (int)(xc > 0 ? ((int64_t)xc + Z / 2) / Z : ((int64_t)xc - Z / 2) / Z);
-    yc = (int)(yc > 0 ? ((int64_t)yc + Z / 2) / Z : ((int64_t)yc - Z / 2) / Z);
-    tx = convert_to_trans_prec(allow_hp, xc) - (x << 3);
-    ty = convert_to_trans_prec(allow_hp, yc) - (y << 3);
-  } else {
-    const int xc =
-        (mat[2] - (1 << WARPEDMODEL_PREC_BITS)) * x + mat[3] * y + mat[0];
-    const int yc =
-        mat[4] * x + (mat[5] - (1 << WARPEDMODEL_PREC_BITS)) * y + mat[1];
-    tx = convert_to_trans_prec(allow_hp, xc);
-    ty = convert_to_trans_prec(allow_hp, yc);
-  }
+
+  const int xc =
+      (mat[2] - (1 << WARPEDMODEL_PREC_BITS)) * x + mat[3] * y + mat[0];
+  const int yc =
+      mat[4] * x + (mat[5] - (1 << WARPEDMODEL_PREC_BITS)) * y + mat[1];
+  tx = convert_to_trans_prec(allow_hp, xc);
+  ty = convert_to_trans_prec(allow_hp, yc);
 
   res.as_mv.row = ty;
   res.as_mv.col = tx;
 
-#if CONFIG_AMVR
   if (is_integer) {
     integer_mv_precision(&res.as_mv);
   }
-#endif
   return res;
 }
 
 static INLINE TransformationType get_gmtype(const WarpedMotionParams *gm) {
-  if (gm->wmmat[6] != 0 || gm->wmmat[7] != 0) {
-    if (!gm->wmmat[6] && !gm->wmmat[4]) return HORTRAPEZOID;
-    if (!gm->wmmat[7] && !gm->wmmat[3]) return VERTRAPEZOID;
-    return HOMOGRAPHY;
-  }
   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);
@@ -333,12 +273,10 @@ static INLINE TransformationType get_gmtype(const WarpedMotionParams *gm) {
   else
     return AFFINE;
 }
-#endif  // CONFIG_GLOBAL_MOTION
 
 typedef struct candidate_mv {
   int_mv this_mv;
   int_mv comp_mv;
-  uint8_t pred_diff[2];
   int weight;
 } CANDIDATE_MV;
 

third_party/aom/av1/common/mvref_common.h

@@ -18,103 +18,36 @@
 extern "C" {
 #endif
 
-#define MVREF_NEIGHBOURS 9
-#define MVREF_ROWS 3
-#define MVREF_COLS 4
+#define MVREF_ROW_COLS 3
+
+// Set the upper limit of the motion vector component magnitude.
+// This would make a motion vector fit in 26 bits. Plus 3 bits for the
+// reference frame index. A tuple of motion vector can hence be stored within
+// 32 bit range for efficient load/store operations.
+#define REFMVS_LIMIT ((1 << 12) - 1)
 
 typedef struct position {
   int row;
   int col;
 } POSITION;
 
-typedef enum {
-  BOTH_ZERO = 0,
-  ZERO_PLUS_PREDICTED = 1,
-  BOTH_PREDICTED = 2,
-  NEW_PLUS_NON_INTRA = 3,
-  BOTH_NEW = 4,
-  INTRA_PLUS_NON_INTRA = 5,
-  BOTH_INTRA = 6,
-  INVALID_CASE = 9
-} motion_vector_context;
-
-// This is used to figure out a context for the ref blocks. The code flattens
-// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by
-// adding 9 for each intra block, 3 for each zero mv and 1 for each new
-// motion vector. This single number is then converted into a context
-// with a single lookup ( counter_to_context ).
-static const int mode_2_counter[] = {
-  9,  // DC_PRED
-  9,  // V_PRED
-  9,  // H_PRED
-  9,  // D45_PRED
-  9,  // D135_PRED
-  9,  // D117_PRED
-  9,  // D153_PRED
-  9,  // D207_PRED
-  9,  // D63_PRED
-  9,  // SMOOTH_PRED
-#if CONFIG_SMOOTH_HV
-  9,    // SMOOTH_V_PRED
-  9,    // SMOOTH_H_PRED
-#endif  // CONFIG_SMOOTH_HV
-  9,    // TM_PRED
-  0,    // NEARESTMV
-  0,    // NEARMV
-  3,    // ZEROMV
-  1,    // NEWMV
-#if CONFIG_COMPOUND_SINGLEREF
-  0,    // SR_NEAREST_NEARMV
-        //  1,    // SR_NEAREST_NEWMV
-  1,    // SR_NEAR_NEWMV
-  3,    // SR_ZERO_NEWMV
-  1,    // SR_NEW_NEWMV
-#endif  // CONFIG_COMPOUND_SINGLEREF
-  0,    // NEAREST_NEARESTMV
-  0,    // NEAR_NEARMV
-  1,    // NEAREST_NEWMV
-  1,    // NEW_NEARESTMV
-  1,    // NEAR_NEWMV
-  1,    // NEW_NEARMV
-  3,    // ZERO_ZEROMV
-  1,    // NEW_NEWMV
-};
-
-// There are 3^3 different combinations of 3 counts that can be either 0,1 or
-// 2. However the actual count can never be greater than 2 so the highest
-// counter we need is 18. 9 is an invalid counter that's never used.
-static const int counter_to_context[19] = {
-  BOTH_PREDICTED,        // 0
-  NEW_PLUS_NON_INTRA,    // 1
-  BOTH_NEW,              // 2
-  ZERO_PLUS_PREDICTED,   // 3
-  NEW_PLUS_NON_INTRA,    // 4
-  INVALID_CASE,          // 5
-  BOTH_ZERO,             // 6
-  INVALID_CASE,          // 7
-  INVALID_CASE,          // 8
-  INTRA_PLUS_NON_INTRA,  // 9
-  INTRA_PLUS_NON_INTRA,  // 10
-  INVALID_CASE,          // 11
-  INTRA_PLUS_NON_INTRA,  // 12
-  INVALID_CASE,          // 13
-  INVALID_CASE,          // 14
-  INVALID_CASE,          // 15
-  INVALID_CASE,          // 16
-  INVALID_CASE,          // 17
-  BOTH_INTRA             // 18
-};
-
-static const int idx_n_column_to_subblock[4][2] = {
-  { 1, 2 }, { 1, 3 }, { 3, 2 }, { 3, 3 }
-};
-
 // clamp_mv_ref
-#if CONFIG_EXT_PARTITION
 #define MV_BORDER (16 << 3)  // Allow 16 pels in 1/8th pel units
-#else
-#define MV_BORDER (8 << 3)  // Allow 8 pels in 1/8th pel units
-#endif                      // CONFIG_EXT_PARTITION
+
+static INLINE int get_relative_dist(const AV1_COMMON *cm, int a, int b) {
+  if (!cm->seq_params.enable_order_hint) return 0;
+
+  const int bits = cm->seq_params.order_hint_bits_minus_1 + 1;
+
+  assert(bits >= 1);
+  assert(a >= 0 && a < (1 << bits));
+  assert(b >= 0 && b < (1 << bits));
+
+  int diff = a - b;
+  int m = 1 << (bits - 1);
+  diff = (diff & (m - 1)) - (diff & m);
+  return diff;
+}
 
 static INLINE void clamp_mv_ref(MV *mv, int bw, int bh, const MACROBLOCKD *xd) {
   clamp_mv(mv, xd->mb_to_left_edge - bw * 8 - MV_BORDER,
@@ -125,19 +58,16 @@ static INLINE void clamp_mv_ref(MV *mv, int bw, int bh, const MACROBLOCKD *xd) {
 
 // This function returns either the appropriate sub block or block's mv
 // on whether the block_size < 8x8 and we have check_sub_blocks set.
-static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv,
-                                      int search_col, int block_idx) {
+static INLINE int_mv get_sub_block_mv(const MB_MODE_INFO *candidate,
+                                      int which_mv, int search_col) {
   (void)search_col;
-  (void)block_idx;
-  return candidate->mbmi.mv[which_mv];
+  return candidate->mv[which_mv];
 }
 
-static INLINE int_mv get_sub_block_pred_mv(const MODE_INFO *candidate,
-                                           int which_mv, int search_col,
-                                           int block_idx) {
+static INLINE int_mv get_sub_block_pred_mv(const MB_MODE_INFO *candidate,
+                                           int which_mv, int search_col) {
   (void)search_col;
-  (void)block_idx;
-  return candidate->mbmi.mv[which_mv];
+  return candidate->mv[which_mv];
 }
 
 // Performs mv sign inversion if indicated by the reference frame combination.
@@ -152,48 +82,11 @@ static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
   return mv;
 }
 
-#define CLIP_IN_ADD(mv, bw, bh, xd) clamp_mv_ref(mv, bw, bh, xd)
-
-// This macro is used to add a motion vector mv_ref list if it isn't
-// already in the list.  If it's the second motion vector it will also
-// skip all additional processing and jump to done!
-#define ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, bw, bh, xd, Done)      \
-  do {                                                                       \
-    (mv_ref_list)[(refmv_count)] = (mv);                                     \
-    CLIP_IN_ADD(&(mv_ref_list)[(refmv_count)].as_mv, (bw), (bh), (xd));      \
-    if (refmv_count && (mv_ref_list)[1].as_int != (mv_ref_list)[0].as_int) { \
-      (refmv_count) = 2;                                                     \
-      goto Done;                                                             \
-    }                                                                        \
-    (refmv_count) = 1;                                                       \
-  } while (0)
-
-// If either reference frame is different, not INTRA, and they
-// are different from each other scale and add the mv to our list.
-#define IF_DIFF_REF_FRAME_ADD_MV(mbmi, ref_frame, ref_sign_bias, refmv_count, \
-                                 mv_ref_list, bw, bh, xd, Done)               \
-  do {                                                                        \
-    if (is_inter_block(mbmi)) {                                               \
-      if ((mbmi)->ref_frame[0] != ref_frame)                                  \
-        ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias),        \
-                        refmv_count, mv_ref_list, bw, bh, xd, Done);          \
-      if (has_second_ref(mbmi) && (mbmi)->ref_frame[1] != ref_frame)          \
-        ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias),        \
-                        refmv_count, mv_ref_list, bw, bh, xd, Done);          \
-    }                                                                         \
-  } while (0)
-
 // 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,
-                            int mi_rows, const AV1_COMMON *cm,
-                            const POSITION *mi_pos) {
-#if CONFIG_DEPENDENT_HORZTILES
-  const int dependent_horz_tile_flag = cm->dependent_horz_tiles;
-#else
+                            int mi_rows, const POSITION *mi_pos) {
   const int dependent_horz_tile_flag = 0;
-  (void)cm;
-#endif
   if (dependent_horz_tile_flag && !tile->tg_horz_boundary) {
     return !(mi_row + mi_pos->row < 0 ||
              mi_col + mi_pos->col < tile->mi_col_start ||
@@ -208,14 +101,8 @@ static INLINE int is_inside(const TileInfo *const tile, int mi_col, int mi_row,
 }
 
 static INLINE int find_valid_row_offset(const TileInfo *const tile, int mi_row,
-                                        int mi_rows, const AV1_COMMON *cm,
-                                        int row_offset) {
-#if CONFIG_DEPENDENT_HORZTILES
-  const int dependent_horz_tile_flag = cm->dependent_horz_tiles;
-#else
+                                        int mi_rows, int row_offset) {
   const int dependent_horz_tile_flag = 0;
-  (void)cm;
-#endif
   if (dependent_horz_tile_flag && !tile->tg_horz_boundary)
     return clamp(row_offset, -mi_row, mi_rows - mi_row - 1);
   else
@@ -229,87 +116,49 @@ static INLINE int find_valid_col_offset(const TileInfo *const tile, int mi_col,
                tile->mi_col_end - mi_col - 1);
 }
 
-static INLINE void lower_mv_precision(MV *mv, int allow_hp
-#if CONFIG_AMVR
-                                      ,
-                                      int is_integer
-#endif
-                                      ) {
-#if CONFIG_AMVR
+static INLINE void lower_mv_precision(MV *mv, int allow_hp, int is_integer) {
   if (is_integer) {
     integer_mv_precision(mv);
   } else {
-#endif
     if (!allow_hp) {
       if (mv->row & 1) mv->row += (mv->row > 0 ? -1 : 1);
       if (mv->col & 1) mv->col += (mv->col > 0 ? -1 : 1);
     }
-#if CONFIG_AMVR
   }
-#endif
 }
 
-static INLINE uint8_t av1_get_pred_diff_ctx(const int_mv pred_mv,
-                                            const int_mv this_mv) {
-  if (abs(this_mv.as_mv.row - pred_mv.as_mv.row) <= 4 &&
-      abs(this_mv.as_mv.col - pred_mv.as_mv.col) <= 4)
-    return 2;
-  else
-    return 1;
-}
-
-static INLINE int av1_nmv_ctx(const uint8_t ref_mv_count,
-                              const CANDIDATE_MV *ref_mv_stack, int ref,
-                              int ref_mv_idx) {
-  if (ref_mv_stack[ref_mv_idx].weight >= REF_CAT_LEVEL && ref_mv_count > 0)
-    return ref_mv_stack[ref_mv_idx].pred_diff[ref];
-
-  return 0;
-}
-
-#if CONFIG_EXT_COMP_REFS
-static INLINE int8_t av1_uni_comp_ref_idx(const MV_REFERENCE_FRAME *const rf) {
+static INLINE int8_t get_uni_comp_ref_idx(const MV_REFERENCE_FRAME *const rf) {
   // Single ref pred
   if (rf[1] <= INTRA_FRAME) return -1;
 
   // Bi-directional comp ref pred
   if ((rf[0] < BWDREF_FRAME) && (rf[1] >= BWDREF_FRAME)) return -1;
 
-  for (int8_t ref_idx = 0; ref_idx < UNIDIR_COMP_REFS; ++ref_idx) {
+  for (int8_t ref_idx = 0; ref_idx < TOTAL_UNIDIR_COMP_REFS; ++ref_idx) {
     if (rf[0] == comp_ref0(ref_idx) && rf[1] == comp_ref1(ref_idx))
       return ref_idx;
   }
   return -1;
 }
-#endif  // CONFIG_EXT_COMP_REFS
 
 static INLINE int8_t av1_ref_frame_type(const MV_REFERENCE_FRAME *const rf) {
   if (rf[1] > INTRA_FRAME) {
-#if CONFIG_EXT_COMP_REFS
-    int8_t uni_comp_ref_idx = av1_uni_comp_ref_idx(rf);
-#if !USE_UNI_COMP_REFS
-    // NOTE: uni-directional comp refs disabled
-    assert(uni_comp_ref_idx < 0);
-#endif  // !USE_UNI_COMP_REFS
+    const int8_t uni_comp_ref_idx = get_uni_comp_ref_idx(rf);
     if (uni_comp_ref_idx >= 0) {
-      assert((TOTAL_REFS_PER_FRAME + FWD_REFS * BWD_REFS + uni_comp_ref_idx) <
+      assert((REF_FRAMES + FWD_REFS * BWD_REFS + uni_comp_ref_idx) <
              MODE_CTX_REF_FRAMES);
-      return TOTAL_REFS_PER_FRAME + FWD_REFS * BWD_REFS + uni_comp_ref_idx;
+      return REF_FRAMES + FWD_REFS * BWD_REFS + uni_comp_ref_idx;
     } else {
-#endif  // CONFIG_EXT_COMP_REFS
-      return TOTAL_REFS_PER_FRAME + FWD_RF_OFFSET(rf[0]) +
+      return REF_FRAMES + FWD_RF_OFFSET(rf[0]) +
              BWD_RF_OFFSET(rf[1]) * FWD_REFS;
-#if CONFIG_EXT_COMP_REFS
     }
-#endif  // CONFIG_EXT_COMP_REFS
   }
 
   return rf[0];
 }
 
 // clang-format off
-static MV_REFERENCE_FRAME ref_frame_map[COMP_REFS][2] = {
-#if CONFIG_EXT_REFS
+static MV_REFERENCE_FRAME ref_frame_map[TOTAL_COMP_REFS][2] = {
   { LAST_FRAME, BWDREF_FRAME },  { LAST2_FRAME, BWDREF_FRAME },
   { LAST3_FRAME, BWDREF_FRAME }, { GOLDEN_FRAME, BWDREF_FRAME },
 
@@ -317,58 +166,51 @@ static MV_REFERENCE_FRAME ref_frame_map[COMP_REFS][2] = {
   { LAST3_FRAME, ALTREF2_FRAME }, { GOLDEN_FRAME, ALTREF2_FRAME },
 
   { LAST_FRAME, ALTREF_FRAME },  { LAST2_FRAME, ALTREF_FRAME },
-  { LAST3_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME }
+  { LAST3_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME },
 
-  // TODO(zoeliu): Temporarily disable uni-directional comp refs
-#if CONFIG_EXT_COMP_REFS
-  , { LAST_FRAME, LAST2_FRAME }, { LAST_FRAME, LAST3_FRAME },
-  { LAST_FRAME, GOLDEN_FRAME }, { BWDREF_FRAME, ALTREF_FRAME }
-  // TODO(zoeliu): When ALTREF2 is enabled, we may add:
-  //               {BWDREF_FRAME, ALTREF2_FRAME}
-#endif  // CONFIG_EXT_COMP_REFS
-#else  // !CONFIG_EXT_REFS
-  { LAST_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME }
-#endif  // CONFIG_EXT_REFS
+  { LAST_FRAME, LAST2_FRAME }, { LAST_FRAME, LAST3_FRAME },
+  { LAST_FRAME, GOLDEN_FRAME }, { BWDREF_FRAME, ALTREF_FRAME },
+
+  // NOTE: Following reference frame pairs are not supported to be explicitly
+  //       signalled, but they are possibly chosen by the use of skip_mode,
+  //       which may use the most recent one-sided reference frame pair.
+  { LAST2_FRAME, LAST3_FRAME }, { LAST2_FRAME, GOLDEN_FRAME },
+  { LAST3_FRAME, GOLDEN_FRAME }, {BWDREF_FRAME, ALTREF2_FRAME},
+  { ALTREF2_FRAME, ALTREF_FRAME }
 };
 // clang-format on
 
 static INLINE void av1_set_ref_frame(MV_REFERENCE_FRAME *rf,
                                      int8_t ref_frame_type) {
-  if (ref_frame_type >= TOTAL_REFS_PER_FRAME) {
-    rf[0] = ref_frame_map[ref_frame_type - TOTAL_REFS_PER_FRAME][0];
-    rf[1] = ref_frame_map[ref_frame_type - TOTAL_REFS_PER_FRAME][1];
+  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 {
     rf[0] = ref_frame_type;
     rf[1] = NONE_FRAME;
-#if CONFIG_INTRABC
     assert(ref_frame_type > NONE_FRAME);
-#else
-    assert(ref_frame_type > INTRA_FRAME);
-#endif
-    assert(ref_frame_type < TOTAL_REFS_PER_FRAME);
   }
 }
 
+static uint16_t compound_mode_ctx_map[3][COMP_NEWMV_CTXS] = {
+  { 0, 1, 1, 1, 1 },
+  { 1, 2, 3, 4, 4 },
+  { 4, 4, 5, 6, 7 },
+};
+
 static INLINE int16_t av1_mode_context_analyzer(
-    const int16_t *const mode_context, const MV_REFERENCE_FRAME *const rf,
-    BLOCK_SIZE bsize, int block) {
-  int16_t mode_ctx = 0;
-  int8_t ref_frame_type = av1_ref_frame_type(rf);
+    const int16_t *const mode_context, const MV_REFERENCE_FRAME *const rf) {
+  const int8_t ref_frame = av1_ref_frame_type(rf);
 
-  if (block >= 0) {
-    mode_ctx = mode_context[rf[0]] & 0x00ff;
-#if !CONFIG_CB4X4
-    if (block > 0 && bsize < BLOCK_8X8 && bsize > BLOCK_4X4)
-      mode_ctx |= (1 << SKIP_NEARESTMV_SUB8X8_OFFSET);
-#else
-    (void)block;
-    (void)bsize;
-#endif
+  if (rf[1] <= INTRA_FRAME) return mode_context[ref_frame];
 
-    return mode_ctx;
-  }
+  const int16_t newmv_ctx = mode_context[ref_frame] & NEWMV_CTX_MASK;
+  const int16_t refmv_ctx =
+      (mode_context[ref_frame] >> REFMV_OFFSET) & REFMV_CTX_MASK;
 
-  return mode_context[ref_frame_type];
+  const int16_t comp_ctx = compound_mode_ctx_map[refmv_ctx >> 1][AOMMIN(
+      newmv_ctx, COMP_NEWMV_CTXS - 1)];
+  return comp_ctx;
 }
 
 static INLINE uint8_t av1_drl_ctx(const CANDIDATE_MV *ref_mv_stack,
@@ -379,92 +221,99 @@ static INLINE uint8_t av1_drl_ctx(const CANDIDATE_MV *ref_mv_stack,
 
   if (ref_mv_stack[ref_idx].weight >= REF_CAT_LEVEL &&
       ref_mv_stack[ref_idx + 1].weight < REF_CAT_LEVEL)
-    return 2;
+    return 1;
 
   if (ref_mv_stack[ref_idx].weight < REF_CAT_LEVEL &&
       ref_mv_stack[ref_idx + 1].weight < REF_CAT_LEVEL)
-    return 3;
+    return 2;
 
   return 0;
 }
 
-#if CONFIG_FRAME_MARKER
 void av1_setup_frame_buf_refs(AV1_COMMON *cm);
-#if CONFIG_FRAME_SIGN_BIAS
 void av1_setup_frame_sign_bias(AV1_COMMON *cm);
-#endif  // CONFIG_FRAME_SIGN_BIAS
-#if CONFIG_MFMV
+void av1_setup_skip_mode_allowed(AV1_COMMON *cm);
 void av1_setup_motion_field(AV1_COMMON *cm);
-#endif  // CONFIG_MFMV
-#endif  // CONFIG_FRAME_MARKER
+void av1_set_frame_refs(AV1_COMMON *const cm, int lst_map_idx, int gld_map_idx);
 
-void av1_copy_frame_mvs(const AV1_COMMON *const cm, MODE_INFO *mi, int mi_row,
-                        int mi_col, int x_mis, int y_mis);
+static INLINE void av1_collect_neighbors_ref_counts(MACROBLOCKD *const xd) {
+  av1_zero(xd->neighbors_ref_counts);
+
+  uint8_t *const ref_counts = xd->neighbors_ref_counts;
+
+  const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const int above_in_image = xd->up_available;
+  const int left_in_image = xd->left_available;
+
+  // Above neighbor
+  if (above_in_image && is_inter_block(above_mbmi)) {
+    ref_counts[above_mbmi->ref_frame[0]]++;
+    if (has_second_ref(above_mbmi)) {
+      ref_counts[above_mbmi->ref_frame[1]]++;
+    }
+  }
+
+  // Left neighbor
+  if (left_in_image && is_inter_block(left_mbmi)) {
+    ref_counts[left_mbmi->ref_frame[0]]++;
+    if (has_second_ref(left_mbmi)) {
+      ref_counts[left_mbmi->ref_frame[1]]++;
+    }
+  }
+}
+
+void av1_copy_frame_mvs(const AV1_COMMON *const cm, MB_MODE_INFO *mi,
+                        int mi_row, int mi_col, int x_mis, int y_mis);
 
-typedef void (*find_mv_refs_sync)(void *const data, int mi_row);
 void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
-                      MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
-                      uint8_t *ref_mv_count, CANDIDATE_MV *ref_mv_stack,
-                      int16_t *compound_mode_context, int_mv *mv_ref_list,
-                      int mi_row, int mi_col, find_mv_refs_sync sync,
-                      void *const data, int16_t *mode_context);
+                      MB_MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
+                      uint8_t ref_mv_count[MODE_CTX_REF_FRAMES],
+                      CANDIDATE_MV ref_mv_stack[][MAX_REF_MV_STACK_SIZE],
+                      int_mv mv_ref_list[][MAX_MV_REF_CANDIDATES],
+                      int_mv *global_mvs, int mi_row, int mi_col,
+                      int16_t *mode_context);
 
 // check a list of motion vectors by sad score using a number rows of pixels
 // above and a number cols of pixels in the left to select the one with best
 // score to use as ref motion vector
-#if CONFIG_AMVR
 void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
                            int_mv *near_mv, int is_integer);
-#else
-void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
-                           int_mv *near_mv);
-#endif
 
-void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                   int block, int ref, int mi_row, int mi_col,
-                                   CANDIDATE_MV *ref_mv_stack,
-                                   uint8_t *ref_mv_count, int_mv *mv_list,
-                                   int_mv *nearest_mv, int_mv *near_mv);
-
-// This function keeps a mode count for a given MB/SB
-void av1_update_mv_context(const AV1_COMMON *cm, const MACROBLOCKD *xd,
-                           MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
-                           int_mv *mv_ref_list, int block, int mi_row,
-                           int mi_col, int16_t *mode_context);
-
-#if CONFIG_WARPED_MOTION
-#if WARPED_MOTION_SORT_SAMPLES
-int sortSamples(int *pts_mv, MV *mv, int *pts, int *pts_inref, int len);
-int findSamples(const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
-                int *pts, int *pts_inref, int *pts_mv);
-#else
+int selectSamples(MV *mv, int *pts, int *pts_inref, int len, BLOCK_SIZE bsize);
 int findSamples(const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
                 int *pts, int *pts_inref);
-#endif  // WARPED_MOTION_SORT_SAMPLES
-#endif  // CONFIG_WARPED_MOTION
 
-#if CONFIG_INTRABC
-static INLINE void av1_find_ref_dv(int_mv *ref_dv, int mi_row, int mi_col) {
-  // TODO(aconverse@google.com): Handle tiles and such
+#define INTRABC_DELAY_PIXELS 256  //  Delay of 256 pixels
+#define INTRABC_DELAY_SB64 (INTRABC_DELAY_PIXELS / 64)
+#define USE_WAVE_FRONT 1  // Use only top left area of frame for reference.
+
+static INLINE void av1_find_ref_dv(int_mv *ref_dv, const TileInfo *const tile,
+                                   int mib_size, int mi_row, int mi_col) {
   (void)mi_col;
-  if (mi_row < MAX_MIB_SIZE) {
+  if (mi_row - mib_size < tile->mi_row_start) {
     ref_dv->as_mv.row = 0;
-    ref_dv->as_mv.col = -MI_SIZE * MAX_MIB_SIZE;
+    ref_dv->as_mv.col = -MI_SIZE * mib_size - INTRABC_DELAY_PIXELS;
   } else {
-    ref_dv->as_mv.row = -MI_SIZE * MAX_MIB_SIZE;
+    ref_dv->as_mv.row = -MI_SIZE * mib_size;
     ref_dv->as_mv.col = 0;
   }
+  ref_dv->as_mv.row *= 8;
+  ref_dv->as_mv.col *= 8;
 }
 
-static INLINE int is_dv_valid(const MV dv, const TileInfo *const tile,
-                              int mi_row, int mi_col, BLOCK_SIZE bsize) {
+static INLINE int av1_is_dv_valid(const MV dv, const AV1_COMMON *cm,
+                                  const MACROBLOCKD *xd, int mi_row, int mi_col,
+                                  BLOCK_SIZE bsize, int mib_size_log2) {
   const int bw = block_size_wide[bsize];
   const int bh = block_size_high[bsize];
   const int SCALE_PX_TO_MV = 8;
   // Disallow subpixel for now
   // SUBPEL_MASK is not the correct scale
-  if ((dv.row & (SCALE_PX_TO_MV - 1) || dv.col & (SCALE_PX_TO_MV - 1)))
+  if (((dv.row & (SCALE_PX_TO_MV - 1)) || (dv.col & (SCALE_PX_TO_MV - 1))))
     return 0;
+
+  const TileInfo *const tile = &xd->tile;
   // Is the source top-left inside the current tile?
   const int src_top_edge = mi_row * MI_SIZE * SCALE_PX_TO_MV + dv.row;
   const int tile_top_edge = tile->mi_row_start * MI_SIZE * SCALE_PX_TO_MV;
@@ -479,20 +328,44 @@ static INLINE int is_dv_valid(const MV dv, const TileInfo *const tile,
   const int src_right_edge = (mi_col * MI_SIZE + bw) * SCALE_PX_TO_MV + dv.col;
   const int tile_right_edge = tile->mi_col_end * MI_SIZE * SCALE_PX_TO_MV;
   if (src_right_edge > tile_right_edge) return 0;
-  // Is the bottom right within an already coded SB?
-  const int active_sb_top_edge =
-      (mi_row & ~MAX_MIB_MASK) * MI_SIZE * SCALE_PX_TO_MV;
-  const int active_sb_bottom_edge =
-      ((mi_row & ~MAX_MIB_MASK) + MAX_MIB_SIZE) * MI_SIZE * SCALE_PX_TO_MV;
-  const int active_sb_left_edge =
-      (mi_col & ~MAX_MIB_MASK) * MI_SIZE * SCALE_PX_TO_MV;
-  if (src_bottom_edge > active_sb_bottom_edge) return 0;
-  if (src_bottom_edge > active_sb_top_edge &&
-      src_right_edge > active_sb_left_edge)
+
+  // Special case for sub 8x8 chroma cases, to prevent referring to chroma
+  // pixels outside current tile.
+  for (int plane = 1; plane < av1_num_planes(cm); ++plane) {
+    const struct macroblockd_plane *const pd = &xd->plane[plane];
+    if (is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
+                            pd->subsampling_y)) {
+      if (bw < 8 && pd->subsampling_x)
+        if (src_left_edge < tile_left_edge + 4 * SCALE_PX_TO_MV) return 0;
+      if (bh < 8 && pd->subsampling_y)
+        if (src_top_edge < tile_top_edge + 4 * SCALE_PX_TO_MV) return 0;
+    }
+  }
+
+  // Is the bottom right within an already coded SB? Also consider additional
+  // constraints to facilitate HW decoder.
+  const int max_mib_size = 1 << mib_size_log2;
+  const int active_sb_row = mi_row >> mib_size_log2;
+  const int active_sb64_col = (mi_col * MI_SIZE) >> 6;
+  const int sb_size = max_mib_size * MI_SIZE;
+  const int src_sb_row = ((src_bottom_edge >> 3) - 1) / sb_size;
+  const int src_sb64_col = ((src_right_edge >> 3) - 1) >> 6;
+  const int total_sb64_per_row =
+      ((tile->mi_col_end - tile->mi_col_start - 1) >> 4) + 1;
+  const int active_sb64 = active_sb_row * total_sb64_per_row + active_sb64_col;
+  const int src_sb64 = src_sb_row * total_sb64_per_row + src_sb64_col;
+  if (src_sb64 >= active_sb64 - INTRABC_DELAY_SB64) return 0;
+
+#if USE_WAVE_FRONT
+  const int gradient = 1 + INTRABC_DELAY_SB64 + (sb_size > 64);
+  const int wf_offset = gradient * (active_sb_row - src_sb_row);
+  if (src_sb_row > active_sb_row ||
+      src_sb64_col >= active_sb64_col - INTRABC_DELAY_SB64 + wf_offset)
     return 0;
+#endif
+
   return 1;
 }
-#endif  // CONFIG_INTRABC
 
 #ifdef __cplusplus
 }  // extern "C"

third_party/aom/av1/common/ncobmc_kernels.h

@@ -1,22 +0,0 @@
-/*
- * 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 <stdio.h>
-#include "av1/common/enums.h"
-#include "av1/common/onyxc_int.h"
-#include "av1/common/common.h"
-
-#ifndef AV1_COMMON_NCOBMC_KERNELS_H_
-#define AV1_COMMON_NCOBMC_KERNELS_H_
-
-void get_default_ncobmc_kernels(AV1_COMMON *cm);
-
-#endif  // AV1_COMMON_NCOBMC_KERNELS_H_

third_party/aom/av1/common/obmc.h

@@ -12,31 +12,31 @@
 #ifndef AV1_COMMON_OBMC_H_
 #define AV1_COMMON_OBMC_H_
 
-#if CONFIG_MOTION_VAR
 typedef void (*overlappable_nb_visitor_t)(MACROBLOCKD *xd, int rel_mi_pos,
-                                          uint8_t nb_mi_size, MODE_INFO *nb_mi,
-                                          void *fun_ctxt);
+                                          uint8_t nb_mi_size,
+                                          MB_MODE_INFO *nb_mi, void *fun_ctxt,
+                                          const int num_planes);
 
 static INLINE void foreach_overlappable_nb_above(const AV1_COMMON *cm,
                                                  MACROBLOCKD *xd, int mi_col,
                                                  int nb_max,
                                                  overlappable_nb_visitor_t fun,
                                                  void *fun_ctxt) {
+  const int num_planes = av1_num_planes(cm);
   if (!xd->up_available) return;
 
   int nb_count = 0;
 
   // prev_row_mi points into the mi array, starting at the beginning of the
   // previous row.
-  MODE_INFO **prev_row_mi = xd->mi - mi_col - 1 * xd->mi_stride;
+  MB_MODE_INFO **prev_row_mi = xd->mi - mi_col - 1 * xd->mi_stride;
   const int end_col = AOMMIN(mi_col + xd->n8_w, cm->mi_cols);
   uint8_t mi_step;
   for (int above_mi_col = mi_col; above_mi_col < end_col && nb_count < nb_max;
        above_mi_col += mi_step) {
-    MODE_INFO **above_mi = prev_row_mi + above_mi_col;
-    mi_step = AOMMIN(mi_size_wide[above_mi[0]->mbmi.sb_type],
-                     mi_size_wide[BLOCK_64X64]);
-#if CONFIG_CHROMA_SUB8X8
+    MB_MODE_INFO **above_mi = prev_row_mi + above_mi_col;
+    mi_step =
+        AOMMIN(mi_size_wide[above_mi[0]->sb_type], mi_size_wide[BLOCK_64X64]);
     // If we're considering a block with width 4, it should be treated as
     // half of a pair of blocks with chroma information in the second. Move
     // above_mi_col back to the start of the pair if needed, set above_mbmi
@@ -47,12 +47,10 @@ static INLINE void foreach_overlappable_nb_above(const AV1_COMMON *cm,
       above_mi = prev_row_mi + above_mi_col + 1;
       mi_step = 2;
     }
-#endif  // CONFIG_CHROMA_SUB8X8
-    MB_MODE_INFO *above_mbmi = &above_mi[0]->mbmi;
-    if (is_neighbor_overlappable(above_mbmi)) {
+    if (is_neighbor_overlappable(*above_mi)) {
       ++nb_count;
       fun(xd, above_mi_col - mi_col, AOMMIN(xd->n8_w, mi_step), *above_mi,
-          fun_ctxt);
+          fun_ctxt, num_planes);
     }
   }
 }
@@ -62,35 +60,32 @@ static INLINE void foreach_overlappable_nb_left(const AV1_COMMON *cm,
                                                 int nb_max,
                                                 overlappable_nb_visitor_t fun,
                                                 void *fun_ctxt) {
+  const int num_planes = av1_num_planes(cm);
   if (!xd->left_available) return;
 
   int nb_count = 0;
 
   // prev_col_mi points into the mi array, starting at the top of the
   // previous column
-  MODE_INFO **prev_col_mi = xd->mi - 1 - mi_row * xd->mi_stride;
+  MB_MODE_INFO **prev_col_mi = xd->mi - 1 - mi_row * xd->mi_stride;
   const int end_row = AOMMIN(mi_row + xd->n8_h, cm->mi_rows);
   uint8_t mi_step;
   for (int left_mi_row = mi_row; left_mi_row < end_row && nb_count < nb_max;
        left_mi_row += mi_step) {
-    MODE_INFO **left_mi = prev_col_mi + left_mi_row * xd->mi_stride;
-    mi_step = AOMMIN(mi_size_high[left_mi[0]->mbmi.sb_type],
-                     mi_size_high[BLOCK_64X64]);
-#if CONFIG_CHROMA_SUB8X8
+    MB_MODE_INFO **left_mi = prev_col_mi + left_mi_row * xd->mi_stride;
+    mi_step =
+        AOMMIN(mi_size_high[left_mi[0]->sb_type], mi_size_high[BLOCK_64X64]);
     if (mi_step == 1) {
       left_mi_row &= ~1;
       left_mi = prev_col_mi + (left_mi_row + 1) * xd->mi_stride;
       mi_step = 2;
     }
-#endif  // CONFIG_CHROMA_SUB8X8
-    MB_MODE_INFO *left_mbmi = &left_mi[0]->mbmi;
-    if (is_neighbor_overlappable(left_mbmi)) {
+    if (is_neighbor_overlappable(*left_mi)) {
       ++nb_count;
       fun(xd, left_mi_row - mi_row, AOMMIN(xd->n8_h, mi_step), *left_mi,
-          fun_ctxt);
+          fun_ctxt, num_planes);
     }
   }
 }
 
-#endif  // CONFIG_MOTION_VAR
 #endif  // AV1_COMMON_OBMC_H_

third_party/aom/av1/common/odintrin.c

@@ -13,16 +13,6 @@
 
 #include "av1/common/odintrin.h"
 
-#if defined(OD_ENABLE_ASSERTIONS)
-# include <stdio.h>
-
-void od_fatal_impl(const char *_str, const char *_file, int _line) {
-  fprintf(stderr, "Fatal (internal) error in %s, line %d: %s\n",
-   _file, _line, _str);
-  abort();
-}
-#endif
-
 /*Constants for use with OD_DIVU_SMALL().
   See \cite{Rob05} for details on computing these constants.
   @INPROCEEDINGS{Rob05,

third_party/aom/av1/common/odintrin.h

@@ -14,10 +14,6 @@
 #ifndef AV1_COMMON_ODINTRIN_H_
 #define AV1_COMMON_ODINTRIN_H_
 
-#if defined(_MSC_VER)
-# define _USE_MATH_DEFINES
-#endif
-#include <math.h>
 #include <stdlib.h>
 #include <string.h>
 
@@ -30,71 +26,8 @@
 extern "C" {
 #endif
 
-# if !defined(M_PI)
-#  define M_PI      (3.1415926535897932384626433832795)
-# endif
-
-# if !defined(M_SQRT2)
-#  define M_SQRT2 (1.41421356237309504880168872420970)
-# endif
-
-# if !defined(M_SQRT1_2)
-#  define M_SQRT1_2 (0.70710678118654752440084436210485)
-# endif
-
-# if !defined(M_LOG2E)
-#  define M_LOG2E (1.4426950408889634073599246810019)
-# endif
-
-# if !defined(M_LN2)
-#  define M_LN2 (0.69314718055994530941723212145818)
-# endif
-
-/*Smallest blocks are 4x4*/
-#define OD_LOG_BSIZE0 (2)
-/*There are 5 block sizes total (4x4, 8x8, 16x16, 32x32 and 64x64).*/
-#define OD_NBSIZES (5)
-
-/*There are 4 transform sizes total in AV1 (4x4, 8x8, 16x16 and 32x32).*/
-#define OD_TXSIZES TX_SIZES
-/*The log of the maximum length of the side of a transform.*/
-#define OD_LOG_TXSIZE_MAX (OD_LOG_BSIZE0 + OD_TXSIZES - 1)
-/*The maximum length of the side of a transform.*/
-#define OD_TXSIZE_MAX (1 << OD_LOG_TXSIZE_MAX)
-
-/**The maximum number of color planes allowed in a single frame.*/
-# define OD_NPLANES_MAX (3)
-
-# define OD_COEFF_SHIFT (4)
-
-# define OD_DISABLE_CFL (1)
-# define OD_DISABLE_FILTER (1)
-
-#if !defined(NDEBUG)
-# define OD_ENABLE_ASSERTIONS (1)
-#endif
-
-# define OD_LOG(a)
-# define OD_LOG_PARTIAL(a)
-
-/*Possible block sizes, note that OD_BLOCK_NXN = log2(N) - 2.*/
-#define OD_BLOCK_4X4 (0)
-#define OD_BLOCK_8X8 (1)
-#define OD_BLOCK_16X16 (2)
-#define OD_BLOCK_32X32 (3)
-#define OD_BLOCK_SIZES (OD_BLOCK_32X32 + 1)
-
-# define OD_LIMIT_BSIZE_MIN (OD_BLOCK_4X4)
-# define OD_LIMIT_BSIZE_MAX (OD_BLOCK_32X32)
-
 typedef int od_coeff;
 
-/*This is the strength reduced version of ((_a)/(1 << (_b))).
-  This will not work for _b == 0, however currently this is only used for
-   b == 1 anyway.*/
-# define OD_UNBIASED_RSHIFT32(_a, _b) \
-  (((int32_t)(((uint32_t)(_a) >> (32 - (_b))) + (_a))) >> (_b))
-
 #define OD_DIVU_DMAX (1024)
 
 extern uint32_t OD_DIVU_SMALL_CONSTS[OD_DIVU_DMAX][2];
@@ -116,14 +49,6 @@ extern uint32_t OD_DIVU_SMALL_CONSTS[OD_DIVU_DMAX][2];
 #define OD_CLZ0 (1)
 #define OD_CLZ(x) (-get_msb(x))
 #define OD_ILOG_NZ(x) (OD_CLZ0 - OD_CLZ(x))
-/*Note that __builtin_clz is not defined when x == 0, according to the gcc
-   documentation (and that of the x86 BSR instruction that implements it), so
-   we have to special-case it.
-  We define a special version of the macro to use when x can be zero.*/
-#define OD_ILOG(x) ((x) ? OD_ILOG_NZ(x) : 0)
-
-#define OD_LOG2(x) (M_LOG2E*log(x))
-#define OD_EXP2(x) (exp(M_LN2*(x)))
 
 /*Enable special features for gcc and compatible compilers.*/
 #if defined(__GNUC__) && defined(__GNUC_MINOR__) && defined(__GNUC_PATCHLEVEL__)
@@ -146,36 +71,6 @@ extern uint32_t OD_DIVU_SMALL_CONSTS[OD_DIVU_DMAX][2];
 #define OD_ARG_NONNULL(x)
 #endif
 
-#if defined(OD_ENABLE_ASSERTIONS)
-#if OD_GNUC_PREREQ(2, 5, 0)
-__attribute__((noreturn))
-#endif
-void od_fatal_impl(const char *_str, const char *_file, int _line);
-
-#define OD_FATAL(_str) (od_fatal_impl(_str, __FILE__, __LINE__))
-
-#define OD_ASSERT(_cond)                     \
-  do {                                       \
-    if (!(_cond)) {                          \
-      OD_FATAL("assertion failed: " #_cond); \
-    }                                        \
-  } while (0)
-
-#define OD_ASSERT2(_cond, _message)                        \
-  do {                                                     \
-    if (!(_cond)) {                                        \
-      OD_FATAL("assertion failed: " #_cond "\n" _message); \
-    }                                                      \
-  } while (0)
-
-#define OD_ALWAYS_TRUE(_cond) OD_ASSERT(_cond)
-
-#else
-#define OD_ASSERT(_cond)
-#define OD_ASSERT2(_cond, _message)
-#define OD_ALWAYS_TRUE(_cond) ((void)(_cond))
-#endif
-
 /** Copy n elements of memory from src to dst. The 0* term provides
     compile-time type checking  */
 #if !defined(OVERRIDE_OD_COPY)
@@ -190,85 +85,10 @@ void od_fatal_impl(const char *_str, const char *_file, int _line);
  (memmove((dst), (src), sizeof(*(dst))*(n) + 0*((dst) - (src)) ))
 #endif
 
-/** Linkage will break without this if using a C++ compiler, and will issue
- * warnings without this for a C compiler*/
-#if defined(__cplusplus)
-# define OD_EXTERN extern
-#else
-# define OD_EXTERN
-#endif
-
-/** Set n elements of dst to zero */
-#if !defined(OVERRIDE_OD_CLEAR)
-# define OD_CLEAR(dst, n) (memset((dst), 0, sizeof(*(dst))*(n)))
-#endif
-
-/** Silence unused parameter/variable warnings */
-# define OD_UNUSED(expr) (void)(expr)
-
-#if defined(OD_FLOAT_PVQ)
-typedef double od_val16;
-typedef double od_val32;
-# define OD_QCONST32(x, bits) (x)
-# define OD_ROUND16(x) (x)
-# define OD_ROUND32(x) (x)
-# define OD_SHL(x, shift) (x)
-# define OD_SHR(x, shift) (x)
-# define OD_SHR_ROUND(x, shift) (x)
-# define OD_ABS(x) (fabs(x))
-# define OD_MULT16_16(a, b) ((a)*(b))
-# define OD_MULT16_32_Q16(a, b) ((a)*(b))
-#else
-typedef int16_t od_val16;
-typedef int32_t od_val32;
-/** Compile-time conversion of float constant to 32-bit value */
-# define OD_QCONST32(x, bits) ((od_val32)(.5 + (x)*(((od_val32)1) << (bits))))
-# define OD_ROUND16(x) (int16_t)(floor(.5 + (x)))
-# define OD_ROUND32(x) (int32_t)(floor(.5 + (x)))
-/*Shift x left by shift*/
-# define OD_SHL(a, shift) ((int32_t)((uint32_t)(a) << (shift)))
-/*Shift x right by shift (without rounding)*/
-# define OD_SHR(x, shift) \
-  ((int32_t)((x) >> (shift)))
-/*Shift x right by shift (with rounding)*/
-# define OD_SHR_ROUND(x, shift) \
-  ((int32_t)(((x) + (1 << (shift) >> 1)) >> (shift)))
-/*Shift x right by shift (without rounding) or left by -shift if shift
-  is negative.*/
-# define OD_VSHR(x, shift) \
-  (((shift) > 0) ? OD_SHR(x, shift) : OD_SHL(x, -(shift)))
-/*Shift x right by shift (with rounding) or left by -shift if shift
-  is negative.*/
-# define OD_VSHR_ROUND(x, shift) \
-  (((shift) > 0) ? OD_SHR_ROUND(x, shift) : OD_SHL(x, -(shift)))
-# define OD_ABS(x) (abs(x))
-/* (od_val32)(od_val16) gives TI compiler a hint that it's 16x16->32 multiply */
-/** 16x16 multiplication where the result fits in 32 bits */
-# define OD_MULT16_16(a, b) \
- (((od_val32)(od_val16)(a))*((od_val32)(od_val16)(b)))
-/* Multiplies 16-bit a by 32-bit b and keeps bits [16:47]. */
-# define OD_MULT16_32_Q16(a, b) ((int16_t)(a)*(int64_t)(int32_t)(b) >> 16)
-/*16x16 multiplication where the result fits in 16 bits, without rounding.*/
-# define OD_MULT16_16_Q15(a, b) \
-  (((int16_t)(a)*((int32_t)(int16_t)(b))) >> 15)
-/*16x16 multiplication where the result fits in 16 bits, without rounding.*/
-# define OD_MULT16_16_Q16(a, b) \
-  ((((int16_t)(a))*((int32_t)(int16_t)(b))) >> 16)
-#endif
-
 /*All of these macros should expect floats as arguments.*/
-/*These two should compile as a single SSE instruction.*/
-# define OD_MINF(a, b) ((a) < (b) ? (a) : (b))
-# define OD_MAXF(a, b) ((a) > (b) ? (a) : (b))
-
-# define OD_DIV_R0(x, y) (((x) + OD_FLIPSIGNI((((y) + 1) >> 1) - 1, (x)))/(y))
-
 # define OD_SIGNMASK(a) (-((a) < 0))
 # define OD_FLIPSIGNI(a, b) (((a) + OD_SIGNMASK(b)) ^ OD_SIGNMASK(b))
 
-# define OD_MULT16_16_Q15(a, b) \
-  (((int16_t)(a)*((int32_t)(int16_t)(b))) >> 15)
-
 #ifdef __cplusplus
 }  // extern "C"
 #endif

third_party/aom/av1/common/partition.c

@@ -1,256 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include "enums.h"
-#include "odintrin.h"
-#include "partition.h"
-#include "zigzag.h"
-
-OD_EXTERN const index_pair *OD_ZIGZAG4[4] = {
-  OD_ZIGZAG4_DCT_DCT,
-  OD_ZIGZAG4_ADST_DCT,
-  OD_ZIGZAG4_DCT_ADST,
-  OD_ZIGZAG4_ADST_ADST
-};
-
-OD_EXTERN const index_pair *OD_ZIGZAG8[4] = {
-  OD_ZIGZAG8_DCT_DCT,
-  OD_ZIGZAG8_ADST_DCT,
-  OD_ZIGZAG8_DCT_ADST,
-  OD_ZIGZAG8_ADST_ADST
-};
-
-OD_EXTERN const index_pair *OD_ZIGZAG16[4] = {
-  OD_ZIGZAG16_DCT_DCT,
-  OD_ZIGZAG16_ADST_DCT,
-  OD_ZIGZAG16_DCT_ADST,
-  OD_ZIGZAG16_ADST_ADST
-};
-
-OD_EXTERN const index_pair *OD_ZIGZAG32[4] = {
-  OD_ZIGZAG32_DCT_DCT,
-  OD_ZIGZAG32_DCT_DCT,
-  OD_ZIGZAG32_DCT_DCT,
-  OD_ZIGZAG32_DCT_DCT
-};
-
-/* The tables below specify how coefficient blocks are translated to
-   and from PVQ partition coding scan order for 4x4, 8x8 and 16x16 */
-
-static const int OD_LAYOUT32_OFFSETS[4] = { 0, 128, 256, 768 };
-const band_layout OD_LAYOUT32 = {
-  OD_ZIGZAG32,
-  32,
-  3,
-  OD_LAYOUT32_OFFSETS
-};
-
-static const int OD_LAYOUT16_OFFSETS[4] = { 0, 32, 64, 192 };
-const band_layout OD_LAYOUT16 = {
-  OD_ZIGZAG16,
-  16,
-  3,
-  OD_LAYOUT16_OFFSETS
-};
-
-const int OD_LAYOUT8_OFFSETS[4] = { 0, 8, 16, 48 };
-const band_layout OD_LAYOUT8 = {
-  OD_ZIGZAG8,
-  8,
-  3,
-  OD_LAYOUT8_OFFSETS
-};
-
-static const int OD_LAYOUT4_OFFSETS[2] = { 0, 15 };
-const band_layout OD_LAYOUT4 = {
-  OD_ZIGZAG4,
-  4,
-  1,
-  OD_LAYOUT4_OFFSETS
-};
-
-/* First element is the number of bands, followed by the list all the band
-  boundaries. */
-static const int OD_BAND_OFFSETS4[] = {1, 1, 16};
-static const int OD_BAND_OFFSETS8[] = {4, 1, 16, 24, 32, 64};
-static const int OD_BAND_OFFSETS16[] = {7, 1, 16, 24, 32, 64, 96, 128, 256};
-static const int OD_BAND_OFFSETS32[] = {10, 1, 16, 24, 32, 64, 96, 128, 256,
- 384, 512, 1024};
-static const int OD_BAND_OFFSETS64[] = {13, 1, 16, 24, 32, 64, 96, 128, 256,
- 384, 512, 1024, 1536, 2048, 4096};
-
-const int *const OD_BAND_OFFSETS[OD_TXSIZES + 1] = {
-  OD_BAND_OFFSETS4,
-  OD_BAND_OFFSETS8,
-  OD_BAND_OFFSETS16,
-  OD_BAND_OFFSETS32,
-  OD_BAND_OFFSETS64
-};
-
-/** Perform a single stage of conversion from a coefficient block in
- * raster order into coding scan order
- *
- * @param [in]     layout  scan order specification
- * @param [out]    dst     destination vector
- * @param [in]     src     source coefficient block
- * @param [int]    int     source vector row stride
- */
-static void od_band_from_raster(const band_layout *layout, tran_low_t *dst,
- const tran_low_t *src, int stride, TX_TYPE tx_type) {
-  int i;
-  int len;
-  len = layout->band_offsets[layout->nb_bands];
-  for (i = 0; i < len; i++) {
-    dst[i] = src[layout->dst_table[tx_type][i][1]*stride + layout->dst_table[tx_type][i][0]];
-  }
-}
-
-/** Perform a single stage of conversion from a vector in coding scan
-    order back into a coefficient block in raster order
- *
- * @param [in]     layout  scan order specification
- * @param [out]    dst     destination coefficient block
- * @param [in]     src     source vector
- * @param [int]    stride  destination vector row stride
- */
-static void od_raster_from_band(const band_layout *layout, tran_low_t *dst,
- int stride, TX_TYPE tx_type, const tran_low_t *src) {
-  int i;
-  int len;
-  len = layout->band_offsets[layout->nb_bands];
-  for (i = 0; i < len; i++) {
-    dst[layout->dst_table[tx_type][i][1]*stride + layout->dst_table[tx_type][i][0]] = src[i];
-  }
-}
-
-static const band_layout *const OD_LAYOUTS[] = {&OD_LAYOUT4, &OD_LAYOUT8,
- &OD_LAYOUT16, &OD_LAYOUT32};
-
-/** Converts a coefficient block in raster order into a vector in
- * coding scan order with the PVQ partitions laid out one after
- * another.  This works in stages; the 4x4 conversion is applied to
- * the coefficients nearest DC, then the 8x8 applied to the 8x8 block
- * nearest DC that was not already coded by 4x4, then 16x16 following
- * the same pattern.
- *
- * @param [out]    dst        destination vector
- * @param [in]     n          block size (along one side)
- * @param [in]     ty_type    transfrom type
- * @param [in]     src        source coefficient block
- * @param [in]     stride     source vector row stride
- */
-void od_raster_to_coding_order(tran_low_t *dst, int n, TX_TYPE ty_type,
- const tran_low_t *src, int stride) {
-  int bs;
-  /* dst + 1 because DC is not included for 4x4 blocks. */
-  od_band_from_raster(OD_LAYOUTS[0], dst + 1, src, stride, ty_type);
-  for (bs = 1; bs < OD_TXSIZES; bs++) {
-    int size;
-    int offset;
-    /* Length of block size > 4. */
-    size = 1 << (OD_LOG_BSIZE0 + bs);
-    /* Offset is the size of the previous block squared. */
-    offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
-    if (n >= size) {
-      /* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
-      od_band_from_raster(OD_LAYOUTS[bs], dst + offset, src, stride, ty_type);
-    }
-  }
-  dst[0] = src[0];
-}
-
-/** Converts a vector in coding scan order witht he PVQ partitions
- * laid out one after another into a coefficient block in raster
- * order. This works in stages in the reverse order of raster->scan
- * order; the 16x16 conversion is applied to the coefficients that
- * don't appear in an 8x8 block, then the 8x8 applied to the 8x8 block
- * sans the 4x4 block it contains, then 4x4 is converted sans DC.
- *
- * @param [out]    dst        destination coefficient block
- * @param [in]     stride     destination vector row stride
- * @param [in]     src        source vector
- * @param [in]     n          block size (along one side)
- */
-void od_coding_order_to_raster(tran_low_t *dst, int stride, TX_TYPE ty_type,
- const tran_low_t *src, int n) {
-  int bs;
-  /* src + 1 because DC is not included for 4x4 blocks. */
-  od_raster_from_band(OD_LAYOUTS[0], dst, stride, ty_type, src + 1);
-  for (bs = 1; bs < OD_TXSIZES; bs++) {
-    int size;
-    int offset;
-    /* Length of block size > 4 */
-    size = 1 << (OD_LOG_BSIZE0 + bs);
-    /* Offset is the size of the previous block squared. */
-    offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
-    if (n >= size) {
-      /* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
-      od_raster_from_band(OD_LAYOUTS[bs], dst, stride, ty_type, src + offset);
-    }
-  }
-  dst[0] = src[0];
-}
-
-/** Perform a single stage of conversion from a coefficient block in
- * raster order into coding scan order
- *
- * @param [in]     layout  scan order specification
- * @param [out]    dst     destination vector
- * @param [in]     src     source coefficient block
- * @param [int]    int     source vector row stride
- */
-static void od_band_from_raster_16(const band_layout *layout, int16_t *dst,
- const int16_t *src, int stride) {
-  int i;
-  int len;
-  len = layout->band_offsets[layout->nb_bands];
-  for (i = 0; i < len; i++) {
-    dst[i] = src[layout->dst_table[DCT_DCT][i][1]*stride + layout->dst_table[DCT_DCT][i][0]];
-  }
-}
-
-/** Converts a coefficient block in raster order into a vector in
- * coding scan order with the PVQ partitions laid out one after
- * another.  This works in stages; the 4x4 conversion is applied to
- * the coefficients nearest DC, then the 8x8 applied to the 8x8 block
- * nearest DC that was not already coded by 4x4, then 16x16 following
- * the same pattern.
- *
- * @param [out]    dst        destination vector
- * @param [in]     n          block size (along one side)
- * @param [in]     src        source coefficient block
- * @param [in]     stride     source vector row stride
- */
-void od_raster_to_coding_order_16(int16_t *dst, int n, const int16_t *src,
- int stride) {
-  int bs;
-  /* dst + 1 because DC is not included for 4x4 blocks. */
-  od_band_from_raster_16(OD_LAYOUTS[0], dst + 1, src, stride);
-  for (bs = 1; bs < OD_TXSIZES; bs++) {
-    int size;
-    int offset;
-    /* Length of block size > 4. */
-    size = 1 << (OD_LOG_BSIZE0 + bs);
-    /* Offset is the size of the previous block squared. */
-    offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
-    if (n >= size) {
-      /* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
-      od_band_from_raster_16(OD_LAYOUTS[bs], dst + offset, src, stride);
-    }
-  }
-  dst[0] = src[0];
-}

third_party/aom/av1/common/partition.h

@@ -1,40 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#if !defined(_partition_H)
-# define _partition_H
-
-#include "av1/common/enums.h"
-#include "odintrin.h"
-
-typedef unsigned char index_pair[2];
-
-typedef struct {
-  const index_pair **const dst_table;
-  int size;
-  int nb_bands;
-  const int *const band_offsets;
-} band_layout;
-
-extern const int *const OD_BAND_OFFSETS[OD_TXSIZES + 1];
-
-void od_raster_to_coding_order(tran_low_t *dst, int n,  TX_TYPE ty_type,
- const tran_low_t *src, int stride);
-
-void od_coding_order_to_raster(tran_low_t *dst, int stride,  TX_TYPE ty_type,
- const tran_low_t *src, int n);
-
-void od_raster_to_coding_order_16(int16_t *dst, int n, const int16_t *src,
- int stride);
-
-#endif

third_party/aom/av1/common/ppc/cfl_ppc.c

@@ -0,0 +1,153 @@
+/*
+ * Copyright (c) 2017, 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 <altivec.h>
+
+#include "config/av1_rtcd.h"
+
+#include "av1/common/cfl.h"
+
+#define OFF_0 0
+#define OFF_1 16
+#define OFF_2 32
+#define OFF_3 48
+#define CFL_BUF_LINE_BYTES 64
+#define CFL_LINE_1 64
+#define CFL_LINE_2 128
+#define CFL_LINE_3 192
+
+typedef vector int8_t int8x16_t;
+typedef vector uint8_t uint8x16_t;
+typedef vector int16_t int16x8_t;
+typedef vector uint16_t uint16x8_t;
+typedef vector int32_t int32x4_t;
+typedef vector uint32_t uint32x4_t;
+typedef vector uint64_t uint64x2_t;
+
+static INLINE void subtract_average_vsx(int16_t *pred_buf, int width,
+                                        int height, int round_offset,
+                                        int num_pel_log2) {
+  const int16_t *end = pred_buf + height * CFL_BUF_LINE;
+  const int16_t *sum_buf = pred_buf;
+  const uint32x4_t div_shift = vec_splats((uint32_t)num_pel_log2);
+  const uint8x16_t mask_64 = { 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
+                               0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07 };
+  const uint8x16_t mask_32 = { 0x14, 0x15, 0x16, 0x17, 0x00, 0x01, 0x02, 0x03,
+                               0x1C, 0x1D, 0x1E, 0x1F, 0x08, 0x09, 0x0A, 0x0B };
+
+  int32x4_t sum_32x4_0 = { 0, 0, 0, round_offset };
+  int32x4_t sum_32x4_1 = { 0, 0, 0, 0 };
+  do {
+    sum_32x4_0 = vec_sum4s(vec_vsx_ld(OFF_0, sum_buf), sum_32x4_0);
+    sum_32x4_1 = vec_sum4s(vec_vsx_ld(OFF_0 + CFL_LINE_1, sum_buf), sum_32x4_1);
+    if (width >= 16) {
+      sum_32x4_0 = vec_sum4s(vec_vsx_ld(OFF_1, sum_buf), sum_32x4_0);
+      sum_32x4_1 =
+          vec_sum4s(vec_vsx_ld(OFF_1 + CFL_LINE_1, sum_buf), sum_32x4_1);
+    }
+    if (width == 32) {
+      sum_32x4_0 = vec_sum4s(vec_vsx_ld(OFF_2, sum_buf), sum_32x4_0);
+      sum_32x4_1 =
+          vec_sum4s(vec_vsx_ld(OFF_2 + CFL_LINE_1, sum_buf), sum_32x4_1);
+      sum_32x4_0 = vec_sum4s(vec_vsx_ld(OFF_3, sum_buf), sum_32x4_0);
+      sum_32x4_1 =
+          vec_sum4s(vec_vsx_ld(OFF_3 + CFL_LINE_1, sum_buf), sum_32x4_1);
+    }
+  } while ((sum_buf += (CFL_BUF_LINE * 2)) < end);
+  int32x4_t sum_32x4 = vec_add(sum_32x4_0, sum_32x4_1);
+
+  const int32x4_t perm_64 = vec_perm(sum_32x4, sum_32x4, mask_64);
+  sum_32x4 = vec_add(sum_32x4, perm_64);
+  const int32x4_t perm_32 = vec_perm(sum_32x4, sum_32x4, mask_32);
+  sum_32x4 = vec_add(sum_32x4, perm_32);
+  const int32x4_t avg = vec_sr(sum_32x4, div_shift);
+  const int16x8_t vec_avg = vec_pack(avg, avg);
+  do {
+    vec_vsx_st(vec_sub(vec_vsx_ld(OFF_0, pred_buf), vec_avg), OFF_0, pred_buf);
+    vec_vsx_st(vec_sub(vec_vsx_ld(OFF_0 + CFL_LINE_1, pred_buf), vec_avg),
+               OFF_0 + CFL_BUF_LINE_BYTES, pred_buf);
+    vec_vsx_st(vec_sub(vec_vsx_ld(OFF_0 + CFL_LINE_2, pred_buf), vec_avg),
+               OFF_0 + CFL_LINE_2, pred_buf);
+    vec_vsx_st(vec_sub(vec_vsx_ld(OFF_0 + CFL_LINE_3, pred_buf), vec_avg),
+               OFF_0 + CFL_LINE_3, pred_buf);
+    if (width >= 16) {
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_1, pred_buf), vec_avg), OFF_1,
+                 pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_1 + CFL_LINE_1, pred_buf), vec_avg),
+                 OFF_1 + CFL_LINE_1, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_1 + CFL_LINE_2, pred_buf), vec_avg),
+                 OFF_1 + CFL_LINE_2, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_1 + CFL_LINE_3, pred_buf), vec_avg),
+                 OFF_1 + CFL_LINE_3, pred_buf);
+    }
+    if (width == 32) {
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_2, pred_buf), vec_avg), OFF_2,
+                 pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_2 + CFL_LINE_1, pred_buf), vec_avg),
+                 OFF_2 + CFL_LINE_1, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_2 + CFL_LINE_2, pred_buf), vec_avg),
+                 OFF_2 + CFL_LINE_2, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_2 + CFL_LINE_3, pred_buf), vec_avg),
+                 OFF_2 + CFL_LINE_3, pred_buf);
+
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_3, pred_buf), vec_avg), OFF_3,
+                 pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_3 + CFL_LINE_1, pred_buf), vec_avg),
+                 OFF_3 + CFL_LINE_1, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_3 + CFL_LINE_2, pred_buf), vec_avg),
+                 OFF_3 + CFL_LINE_2, pred_buf);
+      vec_vsx_st(vec_sub(vec_vsx_ld(OFF_3 + CFL_LINE_3, pred_buf), vec_avg),
+                 OFF_3 + CFL_LINE_3, pred_buf);
+    }
+  } while ((pred_buf += CFL_BUF_LINE * 4) < end);
+}
+
+// Declare wrappers for VSX sizes
+CFL_SUB_AVG_X(vsx, 8, 4, 16, 5)
+CFL_SUB_AVG_X(vsx, 8, 8, 32, 6)
+CFL_SUB_AVG_X(vsx, 8, 16, 64, 7)
+CFL_SUB_AVG_X(vsx, 8, 32, 128, 8)
+CFL_SUB_AVG_X(vsx, 16, 4, 32, 6)
+CFL_SUB_AVG_X(vsx, 16, 8, 64, 7)
+CFL_SUB_AVG_X(vsx, 16, 16, 128, 8)
+CFL_SUB_AVG_X(vsx, 16, 32, 256, 9)
+CFL_SUB_AVG_X(vsx, 32, 8, 128, 8)
+CFL_SUB_AVG_X(vsx, 32, 16, 256, 9)
+CFL_SUB_AVG_X(vsx, 32, 32, 512, 10)
+
+// Based on observation, for small blocks VSX does not outperform C (no 64bit
+// load and store intrinsics). So we call the C code for block widths 4.
+cfl_subtract_average_fn get_subtract_average_fn_vsx(TX_SIZE tx_size) {
+  static const cfl_subtract_average_fn sub_avg[TX_SIZES_ALL] = {
+    subtract_average_4x4_c,     /* 4x4 */
+    subtract_average_8x8_vsx,   /* 8x8 */
+    subtract_average_16x16_vsx, /* 16x16 */
+    subtract_average_32x32_vsx, /* 32x32 */
+    cfl_subtract_average_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) */
+    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) */
+  };
+  // Modulo TX_SIZES_ALL to ensure that an attacker won't be able to
+  // index the function pointer array out of bounds.
+  return sub_avg[tx_size % TX_SIZES_ALL];
+}

third_party/aom/av1/common/pred_common.h

@@ -13,6 +13,7 @@
 #define AV1_COMMON_PRED_COMMON_H_
 
 #include "av1/common/blockd.h"
+#include "av1/common/mvref_common.h"
 #include "av1/common/onyxc_int.h"
 #include "aom_dsp/aom_dsp_common.h"
 
@@ -39,115 +40,174 @@ static INLINE int get_segment_id(const AV1_COMMON *const cm,
   return segment_id;
 }
 
+static INLINE int av1_get_spatial_seg_pred(const AV1_COMMON *const cm,
+                                           const MACROBLOCKD *const xd,
+                                           int mi_row, int mi_col,
+                                           int *cdf_index) {
+  int prev_ul = -1;  // top left segment_id
+  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);
+  }
+  if (xd->up_available) {
+    prev_u = get_segment_id(cm, cm->current_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);
+  }
+
+  // Pick CDF index based on number of matching/out-of-bounds segment IDs.
+  if (prev_ul < 0 || prev_u < 0 || prev_l < 0) /* Edge case */
+    *cdf_index = 0;
+  else if ((prev_ul == prev_u) && (prev_ul == prev_l))
+    *cdf_index = 2;
+  else if ((prev_ul == prev_u) || (prev_ul == prev_l) || (prev_u == prev_l))
+    *cdf_index = 1;
+  else
+    *cdf_index = 0;
+
+  // If 2 or more are identical returns that as predictor, otherwise prev_l.
+  if (prev_u == -1)  // edge case
+    return prev_l == -1 ? 0 : prev_l;
+  if (prev_l == -1)  // edge case
+    return prev_u;
+  return (prev_ul == prev_u) ? prev_u : prev_l;
+}
+
 static INLINE int av1_get_pred_context_seg_id(const MACROBLOCKD *xd) {
-  const MODE_INFO *const above_mi = xd->above_mi;
-  const MODE_INFO *const left_mi = xd->left_mi;
-  const int above_sip =
-      (above_mi != NULL) ? above_mi->mbmi.seg_id_predicted : 0;
-  const int left_sip = (left_mi != NULL) ? left_mi->mbmi.seg_id_predicted : 0;
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+  const int above_sip = (above_mi != NULL) ? above_mi->seg_id_predicted : 0;
+  const int left_sip = (left_mi != NULL) ? left_mi->seg_id_predicted : 0;
 
   return above_sip + left_sip;
 }
 
-static INLINE aom_prob av1_get_pred_prob_seg_id(
-    const struct segmentation_probs *segp, const MACROBLOCKD *xd) {
-  return segp->pred_probs[av1_get_pred_context_seg_id(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;
+  int bck_frame_index = 0, fwd_frame_index = 0;
+  int cur_frame_index = cm->cur_frame->cur_frame_offset;
+
+  if (bck_idx >= 0)
+    bck_frame_index = cm->buffer_pool->frame_bufs[bck_idx].cur_frame_offset;
+
+  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));
+
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+
+  int above_ctx = 0, left_ctx = 0;
+  const int offset = (fwd == bck);
+
+  if (above_mi) {
+    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 (has_second_ref(left_mi))
+      left_ctx = left_mi->compound_idx;
+    else if (left_mi->ref_frame[0] == ALTREF_FRAME)
+      left_ctx = 1;
+  }
+
+  return above_ctx + left_ctx + 3 * offset;
+}
+
+static INLINE int get_comp_group_idx_context(const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+  int above_ctx = 0, left_ctx = 0;
+
+  if (above_mi) {
+    if (has_second_ref(above_mi))
+      above_ctx = above_mi->comp_group_idx;
+    else if (above_mi->ref_frame[0] == ALTREF_FRAME)
+      above_ctx = 3;
+  }
+  if (left_mi) {
+    if (has_second_ref(left_mi))
+      left_ctx = left_mi->comp_group_idx;
+    else if (left_mi->ref_frame[0] == ALTREF_FRAME)
+      left_ctx = 3;
+  }
+
+  return AOMMIN(5, above_ctx + left_ctx);
 }
 
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_seg_id(
     struct segmentation_probs *segp, const MACROBLOCKD *xd) {
   return segp->pred_cdf[av1_get_pred_context_seg_id(xd)];
 }
-#endif
+
+static INLINE int av1_get_skip_mode_context(const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+  const int above_skip_mode = above_mi ? above_mi->skip_mode : 0;
+  const int left_skip_mode = left_mi ? left_mi->skip_mode : 0;
+  return above_skip_mode + left_skip_mode;
+}
 
 static INLINE int av1_get_skip_context(const MACROBLOCKD *xd) {
-  const MODE_INFO *const above_mi = xd->above_mi;
-  const MODE_INFO *const left_mi = xd->left_mi;
-  const int above_skip = (above_mi != NULL) ? above_mi->mbmi.skip : 0;
-  const int left_skip = (left_mi != NULL) ? left_mi->mbmi.skip : 0;
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+  const int above_skip = above_mi ? above_mi->skip : 0;
+  const int left_skip = left_mi ? left_mi->skip : 0;
   return above_skip + left_skip;
 }
 
-static INLINE aom_prob av1_get_skip_prob(const AV1_COMMON *cm,
-                                         const MACROBLOCKD *xd) {
-  return cm->fc->skip_probs[av1_get_skip_context(xd)];
-}
-
-#if CONFIG_DUAL_FILTER
 int av1_get_pred_context_switchable_interp(const MACROBLOCKD *xd, int dir);
-#else
-int av1_get_pred_context_switchable_interp(const MACROBLOCKD *xd);
-#endif
 
-#if CONFIG_EXT_INTRA
-#if CONFIG_INTRA_INTERP
-int av1_get_pred_context_intra_interp(const MACROBLOCKD *xd);
-#endif  // CONFIG_INTRA_INTERP
-#endif  // CONFIG_EXT_INTRA
-
-#if CONFIG_PALETTE_DELTA_ENCODING
 // Get a list of palette base colors that are used in the above and left blocks,
 // referred to as "color cache". The return value is the number of colors in the
 // cache (<= 2 * PALETTE_MAX_SIZE). The color values are stored in "cache"
 // in ascending order.
 int av1_get_palette_cache(const MACROBLOCKD *const xd, int plane,
                           uint16_t *cache);
-#endif  // CONFIG_PALETTE_DELTA_ENCODING
+
+static INLINE int av1_get_palette_bsize_ctx(BLOCK_SIZE bsize) {
+  return num_pels_log2_lookup[bsize] - num_pels_log2_lookup[BLOCK_8X8];
+}
+
+static INLINE int av1_get_palette_mode_ctx(const MACROBLOCKD *xd) {
+  const MB_MODE_INFO *const above_mi = xd->above_mbmi;
+  const MB_MODE_INFO *const left_mi = xd->left_mbmi;
+  int ctx = 0;
+  if (above_mi) ctx += (above_mi->palette_mode_info.palette_size[0] > 0);
+  if (left_mi) ctx += (left_mi->palette_mode_info.palette_size[0] > 0);
+  return ctx;
+}
 
 int av1_get_intra_inter_context(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_intra_inter_prob(const AV1_COMMON *cm,
-                                                const MACROBLOCKD *xd) {
-  return cm->fc->intra_inter_prob[av1_get_intra_inter_context(xd)];
+int av1_get_reference_mode_context(const MACROBLOCKD *xd);
+
+static INLINE aom_cdf_prob *av1_get_reference_mode_cdf(const MACROBLOCKD *xd) {
+  return xd->tile_ctx->comp_inter_cdf[av1_get_reference_mode_context(xd)];
 }
 
-int av1_get_reference_mode_context(const AV1_COMMON *cm, const MACROBLOCKD *xd);
-
-static INLINE aom_prob av1_get_reference_mode_prob(const AV1_COMMON *cm,
-                                                   const MACROBLOCKD *xd) {
-  return cm->fc->comp_inter_prob[av1_get_reference_mode_context(cm, xd)];
-}
-#if CONFIG_NEW_MULTISYMBOL
-static INLINE aom_cdf_prob *av1_get_reference_mode_cdf(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return xd->tile_ctx->comp_inter_cdf[av1_get_reference_mode_context(cm, xd)];
-}
-#endif
-
-#if CONFIG_EXT_COMP_REFS
 int av1_get_comp_reference_type_context(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_comp_reference_type_prob(const AV1_COMMON *cm,
-                                                        const MACROBLOCKD *xd) {
-  return cm->fc->comp_ref_type_prob[av1_get_comp_reference_type_context(xd)];
-}
+// == Uni-directional contexts ==
 
 int av1_get_pred_context_uni_comp_ref_p(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_uni_comp_ref_p(const AV1_COMMON *cm,
-                                                        const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_uni_comp_ref_p(xd);
-  return cm->fc->uni_comp_ref_prob[pred_context][0];
-}
-
 int av1_get_pred_context_uni_comp_ref_p1(const MACROBLOCKD *xd);
 
-static INLINE aom_prob
-av1_get_pred_prob_uni_comp_ref_p1(const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_uni_comp_ref_p1(xd);
-  return cm->fc->uni_comp_ref_prob[pred_context][1];
-}
-
 int av1_get_pred_context_uni_comp_ref_p2(const MACROBLOCKD *xd);
 
-static INLINE aom_prob
-av1_get_pred_prob_uni_comp_ref_p2(const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_uni_comp_ref_p2(xd);
-  return cm->fc->uni_comp_ref_prob[pred_context][2];
-}
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_comp_reference_type_cdf(
     const MACROBLOCKD *xd) {
   const int pred_context = av1_get_comp_reference_type_context(xd);
@@ -171,211 +231,126 @@ static INLINE aom_cdf_prob *av1_get_pred_cdf_uni_comp_ref_p2(
   const int pred_context = av1_get_pred_context_uni_comp_ref_p2(xd);
   return xd->tile_ctx->uni_comp_ref_cdf[pred_context][2];
 }
-#endif  // CONFIG_NEW_MULTISYMBOL
-#endif  // CONFIG_EXT_COMP_REFS
 
-int av1_get_pred_context_comp_ref_p(const AV1_COMMON *cm,
-                                    const MACROBLOCKD *xd);
+// == Bi-directional contexts ==
 
-#if CONFIG_NEW_MULTISYMBOL
-static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_ref_p(const AV1_COMMON *cm,
-                                                        const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p(cm, xd);
+int av1_get_pred_context_comp_ref_p(const MACROBLOCKD *xd);
+
+int av1_get_pred_context_comp_ref_p1(const MACROBLOCKD *xd);
+
+int av1_get_pred_context_comp_ref_p2(const MACROBLOCKD *xd);
+
+int av1_get_pred_context_comp_bwdref_p(const MACROBLOCKD *xd);
+
+int av1_get_pred_context_comp_bwdref_p1(const MACROBLOCKD *xd);
+
+static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_ref_p(const MACROBLOCKD *xd) {
+  const int pred_context = av1_get_pred_context_comp_ref_p(xd);
   return xd->tile_ctx->comp_ref_cdf[pred_context][0];
 }
-#endif
 
-static INLINE aom_prob av1_get_pred_prob_comp_ref_p(const AV1_COMMON *cm,
-                                                    const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p(cm, xd);
-  return cm->fc->comp_ref_prob[pred_context][0];
-}
-
-#if CONFIG_EXT_REFS
-int av1_get_pred_context_comp_ref_p1(const AV1_COMMON *cm,
-                                     const MACROBLOCKD *xd);
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_ref_p1(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p1(cm, xd);
+    const MACROBLOCKD *xd) {
+  const int pred_context = av1_get_pred_context_comp_ref_p1(xd);
   return xd->tile_ctx->comp_ref_cdf[pred_context][1];
 }
-#endif  // CONFIG_NEW_MULTISYMBOL
 
-static INLINE aom_prob av1_get_pred_prob_comp_ref_p1(const AV1_COMMON *cm,
-                                                     const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p1(cm, xd);
-  return cm->fc->comp_ref_prob[pred_context][1];
-}
-
-int av1_get_pred_context_comp_ref_p2(const AV1_COMMON *cm,
-                                     const MACROBLOCKD *xd);
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_ref_p2(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p2(cm, xd);
+    const MACROBLOCKD *xd) {
+  const int pred_context = av1_get_pred_context_comp_ref_p2(xd);
   return xd->tile_ctx->comp_ref_cdf[pred_context][2];
 }
-#endif  // CONFIG_NEW_MULTISYMBOL
 
-static INLINE aom_prob av1_get_pred_prob_comp_ref_p2(const AV1_COMMON *cm,
-                                                     const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_ref_p2(cm, xd);
-  return cm->fc->comp_ref_prob[pred_context][2];
-}
-
-int av1_get_pred_context_comp_bwdref_p(const AV1_COMMON *cm,
-                                       const MACROBLOCKD *xd);
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_bwdref_p(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_bwdref_p(cm, xd);
+    const MACROBLOCKD *xd) {
+  const int pred_context = av1_get_pred_context_comp_bwdref_p(xd);
   return xd->tile_ctx->comp_bwdref_cdf[pred_context][0];
 }
-#endif  // CONFIG_NEW_MULTISYMBOL
 
-static INLINE aom_prob av1_get_pred_prob_comp_bwdref_p(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_bwdref_p(cm, xd);
-  return cm->fc->comp_bwdref_prob[pred_context][0];
-}
-
-int av1_get_pred_context_comp_bwdref_p1(const AV1_COMMON *cm,
-                                        const MACROBLOCKD *xd);
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_comp_bwdref_p1(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_bwdref_p1(cm, xd);
+    const MACROBLOCKD *xd) {
+  const int pred_context = av1_get_pred_context_comp_bwdref_p1(xd);
   return xd->tile_ctx->comp_bwdref_cdf[pred_context][1];
 }
-#endif  // CONFIG_NEW_MULTISYMBOL
 
-static INLINE aom_prob av1_get_pred_prob_comp_bwdref_p1(const AV1_COMMON *cm,
-                                                        const MACROBLOCKD *xd) {
-  const int pred_context = av1_get_pred_context_comp_bwdref_p1(cm, xd);
-  return cm->fc->comp_bwdref_prob[pred_context][1];
-}
-#endif  // CONFIG_EXT_REFS
+// == Single contexts ==
 
 int av1_get_pred_context_single_ref_p1(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p1(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p1(xd)][0];
-}
-
 int av1_get_pred_context_single_ref_p2(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p2(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p2(xd)][1];
-}
-
-#if CONFIG_EXT_REFS
 int av1_get_pred_context_single_ref_p3(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p3(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p3(xd)][2];
-}
-
 int av1_get_pred_context_single_ref_p4(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p4(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p4(xd)][3];
-}
-
 int av1_get_pred_context_single_ref_p5(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p5(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p5(xd)][4];
-}
-
 int av1_get_pred_context_single_ref_p6(const MACROBLOCKD *xd);
 
-static INLINE aom_prob av1_get_pred_prob_single_ref_p6(const AV1_COMMON *cm,
-                                                       const MACROBLOCKD *xd) {
-  return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p6(xd)][5];
-}
-#endif  // CONFIG_EXT_REFS
-
-#if CONFIG_NEW_MULTISYMBOL
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p1(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p1(xd)][0];
 }
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p2(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p2(xd)][1];
 }
-#if CONFIG_EXT_REFS
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p3(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p3(xd)][2];
 }
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p4(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p4(xd)][3];
 }
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p5(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p5(xd)][4];
 }
 static INLINE aom_cdf_prob *av1_get_pred_cdf_single_ref_p6(
-    const AV1_COMMON *cm, const MACROBLOCKD *xd) {
-  (void)cm;
+    const MACROBLOCKD *xd) {
   return xd->tile_ctx
       ->single_ref_cdf[av1_get_pred_context_single_ref_p6(xd)][5];
 }
-#endif  // CONFIG_EXT_REFS
-#endif  // CONFIG_NEW_MULTISYMBOL
-
-#if CONFIG_COMPOUND_SINGLEREF
-int av1_get_inter_mode_context(const MACROBLOCKD *xd);
-
-static INLINE aom_prob av1_get_inter_mode_prob(const AV1_COMMON *cm,
-                                               const MACROBLOCKD *xd) {
-  return cm->fc->comp_inter_mode_prob[av1_get_inter_mode_context(xd)];
-}
-#endif  // CONFIG_COMPOUND_SINGLEREF
 
 // Returns a context number for the given MB prediction signal
 // The mode info data structure has a one element border above and to the
 // left of the entries corresponding to real blocks.
 // The prediction flags in these dummy entries are initialized to 0.
 static INLINE int get_tx_size_context(const MACROBLOCKD *xd) {
-  const int max_tx_size = max_txsize_lookup[xd->mi[0]->mbmi.sb_type];
+  const MB_MODE_INFO *mbmi = xd->mi[0];
   const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
   const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
+  const TX_SIZE max_tx_size = max_txsize_rect_lookup[mbmi->sb_type];
+  const int max_tx_wide = tx_size_wide[max_tx_size];
+  const int max_tx_high = tx_size_high[max_tx_size];
   const int has_above = xd->up_available;
   const int has_left = xd->left_available;
-  int above_ctx = (has_above && !above_mbmi->skip)
-                      ? (int)txsize_sqr_map[above_mbmi->tx_size]
-                      : max_tx_size;
-  int left_ctx = (has_left && !left_mbmi->skip)
-                     ? (int)txsize_sqr_map[left_mbmi->tx_size]
-                     : max_tx_size;
 
-  if (!has_left) left_ctx = above_ctx;
+  int above = xd->above_txfm_context[0] >= max_tx_wide;
+  int left = xd->left_txfm_context[0] >= max_tx_high;
 
-  if (!has_above) above_ctx = left_ctx;
-  return (above_ctx + left_ctx) > max_tx_size + TX_SIZE_LUMA_MIN;
+  if (has_above)
+    if (is_inter_block(above_mbmi))
+      above = block_size_wide[above_mbmi->sb_type] >= max_tx_wide;
+
+  if (has_left)
+    if (is_inter_block(left_mbmi))
+      left = block_size_high[left_mbmi->sb_type] >= max_tx_high;
+
+  if (has_above && has_left)
+    return (above + left);
+  else if (has_above)
+    return above;
+  else if (has_left)
+    return left;
+  else
+    return 0;
 }
 
 #ifdef __cplusplus

third_party/aom/av1/common/pvq.h

@@ -1,179 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#if !defined(_pvq_H)
-# define _pvq_H (1)
-# include "generic_code.h"
-# include "odintrin.h"
-
-extern const uint16_t EXP_CDF_TABLE[][16];
-extern const uint16_t LAPLACE_OFFSET[];
-
-#define AV1_PVQ_ENABLE_ACTIVITY_MASKING (0)
-
-# define PVQ_MAX_PARTITIONS (1 + 3*(OD_TXSIZES-1))
-
-# define OD_NOREF_ADAPT_SPEED (4)
-/* Normalized lambda for PVQ quantizer. Since we normalize the gain by q, the
-   distortion is normalized by q^2 and lambda does not need the q^2 factor.
-   At high rate, this would be log(2)/6, but we're using a slightly more
-   aggressive value, closer to:
-   Li, Xiang, et al. "Laplace distribution based Lagrangian rate distortion
-   optimization for hybrid video coding." Circuits and Systems for Video
-   Technology, IEEE Transactions on 19.2 (2009): 193-205.
-   */
-# define OD_PVQ_LAMBDA (.1146)
-
-#define OD_PVQ_SKIP_ZERO 1
-#define OD_PVQ_SKIP_COPY 2
-
-/* Maximum size for coding a PVQ band. */
-#define OD_MAX_PVQ_SIZE (1024)
-
-#if defined(OD_FLOAT_PVQ)
-#define OD_QM_SHIFT (15)
-#else
-#define OD_QM_SHIFT (11)
-#endif
-#define OD_QM_SCALE (1 << OD_QM_SHIFT)
-#if defined(OD_FLOAT_PVQ)
-#define OD_QM_SCALE_1 (1./OD_QM_SCALE)
-#endif
-#define OD_QM_SCALE_MAX 32767
-#define OD_QM_INV_SHIFT (12)
-#define OD_QM_INV_SCALE (1 << OD_QM_INV_SHIFT)
-#if defined(OD_FLOAT_PVQ)
-#define OD_QM_INV_SCALE_1 (1./OD_QM_INV_SCALE)
-#endif
-#define OD_QM_OFFSET(bs) ((((1 << 2*bs) - 1) << 2*OD_LOG_BSIZE0)/3)
-#define OD_QM_STRIDE (OD_QM_OFFSET(OD_TXSIZES))
-#define OD_QM_BUFFER_SIZE (2*OD_QM_STRIDE)
-
-#if !defined(OD_FLOAT_PVQ)
-#define OD_THETA_SHIFT (15)
-#define OD_THETA_SCALE ((1 << OD_THETA_SHIFT)*2./M_PI)
-#define OD_MAX_THETA_SCALE (1 << OD_THETA_SHIFT)
-#define OD_TRIG_SCALE (32768)
-#define OD_BETA_SHIFT (12)
-#define OD_BETA_SCALE_1 (1./(1 << OD_BETA_SHIFT))
-/*Multiplies 16-bit a by 32-bit b and keeps bits [16:64-OD_BETA_SHIFT-1].*/
-#define OD_MULT16_32_QBETA(a, b) \
- ((int16_t)(a)*(int64_t)(int32_t)(b) >> OD_BETA_SHIFT)
-# define OD_MULT16_16_QBETA(a, b) \
-  ((((int16_t)(a))*((int32_t)(int16_t)(b))) >> OD_BETA_SHIFT)
-#define OD_CGAIN_SHIFT (8)
-#define OD_CGAIN_SCALE (1 << OD_CGAIN_SHIFT)
-#else
-#define OD_BETA_SCALE_1 (1.)
-#define OD_THETA_SCALE (1)
-#define OD_TRIG_SCALE (1)
-#define OD_CGAIN_SCALE (1)
-#endif
-#define OD_THETA_SCALE_1 (1./OD_THETA_SCALE)
-#define OD_TRIG_SCALE_1 (1./OD_TRIG_SCALE)
-#define OD_CGAIN_SCALE_1 (1./OD_CGAIN_SCALE)
-#define OD_CGAIN_SCALE_2 (OD_CGAIN_SCALE_1*OD_CGAIN_SCALE_1)
-
-/* Largest PVQ partition is half the coefficients of largest block size. */
-#define MAXN (OD_TXSIZE_MAX*OD_TXSIZE_MAX/2)
-
-#define OD_COMPAND_SHIFT (8 + OD_COEFF_SHIFT)
-#define OD_COMPAND_SCALE (1 << OD_COMPAND_SHIFT)
-#define OD_COMPAND_SCALE_1 (1./OD_COMPAND_SCALE)
-
-#define OD_QM_SIZE (OD_TXSIZES*(OD_TXSIZES + 1))
-
-#define OD_FLAT_QM 0
-#define OD_HVS_QM  1
-
-# define OD_NSB_ADAPT_CTXS (4)
-
-# define OD_ADAPT_K_Q8        0
-# define OD_ADAPT_SUM_EX_Q8   1
-# define OD_ADAPT_COUNT_Q8    2
-# define OD_ADAPT_COUNT_EX_Q8 3
-
-# define OD_ADAPT_NO_VALUE (-2147483647-1)
-
-typedef enum {
-  PVQ_SKIP = 0x0,
-  DC_CODED = 0x1,
-  AC_CODED = 0x2,
-  AC_DC_CODED = 0x3,
-} PVQ_SKIP_TYPE;
-
-typedef struct od_pvq_adapt_ctx  od_pvq_adapt_ctx;
-typedef struct od_pvq_codeword_ctx od_pvq_codeword_ctx;
-
-struct od_pvq_codeword_ctx {
-  int                 pvq_adapt[2*OD_TXSIZES*OD_NSB_ADAPT_CTXS];
-  /* CDFs are size 16 despite the fact that we're using less than that. */
-  uint16_t            pvq_k1_cdf[12][CDF_SIZE(16)];
-  uint16_t            pvq_split_cdf[22*7][CDF_SIZE(8)];
-};
-
-struct od_pvq_adapt_ctx {
-  od_pvq_codeword_ctx pvq_codeword_ctx;
-  generic_encoder     pvq_param_model[3];
-  int                 pvq_ext[OD_TXSIZES*PVQ_MAX_PARTITIONS];
-  int                 pvq_exg[OD_NPLANES_MAX][OD_TXSIZES][PVQ_MAX_PARTITIONS];
-  uint16_t pvq_gaintheta_cdf[2*OD_TXSIZES*PVQ_MAX_PARTITIONS][CDF_SIZE(16)];
-  uint16_t pvq_skip_dir_cdf[2*(OD_TXSIZES-1)][CDF_SIZE(7)];
-};
-
-typedef struct od_qm_entry {
-  int interp_q;
-  int scale_q8;
-  const unsigned char *qm_q4;
-} od_qm_entry;
-
-extern const od_qm_entry OD_DEFAULT_QMS[2][2][OD_NPLANES_MAX];
-
-void od_adapt_pvq_ctx_reset(od_pvq_adapt_ctx *state, int is_keyframe);
-int od_pvq_size_ctx(int n);
-int od_pvq_k1_ctx(int n, int orig_size);
-
-od_val16 od_pvq_sin(od_val32 x);
-od_val16 od_pvq_cos(od_val32 x);
-#if !defined(OD_FLOAT_PVQ)
-int od_vector_log_mag(const od_coeff *x, int n);
-#endif
-
-void od_interp_qm(unsigned char *out, int q, const od_qm_entry *entry1,
-                  const od_qm_entry *entry2);
-
-int od_qm_get_index(int bs, int band);
-
-extern const od_val16 *const OD_PVQ_BETA[2][OD_NPLANES_MAX][OD_TXSIZES + 1];
-
-void od_init_qm(int16_t *x, int16_t *x_inv, const int *qm);
-int od_compute_householder(od_val16 *r, int n, od_val32 gr, int *sign,
- int shift);
-void od_apply_householder(od_val16 *out, const od_val16 *x, const od_val16 *r,
- int n);
-void od_pvq_synthesis_partial(od_coeff *xcoeff, const od_coeff *ypulse,
-                                  const od_val16 *r, int n,
-                                  int noref, od_val32 g,
-                                  od_val32 theta, int m, int s,
-                                  const int16_t *qm_inv);
-od_val32 od_gain_expand(od_val32 cg, int q0, od_val16 beta);
-od_val32 od_pvq_compute_gain(const od_val16 *x, int n, int q0, od_val32 *g,
- od_val16 beta, int bshift);
-int od_pvq_compute_max_theta(od_val32 qcg, od_val16 beta);
-od_val32 od_pvq_compute_theta(int t, int max_theta);
-int od_pvq_compute_k(od_val32 qcg, int itheta, int noref, int n, od_val16 beta);
-
-int od_vector_is_null(const od_coeff *x, int len);
-int od_qm_offset(int bs, int xydec);
-
-#endif

third_party/aom/av1/common/pvq_state.c

@@ -1,50 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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 "av1/common/pvq_state.h"
-#include "av1/common/odintrin.h"
-
-void od_adapt_ctx_reset(od_adapt_ctx *adapt, int is_keyframe) {
-  int pli;
-  od_adapt_pvq_ctx_reset(&adapt->pvq, is_keyframe);
-  OD_CDFS_INIT_Q15(adapt->skip_cdf);
-  for (pli = 0; pli < OD_NPLANES_MAX; pli++) {
-    int i;
-    OD_CDFS_INIT_DYNAMIC(adapt->model_dc[pli].cdf);
-    for (i = 0; i < OD_TXSIZES; i++) {
-      int j;
-      adapt->ex_g[pli][i] = 8;
-      for (j = 0; j < 3; j++) {
-        adapt->ex_dc[pli][i][j] = pli > 0 ? 8 : 32768;
-      }
-    }
-  }
-}
-
-void od_init_skipped_coeffs(int16_t *d, int16_t *pred, int is_keyframe, int bo,
-                            int n, int w) {
-  int i;
-  int j;
-  if (is_keyframe) {
-    for (i = 0; i < n; i++) {
-      for (j = 0; j < n; j++) {
-        /* skip DC */
-        if (i || j) d[bo + i * w + j] = 0;
-      }
-    }
-  } else {
-    for (i = 0; i < n; i++) {
-      for (j = 0; j < n; j++) {
-        d[bo + i * w + j] = pred[i * n + j];
-      }
-    }
-  }
-}

third_party/aom/av1/common/pvq_state.h

@@ -1,52 +0,0 @@
-/*
- * Copyright (c) 2001-2016, 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.
- */
-
-/* clang-format off */
-
-#if !defined(_state_H)
-# define _state_H (1)
-
-typedef struct od_state     od_state;
-typedef struct od_adapt_ctx od_adapt_ctx;
-
-# include "generic_code.h"
-# include "odintrin.h"
-# include "pvq.h"
-
-/*Adaptation speed of scalar Laplace encoding.*/
-# define OD_SCALAR_ADAPT_SPEED (4)
-
-struct od_adapt_ctx {
-  /* Support for PVQ encode/decode */
-  od_pvq_adapt_ctx pvq;
-
-  generic_encoder model_dc[OD_NPLANES_MAX];
-
-  int ex_dc[OD_NPLANES_MAX][OD_TXSIZES][3];
-  int ex_g[OD_NPLANES_MAX][OD_TXSIZES];
-
-  /* Joint skip flag for DC and AC */
-  uint16_t skip_cdf[OD_TXSIZES*2][CDF_SIZE(4)];
-};
-
-struct od_state {
-  od_adapt_ctx *adapt;
-  unsigned char pvq_qm_q4[OD_NPLANES_MAX][OD_QM_SIZE];
-  /* Quantization matrices and their inverses. */
-  int16_t qm[OD_QM_BUFFER_SIZE];
-  int16_t qm_inv[OD_QM_BUFFER_SIZE];
-};
-
-void od_adapt_ctx_reset(od_adapt_ctx *state, int is_keyframe);
-void od_init_skipped_coeffs(int16_t *d, int16_t *pred, int is_keyframe,
- int bo, int n, int w);
-
-#endif

third_party/aom/av1/common/quant_common.h

@@ -25,82 +25,37 @@ extern "C" {
 #define MAXQ 255
 #define QINDEX_RANGE (MAXQ - MINQ + 1)
 #define QINDEX_BITS 8
-#if CONFIG_AOM_QM
 // Total number of QM sets stored
 #define QM_LEVEL_BITS 4
 #define NUM_QM_LEVELS (1 << QM_LEVEL_BITS)
 /* Range of QMS is between first and last value, with offset applied to inter
  * blocks*/
+#define DEFAULT_QM_Y 10
+#define DEFAULT_QM_U 11
+#define DEFAULT_QM_V 12
 #define DEFAULT_QM_FIRST 5
 #define DEFAULT_QM_LAST 9
-#define DEFAULT_QM_INTER_OFFSET 0
-#endif
 
 struct AV1Common;
 
-int16_t av1_dc_quant(int qindex, int delta, aom_bit_depth_t bit_depth);
-int16_t av1_ac_quant(int qindex, int delta, aom_bit_depth_t bit_depth);
-int16_t av1_qindex_from_ac(int ac, aom_bit_depth_t bit_depth);
+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);
+int16_t av1_dc_quant_QTX(int qindex, int delta, aom_bit_depth_t bit_depth);
+int16_t av1_ac_quant_QTX(int qindex, int delta, aom_bit_depth_t bit_depth);
+int16_t av1_qindex_from_ac_Q3(int ac_Q3, aom_bit_depth_t bit_depth);
 
 int av1_get_qindex(const struct segmentation *seg, int segment_id,
                    int base_qindex);
-#if CONFIG_AOM_QM
 // Reduce the large number of quantizers to a smaller number of levels for which
 // different matrices may be defined
 static INLINE int aom_get_qmlevel(int qindex, int first, int last) {
   return first + (qindex * (last + 1 - first)) / QINDEX_RANGE;
 }
-void aom_qm_init(struct AV1Common *cm);
-qm_val_t *aom_iqmatrix(struct AV1Common *cm, int qindex, int comp,
-                       TX_SIZE tx_size, int is_intra);
-qm_val_t *aom_qmatrix(struct AV1Common *cm, int qindex, int comp,
-                      TX_SIZE tx_size, int is_intra);
-#endif
-
-#if CONFIG_NEW_QUANT
-
-#define QUANT_PROFILES 4
-#define QUANT_RANGES 2
-#define NUQ_KNOTS 3
-
-typedef tran_low_t dequant_val_type_nuq[NUQ_KNOTS + 1];
-typedef tran_low_t cuml_bins_type_nuq[NUQ_KNOTS];
-void av1_get_dequant_val_nuq(int q, int band, tran_low_t *dq,
-                             tran_low_t *cuml_bins, int dq_off_index);
-tran_low_t av1_dequant_abscoeff_nuq(int v, int q, const tran_low_t *dq);
-tran_low_t av1_dequant_coeff_nuq(int v, int q, const tran_low_t *dq);
-
-static INLINE int qindex_to_qrange(int qindex) {
-  return (qindex < 140 ? 1 : 0);
-}
-
-static INLINE int get_dq_profile_from_ctx(int qindex, int q_ctx, int is_inter,
-                                          PLANE_TYPE plane_type) {
-  // intra/inter, Y/UV, ctx, qrange
-  static const int
-      def_dq_profile_lookup[REF_TYPES][PLANE_TYPES][COEFF_CONTEXTS0]
-                           [QUANT_RANGES] = {
-                             {
-                                 // intra
-                                 { { 2, 1 }, { 2, 1 }, { 2, 1 } },  // Y
-                                 { { 3, 1 }, { 3, 1 }, { 3, 1 } },  // UV
-                             },
-                             {
-                                 // inter
-                                 { { 3, 1 }, { 2, 1 }, { 2, 1 } },  // Y
-                                 { { 3, 1 }, { 3, 1 }, { 3, 1 } },  // UV
-                             },
-                           };
-  if (!qindex) return 0;  // lossless
-  return def_dq_profile_lookup[is_inter][plane_type][q_ctx]
-                              [qindex_to_qrange(qindex)];
-}
-#endif  // CONFIG_NEW_QUANT
-
-#if CONFIG_PVQ
-extern const int OD_QM8_Q4_FLAT[];
-extern const int OD_QM8_Q4_HVS[];
-#endif
+void av1_qm_init(struct AV1Common *cm);
+const qm_val_t *av1_iqmatrix(struct AV1Common *cm, int qindex, int comp,
+                             TX_SIZE tx_size);
+const qm_val_t *av1_qmatrix(struct AV1Common *cm, int qindex, int comp,
+                            TX_SIZE tx_size);
 
 #ifdef __cplusplus
 }  // extern "C"

third_party/aom/av1/common/reconinter.h

@@ -15,164 +15,26 @@
 #include "av1/common/filter.h"
 #include "av1/common/onyxc_int.h"
 #include "av1/common/convolve.h"
-#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
 #include "av1/common/warped_motion.h"
-#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
 #include "aom/aom_integer.h"
 
-#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-#define WARP_WM_NEIGHBORS_WITH_OBMC 0
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
-
-#if CONFIG_MOTION_VAR && CONFIG_GLOBAL_MOTION
-#define WARP_GM_NEIGHBORS_WITH_OBMC 0
-#endif  // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
+// Work out how many pixels off the edge of a reference frame we're allowed
+// to go when forming an inter prediction.
+// The outermost row/col of each referernce frame is extended by
+// (AOM_BORDER_IN_PIXELS >> subsampling) pixels, but we need to keep
+// at least AOM_INTERP_EXTEND pixels within that to account for filtering.
+//
+// We have to break this up into two macros to keep both clang-format and
+// tools/lint-hunks.py happy.
+#define AOM_LEFT_TOP_MARGIN_PX(subsampling) \
+  ((AOM_BORDER_IN_PIXELS >> subsampling) - AOM_INTERP_EXTEND)
+#define AOM_LEFT_TOP_MARGIN_SCALED(subsampling) \
+  (AOM_LEFT_TOP_MARGIN_PX(subsampling) << SCALE_SUBPEL_BITS)
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-static INLINE int has_scale(int xs, int ys) {
-  return xs != SCALE_SUBPEL_SHIFTS || ys != SCALE_SUBPEL_SHIFTS;
-}
-
-static INLINE void inter_predictor(const uint8_t *src, int src_stride,
-                                   uint8_t *dst, int dst_stride, int subpel_x,
-                                   int subpel_y, const struct scale_factors *sf,
-                                   int w, int h, ConvolveParams *conv_params,
-                                   InterpFilters interp_filters, int xs,
-                                   int ys) {
-  assert(conv_params->do_average == 0 || conv_params->do_average == 1);
-  assert(sf);
-  if (has_scale(xs, ys)) {
-    // TODO(afergs, debargha): Use a different scale convolve function
-    // that uses higher precision for subpel_x, subpel_y, xs, ys
-    if (conv_params->round == CONVOLVE_OPT_NO_ROUND) {
-#if CONFIG_CONVOLVE_ROUND
-      av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
-                             interp_filters, subpel_x, xs, subpel_y, ys, 1,
-                             conv_params);
-      conv_params->do_post_rounding = 1;
-#else
-      assert(0);
-#endif  // CONFIG_CONVOLVE_ROUND
-    } else {
-      assert(conv_params->round == CONVOLVE_OPT_ROUND);
-      av1_convolve_scale(src, src_stride, dst, dst_stride, w, h, interp_filters,
-                         subpel_x, xs, subpel_y, ys, conv_params);
-    }
-  } else {
-    subpel_x >>= SCALE_EXTRA_BITS;
-    subpel_y >>= SCALE_EXTRA_BITS;
-    xs >>= SCALE_EXTRA_BITS;
-    ys >>= SCALE_EXTRA_BITS;
-    assert(subpel_x < SUBPEL_SHIFTS);
-    assert(subpel_y < SUBPEL_SHIFTS);
-    assert(xs <= SUBPEL_SHIFTS);
-    assert(ys <= SUBPEL_SHIFTS);
-    if (conv_params->round == CONVOLVE_OPT_NO_ROUND) {
-#if CONFIG_CONVOLVE_ROUND
-      av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
-                             interp_filters, subpel_x, xs, subpel_y, ys, 0,
-                             conv_params);
-      conv_params->do_post_rounding = 1;
-#else
-      assert(0);
-#endif  // CONFIG_CONVOLVE_ROUND
-    } else {
-      assert(conv_params->round == CONVOLVE_OPT_ROUND);
-
-      InterpFilterParams filter_params_x, filter_params_y;
-      av1_get_convolve_filter_params(interp_filters, 0, &filter_params_x,
-                                     &filter_params_y);
-
-      if (w <= 2 || h <= 2) {
-        av1_convolve_c(src, src_stride, dst, dst_stride, w, h, interp_filters,
-                       subpel_x, xs, subpel_y, ys, conv_params);
-      } else if (filter_params_x.taps == SUBPEL_TAPS &&
-                 filter_params_y.taps == SUBPEL_TAPS) {
-        const int16_t *kernel_x =
-            av1_get_interp_filter_subpel_kernel(filter_params_x, subpel_x);
-        const int16_t *kernel_y =
-            av1_get_interp_filter_subpel_kernel(filter_params_y, subpel_y);
-        sf->predict[subpel_x != 0][subpel_y != 0][conv_params->do_average](
-            src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h);
-      } else {
-        av1_convolve(src, src_stride, dst, dst_stride, w, h, interp_filters,
-                     subpel_x, xs, subpel_y, ys, conv_params);
-      }
-    }
-  }
-}
-
-#if CONFIG_HIGHBITDEPTH
-static INLINE void highbd_inter_predictor(const uint8_t *src, int src_stride,
-                                          uint8_t *dst, int dst_stride,
-                                          int subpel_x, int subpel_y,
-                                          const struct scale_factors *sf, int w,
-                                          int h, ConvolveParams *conv_params,
-                                          InterpFilters interp_filters, int xs,
-                                          int ys, int bd) {
-  const int avg = conv_params->do_average;
-  assert(avg == 0 || avg == 1);
-
-  if (has_scale(xs, ys)) {
-    if (conv_params->round == CONVOLVE_OPT_NO_ROUND) {
-#if CONFIG_CONVOLVE_ROUND
-      av1_highbd_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
-                                    interp_filters, subpel_x, xs, subpel_y, ys,
-                                    1, conv_params, bd);
-      conv_params->do_post_rounding = 1;
-#else
-      assert(0);
-#endif  // CONFIG_CONVOLVE_ROUND
-    } else {
-      av1_highbd_convolve_scale(src, src_stride, dst, dst_stride, w, h,
-                                interp_filters, subpel_x, xs, subpel_y, ys, avg,
-                                bd);
-    }
-  } else {
-    subpel_x >>= SCALE_EXTRA_BITS;
-    subpel_y >>= SCALE_EXTRA_BITS;
-    xs >>= SCALE_EXTRA_BITS;
-    ys >>= SCALE_EXTRA_BITS;
-    assert(subpel_x < SUBPEL_SHIFTS);
-    assert(subpel_y < SUBPEL_SHIFTS);
-    assert(xs <= SUBPEL_SHIFTS);
-    assert(ys <= SUBPEL_SHIFTS);
-    if (conv_params->round == CONVOLVE_OPT_NO_ROUND) {
-#if CONFIG_CONVOLVE_ROUND
-      av1_highbd_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
-                                    interp_filters, subpel_x, xs, subpel_y, ys,
-                                    0, conv_params, bd);
-      conv_params->do_post_rounding = 1;
-#else
-      assert(0);
-#endif  // CONFIG_CONVOLVE_ROUND
-    } else {
-      InterpFilterParams filter_params_x, filter_params_y;
-      av1_get_convolve_filter_params(interp_filters, 0, &filter_params_x,
-                                     &filter_params_y);
-
-      if (filter_params_x.taps == SUBPEL_TAPS &&
-          filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2) {
-        const int16_t *kernel_x =
-            av1_get_interp_filter_subpel_kernel(filter_params_x, subpel_x);
-        const int16_t *kernel_y =
-            av1_get_interp_filter_subpel_kernel(filter_params_y, subpel_y);
-        sf->highbd_predict[subpel_x != 0][subpel_y != 0][avg](
-            src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h,
-            bd);
-      } else {
-        av1_highbd_convolve(src, src_stride, dst, dst_stride, w, h,
-                            interp_filters, subpel_x, xs, subpel_y, ys, avg,
-                            bd);
-      }
-    }
-  }
-}
-#endif  // CONFIG_HIGHBITDEPTH
-
 // Set to (1 << 5) if the 32-ary codebooks are used for any bock size
 #define MAX_WEDGE_TYPES (1 << 4)
 
@@ -208,38 +70,108 @@ typedef struct {
   int bits;
   const wedge_code_type *codebook;
   uint8_t *signflip;
-  int smoother;
   wedge_masks_type *masks;
 } wedge_params_type;
 
 extern const wedge_params_type wedge_params_lookup[BLOCK_SIZES_ALL];
 
+typedef struct SubpelParams {
+  int xs;
+  int ys;
+  int subpel_x;
+  int subpel_y;
+} SubpelParams;
+
+struct build_prediction_ctxt {
+  const AV1_COMMON *cm;
+  int mi_row;
+  int mi_col;
+  uint8_t **tmp_buf;
+  int *tmp_width;
+  int *tmp_height;
+  int *tmp_stride;
+  int mb_to_far_edge;
+};
+
+static INLINE int has_scale(int xs, int ys) {
+  return xs != SCALE_SUBPEL_SHIFTS || ys != SCALE_SUBPEL_SHIFTS;
+}
+
+static INLINE void revert_scale_extra_bits(SubpelParams *sp) {
+  sp->subpel_x >>= SCALE_EXTRA_BITS;
+  sp->subpel_y >>= SCALE_EXTRA_BITS;
+  sp->xs >>= SCALE_EXTRA_BITS;
+  sp->ys >>= SCALE_EXTRA_BITS;
+  assert(sp->subpel_x < SUBPEL_SHIFTS);
+  assert(sp->subpel_y < SUBPEL_SHIFTS);
+  assert(sp->xs <= SUBPEL_SHIFTS);
+  assert(sp->ys <= SUBPEL_SHIFTS);
+}
+
+static INLINE void inter_predictor(const uint8_t *src, int src_stride,
+                                   uint8_t *dst, int dst_stride,
+                                   const SubpelParams *subpel_params,
+                                   const struct scale_factors *sf, int w, int h,
+                                   ConvolveParams *conv_params,
+                                   InterpFilters interp_filters) {
+  assert(conv_params->do_average == 0 || conv_params->do_average == 1);
+  assert(sf);
+  if (has_scale(subpel_params->xs, subpel_params->ys)) {
+    av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
+                           interp_filters, subpel_params->subpel_x,
+                           subpel_params->xs, subpel_params->subpel_y,
+                           subpel_params->ys, 1, conv_params, sf);
+  } else {
+    SubpelParams sp = *subpel_params;
+    revert_scale_extra_bits(&sp);
+    av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
+                           interp_filters, sp.subpel_x, sp.xs, sp.subpel_y,
+                           sp.ys, 0, conv_params, sf);
+  }
+}
+
+static INLINE void highbd_inter_predictor(
+    const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride,
+    const SubpelParams *subpel_params, const struct scale_factors *sf, int w,
+    int h, ConvolveParams *conv_params, InterpFilters interp_filters, int bd) {
+  assert(conv_params->do_average == 0 || conv_params->do_average == 1);
+  assert(sf);
+  if (has_scale(subpel_params->xs, subpel_params->ys)) {
+    av1_highbd_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
+                                  interp_filters, subpel_params->subpel_x,
+                                  subpel_params->xs, subpel_params->subpel_y,
+                                  subpel_params->ys, 1, conv_params, sf, bd);
+  } else {
+    SubpelParams sp = *subpel_params;
+    revert_scale_extra_bits(&sp);
+    av1_highbd_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
+                                  interp_filters, sp.subpel_x, sp.xs,
+                                  sp.subpel_y, sp.ys, 0, conv_params, sf, bd);
+  }
+}
+
+void av1_modify_neighbor_predictor_for_obmc(MB_MODE_INFO *mbmi);
+int av1_skip_u4x4_pred_in_obmc(BLOCK_SIZE bsize,
+                               const struct macroblockd_plane *pd, int dir);
+
 static INLINE int is_interinter_compound_used(COMPOUND_TYPE type,
                                               BLOCK_SIZE sb_type) {
-  (void)sb_type;
+  const int comp_allowed = is_comp_ref_allowed(sb_type);
   switch (type) {
-#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-    case COMPOUND_AVERAGE: return sb_type >= BLOCK_4X4;
-#else   // CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-    case COMPOUND_AVERAGE: return 1;
-#endif  // CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-#if CONFIG_WEDGE
-    case COMPOUND_WEDGE: return wedge_params_lookup[sb_type].bits > 0;
-#endif  // CONFIG_WEDGE
-#if CONFIG_COMPOUND_SEGMENT
-    case COMPOUND_SEG:
-      return AOMMIN(block_size_wide[sb_type], block_size_high[sb_type]) >= 8;
-#endif  // CONFIG_COMPOUND_SEGMENT
+    case COMPOUND_AVERAGE:
+    case COMPOUND_DIFFWTD: return comp_allowed;
+    case COMPOUND_WEDGE:
+      return comp_allowed && wedge_params_lookup[sb_type].bits > 0;
     default: assert(0); return 0;
   }
 }
 
 static INLINE int is_any_masked_compound_used(BLOCK_SIZE sb_type) {
   COMPOUND_TYPE comp_type;
-#if CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-  if (sb_type < BLOCK_4X4) return 0;
-#endif  // CONFIG_CHROMA_2X2 || CONFIG_CHROMA_SUB8X8
-  for (comp_type = 0; comp_type < COMPOUND_TYPES; comp_type++) {
+  int i;
+  if (!is_comp_ref_allowed(sb_type)) return 0;
+  for (i = 0; i < COMPOUND_TYPES; i++) {
+    comp_type = (COMPOUND_TYPE)i;
     if (is_masked_compound_type(comp_type) &&
         is_interinter_compound_used(comp_type, sb_type))
       return 1;
@@ -257,7 +189,6 @@ static INLINE int get_interinter_wedge_bits(BLOCK_SIZE sb_type) {
 }
 
 static INLINE int is_interintra_wedge_used(BLOCK_SIZE sb_type) {
-  (void)sb_type;
   return wedge_params_lookup[sb_type].bits > 0;
 }
 
@@ -265,60 +196,22 @@ static INLINE int get_interintra_wedge_bits(BLOCK_SIZE sb_type) {
   return wedge_params_lookup[sb_type].bits;
 }
 
-#if CONFIG_COMPOUND_SEGMENT
-void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
-                             const uint8_t *src0, int src0_stride,
-                             const uint8_t *src1, int src1_stride,
-                             BLOCK_SIZE sb_type, int h, int w);
-#if CONFIG_HIGHBITDEPTH
-void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
-                                    const uint8_t *src0, int src0_stride,
-                                    const uint8_t *src1, int src1_stride,
-                                    BLOCK_SIZE sb_type, int h, int w, int bd);
-#endif  // CONFIG_HIGHBITDEPTH
-#endif  // CONFIG_COMPOUND_SEGMENT
+void av1_make_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
+                              int dst_stride, const SubpelParams *subpel_params,
+                              const struct scale_factors *sf, int w, int h,
+                              ConvolveParams *conv_params,
+                              InterpFilters interp_filters,
+                              const WarpTypesAllowed *warp_types, int p_col,
+                              int p_row, int plane, int ref,
+                              const MB_MODE_INFO *mi, int build_for_obmc,
+                              const MACROBLOCKD *xd, int can_use_previous);
 
 void av1_make_masked_inter_predictor(
     const uint8_t *pre, int pre_stride, uint8_t *dst, int dst_stride,
-    const int subpel_x, const int subpel_y, const struct scale_factors *sf,
-    int w, int h, ConvolveParams *conv_params, InterpFilters interp_filters,
-    int xs, int ys,
-#if CONFIG_SUPERTX
-    int wedge_offset_x, int wedge_offset_y,
-#endif  // CONFIG_SUPERTX
-    int plane,
-#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
+    const SubpelParams *subpel_params, const struct scale_factors *sf, int w,
+    int h, ConvolveParams *conv_params, InterpFilters interp_filters, int plane,
     const WarpTypesAllowed *warp_types, int p_col, int p_row, int ref,
-#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
-    MACROBLOCKD *xd);
-
-static INLINE int round_mv_comp_q4(int value) {
-  return (value < 0 ? value - 2 : value + 2) / 4;
-}
-
-static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
-  MV res = {
-    round_mv_comp_q4(
-        mi->bmi[0].as_mv[idx].as_mv.row + mi->bmi[1].as_mv[idx].as_mv.row +
-        mi->bmi[2].as_mv[idx].as_mv.row + mi->bmi[3].as_mv[idx].as_mv.row),
-    round_mv_comp_q4(
-        mi->bmi[0].as_mv[idx].as_mv.col + mi->bmi[1].as_mv[idx].as_mv.col +
-        mi->bmi[2].as_mv[idx].as_mv.col + mi->bmi[3].as_mv[idx].as_mv.col)
-  };
-  return res;
-}
-
-static INLINE int round_mv_comp_q2(int value) {
-  return (value < 0 ? value - 1 : value + 1) / 2;
-}
-
-static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
-  MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
-                              mi->bmi[block1].as_mv[idx].as_mv.row),
-             round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
-                              mi->bmi[block1].as_mv[idx].as_mv.col) };
-  return res;
-}
+    MACROBLOCKD *xd, int can_use_previous);
 
 // TODO(jkoleszar): yet another mv clamping function :-(
 static INLINE MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd,
@@ -331,8 +224,8 @@ static INLINE MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd,
   const int spel_right = spel_left - SUBPEL_SHIFTS;
   const int spel_top = (AOM_INTERP_EXTEND + bh) << SUBPEL_BITS;
   const int spel_bottom = spel_top - SUBPEL_SHIFTS;
-  MV clamped_mv = { src_mv->row * (1 << (1 - ss_y)),
-                    src_mv->col * (1 << (1 - ss_x)) };
+  MV clamped_mv = { (int16_t)(src_mv->row * (1 << (1 - ss_y))),
+                    (int16_t)(src_mv->col * (1 << (1 - ss_x))) };
   assert(ss_x <= 1);
   assert(ss_y <= 1);
 
@@ -344,20 +237,6 @@ static INLINE MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd,
   return clamped_mv;
 }
 
-static INLINE MV average_split_mvs(const struct macroblockd_plane *pd,
-                                   const MODE_INFO *mi, int ref, int block) {
-  const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
-  MV res = { 0, 0 };
-  switch (ss_idx) {
-    case 0: res = mi->bmi[block].as_mv[ref].as_mv; break;
-    case 1: res = mi_mv_pred_q2(mi, ref, block, block + 2); break;
-    case 2: res = mi_mv_pred_q2(mi, ref, block, block + 1); break;
-    case 3: res = mi_mv_pred_q4(mi, ref); break;
-    default: assert(ss_idx <= 3 && ss_idx >= 0);
-  }
-  return res;
-}
-
 void av1_build_inter_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                     int mi_row, int mi_col, BUFFER_SET *ctx,
                                     BLOCK_SIZE bsize);
@@ -370,48 +249,22 @@ void av1_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                    int mi_row, int mi_col, BUFFER_SET *ctx,
                                    BLOCK_SIZE bsize);
 
-#if CONFIG_SUPERTX
-void av1_build_inter_predictor_sb_sub8x8_extend(const AV1_COMMON *cm,
-                                                MACROBLOCKD *xd, int mi_row_ori,
-                                                int mi_col_ori, int mi_row,
-                                                int mi_col, int plane,
-                                                BLOCK_SIZE bsize, int block);
-
-void av1_build_inter_predictor_sb_extend(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                         int mi_row_ori, int mi_col_ori,
-                                         int mi_row, int mi_col, int plane,
-                                         BLOCK_SIZE bsize);
-struct macroblockd_plane;
-void av1_build_masked_inter_predictor_complex(
-    MACROBLOCKD *xd, uint8_t *dst, int dst_stride, const uint8_t *pre,
-    int pre_stride, int mi_row, int mi_col, int mi_row_ori, int mi_col_ori,
-    BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, PARTITION_TYPE partition,
-    int plane);
-#endif  // CONFIG_SUPERTX
-
 void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
                                int dst_stride, const MV *src_mv,
                                const struct scale_factors *sf, int w, int h,
                                ConvolveParams *conv_params,
                                InterpFilters interp_filters,
-#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
                                const WarpTypesAllowed *warp_types, int p_col,
                                int p_row, int plane, int ref,
-#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
                                enum mv_precision precision, int x, int y,
-                               const MACROBLOCKD *xd);
+                               const MACROBLOCKD *xd, int can_use_previous);
 
-#if CONFIG_HIGHBITDEPTH
 void av1_highbd_build_inter_predictor(
     const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride,
     const MV *mv_q3, const struct scale_factors *sf, int w, int h, int do_avg,
-    InterpFilters interp_filters,
-#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
-    const WarpTypesAllowed *warp_types, int p_col, int p_row,
-#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
-    int plane, enum mv_precision precision, int x, int y,
-    const MACROBLOCKD *xd);
-#endif
+    InterpFilters interp_filters, const WarpTypesAllowed *warp_types, int p_col,
+    int p_row, int plane, enum mv_precision precision, int x, int y,
+    const MACROBLOCKD *xd, int can_use_previous);
 
 static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride,
                                        const struct scale_factors *sf) {
@@ -427,15 +280,11 @@ static INLINE void setup_pred_plane(struct buf_2d *dst, BLOCK_SIZE bsize,
                                     int stride, int mi_row, int mi_col,
                                     const struct scale_factors *scale,
                                     int subsampling_x, int subsampling_y) {
-#if CONFIG_CHROMA_SUB8X8
   // Offset the buffer pointer
   if (subsampling_y && (mi_row & 0x01) && (mi_size_high[bsize] == 1))
     mi_row -= 1;
   if (subsampling_x && (mi_col & 0x01) && (mi_size_wide[bsize] == 1))
     mi_col -= 1;
-#else
-  (void)bsize;
-#endif
 
   const int x = (MI_SIZE * mi_col) >> subsampling_x;
   const int y = (MI_SIZE * mi_row) >> subsampling_y;
@@ -447,62 +296,33 @@ static INLINE void setup_pred_plane(struct buf_2d *dst, BLOCK_SIZE bsize,
 }
 
 void av1_setup_dst_planes(struct macroblockd_plane *planes, BLOCK_SIZE bsize,
-                          const YV12_BUFFER_CONFIG *src, int mi_row,
-                          int mi_col);
+                          const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
+                          const int plane_start, const int plane_end);
 
 void av1_setup_pre_planes(MACROBLOCKD *xd, int idx,
                           const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
-                          const struct scale_factors *sf);
+                          const struct scale_factors *sf, const int num_planes);
 
 // Detect if the block have sub-pixel level motion vectors
 // per component.
 #define CHECK_SUBPEL 0
-static INLINE int has_subpel_mv_component(const MODE_INFO *const mi,
+static INLINE int has_subpel_mv_component(const MB_MODE_INFO *const mbmi,
                                           const MACROBLOCKD *const xd,
                                           int dir) {
 #if CHECK_SUBPEL
-  const MB_MODE_INFO *const mbmi = &mi->mbmi;
   const BLOCK_SIZE bsize = mbmi->sb_type;
   int plane;
   int ref = (dir >> 1);
-#if CONFIG_CB4X4
-  const int unify_bsize = 1;
-#else
-  const int unify_bsize = 0;
-#endif
 
-  if (bsize >= BLOCK_8X8 || unify_bsize) {
-    if (dir & 0x01) {
-      if (mbmi->mv[ref].as_mv.col & SUBPEL_MASK) return 1;
-    } else {
-      if (mbmi->mv[ref].as_mv.row & SUBPEL_MASK) return 1;
-    }
+  if (dir & 0x01) {
+    if (mbmi->mv[ref].as_mv.col & SUBPEL_MASK) return 1;
   } else {
-    for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
-      const PARTITION_TYPE bp = BLOCK_8X8 - bsize;
-      const struct macroblockd_plane *const pd = &xd->plane[plane];
-      const int have_vsplit = bp != PARTITION_HORZ;
-      const int have_hsplit = bp != PARTITION_VERT;
-      const int num_4x4_w = 2 >> ((!have_vsplit) | pd->subsampling_x);
-      const int num_4x4_h = 2 >> ((!have_hsplit) | pd->subsampling_y);
-
-      int x, y;
-      for (y = 0; y < num_4x4_h; ++y) {
-        for (x = 0; x < num_4x4_w; ++x) {
-          const MV mv = average_split_mvs(pd, mi, ref, y * 2 + x);
-          if (dir & 0x01) {
-            if (mv.col & SUBPEL_MASK) return 1;
-          } else {
-            if (mv.row & SUBPEL_MASK) return 1;
-          }
-        }
-      }
-    }
+    if (mbmi->mv[ref].as_mv.row & SUBPEL_MASK) return 1;
   }
 
   return 0;
 #else
-  (void)mi;
+  (void)mbmi;
   (void)xd;
   (void)dir;
   return 1;
@@ -516,20 +336,16 @@ static INLINE void set_default_interp_filters(
 }
 
 static INLINE int av1_is_interp_needed(const MACROBLOCKD *const xd) {
-  (void)xd;
-#if CONFIG_WARPED_MOTION
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO *const mbmi = xd->mi[0];
+  if (mbmi->skip_mode) return 0;
   if (mbmi->motion_mode == WARPED_CAUSAL) return 0;
-#endif  // CONFIG_WARPED_MOTION
-#if CONFIG_GLOBAL_MOTION
-  if (is_nontrans_global_motion(xd)) return 0;
-#endif  // CONFIG_GLOBAL_MOTION
+  if (is_nontrans_global_motion(xd, xd->mi[0])) return 0;
   return 1;
 }
 
 static INLINE int av1_is_interp_search_needed(const MACROBLOCKD *const xd) {
-  MODE_INFO *const mi = xd->mi[0];
-  const int is_compound = has_second_ref(&mi->mbmi);
+  MB_MODE_INFO *const mi = xd->mi[0];
+  const int is_compound = has_second_ref(mi);
   int ref;
   for (ref = 0; ref < 1 + is_compound; ++ref) {
     int row_col;
@@ -542,17 +358,15 @@ static INLINE int av1_is_interp_search_needed(const MACROBLOCKD *const xd) {
   }
   return 0;
 }
-
-#if CONFIG_MOTION_VAR
-const uint8_t *av1_get_obmc_mask(int length);
-void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                      int mi_row, int mi_col);
-void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                     int mi_row, int mi_col,
-                                     uint8_t *above[MAX_MB_PLANE],
-                                     int above_stride[MAX_MB_PLANE],
-                                     uint8_t *left[MAX_MB_PLANE],
-                                     int left_stride[MAX_MB_PLANE]);
+void av1_setup_build_prediction_by_above_pred(
+    MACROBLOCKD *xd, int rel_mi_col, uint8_t above_mi_width,
+    MB_MODE_INFO *above_mbmi, struct build_prediction_ctxt *ctxt,
+    const int num_planes);
+void av1_setup_build_prediction_by_left_pred(MACROBLOCKD *xd, int rel_mi_row,
+                                             uint8_t left_mi_height,
+                                             MB_MODE_INFO *left_mbmi,
+                                             struct build_prediction_ctxt *ctxt,
+                                             const int num_planes);
 void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                          int mi_row, int mi_col,
                                          uint8_t *tmp_buf[MAX_MB_PLANE],
@@ -565,13 +379,18 @@ void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                         int tmp_width[MAX_MB_PLANE],
                                         int tmp_height[MAX_MB_PLANE],
                                         int tmp_stride[MAX_MB_PLANE]);
+void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                     int mi_row, int mi_col,
+                                     uint8_t *above[MAX_MB_PLANE],
+                                     int above_stride[MAX_MB_PLANE],
+                                     uint8_t *left[MAX_MB_PLANE],
+                                     int left_stride[MAX_MB_PLANE]);
+
+const uint8_t *av1_get_obmc_mask(int length);
+void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
+                                      int mi_row, int mi_col);
 void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                         int mi_row, int mi_col);
-#if CONFIG_NCOBMC
-void av1_build_ncobmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                          int mi_row, int mi_col);
-#endif
-#endif  // CONFIG_MOTION_VAR
 
 #define MASK_MASTER_SIZE ((MAX_WEDGE_SIZE) << 1)
 #define MASK_MASTER_STRIDE (MASK_MASTER_SIZE)
@@ -584,32 +403,24 @@ static INLINE const uint8_t *av1_get_contiguous_soft_mask(int wedge_index,
   return wedge_params_lookup[sb_type].masks[wedge_sign][wedge_index];
 }
 
-const uint8_t *av1_get_soft_mask(int wedge_index, int wedge_sign,
-                                 BLOCK_SIZE sb_type, int wedge_offset_x,
-                                 int wedge_offset_y);
-
-const uint8_t *av1_get_compound_type_mask_inverse(
-    const INTERINTER_COMPOUND_DATA *const comp_data,
-#if CONFIG_COMPOUND_SEGMENT
-    uint8_t *mask_buffer, int h, int w, int stride,
-#endif
-    BLOCK_SIZE sb_type);
-
 const uint8_t *av1_get_compound_type_mask(
     const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type);
-#if CONFIG_INTERINTRA
+
 void av1_build_interintra_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                      uint8_t *ypred, uint8_t *upred,
                                      uint8_t *vpred, int ystride, int ustride,
                                      int vstride, BUFFER_SET *ctx,
                                      BLOCK_SIZE bsize);
+
 void av1_build_interintra_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                          uint8_t *ypred, int ystride,
                                          BUFFER_SET *ctx, BLOCK_SIZE bsize);
+
 void av1_build_interintra_predictors_sbc(const AV1_COMMON *cm, MACROBLOCKD *xd,
                                          uint8_t *upred, int ustride,
                                          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,
@@ -621,57 +432,27 @@ void av1_build_intra_predictors_for_interintra(
 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);
-#endif  // CONFIG_INTERINTRA
+
 // Encoder only
 void av1_build_inter_predictors_for_planes_single_buf(
     MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane_from, int plane_to, int mi_row,
-    int mi_col, int ref, uint8_t *ext_dst[3], int ext_dst_stride[3]);
-void av1_build_wedge_inter_predictor_from_buf(
-    MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane_from, int plane_to,
-#if CONFIG_SUPERTX
-    int wedge_offset_x, int wedge_offset_y,
-#endif  // CONFIG_SUPERTX
-    uint8_t *ext_dst0[3], int ext_dst_stride0[3], uint8_t *ext_dst1[3],
-    int ext_dst_stride1[3]);
+    int mi_col, int ref, uint8_t *ext_dst[3], int ext_dst_stride[3],
+    int can_use_previous);
+void av1_build_wedge_inter_predictor_from_buf(MACROBLOCKD *xd, BLOCK_SIZE bsize,
+                                              int plane_from, int plane_to,
+                                              uint8_t *ext_dst0[3],
+                                              int ext_dst_stride0[3],
+                                              uint8_t *ext_dst1[3],
+                                              int ext_dst_stride1[3]);
 
-#if CONFIG_NCOBMC_ADAPT_WEIGHT
-#define ASSIGN_ALIGNED_PTRS(p, a, s) \
-  p[0] = a;                          \
-  p[1] = a + s;                      \
-  p[2] = a + 2 * s;
-
-#define ASSIGN_ALIGNED_PTRS_HBD(p, a, s, l) \
-  p[0] = CONVERT_TO_BYTEPTR(a);             \
-  p[1] = CONVERT_TO_BYTEPTR(a + s * l);     \
-  p[2] = CONVERT_TO_BYTEPTR(a + 2 * s * l);
-
-void alloc_ncobmc_pred_buffer(MACROBLOCKD *const xd);
-void free_ncobmc_pred_buffer(MACROBLOCKD *const xd);
-void set_sb_mi_boundaries(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
-                          const int mi_row, const int mi_col);
-
-void reset_xd_boundary(MACROBLOCKD *xd, int mi_row, int bh, int mi_col, int bw,
-                       int mi_rows, int mi_cols);
-
-void get_pred_from_intrpl_buf(MACROBLOCKD *xd, int mi_row, int mi_col,
-                              BLOCK_SIZE bsize, int plane);
-
-void build_ncobmc_intrpl_pred(const AV1_COMMON *const cm, MACROBLOCKD *xd,
-                              int plane, int pxl_row, int pxl_col,
-                              BLOCK_SIZE bsize, uint8_t *preds[][MAX_MB_PLANE],
-                              int ps[MAX_MB_PLANE],  // pred buffer strides
-                              int mode);
-
-void av1_get_ext_blk_preds(const AV1_COMMON *cm, MACROBLOCKD *xd, int bsize,
-                           int mi_row, int mi_col,
-                           uint8_t *dst_buf[][MAX_MB_PLANE],
-                           int dst_stride[MAX_MB_PLANE]);
-
-void av1_get_ori_blk_pred(const AV1_COMMON *cm, MACROBLOCKD *xd, int bsize,
-                          int mi_row, int mi_col,
-                          uint8_t *dst_buf[MAX_MB_PLANE],
-                          int dst_stride[MAX_MB_PLANE]);
-#endif  // CONFIG_NCOBMC_ADAPT_WEIGHT
+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);
+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,
+                   WarpedMotionParams *final_warp_params);
 
 #ifdef __cplusplus
 }  // extern "C"

third_party/aom/av1/common/reconintra.h

@@ -22,15 +22,16 @@ extern "C" {
 
 void av1_init_intra_predictors(void);
 void av1_predict_intra_block_facade(const AV1_COMMON *cm, MACROBLOCKD *xd,
-                                    int plane, int block_idx, int blk_col,
-                                    int blk_row, TX_SIZE tx_size);
+                                    int plane, int blk_col, int blk_row,
+                                    TX_SIZE tx_size);
 void av1_predict_intra_block(const AV1_COMMON *cm, const MACROBLOCKD *xd,
-                             int bw, int bh, BLOCK_SIZE bsize,
-                             PREDICTION_MODE mode, const uint8_t *ref,
-                             int ref_stride, uint8_t *dst, int dst_stride,
-                             int aoff, int loff, int plane);
+                             int bw, int bh, TX_SIZE tx_size,
+                             PREDICTION_MODE mode, int angle_delta,
+                             int use_palette,
+                             FILTER_INTRA_MODE filter_intra_mode,
+                             const uint8_t *ref, int ref_stride, uint8_t *dst,
+                             int dst_stride, int aoff, int loff, int plane);
 
-#if CONFIG_INTERINTRA
 // Mapping of interintra to intra mode for use in the intra component
 static const PREDICTION_MODE interintra_to_intra_mode[INTERINTRA_MODES] = {
   DC_PRED, V_PRED, H_PRED, SMOOTH_PRED
@@ -41,44 +42,67 @@ static const INTERINTRA_MODE intra_to_interintra_mode[INTRA_MODES] = {
   II_DC_PRED, II_V_PRED, II_H_PRED, II_V_PRED,      II_SMOOTH_PRED, II_V_PRED,
   II_H_PRED,  II_H_PRED, II_V_PRED, II_SMOOTH_PRED, II_SMOOTH_PRED
 };
-#endif  // CONFIG_INTERINTRA
 
-#if CONFIG_FILTER_INTRA
-#define FILTER_INTRA_PREC_BITS 10
-#endif  // CONFIG_FILTER_INTRA
+#define FILTER_INTRA_SCALE_BITS 4
 
-#define CONFIG_INTRA_EDGE_UPSAMPLE CONFIG_INTRA_EDGE
-#define CONFIG_USE_ANGLE_DELTA_SUB8X8 0
-
-#if CONFIG_EXT_INTRA
-static INLINE int av1_is_directional_mode(PREDICTION_MODE mode,
-                                          BLOCK_SIZE bsize) {
-#if CONFIG_INTRA_EDGE_UPSAMPLE
-  (void)bsize;
-  return mode >= V_PRED && mode <= D63_PRED;
-#else
-  return mode >= V_PRED && mode <= D63_PRED && bsize >= BLOCK_8X8;
-#endif
+static INLINE int av1_is_directional_mode(PREDICTION_MODE mode) {
+  return mode >= V_PRED && mode <= D67_PRED;
 }
 
 static INLINE int av1_use_angle_delta(BLOCK_SIZE bsize) {
-  (void)bsize;
-#if CONFIG_USE_ANGLE_DELTA_SUB8X8
-  return 1;
-#else
   return bsize >= BLOCK_8X8;
-#endif
 }
-#endif  // CONFIG_EXT_INTRA
 
-#if CONFIG_INTRABC
-static INLINE int av1_allow_intrabc(BLOCK_SIZE bsize,
-                                    const AV1_COMMON *const cm) {
-  return (bsize >= BLOCK_8X8 || bsize == BLOCK_4X4) &&
-         cm->allow_screen_content_tools;
+static INLINE int av1_allow_intrabc(const AV1_COMMON *const cm) {
+  return frame_is_intra_only(cm) && cm->allow_screen_content_tools &&
+         cm->allow_intrabc;
 }
-#endif  // CONFIG_INTRABC
 
+static INLINE int av1_filter_intra_allowed_bsize(const AV1_COMMON *const cm,
+                                                 BLOCK_SIZE bs) {
+  if (!cm->seq_params.enable_filter_intra || bs == BLOCK_INVALID) return 0;
+
+  return block_size_wide[bs] <= 32 && block_size_high[bs] <= 32;
+}
+
+static INLINE int av1_filter_intra_allowed(const AV1_COMMON *const cm,
+                                           const MB_MODE_INFO *mbmi) {
+  return mbmi->mode == DC_PRED &&
+         mbmi->palette_mode_info.palette_size[0] == 0 &&
+         av1_filter_intra_allowed_bsize(cm, mbmi->sb_type);
+}
+
+extern const int8_t av1_filter_intra_taps[FILTER_INTRA_MODES][8][8];
+
+// Get the shift (up-scaled by 256) in X w.r.t a unit change in Y.
+// If angle > 0 && angle < 90, dx = -((int)(256 / t));
+// If angle > 90 && angle < 180, dx = (int)(256 / t);
+// If angle > 180 && angle < 270, dx = 1;
+static INLINE int av1_get_dx(int angle) {
+  if (angle > 0 && angle < 90) {
+    return dr_intra_derivative[angle];
+  } else if (angle > 90 && angle < 180) {
+    return dr_intra_derivative[180 - angle];
+  } else {
+    // In this case, we are not really going to use dx. We may return any value.
+    return 1;
+  }
+}
+
+// Get the shift (up-scaled by 256) in Y w.r.t a unit change in X.
+// If angle > 0 && angle < 90, dy = 1;
+// If angle > 90 && angle < 180, dy = (int)(256 * t);
+// If angle > 180 && angle < 270, dy = -((int)(256 * t));
+static INLINE int av1_get_dy(int angle) {
+  if (angle > 90 && angle < 180) {
+    return dr_intra_derivative[angle - 90];
+  } else if (angle > 180 && angle < 270) {
+    return dr_intra_derivative[270 - angle];
+  } else {
+    // In this case, we are not really going to use dy. We may return any value.
+    return 1;
+  }
+}
 #ifdef __cplusplus
 }  // extern "C"
 #endif