Google Git
Sign in
aomedia / aom / a6fe432e4f6876755f7c0575ae834d5735519756 / . / av1 / decoder / decodeframe.c
blob: d03601f1e0925b0ccc92f94a4e41b6cc6d32b43a [file] [log] [blame]
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <stddef.h>
#include "config/aom_config.h"
#include "config/aom_dsp_rtcd.h"
#include "config/aom_scale_rtcd.h"
#include "config/av1_rtcd.h"
#include "aom/aom_codec.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/binary_codes_reader.h"
#include "aom_dsp/bitreader.h"
#include "aom_dsp/bitreader_buffer.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/aom_timer.h"
#include "aom_ports/mem.h"
#include "aom_ports/mem_ops.h"
#include "aom_scale/aom_scale.h"
#include "aom_util/aom_thread.h"
#if CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
#include "aom_util/debug_util.h"
#endif // CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG
#include "av1/common/alloccommon.h"
#include "av1/common/cdef.h"
#include "av1/common/cfl.h"
#if CONFIG_INSPECTION
#include "av1/decoder/inspection.h"
#endif
#include "av1/common/common.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/entropymv.h"
#include "av1/common/frame_buffers.h"
#include "av1/common/idct.h"
#include "av1/common/mvref_common.h"
#include "av1/common/pred_common.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
#include "av1/common/resize.h"
#include "av1/common/seg_common.h"
#include "av1/common/thread_common.h"
#include "av1/common/tile_common.h"
#include "av1/common/warped_motion.h"
#include "av1/common/obmc.h"
#include "av1/decoder/decodeframe.h"
#include "av1/decoder/decodemv.h"
#include "av1/decoder/decoder.h"
#include "av1/decoder/decodetxb.h"
#include "av1/decoder/detokenize.h"
#define ACCT_STR __func__
#define AOM_MIN_THREADS_PER_TILE 1
#define AOM_MAX_THREADS_PER_TILE 2
// This is needed by ext_tile related unit tests.
#define EXT_TILE_DEBUG 1
#define MC_TEMP_BUF_PELS \
(((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2) * \
((MAX_SB_SIZE)*2 + (AOM_INTERP_EXTEND)*2))
// Checks that the remaining bits start with a 1 and ends with 0s.
// It consumes an additional byte, if already byte aligned before the check.
int av1_check_trailing_bits(AV1Decoder *pbi, struct aom_read_bit_buffer *rb) {
AV1_COMMON *const cm = &pbi->common;
// bit_offset is set to 0 (mod 8) when the reader is already byte aligned
int bits_before_alignment = 8 - rb->bit_offset % 8;
int trailing = aom_rb_read_literal(rb, bits_before_alignment);
if (trailing != (1 << (bits_before_alignment - 1))) {
cm->error.error_code = AOM_CODEC_CORRUPT_FRAME;
return -1;
}
return 0;
}
// Use only_chroma = 1 to only set the chroma planes
static void set_planes_to_neutral_grey(const SequenceHeader *const seq_params,
const YV12_BUFFER_CONFIG *const buf,
int only_chroma) {
if (seq_params->use_highbitdepth) {
const int val = 1 << (seq_params->bit_depth - 1);
for (int plane = only_chroma; plane < MAX_MB_PLANE; plane++) {
const int is_uv = plane > 0;
uint16_t *const base = CONVERT_TO_SHORTPTR(buf->buffers[plane]);
// Set the first row to neutral grey. Then copy the first row to all
// subsequent rows.
if (buf->crop_heights[is_uv] > 0) {
aom_memset16(base, val, buf->crop_widths[is_uv]);
for (int row_idx = 1; row_idx < buf->crop_heights[is_uv]; row_idx++) {
memcpy(&base[row_idx * buf->strides[is_uv]], base,
sizeof(*base) * buf->crop_widths[is_uv]);
}
}
}
} else {
for (int plane = only_chroma; plane < MAX_MB_PLANE; plane++) {
const int is_uv = plane > 0;
for (int row_idx = 0; row_idx < buf->crop_heights[is_uv]; row_idx++) {
memset(&buf->buffers[plane][row_idx * buf->uv_stride], 1 << 7,
buf->crop_widths[is_uv]);
}
}
}
}
static void loop_restoration_read_sb_coeffs(const AV1_COMMON *const cm,
MACROBLOCKD *xd,
aom_reader *const r, int plane,
int runit_idx);
static void setup_compound_reference_mode(AV1_COMMON *cm) {
cm->comp_fwd_ref[0] = LAST_FRAME;
cm->comp_fwd_ref[1] = LAST2_FRAME;
cm->comp_fwd_ref[2] = LAST3_FRAME;
cm->comp_fwd_ref[3] = GOLDEN_FRAME;
cm->comp_bwd_ref[0] = BWDREF_FRAME;
cm->comp_bwd_ref[1] = ALTREF2_FRAME;
cm->comp_bwd_ref[2] = ALTREF_FRAME;
}
static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
return len != 0 && len <= (size_t)(end - start);
}
static TX_MODE read_tx_mode(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
if (cm->coded_lossless) return ONLY_4X4;
return aom_rb_read_bit(rb) ? TX_MODE_SELECT : TX_MODE_LARGEST;
}
static REFERENCE_MODE read_frame_reference_mode(
const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
if (frame_is_intra_only(cm)) {
return SINGLE_REFERENCE;
} else {
return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE;
}
}
static void inverse_transform_block(MACROBLOCKD *xd, int plane,
const TX_TYPE tx_type,
const TX_SIZE tx_size, uint8_t *dst,
int stride, int reduced_tx_set) {
struct macroblockd_plane *const pd = &xd->plane[plane];
tran_low_t *const dqcoeff = pd->dqcoeff_block + xd->cb_offset[plane];
eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
uint16_t scan_line = eob_data->max_scan_line;
uint16_t eob = eob_data->eob;
av1_inverse_transform_block(xd, dqcoeff, plane, tx_type, tx_size, dst, stride,
eob, reduced_tx_set);
memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
}
static void read_coeffs_tx_intra_block(const AV1_COMMON *const cm,
MACROBLOCKD *const xd,
aom_reader *const r, const int plane,
const int row, const int col,
const TX_SIZE tx_size) {
MB_MODE_INFO *mbmi = xd->mi[0];
if (!mbmi->skip) {
#if TXCOEFF_TIMER
struct aom_usec_timer timer;
aom_usec_timer_start(&timer);
#endif
av1_read_coeffs_txb_facade(cm, xd, r, plane, row, col, tx_size);
#if TXCOEFF_TIMER
aom_usec_timer_mark(&timer);
const int64_t elapsed_time = aom_usec_timer_elapsed(&timer);
cm->txcoeff_timer += elapsed_time;
++cm->txb_count;
#endif
}
}
static void decode_block_void(const AV1_COMMON *const cm, MACROBLOCKD *const xd,
aom_reader *const r, const int plane,
const int row, const int col,
const TX_SIZE tx_size) {
(void)cm;
(void)xd;
(void)r;
(void)plane;
(void)row;
(void)col;
(void)tx_size;
}
static void predict_inter_block_void(AV1_COMMON *const cm,
MACROBLOCKD *const xd, int mi_row,
int mi_col, BLOCK_SIZE bsize) {
(void)cm;
(void)xd;
(void)mi_row;
(void)mi_col;
(void)bsize;
}
static void cfl_store_inter_block_void(AV1_COMMON *const cm,
MACROBLOCKD *const xd) {
(void)cm;
(void)xd;
}
static void predict_and_reconstruct_intra_block(
const AV1_COMMON *const cm, MACROBLOCKD *const xd, aom_reader *const r,
const int plane, const int row, const int col, const TX_SIZE tx_size) {
(void)r;
MB_MODE_INFO *mbmi = xd->mi[0];
PLANE_TYPE plane_type = get_plane_type(plane);
av1_predict_intra_block_facade(cm, xd, plane, col, row, tx_size);
if (!mbmi->skip) {
struct macroblockd_plane *const pd = &xd->plane[plane];
// tx_type will be read out in av1_read_coeffs_txb_facade
const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, row, col, tx_size,
cm->reduced_tx_set_used);
eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
if (eob_data->eob) {
uint8_t *dst =
&pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
cm->reduced_tx_set_used);
}
}
if (plane == AOM_PLANE_Y && store_cfl_required(cm, xd)) {
cfl_store_tx(xd, row, col, tx_size, mbmi->sb_type);
}
}
static void inverse_transform_inter_block(const AV1_COMMON *const cm,
MACROBLOCKD *const xd,
aom_reader *const r, const int plane,
const int blk_row, const int blk_col,
const TX_SIZE tx_size) {
(void)r;
PLANE_TYPE plane_type = get_plane_type(plane);
const struct macroblockd_plane *const pd = &xd->plane[plane];
// tx_type will be read out in av1_read_coeffs_txb_facade
const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, blk_row, blk_col,
tx_size, cm->reduced_tx_set_used);
uint8_t *dst =
&pd->dst
.buf[(blk_row * pd->dst.stride + blk_col) << tx_size_wide_log2[0]];
inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
cm->reduced_tx_set_used);
#if CONFIG_MISMATCH_DEBUG
int pixel_c, pixel_r;
BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
int blk_w = block_size_wide[bsize];
int blk_h = block_size_high[bsize];
mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, blk_col, blk_row,
pd->subsampling_x, pd->subsampling_y);
mismatch_check_block_tx(dst, pd->dst.stride, cm->current_frame.order_hint,
plane, pixel_c, pixel_r, blk_w, blk_h,
xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
#endif
}
static void set_cb_buffer_offsets(MACROBLOCKD *const xd, TX_SIZE tx_size,
int plane) {
xd->cb_offset[plane] += tx_size_wide[tx_size] * tx_size_high[tx_size];
xd->txb_offset[plane] =
xd->cb_offset[plane] / (TX_SIZE_W_MIN * TX_SIZE_H_MIN);
}
static void decode_reconstruct_tx(AV1_COMMON *cm, ThreadData *const td,
aom_reader *r, MB_MODE_INFO *const mbmi,
int plane, BLOCK_SIZE plane_bsize,
int blk_row, int blk_col, int block,
TX_SIZE tx_size, int *eob_total) {
MACROBLOCKD *const xd = &td->xd;
const struct macroblockd_plane *const pd = &xd->plane[plane];
const TX_SIZE plane_tx_size =
plane ? av1_get_max_uv_txsize(mbmi->sb_type, pd->subsampling_x,
pd->subsampling_y)
: mbmi->inter_tx_size[av1_get_txb_size_index(plane_bsize, blk_row,
blk_col)];
// Scale to match transform block unit.
const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
if (tx_size == plane_tx_size || plane) {
td->read_coeffs_tx_inter_block_visit(cm, xd, r, plane, blk_row, blk_col,
tx_size);
td->inverse_tx_inter_block_visit(cm, xd, r, plane, blk_row, blk_col,
tx_size);
eob_info *eob_data = pd->eob_data + xd->txb_offset[plane];
*eob_total += eob_data->eob;
set_cb_buffer_offsets(xd, tx_size, plane);
} else {
const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
assert(IMPLIES(tx_size <= TX_4X4, sub_txs == tx_size));
assert(IMPLIES(tx_size > TX_4X4, sub_txs < tx_size));
const int bsw = tx_size_wide_unit[sub_txs];
const int bsh = tx_size_high_unit[sub_txs];
const int sub_step = bsw * bsh;
assert(bsw > 0 && bsh > 0);
for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
const int offsetr = blk_row + row;
const int offsetc = blk_col + col;
if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
decode_reconstruct_tx(cm, td, r, mbmi, plane, plane_bsize, offsetr,
offsetc, block, sub_txs, eob_total);
block += sub_step;
}
}
}
}
static void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
BLOCK_SIZE bsize, int mi_row, int mi_col, int bw,
int bh, int x_mis, int y_mis) {
const int num_planes = av1_num_planes(cm);
const int offset = mi_row * cm->mi_stride + mi_col;
const TileInfo *const tile = &xd->tile;
xd->mi = cm->mi_grid_visible + offset;
xd->mi[0] = &cm->mi[offset];
// TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
// passing bsize from decode_partition().
xd->mi[0]->sb_type = bsize;
#if CONFIG_RD_DEBUG
xd->mi[0]->mi_row = mi_row;
xd->mi[0]->mi_col = mi_col;
#endif
xd->cfl.mi_row = mi_row;
xd->cfl.mi_col = mi_col;
assert(x_mis && y_mis);
for (int x = 1; x < x_mis; ++x) xd->mi[x] = xd->mi[0];
int idx = cm->mi_stride;
for (int y = 1; y < y_mis; ++y) {
memcpy(&xd->mi[idx], &xd->mi[0], x_mis * sizeof(xd->mi[0]));
idx += cm->mi_stride;
}
set_plane_n4(xd, bw, bh, num_planes);
set_skip_context(xd, mi_row, mi_col, num_planes);
// Distance of Mb to the various image edges. These are specified to 8th pel
// as they are always compared to values that are in 1/8th pel units
set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0,
num_planes);
}
static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
int mi_row, int mi_col, aom_reader *r,
PARTITION_TYPE partition, BLOCK_SIZE bsize) {
AV1_COMMON *const cm = &pbi->common;
const SequenceHeader *const seq_params = &cm->seq_params;
const int bw = mi_size_wide[bsize];
const int bh = mi_size_high[bsize];
const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
#if CONFIG_ACCOUNTING
aom_accounting_set_context(&pbi->accounting, mi_col, mi_row);
#endif
set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
xd->mi[0]->partition = partition;
av1_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
if (bsize >= BLOCK_8X8 &&
(seq_params->subsampling_x || seq_params->subsampling_y)) {
const BLOCK_SIZE uv_subsize =
ss_size_lookup[bsize][seq_params->subsampling_x]
[seq_params->subsampling_y];
if (uv_subsize == BLOCK_INVALID)
aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
"Invalid block size.");
}
}
typedef struct PadBlock {
int x0;
int x1;
int y0;
int y1;
} PadBlock;
static void highbd_build_mc_border(const uint8_t *src8, int src_stride,
uint8_t *dst8, int dst_stride, int x, int y,
int b_w, int b_h, int w, int h) {
// Get a pointer to the start of the real data for this row.
const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
const uint16_t *ref_row = src - x - y * src_stride;
if (y >= h)
ref_row += (h - 1) * src_stride;
else if (y > 0)
ref_row += y * src_stride;
do {
int right = 0, copy;
int left = x < 0 ? -x : 0;
if (left > b_w) left = b_w;
if (x + b_w > w) right = x + b_w - w;
if (right > b_w) right = b_w;
copy = b_w - left - right;
if (left) aom_memset16(dst, ref_row[0], left);
if (copy) memcpy(dst + left, ref_row + x + left, copy * sizeof(uint16_t));
if (right) aom_memset16(dst + left + copy, ref_row[w - 1], right);
dst += dst_stride;
++y;
if (y > 0 && y < h) ref_row += src_stride;
} while (--b_h);
}
static void build_mc_border(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int x, int y, int b_w, int b_h,
int w, int h) {
// Get a pointer to the start of the real data for this row.
const uint8_t *ref_row = src - x - y * src_stride;
if (y >= h)
ref_row += (h - 1) * src_stride;
else if (y > 0)
ref_row += y * src_stride;
do {
int right = 0, copy;
int left = x < 0 ? -x : 0;
if (left > b_w) left = b_w;
if (x + b_w > w) right = x + b_w - w;
if (right > b_w) right = b_w;
copy = b_w - left - right;
if (left) memset(dst, ref_row[0], left);
if (copy) memcpy(dst + left, ref_row + x + left, copy);
if (right) memset(dst + left + copy, ref_row[w - 1], right);
dst += dst_stride;
++y;
if (y > 0 && y < h) ref_row += src_stride;
} while (--b_h);
}
static INLINE int update_extend_mc_border_params(
const struct scale_factors *const sf, struct buf_2d *const pre_buf,
MV32 scaled_mv, PadBlock *block, int subpel_x_mv, int subpel_y_mv,
int do_warp, int is_intrabc, int *x_pad, int *y_pad) {
const int is_scaled = av1_is_scaled(sf);
// Get reference width and height.
int frame_width = pre_buf->width;
int frame_height = pre_buf->height;
// Do border extension if there is motion or
// width/height is not a multiple of 8 pixels.
if ((!is_intrabc) && (!do_warp) &&
(is_scaled || scaled_mv.col || scaled_mv.row || (frame_width & 0x7) ||
(frame_height & 0x7))) {
if (subpel_x_mv || (sf->x_step_q4 != SUBPEL_SHIFTS)) {
block->x0 -= AOM_INTERP_EXTEND - 1;
block->x1 += AOM_INTERP_EXTEND;
*x_pad = 1;
}
if (subpel_y_mv || (sf->y_step_q4 != SUBPEL_SHIFTS)) {
block->y0 -= AOM_INTERP_EXTEND - 1;
block->y1 += AOM_INTERP_EXTEND;
*y_pad = 1;
}
// Skip border extension if block is inside the frame.
if (block->x0 < 0 || block->x1 > frame_width - 1 || block->y0 < 0 ||
block->y1 > frame_height - 1) {
return 1;
}
}
return 0;
}
static INLINE void extend_mc_border(const struct scale_factors *const sf,
struct buf_2d *const pre_buf,
MV32 scaled_mv, PadBlock block,
int subpel_x_mv, int subpel_y_mv,
int do_warp, int is_intrabc, int highbd,
uint8_t *mc_buf, uint8_t **pre,
int *src_stride) {
int x_pad = 0, y_pad = 0;
if (update_extend_mc_border_params(sf, pre_buf, scaled_mv, &block,
subpel_x_mv, subpel_y_mv, do_warp,
is_intrabc, &x_pad, &y_pad)) {
// Get reference block pointer.
const uint8_t *const buf_ptr =
pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
int buf_stride = pre_buf->stride;
const int b_w = block.x1 - block.x0;
const int b_h = block.y1 - block.y0;
// Extend the border.
if (highbd) {
highbd_build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0,
block.y0, b_w, b_h, pre_buf->width,
pre_buf->height);
} else {
build_mc_border(buf_ptr, buf_stride, mc_buf, b_w, block.x0, block.y0, b_w,
b_h, pre_buf->width, pre_buf->height);
}
*src_stride = b_w;
*pre = mc_buf + y_pad * (AOM_INTERP_EXTEND - 1) * b_w +
x_pad * (AOM_INTERP_EXTEND - 1);
}
}
static INLINE void dec_calc_subpel_params(
MACROBLOCKD *xd, const struct scale_factors *const sf, const MV mv,
int plane, const int pre_x, const int pre_y, int x, int y,
struct buf_2d *const pre_buf, SubpelParams *subpel_params, int bw, int bh,
PadBlock *block, int mi_x, int mi_y, MV32 *scaled_mv, int *subpel_x_mv,
int *subpel_y_mv) {
struct macroblockd_plane *const pd = &xd->plane[plane];
const int is_scaled = av1_is_scaled(sf);
if (is_scaled) {
int ssx = pd->subsampling_x;
int ssy = pd->subsampling_y;
int orig_pos_y = (pre_y + y) << SUBPEL_BITS;
orig_pos_y += mv.row * (1 << (1 - ssy));
int orig_pos_x = (pre_x + x) << SUBPEL_BITS;
orig_pos_x += mv.col * (1 << (1 - ssx));
int pos_y = sf->scale_value_y(orig_pos_y, sf);
int pos_x = sf->scale_value_x(orig_pos_x, sf);
pos_x += SCALE_EXTRA_OFF;
pos_y += SCALE_EXTRA_OFF;
const int top = -AOM_LEFT_TOP_MARGIN_SCALED(ssy);
const int left = -AOM_LEFT_TOP_MARGIN_SCALED(ssx);
const int bottom = (pre_buf->height + AOM_INTERP_EXTEND)
<< SCALE_SUBPEL_BITS;
const int right = (pre_buf->width + AOM_INTERP_EXTEND) << SCALE_SUBPEL_BITS;
pos_y = clamp(pos_y, top, bottom);
pos_x = clamp(pos_x, left, right);
subpel_params->subpel_x = pos_x & SCALE_SUBPEL_MASK;
subpel_params->subpel_y = pos_y & SCALE_SUBPEL_MASK;
subpel_params->xs = sf->x_step_q4;
subpel_params->ys = sf->y_step_q4;
// Get reference block top left coordinate.
block->x0 = pos_x >> SCALE_SUBPEL_BITS;
block->y0 = pos_y >> SCALE_SUBPEL_BITS;
// Get reference block bottom right coordinate.
block->x1 =
((pos_x + (bw - 1) * subpel_params->xs) >> SCALE_SUBPEL_BITS) + 1;
block->y1 =
((pos_y + (bh - 1) * subpel_params->ys) >> SCALE_SUBPEL_BITS) + 1;
MV temp_mv;
temp_mv = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh, pd->subsampling_x,
pd->subsampling_y);
*scaled_mv = av1_scale_mv(&temp_mv, (mi_x + x), (mi_y + y), sf);
scaled_mv->row += SCALE_EXTRA_OFF;
scaled_mv->col += SCALE_EXTRA_OFF;
*subpel_x_mv = scaled_mv->col & SCALE_SUBPEL_MASK;
*subpel_y_mv = scaled_mv->row & SCALE_SUBPEL_MASK;
} else {
// Get block position in current frame.
int pos_x = (pre_x + x) << SUBPEL_BITS;
int pos_y = (pre_y + y) << SUBPEL_BITS;
const MV mv_q4 = clamp_mv_to_umv_border_sb(
xd, &mv, bw, bh, pd->subsampling_x, pd->subsampling_y);
subpel_params->xs = subpel_params->ys = SCALE_SUBPEL_SHIFTS;
subpel_params->subpel_x = (mv_q4.col & SUBPEL_MASK) << SCALE_EXTRA_BITS;
subpel_params->subpel_y = (mv_q4.row & SUBPEL_MASK) << SCALE_EXTRA_BITS;
// Get reference block top left coordinate.
pos_x += mv_q4.col;
pos_y += mv_q4.row;
block->x0 = pos_x >> SUBPEL_BITS;
block->y0 = pos_y >> SUBPEL_BITS;
// Get reference block bottom right coordinate.
block->x1 = (pos_x >> SUBPEL_BITS) + (bw - 1) + 1;
block->y1 = (pos_y >> SUBPEL_BITS) + (bh - 1) + 1;
scaled_mv->row = mv_q4.row;
scaled_mv->col = mv_q4.col;
*subpel_x_mv = scaled_mv->col & SUBPEL_MASK;
*subpel_y_mv = scaled_mv->row & SUBPEL_MASK;
}
}
static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
MACROBLOCKD *xd, int plane,
const MB_MODE_INFO *mi,
int build_for_obmc, int bw,
int bh, int mi_x, int mi_y) {
struct macroblockd_plane *const pd = &xd->plane[plane];
int is_compound = has_second_ref(mi);
int ref;
const int is_intrabc = is_intrabc_block(mi);
assert(IMPLIES(is_intrabc, !is_compound));
int is_global[2] = { 0, 0 };
for (ref = 0; ref < 1 + is_compound; ++ref) {
const WarpedMotionParams *const wm = &xd->global_motion[mi->ref_frame[ref]];
is_global[ref] = is_global_mv_block(mi, wm->wmtype);
}
const BLOCK_SIZE bsize = mi->sb_type;
const int ss_x = pd->subsampling_x;
const int ss_y = pd->subsampling_y;
int sub8x8_inter = (block_size_wide[bsize] < 8 && ss_x) ||
(block_size_high[bsize] < 8 && ss_y);
if (is_intrabc) sub8x8_inter = 0;
// For sub8x8 chroma blocks, we may be covering more than one luma block's
// worth of pixels. Thus (mi_x, mi_y) may not be the correct coordinates for
// the top-left corner of the prediction source - the correct top-left corner
// is at (pre_x, pre_y).
const int row_start =
(block_size_high[bsize] == 4) && ss_y && !build_for_obmc ? -1 : 0;
const int col_start =
(block_size_wide[bsize] == 4) && ss_x && !build_for_obmc ? -1 : 0;
const int pre_x = (mi_x + MI_SIZE * col_start) >> ss_x;
const int pre_y = (mi_y + MI_SIZE * row_start) >> ss_y;
sub8x8_inter = sub8x8_inter && !build_for_obmc;
if (sub8x8_inter) {
for (int row = row_start; row <= 0 && sub8x8_inter; ++row) {
for (int col = col_start; col <= 0; ++col) {
const MB_MODE_INFO *this_mbmi = xd->mi[row * xd->mi_stride + col];
if (!is_inter_block(this_mbmi)) sub8x8_inter = 0;
if (is_intrabc_block(this_mbmi)) sub8x8_inter = 0;
}
}
}
if (sub8x8_inter) {
// block size
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;
assert(!is_compound);
const struct buf_2d orig_pred_buf[2] = { pd->pre[0], pd->pre[1] };
int row = row_start;
int src_stride;
for (int y = 0; y < b8_h; y += b4_h) {
int col = col_start;
for (int x = 0; x < b8_w; x += b4_w) {
MB_MODE_INFO *this_mbmi = xd->mi[row * xd->mi_stride + col];
is_compound = has_second_ref(this_mbmi);
int tmp_dst_stride = 8;
assert(bw < 8 || bh < 8);
ConvolveParams conv_params = get_conv_params_no_round(
0, plane, xd->tmp_conv_dst, tmp_dst_stride, is_compound, xd->bd);
conv_params.use_dist_wtd_comp_avg = 0;
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x;
ref = 0;
const RefCntBuffer *ref_buf =
get_ref_frame_buf(cm, this_mbmi->ref_frame[ref]);
const struct scale_factors *ref_scale_factors =
get_ref_scale_factors_const(cm, this_mbmi->ref_frame[ref]);
pd->pre[ref].buf0 =
(plane == 1) ? ref_buf->buf.u_buffer : ref_buf->buf.v_buffer;
pd->pre[ref].buf =
pd->pre[ref].buf0 + scaled_buffer_offset(pre_x, pre_y,
ref_buf->buf.uv_stride,
ref_scale_factors);
pd->pre[ref].width = ref_buf->buf.uv_crop_width;
pd->pre[ref].height = ref_buf->buf.uv_crop_height;
pd->pre[ref].stride = ref_buf->buf.uv_stride;
const struct scale_factors *const sf =
is_intrabc ? &cm->sf_identity : ref_scale_factors;
struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
const MV mv = this_mbmi->mv[ref].as_mv;
uint8_t *pre;
SubpelParams subpel_params;
PadBlock block;
MV32 scaled_mv;
int subpel_x_mv, subpel_y_mv;
int highbd;
WarpTypesAllowed warp_types;
warp_types.global_warp_allowed = is_global[ref];
warp_types.local_warp_allowed = this_mbmi->motion_mode == WARPED_CAUSAL;
dec_calc_subpel_params(xd, sf, mv, plane, pre_x, pre_y, x, y, pre_buf,
&subpel_params, bw, bh, &block, mi_x, mi_y,
&scaled_mv, &subpel_x_mv, &subpel_y_mv);
pre = pre_buf->buf0 + block.y0 * pre_buf->stride + block.x0;
src_stride = pre_buf->stride;
highbd = is_cur_buf_hbd(xd);
extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv,
subpel_y_mv, 0, is_intrabc, highbd, xd->mc_buf[ref],
&pre, &src_stride);
conv_params.do_average = ref;
if (is_masked_compound_type(mi->interinter_comp.type)) {
// masked compound type has its own average mechanism
conv_params.do_average = 0;
}
av1_make_inter_predictor(
pre, src_stride, dst, dst_buf->stride, &subpel_params, sf, b4_w,
b4_h, &conv_params, this_mbmi->interp_filters, &warp_types,
(mi_x >> pd->subsampling_x) + x, (mi_y >> pd->subsampling_y) + y,
plane, ref, mi, build_for_obmc, xd, cm->allow_warped_motion);
++col;
}
++row;
}
for (ref = 0; ref < 2; ++ref) pd->pre[ref] = orig_pred_buf[ref];
return;
}
{
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *const dst = dst_buf->buf;
uint8_t *pre[2];
SubpelParams subpel_params[2];
int src_stride[2];
for (ref = 0; ref < 1 + is_compound; ++ref) {
const struct scale_factors *const sf =
is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref];
struct buf_2d *const pre_buf = is_intrabc ? dst_buf : &pd->pre[ref];
const MV mv = mi->mv[ref].as_mv;
PadBlock block;
MV32 scaled_mv;
int subpel_x_mv, subpel_y_mv;
int highbd;
dec_calc_subpel_params(xd, sf, mv, plane, pre_x, pre_y, 0, 0, pre_buf,
&subpel_params[ref], bw, bh, &block, mi_x, mi_y,
&scaled_mv, &subpel_x_mv, &subpel_y_mv);
pre[ref] = pre_buf->buf0 + (int64_t)block.y0 * pre_buf->stride + block.x0;
src_stride[ref] = pre_buf->stride;
highbd = is_cur_buf_hbd(xd);
WarpTypesAllowed warp_types;
warp_types.global_warp_allowed = is_global[ref];
warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
int do_warp = (bw >= 8 && bh >= 8 &&
av1_allow_warp(mi, &warp_types,
&xd->global_motion[mi->ref_frame[ref]],
build_for_obmc, sf, NULL));
do_warp = (do_warp && xd->cur_frame_force_integer_mv == 0);
extend_mc_border(sf, pre_buf, scaled_mv, block, subpel_x_mv, subpel_y_mv,
do_warp, is_intrabc, highbd, xd->mc_buf[ref], &pre[ref],
&src_stride[ref]);
}
ConvolveParams conv_params = get_conv_params_no_round(
0, plane, xd->tmp_conv_dst, MAX_SB_SIZE, is_compound, xd->bd);
av1_dist_wtd_comp_weight_assign(
cm, mi, 0, &conv_params.fwd_offset, &conv_params.bck_offset,
&conv_params.use_dist_wtd_comp_avg, is_compound);
for (ref = 0; ref < 1 + is_compound; ++ref) {
const struct scale_factors *const sf =
is_intrabc ? &cm->sf_identity : xd->block_ref_scale_factors[ref];
WarpTypesAllowed warp_types;
warp_types.global_warp_allowed = is_global[ref];
warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
conv_params.do_average = ref;
if (is_masked_compound_type(mi->interinter_comp.type)) {
// masked compound type has its own average mechanism
conv_params.do_average = 0;
}
if (ref && is_masked_compound_type(mi->interinter_comp.type))
av1_make_masked_inter_predictor(
pre[ref], src_stride[ref], dst, dst_buf->stride,
&subpel_params[ref], sf, bw, bh, &conv_params, mi->interp_filters,
plane, &warp_types, mi_x >> pd->subsampling_x,
mi_y >> pd->subsampling_y, ref, xd, cm->allow_warped_motion);
else
av1_make_inter_predictor(
pre[ref], src_stride[ref], dst, dst_buf->stride,
&subpel_params[ref], sf, bw, bh, &conv_params, mi->interp_filters,
&warp_types, mi_x >> pd->subsampling_x, mi_y >> pd->subsampling_y,
plane, ref, mi, build_for_obmc, xd, cm->allow_warped_motion);
}
}
}
static void dec_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;
const int mi_x = mi_col * MI_SIZE;
const int mi_y = mi_row * MI_SIZE;
for (plane = plane_from; plane <= plane_to; ++plane) {
const struct macroblockd_plane *pd = &xd->plane[plane];
const int bw = pd->width;
const int bh = pd->height;
if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
pd->subsampling_y))
continue;
dec_build_inter_predictors(cm, xd, plane, xd->mi[0], 0, bw, bh, mi_x, mi_y);
}
}
static void dec_build_inter_predictors_sby(const AV1_COMMON *cm,
MACROBLOCKD *xd, int mi_row,
int mi_col, const BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0, 0);
if (is_interintra_pred(xd->mi[0])) {
BUFFER_SET default_ctx = { { xd->plane[0].dst.buf, NULL, NULL },
{ xd->plane[0].dst.stride, 0, 0 } };
if (!ctx) ctx = &default_ctx;
av1_build_interintra_predictors_sbp(cm, xd, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, ctx, 0, bsize);
}
}
static void dec_build_inter_predictors_sbuv(const AV1_COMMON *cm,
MACROBLOCKD *xd, int mi_row,
int mi_col, const BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 1,
MAX_MB_PLANE - 1);
if (is_interintra_pred(xd->mi[0])) {
BUFFER_SET default_ctx = {
{ NULL, xd->plane[1].dst.buf, xd->plane[2].dst.buf },
{ 0, xd->plane[1].dst.stride, xd->plane[2].dst.stride }
};
if (!ctx) ctx = &default_ctx;
av1_build_interintra_predictors_sbuv(
cm, xd, xd->plane[1].dst.buf, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].dst.stride, ctx, bsize);
}
}
static void dec_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize) {
const int num_planes = av1_num_planes(cm);
dec_build_inter_predictors_sby(cm, xd, mi_row, mi_col, ctx, bsize);
if (num_planes > 1)
dec_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, ctx, bsize);
}
static INLINE void dec_build_prediction_by_above_pred(
MACROBLOCKD *xd, int rel_mi_col, uint8_t above_mi_width,
MB_MODE_INFO *above_mbmi, void *fun_ctxt, const int num_planes) {
struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt;
const int above_mi_col = ctxt->mi_col + rel_mi_col;
int mi_x, mi_y;
MB_MODE_INFO backup_mbmi = *above_mbmi;
av1_setup_build_prediction_by_above_pred(xd, rel_mi_col, above_mi_width,
&backup_mbmi, ctxt, num_planes);
mi_x = above_mi_col << MI_SIZE_LOG2;
mi_y = ctxt->mi_row << MI_SIZE_LOG2;
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
for (int j = 0; j < num_planes; ++j) {
const struct macroblockd_plane *pd = &xd->plane[j];
int bw = (above_mi_width * MI_SIZE) >> pd->subsampling_x;
int bh = clamp(block_size_high[bsize] >> (pd->subsampling_y + 1), 4,
block_size_high[BLOCK_64X64] >> (pd->subsampling_y + 1));
if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;
dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
mi_y);
}
}
static void dec_build_prediction_by_above_preds(
const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
if (!xd->up_available) return;
// Adjust mb_to_bottom_edge to have the correct value for the OBMC
// prediction block. This is half the height of the original block,
// except for 128-wide blocks, where we only use a height of 32.
int this_height = xd->n4_h * MI_SIZE;
int pred_height = AOMMIN(this_height / 2, 32);
xd->mb_to_bottom_edge += (this_height - pred_height) * 8;
struct build_prediction_ctxt ctxt = { cm, mi_row,
mi_col, tmp_buf,
tmp_width, tmp_height,
tmp_stride, xd->mb_to_right_edge };
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
foreach_overlappable_nb_above(cm, xd, mi_col,
max_neighbor_obmc[mi_size_wide_log2[bsize]],
dec_build_prediction_by_above_pred, &ctxt);
xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
xd->mb_to_right_edge = ctxt.mb_to_far_edge;
xd->mb_to_bottom_edge -= (this_height - pred_height) * 8;
}
static INLINE void dec_build_prediction_by_left_pred(
MACROBLOCKD *xd, int rel_mi_row, uint8_t left_mi_height,
MB_MODE_INFO *left_mbmi, void *fun_ctxt, const int num_planes) {
struct build_prediction_ctxt *ctxt = (struct build_prediction_ctxt *)fun_ctxt;
const int left_mi_row = ctxt->mi_row + rel_mi_row;
int mi_x, mi_y;
MB_MODE_INFO backup_mbmi = *left_mbmi;
av1_setup_build_prediction_by_left_pred(xd, rel_mi_row, left_mi_height,
&backup_mbmi, ctxt, num_planes);
mi_x = ctxt->mi_col << MI_SIZE_LOG2;
mi_y = left_mi_row << MI_SIZE_LOG2;
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
for (int j = 0; j < num_planes; ++j) {
const struct macroblockd_plane *pd = &xd->plane[j];
int bw = clamp(block_size_wide[bsize] >> (pd->subsampling_x + 1), 4,
block_size_wide[BLOCK_64X64] >> (pd->subsampling_x + 1));
int bh = (left_mi_height << MI_SIZE_LOG2) >> pd->subsampling_y;
if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;
dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
mi_y);
}
}
static void dec_build_prediction_by_left_preds(
const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
if (!xd->left_available) return;
// Adjust mb_to_right_edge to have the correct value for the OBMC
// prediction block. This is half the width of the original block,
// except for 128-wide blocks, where we only use a width of 32.
int this_width = xd->n4_w * MI_SIZE;
int pred_width = AOMMIN(this_width / 2, 32);
xd->mb_to_right_edge += (this_width - pred_width) * 8;
struct build_prediction_ctxt ctxt = { cm, mi_row,
mi_col, tmp_buf,
tmp_width, tmp_height,
tmp_stride, xd->mb_to_bottom_edge };
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
foreach_overlappable_nb_left(cm, xd, mi_row,
max_neighbor_obmc[mi_size_high_log2[bsize]],
dec_build_prediction_by_left_pred, &ctxt);
xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
xd->mb_to_right_edge -= (this_width - pred_width) * 8;
xd->mb_to_bottom_edge = ctxt.mb_to_far_edge;
}
static void dec_build_obmc_inter_predictors_sb(const AV1_COMMON *cm,
MACROBLOCKD *xd, int mi_row,
int mi_col) {
const int num_planes = av1_num_planes(cm);
uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE];
int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
if (is_cur_buf_hbd(xd)) {
int len = sizeof(uint16_t);
dst_buf1[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0]);
dst_buf1[1] =
CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * len);
dst_buf1[2] =
CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2 * len);
dst_buf2[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1]);
dst_buf2[1] =
CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * len);
dst_buf2[2] =
CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2 * len);
} else {
dst_buf1[0] = xd->tmp_obmc_bufs[0];
dst_buf1[1] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE;
dst_buf1[2] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2;
dst_buf2[0] = xd->tmp_obmc_bufs[1];
dst_buf2[1] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE;
dst_buf2[2] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2;
}
dec_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
dst_width1, dst_height1, dst_stride1);
dec_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
dst_width2, dst_height2, dst_stride2);
av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, &cm->cur_frame->buf,
mi_row, mi_col, 0, num_planes);
av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, dst_stride1,
dst_buf2, dst_stride2);
}
static void cfl_store_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd) {
MB_MODE_INFO *mbmi = xd->mi[0];
if (store_cfl_required(cm, xd)) {
cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
}
}
static void predict_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
int mi_row, int mi_col, BLOCK_SIZE bsize) {
MB_MODE_INFO *mbmi = xd->mi[0];
const int num_planes = av1_num_planes(cm);
for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
if (frame < LAST_FRAME) {
assert(is_intrabc_block(mbmi));
assert(frame == INTRA_FRAME);
assert(ref == 0);
} else {
const RefCntBuffer *ref_buf = get_ref_frame_buf(cm, frame);
const struct scale_factors *ref_scale_factors =
get_ref_scale_factors_const(cm, frame);
xd->block_ref_scale_factors[ref] = ref_scale_factors;
av1_setup_pre_planes(xd, ref, &ref_buf->buf, mi_row, mi_col,
ref_scale_factors, num_planes);
}
}
dec_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
if (mbmi->motion_mode == OBMC_CAUSAL) {
dec_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
}
#if CONFIG_MISMATCH_DEBUG
for (int plane = 0; plane < num_planes; ++plane) {
const struct macroblockd_plane *pd = &xd->plane[plane];
int pixel_c, pixel_r;
mi_to_pixel_loc(&pixel_c, &pixel_r, mi_col, mi_row, 0, 0, pd->subsampling_x,
pd->subsampling_y);
if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
pd->subsampling_y))
continue;
mismatch_check_block_pre(pd->dst.buf, pd->dst.stride,
cm->current_frame.order_hint, plane, pixel_c,
pixel_r, pd->width, pd->height,
xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH);
}
#endif
}
static void set_color_index_map_offset(MACROBLOCKD *const xd, int plane,
aom_reader *r) {
(void)r;
Av1ColorMapParam params;
const MB_MODE_INFO *const mbmi = xd->mi[0];
av1_get_block_dimensions(mbmi->sb_type, plane, xd, &params.plane_width,
&params.plane_height, NULL, NULL);
xd->color_index_map_offset[plane] += params.plane_width * params.plane_height;
}
static void decode_token_recon_block(AV1Decoder *const pbi,
ThreadData *const td, int mi_row,
int mi_col, aom_reader *r,
BLOCK_SIZE bsize) {
AV1_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &td->xd;
const int num_planes = av1_num_planes(cm);
MB_MODE_INFO *mbmi = xd->mi[0];
CFL_CTX *const cfl = &xd->cfl;
cfl->is_chroma_reference = is_chroma_reference(
mi_row, mi_col, bsize, cfl->subsampling_x, cfl->subsampling_y);
if (!is_inter_block(mbmi)) {
int row, col;
assert(bsize == get_plane_block_size(bsize, xd->plane[0].subsampling_x,
xd->plane[0].subsampling_y));
const int max_blocks_wide = max_block_wide(xd, bsize, 0);
const int max_blocks_high = max_block_high(xd, bsize, 0);
const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
int mu_blocks_wide =
block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
int mu_blocks_high =
block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
for (int plane = 0; plane < num_planes; ++plane) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
pd->subsampling_y))
continue;
const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
const int stepr = tx_size_high_unit[tx_size];
const int stepc = tx_size_wide_unit[tx_size];
const int unit_height = ROUND_POWER_OF_TWO(
AOMMIN(mu_blocks_high + row, max_blocks_high), pd->subsampling_y);
const int unit_width = ROUND_POWER_OF_TWO(
AOMMIN(mu_blocks_wide + col, max_blocks_wide), pd->subsampling_x);
for (int blk_row = row >> pd->subsampling_y; blk_row < unit_height;
blk_row += stepr) {
for (int blk_col = col >> pd->subsampling_x; blk_col < unit_width;
blk_col += stepc) {
td->read_coeffs_tx_intra_block_visit(cm, xd, r, plane, blk_row,
blk_col, tx_size);
td->predict_and_recon_intra_block_visit(cm, xd, r, plane, blk_row,
blk_col, tx_size);
set_cb_buffer_offsets(xd, tx_size, plane);
}
}
}
}
}
} else {
td->predict_inter_block_visit(cm, xd, mi_row, mi_col, bsize);
// Reconstruction
if (!mbmi->skip) {
int eobtotal = 0;
const int max_blocks_wide = max_block_wide(xd, bsize, 0);
const int max_blocks_high = max_block_high(xd, bsize, 0);
int row, col;
const BLOCK_SIZE max_unit_bsize = BLOCK_64X64;
assert(max_unit_bsize ==
get_plane_block_size(BLOCK_64X64, xd->plane[0].subsampling_x,
xd->plane[0].subsampling_y));
int mu_blocks_wide =
block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
int mu_blocks_high =
block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);
for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
for (int plane = 0; plane < num_planes; ++plane) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
pd->subsampling_y))
continue;
const BLOCK_SIZE bsizec =
scale_chroma_bsize(bsize, pd->subsampling_x, pd->subsampling_y);
const BLOCK_SIZE plane_bsize = get_plane_block_size(
bsizec, pd->subsampling_x, pd->subsampling_y);
const TX_SIZE max_tx_size =
get_vartx_max_txsize(xd, plane_bsize, plane);
const int bh_var_tx = tx_size_high_unit[max_tx_size];
const int bw_var_tx = tx_size_wide_unit[max_tx_size];
int block = 0;
int step =
tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];
int blk_row, blk_col;
const int unit_height = ROUND_POWER_OF_TWO(
AOMMIN(mu_blocks_high + row, max_blocks_high),
pd->subsampling_y);
const int unit_width = ROUND_POWER_OF_TWO(
AOMMIN(mu_blocks_wide + col, max_blocks_wide),
pd->subsampling_x);
for (blk_row = row >> pd->subsampling_y; blk_row < unit_height;
blk_row += bh_var_tx) {
for (blk_col = col >> pd->subsampling_x; blk_col < unit_width;
blk_col += bw_var_tx) {
decode_reconstruct_tx(cm, td, r, mbmi, plane, plane_bsize,
blk_row, blk_col, block, max_tx_size,
&eobtotal);
block += step;
}
}
}
}
}
}
td->cfl_store_inter_block_visit(cm, xd);
}
av1_visit_palette(pbi, xd, mi_row, mi_col, r, bsize,
set_color_index_map_offset);
}
#if LOOP_FILTER_BITMASK
static void store_bitmask_vartx(AV1_COMMON *cm, int mi_row, int mi_col,
BLOCK_SIZE bsize, TX_SIZE tx_size,
MB_MODE_INFO *mbmi);
#endif
static void set_inter_tx_size(MB_MODE_INFO *mbmi, int stride_log2,
int tx_w_log2, int tx_h_log2, int min_txs,
int split_size, int txs, int blk_row,
int blk_col) {
for (int idy = 0; idy < tx_size_high_unit[split_size];
idy += tx_size_high_unit[min_txs]) {
for (int idx = 0; idx < tx_size_wide_unit[split_size];
idx += tx_size_wide_unit[min_txs]) {
const int index = (((blk_row + idy) >> tx_h_log2) << stride_log2) +
((blk_col + idx) >> tx_w_log2);
mbmi->inter_tx_size[index] = txs;
}
}
}
static void read_tx_size_vartx(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
TX_SIZE tx_size, int depth,
#if LOOP_FILTER_BITMASK
AV1_COMMON *cm, int mi_row, int mi_col,
int store_bitmask,
#endif
int blk_row, int blk_col, aom_reader *r) {
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
int is_split = 0;
const BLOCK_SIZE bsize = mbmi->sb_type;
const int max_blocks_high = max_block_high(xd, bsize, 0);
const int max_blocks_wide = max_block_wide(xd, bsize, 0);
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
assert(tx_size > TX_4X4);
TX_SIZE txs = max_txsize_rect_lookup[bsize];
for (int level = 0; level < MAX_VARTX_DEPTH - 1; ++level)
txs = sub_tx_size_map[txs];
const int tx_w_log2 = tx_size_wide_log2[txs] - MI_SIZE_LOG2;
const int tx_h_log2 = tx_size_high_log2[txs] - MI_SIZE_LOG2;
const int bw_log2 = mi_size_wide_log2[bsize];
const int stride_log2 = bw_log2 - tx_w_log2;
if (depth == MAX_VARTX_DEPTH) {
set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
tx_size, blk_row, blk_col);
mbmi->tx_size = tx_size;
txfm_partition_update(xd->above_txfm_context + blk_col,
xd->left_txfm_context + blk_row, tx_size, tx_size);
return;
}
const int ctx = txfm_partition_context(xd->above_txfm_context + blk_col,
xd->left_txfm_context + blk_row,
mbmi->sb_type, tx_size);
is_split = aom_read_symbol(r, ec_ctx->txfm_partition_cdf[ctx], 2, ACCT_STR);
if (is_split) {
const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
const int bsw = tx_size_wide_unit[sub_txs];
const int bsh = tx_size_high_unit[sub_txs];
if (sub_txs == TX_4X4) {
set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
sub_txs, blk_row, blk_col);
mbmi->tx_size = sub_txs;
txfm_partition_update(xd->above_txfm_context + blk_col,
xd->left_txfm_context + blk_row, sub_txs, tx_size);
#if LOOP_FILTER_BITMASK
if (store_bitmask) {
store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
txsize_to_bsize[tx_size], TX_4X4, mbmi);
}
#endif
return;
}
#if LOOP_FILTER_BITMASK
if (depth + 1 == MAX_VARTX_DEPTH && store_bitmask) {
store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
txsize_to_bsize[tx_size], sub_txs, mbmi);
store_bitmask = 0;
}
#endif
assert(bsw > 0 && bsh > 0);
for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
int offsetr = blk_row + row;
int offsetc = blk_col + col;
read_tx_size_vartx(xd, mbmi, sub_txs, depth + 1,
#if LOOP_FILTER_BITMASK
cm, mi_row, mi_col, store_bitmask,
#endif
offsetr, offsetc, r);
}
}
} else {
set_inter_tx_size(mbmi, stride_log2, tx_w_log2, tx_h_log2, txs, tx_size,
tx_size, blk_row, blk_col);
mbmi->tx_size = tx_size;
txfm_partition_update(xd->above_txfm_context + blk_col,
xd->left_txfm_context + blk_row, tx_size, tx_size);
#if LOOP_FILTER_BITMASK
if (store_bitmask) {
store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col,
txsize_to_bsize[tx_size], tx_size, mbmi);
}
#endif
}
}
static TX_SIZE read_selected_tx_size(MACROBLOCKD *xd, aom_reader *r) {
// TODO(debargha): Clean up the logic here. This function should only
// be called for intra.
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
const int32_t tx_size_cat = bsize_to_tx_size_cat(bsize);
const int max_depths = bsize_to_max_depth(bsize);
const int ctx = get_tx_size_context(xd);
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
const int depth = aom_read_symbol(r, ec_ctx->tx_size_cdf[tx_size_cat][ctx],
max_depths + 1, ACCT_STR);
assert(depth >= 0 && depth <= max_depths);
const TX_SIZE tx_size = depth_to_tx_size(depth, bsize);
return tx_size;
}
static TX_SIZE read_tx_size(AV1_COMMON *cm, MACROBLOCKD *xd, int is_inter,
int allow_select_inter, aom_reader *r) {
const TX_MODE tx_mode = cm->tx_mode;
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
if (xd->lossless[xd->mi[0]->segment_id]) return TX_4X4;
if (block_signals_txsize(bsize)) {
if ((!is_inter || allow_select_inter) && tx_mode == TX_MODE_SELECT) {
const TX_SIZE coded_tx_size = read_selected_tx_size(xd, r);
return coded_tx_size;
} else {
return tx_size_from_tx_mode(bsize, tx_mode);
}
} else {
assert(IMPLIES(tx_mode == ONLY_4X4, bsize == BLOCK_4X4));
return max_txsize_rect_lookup[bsize];
}
}
#if LOOP_FILTER_BITMASK
static void store_bitmask_vartx(AV1_COMMON *cm, int mi_row, int mi_col,
BLOCK_SIZE bsize, TX_SIZE tx_size,
MB_MODE_INFO *mbmi) {
LoopFilterMask *lfm = get_loop_filter_mask(cm, mi_row, mi_col);
const TX_SIZE tx_size_y_vert = txsize_vert_map[tx_size];
const TX_SIZE tx_size_y_horz = txsize_horz_map[tx_size];
const TX_SIZE tx_size_uv_vert = txsize_vert_map[av1_get_max_uv_txsize(
mbmi->sb_type, cm->seq_params.subsampling_x,
cm->seq_params.subsampling_y)];
const TX_SIZE tx_size_uv_horz = txsize_horz_map[av1_get_max_uv_txsize(
mbmi->sb_type, cm->seq_params.subsampling_x,
cm->seq_params.subsampling_y)];
const int is_square_transform_size = tx_size <= TX_64X64;
int mask_id = 0;
int offset = 0;
const int half_ratio_tx_size_max32 =
(tx_size > TX_64X64) & (tx_size <= TX_32X16);
if (is_square_transform_size) {
switch (tx_size) {
case TX_4X4: mask_id = mask_id_table_tx_4x4[bsize]; break;
case TX_8X8:
mask_id = mask_id_table_tx_8x8[bsize];
offset = 19;
break;
case TX_16X16:
mask_id = mask_id_table_tx_16x16[bsize];
offset = 33;
break;
case TX_32X32:
mask_id = mask_id_table_tx_32x32[bsize];
offset = 42;
break;
case TX_64X64: mask_id = 46; break;
default: assert(!is_square_transform_size); return;
}
mask_id += offset;
} else if (half_ratio_tx_size_max32) {
int tx_size_equal_block_size = bsize == txsize_to_bsize[tx_size];
mask_id = 47 + 2 * (tx_size - TX_4X8) + (tx_size_equal_block_size ? 0 : 1);
} else if (tx_size == TX_32X64) {
mask_id = 59;
} else if (tx_size == TX_64X32) {
mask_id = 60;
} else { // quarter ratio tx size
mask_id = 61 + (tx_size - TX_4X16);
}
int index = 0;
const int row = mi_row % MI_SIZE_64X64;
const int col = mi_col % MI_SIZE_64X64;
const int shift = get_index_shift(col, row, &index);
const int vert_shift = tx_size_y_vert <= TX_8X8 ? shift : col;
for (int i = 0; i + index < 4; ++i) {
// y vertical.
lfm->tx_size_ver[0][tx_size_y_horz].bits[i + index] |=
(left_mask_univariant_reordered[mask_id].bits[i] << vert_shift);
// y horizontal.
lfm->tx_size_hor[0][tx_size_y_vert].bits[i + index] |=
(above_mask_univariant_reordered[mask_id].bits[i] << shift);
// u/v vertical.
lfm->tx_size_ver[1][tx_size_uv_horz].bits[i + index] |=
(left_mask_univariant_reordered[mask_id].bits[i] << vert_shift);
// u/v horizontal.
lfm->tx_size_hor[1][tx_size_uv_vert].bits[i + index] |=
(above_mask_univariant_reordered[mask_id].bits[i] << shift);
}
}
static void store_bitmask_univariant_tx(AV1_COMMON *cm, int mi_row, int mi_col,
BLOCK_SIZE bsize, MB_MODE_INFO *mbmi) {
// Use a lookup table that provides one bitmask for a given block size and
// a univariant transform size.
int index;
int shift;
int row;
int col;
LoopFilterMask *lfm = get_loop_filter_mask(cm, mi_row, mi_col);
const TX_SIZE tx_size_y_vert = txsize_vert_map[mbmi->tx_size];
const TX_SIZE tx_size_y_horz = txsize_horz_map[mbmi->tx_size];
const TX_SIZE tx_size_uv_vert = txsize_vert_map[av1_get_max_uv_txsize(
mbmi->sb_type, cm->seq_params.subsampling_x,
cm->seq_params.subsampling_y)];
const TX_SIZE tx_size_uv_horz = txsize_horz_map[av1_get_max_uv_txsize(
mbmi->sb_type, cm->seq_params.subsampling_x,
cm->seq_params.subsampling_y)];
const int is_square_transform_size = mbmi->tx_size <= TX_64X64;
int mask_id = 0;
int offset = 0;
const int half_ratio_tx_size_max32 =
(mbmi->tx_size > TX_64X64) & (mbmi->tx_size <= TX_32X16);
if (is_square_transform_size) {
switch (mbmi->tx_size) {
case TX_4X4: mask_id = mask_id_table_tx_4x4[bsize]; break;
case TX_8X8:
mask_id = mask_id_table_tx_8x8[bsize];
offset = 19;
break;
case TX_16X16:
mask_id = mask_id_table_tx_16x16[bsize];
offset = 33;
break;
case TX_32X32:
mask_id = mask_id_table_tx_32x32[bsize];
offset = 42;
break;
case TX_64X64: mask_id = 46; break;
default: assert(!is_square_transform_size); return;
}
mask_id += offset;
} else if (half_ratio_tx_size_max32) {
int tx_size_equal_block_size = bsize == txsize_to_bsize[mbmi->tx_size];
mask_id =
47 + 2 * (mbmi->tx_size - TX_4X8) + (tx_size_equal_block_size ? 0 : 1);
} else if (mbmi->tx_size == TX_32X64) {
mask_id = 59;
} else if (mbmi->tx_size == TX_64X32) {
mask_id = 60;
} else { // quarter ratio tx size
mask_id = 61 + (mbmi->tx_size - TX_4X16);
}
row = mi_row % MI_SIZE_64X64;
col = mi_col % MI_SIZE_64X64;
shift = get_index_shift(col, row, &index);
const int vert_shift = tx_size_y_vert <= TX_8X8 ? shift : col;
for (int i = 0; i + index < 4; ++i) {
// y vertical.
lfm->tx_size_ver[0][tx_size_y_horz].bits[i + index] |=
(left_mask_univariant_reordered[mask_id].bits[i] << vert_shift);
// y horizontal.
lfm->tx_size_hor[0][tx_size_y_vert].bits[i + index] |=
(above_mask_univariant_reordered[mask_id].bits[i] << shift);
// u/v vertical.
lfm->tx_size_ver[1][tx_size_uv_horz].bits[i + index] |=
(left_mask_univariant_reordered[mask_id].bits[i] << vert_shift);
// u/v horizontal.
lfm->tx_size_hor[1][tx_size_uv_vert].bits[i + index] |=
(above_mask_univariant_reordered[mask_id].bits[i] << shift);
}
}
static void store_bitmask_other_info(AV1_COMMON *cm, int mi_row, int mi_col,
BLOCK_SIZE bsize, MB_MODE_INFO *mbmi,
int is_horz_coding_block_border,
int is_vert_coding_block_border) {
int index;
int shift;
int row;
LoopFilterMask *lfm = get_loop_filter_mask(cm, mi_row, mi_col);
const int row_start = mi_row % MI_SIZE_64X64;
const int col_start = mi_col % MI_SIZE_64X64;
shift = get_index_shift(col_start, row_start, &index);
if (is_horz_coding_block_border) {
const int block_shift = shift + mi_size_wide[bsize];
assert(block_shift <= 64);
const uint64_t right_edge_shift =
(block_shift == 64) ? 0xffffffffffffffff : ((uint64_t)1 << block_shift);
const uint64_t left_edge_shift = (block_shift == 64)
? (((uint64_t)1 << shift) - 1)
: ((uint64_t)1 << shift);
assert(right_edge_shift > left_edge_shift);
const uint64_t top_edge_mask = right_edge_shift - left_edge_shift;
lfm->is_horz_border.bits[index] |= top_edge_mask;
}
if (is_vert_coding_block_border) {
const int is_vert_border = mask_id_table_vert_border[bsize];
const int vert_shift = block_size_high[bsize] <= 8 ? shift : col_start;
for (int i = 0; i + index < 4; ++i) {
lfm->is_vert_border.bits[i + index] |=
(left_mask_univariant_reordered[is_vert_border].bits[i]
<< vert_shift);
}
}
const int is_skip = mbmi->skip && is_inter_block(mbmi);
if (is_skip) {
const int is_skip_mask = mask_id_table_tx_4x4[bsize];
for (int i = 0; i + index < 4; ++i) {
lfm->skip.bits[i + index] |=
(above_mask_univariant_reordered[is_skip_mask].bits[i] << shift);
}
}
const uint8_t level_vert_y = get_filter_level(cm, &cm->lf_info, 0, 0, mbmi);
const uint8_t level_horz_y = get_filter_level(cm, &cm->lf_info, 1, 0, mbmi);
const uint8_t level_u = get_filter_level(cm, &cm->lf_info, 0, 1, mbmi);
const uint8_t level_v = get_filter_level(cm, &cm->lf_info, 0, 2, mbmi);
for (int r = mi_row; r < mi_row + mi_size_high[bsize]; r++) {
index = 0;
row = r % MI_SIZE_64X64;
memset(&lfm->lfl_y_ver[row][col_start], level_vert_y,
sizeof(uint8_t) * mi_size_wide[bsize]);
memset(&lfm->lfl_y_hor[row][col_start], level_horz_y,
sizeof(uint8_t) * mi_size_wide[bsize]);
memset(&lfm->lfl_u_ver[row][col_start], level_u,
sizeof(uint8_t) * mi_size_wide[bsize]);
memset(&lfm->lfl_u_hor[row][col_start], level_u,
sizeof(uint8_t) * mi_size_wide[bsize]);
memset(&lfm->lfl_v_ver[row][col_start], level_v,
sizeof(uint8_t) * mi_size_wide[bsize]);
memset(&lfm->lfl_v_hor[row][col_start], level_v,
sizeof(uint8_t) * mi_size_wide[bsize]);
}
}
#endif
static void parse_decode_block(AV1Decoder *const pbi, ThreadData *const td,
int mi_row, int mi_col, aom_reader *r,
PARTITION_TYPE partition, BLOCK_SIZE bsize) {
MACROBLOCKD *const xd = &td->xd;
decode_mbmi_block(pbi, xd, mi_row, mi_col, r, partition, bsize);
av1_visit_palette(pbi, xd, mi_row, mi_col, r, bsize,
av1_decode_palette_tokens);
AV1_COMMON *cm = &pbi->common;
const int num_planes = av1_num_planes(cm);
MB_MODE_INFO *mbmi = xd->mi[0];
int inter_block_tx = is_inter_block(mbmi) || is_intrabc_block(mbmi);
if (cm->tx_mode == TX_MODE_SELECT && block_signals_txsize(bsize) &&
!mbmi->skip && inter_block_tx && !xd->lossless[mbmi->segment_id]) {
const TX_SIZE max_tx_size = max_txsize_rect_lookup[bsize];
const int bh = tx_size_high_unit[max_tx_size];
const int bw = tx_size_wide_unit[max_tx_size];
const int width = block_size_wide[bsize] >> tx_size_wide_log2[0];
const int height = block_size_high[bsize] >> tx_size_high_log2[0];
for (int idy = 0; idy < height; idy += bh)
for (int idx = 0; idx < width; idx += bw)
read_tx_size_vartx(xd, mbmi, max_tx_size, 0,
#if LOOP_FILTER_BITMASK
cm, mi_row, mi_col, 1,
#endif
idy, idx, r);
} else {
mbmi->tx_size = read_tx_size(cm, xd, inter_block_tx, !mbmi->skip, r);
if (inter_block_tx)
memset(mbmi->inter_tx_size, mbmi->tx_size, sizeof(mbmi->inter_tx_size));
set_txfm_ctxs(mbmi->tx_size, xd->n4_w, xd->n4_h,
mbmi->skip && is_inter_block(mbmi), xd);
#if LOOP_FILTER_BITMASK
const int w = mi_size_wide[bsize];
const int h = mi_size_high[bsize];
if (w <= mi_size_wide[BLOCK_64X64] && h <= mi_size_high[BLOCK_64X64]) {
store_bitmask_univariant_tx(cm, mi_row, mi_col, bsize, mbmi);
} else {
for (int row = 0; row < h; row += mi_size_high[BLOCK_64X64]) {
for (int col = 0; col < w; col += mi_size_wide[BLOCK_64X64]) {
store_bitmask_univariant_tx(cm, mi_row + row, mi_col + col,
BLOCK_64X64, mbmi);
}
}
}
#endif
}
#if LOOP_FILTER_BITMASK
const int w = mi_size_wide[bsize];
const int h = mi_size_high[bsize];
if (w <= mi_size_wide[BLOCK_64X64] && h <= mi_size_high[BLOCK_64X64]) {
store_bitmask_other_info(cm, mi_row, mi_col, bsize, mbmi, 1, 1);
} else {
for (int row = 0; row < h; row += mi_size_high[BLOCK_64X64]) {
for (int col = 0; col < w; col += mi_size_wide[BLOCK_64X64]) {
store_bitmask_other_info(cm, mi_row + row, mi_col + col, BLOCK_64X64,
mbmi, row == 0, col == 0);
}
}
}
#endif
if (cm->delta_q_info.delta_q_present_flag) {
for (int i = 0; i < MAX_SEGMENTS; i++) {
const int current_qindex =
av1_get_qindex(&cm->seg, i, xd->current_qindex);
for (int j = 0; j < num_planes; ++j) {
const int dc_delta_q =
j == 0 ? cm->y_dc_delta_q
: (j == 1 ? cm->u_dc_delta_q : cm->v_dc_delta_q);
const int ac_delta_q =
j == 0 ? 0 : (j == 1 ? cm->u_ac_delta_q : cm->v_ac_delta_q);
xd->plane[j].seg_dequant_QTX[i][0] = av1_dc_quant_QTX(
current_qindex, dc_delta_q, cm->seq_params.bit_depth);
xd->plane[j].seg_dequant_QTX[i][1] = av1_ac_quant_QTX(
current_qindex, ac_delta_q, cm->seq_params.bit_depth);
}
}
}
if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize, num_planes);
decode_token_recon_block(pbi, td, mi_row, mi_col, r, bsize);
}
static void set_offsets_for_pred_and_recon(AV1Decoder *const pbi,
ThreadData *const td, int mi_row,
int mi_col, BLOCK_SIZE bsize) {
AV1_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &td->xd;
const int bw = mi_size_wide[bsize];
const int bh = mi_size_high[bsize];
const int num_planes = av1_num_planes(cm);
const int offset = mi_row * cm->mi_stride + mi_col;
const TileInfo *const tile = &xd->tile;
xd->mi = cm->mi_grid_visible + offset;
xd->cfl.mi_row = mi_row;
xd->cfl.mi_col = mi_col;
set_plane_n4(xd, bw, bh, num_planes);
// Distance of Mb to the various image edges. These are specified to 8th pel
// as they are always compared to values that are in 1/8th pel units
set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row, mi_col, 0,
num_planes);
}
static void decode_block(AV1Decoder *const pbi, ThreadData *const td,
int mi_row, int mi_col, aom_reader *r,
PARTITION_TYPE partition, BLOCK_SIZE bsize) {
(void)partition;
set_offsets_for_pred_and_recon(pbi, td, mi_row, mi_col, bsize);
decode_token_recon_block(pbi, td, mi_row, mi_col, r, bsize);
}
static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col,
aom_reader *r, int has_rows, int has_cols,
BLOCK_SIZE bsize) {
const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
if (!has_rows && !has_cols) return PARTITION_SPLIT;
assert(ctx >= 0);
aom_cdf_prob *partition_cdf = ec_ctx->partition_cdf[ctx];
if (has_rows && has_cols) {
return (PARTITION_TYPE)aom_read_symbol(
r, partition_cdf, partition_cdf_length(bsize), ACCT_STR);
} else if (!has_rows && has_cols) {
assert(bsize > BLOCK_8X8);
aom_cdf_prob cdf[2];
partition_gather_vert_alike(cdf, partition_cdf, bsize);
assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ;
} else {
assert(has_rows && !has_cols);
assert(bsize > BLOCK_8X8);
aom_cdf_prob cdf[2];
partition_gather_horz_alike(cdf, partition_cdf, bsize);
assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
return aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT;
}
}
// TODO(slavarnway): eliminate bsize and subsize in future commits
static void decode_partition(AV1Decoder *const pbi, ThreadData *const td,
int mi_row, int mi_col, aom_reader *reader,
BLOCK_SIZE bsize, int parse_decode_flag) {
assert(bsize < BLOCK_SIZES_ALL);
AV1_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &td->xd;
const int bw = mi_size_wide[bsize];
const int hbs = bw >> 1;
PARTITION_TYPE partition;
BLOCK_SIZE subsize;
const int quarter_step = bw / 4;
BLOCK_SIZE bsize2 = get_partition_subsize(bsize, PARTITION_SPLIT);
const int has_rows = (mi_row + hbs) < cm->mi_rows;
const int has_cols = (mi_col + hbs) < cm->mi_cols;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
// parse_decode_flag takes the following values :
// 01 - do parse only
// 10 - do decode only
// 11 - do parse and decode
static const block_visitor_fn_t block_visit[4] = { NULL, parse_decode_block,
decode_block,
parse_decode_block };
if (parse_decode_flag & 1) {
const int num_planes = av1_num_planes(cm);
for (int plane = 0; plane < num_planes; ++plane) {
int rcol0, rcol1, rrow0, rrow1;
if (av1_loop_restoration_corners_in_sb(cm, plane, mi_row, mi_col, bsize,
&rcol0, &rcol1, &rrow0, &rrow1)) {
const int rstride = cm->rst_info[plane].horz_units_per_tile;
for (int rrow = rrow0; rrow < rrow1; ++rrow) {
for (int rcol = rcol0; rcol < rcol1; ++rcol) {
const int runit_idx = rcol + rrow * rstride;
loop_restoration_read_sb_coeffs(cm, xd, reader, plane, runit_idx);
}
}
}
}
partition = (bsize < BLOCK_8X8) ? PARTITION_NONE
: read_partition(xd, mi_row, mi_col, reader,
has_rows, has_cols, bsize);
} else {
partition = get_partition(cm, mi_row, mi_col, bsize);
}
subsize = get_partition_subsize(bsize, partition);
if (subsize == BLOCK_INVALID) {
aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
"Partition is invalid for block size %dx%d",
block_size_wide[bsize], block_size_high[bsize]);
}
// Check the bitstream is conformant: if there is subsampling on the
// chroma planes, subsize must subsample to a valid block size.
const struct macroblockd_plane *const pd_u = &xd->plane[1];
if (get_plane_block_size(subsize, pd_u->subsampling_x, pd_u->subsampling_y) ==
BLOCK_INVALID) {
aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
"Block size %dx%d invalid with this subsampling mode",
block_size_wide[subsize], block_size_high[subsize]);
}
#define DEC_BLOCK_STX_ARG
#define DEC_BLOCK_EPT_ARG partition,
#define DEC_BLOCK(db_r, db_c, db_subsize) \
block_visit[parse_decode_flag](pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), \
reader, DEC_BLOCK_EPT_ARG(db_subsize))
#define DEC_PARTITION(db_r, db_c, db_subsize) \
decode_partition(pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), reader, \
(db_subsize), parse_decode_flag)
switch (partition) {
case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break;
case PARTITION_HORZ:
DEC_BLOCK(mi_row, mi_col, subsize);
if (has_rows) DEC_BLOCK(mi_row + hbs, mi_col, subsize);
break;
case PARTITION_VERT:
DEC_BLOCK(mi_row, mi_col, subsize);
if (has_cols) DEC_BLOCK(mi_row, mi_col + hbs, subsize);
break;
case PARTITION_SPLIT:
DEC_PARTITION(mi_row, mi_col, subsize);
DEC_PARTITION(mi_row, mi_col + hbs, subsize);
DEC_PARTITION(mi_row + hbs, mi_col, subsize);
DEC_PARTITION(mi_row + hbs, mi_col + hbs, subsize);
break;
case PARTITION_HORZ_A:
DEC_BLOCK(mi_row, mi_col, bsize2);
DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
DEC_BLOCK(mi_row + hbs, mi_col, subsize);
break;
case PARTITION_HORZ_B:
DEC_BLOCK(mi_row, mi_col, subsize);
DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
break;
case PARTITION_VERT_A:
DEC_BLOCK(mi_row, mi_col, bsize2);
DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
DEC_BLOCK(mi_row, mi_col + hbs, subsize);
break;
case PARTITION_VERT_B:
DEC_BLOCK(mi_row, mi_col, subsize);
DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
break;
case PARTITION_HORZ_4:
for (int i = 0; i < 4; ++i) {
int this_mi_row = mi_row + i * quarter_step;
if (i > 0 && this_mi_row >= cm->mi_rows) break;
DEC_BLOCK(this_mi_row, mi_col, subsize);
}
break;
case PARTITION_VERT_4:
for (int i = 0; i < 4; ++i) {
int this_mi_col = mi_col + i * quarter_step;
if (i > 0 && this_mi_col >= cm->mi_cols) break;
DEC_BLOCK(mi_row, this_mi_col, subsize);
}
break;
default: assert(0 && "Invalid partition type");
}
#undef DEC_PARTITION
#undef DEC_BLOCK
#undef DEC_BLOCK_EPT_ARG
#undef DEC_BLOCK_STX_ARG
if (parse_decode_flag & 1)
update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition);
}
static void setup_bool_decoder(const uint8_t *data, const uint8_t *data_end,
const size_t read_size,
struct aom_internal_error_info *error_info,
aom_reader *r, uint8_t allow_update_cdf) {
// Validate the calculated partition length. If the buffer
// described by the partition can't be fully read, then restrict
// it to the portion that can be (for EC mode) or throw an error.
if (!read_is_valid(data, read_size, data_end))
aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME,
"Truncated packet or corrupt tile length");
if (aom_reader_init(r, data, read_size))
aom_internal_error(error_info, AOM_CODEC_MEM_ERROR,
"Failed to allocate bool decoder %d", 1);
r->allow_update_cdf = allow_update_cdf;
}
static void setup_segmentation(AV1_COMMON *const cm,
struct aom_read_bit_buffer *rb) {
struct segmentation *const seg = &cm->seg;
seg->update_map = 0;
seg->update_data = 0;
seg->temporal_update = 0;
seg->enabled = aom_rb_read_bit(rb);
if (!seg->enabled) {
if (cm->cur_frame->seg_map)
memset(cm->cur_frame->seg_map, 0, (cm->mi_rows * cm->mi_cols));
memset(seg, 0, sizeof(*seg));
segfeatures_copy(&cm->cur_frame->seg, seg);
return;
}
if (cm->seg.enabled && cm->prev_frame &&
(cm->mi_rows == cm->prev_frame->mi_rows) &&
(cm->mi_cols == cm->prev_frame->mi_cols)) {
cm->last_frame_seg_map = cm->prev_frame->seg_map;
} else {
cm->last_frame_seg_map = NULL;
}
// Read update flags
if (cm->primary_ref_frame == PRIMARY_REF_NONE) {
// These frames can't use previous frames, so must signal map + features
seg->update_map = 1;
seg->temporal_update = 0;
seg->update_data = 1;
} else {
seg->update_map = aom_rb_read_bit(rb);
if (seg->update_map) {
seg->temporal_update = aom_rb_read_bit(rb);
} else {
seg->temporal_update = 0;
}
seg->update_data = aom_rb_read_bit(rb);
}
// Segmentation data update
if (seg->update_data) {
av1_clearall_segfeatures(seg);
for (int i = 0; i < MAX_SEGMENTS; i++) {
for (int j = 0; j < SEG_LVL_MAX; j++) {
int data = 0;
const int feature_enabled = aom_rb_read_bit(rb);
if (feature_enabled) {
av1_enable_segfeature(seg, i, j);
const int data_max = av1_seg_feature_data_max(j);
const int data_min = -data_max;
const int ubits = get_unsigned_bits(data_max);
if (av1_is_segfeature_signed(j)) {
data = aom_rb_read_inv_signed_literal(rb, ubits);
} else {
data = aom_rb_read_literal(rb, ubits);
}
data = clamp(data, data_min, data_max);
}
av1_set_segdata(seg, i, j, data);
}
}
calculate_segdata(seg);
} else if (cm->prev_frame) {
segfeatures_copy(seg, &cm->prev_frame->seg);
}
segfeatures_copy(&cm->cur_frame->seg, seg);
}
static void decode_restoration_mode(AV1_COMMON *cm,
struct aom_read_bit_buffer *rb) {
assert(!cm->all_lossless);
const int num_planes = av1_num_planes(cm);
if (cm->allow_intrabc) return;
int all_none = 1, chroma_none = 1;
for (int p = 0; p < num_planes; ++p) {
RestorationInfo *rsi = &cm->rst_info[p];
if (aom_rb_read_bit(rb)) {
rsi->frame_restoration_type =
aom_rb_read_bit(rb) ? RESTORE_SGRPROJ : RESTORE_WIENER;
} else {
rsi->frame_restoration_type =
aom_rb_read_bit(rb) ? RESTORE_SWITCHABLE : RESTORE_NONE;
}
if (rsi->frame_restoration_type != RESTORE_NONE) {
all_none = 0;
chroma_none &= p == 0;
}
}
if (!all_none) {
assert(cm->seq_params.sb_size == BLOCK_64X64 ||
cm->seq_params.sb_size == BLOCK_128X128);
const int sb_size = cm->seq_params.sb_size == BLOCK_128X128 ? 128 : 64;
for (int p = 0; p < num_planes; ++p)
cm->rst_info[p].restoration_unit_size = sb_size;
RestorationInfo *rsi = &cm->rst_info[0];
if (sb_size == 64) {
rsi->restoration_unit_size <<= aom_rb_read_bit(rb);
}
if (rsi->restoration_unit_size > 64) {
rsi->restoration_unit_size <<= aom_rb_read_bit(rb);
}
} else {
const int size = RESTORATION_UNITSIZE_MAX;
for (int p = 0; p < num_planes; ++p)
cm->rst_info[p].restoration_unit_size = size;
}
if (num_planes > 1) {
int s = AOMMIN(cm->seq_params.subsampling_x, cm->seq_params.subsampling_y);
if (s && !chroma_none) {
cm->rst_info[1].restoration_unit_size =
cm->rst_info[0].restoration_unit_size >> (aom_rb_read_bit(rb) * s);
} else {
cm->rst_info[1].restoration_unit_size =
cm->rst_info[0].restoration_unit_size;
}
cm->rst_info[2].restoration_unit_size =
cm->rst_info[1].restoration_unit_size;
}
}
static void read_wiener_filter(int wiener_win, WienerInfo *wiener_info,
WienerInfo *ref_wiener_info, aom_reader *rb) {
memset(wiener_info->vfilter, 0, sizeof(wiener_info->vfilter));
memset(wiener_info->hfilter, 0, sizeof(wiener_info->hfilter));
if (wiener_win == WIENER_WIN)
wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
WIENER_FILT_TAP0_SUBEXP_K,
ref_wiener_info->vfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) +
WIENER_FILT_TAP0_MINV;
else
wiener_info->vfilter[0] = wiener_info->vfilter[WIENER_WIN - 1] = 0;
wiener_info->vfilter[1] = wiener_info->vfilter[WIENER_WIN - 2] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
WIENER_FILT_TAP1_SUBEXP_K,
ref_wiener_info->vfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) +
WIENER_FILT_TAP1_MINV;
wiener_info->vfilter[2] = wiener_info->vfilter[WIENER_WIN - 3] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
WIENER_FILT_TAP2_SUBEXP_K,
ref_wiener_info->vfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) +
WIENER_FILT_TAP2_MINV;
// The central element has an implicit +WIENER_FILT_STEP
wiener_info->vfilter[WIENER_HALFWIN] =
-2 * (wiener_info->vfilter[0] + wiener_info->vfilter[1] +
wiener_info->vfilter[2]);
if (wiener_win == WIENER_WIN)
wiener_info->hfilter[0] = wiener_info->hfilter[WIENER_WIN - 1] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP0_MAXV - WIENER_FILT_TAP0_MINV + 1,
WIENER_FILT_TAP0_SUBEXP_K,
ref_wiener_info->hfilter[0] - WIENER_FILT_TAP0_MINV, ACCT_STR) +
WIENER_FILT_TAP0_MINV;
else
wiener_info->hfilter[0] = wiener_info->hfilter[WIENER_WIN - 1] = 0;
wiener_info->hfilter[1] = wiener_info->hfilter[WIENER_WIN - 2] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP1_MAXV - WIENER_FILT_TAP1_MINV + 1,
WIENER_FILT_TAP1_SUBEXP_K,
ref_wiener_info->hfilter[1] - WIENER_FILT_TAP1_MINV, ACCT_STR) +
WIENER_FILT_TAP1_MINV;
wiener_info->hfilter[2] = wiener_info->hfilter[WIENER_WIN - 3] =
aom_read_primitive_refsubexpfin(
rb, WIENER_FILT_TAP2_MAXV - WIENER_FILT_TAP2_MINV + 1,
WIENER_FILT_TAP2_SUBEXP_K,
ref_wiener_info->hfilter[2] - WIENER_FILT_TAP2_MINV, ACCT_STR) +
WIENER_FILT_TAP2_MINV;
// The central element has an implicit +WIENER_FILT_STEP
wiener_info->hfilter[WIENER_HALFWIN] =
-2 * (wiener_info->hfilter[0] + wiener_info->hfilter[1] +
wiener_info->hfilter[2]);
memcpy(ref_wiener_info, wiener_info, sizeof(*wiener_info));
}
static void read_sgrproj_filter(SgrprojInfo *sgrproj_info,
SgrprojInfo *ref_sgrproj_info, aom_reader *rb) {
sgrproj_info->ep = aom_read_literal(rb, SGRPROJ_PARAMS_BITS, ACCT_STR);
const sgr_params_type *params = &sgr_params[sgrproj_info->ep];
if (params->r[0] == 0) {
sgrproj_info->xqd[0] = 0;
sgrproj_info->xqd[1] =
aom_read_primitive_refsubexpfin(
rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
SGRPROJ_PRJ_MIN1;
} else if (params->r[1] == 0) {
sgrproj_info->xqd[0] =
aom_read_primitive_refsubexpfin(
rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
SGRPROJ_PRJ_MIN0;
sgrproj_info->xqd[1] = clamp((1 << SGRPROJ_PRJ_BITS) - sgrproj_info->xqd[0],
SGRPROJ_PRJ_MIN1, SGRPROJ_PRJ_MAX1);
} else {
sgrproj_info->xqd[0] =
aom_read_primitive_refsubexpfin(
rb, SGRPROJ_PRJ_MAX0 - SGRPROJ_PRJ_MIN0 + 1, SGRPROJ_PRJ_SUBEXP_K,
ref_sgrproj_info->xqd[0] - SGRPROJ_PRJ_MIN0, ACCT_STR) +
SGRPROJ_PRJ_MIN0;
sgrproj_info->xqd[1] =
aom_read_primitive_refsubexpfin(
rb, SGRPROJ_PRJ_MAX1 - SGRPROJ_PRJ_MIN1 + 1, SGRPROJ_PRJ_SUBEXP_K,
ref_sgrproj_info->xqd[1] - SGRPROJ_PRJ_MIN1, ACCT_STR) +
SGRPROJ_PRJ_MIN1;
}
memcpy(ref_sgrproj_info, sgrproj_info, sizeof(*sgrproj_info));
}
static void loop_restoration_read_sb_coeffs(const AV1_COMMON *const cm,
MACROBLOCKD *xd,
aom_reader *const r, int plane,
int runit_idx) {
const RestorationInfo *rsi = &cm->rst_info[plane];
RestorationUnitInfo *rui = &rsi->unit_info[runit_idx];
if (rsi->frame_restoration_type == RESTORE_NONE) return;
assert(!cm->all_lossless);
const int wiener_win = (plane > 0) ? WIENER_WIN_CHROMA : WIENER_WIN;
WienerInfo *wiener_info = xd->wiener_info + plane;
SgrprojInfo *sgrproj_info = xd->sgrproj_info + plane;
if (rsi->frame_restoration_type == RESTORE_SWITCHABLE) {
rui->restoration_type =
aom_read_symbol(r, xd->tile_ctx->switchable_restore_cdf,
RESTORE_SWITCHABLE_TYPES, ACCT_STR);
switch (rui->restoration_type) {
case RESTORE_WIENER:
read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r);
break;
case RESTORE_SGRPROJ:
read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r);
break;
default: assert(rui->restoration_type == RESTORE_NONE); break;
}
} else if (rsi->frame_restoration_type == RESTORE_WIENER) {
if (aom_read_symbol(r, xd->tile_ctx->wiener_restore_cdf, 2, ACCT_STR)) {
rui->restoration_type = RESTORE_WIENER;
read_wiener_filter(wiener_win, &rui->wiener_info, wiener_info, r);
} else {
rui->restoration_type = RESTORE_NONE;
}
} else if (rsi->frame_restoration_type == RESTORE_SGRPROJ) {
if (aom_read_symbol(r, xd->tile_ctx->sgrproj_restore_cdf, 2, ACCT_STR)) {
rui->restoration_type = RESTORE_SGRPROJ;
read_sgrproj_filter(&rui->sgrproj_info, sgrproj_info, r);
} else {
rui->restoration_type = RESTORE_NONE;
}
}
}
static void setup_loopfilter(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
const int num_planes = av1_num_planes(cm);
struct loopfilter *lf = &cm->lf;
if (cm->allow_intrabc || cm->coded_lossless) {
// write default deltas to frame buffer
av1_set_default_ref_deltas(cm->cur_frame->ref_deltas);
av1_set_default_mode_deltas(cm->cur_frame->mode_deltas);
return;
}
assert(!cm->coded_lossless);
if (cm->prev_frame) {
// write deltas to frame buffer
memcpy(lf->ref_deltas, cm->prev_frame->ref_deltas, REF_FRAMES);
memcpy(lf->mode_deltas, cm->prev_frame->mode_deltas, MAX_MODE_LF_DELTAS);
} else {
av1_set_default_ref_deltas(lf->ref_deltas);
av1_set_default_mode_deltas(lf->mode_deltas);
}
lf->filter_level[0] = aom_rb_read_literal(rb, 6);
lf->filter_level[1] = aom_rb_read_literal(rb, 6);
if (num_planes > 1) {
if (lf->filter_level[0] || lf->filter_level[1]) {
lf->filter_level_u = aom_rb_read_literal(rb, 6);
lf->filter_level_v = aom_rb_read_literal(rb, 6);
}
}
lf->sharpness_level = aom_rb_read_literal(rb, 3);
// Read in loop filter deltas applied at the MB level based on mode or ref
// frame.
lf->mode_ref_delta_update = 0;
lf->mode_ref_delta_enabled = aom_rb_read_bit(rb);
if (lf->mode_ref_delta_enabled) {
lf->mode_ref_delta_update = aom_rb_read_bit(rb);
if (lf->mode_ref_delta_update) {
for (int i = 0; i < REF_FRAMES; i++)
if (aom_rb_read_bit(rb))
lf->ref_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
for (int i = 0; i < MAX_MODE_LF_DELTAS; i++)
if (aom_rb_read_bit(rb))
lf->mode_deltas[i] = aom_rb_read_inv_signed_literal(rb, 6);
}
}
// write deltas to frame buffer
memcpy(cm->cur_frame->ref_deltas, lf->ref_deltas, REF_FRAMES);
memcpy(cm->cur_frame->mode_deltas, lf->mode_deltas, MAX_MODE_LF_DELTAS);
}
static void setup_cdef(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
const int num_planes = av1_num_planes(cm);
CdefInfo *const cdef_info = &cm->cdef_info;
if (cm->allow_intrabc) return;
cdef_info->cdef_pri_damping = aom_rb_read_literal(rb, 2) + 3;
cdef_info->cdef_sec_damping = cdef_info->cdef_pri_damping;
cdef_info->cdef_bits = aom_rb_read_literal(rb, 2);
cdef_info->nb_cdef_strengths = 1 << cdef_info->cdef_bits;
for (int i = 0; i < cdef_info->nb_cdef_strengths; i++) {
cdef_info->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS);
cdef_info->cdef_uv_strengths[i] =
num_planes > 1 ? aom_rb_read_literal(rb, CDEF_STRENGTH_BITS) : 0;
}
}
static INLINE int read_delta_q(struct aom_read_bit_buffer *rb) {
return aom_rb_read_bit(rb) ? aom_rb_read_inv_signed_literal(rb, 6) : 0;
}
static void setup_quantization(AV1_COMMON *const cm,
struct aom_read_bit_buffer *rb) {
const SequenceHeader *const seq_params = &cm->seq_params;
const int num_planes = av1_num_planes(cm);
cm->base_qindex = aom_rb_read_literal(rb, QINDEX_BITS);
cm->y_dc_delta_q = read_delta_q(rb);
if (num_planes > 1) {
int diff_uv_delta = 0;
if (seq_params->separate_uv_delta_q) diff_uv_delta = aom_rb_read_bit(rb);
cm->u_dc_delta_q = read_delta_q(rb);
cm->u_ac_delta_q = read_delta_q(rb);
if (diff_uv_delta) {
cm->v_dc_delta_q = read_delta_q(rb);
cm->v_ac_delta_q = read_delta_q(rb);
} else {
cm->v_dc_delta_q = cm->u_dc_delta_q;
cm->v_ac_delta_q = cm->u_ac_delta_q;
}
} else {
cm->u_dc_delta_q = 0;
cm->u_ac_delta_q = 0;
cm->v_dc_delta_q = 0;
cm->v_ac_delta_q = 0;
}
cm->using_qmatrix = aom_rb_read_bit(rb);
if (cm->using_qmatrix) {
cm->qm_y = aom_rb_read_literal(rb, QM_LEVEL_BITS);
cm->qm_u = aom_rb_read_literal(rb, QM_LEVEL_BITS);
if (!seq_params->separate_uv_delta_q)
cm->qm_v = cm->qm_u;
else
cm->qm_v = aom_rb_read_literal(rb, QM_LEVEL_BITS);
} else {
cm->qm_y = 0;
cm->qm_u = 0;
cm->qm_v = 0;
}
}
// Build y/uv dequant values based on segmentation.
static void setup_segmentation_dequant(AV1_COMMON *const cm,
MACROBLOCKD *const xd) {
const int bit_depth = cm->seq_params.bit_depth;
const int using_qm = cm->using_qmatrix;
// When segmentation is disabled, only the first value is used. The
// remaining are don't cares.
const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1;
for (int i = 0; i < max_segments; ++i) {
const int qindex = xd->qindex[i];
cm->y_dequant_QTX[i][0] =
av1_dc_quant_QTX(qindex, cm->y_dc_delta_q, bit_depth);
cm->y_dequant_QTX[i][1] = av1_ac_quant_QTX(qindex, 0, bit_depth);
cm->u_dequant_QTX[i][0] =
av1_dc_quant_QTX(qindex, cm->u_dc_delta_q, bit_depth);
cm->u_dequant_QTX[i][1] =
av1_ac_quant_QTX(qindex, cm->u_ac_delta_q, bit_depth);
cm->v_dequant_QTX[i][0] =
av1_dc_quant_QTX(qindex, cm->v_dc_delta_q, bit_depth);
cm->v_dequant_QTX[i][1] =
av1_ac_quant_QTX(qindex, cm->v_ac_delta_q, bit_depth);
const int lossless = xd->lossless[i];
// NB: depends on base index so there is only 1 set per frame
// No quant weighting when lossless or signalled not using QM
int qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_y;
for (int j = 0; j < TX_SIZES_ALL; ++j) {
cm->y_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_Y, j);
}
qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_u;
for (int j = 0; j < TX_SIZES_ALL; ++j) {
cm->u_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_U, j);
}
qmlevel = (lossless || using_qm == 0) ? NUM_QM_LEVELS - 1 : cm->qm_v;
for (int j = 0; j < TX_SIZES_ALL; ++j) {
cm->v_iqmatrix[i][j] = av1_iqmatrix(cm, qmlevel, AOM_PLANE_V, j);
}
}
}
static InterpFilter read_frame_interp_filter(struct aom_read_bit_buffer *rb) {
return aom_rb_read_bit(rb) ? SWITCHABLE
: aom_rb_read_literal(rb, LOG_SWITCHABLE_FILTERS);
}
static void setup_render_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
cm->render_width = cm->superres_upscaled_width;
cm->render_height = cm->superres_upscaled_height;
if (aom_rb_read_bit(rb))
av1_read_frame_size(rb, 16, 16, &cm->render_width, &cm->render_height);
}
// TODO(afergs): make "struct aom_read_bit_buffer *const rb"?
static void setup_superres(AV1_COMMON *const cm, struct aom_read_bit_buffer *rb,
int *width, int *height) {
cm->superres_upscaled_width = *width;
cm->superres_upscaled_height = *height;
const SequenceHeader *const seq_params = &cm->seq_params;
if (!seq_params->enable_superres) return;
if (aom_rb_read_bit(rb)) {
cm->superres_scale_denominator =
(uint8_t)aom_rb_read_literal(rb, SUPERRES_SCALE_BITS);
cm->superres_scale_denominator += SUPERRES_SCALE_DENOMINATOR_MIN;
// Don't edit cm->width or cm->height directly, or the buffers won't get
// resized correctly
av1_calculate_scaled_superres_size(width, height,
cm->superres_scale_denominator);
} else {
// 1:1 scaling - ie. no scaling, scale not provided
cm->superres_scale_denominator = SCALE_NUMERATOR;
}
}
static void resize_context_buffers(AV1_COMMON *cm, int width, int height) {
#if CONFIG_SIZE_LIMIT
if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT)
aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
"Dimensions of %dx%d beyond allowed size of %dx%d.",
width, height, DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT);
#endif
if (cm->width != width || cm->height != height) {
const int new_mi_rows =
ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
const int new_mi_cols =
ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2) >> MI_SIZE_LOG2;
// Allocations in av1_alloc_context_buffers() depend on individual
// dimensions as well as the overall size.
if (new_mi_cols > cm->mi_cols || new_mi_rows > cm->mi_rows) {
if (av1_alloc_context_buffers(cm, width, height)) {
// The cm->mi_* values have been cleared and any existing context
// buffers have been freed. Clear cm->width and cm->height to be
// consistent and to force a realloc next time.
cm->width = 0;
cm->height = 0;
aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
"Failed to allocate context buffers");
}
} else {
av1_set_mb_mi(cm, width, height);
}
av1_init_context_buffers(cm);
cm->width = width;
cm->height = height;
}
ensure_mv_buffer(cm->cur_frame, cm);
cm->cur_frame->width = cm->width;
cm->cur_frame->height = cm->height;
}
static void setup_buffer_pool(AV1_COMMON *cm) {
BufferPool *const pool = cm->buffer_pool;
const SequenceHeader *const seq_params = &cm->seq_params;
lock_buffer_pool(pool);
if (aom_realloc_frame_buffer(
&cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x,
seq_params->subsampling_y, seq_params->use_highbitdepth,
AOM_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
&cm->cur_frame->raw_frame_buffer, pool->get_fb_cb, pool->cb_priv)) {
unlock_buffer_pool(pool);
aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR,
"Failed to allocate frame buffer");
}
unlock_buffer_pool(pool);
cm->cur_frame->buf.bit_depth = (unsigned int)seq_params->bit_depth;
cm->cur_frame->buf.color_primaries = seq_params->color_primaries;
cm->cur_frame->buf.transfer_characteristics =
seq_params->transfer_characteristics;
cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients;
cm->cur_frame->buf.monochrome = seq_params->monochrome;
cm->cur_frame->buf.chroma_sample_position =
seq_params->chroma_sample_position;
cm->cur_frame->buf.color_range = seq_params->color_range;
cm->cur_frame->buf.render_width = cm->render_width;
cm->cur_frame->buf.render_height = cm->render_height;
}
static void setup_frame_size(AV1_COMMON *cm, int frame_size_override_flag,
struct aom_read_bit_buffer *rb) {
const SequenceHeader *const seq_params = &cm->seq_params;
int width, height;
if (frame_size_override_flag) {
int num_bits_width = seq_params->num_bits_width;
int num_bits_height = seq_params->num_bits_height;
av1_read_frame_size(rb, num_bits_width, num_bits_height, &width, &height);
if (width > seq_params->max_frame_width ||