wuggy/Dactyloidae
wuggy/Dactyloidae
1
0
Fork 0
mirror of https://repo.dactyloidae.xyz/Dactyloidae/UXP.git synced 2026-10-06 23:37:31 +09:00
Code Issues Projects Releases Packages Wiki Activity

Update aom to slightly newer commit ID

Browse source
This commit is contained in:
trav90 2018-10-17 05:59:08 -05:00 • committed by Roy Tam
commit 76f1e6edca
311 changed files with 55292 additions and 33790 deletions
Download patch file Download diff file Expand all files Collapse all files

65
third_party/aom/av1/av1.cmake vendored
View file

@ -8,6 +8,9 @@
## 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.
##
if (NOT AOM_AV1_AV1_CMAKE_)
set(AOM_AV1_AV1_CMAKE_ 1)
set(AOM_AV1_COMMON_SOURCES
"${AOM_ROOT}/av1/av1_iface_common.h"
"${AOM_ROOT}/av1/common/alloccommon.c"
@ -16,11 +19,11 @@ set(AOM_AV1_COMMON_SOURCES
"${AOM_ROOT}/av1/common/av1_fwd_txfm1d.c"
"${AOM_ROOT}/av1/common/av1_fwd_txfm1d.h"
"${AOM_ROOT}/av1/common/av1_fwd_txfm2d.c"
"${AOM_ROOT}/av1/common/av1_fwd_txfm2d_cfg.h"
"${AOM_ROOT}/av1/common/av1_fwd_txfm1d_cfg.h"
"${AOM_ROOT}/av1/common/av1_inv_txfm1d.c"
"${AOM_ROOT}/av1/common/av1_inv_txfm1d.h"
"${AOM_ROOT}/av1/common/av1_inv_txfm2d.c"
"${AOM_ROOT}/av1/common/av1_inv_txfm2d_cfg.h"
"${AOM_ROOT}/av1/common/av1_inv_txfm1d_cfg.h"
"${AOM_ROOT}/av1/common/av1_loopfilter.c"
"${AOM_ROOT}/av1/common/av1_loopfilter.h"
"${AOM_ROOT}/av1/common/av1_txfm.h"
@ -60,7 +63,6 @@ set(AOM_AV1_COMMON_SOURCES
"${AOM_ROOT}/av1/common/reconintra.h"
"${AOM_ROOT}/av1/common/resize.c"
"${AOM_ROOT}/av1/common/resize.h"
"${AOM_ROOT}/av1/common/restoration.h"
"${AOM_ROOT}/av1/common/scale.c"
"${AOM_ROOT}/av1/common/scale.h"
"${AOM_ROOT}/av1/common/scan.c"
@ -146,13 +148,9 @@ set(AOM_AV1_ENCODER_SOURCES
"${AOM_ROOT}/av1/encoder/tokenize.c"
"${AOM_ROOT}/av1/encoder/tokenize.h"
"${AOM_ROOT}/av1/encoder/treewriter.c"
"${AOM_ROOT}/av1/encoder/treewriter.h"
"${AOM_ROOT}/av1/encoder/variance_tree.c"
"${AOM_ROOT}/av1/encoder/variance_tree.h")
"${AOM_ROOT}/av1/encoder/treewriter.h")
set(AOM_AV1_COMMON_INTRIN_SSE2
# Requires CONFIG_GLOBAL_MOTION or CONFIG_WARPED_MOTION
#"${AOM_ROOT}/av1/common/x86/warp_plane_sse2.c"
"${AOM_ROOT}/av1/common/x86/idct_intrin_sse2.c")
set(AOM_AV1_COMMON_INTRIN_SSSE3
@ -293,6 +291,27 @@ if (CONFIG_ACCOUNTING)
"${AOM_ROOT}/av1/decoder/accounting.h")
endif ()
if (CONFIG_GLOBAL_MOTION)
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/corner_detect.c"
"${AOM_ROOT}/av1/encoder/corner_detect.h"
"${AOM_ROOT}/av1/encoder/corner_match.c"
"${AOM_ROOT}/av1/encoder/corner_match.h"
"${AOM_ROOT}/av1/encoder/global_motion.c"
"${AOM_ROOT}/av1/encoder/global_motion.h"
"${AOM_ROOT}/av1/encoder/ransac.c"
"${AOM_ROOT}/av1/encoder/ransac.h"
"${AOM_ROOT}/third_party/fastfeat/fast_9.c"
"${AOM_ROOT}/third_party/fastfeat/fast.c"
"${AOM_ROOT}/third_party/fastfeat/fast.h"
"${AOM_ROOT}/third_party/fastfeat/nonmax.c")
set(AOM_AV1_ENCODER_INTRIN_SSE4_1
${AOM_AV1_ENCODER_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/encoder/x86/corner_match_sse4.c")
endif ()
if (CONFIG_INSPECTION)
set(AOM_AV1_DECODER_SOURCES
${AOM_AV1_DECODER_SOURCES}
@ -320,6 +339,22 @@ if (CONFIG_CFL)
"${AOM_ROOT}/av1/common/cfl.h")
endif ()
if (CONFIG_LOOP_RESTORATION)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/restoration.c"
"${AOM_ROOT}/av1/common/restoration.h")
set(AOM_AV1_COMMON_INTRIN_SSE4_1
${AOM_AV1_COMMON_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/common/x86/selfguided_sse4.c")
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/pickrst.c"
"${AOM_ROOT}/av1/encoder/pickrst.h")
endif ()
if (CONFIG_PVQ)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
@ -382,7 +417,7 @@ if (CONFIG_PVQ)
endif ()
endif ()
if (CONFIG_WARPED_MOTION)
if (CONFIG_WARPED_MOTION OR CONFIG_GLOBAL_MOTION)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/warped_motion.c"
@ -391,6 +426,16 @@ if (CONFIG_WARPED_MOTION)
set(AOM_AV1_COMMON_INTRIN_SSE2
${AOM_AV1_COMMON_INTRIN_SSE2}
"${AOM_ROOT}/av1/common/x86/warp_plane_sse2.c")
set(AOM_AV1_COMMON_INTRIN_SSSE3
${AOM_AV1_COMMON_INTRIN_SSSE3}
"${AOM_ROOT}/av1/common/x86/warp_plane_ssse3.c")
if (CONFIG_HIGHBITDEPTH)
set(AOM_AV1_COMMON_INTRIN_SSSE3
${AOM_AV1_COMMON_INTRIN_SSSE3}
"${AOM_ROOT}/av1/common/x86/highbd_warp_plane_ssse3.c")
endif ()
endif ()
# Setup AV1 common/decoder/encoder targets. The libaom target must exist before
@ -516,3 +561,5 @@ endfunction ()
function (setup_av1_test_targets)
endfunction ()
endif () # AOM_AV1_AV1_CMAKE_

6
third_party/aom/av1/av1_common.mk vendored
View file

@ -58,7 +58,6 @@ AV1_COMMON_SRCS-yes += common/reconinter.c
AV1_COMMON_SRCS-yes += common/reconintra.c
AV1_COMMON_SRCS-yes += common/resize.c
AV1_COMMON_SRCS-yes += common/resize.h
AV1_COMMON_SRCS-yes += common/restoration.h
AV1_COMMON_SRCS-yes += common/common_data.h
AV1_COMMON_SRCS-yes += common/scan.c
AV1_COMMON_SRCS-yes += common/scan.h
@ -69,9 +68,9 @@ AV1_COMMON_SRCS-yes += common/av1_fwd_txfm1d.c
AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d.h
AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d.c
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm2d.c
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm2d_cfg.h
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm1d_cfg.h
AV1_COMMON_SRCS-yes += common/av1_inv_txfm2d.c
AV1_COMMON_SRCS-yes += common/av1_inv_txfm2d_cfg.h
AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d_cfg.h
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/av1_convolve_ssse3.c
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_highbd_convolve_sse4.c
@ -172,6 +171,7 @@ endif
ifneq ($(findstring yes,$(CONFIG_GLOBAL_MOTION) $(CONFIG_WARPED_MOTION)),)
AV1_COMMON_SRCS-$(HAVE_SSE2) += common/x86/warp_plane_sse2.c
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/warp_plane_ssse3.c
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/highbd_warp_plane_ssse3.c
endif

7
third_party/aom/av1/av1_cx.mk vendored
View file

@ -23,8 +23,6 @@ AV1_CX_SRCS-yes += encoder/av1_quantize.h
AV1_CX_SRCS-yes += encoder/bitstream.c
AV1_CX_SRCS-yes += encoder/context_tree.c
AV1_CX_SRCS-yes += encoder/context_tree.h
AV1_CX_SRCS-yes += encoder/variance_tree.c
AV1_CX_SRCS-yes += encoder/variance_tree.h
AV1_CX_SRCS-yes += encoder/cost.h
AV1_CX_SRCS-yes += encoder/cost.c
AV1_CX_SRCS-yes += encoder/dct.c
@ -38,6 +36,7 @@ AV1_CX_SRCS-yes += encoder/ethread.h
AV1_CX_SRCS-yes += encoder/ethread.c
AV1_CX_SRCS-yes += encoder/extend.c
AV1_CX_SRCS-yes += encoder/firstpass.c
AV1_CX_SRCS-yes += encoder/mathutils.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/fast.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/nonmax.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/fast_9.c
@ -162,4 +161,8 @@ AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct16x16_msa.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct_msa.h
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/temporal_filter_msa.c
ifeq ($(CONFIG_GLOBAL_MOTION),yes)
AV1_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/corner_match_sse4.c
endif
AV1_CX_SRCS-yes := $(filter-out $(AV1_CX_SRCS_REMOVE-yes),$(AV1_CX_SRCS-yes))

10
third_party/aom/av1/av1_cx_iface.c vendored
View file

@ -483,6 +483,16 @@ static aom_codec_err_t set_encoder_config(
oxcf->resize_mode = RESIZE_NONE;
}
// Initialize to input resolution if not specified.
if (oxcf->resize_mode != RESIZE_FIXED) {
oxcf->scaled_frame_width = oxcf->width;
oxcf->scaled_frame_height = oxcf->height;
}
#if CONFIG_FRAME_SUPERRES
oxcf->superres_enabled = 1; // TODO(afergs): Check the config
#endif // CONFIG_FRAME_SUPERRES
oxcf->maximum_buffer_size_ms = is_vbr ? 240000 : cfg->rc_buf_sz;
oxcf->starting_buffer_level_ms = is_vbr ? 60000 : cfg->rc_buf_initial_sz;
oxcf->optimal_buffer_level_ms = is_vbr ? 60000 : cfg->rc_buf_optimal_sz;

11
third_party/aom/av1/av1_dx_iface.c vendored
View file

@ -31,8 +31,6 @@
#include "av1/av1_iface_common.h"
typedef aom_codec_stream_info_t av1_stream_info_t;
// This limit is due to framebuffer numbers.
// TODO(hkuang): Remove this limit after implementing ondemand framebuffers.
#define FRAME_CACHE_SIZE 6 // Cache maximum 6 decoded frames.
@ -45,7 +43,7 @@ typedef struct cache_frame {
struct aom_codec_alg_priv {
aom_codec_priv_t base;
aom_codec_dec_cfg_t cfg;
av1_stream_info_t si;
aom_codec_stream_info_t si;
int postproc_cfg_set;
aom_postproc_cfg_t postproc_cfg;
aom_decrypt_cb decrypt_cb;
@ -101,7 +99,6 @@ static aom_codec_err_t decoder_init(aom_codec_ctx_t *ctx,
ctx->priv = (aom_codec_priv_t *)priv;
ctx->priv->init_flags = ctx->init_flags;
priv->si.sz = sizeof(priv->si);
priv->flushed = 0;
// Only do frame parallel decode when threads > 1.
priv->frame_parallel_decode =
@ -263,11 +260,7 @@ static aom_codec_err_t decoder_peek_si(const uint8_t *data,
static aom_codec_err_t decoder_get_si(aom_codec_alg_priv_t *ctx,
aom_codec_stream_info_t *si) {
const size_t sz = (si->sz >= sizeof(av1_stream_info_t))
? sizeof(av1_stream_info_t)
: sizeof(aom_codec_stream_info_t);
memcpy(si, &ctx->si, sz);
si->sz = (unsigned int)sz;
memcpy(si, &ctx->si, sizeof(*si));
return AOM_CODEC_OK;
}

18
third_party/aom/av1/common/alloccommon.c vendored
View file

@ -125,6 +125,11 @@ void av1_free_context_buffers(AV1_COMMON *cm) {
#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;
}
#endif
}
@ -155,7 +160,8 @@ int av1_alloc_context_buffers(AV1_COMMON *cm, int width, int height) {
for (i = 0; i < MAX_MB_PLANE; i++) {
aom_free(cm->above_context[i]);
cm->above_context[i] = (ENTROPY_CONTEXT *)aom_calloc(
2 * aligned_mi_cols, sizeof(*cm->above_context[0]));
aligned_mi_cols << (MI_SIZE_LOG2 - tx_size_wide_log2[0]),
sizeof(*cm->above_context[0]));
if (!cm->above_context[i]) goto fail;
}
@ -167,8 +173,16 @@ int av1_alloc_context_buffers(AV1_COMMON *cm, int width, int height) {
#if CONFIG_VAR_TX
aom_free(cm->above_txfm_context);
cm->above_txfm_context = (TXFM_CONTEXT *)aom_calloc(
aligned_mi_cols, sizeof(*cm->above_txfm_context));
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;

115
third_party/aom/av1/common/av1_fwd_txfm1d.c vendored
View file

@ -10,6 +10,7 @@
*/
#include <stdlib.h>
#include "aom_dsp/inv_txfm.h"
#include "av1/common/av1_fwd_txfm1d.h"
#if CONFIG_COEFFICIENT_RANGE_CHECKING
@ -53,7 +54,7 @@ void av1_fdct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -100,7 +101,7 @@ void av1_fdct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[3];
@ -115,7 +116,7 @@ void av1_fdct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -130,7 +131,7 @@ void av1_fdct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -193,7 +194,7 @@ void av1_fdct16_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[7];
@ -216,7 +217,7 @@ void av1_fdct16_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[3];
@ -239,7 +240,7 @@ void av1_fdct16_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -262,7 +263,7 @@ void av1_fdct16_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -285,7 +286,7 @@ void av1_fdct16_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -380,7 +381,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[15];
@ -419,7 +420,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[7];
@ -458,7 +459,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[3];
@ -497,7 +498,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -536,7 +537,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -575,7 +576,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 7
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -614,7 +615,7 @@ void av1_fdct32_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -713,7 +714,7 @@ void av1_fadst4_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[8], bf0[0], cospi[56], bf0[1], cos_bit[stage]);
@ -734,7 +735,7 @@ void av1_fadst4_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -781,7 +782,7 @@ void av1_fadst8_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
@ -810,7 +811,7 @@ void av1_fadst8_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -839,7 +840,7 @@ void av1_fadst8_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -902,7 +903,7 @@ void av1_fadst16_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
@ -947,7 +948,7 @@ void av1_fadst16_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -992,7 +993,7 @@ void av1_fadst16_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1037,7 +1038,7 @@ void av1_fadst16_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1132,7 +1133,7 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[1], bf0[0], cospi[63], bf0[1], cos_bit[stage]);
@ -1209,7 +1210,7 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1286,7 +1287,7 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1363,7 +1364,7 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1440,7 +1441,7 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
// stage 10
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1516,6 +1517,38 @@ void av1_fadst32_new(const int32_t *input, int32_t *output,
range_check(stage, input, bf1, size, stage_range[stage]);
}
#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)cos_bit;
for (int i = 0; i < 4; ++i)
output[i] = (int32_t)dct_const_round_shift(input[i] * Sqrt2);
range_check(0, input, output, 4, stage_range[0]);
}
void av1_fidentity8_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 8; ++i) output[i] = input[i] * 2;
range_check(0, input, output, 8, stage_range[0]);
}
void av1_fidentity16_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 16; ++i)
output[i] = (int32_t)dct_const_round_shift(input[i] * 2 * Sqrt2);
range_check(0, input, output, 16, stage_range[0]);
}
void av1_fidentity32_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 32; ++i) output[i] = input[i] * 4;
range_check(0, input, output, 32, stage_range[0]);
}
#endif // CONFIG_EXT_TX
#if CONFIG_TX64X64
void av1_fdct64_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
@ -1531,7 +1564,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf1 = output;
bf1[0] = input[0] + input[63];
bf1[1] = input[1] + input[62];
@ -1601,7 +1634,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[31];
@ -1672,7 +1705,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[15];
@ -1743,7 +1776,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[7];
@ -1814,7 +1847,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[3];
@ -1885,7 +1918,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -1956,7 +1989,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 7
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -2027,7 +2060,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -2098,7 +2131,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 9
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -2169,7 +2202,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 10
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -2240,7 +2273,7 @@ void av1_fdct64_new(const int32_t *input, int32_t *output,
// stage 11
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];

10
third_party/aom/av1/common/av1_fwd_txfm1d.h vendored
View file

@ -37,6 +37,16 @@ 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);
#endif // CONFIG_EXT_TX
#ifdef __cplusplus
}

364
third_party/aom/av1/common/av1_fwd_txfm1d_cfg.h vendored Normal file
View file

@ -0,0 +1,364 @@
/*
* 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"
// Identity will always use max bitdepth regardless of size
static const int8_t fwd_stage_range_identity[1] = { 12 };
// ---------------- 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] = { 15, 16, 17, 17 };
static const int8_t fwd_stage_range_row_dct_4[4] = { 17, 18, 18, 18 };
static const int8_t fwd_stage_range_col_adst_4[6] = { 15, 15, 16, 17, 17, 17 };
static const int8_t fwd_stage_range_row_adst_4[6] = { 17, 17, 17, 18, 18, 18 };
// 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] = { 15, 16, 17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_dct_8[6] = { 17, 18, 19, 19, 19, 19 };
static const int8_t fwd_stage_range_col_adst_8[8] = { 15, 15, 16, 17,
17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_adst_8[8] = { 17, 17, 17, 18,
18, 19, 19, 19 };
// 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] = { 15, 16, 17, 18,
19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_16[8] = { 17, 18, 19, 20,
20, 20, 20, 20 };
static const int8_t fwd_stage_range_col_adst_16[10] = { 15, 15, 16, 17, 17,
18, 18, 19, 19, 19 };
static const int8_t fwd_stage_range_row_adst_16[10] = { 17, 17, 17, 18, 18,
19, 19, 20, 20, 20 };
// 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] = { 15, 16, 17, 18, 19,
20, 20, 20, 20, 20 };
static const int8_t fwd_stage_range_row_dct_32[10] = { 16, 17, 18, 19, 20,
20, 20, 20, 20, 20 };
static const int8_t fwd_stage_range_col_adst_32[12] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_stage_range_row_adst_32[12] = {
16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
// 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] = { 13, 14, 15, 16, 17, 18,
19, 19, 19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_64[12] = { 17, 18, 19, 20, 21, 22,
22, 22, 22, 22, 22, 22 };
// 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_identity, // .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_identity, // .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_identity, // .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_identity, // .stage_range
NULL, // .cos_bit
TXFM_TYPE_IDENTITY32, // .txfm_type
};
#endif // CONFIG_EXT_TX
#endif // AV1_FWD_TXFM2D_CFG_H_

170
third_party/aom/av1/common/av1_fwd_txfm2d.c vendored
View file

@ -14,7 +14,7 @@
#include "./av1_rtcd.h"
#include "av1/common/enums.h"
#include "av1/common/av1_fwd_txfm1d.h"
#include "av1/common/av1_fwd_txfm2d_cfg.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) {
@ -27,6 +27,12 @@ static INLINE TxfmFunc fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
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;
#endif // CONFIG_EXT_TX
default: assert(0); return NULL;
}
}
@ -35,14 +41,15 @@ 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 c, r;
const int txfm_size = cfg->cfg->txfm_size;
const int8_t *shift = cfg->cfg->shift;
const int8_t *stage_range_col = cfg->cfg->stage_range_col;
const int8_t *stage_range_row = cfg->cfg->stage_range_row;
const int8_t *cos_bit_col = cfg->cfg->cos_bit_col;
const int8_t *cos_bit_row = cfg->cfg->cos_bit_row;
const TxfmFunc txfm_func_col = fwd_txfm_type_to_func(cfg->cfg->txfm_type_col);
const TxfmFunc txfm_func_row = fwd_txfm_type_to_func(cfg->cfg->txfm_type_row);
// TODO(sarahparker) must correct for rectangular transforms in follow up
const int txfm_size = cfg->row_cfg->txfm_size;
const int8_t *shift = cfg->row_cfg->shift;
const int8_t *stage_range_col = cfg->col_cfg->stage_range;
const int8_t *stage_range_row = cfg->row_cfg->stage_range;
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;
@ -117,96 +124,79 @@ void av1_fwd_txfm2d_64x64_c(const int16_t *input, int32_t *output, int stride,
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
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
static const TXFM_2D_CFG *fwd_txfm_cfg_ls[FLIPADST_ADST + 1][TX_SIZES] = {
// FLIPADST
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
&fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
&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_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
};
#else // CONFIG_EXT_TX
static const TXFM_2D_CFG *fwd_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
&fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
};
&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(int tx_type, int tx_size) {
TXFM_2D_FLIP_CFG cfg;
set_flip_cfg(tx_type, &cfg);
cfg.cfg = fwd_txfm_cfg_ls[tx_type][tx_size];
int tx_type_col = vtx_tab[tx_type];
int tx_type_row = htx_tab[tx_type];
cfg.col_cfg = fwd_txfm_col_cfg_ls[tx_type_col][tx_size];
cfg.row_cfg = fwd_txfm_row_cfg_ls[tx_type_row][tx_size];
return cfg;
}
@ -214,13 +204,11 @@ TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(int tx_type) {
TXFM_2D_FLIP_CFG cfg;
switch (tx_type) {
case DCT_DCT:
cfg.cfg = &fwd_txfm_2d_cfg_dct_dct_64;
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;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
default:
cfg.ud_flip = 0;
cfg.lr_flip = 0;

444
third_party/aom/av1/common/av1_fwd_txfm2d_cfg.h vendored
View file

@ -1,444 +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"
// ---------------- config fwd_dct_dct_4 ----------------
static const int8_t fwd_shift_dct_dct_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_dct_dct_4[4] = { 15, 16, 17, 17 };
static const int8_t fwd_stage_range_row_dct_dct_4[4] = { 17, 18, 18, 18 };
static const int8_t fwd_cos_bit_col_dct_dct_4[4] = { 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_4 = {
4, // .txfm_size
4, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_4, // .shift
fwd_stage_range_col_dct_dct_4, // .stage_range_col
fwd_stage_range_row_dct_dct_4, // .stage_range_row
fwd_cos_bit_col_dct_dct_4, // .cos_bit_col
fwd_cos_bit_row_dct_dct_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_8 ----------------
static const int8_t fwd_shift_dct_dct_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_dct_dct_8[6] = {
15, 16, 17, 18, 18, 18
};
static const int8_t fwd_stage_range_row_dct_dct_8[6] = {
17, 18, 19, 19, 19, 19
};
static const int8_t fwd_cos_bit_col_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_8 = {
8, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_8, // .shift
fwd_stage_range_col_dct_dct_8, // .stage_range_col
fwd_stage_range_row_dct_dct_8, // .stage_range_row
fwd_cos_bit_col_dct_dct_8, // .cos_bit_col
fwd_cos_bit_row_dct_dct_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_16 ----------------
static const int8_t fwd_shift_dct_dct_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_dct_dct_16[8] = { 15, 16, 17, 18,
19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_dct_16[8] = { 17, 18, 19, 20,
20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_dct_dct_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_16 = {
16, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_16, // .shift
fwd_stage_range_col_dct_dct_16, // .stage_range_col
fwd_stage_range_row_dct_dct_16, // .stage_range_row
fwd_cos_bit_col_dct_dct_16, // .cos_bit_col
fwd_cos_bit_row_dct_dct_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_32 ----------------
static const int8_t fwd_shift_dct_dct_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_dct_dct_32[10] = { 15, 16, 17, 18, 19,
20, 20, 20, 20, 20 };
static const int8_t fwd_stage_range_row_dct_dct_32[10] = { 16, 17, 18, 19, 20,
20, 20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const int8_t fwd_cos_bit_row_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_32 = {
32, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
fwd_shift_dct_dct_32, // .shift
fwd_stage_range_col_dct_dct_32, // .stage_range_col
fwd_stage_range_row_dct_dct_32, // .stage_range_row
fwd_cos_bit_col_dct_dct_32, // .cos_bit_col
fwd_cos_bit_row_dct_dct_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_64 ----------------
static const int8_t fwd_shift_dct_dct_64[3] = { 0, -2, -2 };
static const int8_t fwd_stage_range_col_dct_dct_64[12] = {
13, 14, 15, 16, 17, 18, 19, 19, 19, 19, 19, 19
};
static const int8_t fwd_stage_range_row_dct_dct_64[12] = {
17, 18, 19, 20, 21, 22, 22, 22, 22, 22, 22, 22
};
static const int8_t fwd_cos_bit_col_dct_dct_64[12] = { 15, 15, 15, 15, 15, 14,
13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_64[12] = { 15, 14, 13, 12, 11, 10,
10, 10, 10, 10, 10, 10 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_64 = {
64, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
fwd_shift_dct_dct_64, // .shift
fwd_stage_range_col_dct_dct_64, // .stage_range_col
fwd_stage_range_row_dct_dct_64, // .stage_range_row
fwd_cos_bit_col_dct_dct_64, // .cos_bit_col
fwd_cos_bit_row_dct_dct_64, // .cos_bit_row
TXFM_TYPE_DCT64, // .txfm_type_col
TXFM_TYPE_DCT64
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_4 ----------------
static const int8_t fwd_shift_dct_adst_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_dct_adst_4[4] = { 15, 16, 17, 17 };
static const int8_t fwd_stage_range_row_dct_adst_4[6] = {
17, 17, 17, 18, 18, 18
};
static const int8_t fwd_cos_bit_col_dct_adst_4[4] = { 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_4 = {
4, // .txfm_size
4, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_4, // .shift
fwd_stage_range_col_dct_adst_4, // .stage_range_col
fwd_stage_range_row_dct_adst_4, // .stage_range_row
fwd_cos_bit_col_dct_adst_4, // .cos_bit_col
fwd_cos_bit_row_dct_adst_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_8 ----------------
static const int8_t fwd_shift_dct_adst_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_dct_adst_8[6] = {
15, 16, 17, 18, 18, 18
};
static const int8_t fwd_stage_range_row_dct_adst_8[8] = { 17, 17, 17, 18,
18, 19, 19, 19 };
static const int8_t fwd_cos_bit_col_dct_adst_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_8 = {
8, // .txfm_size
6, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_8, // .shift
fwd_stage_range_col_dct_adst_8, // .stage_range_col
fwd_stage_range_row_dct_adst_8, // .stage_range_row
fwd_cos_bit_col_dct_adst_8, // .cos_bit_col
fwd_cos_bit_row_dct_adst_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_16 ----------------
static const int8_t fwd_shift_dct_adst_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_dct_adst_16[8] = { 15, 16, 17, 18,
19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_adst_16[10] = {
17, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_dct_adst_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_16 = {
16, // .txfm_size
8, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_16, // .shift
fwd_stage_range_col_dct_adst_16, // .stage_range_col
fwd_stage_range_row_dct_adst_16, // .stage_range_row
fwd_cos_bit_col_dct_adst_16, // .cos_bit_col
fwd_cos_bit_row_dct_adst_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_32 ----------------
static const int8_t fwd_shift_dct_adst_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_dct_adst_32[10] = {
15, 16, 17, 18, 19, 20, 20, 20, 20, 20
};
static const int8_t fwd_stage_range_row_dct_adst_32[12] = {
16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_dct_adst_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const int8_t fwd_cos_bit_row_dct_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_32 = {
32, // .txfm_size
10, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
fwd_shift_dct_adst_32, // .shift
fwd_stage_range_col_dct_adst_32, // .stage_range_col
fwd_stage_range_row_dct_adst_32, // .stage_range_row
fwd_cos_bit_col_dct_adst_32, // .cos_bit_col
fwd_cos_bit_row_dct_adst_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_4 ----------------
static const int8_t fwd_shift_adst_adst_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_adst_adst_4[6] = { 15, 15, 16,
17, 17, 17 };
static const int8_t fwd_stage_range_row_adst_adst_4[6] = { 17, 17, 17,
18, 18, 18 };
static const int8_t fwd_cos_bit_col_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_4 = {
4, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_4, // .shift
fwd_stage_range_col_adst_adst_4, // .stage_range_col
fwd_stage_range_row_adst_adst_4, // .stage_range_row
fwd_cos_bit_col_adst_adst_4, // .cos_bit_col
fwd_cos_bit_row_adst_adst_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_8 ----------------
static const int8_t fwd_shift_adst_adst_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_adst_adst_8[8] = { 15, 15, 16, 17,
17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_adst_adst_8[8] = { 17, 17, 17, 18,
18, 19, 19, 19 };
static const int8_t fwd_cos_bit_col_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_8 = {
8, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_8, // .shift
fwd_stage_range_col_adst_adst_8, // .stage_range_col
fwd_stage_range_row_adst_adst_8, // .stage_range_row
fwd_cos_bit_col_adst_adst_8, // .cos_bit_col
fwd_cos_bit_row_adst_adst_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_16 ----------------
static const int8_t fwd_shift_adst_adst_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_adst_adst_16[10] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 19
};
static const int8_t fwd_stage_range_row_adst_adst_16[10] = {
17, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_adst_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_16 = {
16, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_16, // .shift
fwd_stage_range_col_adst_adst_16, // .stage_range_col
fwd_stage_range_row_adst_adst_16, // .stage_range_row
fwd_cos_bit_col_adst_adst_16, // .cos_bit_col
fwd_cos_bit_row_adst_adst_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_32 ----------------
static const int8_t fwd_shift_adst_adst_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_adst_adst_32[12] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_stage_range_row_adst_adst_32[12] = {
16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const int8_t fwd_cos_bit_row_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_32 = {
32, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
fwd_shift_adst_adst_32, // .shift
fwd_stage_range_col_adst_adst_32, // .stage_range_col
fwd_stage_range_row_adst_adst_32, // .stage_range_row
fwd_cos_bit_col_adst_adst_32, // .cos_bit_col
fwd_cos_bit_row_adst_adst_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_4 ----------------
static const int8_t fwd_shift_adst_dct_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_adst_dct_4[6] = {
15, 15, 16, 17, 17, 17
};
static const int8_t fwd_stage_range_row_adst_dct_4[4] = { 17, 18, 18, 18 };
static const int8_t fwd_cos_bit_col_adst_dct_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_4 = {
4, // .txfm_size
6, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_4, // .shift
fwd_stage_range_col_adst_dct_4, // .stage_range_col
fwd_stage_range_row_adst_dct_4, // .stage_range_row
fwd_cos_bit_col_adst_dct_4, // .cos_bit_col
fwd_cos_bit_row_adst_dct_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_8 ----------------
static const int8_t fwd_shift_adst_dct_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_adst_dct_8[8] = { 15, 15, 16, 17,
17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_adst_dct_8[6] = {
17, 18, 19, 19, 19, 19
};
static const int8_t fwd_cos_bit_col_adst_dct_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_8 = {
8, // .txfm_size
8, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_8, // .shift
fwd_stage_range_col_adst_dct_8, // .stage_range_col
fwd_stage_range_row_adst_dct_8, // .stage_range_row
fwd_cos_bit_col_adst_dct_8, // .cos_bit_col
fwd_cos_bit_row_adst_dct_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_16 ----------------
static const int8_t fwd_shift_adst_dct_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_adst_dct_16[10] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 19
};
static const int8_t fwd_stage_range_row_adst_dct_16[8] = { 17, 18, 19, 20,
20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_adst_dct_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_16 = {
16, // .txfm_size
10, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_16, // .shift
fwd_stage_range_col_adst_dct_16, // .stage_range_col
fwd_stage_range_row_adst_dct_16, // .stage_range_row
fwd_cos_bit_col_adst_dct_16, // .cos_bit_col
fwd_cos_bit_row_adst_dct_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_32 ----------------
static const int8_t fwd_shift_adst_dct_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_adst_dct_32[12] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_stage_range_row_adst_dct_32[10] = {
16, 17, 18, 19, 20, 20, 20, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_dct_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const int8_t fwd_cos_bit_row_adst_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_32 = {
32, // .txfm_size
12, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
fwd_shift_adst_dct_32, // .shift
fwd_stage_range_col_adst_dct_32, // .stage_range_col
fwd_stage_range_row_adst_dct_32, // .stage_range_row
fwd_cos_bit_col_adst_dct_32, // .cos_bit_col
fwd_cos_bit_row_adst_dct_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
#endif // AV1_FWD_TXFM2D_CFG_H_

124
third_party/aom/av1/common/av1_inv_txfm1d.c vendored
View file

@ -10,6 +10,7 @@
*/
#include <stdlib.h>
#include "aom_dsp/inv_txfm.h"
#include "av1/common/av1_inv_txfm1d.h"
#if CONFIG_COEFFICIENT_RANGE_CHECKING
@ -66,6 +67,7 @@ void av1_idct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = input[2];
@ -75,7 +77,7 @@ void av1_idct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -109,6 +111,7 @@ void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = input[4];
@ -122,7 +125,7 @@ void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -137,7 +140,7 @@ void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -152,7 +155,7 @@ void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[3];
@ -194,6 +197,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = input[8];
@ -215,7 +219,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -238,7 +242,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -261,7 +265,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -284,7 +288,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[3];
@ -307,7 +311,7 @@ void av1_idct16_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[7];
@ -365,6 +369,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = input[16];
@ -402,7 +407,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -441,7 +446,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -480,7 +485,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -519,7 +524,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -558,7 +563,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[3];
@ -597,7 +602,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 7
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[7];
@ -636,7 +641,7 @@ void av1_idct32_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[15];
@ -726,6 +731,7 @@ void av1_iadst4_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = -input[3];
@ -735,7 +741,7 @@ void av1_iadst4_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -756,7 +762,7 @@ void av1_iadst4_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[8], bf0[0], cospi[56], bf0[1], cos_bit[stage]);
@ -790,6 +796,7 @@ void av1_iadst8_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = -input[7];
@ -803,7 +810,7 @@ void av1_iadst8_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -832,7 +839,7 @@ void av1_iadst8_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -861,7 +868,7 @@ void av1_iadst8_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
@ -903,6 +910,7 @@ void av1_iadst16_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = -input[15];
@ -924,7 +932,7 @@ void av1_iadst16_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -969,7 +977,7 @@ void av1_iadst16_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1014,7 +1022,7 @@ void av1_iadst16_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1059,7 +1067,7 @@ void av1_iadst16_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
@ -1117,6 +1125,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = -input[31];
@ -1154,7 +1163,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1231,7 +1240,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1308,7 +1317,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1385,7 +1394,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1462,7 +1471,7 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
// stage 10
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[1], bf0[0], cospi[63], bf0[1], cos_bit[stage]);
@ -1538,6 +1547,38 @@ void av1_iadst32_new(const int32_t *input, int32_t *output,
range_check(stage, input, bf1, size, stage_range[stage]);
}
#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)cos_bit;
for (int i = 0; i < 4; ++i)
output[i] = (int32_t)dct_const_round_shift(input[i] * Sqrt2);
range_check(0, input, output, 4, stage_range[0]);
}
void av1_iidentity8_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 8; ++i) output[i] = input[i] * 2;
range_check(0, input, output, 8, stage_range[0]);
}
void av1_iidentity16_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 16; ++i)
output[i] = (int32_t)dct_const_round_shift(input[i] * 2 * Sqrt2);
range_check(0, input, output, 16, stage_range[0]);
}
void av1_iidentity32_c(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
(void)cos_bit;
for (int i = 0; i < 32; ++i) output[i] = input[i] * 4;
range_check(0, input, output, 32, stage_range[0]);
}
#endif // CONFIG_EXT_TX
#if CONFIG_TX64X64
void av1_idct64_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range) {
@ -1553,7 +1594,8 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 1;
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
assert(output != input);
bf1 = output;
bf1[0] = input[0];
bf1[1] = input[32];
@ -1623,7 +1665,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 2
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1694,7 +1736,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 3
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -1765,7 +1807,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 4
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0];
@ -1836,7 +1878,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 5
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0];
@ -1907,7 +1949,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 6
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = half_btf(cospi[32], bf0[0], cospi[32], bf0[1], cos_bit[stage]);
@ -1978,7 +2020,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 7
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[3];
@ -2049,7 +2091,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 8
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[7];
@ -2120,7 +2162,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 9
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[15];
@ -2191,7 +2233,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 10
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = output;
bf1 = step;
bf1[0] = bf0[0] + bf0[31];
@ -2262,7 +2304,7 @@ void av1_idct64_new(const int32_t *input, int32_t *output,
// stage 11
stage++;
cospi = cospi_arr[cos_bit[stage] - cos_bit_min];
cospi = cospi_arr(cos_bit[stage]);
bf0 = step;
bf1 = output;
bf1[0] = bf0[0] + bf0[63];

10
third_party/aom/av1/common/av1_inv_txfm1d.h vendored
View file

@ -37,6 +37,16 @@ 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);
#endif // CONFIG_EXT_TX
#ifdef __cplusplus
}

362
third_party/aom/av1/common/av1_inv_txfm1d_cfg.h vendored Normal file
View file

@ -0,0 +1,362 @@
/*
* 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_INV_TXFM2D_CFG_H_
#define AV1_INV_TXFM2D_CFG_H_
#include "av1/common/av1_inv_txfm1d.h"
// Identity will always use max bitdepth regardless of size
static const int8_t inv_stage_range_identity[1] = { 12 };
// ---------------- 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] = { 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_dct_4[4] = { 18, 18, 18, 18 };
static const int8_t inv_stage_range_col_adst_4[6] = { 18, 18, 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_adst_4[6] = { 18, 18, 18, 18, 18, 18 };
// 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] = { 19, 19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_8[6] = { 19, 19, 19, 19, 19, 19 };
static const int8_t inv_stage_range_col_adst_8[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_8[8] = { 19, 19, 19, 19,
19, 19, 19, 19 };
// 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
};
// ---------------- 16x16 1D constants -----------------------
// shift
static const int8_t inv_shift_16[2] = { -1, -5 };
// stage range
static const int8_t inv_stage_range_col_dct_16[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_16[8] = { 20, 20, 20, 20,
20, 20, 20, 20 };
static const int8_t inv_stage_range_col_adst_16[10] = { 19, 19, 19, 19, 19,
19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_16[10] = { 20, 20, 20, 20, 20,
20, 20, 20, 20, 20 };
// 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] = { 19, 19, 19, 19, 19,
19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_32[10] = { 20, 20, 20, 20, 20,
20, 20, 20, 20, 20 };
static const int8_t inv_stage_range_col_adst_32[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_32[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
// 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, -7 };
// stage range
static const int8_t inv_stage_range_col_dct_64[12] = { 19, 19, 19, 19, 19, 19,
19, 19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_64[12] = { 20, 20, 20, 20, 20, 20,
20, 20, 20, 20, 20, 20 };
// 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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 1, // .log_scale
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
};
// ---------------- 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
};
// ---------------- row config inv_adst_4 ----------------
static const TXFM_1D_CFG inv_txfm_1d_row_cfg_adst_4 = {
4, // .txfm_size
6, // .stage_num
// 0, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 1, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 1, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 0, // .log_scale
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
// 1, // .log_scale
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
// 0, // .log_scale
inv_shift_4, // .shift
inv_stage_range_identity, // .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
// 0, // .log_scale
inv_shift_8, // .shift
inv_stage_range_identity, // .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
// 0, // .log_scale
inv_shift_16, // .shift
inv_stage_range_identity, // .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
// 1, // .log_scale
inv_shift_32, // .shift
inv_stage_range_identity, // .stage_range
NULL, // .cos_bit
TXFM_TYPE_IDENTITY32, // .txfm_type
};
#endif // CONFIG_EXT_TX
#endif // AV1_INV_TXFM2D_CFG_H_

167
third_party/aom/av1/common/av1_inv_txfm2d.c vendored
View file

@ -13,7 +13,7 @@
#include "av1/common/enums.h"
#include "av1/common/av1_txfm.h"
#include "av1/common/av1_inv_txfm1d.h"
#include "av1/common/av1_inv_txfm2d_cfg.h"
#include "av1/common/av1_inv_txfm1d_cfg.h"
static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
switch (txfm_type) {
@ -25,132 +25,100 @@ static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
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;
#endif // CONFIG_EXT_TX
default: assert(0); return NULL;
}
}
const TXFM_2D_CFG *inv_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
// DCT_DCT
static const TXFM_1D_CFG *inv_txfm_col_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
// DCT
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_dct_dct_4, &inv_txfm_2d_cfg_dct_dct_8,
&inv_txfm_2d_cfg_dct_dct_16, &inv_txfm_2d_cfg_dct_dct_32 },
// ADST_DCT
&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 },
// ADST
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
// DCT_ADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
// ADST_ADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
&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_EXT_TX
// FLIPADST_DCT
// FLIPADST
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
// DCT_FLIPADST
&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 },
// IDENTITY
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
// FLIPADST_FLIPADST
&inv_txfm_1d_cfg_identity_4, &inv_txfm_1d_cfg_identity_8,
&inv_txfm_1d_cfg_identity_16, &inv_txfm_1d_cfg_identity_32 },
#endif // CONFIG_EXT_TX
};
static const TXFM_1D_CFG *inv_txfm_row_cfg_ls[TX_TYPES_1D][TX_SIZES] = {
// DCT
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
// ADST_FLIPADST
&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 },
// ADST
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
// FLIPADST_ADST
&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_EXT_TX
// FLIPADST
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // IDTX
#if CONFIG_CB4X4
NULL,
&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 },
// IDENTITY
{
#if CONFIG_CHROMA_2X2
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // V_DCT
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
{ // H_DCT
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
{ // V_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // H_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // V_FLIP_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // H_FLIP_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
&inv_txfm_1d_cfg_identity_4, &inv_txfm_1d_cfg_identity_8,
&inv_txfm_1d_cfg_identity_16, &inv_txfm_1d_cfg_identity_32 },
#endif // CONFIG_EXT_TX
};
TXFM_2D_FLIP_CFG av1_get_inv_txfm_cfg(int tx_type, int tx_size) {
TXFM_2D_FLIP_CFG cfg;
set_flip_cfg(tx_type, &cfg);
cfg.cfg = inv_txfm_cfg_ls[tx_type][tx_size];
int tx_type_col = vtx_tab[tx_type];
int tx_type_row = htx_tab[tx_type];
// TODO(sarahparker) this is currently only implemented for
// square transforms
cfg.col_cfg = inv_txfm_col_cfg_ls[tx_type_col][tx_size];
cfg.row_cfg = inv_txfm_row_cfg_ls[tx_type_row][tx_size];
return cfg;
}
TXFM_2D_FLIP_CFG av1_get_inv_txfm_64x64_cfg(int tx_type) {
TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL };
TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL, NULL };
switch (tx_type) {
case DCT_DCT:
cfg.cfg = &inv_txfm_2d_cfg_dct_dct_64;
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);
@ -161,14 +129,15 @@ TXFM_2D_FLIP_CFG av1_get_inv_txfm_64x64_cfg(int tx_type) {
static INLINE void inv_txfm2d_add_c(const int32_t *input, int16_t *output,
int stride, TXFM_2D_FLIP_CFG *cfg,
int32_t *txfm_buf) {
const int txfm_size = cfg->cfg->txfm_size;
const int8_t *shift = cfg->cfg->shift;
const int8_t *stage_range_col = cfg->cfg->stage_range_col;
const int8_t *stage_range_row = cfg->cfg->stage_range_row;
const int8_t *cos_bit_col = cfg->cfg->cos_bit_col;
const int8_t *cos_bit_row = cfg->cfg->cos_bit_row;
const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->cfg->txfm_type_col);
const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->cfg->txfm_type_row);
// TODO(sarahparker) must correct for rectangular transforms in follow up
const int txfm_size = cfg->row_cfg->txfm_size;
const int8_t *shift = cfg->row_cfg->shift;
const int8_t *stage_range_col = cfg->col_cfg->stage_range;
const int8_t *stage_range_row = cfg->row_cfg->stage_range;
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);
// txfm_buf's length is txfm_size * txfm_size + 2 * txfm_size
// it is used for intermediate data buffering
@ -216,7 +185,11 @@ static INLINE void inv_txfm2d_add_facade(const int32_t *input, uint16_t *output,
// int16_t*
TXFM_2D_FLIP_CFG cfg = av1_get_inv_txfm_cfg(tx_type, tx_size);
inv_txfm2d_add_c(input, (int16_t *)output, stride, &cfg, txfm_buf);
clamp_block((int16_t *)output, cfg.cfg->txfm_size, stride, 0, (1 << bd) - 1);
// TODO(sarahparker) just using the cfg_row->txfm_size for now because
// we are assumint this is only used for square transforms. This will
// be adjusted in a follow up
clamp_block((int16_t *)output, cfg.row_cfg->txfm_size, stride, 0,
(1 << bd) - 1);
}
void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,

447
third_party/aom/av1/common/av1_inv_txfm2d_cfg.h vendored
View file

@ -1,447 +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_INV_TXFM2D_CFG_H_
#define AV1_INV_TXFM2D_CFG_H_
#include "av1/common/av1_inv_txfm1d.h"
// ---------------- config inv_dct_dct_4 ----------------
static const int8_t inv_shift_dct_dct_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_dct_dct_4[4] = { 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_dct_dct_4[4] = { 18, 18, 18, 18 };
static const int8_t inv_cos_bit_col_dct_dct_4[4] = { 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_4 = {
4, // .txfm_size
4, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_4, // .shift
inv_stage_range_col_dct_dct_4, // .stage_range_col
inv_stage_range_row_dct_dct_4, // .stage_range_row
inv_cos_bit_col_dct_dct_4, // .cos_bit_col
inv_cos_bit_row_dct_dct_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config inv_dct_dct_8 ----------------
static const int8_t inv_shift_dct_dct_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_dct_dct_8[6] = {
19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_dct_8[6] = {
19, 19, 19, 19, 19, 19
};
static const int8_t inv_cos_bit_col_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_8 = {
8, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_8, // .shift
inv_stage_range_col_dct_dct_8, // .stage_range_col
inv_stage_range_row_dct_dct_8, // .stage_range_row
inv_cos_bit_col_dct_dct_8, // .cos_bit_col
inv_cos_bit_row_dct_dct_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config inv_dct_dct_16 ----------------
static const int8_t inv_shift_dct_dct_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_dct_16[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_dct_16[8] = { 20, 20, 20, 20,
20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_dct_dct_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_16 = {
16, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_16, // .shift
inv_stage_range_col_dct_dct_16, // .stage_range_col
inv_stage_range_row_dct_dct_16, // .stage_range_row
inv_cos_bit_col_dct_dct_16, // .cos_bit_col
inv_cos_bit_row_dct_dct_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config inv_dct_dct_32 ----------------
static const int8_t inv_shift_dct_dct_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_dct_32[10] = { 19, 19, 19, 19, 19,
19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_dct_32[10] = { 20, 20, 20, 20, 20,
20, 20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_dct_dct_32[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_32 = {
32, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
inv_shift_dct_dct_32, // .shift
inv_stage_range_col_dct_dct_32, // .stage_range_col
inv_stage_range_row_dct_dct_32, // .stage_range_row
inv_cos_bit_col_dct_dct_32, // .cos_bit_col
inv_cos_bit_row_dct_dct_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
// ---------------- config inv_dct_dct_64 ----------------
static const int8_t inv_shift_dct_dct_64[2] = { -1, -7 };
static const int8_t inv_stage_range_col_dct_dct_64[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_dct_64[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_dct_64[12] = { 13, 13, 13, 13, 13, 13,
13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_64[12] = { 12, 12, 12, 12, 12, 12,
12, 12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_64 = {
64, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
inv_shift_dct_dct_64, // .shift
inv_stage_range_col_dct_dct_64, // .stage_range_col
inv_stage_range_row_dct_dct_64, // .stage_range_row
inv_cos_bit_col_dct_dct_64, // .cos_bit_col
inv_cos_bit_row_dct_dct_64, // .cos_bit_row
TXFM_TYPE_DCT64, // .txfm_type_col
TXFM_TYPE_DCT64
}; // .txfm_type_row
// ---------------- config inv_dct_adst_4 ----------------
static const int8_t inv_shift_dct_adst_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_dct_adst_4[4] = { 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_dct_adst_4[6] = {
18, 18, 18, 18, 18, 18
};
static const int8_t inv_cos_bit_col_dct_adst_4[4] = { 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_4 = {
4, // .txfm_size
4, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_4, // .shift
inv_stage_range_col_dct_adst_4, // .stage_range_col
inv_stage_range_row_dct_adst_4, // .stage_range_row
inv_cos_bit_col_dct_adst_4, // .cos_bit_col
inv_cos_bit_row_dct_adst_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config inv_dct_adst_8 ----------------
static const int8_t inv_shift_dct_adst_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_dct_adst_8[6] = {
19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_adst_8[8] = { 19, 19, 19, 19,
19, 19, 19, 19 };
static const int8_t inv_cos_bit_col_dct_adst_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_8 = {
8, // .txfm_size
6, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_8, // .shift
inv_stage_range_col_dct_adst_8, // .stage_range_col
inv_stage_range_row_dct_adst_8, // .stage_range_row
inv_cos_bit_col_dct_adst_8, // .cos_bit_col
inv_cos_bit_row_dct_adst_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config inv_dct_adst_16 ----------------
static const int8_t inv_shift_dct_adst_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_adst_16[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_adst_16[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_adst_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_16 = {
16, // .txfm_size
8, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_16, // .shift
inv_stage_range_col_dct_adst_16, // .stage_range_col
inv_stage_range_row_dct_adst_16, // .stage_range_row
inv_cos_bit_col_dct_adst_16, // .cos_bit_col
inv_cos_bit_row_dct_adst_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config inv_dct_adst_32 ----------------
static const int8_t inv_shift_dct_adst_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_adst_32[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_adst_32[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_adst_32[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_32 = {
32, // .txfm_size
10, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
inv_shift_dct_adst_32, // .shift
inv_stage_range_col_dct_adst_32, // .stage_range_col
inv_stage_range_row_dct_adst_32, // .stage_range_row
inv_cos_bit_col_dct_adst_32, // .cos_bit_col
inv_cos_bit_row_dct_adst_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config inv_adst_adst_4 ----------------
static const int8_t inv_shift_adst_adst_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_adst_adst_4[6] = { 18, 18, 18,
18, 17, 17 };
static const int8_t inv_stage_range_row_adst_adst_4[6] = { 18, 18, 18,
18, 18, 18 };
static const int8_t inv_cos_bit_col_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_4 = {
4, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_4, // .shift
inv_stage_range_col_adst_adst_4, // .stage_range_col
inv_stage_range_row_adst_adst_4, // .stage_range_row
inv_cos_bit_col_adst_adst_4, // .cos_bit_col
inv_cos_bit_row_adst_adst_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config inv_adst_adst_8 ----------------
static const int8_t inv_shift_adst_adst_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_adst_adst_8[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_adst_8[8] = { 19, 19, 19, 19,
19, 19, 19, 19 };
static const int8_t inv_cos_bit_col_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_8 = {
8, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_8, // .shift
inv_stage_range_col_adst_adst_8, // .stage_range_col
inv_stage_range_row_adst_adst_8, // .stage_range_row
inv_cos_bit_col_adst_adst_8, // .cos_bit_col
inv_cos_bit_row_adst_adst_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config inv_adst_adst_16 ----------------
static const int8_t inv_shift_adst_adst_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_adst_16[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_adst_16[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_adst_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_16 = {
16, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_16, // .shift
inv_stage_range_col_adst_adst_16, // .stage_range_col
inv_stage_range_row_adst_adst_16, // .stage_range_row
inv_cos_bit_col_adst_adst_16, // .cos_bit_col
inv_cos_bit_row_adst_adst_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config inv_adst_adst_32 ----------------
static const int8_t inv_shift_adst_adst_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_adst_32[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_adst_32[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_adst_32[12] = {
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13
};
static const int8_t inv_cos_bit_row_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_32 = {
32, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
inv_shift_adst_adst_32, // .shift
inv_stage_range_col_adst_adst_32, // .stage_range_col
inv_stage_range_row_adst_adst_32, // .stage_range_row
inv_cos_bit_col_adst_adst_32, // .cos_bit_col
inv_cos_bit_row_adst_adst_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config inv_adst_dct_4 ----------------
static const int8_t inv_shift_adst_dct_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_adst_dct_4[6] = {
18, 18, 18, 18, 17, 17
};
static const int8_t inv_stage_range_row_adst_dct_4[4] = { 18, 18, 18, 18 };
static const int8_t inv_cos_bit_col_adst_dct_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_4 = {
4, // .txfm_size
6, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_4, // .shift
inv_stage_range_col_adst_dct_4, // .stage_range_col
inv_stage_range_row_adst_dct_4, // .stage_range_row
inv_cos_bit_col_adst_dct_4, // .cos_bit_col
inv_cos_bit_row_adst_dct_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config inv_adst_dct_8 ----------------
static const int8_t inv_shift_adst_dct_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_adst_dct_8[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_dct_8[6] = {
19, 19, 19, 19, 19, 19
};
static const int8_t inv_cos_bit_col_adst_dct_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_8 = {
8, // .txfm_size
8, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_8, // .shift
inv_stage_range_col_adst_dct_8, // .stage_range_col
inv_stage_range_row_adst_dct_8, // .stage_range_row
inv_cos_bit_col_adst_dct_8, // .cos_bit_col
inv_cos_bit_row_adst_dct_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config inv_adst_dct_16 ----------------
static const int8_t inv_shift_adst_dct_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_dct_16[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_dct_16[8] = { 20, 20, 20, 20,
20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_adst_dct_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_16 = {
16, // .txfm_size
10, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_16, // .shift
inv_stage_range_col_adst_dct_16, // .stage_range_col
inv_stage_range_row_adst_dct_16, // .stage_range_row
inv_cos_bit_col_adst_dct_16, // .cos_bit_col
inv_cos_bit_row_adst_dct_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config inv_adst_dct_32 ----------------
static const int8_t inv_shift_adst_dct_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_dct_32[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_dct_32[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_dct_32[12] = {
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13
};
static const int8_t inv_cos_bit_row_adst_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_32 = {
32, // .txfm_size
12, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
inv_shift_adst_dct_32, // .shift
inv_stage_range_col_adst_dct_32, // .stage_range_col
inv_stage_range_row_adst_dct_32, // .stage_range_row
inv_cos_bit_col_adst_dct_32, // .cos_bit_col
inv_cos_bit_row_adst_dct_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
extern const TXFM_2D_CFG *inv_txfm_cfg_ls[TX_TYPES][TX_SIZES];
#endif // AV1_INV_TXFM2D_CFG_H_

343
third_party/aom/av1/common/av1_loopfilter.c vendored
View file

@ -22,7 +22,7 @@
#include "av1/common/seg_common.h"
#define CONFIG_PARALLEL_DEBLOCKING_15TAPLUMAONLY 0
#define PARALLEL_DEBLOCKING_15TAPLUMAONLY 1
// 64 bit masks for left transform size. Each 1 represents a position where
// we should apply a loop filter across the left border of an 8x8 block
@ -42,7 +42,7 @@
//
// A loopfilter should be applied to every other 8x8 horizontally.
static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
0xffffffffffffffffULL, // TX_2X2
#endif
0xffffffffffffffffULL, // TX_4X4
@ -72,7 +72,7 @@ static const uint64_t left_64x64_txform_mask[TX_SIZES] = {
//
// A loopfilter should be applied to every other 4 the row vertically.
static const uint64_t above_64x64_txform_mask[TX_SIZES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
0xffffffffffffffffULL, // TX_4X4
#endif
0xffffffffffffffffULL, // TX_4X4
@ -171,7 +171,7 @@ static const uint64_t above_border = 0x000000ff000000ffULL;
// 16 bit masks for uv transform sizes.
static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
0xffff, // TX_2X2
#endif
0xffff, // TX_4X4
@ -184,7 +184,7 @@ static const uint16_t left_64x64_txform_mask_uv[TX_SIZES] = {
};
static const uint16_t above_64x64_txform_mask_uv[TX_SIZES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
0xffff, // TX_2X2
#endif
0xffff, // TX_4X4
@ -267,11 +267,14 @@ static const int mode_lf_lut[] = {
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // INTRA_MODES
#if CONFIG_ALT_INTRA
0,
#endif
#if CONFIG_SMOOTH_HV
0, 0,
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
1, 1, 0, 1, // INTER_MODES (ZEROMV == 0)
#if CONFIG_EXT_INTER
1, 1, 1, 1, 1, 1, 1, 1, 0, 1 // INTER_COMPOUND_MODES (ZERO_ZEROMV == 0)
#endif // CONFIG_EXT_INTER
1, 1, 1, 1, 1, 1, 0, 1 // INTER_COMPOUND_MODES (ZERO_ZEROMV == 0)
#endif // CONFIG_EXT_INTER
};
static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
@ -345,8 +348,6 @@ static uint8_t get_filter_level(const loop_filter_info_n *lfi_n,
}
#endif
#define NELEMENTS(x) (sizeof((x)) / sizeof((x)[0]))
void av1_loop_filter_init(AV1_COMMON *cm) {
assert(MB_MODE_COUNT == NELEMENTS(mode_lf_lut));
loop_filter_info_n *lfi = &cm->lf_info;
@ -1362,9 +1363,10 @@ typedef struct {
// the non420 case).
// Note: 'row_masks_ptr' and/or 'col_masks_ptr' can be passed NULL.
static void get_filter_level_and_masks_non420(
AV1_COMMON *const cm, const struct macroblockd_plane *const plane,
AV1_COMMON *const cm, const struct macroblockd_plane *const plane, int pl,
MODE_INFO **mib, int mi_row, int mi_col, int idx_r, uint8_t *const lfl_r,
unsigned int *const mask_4x4_int_r, FilterMasks *const row_masks_ptr,
unsigned int *const mask_4x4_int_r_ptr,
unsigned int *const mask_4x4_int_c_ptr, FilterMasks *const row_masks_ptr,
FilterMasks *const col_masks_ptr) {
const int ss_x = plane->subsampling_x;
const int ss_y = plane->subsampling_y;
@ -1372,7 +1374,7 @@ static void get_filter_level_and_masks_non420(
FilterMasks row_masks, col_masks;
memset(&row_masks, 0, sizeof(row_masks));
memset(&col_masks, 0, sizeof(col_masks));
*mask_4x4_int_r = 0;
unsigned int mask_4x4_int_r = 0, mask_4x4_int_c = 0;
const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
// Determine the vertical edges that need filtering
@ -1398,10 +1400,6 @@ static void get_filter_level_and_masks_non420(
(num_4x4_blocks_high_lookup[sb_type] > 1) ? !blk_row : 1;
const int skip_this_r = skip_this && !block_edge_above;
#if CONFIG_VAR_TX
const TX_SIZE mb_tx_size = mbmi->inter_tx_size[blk_row][blk_col];
#endif
TX_SIZE tx_size = (plane->plane_type == PLANE_TYPE_UV)
? get_uv_tx_size(mbmi, plane)
: mbmi->tx_size;
@ -1411,9 +1409,6 @@ static void get_filter_level_and_masks_non420(
const int skip_border_4x4_r =
ss_y && mi_row + idx_r >= cm->mi_rows - mi_size_high[BLOCK_8X8];
TX_SIZE tx_size_c = txsize_horz_map[tx_size];
TX_SIZE tx_size_r = txsize_vert_map[tx_size];
int tx_size_mask = 0;
const int c_step = (c >> ss_x);
const int r_step = (r >> ss_y);
@ -1421,8 +1416,15 @@ static void get_filter_level_and_masks_non420(
#if CONFIG_VAR_TX
if (is_inter_block(mbmi) && !mbmi->skip) {
const int tx_row_idx =
(blk_row * mi_size_high[BLOCK_8X8] << TX_UNIT_HIGH_LOG2) >> 1;
const int tx_col_idx =
(blk_col * mi_size_wide[BLOCK_8X8] << TX_UNIT_WIDE_LOG2) >> 1;
const BLOCK_SIZE bsize =
AOMMAX(BLOCK_4X4, get_plane_block_size(mbmi->sb_type, plane));
const TX_SIZE mb_tx_size = mbmi->inter_tx_size[tx_row_idx][tx_col_idx];
tx_size = (plane->plane_type == PLANE_TYPE_UV)
? uv_txsize_lookup[sb_type][mb_tx_size][ss_x][ss_y]
? uv_txsize_lookup[bsize][mb_tx_size][0][0]
: mb_tx_size;
}
#endif
@ -1435,12 +1437,29 @@ static void get_filter_level_and_masks_non420(
#endif
#if CONFIG_VAR_TX
tx_size_r = AOMMIN(tx_size, cm->above_txfm_context[mi_col + c]);
tx_size_c =
AOMMIN(tx_size, cm->left_txfm_context[(mi_row + r) & MAX_MIB_MASK]);
TX_SIZE tx_size_r, tx_size_c;
cm->above_txfm_context[mi_col + c] = tx_size;
cm->left_txfm_context[(mi_row + r) & MAX_MIB_MASK] = tx_size;
const int tx_wide =
AOMMIN(tx_size_wide[tx_size],
tx_size_wide[cm->top_txfm_context[pl][(mi_col + idx_c)
<< TX_UNIT_WIDE_LOG2]]);
const int tx_high = AOMMIN(
tx_size_high[tx_size],
tx_size_high[cm->left_txfm_context[pl][((mi_row + idx_r) & MAX_MIB_MASK)
<< TX_UNIT_HIGH_LOG2]]);
tx_size_c = get_sqr_tx_size(tx_wide);
tx_size_r = get_sqr_tx_size(tx_high);
memset(cm->top_txfm_context[pl] + ((mi_col + idx_c) << TX_UNIT_WIDE_LOG2),
tx_size, mi_size_wide[BLOCK_8X8] << TX_UNIT_WIDE_LOG2);
memset(cm->left_txfm_context[pl] +
(((mi_row + idx_r) & MAX_MIB_MASK) << TX_UNIT_HIGH_LOG2),
tx_size, mi_size_high[BLOCK_8X8] << TX_UNIT_HIGH_LOG2);
#else
TX_SIZE tx_size_c = txsize_horz_map[tx_size];
TX_SIZE tx_size_r = txsize_vert_map[tx_size];
(void)pl;
#endif // CONFIG_VAR_TX
if (tx_size_c == TX_32X32)
@ -1477,7 +1496,7 @@ static void get_filter_level_and_masks_non420(
if (!skip_this && tx_size_c < TX_8X8 && !skip_border_4x4_c &&
(c_step & tx_size_mask) == 0)
*mask_4x4_int_r |= col_mask;
mask_4x4_int_c |= col_mask;
}
if (tx_size_r == TX_32X32)
@ -1511,20 +1530,22 @@ static void get_filter_level_and_masks_non420(
row_masks.m4x4 |= col_mask;
}
if (!skip_this && tx_size_r < TX_8X8 && !skip_border_4x4_c &&
if (!skip_this && tx_size_r < TX_8X8 && !skip_border_4x4_r &&
((r >> ss_y) & tx_size_mask) == 0)
*mask_4x4_int_r |= col_mask;
mask_4x4_int_r |= col_mask;
}
}
if (row_masks_ptr) *row_masks_ptr = row_masks;
if (col_masks_ptr) *col_masks_ptr = col_masks;
if (mask_4x4_int_c_ptr) *mask_4x4_int_c_ptr = mask_4x4_int_c;
if (mask_4x4_int_r_ptr) *mask_4x4_int_r_ptr = mask_4x4_int_r;
}
void av1_filter_block_plane_non420_ver(AV1_COMMON *const cm,
struct macroblockd_plane *plane,
MODE_INFO **mib, int mi_row,
int mi_col) {
MODE_INFO **mib, int mi_row, int mi_col,
int pl) {
const int ss_y = plane->subsampling_y;
const int row_step = mi_size_high[BLOCK_8X8] << ss_y;
struct buf_2d *const dst = &plane->dst;
@ -1537,8 +1558,8 @@ void av1_filter_block_plane_non420_ver(AV1_COMMON *const cm,
unsigned int mask_4x4_int;
FilterMasks col_masks;
const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
get_filter_level_and_masks_non420(cm, plane, mib, mi_row, mi_col, idx_r,
&lfl[r][0], &mask_4x4_int, NULL,
get_filter_level_and_masks_non420(cm, plane, pl, mib, mi_row, mi_col, idx_r,
&lfl[r][0], NULL, &mask_4x4_int, NULL,
&col_masks);
// Disable filtering on the leftmost column or tile boundary
@ -1572,8 +1593,8 @@ void av1_filter_block_plane_non420_ver(AV1_COMMON *const cm,
void av1_filter_block_plane_non420_hor(AV1_COMMON *const cm,
struct macroblockd_plane *plane,
MODE_INFO **mib, int mi_row,
int mi_col) {
MODE_INFO **mib, int mi_row, int mi_col,
int pl) {
const int ss_y = plane->subsampling_y;
const int row_step = mi_size_high[BLOCK_8X8] << ss_y;
struct buf_2d *const dst = &plane->dst;
@ -1585,16 +1606,13 @@ void av1_filter_block_plane_non420_hor(AV1_COMMON *const cm,
for (idx_r = 0; idx_r < cm->mib_size && mi_row + idx_r < cm->mi_rows;
idx_r += row_step) {
const int r = idx_r >> mi_height_log2_lookup[BLOCK_8X8];
get_filter_level_and_masks_non420(cm, plane, mib, mi_row, mi_col, idx_r,
&lfl[r][0], mask_4x4_int + r,
get_filter_level_and_masks_non420(cm, plane, pl, mib, mi_row, mi_col, idx_r,
&lfl[r][0], mask_4x4_int + r, NULL,
row_masks_array + r, NULL);
}
for (idx_r = 0; idx_r < cm->mib_size && mi_row + idx_r < cm->mi_rows;
idx_r += row_step) {
const int skip_border_4x4_r =
ss_y && mi_row + idx_r >= cm->mi_rows - mi_size_wide[BLOCK_8X8];
const int r = idx_r >> mi_width_log2_lookup[BLOCK_8X8];
const unsigned int mask_4x4_int_r = skip_border_4x4_r ? 0 : mask_4x4_int[r];
FilterMasks row_masks;
#if CONFIG_LOOPFILTERING_ACROSS_TILES
@ -1615,12 +1633,12 @@ void av1_filter_block_plane_non420_hor(AV1_COMMON *const cm,
if (cm->use_highbitdepth)
highbd_filter_selectively_horiz(
CONVERT_TO_SHORTPTR(dst->buf), dst->stride, row_masks.m16x16,
row_masks.m8x8, row_masks.m4x4, mask_4x4_int_r, &cm->lf_info,
row_masks.m8x8, row_masks.m4x4, mask_4x4_int[r], &cm->lf_info,
&lfl[r][0], (int)cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
filter_selectively_horiz(dst->buf, dst->stride, row_masks.m16x16,
row_masks.m8x8, row_masks.m4x4, mask_4x4_int_r,
row_masks.m8x8, row_masks.m4x4, mask_4x4_int[r],
&cm->lf_info, &lfl[r][0]);
dst->buf += 8 * dst->stride;
}
@ -1839,8 +1857,6 @@ void av1_filter_block_plane_ss11_hor(AV1_COMMON *const cm,
dst->buf = dst0;
}
#if !(CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES || \
CONFIG_CB4X4)
#if CONFIG_PARALLEL_DEBLOCKING
typedef enum EDGE_DIR { VERT_EDGE = 0, HORZ_EDGE = 1, NUM_EDGE_DIRS } EDGE_DIR;
static const uint32_t av1_prediction_masks[NUM_EDGE_DIRS][BLOCK_SIZES] = {
@ -1900,7 +1916,7 @@ static const uint32_t av1_prediction_masks[NUM_EDGE_DIRS][BLOCK_SIZES] = {
static const uint32_t av1_transform_masks[NUM_EDGE_DIRS][TX_SIZES_ALL] = {
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
2 - 1, // TX_2X2
#endif
4 - 1, // TX_4X4
@ -1922,7 +1938,7 @@ static const uint32_t av1_transform_masks[NUM_EDGE_DIRS][TX_SIZES_ALL] = {
32 - 1 // TX_32X8
},
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
2 - 1, // TX_2X2
#endif
4 - 1, // TX_4X4
@ -1992,10 +2008,17 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
// not sure if changes are required.
assert(0 && "Not yet updated");
#endif // CONFIG_EXT_PARTITION
{
const TX_SIZE ts =
av1_get_transform_size(ppCurr[0], edgeDir, scaleHorz, scaleVert);
#if CONFIG_EXT_DELTA_Q
const uint32_t currLevel =
get_filter_level(cm, &cm->lf_info, &ppCurr[0]->mbmi);
#else
const uint32_t currLevel = get_filter_level(&cm->lf_info, &ppCurr[0]->mbmi);
#endif // CONFIG_EXT_DELTA_Q
const int currSkipped =
ppCurr[0]->mbmi.skip && is_inter_block(&ppCurr[0]->mbmi);
const uint32_t coord = (VERT_EDGE == edgeDir) ? (x) : (y);
@ -2016,7 +2039,13 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
const MODE_INFO *const pPrev = *(ppCurr - modeStep);
const TX_SIZE pvTs =
av1_get_transform_size(pPrev, edgeDir, scaleHorz, scaleVert);
#if CONFIG_EXT_DELTA_Q
const uint32_t pvLvl =
get_filter_level(cm, &cm->lf_info, &pPrev->mbmi);
#else
const uint32_t pvLvl = get_filter_level(&cm->lf_info, &pPrev->mbmi);
#endif // CONFIG_EXT_DELTA_Q
const int pvSkip = pPrev->mbmi.skip && is_inter_block(&pPrev->mbmi);
const int32_t puEdge =
(coord &
@ -2028,7 +2057,7 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
// if the current and the previous blocks are skipped,
// deblock the edge if the edge belongs to a PU's edge only.
if ((currLevel || pvLvl) && (!pvSkip || !currSkipped || puEdge)) {
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || CONFIG_PARALLEL_DEBLOCKING_15TAPLUMAONLY
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
const TX_SIZE minTs = AOMMIN(ts, pvTs);
if (TX_4X4 >= minTs) {
pParams->filterLength = 4;
@ -2036,7 +2065,7 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
pParams->filterLength = 8;
} else {
pParams->filterLength = 16;
#if CONFIG_PARALLEL_DEBLOCKING_15TAPLUMAONLY
#if PARALLEL_DEBLOCKING_15TAPLUMAONLY
// No wide filtering for chroma plane
if (scaleHorz || scaleVert) {
pParams->filterLength = 8;
@ -2046,7 +2075,7 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
#else
pParams->filterLength = (TX_4X4 >= AOMMIN(ts, pvTs)) ? (4) : (8);
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
// update the level if the current block is skipped,
// but the previous one is not
@ -2054,10 +2083,14 @@ static void set_lpf_parameters(AV1_DEBLOCKING_PARAMETERS *const pParams,
}
}
}
#if !CONFIG_CB4X4
// prepare internal edge parameters
if (currLevel && !currSkipped) {
pParams->filterLengthInternal = (TX_4X4 >= ts) ? (4) : (0);
}
#endif
// prepare common parameters
if (pParams->filterLength || pParams->filterLengthInternal) {
const loop_filter_thresh *const limits = cm->lf_info.lfthr + level;
@ -2075,15 +2108,21 @@ static void av1_filter_block_plane_vert(const AV1_COMMON *const cm,
const ptrdiff_t modeStride,
const uint32_t cuX,
const uint32_t cuY) {
const int col_step = MI_SIZE >> MI_SIZE_LOG2;
const int row_step = MI_SIZE >> MI_SIZE_LOG2;
const uint32_t scaleHorz = pPlane->subsampling_x;
const uint32_t scaleVert = pPlane->subsampling_y;
const uint32_t width = pPlane->dst.width;
const uint32_t height = pPlane->dst.height;
uint8_t *const pDst = pPlane->dst.buf;
const int dstStride = pPlane->dst.stride;
for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += 1) {
for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += row_step) {
uint8_t *p = pDst + y * MI_SIZE * dstStride;
for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += 1) {
for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += col_step) {
// inner loop always filter vertical edges in a MI block. If MI size
// is 8x8, it will filter the vertical edge aligned with a 8x8 block.
// If 4x4 trasnform is used, it will then filter the internal edge
// aligned with a 4x4 block
const MODE_INFO **const pCurr =
ppModeInfo + (y << scaleVert) * modeStride + (x << scaleHorz);
AV1_DEBLOCKING_PARAMETERS params;
@ -2094,31 +2133,59 @@ static void av1_filter_block_plane_vert(const AV1_COMMON *const cm,
switch (params.filterLength) {
// apply 4-tap filtering
case 4:
aom_lpf_vertical_4(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_vertical_4_c(CONVERT_TO_SHORTPTR(p), dstStride,
params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_vertical_4_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
// apply 8-tap filtering
case 8:
aom_lpf_vertical_8(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_vertical_8_c(CONVERT_TO_SHORTPTR(p), dstStride,
params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_vertical_8_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || CONFIG_PARALLEL_DEBLOCKING_15TAPLUMAONLY
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
// apply 16-tap filtering
case 16:
aom_lpf_vertical_16(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_vertical_16_c(CONVERT_TO_SHORTPTR(p), dstStride,
params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_vertical_16_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
// no filtering
default: break;
}
// process the internal edge
if (params.filterLengthInternal) {
aom_lpf_vertical_4(p + 4, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_vertical_4_c(CONVERT_TO_SHORTPTR(p + 4), dstStride,
params.mblim, params.lim, params.hev_thr,
cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_vertical_4_c(p + 4, dstStride, params.mblim, params.lim,
params.hev_thr);
}
// advance the destination pointer
p += 8;
p += MI_SIZE;
}
}
}
@ -2129,15 +2196,21 @@ static void av1_filter_block_plane_horz(const AV1_COMMON *const cm,
const ptrdiff_t modeStride,
const uint32_t cuX,
const uint32_t cuY) {
const int col_step = MI_SIZE >> MI_SIZE_LOG2;
const int row_step = MI_SIZE >> MI_SIZE_LOG2;
const uint32_t scaleHorz = pPlane->subsampling_x;
const uint32_t scaleVert = pPlane->subsampling_y;
const uint32_t width = pPlane->dst.width;
const uint32_t height = pPlane->dst.height;
uint8_t *const pDst = pPlane->dst.buf;
const int dstStride = pPlane->dst.stride;
for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += 1) {
for (int y = 0; y < (MAX_MIB_SIZE >> scaleVert); y += row_step) {
uint8_t *p = pDst + y * MI_SIZE * dstStride;
for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += 1) {
for (int x = 0; x < (MAX_MIB_SIZE >> scaleHorz); x += col_step) {
// inner loop always filter vertical edges in a MI block. If MI size
// is 8x8, it will first filter the vertical edge aligned with a 8x8
// block. If 4x4 trasnform is used, it will then filter the internal
// edge aligned with a 4x4 block
const MODE_INFO **const pCurr =
ppModeInfo + (y << scaleVert) * modeStride + (x << scaleHorz);
AV1_DEBLOCKING_PARAMETERS params;
@ -2148,52 +2221,84 @@ static void av1_filter_block_plane_horz(const AV1_COMMON *const cm,
switch (params.filterLength) {
// apply 4-tap filtering
case 4:
aom_lpf_horizontal_4(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_horizontal_4_c(CONVERT_TO_SHORTPTR(p), dstStride,
params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_horizontal_4_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
// apply 8-tap filtering
case 8:
aom_lpf_horizontal_8(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_horizontal_8_c(CONVERT_TO_SHORTPTR(p), dstStride,
params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_horizontal_8_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || CONFIG_PARALLEL_DEBLOCKING_15TAPLUMAONLY
#if CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
// apply 16-tap filtering
case 16:
aom_lpf_horizontal_edge_16(p, dstStride, params.mblim, params.lim,
params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_horizontal_edge_16_c(
CONVERT_TO_SHORTPTR(p), dstStride, params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_horizontal_edge_16_c(p, dstStride, params.mblim, params.lim,
params.hev_thr);
break;
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP
#endif // CONFIG_PARALLEL_DEBLOCKING_15TAP || PARALLEL_DEBLOCKING_15TAPLUMAONLY
// no filtering
default: break;
}
// process the internal edge
if (params.filterLengthInternal) {
aom_lpf_horizontal_4(p + 4 * dstStride, dstStride, params.mblim,
params.lim, params.hev_thr);
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
aom_highbd_lpf_horizontal_4_c(CONVERT_TO_SHORTPTR(p + 4 * dstStride),
dstStride, params.mblim, params.lim,
params.hev_thr, cm->bit_depth);
else
#endif // CONFIG_HIGHBITDEPTH
aom_lpf_horizontal_4_c(p + 4 * dstStride, dstStride, params.mblim,
params.lim, params.hev_thr);
}
// advance the destination pointer
p += 8;
p += MI_SIZE;
}
}
}
#endif // CONFIG_PARALLEL_DEBLOCKING
#endif
void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only) {
#if CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES || \
CONFIG_CB4X4
const int num_planes = y_only ? 1 : MAX_MB_PLANE;
int mi_row, mi_col;
#if CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES || \
CONFIG_CB4X4
#if !CONFIG_PARALLEL_DEBLOCKING
#if CONFIG_VAR_TX
memset(cm->above_txfm_context, TX_SIZES, cm->mi_cols);
for (int i = 0; i < MAX_MB_PLANE; ++i)
memset(cm->top_txfm_context[i], TX_32X32, cm->mi_cols << TX_UNIT_WIDE_LOG2);
#endif // CONFIG_VAR_TX
for (mi_row = start; mi_row < stop; mi_row += cm->mib_size) {
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
#if CONFIG_VAR_TX
memset(cm->left_txfm_context, TX_SIZES, MAX_MIB_SIZE);
for (int i = 0; i < MAX_MB_PLANE; ++i)
memset(cm->left_txfm_context[i], TX_32X32, MAX_MIB_SIZE
<< TX_UNIT_WIDE_LOG2);
#endif // CONFIG_VAR_TX
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += cm->mib_size) {
int plane;
@ -2202,33 +2307,23 @@ void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
for (plane = 0; plane < num_planes; ++plane) {
av1_filter_block_plane_non420_ver(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
mi_row, mi_col, plane);
av1_filter_block_plane_non420_hor(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
mi_row, mi_col, plane);
}
}
}
#else // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
const int num_planes = y_only ? 1 : MAX_MB_PLANE;
int mi_row, mi_col;
#if !CONFIG_PARALLEL_DEBLOCKING
enum lf_path path;
LOOP_FILTER_MASK lfm;
#else
#if CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
assert(0 && "Not yet updated. ToDo as next steps");
#endif // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
if (y_only)
path = LF_PATH_444;
else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
path = LF_PATH_420;
else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
path = LF_PATH_444;
else
path = LF_PATH_SLOW;
#endif
#if CONFIG_PARALLEL_DEBLOCKING
for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
// filter all vertical edges in every 64x64 super block
for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
const int32_t scaleHorz = planes[planeIdx].subsampling_x;
const int32_t scaleVert = planes[planeIdx].subsampling_y;
@ -2243,6 +2338,42 @@ void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
// filter all horizontal edges in every 64x64 super block
for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
const int32_t scaleHorz = planes[planeIdx].subsampling_x;
const int32_t scaleVert = planes[planeIdx].subsampling_y;
av1_filter_block_plane_horz(
cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
(mi_row * MI_SIZE) >> scaleVert);
}
}
}
#endif // CONFIG_PARALLEL_DEBLOCKING
#else // CONFIG_VAR_TX || CONFIG_EXT_PARTITION || CONFIG_EXT_PARTITION_TYPES
#if CONFIG_PARALLEL_DEBLOCKING
for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
// filter all vertical edges in every 64x64 super block
for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
const int32_t scaleHorz = planes[planeIdx].subsampling_x;
const int32_t scaleVert = planes[planeIdx].subsampling_y;
av1_filter_block_plane_vert(
cm, planes + planeIdx, (const MODE_INFO **)(mi + mi_col),
cm->mi_stride, (mi_col * MI_SIZE) >> scaleHorz,
(mi_row * MI_SIZE) >> scaleVert);
}
}
}
for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
av1_setup_dst_planes(planes, cm->sb_size, frame_buffer, mi_row, mi_col);
// filter all horizontal edges in every 64x64 super block
for (int planeIdx = 0; planeIdx < num_planes; planeIdx += 1) {
const int32_t scaleHorz = planes[planeIdx].subsampling_x;
const int32_t scaleVert = planes[planeIdx].subsampling_y;
@ -2254,6 +2385,18 @@ void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
}
}
#else // CONFIG_PARALLEL_DEBLOCKING
enum lf_path path;
LOOP_FILTER_MASK lfm;
if (y_only)
path = LF_PATH_444;
else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
path = LF_PATH_420;
else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
path = LF_PATH_444;
else
path = LF_PATH_SLOW;
for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) {
MODE_INFO **mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MAX_MIB_SIZE) {
@ -2278,9 +2421,9 @@ void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer, AV1_COMMON *cm,
break;
case LF_PATH_SLOW:
av1_filter_block_plane_non420_ver(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
mi_row, mi_col, plane);
av1_filter_block_plane_non420_hor(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
mi_row, mi_col, plane);
break;
}

4
third_party/aom/av1/common/av1_loopfilter.h vendored
View file

@ -115,11 +115,11 @@ void av1_filter_block_plane_ss11_hor(struct AV1Common *const cm,
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 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 mi_col, int pl);
void av1_loop_filter_init(struct AV1Common *cm);

20
third_party/aom/av1/common/av1_rtcd_defs.pl vendored
View file

@ -312,6 +312,7 @@ if (aom_config("CONFIG_AOM_QM") eq "yes") {
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";
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/;
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";
@ -399,6 +400,13 @@ if (aom_config("CONFIG_HIGHBITDEPTH") ne "yes") {
add_proto qw/void av1_fwd_idtx/, "const int16_t *src_diff, tran_low_t *coeff, int stride, int bs, int tx_type";
if (aom_config("CONFIG_DPCM_INTRA") eq "yes") {
@sizes = (4, 8, 16, 32);
foreach $size (@sizes) {
add_proto "void", "av1_dpcm_ft$size", "const int16_t *input, int stride, TX_TYPE_1D tx_type, tran_low_t *output";
}
}
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
#fwd txfm
add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
@ -608,15 +616,21 @@ if (aom_config("CONFIG_PVQ") eq "yes") {
if ((aom_config("CONFIG_WARPED_MOTION") eq "yes") ||
(aom_config("CONFIG_GLOBAL_MOTION") eq "yes")) {
add_proto qw/void av1_warp_affine/, "int32_t *mat, 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, int ref_frm, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
specialize qw/av1_warp_affine sse2/;
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, int comp_avg, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
specialize qw/av1_warp_affine sse2 ssse3/;
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void av1_highbd_warp_affine/, "int32_t *mat, 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, int ref_frm, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
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, int comp_avg, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
specialize qw/av1_highbd_warp_affine ssse3/;
}
}
if (aom_config("CONFIG_GLOBAL_MOTION") eq "yes" &&
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") {

49
third_party/aom/av1/common/av1_txfm.h vendored
View file

@ -24,7 +24,7 @@ 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[7][64] = {
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,
@ -68,6 +68,10 @@ static const int32_t cospi_arr[7][64] = {
14359, 12785, 11204, 9616, 8022, 6424, 4821, 3216, 1608 }
};
static INLINE const int32_t *cospi_arr(int n) {
return 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;
@ -139,26 +143,27 @@ typedef enum TXFM_TYPE {
TXFM_TYPE_ADST8,
TXFM_TYPE_ADST16,
TXFM_TYPE_ADST32,
TXFM_TYPE_IDENTITY4,
TXFM_TYPE_IDENTITY8,
TXFM_TYPE_IDENTITY16,
TXFM_TYPE_IDENTITY32,
} TXFM_TYPE;
typedef struct TXFM_2D_CFG {
typedef struct TXFM_1D_CFG {
const int txfm_size;
const int stage_num_col;
const int stage_num_row;
const int stage_num;
const int8_t *shift;
const int8_t *stage_range_col;
const int8_t *stage_range_row;
const int8_t *cos_bit_col;
const int8_t *cos_bit_row;
const TXFM_TYPE txfm_type_col;
const TXFM_TYPE txfm_type_row;
} TXFM_2D_CFG;
const int8_t *stage_range;
const int8_t *cos_bit;
const TXFM_TYPE txfm_type;
} TXFM_1D_CFG;
typedef struct TXFM_2D_FLIP_CFG {
int ud_flip; // flip upside down
int lr_flip; // flip left to right
const TXFM_2D_CFG *cfg;
const TXFM_1D_CFG *col_cfg;
const TXFM_1D_CFG *row_cfg;
} TXFM_2D_FLIP_CFG;
static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
@ -171,11 +176,23 @@ static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
cfg->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;
break;
case FLIPADST_DCT:
case FLIPADST_ADST:
case V_FLIPADST:
cfg->ud_flip = 1;
cfg->lr_flip = 0;
break;
case DCT_FLIPADST:
case ADST_FLIPADST:
case H_FLIPADST:
cfg->ud_flip = 0;
cfg->lr_flip = 1;
break;
@ -183,14 +200,6 @@ static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
cfg->ud_flip = 1;
cfg->lr_flip = 1;
break;
case ADST_FLIPADST:
cfg->ud_flip = 0;
cfg->lr_flip = 1;
break;
case FLIPADST_ADST:
cfg->ud_flip = 1;
cfg->lr_flip = 0;
break;
#endif // CONFIG_EXT_TX
default:
cfg->ud_flip = 0;

56
third_party/aom/av1/common/blockd.c vendored
View file

@ -179,15 +179,15 @@ void av1_foreach_transformed_block(const MACROBLOCKD *const xd,
#endif
#if CONFIG_DAALA_DIST
void av1_foreach_8x8_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
void av1_foreach_8x8_transformed_block_in_yplane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit,
foreach_transformed_block_visitor mi_visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const struct macroblockd_plane *const pd = &xd->plane[0];
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 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 = get_tx_size(plane, xd);
const TX_SIZE tx_size = get_tx_size(0, xd);
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
const uint8_t txw_unit = tx_size_wide_unit[tx_size];
const uint8_t txh_unit = tx_size_high_unit[tx_size];
@ -197,18 +197,24 @@ void av1_foreach_8x8_transformed_block_in_plane(
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
const int max_blocks_wide = max_block_wide(xd, plane_bsize, 0);
const int max_blocks_high = max_block_high(xd, plane_bsize, 0);
const int skip_check_r = tx_size_high[tx_size] == 8 ? 1 : 0;
const int skip_check_c = tx_size_wide[tx_size] == 8 ? 1 : 0;
assert(plane_bsize >= BLOCK_8X8);
assert(tx_size == TX_4X4 || tx_size == TX_4X8 || tx_size == TX_8X4);
// Keep track of the row and column of the blocks we use so that we know
// if we are in the unrestricted motion border.
for (r = 0; r < max_blocks_high; r += txh_unit) {
// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += txw_unit) {
visit(plane, i, r, c, plane_bsize, tx_size, arg);
// Call whenever each 8x8 block is done
if ((r & 1) && (c & 1))
mi_visit(plane, i, r - 1, c - 1, plane_bsize, TX_8X8, arg);
visit(0, i, r, c, plane_bsize, tx_size, arg);
// Call whenever each 8x8 tx block is done
if (((r & txh_unit) || skip_check_r) && ((c & txw_unit) || skip_check_c))
mi_visit(0, i, r - (1 - skip_check_r) * txh_unit,
c - (1 - skip_check_c) * txw_unit, plane_bsize, tx_size, arg);
i += step;
}
}
@ -259,6 +265,36 @@ void av1_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
}
#endif
void av1_reset_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize) {
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
for (i = 0; i < nplanes; i++) {
struct macroblockd_plane *const pd = &xd->plane[i];
#if CONFIG_CHROMA_2X2 || !CONFIG_CB4X4
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
#else
const BLOCK_SIZE plane_bsize =
AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
#endif
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);
memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * txs_high);
}
}
void av1_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y) {
int i;

206
third_party/aom/av1/common/blockd.h vendored
View file

@ -38,7 +38,11 @@
extern "C" {
#endif
#define SUB8X8_COMP_REF (!(CONFIG_CB4X4 && CONFIG_CHROMA_2X2))
#if (CONFIG_CHROMA_SUB8X8 || CONFIG_CHROMA_2X2)
#define SUB8X8_COMP_REF 0
#else
#define SUB8X8_COMP_REF 1
#endif
#define MAX_MB_PLANE 3
@ -49,26 +53,20 @@ extern "C" {
// 0: Uniform
// 1: Difference weighted
#define COMPOUND_SEGMENT_TYPE 1
#if COMPOUND_SEGMENT_TYPE == 0
#define MAX_SEG_MASK_BITS 1
// SEG_MASK_TYPES should not surpass 1 << MAX_SEG_MASK_BITS
typedef enum {
#if COMPOUND_SEGMENT_TYPE == 0
UNIFORM_45 = 0,
UNIFORM_45_INV,
SEG_MASK_TYPES,
} SEG_MASK_TYPE;
#elif COMPOUND_SEGMENT_TYPE == 1
#define MAX_SEG_MASK_BITS 1
// SEG_MASK_TYPES should not surpass 1 << MAX_SEG_MASK_BITS
typedef enum {
DIFFWTD_42 = 0,
DIFFWTD_42_INV,
DIFFWTD_38 = 0,
DIFFWTD_38_INV,
#endif // COMPOUND_SEGMENT_TYPE
SEG_MASK_TYPES,
} SEG_MASK_TYPE;
#endif // COMPOUND_SEGMENT_TYPE
#endif // CONFIG_COMPOUND_SEGMENT
#endif // CONFIG_EXT_INTER
@ -131,7 +129,7 @@ static INLINE int is_inter_compound_mode(PREDICTION_MODE mode) {
}
static INLINE PREDICTION_MODE compound_ref0_mode(PREDICTION_MODE mode) {
static PREDICTION_MODE lut[MB_MODE_COUNT] = {
static PREDICTION_MODE lut[] = {
MB_MODE_COUNT, // DC_PRED
MB_MODE_COUNT, // V_PRED
MB_MODE_COUNT, // H_PRED
@ -143,6 +141,10 @@ static INLINE PREDICTION_MODE compound_ref0_mode(PREDICTION_MODE mode) {
MB_MODE_COUNT, // D63_PRED
#if CONFIG_ALT_INTRA
MB_MODE_COUNT, // SMOOTH_PRED
#if CONFIG_SMOOTH_HV
MB_MODE_COUNT, // SMOOTH_V_PRED
MB_MODE_COUNT, // SMOOTH_H_PRED
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
MB_MODE_COUNT, // TM_PRED
MB_MODE_COUNT, // NEARESTMV
@ -157,8 +159,6 @@ static INLINE PREDICTION_MODE compound_ref0_mode(PREDICTION_MODE mode) {
NEWMV, // SR_NEW_NEWMV
#endif // CONFIG_COMPOUND_SINGLEREF
NEARESTMV, // NEAREST_NEARESTMV
NEARESTMV, // NEAREST_NEARMV
NEARMV, // NEAR_NEARESTMV
NEARMV, // NEAR_NEARMV
NEARESTMV, // NEAREST_NEWMV
NEWMV, // NEW_NEARESTMV
@ -167,12 +167,13 @@ static INLINE PREDICTION_MODE compound_ref0_mode(PREDICTION_MODE mode) {
ZEROMV, // ZERO_ZEROMV
NEWMV, // NEW_NEWMV
};
assert(NELEMENTS(lut) == MB_MODE_COUNT);
assert(is_inter_compound_mode(mode));
return lut[mode];
}
static INLINE PREDICTION_MODE compound_ref1_mode(PREDICTION_MODE mode) {
static PREDICTION_MODE lut[MB_MODE_COUNT] = {
static PREDICTION_MODE lut[] = {
MB_MODE_COUNT, // DC_PRED
MB_MODE_COUNT, // V_PRED
MB_MODE_COUNT, // H_PRED
@ -184,6 +185,10 @@ static INLINE PREDICTION_MODE compound_ref1_mode(PREDICTION_MODE mode) {
MB_MODE_COUNT, // D63_PRED
#if CONFIG_ALT_INTRA
MB_MODE_COUNT, // SMOOTH_PRED
#if CONFIG_SMOOTH_HV
MB_MODE_COUNT, // SMOOTH_V_PRED
MB_MODE_COUNT, // SMOOTH_H_PRED
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
MB_MODE_COUNT, // TM_PRED
MB_MODE_COUNT, // NEARESTMV
@ -198,8 +203,6 @@ static INLINE PREDICTION_MODE compound_ref1_mode(PREDICTION_MODE mode) {
NEWMV, // SR_NEW_NEWMV
#endif // CONFIG_COMPOUND_SINGLEREF
NEARESTMV, // NEAREST_NEARESTMV
NEARMV, // NEAREST_NEARMV
NEARESTMV, // NEAR_NEARESTMV
NEARMV, // NEAR_NEARMV
NEWMV, // NEAREST_NEWMV
NEARESTMV, // NEW_NEARESTMV
@ -208,13 +211,14 @@ static INLINE PREDICTION_MODE compound_ref1_mode(PREDICTION_MODE mode) {
ZEROMV, // ZERO_ZEROMV
NEWMV, // NEW_NEWMV
};
assert(NELEMENTS(lut) == MB_MODE_COUNT);
assert(is_inter_compound_mode(mode));
return lut[mode];
}
static INLINE int have_nearmv_in_inter_mode(PREDICTION_MODE mode) {
return (mode == NEARMV || mode == NEAR_NEARMV || mode == NEAREST_NEARMV ||
mode == NEAR_NEARESTMV || mode == NEAR_NEWMV || mode == NEW_NEARMV);
return (mode == NEARMV || mode == NEAR_NEARMV || mode == NEAR_NEWMV ||
mode == NEW_NEARMV);
}
static INLINE int have_newmv_in_inter_mode(PREDICTION_MODE mode) {
@ -260,9 +264,7 @@ static INLINE int have_newmv_in_inter_mode(PREDICTION_MODE mode) {
typedef struct {
PREDICTION_MODE as_mode;
int_mv as_mv[2]; // first, second inter predictor motion vectors
#if CONFIG_REF_MV
int_mv pred_mv[2];
#endif
#if CONFIG_EXT_INTER
int_mv ref_mv[2];
#endif // CONFIG_EXT_INTER
@ -274,12 +276,8 @@ typedef int8_t MV_REFERENCE_FRAME;
typedef struct {
// Number of base colors for Y (0) and UV (1)
uint8_t palette_size[2];
// Value of base colors for Y, U, and V
#if CONFIG_HIGHBITDEPTH
// Value of base colors for Y, U, and V
uint16_t palette_colors[3 * PALETTE_MAX_SIZE];
#else
uint8_t palette_colors[3 * PALETTE_MAX_SIZE];
#endif // CONFIG_HIGHBITDEPTH
// Only used by encoder to store the color index of the top left pixel.
// TODO(huisu): move this to encoder
uint8_t palette_first_color_idx[2];
@ -312,6 +310,9 @@ typedef struct RD_STATS {
int64_t rdcost;
int64_t sse;
int skip; // sse should equal to dist when skip == 1
#if CONFIG_DAALA_DIST && CONFIG_CB4X4
int64_t dist_y;
#endif
#if CONFIG_RD_DEBUG
int txb_coeff_cost[MAX_MB_PLANE];
#if CONFIG_VAR_TX
@ -338,7 +339,7 @@ typedef struct {
#endif // CONFIG_EXT_INTER
// This structure now relates to 8x8 block regions.
typedef struct {
typedef struct MB_MODE_INFO {
// Common for both INTER and INTRA blocks
BLOCK_SIZE sb_type;
PREDICTION_MODE mode;
@ -391,8 +392,10 @@ typedef struct {
#endif // CONFIG_EXT_INTRA
#if CONFIG_EXT_INTER
#if CONFIG_INTERINTRA
// interintra members
INTERINTRA_MODE interintra_mode;
#endif
// TODO(debargha): Consolidate these flags
int use_wedge_interintra;
int interintra_wedge_index;
@ -413,9 +416,7 @@ typedef struct {
#endif // CONFIG_MOTION_VAR
int_mv mv[2];
int_mv pred_mv[2];
#if CONFIG_REF_MV
uint8_t ref_mv_idx;
#endif
#if CONFIG_EXT_PARTITION_TYPES
PARTITION_TYPE partition;
#endif
@ -441,6 +442,13 @@ typedef struct {
WarpedMotionParams wm_params[2];
#endif // CONFIG_WARPED_MOTION
#if CONFIG_CFL
// Index of the alpha Cb and alpha Cr combination
int cfl_alpha_idx;
// Signs of alpha Cb and alpha Cr
CFL_SIGN_TYPE cfl_alpha_signs[CFL_PRED_PLANES];
#endif
BOUNDARY_TYPE boundary_info;
} MB_MODE_INFO;
@ -536,16 +544,14 @@ typedef struct macroblockd_plane {
uint8_t width, height;
#if CONFIG_AOM_QM
const qm_val_t *seg_iqmatrix[MAX_SEGMENTS][2][TX_SIZES];
const qm_val_t *seg_iqmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
const qm_val_t *seg_qmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
#endif
// encoder
const int16_t *dequant;
#if CONFIG_NEW_QUANT
const dequant_val_type_nuq *dequant_val_nuq[QUANT_PROFILES];
#endif // CONFIG_NEW_QUANT
#if CONFIG_AOM_QM
const qm_val_t *seg_qmatrix[MAX_SEGMENTS][2][TX_SIZES];
#endif
#if CONFIG_PVQ || CONFIG_DAALA_DIST
DECLARE_ALIGNED(16, int16_t, pred[MAX_SB_SQUARE]);
@ -558,14 +564,14 @@ typedef struct macroblockd_plane {
((x) + (i) * (1 << (tx_size_wide_log2[0] + tx_size_high_log2[0])))
typedef struct RefBuffer {
// TODO(dkovalev): idx is not really required and should be removed, now it
// is used in av1_onyxd_if.c
int idx;
YV12_BUFFER_CONFIG *buf;
struct scale_factors sf;
} RefBuffer;
typedef int16_t EobThresholdMD[TX_SIZES_ALL][TX_TYPES];
#if CONFIG_ADAPT_SCAN
typedef int16_t EobThresholdMD[TX_TYPES][EOB_THRESHOLD_NUM];
#endif
typedef struct macroblockd {
struct macroblockd_plane plane[MAX_MB_PLANE];
@ -606,6 +612,11 @@ typedef struct macroblockd {
/* pointer to current frame */
const YV12_BUFFER_CONFIG *cur_buf;
#if CONFIG_INTRABC
/* Scale of the current frame with respect to itself */
struct scale_factors sf_identity;
#endif
ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
ENTROPY_CONTEXT left_context[MAX_MB_PLANE][2 * MAX_MIB_SIZE];
@ -615,7 +626,7 @@ typedef struct macroblockd {
#if CONFIG_VAR_TX
TXFM_CONTEXT *above_txfm_context;
TXFM_CONTEXT *left_txfm_context;
TXFM_CONTEXT left_txfm_context_buffer[MAX_MIB_SIZE];
TXFM_CONTEXT left_txfm_context_buffer[2 * MAX_MIB_SIZE];
TX_SIZE max_tx_size;
#if CONFIG_SUPERTX
@ -626,11 +637,9 @@ typedef struct macroblockd {
// block dimension in the unit of mode_info.
uint8_t n8_w, n8_h;
#if CONFIG_REF_MV
uint8_t ref_mv_count[MODE_CTX_REF_FRAMES];
CANDIDATE_MV ref_mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE];
uint8_t is_sec_rect;
#endif
#if CONFIG_PVQ
daala_dec_ctx daala_dec;
@ -638,10 +647,8 @@ typedef struct macroblockd {
#if CONFIG_EC_ADAPT
FRAME_CONTEXT *tile_ctx;
#endif
#if CONFIG_HIGHBITDEPTH
/* Bit depth: 8, 10, 12 */
int bd;
#endif
int qindex[MAX_SEGMENTS];
int lossless[MAX_SEGMENTS];
@ -699,6 +706,10 @@ static const TX_TYPE intra_mode_to_tx_type_context[INTRA_MODES] = {
ADST_DCT, // D63
#if CONFIG_ALT_INTRA
ADST_ADST, // SMOOTH
#if CONFIG_SMOOTH_HV
ADST_DCT, // SMOOTH_V
DCT_ADST, // SMOOTH_H
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
ADST_ADST, // TM
};
@ -761,24 +772,24 @@ static const int ext_tx_set_index_inter[EXT_TX_SET_TYPES] = {
static INLINE TxSetType get_ext_tx_set_type(TX_SIZE tx_size, BLOCK_SIZE bs,
int is_inter, int use_reduced_set) {
const TX_SIZE tx_size2 = txsize_sqr_up_map[tx_size];
tx_size = txsize_sqr_map[tx_size];
const TX_SIZE tx_size_sqr_up = txsize_sqr_up_map[tx_size];
const TX_SIZE tx_size_sqr = txsize_sqr_map[tx_size];
#if CONFIG_CB4X4 && USE_TXTYPE_SEARCH_FOR_SUB8X8_IN_CB4X4
(void)bs;
if (tx_size > TX_32X32) return EXT_TX_SET_DCTONLY;
if (tx_size_sqr > TX_32X32) return EXT_TX_SET_DCTONLY;
#else
if (tx_size > TX_32X32 || bs < BLOCK_8X8) return EXT_TX_SET_DCTONLY;
if (tx_size_sqr > TX_32X32 || bs < BLOCK_8X8) return EXT_TX_SET_DCTONLY;
#endif
if (use_reduced_set)
return is_inter ? EXT_TX_SET_DCT_IDTX : EXT_TX_SET_DTT4_IDTX;
if (tx_size2 == TX_32X32)
if (tx_size_sqr_up == TX_32X32)
return is_inter ? EXT_TX_SET_DCT_IDTX : EXT_TX_SET_DCTONLY;
if (is_inter)
return (tx_size == TX_16X16 ? EXT_TX_SET_DTT9_IDTX_1DDCT
: EXT_TX_SET_ALL16);
return (tx_size_sqr == TX_16X16 ? EXT_TX_SET_DTT9_IDTX_1DDCT
: EXT_TX_SET_ALL16);
else
return (tx_size == TX_16X16 ? EXT_TX_SET_DTT4_IDTX
: EXT_TX_SET_DTT4_IDTX_1DDCT);
return (tx_size_sqr == TX_16X16 ? EXT_TX_SET_DTT4_IDTX
: EXT_TX_SET_DTT4_IDTX_1DDCT);
}
static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs, int is_inter,
@ -791,7 +802,7 @@ static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs, int is_inter,
static const int use_intra_ext_tx_for_txsize[EXT_TX_SETS_INTRA][EXT_TX_SIZES] =
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ 1, 1, 1, 1, 1 }, // unused
{ 0, 1, 1, 0, 0 },
{ 0, 0, 0, 1, 0 },
@ -799,12 +810,12 @@ static const int use_intra_ext_tx_for_txsize[EXT_TX_SETS_INTRA][EXT_TX_SIZES] =
{ 1, 1, 1, 1 }, // unused
{ 1, 1, 0, 0 },
{ 0, 0, 1, 0 },
#endif // CONFIG_CB4X4
#endif // CONFIG_CHROMA_2X2
};
static const int use_inter_ext_tx_for_txsize[EXT_TX_SETS_INTER][EXT_TX_SIZES] =
{
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ 1, 1, 1, 1, 1 }, // unused
{ 0, 1, 1, 0, 0 },
{ 0, 0, 0, 1, 0 },
@ -814,7 +825,7 @@ static const int use_inter_ext_tx_for_txsize[EXT_TX_SETS_INTER][EXT_TX_SIZES] =
{ 1, 1, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 0, 0, 1 },
#endif // CONFIG_CB4X4
#endif // CONFIG_CHROMA_2X2
};
// Transform types used in each intra set
@ -887,6 +898,45 @@ static INLINE int is_rect_tx_allowed(const MACROBLOCKD *xd,
return is_rect_tx_allowed_bsize(mbmi->sb_type) &&
!xd->lossless[mbmi->segment_id];
}
#if CONFIG_RECT_TX_EXT
static INLINE int is_quarter_tx_allowed_bsize(BLOCK_SIZE bsize) {
static const char LUT_QTTX[BLOCK_SIZES] = {
#if CONFIG_CB4X4
0, // BLOCK_2X2
0, // BLOCK_2X4
0, // BLOCK_4X2
#endif
0, // BLOCK_4X4
0, // BLOCK_4X8
0, // BLOCK_8X4
0, // BLOCK_8X8
1, // BLOCK_8X16
1, // BLOCK_16X8
0, // BLOCK_16X16
0, // BLOCK_16X32
0, // BLOCK_32X16
0, // BLOCK_32X32
0, // BLOCK_32X64
0, // BLOCK_64X32
0, // BLOCK_64X64
#if CONFIG_EXT_PARTITION
0, // BLOCK_64X128
0, // BLOCK_128X64
0, // BLOCK_128X128
#endif // CONFIG_EXT_PARTITION
};
return LUT_QTTX[bsize];
}
static INLINE int is_quarter_tx_allowed(const MACROBLOCKD *xd,
const MB_MODE_INFO *mbmi,
int is_inter) {
return is_quarter_tx_allowed_bsize(mbmi->sb_type) && is_inter &&
!xd->lossless[mbmi->segment_id];
}
#endif // CONFIG_RECT_TX_EXT
#endif // CONFIG_RECT_TX
#endif // CONFIG_EXT_TX
@ -926,8 +976,14 @@ static INLINE TX_SIZE tx_size_from_tx_mode(BLOCK_SIZE bsize, TX_MODE tx_mode,
#define MAX_ANGLE_DELTA 3
#define ANGLE_STEP 3
extern const int16_t dr_intra_derivative[90];
static const uint8_t mode_to_angle_map[INTRA_MODES] = {
static const uint8_t mode_to_angle_map[] = {
0, 90, 180, 45, 135, 111, 157, 203, 67, 0,
#if CONFIG_ALT_INTRA
0,
#if CONFIG_SMOOTH_HV
0, 0,
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
};
#if CONFIG_INTRA_INTERP
// Returns whether filter selection is needed for a given
@ -977,10 +1033,10 @@ static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, const MACROBLOCKD *xd,
int block, TX_SIZE tx_size) {
const MODE_INFO *const mi = xd->mi[0];
const MB_MODE_INFO *const mbmi = &mi->mbmi;
#if CONFIG_INTRABC
// TODO(aconverse@google.com): Revisit this decision
#if CONFIG_INTRABC && (!CONFIG_EXT_TX || CONFIG_TXK_SEL)
// TODO(aconverse@google.com): Handle INTRABC + EXT_TX + TXK_SEL
if (is_intrabc_block(mbmi)) return DCT_DCT;
#endif // CONFIG_INTRABC
#endif // CONFIG_INTRABC && (!CONFIG_EXT_TX || CONFIG_TXK_SEL)
#if !CONFIG_TXK_SEL
#if FIXED_TX_TYPE
const int block_raster_idx = av1_block_index_to_raster_order(tx_size, block);
@ -1002,7 +1058,7 @@ static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, const MACROBLOCKD *xd,
if (is_inter_block(mbmi)) {
// UV Inter only
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
if (tx_size < TX_4X4) return DCT_DCT;
#endif
return (mbmi->tx_type == IDTX && txsize_sqr_map[tx_size] >= TX_32X32)
@ -1013,11 +1069,13 @@ static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, const MACROBLOCKD *xd,
#if CONFIG_CB4X4
(void)block;
#if CONFIG_CHROMA_2X2
if (tx_size < TX_4X4)
return DCT_DCT;
else
#endif
return intra_mode_to_tx_type_context[mbmi->uv_mode];
#else
#else // CONFIG_CB4X4
// Sub8x8-Inter/Intra OR UV-Intra
if (is_inter_block(mbmi)) // Sub8x8-Inter
@ -1051,17 +1109,17 @@ static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type, const MACROBLOCKD *xd,
void av1_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y);
static INLINE int tx_size_to_depth(TX_SIZE tx_size) {
return (int)(tx_size - TX_4X4);
return (int)(tx_size - TX_SIZE_LUMA_MIN);
}
static INLINE TX_SIZE depth_to_tx_size(int depth) {
return (TX_SIZE)(depth + TX_4X4);
return (TX_SIZE)(depth + TX_SIZE_LUMA_MIN);
}
static INLINE TX_SIZE get_uv_tx_size(const MB_MODE_INFO *mbmi,
const struct macroblockd_plane *pd) {
TX_SIZE uv_txsize;
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
assert(mbmi->tx_size > TX_2X2);
#endif
@ -1073,9 +1131,6 @@ static INLINE TX_SIZE get_uv_tx_size(const MB_MODE_INFO *mbmi,
uv_txsize = uv_txsize_lookup[mbmi->sb_type][mbmi->tx_size][pd->subsampling_x]
[pd->subsampling_y];
#if CONFIG_CB4X4 && !CONFIG_CHROMA_2X2
uv_txsize = AOMMAX(uv_txsize, TX_4X4);
#endif
assert(uv_txsize != TX_INVALID);
return uv_txsize;
}
@ -1092,17 +1147,8 @@ get_plane_block_size(BLOCK_SIZE bsize, const struct macroblockd_plane *pd) {
return ss_size_lookup[bsize][pd->subsampling_x][pd->subsampling_y];
}
static INLINE void reset_skip_context(MACROBLOCKD *xd, BLOCK_SIZE bsize) {
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {
struct macroblockd_plane *const pd = &xd->plane[i];
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
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);
memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * txs_high);
}
}
void av1_reset_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize);
typedef void (*foreach_transformed_block_visitor)(int plane, int block,
int blk_row, int blk_col,
@ -1121,8 +1167,8 @@ void av1_foreach_transformed_block(const MACROBLOCKD *const xd,
#endif
#if CONFIG_DAALA_DIST
void av1_foreach_8x8_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
void av1_foreach_8x8_transformed_block_in_yplane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit,
foreach_transformed_block_visitor mi_visit, void *arg);
#endif

20
third_party/aom/av1/common/cdef.c vendored
View file

@ -26,13 +26,9 @@ int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col) {
int skip = 1;
maxc = cm->mi_cols - mi_col;
maxr = cm->mi_rows - mi_row;
#if CONFIG_EXT_PARTITION
if (maxr > cm->mib_size_log2) maxr = cm->mib_size_log2;
if (maxc > cm->mib_size_log2) maxc = cm->mib_size_log2;
#else
if (maxr > MAX_MIB_SIZE) maxr = MAX_MIB_SIZE;
if (maxc > MAX_MIB_SIZE) maxc = MAX_MIB_SIZE;
#endif
maxr = AOMMIN(maxr, cm->mib_size);
maxc = AOMMIN(maxc, cm->mib_size);
for (r = 0; r < maxr; r++) {
for (c = 0; c < maxc; c++) {
@ -63,13 +59,9 @@ int sb_compute_dering_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
grid = cm->mi_grid_visible;
maxc = cm->mi_cols - mi_col;
maxr = cm->mi_rows - mi_row;
#if CONFIG_EXT_PARTITION
if (maxr > cm->mib_size_log2) maxr = cm->mib_size_log2;
if (maxc > cm->mib_size_log2) maxc = cm->mib_size_log2;
#else
if (maxr > MAX_MIB_SIZE) maxr = MAX_MIB_SIZE;
if (maxc > MAX_MIB_SIZE) maxc = MAX_MIB_SIZE;
#endif
maxr = AOMMIN(maxr, cm->mib_size);
maxc = AOMMIN(maxc, cm->mib_size);
const int r_step = mi_size_high[BLOCK_8X8];
const int c_step = mi_size_wide[BLOCK_8X8];

85
third_party/aom/av1/common/cfl.c vendored
View file

@ -47,23 +47,27 @@ void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
: tx_size_high[tx_size];
// Number of pixel on the top and left borders.
const int num_pel = block_width + block_height;
const double num_pel = block_width + block_height;
int sum_u = 0;
int sum_v = 0;
// Match behavior of build_intra_predictors (reconintra.c) at superblock
// boundaries:
//
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// ..
// Match behavior of build_intra_predictors (reconintra.c) at superblock
// boundaries:
//
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// ..
// TODO(ltrudeau) replace this with DC_PRED assembly
#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
// TODO(ltrudeau) replace this with DC_PRED assembly
for (int i = 0; i < block_width; i++) {
sum_u += dst_u[-dst_u_stride + i];
sum_v += dst_v[-dst_v_stride + i];
@ -73,7 +77,11 @@ void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
sum_v = block_width * 127;
}
#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
for (int i = 0; i < block_height; i++) {
sum_u += dst_u[i * dst_u_stride - 1];
sum_v += dst_v[i * dst_v_stride - 1];
@ -83,25 +91,22 @@ void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
sum_v += block_height * 129;
}
xd->cfl->dc_pred[CFL_PRED_U] = (sum_u + (num_pel >> 1)) / num_pel;
xd->cfl->dc_pred[CFL_PRED_V] = (sum_v + (num_pel >> 1)) / num_pel;
xd->cfl->dc_pred[CFL_PRED_U] = sum_u / num_pel;
xd->cfl->dc_pred[CFL_PRED_V] = sum_v / num_pel;
}
// Predict the current transform block using CfL.
// it is assumed that dst points at the start of the transform block
void cfl_predict_block(const CFL_CTX *cfl, uint8_t *dst, int dst_stride,
int row, int col, TX_SIZE tx_size, int dc_pred) {
const int tx_block_width = tx_size_wide[tx_size];
const int tx_block_height = tx_size_high[tx_size];
int row, int col, TX_SIZE tx_size, double dc_pred,
double alpha) {
const int width = tx_size_wide[tx_size];
const int height = tx_size_high[tx_size];
// TODO(ltrudeau) implement alpha
// Place holder for alpha
const double alpha = 0;
const double y_avg = cfl_load(cfl, dst, dst_stride, row, col, tx_size);
const double y_avg = cfl_load(cfl, dst, dst_stride, row, col, width, height);
for (int j = 0; j < tx_block_height; j++) {
for (int i = 0; i < tx_block_width; i++) {
dst[i] = (uint8_t)(alpha * y_avg + dc_pred + 0.5);
for (int j = 0; j < height; j++) {
for (int i = 0; i < width; i++) {
dst[i] = (uint8_t)(alpha * (dst[i] - y_avg) + dc_pred + 0.5);
}
dst += dst_stride;
}
@ -142,9 +147,7 @@ void cfl_store(CFL_CTX *cfl, const uint8_t *input, int input_stride, int row,
// Load from the CfL pixel buffer into output
double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
int col, TX_SIZE tx_size) {
const int tx_width = tx_size_wide[tx_size];
const int tx_height = tx_size_high[tx_size];
int col, int width, int height) {
const int sub_x = cfl->subsampling_x;
const int sub_y = cfl->subsampling_y;
const int tx_off_log2 = tx_size_wide_log2[0];
@ -161,12 +164,12 @@ double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
// TODO(ltrudeau) add support for 4:2:2
if (sub_y == 0 && sub_x == 0) {
y_pix = &cfl->y_pix[(row * MAX_SB_SIZE + col) << tx_off_log2];
int uv_width = (col << tx_off_log2) + tx_width;
int uv_width = (col << tx_off_log2) + width;
diff_width = uv_width - cfl->y_width;
int uv_height = (row << tx_off_log2) + tx_width;
int uv_height = (row << tx_off_log2) + height;
diff_height = uv_height - cfl->y_height;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
for (int j = 0; j < height; j++) {
for (int i = 0; i < width; i++) {
// In 4:4:4, pixels match 1 to 1
output[output_row_offset + i] = y_pix[pred_row_offset + i];
}
@ -175,12 +178,12 @@ double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
}
} else if (sub_y == 1 && sub_x == 1) {
y_pix = &cfl->y_pix[(row * MAX_SB_SIZE + col) << (tx_off_log2 + sub_y)];
int uv_width = ((col << tx_off_log2) + tx_width) << sub_x;
int uv_width = ((col << tx_off_log2) + width) << sub_x;
diff_width = (uv_width - cfl->y_width) >> sub_x;
int uv_height = ((row << tx_off_log2) + tx_width) << sub_y;
int uv_height = ((row << tx_off_log2) + height) << sub_y;
diff_height = (uv_height - cfl->y_height) >> sub_y;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
for (int j = 0; j < height; j++) {
for (int i = 0; i < width; i++) {
top_left = (pred_row_offset + i) << sub_y;
bot_left = top_left + MAX_SB_SIZE;
// In 4:2:0, average pixels in 2x2 grid
@ -206,9 +209,9 @@ double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
// frame, the columns will be copied over them.
if (diff_width > 0) {
int last_pixel;
output_row_offset = tx_width - diff_width;
output_row_offset = width - diff_width;
for (int j = 0; j < tx_height; j++) {
for (int j = 0; j < height; j++) {
last_pixel = output_row_offset - 1;
for (int i = 0; i < diff_width; i++) {
output[output_row_offset + i] = output[last_pixel];
@ -221,7 +224,7 @@ double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
output_row_offset = diff_height * output_stride;
const int last_row_offset = output_row_offset - output_stride;
for (int j = 0; j < diff_height; j++) {
for (int i = 0; i < tx_width; i++) {
for (int i = 0; i < width; i++) {
output[output_row_offset + i] = output[last_row_offset + i];
}
output_row_offset += output_stride;
@ -230,11 +233,11 @@ double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
int avg = 0;
output_row_offset = 0;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
for (int j = 0; j < height; j++) {
for (int i = 0; i < width; i++) {
avg += output[output_row_offset + i];
}
output_row_offset += output_stride;
}
return avg / (double)(tx_width * tx_height);
return avg / (double)(width * height);
}

41
third_party/aom/av1/common/cfl.h vendored
View file

@ -12,6 +12,8 @@
#ifndef AV1_COMMON_CFL_H_
#define AV1_COMMON_CFL_H_
#include <assert.h>
#include "av1/common/enums.h"
// Forward declaration of AV1_COMMON, in order to avoid creating a cyclic
@ -33,20 +35,53 @@ typedef struct {
int subsampling_x, subsampling_y;
// CfL Performs its own block level DC_PRED for each chromatic plane
int dc_pred[CFL_PRED_PLANES];
double dc_pred[CFL_PRED_PLANES];
// The rate associated with each alpha codeword
int costs[CFL_ALPHABET_SIZE];
// Count the number of TX blocks in a predicted block to know when you are at
// the last one, so you can check for skips.
// TODO(any) Is there a better way to do this?
int num_tx_blk[CFL_PRED_PLANES];
} CFL_CTX;
static const double cfl_alpha_mags[CFL_MAGS_SIZE] = {
0., 0.125, -0.125, 0.25, -0.25, 0.5, -0.5
};
static const int cfl_alpha_codes[CFL_ALPHABET_SIZE][CFL_PRED_PLANES] = {
// barrbrain's simple 1D quant ordered by subset 3 likelihood
{ 0, 0 }, { 1, 1 }, { 3, 0 }, { 3, 1 }, { 1, 0 }, { 3, 3 },
{ 0, 1 }, { 5, 5 }, { 5, 3 }, { 1, 3 }, { 5, 3 }, { 3, 5 },
{ 0, 3 }, { 5, 1 }, { 1, 5 }, { 0, 5 }
};
void cfl_init(CFL_CTX *cfl, AV1_COMMON *cm, int subsampling_x,
int subsampling_y);
void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size);
static INLINE double cfl_idx_to_alpha(int alpha_idx, CFL_SIGN_TYPE alpha_sign,
CFL_PRED_TYPE pred_type) {
const int mag_idx = cfl_alpha_codes[alpha_idx][pred_type];
const double abs_alpha = cfl_alpha_mags[mag_idx];
if (alpha_sign == CFL_SIGN_POS) {
return abs_alpha;
} else {
assert(abs_alpha != 0.0);
assert(cfl_alpha_mags[mag_idx + 1] == -abs_alpha);
return -abs_alpha;
}
}
void cfl_predict_block(const CFL_CTX *cfl, uint8_t *dst, int dst_stride,
int row, int col, TX_SIZE tx_size, int dc_pred);
int row, int col, TX_SIZE tx_size, double dc_pred,
double alpha);
void cfl_store(CFL_CTX *cfl, const uint8_t *input, int input_stride, int row,
int col, TX_SIZE tx_size);
double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
int col, TX_SIZE tx_size);
int col, int width, int height);
#endif // AV1_COMMON_CFL_H_

312
third_party/aom/av1/common/common_data.h vendored
View file

@ -21,105 +21,117 @@ extern "C" {
#endif
#if CONFIG_EXT_PARTITION
#define IF_EXT_PARTITION(...) __VA_ARGS__
#define IF_EXT_PARTITION(...) __VA_ARGS__,
#else
#define IF_EXT_PARTITION(...)
#endif
// Log 2 conversion lookup tables for block width and height
static const uint8_t b_width_log2_lookup[BLOCK_SIZES] = {
static const uint8_t b_width_log2_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
0, 0, 0,
#endif
0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, IF_EXT_PARTITION(4, 5, 5)
0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, IF_EXT_PARTITION(4, 5, 5) 0, 2, 1, 3
};
static const uint8_t b_height_log2_lookup[BLOCK_SIZES] = {
static const uint8_t b_height_log2_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
0, 0, 0,
#endif
0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4, IF_EXT_PARTITION(5, 4, 5)
0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4, IF_EXT_PARTITION(5, 4, 5) 2, 0, 3, 1
};
// Log 2 conversion lookup tables for modeinfo width and height
static const uint8_t mi_width_log2_lookup[BLOCK_SIZES] = {
static const uint8_t mi_width_log2_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, IF_EXT_PARTITION(4, 5, 5)
0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, IF_EXT_PARTITION(4, 5, 5) 0,
2, 1, 3
#else
0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, IF_EXT_PARTITION(3, 4, 4)
0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, IF_EXT_PARTITION(3, 4, 4) 0, 1, 0, 2
#endif
};
static const uint8_t mi_height_log2_lookup[BLOCK_SIZES] = {
static const uint8_t mi_height_log2_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
0, 0, 0, 0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4, IF_EXT_PARTITION(5, 4, 5)
0, 0, 0, 0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4, IF_EXT_PARTITION(5, 4, 5) 2,
0, 3, 1
#else
0, 0, 0, 0, 1, 0, 1, 2, 1, 2, 3, 2, 3, IF_EXT_PARTITION(4, 3, 4)
0, 0, 0, 0, 1, 0, 1, 2, 1, 2, 3, 2, 3, IF_EXT_PARTITION(4, 3, 4) 1, 0, 2, 0
#endif
};
static const uint8_t mi_size_wide[BLOCK_SIZES] = {
/* clang-format off */
static const uint8_t mi_size_wide[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16, IF_EXT_PARTITION(16, 32, 32)
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16,
IF_EXT_PARTITION(16, 32, 32) 1, 4, 2, 8
#else
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, IF_EXT_PARTITION(8, 16, 16)
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, IF_EXT_PARTITION(8, 16, 16) 1, 2, 1, 4
#endif
};
static const uint8_t mi_size_high[BLOCK_SIZES] = {
static const uint8_t mi_size_high[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16, IF_EXT_PARTITION(32, 16, 32)
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16,
IF_EXT_PARTITION(32, 16, 32) 4, 1, 8, 2
#else
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, IF_EXT_PARTITION(16, 8, 16)
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, IF_EXT_PARTITION(16, 8, 16) 2, 1, 4, 1
#endif
};
/* clang-format on */
// Width/height lookup tables in units of various block sizes
static const uint8_t block_size_wide[BLOCK_SIZES] = {
static const uint8_t block_size_wide[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
2, 2, 4,
2, 2, 4,
#endif
4, 4, 8, 8, 8, 16, 16, 16, 32, 32, 32, 64, 64, IF_EXT_PARTITION(64, 128, 128)
4, 4, 8, 8, 8, 16, 16,
16, 32, 32, 32, 64, 64, IF_EXT_PARTITION(64, 128, 128) 4,
16, 8, 32
};
static const uint8_t block_size_high[BLOCK_SIZES] = {
static const uint8_t block_size_high[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
2, 4, 2,
2, 4, 2,
#endif
4, 8, 4, 8, 16, 8, 16, 32, 16, 32, 64, 32, 64, IF_EXT_PARTITION(128, 64, 128)
4, 8, 4, 8, 16, 8, 16,
32, 16, 32, 64, 32, 64, IF_EXT_PARTITION(128, 64, 128) 16,
4, 32, 8
};
static const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES] = {
static const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16, IF_EXT_PARTITION(16, 32, 32)
1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16, IF_EXT_PARTITION(16, 32, 32) 1,
4, 2, 8
};
static const uint8_t num_4x4_blocks_high_lookup[BLOCK_SIZES] = {
static const uint8_t num_4x4_blocks_high_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16, IF_EXT_PARTITION(32, 16, 32)
1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16, IF_EXT_PARTITION(32, 16, 32) 4,
1, 8, 2
};
static const uint8_t num_8x8_blocks_wide_lookup[BLOCK_SIZES] = {
static const uint8_t num_8x8_blocks_wide_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, IF_EXT_PARTITION(8, 16, 16)
1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8, IF_EXT_PARTITION(8, 16, 16) 1, 2, 1, 4
};
static const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES] = {
static const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, IF_EXT_PARTITION(16, 8, 16)
1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8, IF_EXT_PARTITION(16, 8, 16) 2, 1, 4, 1
};
static const uint8_t num_16x16_blocks_wide_lookup[BLOCK_SIZES] = {
static const uint8_t num_16x16_blocks_wide_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 4, 4, IF_EXT_PARTITION(4, 8, 8)
1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 4, 4, IF_EXT_PARTITION(4, 8, 8) 1, 1, 1, 2
};
static const uint8_t num_16x16_blocks_high_lookup[BLOCK_SIZES] = {
static const uint8_t num_16x16_blocks_high_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
1, 1, 1,
#endif
1, 1, 1, 1, 1, 1, 1, 2, 1, 2, 4, 2, 4, IF_EXT_PARTITION(8, 4, 8)
1, 1, 1, 1, 1, 1, 1, 2, 1, 2, 4, 2, 4, IF_EXT_PARTITION(8, 4, 8) 1, 1, 2, 1
};
// AOMMIN(3, AOMMIN(b_width_log2(bsize), b_height_log2(bsize)))
@ -130,11 +142,12 @@ static const uint8_t size_group_lookup[BLOCK_SIZES] = {
0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, IF_EXT_PARTITION(3, 3, 3)
};
static const uint8_t num_pels_log2_lookup[BLOCK_SIZES] = {
static const uint8_t num_pels_log2_lookup[BLOCK_SIZES_ALL] = {
#if CONFIG_CB4X4
2, 3, 3,
#endif
4, 5, 5, 6, 7, 7, 8, 9, 9, 10, 11, 11, 12, IF_EXT_PARTITION(13, 13, 14)
4, 5, 5, 6, 7, 7, 8, 9, 9, 10, 11, 11, 12, IF_EXT_PARTITION(13, 13, 14) 6,
6, 8, 8
};
/* clang-format off */
@ -424,9 +437,11 @@ static const BLOCK_SIZE subsize_lookup[PARTITION_TYPES][BLOCK_SIZES] =
};
static const TX_SIZE max_txsize_lookup[BLOCK_SIZES] = {
#if CONFIG_CB4X4
// 2X2, 2X4, 4X2,
#if CONFIG_CHROMA_2X2
TX_2X2, TX_2X2, TX_2X2,
#elif CONFIG_CB4X4
TX_4X4, TX_4X4, TX_4X4,
#endif
// 4X4
TX_4X4,
@ -457,10 +472,12 @@ static const TX_SIZE max_txsize_lookup[BLOCK_SIZES] = {
#if CONFIG_RECT_TX && (CONFIG_EXT_TX || CONFIG_VAR_TX)
static const TX_SIZE max_txsize_rect_lookup[BLOCK_SIZES] = {
#if CONFIG_CB4X4
// 2X2, 2X4, 4X2,
#if CONFIG_CHROMA_2X2
TX_2X2, TX_2X2, TX_2X2,
#endif // CONFIG_CB4X4
#elif CONFIG_CB4X4
TX_4X4, TX_4X4, TX_4X4,
#endif
// 4X4
TX_4X4,
// 4X8, 8X4, 8X8
@ -487,10 +504,49 @@ static const TX_SIZE max_txsize_rect_lookup[BLOCK_SIZES] = {
#endif // CONFIG_EXT_PARTITION
#endif // CONFIG_TX64X64
};
#if CONFIG_EXT_TX && CONFIG_RECT_TX_EXT
static const TX_SIZE quarter_txsize_lookup[BLOCK_SIZES] = {
#if CONFIG_CB4X4
// 2X2, 2X4, 4X2,
TX_INVALID, TX_INVALID, TX_INVALID,
#endif
// 4x4,
TX_INVALID,
// 4x8, 8x4, 8x8,
TX_INVALID, TX_INVALID, TX_INVALID,
// 8x16, 16x8, 16x16,
TX_4X16, TX_16X4, TX_INVALID,
// 16x32, 32x16, 32x32,
TX_8X32, TX_32X8, TX_INVALID,
// 32x64, 64x32, 64x64
TX_INVALID, TX_INVALID, TX_INVALID,
#if CONFIG_EXT_PARTITION
// 64x128, 128x64, 128x128
TX_INVALID, TX_INVALID, TX_INVALID,
#endif
};
#endif // CONFIG_EXT_TX && CONFIG_RECT_TX_EXT
#else
#define max_txsize_rect_lookup max_txsize_lookup
#endif // CONFIG_RECT_TX && (CONFIG_EXT_TX || CONFIG_VAR_TX)
static const TX_TYPE_1D vtx_tab[TX_TYPES] = {
DCT_1D, ADST_1D, DCT_1D, ADST_1D,
#if CONFIG_EXT_TX
FLIPADST_1D, DCT_1D, FLIPADST_1D, ADST_1D, FLIPADST_1D, IDTX_1D,
DCT_1D, IDTX_1D, ADST_1D, IDTX_1D, FLIPADST_1D, IDTX_1D,
#endif // CONFIG_EXT_TX
};
static const TX_TYPE_1D htx_tab[TX_TYPES] = {
DCT_1D, DCT_1D, ADST_1D, ADST_1D,
#if CONFIG_EXT_TX
DCT_1D, FLIPADST_1D, FLIPADST_1D, FLIPADST_1D, ADST_1D, IDTX_1D,
IDTX_1D, DCT_1D, IDTX_1D, ADST_1D, IDTX_1D, FLIPADST_1D,
#endif // CONFIG_EXT_TX
};
#if CONFIG_RECT_TX && (CONFIG_EXT_TX || CONFIG_VAR_TX)
// Same as "max_txsize_lookup[bsize] - TX_8X8", except for rectangular
// block which may use a rectangular transform, in which case it is
@ -571,7 +627,7 @@ static const int32_t intra_tx_size_cat_lookup[BLOCK_SIZES] = {
/* clang-format on */
static const TX_SIZE sub_tx_size_map[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // TX_2X2
#endif
TX_4X4, // TX_4X4
@ -594,7 +650,7 @@ static const TX_SIZE sub_tx_size_map[TX_SIZES_ALL] = {
};
static const TX_SIZE txsize_horz_map[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // TX_2X2
#endif
TX_4X4, // TX_4X4
@ -617,7 +673,7 @@ static const TX_SIZE txsize_horz_map[TX_SIZES_ALL] = {
};
static const TX_SIZE txsize_vert_map[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // TX_2X2
#endif
TX_4X4, // TX_4X4
@ -639,7 +695,7 @@ static const TX_SIZE txsize_vert_map[TX_SIZES_ALL] = {
TX_8X8, // TX_32X8
};
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
#define TX_SIZE_W_MIN 2
#else
#define TX_SIZE_W_MIN 4
@ -647,7 +703,7 @@ static const TX_SIZE txsize_vert_map[TX_SIZES_ALL] = {
// Transform block width in pixels
static const int tx_size_wide[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
2,
#endif
4, 8, 16, 32,
@ -657,7 +713,7 @@ static const int tx_size_wide[TX_SIZES_ALL] = {
4, 8, 8, 16, 16, 32, 4, 16, 8, 32
};
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
#define TX_SIZE_H_MIN 2
#else
#define TX_SIZE_H_MIN 4
@ -665,7 +721,7 @@ static const int tx_size_wide[TX_SIZES_ALL] = {
// Transform block height in pixels
static const int tx_size_high[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
2,
#endif
4, 8, 16, 32,
@ -677,41 +733,41 @@ static const int tx_size_high[TX_SIZES_ALL] = {
// Transform block width in unit
static const int tx_size_wide_unit[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
1, 2, 4, 8, 16,
#if CONFIG_TX64X64
32,
#endif // CONFIG_TX64X64
2, 4, 4, 8, 8, 16, 2, 8, 4, 16
#else // CONFIG_CB4X4
#else // CONFIG_CHROMA_2X2
1, 2, 4, 8,
#if CONFIG_TX64X64
16,
#endif // CONFIG_TX64X64
1, 2, 2, 4, 4, 8, 1, 4, 2, 8
#endif // CONFIG_CB4X4
#endif // CONFIG_CHROMA_2X2
};
// Transform block height in unit
static const int tx_size_high_unit[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
1, 2, 4, 8, 16,
#if CONFIG_TX64X64
32,
#endif // CONFIG_TX64X64
4, 2, 8, 4, 16, 8, 8, 2, 16, 4
#else // CONFIG_CB4X4
#else // CONFIG_CHROMA_2X2
1, 2, 4, 8,
#if CONFIG_TX64X64
16,
#endif // CONFIG_TX64X64
2, 1, 4, 2, 8, 4, 4, 1, 8, 2
#endif // CONFIG_CB4X4
#endif // CONFIG_CHROMA_2X2
};
// Transform block width in log2
static const int tx_size_wide_log2[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
1,
#endif
2, 3, 4, 5,
@ -723,7 +779,7 @@ static const int tx_size_wide_log2[TX_SIZES_ALL] = {
// Transform block height in log2
static const int tx_size_high_log2[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
1,
#endif
2, 3, 4, 5,
@ -733,8 +789,11 @@ static const int tx_size_high_log2[TX_SIZES_ALL] = {
3, 2, 4, 3, 5, 4, 4, 2, 5, 3
};
#define TX_UNIT_WIDE_LOG2 (MI_SIZE_LOG2 - tx_size_wide_log2[0])
#define TX_UNIT_HIGH_LOG2 (MI_SIZE_LOG2 - tx_size_high_log2[0])
static const int tx_size_2d[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
4,
#endif
16, 64, 256, 1024,
@ -745,7 +804,7 @@ static const int tx_size_2d[TX_SIZES_ALL] = {
};
static const BLOCK_SIZE txsize_to_bsize[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
BLOCK_2X2, // TX_2X2
#endif
BLOCK_4X4, // TX_4X4
@ -753,22 +812,22 @@ static const BLOCK_SIZE txsize_to_bsize[TX_SIZES_ALL] = {
BLOCK_16X16, // TX_16X16
BLOCK_32X32, // TX_32X32
#if CONFIG_TX64X64
BLOCK_64X64, // TX_64X64
#endif // CONFIG_TX64X64
BLOCK_4X8, // TX_4X8
BLOCK_8X4, // TX_8X4
BLOCK_8X16, // TX_8X16
BLOCK_16X8, // TX_16X8
BLOCK_16X32, // TX_16X32
BLOCK_32X16, // TX_32X16
BLOCK_INVALID, // TX_4X16
BLOCK_INVALID, // TX_16X4
BLOCK_INVALID, // TX_8X32
BLOCK_INVALID, // TX_32X8
BLOCK_64X64, // TX_64X64
#endif // CONFIG_TX64X64
BLOCK_4X8, // TX_4X8
BLOCK_8X4, // TX_8X4
BLOCK_8X16, // TX_8X16
BLOCK_16X8, // TX_16X8
BLOCK_16X32, // TX_16X32
BLOCK_32X16, // TX_32X16
BLOCK_4X16, // TX_4X16
BLOCK_16X4, // TX_16X4
BLOCK_8X32, // TX_8X32
BLOCK_32X8, // TX_32X8
};
static const TX_SIZE txsize_sqr_map[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // TX_2X2
#endif
TX_4X4, // TX_4X4
@ -791,7 +850,7 @@ static const TX_SIZE txsize_sqr_map[TX_SIZES_ALL] = {
};
static const TX_SIZE txsize_sqr_up_map[TX_SIZES_ALL] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // TX_2X2
#endif
TX_4X4, // TX_4X4
@ -864,6 +923,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
// ss_x == 0 ss_x == 0 ss_x == 1 ss_x == 1
// ss_y == 0 ss_y == 1 ss_y == 0 ss_y == 1
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{
// BLOCK_2X2
{ { TX_2X2, TX_2X2 }, { TX_2X2, TX_2X2 } },
@ -927,15 +987,77 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
#else // CONFIG_CHROMA_2X2
{
// BLOCK_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#if CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
{
// BLOCK_2X4
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#if CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
{
// BLOCK_2X4
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#if CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_TX64X64
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
#endif // CONFIG_CHROMA_2X2
#endif
{
// BLOCK_4X4
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_2X2, TX_2X2 }, { TX_2X2, TX_2X2 } },
{ { TX_4X4, TX_2X2 }, { TX_2X2, TX_2X2 } },
#else
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_CB4X4
#endif // CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -955,7 +1077,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_4X8
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_2X2, TX_2X2 }, { TX_2X2, TX_2X2 } },
{ { TX_4X4, TX_2X2 }, { TX_2X2, TX_2X2 } },
#else
@ -967,7 +1089,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
#if CONFIG_TX64X64
{ { TX_4X8, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_TX64X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X8, TX_4X4 }, { TX_2X2, TX_2X2 } }, // used
#else
{ { TX_4X8, TX_4X4 }, { TX_4X4, TX_4X4 } }, // used
@ -984,7 +1106,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_8X4
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_2X2, TX_2X2 }, { TX_2X2, TX_2X2 } },
{ { TX_4X4, TX_2X2 }, { TX_2X2, TX_2X2 } },
#else
@ -997,7 +1119,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_8X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif // CONFIG_TX64X64
{ { TX_8X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_8X4, TX_2X2 }, { TX_4X4, TX_2X2 } }, // used
#else
{ { TX_8X4, TX_4X4 }, { TX_4X4, TX_4X4 } }, // used
@ -1013,7 +1135,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_8X8
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_2X2, TX_2X2 }, { TX_2X2, TX_2X2 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1036,7 +1158,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_8X16
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1052,14 +1174,14 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_8X16, TX_8X8 }, { TX_4X8, TX_4X8 } },
{ { TX_8X16, TX_8X8 }, { TX_4X8, TX_4X8 } },
{ { TX_8X16, TX_8X8 }, { TX_4X8, TX_4X8 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_4X16, TX_4X8 }, { TX_4X16, TX_4X8 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
{
// BLOCK_16X8
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1076,13 +1198,13 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_16X8, TX_8X4 }, { TX_8X8, TX_8X4 } },
{ { TX_16X8, TX_8X4 }, { TX_8X8, TX_8X4 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_16X4, TX_16X4 }, { TX_8X4, TX_8X4 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
{
// BLOCK_16X16
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1105,7 +1227,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_16X32
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1123,12 +1245,12 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_16X32, TX_16X16 }, { TX_8X16, TX_8X16 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_8X32, TX_8X16 }, { TX_4X16, TX_4X16 } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
},
{
// BLOCK_32X16
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1147,11 +1269,11 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_INVALID, TX_INVALID }, { TX_INVALID, TX_INVALID } },
{ { TX_32X8, TX_32X8 }, { TX_16X8, TX_16X4 } },
},
{
// BLOCK_32X32
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1174,7 +1296,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_32X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1197,7 +1319,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_64X32
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1220,7 +1342,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_64X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1244,7 +1366,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
#if CONFIG_EXT_PARTITION
{
// BLOCK_64X128
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1267,7 +1389,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_128X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1290,7 +1412,7 @@ static const TX_SIZE uv_txsize_lookup[BLOCK_SIZES][TX_SIZES_ALL][2][2] = {
},
{
// BLOCK_128X128
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
@ -1371,7 +1493,7 @@ static const struct {
static const TX_SIZE uvsupertx_size_lookup[TX_SIZES][2][2] = {
// ss_x == 0 ss_x == 0 ss_x == 1 ss_x == 1
// ss_y == 0 ss_y == 1 ss_y == 0 ss_y == 1
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },
#endif
{ { TX_4X4, TX_4X4 }, { TX_4X4, TX_4X4 } },

148
third_party/aom/av1/common/convolve.c vendored
View file

@ -290,12 +290,14 @@ void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
#if USE_EXTRA_FILTER
if (filter_params_x.interp_filter == MULTITAP_SHARP &&
filter_params_y.interp_filter == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
#endif // USE_EXTRA_FILTER
#else
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(*interp_filter);
@ -331,6 +333,140 @@ void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
}
}
#if CONFIG_HIGHBITDEPTH
static INLINE void transpose_uint16(uint16_t *dst, int dst_stride,
const uint16_t *src, int src_stride, int w,
int h) {
int r, c;
for (r = 0; r < h; ++r)
for (c = 0; c < w; ++c) dst[c * dst_stride + r] = src[r * src_stride + c];
}
void av1_highbd_convolve_rounding(const int32_t *src, int src_stride,
uint8_t *dst8, int dst_stride, int w, int h,
int bits, int bd) {
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
int r, c;
for (r = 0; r < h; ++r) {
for (c = 0; c < w; ++c) {
dst[r * dst_stride + c] = clip_pixel_highbd(
ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits), bd);
}
}
}
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) {
int x, y, k;
CONV_BUF_TYPE im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
int im_h = h + filter_params_y->taps - 1;
int im_stride = w;
const int fo_vert = filter_params_y->taps / 2 - 1;
const int fo_horiz = filter_params_x->taps / 2 - 1;
(void)conv_params;
// horizontal filter
const uint16_t *src_horiz = src - fo_vert * src_stride;
const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
*filter_params_x, subpel_x_q4 & SUBPEL_MASK);
for (y = 0; y < im_h; ++y) {
for (x = 0; x < w; ++x) {
CONV_BUF_TYPE sum = 0;
for (k = 0; k < filter_params_x->taps; ++k) {
sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
}
#if CONFIG_COMPOUND_ROUND
im_block[y * im_stride + x] = clip_pixel_highbd(
ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0), bd);
#else
(void)bd;
im_block[y * im_stride + x] =
ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0);
#endif
}
}
// vertical filter
CONV_BUF_TYPE *src_vert = im_block + fo_vert * im_stride;
const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
*filter_params_y, subpel_y_q4 & SUBPEL_MASK);
for (y = 0; y < h; ++y) {
for (x = 0; x < w; ++x) {
CONV_BUF_TYPE sum = 0;
for (k = 0; k < filter_params_y->taps; ++k) {
sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
}
dst[y * dst_stride + x] +=
ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_1);
}
}
}
void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
uint8_t *dst, int dst_stride, int w, int h,
const InterpFilter *interp_filter,
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params, int bd) {
(void)x_step_q4;
(void)y_step_q4;
(void)dst;
(void)dst_stride;
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
#if USE_EXTRA_FILTER
if (filter_params_x.interp_filter == MULTITAP_SHARP &&
filter_params_y.interp_filter == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
#endif
#else
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(*interp_filter);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(*interp_filter);
#endif
const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
if (filter_params_y.taps < filter_params_x.taps) {
uint16_t tr_src[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) *
(MAX_SB_SIZE + MAX_FILTER_TAP - 1)];
int tr_src_stride = MAX_SB_SIZE + MAX_FILTER_TAP - 1;
CONV_BUF_TYPE tr_dst[MAX_SB_SIZE * MAX_SB_SIZE];
int tr_dst_stride = MAX_SB_SIZE;
int fo_vert = filter_params_y.taps / 2 - 1;
int fo_horiz = filter_params_x.taps / 2 - 1;
transpose_uint16(
tr_src, tr_src_stride, src - fo_vert * src_stride - fo_horiz,
src_stride, w + filter_params_x.taps - 1, h + filter_params_y.taps - 1);
transpose_int32(tr_dst, tr_dst_stride, conv_params->dst,
conv_params->dst_stride, w, h);
// horizontal and vertical parameters are swapped because of the transpose
av1_highbd_convolve_2d(tr_src + fo_horiz * tr_src_stride + fo_vert,
tr_src_stride, tr_dst, tr_dst_stride, h, w,
&filter_params_y, &filter_params_x, subpel_y_q4,
subpel_x_q4, conv_params, bd);
transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
tr_dst_stride, h, w);
} else {
av1_highbd_convolve_2d(src, src_stride, conv_params->dst,
conv_params->dst_stride, w, h, &filter_params_x,
&filter_params_y, subpel_x_q4, subpel_y_q4,
conv_params, bd);
}
}
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_CONVOLVE_ROUND
typedef void (*ConvolveFunc)(const uint8_t *src, int src_stride, uint8_t *dst,
@ -394,7 +530,7 @@ static void convolve_helper(const uint8_t *src, int src_stride, uint8_t *dst,
temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
int filter_size;
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
if (interp_filter[0 + 2 * conv_params->ref] == MULTITAP_SHARP &&
interp_filter[1 + 2 * conv_params->ref] == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
@ -427,7 +563,7 @@ static void convolve_helper(const uint8_t *src, int src_stride, uint8_t *dst,
convolve_horiz(temp + (filter_size / 2 - 1), temp_stride, dst, dst_stride,
w, h, filter_params, subpel_x_q4, x_step_q4, conv_params);
} else
#endif // CONFIG_DUAL_FILTER
#endif // CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
{
int intermediate_height;
int temp_stride = MAX_SB_SIZE;
@ -708,15 +844,15 @@ void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
#endif
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
if (interp_filter[0 + 2 * ref_idx] == MULTITAP_SHARP &&
interp_filter[1 + 2 * ref_idx] == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
#endif
#if CONFIG_DUAL_FILTER
if (filter_params_y.taps < filter_params_x.taps) {
int intermediate_width;
int temp_stride = max_intermediate_size;
@ -739,7 +875,7 @@ void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
temp8 + (filter_size / 2 - 1), temp_stride, dst8, dst_stride, w, h,
filter_params, subpel_x_q4, x_step_q4, ref_idx, bd);
} else
#endif // CONFIG_DUAL_FILTER
#endif // CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
{
int intermediate_height;
int temp_stride = MAX_SB_SIZE;

23
third_party/aom/av1/common/convolve.h vendored
View file

@ -33,6 +33,7 @@ typedef struct ConvolveParams {
int round_0;
int round_1;
int plane;
int do_post_rounding;
} ConvolveParams;
static INLINE ConvolveParams get_conv_params(int ref, int plane) {
@ -40,6 +41,7 @@ static INLINE ConvolveParams get_conv_params(int ref, int plane) {
conv_params.ref = ref;
conv_params.round = CONVOLVE_OPT_ROUND;
conv_params.plane = plane;
conv_params.do_post_rounding = 0;
return conv_params;
}
struct AV1Common;
@ -73,11 +75,32 @@ static INLINE ConvolveParams get_conv_params_no_round(int ref, int plane,
conv_params.dst = dst;
conv_params.dst_stride = dst_stride;
conv_params.plane = plane;
conv_params.do_post_rounding = 0;
return conv_params;
}
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h, int bits);
#if CONFIG_HIGHBITDEPTH
void av1_highbd_convolve_rounding(const int32_t *src, int src_stride,
uint8_t *dst8, int dst_stride, int w, int h,
int bits, int bd);
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);
void av1_highbd_convolve_2d_facade(const uint8_t *src8, int src_stride,
uint8_t *dst, int dst_stride, int w, int h,
const InterpFilter *interp_filter,
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params, int bd);
#endif
#endif // CONFIG_CONVOLVE_ROUND
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,

10
third_party/aom/av1/common/debugmodes.c vendored
View file

@ -12,6 +12,7 @@
#include <stdio.h>
#include "av1/common/blockd.h"
#include "av1/common/enums.h"
#include "av1/common/onyxc_int.h"
static void log_frame_info(AV1_COMMON *cm, const char *str, FILE *f) {
@ -35,11 +36,12 @@ 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 ", *((int *)((char *)(&mi[0]->mbmi) + member_offset)));
fprintf(file, "%2d ",
*((char *)((char *)(&mi[0]->mbmi) + member_offset)));
mi++;
}
fprintf(file, "\n");
mi += 8;
mi += MAX_MIB_SIZE;
}
fprintf(file, "\n");
}
@ -67,7 +69,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
mi++;
}
fprintf(mvs, "\n");
mi += 8;
mi += MAX_MIB_SIZE;
}
fprintf(mvs, "\n");
@ -82,7 +84,7 @@ void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
mi++;
}
fprintf(mvs, "\n");
mi += 8;
mi += MAX_MIB_SIZE;
}
fprintf(mvs, "\n");

722
third_party/aom/av1/common/entropy.c vendored
View file

@ -135,7 +135,7 @@ const aom_prob av1_cat6_prob[] = {
};
const uint16_t band_count_table[TX_SIZES_ALL][8] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ 1, 2, 2, 3, 0, 0, 0 },
#endif
{ 1, 2, 3, 4, 3, 16 - 13, 0 }, { 1, 2, 3, 4, 11, 64 - 21, 0 },
@ -151,7 +151,7 @@ const uint16_t band_count_table[TX_SIZES_ALL][8] = {
};
const uint16_t band_cum_count_table[TX_SIZES_ALL][8] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ 0, 1, 3, 6, 10, 13, 16, 0 },
#endif
{ 0, 1, 3, 6, 10, 13, 16, 0 }, { 0, 1, 3, 6, 10, 21, 64, 0 },
@ -627,7 +627,6 @@ const aom_prob av1_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES] = {
// The full source code of the generating program is available in:
// tools/gen_constrained_tokenset.py
//
#if CONFIG_NEW_TOKENSET
// Values for tokens TWO_TOKEN through CATEGORY6_TOKEN included
// in the table here : the ONE_TOKEN probability is
// removed and the probabilities rescaled.
@ -891,275 +890,13 @@ const aom_cdf_prob av1_pareto8_tail_probs[COEFF_PROB_MODELS][TAIL_NODES] = {
{ 31131, 1448, 152, 31, 2, 1, 1, 1, 1 },
{ 31486, 1150, 107, 20, 1, 1, 1, 1, 1 },
};
#elif CONFIG_EC_MULTISYMBOL
// Values for tokens ONE_TOKEN through CATEGORY6_TOKEN included here.
// ZERO_TOKEN and EOB_TOKEN are coded as flags outside this coder.
const aom_cdf_prob
av1_pareto8_token_probs[COEFF_PROB_MODELS][ENTROPY_TOKENS - 2] = {
{ 128, 127, 127, 126, 251, 495, 965, 1832, 3305, 25412 },
{ 256, 254, 252, 249, 492, 959, 1820, 3283, 5365, 19838 },
{ 384, 379, 374, 369, 724, 1392, 2574, 4417, 6568, 15587 },
{ 512, 503, 494, 486, 946, 1795, 3238, 5289, 7184, 12321 },
{ 640, 626, 612, 599, 1159, 2172, 3818, 5946, 7401, 9795 },
{ 768, 748, 728, 709, 1363, 2522, 4324, 6424, 7352, 7830 },
{ 896, 869, 842, 816, 1559, 2847, 4762, 6755, 7131, 6291 },
{ 1024, 988, 954, 921, 1747, 3148, 5139, 6966, 6803, 5078 },
{ 1152, 1107, 1063, 1022, 1926, 3427, 5460, 7080, 6412, 4119 },
{ 1280, 1224, 1171, 1120, 2098, 3685, 5730, 7113, 5991, 3356 },
{ 1408, 1340, 1276, 1216, 2261, 3923, 5955, 7083, 5560, 2746 },
{ 1536, 1455, 1380, 1308, 2418, 4142, 6140, 7001, 5133, 2255 },
{ 1664, 1569, 1481, 1398, 2567, 4342, 6287, 6879, 4721, 1860 },
{ 1792, 1683, 1580, 1485, 2709, 4525, 6401, 6725, 4329, 1539 },
{ 1920, 1794, 1678, 1570, 2845, 4692, 6486, 6546, 3959, 1278 },
{ 2048, 1905, 1773, 1651, 2974, 4844, 6543, 6350, 3615, 1065 },
{ 2176, 2015, 1867, 1731, 3096, 4980, 6576, 6140, 3296, 891 },
{ 2304, 2123, 1958, 1807, 3212, 5104, 6589, 5922, 3002, 747 },
{ 2432, 2231, 2048, 1882, 3322, 5214, 6581, 5698, 2732, 628 },
{ 2560, 2337, 2136, 1953, 3427, 5311, 6557, 5472, 2485, 530 },
{ 2688, 2442, 2222, 2023, 3525, 5397, 6518, 5246, 2259, 448 },
{ 2816, 2547, 2306, 2090, 3618, 5472, 6465, 5021, 2053, 380 },
{ 2944, 2650, 2388, 2154, 3706, 5537, 6401, 4799, 1866, 323 },
{ 3072, 2752, 2468, 2217, 3788, 5591, 6327, 4581, 1696, 276 },
{ 3200, 2853, 2547, 2277, 3866, 5637, 6243, 4369, 1541, 235 },
{ 3328, 2952, 2624, 2335, 3938, 5673, 6152, 4163, 1401, 202 },
{ 3456, 3051, 2699, 2391, 4006, 5702, 6054, 3962, 1274, 173 },
{ 3584, 3149, 2772, 2444, 4070, 5723, 5950, 3769, 1158, 149 },
{ 3712, 3246, 2843, 2496, 4128, 5736, 5842, 3583, 1054, 128 },
{ 3840, 3341, 2913, 2545, 4183, 5743, 5729, 3404, 959, 111 },
{ 3968, 3436, 2981, 2593, 4233, 5743, 5614, 3232, 872, 96 },
{ 4096, 3529, 3048, 2638, 4280, 5737, 5496, 3067, 794, 83 },
{ 4224, 3621, 3113, 2682, 4322, 5726, 5375, 2909, 724, 72 },
{ 4352, 3712, 3176, 2724, 4361, 5709, 5253, 2759, 659, 63 },
{ 4480, 3803, 3237, 2764, 4396, 5687, 5130, 2615, 601, 55 },
{ 4608, 3892, 3297, 2801, 4428, 5661, 5007, 2478, 548, 48 },
{ 4736, 3980, 3355, 2838, 4456, 5631, 4883, 2347, 500, 42 },
{ 4864, 4067, 3412, 2872, 4481, 5596, 4760, 2223, 456, 37 },
{ 4992, 4152, 3467, 2905, 4503, 5558, 4637, 2105, 417, 32 },
{ 5120, 4237, 3521, 2936, 4521, 5516, 4515, 1993, 381, 28 },
{ 5248, 4321, 3573, 2966, 4537, 5471, 4393, 1886, 348, 25 },
{ 5376, 4404, 3623, 2993, 4550, 5424, 4273, 1785, 318, 22 },
{ 5504, 4486, 3672, 3020, 4560, 5373, 4155, 1688, 291, 19 },
{ 5632, 4566, 3720, 3044, 4568, 5321, 4037, 1597, 266, 17 },
{ 5760, 4646, 3766, 3067, 4572, 5265, 3922, 1511, 244, 15 },
{ 5888, 4724, 3811, 3089, 4575, 5208, 3808, 1429, 223, 13 },
{ 6016, 4802, 3854, 3109, 4575, 5148, 3696, 1352, 204, 12 },
{ 6144, 4878, 3895, 3128, 4573, 5088, 3587, 1278, 187, 10 },
{ 6272, 4953, 3936, 3145, 4568, 5025, 3479, 1209, 172, 9 },
{ 6400, 5028, 3975, 3161, 4561, 4961, 3373, 1143, 158, 8 },
{ 6528, 5101, 4012, 3175, 4553, 4896, 3270, 1081, 145, 7 },
{ 6656, 5173, 4048, 3189, 4542, 4830, 3168, 1022, 133, 7 },
{ 6784, 5244, 4083, 3201, 4530, 4763, 3069, 966, 122, 6 },
{ 6912, 5314, 4117, 3212, 4516, 4694, 2973, 913, 112, 5 },
{ 7040, 5383, 4149, 3221, 4500, 4626, 2878, 863, 103, 5 },
{ 7168, 5452, 4180, 3229, 4482, 4556, 2786, 816, 95, 4 },
{ 7296, 5519, 4210, 3236, 4463, 4486, 2696, 771, 87, 4 },
{ 7424, 5585, 4238, 3242, 4442, 4416, 2609, 729, 80, 3 },
{ 7552, 5650, 4265, 3247, 4420, 4345, 2523, 689, 74, 3 },
{ 7680, 5714, 4291, 3251, 4396, 4274, 2440, 651, 68, 3 },
{ 7808, 5777, 4315, 3254, 4371, 4203, 2359, 616, 63, 2 },
{ 7936, 5838, 4339, 3255, 4345, 4132, 2281, 582, 58, 2 },
{ 8064, 5899, 4361, 3256, 4318, 4061, 2204, 550, 53, 2 },
{ 8192, 5959, 4382, 3255, 4289, 3990, 2130, 520, 49, 2 },
{ 8320, 6018, 4402, 3254, 4259, 3919, 2057, 492, 45, 2 },
{ 8448, 6075, 4421, 3252, 4229, 3848, 1987, 465, 42, 1 },
{ 8576, 6133, 4438, 3248, 4197, 3778, 1919, 439, 39, 1 },
{ 8704, 6188, 4455, 3244, 4164, 3708, 1853, 415, 36, 1 },
{ 8832, 6243, 4470, 3239, 4131, 3638, 1789, 392, 33, 1 },
{ 8960, 6297, 4484, 3233, 4096, 3569, 1727, 371, 30, 1 },
{ 9088, 6349, 4497, 3226, 4061, 3500, 1667, 351, 28, 1 },
{ 9216, 6401, 4509, 3219, 4025, 3432, 1608, 331, 26, 1 },
{ 9344, 6452, 4520, 3210, 3989, 3364, 1551, 313, 24, 1 },
{ 9472, 6501, 4530, 3201, 3952, 3297, 1496, 296, 22, 1 },
{ 9600, 6550, 4539, 3191, 3914, 3230, 1443, 280, 20, 1 },
{ 9728, 6597, 4547, 3180, 3875, 3164, 1392, 265, 19, 1 },
{ 9856, 6644, 4554, 3169, 3836, 3098, 1342, 250, 18, 1 },
{ 9984, 6690, 4560, 3157, 3796, 3034, 1293, 237, 16, 1 },
{ 10112, 6734, 4565, 3144, 3756, 2970, 1247, 224, 15, 1 },
{ 10240, 6778, 4568, 3131, 3716, 2907, 1202, 211, 14, 1 },
{ 10368, 6821, 4571, 3117, 3675, 2844, 1158, 200, 13, 1 },
{ 10496, 6862, 4573, 3102, 3634, 2783, 1116, 189, 12, 1 },
{ 10624, 6903, 4574, 3087, 3592, 2722, 1075, 179, 11, 1 },
{ 10752, 6942, 4575, 3071, 3551, 2662, 1035, 169, 10, 1 },
{ 10880, 6981, 4574, 3054, 3508, 2603, 997, 160, 10, 1 },
{ 11008, 7019, 4572, 3038, 3466, 2544, 960, 151, 9, 1 },
{ 11136, 7055, 4570, 3020, 3424, 2487, 924, 143, 8, 1 },
{ 11264, 7091, 4566, 3002, 3381, 2430, 890, 135, 8, 1 },
{ 11392, 7126, 4563, 2984, 3338, 2374, 856, 127, 7, 1 },
{ 11520, 7159, 4557, 2965, 3295, 2319, 824, 121, 7, 1 },
{ 11648, 7193, 4552, 2945, 3252, 2264, 793, 114, 6, 1 },
{ 11776, 7224, 4545, 2925, 3209, 2211, 763, 108, 6, 1 },
{ 11904, 7255, 4538, 2905, 3165, 2159, 734, 102, 5, 1 },
{ 12032, 7285, 4530, 2884, 3122, 2107, 706, 96, 5, 1 },
{ 12160, 7314, 4520, 2863, 3079, 2056, 679, 91, 5, 1 },
{ 12288, 7341, 4511, 2842, 3036, 2006, 653, 86, 4, 1 },
{ 12416, 7368, 4500, 2820, 2993, 1957, 628, 81, 4, 1 },
{ 12544, 7394, 4489, 2797, 2949, 1909, 604, 77, 4, 1 },
{ 12672, 7419, 4477, 2775, 2906, 1861, 581, 73, 3, 1 },
{ 12800, 7443, 4464, 2752, 2863, 1815, 558, 69, 3, 1 },
{ 12928, 7466, 4451, 2729, 2820, 1769, 536, 65, 3, 1 },
{ 13056, 7488, 4437, 2705, 2777, 1724, 516, 61, 3, 1 },
{ 13184, 7509, 4422, 2682, 2734, 1680, 495, 58, 3, 1 },
{ 13312, 7529, 4406, 2658, 2692, 1637, 476, 55, 2, 1 },
{ 13440, 7548, 4390, 2633, 2650, 1595, 457, 52, 2, 1 },
{ 13568, 7567, 4373, 2609, 2607, 1553, 439, 49, 2, 1 },
{ 13696, 7583, 4356, 2584, 2565, 1513, 422, 46, 2, 1 },
{ 13824, 7600, 4337, 2559, 2523, 1473, 405, 44, 2, 1 },
{ 13952, 7615, 4319, 2533, 2482, 1434, 389, 41, 2, 1 },
{ 14080, 7629, 4300, 2508, 2441, 1395, 373, 39, 2, 1 },
{ 14208, 7643, 4280, 2482, 2400, 1358, 358, 37, 1, 1 },
{ 14336, 7655, 4259, 2457, 2359, 1321, 344, 35, 1, 1 },
{ 14464, 7667, 4238, 2431, 2318, 1285, 330, 33, 1, 1 },
{ 14592, 7677, 4217, 2405, 2278, 1250, 316, 31, 1, 1 },
{ 14720, 7687, 4195, 2378, 2238, 1215, 304, 29, 1, 1 },
{ 14848, 7696, 4172, 2352, 2198, 1181, 291, 28, 1, 1 },
{ 14976, 7703, 4149, 2326, 2159, 1148, 279, 26, 1, 1 },
{ 15104, 7710, 4125, 2299, 2119, 1116, 268, 25, 1, 1 },
{ 15232, 7715, 4101, 2272, 2081, 1085, 257, 23, 1, 1 },
{ 15360, 7721, 4076, 2245, 2042, 1054, 246, 22, 1, 1 },
{ 15488, 7724, 4051, 2219, 2004, 1023, 236, 21, 1, 1 },
{ 15616, 7727, 4025, 2192, 1966, 994, 226, 20, 1, 1 },
{ 15744, 7729, 3999, 2164, 1929, 965, 217, 19, 1, 1 },
{ 15872, 7731, 3972, 2137, 1892, 937, 207, 18, 1, 1 },
{ 16000, 7731, 3945, 2110, 1855, 909, 199, 17, 1, 1 },
{ 16128, 7730, 3918, 2083, 1819, 882, 190, 16, 1, 1 },
{ 16256, 7728, 3890, 2056, 1783, 856, 182, 15, 1, 1 },
{ 16384, 7725, 3862, 2029, 1747, 831, 174, 14, 1, 1 },
{ 16512, 7721, 3833, 2002, 1712, 806, 167, 13, 1, 1 },
{ 16640, 7717, 3804, 1975, 1677, 781, 160, 12, 1, 1 },
{ 16768, 7712, 3775, 1947, 1642, 757, 153, 12, 1, 1 },
{ 16896, 7706, 3745, 1920, 1608, 734, 146, 11, 1, 1 },
{ 17024, 7699, 3714, 1893, 1575, 711, 140, 10, 1, 1 },
{ 17152, 7690, 3684, 1866, 1541, 689, 134, 10, 1, 1 },
{ 17280, 7681, 3653, 1839, 1508, 668, 128, 9, 1, 1 },
{ 17408, 7671, 3621, 1812, 1476, 647, 122, 9, 1, 1 },
{ 17536, 7660, 3590, 1785, 1444, 626, 117, 8, 1, 1 },
{ 17664, 7648, 3558, 1758, 1412, 606, 112, 8, 1, 1 },
{ 17792, 7635, 3526, 1731, 1381, 587, 107, 7, 1, 1 },
{ 17920, 7622, 3493, 1704, 1350, 568, 102, 7, 1, 1 },
{ 18048, 7607, 3461, 1678, 1319, 549, 98, 6, 1, 1 },
{ 18176, 7592, 3428, 1651, 1289, 531, 93, 6, 1, 1 },
{ 18304, 7575, 3394, 1625, 1259, 514, 89, 6, 1, 1 },
{ 18432, 7558, 3361, 1598, 1230, 497, 85, 5, 1, 1 },
{ 18560, 7540, 3327, 1572, 1201, 480, 81, 5, 1, 1 },
{ 18688, 7520, 3293, 1546, 1173, 464, 77, 5, 1, 1 },
{ 18816, 7500, 3258, 1520, 1145, 448, 74, 5, 1, 1 },
{ 18944, 7480, 3224, 1494, 1117, 433, 70, 4, 1, 1 },
{ 19072, 7458, 3189, 1468, 1090, 418, 67, 4, 1, 1 },
{ 19200, 7435, 3154, 1442, 1063, 404, 64, 4, 1, 1 },
{ 19328, 7410, 3119, 1417, 1037, 390, 61, 4, 1, 1 },
{ 19456, 7386, 3084, 1392, 1011, 376, 58, 3, 1, 1 },
{ 19584, 7361, 3048, 1366, 986, 363, 55, 3, 1, 1 },
{ 19712, 7335, 3012, 1341, 960, 350, 53, 3, 1, 1 },
{ 19840, 7307, 2977, 1316, 936, 337, 50, 3, 1, 1 },
{ 19968, 7279, 2941, 1291, 911, 325, 48, 3, 1, 1 },
{ 20096, 7251, 2905, 1267, 887, 313, 45, 2, 1, 1 },
{ 20224, 7220, 2868, 1243, 864, 302, 43, 2, 1, 1 },
{ 20352, 7189, 2832, 1218, 841, 291, 41, 2, 1, 1 },
{ 20480, 7158, 2795, 1194, 818, 280, 39, 2, 1, 1 },
{ 20608, 7124, 2759, 1170, 796, 270, 37, 2, 1, 1 },
{ 20736, 7091, 2722, 1147, 774, 259, 35, 2, 1, 1 },
{ 20864, 7056, 2685, 1123, 752, 250, 34, 2, 1, 1 },
{ 20992, 7021, 2648, 1100, 731, 240, 32, 2, 1, 1 },
{ 21120, 6985, 2612, 1077, 710, 231, 30, 1, 1, 1 },
{ 21248, 6948, 2574, 1054, 690, 222, 29, 1, 1, 1 },
{ 21376, 6911, 2537, 1031, 670, 213, 27, 1, 1, 1 },
{ 21504, 6872, 2500, 1008, 650, 205, 26, 1, 1, 1 },
{ 21632, 6831, 2463, 986, 631, 197, 25, 1, 1, 1 },
{ 21760, 6791, 2426, 964, 612, 189, 23, 1, 1, 1 },
{ 21888, 6749, 2389, 942, 594, 181, 22, 1, 1, 1 },
{ 22016, 6707, 2351, 921, 575, 174, 21, 1, 1, 1 },
{ 22144, 6663, 2314, 899, 558, 167, 20, 1, 1, 1 },
{ 22272, 6619, 2277, 878, 540, 160, 19, 1, 1, 1 },
{ 22400, 6574, 2240, 857, 523, 153, 18, 1, 1, 1 },
{ 22528, 6529, 2202, 836, 507, 146, 17, 1, 1, 1 },
{ 22656, 6482, 2165, 816, 490, 140, 16, 1, 1, 1 },
{ 22784, 6435, 2128, 795, 474, 134, 15, 1, 1, 1 },
{ 22912, 6386, 2091, 775, 459, 128, 14, 1, 1, 1 },
{ 23040, 6336, 2054, 756, 443, 123, 13, 1, 1, 1 },
{ 23168, 6286, 2017, 736, 428, 117, 13, 1, 1, 1 },
{ 23296, 6234, 1980, 717, 414, 112, 12, 1, 1, 1 },
{ 23424, 6183, 1943, 698, 399, 107, 11, 1, 1, 1 },
{ 23552, 6130, 1906, 679, 385, 102, 11, 1, 1, 1 },
{ 23680, 6077, 1869, 660, 372, 97, 10, 1, 1, 1 },
{ 23808, 6022, 1833, 642, 358, 93, 9, 1, 1, 1 },
{ 23936, 5966, 1796, 624, 345, 89, 9, 1, 1, 1 },
{ 24064, 5910, 1760, 606, 333, 84, 8, 1, 1, 1 },
{ 24192, 5853, 1724, 588, 320, 80, 8, 1, 1, 1 },
{ 24320, 5796, 1687, 571, 308, 76, 7, 1, 1, 1 },
{ 24448, 5735, 1651, 554, 297, 73, 7, 1, 1, 1 },
{ 24576, 5677, 1615, 537, 285, 69, 6, 1, 1, 1 },
{ 24704, 5615, 1579, 521, 274, 66, 6, 1, 1, 1 },
{ 24832, 5554, 1544, 504, 263, 62, 6, 1, 1, 1 },
{ 24960, 5492, 1508, 488, 253, 59, 5, 1, 1, 1 },
{ 25088, 5428, 1473, 473, 242, 56, 5, 1, 1, 1 },
{ 25216, 5364, 1438, 457, 232, 53, 5, 1, 1, 1 },
{ 25344, 5300, 1403, 442, 222, 50, 4, 1, 1, 1 },
{ 25472, 5233, 1368, 427, 213, 48, 4, 1, 1, 1 },
{ 25600, 5166, 1334, 412, 204, 45, 4, 1, 1, 1 },
{ 25728, 5098, 1299, 398, 195, 43, 4, 1, 1, 1 },
{ 25856, 5030, 1266, 384, 186, 40, 3, 1, 1, 1 },
{ 25984, 4960, 1232, 370, 178, 38, 3, 1, 1, 1 },
{ 26112, 4890, 1198, 356, 170, 36, 3, 1, 1, 1 },
{ 26240, 4819, 1164, 343, 162, 34, 3, 1, 1, 1 },
{ 26368, 4748, 1132, 329, 154, 32, 2, 1, 1, 1 },
{ 26496, 4675, 1098, 317, 147, 30, 2, 1, 1, 1 },
{ 26624, 4602, 1066, 304, 139, 28, 2, 1, 1, 1 },
{ 26752, 4527, 1034, 292, 132, 26, 2, 1, 1, 1 },
{ 26880, 4451, 1001, 280, 126, 25, 2, 1, 1, 1 },
{ 27008, 4375, 970, 268, 119, 23, 2, 1, 1, 1 },
{ 27136, 4299, 938, 256, 113, 21, 2, 1, 1, 1 },
{ 27264, 4221, 907, 245, 107, 20, 1, 1, 1, 1 },
{ 27392, 4142, 876, 234, 101, 19, 1, 1, 1, 1 },
{ 27520, 4063, 846, 223, 95, 17, 1, 1, 1, 1 },
{ 27648, 3982, 815, 213, 90, 16, 1, 1, 1, 1 },
{ 27776, 3900, 786, 202, 85, 15, 1, 1, 1, 1 },
{ 27904, 3818, 756, 192, 80, 14, 1, 1, 1, 1 },
{ 28032, 3734, 727, 183, 75, 13, 1, 1, 1, 1 },
{ 28160, 3651, 698, 173, 70, 12, 1, 1, 1, 1 },
{ 28288, 3566, 669, 164, 66, 11, 1, 1, 1, 1 },
{ 28416, 3481, 641, 155, 61, 10, 1, 1, 1, 1 },
{ 28544, 3393, 614, 147, 57, 9, 1, 1, 1, 1 },
{ 28672, 3306, 586, 138, 53, 9, 1, 1, 1, 1 },
{ 28800, 3217, 559, 130, 50, 8, 1, 1, 1, 1 },
{ 28928, 3128, 533, 122, 46, 7, 1, 1, 1, 1 },
{ 29056, 3037, 507, 114, 43, 7, 1, 1, 1, 1 },
{ 29184, 2947, 481, 107, 39, 6, 1, 1, 1, 1 },
{ 29312, 2855, 456, 100, 36, 5, 1, 1, 1, 1 },
{ 29440, 2762, 431, 93, 33, 5, 1, 1, 1, 1 },
{ 29568, 2668, 407, 86, 31, 4, 1, 1, 1, 1 },
{ 29696, 2573, 383, 80, 28, 4, 1, 1, 1, 1 },
{ 29824, 2478, 359, 74, 25, 4, 1, 1, 1, 1 },
{ 29952, 2381, 337, 68, 23, 3, 1, 1, 1, 1 },
{ 30080, 2284, 314, 62, 21, 3, 1, 1, 1, 1 },
{ 30208, 2185, 293, 57, 19, 2, 1, 1, 1, 1 },
{ 30336, 2086, 271, 52, 17, 2, 1, 1, 1, 1 },
{ 30464, 1986, 250, 47, 15, 2, 1, 1, 1, 1 },
{ 30592, 1885, 230, 42, 13, 2, 1, 1, 1, 1 },
{ 30720, 1782, 211, 38, 12, 1, 1, 1, 1, 1 },
{ 30848, 1679, 192, 34, 10, 1, 1, 1, 1, 1 },
{ 30976, 1575, 173, 30, 9, 1, 1, 1, 1, 1 },
{ 31104, 1469, 156, 26, 8, 1, 1, 1, 1, 1 },
{ 31232, 1364, 138, 23, 6, 1, 1, 1, 1, 1 },
{ 31360, 1257, 122, 19, 5, 1, 1, 1, 1, 1 },
{ 31488, 1149, 106, 16, 4, 1, 1, 1, 1, 1 },
{ 31616, 1038, 91, 14, 4, 1, 1, 1, 1, 1 },
{ 31744, 928, 77, 11, 3, 1, 1, 1, 1, 1 },
{ 31872, 816, 64, 9, 2, 1, 1, 1, 1, 1 },
{ 32000, 703, 51, 7, 2, 1, 1, 1, 1, 1 },
{ 32128, 589, 40, 5, 1, 1, 1, 1, 1, 1 },
{ 32256, 473, 29, 4, 1, 1, 1, 1, 1, 1 },
{ 32384, 357, 19, 2, 1, 1, 1, 1, 1, 1 },
{ 32512, 238, 11, 1, 1, 1, 1, 1, 1, 1 },
{ 32640, 117, 4, 1, 1, 1, 1, 1, 1, 1 },
};
#endif // CONFIG_NEW_TOKENSET
/* clang-format off */
#if CONFIG_Q_ADAPT_PROBS
const av1_coeff_probs_model
default_qctx_coef_probs[QCTX_BINS][TX_SIZES][PLANE_TYPES] = {
{ // Q_Index 0
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ // TX_SIZE 0
{ // Y plane
{ // Intra
@ -1825,7 +1562,7 @@ default_qctx_coef_probs[QCTX_BINS][TX_SIZES][PLANE_TYPES] = {
#endif // CONFIG_TX64X64
},
{ // Q_Index 1
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ // TX_SIZE 0
{ // Y plane
{ // Intra
@ -2491,7 +2228,7 @@ default_qctx_coef_probs[QCTX_BINS][TX_SIZES][PLANE_TYPES] = {
#endif // CONFIG_TX64X64
},
{ // Q_Index 2
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ // TX_SIZE 0
{ // Y plane
{ // Intra
@ -3157,7 +2894,7 @@ default_qctx_coef_probs[QCTX_BINS][TX_SIZES][PLANE_TYPES] = {
#endif // CONFIG_TX64X64
},
{ // Q_Index 3
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ // TX_SIZE 0
{ // Y plane
{ // Intra
@ -3824,7 +3561,6 @@ default_qctx_coef_probs[QCTX_BINS][TX_SIZES][PLANE_TYPES] = {
},
};
#else
#if CONFIG_NEW_TOKENSET
static const av1_coeff_probs_model default_coef_probs_4x4[PLANE_TYPES] = {
{ // Y plane
{ // Intra
@ -4249,343 +3985,6 @@ static const av1_coeff_probs_model default_coef_probs_32x32[PLANE_TYPES] = {
}
}
};
#else // CONFIG_NEW_TOKENSET
static const av1_coeff_probs_model default_coef_probs_4x4[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 195, 29, 183 }, { 84, 49, 136 }, { 8, 42, 71 }
}, { // Band 1
{ 31, 107, 169 }, { 35, 99, 159 }, { 17, 82, 140 },
{ 8, 66, 114 }, { 2, 44, 76 }, { 1, 19, 32 }
}, { // Band 2
{ 40, 132, 201 }, { 29, 114, 187 }, { 13, 91, 157 },
{ 7, 75, 127 }, { 3, 58, 95 }, { 1, 28, 47 }
}, { // Band 3
{ 69, 142, 221 }, { 42, 122, 201 }, { 15, 91, 159 },
{ 6, 67, 121 }, { 1, 42, 77 }, { 1, 17, 31 }
}, { // Band 4
{ 102, 148, 228 }, { 67, 117, 204 }, { 17, 82, 154 },
{ 6, 59, 114 }, { 2, 39, 75 }, { 1, 15, 29 }
}, { // Band 5
{ 156, 57, 233 }, { 119, 57, 212 }, { 58, 48, 163 },
{ 29, 40, 124 }, { 12, 30, 81 }, { 3, 12, 31 }
}
}, { // Inter
{ // Band 0
{ 191, 107, 226 }, { 124, 117, 204 }, { 25, 99, 155 }
}, { // Band 1
{ 29, 148, 210 }, { 37, 126, 194 }, { 8, 93, 157 },
{ 2, 68, 118 }, { 1, 39, 69 }, { 1, 17, 33 }
}, { // Band 2
{ 41, 151, 213 }, { 27, 123, 193 }, { 3, 82, 144 },
{ 1, 58, 105 }, { 1, 32, 60 }, { 1, 13, 26 }
}, { // Band 3
{ 59, 159, 220 }, { 23, 126, 198 }, { 4, 88, 151 },
{ 1, 66, 114 }, { 1, 38, 71 }, { 1, 18, 34 }
}, { // Band 4
{ 114, 136, 232 }, { 51, 114, 207 }, { 11, 83, 155 },
{ 3, 56, 105 }, { 1, 33, 65 }, { 1, 17, 34 }
}, { // Band 5
{ 149, 65, 234 }, { 121, 57, 215 }, { 61, 49, 166 },
{ 28, 36, 114 }, { 12, 25, 76 }, { 3, 16, 42 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 214, 49, 220 }, { 132, 63, 188 }, { 42, 65, 137 }
}, { // Band 1
{ 85, 137, 221 }, { 104, 131, 216 }, { 49, 111, 192 },
{ 21, 87, 155 }, { 2, 49, 87 }, { 1, 16, 28 }
}, { // Band 2
{ 89, 163, 230 }, { 90, 137, 220 }, { 29, 100, 183 },
{ 10, 70, 135 }, { 2, 42, 81 }, { 1, 17, 33 }
}, { // Band 3
{ 108, 167, 237 }, { 55, 133, 222 }, { 15, 97, 179 },
{ 4, 72, 135 }, { 1, 45, 85 }, { 1, 19, 38 }
}, { // Band 4
{ 124, 146, 240 }, { 66, 124, 224 }, { 17, 88, 175 },
{ 4, 58, 122 }, { 1, 36, 75 }, { 1, 18, 37 }
}, { // Band 5
{ 141, 79, 241 }, { 126, 70, 227 }, { 66, 58, 182 },
{ 30, 44, 136 }, { 12, 34, 96 }, { 2, 20, 47 }
}
}, { // Inter
{ // Band 0
{ 229, 99, 249 }, { 143, 111, 235 }, { 46, 109, 192 }
}, { // Band 1
{ 82, 158, 236 }, { 94, 146, 224 }, { 25, 117, 191 },
{ 9, 87, 149 }, { 3, 56, 99 }, { 1, 33, 57 }
}, { // Band 2
{ 83, 167, 237 }, { 68, 145, 222 }, { 10, 103, 177 },
{ 2, 72, 131 }, { 1, 41, 79 }, { 1, 20, 39 }
}, { // Band 3
{ 99, 167, 239 }, { 47, 141, 224 }, { 10, 104, 178 },
{ 2, 73, 133 }, { 1, 44, 85 }, { 1, 22, 47 }
}, { // Band 4
{ 127, 145, 243 }, { 71, 129, 228 }, { 17, 93, 177 },
{ 3, 61, 124 }, { 1, 41, 84 }, { 1, 21, 52 }
}, { // Band 5
{ 157, 78, 244 }, { 140, 72, 231 }, { 69, 58, 184 },
{ 31, 44, 137 }, { 14, 38, 105 }, { 8, 23, 61 }
}
}
}
};
static const av1_coeff_probs_model default_coef_probs_8x8[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 125, 34, 187 }, { 52, 41, 133 }, { 6, 31, 56 }
}, { // Band 1
{ 37, 109, 153 }, { 51, 102, 147 }, { 23, 87, 128 },
{ 8, 67, 101 }, { 1, 41, 63 }, { 1, 19, 29 }
}, { // Band 2
{ 31, 154, 185 }, { 17, 127, 175 }, { 6, 96, 145 },
{ 2, 73, 114 }, { 1, 51, 82 }, { 1, 28, 45 }
}, { // Band 3
{ 23, 163, 200 }, { 10, 131, 185 }, { 2, 93, 148 },
{ 1, 67, 111 }, { 1, 41, 69 }, { 1, 14, 24 }
}, { // Band 4
{ 29, 176, 217 }, { 12, 145, 201 }, { 3, 101, 156 },
{ 1, 69, 111 }, { 1, 39, 63 }, { 1, 14, 23 }
}, { // Band 5
{ 57, 192, 233 }, { 25, 154, 215 }, { 6, 109, 167 },
{ 3, 78, 118 }, { 1, 48, 69 }, { 1, 21, 29 }
}
}, { // Inter
{ // Band 0
{ 202, 105, 245 }, { 108, 106, 216 }, { 18, 90, 144 }
}, { // Band 1
{ 33, 172, 219 }, { 64, 149, 206 }, { 14, 117, 177 },
{ 5, 90, 141 }, { 2, 61, 95 }, { 1, 37, 57 }
}, { // Band 2
{ 33, 179, 220 }, { 11, 140, 198 }, { 1, 89, 148 },
{ 1, 60, 104 }, { 1, 33, 57 }, { 1, 12, 21 }
}, { // Band 3
{ 30, 181, 221 }, { 8, 141, 198 }, { 1, 87, 145 },
{ 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 20 }
}, { // Band 4
{ 32, 186, 224 }, { 7, 142, 198 }, { 1, 86, 143 },
{ 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 22 }
}, { // Band 5
{ 57, 192, 227 }, { 20, 143, 204 }, { 3, 96, 154 },
{ 1, 68, 112 }, { 1, 42, 69 }, { 1, 19, 32 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 212, 35, 215 }, { 113, 47, 169 }, { 29, 48, 105 }
}, { // Band 1
{ 74, 129, 203 }, { 106, 120, 203 }, { 49, 107, 178 },
{ 19, 84, 144 }, { 4, 50, 84 }, { 1, 15, 25 }
}, { // Band 2
{ 71, 172, 217 }, { 44, 141, 209 }, { 15, 102, 173 },
{ 6, 76, 133 }, { 2, 51, 89 }, { 1, 24, 42 }
}, { // Band 3
{ 64, 185, 231 }, { 31, 148, 216 }, { 8, 103, 175 },
{ 3, 74, 131 }, { 1, 46, 81 }, { 1, 18, 30 }
}, { // Band 4
{ 65, 196, 235 }, { 25, 157, 221 }, { 5, 105, 174 },
{ 1, 67, 120 }, { 1, 38, 69 }, { 1, 15, 30 }
}, { // Band 5
{ 65, 204, 238 }, { 30, 156, 224 }, { 7, 107, 177 },
{ 2, 70, 124 }, { 1, 42, 73 }, { 1, 18, 34 }
}
}, { // Inter
{ // Band 0
{ 225, 86, 251 }, { 144, 104, 235 }, { 42, 99, 181 }
}, { // Band 1
{ 85, 175, 239 }, { 112, 165, 229 }, { 29, 136, 200 },
{ 12, 103, 162 }, { 6, 77, 123 }, { 2, 53, 84 }
}, { // Band 2
{ 75, 183, 239 }, { 30, 155, 221 }, { 3, 106, 171 },
{ 1, 74, 128 }, { 1, 44, 76 }, { 1, 17, 28 }
}, { // Band 3
{ 73, 185, 240 }, { 27, 159, 222 }, { 2, 107, 172 },
{ 1, 75, 127 }, { 1, 42, 73 }, { 1, 17, 29 }
}, { // Band 4
{ 62, 190, 238 }, { 21, 159, 222 }, { 2, 107, 172 },
{ 1, 72, 122 }, { 1, 40, 71 }, { 1, 18, 32 }
}, { // Band 5
{ 61, 199, 240 }, { 27, 161, 226 }, { 4, 113, 180 },
{ 1, 76, 129 }, { 1, 46, 80 }, { 1, 23, 41 }
}
}
}
};
static const av1_coeff_probs_model default_coef_probs_16x16[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 7, 27, 153 }, { 5, 30, 95 }, { 1, 16, 30 }
}, { // Band 1
{ 50, 75, 127 }, { 57, 75, 124 }, { 27, 67, 108 },
{ 10, 54, 86 }, { 1, 33, 52 }, { 1, 12, 18 }
}, { // Band 2
{ 43, 125, 151 }, { 26, 108, 148 }, { 7, 83, 122 },
{ 2, 59, 89 }, { 1, 38, 60 }, { 1, 17, 27 }
}, { // Band 3
{ 23, 144, 163 }, { 13, 112, 154 }, { 2, 75, 117 },
{ 1, 50, 81 }, { 1, 31, 51 }, { 1, 14, 23 }
}, { // Band 4
{ 18, 162, 185 }, { 6, 123, 171 }, { 1, 78, 125 },
{ 1, 51, 86 }, { 1, 31, 54 }, { 1, 14, 23 }
}, { // Band 5
{ 15, 199, 227 }, { 3, 150, 204 }, { 1, 91, 146 },
{ 1, 55, 95 }, { 1, 30, 53 }, { 1, 11, 20 }
}
}, { // Inter
{ // Band 0
{ 19, 55, 240 }, { 19, 59, 196 }, { 3, 52, 105 }
}, { // Band 1
{ 41, 166, 207 }, { 104, 153, 199 }, { 31, 123, 181 },
{ 14, 101, 152 }, { 5, 72, 106 }, { 1, 36, 52 }
}, { // Band 2
{ 35, 176, 211 }, { 12, 131, 190 }, { 2, 88, 144 },
{ 1, 60, 101 }, { 1, 36, 60 }, { 1, 16, 28 }
}, { // Band 3
{ 28, 183, 213 }, { 8, 134, 191 }, { 1, 86, 142 },
{ 1, 56, 96 }, { 1, 30, 53 }, { 1, 12, 20 }
}, { // Band 4
{ 20, 190, 215 }, { 4, 135, 192 }, { 1, 84, 139 },
{ 1, 53, 91 }, { 1, 28, 49 }, { 1, 11, 20 }
}, { // Band 5
{ 13, 196, 216 }, { 2, 137, 192 }, { 1, 86, 143 },
{ 1, 57, 99 }, { 1, 32, 56 }, { 1, 13, 24 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 211, 29, 217 }, { 96, 47, 156 }, { 22, 43, 87 }
}, { // Band 1
{ 78, 120, 193 }, { 111, 116, 186 }, { 46, 102, 164 },
{ 15, 80, 128 }, { 2, 49, 76 }, { 1, 18, 28 }
}, { // Band 2
{ 71, 161, 203 }, { 42, 132, 192 }, { 10, 98, 150 },
{ 3, 69, 109 }, { 1, 44, 70 }, { 1, 18, 29 }
}, { // Band 3
{ 57, 186, 211 }, { 30, 140, 196 }, { 4, 93, 146 },
{ 1, 62, 102 }, { 1, 38, 65 }, { 1, 16, 27 }
}, { // Band 4
{ 47, 199, 217 }, { 14, 145, 196 }, { 1, 88, 142 },
{ 1, 57, 98 }, { 1, 36, 62 }, { 1, 15, 26 }
}, { // Band 5
{ 26, 219, 229 }, { 5, 155, 207 }, { 1, 94, 151 },
{ 1, 60, 104 }, { 1, 36, 62 }, { 1, 16, 28 }
}
}, { // Inter
{ // Band 0
{ 233, 29, 248 }, { 146, 47, 220 }, { 43, 52, 140 }
}, { // Band 1
{ 100, 163, 232 }, { 179, 161, 222 }, { 63, 142, 204 },
{ 37, 113, 174 }, { 26, 89, 137 }, { 18, 68, 97 }
}, { // Band 2
{ 85, 181, 230 }, { 32, 146, 209 }, { 7, 100, 164 },
{ 3, 71, 121 }, { 1, 45, 77 }, { 1, 18, 30 }
}, { // Band 3
{ 65, 187, 230 }, { 20, 148, 207 }, { 2, 97, 159 },
{ 1, 68, 116 }, { 1, 40, 70 }, { 1, 14, 29 }
}, { // Band 4
{ 40, 194, 227 }, { 8, 147, 204 }, { 1, 94, 155 },
{ 1, 65, 112 }, { 1, 39, 66 }, { 1, 14, 26 }
}, { // Band 5
{ 16, 208, 228 }, { 3, 151, 207 }, { 1, 98, 160 },
{ 1, 67, 117 }, { 1, 41, 74 }, { 1, 17, 31 }
}
}
}
};
static const av1_coeff_probs_model default_coef_probs_32x32[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 17, 38, 140 }, { 7, 34, 80 }, { 1, 17, 29 }
}, { // Band 1
{ 37, 75, 128 }, { 41, 76, 128 }, { 26, 66, 116 },
{ 12, 52, 94 }, { 2, 32, 55 }, { 1, 10, 16 }
}, { // Band 2
{ 50, 127, 154 }, { 37, 109, 152 }, { 16, 82, 121 },
{ 5, 59, 85 }, { 1, 35, 54 }, { 1, 13, 20 }
}, { // Band 3
{ 40, 142, 167 }, { 17, 110, 157 }, { 2, 71, 112 },
{ 1, 44, 72 }, { 1, 27, 45 }, { 1, 11, 17 }
}, { // Band 4
{ 30, 175, 188 }, { 9, 124, 169 }, { 1, 74, 116 },
{ 1, 48, 78 }, { 1, 30, 49 }, { 1, 11, 18 }
}, { // Band 5
{ 10, 222, 223 }, { 2, 150, 194 }, { 1, 83, 128 },
{ 1, 48, 79 }, { 1, 27, 45 }, { 1, 11, 17 }
}
}, { // Inter
{ // Band 0
{ 36, 41, 235 }, { 29, 36, 193 }, { 10, 27, 111 }
}, { // Band 1
{ 85, 165, 222 }, { 177, 162, 215 }, { 110, 135, 195 },
{ 57, 113, 168 }, { 23, 83, 120 }, { 10, 49, 61 }
}, { // Band 2
{ 85, 190, 223 }, { 36, 139, 200 }, { 5, 90, 146 },
{ 1, 60, 103 }, { 1, 38, 65 }, { 1, 18, 30 }
}, { // Band 3
{ 72, 202, 223 }, { 23, 141, 199 }, { 2, 86, 140 },
{ 1, 56, 97 }, { 1, 36, 61 }, { 1, 16, 27 }
}, { // Band 4
{ 55, 218, 225 }, { 13, 145, 200 }, { 1, 86, 141 },
{ 1, 57, 99 }, { 1, 35, 61 }, { 1, 13, 22 }
}, { // Band 5
{ 15, 235, 212 }, { 1, 132, 184 }, { 1, 84, 139 },
{ 1, 57, 97 }, { 1, 34, 56 }, { 1, 14, 23 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 181, 21, 201 }, { 61, 37, 123 }, { 10, 38, 71 }
}, { // Band 1
{ 47, 106, 172 }, { 95, 104, 173 }, { 42, 93, 159 },
{ 18, 77, 131 }, { 4, 50, 81 }, { 1, 17, 23 }
}, { // Band 2
{ 62, 147, 199 }, { 44, 130, 189 }, { 28, 102, 154 },
{ 18, 75, 115 }, { 2, 44, 65 }, { 1, 12, 19 }
}, { // Band 3
{ 55, 153, 210 }, { 24, 130, 194 }, { 3, 93, 146 },
{ 1, 61, 97 }, { 1, 31, 50 }, { 1, 10, 16 }
}, { // Band 4
{ 49, 186, 223 }, { 17, 148, 204 }, { 1, 96, 142 },
{ 1, 53, 83 }, { 1, 26, 44 }, { 1, 11, 17 }
}, { // Band 5
{ 13, 217, 212 }, { 2, 136, 180 }, { 1, 78, 124 },
{ 1, 50, 83 }, { 1, 29, 49 }, { 1, 14, 23 }
}
}, { // Inter
{ // Band 0
{ 197, 13, 247 }, { 82, 17, 222 }, { 25, 17, 162 }
}, { // Band 1
{ 126, 186, 247 }, { 234, 191, 243 }, { 176, 177, 234 },
{ 104, 158, 220 }, { 66, 128, 186 }, { 55, 90, 137 }
}, { // Band 2
{ 111, 197, 242 }, { 46, 158, 219 }, { 9, 104, 171 },
{ 2, 65, 125 }, { 1, 44, 80 }, { 1, 17, 91 }
}, { // Band 3
{ 104, 208, 245 }, { 39, 168, 224 }, { 3, 109, 162 },
{ 1, 79, 124 }, { 1, 50, 102 }, { 1, 43, 102 }
}, { // Band 4
{ 84, 220, 246 }, { 31, 177, 231 }, { 2, 115, 180 },
{ 1, 79, 134 }, { 1, 55, 77 }, { 1, 60, 79 }
}, { // Band 5
{ 43, 243, 240 }, { 8, 180, 217 }, { 1, 115, 166 },
{ 1, 84, 121 }, { 1, 51, 67 }, { 1, 16, 6 }
}
}
}
};
#endif // CONFIG_NEW_TOKENSET
#if CONFIG_TX64X64
// FIXME. Optimize for EC_MULTISYMBOL
@ -4674,9 +4073,20 @@ static const av1_coeff_probs_model default_coef_probs_64x64[PLANE_TYPES] = {
};
#endif // CONFIG_TX64X64
#endif // CONFIG_Q_ADAPT_PROBS
#if CONFIG_NEW_TOKENSET
static const aom_prob av1_default_blockzero_probs[TX_SIZES][PLANE_TYPES]
[REF_TYPES][BLOCKZ_CONTEXTS] = {
#if CONFIG_CHROMA_2X2
{ // TX_2x2
{ // Y plane
{ 195, 84, 8, }, // Intra
{ 191, 124, 25, }, // Inter
},
{ // UV plane
{ 214, 132, 42, }, // Intra
{ 229, 143, 46, }, // Inter
},
},
#endif
{ // TX_4x4
{ // Y plane
{ 195, 84, 8, }, // Intra
@ -5915,7 +5325,6 @@ static const coeff_cdf_model default_coef_head_cdf_32x32[PLANE_TYPES] = {
{AOM_ICDF(3009), AOM_ICDF(3246), AOM_ICDF(10158), AOM_ICDF(10533),
AOM_ICDF(32768) } } } }
};
#endif // CONFIG_NEW_TOKENSET
/* clang-format on */
@ -5930,8 +5339,6 @@ void av1_model_to_full_probs(const aom_prob *model, aom_prob *full) {
extend_to_full_distribution(&full[UNCONSTRAINED_NODES], model[PIVOT_NODE]);
}
#if CONFIG_NEW_TOKENSET
static void build_tail_cdfs(aom_cdf_prob cdf_tail[CDF_SIZE(ENTROPY_TOKENS)],
aom_cdf_prob cdf_head[CDF_SIZE(ENTROPY_TOKENS)],
int band_zero) {
@ -6041,7 +5448,7 @@ static void av1_default_coef_cdfs(FRAME_CONTEXT *fc) {
for (j = 0; j < REF_TYPES; ++j)
for (k = 0; k < COEF_BANDS; ++k)
for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
av1_copy(fc->coef_head_cdfs[TX_2X2][i][j][k][l],
default_coef_head_cdf_4x4[i][j][k][l]);
#endif
@ -6074,19 +5481,6 @@ void av1_coef_head_cdfs(FRAME_CONTEXT *fc) {
}
}
#elif CONFIG_EC_MULTISYMBOL
static void build_token_cdfs(const aom_prob *pdf_model,
aom_cdf_prob cdf[ENTROPY_TOKENS + 1]) {
int i, sum = 0;
assert(pdf_model[2] != 0);
for (i = 0; i < ENTROPY_TOKENS - 2; ++i) {
sum += av1_pareto8_token_probs[pdf_model[2] - 1][i];
cdf[i] = AOM_ICDF(sum);
}
}
#endif // CONFIG_NEW_TOKENSET
#if CONFIG_EC_MULTISYMBOL
void av1_coef_pareto_cdfs(FRAME_CONTEXT *fc) {
/* Build the tail based on a Pareto distribution */
TX_SIZE t;
@ -6096,15 +5490,9 @@ void av1_coef_pareto_cdfs(FRAME_CONTEXT *fc) {
for (j = 0; j < REF_TYPES; ++j)
for (k = 0; k < COEF_BANDS; ++k)
for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
#if CONFIG_NEW_TOKENSET
build_tail_cdfs(fc->coef_tail_cdfs[t][i][j][k][l],
fc->coef_head_cdfs[t][i][j][k][l], k == 0);
#else
build_token_cdfs(fc->coef_probs[t][i][j][k][l],
fc->coef_cdfs[t][i][j][k][l]);
#endif
}
#endif
void av1_default_coef_probs(AV1_COMMON *cm) {
#if CONFIG_Q_ADAPT_PROBS
@ -6112,7 +5500,7 @@ void av1_default_coef_probs(AV1_COMMON *cm) {
ROUND_POWER_OF_TWO(cm->base_qindex, 8 - QCTX_BIN_BITS), QCTX_BINS - 1);
av1_copy(cm->fc->coef_probs, default_qctx_coef_probs[index]);
#else
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
av1_copy(cm->fc->coef_probs[TX_2X2], default_coef_probs_4x4);
#endif
av1_copy(cm->fc->coef_probs[TX_4X4], default_coef_probs_4x4);
@ -6123,44 +5511,29 @@ void av1_default_coef_probs(AV1_COMMON *cm) {
av1_copy(cm->fc->coef_probs[TX_64X64], default_coef_probs_64x64);
#endif // CONFIG_TX64X64
#endif // CONFIG_Q_ADAPT_PROBS
#if CONFIG_NEW_TOKENSET
av1_copy(cm->fc->blockzero_probs, av1_default_blockzero_probs);
#endif
#if CONFIG_NEW_TOKENSET
/* Load the head tokens */
av1_default_coef_cdfs(cm->fc);
#endif
#if CONFIG_EC_MULTISYMBOL
av1_coef_pareto_cdfs(cm->fc);
#endif // CONFIG_EC_MULTISYMBOL
}
#if !CONFIG_LV_MAP
static void adapt_coef_probs(AV1_COMMON *cm, TX_SIZE tx_size,
unsigned int count_sat,
unsigned int update_factor) {
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
const FRAME_CONTEXT *pre_fc = cm->pre_fc;
av1_coeff_probs_model *const probs = cm->fc->coef_probs[tx_size];
#if CONFIG_SUBFRAME_PROB_UPDATE
const av1_coeff_probs_model *const pre_probs =
cm->partial_prob_update
? (const av1_coeff_probs_model *)cm->starting_coef_probs[tx_size]
: pre_fc->coef_probs[tx_size];
#else
const av1_coeff_probs_model *const pre_probs = pre_fc->coef_probs[tx_size];
#endif // CONFIG_SUBFRAME_PROB_UPDATE
const av1_coeff_count_model *const counts =
(const av1_coeff_count_model *)cm->counts.coef[tx_size];
const unsigned int(*eob_counts)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] =
(const unsigned int(*)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS])
cm->counts.eob_branch[tx_size];
#if CONFIG_NEW_TOKENSET
const av1_blockz_probs_model *const pre_blockz_probs =
pre_fc->blockzero_probs[tx_size];
av1_blockz_probs_model *const blockz_probs = cm->fc->blockzero_probs[tx_size];
const av1_blockz_count_model *const blockz_counts =
(const av1_blockz_count_model *)&cm->counts.blockz_count[tx_size][0];
#endif
int i, j, k, l, m;
#if CONFIG_RECT_TX
assert(!is_rect_tx(tx_size));
@ -6183,7 +5556,6 @@ static void adapt_coef_probs(AV1_COMMON *cm, TX_SIZE tx_size,
count_sat, update_factor);
}
#if CONFIG_NEW_TOKENSET
for (i = 0; i < PLANE_TYPES; ++i) {
for (j = 0; j < REF_TYPES; ++j) {
for (k = 0; k < BLOCKZ_CONTEXTS; ++k) {
@ -6195,7 +5567,6 @@ static void adapt_coef_probs(AV1_COMMON *cm, TX_SIZE tx_size,
}
}
}
#endif
}
#endif // !CONFIG_LV_MAP
@ -6209,9 +5580,6 @@ void av1_adapt_coef_probs(AV1_COMMON *cm) {
update_factor = COEF_MAX_UPDATE_FACTOR;
count_sat = COEF_COUNT_SAT;
}
#if CONFIG_SUBFRAME_PROB_UPDATE
if (cm->partial_prob_update == 1) update_factor = COEF_MAX_UPDATE_FACTOR;
#endif // CONFIG_SUBFRAME_PROB_UPDATE
#if CONFIG_LV_MAP
av1_adapt_txb_probs(cm, count_sat, update_factor);
@ -6222,18 +5590,6 @@ void av1_adapt_coef_probs(AV1_COMMON *cm) {
#endif
}
#if CONFIG_SUBFRAME_PROB_UPDATE
void av1_partial_adapt_probs(AV1_COMMON *cm, int mi_row, int mi_col) {
(void)mi_row;
(void)mi_col;
if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
cm->partial_prob_update = 1;
av1_adapt_coef_probs(cm);
}
}
#endif // CONFIG_SUBFRAME_PROB_UPDATE
#if CONFIG_EC_ADAPT
static void av1_average_cdf(aom_cdf_prob *cdf_ptr[], aom_cdf_prob *fc_cdf_ptr,
int cdf_size, const int num_tiles) {
@ -6263,26 +5619,21 @@ static void av1_average_cdf(aom_cdf_prob *cdf_ptr[], aom_cdf_prob *fc_cdf_ptr,
av1_average_cdf(cdf_ptr, fc_cdf_ptr, cdf_size, num_tiles);
void av1_average_tile_coef_cdfs(FRAME_CONTEXT *fc, FRAME_CONTEXT *ec_ctxs[],
aom_cdf_prob *cdf_ptr[], const int num_tiles) {
aom_cdf_prob *cdf_ptr[], int num_tiles) {
int i, cdf_size;
aom_cdf_prob *fc_cdf_ptr;
#if CONFIG_NEW_TOKENSET
AVERAGE_TILE_CDFS(coef_head_cdfs)
AVERAGE_TILE_CDFS(coef_tail_cdfs)
#else
AVERAGE_TILE_CDFS(coef_cdfs)
#endif
}
void av1_average_tile_mv_cdfs(FRAME_CONTEXT *fc, FRAME_CONTEXT *ec_ctxs[],
aom_cdf_prob *cdf_ptr[], const int num_tiles) {
aom_cdf_prob *cdf_ptr[], int num_tiles) {
int i, k, cdf_size;
aom_cdf_prob *fc_cdf_ptr;
#if CONFIG_REF_MV
int j;
for (j = 0; j < NMV_CONTEXTS; ++j) {
AVERAGE_TILE_CDFS(nmvc[j].joint_cdf)
@ -6293,19 +5644,10 @@ void av1_average_tile_mv_cdfs(FRAME_CONTEXT *fc, FRAME_CONTEXT *ec_ctxs[],
AVERAGE_TILE_CDFS(nmvc[j].comps[k].fp_cdf);
}
}
#else
AVERAGE_TILE_CDFS(nmvc.joint_cdf)
for (k = 0; k < 2; ++k) {
AVERAGE_TILE_CDFS(nmvc.comps[k].class_cdf)
AVERAGE_TILE_CDFS(nmvc.comps[k].class0_fp_cdf)
AVERAGE_TILE_CDFS(nmvc.comps[k].fp_cdf)
}
#endif
}
void av1_average_tile_intra_cdfs(FRAME_CONTEXT *fc, FRAME_CONTEXT *ec_ctxs[],
aom_cdf_prob *cdf_ptr[], const int num_tiles) {
aom_cdf_prob *cdf_ptr[], int num_tiles) {
int i, cdf_size;
aom_cdf_prob *fc_cdf_ptr;
@ -6335,28 +5677,22 @@ void av1_average_tile_intra_cdfs(FRAME_CONTEXT *fc, FRAME_CONTEXT *ec_ctxs[],
#if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
AVERAGE_TILE_CDFS(intra_filter_cdf)
#endif // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
#if CONFIG_FILTER_INTRA
#endif // CONFIG_FILTER_INTRA
}
void av1_average_tile_inter_cdfs(AV1_COMMON *cm, FRAME_CONTEXT *fc,
FRAME_CONTEXT *ec_ctxs[],
aom_cdf_prob *cdf_ptr[], const int num_tiles) {
aom_cdf_prob *cdf_ptr[], int num_tiles) {
int i, cdf_size;
aom_cdf_prob *fc_cdf_ptr;
// FIXME: comp_inter_cdf not defined
// FIXME: comp_inter_cdf not defined
// FIXME: comp_ref_cdf and comp_bwd_ref not defined
// FIXME: comp_ref_cdf and comp_bwd_ref not defined
// FIXME: single_ref_cdf not defined
// FIXME: single_ref_cdf not defined
#if CONFIG_REF_MV
// FIXME: cdfs not defined for newmv_mode, zeromv_mode, drl_mode, new2mv_mode
#else
AVERAGE_TILE_CDFS(inter_mode_cdf)
#endif
// FIXME: cdfs not defined for newmv_mode, zeromv_mode, drl_mode, new2mv_mode
// FIXME: cdfs not defined for motion_mode_prob, obmc_prob

74
third_party/aom/av1/common/entropy.h vendored
View file

@ -32,10 +32,6 @@ extern "C" {
#define QCTX_BINS (1 << QCTX_BIN_BITS)
#endif // CONFIG_Q_ADAPT_PROBS
#if CONFIG_SUBFRAME_PROB_UPDATE
#define COEF_PROBS_BUFS 16
#endif // CONFIG_SUBFRAME_PROB_UPDATE
// Coefficient token alphabet
#define ZERO_TOKEN 0 // 0 Extra Bits 0+0
#define ONE_TOKEN 1 // 1 Extra Bits 0+1
@ -49,18 +45,16 @@ extern "C" {
#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
#if CONFIG_NEW_TOKENSET
#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 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
#endif
#define ENTROPY_TOKENS 12
#define ENTROPY_NODES 11
@ -131,7 +125,7 @@ static INLINE int av1_get_cat6_extrabits_size(TX_SIZE tx_size,
// TODO(debargha): Does TX_64X64 require an additional extrabit?
if (tx_size > TX_32X32) tx_size = TX_32X32;
#endif
#if CONFIG_CB4X4
#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);
@ -175,9 +169,7 @@ static INLINE int av1_get_cat6_extrabits_size(TX_SIZE tx_size,
distinct bands). */
#define COEFF_CONTEXTS 6
#if CONFIG_EC_MULTISYMBOL
#define BLOCKZ_CONTEXTS 3
#endif
#define COEFF_CONTEXTS0 3 // for band 0
#define BAND_COEFF_CONTEXTS(band) \
((band) == 0 ? COEFF_CONTEXTS0 : COEFF_CONTEXTS)
@ -199,9 +191,6 @@ void av1_adapt_coef_probs(struct AV1Common *cm);
#if CONFIG_EC_ADAPT
void av1_adapt_coef_cdfs(struct AV1Common *cm, struct frame_contexts *pre_fc);
#endif
#if CONFIG_SUBFRAME_PROB_UPDATE
void av1_partial_adapt_probs(struct AV1Common *cm, int mi_row, int mi_col);
#endif // CONFIG_SUBFRAME_PROB_UPDATE
// This is the index in the scan order beyond which all coefficients for
// 8x8 transform and above are in the top band.
@ -250,7 +239,6 @@ typedef unsigned int av1_coeff_count_model[REF_TYPES][COEF_BANDS]
void av1_model_to_full_probs(const aom_prob *model, aom_prob *full);
#if CONFIG_EC_MULTISYMBOL
typedef aom_cdf_prob coeff_cdf_model[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[CDF_SIZE(ENTROPY_TOKENS)];
typedef aom_prob av1_blockz_probs_model[REF_TYPES][BLOCKZ_CONTEXTS];
@ -260,11 +248,9 @@ extern const aom_cdf_prob av1_pareto8_token_probs[COEFF_PROB_MODELS]
extern const aom_cdf_prob av1_pareto8_tail_probs[COEFF_PROB_MODELS]
[ENTROPY_TOKENS - 3];
struct frame_contexts;
#if CONFIG_NEW_TOKENSET
void av1_coef_head_cdfs(struct frame_contexts *fc);
#endif
void av1_coef_pareto_cdfs(struct frame_contexts *fc);
#endif // CONFIG_EC_MULTISYMBOL
typedef char ENTROPY_CONTEXT;
@ -277,7 +263,7 @@ 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_CB4X4
#if CONFIG_CHROMA_2X2
switch (tx_size) {
case TX_2X2:
above_ec = a[0] != 0;
@ -323,10 +309,36 @@ 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 | *(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;
#endif // CONFIG_TX64X64
#if CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
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 // CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
default: assert(0 && "Invalid transform size."); break;
}
return combine_entropy_contexts(above_ec, left_ec);
#endif
#endif // CONFIG_CHROMA_2X2
switch (tx_size) {
case TX_4X4:
@ -375,6 +387,24 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
break;
#endif // CONFIG_TX64X64
#if CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
case TX_4X16:
above_ec = a[0] != 0;
left_ec = !!*(const uint32_t *)l;
break;
case TX_16X4:
above_ec = !!*(const uint32_t *)a;
left_ec = l[0] != 0;
break;
case TX_8X32:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
case TX_32X8:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
#endif // CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
default: assert(0 && "Invalid transform size."); break;
}
return combine_entropy_contexts(above_ec, left_ec);

1945
third_party/aom/av1/common/entropymode.c vendored
View file

File diff suppressed because it is too large Load diff

61
third_party/aom/av1/common/entropymode.h vendored
View file

@ -73,7 +73,7 @@ extern "C" {
#endif // CONFIG_PALETTE
#if CONFIG_INTRABC
#define INTRABC_PROB 192
#define INTRABC_PROB_DEFAULT 192
#endif // CONFIG_INTRABC
struct AV1Common;
@ -99,18 +99,14 @@ typedef struct frame_contexts {
aom_prob partition_prob[PARTITION_CONTEXTS][PARTITION_TYPES - 1];
#endif
av1_coeff_probs_model coef_probs[TX_SIZES][PLANE_TYPES];
#if CONFIG_NEW_TOKENSET
coeff_cdf_model coef_tail_cdfs[TX_SIZES][PLANE_TYPES];
coeff_cdf_model coef_head_cdfs[TX_SIZES][PLANE_TYPES];
aom_prob blockzero_probs[TX_SIZES][PLANE_TYPES][REF_TYPES][BLOCKZ_CONTEXTS];
#elif CONFIG_EC_MULTISYMBOL
coeff_cdf_model coef_cdfs[TX_SIZES][PLANE_TYPES];
#endif // CONFIG_NEW_TOKENSET
aom_prob switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS - 1];
#if CONFIG_ADAPT_SCAN
// TODO(angiebird): try aom_prob
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
uint32_t non_zero_prob_2x2[TX_TYPES][4];
#endif
uint32_t non_zero_prob_4X4[TX_TYPES][16];
@ -125,7 +121,7 @@ typedef struct frame_contexts {
uint32_t non_zero_prob_32X16[TX_TYPES][512];
uint32_t non_zero_prob_16X32[TX_TYPES][512];
#if CONFIG_CB4X4
#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]);
@ -140,7 +136,7 @@ typedef struct frame_contexts {
DECLARE_ALIGNED(16, int16_t, scan_16X32[TX_TYPES][512]);
DECLARE_ALIGNED(16, int16_t, scan_32X16[TX_TYPES][512]);
#if CONFIG_CB4X4
#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]);
@ -155,7 +151,7 @@ typedef struct frame_contexts {
DECLARE_ALIGNED(16, int16_t, iscan_16X32[TX_TYPES][512]);
DECLARE_ALIGNED(16, int16_t, iscan_32X16[TX_TYPES][512]);
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
int16_t nb_2x2[TX_TYPES][(4 + 1) * 2];
#endif
int16_t nb_4X4[TX_TYPES][(16 + 1) * 2];
@ -185,12 +181,10 @@ typedef struct frame_contexts {
aom_prob coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS];
#endif
#if CONFIG_REF_MV
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];
#endif // CONFIG_REF_MV
aom_prob inter_mode_probs[INTER_MODE_CONTEXTS][INTER_MODES - 1];
#if CONFIG_EXT_INTER
@ -201,9 +195,11 @@ typedef struct frame_contexts {
[INTER_SINGLEREF_COMP_MODES - 1];
#endif // CONFIG_COMPOUND_SINGLEREF
aom_prob compound_type_prob[BLOCK_SIZES][COMPOUND_TYPES - 1];
#if CONFIG_INTERINTRA
aom_prob interintra_prob[BLOCK_SIZE_GROUPS];
aom_prob interintra_mode_prob[BLOCK_SIZE_GROUPS][INTERINTRA_MODES - 1];
aom_prob wedge_interintra_prob[BLOCK_SIZES];
#endif // CONFIG_INTERINTRA
#endif // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
aom_prob motion_mode_prob[BLOCK_SIZES][MOTION_MODES - 1];
@ -224,17 +220,17 @@ typedef struct frame_contexts {
aom_prob comp_inter_mode_prob[COMP_INTER_MODE_CONTEXTS];
#endif // CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
aom_prob tx_size_probs[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][MAX_TX_DEPTH];
#if CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
aom_prob quarter_tx_size_prob;
#endif // CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
#if CONFIG_VAR_TX
aom_prob txfm_partition_prob[TXFM_PARTITION_CONTEXTS];
#endif
aom_prob skip_probs[SKIP_CONTEXTS];
#if CONFIG_REF_MV
nmv_context nmvc[NMV_CONTEXTS];
#else
nmv_context nmvc;
#endif
#if CONFIG_INTRABC
nmv_context ndvc;
aom_prob intrabc_prob;
#endif
int initialized;
#if CONFIG_EXT_TX
@ -263,7 +259,6 @@ typedef struct frame_contexts {
#if CONFIG_LOOP_RESTORATION
aom_prob switchable_restore_prob[RESTORE_SWITCHABLE_TYPES - 1];
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob y_mode_cdf[BLOCK_SIZE_GROUPS][CDF_SIZE(INTRA_MODES)];
aom_cdf_prob uv_mode_cdf[INTRA_MODES][CDF_SIZE(INTRA_MODES)];
#if CONFIG_EXT_PARTITION_TYPES
@ -297,7 +292,6 @@ typedef struct frame_contexts {
#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
#endif // CONFIG_EC_MULTISYMBOL
#if CONFIG_DELTA_Q
aom_prob delta_q_prob[DELTA_Q_PROBS];
#if CONFIG_EXT_DELTA_Q
@ -309,6 +303,9 @@ typedef struct frame_contexts {
// 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_alpha_cdf[CDF_SIZE(CFL_ALPHABET_SIZE)];
#endif
} FRAME_CONTEXT;
typedef struct FRAME_COUNTS {
@ -328,9 +325,9 @@ typedef struct FRAME_COUNTS {
unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS];
#if CONFIG_ADAPT_SCAN
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
unsigned int non_zero_count_2x2[TX_TYPES][4];
#endif // CONFIG_CB4X4
#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];
@ -356,16 +353,12 @@ typedef struct FRAME_COUNTS {
unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS][2];
#endif // CONFIG_LV_MAP
#if CONFIG_EC_MULTISYMBOL
av1_blockz_count_model blockz_count[TX_SIZES][PLANE_TYPES];
#endif
#if CONFIG_REF_MV
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];
#endif
unsigned int inter_mode[INTER_MODE_CONTEXTS][INTER_MODES];
#if CONFIG_EXT_INTER
@ -374,9 +367,11 @@ typedef struct FRAME_COUNTS {
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][2];
#endif // CONFIG_INTERINTRA
unsigned int compound_interinter[BLOCK_SIZES][COMPOUND_TYPES];
#endif // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
@ -401,17 +396,17 @@ typedef struct FRAME_COUNTS {
// to use forward updates for the coeff probs, and as such it does not really
// belong into this structure.
unsigned int tx_size_totals[TX_SIZES];
unsigned int tx_size[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][TX_SIZES];
unsigned int tx_size[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][MAX_TX_DEPTH + 1];
#if CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
unsigned int quarter_tx_size[2];
#endif // CONFIG_EXT_TX && CONFIG_RECT_TX && CONFIG_RECT_TX_EXT
#if CONFIG_VAR_TX
unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
#endif
unsigned int skip[SKIP_CONTEXTS][2];
#if CONFIG_REF_MV
nmv_context_counts mv[NMV_CONTEXTS];
#else
nmv_context_counts mv;
#endif
#if CONFIG_INTRABC
unsigned int intrabc[2];
nmv_context_counts dv;
#endif
#if CONFIG_DELTA_Q
@ -451,11 +446,9 @@ typedef struct FRAME_COUNTS {
// Contexts used: Intra mode (Y plane) of 'above' and 'left' blocks.
extern const aom_prob av1_kf_y_mode_prob[INTRA_MODES][INTRA_MODES]
[INTRA_MODES - 1];
#if CONFIG_EC_MULTISYMBOL
// CDF version of 'av1_kf_y_mode_prob'.
extern const aom_cdf_prob av1_kf_y_mode_cdf[INTRA_MODES][INTRA_MODES]
[CDF_SIZE(INTRA_MODES)];
#endif
#if CONFIG_PALETTE
extern const aom_prob av1_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES]
@ -474,7 +467,6 @@ extern const aom_prob av1_default_palette_uv_color_index_prob
extern const aom_tree_index av1_intra_mode_tree[TREE_SIZE(INTRA_MODES)];
extern const aom_tree_index av1_inter_mode_tree[TREE_SIZE(INTER_MODES)];
#if CONFIG_EC_MULTISYMBOL
extern int av1_intra_mode_ind[INTRA_MODES];
extern int av1_intra_mode_inv[INTRA_MODES];
extern int av1_inter_mode_ind[INTER_MODES];
@ -485,11 +477,12 @@ extern int av1_ext_tx_intra_inv[EXT_TX_SETS_INTRA][TX_TYPES];
extern int av1_ext_tx_inter_ind[EXT_TX_SETS_INTER][TX_TYPES];
extern int av1_ext_tx_inter_inv[EXT_TX_SETS_INTER][TX_TYPES];
#endif
#endif
#if CONFIG_EXT_INTER
#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
@ -534,10 +527,10 @@ extern const aom_tree_index av1_motion_mode_tree[TREE_SIZE(MOTION_MODES)];
extern const aom_tree_index
av1_switchable_restore_tree[TREE_SIZE(RESTORE_SWITCHABLE_TYPES)];
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_EC_MULTISYMBOL
extern int av1_switchable_interp_ind[SWITCHABLE_FILTERS];
extern int av1_switchable_interp_inv[SWITCHABLE_FILTERS];
#if !CONFIG_EC_ADAPT
void av1_set_mode_cdfs(struct AV1Common *cm);
#endif
@ -545,7 +538,7 @@ 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);
#if CONFIG_EC_MULTISYMBOL && !CONFIG_EXT_TX
#if !CONFIG_EXT_TX
extern int av1_ext_tx_ind[TX_TYPES];
extern int av1_ext_tx_inv[TX_TYPES];
#endif

61
third_party/aom/av1/common/entropymv.c vendored
View file

@ -43,57 +43,47 @@ const aom_tree_index av1_mv_fp_tree[TREE_SIZE(MV_FP_SIZE)] = { -0, 2, -1,
static const nmv_context default_nmv_context = {
{ 32, 64, 96 }, // joints
#if CONFIG_EC_MULTISYMBOL
{ AOM_ICDF(4096), AOM_ICDF(11264), AOM_ICDF(19328), AOM_ICDF(32768),
0 }, // joint_cdf
#endif
{ {
// Vertical component
128, // sign
{ 224, 144, 192, 168, 192, 176, 192, 198, 198, 245 }, // class
#if CONFIG_EC_MULTISYMBOL
{ 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
#endif
{ 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
#if CONFIG_EC_MULTISYMBOL
{ 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
#endif
160, // class0_hp bit
128, // hp
160, // class0_hp bit
128, // hp
},
{
// Horizontal component
128, // sign
{ 216, 128, 176, 160, 176, 176, 192, 198, 198, 208 }, // class
#if CONFIG_EC_MULTISYMBOL
{ 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
#endif
{ 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
#if CONFIG_EC_MULTISYMBOL
{ 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
#endif
160, // class0_hp bit
128, // hp
160, // class0_hp bit
128, // hp
} },
};
@ -232,23 +222,16 @@ void av1_inc_mv(const MV *mv, nmv_context_counts *counts, const int usehp) {
void av1_adapt_mv_probs(AV1_COMMON *cm, int allow_hp) {
int i, j;
#if CONFIG_REF_MV
int idx;
for (idx = 0; idx < NMV_CONTEXTS; ++idx) {
nmv_context *fc = &cm->fc->nmvc[idx];
const nmv_context *pre_fc =
&cm->frame_contexts[cm->frame_context_idx].nmvc[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];
#else
nmv_context *fc = &cm->fc->nmvc;
const nmv_context *pre_fc = &cm->frame_contexts[cm->frame_context_idx].nmvc;
const nmv_context_counts *counts = &cm->counts.mv;
#endif // CONFIG_REF_MV
aom_tree_merge_probs(av1_mv_joint_tree, pre_fc->joints, counts->joints,
fc->joints);
aom_tree_merge_probs(av1_mv_joint_tree, pre_nmvc->joints, counts->joints,
nmvc->joints);
for (i = 0; i < 2; ++i) {
nmv_component *comp = &fc->comps[i];
const nmv_component *pre_comp = &pre_fc->comps[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);
@ -272,12 +255,10 @@ void av1_adapt_mv_probs(AV1_COMMON *cm, int allow_hp) {
comp->hp = av1_mode_mv_merge_probs(pre_comp->hp, c->hp);
}
}
#if CONFIG_REF_MV
}
#endif // CONFIG_REF_MV
}
#if CONFIG_EC_MULTISYMBOL && !CONFIG_EC_ADAPT
#if !CONFIG_EC_ADAPT
void av1_set_mv_cdfs(nmv_context *ctx) {
int i;
int j;
@ -297,15 +278,11 @@ void av1_set_mv_cdfs(nmv_context *ctx) {
#endif
void av1_init_mv_probs(AV1_COMMON *cm) {
#if CONFIG_REF_MV
int i;
for (i = 0; i < NMV_CONTEXTS; ++i) {
// NB: this sets CDFs too
cm->fc->nmvc[i] = default_nmv_context;
}
#else
cm->fc->nmvc = default_nmv_context;
#endif // CONFIG_REF_MV
#if CONFIG_INTRABC
cm->fc->ndvc = default_nmv_context;
#endif // CONFIG_INTRABC

8
third_party/aom/av1/common/entropymv.h vendored
View file

@ -84,26 +84,20 @@ extern const aom_tree_index av1_mv_fp_tree[];
typedef struct {
aom_prob sign;
aom_prob classes[MV_CLASSES - 1];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob class_cdf[CDF_SIZE(MV_CLASSES)];
#endif
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];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob class0_fp_cdf[CLASS0_SIZE][CDF_SIZE(MV_FP_SIZE)];
aom_cdf_prob fp_cdf[CDF_SIZE(MV_FP_SIZE)];
#endif
aom_prob class0_hp;
aom_prob hp;
} nmv_component;
typedef struct {
aom_prob joints[MV_JOINTS - 1];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob joint_cdf[CDF_SIZE(MV_JOINTS)];
#endif
nmv_component comps[2];
} nmv_context;
@ -138,7 +132,7 @@ void av1_inc_mv(const MV *mv, nmv_context_counts *mvctx, const int usehp);
extern const aom_tree_index
av1_global_motion_types_tree[TREE_SIZE(GLOBAL_TRANS_TYPES)];
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_EC_MULTISYMBOL
#if !CONFIG_EC_ADAPT
void av1_set_mv_cdfs(nmv_context *ctx);
#endif

66
third_party/aom/av1/common/enums.h vendored
View file

@ -51,7 +51,6 @@ extern "C" {
// Mask to extract MI offset within max MIB
#define MAX_MIB_MASK (MAX_MIB_SIZE - 1)
#define MAX_MIB_MASK_2 (MAX_MIB_SIZE * 2 - 1)
// Maximum number of tile rows and tile columns
#if CONFIG_EXT_TILE
@ -111,8 +110,13 @@ typedef enum ATTRIBUTE_PACKED {
BLOCK_128X64,
BLOCK_128X128,
#endif // CONFIG_EXT_PARTITION
BLOCK_SIZES,
BLOCK_INVALID = BLOCK_SIZES,
BLOCK_4X16,
BLOCK_16X4,
BLOCK_8X32,
BLOCK_32X8,
BLOCK_SIZES_ALL,
BLOCK_SIZES = BLOCK_4X16,
BLOCK_INVALID = 255,
BLOCK_LARGEST = (BLOCK_SIZES - 1)
} BLOCK_SIZE;
@ -145,7 +149,7 @@ typedef char PARTITION_CONTEXT;
// block transform size
typedef enum ATTRIBUTE_PACKED {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
TX_2X2, // 2x2 transform
#endif
TX_4X4, // 4x4 transform
@ -170,7 +174,12 @@ typedef enum ATTRIBUTE_PACKED {
TX_INVALID = 255 // Invalid transform size
} TX_SIZE;
#define MAX_TX_DEPTH (TX_SIZES - 1 - TX_4X4)
#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)
#define MAX_TX_SIZE_LOG2 (5 + CONFIG_TX64X64)
#define MAX_TX_SIZE (1 << MAX_TX_SIZE_LOG2)
@ -240,15 +249,15 @@ typedef enum {
} BOUNDARY_TYPE;
#if CONFIG_EXT_TX
#if CONFIG_CB4X4
#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_CB4X4
#endif // CONFIG_CHROMA_2X2
#define EXT_TX_SETS_INTER 4 // Sets of transform selections for INTER
#define EXT_TX_SETS_INTRA 3 // Sets of transform selections for INTRA
#else
#if CONFIG_CB4X4
#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
@ -274,7 +283,14 @@ typedef enum {
typedef enum { PLANE_TYPE_Y = 0, PLANE_TYPE_UV = 1, PLANE_TYPES } PLANE_TYPE;
#if CONFIG_CFL
// TODO(ltrudeau) this should change based on QP size
#define CB_ALPHABET_SIZE 4
#define CR_ALPHABET_SIZE 4
#define CFL_ALPHABET_SIZE (CB_ALPHABET_SIZE * CR_ALPHABET_SIZE)
#define CFL_MAGS_SIZE 7
typedef enum { CFL_PRED_U = 0, CFL_PRED_V = 1, CFL_PRED_PLANES } CFL_PRED_TYPE;
typedef enum { CFL_SIGN_NEG = 0, CFL_SIGN_POS = 1, CFL_SIGNS } CFL_SIGN_TYPE;
#endif
#if CONFIG_PALETTE
@ -314,8 +330,12 @@ typedef enum ATTRIBUTE_PACKED {
D63_PRED, // Directional 63 deg = round(arctan(2/1) * 180/pi)
#if CONFIG_ALT_INTRA
SMOOTH_PRED, // Combination of horizontal and vertical interpolation
#endif // CONFIG_ALT_INTRA
TM_PRED, // True-motion
#if CONFIG_SMOOTH_HV
SMOOTH_V_PRED, // Vertical interpolation
SMOOTH_H_PRED, // Horizontal interpolation
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
TM_PRED, // True-motion
NEARESTMV,
NEARMV,
ZEROMV,
@ -331,8 +351,6 @@ typedef enum ATTRIBUTE_PACKED {
#endif // CONFIG_COMPOUND_SINGLEREF
// Compound ref compound modes
NEAREST_NEARESTMV,
NEAREST_NEARMV,
NEAR_NEARESTMV,
NEAR_NEARMV,
NEAREST_NEWMV,
NEW_NEARESTMV,
@ -357,22 +375,20 @@ typedef enum {
MOTION_MODES
} MOTION_MODE;
// TODO(urvang): Consider adding II_SMOOTH_PRED if it's helpful.
#if CONFIG_EXT_INTER
#if CONFIG_INTERINTRA
typedef enum {
II_DC_PRED = 0,
II_V_PRED,
II_H_PRED,
II_D45_PRED,
II_D135_PRED,
II_D117_PRED,
II_D153_PRED,
II_D207_PRED,
II_D63_PRED,
#if CONFIG_ALT_INTRA
II_SMOOTH_PRED,
#else
II_TM_PRED,
#endif // CONFIG_ALT_INTRA
INTERINTRA_MODES
} INTERINTRA_MODE;
#endif
typedef enum {
COMPOUND_AVERAGE = 0,
@ -404,7 +420,7 @@ typedef enum {
#endif // CONFIG_FILTER_INTRA
#if CONFIG_EXT_INTRA
#define DIRECTIONAL_MODES (INTRA_MODES - 2)
#define DIRECTIONAL_MODES 8
#endif // CONFIG_EXT_INTRA
#define INTER_MODES (1 + NEWMV - NEARESTMV)
@ -419,7 +435,6 @@ typedef enum {
#define SKIP_CONTEXTS 3
#if CONFIG_REF_MV
#define NMV_CONTEXTS 3
#define NEWMV_MODE_CONTEXTS 7
@ -438,7 +453,6 @@ typedef enum {
#define SKIP_NEARESTMV_OFFSET 9
#define SKIP_NEARMV_OFFSET 10
#define SKIP_NEARESTMV_SUB8X8_OFFSET 11
#endif
#define INTER_MODE_CONTEXTS 7
#if CONFIG_DELTA_Q
@ -455,14 +469,12 @@ typedef enum {
/* Segment Feature Masks */
#define MAX_MV_REF_CANDIDATES 2
#if CONFIG_REF_MV
#define MAX_REF_MV_STACK_SIZE 16
#if CONFIG_EXT_PARTITION
#define REF_CAT_LEVEL 640
#else
#define REF_CAT_LEVEL 255
#endif // CONFIG_EXT_PARTITION
#endif // CONFIG_REF_MV
#define INTRA_INTER_CONTEXTS 4
#define COMP_INTER_CONTEXTS 5
@ -508,11 +520,7 @@ typedef uint8_t TXFM_CONTEXT;
#define SINGLE_REFS (FWD_REFS + BWD_REFS)
#define COMP_REFS (FWD_REFS * BWD_REFS)
#if CONFIG_REF_MV
#define MODE_CTX_REF_FRAMES (TOTAL_REFS_PER_FRAME + COMP_REFS)
#else
#define MODE_CTX_REF_FRAMES TOTAL_REFS_PER_FRAME
#endif
#if CONFIG_SUPERTX
#define PARTITION_SUPERTX_CONTEXTS 2

64
third_party/aom/av1/common/filter.c vendored
View file

@ -48,9 +48,19 @@ DECLARE_ALIGNED(16, static const int16_t,
};
#endif // USE_TEMPORALFILTER_12TAP
#if CONFIG_DUAL_FILTER
#if USE_EXTRA_FILTER
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 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 }
#else
// intfilt 0.575
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -5, 126, 8, -3, 1, 0 },
{ -1, 3, -10, 123, 18, -6, 2, -1 }, { -1, 4, -14, 118, 27, -9, 3, 0 },
@ -60,10 +70,21 @@ DECLARE_ALIGNED(256, static const InterpKernel,
{ -1, 5, -17, 58, 97, -19, 6, -1 }, { -1, 4, -14, 48, 105, -18, 5, -1 },
{ -1, 4, -12, 37, 112, -16, 5, -1 }, { 0, 3, -9, 27, 118, -14, 4, -1 },
{ -1, 2, -6, 18, 123, -10, 3, -1 }, { 0, 1, -3, 8, 126, -5, 1, 0 },
#endif
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_regular_uv[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 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 }
#else
// intfilt 0.575
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -5, 126, 8, -3, 1, 0 },
{ -1, 3, -10, 123, 18, -6, 2, -1 }, { -1, 4, -14, 118, 27, -9, 3, 0 },
@ -73,6 +94,7 @@ DECLARE_ALIGNED(256, static const InterpKernel,
{ -1, 5, -17, 58, 97, -19, 6, -1 }, { -1, 4, -14, 48, 105, -18, 5, -1 },
{ -1, 4, -12, 37, 112, -16, 5, -1 }, { 0, 3, -9, 27, 118, -14, 4, -1 },
{ -1, 2, -6, 18, 123, -10, 3, -1 }, { 0, 1, -3, 8, 126, -5, 1, 0 },
#endif
};
#if USE_12TAP_FILTER
@ -162,6 +184,16 @@ DECLARE_ALIGNED(256, static const InterpKernel,
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8smooth[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 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
// freqmultiplier = 0.8
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, -5, 13, 102, 24, -7, 1, 0 },
{ 0, -4, 8, 100, 31, -8, 1, 0 }, { 0, -3, 4, 97, 37, -8, 1, 0 },
@ -171,10 +203,21 @@ DECLARE_ALIGNED(256, static const InterpKernel,
{ 0, 0, -9, 59, 84, -5, -1, 0 }, { 0, 1, -9, 51, 90, -3, -2, 0 },
{ 0, 1, -9, 44, 94, 0, -2, 0 }, { 0, 1, -8, 37, 97, 4, -3, 0 },
{ 0, 1, -8, 31, 100, 8, -4, 0 }, { 0, 1, -7, 24, 102, 13, -5, 0 },
#endif
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_smooth_uv[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 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
// freqmultiplier = 0.8
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, -5, 13, 102, 24, -7, 1, 0 },
{ 0, -4, 8, 100, 31, -8, 1, 0 }, { 0, -3, 4, 97, 37, -8, 1, 0 },
@ -184,8 +227,9 @@ DECLARE_ALIGNED(256, static const InterpKernel,
{ 0, 0, -9, 59, 84, -5, -1, 0 }, { 0, 1, -9, 51, 90, -3, -2, 0 },
{ 0, 1, -9, 44, 94, 0, -2, 0 }, { 0, 1, -8, 37, 97, 4, -3, 0 },
{ 0, 1, -8, 31, 100, 8, -4, 0 }, { 0, 1, -7, 24, 102, 13, -5, 0 },
#endif
};
#else // CONFIG_DUAL_FILTER
#else // USE_EXTRA_FILTER
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8[SUBPEL_SHIFTS]) = {
@ -255,7 +299,7 @@ DECLARE_ALIGNED(256, static const InterpKernel,
{ 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 1, 38, 64, 32, -1, -3 }
#endif
};
#endif // CONFIG_DUAL_FILTER
#endif // USE_EXTRA_FILTER
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
@ -268,7 +312,7 @@ const InterpKernel *av1_intra_filter_kernels[INTRA_FILTERS] = {
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_DUAL_FILTER
#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,
@ -309,7 +353,7 @@ static const InterpFilterParams
{ (const int16_t *)bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
BILINEAR }
};
#endif // CONFIG_DUAL_FILTER
#endif // USE_EXTRA_FILTER
#if USE_TEMPORALFILTER_12TAP
static const InterpFilterParams av1_interp_temporalfilter_12tap = {
@ -340,11 +384,17 @@ const int16_t *av1_get_interp_filter_kernel(const InterpFilter interp_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 {
@ -356,5 +406,9 @@ InterpFilter av1_get_plane_interp_filter(InterpFilter interp_filter,
default: return interp_filter;
}
}
#else // USE_EXTRA_FILTER
(void)plane;
return interp_filter;
#endif // USE_EXTRA_FILTER
}
#endif

9
third_party/aom/av1/common/filter.h vendored
View file

@ -25,22 +25,23 @@ extern "C" {
#define MAX_FILTER_TAP 12
#define USE_12TAP_FILTER 0
#define USE_EXTRA_FILTER 0
typedef enum {
EIGHTTAP_REGULAR,
EIGHTTAP_SMOOTH,
MULTITAP_SHARP,
#if CONFIG_DUAL_FILTER
#if USE_EXTRA_FILTER
EIGHTTAP_SMOOTH2,
#endif // CONFIG_DUAL_FILTER
#endif // USE_EXTRA_FILTER
BILINEAR,
#if CONFIG_DUAL_FILTER
#if USE_EXTRA_FILTER
EIGHTTAP_SHARP,
FILTER_REGULAR_UV,
FILTER_SMOOTH_UV,
FILTER_SHARP_UV,
FILTER_SMOOTH2_UV,
#endif // CONFIG_DUAL_FILTER
#endif // USE_EXTRA_FILTER
INTERP_FILTERS_ALL,
SWITCHABLE_FILTERS = BILINEAR,
SWITCHABLE = SWITCHABLE_FILTERS + 1, /* the last switchable one */

1222
third_party/aom/av1/common/idct.c vendored
View file

File diff suppressed because it is too large Load diff

37
third_party/aom/av1/common/idct.h vendored
View file

@ -46,9 +46,7 @@ typedef struct {
} transform_2d;
#if CONFIG_HIGHBITDEPTH
typedef void (*highbd_transform_1d)(const tran_low_t *, tran_low_t *,
const int8_t *cos_bit,
const int8_t *stage_range, int bd);
typedef void (*highbd_transform_1d)(const tran_low_t *, tran_low_t *, int bd);
typedef struct {
highbd_transform_1d cols, rows; // vertical and horizontal
@ -63,12 +61,6 @@ void av1_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
void av1_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void av1_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type, int lossless);
void av1_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type);
void av1_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type);
void av1_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
void av1_inverse_transform_block(const MACROBLOCKD *xd,
@ -92,6 +84,33 @@ void av1_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
void av1_highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_DPCM_INTRA
void av1_dpcm_inv_txfm_add_4_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, uint8_t *dest);
void av1_dpcm_inv_txfm_add_8_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, uint8_t *dest);
void av1_dpcm_inv_txfm_add_16_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, uint8_t *dest);
void av1_dpcm_inv_txfm_add_32_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, uint8_t *dest);
typedef void (*dpcm_inv_txfm_add_func)(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, uint8_t *dest);
dpcm_inv_txfm_add_func av1_get_dpcm_inv_txfm_add_func(int tx_length);
#if CONFIG_HIGHBITDEPTH
void av1_hbd_dpcm_inv_txfm_add_4_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, int bd, uint16_t *dest);
void av1_hbd_dpcm_inv_txfm_add_8_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, int bd, uint16_t *dest);
void av1_hbd_dpcm_inv_txfm_add_16_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, int bd, uint16_t *dest);
void av1_hbd_dpcm_inv_txfm_add_32_c(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, int bd, uint16_t *dest);
typedef void (*hbd_dpcm_inv_txfm_add_func)(const tran_low_t *input, int stride,
TX_TYPE_1D tx_type, int bd,
uint16_t *dest);
hbd_dpcm_inv_txfm_add_func av1_get_hbd_dpcm_inv_txfm_add_func(int tx_length);
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_DPCM_INTRA
#ifdef __cplusplus
} // extern "C"
#endif

4
third_party/aom/av1/common/mv.h vendored
View file

@ -58,6 +58,8 @@ typedef struct mv32 {
// 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 \
@ -269,14 +271,12 @@ static INLINE TransformationType get_gmtype(const WarpedMotionParams *gm) {
}
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_REF_MV
typedef struct candidate_mv {
int_mv this_mv;
int_mv comp_mv;
uint8_t pred_diff[2];
int weight;
} CANDIDATE_MV;
#endif
static INLINE int is_zero_mv(const MV *mv) {
return *((const uint32_t *)mv) == 0;

78
third_party/aom/av1/common/mvref_common.c vendored
View file

@ -14,8 +14,6 @@
#include "av1/common/warped_motion.h"
#endif // CONFIG_WARPED_MOTION
#if CONFIG_REF_MV
static uint8_t add_ref_mv_candidate(
const MODE_INFO *const candidate_mi, const MB_MODE_INFO *const candidate,
const MV_REFERENCE_FRAME rf[2], uint8_t *refmv_count,
@ -556,7 +554,6 @@ static void setup_ref_mv_list(const AV1_COMMON *cm, const MACROBLOCKD *xd,
}
}
}
#endif
// This function searches the neighbourhood of a given MB/SB
// to try and find candidate reference vectors.
@ -568,9 +565,6 @@ static void find_mv_refs_idx(const AV1_COMMON *cm, const MACROBLOCKD *xd,
int_mv zeromv) {
const int *ref_sign_bias = cm->ref_frame_sign_bias;
int i, refmv_count = 0;
#if !CONFIG_REF_MV
const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
#endif
int different_ref_found = 0;
int context_counter = 0;
#if CONFIG_MV_COMPRESS
@ -591,12 +585,14 @@ static void find_mv_refs_idx(const AV1_COMMON *cm, const MACROBLOCKD *xd,
cm->use_prev_frame_mvs
? cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col
: NULL;
#endif
#if CONFIG_INTRABC
assert(IMPLIES(ref_frame == INTRA_FRAME, cm->use_prev_frame_mvs == 0));
#endif
const TileInfo *const tile = &xd->tile;
const BLOCK_SIZE bsize = mi->mbmi.sb_type;
const int bw = block_size_wide[AOMMAX(bsize, BLOCK_8X8)];
const int bh = block_size_high[AOMMAX(bsize, BLOCK_8X8)];
#if CONFIG_REF_MV
POSITION mv_ref_search[MVREF_NEIGHBOURS];
const int num_8x8_blocks_wide = num_8x8_blocks_wide_lookup[bsize];
const int num_8x8_blocks_high = num_8x8_blocks_high_lookup[bsize];
@ -639,7 +635,6 @@ static void find_mv_refs_idx(const AV1_COMMON *cm, const MACROBLOCKD *xd,
mv_ref_search[i].col *= 2;
}
#endif // CONFIG_CB4X4
#endif // CONFIG_REF_MV
// The nearest 2 blocks are treated differently
// if the size < 8x8 we get the mv from the bmi substructure,
@ -673,12 +668,10 @@ static void find_mv_refs_idx(const AV1_COMMON *cm, const MACROBLOCKD *xd,
!xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]
? NULL
: &xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]->mbmi;
#if CONFIG_REF_MV
if (candidate == NULL) continue;
if ((mi_row % MAX_MIB_SIZE) + mv_ref->row >= MAX_MIB_SIZE ||
(mi_col % MAX_MIB_SIZE) + mv_ref->col >= MAX_MIB_SIZE)
continue;
#endif
different_ref_found = 1;
if (candidate->ref_frame[0] == ref_frame)
@ -727,12 +720,10 @@ static void find_mv_refs_idx(const AV1_COMMON *cm, const MACROBLOCKD *xd,
!xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]
? NULL
: &xd->mi[mv_ref->col + mv_ref->row * xd->mi_stride]->mbmi;
#if CONFIG_REF_MV
if (candidate == NULL) continue;
if ((mi_row % MAX_MIB_SIZE) + mv_ref->row >= MAX_MIB_SIZE ||
(mi_col % MAX_MIB_SIZE) + mv_ref->col >= MAX_MIB_SIZE)
continue;
#endif
// If the candidate is INTRA we don't want to consider its mv.
IF_DIFF_REF_FRAME_ADD_MV(candidate, ref_frame, ref_sign_bias,
@ -780,50 +771,18 @@ void av1_update_mv_context(const AV1_COMMON *cm, const MACROBLOCKD *xd,
int_mv *mv_ref_list, int block, int mi_row,
int mi_col, int16_t *mode_context) {
int i, refmv_count = 0;
#if !CONFIG_REF_MV
const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
#endif
int context_counter = 0;
const int bw = block_size_wide[mi->mbmi.sb_type];
const int bh = block_size_high[mi->mbmi.sb_type];
const TileInfo *const tile = &xd->tile;
#if CONFIG_REF_MV
POSITION mv_ref_search[MVREF_NEIGHBOURS];
POSITION mv_ref_search[2];
const int num_8x8_blocks_wide = mi_size_wide[mi->mbmi.sb_type];
const int num_8x8_blocks_high = mi_size_high[mi->mbmi.sb_type];
mv_ref_search[0].row = num_8x8_blocks_high - 1;
mv_ref_search[0].col = -1;
mv_ref_search[1].row = -1;
mv_ref_search[1].col = num_8x8_blocks_wide - 1;
mv_ref_search[2].row = -1;
mv_ref_search[2].col = (num_8x8_blocks_wide - 1) >> 1;
mv_ref_search[3].row = (num_8x8_blocks_high - 1) >> 1;
mv_ref_search[3].col = -1;
mv_ref_search[4].row = -1;
mv_ref_search[4].col = -1;
#if CONFIG_EXT_PARTITION_TYPES
if (num_8x8_blocks_wide == num_8x8_blocks_high) {
mv_ref_search[5].row = -1;
mv_ref_search[5].col = 0;
mv_ref_search[6].row = 0;
mv_ref_search[6].col = -1;
} else {
mv_ref_search[5].row = -1;
mv_ref_search[5].col = num_8x8_blocks_wide;
mv_ref_search[6].row = num_8x8_blocks_high;
mv_ref_search[6].col = -1;
}
#else
mv_ref_search[5].row = -1;
mv_ref_search[5].col = num_8x8_blocks_wide;
mv_ref_search[6].row = num_8x8_blocks_high;
mv_ref_search[6].col = -1;
#endif // CONFIG_EXT_PARTITION_TYPES
mv_ref_search[7].row = -1;
mv_ref_search[7].col = -3;
mv_ref_search[8].row = num_8x8_blocks_high - 1;
mv_ref_search[8].col = -3;
#endif
// Blank the reference vector list
memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
@ -859,12 +818,10 @@ Done:
void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
#if CONFIG_REF_MV
uint8_t *ref_mv_count, CANDIDATE_MV *ref_mv_stack,
#if CONFIG_EXT_INTER
int16_t *compound_mode_context,
#endif // CONFIG_EXT_INTER
#endif
int_mv *mv_ref_list, int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
int16_t *mode_context) {
@ -872,23 +829,17 @@ void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
#if CONFIG_GLOBAL_MOTION
BLOCK_SIZE bsize = mi->mbmi.sb_type;
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_REF_MV
int idx, all_zero = 1;
#if CONFIG_GLOBAL_MOTION
MV_REFERENCE_FRAME rf[2];
#endif // CONFIG_GLOBAL_MOTION
#endif // CONFIG_REF_MV
#if CONFIG_EXT_INTER
av1_update_mv_context(cm, xd, mi, ref_frame, mv_ref_list, -1, mi_row, mi_col,
#if CONFIG_REF_MV
compound_mode_context);
#else
mode_context);
#endif // CONFIG_REF_MV
#endif // CONFIG_EXT_INTER
#if CONFIG_GLOBAL_MOTION
#if CONFIG_REF_MV
av1_set_ref_frame(rf, ref_frame);
zeromv[0].as_int = gm_get_motion_vector(&cm->global_motion[rf[0]],
cm->allow_high_precision_mv, bsize,
@ -900,24 +851,14 @@ void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
bsize, mi_col, mi_row, 0)
.as_int
: 0;
#else
zeromv[0].as_int = gm_get_motion_vector(&cm->global_motion[ref_frame],
cm->allow_high_precision_mv, bsize,
mi_col, mi_row, 0)
.as_int;
zeromv[1].as_int = 0;
#endif // CONFIG_REF_MV
#else
zeromv[0].as_int = zeromv[1].as_int = 0;
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_REF_MV
if (ref_frame <= ALTREF_FRAME)
#endif // CONFIG_REF_MV
find_mv_refs_idx(cm, xd, mi, ref_frame, mv_ref_list, -1, mi_row, mi_col,
sync, data, mode_context, zeromv[0]);
#if CONFIG_REF_MV
setup_ref_mv_list(cm, xd, ref_frame, ref_mv_count, ref_mv_stack, mv_ref_list,
-1, mi_row, mi_col, mode_context);
/* Note: If global motion is enabled, then we want to set the ALL_ZERO flag
@ -947,7 +888,6 @@ void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
}
if (all_zero) mode_context[ref_frame] |= (1 << ALL_ZERO_FLAG_OFFSET);
#endif
}
void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
@ -963,10 +903,8 @@ void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
int block, int ref, int mi_row, int mi_col,
#if CONFIG_REF_MV
CANDIDATE_MV *ref_mv_stack,
uint8_t *ref_mv_count,
#endif
#if CONFIG_EXT_INTER
int_mv *mv_list,
#endif // CONFIG_EXT_INTER
@ -978,13 +916,11 @@ void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
b_mode_info *bmi = mi->bmi;
int n;
int_mv zeromv;
#if CONFIG_REF_MV
CANDIDATE_MV tmp_mv;
uint8_t idx;
uint8_t above_count = 0, left_count = 0;
MV_REFERENCE_FRAME rf[2] = { mi->mbmi.ref_frame[ref], NONE_FRAME };
*ref_mv_count = 0;
#endif
assert(MAX_MV_REF_CANDIDATES == 2);
@ -999,7 +935,6 @@ void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
find_mv_refs_idx(cm, xd, mi, mi->mbmi.ref_frame[ref], mv_list, block, mi_row,
mi_col, NULL, NULL, NULL, zeromv);
#if CONFIG_REF_MV
scan_blk_mbmi(cm, xd, mi_row, mi_col, block, rf, -1, 0, ref_mv_stack,
ref_mv_count);
above_count = *ref_mv_count;
@ -1020,7 +955,6 @@ void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
for (idx = 0; idx < AOMMIN(MAX_MV_REF_CANDIDATES, *ref_mv_count); ++idx)
mv_list[idx].as_int = ref_mv_stack[idx].this_mv.as_int;
#endif
near_mv->as_int = 0;
switch (block) {

190
third_party/aom/av1/common/mvref_common.h vendored
View file

@ -18,11 +18,7 @@
extern "C" {
#endif
#if CONFIG_REF_MV
#define MVREF_NEIGHBOURS 9
#else
#define MVREF_NEIGHBOURS 8
#endif
typedef struct position {
int row;
@ -45,7 +41,7 @@ typedef enum {
// 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[MB_MODE_COUNT] = {
static const int mode_2_counter[] = {
9, // DC_PRED
9, // V_PRED
9, // H_PRED
@ -56,7 +52,11 @@ static const int mode_2_counter[MB_MODE_COUNT] = {
9, // D207_PRED
9, // D63_PRED
#if CONFIG_ALT_INTRA
9, // SMOOTH_PRED
9, // SMOOTH_PRED
#if CONFIG_SMOOTH_HV
9, // SMOOTH_V_PRED
9, // SMOOTH_H_PRED
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
9, // TM_PRED
0, // NEARESTMV
@ -72,8 +72,6 @@ static const int mode_2_counter[MB_MODE_COUNT] = {
1, // SR_NEW_NEWMV
#endif // CONFIG_COMPOUND_SINGLEREF
0, // NEAREST_NEARESTMV
0, // NEAREST_NEARMV
0, // NEAR_NEARESTMV
0, // NEAR_NEARMV
1, // NEAREST_NEWMV
1, // NEW_NEARESTMV
@ -109,158 +107,6 @@ static const int counter_to_context[19] = {
BOTH_INTRA // 18
};
#if !CONFIG_REF_MV
static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = {
// 4X4
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 4X8
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X4
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X8
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X16
{ { 0, -1 },
{ -1, 0 },
{ 1, -1 },
{ -1, -1 },
{ 0, -2 },
{ -2, 0 },
{ -2, -1 },
{ -1, -2 } },
// 16X8
{ { -1, 0 },
{ 0, -1 },
{ -1, 1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -1, -2 },
{ -2, -1 } },
// 16X16
{ { -1, 0 },
{ 0, -1 },
{ -1, 1 },
{ 1, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 16X32
{ { 0, -1 },
{ -1, 0 },
{ 2, -1 },
{ -1, -1 },
{ -1, 1 },
{ 0, -3 },
{ -3, 0 },
{ -3, -3 } },
// 32X16
{ { -1, 0 },
{ 0, -1 },
{ -1, 2 },
{ -1, -1 },
{ 1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 32X32
{ { -1, 1 },
{ 1, -1 },
{ -1, 2 },
{ 2, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 32X64
{ { 0, -1 },
{ -1, 0 },
{ 4, -1 },
{ -1, 2 },
{ -1, -1 },
{ 0, -3 },
{ -3, 0 },
{ 2, -1 } },
// 64X32
{ { -1, 0 },
{ 0, -1 },
{ -1, 4 },
{ 2, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -1, 2 } },
// 64X64
{ { -1, 3 },
{ 3, -1 },
{ -1, 4 },
{ 4, -1 },
{ -1, -1 },
{ -1, 0 },
{ 0, -1 },
{ -1, 6 } },
#if CONFIG_EXT_PARTITION
// TODO(debargha/jingning) Making them twice the 32x64, .. ones above
// 64x128
{ { 0, -2 },
{ -2, 0 },
{ 8, -2 },
{ -2, 4 },
{ -2, -2 },
{ 0, -6 },
{ -6, 0 },
{ 4, -2 } },
// 128x64
{ { -2, 0 },
{ 0, -2 },
{ -2, 8 },
{ 4, -2 },
{ -2, -2 },
{ -6, 0 },
{ 0, -6 },
{ -2, 4 } },
// 128x128
{ { -2, 6 },
{ 6, -2 },
{ -2, 8 },
{ 8, -2 },
{ -2, -2 },
{ -2, 0 },
{ 0, -2 },
{ -2, 12 } },
#endif // CONFIG_EXT_PARTITION
};
#endif
static const int idx_n_column_to_subblock[4][2] = {
{ 1, 2 }, { 1, 3 }, { 3, 2 }, { 3, 3 }
};
@ -283,20 +129,11 @@ static INLINE void clamp_mv_ref(MV *mv, int bw, int bh, const MACROBLOCKD *xd) {
// 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) {
#if CONFIG_REF_MV
(void)search_col;
(void)block_idx;
return candidate->mbmi.mv[which_mv];
#else
return block_idx >= 0 && candidate->mbmi.sb_type < BLOCK_8X8
? candidate
->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]]
.as_mv[which_mv]
: candidate->mbmi.mv[which_mv];
#endif
}
#if CONFIG_REF_MV
static INLINE int_mv get_sub_block_pred_mv(const MODE_INFO *candidate,
int which_mv, int search_col,
int block_idx) {
@ -304,7 +141,6 @@ static INLINE int_mv get_sub_block_pred_mv(const MODE_INFO *candidate,
(void)block_idx;
return candidate->mbmi.mv[which_mv];
}
#endif
// Performs mv sign inversion if indicated by the reference frame combination.
static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
@ -384,7 +220,6 @@ static INLINE void lower_mv_precision(MV *mv, int allow_hp) {
}
}
#if CONFIG_REF_MV
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 &&
@ -434,8 +269,12 @@ static INLINE void av1_set_ref_frame(MV_REFERENCE_FRAME *rf,
} else {
rf[0] = ref_frame_type;
rf[1] = NONE_FRAME;
assert(ref_frame_type > INTRA_FRAME &&
ref_frame_type < TOTAL_REFS_PER_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);
}
}
@ -477,17 +316,14 @@ static INLINE uint8_t av1_drl_ctx(const CANDIDATE_MV *ref_mv_stack,
return 0;
}
#endif
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,
#if CONFIG_REF_MV
uint8_t *ref_mv_count, CANDIDATE_MV *ref_mv_stack,
#if CONFIG_EXT_INTER
int16_t *compound_mode_context,
#endif // CONFIG_EXT_INTER
#endif
int_mv *mv_ref_list, int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
int16_t *mode_context);
@ -500,10 +336,8 @@ void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
int block, int ref, int mi_row, int mi_col,
#if CONFIG_REF_MV
CANDIDATE_MV *ref_mv_stack,
uint8_t *ref_mv_count,
#endif
#if CONFIG_EXT_INTER
int_mv *mv_list,
#endif // CONFIG_EXT_INTER

3
third_party/aom/av1/common/odintrin.h vendored
View file

@ -14,6 +14,9 @@
#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>

112
third_party/aom/av1/common/onyxc_int.h vendored
View file

@ -100,9 +100,7 @@ typedef enum {
typedef struct {
int_mv mv[2];
#if CONFIG_REF_MV
int_mv pred_mv[2];
#endif
MV_REFERENCE_FRAME ref_frame[2];
} MV_REF;
@ -163,12 +161,6 @@ typedef struct AV1Common {
int last_width;
int last_height;
#if CONFIG_FRAME_SUPERRES
// The numerator of the superres scale, the denominator is fixed
uint8_t superres_scale_numerator;
int superres_width, superres_height;
#endif // CONFIG_FRAME_SUPERRES
// TODO(jkoleszar): this implies chroma ss right now, but could vary per
// plane. Revisit as part of the future change to YV12_BUFFER_CONFIG to
// support additional planes.
@ -216,9 +208,17 @@ typedef struct AV1Common {
int allow_high_precision_mv;
#if CONFIG_PALETTE
#if CONFIG_PALETTE || CONFIG_INTRABC
int allow_screen_content_tools;
#endif // CONFIG_PALETTE
#endif // CONFIG_PALETTE || CONFIG_INTRABC
#if CONFIG_EXT_INTER
#if CONFIG_INTERINTRA
int allow_interintra_compound;
#endif // CONFIG_INTERINTRA
#if CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
int allow_masked_compound;
#endif // CONFIG_WEDGE || CONFIG_COMPOUND_SEGMENT
#endif // CONFIG_EXT_INTER
// Flag signaling which frame contexts should be reset to default values.
RESET_FRAME_CONTEXT_MODE reset_frame_context;
@ -242,15 +242,15 @@ typedef struct AV1Common {
#if CONFIG_AOM_QM
// Global quant matrix tables
qm_val_t *giqmatrix[NUM_QM_LEVELS][2][2][TX_SIZES];
qm_val_t *gqmatrix[NUM_QM_LEVELS][2][2][TX_SIZES];
qm_val_t *giqmatrix[NUM_QM_LEVELS][2][2][TX_SIZES_ALL];
qm_val_t *gqmatrix[NUM_QM_LEVELS][2][2][TX_SIZES_ALL];
// Local quant matrix tables for each frame
qm_val_t *y_iqmatrix[MAX_SEGMENTS][2][TX_SIZES];
qm_val_t *uv_iqmatrix[MAX_SEGMENTS][2][TX_SIZES];
qm_val_t *y_iqmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
qm_val_t *uv_iqmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
// Encoder
qm_val_t *y_qmatrix[MAX_SEGMENTS][2][TX_SIZES];
qm_val_t *uv_qmatrix[MAX_SEGMENTS][2][TX_SIZES];
qm_val_t *y_qmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
qm_val_t *uv_qmatrix[MAX_SEGMENTS][2][TX_SIZES_ALL];
int using_qmatrix;
int min_qmlevel;
@ -299,6 +299,10 @@ typedef struct AV1Common {
InterpFilter interp_filter;
loop_filter_info_n lf_info;
#if CONFIG_FRAME_SUPERRES
// The numerator of the superres scale; the denominator is fixed.
uint8_t superres_scale_numerator;
#endif // CONFIG_FRAME_SUPERRES
#if CONFIG_LOOP_RESTORATION
RestorationInfo rst_info[MAX_MB_PLANE];
RestorationInternal rst_internal;
@ -331,21 +335,10 @@ typedef struct AV1Common {
FRAME_CONTEXT *fc; /* this frame entropy */
FRAME_CONTEXT *frame_contexts; // FRAME_CONTEXTS
FRAME_CONTEXT *pre_fc; // Context referenced in this frame
unsigned int frame_context_idx; /* Context to use/update */
FRAME_COUNTS counts;
#if CONFIG_SUBFRAME_PROB_UPDATE
// The initial probabilities for a frame, before any subframe backward update,
// and after forward update.
av1_coeff_probs_model starting_coef_probs[TX_SIZES][PLANE_TYPES];
// Number of subframe backward updates already done
uint8_t coef_probs_update_idx;
// Signal if the backward update is subframe or end-of-frame
uint8_t partial_prob_update;
// Frame level flag to turn on/off subframe backward update
uint8_t do_subframe_update;
#endif // CONFIG_SUBFRAME_PROB_UPDATE
unsigned int current_video_frame;
BITSTREAM_PROFILE profile;
@ -393,7 +386,8 @@ typedef struct AV1Common {
ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
#if CONFIG_VAR_TX
TXFM_CONTEXT *above_txfm_context;
TXFM_CONTEXT left_txfm_context[MAX_MIB_SIZE];
TXFM_CONTEXT *top_txfm_context[MAX_MB_PLANE];
TXFM_CONTEXT left_txfm_context[MAX_MB_PLANE][2 * MAX_MIB_SIZE];
#endif
int above_context_alloc_cols;
@ -576,17 +570,20 @@ static INLINE void av1_init_macroblockd(AV1_COMMON *cm, MACROBLOCKD *xd,
static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
int i;
int row_offset = mi_row;
int col_offset = mi_col;
for (i = 0; i < MAX_MB_PLANE; ++i) {
struct macroblockd_plane *const pd = &xd->plane[i];
#if CONFIG_CHROMA_SUB8X8
if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
// Offset the buffer pointer
if (pd->subsampling_y && (mi_row & 0x01)) mi_row -= 1;
if (pd->subsampling_x && (mi_col & 0x01)) mi_col -= 1;
if (pd->subsampling_y && (mi_row & 0x01)) row_offset = mi_row - 1;
if (pd->subsampling_x && (mi_col & 0x01)) col_offset = mi_col - 1;
}
#endif
int above_idx = mi_col * 2;
int left_idx = (mi_row * 2) & MAX_MIB_MASK_2;
int above_idx = col_offset << (MI_SIZE_LOG2 - tx_size_wide_log2[0]);
int left_idx = (row_offset & MAX_MIB_MASK)
<< (MI_SIZE_LOG2 - tx_size_high_log2[0]);
pd->above_context = &xd->above_context[i][above_idx >> pd->subsampling_x];
pd->left_context = &xd->left_context[i][left_idx >> pd->subsampling_y];
}
@ -668,14 +665,12 @@ static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile,
xd->n8_h = bh;
xd->n8_w = bw;
#if CONFIG_REF_MV
xd->is_sec_rect = 0;
if (xd->n8_w < xd->n8_h)
if (mi_col & (xd->n8_h - 1)) xd->is_sec_rect = 1;
if (xd->n8_w > xd->n8_h)
if (mi_row & (xd->n8_w - 1)) xd->is_sec_rect = 1;
#endif // CONFIG_REF_MV
}
static INLINE const aom_prob *get_y_mode_probs(const AV1_COMMON *cm,
@ -688,7 +683,6 @@ static INLINE const aom_prob *get_y_mode_probs(const AV1_COMMON *cm,
return cm->kf_y_prob[above][left];
}
#if CONFIG_EC_MULTISYMBOL
static INLINE aom_cdf_prob *get_y_mode_cdf(FRAME_CONTEXT *tile_ctx,
const MODE_INFO *mi,
const MODE_INFO *above_mi,
@ -698,7 +692,6 @@ static INLINE aom_cdf_prob *get_y_mode_cdf(FRAME_CONTEXT *tile_ctx,
const PREDICTION_MODE left = av1_left_block_mode(mi, left_mi, block);
return tile_ctx->kf_y_cdf[above][left];
}
#endif
static INLINE void update_partition_context(MACROBLOCKD *xd, int mi_row,
int mi_col, BLOCK_SIZE subsize,
@ -873,8 +866,8 @@ static INLINE void av1_zero_above_context(AV1_COMMON *const cm,
const int width = mi_col_end - mi_col_start;
const int aligned_width = ALIGN_POWER_OF_TWO(width, cm->mib_size_log2);
const int offset_y = 2 * mi_col_start;
const int width_y = 2 * aligned_width;
const int offset_y = mi_col_start << (MI_SIZE_LOG2 - tx_size_wide_log2[0]);
const int width_y = aligned_width << (MI_SIZE_LOG2 - tx_size_wide_log2[0]);
const int offset_uv = offset_y >> cm->subsampling_x;
const int width_uv = width_y >> cm->subsampling_x;
@ -885,7 +878,8 @@ static INLINE void av1_zero_above_context(AV1_COMMON *const cm,
av1_zero_array(cm->above_seg_context + mi_col_start, aligned_width);
#if CONFIG_VAR_TX
av1_zero_array(cm->above_txfm_context + mi_col_start, aligned_width);
av1_zero_array(cm->above_txfm_context + (mi_col_start << TX_UNIT_WIDE_LOG2),
aligned_width << TX_UNIT_WIDE_LOG2);
#endif // CONFIG_VAR_TX
}
@ -899,7 +893,7 @@ static INLINE void av1_zero_left_context(MACROBLOCKD *const xd) {
#if CONFIG_VAR_TX
static INLINE TX_SIZE get_min_tx_size(TX_SIZE tx_size) {
if (tx_size >= TX_SIZES_ALL) assert(0);
assert(tx_size < TX_SIZES_ALL);
return txsize_sqr_map[tx_size];
}
@ -918,16 +912,16 @@ static INLINE void set_txfm_ctxs(TX_SIZE tx_size, int n8_w, int n8_h, int skip,
bh = n8_h * MI_SIZE;
}
set_txfm_ctx(xd->above_txfm_context, bw, n8_w);
set_txfm_ctx(xd->left_txfm_context, bh, n8_h);
set_txfm_ctx(xd->above_txfm_context, bw, n8_w << TX_UNIT_WIDE_LOG2);
set_txfm_ctx(xd->left_txfm_context, bh, n8_h << TX_UNIT_HIGH_LOG2);
}
static INLINE void txfm_partition_update(TXFM_CONTEXT *above_ctx,
TXFM_CONTEXT *left_ctx,
TX_SIZE tx_size, TX_SIZE txb_size) {
BLOCK_SIZE bsize = txsize_to_bsize[txb_size];
int bh = mi_size_high[bsize];
int bw = mi_size_wide[bsize];
int bh = mi_size_high[bsize] << TX_UNIT_HIGH_LOG2;
int bw = mi_size_wide[bsize] << TX_UNIT_WIDE_LOG2;
uint8_t txw = tx_size_wide[tx_size];
uint8_t txh = tx_size_high[tx_size];
int i;
@ -935,6 +929,21 @@ static INLINE void txfm_partition_update(TXFM_CONTEXT *above_ctx,
for (i = 0; i < bw; ++i) above_ctx[i] = txw;
}
static INLINE TX_SIZE get_sqr_tx_size(int tx_dim) {
TX_SIZE tx_size;
switch (tx_dim) {
#if CONFIG_EXT_PARTITION
case 128:
#endif
case 64:
case 32: tx_size = TX_32X32; break;
case 16: tx_size = TX_16X16; break;
case 8: tx_size = TX_8X8; break;
default: tx_size = TX_4X4;
}
return tx_size;
}
static INLINE int txfm_partition_context(TXFM_CONTEXT *above_ctx,
TXFM_CONTEXT *left_ctx,
BLOCK_SIZE bsize, TX_SIZE tx_size) {
@ -942,22 +951,13 @@ static INLINE int txfm_partition_context(TXFM_CONTEXT *above_ctx,
const uint8_t txh = tx_size_high[tx_size];
const int above = *above_ctx < txw;
const int left = *left_ctx < txh;
TX_SIZE max_tx_size = max_txsize_lookup[bsize];
int category = TXFM_PARTITION_CONTEXTS - 1;
// dummy return, not used by others.
if (tx_size <= TX_4X4) return 0;
switch (AOMMAX(block_size_wide[bsize], block_size_high[bsize])) {
#if CONFIG_EXT_PARTITION
case 128:
#endif
case 64:
case 32: max_tx_size = TX_32X32; break;
case 16: max_tx_size = TX_16X16; break;
case 8: max_tx_size = TX_8X8; break;
default: assert(0);
}
TX_SIZE max_tx_size =
get_sqr_tx_size(AOMMAX(block_size_wide[bsize], block_size_high[bsize]));
if (max_tx_size >= TX_8X8) {
category = (tx_size != max_tx_size && max_tx_size > TX_8X8) +

58
third_party/aom/av1/common/pred_common.c vendored
View file

@ -153,6 +153,48 @@ int av1_get_pred_context_intra_interp(const MACROBLOCKD *xd) {
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_PALETTE && CONFIG_PALETTE_DELTA_ENCODING
int av1_get_palette_cache(const MODE_INFO *above_mi, const MODE_INFO *left_mi,
int plane, uint16_t *cache) {
int above_n = 0, left_n = 0;
if (above_mi)
above_n = above_mi->mbmi.palette_mode_info.palette_size[plane != 0];
if (left_mi)
left_n = left_mi->mbmi.palette_mode_info.palette_size[plane != 0];
if (above_n == 0 && left_n == 0) return 0;
int above_idx = plane * PALETTE_MAX_SIZE;
int left_idx = plane * PALETTE_MAX_SIZE;
int n = 0;
const uint16_t *above_colors =
above_mi->mbmi.palette_mode_info.palette_colors;
const uint16_t *left_colors = left_mi->mbmi.palette_mode_info.palette_colors;
// Merge the sorted lists of base colors from above and left to get
// combined sorted color cache.
while (above_n > 0 && left_n > 0) {
uint16_t v_above = above_colors[above_idx];
uint16_t v_left = left_colors[left_idx];
if (v_left < v_above) {
if (n == 0 || v_left != cache[n - 1]) cache[n++] = v_left;
++left_idx, --left_n;
} else {
if (n == 0 || v_above != cache[n - 1]) cache[n++] = v_above;
++above_idx, --above_n;
if (v_left == v_above) ++left_idx, --left_n;
}
}
while (above_n-- > 0) {
uint16_t val = above_colors[above_idx++];
if (n == 0 || val != cache[n - 1]) cache[n++] = val;
}
while (left_n-- > 0) {
uint16_t val = left_colors[left_idx++];
if (n == 0 || val != cache[n - 1]) cache[n++] = val;
}
assert(n <= 2 * PALETTE_MAX_SIZE);
return n;
}
#endif // CONFIG_PALETTE && CONFIG_PALETTE_DELTA_ENCODING
// The mode info data structure has a one element border above and to the
// left of the entries corresponding to real macroblocks.
// The prediction flags in these dummy entries are initialized to 0.
@ -278,7 +320,7 @@ int av1_get_reference_mode_context(const AV1_COMMON *cm,
//
// NOTE(zoeliu): The probability of ref_frame[0] is either
// GOLDEN_FRAME or LAST3_FRAME.
#if CONFIG_LOWDELAY_COMPOUND
#if CONFIG_ONE_SIDED_COMPOUND
int av1_get_pred_context_comp_ref_p(UNUSED const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
#else
@ -295,7 +337,7 @@ int av1_get_pred_context_comp_ref_p(const AV1_COMMON *cm,
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
#if CONFIG_LOWDELAY_COMPOUND // No change to bitstream
#if CONFIG_ONE_SIDED_COMPOUND // No change to bitstream
// Code seems to assume that signbias of cm->comp_bwd_ref[0] is always 1
const int bwd_ref_sign_idx = 1;
#else
@ -388,7 +430,7 @@ int av1_get_pred_context_comp_ref_p(const AV1_COMMON *cm,
//
// NOTE(zoeliu): The probability of ref_frame[0] is LAST_FRAME,
// conditioning on it is either LAST_FRAME or LAST2_FRAME.
#if CONFIG_LOWDELAY_COMPOUND
#if CONFIG_ONE_SIDED_COMPOUND
int av1_get_pred_context_comp_ref_p1(UNUSED const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
#else
@ -405,7 +447,7 @@ int av1_get_pred_context_comp_ref_p1(const AV1_COMMON *cm,
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
#if CONFIG_LOWDELAY_COMPOUND // No change to bitstream
#if CONFIG_ONE_SIDED_COMPOUND // No change to bitstream
// Code seems to assume that signbias of cm->comp_bwd_ref[0] is always 1
const int bwd_ref_sign_idx = 1;
#else
@ -499,7 +541,7 @@ int av1_get_pred_context_comp_ref_p1(const AV1_COMMON *cm,
//
// NOTE(zoeliu): The probability of ref_frame[0] is GOLDEN_FRAME,
// conditioning on it is either GOLDEN or LAST3.
#if CONFIG_LOWDELAY_COMPOUND
#if CONFIG_ONE_SIDED_COMPOUND
int av1_get_pred_context_comp_ref_p2(UNUSED const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
#else
@ -516,7 +558,7 @@ int av1_get_pred_context_comp_ref_p2(const AV1_COMMON *cm,
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
#if CONFIG_LOWDELAY_COMPOUND // No change to bitstream
#if CONFIG_ONE_SIDED_COMPOUND // No change to bitstream
// Code seems to assume that signbias of cm->comp_bwd_ref[0] is always 1
const int bwd_ref_sign_idx = 1;
#else
@ -604,7 +646,7 @@ int av1_get_pred_context_comp_ref_p2(const AV1_COMMON *cm,
}
// Returns a context number for the given MB prediction signal
#if CONFIG_LOWDELAY_COMPOUND
#if CONFIG_ONE_SIDED_COMPOUND
int av1_get_pred_context_comp_bwdref_p(UNUSED const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
#else
@ -621,7 +663,7 @@ int av1_get_pred_context_comp_bwdref_p(const AV1_COMMON *cm,
// The mode info data structure has a one element border above and to the
// left of the entries corresponding to real macroblocks.
// The prediction flags in these dummy entries are initialized to 0.
#if CONFIG_LOWDELAY_COMPOUND // No change to bitstream
#if CONFIG_ONE_SIDED_COMPOUND // No change to bitstream
// Code seems to assume that signbias of cm->comp_bwd_ref[0] is always 1
const int bwd_ref_sign_idx = 1;
#else

21
third_party/aom/av1/common/pred_common.h vendored
View file

@ -79,6 +79,15 @@ int av1_get_pred_context_intra_interp(const MACROBLOCKD *xd);
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_PALETTE && 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 MODE_INFO *above_mi, const MODE_INFO *left_mi,
int plane, uint16_t *cache);
#endif // CONFIG_PALETTE && CONFIG_PALETTE_DELTA_ENCODING
int av1_get_intra_inter_context(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_intra_inter_prob(const AV1_COMMON *cm,
@ -197,13 +206,7 @@ static INLINE int get_tx_size_context(const MACROBLOCKD *xd) {
if (!has_left) left_ctx = above_ctx;
if (!has_above) above_ctx = left_ctx;
#if CONFIG_CB4X4
// TODO(jingning): Temporary setup. Will rework this after the cb4x4
// framework is up running.
return (above_ctx + left_ctx) > max_tx_size + 1;
#else
return (above_ctx + left_ctx) > max_tx_size;
#endif
return (above_ctx + left_ctx) > max_tx_size + TX_SIZE_LUMA_MIN;
}
#if CONFIG_VAR_TX
@ -222,7 +225,9 @@ static void update_tx_counts(AV1_COMMON *cm, MACROBLOCKD *xd,
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
if (tx_size == plane_tx_size) {
++xd->counts->tx_size[max_tx_size - TX_8X8][ctx][tx_size];
int depth;
depth = tx_size_to_depth(tx_size);
++xd->counts->tx_size[max_tx_size - TX_SIZE_CTX_MIN][ctx][depth];
mbmi->tx_size = tx_size;
} else {
int bsl = b_width_log2_lookup[bsize];

24843
third_party/aom/av1/common/quant_common.c vendored
View file

File diff suppressed because it is too large Load diff

14
third_party/aom/av1/common/quant_common.h vendored
View file

@ -29,11 +29,11 @@ extern "C" {
// Total number of QM sets stored
#define QM_LEVEL_BITS 4
#define NUM_QM_LEVELS (1 << QM_LEVEL_BITS)
/* Offset into the list of QMs. Actual number of levels used is
(NUM_QM_LEVELS-AOM_QM_OFFSET)
Lower value of AOM_QM_OFFSET implies more heavily weighted matrices.*/
#define DEFAULT_QM_FIRST (NUM_QM_LEVELS / 2)
#define DEFAULT_QM_LAST (NUM_QM_LEVELS - 1)
/* Range of QMS is between first and last value, with offset applied to inter
* blocks*/
#define DEFAULT_QM_FIRST 5
#define DEFAULT_QM_LAST 9
#define DEFAULT_QM_INTER_OFFSET 0
#endif
struct AV1Common;
@ -54,9 +54,9 @@ static INLINE int aom_get_qmlevel(int qindex, int first, int last) {
}
void aom_qm_init(struct AV1Common *cm);
qm_val_t *aom_iqmatrix(struct AV1Common *cm, int qindex, int comp,
int log2sizem2, int is_intra);
TX_SIZE tx_size, int is_intra);
qm_val_t *aom_qmatrix(struct AV1Common *cm, int qindex, int comp,
int log2sizem2, int is_intra);
TX_SIZE tx_size, int is_intra);
#endif
#if CONFIG_NEW_QUANT

502
third_party/aom/av1/common/reconinter.c vendored
View file

@ -371,12 +371,12 @@ void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
const uint8_t *src1, int src1_stride,
BLOCK_SIZE sb_type, int h, int w) {
switch (mask_type) {
case DIFFWTD_42:
diffwtd_mask(mask, 0, 42, src0, src0_stride, src1, src1_stride, sb_type,
case DIFFWTD_38:
diffwtd_mask(mask, 0, 38, src0, src0_stride, src1, src1_stride, sb_type,
h, w);
break;
case DIFFWTD_42_INV:
diffwtd_mask(mask, 1, 42, src0, src0_stride, src1, src1_stride, sb_type,
case DIFFWTD_38_INV:
diffwtd_mask(mask, 1, 38, src0, src0_stride, src1, src1_stride, sb_type,
h, w);
break;
default: assert(0);
@ -407,12 +407,12 @@ void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
const uint8_t *src1, int src1_stride,
BLOCK_SIZE sb_type, int h, int w, int bd) {
switch (mask_type) {
case DIFFWTD_42:
case DIFFWTD_38:
diffwtd_mask_highbd(mask, 0, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
bd);
break;
case DIFFWTD_42_INV:
case DIFFWTD_38_INV:
diffwtd_mask_highbd(mask, 1, 42, CONVERT_TO_SHORTPTR(src0), src0_stride,
CONVERT_TO_SHORTPTR(src1), src1_stride, sb_type, h, w,
bd);
@ -459,7 +459,7 @@ static void shift_copy(const uint8_t *src, uint8_t *dst, int shift, int width) {
}
}
#else
static const double smoother_param[NSMOOTHERS] = { 2.83 };
static const double smoother_param[NSMOOTHERS] = { 3.0 };
#endif // MASK_MASTER_SIZE == 64
static void init_wedge_master_masks() {
@ -468,21 +468,22 @@ static void init_wedge_master_masks() {
const int h = MASK_MASTER_SIZE;
const int stride = MASK_MASTER_STRIDE;
for (s = 0; s < NSMOOTHERS; s++) {
// Note: index [0] stores the masters, and [1] its complement.
#if MASK_MASTER_SIZE == 64
// Generate prototype by shifting the masters
int shift = h / 4;
for (i = 0; i < h; i += 2) {
shift_copy(wedge_master_oblique_even[s],
&wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride], shift,
&wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride], shift,
MASK_MASTER_SIZE);
shift--;
shift_copy(wedge_master_oblique_odd[s],
&wedge_mask_obl[s][1][WEDGE_OBLIQUE63][(i + 1) * stride],
&wedge_mask_obl[s][0][WEDGE_OBLIQUE63][(i + 1) * stride],
shift, MASK_MASTER_SIZE);
memcpy(&wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride],
memcpy(&wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride],
wedge_master_vertical[s],
MASK_MASTER_SIZE * sizeof(wedge_master_vertical[s][0]));
memcpy(&wedge_mask_obl[s][1][WEDGE_VERTICAL][(i + 1) * stride],
memcpy(&wedge_mask_obl[s][0][WEDGE_VERTICAL][(i + 1) * stride],
wedge_master_vertical[s],
MASK_MASTER_SIZE * sizeof(wedge_master_vertical[s][0]));
}
@ -495,29 +496,29 @@ static void init_wedge_master_masks() {
int y = (2 * i + 1 - h);
double d = (a[0] * x + a[1] * y) / asqrt;
const int msk = (int)rint((1.0 + tanh(d / smoother_param[s])) * 32);
wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride + j] = msk;
wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j] = msk;
const int mskx = (int)rint((1.0 + tanh(x / smoother_param[s])) * 32);
wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j] = mskx;
wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j] = mskx;
}
}
#endif // MASK_MASTER_SIZE == 64
for (i = 0; i < h; ++i) {
for (j = 0; j < w; ++j) {
const int msk = wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride + j];
wedge_mask_obl[s][1][WEDGE_OBLIQUE27][j * stride + i] = msk;
wedge_mask_obl[s][1][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
wedge_mask_obl[s][1][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
(1 << WEDGE_WEIGHT_BITS) - msk;
wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j] =
wedge_mask_obl[s][0][WEDGE_OBLIQUE27][j * stride + i] =
(1 << WEDGE_WEIGHT_BITS) - msk;
const int msk = wedge_mask_obl[s][0][WEDGE_OBLIQUE63][i * stride + j];
wedge_mask_obl[s][0][WEDGE_OBLIQUE27][j * stride + i] = msk;
wedge_mask_obl[s][0][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
wedge_mask_obl[s][0][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
(1 << WEDGE_WEIGHT_BITS) - msk;
wedge_mask_obl[s][1][WEDGE_OBLIQUE63][i * stride + j] =
wedge_mask_obl[s][1][WEDGE_OBLIQUE27][j * stride + i] =
(1 << WEDGE_WEIGHT_BITS) - msk;
wedge_mask_obl[s][1][WEDGE_OBLIQUE117][i * stride + w - 1 - j] =
wedge_mask_obl[s][1][WEDGE_OBLIQUE153][(w - 1 - j) * stride + i] =
msk;
const int mskx = wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j];
wedge_mask_obl[s][1][WEDGE_HORIZONTAL][j * stride + i] = mskx;
wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j] =
wedge_mask_obl[s][0][WEDGE_HORIZONTAL][j * stride + i] =
const int mskx = wedge_mask_obl[s][0][WEDGE_VERTICAL][i * stride + j];
wedge_mask_obl[s][0][WEDGE_HORIZONTAL][j * stride + i] = mskx;
wedge_mask_obl[s][1][WEDGE_VERTICAL][i * stride + j] =
wedge_mask_obl[s][1][WEDGE_HORIZONTAL][j * stride + i] =
(1 << WEDGE_WEIGHT_BITS) - mskx;
}
}
@ -539,12 +540,23 @@ static void init_wedge_signs() {
int i, w;
if (wbits == 0) continue;
for (w = 0; w < wtypes; ++w) {
// Get the mask master, i.e. index [0]
const uint8_t *mask = get_wedge_mask_inplace(w, 0, sb_type);
int sum = 0;
for (i = 0; i < bw; ++i) sum += mask[i];
for (i = 0; i < bh; ++i) sum += mask[i * MASK_MASTER_STRIDE];
sum = (sum + (bw + bh) / 2) / (bw + bh);
wedge_params.signflip[w] = (sum < 32);
int avg = 0;
for (i = 0; i < bw; ++i) avg += mask[i];
for (i = 1; i < bh; ++i) avg += mask[i * MASK_MASTER_STRIDE];
avg = (avg + (bw + bh - 1) / 2) / (bw + bh - 1);
// Default sign of this wedge is 1 if the average < 32, 0 otherwise.
// If default sign is 1:
// If sign requested is 0, we need to flip the sign and return
// the complement i.e. index [1] instead. If sign requested is 1
// we need to flip the sign and return index [0] instead.
// If default sign is 0:
// If sign requested is 0, we need to return index [0] the master
// if sign requested is 1, we need to return the complement index [1]
// instead.
wedge_params.signflip[w] = (avg < 32);
// printf("%d[%d] = %d\n", sb_type, w, wedge_params.signflip[w]);
}
}
}
@ -880,7 +892,7 @@ typedef struct SubpelParams {
int subpel_y;
} SubpelParams;
void build_inter_predictors(MACROBLOCKD *xd, int plane,
void build_inter_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd, int plane,
#if CONFIG_MOTION_VAR
int mi_col_offset, int mi_row_offset,
#endif // CONFIG_MOTION_VAR
@ -893,28 +905,17 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
struct macroblockd_plane *const pd = &xd->plane[plane];
#if CONFIG_MOTION_VAR
const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
#if !CONFIG_CB4X4 || CONFIG_SUB8X8_MC
const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
#endif // !CONFIG_CB4X4 || CONFIG_SUB8X8_MC
#else
const MODE_INFO *mi = xd->mi[0];
#endif // CONFIG_MOTION_VAR
const int is_compound = has_second_ref(&mi->mbmi);
int is_compound = has_second_ref(&mi->mbmi);
int ref;
#if CONFIG_INTRABC
const int is_intrabc = is_intrabc_block(&mi->mbmi);
struct scale_factors sf_identity;
#if CONFIG_HIGHBITDEPTH
av1_setup_scale_factors_for_frame(
&sf_identity, 64, 64, 64, 64,
xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
#else
av1_setup_scale_factors_for_frame(&sf_identity, 64, 64, 64, 64);
#endif // CONFIG_HIGHBITDEPTH
assert(IMPLIES(is_intrabc, !is_compound));
#endif // CONFIG_INTRABC
#if CONFIG_GLOBAL_MOTION
int is_global[2];
int is_global[2] = { 0, 0 };
for (ref = 0; ref < 1 + is_compound; ++ref) {
WarpedMotionParams *const wm = &xd->global_motion[mi->mbmi.ref_frame[ref]];
is_global[ref] = is_global_mv_block(mi, block, wm->wmtype);
@ -923,47 +924,87 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
#if CONFIG_CB4X4
(void)block;
(void)cm;
#endif
#if CONFIG_SUB8X8_MC
#if CONFIG_MOTION_VAR
if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
#else
if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
#endif // CONFIG_MOTION_VAR
// block size in log2
const int b4_wl = b_width_log2_lookup[mi->mbmi.sb_type];
const int b4_hl = b_height_log2_lookup[mi->mbmi.sb_type];
const int b8_sl = b_width_log2_lookup[BLOCK_8X8];
#if CONFIG_MOTION_VAR && (CONFIG_CHROMA_SUB8X8 || !CONFIG_CB4X4)
const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
#endif // CONFIG_MOTION_VAR && (CONFIG_CHROMA_SUB8X8 || !CONFIG_CB4X4)
#if CONFIG_CHROMA_SUB8X8
const BLOCK_SIZE bsize = mi->mbmi.sb_type;
const int ss_x = pd->subsampling_x;
const int ss_y = pd->subsampling_y;
int sub8x8_inter = bsize < BLOCK_8X8 && (ss_x || ss_y);
const int row_start = (block_size_high[bsize] == 4) && ss_y ? -1 : 0;
const int col_start = (block_size_wide[bsize] == 4) && ss_x ? -1 : 0;
#if CONFIG_MOTION_VAR
if (!build_for_obmc && sub8x8_inter) {
#else
if (sub8x8_inter) {
#endif // CONFIG_MOTION_VAR
for (int row = row_start; row <= 0 && sub8x8_inter; ++row)
for (int col = col_start; col <= 0; ++col)
if (!is_inter_block(&xd->mi[row * xd->mi_stride + col]->mbmi))
sub8x8_inter = 0;
}
#if CONFIG_MOTION_VAR
if (!build_for_obmc && sub8x8_inter) {
#else
if (sub8x8_inter) {
#endif // CONFIG_MOTION_VAR
// block size
const int b4_w = 1 << b4_wl;
const int b4_h = 1 << b4_hl;
const int b8_s = 1 << b8_sl;
const int b4_w = block_size_wide[bsize] >> ss_x;
const int b4_h = block_size_high[bsize] >> ss_y;
const BLOCK_SIZE plane_bsize = scale_chroma_bsize(bsize, ss_x, ss_y);
const int b8_w = block_size_wide[plane_bsize] >> ss_x;
const int b8_h = block_size_high[plane_bsize] >> ss_y;
int idx, idy;
const int x_base = x;
const int y_base = y;
// processing unit size
const int x_step = w >> (b8_sl - b4_wl);
const int y_step = h >> (b8_sl - b4_hl);
const struct buf_2d orig_pred_buf[2] = { pd->pre[0], pd->pre[1] };
int row = row_start;
for (idy = 0; idy < b8_h; idy += b4_h) {
int col = col_start;
for (idx = 0; idx < b8_w; idx += b4_w) {
MB_MODE_INFO *this_mbmi = &xd->mi[row * xd->mi_stride + col]->mbmi;
is_compound = has_second_ref(this_mbmi);
for (idy = 0; idy < b8_s; idy += b4_h) {
for (idx = 0; idx < b8_s; idx += b4_w) {
const int chr_idx = (idy * 2) + idx;
for (ref = 0; ref < 1 + is_compound; ++ref) {
struct buf_2d *const dst_buf = &pd->dst;
const RefBuffer *ref_buf =
&cm->frame_refs[this_mbmi->ref_frame[ref] - LAST_FRAME];
const int c_offset = (mi_x + MI_SIZE * col_start) >> ss_x;
const int r_offset = (mi_y + MI_SIZE * row_start) >> ss_y;
pd->pre[ref].buf0 =
(plane == 1) ? ref_buf->buf->u_buffer : ref_buf->buf->v_buffer;
pd->pre[ref].buf =
pd->pre[ref].buf0 + scaled_buffer_offset(c_offset, r_offset,
ref_buf->buf->uv_stride,
&ref_buf->sf);
pd->pre[ref].width = ref_buf->buf->uv_crop_width;
pd->pre[ref].height = ref_buf->buf->uv_crop_height;
pd->pre[ref].stride = ref_buf->buf->uv_stride;
#if CONFIG_INTRABC
const struct scale_factors *const sf =
is_intrabc ? &sf_identity : &xd->block_refs[ref]->sf;
is_intrabc ? &xd->sf_identity : &xd->block_refs[ref]->sf;
struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
#else
const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
struct buf_2d *const pre_buf = &pd->pre[ref];
#endif // CONFIG_INTRABC
uint8_t *dst = dst_buf->buf;
const MV mv = mi->bmi[chr_idx].as_mv[ref].as_mv;
const MV mv = this_mbmi->mv[ref].as_mv;
const MV mv_q4 = clamp_mv_to_umv_border_sb(
xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
uint8_t *pre;
@ -977,12 +1018,13 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
warp_types.global_warp_allowed = is_global[ref];
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_WARPED_MOTION
warp_types.local_warp_allowed = mi->mbmi.motion_mode == WARPED_CAUSAL;
warp_types.local_warp_allowed =
this_mbmi->motion_mode == WARPED_CAUSAL;
#endif // CONFIG_WARPED_MOTION
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
x = x_base + idx * x_step;
y = y_base + idy * y_step;
x = x_base + idx;
y = y_base + idy;
dst += dst_buf->stride * y + x;
@ -1022,7 +1064,7 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
#endif // CONFIG_EXT_INTER
av1_make_inter_predictor(
pre, pre_buf->stride, dst, dst_buf->stride, subpel_x, subpel_y,
sf, x_step, y_step, &conv_params, mi->mbmi.interp_filter,
sf, b4_w, b4_h, &conv_params, this_mbmi->interp_filter,
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
&warp_types, (mi_x >> pd->subsampling_x) + x,
(mi_y >> pd->subsampling_y) + y, plane, ref,
@ -1032,11 +1074,17 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
#endif
xs, ys, xd);
}
++col;
}
++row;
}
for (ref = 0; ref < 2; ++ref) pd->pre[ref] = orig_pred_buf[ref];
return;
}
#endif
#else
(void)cm;
#endif // CONFIG_CHROMA_SUB8X8
{
struct buf_2d *const dst_buf = &pd->dst;
@ -1052,7 +1100,7 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
for (ref = 0; ref < 1 + is_compound; ++ref) {
#if CONFIG_INTRABC
const struct scale_factors *const sf =
is_intrabc ? &sf_identity : &xd->block_refs[ref]->sf;
is_intrabc ? &xd->sf_identity : &xd->block_refs[ref]->sf;
struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
#else
const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
@ -1110,7 +1158,7 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
for (ref = 0; ref < 1 + is_compound; ++ref) {
#if CONFIG_INTRABC
const struct scale_factors *const sf =
is_intrabc ? &sf_identity : &xd->block_refs[ref]->sf;
is_intrabc ? &xd->sf_identity : &xd->block_refs[ref]->sf;
struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
#else
const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
@ -1159,19 +1207,28 @@ void build_inter_predictors(MACROBLOCKD *xd, int plane,
}
#if CONFIG_CONVOLVE_ROUND
// TODO(angiebird): This part needs optimization
// TODO(angiebird): This part needs optimization
if (conv_params.do_post_rounding) {
#if CONFIG_HIGHBITDEPTH
if (!(xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH))
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
av1_highbd_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride,
w, h, FILTER_BITS * 2 + is_compound -
conv_params.round_0 -
conv_params.round_1,
xd->bd);
else
#endif // CONFIG_HIGHBITDEPTH
av1_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride, w, h,
FILTER_BITS * 2 + is_compound -
conv_params.round_0 - conv_params.round_1);
av1_convolve_rounding(tmp_dst, MAX_SB_SIZE, dst, dst_buf->stride, w, h,
FILTER_BITS * 2 + is_compound -
conv_params.round_0 - conv_params.round_1);
}
#endif // CONFIG_CONVOLVE_ROUND
}
}
void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
int ic, int mi_row, int mi_col) {
void av1_build_inter_predictor_sub8x8(const AV1_COMMON *cm, MACROBLOCKD *xd,
int plane, int i, int ir, int ic,
int mi_row, int mi_col) {
struct macroblockd_plane *const pd = &xd->plane[plane];
MODE_INFO *const mi = xd->mi[0];
const BLOCK_SIZE plane_bsize = get_plane_block_size(mi->mbmi.sb_type, pd);
@ -1180,6 +1237,8 @@ void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
uint8_t *const dst = &pd->dst.buf[(ir * pd->dst.stride + ic) << 2];
int ref;
const int is_compound = has_second_ref(&mi->mbmi);
(void)cm;
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
WarpTypesAllowed warp_types;
const int p_col = ((mi_col * MI_SIZE) >> pd->subsampling_x) + 4 * ic;
@ -1229,7 +1288,8 @@ void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
}
}
static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
static void build_inter_predictors_for_planes(const AV1_COMMON *cm,
MACROBLOCKD *xd, BLOCK_SIZE bsize,
int mi_row, int mi_col,
int plane_from, int plane_to) {
int plane;
@ -1265,7 +1325,7 @@ static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
assert(pw * num_4x4_w == bw && ph * num_4x4_h == bh);
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x)
build_inter_predictors(xd, plane,
build_inter_predictors(cm, xd, plane,
#if CONFIG_MOTION_VAR
0, 0,
#endif // CONFIG_MOTION_VAR
@ -1275,7 +1335,7 @@ static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
mi_x, mi_y);
} else {
build_inter_predictors(xd, plane,
build_inter_predictors(cm, xd, plane,
#if CONFIG_MOTION_VAR
0, 0,
#endif // CONFIG_MOTION_VAR
@ -1288,10 +1348,11 @@ static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
}
}
void av1_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
#if CONFIG_EXT_INTER
void av1_build_inter_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0, 0);
#if CONFIG_EXT_INTER && CONFIG_INTERINTRA
if (is_interintra_pred(&xd->mi[0]->mbmi)) {
BUFFER_SET default_ctx = { { xd->plane[0].dst.buf, NULL, NULL },
{ xd->plane[0].dst.stride, 0, 0 } };
@ -1301,14 +1362,15 @@ void av1_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
}
#else
(void)ctx;
#endif // CONFIG_EXT_INTER
#endif // CONFIG_EXT_INTER && CONFIG_INTERINTRA
}
void av1_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
void av1_build_inter_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 1,
MAX_MB_PLANE - 1);
#if CONFIG_EXT_INTER
#if CONFIG_EXT_INTER && CONFIG_INTERINTRA
if (is_interintra_pred(&xd->mi[0]->mbmi)) {
BUFFER_SET default_ctx = {
{ NULL, xd->plane[1].dst.buf, xd->plane[2].dst.buf },
@ -1321,15 +1383,16 @@ void av1_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
}
#else
(void)ctx;
#endif // CONFIG_EXT_INTER
#endif // CONFIG_EXT_INTER && CONFIG_INTERINTRA
}
// TODO(afergs): Check if ctx can be made constant
void av1_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
void av1_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0,
MAX_MB_PLANE - 1);
#if CONFIG_EXT_INTER
#if CONFIG_EXT_INTER && CONFIG_INTERINTRA
if (is_interintra_pred(&xd->mi[0]->mbmi)) {
BUFFER_SET default_ctx = {
{ xd->plane[0].dst.buf, xd->plane[1].dst.buf, xd->plane[2].dst.buf },
@ -1344,7 +1407,7 @@ void av1_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
}
#else
(void)ctx;
#endif // CONFIG_EXT_INTER
#endif // CONFIG_EXT_INTER && CONFIG_INTERINTRA
}
void av1_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
@ -1522,7 +1585,8 @@ void av1_build_masked_inter_predictor_complex(
} while (--h_remain);
}
void av1_build_inter_predictors_sb_sub8x8_extend(MACROBLOCKD *xd,
void av1_build_inter_predictors_sb_sub8x8_extend(const AV1_COMMON *cm,
MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
@ -1553,7 +1617,7 @@ void av1_build_inter_predictors_sb_sub8x8_extend(MACROBLOCKD *xd,
const int bw = 4 * num_4x4_w;
const int bh = 4 * num_4x4_h;
build_inter_predictors(xd, plane,
build_inter_predictors(cm, xd, plane,
#if CONFIG_MOTION_VAR
0, 0,
#endif // CONFIG_MOTION_VAR
@ -1577,7 +1641,7 @@ void av1_build_inter_predictors_sb_sub8x8_extend(MACROBLOCKD *xd,
#endif // CONFIG_EXT_INTER
}
void av1_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
void av1_build_inter_predictors_sb_extend(const AV1_COMMON *cm, MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
@ -1596,7 +1660,7 @@ void av1_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
const int bw = block_size_wide[plane_bsize];
const int bh = block_size_high[plane_bsize];
build_inter_predictors(xd, plane,
build_inter_predictors(cm, xd, plane,
#if CONFIG_MOTION_VAR
0, 0,
#endif // CONFIG_MOTION_VAR
@ -1697,10 +1761,13 @@ const uint8_t *av1_get_obmc_mask_flipped(int length) {
void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col) {
int i, mi_step;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
xd->mi[0]->mbmi.overlappable_neighbors[0] = 0;
xd->mi[0]->mbmi.overlappable_neighbors[1] = 0;
if (!is_motion_variation_allowed_bsize(mbmi->sb_type)) return;
if (xd->up_available) {
const int ilimit = AOMMIN(xd->n8_w, cm->mi_cols - mi_col);
for (i = 0; i < ilimit; i += mi_step) {
@ -1709,8 +1776,15 @@ void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
MODE_INFO *above_mi =
xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride];
MB_MODE_INFO *above_mbmi = &above_mi->mbmi;
mi_step = AOMMIN(xd->n8_w, mi_size_wide[above_mbmi->sb_type]);
#if CONFIG_CHROMA_SUB8X8
if (above_mbmi->sb_type < BLOCK_8X8) {
++mi_col_offset;
above_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
BLOCK_SIZE above_bsize = AOMMAX(above_mbmi->sb_type, BLOCK_8X8);
mi_step = AOMMIN(xd->n8_w, mi_size_wide[above_bsize]);
if (is_neighbor_overlappable(above_mbmi))
xd->mi[0]->mbmi.overlappable_neighbors[0]++;
@ -1726,7 +1800,15 @@ void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride];
MB_MODE_INFO *left_mbmi = &left_mi->mbmi;
mi_step = AOMMIN(xd->n8_h, mi_size_high[left_mbmi->sb_type]);
#if CONFIG_CHROMA_SUB8X8
if (left_mbmi->sb_type < BLOCK_8X8) {
++mi_row_offset;
left_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
BLOCK_SIZE left_bsize = AOMMAX(left_mbmi->sb_type, BLOCK_8X8);
mi_step = AOMMIN(xd->n8_h, mi_size_high[left_bsize]);
if (is_neighbor_overlappable(left_mbmi))
xd->mi[0]->mbmi.overlappable_neighbors[1]++;
@ -1757,7 +1839,7 @@ int skip_u4x4_pred_in_obmc(BLOCK_SIZE bsize, const struct macroblockd_plane *pd,
#else
case BLOCK_4X4:
case BLOCK_8X4:
case BLOCK_4X8: return dir == 1; break;
case BLOCK_4X8: return dir == 0; break;
#endif
default: return 0;
}
@ -1791,10 +1873,18 @@ void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
i = 0;
do { // for each mi in the above row
const int mi_col_offset = i;
const MB_MODE_INFO *const above_mbmi =
int mi_col_offset = i;
MB_MODE_INFO *above_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
const BLOCK_SIZE a_bsize = above_mbmi->sb_type;
#if CONFIG_CHROMA_SUB8X8
if (above_mbmi->sb_type < BLOCK_8X8) {
++mi_col_offset;
above_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
const BLOCK_SIZE a_bsize = AOMMAX(BLOCK_8X8, above_mbmi->sb_type);
const int mi_step = AOMMIN(xd->n8_w, mi_size_wide[a_bsize]);
if (is_neighbor_overlappable(above_mbmi)) {
@ -1840,10 +1930,18 @@ void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
i = 0;
do { // for each mi in the left column
const int mi_row_offset = i;
const MB_MODE_INFO *const left_mbmi =
int mi_row_offset = i;
MB_MODE_INFO *left_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
const BLOCK_SIZE l_bsize = left_mbmi->sb_type;
#if CONFIG_CHROMA_SUB8X8
if (left_mbmi->sb_type < BLOCK_8X8) {
++mi_row_offset;
left_mbmi =
&xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
const BLOCK_SIZE l_bsize = AOMMAX(BLOCK_8X8, left_mbmi->sb_type);
const int mi_step = AOMMIN(xd->n8_h, mi_size_high[l_bsize]);
if (is_neighbor_overlappable(left_mbmi)) {
@ -1916,7 +2014,15 @@ void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_x, mi_y, bw, bh;
MODE_INFO *above_mi = xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride];
MB_MODE_INFO *above_mbmi = &above_mi->mbmi;
const BLOCK_SIZE a_bsize = above_mbmi->sb_type;
#if CONFIG_CHROMA_SUB8X8
if (above_mbmi->sb_type < BLOCK_8X8) {
++mi_col_offset;
above_mbmi = &xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
const BLOCK_SIZE a_bsize = AOMMAX(BLOCK_8X8, above_mbmi->sb_type);
MB_MODE_INFO backup_mbmi;
mi_step = AOMMIN(xd->n8_w, mi_size_wide[a_bsize]);
@ -1931,8 +2037,8 @@ void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
for (j = 0; j < MAX_MB_PLANE; ++j) {
struct macroblockd_plane *const pd = &xd->plane[j];
setup_pred_plane(&pd->dst, AOMMAX(a_bsize, BLOCK_8X8), tmp_buf[j],
tmp_width[j], tmp_height[j], tmp_stride[j], 0, i, NULL,
setup_pred_plane(&pd->dst, a_bsize, tmp_buf[j], tmp_width[j],
tmp_height[j], tmp_stride[j], 0, i, NULL,
pd->subsampling_x, pd->subsampling_y);
}
for (ref = 0; ref < 1 + has_second_ref(above_mbmi); ++ref) {
@ -1960,8 +2066,8 @@ void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
4);
if (skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset, 0, bw, bh, 0,
0, bw, bh,
build_inter_predictors(cm, xd, j, mi_col_offset, mi_row_offset, 0, bw, bh,
0, 0, bw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
@ -1997,7 +2103,15 @@ void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_x, mi_y, bw, bh;
MODE_INFO *left_mi = xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride];
MB_MODE_INFO *left_mbmi = &left_mi->mbmi;
const BLOCK_SIZE l_bsize = left_mbmi->sb_type;
#if CONFIG_CHROMA_SUB8X8
if (left_mbmi->sb_type < BLOCK_8X8) {
++mi_row_offset;
left_mbmi = &xd->mi[mi_col_offset + mi_row_offset * xd->mi_stride]->mbmi;
}
#endif
const BLOCK_SIZE l_bsize = AOMMAX(left_mbmi->sb_type, BLOCK_8X8);
MB_MODE_INFO backup_mbmi;
mi_step = AOMMIN(xd->n8_h, mi_size_high[l_bsize]);
@ -2012,8 +2126,8 @@ void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
for (j = 0; j < MAX_MB_PLANE; ++j) {
struct macroblockd_plane *const pd = &xd->plane[j];
setup_pred_plane(&pd->dst, AOMMAX(l_bsize, BLOCK_8X8), tmp_buf[j],
tmp_width[j], tmp_height[j], tmp_stride[j], i, 0, NULL,
setup_pred_plane(&pd->dst, l_bsize, tmp_buf[j], tmp_width[j],
tmp_height[j], tmp_stride[j], i, 0, NULL,
pd->subsampling_x, pd->subsampling_y);
}
for (ref = 0; ref < 1 + has_second_ref(left_mbmi); ++ref) {
@ -2041,8 +2155,8 @@ void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
bh = (mi_step << MI_SIZE_LOG2) >> pd->subsampling_y;
if (skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset, 0, bw, bh, 0,
0, bw, bh,
build_inter_predictors(cm, xd, j, mi_col_offset, mi_row_offset, 0, bw, bh,
0, 0, bw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
@ -2185,7 +2299,7 @@ void av1_build_prediction_by_bottom_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
continue;
build_inter_predictors(
xd, j, mi_col_offset, mi_row_offset, y * 2 + x, bw, bh,
cm, xd, j, mi_col_offset, mi_row_offset, y * 2 + x, bw, bh,
(4 * x) >> pd->subsampling_x,
xd->n8_h == 1 ? (4 >> pd->subsampling_y) : 0, pw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
@ -2194,13 +2308,13 @@ void av1_build_prediction_by_bottom_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
mi_x, mi_y);
}
} else {
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset, 0, bw, bh,
0, xd->n8_h == 1 ? (4 >> pd->subsampling_y) : 0,
bw, bh,
build_inter_predictors(
cm, xd, j, mi_col_offset, mi_row_offset, 0, bw, bh, 0,
xd->n8_h == 1 ? (4 >> pd->subsampling_y) : 0, bw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
mi_x, mi_y);
mi_x, mi_y);
}
}
#if CONFIG_EXT_INTER
@ -2292,7 +2406,7 @@ void av1_build_prediction_by_right_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
if ((bp == PARTITION_VERT || bp == PARTITION_SPLIT) && x != 0)
continue;
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset,
build_inter_predictors(cm, xd, j, mi_col_offset, mi_row_offset,
y * 2 + x, bw, bh,
xd->n8_w == 1 ? 4 >> pd->subsampling_x : 0,
(4 * y) >> pd->subsampling_y, bw, ph,
@ -2302,9 +2416,9 @@ void av1_build_prediction_by_right_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
mi_x, mi_y);
}
} else {
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset, 0, bw, bh,
xd->n8_w == 1 ? 4 >> pd->subsampling_x : 0, 0,
bw, bh,
build_inter_predictors(cm, xd, j, mi_col_offset, mi_row_offset, 0, bw,
bh, xd->n8_w == 1 ? 4 >> pd->subsampling_x : 0,
0, bw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
@ -2489,15 +2603,16 @@ void av1_build_ncobmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
/* clang-format off */
#if CONFIG_INTERINTRA
#if CONFIG_EXT_PARTITION
static const int ii_weights1d[MAX_SB_SIZE] = {
26, 25, 24, 24, 23, 23, 22, 22, 21, 21, 20, 20, 19, 19, 18, 18, 17, 17, 17,
16, 16, 16, 15, 15, 15, 15, 14, 14, 14, 13, 13, 13, 13, 13, 12, 12, 12, 12,
12, 11, 11, 11, 11, 11, 11, 10, 10, 10, 10, 10, 10, 10, 10, 10, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
60, 58, 56, 54, 52, 50, 48, 47, 45, 44, 42, 41, 39, 38, 37, 35, 34, 33, 32,
31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 22, 21, 20, 19, 19, 18, 18, 17, 16,
16, 15, 15, 14, 14, 13, 13, 12, 12, 12, 11, 11, 10, 10, 10, 9, 9, 9, 8,
8, 8, 8, 7, 7, 7, 7, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 4, 4,
4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
};
static int ii_size_scales[BLOCK_SIZES] = {
#if CONFIG_CB4X4
@ -2508,10 +2623,10 @@ static int ii_size_scales[BLOCK_SIZES] = {
};
#else
static const int ii_weights1d[MAX_SB_SIZE] = {
26, 25, 24, 24, 23, 23, 22, 22, 21, 21, 20, 20, 19, 19, 18, 18,
17, 17, 17, 16, 16, 16, 15, 15, 15, 15, 14, 14, 14, 13, 13, 13,
13, 13, 12, 12, 12, 12, 12, 11, 11, 11, 11, 11, 11, 10, 10, 10,
10, 10, 10, 10, 10, 10, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9
60, 56, 52, 48, 45, 42, 39, 37, 34, 32, 30, 28, 26, 24, 22, 21,
19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 10, 9, 8, 8, 7, 7,
6, 6, 6, 5, 5, 4, 4, 4, 4, 3, 3, 3, 3, 3, 2, 2,
2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
};
static int ii_size_scales[BLOCK_SIZES] = {
#if CONFIG_CB4X4
@ -2570,35 +2685,8 @@ static void combine_interintra(INTERINTRA_MODE mode, int use_wedge_interintra,
}
break;
case II_D63_PRED:
case II_D117_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = (ii_weights1d[i * size_scale] * 3 +
ii_weights1d[j * size_scale]) >>
2;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
case II_D207_PRED:
case II_D153_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = (ii_weights1d[j * size_scale] * 3 +
ii_weights1d[i * size_scale]) >>
2;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
case II_D135_PRED:
#if CONFIG_ALT_INTRA
case II_SMOOTH_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = ii_weights1d[(i < j ? i : j) * size_scale];
@ -2608,21 +2696,11 @@ static void combine_interintra(INTERINTRA_MODE mode, int use_wedge_interintra,
}
}
break;
#endif
case II_D45_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale =
(ii_weights1d[i * size_scale] + ii_weights1d[j * size_scale]) >>
1;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
#if !CONFIG_ALT_INTRA
case II_TM_PRED:
#endif
case II_DC_PRED:
default:
for (i = 0; i < bh; ++i) {
@ -2686,35 +2764,8 @@ static void combine_interintra_highbd(
}
break;
case II_D63_PRED:
case II_D117_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = (ii_weights1d[i * size_scale] * 3 +
ii_weights1d[j * size_scale]) >>
2;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
case II_D207_PRED:
case II_D153_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = (ii_weights1d[j * size_scale] * 3 +
ii_weights1d[i * size_scale]) >>
2;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
case II_D135_PRED:
#if CONFIG_ALT_INTRA
case II_SMOOTH_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale = ii_weights1d[(i < j ? i : j) * size_scale];
@ -2724,21 +2775,11 @@ static void combine_interintra_highbd(
}
}
break;
#endif
case II_D45_PRED:
for (i = 0; i < bh; ++i) {
for (j = 0; j < bw; ++j) {
int scale =
(ii_weights1d[i * size_scale] + ii_weights1d[j * size_scale]) >>
1;
comppred[i * compstride + j] =
AOM_BLEND_A64(scale, intrapred[i * intrastride + j],
interpred[i * interstride + j]);
}
}
break;
#if !CONFIG_ALT_INTRA
case II_TM_PRED:
#endif
case II_DC_PRED:
default:
for (i = 0; i < bh; ++i) {
@ -2850,6 +2891,7 @@ void av1_build_interintra_predictors(MACROBLOCKD *xd, uint8_t *ypred,
av1_build_interintra_predictors_sbuv(xd, upred, vpred, ustride, vstride, ctx,
bsize);
}
#endif // CONFIG_INTERINTRA
// Builds the inter-predictor for the single ref case
// for use in the encoder to search the wedges efficiently.
@ -2939,13 +2981,13 @@ void av1_build_inter_predictors_for_planes_single_buf(
for (plane = plane_from; plane <= plane_to; ++plane) {
const BLOCK_SIZE plane_bsize =
get_plane_block_size(bsize, &xd->plane[plane]);
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
const int bw = block_size_wide[plane_bsize];
const int bh = block_size_high[plane_bsize];
if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8 && !CONFIG_CB4X4) {
int x, y;
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
assert(bsize == BLOCK_8X8);
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x)
@ -3052,11 +3094,11 @@ void av1_build_wedge_inter_predictor_from_buf(
for (plane = plane_from; plane <= plane_to; ++plane) {
const BLOCK_SIZE plane_bsize =
get_plane_block_size(bsize, &xd->plane[plane]);
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8 && !CONFIG_CB4X4) {
int x, y;
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
assert(bsize == BLOCK_8X8);
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x)

191
third_party/aom/av1/common/reconinter.h vendored
View file

@ -32,11 +32,12 @@
extern "C" {
#endif
static INLINE int has_scale(int xs, int ys) { return xs != 16 || ys != 16; }
static INLINE void inter_predictor(const uint8_t *src, int src_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,
uint8_t *dst, int dst_stride, int subpel_x,
int subpel_y, const struct scale_factors *sf,
int w, int h, ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
@ -44,63 +45,53 @@ static INLINE void inter_predictor(const uint8_t *src, int src_stride,
#endif
int xs, int ys) {
#if CONFIG_DUAL_FILTER
InterpFilter filter_x = av1_get_plane_interp_filter(
const InterpFilter filter_x = av1_get_plane_interp_filter(
interp_filter[1 + 2 * conv_params->ref], conv_params->plane);
InterpFilter filter_y = av1_get_plane_interp_filter(
const InterpFilter filter_y = av1_get_plane_interp_filter(
interp_filter[0 + 2 * conv_params->ref], conv_params->plane);
InterpFilterParams interp_filter_params_x =
const InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(filter_x);
InterpFilterParams interp_filter_params_y =
const InterpFilterParams interp_filter_params_y =
av1_get_interp_filter_params(filter_y);
#else
InterpFilterParams interp_filter_params =
const InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(interp_filter);
const InterpFilterParams interp_filter_params_y = interp_filter_params_x;
#endif
assert(sf);
#if CONFIG_DUAL_FILTER
if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2 &&
conv_params->round == CONVOLVE_OPT_ROUND && xs == 16 && ys == 16) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
#else
if (interp_filter_params.taps == SUBPEL_TAPS && w > 2 && h > 2 &&
conv_params->round == CONVOLVE_OPT_ROUND && xs == 16 && ys == 16) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_y);
#endif
sf->predict[subpel_x != 0][subpel_y != 0][conv_params->ref](
src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h);
} else {
// ref_idx > 0 means this is the second reference frame
// first reference frame's prediction result is already in dst
// therefore we need to average the first and second results
if (has_scale(xs, ys)) {
av1_convolve_c(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, conv_params);
} else if (conv_params->round == CONVOLVE_OPT_NO_ROUND) {
#if CONFIG_CONVOLVE_ROUND
if (conv_params->round == CONVOLVE_OPT_NO_ROUND && xs == 16 && ys == 16)
av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
#if CONFIG_DUAL_FILTER
interp_filter,
interp_filter,
#else // CONFIG_DUAL_FILTER
&interp_filter,
#endif // CONFIG_DUAL_FILTER
subpel_x, xs, subpel_y, ys, conv_params);
conv_params->do_post_rounding = 1;
#else
&interp_filter,
#endif
subpel_x, xs, subpel_y, ys, conv_params);
else
#endif
{
if (xs == 16 && ys == 16) {
av1_convolve(src, src_stride, dst, dst_stride, w, h, interp_filter,
assert(0);
#endif // CONFIG_CONVOLVE_ROUND
} else {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (w <= 2 || h <= 2) {
av1_convolve_c(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, conv_params);
} else {
// If xs == 16 || ys == 16 scaling is happening and the SSE2
// instructions don't support scaling; use the C versions to be safe.
av1_convolve_c(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, conv_params);
}
} else if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
sf->predict[subpel_x != 0][subpel_y != 0][conv_params->ref](
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_filter,
subpel_x, xs, subpel_y, ys, conv_params);
}
}
}
@ -111,46 +102,59 @@ static INLINE void highbd_inter_predictor(const uint8_t *src, int src_stride,
const int subpel_x,
const int subpel_y,
const struct scale_factors *sf, int w,
int h, int ref,
int h, ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
int xs, int ys, int bd) {
const int ref = conv_params->ref;
// ref > 0 means this is the second reference frame
// first reference frame's prediction result is already in dst
// therefore we need to average the first and second results
const int avg = ref > 0;
#if CONFIG_DUAL_FILTER
InterpFilterParams interp_filter_params_x =
const InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * ref]);
InterpFilterParams interp_filter_params_y =
const InterpFilterParams interp_filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * ref]);
#else
InterpFilterParams interp_filter_params =
const InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(interp_filter);
const InterpFilterParams interp_filter_params_y = interp_filter_params_x;
#endif
#if CONFIG_DUAL_FILTER
if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
#else
if (interp_filter_params.taps == SUBPEL_TAPS && w > 2 && h > 2) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_y);
#endif // CONFIG_DUAL_FILTER
sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h, bd);
} else {
// ref > 0 means this is the second reference frame
// first reference frame's prediction result is already in dst
// therefore we need to average the first and second results
int avg = ref > 0;
if (has_scale(xs, ys)) {
av1_highbd_convolve(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, avg, bd);
} else 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,
#if CONFIG_DUAL_FILTER
interp_filter,
#else // CONFIG_DUAL_FILTER
&interp_filter,
#endif // CONFIG_DUAL_FILTER
subpel_x, xs, subpel_y, ys, conv_params, bd);
conv_params->do_post_rounding = 1;
#else
assert(0);
#endif // CONFIG_CONVOLVE_ROUND
} else {
if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
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_filter,
subpel_x, xs, subpel_y, ys, avg, bd);
}
}
}
#endif // CONFIG_HIGHBITDEPTH
@ -254,7 +258,7 @@ void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
#endif // CONFIG_COMPOUND_SEGMENT
#endif // CONFIG_EXT_INTER
void build_inter_predictors(MACROBLOCKD *xd, int plane,
void build_inter_predictors(const AV1_COMMON *cm, MACROBLOCKD *xd, int plane,
#if CONFIG_MOTION_VAR
int mi_col_offset, int mi_row_offset,
#endif // CONFIG_MOTION_VAR
@ -415,20 +419,26 @@ static INLINE void av1_make_inter_predictor(
if (do_warp) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const struct buf_2d *const pre_buf = &pd->pre[ref];
#if CONFIG_EXT_INTER
int compute_avg =
ref && mi->mbmi.interinter_compound_type == COMPOUND_AVERAGE;
#else
int compute_avg = ref;
#endif // CONFIG_EXT_INTER
av1_warp_plane(&final_warp_params,
#if CONFIG_HIGHBITDEPTH
xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
#endif // CONFIG_HIGHBITDEPTH
pre_buf->buf0, pre_buf->width, pre_buf->height,
pre_buf->stride, dst, p_col, p_row, w, h, dst_stride,
pd->subsampling_x, pd->subsampling_y, xs, ys, ref);
pd->subsampling_x, pd->subsampling_y, xs, ys, compute_avg);
return;
}
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#if CONFIG_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
sf, w, h, conv_params->ref, interp_filter, xs, ys,
sf, w, h, conv_params, interp_filter, xs, ys,
xd->bd);
return;
}
@ -526,27 +536,32 @@ static INLINE MV average_split_mvs(const struct macroblockd_plane *pd,
return res;
}
void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
int ic, int mi_row, int mi_col);
void av1_build_inter_predictor_sub8x8(const AV1_COMMON *cm, MACROBLOCKD *xd,
int plane, int i, int ir, int ic,
int mi_row, int mi_col);
void av1_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
void av1_build_inter_predictors_sby(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize);
void av1_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
void av1_build_inter_predictors_sbuv(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize);
void av1_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
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_predictors_sb_sub8x8_extend(MACROBLOCKD *xd,
void av1_build_inter_predictors_sb_sub8x8_extend(const AV1_COMMON *cm,
MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
int mi_row, int mi_col,
BLOCK_SIZE bsize, int block);
void av1_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
void av1_build_inter_predictors_sb_extend(const AV1_COMMON *cm, MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
@ -783,7 +798,7 @@ const uint8_t *av1_get_compound_type_mask_inverse(
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(MACROBLOCKD *xd, uint8_t *ypred,
uint8_t *upred, uint8_t *vpred,
int ystride, int ustride, int vstride,
@ -807,7 +822,7 @@ void av1_build_intra_predictors_for_interintra(MACROBLOCKD *xd,
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,

481
third_party/aom/av1/common/reconintra.c vendored
View file

@ -47,7 +47,11 @@ static const uint8_t extend_modes[INTRA_MODES] = {
NEED_LEFT | NEED_BOTTOMLEFT, // D207
NEED_ABOVE | NEED_ABOVERIGHT, // D63
#if CONFIG_ALT_INTRA
NEED_LEFT | NEED_ABOVE, // SMOOTH
NEED_LEFT | NEED_ABOVE, // SMOOTH
#if CONFIG_SMOOTH_HV
NEED_LEFT | NEED_ABOVE, // SMOOTH_V
NEED_LEFT | NEED_ABOVE, // SMOOTH_H
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT, // TM
};
@ -443,8 +447,10 @@ static int has_top_right(BLOCK_SIZE bsize, int mi_row, int mi_col,
const int plane_bw_unit = AOMMAX(bw_unit >> ss_x, 1);
const int top_right_count_unit = tx_size_wide_unit[txsz];
#if !CONFIG_CB4X4
// Special handling for block sizes 4x8 and 4x4.
if (ss_x == 0 && bw_unit < 2 && col_off == 0) return 1;
#endif
if (row_off > 0) { // Just need to check if enough pixels on the right.
return col_off + top_right_count_unit < plane_bw_unit;
@ -514,9 +520,13 @@ static int has_bottom_left(BLOCK_SIZE bsize, int mi_row, int mi_col,
// and/or bottom-left superblocks. But only the left superblock is
// available, so check if all required pixels fall in that superblock.
if (blk_col_in_sb == 0) {
const int blk_start_row_off = blk_row_in_sb << (bh_in_mi_log2 + !ss_y);
const int blk_start_row_off = blk_row_in_sb
<< (bh_in_mi_log2 + MI_SIZE_LOG2 -
tx_size_wide_log2[0]) >>
ss_y;
const int row_off_in_sb = blk_start_row_off + row_off;
const int sb_height_unit = MAX_MIB_SIZE << !ss_y;
const int sb_height_unit =
MAX_MIB_SIZE << (MI_SIZE_LOG2 - tx_size_wide_log2[0]) >> ss_y;
return row_off_in_sb + bottom_left_count_unit < sb_height_unit;
}
@ -554,6 +564,9 @@ static intra_high_pred_fn dc_pred_high[2][2][TX_SIZES];
#endif // CONFIG_HIGHBITDEPTH
static void av1_init_intra_predictors_internal(void) {
#if CONFIG_EXT_INTRA
assert(NELEMENTS(mode_to_angle_map) == INTRA_MODES);
#endif // CONFIG_EXT_INTRA
#if CONFIG_TX64X64
#define INIT_NO_4X4(p, type) \
p[TX_8X8] = aom_##type##_predictor_8x8; \
@ -567,7 +580,7 @@ static void av1_init_intra_predictors_internal(void) {
p[TX_32X32] = aom_##type##_predictor_32x32
#endif // CONFIG_TX64X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
#define INIT_ALL_SIZES(p, type) \
p[TX_2X2] = aom_##type##_predictor_2x2; \
p[TX_4X4] = aom_##type##_predictor_4x4; \
@ -590,6 +603,10 @@ static void av1_init_intra_predictors_internal(void) {
#if CONFIG_ALT_INTRA
INIT_ALL_SIZES(pred[TM_PRED], paeth);
INIT_ALL_SIZES(pred[SMOOTH_PRED], smooth);
#if CONFIG_SMOOTH_HV
INIT_ALL_SIZES(pred[SMOOTH_V_PRED], smooth_v);
INIT_ALL_SIZES(pred[SMOOTH_H_PRED], smooth_h);
#endif // CONFIG_SMOOTH_HV
#else
INIT_ALL_SIZES(pred[TM_PRED], tm);
#endif // CONFIG_ALT_INTRA
@ -612,6 +629,10 @@ static void av1_init_intra_predictors_internal(void) {
#if CONFIG_ALT_INTRA
INIT_ALL_SIZES(pred_high[TM_PRED], highbd_paeth);
INIT_ALL_SIZES(pred_high[SMOOTH_PRED], highbd_smooth);
#if CONFIG_SMOOTH_HV
INIT_ALL_SIZES(pred_high[SMOOTH_V_PRED], highbd_smooth_v);
INIT_ALL_SIZES(pred_high[SMOOTH_H_PRED], highbd_smooth_h);
#endif // CONFIG_SMOOTH_HV
#else
INIT_ALL_SIZES(pred_high[TM_PRED], highbd_tm);
#endif // CONFIG_ALT_INTRA
@ -786,6 +807,7 @@ static void dr_prediction_z2(uint8_t *dst, ptrdiff_t stride, int bs,
#endif // CONFIG_INTRA_INTERP
} else {
base2 = y >> 8;
assert(base2 >= -1);
shift2 = y & 0xFF;
#if CONFIG_INTRA_INTERP
val = intra_subpel_interp(base2, shift2, left, -1, bs - 1, filter_type);
@ -1191,8 +1213,8 @@ static void highbd_dr_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
#if CONFIG_FILTER_INTRA
#if USE_3TAP_INTRA_FILTER
int av1_filter_intra_taps_3[TX_SIZES][INTRA_MODES][3] = {
#if CONFIG_CB4X4
static int filter_intra_taps_3[TX_SIZES][FILTER_INTRA_MODES][3] = {
#if CONFIG_CHROMA_2X2
{
{ 697, 836, -509 },
{ 993, 513, -482 },
@ -1203,7 +1225,7 @@ int av1_filter_intra_taps_3[TX_SIZES][INTRA_MODES][3] = {
{ 358, 687, -21 },
{ 411, 1083, -470 },
{ 912, 814, -702 },
{ 883, 902, 761 },
{ 883, 902, -761 },
},
#endif
{
@ -1216,7 +1238,7 @@ int av1_filter_intra_taps_3[TX_SIZES][INTRA_MODES][3] = {
{ 358, 687, -21 },
{ 411, 1083, -470 },
{ 912, 814, -702 },
{ 883, 902, 761 },
{ 883, 902, -761 },
},
{
{ 659, 816, -451 },
@ -1270,8 +1292,8 @@ int av1_filter_intra_taps_3[TX_SIZES][INTRA_MODES][3] = {
#endif // CONFIG_TX64X64
};
#else
int av1_filter_intra_taps_4[TX_SIZES][INTRA_MODES][4] = {
#if CONFIG_CB4X4
static int filter_intra_taps_4[TX_SIZES][FILTER_INTRA_MODES][4] = {
#if CONFIG_CHROMA_2X2
{
{ 735, 881, -537, -54 },
{ 1005, 519, -488, -11 },
@ -1375,9 +1397,9 @@ static void filter_intra_predictors_3tap(uint8_t *dst, ptrdiff_t stride, int bs,
int buffer[33][33];
#endif // CONFIG_TX64X64
const TX_SIZE tx_size = get_txsize_from_blocklen(bs);
const int c0 = av1_filter_intra_taps_3[tx_size][mode][0];
const int c1 = av1_filter_intra_taps_3[tx_size][mode][1];
const int c2 = av1_filter_intra_taps_3[tx_size][mode][2];
const int c0 = filter_intra_taps_3[tx_size][mode][0];
const int c1 = filter_intra_taps_3[tx_size][mode][1];
const int c2 = filter_intra_taps_3[tx_size][mode][2];
k = 0;
mean = 0;
@ -1397,13 +1419,11 @@ static void filter_intra_predictors_3tap(uint8_t *dst, ptrdiff_t stride, int bs,
ipred = c0 * buffer[r - 1][c] + c1 * buffer[r][c - 1] +
c2 * buffer[r - 1][c - 1];
buffer[r][c] = ROUND_POWER_OF_TWO_SIGNED(ipred, FILTER_INTRA_PREC_BITS);
buffer[r][c] = clip_pixel(buffer[r][c] + mean) - mean;
}
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
ipred = buffer[r + 1][c + 1] + mean;
dst[c] = clip_pixel(ipred);
}
for (c = 0; c < bs; ++c) dst[c] = clip_pixel(buffer[r + 1][c + 1] + mean);
dst += stride;
}
}
@ -1419,10 +1439,10 @@ static void filter_intra_predictors_4tap(uint8_t *dst, ptrdiff_t stride, int bs,
int buffer[33][65];
#endif // CONFIG_TX64X64
const TX_SIZE tx_size = get_txsize_from_blocklen(bs);
const int c0 = av1_filter_intra_taps_4[tx_size][mode][0];
const int c1 = av1_filter_intra_taps_4[tx_size][mode][1];
const int c2 = av1_filter_intra_taps_4[tx_size][mode][2];
const int c3 = av1_filter_intra_taps_4[tx_size][mode][3];
const int c0 = filter_intra_taps_4[tx_size][mode][0];
const int c1 = filter_intra_taps_4[tx_size][mode][1];
const int c2 = filter_intra_taps_4[tx_size][mode][2];
const int c3 = filter_intra_taps_4[tx_size][mode][3];
k = 0;
mean = 0;
@ -1442,13 +1462,11 @@ static void filter_intra_predictors_4tap(uint8_t *dst, ptrdiff_t stride, int bs,
ipred = c0 * buffer[r - 1][c] + c1 * buffer[r][c - 1] +
c2 * buffer[r - 1][c - 1] + c3 * buffer[r - 1][c + 1];
buffer[r][c] = ROUND_POWER_OF_TWO_SIGNED(ipred, FILTER_INTRA_PREC_BITS);
buffer[r][c] = clip_pixel(buffer[r][c] + mean) - mean;
}
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
ipred = buffer[r + 1][c + 1] + mean;
dst[c] = clip_pixel(ipred);
}
for (c = 0; c < bs; ++c) dst[c] = clip_pixel(buffer[r + 1][c + 1] + mean);
dst += stride;
}
}
@ -1457,90 +1475,90 @@ static void filter_intra_predictors_4tap(uint8_t *dst, ptrdiff_t stride, int bs,
void av1_dc_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, DC_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_DC_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, DC_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_DC_PRED);
#endif
}
void av1_v_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, V_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_V_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, V_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_V_PRED);
#endif
}
void av1_h_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, H_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_H_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, H_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_H_PRED);
#endif
}
void av1_d45_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D45_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D45_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D45_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D45_PRED);
#endif
}
void av1_d135_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D135_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D135_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D135_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D135_PRED);
#endif
}
void av1_d117_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D117_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D117_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D117_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D117_PRED);
#endif
}
void av1_d153_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D153_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D153_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D153_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D153_PRED);
#endif
}
void av1_d207_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D207_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D207_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D207_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D207_PRED);
#endif
}
void av1_d63_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, D63_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_D63_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, D63_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_D63_PRED);
#endif
}
void av1_tm_filter_predictor_c(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {
#if USE_3TAP_INTRA_FILTER
filter_intra_predictors_3tap(dst, stride, bs, above, left, TM_PRED);
filter_intra_predictors_3tap(dst, stride, bs, above, left, FILTER_TM_PRED);
#else
filter_intra_predictors_4tap(dst, stride, bs, above, left, TM_PRED);
filter_intra_predictors_4tap(dst, stride, bs, above, left, FILTER_TM_PRED);
#endif
}
@ -1595,9 +1613,9 @@ static void highbd_filter_intra_predictors_3tap(uint16_t *dst, ptrdiff_t stride,
int preds[33][33];
#endif // CONFIG_TX64X64
const TX_SIZE tx_size = get_txsize_from_blocklen(bs);
const int c0 = av1_filter_intra_taps_3[tx_size][mode][0];
const int c1 = av1_filter_intra_taps_3[tx_size][mode][1];
const int c2 = av1_filter_intra_taps_3[tx_size][mode][2];
const int c0 = filter_intra_taps_3[tx_size][mode][0];
const int c1 = filter_intra_taps_3[tx_size][mode][1];
const int c2 = filter_intra_taps_3[tx_size][mode][2];
k = 0;
mean = 0;
@ -1617,13 +1635,12 @@ static void highbd_filter_intra_predictors_3tap(uint16_t *dst, ptrdiff_t stride,
ipred = c0 * preds[r - 1][c] + c1 * preds[r][c - 1] +
c2 * preds[r - 1][c - 1];
preds[r][c] = ROUND_POWER_OF_TWO_SIGNED(ipred, FILTER_INTRA_PREC_BITS);
preds[r][c] = clip_pixel_highbd(preds[r][c] + mean, bd) - mean;
}
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
ipred = preds[r + 1][c + 1] + mean;
dst[c] = clip_pixel_highbd(ipred, bd);
}
for (c = 0; c < bs; ++c)
dst[c] = clip_pixel_highbd(preds[r + 1][c + 1] + mean, bd);
dst += stride;
}
}
@ -1640,10 +1657,10 @@ static void highbd_filter_intra_predictors_4tap(uint16_t *dst, ptrdiff_t stride,
int preds[33][65];
#endif // CONFIG_TX64X64
const TX_SIZE tx_size = get_txsize_from_blocklen(bs);
const int c0 = av1_filter_intra_taps_4[tx_size][mode][0];
const int c1 = av1_filter_intra_taps_4[tx_size][mode][1];
const int c2 = av1_filter_intra_taps_4[tx_size][mode][2];
const int c3 = av1_filter_intra_taps_4[tx_size][mode][3];
const int c0 = filter_intra_taps_4[tx_size][mode][0];
const int c1 = filter_intra_taps_4[tx_size][mode][1];
const int c2 = filter_intra_taps_4[tx_size][mode][2];
const int c3 = filter_intra_taps_4[tx_size][mode][3];
k = 0;
mean = 0;
@ -1663,13 +1680,12 @@ static void highbd_filter_intra_predictors_4tap(uint16_t *dst, ptrdiff_t stride,
ipred = c0 * preds[r - 1][c] + c1 * preds[r][c - 1] +
c2 * preds[r - 1][c - 1] + c3 * preds[r - 1][c + 1];
preds[r][c] = ROUND_POWER_OF_TWO_SIGNED(ipred, FILTER_INTRA_PREC_BITS);
preds[r][c] = clip_pixel_highbd(preds[r][c] + mean, bd) - mean;
}
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
ipred = preds[r + 1][c + 1] + mean;
dst[c] = clip_pixel_highbd(ipred, bd);
}
for (c = 0; c < bs; ++c)
dst[c] = clip_pixel_highbd(preds[r + 1][c + 1] + mean, bd);
dst += stride;
}
}
@ -1679,11 +1695,11 @@ void av1_highbd_dc_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, DC_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_DC_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, DC_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_DC_PRED, bd);
#endif
}
@ -1691,9 +1707,11 @@ void av1_highbd_v_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, V_PRED, bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_V_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, V_PRED, bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_V_PRED, bd);
#endif
}
@ -1701,9 +1719,11 @@ void av1_highbd_h_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, H_PRED, bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_H_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, H_PRED, bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_H_PRED, bd);
#endif
}
@ -1711,11 +1731,11 @@ void av1_highbd_d45_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D45_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D45_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D45_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D45_PRED, bd);
#endif
}
@ -1723,11 +1743,11 @@ void av1_highbd_d135_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D135_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D135_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D135_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D135_PRED, bd);
#endif
}
@ -1735,11 +1755,11 @@ void av1_highbd_d117_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D117_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D117_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D117_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D117_PRED, bd);
#endif
}
@ -1747,11 +1767,11 @@ void av1_highbd_d153_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D153_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D153_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D153_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D153_PRED, bd);
#endif
}
@ -1759,11 +1779,11 @@ void av1_highbd_d207_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D207_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D207_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D207_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D207_PRED, bd);
#endif
}
@ -1771,11 +1791,11 @@ void av1_highbd_d63_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, D63_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_D63_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, D63_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_D63_PRED, bd);
#endif
}
@ -1783,11 +1803,11 @@ void av1_highbd_tm_filter_predictor_c(uint16_t *dst, ptrdiff_t stride, int bs,
const uint16_t *above,
const uint16_t *left, int bd) {
#if USE_3TAP_INTRA_FILTER
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left, TM_PRED,
bd);
highbd_filter_intra_predictors_3tap(dst, stride, bs, above, left,
FILTER_TM_PRED, bd);
#else
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left, TM_PRED,
bd);
highbd_filter_intra_predictors_4tap(dst, stride, bs, above, left,
FILTER_TM_PRED, bd);
#endif
}
@ -1832,6 +1852,87 @@ static void highbd_filter_intra_predictors(FILTER_INTRA_MODE mode,
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_FILTER_INTRA
#if CONFIG_INTRA_EDGE
static int intra_edge_filter_strength(int bsz, int delta) {
const int d = abs(delta);
int strength = 0;
if (bsz == 8) {
if (d < 8) {
strength = 0;
} else if (d < 32) {
strength = 1;
} else if (d < 90) {
strength = 3;
}
} else if (bsz == 16) {
if (d < 4) {
strength = 0;
} else if (d < 16) {
strength = 1;
} else if (d < 90) {
strength = 3;
}
} else if (bsz == 32) {
if (d < 16) {
strength = 2;
} else if (d < 90) {
strength = 3;
}
}
return strength;
}
static void filter_intra_edge(uint8_t *p, int sz, int strength) {
if (!strength) return;
const int kernel[3][5] = {
{ 0, 4, 8, 4, 0 }, { 0, 5, 6, 5, 0 }, { 2, 4, 4, 4, 2 }
};
const int filt = strength - 1;
uint8_t edge[129];
memcpy(edge, p, sz * sizeof(*p));
for (int i = 1; i < sz - 1; i++) {
int s = 0;
for (int j = 0; j < 5; j++) {
int k = i - 2 + j;
k = (k < 0) ? 0 : k;
k = (k > sz - 1) ? sz - 1 : k;
s += edge[k] * kernel[filt][j];
}
s = (s + 8) >> 4;
p[i] = s;
}
}
#if CONFIG_HIGHBITDEPTH
static void filter_intra_edge_high(uint16_t *p, int sz, int strength) {
if (!strength) return;
const int kernel[3][5] = {
{ 0, 4, 8, 4, 0 }, { 0, 5, 6, 5, 0 }, { 2, 4, 4, 4, 2 }
};
const int filt = strength - 1;
uint16_t edge[129];
memcpy(edge, p, sz * sizeof(*p));
for (int i = 1; i < sz - 1; i++) {
int s = 0;
for (int j = 0; j < 5; j++) {
int k = i - 2 + j;
k = (k < 0) ? 0 : k;
k = (k > sz - 1) ? sz - 1 : k;
s += edge[k] * kernel[filt][j];
}
s = (s + 8) >> 4;
p[i] = s;
}
}
#endif // CONFIG_INTRA_EDGE
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_HIGHBITDEPTH
static void build_intra_predictors_high(
const MACROBLOCKD *xd, const uint8_t *ref8, int ref_stride, uint8_t *dst8,
@ -1842,9 +1943,8 @@ static void build_intra_predictors_high(
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
DECLARE_ALIGNED(16, uint16_t, left_data[MAX_TX_SIZE * 2 + 16]);
DECLARE_ALIGNED(16, uint16_t, above_data[MAX_TX_SIZE * 2 + 16]);
uint16_t *above_row = above_data + 16;
uint16_t *left_col = left_data + 16;
const uint16_t *const_above_row = above_row;
uint16_t *const above_row = above_data + 16;
uint16_t *const left_col = left_data + 16;
const int bs = tx_size_wide[tx_size];
int need_left = extend_modes[mode] & NEED_LEFT;
int need_above = extend_modes[mode] & NEED_ABOVE;
@ -1894,7 +1994,16 @@ static void build_intra_predictors_high(
assert(n_bottomleft_px >= 0);
if ((!need_above && n_left_px == 0) || (!need_left && n_top_px == 0)) {
#if CONFIG_INTRA_EDGE
int val;
if (need_left) {
val = (n_top_px > 0) ? above_ref[0] : base + 1;
} else {
val = (n_left_px > 0) ? ref[-1] : base - 1;
}
#else
const int val = need_left ? base + 1 : base - 1;
#endif
for (i = 0; i < bs; ++i) {
aom_memset16(dst, val, bs);
dst += dst_stride;
@ -1927,7 +2036,15 @@ static void build_intra_predictors_high(
if (i < (bs << need_bottom))
aom_memset16(&left_col[i], left_col[i - 1], (bs << need_bottom) - i);
} else {
aom_memset16(left_col, base + 1, bs << need_bottom);
#if CONFIG_INTRA_EDGE
if (n_top_px > 0) {
aom_memset16(left_col, above_ref[0], bs << need_bottom);
} else {
#endif
aom_memset16(left_col, base + 1, bs << need_bottom);
#if CONFIG_INTRA_EDGE
}
#endif
}
}
@ -1957,20 +2074,58 @@ static void build_intra_predictors_high(
if (i < (bs << need_right))
aom_memset16(&above_row[i], above_row[i - 1], (bs << need_right) - i);
} else {
aom_memset16(above_row, base - 1, bs << need_right);
#if CONFIG_INTRA_EDGE
if (n_left_px > 0) {
aom_memset16(above_row, ref[-1], bs << need_right);
} else {
#endif
aom_memset16(above_row, base - 1, bs << need_right);
#if CONFIG_INTRA_EDGE
}
#endif
}
}
if (need_above_left) {
#if CONFIG_INTRA_EDGE
if (n_top_px > 0 && n_left_px > 0) {
above_row[-1] = above_ref[-1];
} else if (n_top_px > 0) {
above_row[-1] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-1] = ref[-1];
} else {
above_row[-1] = base;
}
#else
above_row[-1] =
n_top_px > 0 ? (n_left_px > 0 ? above_ref[-1] : base + 1) : base - 1;
#endif
left_col[-1] = above_row[-1];
}
#if CONFIG_EXT_INTRA && CONFIG_INTRA_EDGE
if (is_dr_mode && p_angle != 90 && p_angle != 180) {
const int ab_le = need_above_left ? 1 : 0;
if (need_above && n_top_px > 0) {
const int strength = intra_edge_filter_strength(bs, p_angle - 90);
const int need_right = p_angle < 90;
const int n_px = n_top_px + ab_le + (need_right ? n_topright_px : 0);
filter_intra_edge_high(above_row - ab_le, n_px, strength);
}
if (need_left && n_left_px > 0) {
const int strength = intra_edge_filter_strength(bs, p_angle - 180);
const int need_bottom = p_angle > 180;
const int n_px = n_left_px + ab_le + (need_bottom ? n_bottomleft_px : 0);
filter_intra_edge_high(left_col - ab_le, n_px, strength);
}
}
#endif
#if CONFIG_FILTER_INTRA
if (filter_intra_mode_info->use_filter_intra_mode[plane != 0]) {
highbd_filter_intra_predictors(filter_intra_mode, dst, dst_stride, bs,
const_above_row, left_col, xd->bd);
above_row, left_col, xd->bd);
return;
}
#endif // CONFIG_FILTER_INTRA
@ -1982,7 +2137,7 @@ static void build_intra_predictors_high(
if (plane == 0 && av1_is_intra_filter_switchable(p_angle))
filter = xd->mi[0]->mbmi.intra_filter;
#endif // CONFIG_INTRA_INTERP
highbd_dr_predictor(dst, dst_stride, bs, const_above_row, left_col,
highbd_dr_predictor(dst, dst_stride, bs, above_row, left_col,
#if CONFIG_INTRA_INTERP
filter,
#endif // CONFIG_INTRA_INTERP
@ -1994,10 +2149,9 @@ static void build_intra_predictors_high(
// predict
if (mode == DC_PRED) {
dc_pred_high[n_left_px > 0][n_top_px > 0][tx_size](
dst, dst_stride, const_above_row, left_col, xd->bd);
dst, dst_stride, above_row, left_col, xd->bd);
} else {
pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
xd->bd);
pred_high[mode][tx_size](dst, dst_stride, above_row, left_col, xd->bd);
}
}
#endif // CONFIG_HIGHBITDEPTH
@ -2012,9 +2166,8 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
const uint8_t *above_ref = ref - ref_stride;
DECLARE_ALIGNED(16, uint8_t, left_data[MAX_TX_SIZE * 2 + 16]);
DECLARE_ALIGNED(16, uint8_t, above_data[MAX_TX_SIZE * 2 + 16]);
uint8_t *above_row = above_data + 16;
uint8_t *left_col = left_data + 16;
const uint8_t *const_above_row = above_row;
uint8_t *const above_row = above_data + 16;
uint8_t *const left_col = left_data + 16;
const int bs = tx_size_wide[tx_size];
int need_left = extend_modes[mode] & NEED_LEFT;
int need_above = extend_modes[mode] & NEED_ABOVE;
@ -2065,7 +2218,16 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
assert(n_bottomleft_px >= 0);
if ((!need_above && n_left_px == 0) || (!need_left && n_top_px == 0)) {
#if CONFIG_INTRA_EDGE
int val;
if (need_left) {
val = (n_top_px > 0) ? above_ref[0] : 129;
} else {
val = (n_left_px > 0) ? ref[-1] : 127;
}
#else
const int val = need_left ? 129 : 127;
#endif
for (i = 0; i < bs; ++i) {
memset(dst, val, bs);
dst += dst_stride;
@ -2098,7 +2260,15 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
if (i < (bs << need_bottom))
memset(&left_col[i], left_col[i - 1], (bs << need_bottom) - i);
} else {
memset(left_col, 129, bs << need_bottom);
#if CONFIG_INTRA_EDGE
if (n_top_px > 0) {
memset(left_col, above_ref[0], bs << need_bottom);
} else {
#endif
memset(left_col, 129, bs << need_bottom);
#if CONFIG_INTRA_EDGE
}
#endif
}
}
@ -2127,19 +2297,57 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
if (i < (bs << need_right))
memset(&above_row[i], above_row[i - 1], (bs << need_right) - i);
} else {
memset(above_row, 127, bs << need_right);
#if CONFIG_INTRA_EDGE
if (n_left_px > 0) {
memset(above_row, ref[-1], bs << need_right);
} else {
#endif
memset(above_row, 127, bs << need_right);
#if CONFIG_INTRA_EDGE
}
#endif
}
}
if (need_above_left) {
#if CONFIG_INTRA_EDGE
if (n_top_px > 0 && n_left_px > 0) {
above_row[-1] = above_ref[-1];
} else if (n_top_px > 0) {
above_row[-1] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-1] = ref[-1];
} else {
above_row[-1] = 128;
}
#else
above_row[-1] = n_top_px > 0 ? (n_left_px > 0 ? above_ref[-1] : 129) : 127;
#endif
left_col[-1] = above_row[-1];
}
#if CONFIG_EXT_INTRA && CONFIG_INTRA_EDGE
if (is_dr_mode && p_angle != 90 && p_angle != 180) {
const int ab_le = need_above_left ? 1 : 0;
if (need_above && n_top_px > 0) {
const int strength = intra_edge_filter_strength(bs, p_angle - 90);
const int need_right = p_angle < 90;
const int n_px = n_top_px + ab_le + (need_right ? n_topright_px : 0);
filter_intra_edge(above_row - ab_le, n_px, strength);
}
if (need_left && n_left_px > 0) {
const int strength = intra_edge_filter_strength(bs, p_angle - 180);
const int need_bottom = p_angle > 180;
const int n_px = n_left_px + ab_le + (need_bottom ? n_bottomleft_px : 0);
filter_intra_edge(left_col - ab_le, n_px, strength);
}
}
#endif
#if CONFIG_FILTER_INTRA
if (filter_intra_mode_info->use_filter_intra_mode[plane != 0]) {
filter_intra_predictors(filter_intra_mode, dst, dst_stride, bs,
const_above_row, left_col);
filter_intra_predictors(filter_intra_mode, dst, dst_stride, bs, above_row,
left_col);
return;
}
#endif // CONFIG_FILTER_INTRA
@ -2150,7 +2358,7 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
if (plane == 0 && av1_is_intra_filter_switchable(p_angle))
filter = xd->mi[0]->mbmi.intra_filter;
#endif // CONFIG_INTRA_INTERP
dr_predictor(dst, dst_stride, tx_size, const_above_row, left_col,
dr_predictor(dst, dst_stride, tx_size, above_row, left_col,
#if CONFIG_INTRA_INTERP
filter,
#endif // CONFIG_INTRA_INTERP
@ -2165,14 +2373,14 @@ static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
// CFL predict its own DC_PRED for Chromatic planes
if (plane == AOM_PLANE_Y) {
#endif
dc_pred[n_left_px > 0][n_top_px > 0][tx_size](dst, dst_stride,
const_above_row, left_col);
dc_pred[n_left_px > 0][n_top_px > 0][tx_size](dst, dst_stride, above_row,
left_col);
#if CONFIG_CFL
}
#endif
} else {
pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
pred[mode][tx_size](dst, dst_stride, above_row, left_col);
}
}
@ -2216,9 +2424,9 @@ static void predict_square_intra_block(const MACROBLOCKD *xd, int wpx, int hpx,
// the frame bottom edge
const int yd = (xd->mb_to_bottom_edge >> (3 + pd->subsampling_y)) +
(hpx - y - txhpx) - yd_chr_offset;
const int right_available =
(mi_col + ((col_off + txw) >> (1 - pd->subsampling_x))) <
xd->tile.mi_col_end;
const int right_available = mi_col + ((col_off + txw) << pd->subsampling_x >>
(MI_SIZE_LOG2 - tx_size_wide_log2[0])) <
xd->tile.mi_col_end;
const int bottom_available = (yd > 0);
#if CONFIG_EXT_PARTITION_TYPES
const PARTITION_TYPE partition = xd->mi[0]->mbmi.partition;
@ -2246,30 +2454,27 @@ static void predict_square_intra_block(const MACROBLOCKD *xd, int wpx, int hpx,
const int stride = wpx;
int r, c;
const uint8_t *const map = xd->plane[plane != 0].color_index_map;
#if CONFIG_HIGHBITDEPTH
uint16_t *palette = xd->mi[0]->mbmi.palette_mode_info.palette_colors +
plane * PALETTE_MAX_SIZE;
#else
uint8_t *palette = xd->mi[0]->mbmi.palette_mode_info.palette_colors +
plane * PALETTE_MAX_SIZE;
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
for (r = 0; r < bs; ++r)
for (c = 0; c < bs; ++c)
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
dst16[r * dst_stride + c] = palette[map[(r + y) * stride + c + x]];
}
}
} else {
for (r = 0; r < bs; ++r)
for (c = 0; c < bs; ++c)
#endif // CONFIG_HIGHBITDEPTH
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) {
dst[r * dst_stride + c] =
(uint8_t)(palette[map[(r + y) * stride + c + x]]);
(uint8_t)palette[map[(r + y) * stride + c + x]];
}
}
#if CONFIG_HIGHBITDEPTH
}
#else
for (r = 0; r < bs; ++r)
for (c = 0; c < bs; ++c)
dst[r * dst_stride + c] = palette[map[(r + y) * stride + c + x]];
#endif // CONFIG_HIGHBITDEPTH
return;
}
@ -2316,8 +2521,12 @@ void av1_predict_intra_block_facade(MACROBLOCKD *xd, int plane, int block_idx,
// is signaled.
cfl_dc_pred(xd, get_plane_block_size(block_idx, pd), tx_size);
}
cfl_predict_block(xd->cfl, dst, pd->dst.stride, blk_row, blk_col, tx_size,
xd->cfl->dc_pred[plane - 1]);
cfl_predict_block(
xd->cfl, dst, pd->dst.stride, blk_row, blk_col, tx_size,
xd->cfl->dc_pred[plane - 1],
cfl_idx_to_alpha(mbmi->cfl_alpha_idx, mbmi->cfl_alpha_signs[plane - 1],
plane - 1));
}
#endif
}

52
third_party/aom/av1/common/reconintra.h vendored
View file

@ -19,6 +19,20 @@
extern "C" {
#endif
#if CONFIG_DPCM_INTRA
static INLINE int av1_use_dpcm_intra(int plane, PREDICTION_MODE mode,
TX_TYPE tx_type,
const MB_MODE_INFO *const mbmi) {
(void)mbmi;
(void)plane;
#if CONFIG_EXT_INTRA
if (mbmi->sb_type >= BLOCK_8X8 && mbmi->angle_delta[plane != 0]) return 0;
#endif // CONFIG_EXT_INTRA
return (mode == V_PRED && (tx_type == IDTX || tx_type == H_DCT)) ||
(mode == H_PRED && (tx_type == IDTX || tx_type == V_DCT));
}
#endif // CONFIG_DPCM_INTRA
void av1_init_intra_predictors(void);
void av1_predict_intra_block_facade(MACROBLOCKD *xd, int plane, int block_idx,
int blk_col, int blk_row, TX_SIZE tx_size);
@ -27,30 +41,36 @@ void av1_predict_intra_block(const MACROBLOCKD *xd, int bw, int bh,
const uint8_t *ref, int ref_stride, uint8_t *dst,
int dst_stride, int aoff, int loff, int plane);
#if CONFIG_EXT_INTER
#if CONFIG_EXT_INTER && 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, D45_PRED, D135_PRED,
D117_PRED, D153_PRED, D207_PRED, D63_PRED, TM_PRED
DC_PRED, V_PRED, H_PRED,
#if CONFIG_ALT_INTRA
SMOOTH_PRED
#else
TM_PRED
#endif
};
// Mapping of intra mode to the interintra mode
static const INTERINTRA_MODE intra_to_interintra_mode[INTRA_MODES] = {
II_DC_PRED, II_V_PRED, II_H_PRED, II_D45_PRED, II_D135_PRED,
II_D117_PRED, II_D153_PRED, II_D207_PRED, II_D63_PRED,
II_DC_PRED, II_V_PRED, II_H_PRED, II_V_PRED,
#if CONFIG_ALT_INTRA
II_DC_PRED, // Note: Filler value, as there's no II_SMOOTH_PRED.
#endif // CONFIG_ALT_INTRA
II_TM_PRED
};
#endif // CONFIG_EXT_INTER
#ifdef __cplusplus
} // extern "C"
II_SMOOTH_PRED,
#else
II_TM_PRED,
#endif
II_V_PRED, II_H_PRED, II_H_PRED, II_V_PRED,
#if CONFIG_ALT_INTRA
II_SMOOTH_PRED, II_SMOOTH_PRED
#else
II_TM_PRED
#endif
};
#endif // CONFIG_EXT_INTER && CONFIG_INTERINTRA
#if CONFIG_FILTER_INTRA
#define FILTER_INTRA_PREC_BITS 10
extern int av1_filter_intra_taps_4[TX_SIZES][INTRA_MODES][4];
#endif // CONFIG_FILTER_INTRA
#if CONFIG_EXT_INTRA
@ -59,9 +79,15 @@ static INLINE int av1_is_directional_mode(PREDICTION_MODE mode,
return mode != DC_PRED && mode != TM_PRED &&
#if CONFIG_ALT_INTRA
mode != SMOOTH_PRED &&
#if CONFIG_SMOOTH_HV
mode != SMOOTH_V_PRED && mode != SMOOTH_H_PRED &&
#endif // CONFIG_SMOOTH_HV
#endif // CONFIG_ALT_INTRA
bsize >= BLOCK_8X8;
}
#endif // CONFIG_EXT_INTRA
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_RECONINTRA_H_

193
third_party/aom/av1/common/resize.c vendored
View file

@ -21,20 +21,23 @@
#include "aom_dsp/aom_dsp_common.h"
#endif // CONFIG_HIGHBITDEPTH
#include "aom_ports/mem.h"
#include "aom_scale/aom_scale.h"
#include "av1/common/common.h"
#include "av1/common/resize.h"
#include "./aom_scale_rtcd.h"
#define FILTER_BITS 7
#define INTERP_TAPS 8
#define SUBPEL_BITS 5
#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
#define SUBPEL_BITS_RS 5
#define SUBPEL_MASK_RS ((1 << SUBPEL_BITS_RS) - 1)
#define INTERP_PRECISION_BITS 32
typedef int16_t interp_kernel[INTERP_TAPS];
// Filters for interpolation (0.5-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS)] = {
static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS_RS)] = {
{ -3, 0, 35, 64, 35, 0, -3, 0 }, { -3, -1, 34, 64, 36, 1, -3, 0 },
{ -3, -1, 32, 64, 38, 1, -3, 0 }, { -2, -2, 31, 63, 39, 2, -3, 0 },
{ -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 28, 63, 42, 3, -4, 0 },
@ -54,7 +57,7 @@ static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS)] = {
};
// Filters for interpolation (0.625-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS)] = {
static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS_RS)] = {
{ -1, -8, 33, 80, 33, -8, -1, 0 }, { -1, -8, 30, 80, 35, -8, -1, 1 },
{ -1, -8, 28, 80, 37, -7, -2, 1 }, { 0, -8, 26, 79, 39, -7, -2, 1 },
{ 0, -8, 24, 79, 41, -7, -2, 1 }, { 0, -8, 22, 78, 43, -6, -2, 1 },
@ -74,7 +77,7 @@ static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS)] = {
};
// Filters for interpolation (0.75-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS)] = {
static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS_RS)] = {
{ 2, -11, 25, 96, 25, -11, 2, 0 }, { 2, -11, 22, 96, 28, -11, 2, 0 },
{ 2, -10, 19, 95, 31, -11, 2, 0 }, { 2, -10, 17, 95, 34, -12, 2, 0 },
{ 2, -9, 14, 94, 37, -12, 2, 0 }, { 2, -8, 12, 93, 40, -12, 1, 0 },
@ -94,7 +97,7 @@ static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS)] = {
};
// Filters for interpolation (0.875-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS)] = {
static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS_RS)] = {
{ 3, -8, 13, 112, 13, -8, 3, 0 }, { 3, -7, 10, 112, 17, -9, 3, -1 },
{ 2, -6, 7, 111, 21, -9, 3, -1 }, { 2, -5, 4, 111, 24, -10, 3, -1 },
{ 2, -4, 1, 110, 28, -11, 3, -1 }, { 1, -3, -1, 108, 32, -12, 4, -1 },
@ -114,7 +117,7 @@ static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS)] = {
};
// Filters for interpolation (full-band) - no filtering for integer pixels
static const interp_kernel filteredinterp_filters1000[(1 << SUBPEL_BITS)] = {
static const interp_kernel filteredinterp_filters1000[(1 << SUBPEL_BITS_RS)] = {
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -3, 128, 3, -1, 0, 0 },
{ -1, 2, -6, 127, 7, -2, 1, 0 }, { -1, 3, -9, 126, 12, -4, 1, 0 },
{ -1, 4, -12, 125, 16, -5, 1, 0 }, { -1, 4, -14, 123, 20, -6, 2, 0 },
@ -187,13 +190,13 @@ static void interpolate(const uint8_t *const input, int inlength,
for (x = 0, y = offset; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
sum += filter[k] * input[AOMMAX(AOMMIN(pk, inlength - 1), 0)];
}
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
@ -202,20 +205,20 @@ static void interpolate(const uint8_t *const input, int inlength,
for (x = 0, y = offset; x < x1; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[AOMMAX(int_pel - INTERP_TAPS / 2 + 1 + k, 0)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
// Middle part.
for (; x <= x2; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
@ -226,18 +229,19 @@ static void interpolate(const uint8_t *const input, int inlength,
for (; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] *
input[AOMMIN(int_pel - INTERP_TAPS / 2 + 1 + k, inlength - 1)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
}
}
#ifndef __clang_analyzer__
static void down2_symeven(const uint8_t *const input, int length,
uint8_t *output) {
// Actual filter len = 2 * filter_len_half.
@ -254,9 +258,9 @@ static void down2_symeven(const uint8_t *const input, int length,
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
sum +=
(input[AOMMAX(i - j, 0)] + input[AOMMIN(i + 1 + j, length - 1)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@ -266,7 +270,7 @@ static void down2_symeven(const uint8_t *const input, int length,
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
sum += (input[AOMMAX(i - j, 0)] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@ -284,15 +288,15 @@ static void down2_symeven(const uint8_t *const input, int length,
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
sum +=
(input[i - j] + input[AOMMIN(i + 1 + j, length - 1)]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
}
}
#endif
static void down2_symodd(const uint8_t *const input, int length,
uint8_t *output) {
@ -367,22 +371,19 @@ static int get_down2_steps(int in_length, int out_length) {
static void resize_multistep(const uint8_t *const input, int length,
uint8_t *output, int olength, uint8_t *otmp) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
return;
}
steps = get_down2_steps(length, olength);
const int steps = get_down2_steps(length, olength);
if (steps > 0) {
int s;
uint8_t *out = NULL;
uint8_t *otmp2;
int filteredlength = length;
assert(otmp != NULL);
otmp2 = otmp + get_down2_length(length, 1);
for (s = 0; s < steps; ++s) {
uint8_t *otmp2 = otmp + get_down2_length(length, 1);
for (int s = 0; s < steps; ++s) {
const int proj_filteredlength = get_down2_length(filteredlength, 1);
const uint8_t *const in = (s == 0 ? input : out);
if (s == steps - 1 && proj_filteredlength == olength)
@ -426,8 +427,7 @@ void av1_resize_plane(const uint8_t *const input, int height, int width,
int out_stride) {
int i;
uint8_t *intbuf = (uint8_t *)malloc(sizeof(uint8_t) * width2 * height);
uint8_t *tmpbuf =
(uint8_t *)malloc(sizeof(uint8_t) * (width < height ? height : width));
uint8_t *tmpbuf = (uint8_t *)malloc(sizeof(uint8_t) * AOMMAX(width, height));
uint8_t *arrbuf = (uint8_t *)malloc(sizeof(uint8_t) * height);
uint8_t *arrbuf2 = (uint8_t *)malloc(sizeof(uint8_t) * height2);
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
@ -489,13 +489,13 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
for (x = 0, y = offset; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
sum += filter[k] * input[AOMMAX(AOMMIN(pk, inlength - 1), 0)];
}
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
@ -504,20 +504,20 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
for (x = 0, y = offset; x < x1; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[AOMMAX(int_pel - INTERP_TAPS / 2 + 1 + k, 0)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
// Middle part.
for (; x <= x2; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
@ -528,18 +528,18 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
for (; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
sub_pel =
(y >> (INTERP_PRECISION_BITS - SUBPEL_BITS_RS)) & SUBPEL_MASK_RS;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] *
input[AOMMIN(int_pel - INTERP_TAPS / 2 + 1 + k, inlength - 1)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
}
}
#ifndef __clang_analyzer__
static void highbd_down2_symeven(const uint16_t *const input, int length,
uint16_t *output, int bd) {
// Actual filter len = 2 * filter_len_half.
@ -556,9 +556,9 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
sum +=
(input[AOMMAX(0, i - j)] + input[AOMMIN(i + 1 + j, length - 1)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@ -568,7 +568,7 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
sum += (input[AOMMAX(0, i - j)] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@ -586,9 +586,8 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
sum +=
(input[i - j] + input[AOMMIN(i + 1 + j, length - 1)]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@ -612,8 +611,7 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + j >= length ? length - 1 : i + j)]) *
sum += (input[AOMMAX(i - j, 0)] + input[AOMMIN(i + j, length - 1)]) *
filter[j];
}
sum >>= FILTER_BITS;
@ -624,7 +622,7 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
sum += (input[AOMMAX(i - j, 0)] + input[i + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@ -642,34 +640,31 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
sum += (input[i - j] + input[AOMMIN(i + j, length - 1)]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
}
}
#endif
static void highbd_resize_multistep(const uint16_t *const input, int length,
uint16_t *output, int olength,
uint16_t *otmp, int bd) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
return;
}
steps = get_down2_steps(length, olength);
const int steps = get_down2_steps(length, olength);
if (steps > 0) {
int s;
uint16_t *out = NULL;
uint16_t *otmp2;
int filteredlength = length;
assert(otmp != NULL);
otmp2 = otmp + get_down2_length(length, 1);
for (s = 0; s < steps; ++s) {
uint16_t *otmp2 = otmp + get_down2_length(length, 1);
for (int s = 0; s < steps; ++s) {
const int proj_filteredlength = get_down2_length(filteredlength, 1);
const uint16_t *const in = (s == 0 ? input : out);
if (s == steps - 1 && proj_filteredlength == olength)
@ -716,7 +711,7 @@ void av1_highbd_resize_plane(const uint8_t *const input, int height, int width,
int i;
uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
uint16_t *tmpbuf =
(uint16_t *)malloc(sizeof(uint16_t) * (width < height ? height : width));
(uint16_t *)malloc(sizeof(uint16_t) * AOMMAX(width, height));
uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * height);
uint16_t *arrbuf2 = (uint16_t *)malloc(sizeof(uint16_t) * height2);
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
@ -819,3 +814,69 @@ void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
ouv_stride, bd);
}
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_HIGHBITDEPTH
static void resize_and_extend_frame(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst, int bd) {
#else
static void resize_and_extend_frame(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst) {
#endif // CONFIG_HIGHBITDEPTH
// TODO(dkovalev): replace YV12_BUFFER_CONFIG with aom_image_t
int i;
const uint8_t *const srcs[3] = { src->y_buffer, src->u_buffer,
src->v_buffer };
const int src_strides[3] = { src->y_stride, src->uv_stride, src->uv_stride };
const int src_widths[3] = { src->y_crop_width, src->uv_crop_width,
src->uv_crop_width };
const int src_heights[3] = { src->y_crop_height, src->uv_crop_height,
src->uv_crop_height };
uint8_t *const dsts[3] = { dst->y_buffer, dst->u_buffer, dst->v_buffer };
const int dst_strides[3] = { dst->y_stride, dst->uv_stride, dst->uv_stride };
const int dst_widths[3] = { dst->y_crop_width, dst->uv_crop_width,
dst->uv_crop_width };
const int dst_heights[3] = { dst->y_crop_height, dst->uv_crop_height,
dst->uv_crop_height };
for (i = 0; i < MAX_MB_PLANE; ++i) {
#if CONFIG_HIGHBITDEPTH
if (src->flags & YV12_FLAG_HIGHBITDEPTH)
av1_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
src_strides[i], dsts[i], dst_heights[i],
dst_widths[i], dst_strides[i], bd);
else
#endif // CONFIG_HIGHBITDEPTH
av1_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
}
aom_extend_frame_borders(dst);
}
YV12_BUFFER_CONFIG *av1_scale_if_required_fast(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled) {
if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
cm->mi_rows * MI_SIZE != unscaled->y_height) {
// For 2x2 scaling down.
aom_scale_frame(unscaled, scaled, unscaled->y_buffer, 9, 2, 1, 2, 1, 0);
aom_extend_frame_borders(scaled);
return scaled;
} else {
return unscaled;
}
}
YV12_BUFFER_CONFIG *av1_scale_if_required(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled) {
if (cm->width != unscaled->y_width || cm->height != unscaled->y_height) {
#if CONFIG_HIGHBITDEPTH
resize_and_extend_frame(unscaled, scaled, (int)cm->bit_depth);
#else
resize_and_extend_frame(unscaled, scaled);
#endif // CONFIG_HIGHBITDEPTH
return scaled;
} else {
return unscaled;
}
}

9
third_party/aom/av1/common/resize.h vendored
View file

@ -14,6 +14,7 @@
#include <stdio.h>
#include "aom/aom_integer.h"
#include "av1/common/onyxc_int.h"
#ifdef __cplusplus
extern "C" {
@ -62,6 +63,14 @@ void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
int owidth, int bd);
#endif // CONFIG_HIGHBITDEPTH
YV12_BUFFER_CONFIG *av1_scale_if_required_fast(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled);
YV12_BUFFER_CONFIG *av1_scale_if_required(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled);
#ifdef __cplusplus
} // extern "C"
#endif

14
third_party/aom/av1/common/restoration.c vendored
View file

@ -160,9 +160,16 @@ static void loop_wiener_filter_tile(uint8_t *data, int tile_idx, int width,
int h = AOMMIN(MAX_SB_SIZE, (v_end - i + 15) & ~15);
const uint8_t *data_p = data + i * stride + j;
uint8_t *dst_p = dst + i * dst_stride + j;
#if USE_WIENER_HIGH_INTERMEDIATE_PRECISION
aom_convolve8_add_src_hip(data_p, stride, dst_p, dst_stride,
rst->rsi->wiener_info[tile_idx].hfilter, 16,
rst->rsi->wiener_info[tile_idx].vfilter, 16, w,
h);
#else
aom_convolve8_add_src(data_p, stride, dst_p, dst_stride,
rst->rsi->wiener_info[tile_idx].hfilter, 16,
rst->rsi->wiener_info[tile_idx].vfilter, 16, w, h);
#endif // USE_WIENER_HIGH_INTERMEDIATE_PRECISION
}
}
@ -992,10 +999,17 @@ static void loop_wiener_filter_tile_highbd(uint16_t *data, int tile_idx,
int h = AOMMIN(MAX_SB_SIZE, (v_end - i + 15) & ~15);
const uint16_t *data_p = data + i * stride + j;
uint16_t *dst_p = dst + i * dst_stride + j;
#if USE_WIENER_HIGH_INTERMEDIATE_PRECISION
aom_highbd_convolve8_add_src_hip(
CONVERT_TO_BYTEPTR(data_p), stride, CONVERT_TO_BYTEPTR(dst_p),
dst_stride, rst->rsi->wiener_info[tile_idx].hfilter, 16,
rst->rsi->wiener_info[tile_idx].vfilter, 16, w, h, bit_depth);
#else
aom_highbd_convolve8_add_src(
CONVERT_TO_BYTEPTR(data_p), stride, CONVERT_TO_BYTEPTR(dst_p),
dst_stride, rst->rsi->wiener_info[tile_idx].hfilter, 16,
rst->rsi->wiener_info[tile_idx].vfilter, 16, w, h, bit_depth);
#endif // USE_WIENER_HIGH_INTERMEDIATE_PRECISION
}
}

7
third_party/aom/av1/common/restoration.h vendored
View file

@ -81,10 +81,17 @@ extern "C" {
#define WIENER_FILT_PREC_BITS 7
#define WIENER_FILT_STEP (1 << WIENER_FILT_PREC_BITS)
// Whether to use high intermediate precision filtering
#define USE_WIENER_HIGH_INTERMEDIATE_PRECISION 1
// Central values for the taps
#define WIENER_FILT_TAP0_MIDV (3)
#define WIENER_FILT_TAP1_MIDV (-7)
#define WIENER_FILT_TAP2_MIDV (15)
#define WIENER_FILT_TAP3_MIDV \
(WIENER_FILT_STEP - \
2 * (WIENER_FILT_TAP0_MIDV + WIENER_FILT_TAP1_MIDV + \
WIENER_FILT_TAP2_MIDV))
#define WIENER_FILT_TAP0_BITS 4
#define WIENER_FILT_TAP1_BITS 5

106
third_party/aom/av1/common/scan.c vendored
View file

@ -14,7 +14,7 @@
#include "av1/common/common_data.h"
#include "av1/common/scan.h"
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
DECLARE_ALIGNED(16, static const int16_t, default_scan_2x2[4]) = {
0, 1, 2, 3,
};
@ -73,17 +73,17 @@ DECLARE_ALIGNED(16, static const int16_t, mrow_scan_8x4[32]) = {
};
DECLARE_ALIGNED(16, static const int16_t, default_scan_4x16[64]) = {
0, 1, 4, 2, 5, 8, 3, 6, 9, 12, 7, 10, 13, 16, 11, 14,
17, 20, 15, 18, 21, 24, 19, 22, 25, 28, 23, 26, 29, 32, 27, 30,
33, 36, 31, 34, 37, 40, 35, 38, 41, 44, 39, 42, 45, 48, 43, 46,
49, 52, 47, 50, 53, 56, 51, 54, 57, 60, 55, 58, 61, 59, 62, 63,
0, 1, 4, 5, 2, 8, 6, 9, 10, 3, 12, 7, 13, 11, 14, 16,
17, 15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
};
DECLARE_ALIGNED(16, static const int16_t, default_scan_16x4[64]) = {
0, 1, 16, 2, 17, 32, 3, 18, 33, 48, 4, 19, 34, 49, 5, 20,
35, 50, 6, 21, 36, 51, 7, 22, 37, 52, 8, 23, 38, 53, 9, 24,
39, 54, 10, 25, 40, 55, 11, 26, 41, 56, 12, 27, 42, 57, 13, 28,
43, 58, 14, 29, 44, 59, 15, 30, 45, 60, 31, 46, 61, 47, 62, 63,
0, 1, 16, 17, 2, 32, 18, 33, 34, 3, 48, 19, 49, 35, 50, 4,
20, 51, 36, 5, 52, 21, 37, 53, 6, 22, 38, 54, 7, 23, 39, 55,
8, 24, 40, 56, 9, 25, 41, 57, 10, 26, 42, 58, 11, 27, 43, 59,
12, 28, 44, 60, 13, 29, 45, 61, 14, 30, 46, 62, 15, 31, 47, 63,
};
#if CONFIG_EXT_TX
@ -1491,7 +1491,7 @@ DECLARE_ALIGNED(16, static const int16_t, default_scan_64x64[4096]) = {
};
#endif // CONFIG_TX64X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
DECLARE_ALIGNED(16, static const int16_t,
default_scan_2x2_neighbors[5 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 1, 1, 2, 0, 0,
@ -1582,36 +1582,36 @@ DECLARE_ALIGNED(16, static const int16_t,
DECLARE_ALIGNED(16, static const int16_t,
default_scan_4x16_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 1, 1, 1, 4, 4, 4, 2, 2, 2, 5, 5, 8, 8,
8, 3, 6, 6, 9, 9, 12, 12, 12, 7, 10, 10, 13, 13, 16, 16, 16, 11, 14,
14, 17, 17, 20, 20, 20, 15, 18, 18, 21, 21, 24, 24, 24, 19, 22, 22, 25, 25,
28, 28, 28, 23, 26, 26, 29, 29, 32, 32, 32, 27, 30, 30, 33, 33, 36, 36, 36,
31, 34, 34, 37, 37, 40, 40, 40, 35, 38, 38, 41, 41, 44, 44, 44, 39, 42, 42,
45, 45, 48, 48, 48, 43, 46, 46, 49, 49, 52, 52, 52, 47, 50, 50, 53, 53, 56,
56, 56, 51, 54, 54, 57, 57, 60, 55, 58, 58, 61, 59, 62, 0, 0
0, 0, 0, 0, 0, 0, 1, 4, 1, 1, 4, 4, 2, 5, 5, 8, 6, 9, 2,
2, 8, 8, 3, 6, 9, 12, 7, 10, 10, 13, 12, 12, 13, 16, 11, 14, 14, 17,
15, 18, 16, 16, 17, 20, 18, 21, 19, 22, 20, 20, 21, 24, 22, 25, 23, 26, 24,
24, 25, 28, 26, 29, 27, 30, 28, 28, 29, 32, 30, 33, 31, 34, 32, 32, 33, 36,
34, 37, 35, 38, 36, 36, 37, 40, 38, 41, 39, 42, 40, 40, 41, 44, 42, 45, 43,
46, 44, 44, 45, 48, 46, 49, 47, 50, 48, 48, 49, 52, 50, 53, 51, 54, 52, 52,
53, 56, 54, 57, 55, 58, 56, 56, 57, 60, 58, 61, 59, 62, 0, 0
};
DECLARE_ALIGNED(16, static const int16_t,
default_scan_16x4_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 1, 1, 1, 16, 16, 16, 2, 2, 2, 17, 17, 32, 32,
32, 3, 3, 3, 18, 18, 33, 33, 48, 4, 4, 4, 19, 19, 34, 34, 49, 5, 5,
5, 20, 20, 35, 35, 50, 6, 6, 6, 21, 21, 36, 36, 51, 7, 7, 7, 22, 22,
37, 37, 52, 8, 8, 8, 23, 23, 38, 38, 53, 9, 9, 9, 24, 24, 39, 39, 54,
10, 10, 10, 25, 25, 40, 40, 55, 11, 11, 11, 26, 26, 41, 41, 56, 12, 12, 12,
27, 27, 42, 42, 57, 13, 13, 13, 28, 28, 43, 43, 58, 14, 14, 14, 29, 29, 44,
44, 59, 15, 30, 30, 45, 45, 60, 31, 46, 46, 61, 47, 62, 0, 0
0, 0, 0, 0, 0, 0, 1, 16, 1, 1, 16, 16, 2, 17, 17, 32, 18, 33, 2,
2, 32, 32, 3, 18, 33, 48, 19, 34, 34, 49, 3, 3, 4, 19, 35, 50, 20, 35,
4, 4, 36, 51, 5, 20, 21, 36, 37, 52, 5, 5, 6, 21, 22, 37, 38, 53, 6,
6, 7, 22, 23, 38, 39, 54, 7, 7, 8, 23, 24, 39, 40, 55, 8, 8, 9, 24,
25, 40, 41, 56, 9, 9, 10, 25, 26, 41, 42, 57, 10, 10, 11, 26, 27, 42, 43,
58, 11, 11, 12, 27, 28, 43, 44, 59, 12, 12, 13, 28, 29, 44, 45, 60, 13, 13,
14, 29, 30, 45, 46, 61, 14, 14, 15, 30, 31, 46, 47, 62, 0, 0
};
#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, static const int16_t,
mrow_scan_4x16_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 1, 1, 1, 16, 16, 16, 2, 2, 2, 17, 17, 32, 32,
32, 3, 3, 3, 18, 18, 33, 33, 48, 4, 4, 4, 19, 19, 34, 34, 49, 5, 5,
5, 20, 20, 35, 35, 50, 6, 6, 6, 21, 21, 36, 36, 51, 7, 7, 7, 22, 22,
37, 37, 52, 8, 8, 8, 23, 23, 38, 38, 53, 9, 9, 9, 24, 24, 39, 39, 54,
10, 10, 10, 25, 25, 40, 40, 55, 11, 11, 11, 26, 26, 41, 41, 56, 12, 12, 12,
27, 27, 42, 42, 57, 13, 13, 13, 28, 28, 43, 43, 58, 14, 14, 14, 29, 29, 44,
44, 59, 15, 30, 30, 45, 45, 60, 31, 46, 46, 61, 47, 62, 0, 0
0, 0, 0, 0, 1, 1, 2, 2, 0, 0, 1, 4, 2, 5, 3, 6, 4, 4, 5,
8, 6, 9, 7, 10, 8, 8, 9, 12, 10, 13, 11, 14, 12, 12, 13, 16, 14, 17,
15, 18, 16, 16, 17, 20, 18, 21, 19, 22, 20, 20, 21, 24, 22, 25, 23, 26, 24,
24, 25, 28, 26, 29, 27, 30, 28, 28, 29, 32, 30, 33, 31, 34, 32, 32, 33, 36,
34, 37, 35, 38, 36, 36, 37, 40, 38, 41, 39, 42, 40, 40, 41, 44, 42, 45, 43,
46, 44, 44, 45, 48, 46, 49, 47, 50, 48, 48, 49, 52, 50, 53, 51, 54, 52, 52,
53, 56, 54, 57, 55, 58, 56, 56, 57, 60, 58, 61, 59, 62, 0, 0
};
DECLARE_ALIGNED(16, static const int16_t,
@ -4249,7 +4249,7 @@ DECLARE_ALIGNED(16, static const int16_t,
};
#endif // CONFIG_TX64X64
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
DECLARE_ALIGNED(16, static const int16_t, av1_default_iscan_2x2[4]) = { 0, 1, 2,
3 };
#endif
@ -4307,17 +4307,17 @@ DECLARE_ALIGNED(16, static const int16_t, av1_mrow_iscan_8x4[32]) = {
};
DECLARE_ALIGNED(16, static const int16_t, av1_default_iscan_4x16[64]) = {
0, 1, 3, 6, 2, 4, 7, 10, 5, 8, 11, 14, 9, 12, 15, 18,
13, 16, 19, 22, 17, 20, 23, 26, 21, 24, 27, 30, 25, 28, 31, 34,
29, 32, 35, 38, 33, 36, 39, 42, 37, 40, 43, 46, 41, 44, 47, 50,
45, 48, 51, 54, 49, 52, 55, 58, 53, 56, 59, 61, 57, 60, 62, 63,
0, 1, 4, 9, 2, 3, 6, 11, 5, 7, 8, 13, 10, 12, 14, 17,
15, 16, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
};
DECLARE_ALIGNED(16, static const int16_t, av1_default_iscan_16x4[64]) = {
0, 1, 3, 6, 10, 14, 18, 22, 26, 30, 34, 38, 42, 46, 50, 54,
2, 4, 7, 11, 15, 19, 23, 27, 31, 35, 39, 43, 47, 51, 55, 58,
5, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 59, 61,
9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 60, 62, 63,
0, 1, 4, 9, 15, 19, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60,
2, 3, 6, 11, 16, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61,
5, 7, 8, 13, 18, 22, 26, 30, 34, 38, 42, 46, 50, 54, 58, 62,
10, 12, 14, 17, 20, 23, 27, 31, 35, 39, 43, 47, 51, 55, 59, 63,
};
#if CONFIG_EXT_TX
@ -4343,10 +4343,10 @@ DECLARE_ALIGNED(16, static const int16_t, av1_mcol_iscan_4x16[64]) = {
};
DECLARE_ALIGNED(16, static const int16_t, av1_mcol_iscan_16x4[64]) = {
0, 16, 32, 48, 1, 17, 33, 49, 2, 18, 34, 50, 3, 19, 35, 51,
4, 20, 36, 52, 5, 21, 37, 53, 6, 22, 38, 54, 7, 23, 39, 55,
8, 24, 40, 56, 9, 25, 41, 57, 10, 26, 42, 58, 11, 27, 43, 59,
12, 28, 44, 60, 13, 29, 45, 61, 14, 30, 46, 62, 15, 31, 47, 63,
0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60,
1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61,
2, 6, 10, 14, 18, 22, 26, 30, 34, 38, 42, 46, 50, 54, 58, 62,
3, 7, 11, 15, 19, 23, 27, 31, 35, 39, 43, 47, 51, 55, 59, 63,
};
#endif // CONFIG_EXT_TX
@ -5719,7 +5719,7 @@ DECLARE_ALIGNED(16, static const int16_t, av1_default_iscan_64x64[4096]) = {
#endif // CONFIG_TX64X64
const SCAN_ORDER av1_default_scan_orders[TX_SIZES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{ default_scan_2x2, av1_default_iscan_2x2, default_scan_2x2_neighbors },
#endif
{ default_scan_4x4, av1_default_iscan_4x4, default_scan_4x4_neighbors },
@ -5732,7 +5732,7 @@ const SCAN_ORDER av1_default_scan_orders[TX_SIZES] = {
};
const SCAN_ORDER av1_intra_scan_orders[TX_SIZES_ALL][TX_TYPES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{
// TX_2X2
{ default_scan_2x2, av1_default_iscan_2x2, default_scan_2x2_neighbors },
@ -6043,7 +6043,7 @@ const SCAN_ORDER av1_intra_scan_orders[TX_SIZES_ALL][TX_TYPES] = {
};
const SCAN_ORDER av1_inter_scan_orders[TX_SIZES_ALL][TX_TYPES] = {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
{
// TX_2X2
{ default_scan_2x2, av1_default_iscan_2x2, default_scan_2x2_neighbors },
@ -6493,7 +6493,7 @@ const SCAN_ORDER av1_inter_scan_orders[TX_SIZES_ALL][TX_TYPES] = {
static uint32_t *get_non_zero_prob(FRAME_CONTEXT *fc, TX_SIZE tx_size,
TX_TYPE tx_type) {
switch (tx_size) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
case TX_2X2: return fc->non_zero_prob_2x2[tx_type];
#endif
case TX_4X4: return fc->non_zero_prob_4X4[tx_type];
@ -6515,7 +6515,7 @@ static uint32_t *get_non_zero_prob(FRAME_CONTEXT *fc, TX_SIZE tx_size,
static int16_t *get_adapt_scan(FRAME_CONTEXT *fc, TX_SIZE tx_size,
TX_TYPE tx_type) {
switch (tx_size) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
case TX_2X2: return fc->scan_2x2[tx_type];
#endif
case TX_4X4: return fc->scan_4X4[tx_type];
@ -6537,7 +6537,7 @@ static int16_t *get_adapt_scan(FRAME_CONTEXT *fc, TX_SIZE tx_size,
static int16_t *get_adapt_iscan(FRAME_CONTEXT *fc, TX_SIZE tx_size,
TX_TYPE tx_type) {
switch (tx_size) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
case TX_2X2: return fc->iscan_2x2[tx_type];
#endif
case TX_4X4: return fc->iscan_4X4[tx_type];
@ -6559,7 +6559,7 @@ static int16_t *get_adapt_iscan(FRAME_CONTEXT *fc, TX_SIZE tx_size,
static int16_t *get_adapt_nb(FRAME_CONTEXT *fc, TX_SIZE tx_size,
TX_TYPE tx_type) {
switch (tx_size) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
case TX_2X2: return fc->nb_2x2[tx_type];
#endif
case TX_4X4: return fc->nb_4X4[tx_type];
@ -6581,7 +6581,7 @@ static int16_t *get_adapt_nb(FRAME_CONTEXT *fc, TX_SIZE tx_size,
static uint32_t *get_non_zero_counts(FRAME_COUNTS *counts, TX_SIZE tx_size,
TX_TYPE tx_type) {
switch (tx_size) {
#if CONFIG_CB4X4
#if CONFIG_CHROMA_2X2
case TX_2X2: return counts->non_zero_count_2x2[tx_type];
#endif
case TX_4X4: return counts->non_zero_count_4X4[tx_type];
@ -6606,7 +6606,7 @@ static INLINE int clamp_64(int64_t value, int low, int high) {
static void update_scan_prob(AV1_COMMON *cm, TX_SIZE tx_size, TX_TYPE tx_type,
int rate_16) {
FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
FRAME_CONTEXT *pre_fc = cm->pre_fc;
uint32_t *prev_non_zero_prob = get_non_zero_prob(pre_fc, tx_size, tx_type);
uint32_t *non_zero_prob = get_non_zero_prob(cm->fc, tx_size, tx_type);
uint32_t *non_zero_count = get_non_zero_counts(&cm->counts, tx_size, tx_type);

2
third_party/aom/av1/common/seg_common.h vendored
View file

@ -48,9 +48,7 @@ struct segmentation {
struct segmentation_probs {
aom_prob tree_probs[SEG_TREE_PROBS];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob tree_cdf[CDF_SIZE(MAX_SEGMENTS)];
#endif
aom_prob pred_probs[PREDICTION_PROBS];
};

12
third_party/aom/av1/common/thread_common.c vendored
View file

@ -113,7 +113,7 @@ static INLINE void loop_filter_block_plane_ver(
break;
case LF_PATH_SLOW:
av1_filter_block_plane_non420_ver(cm, &planes[plane], mi, mi_row,
mi_col);
mi_col, plane);
break;
}
}
@ -135,7 +135,7 @@ static INLINE void loop_filter_block_plane_hor(
break;
case LF_PATH_SLOW:
av1_filter_block_plane_non420_hor(cm, &planes[plane], mi, mi_row,
mi_col);
mi_col, plane);
break;
}
}
@ -168,7 +168,7 @@ static int loop_filter_ver_row_worker(AV1LfSync *const lf_sync,
#if CONFIG_EXT_PARTITION_TYPES
for (plane = 0; plane < num_planes; ++plane)
av1_filter_block_plane_non420_ver(lf_data->cm, &lf_data->planes[plane],
mi + mi_col, mi_row, mi_col);
mi + mi_col, mi_row, mi_col, plane);
#else
for (plane = 0; plane < num_planes; ++plane)
@ -213,7 +213,7 @@ static int loop_filter_hor_row_worker(AV1LfSync *const lf_sync,
#if CONFIG_EXT_PARTITION_TYPES
for (plane = 0; plane < num_planes; ++plane)
av1_filter_block_plane_non420_hor(lf_data->cm, &lf_data->planes[plane],
mi + mi_col, mi_row, mi_col);
mi + mi_col, mi_row, mi_col, plane);
#else
for (plane = 0; plane < num_planes; ++plane)
loop_filter_block_plane_hor(lf_data->cm, lf_data->planes, plane,
@ -263,9 +263,9 @@ static int loop_filter_row_worker(AV1LfSync *const lf_sync,
#if CONFIG_EXT_PARTITION_TYPES
for (plane = 0; plane < num_planes; ++plane) {
av1_filter_block_plane_non420_ver(lf_data->cm, &lf_data->planes[plane],
mi + mi_col, mi_row, mi_col);
mi + mi_col, mi_row, mi_col, plane);
av1_filter_block_plane_non420_hor(lf_data->cm, &lf_data->planes[plane],
mi + mi_col, mi_row, mi_col);
mi + mi_col, mi_row, mi_col, plane);
}
#else
av1_setup_mask(lf_data->cm, mi_row, mi_col, mi + mi_col,

15
third_party/aom/av1/common/tile_common.c vendored
View file

@ -86,17 +86,28 @@ void av1_update_boundary_info(const struct AV1Common *cm,
for (col = mi_col; col < (mi_col + cm->mib_size); col++) {
MODE_INFO *const mi = cm->mi + row * cm->mi_stride + col;
mi->mbmi.boundary_info = 0;
// If horizontal dependent tile is enabled, then the horizontal
// tile boundary is not treated as real tile boundary for loop
// filtering, only the horizontal tile group boundary is treated
// as tile boundary.
// Otherwise, tile group boundary is treated the same as tile boundary.
// Loop filtering operation is done based on the
// loopfilter_across_tiles_enabled flag for both tile boundary and tile
// group boundary.
if (cm->tile_cols * cm->tile_rows > 1) {
#if CONFIG_DEPENDENT_HORZTILES
#if CONFIG_TILE_GROUPS
if (row == tile_info->mi_row_start &&
(!cm->dependent_horz_tiles || tile_info->tg_horz_boundary))
#if CONFIG_TILE_GROUPS
#else
if (row == tile_info->mi_row_start && !cm->dependent_horz_tiles)
if (row == tile_info->mi_row_start && !cm->dependent_horz_tiles)
#endif // CONFIG_TILE_GROUPS
#else
if (row == tile_info->mi_row_start)
#endif // CONFIG_DEPENDENT_HORZTILES
mi->mbmi.boundary_info |= TILE_ABOVE_BOUNDARY;
if (col == tile_info->mi_col_start)
mi->mbmi.boundary_info |= TILE_LEFT_BOUNDARY;

2
third_party/aom/av1/common/tile_common.h vendored
View file

@ -38,7 +38,7 @@ void av1_tile_init(TileInfo *tile, const struct AV1Common *cm, int row,
void av1_tile_set_row(TileInfo *tile, const struct AV1Common *cm, int row);
void av1_tile_set_col(TileInfo *tile, const struct AV1Common *cm, int col);
#if CONFIG_DEPENDENT_HORZTILES && CONFIG_TILE_GROUPS
void av1_tile_set_tg_boundary(TileInfo *tile, const struct AV1Common *cm,
void av1_tile_set_tg_boundary(TileInfo *tile, const struct AV1Common *const cm,
int row, int col);
#endif
void av1_get_tile_n_bits(int mi_cols, int *min_log2_tile_cols,

2
third_party/aom/av1/common/txb_common.c vendored
View file

@ -98,7 +98,7 @@ const int16_t av1_coeff_band_32x32[1024] = {
void av1_adapt_txb_probs(AV1_COMMON *cm, unsigned int count_sat,
unsigned int update_factor) {
FRAME_CONTEXT *fc = cm->fc;
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
const FRAME_CONTEXT *pre_fc = cm->pre_fc;
const FRAME_COUNTS *counts = &cm->counts;
TX_SIZE tx_size;
int plane, ctx, level;

290
third_party/aom/av1/common/txb_common.h vendored
View file

@ -32,48 +32,95 @@ static int base_ref_offset[BASE_CONTEXT_POSITION_NUM][2] = {
/* clang-format on*/
};
static INLINE int get_level_count(const tran_low_t *tcoeffs, int stride,
int row, int col, int level,
int (*nb_offset)[2], int nb_num) {
int count = 0;
for (int idx = 0; idx < nb_num; ++idx) {
const int ref_row = row + nb_offset[idx][0];
const int ref_col = col + nb_offset[idx][1];
const int pos = ref_row * stride + ref_col;
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
tran_low_t abs_coeff = abs(tcoeffs[pos]);
count += abs_coeff > level;
}
return count;
}
static INLINE void get_mag(int *mag, const tran_low_t *tcoeffs, int stride,
int row, int col, int (*nb_offset)[2], int nb_num) {
mag[0] = 0;
mag[1] = 0;
for (int idx = 0; idx < nb_num; ++idx) {
const int ref_row = row + nb_offset[idx][0];
const int ref_col = col + nb_offset[idx][1];
const int pos = ref_row * stride + ref_col;
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
tran_low_t abs_coeff = abs(tcoeffs[pos]);
if (nb_offset[idx][0] >= 0 && nb_offset[idx][1] >= 0) {
if (abs_coeff > mag[0]) {
mag[0] = abs_coeff;
mag[1] = 1;
} else if (abs_coeff == mag[0]) {
++mag[1];
}
}
}
}
static INLINE int get_level_count_mag(int *mag, const tran_low_t *tcoeffs,
int stride, int row, int col, int level,
int (*nb_offset)[2], int nb_num) {
int count = 0;
*mag = 0;
for (int idx = 0; idx < nb_num; ++idx) {
const int ref_row = row + nb_offset[idx][0];
const int ref_col = col + nb_offset[idx][1];
const int pos = ref_row * stride + ref_col;
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
tran_low_t abs_coeff = abs(tcoeffs[pos]);
count += abs_coeff > level;
if (nb_offset[idx][0] >= 0 && nb_offset[idx][1] >= 0)
*mag = AOMMAX(*mag, abs_coeff);
}
return count;
}
static INLINE int get_base_ctx_from_count_mag(int row, int col, int count,
int mag, int level) {
const int ctx = (count + 1) >> 1;
const int sig_mag = mag > level;
int ctx_idx = -1;
if (row == 0 && col == 0) {
ctx_idx = (ctx << 1) + sig_mag;
assert(ctx_idx < 8);
} else if (row == 0) {
ctx_idx = 8 + (ctx << 1) + sig_mag;
assert(ctx_idx < 18);
} else if (col == 0) {
ctx_idx = 8 + 10 + (ctx << 1) + sig_mag;
assert(ctx_idx < 28);
} else {
ctx_idx = 8 + 10 + 10 + (ctx << 1) + sig_mag;
assert(ctx_idx < COEFF_BASE_CONTEXTS);
}
return ctx_idx;
}
static INLINE int get_base_ctx(const tran_low_t *tcoeffs,
int c, // raster order
const int bwl, const int level) {
const int stride = 1 << bwl;
const int row = c >> bwl;
const int col = c - (row << bwl);
const int stride = 1 << bwl;
const int level_minus_1 = level - 1;
int ctx = 0;
int mag = 0;
int idx;
int ctx_idx = -1;
tran_low_t abs_coeff;
ctx = 0;
for (idx = 0; idx < BASE_CONTEXT_POSITION_NUM; ++idx) {
int ref_row = row + base_ref_offset[idx][0];
int ref_col = col + base_ref_offset[idx][1];
int pos = (ref_row << bwl) + ref_col;
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
abs_coeff = abs(tcoeffs[pos]);
ctx += abs_coeff > level_minus_1;
if (base_ref_offset[idx][0] >= 0 && base_ref_offset[idx][1] >= 0)
mag |= abs_coeff > level;
}
ctx = (ctx + 1) >> 1;
if (row == 0 && col == 0) {
ctx_idx = (ctx << 1) + mag;
assert(ctx_idx < 8);
} else if (row == 0) {
ctx_idx = 8 + (ctx << 1) + mag;
assert(ctx_idx < 18);
} else if (col == 0) {
ctx_idx = 8 + 10 + (ctx << 1) + mag;
assert(ctx_idx < 28);
} else {
ctx_idx = 8 + 10 + 10 + (ctx << 1) + mag;
assert(ctx_idx < COEFF_BASE_CONTEXTS);
}
int mag;
int count =
get_level_count_mag(&mag, tcoeffs, stride, row, col, level_minus_1,
base_ref_offset, BASE_CONTEXT_POSITION_NUM);
int ctx_idx = get_base_ctx_from_count_mag(row, col, count, mag, level);
return ctx_idx;
}
@ -89,34 +136,13 @@ static int br_level_map[9] = {
0, 0, 1, 1, 2, 2, 3, 3, 3,
};
static INLINE int get_level_ctx(const tran_low_t *tcoeffs,
const int c, // raster order
const int bwl) {
const int row = c >> bwl;
const int col = c - (row << bwl);
const int stride = 1 << bwl;
const int level_minus_1 = NUM_BASE_LEVELS;
int ctx = 0;
int idx;
tran_low_t abs_coeff;
int mag = 0, offset = 0;
for (idx = 0; idx < BR_CONTEXT_POSITION_NUM; ++idx) {
int ref_row = row + br_ref_offset[idx][0];
int ref_col = col + br_ref_offset[idx][1];
int pos = (ref_row << bwl) + ref_col;
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
abs_coeff = abs(tcoeffs[pos]);
ctx += abs_coeff > level_minus_1;
if (br_ref_offset[idx][0] >= 0 && br_ref_offset[idx][1] >= 0)
mag = AOMMAX(mag, abs_coeff);
}
if (mag <= 1)
#define BR_MAG_OFFSET 1
// TODO(angiebird): optimize this function by using a table to map from
// count/mag to ctx
static INLINE int get_br_ctx_from_count_mag(int row, int col, int count,
int mag) {
int offset = 0;
if (mag <= BR_MAG_OFFSET)
offset = 0;
else if (mag <= 3)
offset = 1;
@ -125,8 +151,7 @@ static INLINE int get_level_ctx(const tran_low_t *tcoeffs,
else
offset = 3;
ctx = br_level_map[ctx];
int ctx = br_level_map[count];
ctx += offset * BR_TMP_OFFSET;
// DC: 0 - 1
@ -142,17 +167,33 @@ static INLINE int get_level_ctx(const tran_low_t *tcoeffs,
return 8 + ctx;
}
static int sig_ref_offset[11][2] = {
static INLINE int get_br_ctx(const tran_low_t *tcoeffs,
const int c, // raster order
const int bwl) {
const int stride = 1 << bwl;
const int row = c >> bwl;
const int col = c - (row << bwl);
const int level_minus_1 = NUM_BASE_LEVELS;
int mag;
const int count =
get_level_count_mag(&mag, tcoeffs, stride, row, col, level_minus_1,
br_ref_offset, BR_CONTEXT_POSITION_NUM);
const int ctx = get_br_ctx_from_count_mag(row, col, count, mag);
return ctx;
}
#define SIG_REF_OFFSET_NUM 11
static int sig_ref_offset[SIG_REF_OFFSET_NUM][2] = {
{ -2, -1 }, { -2, 0 }, { -2, 1 }, { -1, -2 }, { -1, -1 }, { -1, 0 },
{ -1, 1 }, { 0, -2 }, { 0, -1 }, { 1, -2 }, { 1, -1 },
};
static INLINE int get_nz_map_ctx(const tran_low_t *tcoeffs,
const uint8_t *txb_mask,
const int c, // raster order
const int coeff_idx, // raster order
const int bwl) {
const int row = c >> bwl;
const int col = c - (row << bwl);
const int row = coeff_idx >> bwl;
const int col = coeff_idx - (row << bwl);
int ctx = 0;
int idx;
int stride = 1 << bwl;
@ -178,7 +219,7 @@ static INLINE int get_nz_map_ctx(const tran_low_t *tcoeffs,
return 5 + ctx;
}
for (idx = 0; idx < 11; ++idx) {
for (idx = 0; idx < SIG_REF_OFFSET_NUM; ++idx) {
int ref_row = row + sig_ref_offset[idx][0];
int ref_col = col + sig_ref_offset[idx][1];
int pos;
@ -212,14 +253,93 @@ static INLINE int get_nz_map_ctx(const tran_low_t *tcoeffs,
return 14 + ctx;
}
static INLINE int get_nz_count(const tran_low_t *tcoeffs, int stride, int row,
int col, const int16_t *iscan) {
int count = 0;
const int pos = row * stride + col;
for (int idx = 0; idx < SIG_REF_OFFSET_NUM; ++idx) {
const int ref_row = row + sig_ref_offset[idx][0];
const int ref_col = col + sig_ref_offset[idx][1];
if (ref_row < 0 || ref_col < 0 || ref_row >= stride || ref_col >= stride)
continue;
const int nb_pos = ref_row * stride + ref_col;
if (iscan[nb_pos] < iscan[pos]) count += (tcoeffs[nb_pos] != 0);
}
return count;
}
// TODO(angiebird): optimize this function by generate a table that maps from
// count to ctx
static INLINE int get_nz_map_ctx_from_count(int count,
const tran_low_t *tcoeffs,
int coeff_idx, // raster order
int bwl, const int16_t *iscan) {
const int row = coeff_idx >> bwl;
const int col = coeff_idx - (row << bwl);
int ctx = 0;
if (row == 0 && col == 0) return 0;
if (row == 0 && col == 1) return 1 + (tcoeffs[0] != 0);
if (row == 1 && col == 0) return 3 + (tcoeffs[0] != 0);
if (row == 1 && col == 1) {
int pos;
ctx = (tcoeffs[0] != 0);
if (iscan[1] < iscan[coeff_idx]) ctx += (tcoeffs[1] != 0);
pos = 1 << bwl;
if (iscan[pos] < iscan[coeff_idx]) ctx += (tcoeffs[pos] != 0);
ctx = (ctx + 1) >> 1;
assert(5 + ctx <= 7);
return 5 + ctx;
}
if (row == 0) {
ctx = (count + 1) >> 1;
assert(ctx < 3);
return 8 + ctx;
}
if (col == 0) {
ctx = (count + 1) >> 1;
assert(ctx < 3);
return 11 + ctx;
}
ctx = count >> 1;
assert(14 + ctx < 20);
return 14 + ctx;
}
// TODO(angiebird): merge this function with get_nz_map_ctx() after proper
// testing
static INLINE int get_nz_map_ctx2(const tran_low_t *tcoeffs,
const int coeff_idx, // raster order
const int bwl, const int16_t *iscan) {
int stride = 1 << bwl;
const int row = coeff_idx >> bwl;
const int col = coeff_idx - (row << bwl);
int count = get_nz_count(tcoeffs, stride, row, col, iscan);
return get_nz_map_ctx_from_count(count, tcoeffs, coeff_idx, bwl, iscan);
}
static INLINE int get_eob_ctx(const tran_low_t *tcoeffs,
const int c, // raster order
const int coeff_idx, // raster order
const int bwl) {
(void)tcoeffs;
if (bwl == 2) return av1_coeff_band_4x4[c];
if (bwl == 3) return av1_coeff_band_8x8[c];
if (bwl == 4) return av1_coeff_band_16x16[c];
if (bwl == 5) return av1_coeff_band_32x32[c];
if (bwl == 2) return av1_coeff_band_4x4[coeff_idx];
if (bwl == 3) return av1_coeff_band_8x8[coeff_idx];
if (bwl == 4) return av1_coeff_band_16x16[coeff_idx];
if (bwl == 5) return av1_coeff_band_32x32[coeff_idx];
assert(0);
return 0;
@ -245,14 +365,14 @@ static INLINE int get_dc_sign_ctx(int dc_sign) {
static INLINE void get_txb_ctx(BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
int plane, const ENTROPY_CONTEXT *a,
const ENTROPY_CONTEXT *l, TXB_CTX *txb_ctx) {
const int tx_size_in_blocks = 1 << tx_size;
const int txb_w_unit = tx_size_wide_unit[tx_size];
const int txb_h_unit = tx_size_high_unit[tx_size];
int ctx_offset = (plane == 0) ? 0 : 7;
int k;
if (plane_bsize > txsize_to_bsize[tx_size]) ctx_offset += 3;
int dc_sign = 0;
for (k = 0; k < tx_size_in_blocks; ++k) {
for (int k = 0; k < txb_w_unit; ++k) {
int sign = ((uint8_t)a[k]) >> COEFF_CONTEXT_BITS;
if (sign == 1)
--dc_sign;
@ -260,8 +380,10 @@ static INLINE void get_txb_ctx(BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
++dc_sign;
else if (sign != 0)
assert(0);
}
sign = ((uint8_t)l[k]) >> COEFF_CONTEXT_BITS;
for (int k = 0; k < txb_h_unit; ++k) {
int sign = ((uint8_t)l[k]) >> COEFF_CONTEXT_BITS;
if (sign == 1)
--dc_sign;
else if (sign == 2)
@ -269,15 +391,21 @@ static INLINE void get_txb_ctx(BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
else if (sign != 0)
assert(0);
}
txb_ctx->dc_sign_ctx = get_dc_sign_ctx(dc_sign);
if (plane == 0) {
int top = 0;
int left = 0;
for (k = 0; k < tx_size_in_blocks; ++k) {
for (int k = 0; k < txb_w_unit; ++k) {
top = AOMMAX(top, ((uint8_t)a[k] & COEFF_CONTEXT_MASK));
}
for (int k = 0; k < txb_h_unit; ++k) {
left = AOMMAX(left, ((uint8_t)l[k] & COEFF_CONTEXT_MASK));
}
top = AOMMIN(top, 255);
left = AOMMIN(left, 255);

809
third_party/aom/av1/common/warped_motion.c vendored
View file

File diff suppressed because it is too large Load diff

53
third_party/aom/av1/common/warped_motion.h vendored
View file

@ -35,59 +35,72 @@
extern const int16_t warped_filter[WARPEDPIXEL_PREC_SHIFTS * 3 + 1][8];
typedef void (*ProjectPointsFunc)(int32_t *mat, int *points, int *proj,
typedef void (*ProjectPointsFunc)(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj,
const int subsampling_x,
const int subsampling_y);
void project_points_translation(int32_t *mat, int *points, int *proj,
void project_points_translation(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points_rotzoom(int32_t *mat, int *points, int *proj, const int n,
const int stride_points, const int stride_proj,
const int subsampling_x, const int subsampling_y);
void project_points_rotzoom(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points_affine(int32_t *mat, int *points, int *proj, const int n,
const int stride_points, const int stride_proj,
const int subsampling_x, const int subsampling_y);
void project_points_affine(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points_hortrapezoid(int32_t *mat, int *points, int *proj,
void project_points_hortrapezoid(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points_vertrapezoid(int32_t *mat, int *points, int *proj,
void project_points_vertrapezoid(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points_homography(int32_t *mat, int *points, int *proj,
void project_points_homography(const int32_t *mat, int *points, int *proj,
const int n, const int stride_points,
const int stride_proj, const int subsampling_x,
const int subsampling_y);
void project_points(WarpedMotionParams *wm_params, int *points, int *proj,
void project_points(const WarpedMotionParams *wm_params, int *points, int *proj,
const int n, const int stride_points, const int stride_proj,
const int subsampling_x, const int subsampling_y);
double av1_warp_erroradv(WarpedMotionParams *wm,
// Returns the error between the result of applying motion 'wm' to the frame
// described by 'ref' and the frame described by 'dst'.
int64_t av1_warp_error(WarpedMotionParams *wm,
#if CONFIG_HIGHBITDEPTH
int use_hbd, int bd,
int use_hbd, int bd,
#endif // CONFIG_HIGHBITDEPTH
uint8_t *ref, int width, int height, int stride,
uint8_t *dst, int p_col, int p_row, int p_width,
int p_height, int p_stride, int subsampling_x,
int subsampling_y, int x_scale, int y_scale);
const uint8_t *ref, int width, int height, int stride,
uint8_t *dst, int p_col, int p_row, int p_width,
int p_height, int p_stride, int subsampling_x,
int subsampling_y, int x_scale, int y_scale);
// Returns the error between the frame described by 'ref' and the frame
// described by 'dst'.
int64_t av1_frame_error(
#if CONFIG_HIGHBITDEPTH
int use_hbd, int bd,
#endif // CONFIG_HIGHBITDEPTH
const uint8_t *ref, int stride, uint8_t *dst, int p_col, int p_row,
int p_width, int p_height, int p_stride);
void av1_warp_plane(WarpedMotionParams *wm,
#if CONFIG_HIGHBITDEPTH
int use_hbd, int bd,
#endif // CONFIG_HIGHBITDEPTH
uint8_t *ref, int width, int height, int stride,
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, int x_scale, int y_scale, int ref_frm);
int subsampling_y, int x_scale, int y_scale, int comp_avg);
int find_projection(int np, int *pts1, int *pts2, BLOCK_SIZE bsize, int mvy,
int mvx, WarpedMotionParams *wm_params, int mi_row,

15
third_party/aom/av1/common/x86/av1_convolve_ssse3.c vendored
View file

@ -19,13 +19,13 @@
#define WIDTH_BOUND (16)
#define HEIGHT_BOUND (16)
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
DECLARE_ALIGNED(16, static int8_t,
sub_pel_filters_12sharp_signal_dir[15][2][16]);
DECLARE_ALIGNED(16, static int8_t,
sub_pel_filters_12sharp_ver_signal_dir[15][6][16]);
#endif // CONFIG_DUAL_FILTER
#endif // CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
#if USE_TEMPORALFILTER_12TAP
DECLARE_ALIGNED(16, static int8_t,
@ -39,7 +39,7 @@ typedef int8_t (*SubpelFilterCoeffs)[16];
static INLINE SubpelFilterCoeffs
get_subpel_filter_signal_dir(const InterpFilterParams p, int index) {
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
if (p.interp_filter == MULTITAP_SHARP) {
return &sub_pel_filters_12sharp_signal_dir[index][0];
}
@ -56,7 +56,7 @@ get_subpel_filter_signal_dir(const InterpFilterParams p, int index) {
static INLINE SubpelFilterCoeffs
get_subpel_filter_ver_signal_dir(const InterpFilterParams p, int index) {
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
if (p.interp_filter == MULTITAP_SHARP) {
return &sub_pel_filters_12sharp_ver_signal_dir[index][0];
}
@ -143,6 +143,7 @@ static void horiz_w4_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
const __m128i k_256 = _mm_set1_epi16(1 << 8);
const __m128i zero = _mm_setzero_si128();
assert(tapsNum == 10 || tapsNum == 12);
if (10 == tapsNum) {
src -= 1;
}
@ -470,6 +471,7 @@ static void filter_horiz_v8p_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
__m128i min_x2x3, max_x2x3;
__m128i temp;
assert(tapsNum == 10 || tapsNum == 12);
if (tapsNum == 10) {
src_ptr -= 1;
}
@ -612,6 +614,7 @@ static void filter_horiz_v4p_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
__m128i x0, x1, x2, x3, x4, x5;
__m128i min_x2x3, max_x2x3, temp;
assert(tapsNum == 10 || tapsNum == 12);
if (tapsNum == 10) {
src_ptr -= 1;
}
@ -982,7 +985,7 @@ typedef struct SimdFilter {
int8_t (*simd_vert_filter)[6][16];
} SimdFilter;
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
#define MULTITAP_FILTER_NUM 1
SimdFilter simd_filters[MULTITAP_FILTER_NUM] = {
{ MULTITAP_SHARP, &sub_pel_filters_12sharp_signal_dir[0],
@ -1010,7 +1013,7 @@ void av1_lowbd_convolve_init_ssse3(void) {
temporal_simd_filter.simd_vert_filter);
}
#endif
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
{
int i;
for (i = 0; i < MULTITAP_FILTER_NUM; ++i) {

28
third_party/aom/av1/common/x86/av1_fwd_txfm1d_sse4.c vendored
View file

@ -57,7 +57,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 2
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = _mm_add_epi32(buf1[0], buf1[15]);
buf0[15] = _mm_sub_epi32(buf1[0], buf1[15]);
buf0[1] = _mm_add_epi32(buf1[1], buf1[14]);
@ -94,7 +94,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 3
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf1[0] = _mm_add_epi32(buf0[0], buf0[7]);
buf1[7] = _mm_sub_epi32(buf0[0], buf0[7]);
buf1[1] = _mm_add_epi32(buf0[1], buf0[6]);
@ -131,7 +131,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 4
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = _mm_add_epi32(buf1[0], buf1[3]);
buf0[3] = _mm_sub_epi32(buf1[0], buf1[3]);
buf0[1] = _mm_add_epi32(buf1[1], buf1[2]);
@ -168,7 +168,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 5
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
btf_32_sse4_1_type0(cospi[32], cospi[32], buf0[0], buf0[1], buf1[0],
buf1[1], bit);
btf_32_sse4_1_type1(cospi[48], cospi[16], buf0[2], buf0[3], buf1[2],
@ -205,7 +205,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 6
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
buf0[2] = buf1[2];
@ -242,7 +242,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 7
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf1[0] = buf0[0];
buf1[1] = buf0[1];
buf1[2] = buf0[2];
@ -279,7 +279,7 @@ void av1_fdct32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 8
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
buf0[2] = buf1[2];
@ -383,7 +383,7 @@ void av1_fadst4_new_sse4_1(const __m128i *input, __m128i *output,
// stage 2
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
btf_32_sse4_1_type0(cospi[8], cospi[56], buf1[0], buf1[1], buf0[0], buf0[1],
bit);
btf_32_sse4_1_type0(cospi[40], cospi[24], buf1[2], buf1[3], buf0[2],
@ -399,7 +399,7 @@ void av1_fadst4_new_sse4_1(const __m128i *input, __m128i *output,
// stage 4
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
btf_32_sse4_1_type0(cospi[32], cospi[32], buf1[2], buf1[3], buf0[2],
@ -475,7 +475,7 @@ void av1_fadst32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 2
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
btf_32_sse4_1_type0(cospi[1], cospi[63], buf1[0], buf1[1], buf0[0], buf0[1],
bit);
btf_32_sse4_1_type0(cospi[5], cospi[59], buf1[2], buf1[3], buf0[2], buf0[3],
@ -547,7 +547,7 @@ void av1_fadst32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 4
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
buf0[2] = buf1[2];
@ -619,7 +619,7 @@ void av1_fadst32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 6
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
buf0[2] = buf1[2];
@ -691,7 +691,7 @@ void av1_fadst32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 8
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
buf0[2] = buf1[2];
@ -763,7 +763,7 @@ void av1_fadst32_new_sse4_1(const __m128i *input, __m128i *output,
// stage 10
stage_idx++;
bit = cos_bit[stage_idx];
cospi = cospi_arr[bit - cos_bit_min];
cospi = cospi_arr(bit);
buf0[0] = buf1[0];
buf0[1] = buf1[1];
btf_32_sse4_1_type0(cospi[32], cospi[32], buf1[2], buf1[3], buf0[2],

26
third_party/aom/av1/common/x86/av1_fwd_txfm2d_sse4.c vendored
View file

@ -37,16 +37,20 @@ static INLINE TxfmFuncSSE2 fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
}
static INLINE void fwd_txfm2d_sse4_1(const int16_t *input, int32_t *output,
const int stride, const TXFM_2D_CFG *cfg,
const int stride,
const TXFM_2D_FLIP_CFG *cfg,
int32_t *txfm_buf) {
const int txfm_size = cfg->txfm_size;
const int8_t *shift = cfg->shift;
const int8_t *stage_range_col = cfg->stage_range_col;
const int8_t *stage_range_row = cfg->stage_range_row;
const int8_t *cos_bit_col = cfg->cos_bit_col;
const int8_t *cos_bit_row = cfg->cos_bit_row;
const TxfmFuncSSE2 txfm_func_col = fwd_txfm_type_to_func(cfg->txfm_type_col);
const TxfmFuncSSE2 txfm_func_row = fwd_txfm_type_to_func(cfg->txfm_type_row);
// TODO(sarahparker) must correct for rectangular transforms in follow up
const int txfm_size = cfg->row_cfg->txfm_size;
const int8_t *shift = cfg->row_cfg->shift;
const int8_t *stage_range_col = cfg->col_cfg->stage_range;
const int8_t *stage_range_row = cfg->row_cfg->stage_range;
const int8_t *cos_bit_col = cfg->col_cfg->cos_bit;
const int8_t *cos_bit_row = cfg->row_cfg->cos_bit;
const TxfmFuncSSE2 txfm_func_col =
fwd_txfm_type_to_func(cfg->col_cfg->txfm_type);
const TxfmFuncSSE2 txfm_func_row =
fwd_txfm_type_to_func(cfg->row_cfg->txfm_type);
__m128i *buf_128 = (__m128i *)txfm_buf;
__m128i *out_128 = (__m128i *)output;
@ -69,7 +73,7 @@ void av1_fwd_txfm2d_32x32_sse4_1(const int16_t *input, int32_t *output,
DECLARE_ALIGNED(16, int32_t, txfm_buf[1024]);
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_32X32);
(void)bd;
fwd_txfm2d_sse4_1(input, output, stride, cfg.cfg, txfm_buf);
fwd_txfm2d_sse4_1(input, output, stride, &cfg, txfm_buf);
}
void av1_fwd_txfm2d_64x64_sse4_1(const int16_t *input, int32_t *output,
@ -77,5 +81,5 @@ void av1_fwd_txfm2d_64x64_sse4_1(const int16_t *input, int32_t *output,
DECLARE_ALIGNED(16, int32_t, txfm_buf[4096]);
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_64x64_cfg(tx_type);
(void)bd;
fwd_txfm2d_sse4_1(input, output, stride, cfg.cfg, txfm_buf);
fwd_txfm2d_sse4_1(input, output, stride, &cfg, txfm_buf);
}

8
third_party/aom/av1/common/x86/av1_highbd_convolve_sse4.c vendored
View file

@ -15,7 +15,7 @@
#include "./av1_rtcd.h"
#include "av1/common/filter.h"
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
DECLARE_ALIGNED(16, static int16_t, subpel_filters_sharp[15][6][8]);
#endif
@ -31,7 +31,7 @@ typedef void (*TransposeSave)(int width, int pixelsNum, uint32_t *src,
static INLINE HbdSubpelFilterCoeffs
hbd_get_subpel_filter_ver_signal_dir(const InterpFilterParams p, int index) {
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
if (p.interp_filter == MULTITAP_SHARP) {
return &subpel_filters_sharp[index][0];
}
@ -76,7 +76,7 @@ void av1_highbd_convolve_init_sse4_1(void) {
init_simd_filter(filter_ptr, taps, subpel_temporalfilter);
}
#endif
#if CONFIG_DUAL_FILTER
#if CONFIG_DUAL_FILTER && USE_EXTRA_FILTER
{
InterpFilterParams filter_params =
av1_get_interp_filter_params(MULTITAP_SHARP);
@ -246,6 +246,7 @@ static void highbd_filter_horiz(const uint16_t *src, int src_stride, __m128i *f,
int tapsNum, uint32_t *buf) {
__m128i u[8], v[6];
assert(tapsNum == 10 || tapsNum == 12);
if (tapsNum == 10) {
src -= 1;
}
@ -412,6 +413,7 @@ static void filter_vert_horiz_parallel(const uint16_t *src, int src_stride,
int r = 0;
// TODO(luoyi) treat s[12] as a circular buffer in width = 2 case
assert(taps == 10 || taps == 12);
if (10 == taps) {
i += 1;
s[0] = zero;

222
third_party/aom/av1/common/x86/highbd_inv_txfm_avx2.c vendored
View file

@ -13,7 +13,7 @@
#include "./av1_rtcd.h"
#include "./aom_config.h"
#include "av1/common/av1_inv_txfm2d_cfg.h"
#include "av1/common/av1_inv_txfm1d_cfg.h"
// Note:
// Total 32x4 registers to represent 32x32 block coefficients.
@ -154,20 +154,21 @@ static void write_buffer_32x32(__m256i *in, uint16_t *output, int stride,
}
}
static INLINE __m256i half_btf_avx2(__m256i w0, __m256i n0, __m256i w1,
__m256i n1, __m256i rounding, int bit) {
static INLINE __m256i half_btf_avx2(const __m256i *w0, const __m256i *n0,
const __m256i *w1, const __m256i *n1,
const __m256i *rounding, int bit) {
__m256i x, y;
x = _mm256_mullo_epi32(w0, n0);
y = _mm256_mullo_epi32(w1, n1);
x = _mm256_mullo_epi32(*w0, *n0);
y = _mm256_mullo_epi32(*w1, *n1);
x = _mm256_add_epi32(x, y);
x = _mm256_add_epi32(x, rounding);
x = _mm256_add_epi32(x, *rounding);
x = _mm256_srai_epi32(x, bit);
return x;
}
static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m256i cospi62 = _mm256_set1_epi32(cospi[62]);
const __m256i cospi30 = _mm256_set1_epi32(cospi[30]);
const __m256i cospi46 = _mm256_set1_epi32(cospi[46]);
@ -275,22 +276,38 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[13] = bf1[13];
bf0[14] = bf1[14];
bf0[15] = bf1[15];
bf0[16] = half_btf_avx2(cospi62, bf1[16], cospim2, bf1[31], rounding, bit);
bf0[17] = half_btf_avx2(cospi30, bf1[17], cospim34, bf1[30], rounding, bit);
bf0[18] = half_btf_avx2(cospi46, bf1[18], cospim18, bf1[29], rounding, bit);
bf0[19] = half_btf_avx2(cospi14, bf1[19], cospim50, bf1[28], rounding, bit);
bf0[20] = half_btf_avx2(cospi54, bf1[20], cospim10, bf1[27], rounding, bit);
bf0[21] = half_btf_avx2(cospi22, bf1[21], cospim42, bf1[26], rounding, bit);
bf0[22] = half_btf_avx2(cospi38, bf1[22], cospim26, bf1[25], rounding, bit);
bf0[23] = half_btf_avx2(cospi6, bf1[23], cospim58, bf1[24], rounding, bit);
bf0[24] = half_btf_avx2(cospi58, bf1[23], cospi6, bf1[24], rounding, bit);
bf0[25] = half_btf_avx2(cospi26, bf1[22], cospi38, bf1[25], rounding, bit);
bf0[26] = half_btf_avx2(cospi42, bf1[21], cospi22, bf1[26], rounding, bit);
bf0[27] = half_btf_avx2(cospi10, bf1[20], cospi54, bf1[27], rounding, bit);
bf0[28] = half_btf_avx2(cospi50, bf1[19], cospi14, bf1[28], rounding, bit);
bf0[29] = half_btf_avx2(cospi18, bf1[18], cospi46, bf1[29], rounding, bit);
bf0[30] = half_btf_avx2(cospi34, bf1[17], cospi30, bf1[30], rounding, bit);
bf0[31] = half_btf_avx2(cospi2, bf1[16], cospi62, bf1[31], rounding, bit);
bf0[16] =
half_btf_avx2(&cospi62, &bf1[16], &cospim2, &bf1[31], &rounding, bit);
bf0[17] =
half_btf_avx2(&cospi30, &bf1[17], &cospim34, &bf1[30], &rounding, bit);
bf0[18] =
half_btf_avx2(&cospi46, &bf1[18], &cospim18, &bf1[29], &rounding, bit);
bf0[19] =
half_btf_avx2(&cospi14, &bf1[19], &cospim50, &bf1[28], &rounding, bit);
bf0[20] =
half_btf_avx2(&cospi54, &bf1[20], &cospim10, &bf1[27], &rounding, bit);
bf0[21] =
half_btf_avx2(&cospi22, &bf1[21], &cospim42, &bf1[26], &rounding, bit);
bf0[22] =
half_btf_avx2(&cospi38, &bf1[22], &cospim26, &bf1[25], &rounding, bit);
bf0[23] =
half_btf_avx2(&cospi6, &bf1[23], &cospim58, &bf1[24], &rounding, bit);
bf0[24] =
half_btf_avx2(&cospi58, &bf1[23], &cospi6, &bf1[24], &rounding, bit);
bf0[25] =
half_btf_avx2(&cospi26, &bf1[22], &cospi38, &bf1[25], &rounding, bit);
bf0[26] =
half_btf_avx2(&cospi42, &bf1[21], &cospi22, &bf1[26], &rounding, bit);
bf0[27] =
half_btf_avx2(&cospi10, &bf1[20], &cospi54, &bf1[27], &rounding, bit);
bf0[28] =
half_btf_avx2(&cospi50, &bf1[19], &cospi14, &bf1[28], &rounding, bit);
bf0[29] =
half_btf_avx2(&cospi18, &bf1[18], &cospi46, &bf1[29], &rounding, bit);
bf0[30] =
half_btf_avx2(&cospi34, &bf1[17], &cospi30, &bf1[30], &rounding, bit);
bf0[31] =
half_btf_avx2(&cospi2, &bf1[16], &cospi62, &bf1[31], &rounding, bit);
// stage 3
bf1[0] = bf0[0];
@ -301,14 +318,22 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf1[5] = bf0[5];
bf1[6] = bf0[6];
bf1[7] = bf0[7];
bf1[8] = half_btf_avx2(cospi60, bf0[8], cospim4, bf0[15], rounding, bit);
bf1[9] = half_btf_avx2(cospi28, bf0[9], cospim36, bf0[14], rounding, bit);
bf1[10] = half_btf_avx2(cospi44, bf0[10], cospim20, bf0[13], rounding, bit);
bf1[11] = half_btf_avx2(cospi12, bf0[11], cospim52, bf0[12], rounding, bit);
bf1[12] = half_btf_avx2(cospi52, bf0[11], cospi12, bf0[12], rounding, bit);
bf1[13] = half_btf_avx2(cospi20, bf0[10], cospi44, bf0[13], rounding, bit);
bf1[14] = half_btf_avx2(cospi36, bf0[9], cospi28, bf0[14], rounding, bit);
bf1[15] = half_btf_avx2(cospi4, bf0[8], cospi60, bf0[15], rounding, bit);
bf1[8] =
half_btf_avx2(&cospi60, &bf0[8], &cospim4, &bf0[15], &rounding, bit);
bf1[9] =
half_btf_avx2(&cospi28, &bf0[9], &cospim36, &bf0[14], &rounding, bit);
bf1[10] =
half_btf_avx2(&cospi44, &bf0[10], &cospim20, &bf0[13], &rounding, bit);
bf1[11] =
half_btf_avx2(&cospi12, &bf0[11], &cospim52, &bf0[12], &rounding, bit);
bf1[12] =
half_btf_avx2(&cospi52, &bf0[11], &cospi12, &bf0[12], &rounding, bit);
bf1[13] =
half_btf_avx2(&cospi20, &bf0[10], &cospi44, &bf0[13], &rounding, bit);
bf1[14] =
half_btf_avx2(&cospi36, &bf0[9], &cospi28, &bf0[14], &rounding, bit);
bf1[15] =
half_btf_avx2(&cospi4, &bf0[8], &cospi60, &bf0[15], &rounding, bit);
bf1[16] = _mm256_add_epi32(bf0[16], bf0[17]);
bf1[17] = _mm256_sub_epi32(bf0[16], bf0[17]);
bf1[18] = _mm256_sub_epi32(bf0[19], bf0[18]);
@ -331,10 +356,13 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[1] = bf1[1];
bf0[2] = bf1[2];
bf0[3] = bf1[3];
bf0[4] = half_btf_avx2(cospi56, bf1[4], cospim8, bf1[7], rounding, bit);
bf0[5] = half_btf_avx2(cospi24, bf1[5], cospim40, bf1[6], rounding, bit);
bf0[6] = half_btf_avx2(cospi40, bf1[5], cospi24, bf1[6], rounding, bit);
bf0[7] = half_btf_avx2(cospi8, bf1[4], cospi56, bf1[7], rounding, bit);
bf0[4] =
half_btf_avx2(&cospi56, &bf1[4], &cospim8, &bf1[7], &rounding, bit);
bf0[5] =
half_btf_avx2(&cospi24, &bf1[5], &cospim40, &bf1[6], &rounding, bit);
bf0[6] =
half_btf_avx2(&cospi40, &bf1[5], &cospi24, &bf1[6], &rounding, bit);
bf0[7] = half_btf_avx2(&cospi8, &bf1[4], &cospi56, &bf1[7], &rounding, bit);
bf0[8] = _mm256_add_epi32(bf1[8], bf1[9]);
bf0[9] = _mm256_sub_epi32(bf1[8], bf1[9]);
bf0[10] = _mm256_sub_epi32(bf1[11], bf1[10]);
@ -344,40 +372,54 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[14] = _mm256_sub_epi32(bf1[15], bf1[14]);
bf0[15] = _mm256_add_epi32(bf1[14], bf1[15]);
bf0[16] = bf1[16];
bf0[17] = half_btf_avx2(cospim8, bf1[17], cospi56, bf1[30], rounding, bit);
bf0[18] = half_btf_avx2(cospim56, bf1[18], cospim8, bf1[29], rounding, bit);
bf0[17] =
half_btf_avx2(&cospim8, &bf1[17], &cospi56, &bf1[30], &rounding, bit);
bf0[18] =
half_btf_avx2(&cospim56, &bf1[18], &cospim8, &bf1[29], &rounding, bit);
bf0[19] = bf1[19];
bf0[20] = bf1[20];
bf0[21] = half_btf_avx2(cospim40, bf1[21], cospi24, bf1[26], rounding, bit);
bf0[21] =
half_btf_avx2(&cospim40, &bf1[21], &cospi24, &bf1[26], &rounding, bit);
bf0[22] =
half_btf_avx2(cospim24, bf1[22], cospim40, bf1[25], rounding, bit);
half_btf_avx2(&cospim24, &bf1[22], &cospim40, &bf1[25], &rounding, bit);
bf0[23] = bf1[23];
bf0[24] = bf1[24];
bf0[25] = half_btf_avx2(cospim40, bf1[22], cospi24, bf1[25], rounding, bit);
bf0[26] = half_btf_avx2(cospi24, bf1[21], cospi40, bf1[26], rounding, bit);
bf0[25] =
half_btf_avx2(&cospim40, &bf1[22], &cospi24, &bf1[25], &rounding, bit);
bf0[26] =
half_btf_avx2(&cospi24, &bf1[21], &cospi40, &bf1[26], &rounding, bit);
bf0[27] = bf1[27];
bf0[28] = bf1[28];
bf0[29] = half_btf_avx2(cospim8, bf1[18], cospi56, bf1[29], rounding, bit);
bf0[30] = half_btf_avx2(cospi56, bf1[17], cospi8, bf1[30], rounding, bit);
bf0[29] =
half_btf_avx2(&cospim8, &bf1[18], &cospi56, &bf1[29], &rounding, bit);
bf0[30] =
half_btf_avx2(&cospi56, &bf1[17], &cospi8, &bf1[30], &rounding, bit);
bf0[31] = bf1[31];
// stage 5
bf1[0] = half_btf_avx2(cospi32, bf0[0], cospi32, bf0[1], rounding, bit);
bf1[1] = half_btf_avx2(cospi32, bf0[0], cospim32, bf0[1], rounding, bit);
bf1[2] = half_btf_avx2(cospi48, bf0[2], cospim16, bf0[3], rounding, bit);
bf1[3] = half_btf_avx2(cospi16, bf0[2], cospi48, bf0[3], rounding, bit);
bf1[0] =
half_btf_avx2(&cospi32, &bf0[0], &cospi32, &bf0[1], &rounding, bit);
bf1[1] =
half_btf_avx2(&cospi32, &bf0[0], &cospim32, &bf0[1], &rounding, bit);
bf1[2] =
half_btf_avx2(&cospi48, &bf0[2], &cospim16, &bf0[3], &rounding, bit);
bf1[3] =
half_btf_avx2(&cospi16, &bf0[2], &cospi48, &bf0[3], &rounding, bit);
bf1[4] = _mm256_add_epi32(bf0[4], bf0[5]);
bf1[5] = _mm256_sub_epi32(bf0[4], bf0[5]);
bf1[6] = _mm256_sub_epi32(bf0[7], bf0[6]);
bf1[7] = _mm256_add_epi32(bf0[6], bf0[7]);
bf1[8] = bf0[8];
bf1[9] = half_btf_avx2(cospim16, bf0[9], cospi48, bf0[14], rounding, bit);
bf1[9] =
half_btf_avx2(&cospim16, &bf0[9], &cospi48, &bf0[14], &rounding, bit);
bf1[10] =
half_btf_avx2(cospim48, bf0[10], cospim16, bf0[13], rounding, bit);
half_btf_avx2(&cospim48, &bf0[10], &cospim16, &bf0[13], &rounding, bit);
bf1[11] = bf0[11];
bf1[12] = bf0[12];
bf1[13] = half_btf_avx2(cospim16, bf0[10], cospi48, bf0[13], rounding, bit);
bf1[14] = half_btf_avx2(cospi48, bf0[9], cospi16, bf0[14], rounding, bit);
bf1[13] =
half_btf_avx2(&cospim16, &bf0[10], &cospi48, &bf0[13], &rounding, bit);
bf1[14] =
half_btf_avx2(&cospi48, &bf0[9], &cospi16, &bf0[14], &rounding, bit);
bf1[15] = bf0[15];
bf1[16] = _mm256_add_epi32(bf0[16], bf0[19]);
bf1[17] = _mm256_add_epi32(bf0[17], bf0[18]);
@ -402,8 +444,10 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[2] = _mm256_sub_epi32(bf1[1], bf1[2]);
bf0[3] = _mm256_sub_epi32(bf1[0], bf1[3]);
bf0[4] = bf1[4];
bf0[5] = half_btf_avx2(cospim32, bf1[5], cospi32, bf1[6], rounding, bit);
bf0[6] = half_btf_avx2(cospi32, bf1[5], cospi32, bf1[6], rounding, bit);
bf0[5] =
half_btf_avx2(&cospim32, &bf1[5], &cospi32, &bf1[6], &rounding, bit);
bf0[6] =
half_btf_avx2(&cospi32, &bf1[5], &cospi32, &bf1[6], &rounding, bit);
bf0[7] = bf1[7];
bf0[8] = _mm256_add_epi32(bf1[8], bf1[11]);
bf0[9] = _mm256_add_epi32(bf1[9], bf1[10]);
@ -415,20 +459,26 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[15] = _mm256_add_epi32(bf1[12], bf1[15]);
bf0[16] = bf1[16];
bf0[17] = bf1[17];
bf0[18] = half_btf_avx2(cospim16, bf1[18], cospi48, bf1[29], rounding, bit);
bf0[19] = half_btf_avx2(cospim16, bf1[19], cospi48, bf1[28], rounding, bit);
bf0[18] =
half_btf_avx2(&cospim16, &bf1[18], &cospi48, &bf1[29], &rounding, bit);
bf0[19] =
half_btf_avx2(&cospim16, &bf1[19], &cospi48, &bf1[28], &rounding, bit);
bf0[20] =
half_btf_avx2(cospim48, bf1[20], cospim16, bf1[27], rounding, bit);
half_btf_avx2(&cospim48, &bf1[20], &cospim16, &bf1[27], &rounding, bit);
bf0[21] =
half_btf_avx2(cospim48, bf1[21], cospim16, bf1[26], rounding, bit);
half_btf_avx2(&cospim48, &bf1[21], &cospim16, &bf1[26], &rounding, bit);
bf0[22] = bf1[22];
bf0[23] = bf1[23];
bf0[24] = bf1[24];
bf0[25] = bf1[25];
bf0[26] = half_btf_avx2(cospim16, bf1[21], cospi48, bf1[26], rounding, bit);
bf0[27] = half_btf_avx2(cospim16, bf1[20], cospi48, bf1[27], rounding, bit);
bf0[28] = half_btf_avx2(cospi48, bf1[19], cospi16, bf1[28], rounding, bit);
bf0[29] = half_btf_avx2(cospi48, bf1[18], cospi16, bf1[29], rounding, bit);
bf0[26] =
half_btf_avx2(&cospim16, &bf1[21], &cospi48, &bf1[26], &rounding, bit);
bf0[27] =
half_btf_avx2(&cospim16, &bf1[20], &cospi48, &bf1[27], &rounding, bit);
bf0[28] =
half_btf_avx2(&cospi48, &bf1[19], &cospi16, &bf1[28], &rounding, bit);
bf0[29] =
half_btf_avx2(&cospi48, &bf1[18], &cospi16, &bf1[29], &rounding, bit);
bf0[30] = bf1[30];
bf0[31] = bf1[31];
@ -443,10 +493,14 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf1[7] = _mm256_sub_epi32(bf0[0], bf0[7]);
bf1[8] = bf0[8];
bf1[9] = bf0[9];
bf1[10] = half_btf_avx2(cospim32, bf0[10], cospi32, bf0[13], rounding, bit);
bf1[11] = half_btf_avx2(cospim32, bf0[11], cospi32, bf0[12], rounding, bit);
bf1[12] = half_btf_avx2(cospi32, bf0[11], cospi32, bf0[12], rounding, bit);
bf1[13] = half_btf_avx2(cospi32, bf0[10], cospi32, bf0[13], rounding, bit);
bf1[10] =
half_btf_avx2(&cospim32, &bf0[10], &cospi32, &bf0[13], &rounding, bit);
bf1[11] =
half_btf_avx2(&cospim32, &bf0[11], &cospi32, &bf0[12], &rounding, bit);
bf1[12] =
half_btf_avx2(&cospi32, &bf0[11], &cospi32, &bf0[12], &rounding, bit);
bf1[13] =
half_btf_avx2(&cospi32, &bf0[10], &cospi32, &bf0[13], &rounding, bit);
bf1[14] = bf0[14];
bf1[15] = bf0[15];
bf1[16] = _mm256_add_epi32(bf0[16], bf0[23]);
@ -487,14 +541,22 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
bf0[17] = bf1[17];
bf0[18] = bf1[18];
bf0[19] = bf1[19];
bf0[20] = half_btf_avx2(cospim32, bf1[20], cospi32, bf1[27], rounding, bit);
bf0[21] = half_btf_avx2(cospim32, bf1[21], cospi32, bf1[26], rounding, bit);
bf0[22] = half_btf_avx2(cospim32, bf1[22], cospi32, bf1[25], rounding, bit);
bf0[23] = half_btf_avx2(cospim32, bf1[23], cospi32, bf1[24], rounding, bit);
bf0[24] = half_btf_avx2(cospi32, bf1[23], cospi32, bf1[24], rounding, bit);
bf0[25] = half_btf_avx2(cospi32, bf1[22], cospi32, bf1[25], rounding, bit);
bf0[26] = half_btf_avx2(cospi32, bf1[21], cospi32, bf1[26], rounding, bit);
bf0[27] = half_btf_avx2(cospi32, bf1[20], cospi32, bf1[27], rounding, bit);
bf0[20] =
half_btf_avx2(&cospim32, &bf1[20], &cospi32, &bf1[27], &rounding, bit);
bf0[21] =
half_btf_avx2(&cospim32, &bf1[21], &cospi32, &bf1[26], &rounding, bit);
bf0[22] =
half_btf_avx2(&cospim32, &bf1[22], &cospi32, &bf1[25], &rounding, bit);
bf0[23] =
half_btf_avx2(&cospim32, &bf1[23], &cospi32, &bf1[24], &rounding, bit);
bf0[24] =
half_btf_avx2(&cospi32, &bf1[23], &cospi32, &bf1[24], &rounding, bit);
bf0[25] =
half_btf_avx2(&cospi32, &bf1[22], &cospi32, &bf1[25], &rounding, bit);
bf0[26] =
half_btf_avx2(&cospi32, &bf1[21], &cospi32, &bf1[26], &rounding, bit);
bf0[27] =
half_btf_avx2(&cospi32, &bf1[20], &cospi32, &bf1[27], &rounding, bit);
bf0[28] = bf1[28];
bf0[29] = bf1[29];
bf0[30] = bf1[30];
@ -539,18 +601,20 @@ static void idct32_avx2(__m256i *in, __m256i *out, int bit) {
void av1_inv_txfm2d_add_32x32_avx2(const int32_t *coeff, uint16_t *output,
int stride, int tx_type, int bd) {
__m256i in[128], out[128];
const TXFM_2D_CFG *cfg = NULL;
const TXFM_1D_CFG *row_cfg = NULL;
const TXFM_1D_CFG *col_cfg = NULL;
switch (tx_type) {
case DCT_DCT:
cfg = &inv_txfm_2d_cfg_dct_dct_32;
row_cfg = &inv_txfm_1d_row_cfg_dct_32;
col_cfg = &inv_txfm_1d_col_cfg_dct_32;
load_buffer_32x32(coeff, in);
transpose_32x32(in, out);
idct32_avx2(out, in, cfg->cos_bit_row[2]);
round_shift_32x32(in, -cfg->shift[0]);
idct32_avx2(out, in, row_cfg->cos_bit[2]);
round_shift_32x32(in, -row_cfg->shift[0]);
transpose_32x32(in, out);
idct32_avx2(out, in, cfg->cos_bit_col[2]);
write_buffer_32x32(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct32_avx2(out, in, col_cfg->cos_bit[2]);
write_buffer_32x32(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
default: assert(0);
}

384
third_party/aom/av1/common/x86/highbd_inv_txfm_sse4.c vendored
View file

@ -13,7 +13,7 @@
#include "./av1_rtcd.h"
#include "./aom_config.h"
#include "av1/common/av1_inv_txfm2d_cfg.h"
#include "av1/common/av1_inv_txfm1d_cfg.h"
#include "av1/common/x86/highbd_txfm_utility_sse4.h"
static INLINE void load_buffer_4x4(const int32_t *coeff, __m128i *in) {
@ -24,7 +24,7 @@ static INLINE void load_buffer_4x4(const int32_t *coeff, __m128i *in) {
}
static void idct4x4_sse4_1(__m128i *in, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
@ -72,7 +72,7 @@ static void idct4x4_sse4_1(__m128i *in, int bit) {
}
static void iadst4x4_sse4_1(__m128i *in, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
const __m128i cospim8 = _mm_set1_epi32(-cospi[8]);
@ -232,72 +232,82 @@ static void write_buffer_4x4(__m128i *in, uint16_t *output, int stride,
void av1_inv_txfm2d_add_4x4_sse4_1(const int32_t *coeff, uint16_t *output,
int stride, int tx_type, int bd) {
__m128i in[4];
const TXFM_2D_CFG *cfg = NULL;
const TXFM_1D_CFG *row_cfg = NULL;
const TXFM_1D_CFG *col_cfg = NULL;
switch (tx_type) {
case DCT_DCT:
cfg = &inv_txfm_2d_cfg_dct_dct_4;
row_cfg = &inv_txfm_1d_row_cfg_dct_4;
col_cfg = &inv_txfm_1d_col_cfg_dct_4;
load_buffer_4x4(coeff, in);
idct4x4_sse4_1(in, cfg->cos_bit_row[2]);
idct4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct4x4_sse4_1(in, row_cfg->cos_bit[2]);
idct4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_4;
row_cfg = &inv_txfm_1d_row_cfg_dct_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
idct4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case DCT_ADST:
cfg = &inv_txfm_2d_cfg_dct_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_dct_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
idct4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
idct4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_4;
row_cfg = &inv_txfm_1d_row_cfg_dct_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
idct4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 1, -cfg->shift[1], bd);
idct4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
case DCT_FLIPADST:
cfg = &inv_txfm_2d_cfg_dct_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_dct_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
idct4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 1, 0, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
idct4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case FLIPADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 1, 1, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 1, 1, -row_cfg->shift[1], bd);
break;
case ADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 1, 0, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case FLIPADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_4;
row_cfg = &inv_txfm_1d_row_cfg_adst_4;
col_cfg = &inv_txfm_1d_col_cfg_adst_4;
load_buffer_4x4(coeff, in);
iadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
iadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
write_buffer_4x4(in, output, stride, 0, 1, -cfg->shift[1], bd);
iadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
iadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
write_buffer_4x4(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
#endif // CONFIG_EXT_TX
default: assert(0);
@ -325,7 +335,7 @@ static void load_buffer_8x8(const int32_t *coeff, __m128i *in) {
}
static void idct8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
const __m128i cospim8 = _mm_set1_epi32(-cospi[8]);
const __m128i cospi24 = _mm_set1_epi32(cospi[24]);
@ -439,7 +449,7 @@ static void idct8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
}
static void iadst8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
const __m128i cospim16 = _mm_set1_epi32(-cospi[16]);
@ -698,90 +708,100 @@ static void write_buffer_8x8(__m128i *in, uint16_t *output, int stride,
void av1_inv_txfm2d_add_8x8_sse4_1(const int32_t *coeff, uint16_t *output,
int stride, int tx_type, int bd) {
__m128i in[16], out[16];
const TXFM_2D_CFG *cfg = NULL;
const TXFM_1D_CFG *row_cfg = NULL;
const TXFM_1D_CFG *col_cfg = NULL;
switch (tx_type) {
case DCT_DCT:
cfg = &inv_txfm_2d_cfg_dct_dct_8;
row_cfg = &inv_txfm_1d_row_cfg_dct_8;
col_cfg = &inv_txfm_1d_col_cfg_dct_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
idct8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case DCT_ADST:
cfg = &inv_txfm_2d_cfg_dct_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_dct_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_8;
row_cfg = &inv_txfm_1d_row_cfg_dct_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
idct8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_8;
row_cfg = &inv_txfm_1d_row_cfg_dct_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
idct8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 1, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
case DCT_FLIPADST:
cfg = &inv_txfm_2d_cfg_dct_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_dct_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
idct8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 1, 0, -cfg->shift[1], bd);
idct8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case ADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 1, 0, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case FLIPADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 1, 1, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 1, 1, -row_cfg->shift[1], bd);
break;
case FLIPADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_8;
row_cfg = &inv_txfm_1d_row_cfg_adst_8;
col_cfg = &inv_txfm_1d_col_cfg_adst_8;
load_buffer_8x8(coeff, in);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_row[2]);
iadst8x8_sse4_1(out, in, row_cfg->cos_bit[2]);
transpose_8x8(in, out);
iadst8x8_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_8x8(in, output, stride, 0, 1, -cfg->shift[1], bd);
iadst8x8_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_8x8(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
#endif // CONFIG_EXT_TX
default: assert(0);
@ -849,7 +869,7 @@ static void write_buffer_16x16(__m128i *in, uint16_t *output, int stride,
}
static void idct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi60 = _mm_set1_epi32(cospi[60]);
const __m128i cospim4 = _mm_set1_epi32(-cospi[4]);
const __m128i cospi28 = _mm_set1_epi32(cospi[28]);
@ -907,24 +927,24 @@ static void idct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
v[6] = u[6];
v[7] = u[7];
v[8] = half_btf_sse4_1(cospi60, u[8], cospim4, u[15], rnding, bit);
v[9] = half_btf_sse4_1(cospi28, u[9], cospim36, u[14], rnding, bit);
v[10] = half_btf_sse4_1(cospi44, u[10], cospim20, u[13], rnding, bit);
v[11] = half_btf_sse4_1(cospi12, u[11], cospim52, u[12], rnding, bit);
v[12] = half_btf_sse4_1(cospi52, u[11], cospi12, u[12], rnding, bit);
v[13] = half_btf_sse4_1(cospi20, u[10], cospi44, u[13], rnding, bit);
v[14] = half_btf_sse4_1(cospi36, u[9], cospi28, u[14], rnding, bit);
v[15] = half_btf_sse4_1(cospi4, u[8], cospi60, u[15], rnding, bit);
v[8] = half_btf_sse4_1(&cospi60, &u[8], &cospim4, &u[15], &rnding, bit);
v[9] = half_btf_sse4_1(&cospi28, &u[9], &cospim36, &u[14], &rnding, bit);
v[10] = half_btf_sse4_1(&cospi44, &u[10], &cospim20, &u[13], &rnding, bit);
v[11] = half_btf_sse4_1(&cospi12, &u[11], &cospim52, &u[12], &rnding, bit);
v[12] = half_btf_sse4_1(&cospi52, &u[11], &cospi12, &u[12], &rnding, bit);
v[13] = half_btf_sse4_1(&cospi20, &u[10], &cospi44, &u[13], &rnding, bit);
v[14] = half_btf_sse4_1(&cospi36, &u[9], &cospi28, &u[14], &rnding, bit);
v[15] = half_btf_sse4_1(&cospi4, &u[8], &cospi60, &u[15], &rnding, bit);
// stage 3
u[0] = v[0];
u[1] = v[1];
u[2] = v[2];
u[3] = v[3];
u[4] = half_btf_sse4_1(cospi56, v[4], cospim8, v[7], rnding, bit);
u[5] = half_btf_sse4_1(cospi24, v[5], cospim40, v[6], rnding, bit);
u[6] = half_btf_sse4_1(cospi40, v[5], cospi24, v[6], rnding, bit);
u[7] = half_btf_sse4_1(cospi8, v[4], cospi56, v[7], rnding, bit);
u[4] = half_btf_sse4_1(&cospi56, &v[4], &cospim8, &v[7], &rnding, bit);
u[5] = half_btf_sse4_1(&cospi24, &v[5], &cospim40, &v[6], &rnding, bit);
u[6] = half_btf_sse4_1(&cospi40, &v[5], &cospi24, &v[6], &rnding, bit);
u[7] = half_btf_sse4_1(&cospi8, &v[4], &cospi56, &v[7], &rnding, bit);
u[8] = _mm_add_epi32(v[8], v[9]);
u[9] = _mm_sub_epi32(v[8], v[9]);
u[10] = _mm_sub_epi32(v[11], v[10]);
@ -945,19 +965,19 @@ static void idct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
v[1] = _mm_add_epi32(v[1], rnding);
v[1] = _mm_srai_epi32(v[1], bit);
v[2] = half_btf_sse4_1(cospi48, u[2], cospim16, u[3], rnding, bit);
v[3] = half_btf_sse4_1(cospi16, u[2], cospi48, u[3], rnding, bit);
v[2] = half_btf_sse4_1(&cospi48, &u[2], &cospim16, &u[3], &rnding, bit);
v[3] = half_btf_sse4_1(&cospi16, &u[2], &cospi48, &u[3], &rnding, bit);
v[4] = _mm_add_epi32(u[4], u[5]);
v[5] = _mm_sub_epi32(u[4], u[5]);
v[6] = _mm_sub_epi32(u[7], u[6]);
v[7] = _mm_add_epi32(u[6], u[7]);
v[8] = u[8];
v[9] = half_btf_sse4_1(cospim16, u[9], cospi48, u[14], rnding, bit);
v[10] = half_btf_sse4_1(cospim48, u[10], cospim16, u[13], rnding, bit);
v[9] = half_btf_sse4_1(&cospim16, &u[9], &cospi48, &u[14], &rnding, bit);
v[10] = half_btf_sse4_1(&cospim48, &u[10], &cospim16, &u[13], &rnding, bit);
v[11] = u[11];
v[12] = u[12];
v[13] = half_btf_sse4_1(cospim16, u[10], cospi48, u[13], rnding, bit);
v[14] = half_btf_sse4_1(cospi48, u[9], cospi16, u[14], rnding, bit);
v[13] = half_btf_sse4_1(&cospim16, &u[10], &cospi48, &u[13], &rnding, bit);
v[14] = half_btf_sse4_1(&cospi48, &u[9], &cospi16, &u[14], &rnding, bit);
v[15] = u[15];
// stage 5
@ -1043,7 +1063,7 @@ static void idct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
}
static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
const int32_t *cospi = cospi_arr[bit - cos_bit_min];
const int32_t *cospi = cospi_arr(bit);
const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
@ -1183,18 +1203,18 @@ static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
v[1] = u[1];
v[2] = u[2];
v[3] = u[3];
v[4] = half_btf_sse4_1(cospi16, u[4], cospi48, u[5], rnding, bit);
v[5] = half_btf_sse4_1(cospi48, u[4], cospim16, u[5], rnding, bit);
v[6] = half_btf_sse4_1(cospim48, u[6], cospi16, u[7], rnding, bit);
v[7] = half_btf_sse4_1(cospi16, u[6], cospi48, u[7], rnding, bit);
v[4] = half_btf_sse4_1(&cospi16, &u[4], &cospi48, &u[5], &rnding, bit);
v[5] = half_btf_sse4_1(&cospi48, &u[4], &cospim16, &u[5], &rnding, bit);
v[6] = half_btf_sse4_1(&cospim48, &u[6], &cospi16, &u[7], &rnding, bit);
v[7] = half_btf_sse4_1(&cospi16, &u[6], &cospi48, &u[7], &rnding, bit);
v[8] = u[8];
v[9] = u[9];
v[10] = u[10];
v[11] = u[11];
v[12] = half_btf_sse4_1(cospi16, u[12], cospi48, u[13], rnding, bit);
v[13] = half_btf_sse4_1(cospi48, u[12], cospim16, u[13], rnding, bit);
v[14] = half_btf_sse4_1(cospim48, u[14], cospi16, u[15], rnding, bit);
v[15] = half_btf_sse4_1(cospi16, u[14], cospi48, u[15], rnding, bit);
v[12] = half_btf_sse4_1(&cospi16, &u[12], &cospi48, &u[13], &rnding, bit);
v[13] = half_btf_sse4_1(&cospi48, &u[12], &cospim16, &u[13], &rnding, bit);
v[14] = half_btf_sse4_1(&cospim48, &u[14], &cospi16, &u[15], &rnding, bit);
v[15] = half_btf_sse4_1(&cospi16, &u[14], &cospi48, &u[15], &rnding, bit);
// stage 5
u[0] = _mm_add_epi32(v[0], v[4]);
@ -1223,14 +1243,14 @@ static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
v[5] = u[5];
v[6] = u[6];
v[7] = u[7];
v[8] = half_btf_sse4_1(cospi8, u[8], cospi56, u[9], rnding, bit);
v[9] = half_btf_sse4_1(cospi56, u[8], cospim8, u[9], rnding, bit);
v[10] = half_btf_sse4_1(cospi40, u[10], cospi24, u[11], rnding, bit);
v[11] = half_btf_sse4_1(cospi24, u[10], cospim40, u[11], rnding, bit);
v[12] = half_btf_sse4_1(cospim56, u[12], cospi8, u[13], rnding, bit);
v[13] = half_btf_sse4_1(cospi8, u[12], cospi56, u[13], rnding, bit);
v[14] = half_btf_sse4_1(cospim24, u[14], cospi40, u[15], rnding, bit);
v[15] = half_btf_sse4_1(cospi40, u[14], cospi24, u[15], rnding, bit);
v[8] = half_btf_sse4_1(&cospi8, &u[8], &cospi56, &u[9], &rnding, bit);
v[9] = half_btf_sse4_1(&cospi56, &u[8], &cospim8, &u[9], &rnding, bit);
v[10] = half_btf_sse4_1(&cospi40, &u[10], &cospi24, &u[11], &rnding, bit);
v[11] = half_btf_sse4_1(&cospi24, &u[10], &cospim40, &u[11], &rnding, bit);
v[12] = half_btf_sse4_1(&cospim56, &u[12], &cospi8, &u[13], &rnding, bit);
v[13] = half_btf_sse4_1(&cospi8, &u[12], &cospi56, &u[13], &rnding, bit);
v[14] = half_btf_sse4_1(&cospim24, &u[14], &cospi40, &u[15], &rnding, bit);
v[15] = half_btf_sse4_1(&cospi40, &u[14], &cospi24, &u[15], &rnding, bit);
// stage 7
u[0] = _mm_add_epi32(v[0], v[8]);
@ -1251,22 +1271,22 @@ static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
u[15] = _mm_sub_epi32(v[7], v[15]);
// stage 8
v[0] = half_btf_sse4_1(cospi2, u[0], cospi62, u[1], rnding, bit);
v[1] = half_btf_sse4_1(cospi62, u[0], cospim2, u[1], rnding, bit);
v[2] = half_btf_sse4_1(cospi10, u[2], cospi54, u[3], rnding, bit);
v[3] = half_btf_sse4_1(cospi54, u[2], cospim10, u[3], rnding, bit);
v[4] = half_btf_sse4_1(cospi18, u[4], cospi46, u[5], rnding, bit);
v[5] = half_btf_sse4_1(cospi46, u[4], cospim18, u[5], rnding, bit);
v[6] = half_btf_sse4_1(cospi26, u[6], cospi38, u[7], rnding, bit);
v[7] = half_btf_sse4_1(cospi38, u[6], cospim26, u[7], rnding, bit);
v[8] = half_btf_sse4_1(cospi34, u[8], cospi30, u[9], rnding, bit);
v[9] = half_btf_sse4_1(cospi30, u[8], cospim34, u[9], rnding, bit);
v[10] = half_btf_sse4_1(cospi42, u[10], cospi22, u[11], rnding, bit);
v[11] = half_btf_sse4_1(cospi22, u[10], cospim42, u[11], rnding, bit);
v[12] = half_btf_sse4_1(cospi50, u[12], cospi14, u[13], rnding, bit);
v[13] = half_btf_sse4_1(cospi14, u[12], cospim50, u[13], rnding, bit);
v[14] = half_btf_sse4_1(cospi58, u[14], cospi6, u[15], rnding, bit);
v[15] = half_btf_sse4_1(cospi6, u[14], cospim58, u[15], rnding, bit);
v[0] = half_btf_sse4_1(&cospi2, &u[0], &cospi62, &u[1], &rnding, bit);
v[1] = half_btf_sse4_1(&cospi62, &u[0], &cospim2, &u[1], &rnding, bit);
v[2] = half_btf_sse4_1(&cospi10, &u[2], &cospi54, &u[3], &rnding, bit);
v[3] = half_btf_sse4_1(&cospi54, &u[2], &cospim10, &u[3], &rnding, bit);
v[4] = half_btf_sse4_1(&cospi18, &u[4], &cospi46, &u[5], &rnding, bit);
v[5] = half_btf_sse4_1(&cospi46, &u[4], &cospim18, &u[5], &rnding, bit);
v[6] = half_btf_sse4_1(&cospi26, &u[6], &cospi38, &u[7], &rnding, bit);
v[7] = half_btf_sse4_1(&cospi38, &u[6], &cospim26, &u[7], &rnding, bit);
v[8] = half_btf_sse4_1(&cospi34, &u[8], &cospi30, &u[9], &rnding, bit);
v[9] = half_btf_sse4_1(&cospi30, &u[8], &cospim34, &u[9], &rnding, bit);
v[10] = half_btf_sse4_1(&cospi42, &u[10], &cospi22, &u[11], &rnding, bit);
v[11] = half_btf_sse4_1(&cospi22, &u[10], &cospim42, &u[11], &rnding, bit);
v[12] = half_btf_sse4_1(&cospi50, &u[12], &cospi14, &u[13], &rnding, bit);
v[13] = half_btf_sse4_1(&cospi14, &u[12], &cospim50, &u[13], &rnding, bit);
v[14] = half_btf_sse4_1(&cospi58, &u[14], &cospi6, &u[15], &rnding, bit);
v[15] = half_btf_sse4_1(&cospi6, &u[14], &cospim58, &u[15], &rnding, bit);
// stage 9
out[0 * 4 + col] = v[1];
@ -1298,99 +1318,109 @@ static void round_shift_16x16(__m128i *in, int shift) {
void av1_inv_txfm2d_add_16x16_sse4_1(const int32_t *coeff, uint16_t *output,
int stride, int tx_type, int bd) {
__m128i in[64], out[64];
const TXFM_2D_CFG *cfg = NULL;
const TXFM_1D_CFG *row_cfg = NULL;
const TXFM_1D_CFG *col_cfg = NULL;
switch (tx_type) {
case DCT_DCT:
cfg = &inv_txfm_2d_cfg_dct_dct_16;
row_cfg = &inv_txfm_1d_row_cfg_dct_16;
col_cfg = &inv_txfm_1d_col_cfg_dct_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
idct16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case DCT_ADST:
cfg = &inv_txfm_2d_cfg_dct_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_dct_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 0, -cfg->shift[1], bd);
idct16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_16;
row_cfg = &inv_txfm_1d_row_cfg_dct_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
idct16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
case ADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 0, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 0, -row_cfg->shift[1], bd);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
cfg = &inv_txfm_2d_cfg_adst_dct_16;
row_cfg = &inv_txfm_1d_row_cfg_dct_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
idct16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 1, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
case DCT_FLIPADST:
cfg = &inv_txfm_2d_cfg_dct_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_dct_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
idct16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 1, 0, -cfg->shift[1], bd);
idct16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case ADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 1, 0, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 1, 0, -row_cfg->shift[1], bd);
break;
case FLIPADST_FLIPADST:
cfg = &inv_txfm_2d_cfg_adst_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 1, 1, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 1, 1, -row_cfg->shift[1], bd);
break;
case FLIPADST_ADST:
cfg = &inv_txfm_2d_cfg_adst_adst_16;
row_cfg = &inv_txfm_1d_row_cfg_adst_16;
col_cfg = &inv_txfm_1d_col_cfg_adst_16;
load_buffer_16x16(coeff, in);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_row[2]);
round_shift_16x16(in, -cfg->shift[0]);
iadst16x16_sse4_1(out, in, row_cfg->cos_bit[2]);
round_shift_16x16(in, -row_cfg->shift[0]);
transpose_16x16(in, out);
iadst16x16_sse4_1(out, in, cfg->cos_bit_col[2]);
write_buffer_16x16(in, output, stride, 0, 1, -cfg->shift[1], bd);
iadst16x16_sse4_1(out, in, col_cfg->cos_bit[2]);
write_buffer_16x16(in, output, stride, 0, 1, -row_cfg->shift[1], bd);
break;
#endif
default: assert(0);

11
third_party/aom/av1/common/x86/highbd_txfm_utility_sse4.h vendored
View file

@ -77,14 +77,15 @@ static INLINE void transpose_16x16(const __m128i *in, __m128i *out) {
// Note:
// rounding = 1 << (bit - 1)
static INLINE __m128i half_btf_sse4_1(__m128i w0, __m128i n0, __m128i w1,
__m128i n1, __m128i rounding, int bit) {
static INLINE __m128i half_btf_sse4_1(const __m128i *w0, const __m128i *n0,
const __m128i *w1, const __m128i *n1,
const __m128i *rounding, int bit) {
__m128i x, y;
x = _mm_mullo_epi32(w0, n0);
y = _mm_mullo_epi32(w1, n1);
x = _mm_mullo_epi32(*w0, *n0);
y = _mm_mullo_epi32(*w1, *n1);
x = _mm_add_epi32(x, y);
x = _mm_add_epi32(x, rounding);
x = _mm_add_epi32(x, *rounding);
x = _mm_srai_epi32(x, bit);
return x;
}

320
third_party/aom/av1/common/x86/highbd_warp_plane_ssse3.c vendored
View file

@ -14,16 +14,13 @@
#include "./av1_rtcd.h"
#include "av1/common/warped_motion.h"
static const __m128i *const filter = (const __m128i *const)warped_filter;
/* SSE2 version of the rotzoom/affine warp filter */
void av1_highbd_warp_affine_ssse3(int32_t *mat, 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, int ref_frm,
int16_t alpha, int16_t beta, int16_t gamma,
int16_t delta) {
void av1_highbd_warp_affine_ssse3(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,
int comp_avg, int16_t alpha, int16_t beta,
int16_t gamma, int16_t delta) {
#if HORSHEAR_REDUCE_PREC_BITS >= 5
__m128i tmp[15];
#else
@ -47,23 +44,21 @@ void av1_highbd_warp_affine_ssse3(int32_t *mat, uint16_t *ref, int width,
for (j = 0; j < p_width; j += 8) {
// (x, y) coordinates of the center of this block in the destination
// image
int32_t dst_x = p_col + j + 4;
int32_t dst_y = p_row + i + 4;
const int32_t dst_x = p_col + j + 4;
const int32_t dst_y = p_row + i + 4;
int32_t x4, y4, ix4, sx4, iy4, sy4;
if (subsampling_x)
x4 = ROUND_POWER_OF_TWO_SIGNED(
mat[2] * 2 * dst_x + mat[3] * 2 * dst_y + mat[0] +
(mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
1);
x4 = (mat[2] * 4 * dst_x + mat[3] * 4 * dst_y + mat[0] * 2 +
(mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
x4 = mat[2] * dst_x + mat[3] * dst_y + mat[0];
if (subsampling_y)
y4 = ROUND_POWER_OF_TWO_SIGNED(
mat[4] * 2 * dst_x + mat[5] * 2 * dst_y + mat[1] +
(mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
1);
y4 = (mat[4] * 4 * dst_x + mat[5] * 4 * dst_y + mat[1] * 2 +
(mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
y4 = mat[4] * dst_x + mat[5] * dst_y + mat[1];
@ -72,71 +67,103 @@ void av1_highbd_warp_affine_ssse3(int32_t *mat, uint16_t *ref, int width,
iy4 = y4 >> WARPEDMODEL_PREC_BITS;
sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
// Horizontal filter
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
// Add in all the constant terms, including rounding and offset
sx4 += alpha * (-4) + beta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
sy4 += gamma * (-4) + delta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
// If the block is aligned such that, after clamping, every sample
// would be taken from the leftmost/rightmost column, then we can
// skip the expensive horizontal filter.
if (ix4 <= -7) {
sx4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
// Horizontal filter
// If the block is aligned such that, after clamping, every sample
// would be taken from the leftmost/rightmost column, then we can
// skip the expensive horizontal filter.
if (ix4 <= -7) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
} else if (ix4 >= width + 6) {
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else if (ix4 >= width + 6) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride + (width - 1)] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
} else {
int sx = sx4 + alpha * (-4) + beta * k +
// Include rounding and offset here
(1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
int sx = sx4 + beta * (k + 4);
// Load source pixels
__m128i src =
const __m128i src =
_mm_loadu_si128((__m128i *)(ref + iy * stride + ix4 - 7));
__m128i src2 =
const __m128i src2 =
_mm_loadu_si128((__m128i *)(ref + iy * stride + ix4 + 1));
// Filter even-index pixels
__m128i tmp_0 = filter[(sx + 0 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_2 = filter[(sx + 2 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_4 = filter[(sx + 4 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_6 = filter[(sx + 6 * alpha) >> WARPEDDIFF_PREC_BITS];
const __m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 0 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 2 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 4 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 6 * alpha) >> WARPEDDIFF_PREC_BITS)));
// coeffs 0 1 0 1 2 3 2 3 for pixels 0, 2
__m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
const __m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
// coeffs 0 1 0 1 2 3 2 3 for pixels 4, 6
__m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
const __m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
// coeffs 4 5 4 5 6 7 6 7 for pixels 0, 2
__m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
const __m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
// coeffs 4 5 4 5 6 7 6 7 for pixels 4, 6
__m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
const __m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
// coeffs 0 1 0 1 0 1 0 1 for pixels 0, 2, 4, 6
__m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
const __m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
// coeffs 2 3 2 3 2 3 2 3 for pixels 0, 2, 4, 6
__m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
const __m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
// coeffs 4 5 4 5 4 5 4 5 for pixels 0, 2, 4, 6
__m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
const __m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
// coeffs 6 7 6 7 6 7 6 7 for pixels 0, 2, 4, 6
__m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
const __m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
__m128i round_const =
_mm_set1_epi32((1 << HORSHEAR_REDUCE_PREC_BITS) >> 1);
const __m128i round_const =
_mm_set1_epi32((1 << (bd + WARPEDPIXEL_FILTER_BITS - 1)) +
((1 << HORSHEAR_REDUCE_PREC_BITS) >> 1));
// Calculate filtered results
__m128i res_0 = _mm_madd_epi16(src, coeff_0);
__m128i res_2 =
const __m128i res_0 = _mm_madd_epi16(src, coeff_0);
const __m128i res_2 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 4), coeff_2);
__m128i res_4 =
const __m128i res_4 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 8), coeff_4);
__m128i res_6 =
const __m128i res_6 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 12), coeff_6);
__m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
@ -145,28 +172,36 @@ void av1_highbd_warp_affine_ssse3(int32_t *mat, uint16_t *ref, int width,
HORSHEAR_REDUCE_PREC_BITS);
// Filter odd-index pixels
__m128i tmp_1 = filter[(sx + 1 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_3 = filter[(sx + 3 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_5 = filter[(sx + 5 * alpha) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_7 = filter[(sx + 7 * alpha) >> WARPEDDIFF_PREC_BITS];
const __m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 1 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 3 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 5 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 7 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
__m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
__m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
__m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
const __m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
const __m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
__m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
__m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
__m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
__m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
const __m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
const __m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
const __m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
const __m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
__m128i res_1 =
const __m128i res_1 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 2), coeff_1);
__m128i res_3 =
const __m128i res_3 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 6), coeff_3);
__m128i res_5 =
const __m128i res_5 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 10), coeff_5);
__m128i res_7 =
const __m128i res_7 =
_mm_madd_epi16(_mm_alignr_epi8(src2, src, 14), coeff_7);
__m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
@ -183,101 +218,118 @@ void av1_highbd_warp_affine_ssse3(int32_t *mat, uint16_t *ref, int width,
// Vertical filter
for (k = -4; k < AOMMIN(4, p_height - i - 4); ++k) {
int sy = sy4 + gamma * (-4) + delta * k +
(1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
int sy = sy4 + delta * (k + 4);
// Load from tmp and rearrange pairs of consecutive rows into the
// column order 0 0 2 2 4 4 6 6; 1 1 3 3 5 5 7 7
__m128i *src = tmp + (k + 4);
__m128i src_0 = _mm_unpacklo_epi16(src[0], src[1]);
__m128i src_2 = _mm_unpacklo_epi16(src[2], src[3]);
__m128i src_4 = _mm_unpacklo_epi16(src[4], src[5]);
__m128i src_6 = _mm_unpacklo_epi16(src[6], src[7]);
const __m128i *src = tmp + (k + 4);
const __m128i src_0 = _mm_unpacklo_epi16(src[0], src[1]);
const __m128i src_2 = _mm_unpacklo_epi16(src[2], src[3]);
const __m128i src_4 = _mm_unpacklo_epi16(src[4], src[5]);
const __m128i src_6 = _mm_unpacklo_epi16(src[6], src[7]);
// Filter even-index pixels
__m128i tmp_0 = filter[(sy + 0 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_2 = filter[(sy + 2 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_4 = filter[(sy + 4 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_6 = filter[(sy + 6 * gamma) >> WARPEDDIFF_PREC_BITS];
const __m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 0 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 2 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 4 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 6 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
__m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
__m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
__m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
const __m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
const __m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
const __m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
const __m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
__m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
__m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
__m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
__m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
const __m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
const __m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
const __m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
const __m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
__m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
__m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
__m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
__m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
const __m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
const __m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
const __m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
const __m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
__m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
_mm_add_epi32(res_4, res_6));
const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
_mm_add_epi32(res_4, res_6));
// Filter odd-index pixels
__m128i src_1 = _mm_unpackhi_epi16(src[0], src[1]);
__m128i src_3 = _mm_unpackhi_epi16(src[2], src[3]);
__m128i src_5 = _mm_unpackhi_epi16(src[4], src[5]);
__m128i src_7 = _mm_unpackhi_epi16(src[6], src[7]);
const __m128i src_1 = _mm_unpackhi_epi16(src[0], src[1]);
const __m128i src_3 = _mm_unpackhi_epi16(src[2], src[3]);
const __m128i src_5 = _mm_unpackhi_epi16(src[4], src[5]);
const __m128i src_7 = _mm_unpackhi_epi16(src[6], src[7]);
__m128i tmp_1 = filter[(sy + 1 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_3 = filter[(sy + 3 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_5 = filter[(sy + 5 * gamma) >> WARPEDDIFF_PREC_BITS];
__m128i tmp_7 = filter[(sy + 7 * gamma) >> WARPEDDIFF_PREC_BITS];
const __m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 1 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 3 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 5 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 7 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
__m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
__m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
__m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
const __m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
const __m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
__m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
__m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
__m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
__m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
const __m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
const __m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
const __m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
const __m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
__m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
__m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
__m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
__m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
const __m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
const __m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
const __m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
const __m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
__m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
_mm_add_epi32(res_5, res_7));
const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
_mm_add_epi32(res_5, res_7));
// Rearrange pixels back into the order 0 ... 7
__m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
__m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
// Round and pack into 8 bits
__m128i round_const =
_mm_set1_epi32((1 << VERSHEAR_REDUCE_PREC_BITS) >> 1);
const __m128i round_const =
_mm_set1_epi32(-(1 << (bd + VERSHEAR_REDUCE_PREC_BITS - 1)) +
((1 << VERSHEAR_REDUCE_PREC_BITS) >> 1));
__m128i res_lo_round = _mm_srai_epi32(
const __m128i res_lo_round = _mm_srai_epi32(
_mm_add_epi32(res_lo, round_const), VERSHEAR_REDUCE_PREC_BITS);
__m128i res_hi_round = _mm_srai_epi32(
const __m128i res_hi_round = _mm_srai_epi32(
_mm_add_epi32(res_hi, round_const), VERSHEAR_REDUCE_PREC_BITS);
__m128i res_16bit = _mm_packs_epi32(res_lo_round, res_hi_round);
// Clamp res_16bit to the range [0, 2^bd - 1]
__m128i max_val = _mm_set1_epi16((1 << bd) - 1);
__m128i zero = _mm_setzero_si128();
const __m128i max_val = _mm_set1_epi16((1 << bd) - 1);
const __m128i zero = _mm_setzero_si128();
res_16bit = _mm_max_epi16(_mm_min_epi16(res_16bit, max_val), zero);
// Store, blending with 'pred' if needed
__m128i *p = (__m128i *)&pred[(i + k + 4) * p_stride + j];
__m128i *const p = (__m128i *)&pred[(i + k + 4) * p_stride + j];
// Note: If we're outputting a 4x4 block, we need to be very careful
// to only output 4 pixels at this point, to avoid encode/decode
// mismatches when encoding with multiple threads.
if (p_width == 4) {
if (ref_frm) res_16bit = _mm_avg_epu16(res_16bit, _mm_loadl_epi64(p));
if (comp_avg)
res_16bit = _mm_avg_epu16(res_16bit, _mm_loadl_epi64(p));
_mm_storel_epi64(p, res_16bit);
} else {
if (ref_frm) res_16bit = _mm_avg_epu16(res_16bit, _mm_loadu_si128(p));
if (comp_avg)
res_16bit = _mm_avg_epu16(res_16bit, _mm_loadu_si128(p));
_mm_storeu_si128(p, res_16bit);
}
}

74
third_party/aom/av1/common/x86/hybrid_inv_txfm_avx2.c vendored
View file

@ -14,67 +14,9 @@
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "aom_dsp/x86/txfm_common_avx2.h"
#include "aom_dsp/x86/inv_txfm_common_avx2.h"
static INLINE void load_coeff(const tran_low_t *coeff, __m256i *in) {
#if CONFIG_HIGHBITDEPTH
*in = _mm256_setr_epi16(
(int16_t)coeff[0], (int16_t)coeff[1], (int16_t)coeff[2],
(int16_t)coeff[3], (int16_t)coeff[4], (int16_t)coeff[5],
(int16_t)coeff[6], (int16_t)coeff[7], (int16_t)coeff[8],
(int16_t)coeff[9], (int16_t)coeff[10], (int16_t)coeff[11],
(int16_t)coeff[12], (int16_t)coeff[13], (int16_t)coeff[14],
(int16_t)coeff[15]);
#else
*in = _mm256_loadu_si256((const __m256i *)coeff);
#endif
}
static void load_buffer_16x16(const tran_low_t *coeff, __m256i *in) {
int i = 0;
while (i < 16) {
load_coeff(coeff + (i << 4), &in[i]);
i += 1;
}
}
static void recon_and_store(const __m256i *res, uint8_t *output) {
const __m128i zero = _mm_setzero_si128();
__m128i x = _mm_loadu_si128((__m128i const *)output);
__m128i p0 = _mm_unpacklo_epi8(x, zero);
__m128i p1 = _mm_unpackhi_epi8(x, zero);
p0 = _mm_add_epi16(p0, _mm256_castsi256_si128(*res));
p1 = _mm_add_epi16(p1, _mm256_extractf128_si256(*res, 1));
x = _mm_packus_epi16(p0, p1);
_mm_storeu_si128((__m128i *)output, x);
}
#define IDCT_ROUNDING_POS (6)
static void write_buffer_16x16(__m256i *in, const int stride, uint8_t *output) {
const __m256i rounding = _mm256_set1_epi16(1 << (IDCT_ROUNDING_POS - 1));
int i = 0;
while (i < 16) {
in[i] = _mm256_add_epi16(in[i], rounding);
in[i] = _mm256_srai_epi16(in[i], IDCT_ROUNDING_POS);
recon_and_store(&in[i], output + i * stride);
i += 1;
}
}
static INLINE void unpack_butter_fly(const __m256i *a0, const __m256i *a1,
const __m256i *c0, const __m256i *c1,
__m256i *b0, __m256i *b1) {
__m256i x0, x1;
x0 = _mm256_unpacklo_epi16(*a0, *a1);
x1 = _mm256_unpackhi_epi16(*a0, *a1);
*b0 = butter_fly(x0, x1, *c0);
*b1 = butter_fly(x0, x1, *c1);
}
static void idct16_avx2(__m256i *in) {
void av1_idct16_avx2(__m256i *in) {
const __m256i cospi_p30_m02 = pair256_set_epi16(cospi_30_64, -cospi_2_64);
const __m256i cospi_p02_p30 = pair256_set_epi16(cospi_2_64, cospi_30_64);
const __m256i cospi_p14_m18 = pair256_set_epi16(cospi_14_64, -cospi_18_64);
@ -216,8 +158,8 @@ static void idct16_avx2(__m256i *in) {
}
static void idct16(__m256i *in) {
mm256_transpose_16x16(in);
idct16_avx2(in);
mm256_transpose_16x16(in, in);
av1_idct16_avx2(in);
}
static INLINE void butterfly_32b(const __m256i *a0, const __m256i *a1,
@ -398,7 +340,7 @@ static void iadst16_avx2(__m256i *in) {
}
static void iadst16(__m256i *in) {
mm256_transpose_16x16(in);
mm256_transpose_16x16(in, in);
iadst16_avx2(in);
}
@ -416,8 +358,8 @@ static void flip_col(uint8_t **dest, int *stride, int rows) {
}
static void iidtx16(__m256i *in) {
mm256_transpose_16x16(in);
txfm_scaling16_avx2(Sqrt2, in);
mm256_transpose_16x16(in, in);
txfm_scaling16_avx2((int16_t)Sqrt2, in);
}
#endif
@ -503,5 +445,5 @@ void av1_iht16x16_256_add_avx2(const tran_low_t *input, uint8_t *dest,
#endif // CONFIG_EXT_TX
default: assert(0); break;
}
write_buffer_16x16(in, stride, dest);
store_buffer_16xN(in, stride, dest, 16);
}

8
third_party/aom/av1/common/x86/idct_intrin_sse2.c vendored
View file

@ -17,14 +17,14 @@
#include "av1/common/enums.h"
#if CONFIG_EXT_TX
static INLINE void fliplr_4x4(__m128i in[2]) {
static INLINE void fliplr_4x4(__m128i *in /*in[2]*/) {
in[0] = _mm_shufflelo_epi16(in[0], 0x1b);
in[0] = _mm_shufflehi_epi16(in[0], 0x1b);
in[1] = _mm_shufflelo_epi16(in[1], 0x1b);
in[1] = _mm_shufflehi_epi16(in[1], 0x1b);
}
static INLINE void fliplr_8x8(__m128i in[8]) {
static INLINE void fliplr_8x8(__m128i *in /*in[8]*/) {
in[0] = mm_reverse_epi16(in[0]);
in[1] = mm_reverse_epi16(in[1]);
in[2] = mm_reverse_epi16(in[2]);
@ -36,7 +36,7 @@ static INLINE void fliplr_8x8(__m128i in[8]) {
in[7] = mm_reverse_epi16(in[7]);
}
static INLINE void fliplr_16x8(__m128i in[16]) {
static INLINE void fliplr_16x8(__m128i *in /*in[16]*/) {
fliplr_8x8(&in[0]);
fliplr_8x8(&in[8]);
}
@ -356,7 +356,7 @@ static void iidtx8_sse2(__m128i *in) {
}
static INLINE void iidtx4_sse2(__m128i *in) {
const __m128i v_scale_w = _mm_set1_epi16(Sqrt2);
const __m128i v_scale_w = _mm_set1_epi16((int16_t)Sqrt2);
const __m128i v_p0l_w = _mm_mullo_epi16(in[0], v_scale_w);
const __m128i v_p0h_w = _mm_mulhi_epi16(in[0], v_scale_w);

345
third_party/aom/av1/common/x86/warp_plane_sse2.c vendored
View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
* 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
@ -14,17 +14,15 @@
#include "./av1_rtcd.h"
#include "av1/common/warped_motion.h"
static const __m128i *const filter = (const __m128i *const)warped_filter;
/* SSE2 version of the rotzoom/affine warp filter */
void av1_warp_affine_sse2(int32_t *mat, 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, int ref_frm,
void av1_warp_affine_sse2(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, int comp_avg,
int16_t alpha, int16_t beta, int16_t gamma,
int16_t delta) {
__m128i tmp[15];
int i, j, k;
const int bd = 8;
/* Note: For this code to work, the left/right frame borders need to be
extended by at least 13 pixels each. By the time we get here, other
@ -42,23 +40,21 @@ void av1_warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
for (j = 0; j < p_width; j += 8) {
// (x, y) coordinates of the center of this block in the destination
// image
int32_t dst_x = p_col + j + 4;
int32_t dst_y = p_row + i + 4;
const int32_t dst_x = p_col + j + 4;
const int32_t dst_y = p_row + i + 4;
int32_t x4, y4, ix4, sx4, iy4, sy4;
if (subsampling_x)
x4 = ROUND_POWER_OF_TWO_SIGNED(
mat[2] * 2 * dst_x + mat[3] * 2 * dst_y + mat[0] +
(mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
1);
x4 = (mat[2] * 4 * dst_x + mat[3] * 4 * dst_y + mat[0] * 2 +
(mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
x4 = mat[2] * dst_x + mat[3] * dst_y + mat[0];
if (subsampling_y)
y4 = ROUND_POWER_OF_TWO_SIGNED(
mat[4] * 2 * dst_x + mat[5] * 2 * dst_y + mat[1] +
(mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
1);
y4 = (mat[4] * 4 * dst_x + mat[5] * 4 * dst_y + mat[1] * 2 +
(mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
y4 = mat[4] * dst_x + mat[5] * dst_y + mat[1];
@ -67,76 +63,104 @@ void av1_warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
iy4 = y4 >> WARPEDMODEL_PREC_BITS;
sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
// Horizontal filter
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
// Add in all the constant terms, including rounding and offset
sx4 += alpha * (-4) + beta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
sy4 += gamma * (-4) + delta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
// If the block is aligned such that, after clamping, every sample
// would be taken from the leftmost/rightmost column, then we can
// skip the expensive horizontal filter.
if (ix4 <= -7) {
sx4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
// Horizontal filter
// If the block is aligned such that, after clamping, every sample
// would be taken from the leftmost/rightmost column, then we can
// skip the expensive horizontal filter.
if (ix4 <= -7) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
} else if (ix4 >= width + 6) {
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else if (ix4 >= width + 6) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride + (width - 1)] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
} else {
int sx = sx4 + alpha * (-4) + beta * k +
// Include rounding and offset here
(1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
int sx = sx4 + beta * (k + 4);
// Load source pixels
__m128i zero = _mm_setzero_si128();
__m128i src =
const __m128i zero = _mm_setzero_si128();
const __m128i src =
_mm_loadu_si128((__m128i *)(ref + iy * stride + ix4 - 7));
// Filter even-index pixels
__m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 0 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 2 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 4 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 6 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 0 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 2 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 4 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 6 * alpha) >> WARPEDDIFF_PREC_BITS)));
// coeffs 0 1 0 1 2 3 2 3 for pixels 0, 2
__m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
const __m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
// coeffs 0 1 0 1 2 3 2 3 for pixels 4, 6
__m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
const __m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
// coeffs 4 5 4 5 6 7 6 7 for pixels 0, 2
__m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
const __m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
// coeffs 4 5 4 5 6 7 6 7 for pixels 4, 6
__m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
const __m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
// coeffs 0 1 0 1 0 1 0 1 for pixels 0, 2, 4, 6
__m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
const __m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
// coeffs 2 3 2 3 2 3 2 3 for pixels 0, 2, 4, 6
__m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
const __m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
// coeffs 4 5 4 5 4 5 4 5 for pixels 0, 2, 4, 6
__m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
const __m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
// coeffs 6 7 6 7 6 7 6 7 for pixels 0, 2, 4, 6
__m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
const __m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
__m128i round_const =
_mm_set1_epi32((1 << HORSHEAR_REDUCE_PREC_BITS) >> 1);
const __m128i round_const =
_mm_set1_epi32((1 << (bd + WARPEDPIXEL_FILTER_BITS - 1)) +
((1 << HORSHEAR_REDUCE_PREC_BITS) >> 1));
// Calculate filtered results
__m128i src_0 = _mm_unpacklo_epi8(src, zero);
__m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
__m128i src_2 = _mm_unpacklo_epi8(_mm_srli_si128(src, 2), zero);
__m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
__m128i src_4 = _mm_unpacklo_epi8(_mm_srli_si128(src, 4), zero);
__m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
__m128i src_6 = _mm_unpacklo_epi8(_mm_srli_si128(src, 6), zero);
__m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
const __m128i src_0 = _mm_unpacklo_epi8(src, zero);
const __m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
const __m128i src_2 = _mm_unpacklo_epi8(_mm_srli_si128(src, 2), zero);
const __m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
const __m128i src_4 = _mm_unpacklo_epi8(_mm_srli_si128(src, 4), zero);
const __m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
const __m128i src_6 = _mm_unpacklo_epi8(_mm_srli_si128(src, 6), zero);
const __m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
__m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
_mm_add_epi32(res_2, res_6));
@ -144,33 +168,37 @@ void av1_warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
HORSHEAR_REDUCE_PREC_BITS);
// Filter odd-index pixels
__m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 1 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 3 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 5 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(filter + ((sx + 7 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 1 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 3 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 5 * alpha) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sx + 7 * alpha) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
__m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
__m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
__m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
const __m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
const __m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
__m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
__m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
__m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
__m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
const __m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
const __m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
const __m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
const __m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
__m128i src_1 = _mm_unpacklo_epi8(_mm_srli_si128(src, 1), zero);
__m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
__m128i src_3 = _mm_unpacklo_epi8(_mm_srli_si128(src, 3), zero);
__m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
__m128i src_5 = _mm_unpacklo_epi8(_mm_srli_si128(src, 5), zero);
__m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
__m128i src_7 = _mm_unpacklo_epi8(_mm_srli_si128(src, 7), zero);
__m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
const __m128i src_1 = _mm_unpacklo_epi8(_mm_srli_si128(src, 1), zero);
const __m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
const __m128i src_3 = _mm_unpacklo_epi8(_mm_srli_si128(src, 3), zero);
const __m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
const __m128i src_5 = _mm_unpacklo_epi8(_mm_srli_si128(src, 5), zero);
const __m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
const __m128i src_7 = _mm_unpacklo_epi8(_mm_srli_si128(src, 7), zero);
const __m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
__m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
_mm_add_epi32(res_3, res_7));
@ -186,109 +214,116 @@ void av1_warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
// Vertical filter
for (k = -4; k < AOMMIN(4, p_height - i - 4); ++k) {
int sy = sy4 + gamma * (-4) + delta * k +
(1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
int sy = sy4 + delta * (k + 4);
// Load from tmp and rearrange pairs of consecutive rows into the
// column order 0 0 2 2 4 4 6 6; 1 1 3 3 5 5 7 7
__m128i *src = tmp + (k + 4);
__m128i src_0 = _mm_unpacklo_epi16(src[0], src[1]);
__m128i src_2 = _mm_unpacklo_epi16(src[2], src[3]);
__m128i src_4 = _mm_unpacklo_epi16(src[4], src[5]);
__m128i src_6 = _mm_unpacklo_epi16(src[6], src[7]);
const __m128i *src = tmp + (k + 4);
const __m128i src_0 = _mm_unpacklo_epi16(src[0], src[1]);
const __m128i src_2 = _mm_unpacklo_epi16(src[2], src[3]);
const __m128i src_4 = _mm_unpacklo_epi16(src[4], src[5]);
const __m128i src_6 = _mm_unpacklo_epi16(src[6], src[7]);
// Filter even-index pixels
__m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 0 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 2 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 4 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 6 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 0 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 2 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 4 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 6 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
__m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
__m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
__m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
const __m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
const __m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
const __m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
const __m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
__m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
__m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
__m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
__m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
const __m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
const __m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
const __m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
const __m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
__m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
__m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
__m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
__m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
const __m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
const __m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
const __m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
const __m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
__m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
_mm_add_epi32(res_4, res_6));
const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
_mm_add_epi32(res_4, res_6));
// Filter odd-index pixels
__m128i src_1 = _mm_unpackhi_epi16(src[0], src[1]);
__m128i src_3 = _mm_unpackhi_epi16(src[2], src[3]);
__m128i src_5 = _mm_unpackhi_epi16(src[4], src[5]);
__m128i src_7 = _mm_unpackhi_epi16(src[6], src[7]);
const __m128i src_1 = _mm_unpackhi_epi16(src[0], src[1]);
const __m128i src_3 = _mm_unpackhi_epi16(src[2], src[3]);
const __m128i src_5 = _mm_unpackhi_epi16(src[4], src[5]);
const __m128i src_7 = _mm_unpackhi_epi16(src[6], src[7]);
__m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 1 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 3 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 5 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(filter + ((sy + 7 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 1 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 3 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 5 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 7 * gamma) >> WARPEDDIFF_PREC_BITS)));
__m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
__m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
__m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
__m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
const __m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
const __m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
__m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
__m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
__m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
__m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
const __m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
const __m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
const __m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
const __m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
__m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
__m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
__m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
__m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
const __m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
const __m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
const __m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
const __m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
__m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
_mm_add_epi32(res_5, res_7));
const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
_mm_add_epi32(res_5, res_7));
// Rearrange pixels back into the order 0 ... 7
__m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
__m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
// Round and pack into 8 bits
__m128i round_const =
_mm_set1_epi32((1 << VERSHEAR_REDUCE_PREC_BITS) >> 1);
const __m128i round_const =
_mm_set1_epi32(-(1 << (bd + VERSHEAR_REDUCE_PREC_BITS - 1)) +
((1 << VERSHEAR_REDUCE_PREC_BITS) >> 1));
__m128i res_lo_round = _mm_srai_epi32(
const __m128i res_lo_round = _mm_srai_epi32(
_mm_add_epi32(res_lo, round_const), VERSHEAR_REDUCE_PREC_BITS);
__m128i res_hi_round = _mm_srai_epi32(
const __m128i res_hi_round = _mm_srai_epi32(
_mm_add_epi32(res_hi, round_const), VERSHEAR_REDUCE_PREC_BITS);
__m128i res_16bit = _mm_packs_epi32(res_lo_round, res_hi_round);
const __m128i res_16bit = _mm_packs_epi32(res_lo_round, res_hi_round);
__m128i res_8bit = _mm_packus_epi16(res_16bit, res_16bit);
// Store, blending with 'pred' if needed
__m128i *p = (__m128i *)&pred[(i + k + 4) * p_stride + j];
__m128i *const p = (__m128i *)&pred[(i + k + 4) * p_stride + j];
// Note: If we're outputting a 4x4 block, we need to be very careful
// to only output 4 pixels at this point, to avoid encode/decode
// mismatches when encoding with multiple threads.
if (p_width == 4) {
if (ref_frm) {
if (comp_avg) {
const __m128i orig = _mm_cvtsi32_si128(*(uint32_t *)p);
res_8bit = _mm_avg_epu8(res_8bit, orig);
}
*(uint32_t *)p = _mm_cvtsi128_si32(res_8bit);
} else {
if (ref_frm) res_8bit = _mm_avg_epu8(res_8bit, _mm_loadl_epi64(p));
if (comp_avg) res_8bit = _mm_avg_epu8(res_8bit, _mm_loadl_epi64(p));
_mm_storel_epi64(p, res_8bit);
}
}

508
third_party/aom/av1/common/x86/warp_plane_ssse3.c vendored Normal file
View file

@ -0,0 +1,508 @@
/*
* 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 <tmmintrin.h>
#include "./av1_rtcd.h"
#include "av1/common/warped_motion.h"
/* This is a modified version of 'warped_filter' from warped_motion.c:
* Each coefficient is stored in 8 bits instead of 16 bits
* The coefficients are rearranged in the column order 0, 2, 4, 6, 1, 3, 5, 7
This is done in order to avoid overflow: Since the tap with the largest
coefficient could be any of taps 2, 3, 4 or 5, we can't use the summation
order ((0 + 1) + (4 + 5)) + ((2 + 3) + (6 + 7)) used in the regular
convolve functions.
Instead, we use the summation order
((0 + 2) + (4 + 6)) + ((1 + 3) + (5 + 7)).
The rearrangement of coefficients in this table is so that we can get the
coefficients into the correct order more quickly.
*/
/* clang-format off */
DECLARE_ALIGNED(8, static const int8_t,
filter_8bit[WARPEDPIXEL_PREC_SHIFTS * 3 + 1][8]) = {
#if WARPEDPIXEL_PREC_BITS == 6
// [-1, 0)
{ 0, 127, 0, 0, 0, 1, 0, 0}, { 0, 127, 0, 0, -1, 2, 0, 0},
{ 1, 127, -1, 0, -3, 4, 0, 0}, { 1, 126, -2, 0, -4, 6, 1, 0},
{ 1, 126, -3, 0, -5, 8, 1, 0}, { 1, 125, -4, 0, -6, 11, 1, 0},
{ 1, 124, -4, 0, -7, 13, 1, 0}, { 2, 123, -5, 0, -8, 15, 1, 0},
{ 2, 122, -6, 0, -9, 18, 1, 0}, { 2, 121, -6, 0, -10, 20, 1, 0},
{ 2, 120, -7, 0, -11, 22, 2, 0}, { 2, 119, -8, 0, -12, 25, 2, 0},
{ 3, 117, -8, 0, -13, 27, 2, 0}, { 3, 116, -9, 0, -13, 29, 2, 0},
{ 3, 114, -10, 0, -14, 32, 3, 0}, { 3, 113, -10, 0, -15, 35, 2, 0},
{ 3, 111, -11, 0, -15, 37, 3, 0}, { 3, 109, -11, 0, -16, 40, 3, 0},
{ 3, 108, -12, 0, -16, 42, 3, 0}, { 4, 106, -13, 0, -17, 45, 3, 0},
{ 4, 104, -13, 0, -17, 47, 3, 0}, { 4, 102, -14, 0, -17, 50, 3, 0},
{ 4, 100, -14, 0, -17, 52, 3, 0}, { 4, 98, -15, 0, -18, 55, 4, 0},
{ 4, 96, -15, 0, -18, 58, 3, 0}, { 4, 94, -16, 0, -18, 60, 4, 0},
{ 4, 91, -16, 0, -18, 63, 4, 0}, { 4, 89, -16, 0, -18, 65, 4, 0},
{ 4, 87, -17, 0, -18, 68, 4, 0}, { 4, 85, -17, 0, -18, 70, 4, 0},
{ 4, 82, -17, 0, -18, 73, 4, 0}, { 4, 80, -17, 0, -18, 75, 4, 0},
{ 4, 78, -18, 0, -18, 78, 4, 0}, { 4, 75, -18, 0, -17, 80, 4, 0},
{ 4, 73, -18, 0, -17, 82, 4, 0}, { 4, 70, -18, 0, -17, 85, 4, 0},
{ 4, 68, -18, 0, -17, 87, 4, 0}, { 4, 65, -18, 0, -16, 89, 4, 0},
{ 4, 63, -18, 0, -16, 91, 4, 0}, { 4, 60, -18, 0, -16, 94, 4, 0},
{ 3, 58, -18, 0, -15, 96, 4, 0}, { 4, 55, -18, 0, -15, 98, 4, 0},
{ 3, 52, -17, 0, -14, 100, 4, 0}, { 3, 50, -17, 0, -14, 102, 4, 0},
{ 3, 47, -17, 0, -13, 104, 4, 0}, { 3, 45, -17, 0, -13, 106, 4, 0},
{ 3, 42, -16, 0, -12, 108, 3, 0}, { 3, 40, -16, 0, -11, 109, 3, 0},
{ 3, 37, -15, 0, -11, 111, 3, 0}, { 2, 35, -15, 0, -10, 113, 3, 0},
{ 3, 32, -14, 0, -10, 114, 3, 0}, { 2, 29, -13, 0, -9, 116, 3, 0},
{ 2, 27, -13, 0, -8, 117, 3, 0}, { 2, 25, -12, 0, -8, 119, 2, 0},
{ 2, 22, -11, 0, -7, 120, 2, 0}, { 1, 20, -10, 0, -6, 121, 2, 0},
{ 1, 18, -9, 0, -6, 122, 2, 0}, { 1, 15, -8, 0, -5, 123, 2, 0},
{ 1, 13, -7, 0, -4, 124, 1, 0}, { 1, 11, -6, 0, -4, 125, 1, 0},
{ 1, 8, -5, 0, -3, 126, 1, 0}, { 1, 6, -4, 0, -2, 126, 1, 0},
{ 0, 4, -3, 0, -1, 127, 1, 0}, { 0, 2, -1, 0, 0, 127, 0, 0},
// [0, 1)
{ 0, 0, 1, 0, 0, 127, 0, 0}, { 0, -1, 2, 0, 0, 127, 0, 0},
{ 0, -3, 4, 1, 1, 127, -2, 0}, { 0, -5, 6, 1, 1, 127, -2, 0},
{ 0, -6, 8, 1, 2, 126, -3, 0}, {-1, -7, 11, 2, 2, 126, -4, -1},
{-1, -8, 13, 2, 3, 125, -5, -1}, {-1, -10, 16, 3, 3, 124, -6, -1},
{-1, -11, 18, 3, 4, 123, -7, -1}, {-1, -12, 20, 3, 4, 122, -7, -1},
{-1, -13, 23, 3, 4, 121, -8, -1}, {-2, -14, 25, 4, 5, 120, -9, -1},
{-1, -15, 27, 4, 5, 119, -10, -1}, {-1, -16, 30, 4, 5, 118, -11, -1},
{-2, -17, 33, 5, 6, 116, -12, -1}, {-2, -17, 35, 5, 6, 114, -12, -1},
{-2, -18, 38, 5, 6, 113, -13, -1}, {-2, -19, 41, 6, 7, 111, -14, -2},
{-2, -19, 43, 6, 7, 110, -15, -2}, {-2, -20, 46, 6, 7, 108, -15, -2},
{-2, -20, 49, 6, 7, 106, -16, -2}, {-2, -21, 51, 7, 7, 104, -16, -2},
{-2, -21, 54, 7, 7, 102, -17, -2}, {-2, -21, 56, 7, 8, 100, -18, -2},
{-2, -22, 59, 7, 8, 98, -18, -2}, {-2, -22, 62, 7, 8, 96, -19, -2},
{-2, -22, 64, 7, 8, 94, -19, -2}, {-2, -22, 67, 8, 8, 91, -20, -2},
{-2, -22, 69, 8, 8, 89, -20, -2}, {-2, -22, 72, 8, 8, 87, -21, -2},
{-2, -21, 74, 8, 8, 84, -21, -2}, {-2, -22, 77, 8, 8, 82, -21, -2},
{-2, -21, 79, 8, 8, 79, -21, -2}, {-2, -21, 82, 8, 8, 77, -22, -2},
{-2, -21, 84, 8, 8, 74, -21, -2}, {-2, -21, 87, 8, 8, 72, -22, -2},
{-2, -20, 89, 8, 8, 69, -22, -2}, {-2, -20, 91, 8, 8, 67, -22, -2},
{-2, -19, 94, 8, 7, 64, -22, -2}, {-2, -19, 96, 8, 7, 62, -22, -2},
{-2, -18, 98, 8, 7, 59, -22, -2}, {-2, -18, 100, 8, 7, 56, -21, -2},
{-2, -17, 102, 7, 7, 54, -21, -2}, {-2, -16, 104, 7, 7, 51, -21, -2},
{-2, -16, 106, 7, 6, 49, -20, -2}, {-2, -15, 108, 7, 6, 46, -20, -2},
{-2, -15, 110, 7, 6, 43, -19, -2}, {-2, -14, 111, 7, 6, 41, -19, -2},
{-1, -13, 113, 6, 5, 38, -18, -2}, {-1, -12, 114, 6, 5, 35, -17, -2},
{-1, -12, 116, 6, 5, 33, -17, -2}, {-1, -11, 118, 5, 4, 30, -16, -1},
{-1, -10, 119, 5, 4, 27, -15, -1}, {-1, -9, 120, 5, 4, 25, -14, -2},
{-1, -8, 121, 4, 3, 23, -13, -1}, {-1, -7, 122, 4, 3, 20, -12, -1},
{-1, -7, 123, 4, 3, 18, -11, -1}, {-1, -6, 124, 3, 3, 16, -10, -1},
{-1, -5, 125, 3, 2, 13, -8, -1}, {-1, -4, 126, 2, 2, 11, -7, -1},
{ 0, -3, 126, 2, 1, 8, -6, 0}, { 0, -2, 127, 1, 1, 6, -5, 0},
{ 0, -2, 127, 1, 1, 4, -3, 0}, { 0, 0, 127, 0, 0, 2, -1, 0},
// [1, 2)
{ 0, 0, 127, 0, 0, 1, 0, 0}, { 0, 0, 127, 0, 0, -1, 2, 0},
{ 0, 1, 127, -1, 0, -3, 4, 0}, { 0, 1, 126, -2, 0, -4, 6, 1},
{ 0, 1, 126, -3, 0, -5, 8, 1}, { 0, 1, 125, -4, 0, -6, 11, 1},
{ 0, 1, 124, -4, 0, -7, 13, 1}, { 0, 2, 123, -5, 0, -8, 15, 1},
{ 0, 2, 122, -6, 0, -9, 18, 1}, { 0, 2, 121, -6, 0, -10, 20, 1},
{ 0, 2, 120, -7, 0, -11, 22, 2}, { 0, 2, 119, -8, 0, -12, 25, 2},
{ 0, 3, 117, -8, 0, -13, 27, 2}, { 0, 3, 116, -9, 0, -13, 29, 2},
{ 0, 3, 114, -10, 0, -14, 32, 3}, { 0, 3, 113, -10, 0, -15, 35, 2},
{ 0, 3, 111, -11, 0, -15, 37, 3}, { 0, 3, 109, -11, 0, -16, 40, 3},
{ 0, 3, 108, -12, 0, -16, 42, 3}, { 0, 4, 106, -13, 0, -17, 45, 3},
{ 0, 4, 104, -13, 0, -17, 47, 3}, { 0, 4, 102, -14, 0, -17, 50, 3},
{ 0, 4, 100, -14, 0, -17, 52, 3}, { 0, 4, 98, -15, 0, -18, 55, 4},
{ 0, 4, 96, -15, 0, -18, 58, 3}, { 0, 4, 94, -16, 0, -18, 60, 4},
{ 0, 4, 91, -16, 0, -18, 63, 4}, { 0, 4, 89, -16, 0, -18, 65, 4},
{ 0, 4, 87, -17, 0, -18, 68, 4}, { 0, 4, 85, -17, 0, -18, 70, 4},
{ 0, 4, 82, -17, 0, -18, 73, 4}, { 0, 4, 80, -17, 0, -18, 75, 4},
{ 0, 4, 78, -18, 0, -18, 78, 4}, { 0, 4, 75, -18, 0, -17, 80, 4},
{ 0, 4, 73, -18, 0, -17, 82, 4}, { 0, 4, 70, -18, 0, -17, 85, 4},
{ 0, 4, 68, -18, 0, -17, 87, 4}, { 0, 4, 65, -18, 0, -16, 89, 4},
{ 0, 4, 63, -18, 0, -16, 91, 4}, { 0, 4, 60, -18, 0, -16, 94, 4},
{ 0, 3, 58, -18, 0, -15, 96, 4}, { 0, 4, 55, -18, 0, -15, 98, 4},
{ 0, 3, 52, -17, 0, -14, 100, 4}, { 0, 3, 50, -17, 0, -14, 102, 4},
{ 0, 3, 47, -17, 0, -13, 104, 4}, { 0, 3, 45, -17, 0, -13, 106, 4},
{ 0, 3, 42, -16, 0, -12, 108, 3}, { 0, 3, 40, -16, 0, -11, 109, 3},
{ 0, 3, 37, -15, 0, -11, 111, 3}, { 0, 2, 35, -15, 0, -10, 113, 3},
{ 0, 3, 32, -14, 0, -10, 114, 3}, { 0, 2, 29, -13, 0, -9, 116, 3},
{ 0, 2, 27, -13, 0, -8, 117, 3}, { 0, 2, 25, -12, 0, -8, 119, 2},
{ 0, 2, 22, -11, 0, -7, 120, 2}, { 0, 1, 20, -10, 0, -6, 121, 2},
{ 0, 1, 18, -9, 0, -6, 122, 2}, { 0, 1, 15, -8, 0, -5, 123, 2},
{ 0, 1, 13, -7, 0, -4, 124, 1}, { 0, 1, 11, -6, 0, -4, 125, 1},
{ 0, 1, 8, -5, 0, -3, 126, 1}, { 0, 1, 6, -4, 0, -2, 126, 1},
{ 0, 0, 4, -3, 0, -1, 127, 1}, { 0, 0, 2, -1, 0, 0, 127, 0},
// dummy (replicate row index 191)
{ 0, 0, 2, -1, 0, 0, 127, 0},
#else
// [-1, 0)
{ 0, 127, 0, 0, 0, 1, 0, 0}, { 1, 127, -1, 0, -3, 4, 0, 0},
{ 1, 126, -3, 0, -5, 8, 1, 0}, { 1, 124, -4, 0, -7, 13, 1, 0},
{ 2, 122, -6, 0, -9, 18, 1, 0}, { 2, 120, -7, 0, -11, 22, 2, 0},
{ 3, 117, -8, 0, -13, 27, 2, 0}, { 3, 114, -10, 0, -14, 32, 3, 0},
{ 3, 111, -11, 0, -15, 37, 3, 0}, { 3, 108, -12, 0, -16, 42, 3, 0},
{ 4, 104, -13, 0, -17, 47, 3, 0}, { 4, 100, -14, 0, -17, 52, 3, 0},
{ 4, 96, -15, 0, -18, 58, 3, 0}, { 4, 91, -16, 0, -18, 63, 4, 0},
{ 4, 87, -17, 0, -18, 68, 4, 0}, { 4, 82, -17, 0, -18, 73, 4, 0},
{ 4, 78, -18, 0, -18, 78, 4, 0}, { 4, 73, -18, 0, -17, 82, 4, 0},
{ 4, 68, -18, 0, -17, 87, 4, 0}, { 4, 63, -18, 0, -16, 91, 4, 0},
{ 3, 58, -18, 0, -15, 96, 4, 0}, { 3, 52, -17, 0, -14, 100, 4, 0},
{ 3, 47, -17, 0, -13, 104, 4, 0}, { 3, 42, -16, 0, -12, 108, 3, 0},
{ 3, 37, -15, 0, -11, 111, 3, 0}, { 3, 32, -14, 0, -10, 114, 3, 0},
{ 2, 27, -13, 0, -8, 117, 3, 0}, { 2, 22, -11, 0, -7, 120, 2, 0},
{ 1, 18, -9, 0, -6, 122, 2, 0}, { 1, 13, -7, 0, -4, 124, 1, 0},
{ 1, 8, -5, 0, -3, 126, 1, 0}, { 0, 4, -3, 0, -1, 127, 1, 0},
// [0, 1)
{ 0, 0, 1, 0, 0, 127, 0, 0}, { 0, -3, 4, 1, 1, 127, -2, 0},
{ 0, -6, 8, 1, 2, 126, -3, 0}, {-1, -8, 13, 2, 3, 125, -5, -1},
{-1, -11, 18, 3, 4, 123, -7, -1}, {-1, -13, 23, 3, 4, 121, -8, -1},
{-1, -15, 27, 4, 5, 119, -10, -1}, {-2, -17, 33, 5, 6, 116, -12, -1},
{-2, -18, 38, 5, 6, 113, -13, -1}, {-2, -19, 43, 6, 7, 110, -15, -2},
{-2, -20, 49, 6, 7, 106, -16, -2}, {-2, -21, 54, 7, 7, 102, -17, -2},
{-2, -22, 59, 7, 8, 98, -18, -2}, {-2, -22, 64, 7, 8, 94, -19, -2},
{-2, -22, 69, 8, 8, 89, -20, -2}, {-2, -21, 74, 8, 8, 84, -21, -2},
{-2, -21, 79, 8, 8, 79, -21, -2}, {-2, -21, 84, 8, 8, 74, -21, -2},
{-2, -20, 89, 8, 8, 69, -22, -2}, {-2, -19, 94, 8, 7, 64, -22, -2},
{-2, -18, 98, 8, 7, 59, -22, -2}, {-2, -17, 102, 7, 7, 54, -21, -2},
{-2, -16, 106, 7, 6, 49, -20, -2}, {-2, -15, 110, 7, 6, 43, -19, -2},
{-1, -13, 113, 6, 5, 38, -18, -2}, {-1, -12, 116, 6, 5, 33, -17, -2},
{-1, -10, 119, 5, 4, 27, -15, -1}, {-1, -8, 121, 4, 3, 23, -13, -1},
{-1, -7, 123, 4, 3, 18, -11, -1}, {-1, -5, 125, 3, 2, 13, -8, -1},
{ 0, -3, 126, 2, 1, 8, -6, 0}, { 0, -2, 127, 1, 1, 4, -3, 0},
// [1, 2)
{ 0, 0, 127, 0, 0, 1, 0, 0}, { 0, 1, 127, -1, 0, -3, 4, 0},
{ 0, 1, 126, -3, 0, -5, 8, 1}, { 0, 1, 124, -4, 0, -7, 13, 1},
{ 0, 2, 122, -6, 0, -9, 18, 1}, { 0, 2, 120, -7, 0, -11, 22, 2},
{ 0, 3, 117, -8, 0, -13, 27, 2}, { 0, 3, 114, -10, 0, -14, 32, 3},
{ 0, 3, 111, -11, 0, -15, 37, 3}, { 0, 3, 108, -12, 0, -16, 42, 3},
{ 0, 4, 104, -13, 0, -17, 47, 3}, { 0, 4, 100, -14, 0, -17, 52, 3},
{ 0, 4, 96, -15, 0, -18, 58, 3}, { 0, 4, 91, -16, 0, -18, 63, 4},
{ 0, 4, 87, -17, 0, -18, 68, 4}, { 0, 4, 82, -17, 0, -18, 73, 4},
{ 0, 4, 78, -18, 0, -18, 78, 4}, { 0, 4, 73, -18, 0, -17, 82, 4},
{ 0, 4, 68, -18, 0, -17, 87, 4}, { 0, 4, 63, -18, 0, -16, 91, 4},
{ 0, 3, 58, -18, 0, -15, 96, 4}, { 0, 3, 52, -17, 0, -14, 100, 4},
{ 0, 3, 47, -17, 0, -13, 104, 4}, { 0, 3, 42, -16, 0, -12, 108, 3},
{ 0, 3, 37, -15, 0, -11, 111, 3}, { 0, 3, 32, -14, 0, -10, 114, 3},
{ 0, 2, 27, -13, 0, -8, 117, 3}, { 0, 2, 22, -11, 0, -7, 120, 2},
{ 0, 1, 18, -9, 0, -6, 122, 2}, { 0, 1, 13, -7, 0, -4, 124, 1},
{ 0, 1, 8, -5, 0, -3, 126, 1}, { 0, 0, 4, -3, 0, -1, 127, 1},
// dummy (replicate row index 95)
{ 0, 0, 4, -3, 0, -1, 127, 1},
#endif // WARPEDPIXEL_PREC_BITS == 6
};
/* clang-format on */
// Shuffle masks: we want to convert a sequence of bytes 0, 1, 2, ..., 15
// in an SSE register into two sequences:
// 0, 2, 2, 4, ..., 12, 12, 14, <don't care>
// 1, 3, 3, 5, ..., 13, 13, 15, <don't care>
static const uint8_t even_mask[16] = { 0, 2, 2, 4, 4, 6, 6, 8,
8, 10, 10, 12, 12, 14, 14, 0 };
static const uint8_t odd_mask[16] = { 1, 3, 3, 5, 5, 7, 7, 9,
9, 11, 11, 13, 13, 15, 15, 0 };
void av1_warp_affine_ssse3(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, int comp_avg,
int16_t alpha, int16_t beta, int16_t gamma,
int16_t delta) {
__m128i tmp[15];
int i, j, k;
const int bd = 8;
/* Note: For this code to work, the left/right frame borders need to be
extended by at least 13 pixels each. By the time we get here, other
code will have set up this border, but we allow an explicit check
for debugging purposes.
*/
/*for (i = 0; i < height; ++i) {
for (j = 0; j < 13; ++j) {
assert(ref[i * stride - 13 + j] == ref[i * stride]);
assert(ref[i * stride + width + j] == ref[i * stride + (width - 1)]);
}
}*/
for (i = 0; i < p_height; i += 8) {
for (j = 0; j < p_width; j += 8) {
// (x, y) coordinates of the center of this block in the destination
// image
const int32_t dst_x = p_col + j + 4;
const int32_t dst_y = p_row + i + 4;
int32_t x4, y4, ix4, sx4, iy4, sy4;
if (subsampling_x)
x4 = (mat[2] * 4 * dst_x + mat[3] * 4 * dst_y + mat[0] * 2 +
(mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
x4 = mat[2] * dst_x + mat[3] * dst_y + mat[0];
if (subsampling_y)
y4 = (mat[4] * 4 * dst_x + mat[5] * 4 * dst_y + mat[1] * 2 +
(mat[4] + mat[5] - (1 << WARPEDMODEL_PREC_BITS))) /
4;
else
y4 = mat[4] * dst_x + mat[5] * dst_y + mat[1];
ix4 = x4 >> WARPEDMODEL_PREC_BITS;
sx4 = x4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
iy4 = y4 >> WARPEDMODEL_PREC_BITS;
sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
// Add in all the constant terms, including rounding and offset
sx4 += alpha * (-4) + beta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
sy4 += gamma * (-4) + delta * (-4) + (1 << (WARPEDDIFF_PREC_BITS - 1)) +
(WARPEDPIXEL_PREC_SHIFTS << WARPEDDIFF_PREC_BITS);
sx4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
// Horizontal filter
// If the block is aligned such that, after clamping, every sample
// would be taken from the leftmost/rightmost column, then we can
// skip the expensive horizontal filter.
if (ix4 <= -7) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else if (ix4 >= width + 6) {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
tmp[k + 7] = _mm_set1_epi16(
(1 << (bd + WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS -
1)) +
ref[iy * stride + (width - 1)] *
(1 << (WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)));
}
} else {
for (k = -7; k < AOMMIN(8, p_height - i); ++k) {
int iy = iy4 + k;
if (iy < 0)
iy = 0;
else if (iy > height - 1)
iy = height - 1;
int sx = sx4 + beta * (k + 4);
// Load source pixels
const __m128i src =
_mm_loadu_si128((__m128i *)(ref + iy * stride + ix4 - 7));
const __m128i src_even =
_mm_shuffle_epi8(src, _mm_loadu_si128((__m128i *)even_mask));
const __m128i src_odd =
_mm_shuffle_epi8(src, _mm_loadu_si128((__m128i *)odd_mask));
// Filter even-index pixels
const __m128i tmp_0 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 0 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_1 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 1 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_2 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 2 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_3 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 3 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_4 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 4 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_5 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 5 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_6 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 6 * alpha) >> WARPEDDIFF_PREC_BITS]);
const __m128i tmp_7 = _mm_loadl_epi64((
__m128i *)&filter_8bit[(sx + 7 * alpha) >> WARPEDDIFF_PREC_BITS]);
// Coeffs 0 2 0 2 4 6 4 6 1 3 1 3 5 7 5 7 for pixels 0 2
const __m128i tmp_8 = _mm_unpacklo_epi16(tmp_0, tmp_2);
// Coeffs 0 2 0 2 4 6 4 6 1 3 1 3 5 7 5 7 for pixels 1 3
const __m128i tmp_9 = _mm_unpacklo_epi16(tmp_1, tmp_3);
// Coeffs 0 2 0 2 4 6 4 6 1 3 1 3 5 7 5 7 for pixels 4 6
const __m128i tmp_10 = _mm_unpacklo_epi16(tmp_4, tmp_6);
// Coeffs 0 2 0 2 4 6 4 6 1 3 1 3 5 7 5 7 for pixels 5 7
const __m128i tmp_11 = _mm_unpacklo_epi16(tmp_5, tmp_7);
// Coeffs 0 2 0 2 0 2 0 2 4 6 4 6 4 6 4 6 for pixels 0 2 4 6
const __m128i tmp_12 = _mm_unpacklo_epi32(tmp_8, tmp_10);
// Coeffs 1 3 1 3 1 3 1 3 5 7 5 7 5 7 5 7 for pixels 0 2 4 6
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_8, tmp_10);
// Coeffs 0 2 0 2 0 2 0 2 4 6 4 6 4 6 4 6 for pixels 1 3 5 7
const __m128i tmp_14 = _mm_unpacklo_epi32(tmp_9, tmp_11);
// Coeffs 1 3 1 3 1 3 1 3 5 7 5 7 5 7 5 7 for pixels 1 3 5 7
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_9, tmp_11);
// Coeffs 0 2 for pixels 0 2 4 6 1 3 5 7
const __m128i coeff_02 = _mm_unpacklo_epi64(tmp_12, tmp_14);
// Coeffs 4 6 for pixels 0 2 4 6 1 3 5 7
const __m128i coeff_46 = _mm_unpackhi_epi64(tmp_12, tmp_14);
// Coeffs 1 3 for pixels 0 2 4 6 1 3 5 7
const __m128i coeff_13 = _mm_unpacklo_epi64(tmp_13, tmp_15);
// Coeffs 5 7 for pixels 0 2 4 6 1 3 5 7
const __m128i coeff_57 = _mm_unpackhi_epi64(tmp_13, tmp_15);
// The pixel order we need for 'src' is:
// 0 2 2 4 4 6 6 8 1 3 3 5 5 7 7 9
const __m128i src_02 = _mm_unpacklo_epi64(src_even, src_odd);
const __m128i res_02 = _mm_maddubs_epi16(src_02, coeff_02);
// 4 6 6 8 8 10 10 12 5 7 7 9 9 11 11 13
const __m128i src_46 = _mm_unpacklo_epi64(_mm_srli_si128(src_even, 4),
_mm_srli_si128(src_odd, 4));
const __m128i res_46 = _mm_maddubs_epi16(src_46, coeff_46);
// 1 3 3 5 5 7 7 9 2 4 4 6 6 8 8 10
const __m128i src_13 =
_mm_unpacklo_epi64(src_odd, _mm_srli_si128(src_even, 2));
const __m128i res_13 = _mm_maddubs_epi16(src_13, coeff_13);
// 5 7 7 9 9 11 11 13 6 8 8 10 10 12 12 14
const __m128i src_57 = _mm_unpacklo_epi64(
_mm_srli_si128(src_odd, 4), _mm_srli_si128(src_even, 6));
const __m128i res_57 = _mm_maddubs_epi16(src_57, coeff_57);
const __m128i round_const =
_mm_set1_epi16((1 << (bd + WARPEDPIXEL_FILTER_BITS - 1)) +
((1 << HORSHEAR_REDUCE_PREC_BITS) >> 1));
// Note: The values res_02 + res_46 and res_13 + res_57 both
// fit into int16s at this point, but their sum may be too wide to fit
// into an int16. However, once we also add round_const, the sum of
// all of these fits into a uint16.
//
// The wrapping behaviour of _mm_add_* is used here to make sure we
// get the correct result despite converting between different
// (implicit) types.
const __m128i res_even = _mm_add_epi16(res_02, res_46);
const __m128i res_odd = _mm_add_epi16(res_13, res_57);
const __m128i res =
_mm_add_epi16(_mm_add_epi16(res_even, res_odd), round_const);
tmp[k + 7] = _mm_srli_epi16(res, HORSHEAR_REDUCE_PREC_BITS);
}
}
// Vertical filter
for (k = -4; k < AOMMIN(4, p_height - i - 4); ++k) {
int sy = sy4 + delta * (k + 4);
// Load from tmp and rearrange pairs of consecutive rows into the
// column order 0 0 2 2 4 4 6 6; 1 1 3 3 5 5 7 7
const __m128i *src = tmp + (k + 4);
const __m128i src_0 = _mm_unpacklo_epi16(src[0], src[1]);
const __m128i src_2 = _mm_unpacklo_epi16(src[2], src[3]);
const __m128i src_4 = _mm_unpacklo_epi16(src[4], src[5]);
const __m128i src_6 = _mm_unpacklo_epi16(src[6], src[7]);
// Filter even-index pixels
const __m128i tmp_0 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 0 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_2 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 2 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_4 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 4 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_6 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 6 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_8 = _mm_unpacklo_epi32(tmp_0, tmp_2);
const __m128i tmp_10 = _mm_unpacklo_epi32(tmp_4, tmp_6);
const __m128i tmp_12 = _mm_unpackhi_epi32(tmp_0, tmp_2);
const __m128i tmp_14 = _mm_unpackhi_epi32(tmp_4, tmp_6);
const __m128i coeff_0 = _mm_unpacklo_epi64(tmp_8, tmp_10);
const __m128i coeff_2 = _mm_unpackhi_epi64(tmp_8, tmp_10);
const __m128i coeff_4 = _mm_unpacklo_epi64(tmp_12, tmp_14);
const __m128i coeff_6 = _mm_unpackhi_epi64(tmp_12, tmp_14);
const __m128i res_0 = _mm_madd_epi16(src_0, coeff_0);
const __m128i res_2 = _mm_madd_epi16(src_2, coeff_2);
const __m128i res_4 = _mm_madd_epi16(src_4, coeff_4);
const __m128i res_6 = _mm_madd_epi16(src_6, coeff_6);
const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
_mm_add_epi32(res_4, res_6));
// Filter odd-index pixels
const __m128i src_1 = _mm_unpackhi_epi16(src[0], src[1]);
const __m128i src_3 = _mm_unpackhi_epi16(src[2], src[3]);
const __m128i src_5 = _mm_unpackhi_epi16(src[4], src[5]);
const __m128i src_7 = _mm_unpackhi_epi16(src[6], src[7]);
const __m128i tmp_1 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 1 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_3 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 3 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_5 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 5 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_7 = _mm_loadu_si128(
(__m128i *)(warped_filter +
((sy + 7 * gamma) >> WARPEDDIFF_PREC_BITS)));
const __m128i tmp_9 = _mm_unpacklo_epi32(tmp_1, tmp_3);
const __m128i tmp_11 = _mm_unpacklo_epi32(tmp_5, tmp_7);
const __m128i tmp_13 = _mm_unpackhi_epi32(tmp_1, tmp_3);
const __m128i tmp_15 = _mm_unpackhi_epi32(tmp_5, tmp_7);
const __m128i coeff_1 = _mm_unpacklo_epi64(tmp_9, tmp_11);
const __m128i coeff_3 = _mm_unpackhi_epi64(tmp_9, tmp_11);
const __m128i coeff_5 = _mm_unpacklo_epi64(tmp_13, tmp_15);
const __m128i coeff_7 = _mm_unpackhi_epi64(tmp_13, tmp_15);
const __m128i res_1 = _mm_madd_epi16(src_1, coeff_1);
const __m128i res_3 = _mm_madd_epi16(src_3, coeff_3);
const __m128i res_5 = _mm_madd_epi16(src_5, coeff_5);
const __m128i res_7 = _mm_madd_epi16(src_7, coeff_7);
const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
_mm_add_epi32(res_5, res_7));
// Rearrange pixels back into the order 0 ... 7
const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
// Round and pack into 8 bits
const __m128i round_const =
_mm_set1_epi32(-(1 << (bd + VERSHEAR_REDUCE_PREC_BITS - 1)) +
((1 << VERSHEAR_REDUCE_PREC_BITS) >> 1));
const __m128i res_lo_round = _mm_srai_epi32(
_mm_add_epi32(res_lo, round_const), VERSHEAR_REDUCE_PREC_BITS);
const __m128i res_hi_round = _mm_srai_epi32(
_mm_add_epi32(res_hi, round_const), VERSHEAR_REDUCE_PREC_BITS);
const __m128i res_16bit = _mm_packs_epi32(res_lo_round, res_hi_round);
__m128i res_8bit = _mm_packus_epi16(res_16bit, res_16bit);
// Store, blending with 'pred' if needed
__m128i *const p = (__m128i *)&pred[(i + k + 4) * p_stride + j];
// Note: If we're outputting a 4x4 block, we need to be very careful
// to only output 4 pixels at this point, to avoid encode/decode
// mismatches when encoding with multiple threads.
if (p_width == 4) {
if (comp_avg) {
const __m128i orig = _mm_cvtsi32_si128(*(uint32_t *)p);
res_8bit = _mm_avg_epu8(res_8bit, orig);
}
*(uint32_t *)p = _mm_cvtsi128_si32(res_8bit);
} else {
if (comp_avg) res_8bit = _mm_avg_epu8(res_8bit, _mm_loadl_epi64(p));
_mm_storel_epi64(p, res_8bit);
}
}
}
}
}

610
third_party/aom/av1/decoder/decodeframe.c vendored
View file

File diff suppressed because it is too large Load diff

650
third_party/aom/av1/decoder/decodemv.c vendored
View file

File diff suppressed because it is too large Load diff

2
third_party/aom/av1/decoder/decoder.c vendored
View file

@ -50,7 +50,6 @@ static void initialize_dec(void) {
av1_init_wedge_masks();
#endif // CONFIG_EXT_INTER
init_done = 1;
#if CONFIG_EC_MULTISYMBOL
av1_indices_from_tree(av1_intra_mode_ind, av1_intra_mode_inv,
av1_intra_mode_tree);
av1_indices_from_tree(av1_switchable_interp_ind, av1_switchable_interp_inv,
@ -68,7 +67,6 @@ static void initialize_dec(void) {
#endif
av1_indices_from_tree(av1_inter_mode_ind, av1_inter_mode_inv,
av1_inter_mode_tree);
#endif
}
}

2
third_party/aom/av1/decoder/decoder.h vendored
View file

@ -203,7 +203,7 @@ static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs,
}
}
#if CONFIG_EXT_REFS
#if CONFIG_EXT_REFS || CONFIG_TEMPMV_SIGNALING
static INLINE int dec_is_ref_frame_buf(AV1Decoder *const pbi,
RefCntBuffer *frame_buf) {
AV1_COMMON *const cm = &pbi->common;

2
third_party/aom/av1/decoder/decodetxb.c vendored
View file

@ -169,7 +169,7 @@ uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
sign = aom_read_bit(r, ACCT_STR);
}
ctx = get_level_ctx(tcoeffs, scan[c], bwl);
ctx = get_br_ctx(tcoeffs, scan[c], bwl);
if (cm->fc->coeff_lps[tx_size][plane_type][ctx] == 0) exit(0);

122
third_party/aom/av1/decoder/detokenize.c vendored
View file

@ -106,7 +106,7 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
dequant_val_type_nuq *dq_val,
#endif // CONFIG_NEW_QUANT
#if CONFIG_AOM_QM
const qm_val_t *iqm[2][TX_SIZES],
const qm_val_t *iqm[2][TX_SIZES_ALL],
#endif // CONFIG_AOM_QM
int ctx, const int16_t *scan, const int16_t *nb,
int16_t *max_scan_line, aom_reader *r) {
@ -123,7 +123,6 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
#endif // CONFIG_AOM_QM
int band, c = 0;
const int tx_size_ctx = txsize_sqr_map[tx_size];
#if CONFIG_NEW_TOKENSET
aom_cdf_prob(*coef_head_cdfs)[COEFF_CONTEXTS][CDF_SIZE(ENTROPY_TOKENS)] =
ec_ctx->coef_head_cdfs[tx_size_ctx][type][ref];
aom_cdf_prob(*coef_tail_cdfs)[COEFF_CONTEXTS][CDF_SIZE(ENTROPY_TOKENS)] =
@ -135,18 +134,6 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
unsigned int(*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1] = NULL;
unsigned int(*eob_branch_count)[COEFF_CONTEXTS] = NULL;
#endif
#else
aom_prob(*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
ec_ctx->coef_probs[tx_size_ctx][type][ref];
const aom_prob *prob;
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob(*coef_cdfs)[COEFF_CONTEXTS][CDF_SIZE(ENTROPY_TOKENS)] =
ec_ctx->coef_cdfs[tx_size_ctx][type][ref];
aom_cdf_prob(*cdf)[CDF_SIZE(ENTROPY_TOKENS)];
#endif // CONFIG_EC_MULTISYMBOL
unsigned int(*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1] = NULL;
unsigned int(*eob_branch_count)[COEFF_CONTEXTS] = NULL;
#endif // CONFIG_NEW_TOKENSET
uint8_t token_cache[MAX_TX_SQUARE];
const uint8_t *band_translate = get_band_translate(tx_size);
int dq_shift;
@ -156,23 +143,17 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
const tran_low_t *dqv_val = &dq_val[0][0];
#endif // CONFIG_NEW_QUANT
(void)tx_type;
#if CONFIG_AOM_QM
(void)iqmatrix;
#endif // CONFIG_AOM_QM
if (counts) {
#if !CONFIG_NEW_TOKENSET || !CONFIG_EC_ADAPT
#if !CONFIG_EC_ADAPT
coef_counts = counts->coef[tx_size_ctx][type][ref];
eob_branch_count = counts->eob_branch[tx_size_ctx][type][ref];
#endif
#if CONFIG_NEW_TOKENSET && !CONFIG_EC_ADAPT
blockz_count = counts->blockz_count[tx_size_ctx][type][ref][ctx];
#endif
}
dq_shift = av1_get_tx_scale(tx_size);
#if CONFIG_NEW_TOKENSET
band = *band_translate++;
int more_data = 1;
@ -238,12 +219,7 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
*max_scan_line = AOMMAX(*max_scan_line, scan[c]);
token_cache[scan[c]] = av1_pt_energy_class[token];
val = token_to_value(r, token, tx_size,
#if CONFIG_HIGHBITDEPTH
xd->bd);
#else
8);
#endif // CONFIG_HIGHBITDEPTH
val = token_to_value(r, token, tx_size, xd->bd);
#if CONFIG_NEW_QUANT
v = av1_dequant_abscoeff_nuq(val, dqv, dqv_val);
@ -258,11 +234,7 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
v = aom_read_bit(r, ACCT_STR) ? -v : v;
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#if CONFIG_HIGHBITDEPTH
check_range(v, xd->bd);
#else
check_range(v, 8);
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_COEFFICIENT_RANGE_CHECKING
dqcoeff[scan[c]] = v;
@ -273,94 +245,6 @@ static int decode_coefs(MACROBLOCKD *xd, PLANE_TYPE type, tran_low_t *dqcoeff,
dqv = dq[1];
ctx = get_coef_context(nb, token_cache, c);
band = *band_translate++;
#else // CONFIG_NEW_TOKENSET
while (c < max_eob) {
int val = -1;
band = *band_translate++;
prob = coef_probs[band][ctx];
if (counts) ++eob_branch_count[band][ctx];
if (!aom_read(r, prob[EOB_CONTEXT_NODE], ACCT_STR)) {
INCREMENT_COUNT(EOB_MODEL_TOKEN);
break;
}
#if CONFIG_NEW_QUANT
dqv_val = &dq_val[band][0];
#endif // CONFIG_NEW_QUANT
while (!aom_read(r, prob[ZERO_CONTEXT_NODE], ACCT_STR)) {
INCREMENT_COUNT(ZERO_TOKEN);
dqv = dq[1];
token_cache[scan[c]] = 0;
++c;
if (c >= max_eob) return c; // zero tokens at the end (no eob token)
ctx = get_coef_context(nb, token_cache, c);
band = *band_translate++;
prob = coef_probs[band][ctx];
#if CONFIG_NEW_QUANT
dqv_val = &dq_val[band][0];
#endif // CONFIG_NEW_QUANT
}
*max_scan_line = AOMMAX(*max_scan_line, scan[c]);
#if CONFIG_EC_MULTISYMBOL
cdf = &coef_cdfs[band][ctx];
token = ONE_TOKEN +
aom_read_symbol(r, *cdf, CATEGORY6_TOKEN - ONE_TOKEN + 1, ACCT_STR);
INCREMENT_COUNT(ONE_TOKEN + (token > ONE_TOKEN));
assert(token != ZERO_TOKEN);
val = token_to_value(r, token, tx_size,
#if CONFIG_HIGHBITDEPTH
xd->bd);
#else
8);
#endif // CONFIG_HIGHBITDEPTH
#else // CONFIG_EC_MULTISYMBOL
if (!aom_read(r, prob[ONE_CONTEXT_NODE], ACCT_STR)) {
INCREMENT_COUNT(ONE_TOKEN);
token = ONE_TOKEN;
val = 1;
} else {
INCREMENT_COUNT(TWO_TOKEN);
token = aom_read_tree(r, av1_coef_con_tree,
av1_pareto8_full[prob[PIVOT_NODE] - 1], ACCT_STR);
assert(token != ZERO_TOKEN && token != ONE_TOKEN);
val = token_to_value(r, token, tx_size,
#if CONFIG_HIGHBITDEPTH
xd->bd);
#else
8);
#endif // CONFIG_HIGHBITDEPTH
}
#endif // CONFIG_EC_MULTISYMBOL
#if CONFIG_NEW_QUANT
v = av1_dequant_abscoeff_nuq(val, dqv, dqv_val);
v = dq_shift ? ROUND_POWER_OF_TWO(v, dq_shift) : v;
#else
#if CONFIG_AOM_QM
dqv = ((iqmatrix[scan[c]] * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >>
AOM_QM_BITS;
#endif
v = (val * dqv) >> dq_shift;
#endif // CONFIG_NEW_QUANT
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#if CONFIG_HIGHBITDEPTH
dqcoeff[scan[c]] =
highbd_check_range((aom_read_bit(r, ACCT_STR) ? -v : v), xd->bd);
#else
dqcoeff[scan[c]] = check_range(aom_read_bit(r, ACCT_STR) ? -v : v, 8);
#endif // CONFIG_HIGHBITDEPTH
#else
dqcoeff[scan[c]] = aom_read_bit(r, ACCT_STR) ? -v : v;
#endif // CONFIG_COEFFICIENT_RANGE_CHECKING
token_cache[scan[c]] = av1_pt_energy_class[token];
++c;
ctx = get_coef_context(nb, token_cache, c);
dqv = dq[1];
#endif // CONFIG_NEW_TOKENSET
}
return c;

13
third_party/aom/av1/decoder/inspection.c vendored
View file

@ -14,6 +14,9 @@
#if CONFIG_CDEF
#include "av1/common/cdef.h"
#endif
#if CONFIG_CFL
#include "av1/common/cfl.h"
#endif
void ifd_init(insp_frame_data *fd, int frame_width, int frame_height) {
fd->mi_cols = ALIGN_POWER_OF_TWO(frame_width, 3) >> MI_SIZE_LOG2;
@ -96,6 +99,16 @@ int ifd_inspect(insp_frame_data *fd, void *decoder) {
mi->cdef_strength =
cm->cdef_strengths[mbmi->cdef_strength] % CLPF_STRENGTHS;
mi->cdef_strength += mi->cdef_strength == 3;
#endif
#if CONFIG_CFL
if (mbmi->uv_mode == DC_PRED) {
mi->cfl_alpha_idx = mbmi->cfl_alpha_idx;
mi->cfl_alpha_sign = (mbmi->cfl_alpha_signs[CFL_PRED_V] << CFL_PRED_V) +
mbmi->cfl_alpha_signs[CFL_PRED_U];
} else {
mi->cfl_alpha_idx = 0;
mi->cfl_alpha_sign = 0;
}
#endif
}
}

5
third_party/aom/av1/decoder/inspection.h vendored
View file

@ -15,6 +15,7 @@
extern "C" {
#endif // __cplusplus
#include "av1/common/seg_common.h"
#if CONFIG_ACCOUNTING
#include "av1/decoder/accounting.h"
#endif
@ -49,6 +50,10 @@ struct insp_mi_data {
int8_t cdef_level;
int8_t cdef_strength;
#endif
#if CONFIG_CFL
int8_t cfl_alpha_idx;
int8_t cfl_alpha_sign;
#endif
};
typedef struct insp_frame_data insp_frame_data;

4
third_party/aom/av1/decoder/laplace_decoder.c vendored
View file

@ -39,10 +39,10 @@ static int aom_decode_pvq_split_(aom_reader *r, od_pvq_codeword_ctx *adapt,
count += msbs << shift;
if (count > sum) {
count = sum;
#if CONFIG_DAALA_EC
#if !CONFIG_ANS
r->ec.error = 1;
#else
# error "CONFIG_PVQ currently requires CONFIG_DAALA_EC."
# error "CONFIG_PVQ currently requires !CONFIG_ANS."
#endif
}
return count;

4
third_party/aom/av1/encoder/aq_cyclicrefresh.c vendored
View file

@ -353,8 +353,8 @@ void av1_cyclic_refresh_check_golden_update(AV1_COMP *const cpi) {
// frame because of the camera movement, set this frame as the golden frame.
// Use 70% and 5% as the thresholds for golden frame refreshing.
// Also, force this frame as a golden update frame if this frame will change
// the resolution (resize_pending != 0).
if (cpi->resize_pending != 0 ||
// the resolution (av1_resize_pending != 0).
if (av1_resize_pending(cpi) ||
(cnt1 * 10 > (70 * rows * cols) && cnt2 * 20 < cnt1)) {
av1_cyclic_refresh_set_golden_update(cpi);
rc->frames_till_gf_update_due = rc->baseline_gf_interval;

54
third_party/aom/av1/encoder/av1_quantize.c vendored
View file

@ -1594,50 +1594,48 @@ static int get_qzbin_factor(int q, aom_bit_depth_t bit_depth) {
#endif
}
void av1_init_quantizer(AV1_COMP *cpi) {
AV1_COMMON *const cm = &cpi->common;
QUANTS *const quants = &cpi->quants;
void av1_build_quantizer(aom_bit_depth_t bit_depth, int y_dc_delta_q,
int uv_dc_delta_q, int uv_ac_delta_q,
QUANTS *const quants, Dequants *const deq) {
int i, q, quant;
#if CONFIG_NEW_QUANT
int dq;
#endif
for (q = 0; q < QINDEX_RANGE; q++) {
const int qzbin_factor = get_qzbin_factor(q, cm->bit_depth);
const int qzbin_factor = get_qzbin_factor(q, bit_depth);
const int qrounding_factor = q == 0 ? 64 : 48;
for (i = 0; i < 2; ++i) {
int qrounding_factor_fp = 64;
// y
quant = i == 0 ? av1_dc_quant(q, cm->y_dc_delta_q, cm->bit_depth)
: av1_ac_quant(q, 0, cm->bit_depth);
quant = i == 0 ? av1_dc_quant(q, y_dc_delta_q, bit_depth)
: av1_ac_quant(q, 0, bit_depth);
invert_quant(&quants->y_quant[q][i], &quants->y_quant_shift[q][i], quant);
quants->y_quant_fp[q][i] = (1 << 16) / quant;
quants->y_round_fp[q][i] = (qrounding_factor_fp * quant) >> 7;
quants->y_zbin[q][i] = ROUND_POWER_OF_TWO(qzbin_factor * quant, 7);
quants->y_round[q][i] = (qrounding_factor * quant) >> 7;
cpi->y_dequant[q][i] = quant;
deq->y_dequant[q][i] = quant;
// uv
quant = i == 0 ? av1_dc_quant(q, cm->uv_dc_delta_q, cm->bit_depth)
: av1_ac_quant(q, cm->uv_ac_delta_q, cm->bit_depth);
quant = i == 0 ? av1_dc_quant(q, uv_dc_delta_q, bit_depth)
: av1_ac_quant(q, uv_ac_delta_q, bit_depth);
invert_quant(&quants->uv_quant[q][i], &quants->uv_quant_shift[q][i],
quant);
quants->uv_quant_fp[q][i] = (1 << 16) / quant;
quants->uv_round_fp[q][i] = (qrounding_factor_fp * quant) >> 7;
quants->uv_zbin[q][i] = ROUND_POWER_OF_TWO(qzbin_factor * quant, 7);
quants->uv_round[q][i] = (qrounding_factor * quant) >> 7;
cpi->uv_dequant[q][i] = quant;
deq->uv_dequant[q][i] = quant;
}
#if CONFIG_NEW_QUANT
int dq;
for (dq = 0; dq < QUANT_PROFILES; dq++) {
for (i = 0; i < COEF_BANDS; i++) {
const int y_quant = cpi->y_dequant[q][i != 0];
const int uvquant = cpi->uv_dequant[q][i != 0];
av1_get_dequant_val_nuq(y_quant, i, cpi->y_dequant_val_nuq[dq][q][i],
const int y_quant = deq->y_dequant[q][i != 0];
const int uvquant = deq->uv_dequant[q][i != 0];
av1_get_dequant_val_nuq(y_quant, i, deq->y_dequant_val_nuq[dq][q][i],
quants->y_cuml_bins_nuq[dq][q][i], dq);
av1_get_dequant_val_nuq(uvquant, i, cpi->uv_dequant_val_nuq[dq][q][i],
av1_get_dequant_val_nuq(uvquant, i, deq->uv_dequant_val_nuq[dq][q][i],
quants->uv_cuml_bins_nuq[dq][q][i], dq);
}
}
@ -1650,7 +1648,7 @@ void av1_init_quantizer(AV1_COMP *cpi) {
quants->y_quant_shift[q][i] = quants->y_quant_shift[q][1];
quants->y_zbin[q][i] = quants->y_zbin[q][1];
quants->y_round[q][i] = quants->y_round[q][1];
cpi->y_dequant[q][i] = cpi->y_dequant[q][1];
deq->y_dequant[q][i] = deq->y_dequant[q][1];
quants->uv_quant[q][i] = quants->uv_quant[q][1];
quants->uv_quant_fp[q][i] = quants->uv_quant_fp[q][1];
@ -1658,11 +1656,19 @@ void av1_init_quantizer(AV1_COMP *cpi) {
quants->uv_quant_shift[q][i] = quants->uv_quant_shift[q][1];
quants->uv_zbin[q][i] = quants->uv_zbin[q][1];
quants->uv_round[q][i] = quants->uv_round[q][1];
cpi->uv_dequant[q][i] = cpi->uv_dequant[q][1];
deq->uv_dequant[q][i] = deq->uv_dequant[q][1];
}
}
}
void av1_init_quantizer(AV1_COMP *cpi) {
AV1_COMMON *const cm = &cpi->common;
QUANTS *const quants = &cpi->quants;
Dequants *const dequants = &cpi->dequants;
av1_build_quantizer(cm->bit_depth, cm->y_dc_delta_q, cm->uv_dc_delta_q,
cm->uv_ac_delta_q, quants, dequants);
}
void av1_init_plane_quantizers(const AV1_COMP *cpi, MACROBLOCK *x,
int segment_id) {
const AV1_COMMON *const cm = &cpi->common;
@ -1712,11 +1718,12 @@ void av1_init_plane_quantizers(const AV1_COMP *cpi, MACROBLOCK *x,
memcpy(&xd->plane[0].seg_iqmatrix[segment_id], cm->giqmatrix[qmlevel][0],
sizeof(cm->giqmatrix[qmlevel][0]));
#endif
xd->plane[0].dequant = cpi->y_dequant[qindex];
xd->plane[0].dequant = cpi->dequants.y_dequant[qindex];
#if CONFIG_NEW_QUANT
for (dq = 0; dq < QUANT_PROFILES; dq++) {
x->plane[0].cuml_bins_nuq[dq] = quants->y_cuml_bins_nuq[dq][qindex];
xd->plane[0].dequant_val_nuq[dq] = cpi->y_dequant_val_nuq[dq][qindex];
xd->plane[0].dequant_val_nuq[dq] =
cpi->dequants.y_dequant_val_nuq[dq][qindex];
}
#endif // CONFIG_NEW_QUANT
@ -1734,11 +1741,12 @@ void av1_init_plane_quantizers(const AV1_COMP *cpi, MACROBLOCK *x,
memcpy(&xd->plane[i].seg_iqmatrix[segment_id], cm->giqmatrix[qmlevel][1],
sizeof(cm->giqmatrix[qmlevel][1]));
#endif
xd->plane[i].dequant = cpi->uv_dequant[qindex];
xd->plane[i].dequant = cpi->dequants.uv_dequant[qindex];
#if CONFIG_NEW_QUANT
for (dq = 0; dq < QUANT_PROFILES; dq++) {
x->plane[i].cuml_bins_nuq[dq] = quants->uv_cuml_bins_nuq[dq][qindex];
xd->plane[i].dequant_val_nuq[dq] = cpi->uv_dequant_val_nuq[dq][qindex];
xd->plane[i].dequant_val_nuq[dq] =
cpi->dequants.uv_dequant_val_nuq[dq][qindex];
}
#endif // CONFIG_NEW_QUANT
}

15
third_party/aom/av1/encoder/av1_quantize.h vendored
View file

@ -69,6 +69,17 @@ typedef struct {
DECLARE_ALIGNED(16, int16_t, uv_round[QINDEX_RANGE][8]);
} QUANTS;
typedef struct {
DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]); // 8: SIMD width
DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]); // 8: SIMD width
#if CONFIG_NEW_QUANT
DECLARE_ALIGNED(16, dequant_val_type_nuq,
y_dequant_val_nuq[QUANT_PROFILES][QINDEX_RANGE][COEF_BANDS]);
DECLARE_ALIGNED(16, dequant_val_type_nuq,
uv_dequant_val_nuq[QUANT_PROFILES][QINDEX_RANGE][COEF_BANDS]);
#endif // CONFIG_NEW_QUANT
} Dequants;
struct AV1_COMP;
struct AV1Common;
@ -77,6 +88,10 @@ void av1_frame_init_quantizer(struct AV1_COMP *cpi);
void av1_init_plane_quantizers(const struct AV1_COMP *cpi, MACROBLOCK *x,
int segment_id);
void av1_build_quantizer(aom_bit_depth_t bit_depth, int y_dc_delta_q,
int uv_dc_delta_q, int uv_ac_delta_q,
QUANTS *const quants, Dequants *const deq);
void av1_init_quantizer(struct AV1_COMP *cpi);
void av1_set_quantizer(struct AV1Common *cm, int q);

1132
third_party/aom/av1/encoder/bitstream.c vendored
View file

File diff suppressed because it is too large Load diff

28
third_party/aom/av1/encoder/block.h vendored
View file

@ -17,9 +17,7 @@
#if CONFIG_PVQ
#include "av1/encoder/encint.h"
#endif
#if CONFIG_REF_MV
#include "av1/common/mvref_common.h"
#endif
#ifdef __cplusplus
extern "C" {
@ -79,13 +77,11 @@ typedef struct {
int dc_sign_ctx[MAX_MB_PLANE]
[MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)];
#endif
#if CONFIG_REF_MV
uint8_t ref_mv_count[MODE_CTX_REF_FRAMES];
CANDIDATE_MV ref_mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE];
#if CONFIG_EXT_INTER
int16_t compound_mode_context[MODE_CTX_REF_FRAMES];
#endif // CONFIG_EXT_INTER
#endif
} MB_MODE_INFO_EXT;
typedef struct {
@ -141,27 +137,18 @@ struct macroblock {
unsigned int pred_sse[TOTAL_REFS_PER_FRAME];
int pred_mv_sad[TOTAL_REFS_PER_FRAME];
#if CONFIG_REF_MV
int *nmvjointcost;
int nmv_vec_cost[NMV_CONTEXTS][MV_JOINTS];
int *nmvcost[NMV_CONTEXTS][2];
int *nmvcost_hp[NMV_CONTEXTS][2];
int **mv_cost_stack[NMV_CONTEXTS];
int *nmvjointsadcost;
#else
int nmvjointcost[MV_JOINTS];
int *nmvcost[2];
int *nmvcost_hp[2];
int nmvjointsadcost[MV_JOINTS];
#endif
int **mvcost;
int *nmvsadcost[2];
int *nmvsadcost_hp[2];
int **mvsadcost;
#if CONFIG_MOTION_VAR
int32_t *wsrc_buf;
int32_t *mask_buf;
uint8_t *above_pred_buf;
uint8_t *left_pred_buf;
#endif // CONFIG_MOTION_VAR
#if CONFIG_PALETTE
@ -174,9 +161,7 @@ struct macroblock {
#if CONFIG_VAR_TX
uint8_t blk_skip[MAX_MB_PLANE][MAX_MIB_SIZE * MAX_MIB_SIZE * 8];
#if CONFIG_REF_MV
uint8_t blk_skip_drl[MAX_MB_PLANE][MAX_MIB_SIZE * MAX_MIB_SIZE * 8];
#endif
#endif
int skip;
@ -226,8 +211,11 @@ struct macroblock {
// This is needed when using the 8x8 Daala distortion metric during RDO,
// because it evaluates distortion in a different order than the underlying
// 4x4 blocks are coded.
int rate_4x4[256];
#endif
int rate_4x4[MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)];
#if CONFIG_CB4X4
DECLARE_ALIGNED(16, uint8_t, decoded_8x8[8 * 8]);
#endif // CONFIG_CB4X4
#endif // CONFIG_DAALA_DIST
#if CONFIG_CFL
// Whether luma needs to be stored during RDO.
int cfl_store_y;

4
third_party/aom/av1/encoder/context_tree.h vendored
View file

@ -34,7 +34,6 @@ typedef struct {
uint8_t *blk_skip[MAX_MB_PLANE];
#endif
// dual buffer pointers, 0: in use, 1: best in store
tran_low_t *coeff[MAX_MB_PLANE];
tran_low_t *qcoeff[MAX_MB_PLANE];
tran_low_t *dqcoeff[MAX_MB_PLANE];
@ -48,9 +47,8 @@ typedef struct {
int num_4x4_blk;
int skip;
int pred_pixel_ready;
// For current partition, only if all Y, U, and V transform blocks'
// coefficients are quantized to 0, skippable is set to 0.
// coefficients are quantized to 0, skippable is set to 1.
int skippable;
int best_mode_index;
int hybrid_pred_diff;

15
third_party/aom/av1/encoder/corner_match.c vendored
View file

@ -9,16 +9,13 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <stdio.h>
#include <stdlib.h>
#include <memory.h>
#include <math.h>
#include "./av1_rtcd.h"
#include "av1/encoder/corner_match.h"
#define MATCH_SZ 13
#define MATCH_SZ_BY2 ((MATCH_SZ - 1) / 2)
#define MATCH_SZ_SQ (MATCH_SZ * MATCH_SZ)
#define SEARCH_SZ 9
#define SEARCH_SZ_BY2 ((SEARCH_SZ - 1) / 2)
@ -28,8 +25,8 @@
centered at (x, y).
*/
static double compute_variance(unsigned char *im, int stride, int x, int y) {
int sum = 0.0;
int sumsq = 0.0;
int sum = 0;
int sumsq = 0;
int var;
int i, j;
for (i = 0; i < MATCH_SZ; ++i)
@ -46,9 +43,9 @@ static double compute_variance(unsigned char *im, int stride, int x, int y) {
correlation/standard deviation are taken over MATCH_SZ by MATCH_SZ windows
of each image, centered at (x1, y1) and (x2, y2) respectively.
*/
static double compute_cross_correlation(unsigned char *im1, int stride1, int x1,
int y1, unsigned char *im2, int stride2,
int x2, int y2) {
double compute_cross_correlation_c(unsigned char *im1, int stride1, int x1,
int y1, unsigned char *im2, int stride2,
int x2, int y2) {
int v1, v2;
int sum1 = 0;
int sum2 = 0;

4
third_party/aom/av1/encoder/corner_match.h vendored
View file

@ -15,6 +15,10 @@
#include <stdlib.h>
#include <memory.h>
#define MATCH_SZ 13
#define MATCH_SZ_BY2 ((MATCH_SZ - 1) / 2)
#define MATCH_SZ_SQ (MATCH_SZ * MATCH_SZ)
typedef struct {
int x, y;
int rx, ry;

8
third_party/aom/av1/encoder/daala_compat_enc.c vendored
View file

@ -12,19 +12,19 @@
#include "encint.h"
void od_encode_checkpoint(const daala_enc_ctx *enc, od_rollback_buffer *rbuf) {
#if CONFIG_DAALA_EC
#if !CONFIG_ANS
od_ec_enc_checkpoint(&rbuf->ec, &enc->w.ec);
#else
#error "CONFIG_PVQ currently requires CONFIG_DAALA_EC."
#error "CONFIG_PVQ currently requires !CONFIG_ANS."
#endif
OD_COPY(&rbuf->adapt, enc->state.adapt, 1);
}
void od_encode_rollback(daala_enc_ctx *enc, const od_rollback_buffer *rbuf) {
#if CONFIG_DAALA_EC
#if !CONFIG_ANS
od_ec_enc_rollback(&enc->w.ec, &rbuf->ec);
#else
#error "CONFIG_PVQ currently requires CONFIG_DAALA_EC."
#error "CONFIG_PVQ currently requires !CONFIG_ANS."
#endif
OD_COPY(enc->state.adapt, &rbuf->adapt, 1);
}

57
third_party/aom/av1/encoder/dct.c vendored
View file

@ -19,7 +19,7 @@
#include "aom_ports/mem.h"
#include "av1/common/blockd.h"
#include "av1/common/av1_fwd_txfm1d.h"
#include "av1/common/av1_fwd_txfm2d_cfg.h"
#include "av1/common/av1_fwd_txfm1d_cfg.h"
#include "av1/common/idct.h"
static INLINE void range_check(const tran_low_t *input, const int size,
@ -1022,6 +1022,10 @@ static void fhalfright32(const tran_low_t *input, tran_low_t *output) {
}
#if CONFIG_EXT_TX
// TODO(sarahparker) these functions will be removed once the highbitdepth
// codepath works properly for rectangular transforms. They have almost
// identical versions in av1_fwd_txfm1d.c, but those are currently only
// being used for square transforms.
static void fidtx4(const tran_low_t *input, tran_low_t *output) {
int i;
for (i = 0; i < 4; ++i)
@ -2133,8 +2137,7 @@ static void fdct64_col(const tran_low_t *input, tran_low_t *output) {
int32_t in[64], out[64];
int i;
for (i = 0; i < 64; ++i) in[i] = (int32_t)input[i];
av1_fdct64_new(in, out, fwd_cos_bit_col_dct_dct_64,
fwd_stage_range_col_dct_dct_64);
av1_fdct64_new(in, out, fwd_cos_bit_col_dct_64, fwd_stage_range_col_dct_64);
for (i = 0; i < 64; ++i) output[i] = (tran_low_t)out[i];
}
@ -2142,8 +2145,7 @@ static void fdct64_row(const tran_low_t *input, tran_low_t *output) {
int32_t in[64], out[64];
int i;
for (i = 0; i < 64; ++i) in[i] = (int32_t)input[i];
av1_fdct64_new(in, out, fwd_cos_bit_row_dct_dct_64,
fwd_stage_range_row_dct_dct_64);
av1_fdct64_new(in, out, fwd_cos_bit_row_dct_64, fwd_stage_range_row_dct_64);
for (i = 0; i < 64; ++i) output[i] = (tran_low_t)out[i];
}
@ -2225,4 +2227,49 @@ void av1_highbd_fht64x64_c(const int16_t *input, tran_low_t *output, int stride,
}
#endif // CONFIG_TX64X64
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_DPCM_INTRA
void av1_dpcm_ft4_c(const int16_t *input, int stride, TX_TYPE_1D tx_type,
tran_low_t *output) {
assert(tx_type < TX_TYPES_1D);
static const transform_1d FHT[] = { fdct4, fadst4, fadst4, fidtx4 };
const transform_1d ft = FHT[tx_type];
tran_low_t temp_in[4];
for (int i = 0; i < 4; ++i)
temp_in[i] = (tran_low_t)fdct_round_shift(input[i * stride] * 4 * Sqrt2);
ft(temp_in, output);
}
void av1_dpcm_ft8_c(const int16_t *input, int stride, TX_TYPE_1D tx_type,
tran_low_t *output) {
assert(tx_type < TX_TYPES_1D);
static const transform_1d FHT[] = { fdct8, fadst8, fadst8, fidtx8 };
const transform_1d ft = FHT[tx_type];
tran_low_t temp_in[8];
for (int i = 0; i < 8; ++i) temp_in[i] = input[i * stride] * 4;
ft(temp_in, output);
}
void av1_dpcm_ft16_c(const int16_t *input, int stride, TX_TYPE_1D tx_type,
tran_low_t *output) {
assert(tx_type < TX_TYPES_1D);
static const transform_1d FHT[] = { fdct16, fadst16, fadst16, fidtx16 };
const transform_1d ft = FHT[tx_type];
tran_low_t temp_in[16];
for (int i = 0; i < 16; ++i)
temp_in[i] = (tran_low_t)fdct_round_shift(input[i * stride] * 2 * Sqrt2);
ft(temp_in, output);
}
void av1_dpcm_ft32_c(const int16_t *input, int stride, TX_TYPE_1D tx_type,
tran_low_t *output) {
assert(tx_type < TX_TYPES_1D);
static const transform_1d FHT[] = { fdct32, fhalfright32, fhalfright32,
fidtx32 };
const transform_1d ft = FHT[tx_type];
tran_low_t temp_in[32];
for (int i = 0; i < 32; ++i) temp_in[i] = input[i * stride];
ft(temp_in, output);
}
#endif // CONFIG_DPCM_INTRA
#endif // !AV1_DCT_GTEST

1335
third_party/aom/av1/encoder/encodeframe.c vendored
View file

File diff suppressed because it is too large Load diff

9
third_party/aom/av1/encoder/encodeframe.h vendored
View file

@ -25,13 +25,6 @@ struct yv12_buffer_config;
struct AV1_COMP;
struct ThreadData;
// Constants used in SOURCE_VAR_BASED_PARTITION
#define VAR_HIST_MAX_BG_VAR 1000
#define VAR_HIST_FACTOR 10
#define VAR_HIST_BINS (VAR_HIST_MAX_BG_VAR / VAR_HIST_FACTOR + 1)
#define VAR_HIST_LARGE_CUT_OFF 75
#define VAR_HIST_SMALL_CUT_OFF 45
void av1_setup_src_planes(struct macroblock *x,
const struct yv12_buffer_config *src, int mi_row,
int mi_col);
@ -42,8 +35,6 @@ void av1_init_tile_data(struct AV1_COMP *cpi);
void av1_encode_tile(struct AV1_COMP *cpi, struct ThreadData *td, int tile_row,
int tile_col);
void av1_set_variance_partition_thresholds(struct AV1_COMP *cpi, int q);
void av1_update_tx_type_count(const struct AV1Common *cm, MACROBLOCKD *xd,
#if CONFIG_TXK_SEL
int block, int plane,

703
third_party/aom/av1/encoder/encodemb.c vendored
View file

@ -115,7 +115,7 @@ static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
#if CONFIG_EC_ADAPT
{ 10, 7 }, { 8, 5 },
#else
{ 10, 6 }, { 8, 5 },
{ 10, 6 }, { 8, 6 },
#endif
};
@ -125,35 +125,31 @@ static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1); \
}
static INLINE int64_t
get_token_bit_costs(unsigned int token_costs[2][COEFF_CONTEXTS][ENTROPY_TOKENS],
int skip_eob, int ctx, int token) {
#if CONFIG_NEW_TOKENSET
static INLINE unsigned int get_token_bit_costs(
unsigned int token_costs[2][COEFF_CONTEXTS][ENTROPY_TOKENS], int skip_eob,
int ctx, int token) {
(void)skip_eob;
return token_costs[token == ZERO_TOKEN || token == EOB_TOKEN][ctx][token];
#else
return token_costs[skip_eob][ctx][token];
#endif
}
#if !CONFIG_LV_MAP
#define USE_GREEDY_OPTIMIZE_B 0
#if USE_GREEDY_OPTIMIZE_B
typedef struct av1_token_state {
typedef struct av1_token_state_greedy {
int16_t token;
tran_low_t qc;
tran_low_t dqc;
} av1_token_state;
} av1_token_state_greedy;
int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
TX_SIZE tx_size, int ctx) {
#if !CONFIG_PVQ
static int optimize_b_greedy(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
int block, TX_SIZE tx_size, int ctx) {
MACROBLOCKD *const xd = &mb->e_mbd;
struct macroblock_plane *const p = &mb->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
const int ref = is_inter_block(&xd->mi[0]->mbmi);
av1_token_state tokens[MAX_TX_SQUARE + 1][2];
av1_token_state_greedy tokens[MAX_TX_SQUARE + 1][2];
uint8_t token_cache[MAX_TX_SQUARE];
const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
@ -176,38 +172,23 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
#if CONFIG_NEW_QUANT
int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
#elif !CONFIG_AOM_QM
const int dq_step[2] = { dequant_ptr[0] >> shift, dequant_ptr[1] >> shift };
#endif // CONFIG_NEW_QUANT
int sz = 0;
const int64_t rddiv = mb->rddiv;
int64_t rd_cost0, rd_cost1;
int16_t t0, t1;
int i, final_eob;
#if CONFIG_HIGHBITDEPTH
const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, xd->bd);
#else
const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, 8);
#endif
unsigned int(*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
mb->token_costs[txsize_sqr_map[tx_size]][plane_type][ref];
const int default_eob = tx_size_2d[tx_size];
assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
assert(mb->qindex > 0);
assert((!plane_type && !plane) || (plane_type && plane));
assert(eob <= default_eob);
int64_t rdmult = (mb->rdmult * plane_rd_mult[ref][plane_type]) >> 1;
/* CpuSpeedTest uses "--min-q=0 --max-q=0" and expects 100dB psnr
* This creates conflict with search for a better EOB position
* The line below is to make sure EOB search is disabled at this corner case.
*/
#if !CONFIG_NEW_QUANT && !CONFIG_AOM_QM
if (dq_step[1] <= 4) {
rdmult = 1;
}
#endif
int64_t rate0, rate1;
for (i = 0; i < eob; i++) {
@ -402,22 +383,10 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
dqc_a = shift ? ROUND_POWER_OF_TWO(dqc_a, shift) : dqc_a;
if (sz) dqc_a = -dqc_a;
#else
// The 32x32 transform coefficient uses half quantization step size.
// Account for the rounding difference in the dequantized coefficeint
// value when the quantization index is dropped from an even number
// to an odd number.
#if CONFIG_AOM_QM
tran_low_t offset = dqv >> shift;
#else
tran_low_t offset = dq_step[rc != 0];
#endif
if (shift & x_a) offset += (dqv & 0x01);
if (sz == 0)
dqc_a = dqcoeff[rc] - offset;
if (x_a < 0)
dqc_a = -((-x_a * dqv) >> shift);
else
dqc_a = dqcoeff[rc] + offset;
dqc_a = (x_a * dqv) >> shift;
#endif // CONFIG_NEW_QUANT
} else {
dqc_a = 0;
@ -483,19 +452,11 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
mb->plane[plane].eobs[block] = final_eob;
return final_eob;
#else // !CONFIG_PVQ
(void)cm;
(void)tx_size;
(void)ctx;
struct macroblock_plane *const p = &mb->plane[plane];
return p->eobs[block];
#endif // !CONFIG_PVQ
}
#else // USE_GREEDY_OPTIMIZE_B
typedef struct av1_token_state {
typedef struct av1_token_state_org {
int64_t error;
int rate;
int16_t next;
@ -503,16 +464,15 @@ typedef struct av1_token_state {
tran_low_t qc;
tran_low_t dqc;
uint8_t best_index;
} av1_token_state;
} av1_token_state_org;
int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
TX_SIZE tx_size, int ctx) {
#if !CONFIG_PVQ
static int optimize_b_org(const AV1_COMMON *cm, MACROBLOCK *mb, int plane,
int block, TX_SIZE tx_size, int ctx) {
MACROBLOCKD *const xd = &mb->e_mbd;
struct macroblock_plane *const p = &mb->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
const int ref = is_inter_block(&xd->mi[0]->mbmi);
av1_token_state tokens[MAX_TX_SQUARE + 1][2];
av1_token_state_org tokens[MAX_TX_SQUARE + 1][2];
uint8_t token_cache[MAX_TX_SQUARE];
const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
@ -536,8 +496,6 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
#if CONFIG_NEW_QUANT
int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
#elif !CONFIG_AOM_QM
const int dq_step[2] = { dequant_ptr[0] >> shift, dequant_ptr[1] >> shift };
#endif // CONFIG_NEW_QUANT
int next = eob, sz = 0;
const int64_t rdmult = (mb->rdmult * plane_rd_mult[ref][plane_type]) >> 1;
@ -549,11 +507,7 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
int best, band = (eob < default_eob) ? band_translate[eob]
: band_translate[eob - 1];
int pt, i, final_eob;
#if CONFIG_HIGHBITDEPTH
const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, xd->bd);
#else
const int cat6_bits = av1_get_cat6_extrabits_size(tx_size, 8);
#endif
unsigned int(*token_costs)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =
mb->token_costs[txsize_sqr_map[tx_size]][plane_type][ref];
const uint16_t *band_counts = &band_count_table[tx_size][band];
@ -566,11 +520,10 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
? av1_get_qindex(&cm->seg, xd->mi[0]->mbmi.segment_id,
cm->base_qindex)
: cm->base_qindex;
if (qindex == 0) {
assert((qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
}
assert(qindex > 0);
(void)qindex;
#else
assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
assert(mb->qindex > 0);
#endif
token_costs += band;
@ -777,22 +730,10 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
: tokens[i][1].dqc;
if (sz) tokens[i][1].dqc = -tokens[i][1].dqc;
#else
// The 32x32 transform coefficient uses half quantization step size.
// Account for the rounding difference in the dequantized coefficeint
// value when the quantization index is dropped from an even number
// to an odd number.
#if CONFIG_AOM_QM
tran_low_t offset = dqv >> shift;
#else
tran_low_t offset = dq_step[rc != 0];
#endif
if (shift & x) offset += (dqv & 0x01);
if (sz == 0)
tokens[i][1].dqc = dqcoeff[rc] - offset;
if (x < 0)
tokens[i][1].dqc = -((-x * dqv) >> shift);
else
tokens[i][1].dqc = dqcoeff[rc] + offset;
tokens[i][1].dqc = (x * dqv) >> shift;
#endif // CONFIG_NEW_QUANT
} else {
tokens[i][1].dqc = 0;
@ -858,16 +799,47 @@ int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
mb->plane[plane].eobs[block] = final_eob;
assert(final_eob <= default_eob);
return final_eob;
#else // !CONFIG_PVQ
(void)cm;
(void)tx_size;
(void)ctx;
struct macroblock_plane *const p = &mb->plane[plane];
return p->eobs[block];
#endif // !CONFIG_PVQ
}
#endif // USE_GREEDY_OPTIMIZE_B
#endif // !CONFIG_LV_MAP
int av1_optimize_b(const AV1_COMMON *cm, MACROBLOCK *mb, int plane, int block,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
const ENTROPY_CONTEXT *a, const ENTROPY_CONTEXT *l) {
MACROBLOCKD *const xd = &mb->e_mbd;
struct macroblock_plane *const p = &mb->plane[plane];
const int eob = p->eobs[block];
assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
if (eob == 0) return eob;
if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return eob;
#if CONFIG_PVQ
(void)cm;
(void)tx_size;
(void)a;
(void)l;
return eob;
#endif
#if !CONFIG_LV_MAP
(void)plane_bsize;
#if CONFIG_VAR_TX
int ctx = get_entropy_context(tx_size, a, l);
#else
int ctx = combine_entropy_contexts(*a, *l);
#endif
#if USE_GREEDY_OPTIMIZE_B
return optimize_b_greedy(cm, mb, plane, block, tx_size, ctx);
#else // USE_GREEDY_OPTIMIZE_B
return optimize_b_org(cm, mb, plane, block, tx_size, ctx);
#endif // USE_GREEDY_OPTIMIZE_B
#else // !CONFIG_LV_MAP
TXB_CTX txb_ctx;
get_txb_ctx(plane_bsize, tx_size, plane, a, l, &txb_ctx);
return av1_optimize_txb(cm, mb, plane, block, tx_size, &txb_ctx);
#endif // !CONFIG_LV_MAP
}
#if !CONFIG_PVQ
#if CONFIG_HIGHBITDEPTH
@ -1158,8 +1130,7 @@ static void encode_block(int plane, int block, int blk_row, int blk_col,
#endif
#if !CONFIG_PVQ
if (p->eobs[block] && !xd->lossless[xd->mi[0]->mbmi.segment_id])
av1_optimize_b(cm, x, plane, block, tx_size, ctx);
av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
av1_set_txb_context(x, plane, block, tx_size, a, l);
@ -1202,12 +1173,13 @@ static void encode_block_inter(int plane, int block, int blk_row, int blk_col,
if (tx_size == plane_tx_size) {
encode_block(plane, block, blk_row, blk_col, plane_bsize, tx_size, arg);
} else {
assert(tx_size < TX_SIZES_ALL);
const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
assert(sub_txs < tx_size);
// This is the square transform block partition entry point.
int bsl = tx_size_wide_unit[sub_txs];
int i;
assert(bsl > 0);
assert(tx_size < TX_SIZES_ALL);
for (i = 0; i < 4; ++i) {
const int offsetr = blk_row + ((i >> 1) * bsl);
@ -1301,8 +1273,8 @@ void av1_encode_sby_pass1(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize) {
encode_block_pass1, &args);
}
void av1_encode_sb(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize,
const int mi_row, const int mi_col) {
void av1_encode_sb(AV1_COMMON *cm, MACROBLOCK *x, BLOCK_SIZE bsize, int mi_row,
int mi_col) {
MACROBLOCKD *const xd = &x->e_mbd;
struct optimize_ctx ctx;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
@ -1433,6 +1405,301 @@ static void encode_block_intra_and_set_context(int plane, int block,
#endif
}
#if CONFIG_DPCM_INTRA
static int get_eob(const tran_low_t *qcoeff, intptr_t n_coeffs,
const int16_t *scan) {
int eob = -1;
for (int i = (int)n_coeffs - 1; i >= 0; i--) {
const int rc = scan[i];
if (qcoeff[rc]) {
eob = i;
break;
}
}
return eob + 1;
}
static void quantize_scaler(int coeff, int16_t zbin, int16_t round_value,
int16_t quant, int16_t quant_shift, int16_t dequant,
int log_scale, tran_low_t *const qcoeff,
tran_low_t *const dqcoeff) {
zbin = ROUND_POWER_OF_TWO(zbin, log_scale);
round_value = ROUND_POWER_OF_TWO(round_value, log_scale);
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
if (abs_coeff >= zbin) {
int tmp = clamp(abs_coeff + round_value, INT16_MIN, INT16_MAX);
tmp = ((((tmp * quant) >> 16) + tmp) * quant_shift) >> (16 - log_scale);
*qcoeff = (tmp ^ coeff_sign) - coeff_sign;
*dqcoeff = (*qcoeff * dequant) / (1 << log_scale);
}
}
typedef void (*dpcm_fwd_tx_func)(const int16_t *input, int stride,
TX_TYPE_1D tx_type, tran_low_t *output);
static dpcm_fwd_tx_func get_dpcm_fwd_tx_func(int tx_length) {
switch (tx_length) {
case 4: return av1_dpcm_ft4_c;
case 8: return av1_dpcm_ft8_c;
case 16: return av1_dpcm_ft16_c;
case 32:
return av1_dpcm_ft32_c;
// TODO(huisu): add support for TX_64X64.
default: assert(0); return NULL;
}
}
static void process_block_dpcm_vert(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
struct macroblockd_plane *const pd,
struct macroblock_plane *const p,
uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int16_t *src_diff,
int diff_stride, tran_low_t *coeff,
tran_low_t *qcoeff, tran_low_t *dqcoeff) {
const int tx1d_width = tx_size_wide[tx_size];
dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_width);
dpcm_inv_txfm_add_func inverse_tx =
av1_get_dpcm_inv_txfm_add_func(tx1d_width);
const int tx1d_height = tx_size_high[tx_size];
const int log_scale = av1_get_tx_scale(tx_size);
int q_idx = 0;
for (int r = 0; r < tx1d_height; ++r) {
// Update prediction.
if (r > 0) memcpy(dst, dst - dst_stride, tx1d_width * sizeof(dst[0]));
// Subtraction.
for (int c = 0; c < tx1d_width; ++c) src_diff[c] = src[c] - dst[c];
// Forward transform.
forward_tx(src_diff, 1, tx_type_1d, coeff);
// Quantization.
for (int c = 0; c < tx1d_width; ++c) {
quantize_scaler(coeff[c], p->zbin[q_idx], p->round[q_idx],
p->quant[q_idx], p->quant_shift[q_idx],
pd->dequant[q_idx], log_scale, &qcoeff[c], &dqcoeff[c]);
q_idx = 1;
}
// Inverse transform.
inverse_tx(dqcoeff, 1, tx_type_1d, dst);
// Move to the next row.
coeff += tx1d_width;
qcoeff += tx1d_width;
dqcoeff += tx1d_width;
src_diff += diff_stride;
dst += dst_stride;
src += src_stride;
}
}
static void process_block_dpcm_horz(TX_SIZE tx_size, TX_TYPE_1D tx_type_1d,
struct macroblockd_plane *const pd,
struct macroblock_plane *const p,
uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int16_t *src_diff,
int diff_stride, tran_low_t *coeff,
tran_low_t *qcoeff, tran_low_t *dqcoeff) {
const int tx1d_height = tx_size_high[tx_size];
dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_height);
dpcm_inv_txfm_add_func inverse_tx =
av1_get_dpcm_inv_txfm_add_func(tx1d_height);
const int tx1d_width = tx_size_wide[tx_size];
const int log_scale = av1_get_tx_scale(tx_size);
int q_idx = 0;
for (int c = 0; c < tx1d_width; ++c) {
for (int r = 0; r < tx1d_height; ++r) {
// Update prediction.
if (c > 0) dst[r * dst_stride] = dst[r * dst_stride - 1];
// Subtraction.
src_diff[r * diff_stride] = src[r * src_stride] - dst[r * dst_stride];
}
// Forward transform.
tran_low_t tx_buff[64];
forward_tx(src_diff, diff_stride, tx_type_1d, tx_buff);
for (int r = 0; r < tx1d_height; ++r) coeff[r * tx1d_width] = tx_buff[r];
// Quantization.
for (int r = 0; r < tx1d_height; ++r) {
quantize_scaler(coeff[r * tx1d_width], p->zbin[q_idx], p->round[q_idx],
p->quant[q_idx], p->quant_shift[q_idx],
pd->dequant[q_idx], log_scale, &qcoeff[r * tx1d_width],
&dqcoeff[r * tx1d_width]);
q_idx = 1;
}
// Inverse transform.
for (int r = 0; r < tx1d_height; ++r) tx_buff[r] = dqcoeff[r * tx1d_width];
inverse_tx(tx_buff, dst_stride, tx_type_1d, dst);
// Move to the next column.
++coeff, ++qcoeff, ++dqcoeff, ++src_diff, ++dst, ++src;
}
}
#if CONFIG_HIGHBITDEPTH
static void hbd_process_block_dpcm_vert(
TX_SIZE tx_size, TX_TYPE_1D tx_type_1d, int bd,
struct macroblockd_plane *const pd, struct macroblock_plane *const p,
uint8_t *src8, int src_stride, uint8_t *dst8, int dst_stride,
int16_t *src_diff, int diff_stride, tran_low_t *coeff, tran_low_t *qcoeff,
tran_low_t *dqcoeff) {
const int tx1d_width = tx_size_wide[tx_size];
dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_width);
hbd_dpcm_inv_txfm_add_func inverse_tx =
av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_width);
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
const int tx1d_height = tx_size_high[tx_size];
const int log_scale = av1_get_tx_scale(tx_size);
int q_idx = 0;
for (int r = 0; r < tx1d_height; ++r) {
// Update prediction.
if (r > 0) memcpy(dst, dst - dst_stride, tx1d_width * sizeof(dst[0]));
// Subtraction.
for (int c = 0; c < tx1d_width; ++c) src_diff[c] = src[c] - dst[c];
// Forward transform.
forward_tx(src_diff, 1, tx_type_1d, coeff);
// Quantization.
for (int c = 0; c < tx1d_width; ++c) {
quantize_scaler(coeff[c], p->zbin[q_idx], p->round[q_idx],
p->quant[q_idx], p->quant_shift[q_idx],
pd->dequant[q_idx], log_scale, &qcoeff[c], &dqcoeff[c]);
q_idx = 1;
}
// Inverse transform.
inverse_tx(dqcoeff, 1, tx_type_1d, bd, dst);
// Move to the next row.
coeff += tx1d_width;
qcoeff += tx1d_width;
dqcoeff += tx1d_width;
src_diff += diff_stride;
dst += dst_stride;
src += src_stride;
}
}
static void hbd_process_block_dpcm_horz(
TX_SIZE tx_size, TX_TYPE_1D tx_type_1d, int bd,
struct macroblockd_plane *const pd, struct macroblock_plane *const p,
uint8_t *src8, int src_stride, uint8_t *dst8, int dst_stride,
int16_t *src_diff, int diff_stride, tran_low_t *coeff, tran_low_t *qcoeff,
tran_low_t *dqcoeff) {
const int tx1d_height = tx_size_high[tx_size];
dpcm_fwd_tx_func forward_tx = get_dpcm_fwd_tx_func(tx1d_height);
hbd_dpcm_inv_txfm_add_func inverse_tx =
av1_get_hbd_dpcm_inv_txfm_add_func(tx1d_height);
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
const int tx1d_width = tx_size_wide[tx_size];
const int log_scale = av1_get_tx_scale(tx_size);
int q_idx = 0;
for (int c = 0; c < tx1d_width; ++c) {
for (int r = 0; r < tx1d_height; ++r) {
// Update prediction.
if (c > 0) dst[r * dst_stride] = dst[r * dst_stride - 1];
// Subtraction.
src_diff[r * diff_stride] = src[r * src_stride] - dst[r * dst_stride];
}
// Forward transform.
tran_low_t tx_buff[64];
forward_tx(src_diff, diff_stride, tx_type_1d, tx_buff);
for (int r = 0; r < tx1d_height; ++r) coeff[r * tx1d_width] = tx_buff[r];
// Quantization.
for (int r = 0; r < tx1d_height; ++r) {
quantize_scaler(coeff[r * tx1d_width], p->zbin[q_idx], p->round[q_idx],
p->quant[q_idx], p->quant_shift[q_idx],
pd->dequant[q_idx], log_scale, &qcoeff[r * tx1d_width],
&dqcoeff[r * tx1d_width]);
q_idx = 1;
}
// Inverse transform.
for (int r = 0; r < tx1d_height; ++r) tx_buff[r] = dqcoeff[r * tx1d_width];
inverse_tx(tx_buff, dst_stride, tx_type_1d, bd, dst);
// Move to the next column.
++coeff, ++qcoeff, ++dqcoeff, ++src_diff, ++dst, ++src;
}
}
#endif // CONFIG_HIGHBITDEPTH
void av1_encode_block_intra_dpcm(const AV1_COMMON *cm, MACROBLOCK *x,
PREDICTION_MODE mode, int plane, int block,
int blk_row, int blk_col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
TX_TYPE tx_type, ENTROPY_CONTEXT *ta,
ENTROPY_CONTEXT *tl, int8_t *skip) {
MACROBLOCKD *const xd = &x->e_mbd;
struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
const int diff_stride = block_size_wide[plane_bsize];
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
const int tx1d_width = tx_size_wide[tx_size];
const int tx1d_height = tx_size_high[tx_size];
const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, 0);
tran_low_t *coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
uint8_t *dst =
&pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
uint8_t *src =
&p->src.buf[(blk_row * src_stride + blk_col) << tx_size_wide_log2[0]];
int16_t *src_diff =
&p->src_diff[(blk_row * diff_stride + blk_col) << tx_size_wide_log2[0]];
uint16_t *eob = &p->eobs[block];
*eob = 0;
memset(qcoeff, 0, tx1d_height * tx1d_width * sizeof(*qcoeff));
memset(dqcoeff, 0, tx1d_height * tx1d_width * sizeof(*dqcoeff));
if (LIKELY(!x->skip_block)) {
TX_TYPE_1D tx_type_1d = DCT_1D;
switch (tx_type) {
case IDTX: tx_type_1d = IDTX_1D; break;
case V_DCT:
assert(mode == H_PRED);
tx_type_1d = DCT_1D;
break;
case H_DCT:
assert(mode == V_PRED);
tx_type_1d = DCT_1D;
break;
default: assert(0);
}
switch (mode) {
case V_PRED:
#if CONFIG_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
hbd_process_block_dpcm_vert(tx_size, tx_type_1d, xd->bd, pd, p, src,
src_stride, dst, dst_stride, src_diff,
diff_stride, coeff, qcoeff, dqcoeff);
} else {
#endif // CONFIG_HIGHBITDEPTH
process_block_dpcm_vert(tx_size, tx_type_1d, pd, p, src, src_stride,
dst, dst_stride, src_diff, diff_stride, coeff,
qcoeff, dqcoeff);
#if CONFIG_HIGHBITDEPTH
}
#endif // CONFIG_HIGHBITDEPTH
break;
case H_PRED:
#if CONFIG_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
hbd_process_block_dpcm_horz(tx_size, tx_type_1d, xd->bd, pd, p, src,
src_stride, dst, dst_stride, src_diff,
diff_stride, coeff, qcoeff, dqcoeff);
} else {
#endif // CONFIG_HIGHBITDEPTH
process_block_dpcm_horz(tx_size, tx_type_1d, pd, p, src, src_stride,
dst, dst_stride, src_diff, diff_stride, coeff,
qcoeff, dqcoeff);
#if CONFIG_HIGHBITDEPTH
}
#endif // CONFIG_HIGHBITDEPTH
break;
default: assert(0);
}
*eob = get_eob(qcoeff, tx1d_height * tx1d_width, scan_order->scan);
}
ta[blk_col] = tl[blk_row] = *eob > 0;
if (*eob) *skip = 0;
}
#endif // CONFIG_DPCM_INTRA
void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *arg) {
@ -1449,7 +1716,33 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
const int dst_stride = pd->dst.stride;
uint8_t *dst =
&pd->dst.buf[(blk_row * dst_stride + blk_col) << tx_size_wide_log2[0]];
#if CONFIG_CFL
#if CONFIG_EC_ADAPT
FRAME_CONTEXT *const ec_ctx = xd->tile_ctx;
#else
FRAME_CONTEXT *const ec_ctx = cm->fc;
#endif // CONFIG_EC_ADAPT
av1_predict_intra_block_encoder_facade(x, ec_ctx, plane, block, blk_col,
blk_row, tx_size, plane_bsize);
#else
av1_predict_intra_block_facade(xd, plane, block, blk_col, blk_row, tx_size);
#endif
#if CONFIG_DPCM_INTRA
const int block_raster_idx = av1_block_index_to_raster_order(tx_size, block);
const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
const PREDICTION_MODE mode =
(plane == 0) ? get_y_mode(xd->mi[0], block_raster_idx) : mbmi->uv_mode;
if (av1_use_dpcm_intra(plane, mode, tx_type, mbmi)) {
av1_encode_block_intra_dpcm(cm, x, mode, plane, block, blk_row, blk_col,
plane_bsize, tx_size, tx_type, args->ta,
args->tl, args->skip);
return;
}
#endif // CONFIG_DPCM_INTRA
av1_subtract_txb(x, plane, plane_bsize, blk_col, blk_row, tx_size);
const ENTROPY_CONTEXT *a = &args->ta[blk_col];
@ -1458,9 +1751,7 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
if (args->enable_optimize_b) {
av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
ctx, AV1_XFORM_QUANT_FP);
if (p->eobs[block]) {
av1_optimize_b(cm, x, plane, block, tx_size, ctx);
}
av1_optimize_b(cm, x, plane, block, plane_bsize, tx_size, a, l);
} else {
av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize, tx_size,
ctx, AV1_XFORM_QUANT_B);
@ -1480,16 +1771,216 @@ void av1_encode_block_intra(int plane, int block, int blk_row, int blk_col,
// Note : *(args->skip) == mbmi->skip
#endif
#if CONFIG_CFL
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
if (plane == AOM_PLANE_Y && x->cfl_store_y) {
cfl_store(xd->cfl, dst, dst_stride, blk_row, blk_col, tx_size);
}
if (mbmi->uv_mode == DC_PRED) {
// TODO(ltrudeau) find a cleaner way to detect last transform block
if (plane == AOM_PLANE_U) {
xd->cfl->num_tx_blk[CFL_PRED_U] =
(blk_row == 0 && blk_col == 0) ? 1
: xd->cfl->num_tx_blk[CFL_PRED_U] + 1;
}
if (plane == AOM_PLANE_V) {
xd->cfl->num_tx_blk[CFL_PRED_V] =
(blk_row == 0 && blk_col == 0) ? 1
: xd->cfl->num_tx_blk[CFL_PRED_V] + 1;
if (mbmi->skip &&
xd->cfl->num_tx_blk[CFL_PRED_U] == xd->cfl->num_tx_blk[CFL_PRED_V]) {
assert(plane_bsize != BLOCK_INVALID);
const int block_width = block_size_wide[plane_bsize];
const int block_height = block_size_high[plane_bsize];
// if SKIP is chosen at the block level, and ind != 0, we must change
// the prediction
if (mbmi->cfl_alpha_idx != 0) {
const struct macroblockd_plane *const pd_cb = &xd->plane[AOM_PLANE_U];
uint8_t *const dst_cb = pd_cb->dst.buf;
const int dst_stride_cb = pd_cb->dst.stride;
uint8_t *const dst_cr = pd->dst.buf;
const int dst_stride_cr = pd->dst.stride;
for (int j = 0; j < block_height; j++) {
for (int i = 0; i < block_width; i++) {
dst_cb[dst_stride_cb * j + i] =
(uint8_t)(xd->cfl->dc_pred[CFL_PRED_U] + 0.5);
dst_cr[dst_stride_cr * j + i] =
(uint8_t)(xd->cfl->dc_pred[CFL_PRED_V] + 0.5);
}
}
mbmi->cfl_alpha_idx = 0;
mbmi->cfl_alpha_signs[CFL_PRED_U] = CFL_SIGN_POS;
mbmi->cfl_alpha_signs[CFL_PRED_V] = CFL_SIGN_POS;
}
}
}
}
#endif
}
#if CONFIG_CFL
static int cfl_alpha_dist(const uint8_t *y_pix, int y_stride, double y_avg,
const uint8_t *src, int src_stride, int blk_width,
int blk_height, double dc_pred, double alpha,
int *dist_neg_out) {
const double dc_pred_bias = dc_pred + 0.5;
int dist = 0;
int diff;
if (alpha == 0.0) {
const int dc_pred_i = (int)dc_pred_bias;
for (int j = 0; j < blk_height; j++) {
for (int i = 0; i < blk_width; i++) {
diff = src[i] - dc_pred_i;
dist += diff * diff;
}
src += src_stride;
}
if (dist_neg_out) *dist_neg_out = dist;
return dist;
}
int dist_neg = 0;
for (int j = 0; j < blk_height; j++) {
for (int i = 0; i < blk_width; i++) {
const double scaled_luma = alpha * (y_pix[i] - y_avg);
const int uv = src[i];
diff = uv - (int)(scaled_luma + dc_pred_bias);
dist += diff * diff;
diff = uv + (int)(scaled_luma - dc_pred_bias);
dist_neg += diff * diff;
}
y_pix += y_stride;
src += src_stride;
}
if (dist_neg_out) *dist_neg_out = dist_neg;
return dist;
}
static int cfl_compute_alpha_ind(MACROBLOCK *const x, const CFL_CTX *const cfl,
BLOCK_SIZE bsize,
CFL_SIGN_TYPE signs_out[CFL_SIGNS]) {
const struct macroblock_plane *const p_u = &x->plane[AOM_PLANE_U];
const struct macroblock_plane *const p_v = &x->plane[AOM_PLANE_V];
const uint8_t *const src_u = p_u->src.buf;
const uint8_t *const src_v = p_v->src.buf;
const int src_stride_u = p_u->src.stride;
const int src_stride_v = p_v->src.stride;
const int block_width = block_size_wide[bsize];
const int block_height = block_size_high[bsize];
const double dc_pred_u = cfl->dc_pred[CFL_PRED_U];
const double dc_pred_v = cfl->dc_pred[CFL_PRED_V];
// Temporary pixel buffer used to store the CfL prediction when we compute the
// alpha index.
uint8_t tmp_pix[MAX_SB_SQUARE];
// Load CfL Prediction over the entire block
const double y_avg =
cfl_load(cfl, tmp_pix, MAX_SB_SIZE, 0, 0, block_width, block_height);
int sse[CFL_PRED_PLANES][CFL_MAGS_SIZE];
sse[CFL_PRED_U][0] =
cfl_alpha_dist(tmp_pix, MAX_SB_SIZE, y_avg, src_u, src_stride_u,
block_width, block_height, dc_pred_u, 0, NULL);
sse[CFL_PRED_V][0] =
cfl_alpha_dist(tmp_pix, MAX_SB_SIZE, y_avg, src_v, src_stride_v,
block_width, block_height, dc_pred_v, 0, NULL);
for (int m = 1; m < CFL_MAGS_SIZE; m += 2) {
assert(cfl_alpha_mags[m + 1] == -cfl_alpha_mags[m]);
sse[CFL_PRED_U][m] = cfl_alpha_dist(
tmp_pix, MAX_SB_SIZE, y_avg, src_u, src_stride_u, block_width,
block_height, dc_pred_u, cfl_alpha_mags[m], &sse[CFL_PRED_U][m + 1]);
sse[CFL_PRED_V][m] = cfl_alpha_dist(
tmp_pix, MAX_SB_SIZE, y_avg, src_v, src_stride_v, block_width,
block_height, dc_pred_v, cfl_alpha_mags[m], &sse[CFL_PRED_V][m + 1]);
}
int dist;
int64_t cost;
int64_t best_cost;
// Compute least squares parameter of the entire block
// IMPORTANT: We assume that the first code is 0,0
int ind = 0;
signs_out[CFL_PRED_U] = CFL_SIGN_POS;
signs_out[CFL_PRED_V] = CFL_SIGN_POS;
dist = sse[CFL_PRED_U][0] + sse[CFL_PRED_V][0];
dist *= 16;
best_cost = RDCOST(x->rdmult, x->rddiv, cfl->costs[0], dist);
for (int c = 1; c < CFL_ALPHABET_SIZE; c++) {
const int idx_u = cfl_alpha_codes[c][CFL_PRED_U];
const int idx_v = cfl_alpha_codes[c][CFL_PRED_V];
for (CFL_SIGN_TYPE sign_u = idx_u == 0; sign_u < CFL_SIGNS; sign_u++) {
for (CFL_SIGN_TYPE sign_v = idx_v == 0; sign_v < CFL_SIGNS; sign_v++) {
dist = sse[CFL_PRED_U][idx_u + (sign_u == CFL_SIGN_NEG)] +
sse[CFL_PRED_V][idx_v + (sign_v == CFL_SIGN_NEG)];
dist *= 16;
cost = RDCOST(x->rdmult, x->rddiv, cfl->costs[c], dist);
if (cost < best_cost) {
best_cost = cost;
ind = c;
signs_out[CFL_PRED_U] = sign_u;
signs_out[CFL_PRED_V] = sign_v;
}
}
}
}
return ind;
}
static inline void cfl_update_costs(CFL_CTX *cfl, FRAME_CONTEXT *ec_ctx) {
assert(ec_ctx->cfl_alpha_cdf[CFL_ALPHABET_SIZE - 1] ==
AOM_ICDF(CDF_PROB_TOP));
const int prob_den = CDF_PROB_TOP;
int prob_num = AOM_ICDF(ec_ctx->cfl_alpha_cdf[0]);
cfl->costs[0] = av1_cost_zero(get_prob(prob_num, prob_den));
for (int c = 1; c < CFL_ALPHABET_SIZE; c++) {
int sign_bit_cost = (cfl_alpha_codes[c][CFL_PRED_U] != 0) +
(cfl_alpha_codes[c][CFL_PRED_V] != 0);
prob_num = AOM_ICDF(ec_ctx->cfl_alpha_cdf[c]) -
AOM_ICDF(ec_ctx->cfl_alpha_cdf[c - 1]);
cfl->costs[c] = av1_cost_zero(get_prob(prob_num, prob_den)) +
av1_cost_literal(sign_bit_cost);
}
}
void av1_predict_intra_block_encoder_facade(MACROBLOCK *x,
FRAME_CONTEXT *ec_ctx, int plane,
int block_idx, int blk_col,
int blk_row, TX_SIZE tx_size,
BLOCK_SIZE plane_bsize) {
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
if (plane != AOM_PLANE_Y && mbmi->uv_mode == DC_PRED) {
if (blk_col == 0 && blk_row == 0 && plane == AOM_PLANE_U) {
CFL_CTX *const cfl = xd->cfl;
cfl_update_costs(cfl, ec_ctx);
cfl_dc_pred(xd, plane_bsize, tx_size);
mbmi->cfl_alpha_idx =
cfl_compute_alpha_ind(x, cfl, plane_bsize, mbmi->cfl_alpha_signs);
}
}
av1_predict_intra_block_facade(xd, plane, block_idx, blk_col, blk_row,
tx_size);
}
#endif
void av1_encode_intra_block_plane(AV1_COMMON *cm, MACROBLOCK *x,
BLOCK_SIZE bsize, int plane,
int enable_optimize_b, const int mi_row,
const int mi_col) {
int enable_optimize_b, int mi_row,
int mi_col) {
const MACROBLOCKD *const xd = &x->e_mbd;
ENTROPY_CONTEXT ta[2 * MAX_MIB_SIZE] = { 0 };
ENTROPY_CONTEXT tl[2 * MAX_MIB_SIZE] = { 0 };
@ -1545,9 +2036,7 @@ PVQ_SKIP_TYPE av1_pvq_encode_helper(MACROBLOCK *x, tran_low_t *const coeff,
DECLARE_ALIGNED(16, int32_t, ref_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
DECLARE_ALIGNED(16, int32_t, out_int32[OD_TXSIZE_MAX * OD_TXSIZE_MAX]);
#if CONFIG_HIGHBITDEPTH
hbd_downshift = x->e_mbd.bd - 8;
#endif
assert(OD_COEFF_SHIFT >= 4);
// DC quantizer for PVQ
@ -1563,10 +2052,10 @@ PVQ_SKIP_TYPE av1_pvq_encode_helper(MACROBLOCK *x, tran_low_t *const coeff,
*eob = 0;
#if CONFIG_DAALA_EC
#if !CONFIG_ANS
tell = od_ec_enc_tell_frac(&daala_enc->w.ec);
#else
#error "CONFIG_PVQ currently requires CONFIG_DAALA_EC."
#error "CONFIG_PVQ currently requires !CONFIG_ANS."
#endif
// Change coefficient ordering for pvq encoding.
@ -1635,11 +2124,11 @@ PVQ_SKIP_TYPE av1_pvq_encode_helper(MACROBLOCK *x, tran_low_t *const coeff,
*eob = tx_blk_size * tx_blk_size;
#if CONFIG_DAALA_EC
#if !CONFIG_ANS
*rate = (od_ec_enc_tell_frac(&daala_enc->w.ec) - tell)
<< (AV1_PROB_COST_SHIFT - OD_BITRES);
#else
#error "CONFIG_PVQ currently requires CONFIG_DAALA_EC."
#error "CONFIG_PVQ currently requires !CONFIG_ANS."
#endif
assert(*rate >= 0);

Some files were not shown because too many files have changed in this diff Show more