diff options
Diffstat (limited to 'media/libaom/src/av1/decoder')
-rw-r--r-- | media/libaom/src/av1/decoder/accounting.c | 138 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/accounting.h | 82 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodeframe.c | 5567 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodeframe.h | 85 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodemv.c | 1560 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodemv.h | 35 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decoder.c | 575 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decoder.h | 317 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodetxb.c | 362 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/decodetxb.h | 32 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/detokenize.c | 78 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/detokenize.h | 29 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/dthread.c | 192 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/dthread.h | 82 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/inspection.c | 117 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/inspection.h | 84 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/obu.c | 839 | ||||
-rw-r--r-- | media/libaom/src/av1/decoder/obu.h | 31 |
18 files changed, 10205 insertions, 0 deletions
diff --git a/media/libaom/src/av1/decoder/accounting.c b/media/libaom/src/av1/decoder/accounting.c new file mode 100644 index 000000000..8d8f3dfdb --- /dev/null +++ b/media/libaom/src/av1/decoder/accounting.c @@ -0,0 +1,138 @@ +/* + * 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 <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include "aom/aom_integer.h" +#include "av1/decoder/accounting.h" + +static int aom_accounting_hash(const char *str) { + uint32_t val; + const unsigned char *ustr; + val = 0; + ustr = (const unsigned char *)str; + /* This is about the worst hash one can design, but it should be good enough + here. */ + while (*ustr) val += *ustr++; + return val % AOM_ACCOUNTING_HASH_SIZE; +} + +/* Dictionary lookup based on an open-addressing hash table. */ +int aom_accounting_dictionary_lookup(Accounting *accounting, const char *str) { + int hash; + size_t len; + AccountingDictionary *dictionary; + dictionary = &accounting->syms.dictionary; + hash = aom_accounting_hash(str); + while (accounting->hash_dictionary[hash] != -1) { + if (strcmp(dictionary->strs[accounting->hash_dictionary[hash]], str) == 0) { + return accounting->hash_dictionary[hash]; + } + hash++; + if (hash == AOM_ACCOUNTING_HASH_SIZE) hash = 0; + } + /* No match found. */ + assert(dictionary->num_strs + 1 < MAX_SYMBOL_TYPES); + accounting->hash_dictionary[hash] = dictionary->num_strs; + len = strlen(str); + dictionary->strs[dictionary->num_strs] = malloc(len + 1); + snprintf(dictionary->strs[dictionary->num_strs], len + 1, "%s", str); + dictionary->num_strs++; + return dictionary->num_strs - 1; +} + +void aom_accounting_init(Accounting *accounting) { + int i; + accounting->num_syms_allocated = 1000; + accounting->syms.syms = + malloc(sizeof(AccountingSymbol) * accounting->num_syms_allocated); + accounting->syms.dictionary.num_strs = 0; + assert(AOM_ACCOUNTING_HASH_SIZE > 2 * MAX_SYMBOL_TYPES); + for (i = 0; i < AOM_ACCOUNTING_HASH_SIZE; i++) + accounting->hash_dictionary[i] = -1; + aom_accounting_reset(accounting); +} + +void aom_accounting_reset(Accounting *accounting) { + accounting->syms.num_syms = 0; + accounting->syms.num_binary_syms = 0; + accounting->syms.num_multi_syms = 0; + accounting->context.x = -1; + accounting->context.y = -1; + accounting->last_tell_frac = 0; +} + +void aom_accounting_clear(Accounting *accounting) { + int i; + AccountingDictionary *dictionary; + free(accounting->syms.syms); + dictionary = &accounting->syms.dictionary; + for (i = 0; i < dictionary->num_strs; i++) { + free(dictionary->strs[i]); + } +} + +void aom_accounting_set_context(Accounting *accounting, int16_t x, int16_t y) { + accounting->context.x = x; + accounting->context.y = y; +} + +void aom_accounting_record(Accounting *accounting, const char *str, + uint32_t bits) { + AccountingSymbol sym; + // Reuse previous symbol if it has the same context and symbol id. + if (accounting->syms.num_syms) { + AccountingSymbol *last_sym; + last_sym = &accounting->syms.syms[accounting->syms.num_syms - 1]; + if (memcmp(&last_sym->context, &accounting->context, + sizeof(AccountingSymbolContext)) == 0) { + uint32_t id; + id = aom_accounting_dictionary_lookup(accounting, str); + if (id == last_sym->id) { + last_sym->bits += bits; + last_sym->samples++; + return; + } + } + } + sym.context = accounting->context; + sym.samples = 1; + sym.bits = bits; + sym.id = aom_accounting_dictionary_lookup(accounting, str); + assert(sym.id <= 255); + if (accounting->syms.num_syms == accounting->num_syms_allocated) { + accounting->num_syms_allocated *= 2; + accounting->syms.syms = + realloc(accounting->syms.syms, + sizeof(AccountingSymbol) * accounting->num_syms_allocated); + assert(accounting->syms.syms != NULL); + } + accounting->syms.syms[accounting->syms.num_syms++] = sym; +} + +void aom_accounting_dump(Accounting *accounting) { + int i; + AccountingSymbol *sym; + printf("\n----- Number of recorded syntax elements = %d -----\n", + accounting->syms.num_syms); + printf("----- Total number of symbol calls = %d (%d binary) -----\n", + accounting->syms.num_multi_syms + accounting->syms.num_binary_syms, + accounting->syms.num_binary_syms); + for (i = 0; i < accounting->syms.num_syms; i++) { + sym = &accounting->syms.syms[i]; + printf("%s x: %d, y: %d bits: %f samples: %d\n", + accounting->syms.dictionary.strs[sym->id], sym->context.x, + sym->context.y, (float)sym->bits / 8.0, sym->samples); + } +} diff --git a/media/libaom/src/av1/decoder/accounting.h b/media/libaom/src/av1/decoder/accounting.h new file mode 100644 index 000000000..288e5e63e --- /dev/null +++ b/media/libaom/src/av1/decoder/accounting.h @@ -0,0 +1,82 @@ +/* + * 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 AOM_AV1_DECODER_ACCOUNTING_H_ +#define AOM_AV1_DECODER_ACCOUNTING_H_ +#include <stdlib.h> +#include "aom/aomdx.h" + +#ifdef __cplusplus +extern "C" { +#endif // __cplusplus + +#define AOM_ACCOUNTING_HASH_SIZE (1021) + +/* Max number of entries for symbol types in the dictionary (increase as + necessary). */ +#define MAX_SYMBOL_TYPES (256) + +/*The resolution of fractional-precision bit usage measurements, i.e., + 3 => 1/8th bits.*/ +#define AOM_ACCT_BITRES (3) + +typedef struct { + int16_t x; + int16_t y; +} AccountingSymbolContext; + +typedef struct { + AccountingSymbolContext context; + uint32_t id; + /** Number of bits in units of 1/8 bit. */ + uint32_t bits; + uint32_t samples; +} AccountingSymbol; + +/** Dictionary for translating strings into id. */ +typedef struct { + char *(strs[MAX_SYMBOL_TYPES]); + int num_strs; +} AccountingDictionary; + +typedef struct { + /** All recorded symbols decoded. */ + AccountingSymbol *syms; + /** Number of syntax actually recorded. */ + int num_syms; + /** Raw symbol decoding calls for non-binary values. */ + int num_multi_syms; + /** Raw binary symbol decoding calls. */ + int num_binary_syms; + /** Dictionary for translating strings into id. */ + AccountingDictionary dictionary; +} AccountingSymbols; + +struct Accounting { + AccountingSymbols syms; + /** Size allocated for symbols (not all may be used). */ + int num_syms_allocated; + int16_t hash_dictionary[AOM_ACCOUNTING_HASH_SIZE]; + AccountingSymbolContext context; + uint32_t last_tell_frac; +}; + +void aom_accounting_init(Accounting *accounting); +void aom_accounting_reset(Accounting *accounting); +void aom_accounting_clear(Accounting *accounting); +void aom_accounting_set_context(Accounting *accounting, int16_t x, int16_t y); +int aom_accounting_dictionary_lookup(Accounting *accounting, const char *str); +void aom_accounting_record(Accounting *accounting, const char *str, + uint32_t bits); +void aom_accounting_dump(Accounting *accounting); +#ifdef __cplusplus +} // extern "C" +#endif // __cplusplus +#endif // AOM_AV1_DECODER_ACCOUNTING_H_ diff --git a/media/libaom/src/av1/decoder/decodeframe.c b/media/libaom/src/av1/decoder/decodeframe.c new file mode 100644 index 000000000..31f14b531 --- /dev/null +++ b/media/libaom/src/av1/decoder/decodeframe.c @@ -0,0 +1,5567 @@ +/* + * 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__ + +// 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; + eob_info *eob_data = pd->eob_data + xd->txb_offset[plane]; + uint16_t scan_line = eob_data->max_scan_line; + uint16_t eob = eob_data->eob; + + memcpy(dqcoeff, pd->dqcoeff_block + xd->cb_offset[plane], + (scan_line + 1) * sizeof(dqcoeff[0])); + av1_inverse_transform_block(xd, dqcoeff, plane, tx_type, tx_size, dst, stride, + eob, reduced_tx_set); + memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0])); +} + +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->frame_offset, 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, get_frame_new_buffer(cm), 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."); + } + + int reader_corrupted_flag = aom_reader_has_error(r); + aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag); +} + +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_jnt_comp_avg = 0; + struct buf_2d *const dst_buf = &pd->dst; + uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x; + + ref = 0; + const RefBuffer *ref_buf = + &cm->frame_refs[this_mbmi->ref_frame[ref] - LAST_FRAME]; + + 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_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; + + const struct scale_factors *const sf = + is_intrabc ? &cm->sf_identity : &ref_buf->sf; + 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 = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH; + 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_refs[ref]->sf; + 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 + block.y0 * pre_buf->stride + block.x0; + src_stride[ref] = pre_buf->stride; + highbd = xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH; + + 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, subpel_params[ref].xs, + subpel_params[ref].ys, 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_jnt_comp_weight_assign(cm, mi, 0, &conv_params.fwd_offset, + &conv_params.bck_offset, + &conv_params.use_jnt_comp_avg, is_compound); + + for (ref = 0; ref < 1 + is_compound; ++ref) { + const struct scale_factors *const sf = + is_intrabc ? &cm->sf_identity : &xd->block_refs[ref]->sf; + 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, 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, 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 (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + 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, get_frame_new_buffer(cm), + 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 { + RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME]; + + xd->block_refs[ref] = ref_buf; + av1_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, &ref_buf->sf, + num_planes); + } + } + + 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->frame_offset, + 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, ¶ms.plane_width, + ¶ms.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 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, +#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); + + if (depth == MAX_VARTX_DEPTH) { + for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) { + for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) { + const int index = + av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx); + mbmi->inter_tx_size[index] = tx_size; + } + } + 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) { + for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) { + for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) { + const int index = + av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx); + mbmi->inter_tx_size[index] = sub_txs; + } + } + mbmi->tx_size = sub_txs; + txfm_partition_update(xd->above_txfm_context + blk_col, + xd->left_txfm_context + blk_row, sub_txs, tx_size); +#if LOOP_FILTER_BITMASK + store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col, BLOCK_8X8, + TX_4X4, mbmi); +#endif + return; + } +#if LOOP_FILTER_BITMASK + if (depth + 1 == MAX_VARTX_DEPTH) { + store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col, + txsize_to_bsize[tx_size], sub_txs, mbmi); + } +#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, +#endif + offsetr, offsetc, r); + } + } + } else { + for (int idy = 0; idy < tx_size_high_unit[tx_size]; ++idy) { + for (int idx = 0; idx < tx_size_wide_unit[tx_size]; ++idx) { + const int index = + av1_get_txb_size_index(bsize, blk_row + idy, blk_col + idx); + mbmi->inter_tx_size[index] = tx_size; + } + } + mbmi->tx_size = tx_size; + txfm_partition_update(xd->above_txfm_context + blk_col, + xd->left_txfm_context + blk_row, tx_size, tx_size); +#if LOOP_FILTER_BITMASK + store_bitmask_vartx(cm, mi_row + blk_row, mi_col + blk_col, + txsize_to_bsize[tx_size], tx_size, mbmi); +#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 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); + const uint64_t top_edge_mask = + ((uint64_t)1 << (shift + mi_size_wide[bsize])) - ((uint64_t)1 << shift); + lfm->is_horz_border.bits[index] |= top_edge_mask; + const int is_vert_border = mask_id_table_vert_border[bsize]; + const int vert_shift = block_size_high[bsize] <= 8 ? shift : col_start; + for (int i = 0; i + index < 4; ++i) { + lfm->is_vert_border.bits[i + index] |= + (left_mask_univariant_reordered[is_vert_border].bits[i] << vert_shift); + } + 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[row][col_start], level_u, + sizeof(uint8_t) * mi_size_wide[bsize]); + memset(&lfm->lfl_v[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, +#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); + } 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); + } + } + } +#endif + + if (cm->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); + + int reader_corrupted_flag = aom_reader_has_error(r); + aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag); +} + +static void set_offsets_for_pred_and_recon(AV1Decoder *const pbi, + 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, get_frame_new_buffer(cm), 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 *r, + BLOCK_SIZE bsize, int parse_decode_flag) { + 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, r, plane, runit_idx); + } + } + } + } + + partition = (bsize < BLOCK_8X8) ? PARTITION_NONE + : read_partition(xd, mi_row, mi_col, r, + has_rows, has_cols, bsize); + } else { + partition = get_partition(cm, mi_row, mi_col, bsize); + } + subsize = get_partition_subsize(bsize, partition); + + // 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), r, \ + DEC_BLOCK_EPT_ARG(db_subsize)) +#define DEC_PARTITION(db_r, db_c, db_subsize) \ + decode_partition(pbi, td, DEC_BLOCK_STX_ARG(db_r), (db_c), r, (db_subsize), \ + parse_decode_flag) + + 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); + if (cm->allow_intrabc) return; + cm->cdef_pri_damping = cm->cdef_sec_damping = aom_rb_read_literal(rb, 2) + 3; + cm->cdef_bits = aom_rb_read_literal(rb, 2); + cm->nb_cdef_strengths = 1 << cm->cdef_bits; + for (int i = 0; i < cm->nb_cdef_strengths; i++) { + cm->cdef_strengths[i] = aom_rb_read_literal(rb, CDEF_STRENGTH_BITS); + cm->cdef_uv_strengths[i] = + 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->dequant_bit_depth = seq_params->bit_depth; + cm->using_qmatrix = aom_rb_read_bit(rb); + if (cm->using_qmatrix) { + cm->qm_y = aom_rb_read_literal(rb, QM_LEVEL_BITS); + 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) { + const int bit_depth = cm->seq_params.bit_depth; + const int using_qm = cm->using_qmatrix; + // When segmentation is disabled, only the first value is used. The + // remaining are don't cares. + const int max_segments = cm->seg.enabled ? MAX_SEGMENTS : 1; + for (int i = 0; i < max_segments; ++i) { + const int qindex = av1_get_qindex(&cm->seg, i, cm->base_qindex); + 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 = qindex == 0 && cm->y_dc_delta_q == 0 && + cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 && + cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0; + // 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( + get_frame_new_buffer(cm), cm->width, cm->height, + seq_params->subsampling_x, seq_params->subsampling_y, + seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS, + cm->byte_alignment, + &pool->frame_bufs[cm->new_fb_idx].raw_frame_buffer, pool->get_fb_cb, + pool->cb_priv)) { + unlock_buffer_pool(pool); + aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = + seq_params->subsampling_x; + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = + seq_params->subsampling_y; + pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = + (unsigned int)seq_params->bit_depth; + pool->frame_bufs[cm->new_fb_idx].buf.color_primaries = + seq_params->color_primaries; + pool->frame_bufs[cm->new_fb_idx].buf.transfer_characteristics = + seq_params->transfer_characteristics; + pool->frame_bufs[cm->new_fb_idx].buf.matrix_coefficients = + seq_params->matrix_coefficients; + pool->frame_bufs[cm->new_fb_idx].buf.monochrome = seq_params->monochrome; + pool->frame_bufs[cm->new_fb_idx].buf.chroma_sample_position = + seq_params->chroma_sample_position; + pool->frame_bufs[cm->new_fb_idx].buf.color_range = seq_params->color_range; + pool->frame_bufs[cm->new_fb_idx].buf.render_width = cm->render_width; + pool->frame_bufs[cm->new_fb_idx].buf.render_height = cm->render_height; +} + +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 || + height > seq_params->max_frame_height) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Frame dimensions are larger than the maximum values"); + } + } else { + width = seq_params->max_frame_width; + height = seq_params->max_frame_height; + } + + setup_superres(cm, rb, &width, &height); + resize_context_buffers(cm, width, height); + setup_render_size(cm, rb); + setup_buffer_pool(cm); +} + +static void setup_sb_size(SequenceHeader *seq_params, + struct aom_read_bit_buffer *rb) { + set_sb_size(seq_params, aom_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64); +} + +static INLINE int valid_ref_frame_img_fmt(aom_bit_depth_t ref_bit_depth, + int ref_xss, int ref_yss, + aom_bit_depth_t this_bit_depth, + int this_xss, int this_yss) { + return ref_bit_depth == this_bit_depth && ref_xss == this_xss && + ref_yss == this_yss; +} + +static void setup_frame_size_with_refs(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb) { + int width, height; + int found = 0; + int has_valid_ref_frame = 0; + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + if (aom_rb_read_bit(rb)) { + YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf; + width = buf->y_crop_width; + height = buf->y_crop_height; + cm->render_width = buf->render_width; + cm->render_height = buf->render_height; + setup_superres(cm, rb, &width, &height); + resize_context_buffers(cm, width, height); + found = 1; + break; + } + } + + const SequenceHeader *const seq_params = &cm->seq_params; + if (!found) { + 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); + setup_superres(cm, rb, &width, &height); + resize_context_buffers(cm, width, height); + setup_render_size(cm, rb); + } + + if (width <= 0 || height <= 0) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Invalid frame size"); + + // Check to make sure at least one of frames that this frame references + // has valid dimensions. + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + RefBuffer *const ref_frame = &cm->frame_refs[i]; + has_valid_ref_frame |= + valid_ref_frame_size(ref_frame->buf->y_crop_width, + ref_frame->buf->y_crop_height, width, height); + } + if (!has_valid_ref_frame) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Referenced frame has invalid size"); + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + RefBuffer *const ref_frame = &cm->frame_refs[i]; + if (!valid_ref_frame_img_fmt( + ref_frame->buf->bit_depth, ref_frame->buf->subsampling_x, + ref_frame->buf->subsampling_y, seq_params->bit_depth, + seq_params->subsampling_x, seq_params->subsampling_y)) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Referenced frame has incompatible color format"); + } + setup_buffer_pool(cm); +} + +// Same function as av1_read_uniform but reading from uncompresses header wb +static int rb_read_uniform(struct aom_read_bit_buffer *const rb, int n) { + const int l = get_unsigned_bits(n); + const int m = (1 << l) - n; + const int v = aom_rb_read_literal(rb, l - 1); + assert(l != 0); + if (v < m) + return v; + else + return (v << 1) - m + aom_rb_read_bit(rb); +} + +static void read_tile_info_max_tile(AV1_COMMON *const cm, + struct aom_read_bit_buffer *const rb) { + int width_mi = ALIGN_POWER_OF_TWO(cm->mi_cols, cm->seq_params.mib_size_log2); + int height_mi = ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2); + int width_sb = width_mi >> cm->seq_params.mib_size_log2; + int height_sb = height_mi >> cm->seq_params.mib_size_log2; + + av1_get_tile_limits(cm); + cm->uniform_tile_spacing_flag = aom_rb_read_bit(rb); + + // Read tile columns + if (cm->uniform_tile_spacing_flag) { + cm->log2_tile_cols = cm->min_log2_tile_cols; + while (cm->log2_tile_cols < cm->max_log2_tile_cols) { + if (!aom_rb_read_bit(rb)) { + break; + } + cm->log2_tile_cols++; + } + } else { + int i; + int start_sb; + for (i = 0, start_sb = 0; width_sb > 0 && i < MAX_TILE_COLS; i++) { + const int size_sb = + 1 + rb_read_uniform(rb, AOMMIN(width_sb, cm->max_tile_width_sb)); + cm->tile_col_start_sb[i] = start_sb; + start_sb += size_sb; + width_sb -= size_sb; + } + cm->tile_cols = i; + cm->tile_col_start_sb[i] = start_sb + width_sb; + } + av1_calculate_tile_cols(cm); + + // Read tile rows + if (cm->uniform_tile_spacing_flag) { + cm->log2_tile_rows = cm->min_log2_tile_rows; + while (cm->log2_tile_rows < cm->max_log2_tile_rows) { + if (!aom_rb_read_bit(rb)) { + break; + } + cm->log2_tile_rows++; + } + } else { + int i; + int start_sb; + for (i = 0, start_sb = 0; height_sb > 0 && i < MAX_TILE_ROWS; i++) { + const int size_sb = + 1 + rb_read_uniform(rb, AOMMIN(height_sb, cm->max_tile_height_sb)); + cm->tile_row_start_sb[i] = start_sb; + start_sb += size_sb; + height_sb -= size_sb; + } + cm->tile_rows = i; + cm->tile_row_start_sb[i] = start_sb + height_sb; + } + av1_calculate_tile_rows(cm); +} + +void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm) { + cm->single_tile_decoding = 0; + if (cm->large_scale_tile) { + struct loopfilter *lf = &cm->lf; + + // Figure out single_tile_decoding by loopfilter_level. + const int no_loopfilter = !(lf->filter_level[0] || lf->filter_level[1]); + const int no_cdef = cm->cdef_bits == 0 && cm->cdef_strengths[0] == 0 && + cm->cdef_uv_strengths[0] == 0; + const int no_restoration = + cm->rst_info[0].frame_restoration_type == RESTORE_NONE && + cm->rst_info[1].frame_restoration_type == RESTORE_NONE && + cm->rst_info[2].frame_restoration_type == RESTORE_NONE; + assert(IMPLIES(cm->coded_lossless, no_loopfilter && no_cdef)); + assert(IMPLIES(cm->all_lossless, no_restoration)); + cm->single_tile_decoding = no_loopfilter && no_cdef && no_restoration; + } +} + +static void read_tile_info(AV1Decoder *const pbi, + struct aom_read_bit_buffer *const rb) { + AV1_COMMON *const cm = &pbi->common; + + read_tile_info_max_tile(cm, rb); + + cm->context_update_tile_id = 0; + if (cm->tile_rows * cm->tile_cols > 1) { + // tile to use for cdf update + cm->context_update_tile_id = + aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols); + if (cm->context_update_tile_id >= cm->tile_rows * cm->tile_cols) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Invalid context_update_tile_id"); + } + // tile size magnitude + pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1; + } +} + +#if EXT_TILE_DEBUG +static void read_ext_tile_info(AV1Decoder *const pbi, + struct aom_read_bit_buffer *const rb) { + AV1_COMMON *const cm = &pbi->common; + + // This information is stored as a separate byte. + int mod = rb->bit_offset % CHAR_BIT; + if (mod > 0) aom_rb_read_literal(rb, CHAR_BIT - mod); + assert(rb->bit_offset % CHAR_BIT == 0); + + if (cm->tile_cols * cm->tile_rows > 1) { + // Read the number of bytes used to store tile size + pbi->tile_col_size_bytes = aom_rb_read_literal(rb, 2) + 1; + pbi->tile_size_bytes = aom_rb_read_literal(rb, 2) + 1; + } +} +#endif // EXT_TILE_DEBUG + +static size_t mem_get_varsize(const uint8_t *src, int sz) { + switch (sz) { + case 1: return src[0]; + case 2: return mem_get_le16(src); + case 3: return mem_get_le24(src); + case 4: return mem_get_le32(src); + default: assert(0 && "Invalid size"); return -1; + } +} + +#if EXT_TILE_DEBUG +// Reads the next tile returning its size and adjusting '*data' accordingly +// based on 'is_last'. On return, '*data' is updated to point to the end of the +// raw tile buffer in the bit stream. +static void get_ls_tile_buffer( + const uint8_t *const data_end, struct aom_internal_error_info *error_info, + const uint8_t **data, TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], + int tile_size_bytes, int col, int row, int tile_copy_mode) { + size_t size; + + size_t copy_size = 0; + const uint8_t *copy_data = NULL; + + if (!read_is_valid(*data, tile_size_bytes, data_end)) + aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + size = mem_get_varsize(*data, tile_size_bytes); + + // If tile_copy_mode = 1, then the top bit of the tile header indicates copy + // mode. + if (tile_copy_mode && (size >> (tile_size_bytes * 8 - 1)) == 1) { + // The remaining bits in the top byte signal the row offset + int offset = (size >> (tile_size_bytes - 1) * 8) & 0x7f; + + // Currently, only use tiles in same column as reference tiles. + copy_data = tile_buffers[row - offset][col].data; + copy_size = tile_buffers[row - offset][col].size; + size = 0; + } else { + size += AV1_MIN_TILE_SIZE_BYTES; + } + + *data += tile_size_bytes; + + if (size > (size_t)(data_end - *data)) + aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile size"); + + if (size > 0) { + tile_buffers[row][col].data = *data; + tile_buffers[row][col].size = size; + } else { + tile_buffers[row][col].data = copy_data; + tile_buffers[row][col].size = copy_size; + } + + *data += size; +} + +// Returns the end of the last tile buffer +// (tile_buffers[cm->tile_rows - 1][cm->tile_cols - 1]). +static const uint8_t *get_ls_tile_buffers( + AV1Decoder *pbi, const uint8_t *data, const uint8_t *data_end, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { + AV1_COMMON *const cm = &pbi->common; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + const int have_tiles = tile_cols * tile_rows > 1; + const uint8_t *raw_data_end; // The end of the last tile buffer + + if (!have_tiles) { + const size_t tile_size = data_end - data; + tile_buffers[0][0].data = data; + tile_buffers[0][0].size = tile_size; + raw_data_end = NULL; + } else { + // We locate only the tile buffers that are required, which are the ones + // specified by pbi->dec_tile_col and pbi->dec_tile_row. Also, we always + // need the last (bottom right) tile buffer, as we need to know where the + // end of the compressed frame buffer is for proper superframe decoding. + + const uint8_t *tile_col_data_end[MAX_TILE_COLS] = { NULL }; + const uint8_t *const data_start = data; + + const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); + const int single_row = pbi->dec_tile_row >= 0; + const int tile_rows_start = single_row ? dec_tile_row : 0; + const int tile_rows_end = single_row ? tile_rows_start + 1 : tile_rows; + const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); + const int single_col = pbi->dec_tile_col >= 0; + const int tile_cols_start = single_col ? dec_tile_col : 0; + const int tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; + + const int tile_col_size_bytes = pbi->tile_col_size_bytes; + const int tile_size_bytes = pbi->tile_size_bytes; + const int tile_copy_mode = + ((AOMMAX(cm->tile_width, cm->tile_height) << MI_SIZE_LOG2) <= 256) ? 1 + : 0; + // Read tile column sizes for all columns (we need the last tile buffer) + for (int c = 0; c < tile_cols; ++c) { + const int is_last = c == tile_cols - 1; + size_t tile_col_size; + + if (!is_last) { + tile_col_size = mem_get_varsize(data, tile_col_size_bytes); + data += tile_col_size_bytes; + tile_col_data_end[c] = data + tile_col_size; + } else { + tile_col_size = data_end - data; + tile_col_data_end[c] = data_end; + } + data += tile_col_size; + } + + data = data_start; + + // Read the required tile sizes. + for (int c = tile_cols_start; c < tile_cols_end; ++c) { + const int is_last = c == tile_cols - 1; + + if (c > 0) data = tile_col_data_end[c - 1]; + + if (!is_last) data += tile_col_size_bytes; + + // Get the whole of the last column, otherwise stop at the required tile. + for (int r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) { + get_ls_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data, + tile_buffers, tile_size_bytes, c, r, tile_copy_mode); + } + } + + // If we have not read the last column, then read it to get the last tile. + if (tile_cols_end != tile_cols) { + const int c = tile_cols - 1; + + data = tile_col_data_end[c - 1]; + + for (int r = 0; r < tile_rows; ++r) { + get_ls_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data, + tile_buffers, tile_size_bytes, c, r, tile_copy_mode); + } + } + raw_data_end = data; + } + return raw_data_end; +} +#endif // EXT_TILE_DEBUG + +static const uint8_t *get_ls_single_tile_buffer( + AV1Decoder *pbi, const uint8_t *data, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { + assert(pbi->dec_tile_row >= 0 && pbi->dec_tile_col >= 0); + tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].data = data; + tile_buffers[pbi->dec_tile_row][pbi->dec_tile_col].size = + (size_t)pbi->coded_tile_data_size; + return data + pbi->coded_tile_data_size; +} + +// Reads the next tile returning its size and adjusting '*data' accordingly +// based on 'is_last'. +static void get_tile_buffer(const uint8_t *const data_end, + const int tile_size_bytes, int is_last, + struct aom_internal_error_info *error_info, + const uint8_t **data, TileBufferDec *const buf) { + size_t size; + + if (!is_last) { + if (!read_is_valid(*data, tile_size_bytes, data_end)) + aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + size = mem_get_varsize(*data, tile_size_bytes) + AV1_MIN_TILE_SIZE_BYTES; + *data += tile_size_bytes; + + if (size > (size_t)(data_end - *data)) + aom_internal_error(error_info, AOM_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile size"); + } else { + size = data_end - *data; + } + + buf->data = *data; + buf->size = size; + + *data += size; +} + +static void get_tile_buffers(AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], + int start_tile, int end_tile) { + AV1_COMMON *const cm = &pbi->common; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + int tc = 0; + int first_tile_in_tg = 0; + + for (int r = 0; r < tile_rows; ++r) { + for (int c = 0; c < tile_cols; ++c, ++tc) { + TileBufferDec *const buf = &tile_buffers[r][c]; + + const int is_last = (tc == end_tile); + const size_t hdr_offset = 0; + + if (tc < start_tile || tc > end_tile) continue; + + if (data + hdr_offset >= data_end) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Data ended before all tiles were read."); + first_tile_in_tg += tc == first_tile_in_tg ? pbi->tg_size : 0; + data += hdr_offset; + get_tile_buffer(data_end, pbi->tile_size_bytes, is_last, + &pbi->common.error, &data, buf); + } + } +} + +static void set_cb_buffer(AV1Decoder *pbi, MACROBLOCKD *const xd, + CB_BUFFER *cb_buffer_base, const int num_planes, + int mi_row, int mi_col) { + AV1_COMMON *const cm = &pbi->common; + int mib_size_log2 = cm->seq_params.mib_size_log2; + int stride = (cm->mi_cols >> mib_size_log2) + 1; + int offset = (mi_row >> mib_size_log2) * stride + (mi_col >> mib_size_log2); + CB_BUFFER *cb_buffer = cb_buffer_base + offset; + + for (int plane = 0; plane < num_planes; ++plane) { + xd->plane[plane].dqcoeff_block = cb_buffer->dqcoeff[plane]; + xd->plane[plane].eob_data = cb_buffer->eob_data[plane]; + xd->cb_offset[plane] = 0; + xd->txb_offset[plane] = 0; + } + xd->plane[0].color_index_map = cb_buffer->color_index_map[0]; + xd->plane[1].color_index_map = cb_buffer->color_index_map[1]; + xd->color_index_map_offset[0] = 0; + xd->color_index_map_offset[1] = 0; +} + +static void decoder_alloc_tile_data(AV1Decoder *pbi, const int n_tiles) { + AV1_COMMON *const cm = &pbi->common; + aom_free(pbi->tile_data); + CHECK_MEM_ERROR(cm, pbi->tile_data, + aom_memalign(32, n_tiles * sizeof(*pbi->tile_data))); + pbi->allocated_tiles = n_tiles; + for (int i = 0; i < n_tiles; i++) { + TileDataDec *const tile_data = pbi->tile_data + i; + av1_zero(tile_data->dec_row_mt_sync); + } + pbi->allocated_row_mt_sync_rows = 0; +} + +// Set up nsync by width. +static INLINE int get_sync_range(int width) { +// nsync numbers are picked by testing. +#if 0 + if (width < 640) + return 1; + else if (width <= 1280) + return 2; + else if (width <= 4096) + return 4; + else + return 8; +#else + (void)width; +#endif + return 1; +} + +// Allocate memory for decoder row synchronization +static void dec_row_mt_alloc(AV1DecRowMTSync *dec_row_mt_sync, AV1_COMMON *cm, + int rows) { + dec_row_mt_sync->allocated_sb_rows = rows; +#if CONFIG_MULTITHREAD + { + int i; + + CHECK_MEM_ERROR(cm, dec_row_mt_sync->mutex_, + aom_malloc(sizeof(*(dec_row_mt_sync->mutex_)) * rows)); + if (dec_row_mt_sync->mutex_) { + for (i = 0; i < rows; ++i) { + pthread_mutex_init(&dec_row_mt_sync->mutex_[i], NULL); + } + } + + CHECK_MEM_ERROR(cm, dec_row_mt_sync->cond_, + aom_malloc(sizeof(*(dec_row_mt_sync->cond_)) * rows)); + if (dec_row_mt_sync->cond_) { + for (i = 0; i < rows; ++i) { + pthread_cond_init(&dec_row_mt_sync->cond_[i], NULL); + } + } + } +#endif // CONFIG_MULTITHREAD + + CHECK_MEM_ERROR(cm, dec_row_mt_sync->cur_sb_col, + aom_malloc(sizeof(*(dec_row_mt_sync->cur_sb_col)) * rows)); + + // Set up nsync. + dec_row_mt_sync->sync_range = get_sync_range(cm->width); +} + +// Deallocate decoder row synchronization related mutex and data +void av1_dec_row_mt_dealloc(AV1DecRowMTSync *dec_row_mt_sync) { + if (dec_row_mt_sync != NULL) { +#if CONFIG_MULTITHREAD + int i; + if (dec_row_mt_sync->mutex_ != NULL) { + for (i = 0; i < dec_row_mt_sync->allocated_sb_rows; ++i) { + pthread_mutex_destroy(&dec_row_mt_sync->mutex_[i]); + } + aom_free(dec_row_mt_sync->mutex_); + } + if (dec_row_mt_sync->cond_ != NULL) { + for (i = 0; i < dec_row_mt_sync->allocated_sb_rows; ++i) { + pthread_cond_destroy(&dec_row_mt_sync->cond_[i]); + } + aom_free(dec_row_mt_sync->cond_); + } +#endif // CONFIG_MULTITHREAD + aom_free(dec_row_mt_sync->cur_sb_col); + + // clear the structure as the source of this call may be a resize in which + // case this call will be followed by an _alloc() which may fail. + av1_zero(*dec_row_mt_sync); + } +} + +static INLINE void sync_read(AV1DecRowMTSync *const dec_row_mt_sync, int r, + int c) { +#if CONFIG_MULTITHREAD + const int nsync = dec_row_mt_sync->sync_range; + + if (r && !(c & (nsync - 1))) { + pthread_mutex_t *const mutex = &dec_row_mt_sync->mutex_[r - 1]; + pthread_mutex_lock(mutex); + + while (c > dec_row_mt_sync->cur_sb_col[r - 1] - nsync) { + pthread_cond_wait(&dec_row_mt_sync->cond_[r - 1], mutex); + } + pthread_mutex_unlock(mutex); + } +#else + (void)dec_row_mt_sync; + (void)r; + (void)c; +#endif // CONFIG_MULTITHREAD +} + +static INLINE void sync_write(AV1DecRowMTSync *const dec_row_mt_sync, int r, + int c, const int sb_cols) { +#if CONFIG_MULTITHREAD + const int nsync = dec_row_mt_sync->sync_range; + int cur; + int sig = 1; + + if (c < sb_cols - 1) { + cur = c; + if (c % nsync) sig = 0; + } else { + cur = sb_cols + nsync; + } + + if (sig) { + pthread_mutex_lock(&dec_row_mt_sync->mutex_[r]); + + dec_row_mt_sync->cur_sb_col[r] = cur; + + pthread_cond_signal(&dec_row_mt_sync->cond_[r]); + pthread_mutex_unlock(&dec_row_mt_sync->mutex_[r]); + } +#else + (void)dec_row_mt_sync; + (void)r; + (void)c; + (void)sb_cols; +#endif // CONFIG_MULTITHREAD +} + +static void decode_tile_sb_row(AV1Decoder *pbi, ThreadData *const td, + TileInfo tile_info, const int mi_row) { + AV1_COMMON *const cm = &pbi->common; + const int num_planes = av1_num_planes(cm); + TileDataDec *const tile_data = + pbi->tile_data + tile_info.tile_row * cm->tile_cols + tile_info.tile_col; + const int sb_cols_in_tile = av1_get_sb_cols_in_tile(cm, tile_info); + const int sb_row_in_tile = + (mi_row - tile_info.mi_row_start) >> cm->seq_params.mib_size_log2; + int sb_col_in_tile = 0; + + for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end; + mi_col += cm->seq_params.mib_size, sb_col_in_tile++) { + set_cb_buffer(pbi, &td->xd, pbi->cb_buffer_base, num_planes, mi_row, + mi_col); + + sync_read(&tile_data->dec_row_mt_sync, sb_row_in_tile, sb_col_in_tile); + + // Decoding of the super-block + decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, + cm->seq_params.sb_size, 0x2); + + sync_write(&tile_data->dec_row_mt_sync, sb_row_in_tile, sb_col_in_tile, + sb_cols_in_tile); + } +} + +static int check_trailing_bits_after_symbol_coder(aom_reader *r) { + if (aom_reader_has_overflowed(r)) return -1; + + uint32_t nb_bits = aom_reader_tell(r); + uint32_t nb_bytes = (nb_bits + 7) >> 3; + const uint8_t *p = aom_reader_find_begin(r) + nb_bytes; + + // aom_reader_tell() returns 1 for a newly initialized decoder, and the + // return value only increases as values are decoded. So nb_bits > 0, and + // thus p > p_begin. Therefore accessing p[-1] is safe. + uint8_t last_byte = p[-1]; + uint8_t pattern = 128 >> ((nb_bits - 1) & 7); + if ((last_byte & (2 * pattern - 1)) != pattern) return -1; + + // Make sure that all padding bytes are zero as required by the spec. + const uint8_t *p_end = aom_reader_find_end(r); + while (p < p_end) { + if (*p != 0) return -1; + p++; + } + return 0; +} + +static void set_decode_func_pointers(ThreadData *td, int parse_decode_flag) { + td->read_coeffs_tx_intra_block_visit = decode_block_void; + td->predict_and_recon_intra_block_visit = decode_block_void; + td->read_coeffs_tx_inter_block_visit = decode_block_void; + td->inverse_tx_inter_block_visit = decode_block_void; + td->predict_inter_block_visit = predict_inter_block_void; + td->cfl_store_inter_block_visit = cfl_store_inter_block_void; + + if (parse_decode_flag & 0x1) { + td->read_coeffs_tx_intra_block_visit = read_coeffs_tx_intra_block; + td->read_coeffs_tx_inter_block_visit = av1_read_coeffs_txb_facade; + } + if (parse_decode_flag & 0x2) { + td->predict_and_recon_intra_block_visit = + predict_and_reconstruct_intra_block; + td->inverse_tx_inter_block_visit = inverse_transform_inter_block; + td->predict_inter_block_visit = predict_inter_block; + td->cfl_store_inter_block_visit = cfl_store_inter_block; + } +} + +static void decode_tile(AV1Decoder *pbi, ThreadData *const td, int tile_row, + int tile_col) { + TileInfo tile_info; + + AV1_COMMON *const cm = &pbi->common; + const int num_planes = av1_num_planes(cm); + + av1_tile_set_row(&tile_info, cm, tile_row); + av1_tile_set_col(&tile_info, cm, tile_col); + av1_zero_above_context(cm, &td->xd, tile_info.mi_col_start, + tile_info.mi_col_end, tile_row); + av1_reset_loop_filter_delta(&td->xd, num_planes); + av1_reset_loop_restoration(&td->xd, num_planes); + + for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end; + mi_row += cm->seq_params.mib_size) { + av1_zero_left_context(&td->xd); + + for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end; + mi_col += cm->seq_params.mib_size) { + set_cb_buffer(pbi, &td->xd, &td->cb_buffer_base, num_planes, 0, 0); + + // Bit-stream parsing and decoding of the superblock + decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, + cm->seq_params.sb_size, 0x3); + + if (aom_reader_has_overflowed(td->bit_reader)) { + aom_merge_corrupted_flag(&td->xd.corrupted, 1); + return; + } + } + } + + int corrupted = + (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0; + aom_merge_corrupted_flag(&td->xd.corrupted, corrupted); +} + +static const uint8_t *decode_tiles(AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, int start_tile, + int end_tile) { + AV1_COMMON *const cm = &pbi->common; + ThreadData *const td = &pbi->td; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + const int n_tiles = tile_cols * tile_rows; + TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; + const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); + const int single_row = pbi->dec_tile_row >= 0; + const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); + const int single_col = pbi->dec_tile_col >= 0; + int tile_rows_start; + int tile_rows_end; + int tile_cols_start; + int tile_cols_end; + int inv_col_order; + int inv_row_order; + int tile_row, tile_col; + uint8_t allow_update_cdf; + const uint8_t *raw_data_end = NULL; + + if (cm->large_scale_tile) { + tile_rows_start = single_row ? dec_tile_row : 0; + tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; + tile_cols_start = single_col ? dec_tile_col : 0; + tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; + inv_col_order = pbi->inv_tile_order && !single_col; + inv_row_order = pbi->inv_tile_order && !single_row; + allow_update_cdf = 0; + } else { + tile_rows_start = 0; + tile_rows_end = tile_rows; + tile_cols_start = 0; + tile_cols_end = tile_cols; + inv_col_order = pbi->inv_tile_order; + inv_row_order = pbi->inv_tile_order; + allow_update_cdf = 1; + } + + // No tiles to decode. + if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || + // First tile is larger than end_tile. + tile_rows_start * cm->tile_cols + tile_cols_start > end_tile || + // Last tile is smaller than start_tile. + (tile_rows_end - 1) * cm->tile_cols + tile_cols_end - 1 < start_tile) + return data; + + allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update; + + assert(tile_rows <= MAX_TILE_ROWS); + assert(tile_cols <= MAX_TILE_COLS); + +#if EXT_TILE_DEBUG + if (cm->large_scale_tile && !pbi->ext_tile_debug) + raw_data_end = get_ls_single_tile_buffer(pbi, data, tile_buffers); + else if (cm->large_scale_tile && pbi->ext_tile_debug) + raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); + else +#endif // EXT_TILE_DEBUG + get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); + + if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { + decoder_alloc_tile_data(pbi, n_tiles); + } +#if CONFIG_ACCOUNTING + if (pbi->acct_enabled) { + aom_accounting_reset(&pbi->accounting); + } +#endif + + set_decode_func_pointers(&pbi->td, 0x3); + + // Load all tile information into thread_data. + td->xd = pbi->mb; + td->xd.corrupted = 0; + td->xd.mc_buf[0] = td->mc_buf[0]; + td->xd.mc_buf[1] = td->mc_buf[1]; + td->xd.tmp_conv_dst = td->tmp_conv_dst; + for (int j = 0; j < 2; ++j) { + td->xd.tmp_obmc_bufs[j] = td->tmp_obmc_bufs[j]; + } + + for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { + const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row; + + for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) { + const int col = inv_col_order ? tile_cols - 1 - tile_col : tile_col; + TileDataDec *const tile_data = pbi->tile_data + row * cm->tile_cols + col; + const TileBufferDec *const tile_bs_buf = &tile_buffers[row][col]; + + if (row * cm->tile_cols + col < start_tile || + row * cm->tile_cols + col > end_tile) + continue; + + td->bit_reader = &tile_data->bit_reader; + av1_zero(td->dqcoeff); + av1_tile_init(&td->xd.tile, cm, row, col); + td->xd.current_qindex = cm->base_qindex; + setup_bool_decoder(tile_bs_buf->data, data_end, tile_bs_buf->size, + &cm->error, td->bit_reader, allow_update_cdf); +#if CONFIG_ACCOUNTING + if (pbi->acct_enabled) { + td->bit_reader->accounting = &pbi->accounting; + td->bit_reader->accounting->last_tell_frac = + aom_reader_tell_frac(td->bit_reader); + } else { + td->bit_reader->accounting = NULL; + } +#endif + av1_init_macroblockd(cm, &td->xd, td->dqcoeff); + av1_init_above_context(cm, &td->xd, row); + + // Initialise the tile context from the frame context + tile_data->tctx = *cm->fc; + td->xd.tile_ctx = &tile_data->tctx; + + // decode tile + decode_tile(pbi, td, row, col); + aom_merge_corrupted_flag(&pbi->mb.corrupted, td->xd.corrupted); + if (pbi->mb.corrupted) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Failed to decode tile data"); + } + } + + if (cm->large_scale_tile) { + if (n_tiles == 1) { + // Find the end of the single tile buffer + return aom_reader_find_end(&pbi->tile_data->bit_reader); + } + // Return the end of the last tile buffer + return raw_data_end; + } + TileDataDec *const tile_data = pbi->tile_data + end_tile; + + return aom_reader_find_end(&tile_data->bit_reader); +} + +static TileJobsDec *get_dec_job_info(AV1DecTileMT *tile_mt_info) { + TileJobsDec *cur_job_info = NULL; +#if CONFIG_MULTITHREAD + pthread_mutex_lock(tile_mt_info->job_mutex); + + if (tile_mt_info->jobs_dequeued < tile_mt_info->jobs_enqueued) { + cur_job_info = tile_mt_info->job_queue + tile_mt_info->jobs_dequeued; + tile_mt_info->jobs_dequeued++; + } + + pthread_mutex_unlock(tile_mt_info->job_mutex); +#else + (void)tile_mt_info; +#endif + return cur_job_info; +} + +static void tile_worker_hook_init(AV1Decoder *const pbi, + DecWorkerData *const thread_data, + const TileBufferDec *const tile_buffer, + TileDataDec *const tile_data, + uint8_t allow_update_cdf) { + AV1_COMMON *cm = &pbi->common; + ThreadData *const td = thread_data->td; + int tile_row = tile_data->tile_info.tile_row; + int tile_col = tile_data->tile_info.tile_col; + + td->bit_reader = &tile_data->bit_reader; + av1_zero(td->dqcoeff); + av1_tile_init(&td->xd.tile, cm, tile_row, tile_col); + td->xd.current_qindex = cm->base_qindex; + setup_bool_decoder(tile_buffer->data, thread_data->data_end, + tile_buffer->size, &thread_data->error_info, + td->bit_reader, allow_update_cdf); +#if CONFIG_ACCOUNTING + if (pbi->acct_enabled) { + td->bit_reader->accounting = &pbi->accounting; + td->bit_reader->accounting->last_tell_frac = + aom_reader_tell_frac(td->bit_reader); + } else { + td->bit_reader->accounting = NULL; + } +#endif + av1_init_macroblockd(cm, &td->xd, td->dqcoeff); + td->xd.error_info = &thread_data->error_info; + av1_init_above_context(cm, &td->xd, tile_row); + + // Initialise the tile context from the frame context + tile_data->tctx = *cm->fc; + td->xd.tile_ctx = &tile_data->tctx; +#if CONFIG_ACCOUNTING + if (pbi->acct_enabled) { + tile_data->bit_reader.accounting->last_tell_frac = + aom_reader_tell_frac(&tile_data->bit_reader); + } +#endif +} + +static int tile_worker_hook(void *arg1, void *arg2) { + DecWorkerData *const thread_data = (DecWorkerData *)arg1; + AV1Decoder *const pbi = (AV1Decoder *)arg2; + AV1_COMMON *cm = &pbi->common; + ThreadData *const td = thread_data->td; + uint8_t allow_update_cdf; + + // The jmp_buf is valid only for the duration of the function that calls + // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 + // before it returns. + if (setjmp(thread_data->error_info.jmp)) { + thread_data->error_info.setjmp = 0; + thread_data->td->xd.corrupted = 1; + return 0; + } + thread_data->error_info.setjmp = 1; + + allow_update_cdf = cm->large_scale_tile ? 0 : 1; + allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update; + + set_decode_func_pointers(td, 0x3); + + assert(cm->tile_cols > 0); + while (1) { + TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info); + + if (cur_job_info != NULL && !td->xd.corrupted) { + const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer; + TileDataDec *const tile_data = cur_job_info->tile_data; + tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data, + allow_update_cdf); + // decode tile + int tile_row = tile_data->tile_info.tile_row; + int tile_col = tile_data->tile_info.tile_col; + decode_tile(pbi, td, tile_row, tile_col); + } else { + break; + } + } + thread_data->error_info.setjmp = 0; + return !td->xd.corrupted; +} + +static int get_next_job_info(AV1Decoder *const pbi, + AV1DecRowMTJobInfo *next_job_info, + int *end_of_frame) { + AV1_COMMON *cm = &pbi->common; + TileDataDec *tile_data; + AV1DecRowMTSync *dec_row_mt_sync; + AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; + TileInfo tile_info; + const int tile_rows_start = frame_row_mt_info->tile_rows_start; + const int tile_rows_end = frame_row_mt_info->tile_rows_end; + const int tile_cols_start = frame_row_mt_info->tile_cols_start; + const int tile_cols_end = frame_row_mt_info->tile_cols_end; + const int start_tile = frame_row_mt_info->start_tile; + const int end_tile = frame_row_mt_info->end_tile; + const int sb_mi_size = mi_size_wide[cm->seq_params.sb_size]; + int num_mis_to_decode, num_threads_working; + int num_mis_waiting_for_decode; + int min_threads_working = INT_MAX; + int max_mis_to_decode = 0; + int tile_row_idx, tile_col_idx; + int tile_row = 0; + int tile_col = 0; + + memset(next_job_info, 0, sizeof(*next_job_info)); + + // Frame decode is completed or error is encountered. + *end_of_frame = (frame_row_mt_info->mi_rows_decode_started == + frame_row_mt_info->mi_rows_to_decode) || + (frame_row_mt_info->row_mt_exit == 1); + if (*end_of_frame) { + return 1; + } + + // Decoding cannot start as bit-stream parsing is not complete. + if (frame_row_mt_info->mi_rows_parse_done - + frame_row_mt_info->mi_rows_decode_started == + 0) + return 0; + + // Choose the tile to decode. + for (tile_row_idx = tile_rows_start; tile_row_idx < tile_rows_end; + ++tile_row_idx) { + for (tile_col_idx = tile_cols_start; tile_col_idx < tile_cols_end; + ++tile_col_idx) { + if (tile_row_idx * cm->tile_cols + tile_col_idx < start_tile || + tile_row_idx * cm->tile_cols + tile_col_idx > end_tile) + continue; + + tile_data = pbi->tile_data + tile_row_idx * cm->tile_cols + tile_col_idx; + dec_row_mt_sync = &tile_data->dec_row_mt_sync; + + num_threads_working = dec_row_mt_sync->num_threads_working; + num_mis_waiting_for_decode = (dec_row_mt_sync->mi_rows_parse_done - + dec_row_mt_sync->mi_rows_decode_started) * + dec_row_mt_sync->mi_cols; + num_mis_to_decode = + (dec_row_mt_sync->mi_rows - dec_row_mt_sync->mi_rows_decode_started) * + dec_row_mt_sync->mi_cols; + + assert(num_mis_to_decode >= num_mis_waiting_for_decode); + + // Pick the tile which has minimum number of threads working on it. + if (num_mis_waiting_for_decode > 0) { + if (num_threads_working < min_threads_working) { + min_threads_working = num_threads_working; + max_mis_to_decode = 0; + } + if (num_threads_working == min_threads_working && + num_mis_to_decode > max_mis_to_decode) { + max_mis_to_decode = num_mis_to_decode; + tile_row = tile_row_idx; + tile_col = tile_col_idx; + } + } + } + } + + tile_data = pbi->tile_data + tile_row * cm->tile_cols + tile_col; + tile_info = tile_data->tile_info; + dec_row_mt_sync = &tile_data->dec_row_mt_sync; + + next_job_info->tile_row = tile_row; + next_job_info->tile_col = tile_col; + next_job_info->mi_row = + dec_row_mt_sync->mi_rows_decode_started + tile_info.mi_row_start; + + dec_row_mt_sync->num_threads_working++; + dec_row_mt_sync->mi_rows_decode_started += sb_mi_size; + frame_row_mt_info->mi_rows_decode_started += sb_mi_size; + + return 1; +} + +static INLINE void signal_parse_sb_row_done(AV1Decoder *const pbi, + TileDataDec *const tile_data, + const int sb_mi_size) { + AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; +#if CONFIG_MULTITHREAD + pthread_mutex_lock(pbi->row_mt_mutex_); +#endif + tile_data->dec_row_mt_sync.mi_rows_parse_done += sb_mi_size; + frame_row_mt_info->mi_rows_parse_done += sb_mi_size; +#if CONFIG_MULTITHREAD + pthread_cond_broadcast(pbi->row_mt_cond_); + pthread_mutex_unlock(pbi->row_mt_mutex_); +#endif +} + +static int row_mt_worker_hook(void *arg1, void *arg2) { + DecWorkerData *const thread_data = (DecWorkerData *)arg1; + AV1Decoder *const pbi = (AV1Decoder *)arg2; + AV1_COMMON *cm = &pbi->common; + ThreadData *const td = thread_data->td; + uint8_t allow_update_cdf; + const int sb_mi_size = mi_size_wide[cm->seq_params.sb_size]; + AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; + td->xd.corrupted = 0; + + // The jmp_buf is valid only for the duration of the function that calls + // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 + // before it returns. + if (setjmp(thread_data->error_info.jmp)) { + thread_data->error_info.setjmp = 0; + thread_data->td->xd.corrupted = 1; +#if CONFIG_MULTITHREAD + pthread_mutex_lock(pbi->row_mt_mutex_); +#endif + frame_row_mt_info->row_mt_exit = 1; +#if CONFIG_MULTITHREAD + pthread_cond_broadcast(pbi->row_mt_cond_); + pthread_mutex_unlock(pbi->row_mt_mutex_); +#endif + return 0; + } + thread_data->error_info.setjmp = 1; + + const int num_planes = av1_num_planes(cm); + allow_update_cdf = cm->large_scale_tile ? 0 : 1; + allow_update_cdf = allow_update_cdf && !cm->disable_cdf_update; + + assert(cm->tile_cols > 0); + while (1) { + TileJobsDec *cur_job_info = get_dec_job_info(&pbi->tile_mt_info); + + if (cur_job_info != NULL && !td->xd.corrupted) { + const TileBufferDec *const tile_buffer = cur_job_info->tile_buffer; + TileDataDec *const tile_data = cur_job_info->tile_data; + tile_worker_hook_init(pbi, thread_data, tile_buffer, tile_data, + allow_update_cdf); + + set_decode_func_pointers(td, 0x1); + + // decode tile + TileInfo tile_info = tile_data->tile_info; + int tile_row = tile_info.tile_row; + + av1_zero_above_context(cm, &td->xd, tile_info.mi_col_start, + tile_info.mi_col_end, tile_row); + av1_reset_loop_filter_delta(&td->xd, num_planes); + av1_reset_loop_restoration(&td->xd, num_planes); + + for (int mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end; + mi_row += cm->seq_params.mib_size) { + av1_zero_left_context(&td->xd); + + for (int mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end; + mi_col += cm->seq_params.mib_size) { + set_cb_buffer(pbi, &td->xd, pbi->cb_buffer_base, num_planes, mi_row, + mi_col); + + // Bit-stream parsing of the superblock + decode_partition(pbi, td, mi_row, mi_col, td->bit_reader, + cm->seq_params.sb_size, 0x1); + } + signal_parse_sb_row_done(pbi, tile_data, sb_mi_size); + } + + int corrupted = + (check_trailing_bits_after_symbol_coder(td->bit_reader)) ? 1 : 0; + aom_merge_corrupted_flag(&td->xd.corrupted, corrupted); + } else { + break; + } + } + + set_decode_func_pointers(td, 0x2); + + while (1) { + AV1DecRowMTJobInfo next_job_info; + int end_of_frame = 0; + +#if CONFIG_MULTITHREAD + pthread_mutex_lock(pbi->row_mt_mutex_); +#endif + while (!get_next_job_info(pbi, &next_job_info, &end_of_frame)) { +#if CONFIG_MULTITHREAD + pthread_cond_wait(pbi->row_mt_cond_, pbi->row_mt_mutex_); +#endif + } +#if CONFIG_MULTITHREAD + pthread_mutex_unlock(pbi->row_mt_mutex_); +#endif + + if (end_of_frame) break; + + int tile_row = next_job_info.tile_row; + int tile_col = next_job_info.tile_col; + int mi_row = next_job_info.mi_row; + + TileDataDec *tile_data = + pbi->tile_data + tile_row * cm->tile_cols + tile_col; + AV1DecRowMTSync *dec_row_mt_sync = &tile_data->dec_row_mt_sync; + TileInfo tile_info = tile_data->tile_info; + + av1_tile_init(&td->xd.tile, cm, tile_row, tile_col); + av1_init_macroblockd(cm, &td->xd, td->dqcoeff); + td->xd.error_info = &thread_data->error_info; + + decode_tile_sb_row(pbi, td, tile_info, mi_row); + +#if CONFIG_MULTITHREAD + pthread_mutex_lock(pbi->row_mt_mutex_); +#endif + dec_row_mt_sync->num_threads_working--; +#if CONFIG_MULTITHREAD + pthread_mutex_unlock(pbi->row_mt_mutex_); +#endif + } + thread_data->error_info.setjmp = 0; + return !td->xd.corrupted; +} + +// sorts in descending order +static int compare_tile_buffers(const void *a, const void *b) { + const TileJobsDec *const buf1 = (const TileJobsDec *)a; + const TileJobsDec *const buf2 = (const TileJobsDec *)b; + return (((int)buf2->tile_buffer->size) - ((int)buf1->tile_buffer->size)); +} + +static void enqueue_tile_jobs(AV1Decoder *pbi, AV1_COMMON *cm, + int tile_rows_start, int tile_rows_end, + int tile_cols_start, int tile_cols_end, + int startTile, int endTile) { + AV1DecTileMT *tile_mt_info = &pbi->tile_mt_info; + TileJobsDec *tile_job_queue = tile_mt_info->job_queue; + tile_mt_info->jobs_enqueued = 0; + tile_mt_info->jobs_dequeued = 0; + + for (int row = tile_rows_start; row < tile_rows_end; row++) { + for (int col = tile_cols_start; col < tile_cols_end; col++) { + if (row * cm->tile_cols + col < startTile || + row * cm->tile_cols + col > endTile) + continue; + tile_job_queue->tile_buffer = &pbi->tile_buffers[row][col]; + tile_job_queue->tile_data = pbi->tile_data + row * cm->tile_cols + col; + tile_job_queue++; + tile_mt_info->jobs_enqueued++; + } + } +} + +static void alloc_dec_jobs(AV1DecTileMT *tile_mt_info, AV1_COMMON *cm, + int tile_rows, int tile_cols) { + tile_mt_info->alloc_tile_rows = tile_rows; + tile_mt_info->alloc_tile_cols = tile_cols; + int num_tiles = tile_rows * tile_cols; +#if CONFIG_MULTITHREAD + { + CHECK_MEM_ERROR(cm, tile_mt_info->job_mutex, + aom_malloc(sizeof(*tile_mt_info->job_mutex) * num_tiles)); + + for (int i = 0; i < num_tiles; i++) { + pthread_mutex_init(&tile_mt_info->job_mutex[i], NULL); + } + } +#endif + CHECK_MEM_ERROR(cm, tile_mt_info->job_queue, + aom_malloc(sizeof(*tile_mt_info->job_queue) * num_tiles)); +} + +void av1_free_mc_tmp_buf(ThreadData *thread_data) { + int ref; + for (ref = 0; ref < 2; ref++) { + if (thread_data->mc_buf_use_highbd) + aom_free(CONVERT_TO_SHORTPTR(thread_data->mc_buf[ref])); + else + aom_free(thread_data->mc_buf[ref]); + thread_data->mc_buf[ref] = NULL; + } + thread_data->mc_buf_size = 0; + thread_data->mc_buf_use_highbd = 0; + + aom_free(thread_data->tmp_conv_dst); + thread_data->tmp_conv_dst = NULL; + for (int i = 0; i < 2; ++i) { + aom_free(thread_data->tmp_obmc_bufs[i]); + thread_data->tmp_obmc_bufs[i] = NULL; + } +} + +static void allocate_mc_tmp_buf(AV1_COMMON *const cm, ThreadData *thread_data, + int buf_size, int use_highbd) { + for (int ref = 0; ref < 2; ref++) { + if (use_highbd) { + uint16_t *hbd_mc_buf; + CHECK_MEM_ERROR(cm, hbd_mc_buf, (uint16_t *)aom_memalign(16, buf_size)); + thread_data->mc_buf[ref] = CONVERT_TO_BYTEPTR(hbd_mc_buf); + } else { + CHECK_MEM_ERROR(cm, thread_data->mc_buf[ref], + (uint8_t *)aom_memalign(16, buf_size)); + } + } + thread_data->mc_buf_size = buf_size; + thread_data->mc_buf_use_highbd = use_highbd; + + CHECK_MEM_ERROR(cm, thread_data->tmp_conv_dst, + aom_memalign(32, MAX_SB_SIZE * MAX_SB_SIZE * + sizeof(*thread_data->tmp_conv_dst))); + for (int i = 0; i < 2; ++i) { + CHECK_MEM_ERROR( + cm, thread_data->tmp_obmc_bufs[i], + aom_memalign(16, 2 * MAX_MB_PLANE * MAX_SB_SQUARE * + sizeof(*thread_data->tmp_obmc_bufs[i]))); + } +} + +static void reset_dec_workers(AV1Decoder *pbi, AVxWorkerHook worker_hook, + int num_workers) { + const AVxWorkerInterface *const winterface = aom_get_worker_interface(); + + // Reset tile decoding hook + for (int worker_idx = 0; worker_idx < num_workers; ++worker_idx) { + AVxWorker *const worker = &pbi->tile_workers[worker_idx]; + DecWorkerData *const thread_data = pbi->thread_data + worker_idx; + thread_data->td->xd = pbi->mb; + thread_data->td->xd.corrupted = 0; + thread_data->td->xd.mc_buf[0] = thread_data->td->mc_buf[0]; + thread_data->td->xd.mc_buf[1] = thread_data->td->mc_buf[1]; + thread_data->td->xd.tmp_conv_dst = thread_data->td->tmp_conv_dst; + for (int j = 0; j < 2; ++j) { + thread_data->td->xd.tmp_obmc_bufs[j] = thread_data->td->tmp_obmc_bufs[j]; + } + winterface->sync(worker); + + worker->hook = worker_hook; + worker->data1 = thread_data; + worker->data2 = pbi; + } +#if CONFIG_ACCOUNTING + if (pbi->acct_enabled) { + aom_accounting_reset(&pbi->accounting); + } +#endif +} + +static void launch_dec_workers(AV1Decoder *pbi, const uint8_t *data_end, + int num_workers) { + const AVxWorkerInterface *const winterface = aom_get_worker_interface(); + + for (int worker_idx = 0; worker_idx < num_workers; ++worker_idx) { + AVxWorker *const worker = &pbi->tile_workers[worker_idx]; + DecWorkerData *const thread_data = (DecWorkerData *)worker->data1; + + thread_data->data_end = data_end; + + worker->had_error = 0; + if (worker_idx == num_workers - 1) { + winterface->execute(worker); + } else { + winterface->launch(worker); + } + } +} + +static void sync_dec_workers(AV1Decoder *pbi, int num_workers) { + const AVxWorkerInterface *const winterface = aom_get_worker_interface(); + int corrupted = 0; + + for (int worker_idx = num_workers; worker_idx > 0; --worker_idx) { + AVxWorker *const worker = &pbi->tile_workers[worker_idx - 1]; + aom_merge_corrupted_flag(&corrupted, !winterface->sync(worker)); + } + + pbi->mb.corrupted = corrupted; +} + +static void decode_mt_init(AV1Decoder *pbi) { + AV1_COMMON *const cm = &pbi->common; + const AVxWorkerInterface *const winterface = aom_get_worker_interface(); + int worker_idx; + + // Create workers and thread_data + if (pbi->num_workers == 0) { + const int num_threads = pbi->max_threads; + CHECK_MEM_ERROR(cm, pbi->tile_workers, + aom_malloc(num_threads * sizeof(*pbi->tile_workers))); + CHECK_MEM_ERROR(cm, pbi->thread_data, + aom_malloc(num_threads * sizeof(*pbi->thread_data))); + + for (worker_idx = 0; worker_idx < num_threads; ++worker_idx) { + AVxWorker *const worker = &pbi->tile_workers[worker_idx]; + DecWorkerData *const thread_data = pbi->thread_data + worker_idx; + ++pbi->num_workers; + + winterface->init(worker); + if (worker_idx < num_threads - 1 && !winterface->reset(worker)) { + aom_internal_error(&cm->error, AOM_CODEC_ERROR, + "Tile decoder thread creation failed"); + } + + if (worker_idx < num_threads - 1) { + // Allocate thread data. + CHECK_MEM_ERROR(cm, thread_data->td, + aom_memalign(32, sizeof(*thread_data->td))); + av1_zero(*thread_data->td); + } else { + // Main thread acts as a worker and uses the thread data in pbi + thread_data->td = &pbi->td; + } + thread_data->error_info.error_code = AOM_CODEC_OK; + thread_data->error_info.setjmp = 0; + } + } + const int use_highbd = cm->seq_params.use_highbitdepth ? 1 : 0; + const int buf_size = MC_TEMP_BUF_PELS << use_highbd; + for (worker_idx = 0; worker_idx < pbi->max_threads - 1; ++worker_idx) { + DecWorkerData *const thread_data = pbi->thread_data + worker_idx; + if (thread_data->td->mc_buf_size != buf_size) { + av1_free_mc_tmp_buf(thread_data->td); + allocate_mc_tmp_buf(cm, thread_data->td, buf_size, use_highbd); + } + } +} + +static void tile_mt_queue(AV1Decoder *pbi, int tile_cols, int tile_rows, + int tile_rows_start, int tile_rows_end, + int tile_cols_start, int tile_cols_end, + int start_tile, int end_tile) { + AV1_COMMON *const cm = &pbi->common; + if (pbi->tile_mt_info.alloc_tile_cols != tile_cols || + pbi->tile_mt_info.alloc_tile_rows != tile_rows) { + av1_dealloc_dec_jobs(&pbi->tile_mt_info); + alloc_dec_jobs(&pbi->tile_mt_info, cm, tile_rows, tile_cols); + } + enqueue_tile_jobs(pbi, cm, tile_rows_start, tile_rows_end, tile_cols_start, + tile_cols_end, start_tile, end_tile); + qsort(pbi->tile_mt_info.job_queue, pbi->tile_mt_info.jobs_enqueued, + sizeof(pbi->tile_mt_info.job_queue[0]), compare_tile_buffers); +} + +static const uint8_t *decode_tiles_mt(AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, int start_tile, + int end_tile) { + AV1_COMMON *const cm = &pbi->common; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + const int n_tiles = tile_cols * tile_rows; + TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; + const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); + const int single_row = pbi->dec_tile_row >= 0; + const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); + const int single_col = pbi->dec_tile_col >= 0; + int tile_rows_start; + int tile_rows_end; + int tile_cols_start; + int tile_cols_end; + int tile_count_tg; + int num_workers; + const uint8_t *raw_data_end = NULL; + + if (cm->large_scale_tile) { + tile_rows_start = single_row ? dec_tile_row : 0; + tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; + tile_cols_start = single_col ? dec_tile_col : 0; + tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; + } else { + tile_rows_start = 0; + tile_rows_end = tile_rows; + tile_cols_start = 0; + tile_cols_end = tile_cols; + } + tile_count_tg = end_tile - start_tile + 1; + num_workers = AOMMIN(pbi->max_threads, tile_count_tg); + + // No tiles to decode. + if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || + // First tile is larger than end_tile. + tile_rows_start * tile_cols + tile_cols_start > end_tile || + // Last tile is smaller than start_tile. + (tile_rows_end - 1) * tile_cols + tile_cols_end - 1 < start_tile) + return data; + + assert(tile_rows <= MAX_TILE_ROWS); + assert(tile_cols <= MAX_TILE_COLS); + assert(tile_count_tg > 0); + assert(num_workers > 0); + assert(start_tile <= end_tile); + assert(start_tile >= 0 && end_tile < n_tiles); + + decode_mt_init(pbi); + + // get tile size in tile group +#if EXT_TILE_DEBUG + if (cm->large_scale_tile) assert(pbi->ext_tile_debug == 1); + if (cm->large_scale_tile) + raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); + else +#endif // EXT_TILE_DEBUG + get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); + + if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { + decoder_alloc_tile_data(pbi, n_tiles); + } + + for (int row = 0; row < tile_rows; row++) { + for (int col = 0; col < tile_cols; col++) { + TileDataDec *tile_data = pbi->tile_data + row * cm->tile_cols + col; + av1_tile_init(&tile_data->tile_info, cm, row, col); + } + } + + tile_mt_queue(pbi, tile_cols, tile_rows, tile_rows_start, tile_rows_end, + tile_cols_start, tile_cols_end, start_tile, end_tile); + + reset_dec_workers(pbi, tile_worker_hook, num_workers); + launch_dec_workers(pbi, data_end, num_workers); + sync_dec_workers(pbi, num_workers); + + if (pbi->mb.corrupted) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Failed to decode tile data"); + + if (cm->large_scale_tile) { + if (n_tiles == 1) { + // Find the end of the single tile buffer + return aom_reader_find_end(&pbi->tile_data->bit_reader); + } + // Return the end of the last tile buffer + return raw_data_end; + } + TileDataDec *const tile_data = pbi->tile_data + end_tile; + + return aom_reader_find_end(&tile_data->bit_reader); +} + +static void dec_alloc_cb_buf(AV1Decoder *pbi) { + AV1_COMMON *const cm = &pbi->common; + int size = ((cm->mi_rows >> cm->seq_params.mib_size_log2) + 1) * + ((cm->mi_cols >> cm->seq_params.mib_size_log2) + 1); + + if (pbi->cb_buffer_alloc_size < size) { + av1_dec_free_cb_buf(pbi); + CHECK_MEM_ERROR(cm, pbi->cb_buffer_base, + aom_memalign(32, sizeof(*pbi->cb_buffer_base) * size)); + pbi->cb_buffer_alloc_size = size; + } +} + +static void row_mt_frame_init(AV1Decoder *pbi, int tile_rows_start, + int tile_rows_end, int tile_cols_start, + int tile_cols_end, int start_tile, int end_tile, + int max_sb_rows) { + AV1_COMMON *const cm = &pbi->common; + AV1DecRowMTInfo *frame_row_mt_info = &pbi->frame_row_mt_info; + + frame_row_mt_info->tile_rows_start = tile_rows_start; + frame_row_mt_info->tile_rows_end = tile_rows_end; + frame_row_mt_info->tile_cols_start = tile_cols_start; + frame_row_mt_info->tile_cols_end = tile_cols_end; + frame_row_mt_info->start_tile = start_tile; + frame_row_mt_info->end_tile = end_tile; + frame_row_mt_info->mi_rows_to_decode = 0; + frame_row_mt_info->mi_rows_parse_done = 0; + frame_row_mt_info->mi_rows_decode_started = 0; + frame_row_mt_info->row_mt_exit = 0; + + for (int tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { + for (int tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) { + if (tile_row * cm->tile_cols + tile_col < start_tile || + tile_row * cm->tile_cols + tile_col > end_tile) + continue; + + TileDataDec *const tile_data = + pbi->tile_data + tile_row * cm->tile_cols + tile_col; + TileInfo tile_info = tile_data->tile_info; + + tile_data->dec_row_mt_sync.mi_rows_parse_done = 0; + tile_data->dec_row_mt_sync.mi_rows_decode_started = 0; + tile_data->dec_row_mt_sync.num_threads_working = 0; + tile_data->dec_row_mt_sync.mi_rows = + ALIGN_POWER_OF_TWO(tile_info.mi_row_end - tile_info.mi_row_start, + cm->seq_params.mib_size_log2); + tile_data->dec_row_mt_sync.mi_cols = + ALIGN_POWER_OF_TWO(tile_info.mi_col_end - tile_info.mi_col_start, + cm->seq_params.mib_size_log2); + + frame_row_mt_info->mi_rows_to_decode += + tile_data->dec_row_mt_sync.mi_rows; + + // Initialize cur_sb_col to -1 for all SB rows. + memset(tile_data->dec_row_mt_sync.cur_sb_col, -1, + sizeof(*tile_data->dec_row_mt_sync.cur_sb_col) * max_sb_rows); + } + } + +#if CONFIG_MULTITHREAD + if (pbi->row_mt_mutex_ == NULL) { + CHECK_MEM_ERROR(cm, pbi->row_mt_mutex_, + aom_malloc(sizeof(*(pbi->row_mt_mutex_)))); + if (pbi->row_mt_mutex_) { + pthread_mutex_init(pbi->row_mt_mutex_, NULL); + } + } + + if (pbi->row_mt_cond_ == NULL) { + CHECK_MEM_ERROR(cm, pbi->row_mt_cond_, + aom_malloc(sizeof(*(pbi->row_mt_cond_)))); + if (pbi->row_mt_cond_) { + pthread_cond_init(pbi->row_mt_cond_, NULL); + } + } +#endif +} + +static const uint8_t *decode_tiles_row_mt(AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + int start_tile, int end_tile) { + AV1_COMMON *const cm = &pbi->common; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + const int n_tiles = tile_cols * tile_rows; + TileBufferDec(*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; + const int dec_tile_row = AOMMIN(pbi->dec_tile_row, tile_rows); + const int single_row = pbi->dec_tile_row >= 0; + const int dec_tile_col = AOMMIN(pbi->dec_tile_col, tile_cols); + const int single_col = pbi->dec_tile_col >= 0; + int tile_rows_start; + int tile_rows_end; + int tile_cols_start; + int tile_cols_end; + int tile_count_tg; + int num_workers; + const uint8_t *raw_data_end = NULL; + int max_sb_rows = 0; + + if (cm->large_scale_tile) { + tile_rows_start = single_row ? dec_tile_row : 0; + tile_rows_end = single_row ? dec_tile_row + 1 : tile_rows; + tile_cols_start = single_col ? dec_tile_col : 0; + tile_cols_end = single_col ? tile_cols_start + 1 : tile_cols; + } else { + tile_rows_start = 0; + tile_rows_end = tile_rows; + tile_cols_start = 0; + tile_cols_end = tile_cols; + } + tile_count_tg = end_tile - start_tile + 1; + num_workers = pbi->max_threads; + + // No tiles to decode. + if (tile_rows_end <= tile_rows_start || tile_cols_end <= tile_cols_start || + // First tile is larger than end_tile. + tile_rows_start * tile_cols + tile_cols_start > end_tile || + // Last tile is smaller than start_tile. + (tile_rows_end - 1) * tile_cols + tile_cols_end - 1 < start_tile) + return data; + + assert(tile_rows <= MAX_TILE_ROWS); + assert(tile_cols <= MAX_TILE_COLS); + assert(tile_count_tg > 0); + assert(num_workers > 0); + assert(start_tile <= end_tile); + assert(start_tile >= 0 && end_tile < n_tiles); + + (void)tile_count_tg; + + decode_mt_init(pbi); + + // get tile size in tile group +#if EXT_TILE_DEBUG + if (cm->large_scale_tile) assert(pbi->ext_tile_debug == 1); + if (cm->large_scale_tile) + raw_data_end = get_ls_tile_buffers(pbi, data, data_end, tile_buffers); + else +#endif // EXT_TILE_DEBUG + get_tile_buffers(pbi, data, data_end, tile_buffers, start_tile, end_tile); + + if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { + for (int i = 0; i < pbi->allocated_tiles; i++) { + TileDataDec *const tile_data = pbi->tile_data + i; + av1_dec_row_mt_dealloc(&tile_data->dec_row_mt_sync); + } + decoder_alloc_tile_data(pbi, n_tiles); + } + + for (int row = 0; row < tile_rows; row++) { + for (int col = 0; col < tile_cols; col++) { + TileDataDec *tile_data = pbi->tile_data + row * cm->tile_cols + col; + av1_tile_init(&tile_data->tile_info, cm, row, col); + + max_sb_rows = AOMMAX(max_sb_rows, + av1_get_sb_rows_in_tile(cm, tile_data->tile_info)); + } + } + + if (pbi->allocated_row_mt_sync_rows != max_sb_rows) { + for (int i = 0; i < n_tiles; ++i) { + TileDataDec *const tile_data = pbi->tile_data + i; + av1_dec_row_mt_dealloc(&tile_data->dec_row_mt_sync); + dec_row_mt_alloc(&tile_data->dec_row_mt_sync, cm, max_sb_rows); + } + pbi->allocated_row_mt_sync_rows = max_sb_rows; + } + + tile_mt_queue(pbi, tile_cols, tile_rows, tile_rows_start, tile_rows_end, + tile_cols_start, tile_cols_end, start_tile, end_tile); + + dec_alloc_cb_buf(pbi); + + row_mt_frame_init(pbi, tile_rows_start, tile_rows_end, tile_cols_start, + tile_cols_end, start_tile, end_tile, max_sb_rows); + + reset_dec_workers(pbi, row_mt_worker_hook, num_workers); + launch_dec_workers(pbi, data_end, num_workers); + sync_dec_workers(pbi, num_workers); + + if (pbi->mb.corrupted) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Failed to decode tile data"); + + if (cm->large_scale_tile) { + if (n_tiles == 1) { + // Find the end of the single tile buffer + return aom_reader_find_end(&pbi->tile_data->bit_reader); + } + // Return the end of the last tile buffer + return raw_data_end; + } + TileDataDec *const tile_data = pbi->tile_data + end_tile; + + return aom_reader_find_end(&tile_data->bit_reader); +} + +static void error_handler(void *data) { + AV1_COMMON *const cm = (AV1_COMMON *)data; + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, "Truncated packet"); +} + +// Reads the high_bitdepth and twelve_bit fields in color_config() and sets +// seq_params->bit_depth based on the values of those fields and +// seq_params->profile. Reports errors by calling rb->error_handler() or +// aom_internal_error(). +static void read_bitdepth(struct aom_read_bit_buffer *rb, + SequenceHeader *seq_params, + struct aom_internal_error_info *error_info) { + const int high_bitdepth = aom_rb_read_bit(rb); + if (seq_params->profile == PROFILE_2 && high_bitdepth) { + const int twelve_bit = aom_rb_read_bit(rb); + seq_params->bit_depth = twelve_bit ? AOM_BITS_12 : AOM_BITS_10; + } else if (seq_params->profile <= PROFILE_2) { + seq_params->bit_depth = high_bitdepth ? AOM_BITS_10 : AOM_BITS_8; + } else { + aom_internal_error(error_info, AOM_CODEC_UNSUP_BITSTREAM, + "Unsupported profile/bit-depth combination"); + } +} + +void av1_read_film_grain_params(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb) { + aom_film_grain_t *pars = &cm->film_grain_params; + const SequenceHeader *const seq_params = &cm->seq_params; + + pars->apply_grain = aom_rb_read_bit(rb); + if (!pars->apply_grain) { + memset(pars, 0, sizeof(*pars)); + return; + } + + pars->random_seed = aom_rb_read_literal(rb, 16); + if (cm->frame_type == INTER_FRAME) + pars->update_parameters = aom_rb_read_bit(rb); + else + pars->update_parameters = 1; + + pars->bit_depth = seq_params->bit_depth; + + if (!pars->update_parameters) { + // inherit parameters from a previous reference frame + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + int film_grain_params_ref_idx = aom_rb_read_literal(rb, 3); + int buf_idx = cm->ref_frame_map[film_grain_params_ref_idx]; + if (buf_idx == INVALID_IDX) { + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Invalid Film grain reference idx"); + } + if (!frame_bufs[buf_idx].film_grain_params_present) { + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Film grain reference parameters not available"); + } + uint16_t random_seed = pars->random_seed; + *pars = frame_bufs[buf_idx].film_grain_params; // inherit paramaters + pars->random_seed = random_seed; // with new random seed + return; + } + + // Scaling functions parameters + pars->num_y_points = aom_rb_read_literal(rb, 4); // max 14 + if (pars->num_y_points > 14) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Number of points for film grain luma scaling function " + "exceeds the maximum value."); + for (int i = 0; i < pars->num_y_points; i++) { + pars->scaling_points_y[i][0] = aom_rb_read_literal(rb, 8); + if (i && pars->scaling_points_y[i - 1][0] >= pars->scaling_points_y[i][0]) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "First coordinate of the scaling function points " + "shall be increasing."); + pars->scaling_points_y[i][1] = aom_rb_read_literal(rb, 8); + } + + if (!seq_params->monochrome) + pars->chroma_scaling_from_luma = aom_rb_read_bit(rb); + else + pars->chroma_scaling_from_luma = 0; + + if (seq_params->monochrome || pars->chroma_scaling_from_luma || + ((seq_params->subsampling_x == 1) && (seq_params->subsampling_y == 1) && + (pars->num_y_points == 0))) { + pars->num_cb_points = 0; + pars->num_cr_points = 0; + } else { + pars->num_cb_points = aom_rb_read_literal(rb, 4); // max 10 + if (pars->num_cb_points > 10) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Number of points for film grain cb scaling function " + "exceeds the maximum value."); + for (int i = 0; i < pars->num_cb_points; i++) { + pars->scaling_points_cb[i][0] = aom_rb_read_literal(rb, 8); + if (i && + pars->scaling_points_cb[i - 1][0] >= pars->scaling_points_cb[i][0]) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "First coordinate of the scaling function points " + "shall be increasing."); + pars->scaling_points_cb[i][1] = aom_rb_read_literal(rb, 8); + } + + pars->num_cr_points = aom_rb_read_literal(rb, 4); // max 10 + if (pars->num_cr_points > 10) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Number of points for film grain cr scaling function " + "exceeds the maximum value."); + for (int i = 0; i < pars->num_cr_points; i++) { + pars->scaling_points_cr[i][0] = aom_rb_read_literal(rb, 8); + if (i && + pars->scaling_points_cr[i - 1][0] >= pars->scaling_points_cr[i][0]) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "First coordinate of the scaling function points " + "shall be increasing."); + pars->scaling_points_cr[i][1] = aom_rb_read_literal(rb, 8); + } + + if ((seq_params->subsampling_x == 1) && (seq_params->subsampling_y == 1) && + (((pars->num_cb_points == 0) && (pars->num_cr_points != 0)) || + ((pars->num_cb_points != 0) && (pars->num_cr_points == 0)))) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "In YCbCr 4:2:0, film grain shall be applied " + "to both chroma components or neither."); + } + + pars->scaling_shift = aom_rb_read_literal(rb, 2) + 8; // 8 + value + + // AR coefficients + // Only sent if the corresponsing scaling function has + // more than 0 points + + pars->ar_coeff_lag = aom_rb_read_literal(rb, 2); + + int num_pos_luma = 2 * pars->ar_coeff_lag * (pars->ar_coeff_lag + 1); + int num_pos_chroma = num_pos_luma; + if (pars->num_y_points > 0) ++num_pos_chroma; + + if (pars->num_y_points) + for (int i = 0; i < num_pos_luma; i++) + pars->ar_coeffs_y[i] = aom_rb_read_literal(rb, 8) - 128; + + if (pars->num_cb_points || pars->chroma_scaling_from_luma) + for (int i = 0; i < num_pos_chroma; i++) + pars->ar_coeffs_cb[i] = aom_rb_read_literal(rb, 8) - 128; + + if (pars->num_cr_points || pars->chroma_scaling_from_luma) + for (int i = 0; i < num_pos_chroma; i++) + pars->ar_coeffs_cr[i] = aom_rb_read_literal(rb, 8) - 128; + + pars->ar_coeff_shift = aom_rb_read_literal(rb, 2) + 6; // 6 + value + + pars->grain_scale_shift = aom_rb_read_literal(rb, 2); + + if (pars->num_cb_points) { + pars->cb_mult = aom_rb_read_literal(rb, 8); + pars->cb_luma_mult = aom_rb_read_literal(rb, 8); + pars->cb_offset = aom_rb_read_literal(rb, 9); + } + + if (pars->num_cr_points) { + pars->cr_mult = aom_rb_read_literal(rb, 8); + pars->cr_luma_mult = aom_rb_read_literal(rb, 8); + pars->cr_offset = aom_rb_read_literal(rb, 9); + } + + pars->overlap_flag = aom_rb_read_bit(rb); + + pars->clip_to_restricted_range = aom_rb_read_bit(rb); +} + +static void read_film_grain(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) { + if (cm->seq_params.film_grain_params_present && + (cm->show_frame || cm->showable_frame)) { + av1_read_film_grain_params(cm, rb); + } else { + memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params)); + } + cm->film_grain_params.bit_depth = cm->seq_params.bit_depth; + memcpy(&cm->cur_frame->film_grain_params, &cm->film_grain_params, + sizeof(aom_film_grain_t)); +} + +void av1_read_color_config(struct aom_read_bit_buffer *rb, + int allow_lowbitdepth, SequenceHeader *seq_params, + struct aom_internal_error_info *error_info) { + read_bitdepth(rb, seq_params, error_info); + + seq_params->use_highbitdepth = + seq_params->bit_depth > AOM_BITS_8 || !allow_lowbitdepth; + // monochrome bit (not needed for PROFILE_1) + const int is_monochrome = + seq_params->profile != PROFILE_1 ? aom_rb_read_bit(rb) : 0; + seq_params->monochrome = is_monochrome; + int color_description_present_flag = aom_rb_read_bit(rb); + if (color_description_present_flag) { + seq_params->color_primaries = aom_rb_read_literal(rb, 8); + seq_params->transfer_characteristics = aom_rb_read_literal(rb, 8); + seq_params->matrix_coefficients = aom_rb_read_literal(rb, 8); + } else { + seq_params->color_primaries = AOM_CICP_CP_UNSPECIFIED; + seq_params->transfer_characteristics = AOM_CICP_TC_UNSPECIFIED; + seq_params->matrix_coefficients = AOM_CICP_MC_UNSPECIFIED; + } + if (is_monochrome) { + // [16,235] (including xvycc) vs [0,255] range + seq_params->color_range = aom_rb_read_bit(rb); + seq_params->subsampling_y = seq_params->subsampling_x = 1; + seq_params->chroma_sample_position = AOM_CSP_UNKNOWN; + seq_params->separate_uv_delta_q = 0; + return; + } + if (seq_params->color_primaries == AOM_CICP_CP_BT_709 && + seq_params->transfer_characteristics == AOM_CICP_TC_SRGB && + seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY) { + // It would be good to remove this dependency. + seq_params->subsampling_y = seq_params->subsampling_x = 0; + seq_params->color_range = 1; // assume full color-range + if (!(seq_params->profile == PROFILE_1 || + (seq_params->profile == PROFILE_2 && + seq_params->bit_depth == AOM_BITS_12))) { + aom_internal_error( + error_info, AOM_CODEC_UNSUP_BITSTREAM, + "sRGB colorspace not compatible with specified profile"); + } + } else { + // [16,235] (including xvycc) vs [0,255] range + seq_params->color_range = aom_rb_read_bit(rb); + if (seq_params->profile == PROFILE_0) { + // 420 only + seq_params->subsampling_x = seq_params->subsampling_y = 1; + } else if (seq_params->profile == PROFILE_1) { + // 444 only + seq_params->subsampling_x = seq_params->subsampling_y = 0; + } else { + assert(seq_params->profile == PROFILE_2); + if (seq_params->bit_depth == AOM_BITS_12) { + seq_params->subsampling_x = aom_rb_read_bit(rb); + if (seq_params->subsampling_x) + seq_params->subsampling_y = aom_rb_read_bit(rb); // 422 or 420 + else + seq_params->subsampling_y = 0; // 444 + } else { + // 422 + seq_params->subsampling_x = 1; + seq_params->subsampling_y = 0; + } + } + if (seq_params->matrix_coefficients == AOM_CICP_MC_IDENTITY && + (seq_params->subsampling_x || seq_params->subsampling_y)) { + aom_internal_error( + error_info, AOM_CODEC_UNSUP_BITSTREAM, + "Identity CICP Matrix incompatible with non 4:4:4 color sampling"); + } + if (seq_params->subsampling_x && seq_params->subsampling_y) { + seq_params->chroma_sample_position = aom_rb_read_literal(rb, 2); + } + } + seq_params->separate_uv_delta_q = aom_rb_read_bit(rb); +} + +void av1_read_timing_info_header(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb) { + cm->timing_info.num_units_in_display_tick = aom_rb_read_unsigned_literal( + rb, 32); // Number of units in a display tick + cm->timing_info.time_scale = + aom_rb_read_unsigned_literal(rb, 32); // Time scale + if (cm->timing_info.num_units_in_display_tick == 0 || + cm->timing_info.time_scale == 0) { + aom_internal_error( + &cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "num_units_in_display_tick and time_scale must be greater than 0."); + } + cm->timing_info.equal_picture_interval = + aom_rb_read_bit(rb); // Equal picture interval bit + if (cm->timing_info.equal_picture_interval) { + cm->timing_info.num_ticks_per_picture = + aom_rb_read_uvlc(rb) + 1; // ticks per picture + if (cm->timing_info.num_ticks_per_picture == 0) { + aom_internal_error( + &cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "num_ticks_per_picture_minus_1 cannot be (1 << 32) − 1."); + } + } +} + +void av1_read_decoder_model_info(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb) { + cm->buffer_model.encoder_decoder_buffer_delay_length = + aom_rb_read_literal(rb, 5) + 1; + cm->buffer_model.num_units_in_decoding_tick = aom_rb_read_unsigned_literal( + rb, 32); // Number of units in a decoding tick + cm->buffer_model.buffer_removal_time_length = aom_rb_read_literal(rb, 5) + 1; + cm->buffer_model.frame_presentation_time_length = + aom_rb_read_literal(rb, 5) + 1; +} + +void av1_read_op_parameters_info(AV1_COMMON *const cm, + struct aom_read_bit_buffer *rb, int op_num) { + // The cm->op_params array has MAX_NUM_OPERATING_POINTS + 1 elements. + if (op_num > MAX_NUM_OPERATING_POINTS) { + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "AV1 does not support %d decoder model operating points", + op_num + 1); + } + + cm->op_params[op_num].decoder_buffer_delay = aom_rb_read_unsigned_literal( + rb, cm->buffer_model.encoder_decoder_buffer_delay_length); + + cm->op_params[op_num].encoder_buffer_delay = aom_rb_read_unsigned_literal( + rb, cm->buffer_model.encoder_decoder_buffer_delay_length); + + cm->op_params[op_num].low_delay_mode_flag = aom_rb_read_bit(rb); +} + +static void av1_read_temporal_point_info(AV1_COMMON *const cm, + struct aom_read_bit_buffer *rb) { + cm->frame_presentation_time = aom_rb_read_unsigned_literal( + rb, cm->buffer_model.frame_presentation_time_length); +} + +void av1_read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb, + SequenceHeader *seq_params) { + const int num_bits_width = aom_rb_read_literal(rb, 4) + 1; + const int num_bits_height = aom_rb_read_literal(rb, 4) + 1; + const int max_frame_width = aom_rb_read_literal(rb, num_bits_width) + 1; + const int max_frame_height = aom_rb_read_literal(rb, num_bits_height) + 1; + + seq_params->num_bits_width = num_bits_width; + seq_params->num_bits_height = num_bits_height; + seq_params->max_frame_width = max_frame_width; + seq_params->max_frame_height = max_frame_height; + + if (seq_params->reduced_still_picture_hdr) { + seq_params->frame_id_numbers_present_flag = 0; + } else { + seq_params->frame_id_numbers_present_flag = aom_rb_read_bit(rb); + } + if (seq_params->frame_id_numbers_present_flag) { + // We must always have delta_frame_id_length < frame_id_length, + // in order for a frame to be referenced with a unique delta. + // Avoid wasting bits by using a coding that enforces this restriction. + seq_params->delta_frame_id_length = aom_rb_read_literal(rb, 4) + 2; + seq_params->frame_id_length = + aom_rb_read_literal(rb, 3) + seq_params->delta_frame_id_length + 1; + if (seq_params->frame_id_length > 16) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Invalid frame_id_length"); + } + + setup_sb_size(seq_params, rb); + + seq_params->enable_filter_intra = aom_rb_read_bit(rb); + seq_params->enable_intra_edge_filter = aom_rb_read_bit(rb); + + if (seq_params->reduced_still_picture_hdr) { + seq_params->enable_interintra_compound = 0; + seq_params->enable_masked_compound = 0; + seq_params->enable_warped_motion = 0; + seq_params->enable_dual_filter = 0; + seq_params->enable_order_hint = 0; + seq_params->enable_jnt_comp = 0; + seq_params->enable_ref_frame_mvs = 0; + seq_params->force_screen_content_tools = 2; // SELECT_SCREEN_CONTENT_TOOLS + seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV + seq_params->order_hint_bits_minus_1 = -1; + } else { + seq_params->enable_interintra_compound = aom_rb_read_bit(rb); + seq_params->enable_masked_compound = aom_rb_read_bit(rb); + seq_params->enable_warped_motion = aom_rb_read_bit(rb); + seq_params->enable_dual_filter = aom_rb_read_bit(rb); + + seq_params->enable_order_hint = aom_rb_read_bit(rb); + seq_params->enable_jnt_comp = + seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0; + seq_params->enable_ref_frame_mvs = + seq_params->enable_order_hint ? aom_rb_read_bit(rb) : 0; + + if (aom_rb_read_bit(rb)) { + seq_params->force_screen_content_tools = + 2; // SELECT_SCREEN_CONTENT_TOOLS + } else { + seq_params->force_screen_content_tools = aom_rb_read_bit(rb); + } + + if (seq_params->force_screen_content_tools > 0) { + if (aom_rb_read_bit(rb)) { + seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV + } else { + seq_params->force_integer_mv = aom_rb_read_bit(rb); + } + } else { + seq_params->force_integer_mv = 2; // SELECT_INTEGER_MV + } + seq_params->order_hint_bits_minus_1 = + seq_params->enable_order_hint ? aom_rb_read_literal(rb, 3) : -1; + } + + seq_params->enable_superres = aom_rb_read_bit(rb); + seq_params->enable_cdef = aom_rb_read_bit(rb); + seq_params->enable_restoration = aom_rb_read_bit(rb); +} + +static int read_global_motion_params(WarpedMotionParams *params, + const WarpedMotionParams *ref_params, + struct aom_read_bit_buffer *rb, + int allow_hp) { + TransformationType type = aom_rb_read_bit(rb); + if (type != IDENTITY) { + if (aom_rb_read_bit(rb)) + type = ROTZOOM; + else + type = aom_rb_read_bit(rb) ? TRANSLATION : AFFINE; + } + + *params = default_warp_params; + params->wmtype = type; + + if (type >= ROTZOOM) { + params->wmmat[2] = aom_rb_read_signed_primitive_refsubexpfin( + rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, + (ref_params->wmmat[2] >> GM_ALPHA_PREC_DIFF) - + (1 << GM_ALPHA_PREC_BITS)) * + GM_ALPHA_DECODE_FACTOR + + (1 << WARPEDMODEL_PREC_BITS); + params->wmmat[3] = aom_rb_read_signed_primitive_refsubexpfin( + rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, + (ref_params->wmmat[3] >> GM_ALPHA_PREC_DIFF)) * + GM_ALPHA_DECODE_FACTOR; + } + + if (type >= AFFINE) { + params->wmmat[4] = aom_rb_read_signed_primitive_refsubexpfin( + rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, + (ref_params->wmmat[4] >> GM_ALPHA_PREC_DIFF)) * + GM_ALPHA_DECODE_FACTOR; + params->wmmat[5] = aom_rb_read_signed_primitive_refsubexpfin( + rb, GM_ALPHA_MAX + 1, SUBEXPFIN_K, + (ref_params->wmmat[5] >> GM_ALPHA_PREC_DIFF) - + (1 << GM_ALPHA_PREC_BITS)) * + GM_ALPHA_DECODE_FACTOR + + (1 << WARPEDMODEL_PREC_BITS); + } else { + params->wmmat[4] = -params->wmmat[3]; + params->wmmat[5] = params->wmmat[2]; + } + + if (type >= TRANSLATION) { + const int trans_bits = (type == TRANSLATION) + ? GM_ABS_TRANS_ONLY_BITS - !allow_hp + : GM_ABS_TRANS_BITS; + const int trans_dec_factor = + (type == TRANSLATION) ? GM_TRANS_ONLY_DECODE_FACTOR * (1 << !allow_hp) + : GM_TRANS_DECODE_FACTOR; + const int trans_prec_diff = (type == TRANSLATION) + ? GM_TRANS_ONLY_PREC_DIFF + !allow_hp + : GM_TRANS_PREC_DIFF; + params->wmmat[0] = aom_rb_read_signed_primitive_refsubexpfin( + rb, (1 << trans_bits) + 1, SUBEXPFIN_K, + (ref_params->wmmat[0] >> trans_prec_diff)) * + trans_dec_factor; + params->wmmat[1] = aom_rb_read_signed_primitive_refsubexpfin( + rb, (1 << trans_bits) + 1, SUBEXPFIN_K, + (ref_params->wmmat[1] >> trans_prec_diff)) * + trans_dec_factor; + } + + if (params->wmtype <= AFFINE) { + int good_shear_params = get_shear_params(params); + if (!good_shear_params) return 0; + } + + return 1; +} + +static void read_global_motion(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) { + for (int frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) { + const WarpedMotionParams *ref_params = + cm->prev_frame ? &cm->prev_frame->global_motion[frame] + : &default_warp_params; + int good_params = read_global_motion_params( + &cm->global_motion[frame], ref_params, rb, cm->allow_high_precision_mv); + if (!good_params) { +#if WARPED_MOTION_DEBUG + printf("Warning: unexpected global motion shear params from aomenc\n"); +#endif + cm->global_motion[frame].invalid = 1; + } + + // TODO(sarahparker, debargha): The logic in the commented out code below + // does not work currently and causes mismatches when resize is on. Fix it + // before turning the optimization back on. + /* + YV12_BUFFER_CONFIG *ref_buf = get_ref_frame(cm, frame); + if (cm->width == ref_buf->y_crop_width && + cm->height == ref_buf->y_crop_height) { + read_global_motion_params(&cm->global_motion[frame], + &cm->prev_frame->global_motion[frame], rb, + cm->allow_high_precision_mv); + } else { + cm->global_motion[frame] = default_warp_params; + } + */ + /* + printf("Dec Ref %d [%d/%d]: %d %d %d %d\n", + frame, cm->current_video_frame, cm->show_frame, + cm->global_motion[frame].wmmat[0], + cm->global_motion[frame].wmmat[1], + cm->global_motion[frame].wmmat[2], + cm->global_motion[frame].wmmat[3]); + */ + } + memcpy(cm->cur_frame->global_motion, cm->global_motion, + REF_FRAMES * sizeof(WarpedMotionParams)); +} + +static void show_existing_frame_reset(AV1Decoder *const pbi, + int existing_frame_idx) { + AV1_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = pool->frame_bufs; + + assert(cm->show_existing_frame); + + cm->frame_type = KEY_FRAME; + + pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1; + + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + cm->frame_refs[i].idx = INVALID_IDX; + cm->frame_refs[i].buf = NULL; + } + + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + + cm->cur_frame->intra_only = 1; + + if (cm->seq_params.frame_id_numbers_present_flag) { + /* If bitmask is set, update reference frame id values and + mark frames as valid for reference. + Note that the displayed frame be valid for referencing + in order to have been selected. + */ + int refresh_frame_flags = pbi->refresh_frame_flags; + int display_frame_id = cm->ref_frame_id[existing_frame_idx]; + for (int i = 0; i < REF_FRAMES; i++) { + if ((refresh_frame_flags >> i) & 1) { + cm->ref_frame_id[i] = display_frame_id; + cm->valid_for_referencing[i] = 1; + } + } + } + + cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; + + // Generate next_ref_frame_map. + lock_buffer_pool(pool); + int ref_index = 0; + for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + if (mask & 1) { + cm->next_ref_frame_map[ref_index] = cm->new_fb_idx; + ++frame_bufs[cm->new_fb_idx].ref_count; + } else { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + } + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + ++ref_index; + } + + for (; ref_index < REF_FRAMES; ++ref_index) { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + } + unlock_buffer_pool(pool); + pbi->hold_ref_buf = 1; + + // Reload the adapted CDFs from when we originally coded this keyframe + *cm->fc = cm->frame_contexts[existing_frame_idx]; +} + +static INLINE void reset_frame_buffers(AV1_COMMON *cm) { + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + int i; + + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map)); + + lock_buffer_pool(cm->buffer_pool); + for (i = 0; i < FRAME_BUFFERS; ++i) { + if (i != cm->new_fb_idx) { + frame_bufs[i].ref_count = 0; + cm->buffer_pool->release_fb_cb(cm->buffer_pool->cb_priv, + &frame_bufs[i].raw_frame_buffer); + } else { + assert(frame_bufs[i].ref_count == 1); + } + frame_bufs[i].cur_frame_offset = 0; + av1_zero(frame_bufs[i].ref_frame_offset); + } + av1_zero_unused_internal_frame_buffers(&cm->buffer_pool->int_frame_buffers); + unlock_buffer_pool(cm->buffer_pool); +} + +// On success, returns 0. On failure, calls aom_internal_error and does not +// return. +static int read_uncompressed_header(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb) { + AV1_COMMON *const cm = &pbi->common; + const SequenceHeader *const seq_params = &cm->seq_params; + MACROBLOCKD *const xd = &pbi->mb; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = pool->frame_bufs; + + if (!pbi->sequence_header_ready) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "No sequence header"); + } + + cm->last_frame_type = cm->frame_type; + cm->last_intra_only = cm->intra_only; + + // NOTE: By default all coded frames to be used as a reference + cm->is_reference_frame = 1; + + if (seq_params->reduced_still_picture_hdr) { + cm->show_existing_frame = 0; + cm->show_frame = 1; + cm->frame_type = KEY_FRAME; + cm->error_resilient_mode = 1; + } else { + cm->show_existing_frame = aom_rb_read_bit(rb); + cm->reset_decoder_state = 0; + + if (cm->show_existing_frame) { + if (pbi->sequence_header_changed) { + aom_internal_error( + &cm->error, AOM_CODEC_CORRUPT_FRAME, + "New sequence header starts with a show_existing_frame."); + } + // Show an existing frame directly. + const int existing_frame_idx = aom_rb_read_literal(rb, 3); + const int frame_to_show = cm->ref_frame_map[existing_frame_idx]; + if (seq_params->decoder_model_info_present_flag && + cm->timing_info.equal_picture_interval == 0) { + av1_read_temporal_point_info(cm, rb); + } + if (seq_params->frame_id_numbers_present_flag) { + int frame_id_length = seq_params->frame_id_length; + int display_frame_id = aom_rb_read_literal(rb, frame_id_length); + /* Compare display_frame_id with ref_frame_id and check valid for + * referencing */ + if (display_frame_id != cm->ref_frame_id[existing_frame_idx] || + cm->valid_for_referencing[existing_frame_idx] == 0) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Reference buffer frame ID mismatch"); + } + lock_buffer_pool(pool); + if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) { + unlock_buffer_pool(pool); + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Buffer %d does not contain a decoded frame", + frame_to_show); + } + ref_cnt_fb(frame_bufs, &cm->new_fb_idx, frame_to_show); + cm->reset_decoder_state = + frame_bufs[frame_to_show].frame_type == KEY_FRAME; + unlock_buffer_pool(pool); + + cm->lf.filter_level[0] = 0; + cm->lf.filter_level[1] = 0; + cm->show_frame = 1; + + if (!frame_bufs[frame_to_show].showable_frame) { + aom_merge_corrupted_flag(&xd->corrupted, 1); + } + if (cm->reset_decoder_state) frame_bufs[frame_to_show].showable_frame = 0; + + cm->film_grain_params = frame_bufs[frame_to_show].film_grain_params; + + if (cm->reset_decoder_state) { + show_existing_frame_reset(pbi, existing_frame_idx); + } else { + pbi->refresh_frame_flags = 0; + } + + return 0; + } + + cm->frame_type = (FRAME_TYPE)aom_rb_read_literal(rb, 2); // 2 bits + if (pbi->sequence_header_changed) { + if (pbi->common.frame_type == KEY_FRAME) { + // This is the start of a new coded video sequence. + pbi->sequence_header_changed = 0; + pbi->decoding_first_frame = 1; + reset_frame_buffers(&pbi->common); + } else { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Sequence header has changed without a keyframe."); + } + } + + cm->show_frame = aom_rb_read_bit(rb); + if (seq_params->still_picture && + (cm->frame_type != KEY_FRAME || !cm->show_frame)) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Still pictures must be coded as shown keyframes"); + } + cm->showable_frame = cm->frame_type != KEY_FRAME; + if (cm->show_frame) { + if (seq_params->decoder_model_info_present_flag && + cm->timing_info.equal_picture_interval == 0) + av1_read_temporal_point_info(cm, rb); + } else { + // See if this frame can be used as show_existing_frame in future + cm->showable_frame = aom_rb_read_bit(rb); + } + cm->cur_frame->showable_frame = cm->showable_frame; + cm->intra_only = cm->frame_type == INTRA_ONLY_FRAME; + cm->error_resilient_mode = + frame_is_sframe(cm) || (cm->frame_type == KEY_FRAME && cm->show_frame) + ? 1 + : aom_rb_read_bit(rb); + } + + cm->disable_cdf_update = aom_rb_read_bit(rb); + if (seq_params->force_screen_content_tools == 2) { + cm->allow_screen_content_tools = aom_rb_read_bit(rb); + } else { + cm->allow_screen_content_tools = seq_params->force_screen_content_tools; + } + + if (cm->allow_screen_content_tools) { + if (seq_params->force_integer_mv == 2) { + cm->cur_frame_force_integer_mv = aom_rb_read_bit(rb); + } else { + cm->cur_frame_force_integer_mv = seq_params->force_integer_mv; + } + } else { + cm->cur_frame_force_integer_mv = 0; + } + + cm->frame_refs_short_signaling = 0; + int frame_size_override_flag = 0; + cm->allow_intrabc = 0; + cm->primary_ref_frame = PRIMARY_REF_NONE; + + if (!seq_params->reduced_still_picture_hdr) { + if (seq_params->frame_id_numbers_present_flag) { + int frame_id_length = seq_params->frame_id_length; + int diff_len = seq_params->delta_frame_id_length; + int prev_frame_id = 0; + int have_prev_frame_id = !pbi->decoding_first_frame && + !(cm->frame_type == KEY_FRAME && cm->show_frame); + if (have_prev_frame_id) { + prev_frame_id = cm->current_frame_id; + } + cm->current_frame_id = aom_rb_read_literal(rb, frame_id_length); + + if (have_prev_frame_id) { + int diff_frame_id; + if (cm->current_frame_id > prev_frame_id) { + diff_frame_id = cm->current_frame_id - prev_frame_id; + } else { + diff_frame_id = + (1 << frame_id_length) + cm->current_frame_id - prev_frame_id; + } + /* Check current_frame_id for conformance */ + if (prev_frame_id == cm->current_frame_id || + diff_frame_id >= (1 << (frame_id_length - 1))) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Invalid value of current_frame_id"); + } + } + /* Check if some frames need to be marked as not valid for referencing */ + for (int i = 0; i < REF_FRAMES; i++) { + if (cm->frame_type == KEY_FRAME && cm->show_frame) { + cm->valid_for_referencing[i] = 0; + } else if (cm->current_frame_id - (1 << diff_len) > 0) { + if (cm->ref_frame_id[i] > cm->current_frame_id || + cm->ref_frame_id[i] < cm->current_frame_id - (1 << diff_len)) + cm->valid_for_referencing[i] = 0; + } else { + if (cm->ref_frame_id[i] > cm->current_frame_id && + cm->ref_frame_id[i] < (1 << frame_id_length) + + cm->current_frame_id - (1 << diff_len)) + cm->valid_for_referencing[i] = 0; + } + } + } + + frame_size_override_flag = frame_is_sframe(cm) ? 1 : aom_rb_read_bit(rb); + + cm->frame_offset = + aom_rb_read_literal(rb, seq_params->order_hint_bits_minus_1 + 1); + cm->current_video_frame = cm->frame_offset; + + if (!cm->error_resilient_mode && !frame_is_intra_only(cm)) { + cm->primary_ref_frame = aom_rb_read_literal(rb, PRIMARY_REF_BITS); + } + } + + if (seq_params->decoder_model_info_present_flag) { + cm->buffer_removal_time_present = aom_rb_read_bit(rb); + if (cm->buffer_removal_time_present) { + for (int op_num = 0; + op_num < seq_params->operating_points_cnt_minus_1 + 1; op_num++) { + if (cm->op_params[op_num].decoder_model_param_present_flag) { + if ((((seq_params->operating_point_idc[op_num] >> + cm->temporal_layer_id) & + 0x1) && + ((seq_params->operating_point_idc[op_num] >> + (cm->spatial_layer_id + 8)) & + 0x1)) || + seq_params->operating_point_idc[op_num] == 0) { + cm->op_frame_timing[op_num].buffer_removal_time = + aom_rb_read_unsigned_literal( + rb, cm->buffer_model.buffer_removal_time_length); + } else { + cm->op_frame_timing[op_num].buffer_removal_time = 0; + } + } else { + cm->op_frame_timing[op_num].buffer_removal_time = 0; + } + } + } + } + if (cm->frame_type == KEY_FRAME) { + if (!cm->show_frame) // unshown keyframe (forward keyframe) + pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES); + else // shown keyframe + pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1; + + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + cm->frame_refs[i].idx = INVALID_IDX; + cm->frame_refs[i].buf = NULL; + } + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + } else { + if (cm->intra_only) { + pbi->refresh_frame_flags = aom_rb_read_literal(rb, REF_FRAMES); + if (pbi->refresh_frame_flags == 0xFF) { + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Intra only frames cannot have refresh flags 0xFF"); + } + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + } else if (pbi->need_resync != 1) { /* Skip if need resync */ + pbi->refresh_frame_flags = + frame_is_sframe(cm) ? 0xFF : aom_rb_read_literal(rb, REF_FRAMES); + if (!pbi->refresh_frame_flags) { + // NOTE: "pbi->refresh_frame_flags == 0" indicates that the coded frame + // will not be used as a reference + cm->is_reference_frame = 0; + } + } + } + + if (!frame_is_intra_only(cm) || pbi->refresh_frame_flags != 0xFF) { + // Read all ref frame order hints if error_resilient_mode == 1 + if (cm->error_resilient_mode && seq_params->enable_order_hint) { + for (int ref_idx = 0; ref_idx < REF_FRAMES; ref_idx++) { + // Read order hint from bit stream + unsigned int frame_offset = + aom_rb_read_literal(rb, seq_params->order_hint_bits_minus_1 + 1); + // Get buffer index + int buf_idx = cm->ref_frame_map[ref_idx]; + assert(buf_idx < FRAME_BUFFERS); + if (buf_idx == -1 || + frame_offset != frame_bufs[buf_idx].cur_frame_offset) { + if (buf_idx >= 0) { + lock_buffer_pool(pool); + decrease_ref_count(buf_idx, frame_bufs, pool); + unlock_buffer_pool(pool); + } + // If no corresponding buffer exists, allocate a new buffer with all + // pixels set to neutral grey. + buf_idx = get_free_fb(cm); + if (buf_idx == INVALID_IDX) { + aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, + "Unable to find free frame buffer"); + } + lock_buffer_pool(pool); + if (aom_realloc_frame_buffer( + &frame_bufs[buf_idx].buf, seq_params->max_frame_width, + seq_params->max_frame_height, seq_params->subsampling_x, + seq_params->subsampling_y, seq_params->use_highbitdepth, + AOM_BORDER_IN_PIXELS, cm->byte_alignment, + &pool->frame_bufs[buf_idx].raw_frame_buffer, pool->get_fb_cb, + pool->cb_priv)) { + unlock_buffer_pool(pool); + aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + set_planes_to_neutral_grey(seq_params, &frame_bufs[buf_idx].buf, 0); + + cm->ref_frame_map[ref_idx] = buf_idx; + frame_bufs[buf_idx].cur_frame_offset = frame_offset; + } + } + } + } + + if (cm->frame_type == KEY_FRAME) { + setup_frame_size(cm, frame_size_override_flag, rb); + + if (cm->allow_screen_content_tools && !av1_superres_scaled(cm)) + cm->allow_intrabc = aom_rb_read_bit(rb); + cm->allow_ref_frame_mvs = 0; + cm->prev_frame = NULL; + } else { + cm->allow_ref_frame_mvs = 0; + + if (cm->intra_only) { + cm->cur_frame->film_grain_params_present = + seq_params->film_grain_params_present; + setup_frame_size(cm, frame_size_override_flag, rb); + if (cm->allow_screen_content_tools && !av1_superres_scaled(cm)) + cm->allow_intrabc = aom_rb_read_bit(rb); + + } else if (pbi->need_resync != 1) { /* Skip if need resync */ + + // Frame refs short signaling is off when error resilient mode is on. + if (seq_params->enable_order_hint) + cm->frame_refs_short_signaling = aom_rb_read_bit(rb); + + if (cm->frame_refs_short_signaling) { + // == LAST_FRAME == + const int lst_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); + const int lst_idx = cm->ref_frame_map[lst_ref]; + + // == GOLDEN_FRAME == + const int gld_ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); + const int gld_idx = cm->ref_frame_map[gld_ref]; + + // Most of the time, streams start with a keyframe. In that case, + // ref_frame_map will have been filled in at that point and will not + // contain any -1's. However, streams are explicitly allowed to start + // with an intra-only frame, so long as they don't then signal a + // reference to a slot that hasn't been set yet. That's what we are + // checking here. + if (lst_idx == -1) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Inter frame requests nonexistent reference"); + if (gld_idx == -1) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Inter frame requests nonexistent reference"); + + av1_set_frame_refs(cm, lst_ref, gld_ref); + } + + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + int ref = 0; + if (!cm->frame_refs_short_signaling) { + ref = aom_rb_read_literal(rb, REF_FRAMES_LOG2); + const int idx = cm->ref_frame_map[ref]; + + // Most of the time, streams start with a keyframe. In that case, + // ref_frame_map will have been filled in at that point and will not + // contain any -1's. However, streams are explicitly allowed to start + // with an intra-only frame, so long as they don't then signal a + // reference to a slot that hasn't been set yet. That's what we are + // checking here. + if (idx == -1) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Inter frame requests nonexistent reference"); + + RefBuffer *const ref_frame = &cm->frame_refs[i]; + ref_frame->idx = idx; + ref_frame->buf = &frame_bufs[idx].buf; + ref_frame->map_idx = ref; + } else { + ref = cm->frame_refs[i].map_idx; + } + + cm->ref_frame_sign_bias[LAST_FRAME + i] = 0; + + if (seq_params->frame_id_numbers_present_flag) { + int frame_id_length = seq_params->frame_id_length; + int diff_len = seq_params->delta_frame_id_length; + int delta_frame_id_minus_1 = aom_rb_read_literal(rb, diff_len); + int ref_frame_id = + ((cm->current_frame_id - (delta_frame_id_minus_1 + 1) + + (1 << frame_id_length)) % + (1 << frame_id_length)); + // Compare values derived from delta_frame_id_minus_1 and + // refresh_frame_flags. Also, check valid for referencing + if (ref_frame_id != cm->ref_frame_id[ref] || + cm->valid_for_referencing[ref] == 0) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Reference buffer frame ID mismatch"); + } + } + + if (!cm->error_resilient_mode && frame_size_override_flag) { + setup_frame_size_with_refs(cm, rb); + } else { + setup_frame_size(cm, frame_size_override_flag, rb); + } + + if (cm->cur_frame_force_integer_mv) { + cm->allow_high_precision_mv = 0; + } else { + cm->allow_high_precision_mv = aom_rb_read_bit(rb); + } + cm->interp_filter = read_frame_interp_filter(rb); + cm->switchable_motion_mode = aom_rb_read_bit(rb); + } + + cm->prev_frame = get_prev_frame(cm); + if (cm->primary_ref_frame != PRIMARY_REF_NONE && + cm->frame_refs[cm->primary_ref_frame].idx < 0) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Reference frame containing this frame's initial " + "frame context is unavailable."); + } + + if (!cm->intra_only && pbi->need_resync != 1) { + if (frame_might_allow_ref_frame_mvs(cm)) + cm->allow_ref_frame_mvs = aom_rb_read_bit(rb); + else + cm->allow_ref_frame_mvs = 0; + + for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { + RefBuffer *const ref_buf = &cm->frame_refs[i]; + av1_setup_scale_factors_for_frame( + &ref_buf->sf, ref_buf->buf->y_crop_width, + ref_buf->buf->y_crop_height, cm->width, cm->height); + if ((!av1_is_valid_scale(&ref_buf->sf))) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Reference frame has invalid dimensions"); + } + } + } + + av1_setup_frame_buf_refs(cm); + + av1_setup_frame_sign_bias(cm); + + cm->cur_frame->intra_only = cm->frame_type == KEY_FRAME || cm->intra_only; + cm->cur_frame->frame_type = cm->frame_type; + + if (seq_params->frame_id_numbers_present_flag) { + /* If bitmask is set, update reference frame id values and + mark frames as valid for reference */ + int refresh_frame_flags = pbi->refresh_frame_flags; + for (int i = 0; i < REF_FRAMES; i++) { + if ((refresh_frame_flags >> i) & 1) { + cm->ref_frame_id[i] = cm->current_frame_id; + cm->valid_for_referencing[i] = 1; + } + } + } + + const int might_bwd_adapt = + !(seq_params->reduced_still_picture_hdr) && !(cm->disable_cdf_update); + if (might_bwd_adapt) { + cm->refresh_frame_context = aom_rb_read_bit(rb) + ? REFRESH_FRAME_CONTEXT_DISABLED + : REFRESH_FRAME_CONTEXT_BACKWARD; + } else { + cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; + } + + get_frame_new_buffer(cm)->bit_depth = seq_params->bit_depth; + get_frame_new_buffer(cm)->color_primaries = seq_params->color_primaries; + get_frame_new_buffer(cm)->transfer_characteristics = + seq_params->transfer_characteristics; + get_frame_new_buffer(cm)->matrix_coefficients = + seq_params->matrix_coefficients; + get_frame_new_buffer(cm)->monochrome = seq_params->monochrome; + get_frame_new_buffer(cm)->chroma_sample_position = + seq_params->chroma_sample_position; + get_frame_new_buffer(cm)->color_range = seq_params->color_range; + get_frame_new_buffer(cm)->render_width = cm->render_width; + get_frame_new_buffer(cm)->render_height = cm->render_height; + + if (pbi->need_resync) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Keyframe / intra-only frame required to reset decoder" + " state"); + } + + // Generate next_ref_frame_map. + lock_buffer_pool(pool); + int ref_index = 0; + for (int mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + if (mask & 1) { + cm->next_ref_frame_map[ref_index] = cm->new_fb_idx; + ++frame_bufs[cm->new_fb_idx].ref_count; + } else { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + } + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + ++ref_index; + } + + for (; ref_index < REF_FRAMES; ++ref_index) { + cm->next_ref_frame_map[ref_index] = cm->ref_frame_map[ref_index]; + + // Current thread holds the reference frame. + if (cm->ref_frame_map[ref_index] >= 0) + ++frame_bufs[cm->ref_frame_map[ref_index]].ref_count; + } + unlock_buffer_pool(pool); + pbi->hold_ref_buf = 1; + + if (cm->allow_intrabc) { + // Set parameters corresponding to no filtering. + struct loopfilter *lf = &cm->lf; + lf->filter_level[0] = 0; + lf->filter_level[1] = 0; + cm->cdef_bits = 0; + cm->cdef_strengths[0] = 0; + cm->nb_cdef_strengths = 1; + cm->cdef_uv_strengths[0] = 0; + cm->rst_info[0].frame_restoration_type = RESTORE_NONE; + cm->rst_info[1].frame_restoration_type = RESTORE_NONE; + cm->rst_info[2].frame_restoration_type = RESTORE_NONE; + } + + read_tile_info(pbi, rb); + setup_quantization(cm, rb); + xd->bd = (int)seq_params->bit_depth; + + if (cm->num_allocated_above_context_planes < av1_num_planes(cm) || + cm->num_allocated_above_context_mi_col < cm->mi_cols || + cm->num_allocated_above_contexts < cm->tile_rows) { + av1_free_above_context_buffers(cm, cm->num_allocated_above_contexts); + if (av1_alloc_above_context_buffers(cm, cm->tile_rows)) + aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, + "Failed to allocate context buffers"); + } + + if (cm->primary_ref_frame == PRIMARY_REF_NONE) { + av1_setup_past_independence(cm); + } + + setup_segmentation(cm, rb); + + cm->delta_q_res = 1; + cm->delta_lf_res = 1; + cm->delta_lf_present_flag = 0; + cm->delta_lf_multi = 0; + cm->delta_q_present_flag = cm->base_qindex > 0 ? aom_rb_read_bit(rb) : 0; + if (cm->delta_q_present_flag) { + xd->current_qindex = cm->base_qindex; + cm->delta_q_res = 1 << aom_rb_read_literal(rb, 2); + if (!cm->allow_intrabc) cm->delta_lf_present_flag = aom_rb_read_bit(rb); + if (cm->delta_lf_present_flag) { + cm->delta_lf_res = 1 << aom_rb_read_literal(rb, 2); + cm->delta_lf_multi = aom_rb_read_bit(rb); + av1_reset_loop_filter_delta(xd, av1_num_planes(cm)); + } + } + + xd->cur_frame_force_integer_mv = cm->cur_frame_force_integer_mv; + + for (int i = 0; i < MAX_SEGMENTS; ++i) { + const int qindex = cm->seg.enabled + ? av1_get_qindex(&cm->seg, i, cm->base_qindex) + : cm->base_qindex; + xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 && + cm->u_dc_delta_q == 0 && cm->u_ac_delta_q == 0 && + cm->v_dc_delta_q == 0 && cm->v_ac_delta_q == 0; + xd->qindex[i] = qindex; + } + cm->coded_lossless = is_coded_lossless(cm, xd); + cm->all_lossless = cm->coded_lossless && !av1_superres_scaled(cm); + setup_segmentation_dequant(cm); + if (cm->coded_lossless) { + cm->lf.filter_level[0] = 0; + cm->lf.filter_level[1] = 0; + } + if (cm->coded_lossless || !seq_params->enable_cdef) { + cm->cdef_bits = 0; + cm->cdef_strengths[0] = 0; + cm->cdef_uv_strengths[0] = 0; + } + if (cm->all_lossless || !seq_params->enable_restoration) { + cm->rst_info[0].frame_restoration_type = RESTORE_NONE; + cm->rst_info[1].frame_restoration_type = RESTORE_NONE; + cm->rst_info[2].frame_restoration_type = RESTORE_NONE; + } + setup_loopfilter(cm, rb); + + if (!cm->coded_lossless && seq_params->enable_cdef) { + setup_cdef(cm, rb); + } + if (!cm->all_lossless && seq_params->enable_restoration) { + decode_restoration_mode(cm, rb); + } + + cm->tx_mode = read_tx_mode(cm, rb); + cm->reference_mode = read_frame_reference_mode(cm, rb); + if (cm->reference_mode != SINGLE_REFERENCE) setup_compound_reference_mode(cm); + + av1_setup_skip_mode_allowed(cm); + cm->skip_mode_flag = cm->is_skip_mode_allowed ? aom_rb_read_bit(rb) : 0; + + if (frame_might_allow_warped_motion(cm)) + cm->allow_warped_motion = aom_rb_read_bit(rb); + else + cm->allow_warped_motion = 0; + + cm->reduced_tx_set_used = aom_rb_read_bit(rb); + + if (cm->allow_ref_frame_mvs && !frame_might_allow_ref_frame_mvs(cm)) { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Frame wrongly requests reference frame MVs"); + } + + if (!frame_is_intra_only(cm)) read_global_motion(cm, rb); + + cm->cur_frame->film_grain_params_present = + seq_params->film_grain_params_present; + read_film_grain(cm, rb); + +#if EXT_TILE_DEBUG + if (pbi->ext_tile_debug && cm->large_scale_tile) { + read_ext_tile_info(pbi, rb); + av1_set_single_tile_decoding_mode(cm); + } +#endif // EXT_TILE_DEBUG + return 0; +} + +struct aom_read_bit_buffer *av1_init_read_bit_buffer( + AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data, + const uint8_t *data_end) { + rb->bit_offset = 0; + rb->error_handler = error_handler; + rb->error_handler_data = &pbi->common; + rb->bit_buffer = data; + rb->bit_buffer_end = data_end; + return rb; +} + +void av1_read_frame_size(struct aom_read_bit_buffer *rb, int num_bits_width, + int num_bits_height, int *width, int *height) { + *width = aom_rb_read_literal(rb, num_bits_width) + 1; + *height = aom_rb_read_literal(rb, num_bits_height) + 1; +} + +BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb) { + int profile = aom_rb_read_literal(rb, PROFILE_BITS); + return (BITSTREAM_PROFILE)profile; +} + +void superres_post_decode(AV1Decoder *pbi) { + AV1_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + + if (!av1_superres_scaled(cm)) return; + assert(!cm->all_lossless); + + lock_buffer_pool(pool); + av1_superres_upscale(cm, pool); + unlock_buffer_pool(pool); +} + +uint32_t av1_decode_frame_headers_and_setup(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb, + const uint8_t *data, + const uint8_t **p_data_end, + int trailing_bits_present) { + AV1_COMMON *const cm = &pbi->common; + const int num_planes = av1_num_planes(cm); + MACROBLOCKD *const xd = &pbi->mb; + +#if CONFIG_BITSTREAM_DEBUG + bitstream_queue_set_frame_read(cm->current_video_frame * 2 + cm->show_frame); +#endif +#if CONFIG_MISMATCH_DEBUG + mismatch_move_frame_idx_r(); +#endif + + for (int i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { + cm->global_motion[i] = default_warp_params; + cm->cur_frame->global_motion[i] = default_warp_params; + } + xd->global_motion = cm->global_motion; + + read_uncompressed_header(pbi, rb); + + if (trailing_bits_present) av1_check_trailing_bits(pbi, rb); + + // If cm->single_tile_decoding = 0, the independent decoding of a single tile + // or a section of a frame is not allowed. + if (!cm->single_tile_decoding && + (pbi->dec_tile_row >= 0 || pbi->dec_tile_col >= 0)) { + pbi->dec_tile_row = -1; + pbi->dec_tile_col = -1; + } + + const uint32_t uncomp_hdr_size = + (uint32_t)aom_rb_bytes_read(rb); // Size of the uncompressed header + YV12_BUFFER_CONFIG *new_fb = get_frame_new_buffer(cm); + xd->cur_buf = new_fb; + if (av1_allow_intrabc(cm)) { + av1_setup_scale_factors_for_frame( + &cm->sf_identity, xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height, + xd->cur_buf->y_crop_width, xd->cur_buf->y_crop_height); + } + + if (cm->show_existing_frame) { + // showing a frame directly + *p_data_end = data + uncomp_hdr_size; + if (cm->reset_decoder_state) { + // Use the default frame context values. + *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS]; + if (!cm->fc->initialized) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Uninitialized entropy context."); + } + return uncomp_hdr_size; + } + + cm->setup_mi(cm); + + cm->current_frame_seg_map = cm->cur_frame->seg_map; + + av1_setup_motion_field(cm); + + av1_setup_block_planes(xd, cm->seq_params.subsampling_x, + cm->seq_params.subsampling_y, num_planes); + if (cm->primary_ref_frame == PRIMARY_REF_NONE) { + // use the default frame context values + *cm->fc = cm->frame_contexts[FRAME_CONTEXT_DEFAULTS]; + } else { + *cm->fc = cm->frame_contexts[cm->frame_refs[cm->primary_ref_frame].idx]; + } + if (!cm->fc->initialized) + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Uninitialized entropy context."); + + xd->corrupted = 0; + return uncomp_hdr_size; +} + +// Once-per-frame initialization +static void setup_frame_info(AV1Decoder *pbi) { + AV1_COMMON *const cm = &pbi->common; + + if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE || + cm->rst_info[1].frame_restoration_type != RESTORE_NONE || + cm->rst_info[2].frame_restoration_type != RESTORE_NONE) { + av1_alloc_restoration_buffers(cm); + } + const int use_highbd = cm->seq_params.use_highbitdepth ? 1 : 0; + const int buf_size = MC_TEMP_BUF_PELS << use_highbd; + if (pbi->td.mc_buf_size != buf_size) { + av1_free_mc_tmp_buf(&pbi->td); + allocate_mc_tmp_buf(cm, &pbi->td, buf_size, use_highbd); + } +} + +void av1_decode_tg_tiles_and_wrapup(AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + const uint8_t **p_data_end, int start_tile, + int end_tile, int initialize_flag) { + AV1_COMMON *const cm = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + const int tile_count_tg = end_tile - start_tile + 1; + + if (initialize_flag) setup_frame_info(pbi); + const int num_planes = av1_num_planes(cm); +#if LOOP_FILTER_BITMASK + av1_loop_filter_frame_init(cm, 0, num_planes); + av1_zero_array(cm->lf.lfm, cm->lf.lfm_num); +#endif + + if (pbi->max_threads > 1 && !(cm->large_scale_tile && !pbi->ext_tile_debug) && + pbi->row_mt) + *p_data_end = + decode_tiles_row_mt(pbi, data, data_end, start_tile, end_tile); + else if (pbi->max_threads > 1 && tile_count_tg > 1 && + !(cm->large_scale_tile && !pbi->ext_tile_debug)) + *p_data_end = decode_tiles_mt(pbi, data, data_end, start_tile, end_tile); + else + *p_data_end = decode_tiles(pbi, data, data_end, start_tile, end_tile); + + // If the bit stream is monochrome, set the U and V buffers to a constant. + if (num_planes < 3) { + set_planes_to_neutral_grey(&cm->seq_params, xd->cur_buf, 1); + } + + if (end_tile != cm->tile_rows * cm->tile_cols - 1) { + return; + } + + if (!cm->allow_intrabc && !cm->single_tile_decoding) { + if (cm->lf.filter_level[0] || cm->lf.filter_level[1]) { +#if LOOP_FILTER_BITMASK + av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 1, 0, + num_planes, 0); +#else + if (pbi->num_workers > 1) { + av1_loop_filter_frame_mt(get_frame_new_buffer(cm), cm, &pbi->mb, 0, + num_planes, 0, pbi->tile_workers, + pbi->num_workers, &pbi->lf_row_sync); + } else { + av1_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, 0, + num_planes, 0); + } +#endif + } + + const int do_loop_restoration = + cm->rst_info[0].frame_restoration_type != RESTORE_NONE || + cm->rst_info[1].frame_restoration_type != RESTORE_NONE || + cm->rst_info[2].frame_restoration_type != RESTORE_NONE; + const int do_cdef = + !cm->skip_loop_filter && !cm->coded_lossless && + (cm->cdef_bits || cm->cdef_strengths[0] || cm->cdef_uv_strengths[0]); + const int do_superres = av1_superres_scaled(cm); + const int optimized_loop_restoration = !do_cdef && !do_superres; + + if (!optimized_loop_restoration) { + if (do_loop_restoration) + av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 0); + + if (do_cdef) av1_cdef_frame(&pbi->cur_buf->buf, cm, &pbi->mb); + + superres_post_decode(pbi); + + if (do_loop_restoration) { + av1_loop_restoration_save_boundary_lines(&pbi->cur_buf->buf, cm, 1); + if (pbi->num_workers > 1) { + av1_loop_restoration_filter_frame_mt( + (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration, + pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync, + &pbi->lr_ctxt); + } else { + av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf, + cm, optimized_loop_restoration, + &pbi->lr_ctxt); + } + } + } else { + // In no cdef and no superres case. Provide an optimized version of + // loop_restoration_filter. + if (do_loop_restoration) { + if (pbi->num_workers > 1) { + av1_loop_restoration_filter_frame_mt( + (YV12_BUFFER_CONFIG *)xd->cur_buf, cm, optimized_loop_restoration, + pbi->tile_workers, pbi->num_workers, &pbi->lr_row_sync, + &pbi->lr_ctxt); + } else { + av1_loop_restoration_filter_frame((YV12_BUFFER_CONFIG *)xd->cur_buf, + cm, optimized_loop_restoration, + &pbi->lr_ctxt); + } + } + } + } + + if (!xd->corrupted) { + if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { + assert(cm->context_update_tile_id < pbi->allocated_tiles); + *cm->fc = pbi->tile_data[cm->context_update_tile_id].tctx; + av1_reset_cdf_symbol_counters(cm->fc); + } + } else { + aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data is corrupted."); + } + +#if CONFIG_INSPECTION + if (pbi->inspect_cb != NULL) { + (*pbi->inspect_cb)(pbi, pbi->inspect_ctx); + } +#endif + + // Non frame parallel update frame context here. + if (!cm->large_scale_tile) { + cm->frame_contexts[cm->new_fb_idx] = *cm->fc; + } +} diff --git a/media/libaom/src/av1/decoder/decodeframe.h b/media/libaom/src/av1/decoder/decodeframe.h new file mode 100644 index 000000000..ddad273f1 --- /dev/null +++ b/media/libaom/src/av1/decoder/decodeframe.h @@ -0,0 +1,85 @@ +/* + * 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 AOM_AV1_DECODER_DECODEFRAME_H_ +#define AOM_AV1_DECODER_DECODEFRAME_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +struct AV1Decoder; +struct aom_read_bit_buffer; +struct ThreadData; + +// Reads the middle part of the sequence header OBU (from +// frame_width_bits_minus_1 to enable_restoration) into seq_params. +// Reports errors by calling rb->error_handler() or aom_internal_error(). +void av1_read_sequence_header(AV1_COMMON *cm, struct aom_read_bit_buffer *rb, + SequenceHeader *seq_params); + +void av1_read_frame_size(struct aom_read_bit_buffer *rb, int num_bits_width, + int num_bits_height, int *width, int *height); +BITSTREAM_PROFILE av1_read_profile(struct aom_read_bit_buffer *rb); + +// Returns 0 on success. Sets pbi->common.error.error_code and returns -1 on +// failure. +int av1_check_trailing_bits(struct AV1Decoder *pbi, + struct aom_read_bit_buffer *rb); + +// On success, returns the frame header size. On failure, calls +// aom_internal_error and does not return. +// TODO(wtc): Figure out and document the p_data_end parameter. +uint32_t av1_decode_frame_headers_and_setup(struct AV1Decoder *pbi, + struct aom_read_bit_buffer *rb, + const uint8_t *data, + const uint8_t **p_data_end, + int trailing_bits_present); + +void av1_decode_tg_tiles_and_wrapup(struct AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + const uint8_t **p_data_end, int startTile, + int endTile, int initialize_flag); + +// Implements the color_config() function in the spec. Reports errors by +// calling rb->error_handler() or aom_internal_error(). +void av1_read_color_config(struct aom_read_bit_buffer *rb, + int allow_lowbitdepth, SequenceHeader *seq_params, + struct aom_internal_error_info *error_info); + +// Implements the timing_info() function in the spec. Reports errors by calling +// rb->error_handler(). +void av1_read_timing_info_header(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb); + +// Implements the decoder_model_info() function in the spec. Reports errors by +// calling rb->error_handler(). +void av1_read_decoder_model_info(AV1_COMMON *cm, + struct aom_read_bit_buffer *rb); + +// Implements the operating_parameters_info() function in the spec. Reports +// errors by calling rb->error_handler() or aom_internal_error(). +void av1_read_op_parameters_info(AV1_COMMON *const cm, + struct aom_read_bit_buffer *rb, int op_num); + +struct aom_read_bit_buffer *av1_init_read_bit_buffer( + struct AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data, + const uint8_t *data_end); + +void av1_free_mc_tmp_buf(struct ThreadData *thread_data); + +void av1_set_single_tile_decoding_mode(AV1_COMMON *const cm); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // AOM_AV1_DECODER_DECODEFRAME_H_ diff --git a/media/libaom/src/av1/decoder/decodemv.c b/media/libaom/src/av1/decoder/decodemv.c new file mode 100644 index 000000000..551e4d543 --- /dev/null +++ b/media/libaom/src/av1/decoder/decodemv.c @@ -0,0 +1,1560 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include <assert.h> + +#include "av1/common/cfl.h" +#include "av1/common/common.h" +#include "av1/common/entropy.h" +#include "av1/common/entropymode.h" +#include "av1/common/entropymv.h" +#include "av1/common/mvref_common.h" +#include "av1/common/pred_common.h" +#include "av1/common/reconinter.h" +#include "av1/common/reconintra.h" +#include "av1/common/seg_common.h" +#include "av1/common/warped_motion.h" + +#include "av1/decoder/decodeframe.h" +#include "av1/decoder/decodemv.h" + +#include "aom_dsp/aom_dsp_common.h" + +#define ACCT_STR __func__ + +#define DEC_MISMATCH_DEBUG 0 + +static PREDICTION_MODE read_intra_mode(aom_reader *r, aom_cdf_prob *cdf) { + return (PREDICTION_MODE)aom_read_symbol(r, cdf, INTRA_MODES, ACCT_STR); +} + +static void read_cdef(AV1_COMMON *cm, aom_reader *r, MACROBLOCKD *const xd, + int mi_col, int mi_row) { + MB_MODE_INFO *const mbmi = xd->mi[0]; + if (cm->coded_lossless) return; + if (cm->allow_intrabc) { + assert(cm->cdef_bits == 0); + return; + } + + if (!(mi_col & (cm->seq_params.mib_size - 1)) && + !(mi_row & (cm->seq_params.mib_size - 1))) { // Top left? + xd->cdef_preset[0] = xd->cdef_preset[1] = xd->cdef_preset[2] = + xd->cdef_preset[3] = -1; + } + // Read CDEF param at the first non-skip coding block + const int mask = (1 << (6 - MI_SIZE_LOG2)); + const int m = ~(mask - 1); + const int index = cm->seq_params.sb_size == BLOCK_128X128 + ? !!(mi_col & mask) + 2 * !!(mi_row & mask) + : 0; + cm->mi_grid_visible[(mi_row & m) * cm->mi_stride + (mi_col & m)] + ->cdef_strength = xd->cdef_preset[index] = + xd->cdef_preset[index] == -1 && !mbmi->skip + ? aom_read_literal(r, cm->cdef_bits, ACCT_STR) + : xd->cdef_preset[index]; +} + +static int read_delta_qindex(AV1_COMMON *cm, const MACROBLOCKD *xd, + aom_reader *r, MB_MODE_INFO *const mbmi, + int mi_col, int mi_row) { + int sign, abs, reduced_delta_qindex = 0; + BLOCK_SIZE bsize = mbmi->sb_type; + const int b_col = mi_col & (cm->seq_params.mib_size - 1); + const int b_row = mi_row & (cm->seq_params.mib_size - 1); + const int read_delta_q_flag = (b_col == 0 && b_row == 0); + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + + if ((bsize != cm->seq_params.sb_size || mbmi->skip == 0) && + read_delta_q_flag) { + abs = aom_read_symbol(r, ec_ctx->delta_q_cdf, DELTA_Q_PROBS + 1, ACCT_STR); + const int smallval = (abs < DELTA_Q_SMALL); + + if (!smallval) { + const int rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1; + const int thr = (1 << rem_bits) + 1; + abs = aom_read_literal(r, rem_bits, ACCT_STR) + thr; + } + + if (abs) { + sign = aom_read_bit(r, ACCT_STR); + } else { + sign = 1; + } + + reduced_delta_qindex = sign ? -abs : abs; + } + return reduced_delta_qindex; +} +static int read_delta_lflevel(const AV1_COMMON *const cm, aom_reader *r, + aom_cdf_prob *const cdf, + const MB_MODE_INFO *const mbmi, int mi_col, + int mi_row) { + int reduced_delta_lflevel = 0; + const BLOCK_SIZE bsize = mbmi->sb_type; + const int b_col = mi_col & (cm->seq_params.mib_size - 1); + const int b_row = mi_row & (cm->seq_params.mib_size - 1); + const int read_delta_lf_flag = (b_col == 0 && b_row == 0); + + if ((bsize != cm->seq_params.sb_size || mbmi->skip == 0) && + read_delta_lf_flag) { + int abs = aom_read_symbol(r, cdf, DELTA_LF_PROBS + 1, ACCT_STR); + const int smallval = (abs < DELTA_LF_SMALL); + if (!smallval) { + const int rem_bits = aom_read_literal(r, 3, ACCT_STR) + 1; + const int thr = (1 << rem_bits) + 1; + abs = aom_read_literal(r, rem_bits, ACCT_STR) + thr; + } + const int sign = abs ? aom_read_bit(r, ACCT_STR) : 1; + reduced_delta_lflevel = sign ? -abs : abs; + } + return reduced_delta_lflevel; +} + +static UV_PREDICTION_MODE read_intra_mode_uv(FRAME_CONTEXT *ec_ctx, + aom_reader *r, + CFL_ALLOWED_TYPE cfl_allowed, + PREDICTION_MODE y_mode) { + const UV_PREDICTION_MODE uv_mode = + aom_read_symbol(r, ec_ctx->uv_mode_cdf[cfl_allowed][y_mode], + UV_INTRA_MODES - !cfl_allowed, ACCT_STR); + return uv_mode; +} + +static int read_cfl_alphas(FRAME_CONTEXT *const ec_ctx, aom_reader *r, + int *signs_out) { + const int joint_sign = + aom_read_symbol(r, ec_ctx->cfl_sign_cdf, CFL_JOINT_SIGNS, "cfl:signs"); + int idx = 0; + // Magnitudes are only coded for nonzero values + if (CFL_SIGN_U(joint_sign) != CFL_SIGN_ZERO) { + aom_cdf_prob *cdf_u = ec_ctx->cfl_alpha_cdf[CFL_CONTEXT_U(joint_sign)]; + idx = aom_read_symbol(r, cdf_u, CFL_ALPHABET_SIZE, "cfl:alpha_u") + << CFL_ALPHABET_SIZE_LOG2; + } + if (CFL_SIGN_V(joint_sign) != CFL_SIGN_ZERO) { + aom_cdf_prob *cdf_v = ec_ctx->cfl_alpha_cdf[CFL_CONTEXT_V(joint_sign)]; + idx += aom_read_symbol(r, cdf_v, CFL_ALPHABET_SIZE, "cfl:alpha_v"); + } + *signs_out = joint_sign; + return idx; +} + +static INTERINTRA_MODE read_interintra_mode(MACROBLOCKD *xd, aom_reader *r, + int size_group) { + const INTERINTRA_MODE ii_mode = (INTERINTRA_MODE)aom_read_symbol( + r, xd->tile_ctx->interintra_mode_cdf[size_group], INTERINTRA_MODES, + ACCT_STR); + return ii_mode; +} + +static PREDICTION_MODE read_inter_mode(FRAME_CONTEXT *ec_ctx, aom_reader *r, + int16_t ctx) { + int16_t mode_ctx = ctx & NEWMV_CTX_MASK; + int is_newmv, is_zeromv, is_refmv; + is_newmv = aom_read_symbol(r, ec_ctx->newmv_cdf[mode_ctx], 2, ACCT_STR) == 0; + if (is_newmv) return NEWMV; + + mode_ctx = (ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK; + is_zeromv = + aom_read_symbol(r, ec_ctx->zeromv_cdf[mode_ctx], 2, ACCT_STR) == 0; + if (is_zeromv) return GLOBALMV; + + mode_ctx = (ctx >> REFMV_OFFSET) & REFMV_CTX_MASK; + is_refmv = aom_read_symbol(r, ec_ctx->refmv_cdf[mode_ctx], 2, ACCT_STR) == 0; + if (is_refmv) + return NEARESTMV; + else + return NEARMV; +} + +static void read_drl_idx(FRAME_CONTEXT *ec_ctx, MACROBLOCKD *xd, + MB_MODE_INFO *mbmi, aom_reader *r) { + uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame); + mbmi->ref_mv_idx = 0; + if (mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV) { + for (int idx = 0; idx < 2; ++idx) { + if (xd->ref_mv_count[ref_frame_type] > idx + 1) { + uint8_t drl_ctx = av1_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx); + int drl_idx = aom_read_symbol(r, ec_ctx->drl_cdf[drl_ctx], 2, ACCT_STR); + mbmi->ref_mv_idx = idx + drl_idx; + if (!drl_idx) return; + } + } + } + if (have_nearmv_in_inter_mode(mbmi->mode)) { + // Offset the NEARESTMV mode. + // TODO(jingning): Unify the two syntax decoding loops after the NEARESTMV + // mode is factored in. + for (int idx = 1; idx < 3; ++idx) { + if (xd->ref_mv_count[ref_frame_type] > idx + 1) { + uint8_t drl_ctx = av1_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx); + int drl_idx = aom_read_symbol(r, ec_ctx->drl_cdf[drl_ctx], 2, ACCT_STR); + mbmi->ref_mv_idx = idx + drl_idx - 1; + if (!drl_idx) return; + } + } + } +} + +static MOTION_MODE read_motion_mode(AV1_COMMON *cm, MACROBLOCKD *xd, + MB_MODE_INFO *mbmi, aom_reader *r) { + if (cm->switchable_motion_mode == 0) return SIMPLE_TRANSLATION; + if (mbmi->skip_mode) return SIMPLE_TRANSLATION; + + const MOTION_MODE last_motion_mode_allowed = + motion_mode_allowed(xd->global_motion, xd, mbmi, cm->allow_warped_motion); + int motion_mode; + + if (last_motion_mode_allowed == SIMPLE_TRANSLATION) return SIMPLE_TRANSLATION; + + if (last_motion_mode_allowed == OBMC_CAUSAL) { + motion_mode = + aom_read_symbol(r, xd->tile_ctx->obmc_cdf[mbmi->sb_type], 2, ACCT_STR); + return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode); + } else { + motion_mode = + aom_read_symbol(r, xd->tile_ctx->motion_mode_cdf[mbmi->sb_type], + MOTION_MODES, ACCT_STR); + return (MOTION_MODE)(SIMPLE_TRANSLATION + motion_mode); + } +} + +static PREDICTION_MODE read_inter_compound_mode(MACROBLOCKD *xd, aom_reader *r, + int16_t ctx) { + const int mode = + aom_read_symbol(r, xd->tile_ctx->inter_compound_mode_cdf[ctx], + INTER_COMPOUND_MODES, ACCT_STR); + assert(is_inter_compound_mode(NEAREST_NEARESTMV + mode)); + return NEAREST_NEARESTMV + mode; +} + +int av1_neg_deinterleave(int diff, int ref, int max) { + if (!ref) return diff; + if (ref >= (max - 1)) return max - diff - 1; + if (2 * ref < max) { + if (diff <= 2 * ref) { + if (diff & 1) + return ref + ((diff + 1) >> 1); + else + return ref - (diff >> 1); + } + return diff; + } else { + if (diff <= 2 * (max - ref - 1)) { + if (diff & 1) + return ref + ((diff + 1) >> 1); + else + return ref - (diff >> 1); + } + return max - (diff + 1); + } +} + +static int read_segment_id(AV1_COMMON *const cm, const MACROBLOCKD *const xd, + int mi_row, int mi_col, aom_reader *r, int skip) { + int cdf_num; + const int pred = av1_get_spatial_seg_pred(cm, xd, mi_row, mi_col, &cdf_num); + if (skip) return pred; + + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + struct segmentation *const seg = &cm->seg; + struct segmentation_probs *const segp = &ec_ctx->seg; + aom_cdf_prob *pred_cdf = segp->spatial_pred_seg_cdf[cdf_num]; + const int coded_id = aom_read_symbol(r, pred_cdf, MAX_SEGMENTS, ACCT_STR); + const int segment_id = + av1_neg_deinterleave(coded_id, pred, seg->last_active_segid + 1); + + if (segment_id < 0 || segment_id > seg->last_active_segid) { + aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, + "Corrupted segment_ids"); + } + return segment_id; +} + +static int dec_get_segment_id(const AV1_COMMON *cm, const uint8_t *segment_ids, + int mi_offset, int x_mis, int y_mis) { + int segment_id = INT_MAX; + + for (int y = 0; y < y_mis; y++) + for (int x = 0; x < x_mis; x++) + segment_id = + AOMMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]); + + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + return segment_id; +} + +static void set_segment_id(AV1_COMMON *cm, int mi_offset, int x_mis, int y_mis, + int segment_id) { + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + + for (int y = 0; y < y_mis; y++) + for (int x = 0; x < x_mis; x++) + cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id; +} + +static int read_intra_segment_id(AV1_COMMON *const cm, + const MACROBLOCKD *const xd, int mi_row, + int mi_col, int bsize, aom_reader *r, + int skip) { + struct segmentation *const seg = &cm->seg; + if (!seg->enabled) return 0; // Default for disabled segmentation + + assert(seg->update_map && !seg->temporal_update); + + const int mi_offset = mi_row * cm->mi_cols + mi_col; + const int bw = mi_size_wide[bsize]; + const int bh = mi_size_high[bsize]; + const int x_mis = AOMMIN(cm->mi_cols - mi_col, bw); + const int y_mis = AOMMIN(cm->mi_rows - mi_row, bh); + const int segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, skip); + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static void copy_segment_id(const AV1_COMMON *cm, + const uint8_t *last_segment_ids, + uint8_t *current_segment_ids, int mi_offset, + int x_mis, int y_mis) { + for (int y = 0; y < y_mis; y++) + for (int x = 0; x < x_mis; x++) + current_segment_ids[mi_offset + y * cm->mi_cols + x] = + last_segment_ids ? last_segment_ids[mi_offset + y * cm->mi_cols + x] + : 0; +} + +static int get_predicted_segment_id(AV1_COMMON *const cm, int mi_offset, + int x_mis, int y_mis) { + return cm->last_frame_seg_map ? dec_get_segment_id(cm, cm->last_frame_seg_map, + mi_offset, x_mis, y_mis) + : 0; +} + +static int read_inter_segment_id(AV1_COMMON *const cm, MACROBLOCKD *const xd, + int mi_row, int mi_col, int preskip, + aom_reader *r) { + struct segmentation *const seg = &cm->seg; + MB_MODE_INFO *const mbmi = xd->mi[0]; + const int mi_offset = mi_row * cm->mi_cols + mi_col; + const int bw = mi_size_wide[mbmi->sb_type]; + const int bh = mi_size_high[mbmi->sb_type]; + + // TODO(slavarnway): move x_mis, y_mis into xd ????? + const int x_mis = AOMMIN(cm->mi_cols - mi_col, bw); + const int y_mis = AOMMIN(cm->mi_rows - mi_row, bh); + + if (!seg->enabled) return 0; // Default for disabled segmentation + + if (!seg->update_map) { + copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map, + mi_offset, x_mis, y_mis); + return get_predicted_segment_id(cm, mi_offset, x_mis, y_mis); + } + + int segment_id; + if (preskip) { + if (!seg->segid_preskip) return 0; + } else { + if (seg->segid_preskip) return mbmi->segment_id; + if (mbmi->skip) { + if (seg->temporal_update) { + mbmi->seg_id_predicted = 0; + } + segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 1); + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; + } + } + + if (seg->temporal_update) { + const int ctx = av1_get_pred_context_seg_id(xd); + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + struct segmentation_probs *const segp = &ec_ctx->seg; + aom_cdf_prob *pred_cdf = segp->pred_cdf[ctx]; + mbmi->seg_id_predicted = aom_read_symbol(r, pred_cdf, 2, ACCT_STR); + if (mbmi->seg_id_predicted) { + segment_id = get_predicted_segment_id(cm, mi_offset, x_mis, y_mis); + } else { + segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 0); + } + } else { + segment_id = read_segment_id(cm, xd, mi_row, mi_col, r, 0); + } + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static int read_skip_mode(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id, + aom_reader *r) { + if (!cm->skip_mode_flag) return 0; + + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) { + return 0; + } + + if (!is_comp_ref_allowed(xd->mi[0]->sb_type)) return 0; + + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME) || + segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) { + // These features imply single-reference mode, while skip mode implies + // compound reference. Hence, the two are mutually exclusive. + // In other words, skip_mode is implicitly 0 here. + return 0; + } + + const int ctx = av1_get_skip_mode_context(xd); + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + const int skip_mode = + aom_read_symbol(r, ec_ctx->skip_mode_cdfs[ctx], 2, ACCT_STR); + return skip_mode; +} + +static int read_skip(AV1_COMMON *cm, const MACROBLOCKD *xd, int segment_id, + aom_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) { + return 1; + } else { + const int ctx = av1_get_skip_context(xd); + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + const int skip = aom_read_symbol(r, ec_ctx->skip_cdfs[ctx], 2, ACCT_STR); + return skip; + } +} + +// Merge the sorted list of cached colors(cached_colors[0...n_cached_colors-1]) +// and the sorted list of transmitted colors(colors[n_cached_colors...n-1]) into +// one single sorted list(colors[...]). +static void merge_colors(uint16_t *colors, uint16_t *cached_colors, + int n_colors, int n_cached_colors) { + if (n_cached_colors == 0) return; + int cache_idx = 0, trans_idx = n_cached_colors; + for (int i = 0; i < n_colors; ++i) { + if (cache_idx < n_cached_colors && + (trans_idx >= n_colors || + cached_colors[cache_idx] <= colors[trans_idx])) { + colors[i] = cached_colors[cache_idx++]; + } else { + assert(trans_idx < n_colors); + colors[i] = colors[trans_idx++]; + } + } +} + +static void read_palette_colors_y(MACROBLOCKD *const xd, int bit_depth, + PALETTE_MODE_INFO *const pmi, aom_reader *r) { + uint16_t color_cache[2 * PALETTE_MAX_SIZE]; + uint16_t cached_colors[PALETTE_MAX_SIZE]; + const int n_cache = av1_get_palette_cache(xd, 0, color_cache); + const int n = pmi->palette_size[0]; + int idx = 0; + for (int i = 0; i < n_cache && idx < n; ++i) + if (aom_read_bit(r, ACCT_STR)) cached_colors[idx++] = color_cache[i]; + if (idx < n) { + const int n_cached_colors = idx; + pmi->palette_colors[idx++] = aom_read_literal(r, bit_depth, ACCT_STR); + if (idx < n) { + const int min_bits = bit_depth - 3; + int bits = min_bits + aom_read_literal(r, 2, ACCT_STR); + int range = (1 << bit_depth) - pmi->palette_colors[idx - 1] - 1; + for (; idx < n; ++idx) { + assert(range >= 0); + const int delta = aom_read_literal(r, bits, ACCT_STR) + 1; + pmi->palette_colors[idx] = clamp(pmi->palette_colors[idx - 1] + delta, + 0, (1 << bit_depth) - 1); + range -= (pmi->palette_colors[idx] - pmi->palette_colors[idx - 1]); + bits = AOMMIN(bits, av1_ceil_log2(range)); + } + } + merge_colors(pmi->palette_colors, cached_colors, n, n_cached_colors); + } else { + memcpy(pmi->palette_colors, cached_colors, n * sizeof(cached_colors[0])); + } +} + +static void read_palette_colors_uv(MACROBLOCKD *const xd, int bit_depth, + PALETTE_MODE_INFO *const pmi, + aom_reader *r) { + const int n = pmi->palette_size[1]; + // U channel colors. + uint16_t color_cache[2 * PALETTE_MAX_SIZE]; + uint16_t cached_colors[PALETTE_MAX_SIZE]; + const int n_cache = av1_get_palette_cache(xd, 1, color_cache); + int idx = 0; + for (int i = 0; i < n_cache && idx < n; ++i) + if (aom_read_bit(r, ACCT_STR)) cached_colors[idx++] = color_cache[i]; + if (idx < n) { + const int n_cached_colors = idx; + idx += PALETTE_MAX_SIZE; + pmi->palette_colors[idx++] = aom_read_literal(r, bit_depth, ACCT_STR); + if (idx < PALETTE_MAX_SIZE + n) { + const int min_bits = bit_depth - 3; + int bits = min_bits + aom_read_literal(r, 2, ACCT_STR); + int range = (1 << bit_depth) - pmi->palette_colors[idx - 1]; + for (; idx < PALETTE_MAX_SIZE + n; ++idx) { + assert(range >= 0); + const int delta = aom_read_literal(r, bits, ACCT_STR); + pmi->palette_colors[idx] = clamp(pmi->palette_colors[idx - 1] + delta, + 0, (1 << bit_depth) - 1); + range -= (pmi->palette_colors[idx] - pmi->palette_colors[idx - 1]); + bits = AOMMIN(bits, av1_ceil_log2(range)); + } + } + merge_colors(pmi->palette_colors + PALETTE_MAX_SIZE, cached_colors, n, + n_cached_colors); + } else { + memcpy(pmi->palette_colors + PALETTE_MAX_SIZE, cached_colors, + n * sizeof(cached_colors[0])); + } + + // V channel colors. + if (aom_read_bit(r, ACCT_STR)) { // Delta encoding. + const int min_bits_v = bit_depth - 4; + const int max_val = 1 << bit_depth; + int bits = min_bits_v + aom_read_literal(r, 2, ACCT_STR); + pmi->palette_colors[2 * PALETTE_MAX_SIZE] = + aom_read_literal(r, bit_depth, ACCT_STR); + for (int i = 1; i < n; ++i) { + int delta = aom_read_literal(r, bits, ACCT_STR); + if (delta && aom_read_bit(r, ACCT_STR)) delta = -delta; + int val = (int)pmi->palette_colors[2 * PALETTE_MAX_SIZE + i - 1] + delta; + if (val < 0) val += max_val; + if (val >= max_val) val -= max_val; + pmi->palette_colors[2 * PALETTE_MAX_SIZE + i] = val; + } + } else { + for (int i = 0; i < n; ++i) { + pmi->palette_colors[2 * PALETTE_MAX_SIZE + i] = + aom_read_literal(r, bit_depth, ACCT_STR); + } + } +} + +static void read_palette_mode_info(AV1_COMMON *const cm, MACROBLOCKD *const xd, + int mi_row, int mi_col, aom_reader *r) { + const int num_planes = av1_num_planes(cm); + MB_MODE_INFO *const mbmi = xd->mi[0]; + const BLOCK_SIZE bsize = mbmi->sb_type; + assert(av1_allow_palette(cm->allow_screen_content_tools, bsize)); + PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info; + const int bsize_ctx = av1_get_palette_bsize_ctx(bsize); + + if (mbmi->mode == DC_PRED) { + const int palette_mode_ctx = av1_get_palette_mode_ctx(xd); + const int modev = aom_read_symbol( + r, xd->tile_ctx->palette_y_mode_cdf[bsize_ctx][palette_mode_ctx], 2, + ACCT_STR); + if (modev) { + pmi->palette_size[0] = + aom_read_symbol(r, xd->tile_ctx->palette_y_size_cdf[bsize_ctx], + PALETTE_SIZES, ACCT_STR) + + 2; + read_palette_colors_y(xd, cm->seq_params.bit_depth, pmi, r); + } + } + if (num_planes > 1 && mbmi->uv_mode == UV_DC_PRED && + is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x, + xd->plane[1].subsampling_y)) { + const int palette_uv_mode_ctx = (pmi->palette_size[0] > 0); + const int modev = aom_read_symbol( + r, xd->tile_ctx->palette_uv_mode_cdf[palette_uv_mode_ctx], 2, ACCT_STR); + if (modev) { + pmi->palette_size[1] = + aom_read_symbol(r, xd->tile_ctx->palette_uv_size_cdf[bsize_ctx], + PALETTE_SIZES, ACCT_STR) + + 2; + read_palette_colors_uv(xd, cm->seq_params.bit_depth, pmi, r); + } + } +} + +static int read_angle_delta(aom_reader *r, aom_cdf_prob *cdf) { + const int sym = aom_read_symbol(r, cdf, 2 * MAX_ANGLE_DELTA + 1, ACCT_STR); + return sym - MAX_ANGLE_DELTA; +} + +static void read_filter_intra_mode_info(const AV1_COMMON *const cm, + MACROBLOCKD *const xd, aom_reader *r) { + MB_MODE_INFO *const mbmi = xd->mi[0]; + FILTER_INTRA_MODE_INFO *filter_intra_mode_info = + &mbmi->filter_intra_mode_info; + + if (av1_filter_intra_allowed(cm, mbmi)) { + filter_intra_mode_info->use_filter_intra = aom_read_symbol( + r, xd->tile_ctx->filter_intra_cdfs[mbmi->sb_type], 2, ACCT_STR); + if (filter_intra_mode_info->use_filter_intra) { + filter_intra_mode_info->filter_intra_mode = aom_read_symbol( + r, xd->tile_ctx->filter_intra_mode_cdf, FILTER_INTRA_MODES, ACCT_STR); + } + } else { + filter_intra_mode_info->use_filter_intra = 0; + } +} + +void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd, int blk_row, + int blk_col, TX_SIZE tx_size, aom_reader *r) { + MB_MODE_INFO *mbmi = xd->mi[0]; + const int txk_type_idx = + av1_get_txk_type_index(mbmi->sb_type, blk_row, blk_col); + TX_TYPE *tx_type = &mbmi->txk_type[txk_type_idx]; + *tx_type = DCT_DCT; + + // No need to read transform type if block is skipped. + if (mbmi->skip || segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) + return; + + // No need to read transform type for lossless mode(qindex==0). + const int qindex = + cm->seg.enabled ? xd->qindex[mbmi->segment_id] : cm->base_qindex; + if (qindex <= 0) return; + + const int inter_block = is_inter_block(mbmi); + if (get_ext_tx_types(tx_size, inter_block, cm->reduced_tx_set_used) > 1) { + const TxSetType tx_set_type = + av1_get_ext_tx_set_type(tx_size, inter_block, cm->reduced_tx_set_used); + const int eset = + get_ext_tx_set(tx_size, inter_block, cm->reduced_tx_set_used); + // eset == 0 should correspond to a set with only DCT_DCT and + // there is no need to read the tx_type + assert(eset != 0); + + const TX_SIZE square_tx_size = txsize_sqr_map[tx_size]; + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + if (inter_block) { + *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol( + r, ec_ctx->inter_ext_tx_cdf[eset][square_tx_size], + av1_num_ext_tx_set[tx_set_type], ACCT_STR)]; + } else { + const PREDICTION_MODE intra_mode = + mbmi->filter_intra_mode_info.use_filter_intra + ? fimode_to_intradir[mbmi->filter_intra_mode_info + .filter_intra_mode] + : mbmi->mode; + *tx_type = av1_ext_tx_inv[tx_set_type][aom_read_symbol( + r, ec_ctx->intra_ext_tx_cdf[eset][square_tx_size][intra_mode], + av1_num_ext_tx_set[tx_set_type], ACCT_STR)]; + } + } +} + +static INLINE void read_mv(aom_reader *r, MV *mv, const MV *ref, + nmv_context *ctx, MvSubpelPrecision precision); + +static INLINE int is_mv_valid(const MV *mv); + +static INLINE int assign_dv(AV1_COMMON *cm, MACROBLOCKD *xd, int_mv *mv, + const int_mv *ref_mv, int mi_row, int mi_col, + BLOCK_SIZE bsize, aom_reader *r) { + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + read_mv(r, &mv->as_mv, &ref_mv->as_mv, &ec_ctx->ndvc, MV_SUBPEL_NONE); + // DV should not have sub-pel. + assert((mv->as_mv.col & 7) == 0); + assert((mv->as_mv.row & 7) == 0); + mv->as_mv.col = (mv->as_mv.col >> 3) * 8; + mv->as_mv.row = (mv->as_mv.row >> 3) * 8; + int valid = is_mv_valid(&mv->as_mv) && + av1_is_dv_valid(mv->as_mv, cm, xd, mi_row, mi_col, bsize, + cm->seq_params.mib_size_log2); + return valid; +} + +static void read_intrabc_info(AV1_COMMON *const cm, MACROBLOCKD *const xd, + int mi_row, int mi_col, aom_reader *r) { + MB_MODE_INFO *const mbmi = xd->mi[0]; + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + mbmi->use_intrabc = aom_read_symbol(r, ec_ctx->intrabc_cdf, 2, ACCT_STR); + if (mbmi->use_intrabc) { + BLOCK_SIZE bsize = mbmi->sb_type; + mbmi->mode = DC_PRED; + mbmi->uv_mode = UV_DC_PRED; + mbmi->interp_filters = av1_broadcast_interp_filter(BILINEAR); + mbmi->motion_mode = SIMPLE_TRANSLATION; + + int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES]; + int_mv ref_mvs[INTRA_FRAME + 1][MAX_MV_REF_CANDIDATES]; + int_mv global_mvs[REF_FRAMES]; + + av1_find_mv_refs(cm, xd, mbmi, INTRA_FRAME, xd->ref_mv_count, + xd->ref_mv_stack, ref_mvs, global_mvs, mi_row, mi_col, + inter_mode_ctx); + + int_mv nearestmv, nearmv; + + av1_find_best_ref_mvs(0, ref_mvs[INTRA_FRAME], &nearestmv, &nearmv, 0); + int_mv dv_ref = nearestmv.as_int == 0 ? nearmv : nearestmv; + if (dv_ref.as_int == 0) + av1_find_ref_dv(&dv_ref, &xd->tile, cm->seq_params.mib_size, mi_row, + mi_col); + // Ref DV should not have sub-pel. + int valid_dv = (dv_ref.as_mv.col & 7) == 0 && (dv_ref.as_mv.row & 7) == 0; + dv_ref.as_mv.col = (dv_ref.as_mv.col >> 3) * 8; + dv_ref.as_mv.row = (dv_ref.as_mv.row >> 3) * 8; + valid_dv = valid_dv && assign_dv(cm, xd, &mbmi->mv[0], &dv_ref, mi_row, + mi_col, bsize, r); + if (!valid_dv) { + // Intra bc motion vectors are not valid - signal corrupt frame + aom_merge_corrupted_flag(&xd->corrupted, 1); + } + } +} + +// If delta q is present, reads delta_q index. +// Also reads delta_q loop filter levels, if present. +static void read_delta_q_params(AV1_COMMON *const cm, MACROBLOCKD *const xd, + const int mi_row, const int mi_col, + aom_reader *r) { + if (cm->delta_q_present_flag) { + MB_MODE_INFO *const mbmi = xd->mi[0]; + xd->current_qindex += + read_delta_qindex(cm, xd, r, mbmi, mi_col, mi_row) * cm->delta_q_res; + /* Normative: Clamp to [1,MAXQ] to not interfere with lossless mode */ + xd->current_qindex = clamp(xd->current_qindex, 1, MAXQ); + FRAME_CONTEXT *const ec_ctx = xd->tile_ctx; + if (cm->delta_lf_present_flag) { + if (cm->delta_lf_multi) { + const int frame_lf_count = + av1_num_planes(cm) > 1 ? FRAME_LF_COUNT : FRAME_LF_COUNT - 2; + for (int lf_id = 0; lf_id < frame_lf_count; ++lf_id) { + const int tmp_lvl = + xd->delta_lf[lf_id] + + read_delta_lflevel(cm, r, ec_ctx->delta_lf_multi_cdf[lf_id], mbmi, + mi_col, mi_row) * + cm->delta_lf_res; + mbmi->delta_lf[lf_id] = xd->delta_lf[lf_id] = + clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER); + } + } else { + const int tmp_lvl = xd->delta_lf_from_base + + read_delta_lflevel(cm, r, ec_ctx->delta_lf_cdf, + mbmi, mi_col, mi_row) * + cm->delta_lf_res; + mbmi->delta_lf_from_base = xd->delta_lf_from_base = + clamp(tmp_lvl, -MAX_LOOP_FILTER, MAX_LOOP_FILTER); + } + } + } +} + +static void read_intra_frame_mode_info(AV1_COMMON *const cm, + MACROBLOCKD *const xd, int mi_row, + int mi_col, aom_reader *r) { + MB_MODE_INFO *const mbmi = xd->mi[0]; + const MB_MODE_INFO *above_mi = xd->above_mbmi; + const MB_MODE_INFO *left_mi = xd->left_mbmi; + const BLOCK_SIZE bsize = mbmi->sb_type; + struct segmentation *const seg = &cm->seg; + + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + + if (seg->segid_preskip) + mbmi->segment_id = + read_intra_segment_id(cm, xd, mi_row, mi_col, bsize, r, 0); + + mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r); + + if (!seg->segid_preskip) + mbmi->segment_id = + read_intra_segment_id(cm, xd, mi_row, mi_col, bsize, r, mbmi->skip); + + read_cdef(cm, r, xd, mi_col, mi_row); + + read_delta_q_params(cm, xd, mi_row, mi_col, r); + + mbmi->current_qindex = xd->current_qindex; + + mbmi->ref_frame[0] = INTRA_FRAME; + mbmi->ref_frame[1] = NONE_FRAME; + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + mbmi->filter_intra_mode_info.use_filter_intra = 0; + + xd->above_txfm_context = cm->above_txfm_context[xd->tile.tile_row] + mi_col; + xd->left_txfm_context = + xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK); + + if (av1_allow_intrabc(cm)) { + read_intrabc_info(cm, xd, mi_row, mi_col, r); + if (is_intrabc_block(mbmi)) return; + } + + mbmi->mode = read_intra_mode(r, get_y_mode_cdf(ec_ctx, above_mi, left_mi)); + + const int use_angle_delta = av1_use_angle_delta(bsize); + mbmi->angle_delta[PLANE_TYPE_Y] = + (use_angle_delta && av1_is_directional_mode(mbmi->mode)) + ? read_angle_delta(r, ec_ctx->angle_delta_cdf[mbmi->mode - V_PRED]) + : 0; + + if (!cm->seq_params.monochrome && + is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x, + xd->plane[1].subsampling_y)) { + xd->cfl.is_chroma_reference = 1; + mbmi->uv_mode = + read_intra_mode_uv(ec_ctx, r, is_cfl_allowed(xd), mbmi->mode); + if (mbmi->uv_mode == UV_CFL_PRED) { + mbmi->cfl_alpha_idx = read_cfl_alphas(ec_ctx, r, &mbmi->cfl_alpha_signs); + } + mbmi->angle_delta[PLANE_TYPE_UV] = + (use_angle_delta && av1_is_directional_mode(get_uv_mode(mbmi->uv_mode))) + ? read_angle_delta(r, + ec_ctx->angle_delta_cdf[mbmi->uv_mode - V_PRED]) + : 0; + } else { + // Avoid decoding angle_info if there is is no chroma prediction + mbmi->uv_mode = UV_DC_PRED; + xd->cfl.is_chroma_reference = 0; + } + xd->cfl.store_y = store_cfl_required(cm, xd); + + if (av1_allow_palette(cm->allow_screen_content_tools, bsize)) + read_palette_mode_info(cm, xd, mi_row, mi_col, r); + + read_filter_intra_mode_info(cm, xd, r); +} + +static int read_mv_component(aom_reader *r, nmv_component *mvcomp, + int use_subpel, int usehp) { + int mag, d, fr, hp; + const int sign = aom_read_symbol(r, mvcomp->sign_cdf, 2, ACCT_STR); + const int mv_class = + aom_read_symbol(r, mvcomp->classes_cdf, MV_CLASSES, ACCT_STR); + const int class0 = mv_class == MV_CLASS_0; + + // Integer part + if (class0) { + d = aom_read_symbol(r, mvcomp->class0_cdf, CLASS0_SIZE, ACCT_STR); + mag = 0; + } else { + const int n = mv_class + CLASS0_BITS - 1; // number of bits + d = 0; + for (int i = 0; i < n; ++i) + d |= aom_read_symbol(r, mvcomp->bits_cdf[i], 2, ACCT_STR) << i; + mag = CLASS0_SIZE << (mv_class + 2); + } + + if (use_subpel) { + // Fractional part + fr = aom_read_symbol(r, class0 ? mvcomp->class0_fp_cdf[d] : mvcomp->fp_cdf, + MV_FP_SIZE, ACCT_STR); + + // High precision part (if hp is not used, the default value of the hp is 1) + hp = usehp ? aom_read_symbol( + r, class0 ? mvcomp->class0_hp_cdf : mvcomp->hp_cdf, 2, + ACCT_STR) + : 1; + } else { + fr = 3; + hp = 1; + } + + // Result + mag += ((d << 3) | (fr << 1) | hp) + 1; + return sign ? -mag : mag; +} + +static INLINE void read_mv(aom_reader *r, MV *mv, const MV *ref, + nmv_context *ctx, MvSubpelPrecision precision) { + MV diff = kZeroMv; + const MV_JOINT_TYPE joint_type = + (MV_JOINT_TYPE)aom_read_symbol(r, ctx->joints_cdf, MV_JOINTS, ACCT_STR); + + if (mv_joint_vertical(joint_type)) + diff.row = read_mv_component(r, &ctx->comps[0], precision > MV_SUBPEL_NONE, + precision > MV_SUBPEL_LOW_PRECISION); + + if (mv_joint_horizontal(joint_type)) + diff.col = read_mv_component(r, &ctx->comps[1], precision > MV_SUBPEL_NONE, + precision > MV_SUBPEL_LOW_PRECISION); + + mv->row = ref->row + diff.row; + mv->col = ref->col + diff.col; +} + +static REFERENCE_MODE read_block_reference_mode(AV1_COMMON *cm, + const MACROBLOCKD *xd, + aom_reader *r) { + if (!is_comp_ref_allowed(xd->mi[0]->sb_type)) return SINGLE_REFERENCE; + if (cm->reference_mode == REFERENCE_MODE_SELECT) { + const int ctx = av1_get_reference_mode_context(xd); + const REFERENCE_MODE mode = (REFERENCE_MODE)aom_read_symbol( + r, xd->tile_ctx->comp_inter_cdf[ctx], 2, ACCT_STR); + return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE + } else { + assert(cm->reference_mode == SINGLE_REFERENCE); + return cm->reference_mode; + } +} + +#define READ_REF_BIT(pname) \ + aom_read_symbol(r, av1_get_pred_cdf_##pname(xd), 2, ACCT_STR) + +static COMP_REFERENCE_TYPE read_comp_reference_type(const MACROBLOCKD *xd, + aom_reader *r) { + const int ctx = av1_get_comp_reference_type_context(xd); + const COMP_REFERENCE_TYPE comp_ref_type = + (COMP_REFERENCE_TYPE)aom_read_symbol( + r, xd->tile_ctx->comp_ref_type_cdf[ctx], 2, ACCT_STR); + return comp_ref_type; // UNIDIR_COMP_REFERENCE or BIDIR_COMP_REFERENCE +} + +static void set_ref_frames_for_skip_mode(AV1_COMMON *const cm, + MV_REFERENCE_FRAME ref_frame[2]) { + ref_frame[0] = LAST_FRAME + cm->ref_frame_idx_0; + ref_frame[1] = LAST_FRAME + cm->ref_frame_idx_1; +} + +// Read the referncence frame +static void read_ref_frames(AV1_COMMON *const cm, MACROBLOCKD *const xd, + aom_reader *r, int segment_id, + MV_REFERENCE_FRAME ref_frame[2]) { + if (xd->mi[0]->skip_mode) { + set_ref_frames_for_skip_mode(cm, ref_frame); + return; + } + + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) { + ref_frame[0] = (MV_REFERENCE_FRAME)get_segdata(&cm->seg, segment_id, + SEG_LVL_REF_FRAME); + ref_frame[1] = NONE_FRAME; + } else if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP) || + segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) { + ref_frame[0] = LAST_FRAME; + ref_frame[1] = NONE_FRAME; + } else { + const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r); + + if (mode == COMPOUND_REFERENCE) { + const COMP_REFERENCE_TYPE comp_ref_type = read_comp_reference_type(xd, r); + + if (comp_ref_type == UNIDIR_COMP_REFERENCE) { + const int bit = READ_REF_BIT(uni_comp_ref_p); + if (bit) { + ref_frame[0] = BWDREF_FRAME; + ref_frame[1] = ALTREF_FRAME; + } else { + const int bit1 = READ_REF_BIT(uni_comp_ref_p1); + if (bit1) { + const int bit2 = READ_REF_BIT(uni_comp_ref_p2); + if (bit2) { + ref_frame[0] = LAST_FRAME; + ref_frame[1] = GOLDEN_FRAME; + } else { + ref_frame[0] = LAST_FRAME; + ref_frame[1] = LAST3_FRAME; + } + } else { + ref_frame[0] = LAST_FRAME; + ref_frame[1] = LAST2_FRAME; + } + } + + return; + } + + assert(comp_ref_type == BIDIR_COMP_REFERENCE); + + const int idx = 1; + const int bit = READ_REF_BIT(comp_ref_p); + // Decode forward references. + if (!bit) { + const int bit1 = READ_REF_BIT(comp_ref_p1); + ref_frame[!idx] = cm->comp_fwd_ref[bit1 ? 1 : 0]; + } else { + const int bit2 = READ_REF_BIT(comp_ref_p2); + ref_frame[!idx] = cm->comp_fwd_ref[bit2 ? 3 : 2]; + } + + // Decode backward references. + const int bit_bwd = READ_REF_BIT(comp_bwdref_p); + if (!bit_bwd) { + const int bit1_bwd = READ_REF_BIT(comp_bwdref_p1); + ref_frame[idx] = cm->comp_bwd_ref[bit1_bwd]; + } else { + ref_frame[idx] = cm->comp_bwd_ref[2]; + } + } else if (mode == SINGLE_REFERENCE) { + const int bit0 = READ_REF_BIT(single_ref_p1); + if (bit0) { + const int bit1 = READ_REF_BIT(single_ref_p2); + if (!bit1) { + const int bit5 = READ_REF_BIT(single_ref_p6); + ref_frame[0] = bit5 ? ALTREF2_FRAME : BWDREF_FRAME; + } else { + ref_frame[0] = ALTREF_FRAME; + } + } else { + const int bit2 = READ_REF_BIT(single_ref_p3); + if (bit2) { + const int bit4 = READ_REF_BIT(single_ref_p5); + ref_frame[0] = bit4 ? GOLDEN_FRAME : LAST3_FRAME; + } else { + const int bit3 = READ_REF_BIT(single_ref_p4); + ref_frame[0] = bit3 ? LAST2_FRAME : LAST_FRAME; + } + } + + ref_frame[1] = NONE_FRAME; + } else { + assert(0 && "Invalid prediction mode."); + } + } +} + +static INLINE void read_mb_interp_filter(AV1_COMMON *const cm, + MACROBLOCKD *const xd, + MB_MODE_INFO *const mbmi, + aom_reader *r) { + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + + if (!av1_is_interp_needed(xd)) { + set_default_interp_filters(mbmi, cm->interp_filter); + return; + } + + if (cm->interp_filter != SWITCHABLE) { + mbmi->interp_filters = av1_broadcast_interp_filter(cm->interp_filter); + } else { + InterpFilter ref0_filter[2] = { EIGHTTAP_REGULAR, EIGHTTAP_REGULAR }; + for (int dir = 0; dir < 2; ++dir) { + const int ctx = av1_get_pred_context_switchable_interp(xd, dir); + ref0_filter[dir] = (InterpFilter)aom_read_symbol( + r, ec_ctx->switchable_interp_cdf[ctx], SWITCHABLE_FILTERS, ACCT_STR); + if (cm->seq_params.enable_dual_filter == 0) { + ref0_filter[1] = ref0_filter[0]; + break; + } + } + // The index system works as: (0, 1) -> (vertical, horizontal) filter types + mbmi->interp_filters = + av1_make_interp_filters(ref0_filter[0], ref0_filter[1]); + } +} + +static void read_intra_block_mode_info(AV1_COMMON *const cm, const int mi_row, + const int mi_col, MACROBLOCKD *const xd, + MB_MODE_INFO *const mbmi, + aom_reader *r) { + const BLOCK_SIZE bsize = mbmi->sb_type; + const int use_angle_delta = av1_use_angle_delta(bsize); + + mbmi->ref_frame[0] = INTRA_FRAME; + mbmi->ref_frame[1] = NONE_FRAME; + + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + + mbmi->mode = read_intra_mode(r, ec_ctx->y_mode_cdf[size_group_lookup[bsize]]); + + mbmi->angle_delta[PLANE_TYPE_Y] = + use_angle_delta && av1_is_directional_mode(mbmi->mode) + ? read_angle_delta(r, ec_ctx->angle_delta_cdf[mbmi->mode - V_PRED]) + : 0; + const int has_chroma = + is_chroma_reference(mi_row, mi_col, bsize, xd->plane[1].subsampling_x, + xd->plane[1].subsampling_y); + xd->cfl.is_chroma_reference = has_chroma; + if (!cm->seq_params.monochrome && has_chroma) { + mbmi->uv_mode = + read_intra_mode_uv(ec_ctx, r, is_cfl_allowed(xd), mbmi->mode); + if (mbmi->uv_mode == UV_CFL_PRED) { + mbmi->cfl_alpha_idx = + read_cfl_alphas(xd->tile_ctx, r, &mbmi->cfl_alpha_signs); + } + mbmi->angle_delta[PLANE_TYPE_UV] = + use_angle_delta && av1_is_directional_mode(get_uv_mode(mbmi->uv_mode)) + ? read_angle_delta(r, + ec_ctx->angle_delta_cdf[mbmi->uv_mode - V_PRED]) + : 0; + } else { + // Avoid decoding angle_info if there is is no chroma prediction + mbmi->uv_mode = UV_DC_PRED; + } + xd->cfl.store_y = store_cfl_required(cm, xd); + + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + if (av1_allow_palette(cm->allow_screen_content_tools, bsize)) + read_palette_mode_info(cm, xd, mi_row, mi_col, r); + + read_filter_intra_mode_info(cm, xd, r); +} + +static INLINE int is_mv_valid(const MV *mv) { + return mv->row > MV_LOW && mv->row < MV_UPP && mv->col > MV_LOW && + mv->col < MV_UPP; +} + +static INLINE int assign_mv(AV1_COMMON *cm, MACROBLOCKD *xd, + PREDICTION_MODE mode, + MV_REFERENCE_FRAME ref_frame[2], int_mv mv[2], + int_mv ref_mv[2], int_mv nearest_mv[2], + int_mv near_mv[2], int mi_row, int mi_col, + int is_compound, int allow_hp, aom_reader *r) { + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + MB_MODE_INFO *mbmi = xd->mi[0]; + BLOCK_SIZE bsize = mbmi->sb_type; + if (cm->cur_frame_force_integer_mv) { + allow_hp = MV_SUBPEL_NONE; + } + switch (mode) { + case NEWMV: { + nmv_context *const nmvc = &ec_ctx->nmvc; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp); + break; + } + case NEARESTMV: { + mv[0].as_int = nearest_mv[0].as_int; + break; + } + case NEARMV: { + mv[0].as_int = near_mv[0].as_int; + break; + } + case GLOBALMV: { + mv[0].as_int = + gm_get_motion_vector(&cm->global_motion[ref_frame[0]], + cm->allow_high_precision_mv, bsize, mi_col, + mi_row, cm->cur_frame_force_integer_mv) + .as_int; + break; + } + case NEW_NEWMV: { + assert(is_compound); + for (int i = 0; i < 2; ++i) { + nmv_context *const nmvc = &ec_ctx->nmvc; + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, nmvc, allow_hp); + } + break; + } + case NEAREST_NEARESTMV: { + assert(is_compound); + mv[0].as_int = nearest_mv[0].as_int; + mv[1].as_int = nearest_mv[1].as_int; + break; + } + case NEAR_NEARMV: { + assert(is_compound); + mv[0].as_int = near_mv[0].as_int; + mv[1].as_int = near_mv[1].as_int; + break; + } + case NEW_NEARESTMV: { + nmv_context *const nmvc = &ec_ctx->nmvc; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp); + assert(is_compound); + mv[1].as_int = nearest_mv[1].as_int; + break; + } + case NEAREST_NEWMV: { + nmv_context *const nmvc = &ec_ctx->nmvc; + mv[0].as_int = nearest_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, allow_hp); + assert(is_compound); + break; + } + case NEAR_NEWMV: { + nmv_context *const nmvc = &ec_ctx->nmvc; + mv[0].as_int = near_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, nmvc, allow_hp); + assert(is_compound); + break; + } + case NEW_NEARMV: { + nmv_context *const nmvc = &ec_ctx->nmvc; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, nmvc, allow_hp); + assert(is_compound); + mv[1].as_int = near_mv[1].as_int; + break; + } + case GLOBAL_GLOBALMV: { + assert(is_compound); + mv[0].as_int = + gm_get_motion_vector(&cm->global_motion[ref_frame[0]], + cm->allow_high_precision_mv, bsize, mi_col, + mi_row, cm->cur_frame_force_integer_mv) + .as_int; + mv[1].as_int = + gm_get_motion_vector(&cm->global_motion[ref_frame[1]], + cm->allow_high_precision_mv, bsize, mi_col, + mi_row, cm->cur_frame_force_integer_mv) + .as_int; + break; + } + default: { return 0; } + } + + int ret = is_mv_valid(&mv[0].as_mv); + if (is_compound) { + ret = ret && is_mv_valid(&mv[1].as_mv); + } + return ret; +} + +static int read_is_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd, + int segment_id, aom_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) { + const int frame = get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME); + if (frame < LAST_FRAME) return 0; + return frame != INTRA_FRAME; + } + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_GLOBALMV)) { + return 1; + } + const int ctx = av1_get_intra_inter_context(xd); + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + const int is_inter = + aom_read_symbol(r, ec_ctx->intra_inter_cdf[ctx], 2, ACCT_STR); + return is_inter; +} + +#if DEC_MISMATCH_DEBUG +static void dec_dump_logs(AV1_COMMON *cm, MB_MODE_INFO *const mbmi, int mi_row, + int mi_col, int16_t mode_ctx) { + int_mv mv[2] = { { 0 } }; + for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) + mv[ref].as_mv = mbmi->mv[ref].as_mv; + + const int16_t newmv_ctx = mode_ctx & NEWMV_CTX_MASK; + int16_t zeromv_ctx = -1; + int16_t refmv_ctx = -1; + if (mbmi->mode != NEWMV) { + zeromv_ctx = (mode_ctx >> GLOBALMV_OFFSET) & GLOBALMV_CTX_MASK; + if (mbmi->mode != GLOBALMV) + refmv_ctx = (mode_ctx >> REFMV_OFFSET) & REFMV_CTX_MASK; + } + +#define FRAME_TO_CHECK 11 + if (cm->current_video_frame == FRAME_TO_CHECK && cm->show_frame == 1) { + printf( + "=== DECODER ===: " + "Frame=%d, (mi_row,mi_col)=(%d,%d), skip_mode=%d, mode=%d, bsize=%d, " + "show_frame=%d, mv[0]=(%d,%d), mv[1]=(%d,%d), ref[0]=%d, " + "ref[1]=%d, motion_mode=%d, mode_ctx=%d, " + "newmv_ctx=%d, zeromv_ctx=%d, refmv_ctx=%d, tx_size=%d\n", + cm->current_video_frame, mi_row, mi_col, mbmi->skip_mode, mbmi->mode, + mbmi->sb_type, cm->show_frame, mv[0].as_mv.row, mv[0].as_mv.col, + mv[1].as_mv.row, mv[1].as_mv.col, mbmi->ref_frame[0], + mbmi->ref_frame[1], mbmi->motion_mode, mode_ctx, newmv_ctx, zeromv_ctx, + refmv_ctx, mbmi->tx_size); + } +} +#endif // DEC_MISMATCH_DEBUG + +static void read_inter_block_mode_info(AV1Decoder *const pbi, + MACROBLOCKD *const xd, + MB_MODE_INFO *const mbmi, int mi_row, + int mi_col, aom_reader *r) { + AV1_COMMON *const cm = &pbi->common; + const BLOCK_SIZE bsize = mbmi->sb_type; + const int allow_hp = cm->allow_high_precision_mv; + int_mv nearestmv[2], nearmv[2]; + int_mv ref_mvs[MODE_CTX_REF_FRAMES][MAX_MV_REF_CANDIDATES] = { { { 0 } } }; + int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES]; + int pts[SAMPLES_ARRAY_SIZE], pts_inref[SAMPLES_ARRAY_SIZE]; + FRAME_CONTEXT *ec_ctx = xd->tile_ctx; + + mbmi->uv_mode = UV_DC_PRED; + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + + av1_collect_neighbors_ref_counts(xd); + + read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame); + const int is_compound = has_second_ref(mbmi); + + MV_REFERENCE_FRAME ref_frame = av1_ref_frame_type(mbmi->ref_frame); + int_mv global_mvs[REF_FRAMES]; + av1_find_mv_refs(cm, xd, mbmi, ref_frame, xd->ref_mv_count, xd->ref_mv_stack, + ref_mvs, global_mvs, mi_row, mi_col, inter_mode_ctx); + + int mode_ctx = av1_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame); + mbmi->ref_mv_idx = 0; + + if (mbmi->skip_mode) { + assert(is_compound); + mbmi->mode = NEAREST_NEARESTMV; + } else { + if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) || + segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_GLOBALMV)) { + mbmi->mode = GLOBALMV; + } else { + if (is_compound) + mbmi->mode = read_inter_compound_mode(xd, r, mode_ctx); + else + mbmi->mode = read_inter_mode(ec_ctx, r, mode_ctx); + if (mbmi->mode == NEWMV || mbmi->mode == NEW_NEWMV || + have_nearmv_in_inter_mode(mbmi->mode)) + read_drl_idx(ec_ctx, xd, mbmi, r); + } + } + + if (is_compound != is_inter_compound_mode(mbmi->mode)) { + aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, + "Prediction mode %d invalid with ref frame %d %d", + mbmi->mode, mbmi->ref_frame[0], mbmi->ref_frame[1]); + } + + if (!is_compound && mbmi->mode != GLOBALMV) { + av1_find_best_ref_mvs(allow_hp, ref_mvs[mbmi->ref_frame[0]], &nearestmv[0], + &nearmv[0], cm->cur_frame_force_integer_mv); + } + + if (is_compound && mbmi->mode != GLOBAL_GLOBALMV) { + int ref_mv_idx = mbmi->ref_mv_idx + 1; + nearestmv[0] = xd->ref_mv_stack[ref_frame][0].this_mv; + nearestmv[1] = xd->ref_mv_stack[ref_frame][0].comp_mv; + nearmv[0] = xd->ref_mv_stack[ref_frame][ref_mv_idx].this_mv; + nearmv[1] = xd->ref_mv_stack[ref_frame][ref_mv_idx].comp_mv; + lower_mv_precision(&nearestmv[0].as_mv, allow_hp, + cm->cur_frame_force_integer_mv); + lower_mv_precision(&nearestmv[1].as_mv, allow_hp, + cm->cur_frame_force_integer_mv); + lower_mv_precision(&nearmv[0].as_mv, allow_hp, + cm->cur_frame_force_integer_mv); + lower_mv_precision(&nearmv[1].as_mv, allow_hp, + cm->cur_frame_force_integer_mv); + } else if (mbmi->ref_mv_idx > 0 && mbmi->mode == NEARMV) { + int_mv cur_mv = + xd->ref_mv_stack[mbmi->ref_frame[0]][1 + mbmi->ref_mv_idx].this_mv; + nearmv[0] = cur_mv; + } + + int_mv ref_mv[2]; + ref_mv[0] = nearestmv[0]; + ref_mv[1] = nearestmv[1]; + + if (is_compound) { + int ref_mv_idx = mbmi->ref_mv_idx; + // Special case: NEAR_NEWMV and NEW_NEARMV modes use + // 1 + mbmi->ref_mv_idx (like NEARMV) instead of + // mbmi->ref_mv_idx (like NEWMV) + if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEW_NEARMV) + ref_mv_idx = 1 + mbmi->ref_mv_idx; + + // TODO(jingning, yunqing): Do we need a lower_mv_precision() call here? + if (compound_ref0_mode(mbmi->mode) == NEWMV) + ref_mv[0] = xd->ref_mv_stack[ref_frame][ref_mv_idx].this_mv; + + if (compound_ref1_mode(mbmi->mode) == NEWMV) + ref_mv[1] = xd->ref_mv_stack[ref_frame][ref_mv_idx].comp_mv; + } else { + if (mbmi->mode == NEWMV) { + if (xd->ref_mv_count[ref_frame] > 1) + ref_mv[0] = xd->ref_mv_stack[ref_frame][mbmi->ref_mv_idx].this_mv; + } + } + + if (mbmi->skip_mode) { + assert(mbmi->mode == NEAREST_NEARESTMV); + mbmi->mv[0].as_int = nearestmv[0].as_int; + mbmi->mv[1].as_int = nearestmv[1].as_int; + } else { + int mv_corrupted_flag = + !assign_mv(cm, xd, mbmi->mode, mbmi->ref_frame, mbmi->mv, ref_mv, + nearestmv, nearmv, mi_row, mi_col, is_compound, allow_hp, r); + aom_merge_corrupted_flag(&xd->corrupted, mv_corrupted_flag); + } + + mbmi->use_wedge_interintra = 0; + if (cm->seq_params.enable_interintra_compound && !mbmi->skip_mode && + is_interintra_allowed(mbmi)) { + const int bsize_group = size_group_lookup[bsize]; + const int interintra = + aom_read_symbol(r, ec_ctx->interintra_cdf[bsize_group], 2, ACCT_STR); + assert(mbmi->ref_frame[1] == NONE_FRAME); + if (interintra) { + const INTERINTRA_MODE interintra_mode = + read_interintra_mode(xd, r, bsize_group); + mbmi->ref_frame[1] = INTRA_FRAME; + mbmi->interintra_mode = interintra_mode; + mbmi->angle_delta[PLANE_TYPE_Y] = 0; + mbmi->angle_delta[PLANE_TYPE_UV] = 0; + mbmi->filter_intra_mode_info.use_filter_intra = 0; + if (is_interintra_wedge_used(bsize)) { + mbmi->use_wedge_interintra = aom_read_symbol( + r, ec_ctx->wedge_interintra_cdf[bsize], 2, ACCT_STR); + if (mbmi->use_wedge_interintra) { + mbmi->interintra_wedge_index = + aom_read_symbol(r, ec_ctx->wedge_idx_cdf[bsize], 16, ACCT_STR); + mbmi->interintra_wedge_sign = 0; + } + } + } + } + + for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) { + const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref]; + RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME]; + + xd->block_refs[ref] = ref_buf; + } + + mbmi->motion_mode = SIMPLE_TRANSLATION; + if (is_motion_variation_allowed_bsize(mbmi->sb_type) && !mbmi->skip_mode && + !has_second_ref(mbmi)) + mbmi->num_proj_ref = findSamples(cm, xd, mi_row, mi_col, pts, pts_inref); + av1_count_overlappable_neighbors(cm, xd, mi_row, mi_col); + + if (mbmi->ref_frame[1] != INTRA_FRAME) + mbmi->motion_mode = read_motion_mode(cm, xd, mbmi, r); + + // init + mbmi->comp_group_idx = 0; + mbmi->compound_idx = 1; + mbmi->interinter_comp.type = COMPOUND_AVERAGE; + + if (has_second_ref(mbmi) && !mbmi->skip_mode) { + // Read idx to indicate current compound inter prediction mode group + const int masked_compound_used = is_any_masked_compound_used(bsize) && + cm->seq_params.enable_masked_compound; + + if (masked_compound_used) { + const int ctx_comp_group_idx = get_comp_group_idx_context(xd); + mbmi->comp_group_idx = aom_read_symbol( + r, ec_ctx->comp_group_idx_cdf[ctx_comp_group_idx], 2, ACCT_STR); + } + + if (mbmi->comp_group_idx == 0) { + if (cm->seq_params.enable_jnt_comp) { + const int comp_index_ctx = get_comp_index_context(cm, xd); + mbmi->compound_idx = aom_read_symbol( + r, ec_ctx->compound_index_cdf[comp_index_ctx], 2, ACCT_STR); + } else { + // Distance-weighted compound is disabled, so always use average + mbmi->compound_idx = 1; + } + } else { + assert(cm->reference_mode != SINGLE_REFERENCE && + is_inter_compound_mode(mbmi->mode) && + mbmi->motion_mode == SIMPLE_TRANSLATION); + assert(masked_compound_used); + + // compound_diffwtd, wedge + if (is_interinter_compound_used(COMPOUND_WEDGE, bsize)) + mbmi->interinter_comp.type = + 1 + aom_read_symbol(r, ec_ctx->compound_type_cdf[bsize], + COMPOUND_TYPES - 1, ACCT_STR); + else + mbmi->interinter_comp.type = COMPOUND_DIFFWTD; + + if (mbmi->interinter_comp.type == COMPOUND_WEDGE) { + assert(is_interinter_compound_used(COMPOUND_WEDGE, bsize)); + mbmi->interinter_comp.wedge_index = + aom_read_symbol(r, ec_ctx->wedge_idx_cdf[bsize], 16, ACCT_STR); + mbmi->interinter_comp.wedge_sign = aom_read_bit(r, ACCT_STR); + } else { + assert(mbmi->interinter_comp.type == COMPOUND_DIFFWTD); + mbmi->interinter_comp.mask_type = + aom_read_literal(r, MAX_DIFFWTD_MASK_BITS, ACCT_STR); + } + } + } + + read_mb_interp_filter(cm, xd, mbmi, r); + + if (mbmi->motion_mode == WARPED_CAUSAL) { + mbmi->wm_params.wmtype = DEFAULT_WMTYPE; + mbmi->wm_params.invalid = 0; + + if (mbmi->num_proj_ref > 1) + mbmi->num_proj_ref = selectSamples(&mbmi->mv[0].as_mv, pts, pts_inref, + mbmi->num_proj_ref, bsize); + + if (find_projection(mbmi->num_proj_ref, pts, pts_inref, bsize, + mbmi->mv[0].as_mv.row, mbmi->mv[0].as_mv.col, + &mbmi->wm_params, mi_row, mi_col)) { +#if WARPED_MOTION_DEBUG + printf("Warning: unexpected warped model from aomenc\n"); +#endif + mbmi->wm_params.invalid = 1; + } + } + + xd->cfl.is_chroma_reference = + is_chroma_reference(mi_row, mi_col, bsize, cm->seq_params.subsampling_x, + cm->seq_params.subsampling_y); + xd->cfl.store_y = store_cfl_required(cm, xd); + +#if DEC_MISMATCH_DEBUG + dec_dump_logs(cm, mi, mi_row, mi_col, mode_ctx); +#endif // DEC_MISMATCH_DEBUG +} + +static void read_inter_frame_mode_info(AV1Decoder *const pbi, + MACROBLOCKD *const xd, int mi_row, + int mi_col, aom_reader *r) { + AV1_COMMON *const cm = &pbi->common; + MB_MODE_INFO *const mbmi = xd->mi[0]; + int inter_block = 1; + + mbmi->mv[0].as_int = 0; + mbmi->mv[1].as_int = 0; + mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 1, r); + + mbmi->skip_mode = read_skip_mode(cm, xd, mbmi->segment_id, r); + + if (mbmi->skip_mode) + mbmi->skip = 1; + else + mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r); + + mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, 0, r); + + read_cdef(cm, r, xd, mi_col, mi_row); + + read_delta_q_params(cm, xd, mi_row, mi_col, r); + + if (!mbmi->skip_mode) + inter_block = read_is_inter_block(cm, xd, mbmi->segment_id, r); + + mbmi->current_qindex = xd->current_qindex; + + xd->above_txfm_context = cm->above_txfm_context[xd->tile.tile_row] + mi_col; + xd->left_txfm_context = + xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK); + + if (inter_block) + read_inter_block_mode_info(pbi, xd, mbmi, mi_row, mi_col, r); + else + read_intra_block_mode_info(cm, mi_row, mi_col, xd, mbmi, r); +} + +static void intra_copy_frame_mvs(AV1_COMMON *const cm, int mi_row, int mi_col, + int x_mis, int y_mis) { + const int frame_mvs_stride = ROUND_POWER_OF_TWO(cm->mi_cols, 1); + MV_REF *frame_mvs = + cm->cur_frame->mvs + (mi_row >> 1) * frame_mvs_stride + (mi_col >> 1); + x_mis = ROUND_POWER_OF_TWO(x_mis, 1); + y_mis = ROUND_POWER_OF_TWO(y_mis, 1); + + for (int h = 0; h < y_mis; h++) { + MV_REF *mv = frame_mvs; + for (int w = 0; w < x_mis; w++) { + mv->ref_frame = NONE_FRAME; + mv++; + } + frame_mvs += frame_mvs_stride; + } +} + +void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd, int mi_row, + int mi_col, aom_reader *r, int x_mis, int y_mis) { + AV1_COMMON *const cm = &pbi->common; + MB_MODE_INFO *const mi = xd->mi[0]; + mi->use_intrabc = 0; + + if (frame_is_intra_only(cm)) { + read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r); + intra_copy_frame_mvs(cm, mi_row, mi_col, x_mis, y_mis); + } else { + read_inter_frame_mode_info(pbi, xd, mi_row, mi_col, r); + av1_copy_frame_mvs(cm, mi, mi_row, mi_col, x_mis, y_mis); + } +} diff --git a/media/libaom/src/av1/decoder/decodemv.h b/media/libaom/src/av1/decoder/decodemv.h new file mode 100644 index 000000000..1625e5bd2 --- /dev/null +++ b/media/libaom/src/av1/decoder/decodemv.h @@ -0,0 +1,35 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#ifndef AOM_AV1_DECODER_DECODEMV_H_ +#define AOM_AV1_DECODER_DECODEMV_H_ + +#include "aom_dsp/bitreader.h" + +#include "av1/decoder/decoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void av1_read_mode_info(AV1Decoder *const pbi, MACROBLOCKD *xd, + + int mi_row, int mi_col, aom_reader *r, int x_mis, + int y_mis); + +#ifdef __cplusplus +} // extern "C" +#endif + +void av1_read_tx_type(const AV1_COMMON *const cm, MACROBLOCKD *xd, int blk_row, + int blk_col, TX_SIZE tx_size, aom_reader *r); + +#endif // AOM_AV1_DECODER_DECODEMV_H_ diff --git a/media/libaom/src/av1/decoder/decoder.c b/media/libaom/src/av1/decoder/decoder.c new file mode 100644 index 000000000..a5f4fd67f --- /dev/null +++ b/media/libaom/src/av1/decoder/decoder.c @@ -0,0 +1,575 @@ +/* + * 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 <limits.h> +#include <stdio.h> + +#include "config/av1_rtcd.h" +#include "config/aom_dsp_rtcd.h" +#include "config/aom_scale_rtcd.h" + +#include "aom_mem/aom_mem.h" +#include "aom_ports/system_state.h" +#include "aom_ports/aom_once.h" +#include "aom_ports/aom_timer.h" +#include "aom_scale/aom_scale.h" +#include "aom_util/aom_thread.h" + +#include "av1/common/alloccommon.h" +#include "av1/common/av1_loopfilter.h" +#include "av1/common/onyxc_int.h" +#include "av1/common/quant_common.h" +#include "av1/common/reconinter.h" +#include "av1/common/reconintra.h" + +#include "av1/decoder/decodeframe.h" +#include "av1/decoder/decoder.h" +#include "av1/decoder/detokenize.h" +#include "av1/decoder/obu.h" + +static void initialize_dec(void) { + av1_rtcd(); + aom_dsp_rtcd(); + aom_scale_rtcd(); + av1_init_intra_predictors(); + av1_init_wedge_masks(); +} + +static void dec_setup_mi(AV1_COMMON *cm) { + cm->mi = cm->mip; + cm->mi_grid_visible = cm->mi_grid_base; + memset(cm->mi_grid_base, 0, + cm->mi_stride * cm->mi_rows * sizeof(*cm->mi_grid_base)); +} + +static int av1_dec_alloc_mi(AV1_COMMON *cm, int mi_size) { + cm->mip = aom_calloc(mi_size, sizeof(*cm->mip)); + if (!cm->mip) return 1; + cm->mi_alloc_size = mi_size; + cm->mi_grid_base = + (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *)); + if (!cm->mi_grid_base) return 1; + return 0; +} + +static void dec_free_mi(AV1_COMMON *cm) { + aom_free(cm->mip); + cm->mip = NULL; + aom_free(cm->mi_grid_base); + cm->mi_grid_base = NULL; + cm->mi_alloc_size = 0; +} + +AV1Decoder *av1_decoder_create(BufferPool *const pool) { + AV1Decoder *volatile const pbi = aom_memalign(32, sizeof(*pbi)); + AV1_COMMON *volatile const cm = pbi ? &pbi->common : NULL; + + if (!cm) return NULL; + + av1_zero(*pbi); + + // The jmp_buf is valid only for the duration of the function that calls + // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 + // before it returns. + if (setjmp(cm->error.jmp)) { + cm->error.setjmp = 0; + av1_decoder_remove(pbi); + return NULL; + } + + cm->error.setjmp = 1; + + CHECK_MEM_ERROR(cm, cm->fc, + (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc))); + CHECK_MEM_ERROR(cm, cm->frame_contexts, + (FRAME_CONTEXT *)aom_memalign( + 32, FRAME_CONTEXTS * sizeof(*cm->frame_contexts))); + memset(cm->fc, 0, sizeof(*cm->fc)); + memset(cm->frame_contexts, 0, FRAME_CONTEXTS * sizeof(*cm->frame_contexts)); + + pbi->need_resync = 1; + aom_once(initialize_dec); + + // Initialize the references to not point to any frame buffers. + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map)); + + cm->current_video_frame = 0; + pbi->decoding_first_frame = 1; + pbi->common.buffer_pool = pool; + + cm->seq_params.bit_depth = AOM_BITS_8; + cm->dequant_bit_depth = AOM_BITS_8; + + cm->alloc_mi = av1_dec_alloc_mi; + cm->free_mi = dec_free_mi; + cm->setup_mi = dec_setup_mi; + + av1_loop_filter_init(cm); + + av1_qm_init(cm); + av1_loop_restoration_precal(); +#if CONFIG_ACCOUNTING + pbi->acct_enabled = 1; + aom_accounting_init(&pbi->accounting); +#endif + + cm->error.setjmp = 0; + + aom_get_worker_interface()->init(&pbi->lf_worker); + + return pbi; +} + +void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_mt_info) { + if (tile_mt_info != NULL) { +#if CONFIG_MULTITHREAD + if (tile_mt_info->job_mutex != NULL) { + pthread_mutex_destroy(tile_mt_info->job_mutex); + aom_free(tile_mt_info->job_mutex); + } +#endif + aom_free(tile_mt_info->job_queue); + // clear the structure as the source of this call may be a resize in which + // case this call will be followed by an _alloc() which may fail. + av1_zero(*tile_mt_info); + } +} + +void av1_dec_free_cb_buf(AV1Decoder *pbi) { + aom_free(pbi->cb_buffer_base); + pbi->cb_buffer_base = NULL; + pbi->cb_buffer_alloc_size = 0; +} + +void av1_decoder_remove(AV1Decoder *pbi) { + int i; + + if (!pbi) return; + + // Free the tile list output buffer. + if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output); + pbi->tile_list_output = NULL; + + aom_get_worker_interface()->end(&pbi->lf_worker); + aom_free(pbi->lf_worker.data1); + + if (pbi->thread_data) { + for (int worker_idx = 0; worker_idx < pbi->max_threads - 1; worker_idx++) { + DecWorkerData *const thread_data = pbi->thread_data + worker_idx; + av1_free_mc_tmp_buf(thread_data->td); + aom_free(thread_data->td); + } + aom_free(pbi->thread_data); + } + + for (i = 0; i < pbi->num_workers; ++i) { + AVxWorker *const worker = &pbi->tile_workers[i]; + aom_get_worker_interface()->end(worker); + } +#if CONFIG_MULTITHREAD + if (pbi->row_mt_mutex_ != NULL) { + pthread_mutex_destroy(pbi->row_mt_mutex_); + aom_free(pbi->row_mt_mutex_); + } + if (pbi->row_mt_cond_ != NULL) { + pthread_cond_destroy(pbi->row_mt_cond_); + aom_free(pbi->row_mt_cond_); + } +#endif + for (i = 0; i < pbi->allocated_tiles; i++) { + TileDataDec *const tile_data = pbi->tile_data + i; + av1_dec_row_mt_dealloc(&tile_data->dec_row_mt_sync); + } + aom_free(pbi->tile_data); + aom_free(pbi->tile_workers); + + if (pbi->num_workers > 0) { + av1_loop_filter_dealloc(&pbi->lf_row_sync); + av1_loop_restoration_dealloc(&pbi->lr_row_sync, pbi->num_workers); + av1_dealloc_dec_jobs(&pbi->tile_mt_info); + } + + av1_dec_free_cb_buf(pbi); +#if CONFIG_ACCOUNTING + aom_accounting_clear(&pbi->accounting); +#endif + av1_free_mc_tmp_buf(&pbi->td); + + aom_free(pbi); +} + +void av1_visit_palette(AV1Decoder *const pbi, MACROBLOCKD *const xd, int mi_row, + int mi_col, aom_reader *r, BLOCK_SIZE bsize, + palette_visitor_fn_t visit) { + if (!is_inter_block(xd->mi[0])) { + for (int plane = 0; plane < AOMMIN(2, av1_num_planes(&pbi->common)); + ++plane) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + if (is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x, + pd->subsampling_y)) { + if (xd->mi[0]->palette_mode_info.palette_size[plane]) + visit(xd, plane, r); + } else { + assert(xd->mi[0]->palette_mode_info.palette_size[plane] == 0); + } + } + } +} + +static int equal_dimensions(const YV12_BUFFER_CONFIG *a, + const YV12_BUFFER_CONFIG *b) { + return a->y_height == b->y_height && a->y_width == b->y_width && + a->uv_height == b->uv_height && a->uv_width == b->uv_width; +} + +aom_codec_err_t av1_copy_reference_dec(AV1Decoder *pbi, int idx, + YV12_BUFFER_CONFIG *sd) { + AV1_COMMON *cm = &pbi->common; + const int num_planes = av1_num_planes(cm); + + const YV12_BUFFER_CONFIG *const cfg = get_ref_frame(cm, idx); + if (cfg == NULL) { + aom_internal_error(&cm->error, AOM_CODEC_ERROR, "No reference frame"); + return AOM_CODEC_ERROR; + } + if (!equal_dimensions(cfg, sd)) + aom_internal_error(&cm->error, AOM_CODEC_ERROR, + "Incorrect buffer dimensions"); + else + aom_yv12_copy_frame(cfg, sd, num_planes); + + return cm->error.error_code; +} + +static int equal_dimensions_and_border(const YV12_BUFFER_CONFIG *a, + const YV12_BUFFER_CONFIG *b) { + return a->y_height == b->y_height && a->y_width == b->y_width && + a->uv_height == b->uv_height && a->uv_width == b->uv_width && + a->y_stride == b->y_stride && a->uv_stride == b->uv_stride && + a->border == b->border && + (a->flags & YV12_FLAG_HIGHBITDEPTH) == + (b->flags & YV12_FLAG_HIGHBITDEPTH); +} + +aom_codec_err_t av1_set_reference_dec(AV1_COMMON *cm, int idx, + int use_external_ref, + YV12_BUFFER_CONFIG *sd) { + const int num_planes = av1_num_planes(cm); + YV12_BUFFER_CONFIG *ref_buf = NULL; + + // Get the destination reference buffer. + ref_buf = get_ref_frame(cm, idx); + + if (ref_buf == NULL) { + aom_internal_error(&cm->error, AOM_CODEC_ERROR, "No reference frame"); + return AOM_CODEC_ERROR; + } + + if (!use_external_ref) { + if (!equal_dimensions(ref_buf, sd)) { + aom_internal_error(&cm->error, AOM_CODEC_ERROR, + "Incorrect buffer dimensions"); + } else { + // Overwrite the reference frame buffer. + aom_yv12_copy_frame(sd, ref_buf, num_planes); + } + } else { + if (!equal_dimensions_and_border(ref_buf, sd)) { + aom_internal_error(&cm->error, AOM_CODEC_ERROR, + "Incorrect buffer dimensions"); + } else { + // Overwrite the reference frame buffer pointers. + // Once we no longer need the external reference buffer, these pointers + // are restored. + ref_buf->store_buf_adr[0] = ref_buf->y_buffer; + ref_buf->store_buf_adr[1] = ref_buf->u_buffer; + ref_buf->store_buf_adr[2] = ref_buf->v_buffer; + ref_buf->y_buffer = sd->y_buffer; + ref_buf->u_buffer = sd->u_buffer; + ref_buf->v_buffer = sd->v_buffer; + ref_buf->use_external_reference_buffers = 1; + } + } + + return cm->error.error_code; +} + +aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm, + YV12_BUFFER_CONFIG *new_frame, + YV12_BUFFER_CONFIG *sd) { + const int num_planes = av1_num_planes(cm); + + if (!equal_dimensions_and_border(new_frame, sd)) + aom_internal_error(&cm->error, AOM_CODEC_ERROR, + "Incorrect buffer dimensions"); + else + aom_yv12_copy_frame(new_frame, sd, num_planes); + + return cm->error.error_code; +} + +/* If any buffer updating is signaled it should be done here. + Consumes a reference to cm->new_fb_idx. +*/ +static void swap_frame_buffers(AV1Decoder *pbi, int frame_decoded) { + int ref_index = 0, mask; + AV1_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + + if (frame_decoded) { + lock_buffer_pool(pool); + + // In ext-tile decoding, the camera frame header is only decoded once. So, + // we don't release the references here. + if (!pbi->camera_frame_header_ready) { + for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + const int old_idx = cm->ref_frame_map[ref_index]; + // Current thread releases the holding of reference frame. + decrease_ref_count(old_idx, frame_bufs, pool); + + // Release the reference frame holding in the reference map for the + // decoding of the next frame. + if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool); + cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index]; + ++ref_index; + } + + // Current thread releases the holding of reference frame. + const int check_on_show_existing_frame = + !cm->show_existing_frame || cm->reset_decoder_state; + for (; ref_index < REF_FRAMES && check_on_show_existing_frame; + ++ref_index) { + const int old_idx = cm->ref_frame_map[ref_index]; + decrease_ref_count(old_idx, frame_bufs, pool); + cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index]; + } + } + + YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm); + + if (cm->show_existing_frame || cm->show_frame) { + if (pbi->output_all_layers) { + // Append this frame to the output queue + if (pbi->num_output_frames >= MAX_NUM_SPATIAL_LAYERS) { + // We can't store the new frame anywhere, so drop it and return an + // error + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + } else { + pbi->output_frames[pbi->num_output_frames] = cur_frame; + pbi->output_frame_index[pbi->num_output_frames] = cm->new_fb_idx; + pbi->num_output_frames++; + } + } else { + // Replace any existing output frame + assert(pbi->num_output_frames == 0 || pbi->num_output_frames == 1); + if (pbi->num_output_frames > 0) { + decrease_ref_count((int)pbi->output_frame_index[0], frame_bufs, pool); + } + pbi->output_frames[0] = cur_frame; + pbi->output_frame_index[0] = cm->new_fb_idx; + pbi->num_output_frames = 1; + } + } else { + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + } + + unlock_buffer_pool(pool); + } else { + // Nothing was decoded, so just drop this frame buffer + lock_buffer_pool(pool); + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + unlock_buffer_pool(pool); + } + + if (!pbi->camera_frame_header_ready) { + pbi->hold_ref_buf = 0; + + // Invalidate these references until the next frame starts. + for (ref_index = 0; ref_index < INTER_REFS_PER_FRAME; ref_index++) { + cm->frame_refs[ref_index].idx = INVALID_IDX; + cm->frame_refs[ref_index].buf = NULL; + } + } +} + +int av1_receive_compressed_data(AV1Decoder *pbi, size_t size, + const uint8_t **psource) { + AV1_COMMON *volatile const cm = &pbi->common; + BufferPool *volatile const pool = cm->buffer_pool; + RefCntBuffer *volatile const frame_bufs = cm->buffer_pool->frame_bufs; + const uint8_t *source = *psource; + cm->error.error_code = AOM_CODEC_OK; + + if (size == 0) { + // This is used to signal that we are missing frames. + // We do not know if the missing frame(s) was supposed to update + // any of the reference buffers, but we act conservative and + // mark only the last buffer as corrupted. + // + // TODO(jkoleszar): Error concealment is undefined and non-normative + // at this point, but if it becomes so, [0] may not always be the correct + // thing to do here. + if (cm->frame_refs[0].idx > 0) { + assert(cm->frame_refs[0].buf != NULL); + cm->frame_refs[0].buf->corrupted = 1; + } + } + + // Find a free buffer for the new frame, releasing the reference previously + // held. + + // Find a free frame buffer. Return error if can not find any. + cm->new_fb_idx = get_free_fb(cm); + if (cm->new_fb_idx == INVALID_IDX) { + cm->error.error_code = AOM_CODEC_MEM_ERROR; + return 1; + } + + // Assign a MV array to the frame buffer. + cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx]; + + if (!pbi->camera_frame_header_ready) pbi->hold_ref_buf = 0; + + pbi->cur_buf = &frame_bufs[cm->new_fb_idx]; + + // The jmp_buf is valid only for the duration of the function that calls + // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 + // before it returns. + if (setjmp(cm->error.jmp)) { + const AVxWorkerInterface *const winterface = aom_get_worker_interface(); + int i; + + cm->error.setjmp = 0; + + // Synchronize all threads immediately as a subsequent decode call may + // cause a resize invalidating some allocations. + winterface->sync(&pbi->lf_worker); + for (i = 0; i < pbi->num_workers; ++i) { + winterface->sync(&pbi->tile_workers[i]); + } + + lock_buffer_pool(pool); + // Release all the reference buffers if worker thread is holding them. + if (pbi->hold_ref_buf == 1) { + int ref_index = 0, mask; + for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) { + const int old_idx = cm->ref_frame_map[ref_index]; + // Current thread releases the holding of reference frame. + decrease_ref_count(old_idx, frame_bufs, pool); + + // Release the reference frame holding in the reference map for the + // decoding of the next frame. + if (mask & 1) decrease_ref_count(old_idx, frame_bufs, pool); + ++ref_index; + } + + // Current thread releases the holding of reference frame. + const int check_on_show_existing_frame = + !cm->show_existing_frame || cm->reset_decoder_state; + for (; ref_index < REF_FRAMES && check_on_show_existing_frame; + ++ref_index) { + const int old_idx = cm->ref_frame_map[ref_index]; + decrease_ref_count(old_idx, frame_bufs, pool); + } + pbi->hold_ref_buf = 0; + } + // Release current frame. + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + unlock_buffer_pool(pool); + + aom_clear_system_state(); + return -1; + } + + cm->error.setjmp = 1; + + int frame_decoded = + aom_decode_frame_from_obus(pbi, source, source + size, psource); + + if (cm->error.error_code != AOM_CODEC_OK) { + lock_buffer_pool(pool); + decrease_ref_count(cm->new_fb_idx, frame_bufs, pool); + unlock_buffer_pool(pool); + cm->error.setjmp = 0; + return 1; + } + +#if TXCOEFF_TIMER + cm->cum_txcoeff_timer += cm->txcoeff_timer; + fprintf(stderr, + "txb coeff block number: %d, frame time: %ld, cum time %ld in us\n", + cm->txb_count, cm->txcoeff_timer, cm->cum_txcoeff_timer); + cm->txcoeff_timer = 0; + cm->txb_count = 0; +#endif + + // Note: At this point, this function holds a reference to cm->new_fb_idx + // in the buffer pool. This reference is consumed by swap_frame_buffers(). + swap_frame_buffers(pbi, frame_decoded); + + if (frame_decoded) { + pbi->decoding_first_frame = 0; + } + + if (cm->error.error_code != AOM_CODEC_OK) { + cm->error.setjmp = 0; + return 1; + } + + aom_clear_system_state(); + + if (!cm->show_existing_frame) { + cm->last_show_frame = cm->show_frame; + + if (cm->seg.enabled) { + if (cm->prev_frame && (cm->mi_rows == cm->prev_frame->mi_rows) && + (cm->mi_cols == cm->prev_frame->mi_cols)) { + cm->last_frame_seg_map = cm->prev_frame->seg_map; + } else { + cm->last_frame_seg_map = NULL; + } + } + } + + // Update progress in frame parallel decode. + cm->last_width = cm->width; + cm->last_height = cm->height; + cm->last_tile_cols = cm->tile_cols; + cm->last_tile_rows = cm->tile_rows; + cm->error.setjmp = 0; + + return 0; +} + +// Get the frame at a particular index in the output queue +int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd, + aom_film_grain_t **grain_params) { + RefCntBuffer *const frame_bufs = pbi->common.buffer_pool->frame_bufs; + + if (index >= pbi->num_output_frames) return -1; + *sd = pbi->output_frames[index]; + *grain_params = &frame_bufs[pbi->output_frame_index[index]].film_grain_params; + aom_clear_system_state(); + return 0; +} + +// Get the highest-spatial-layer output +// TODO(david.barker): What should this do? +int av1_get_frame_to_show(AV1Decoder *pbi, YV12_BUFFER_CONFIG *frame) { + if (pbi->num_output_frames == 0) return -1; + + *frame = *pbi->output_frames[pbi->num_output_frames - 1]; + return 0; +} diff --git a/media/libaom/src/av1/decoder/decoder.h b/media/libaom/src/av1/decoder/decoder.h new file mode 100644 index 000000000..5ca939c24 --- /dev/null +++ b/media/libaom/src/av1/decoder/decoder.h @@ -0,0 +1,317 @@ +/* + * 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 AOM_AV1_DECODER_DECODER_H_ +#define AOM_AV1_DECODER_DECODER_H_ + +#include "config/aom_config.h" + +#include "aom/aom_codec.h" +#include "aom_dsp/bitreader.h" +#include "aom_scale/yv12config.h" +#include "aom_util/aom_thread.h" + +#include "av1/common/thread_common.h" +#include "av1/common/onyxc_int.h" +#include "av1/decoder/dthread.h" +#if CONFIG_ACCOUNTING +#include "av1/decoder/accounting.h" +#endif +#if CONFIG_INSPECTION +#include "av1/decoder/inspection.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +typedef void (*decode_block_visitor_fn_t)(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); + +typedef void (*predict_inter_block_visitor_fn_t)(AV1_COMMON *const cm, + MACROBLOCKD *const xd, + int mi_row, int mi_col, + BLOCK_SIZE bsize); + +typedef void (*cfl_store_inter_block_visitor_fn_t)(AV1_COMMON *const cm, + MACROBLOCKD *const xd); + +typedef struct ThreadData { + aom_reader *bit_reader; + DECLARE_ALIGNED(32, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(32, tran_low_t, dqcoeff[MAX_TX_SQUARE]); + CB_BUFFER cb_buffer_base; + uint8_t *mc_buf[2]; + int32_t mc_buf_size; + int mc_buf_use_highbd; // Boolean: whether the byte pointers stored in + // mc_buf were converted from highbd pointers. + + CONV_BUF_TYPE *tmp_conv_dst; + uint8_t *tmp_obmc_bufs[2]; + + decode_block_visitor_fn_t read_coeffs_tx_intra_block_visit; + decode_block_visitor_fn_t predict_and_recon_intra_block_visit; + decode_block_visitor_fn_t read_coeffs_tx_inter_block_visit; + decode_block_visitor_fn_t inverse_tx_inter_block_visit; + predict_inter_block_visitor_fn_t predict_inter_block_visit; + cfl_store_inter_block_visitor_fn_t cfl_store_inter_block_visit; +} ThreadData; + +typedef struct AV1DecRowMTJobInfo { + int tile_row; + int tile_col; + int mi_row; +} AV1DecRowMTJobInfo; + +typedef struct AV1DecRowMTSyncData { +#if CONFIG_MULTITHREAD + pthread_mutex_t *mutex_; + pthread_cond_t *cond_; +#endif + int allocated_sb_rows; + int *cur_sb_col; + int sync_range; + int mi_rows; + int mi_cols; + int mi_rows_parse_done; + int mi_rows_decode_started; + int num_threads_working; +} AV1DecRowMTSync; + +typedef struct AV1DecRowMTInfo { + int tile_rows_start; + int tile_rows_end; + int tile_cols_start; + int tile_cols_end; + int start_tile; + int end_tile; + int mi_rows_parse_done; + int mi_rows_decode_started; + int mi_rows_to_decode; + int row_mt_exit; +} AV1DecRowMTInfo; + +typedef struct TileDataDec { + TileInfo tile_info; + aom_reader bit_reader; + DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx); + AV1DecRowMTSync dec_row_mt_sync; +} TileDataDec; + +typedef struct TileBufferDec { + const uint8_t *data; + size_t size; +} TileBufferDec; + +typedef struct DataBuffer { + const uint8_t *data; + size_t size; +} DataBuffer; + +typedef struct EXTERNAL_REFERENCES { + YV12_BUFFER_CONFIG refs[MAX_EXTERNAL_REFERENCES]; + int num; +} EXTERNAL_REFERENCES; + +typedef struct TileJobsDec { + TileBufferDec *tile_buffer; + TileDataDec *tile_data; +} TileJobsDec; + +typedef struct AV1DecTileMTData { +#if CONFIG_MULTITHREAD + pthread_mutex_t *job_mutex; +#endif + TileJobsDec *job_queue; + int jobs_enqueued; + int jobs_dequeued; + int alloc_tile_rows; + int alloc_tile_cols; +} AV1DecTileMT; + +typedef struct AV1Decoder { + DECLARE_ALIGNED(32, MACROBLOCKD, mb); + + DECLARE_ALIGNED(32, AV1_COMMON, common); + + int refresh_frame_flags; + + // TODO(hkuang): Combine this with cur_buf in macroblockd as they are + // the same. + RefCntBuffer *cur_buf; // Current decoding frame buffer. + + AVxWorker *frame_worker_owner; // frame_worker that owns this pbi. + AVxWorker lf_worker; + AV1LfSync lf_row_sync; + AV1LrSync lr_row_sync; + AV1LrStruct lr_ctxt; + AVxWorker *tile_workers; + int num_workers; + DecWorkerData *thread_data; + ThreadData td; + TileDataDec *tile_data; + int allocated_tiles; + + TileBufferDec tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS]; + AV1DecTileMT tile_mt_info; + + // Each time the decoder is called, we expect to receive a full temporal unit. + // This can contain up to one shown frame per spatial layer in the current + // operating point (note that some layers may be entirely omitted). + // If the 'output_all_layers' option is true, we save all of these shown + // frames so that they can be returned to the application. If the + // 'output_all_layers' option is false, then we only output one image per + // temporal unit. + // + // Note: The saved buffers are released at the start of the next time the + // application calls aom_codec_decode(). + int output_all_layers; + YV12_BUFFER_CONFIG *output_frames[MAX_NUM_SPATIAL_LAYERS]; + size_t output_frame_index[MAX_NUM_SPATIAL_LAYERS]; // Buffer pool indices + size_t num_output_frames; // How many frames are queued up so far? + + // In order to properly support random-access decoding, we need + // to behave slightly differently for the very first frame we decode. + // So we track whether this is the first frame or not. + int decoding_first_frame; + + int allow_lowbitdepth; + int max_threads; + int inv_tile_order; + int need_resync; // wait for key/intra-only frame. + int hold_ref_buf; // hold the reference buffer. + + int tile_size_bytes; + int tile_col_size_bytes; + int dec_tile_row, dec_tile_col; // always -1 for non-VR tile encoding +#if CONFIG_ACCOUNTING + int acct_enabled; + Accounting accounting; +#endif + int tg_size; // Number of tiles in the current tilegroup + int tg_start; // First tile in the current tilegroup + int tg_size_bit_offset; + int sequence_header_ready; + int sequence_header_changed; +#if CONFIG_INSPECTION + aom_inspect_cb inspect_cb; + void *inspect_ctx; +#endif + int operating_point; + int current_operating_point; + int seen_frame_header; + + // State if the camera frame header is already decoded while + // large_scale_tile = 1. + int camera_frame_header_ready; + size_t frame_header_size; + DataBuffer obu_size_hdr; + int output_frame_width_in_tiles_minus_1; + int output_frame_height_in_tiles_minus_1; + int tile_count_minus_1; + uint32_t coded_tile_data_size; + unsigned int ext_tile_debug; // for ext-tile software debug & testing + unsigned int row_mt; + EXTERNAL_REFERENCES ext_refs; + size_t tile_list_size; + uint8_t *tile_list_output; + size_t buffer_sz; + + CB_BUFFER *cb_buffer_base; + int cb_buffer_alloc_size; + + int allocated_row_mt_sync_rows; + +#if CONFIG_MULTITHREAD + pthread_mutex_t *row_mt_mutex_; + pthread_cond_t *row_mt_cond_; +#endif + + AV1DecRowMTInfo frame_row_mt_info; +} AV1Decoder; + +// Returns 0 on success. Sets pbi->common.error.error_code to a nonzero error +// code and returns a nonzero value on failure. +int av1_receive_compressed_data(struct AV1Decoder *pbi, size_t size, + const uint8_t **dest); + +// Get the frame at a particular index in the output queue +int av1_get_raw_frame(AV1Decoder *pbi, size_t index, YV12_BUFFER_CONFIG **sd, + aom_film_grain_t **grain_params); + +int av1_get_frame_to_show(struct AV1Decoder *pbi, YV12_BUFFER_CONFIG *frame); + +aom_codec_err_t av1_copy_reference_dec(struct AV1Decoder *pbi, int idx, + YV12_BUFFER_CONFIG *sd); + +aom_codec_err_t av1_set_reference_dec(AV1_COMMON *cm, int idx, + int use_external_ref, + YV12_BUFFER_CONFIG *sd); +aom_codec_err_t av1_copy_new_frame_dec(AV1_COMMON *cm, + YV12_BUFFER_CONFIG *new_frame, + YV12_BUFFER_CONFIG *sd); + +struct AV1Decoder *av1_decoder_create(BufferPool *const pool); + +void av1_decoder_remove(struct AV1Decoder *pbi); +void av1_dealloc_dec_jobs(struct AV1DecTileMTData *tile_jobs_sync); + +void av1_dec_row_mt_dealloc(AV1DecRowMTSync *dec_row_mt_sync); + +void av1_dec_free_cb_buf(AV1Decoder *pbi); + +static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs, + BufferPool *const pool) { + if (idx >= 0) { + --frame_bufs[idx].ref_count; + // A worker may only get a free framebuffer index when calling get_free_fb. + // But the private buffer is not set up until finish decoding header. + // So any error happens during decoding header, the frame_bufs will not + // have valid priv buffer. + if (frame_bufs[idx].ref_count == 0 && + frame_bufs[idx].raw_frame_buffer.priv) { + pool->release_fb_cb(pool->cb_priv, &frame_bufs[idx].raw_frame_buffer); + } + } +} + +#define ACCT_STR __func__ +static INLINE int av1_read_uniform(aom_reader *r, int n) { + const int l = get_unsigned_bits(n); + const int m = (1 << l) - n; + const int v = aom_read_literal(r, l - 1, ACCT_STR); + assert(l != 0); + if (v < m) + return v; + else + return (v << 1) - m + aom_read_literal(r, 1, ACCT_STR); +} + +typedef void (*palette_visitor_fn_t)(MACROBLOCKD *const xd, int plane, + aom_reader *r); + +void av1_visit_palette(AV1Decoder *const pbi, MACROBLOCKD *const xd, int mi_row, + int mi_col, aom_reader *r, BLOCK_SIZE bsize, + palette_visitor_fn_t visit); + +typedef void (*block_visitor_fn_t)(AV1Decoder *const pbi, ThreadData *const td, + int mi_row, int mi_col, aom_reader *r, + PARTITION_TYPE partition, BLOCK_SIZE bsize); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // AOM_AV1_DECODER_DECODER_H_ diff --git a/media/libaom/src/av1/decoder/decodetxb.c b/media/libaom/src/av1/decoder/decodetxb.c new file mode 100644 index 000000000..f3ef2d55e --- /dev/null +++ b/media/libaom/src/av1/decoder/decodetxb.c @@ -0,0 +1,362 @@ +/* + * 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 "av1/decoder/decodetxb.h" + +#include "aom_ports/mem.h" +#include "av1/common/idct.h" +#include "av1/common/scan.h" +#include "av1/common/txb_common.h" +#include "av1/decoder/decodemv.h" + +#define ACCT_STR __func__ + +static int read_golomb(MACROBLOCKD *xd, aom_reader *r) { + int x = 1; + int length = 0; + int i = 0; + + while (!i) { + i = aom_read_bit(r, ACCT_STR); + ++length; + if (length > 20) { + aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME, + "Invalid length in read_golomb"); + break; + } + } + + for (i = 0; i < length - 1; ++i) { + x <<= 1; + x += aom_read_bit(r, ACCT_STR); + } + + return x - 1; +} + +static INLINE int rec_eob_pos(const int eob_token, const int extra) { + int eob = k_eob_group_start[eob_token]; + if (eob > 2) { + eob += extra; + } + return eob; +} + +static INLINE int get_dqv(const int16_t *dequant, int coeff_idx, + const qm_val_t *iqmatrix) { + int dqv = dequant[!!coeff_idx]; + if (iqmatrix != NULL) + dqv = + ((iqmatrix[coeff_idx] * dqv) + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS; + return dqv; +} + +static INLINE void read_coeffs_reverse_2d(aom_reader *r, TX_SIZE tx_size, + int start_si, int end_si, + const int16_t *scan, int bwl, + uint8_t *levels, + base_cdf_arr base_cdf, + br_cdf_arr br_cdf) { + for (int c = end_si; c >= start_si; --c) { + const int pos = scan[c]; + const int coeff_ctx = get_lower_levels_ctx_2d(levels, pos, bwl, tx_size); + const int nsymbs = 4; + int level = aom_read_symbol(r, base_cdf[coeff_ctx], nsymbs, ACCT_STR); + if (level > NUM_BASE_LEVELS) { + const int br_ctx = get_br_ctx_2d(levels, pos, bwl); + aom_cdf_prob *cdf = br_cdf[br_ctx]; + for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) { + const int k = aom_read_symbol(r, cdf, BR_CDF_SIZE, ACCT_STR); + level += k; + if (k < BR_CDF_SIZE - 1) break; + } + } + levels[get_padded_idx(pos, bwl)] = level; + } +} + +static INLINE void read_coeffs_reverse(aom_reader *r, TX_SIZE tx_size, + TX_CLASS tx_class, int start_si, + int end_si, const int16_t *scan, int bwl, + uint8_t *levels, base_cdf_arr base_cdf, + br_cdf_arr br_cdf) { + for (int c = end_si; c >= start_si; --c) { + const int pos = scan[c]; + const int coeff_ctx = + get_lower_levels_ctx(levels, pos, bwl, tx_size, tx_class); + const int nsymbs = 4; + int level = aom_read_symbol(r, base_cdf[coeff_ctx], nsymbs, ACCT_STR); + if (level > NUM_BASE_LEVELS) { + const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class); + aom_cdf_prob *cdf = br_cdf[br_ctx]; + for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) { + const int k = aom_read_symbol(r, cdf, BR_CDF_SIZE, ACCT_STR); + level += k; + if (k < BR_CDF_SIZE - 1) break; + } + } + levels[get_padded_idx(pos, bwl)] = level; + } +} + +uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd, + aom_reader *const r, const int blk_row, + const int blk_col, const int plane, + const TXB_CTX *const txb_ctx, + const TX_SIZE tx_size) { + FRAME_CONTEXT *const ec_ctx = xd->tile_ctx; + const int32_t max_value = (1 << (7 + xd->bd)) - 1; + const int32_t min_value = -(1 << (7 + xd->bd)); + const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size); + const PLANE_TYPE plane_type = get_plane_type(plane); + MB_MODE_INFO *const mbmi = xd->mi[0]; + struct macroblockd_plane *const pd = &xd->plane[plane]; + const int16_t *const dequant = pd->seg_dequant_QTX[mbmi->segment_id]; + tran_low_t *const tcoeffs = pd->dqcoeff_block + xd->cb_offset[plane]; + const int shift = av1_get_tx_scale(tx_size); + const int bwl = get_txb_bwl(tx_size); + const int width = get_txb_wide(tx_size); + const int height = get_txb_high(tx_size); + int cul_level = 0; + int dc_val = 0; + uint8_t levels_buf[TX_PAD_2D]; + uint8_t *const levels = set_levels(levels_buf, width); + const int all_zero = aom_read_symbol( + r, ec_ctx->txb_skip_cdf[txs_ctx][txb_ctx->txb_skip_ctx], 2, ACCT_STR); + eob_info *eob_data = pd->eob_data + xd->txb_offset[plane]; + uint16_t *const eob = &(eob_data->eob); + uint16_t *const max_scan_line = &(eob_data->max_scan_line); + *max_scan_line = 0; + *eob = 0; + if (all_zero) { + *max_scan_line = 0; + if (plane == 0) { + const int txk_type_idx = + av1_get_txk_type_index(mbmi->sb_type, blk_row, blk_col); + mbmi->txk_type[txk_type_idx] = DCT_DCT; + } + return 0; + } + + memset(levels_buf, 0, + sizeof(*levels_buf) * + ((width + TX_PAD_HOR) * (height + TX_PAD_VER) + TX_PAD_END)); + if (plane == AOM_PLANE_Y) { + // only y plane's tx_type is transmitted + av1_read_tx_type(cm, xd, blk_row, blk_col, tx_size, r); + } + const TX_TYPE tx_type = av1_get_tx_type(plane_type, xd, blk_row, blk_col, + tx_size, cm->reduced_tx_set_used); + const TX_CLASS tx_class = tx_type_to_class[tx_type]; + const TX_SIZE qm_tx_size = av1_get_adjusted_tx_size(tx_size); + const qm_val_t *iqmatrix = + IS_2D_TRANSFORM(tx_type) + ? pd->seg_iqmatrix[mbmi->segment_id][qm_tx_size] + : cm->giqmatrix[NUM_QM_LEVELS - 1][0][qm_tx_size]; + const SCAN_ORDER *const scan_order = get_scan(tx_size, tx_type); + const int16_t *const scan = scan_order->scan; + int eob_extra = 0; + int eob_pt = 1; + + const int eob_multi_size = txsize_log2_minus4[tx_size]; + const int eob_multi_ctx = (tx_class == TX_CLASS_2D) ? 0 : 1; + switch (eob_multi_size) { + case 0: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf16[plane_type][eob_multi_ctx], + 5, ACCT_STR) + + 1; + break; + case 1: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf32[plane_type][eob_multi_ctx], + 6, ACCT_STR) + + 1; + break; + case 2: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf64[plane_type][eob_multi_ctx], + 7, ACCT_STR) + + 1; + break; + case 3: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf128[plane_type][eob_multi_ctx], + 8, ACCT_STR) + + 1; + break; + case 4: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf256[plane_type][eob_multi_ctx], + 9, ACCT_STR) + + 1; + break; + case 5: + eob_pt = + aom_read_symbol(r, ec_ctx->eob_flag_cdf512[plane_type][eob_multi_ctx], + 10, ACCT_STR) + + 1; + break; + case 6: + default: + eob_pt = aom_read_symbol( + r, ec_ctx->eob_flag_cdf1024[plane_type][eob_multi_ctx], 11, + ACCT_STR) + + 1; + break; + } + + if (k_eob_offset_bits[eob_pt] > 0) { + const int eob_ctx = eob_pt - 3; + int bit = aom_read_symbol( + r, ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_ctx], 2, ACCT_STR); + if (bit) { + eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1)); + } + + for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) { + bit = aom_read_bit(r, ACCT_STR); + if (bit) { + eob_extra += (1 << (k_eob_offset_bits[eob_pt] - 1 - i)); + } + } + } + *eob = rec_eob_pos(eob_pt, eob_extra); + + { + // Read the non-zero coefficient with scan index eob-1 + // TODO(angiebird): Put this into a function + const int c = *eob - 1; + const int pos = scan[c]; + const int coeff_ctx = get_lower_levels_ctx_eob(bwl, height, c); + const int nsymbs = 3; + aom_cdf_prob *cdf = + ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx]; + int level = aom_read_symbol(r, cdf, nsymbs, ACCT_STR) + 1; + if (level > NUM_BASE_LEVELS) { + const int br_ctx = get_br_ctx(levels, pos, bwl, tx_class); + for (int idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) { + const int k = aom_read_symbol( + r, + ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type][br_ctx], + BR_CDF_SIZE, ACCT_STR); + level += k; + if (k < BR_CDF_SIZE - 1) break; + } + } + levels[get_padded_idx(pos, bwl)] = level; + } + if (*eob > 1) { + base_cdf_arr base_cdf = ec_ctx->coeff_base_cdf[txs_ctx][plane_type]; + br_cdf_arr br_cdf = + ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)][plane_type]; + if (tx_class == TX_CLASS_2D) { + read_coeffs_reverse_2d(r, tx_size, 1, *eob - 1 - 1, scan, bwl, levels, + base_cdf, br_cdf); + read_coeffs_reverse(r, tx_size, tx_class, 0, 0, scan, bwl, levels, + base_cdf, br_cdf); + } else { + read_coeffs_reverse(r, tx_size, tx_class, 0, *eob - 1 - 1, scan, bwl, + levels, base_cdf, br_cdf); + } + } + + int16_t num_zero_coeffs = 0; + for (int c = 0; c < *eob; ++c) { + const int pos = scan[c]; + num_zero_coeffs = AOMMAX(num_zero_coeffs, pos); + } + memset(tcoeffs, 0, (num_zero_coeffs + 1) * sizeof(tcoeffs[0])); + + for (int c = 0; c < *eob; ++c) { + const int pos = scan[c]; + uint8_t sign; + tran_low_t level = levels[get_padded_idx(pos, bwl)]; + if (level) { + *max_scan_line = AOMMAX(*max_scan_line, pos); + if (c == 0) { + const int dc_sign_ctx = txb_ctx->dc_sign_ctx; + sign = aom_read_symbol(r, ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx], + 2, ACCT_STR); + } else { + sign = aom_read_bit(r, ACCT_STR); + } + if (level >= MAX_BASE_BR_RANGE) { + level += read_golomb(xd, r); + } + + if (c == 0) dc_val = sign ? -level : level; + + // Bitmasking to clamp level to valid range: + // The valid range for 8/10/12 bit vdieo is at most 14/16/18 bit + level &= 0xfffff; + cul_level += level; + tran_low_t dq_coeff; + // Bitmasking to clamp dq_coeff to valid range: + // The valid range for 8/10/12 bit video is at most 17/19/21 bit + dq_coeff = (tran_low_t)( + (int64_t)level * get_dqv(dequant, scan[c], iqmatrix) & 0xffffff); + dq_coeff = dq_coeff >> shift; + if (sign) { + dq_coeff = -dq_coeff; + } + tcoeffs[pos] = clamp(dq_coeff, min_value, max_value); + } + } + + cul_level = AOMMIN(COEFF_CONTEXT_MASK, cul_level); + + // DC value + set_dc_sign(&cul_level, dc_val); + + return cul_level; +} + +void av1_read_coeffs_txb_facade(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) { +#if TXCOEFF_TIMER + struct aom_usec_timer timer; + aom_usec_timer_start(&timer); +#endif + MB_MODE_INFO *const mbmi = xd->mi[0]; + struct macroblockd_plane *const pd = &xd->plane[plane]; + + const BLOCK_SIZE bsize = mbmi->sb_type; + const BLOCK_SIZE plane_bsize = + get_plane_block_size(bsize, pd->subsampling_x, pd->subsampling_y); + + TXB_CTX txb_ctx; + get_txb_ctx(plane_bsize, tx_size, plane, pd->above_context + col, + pd->left_context + row, &txb_ctx); + const uint8_t cul_level = + av1_read_coeffs_txb(cm, xd, r, row, col, plane, &txb_ctx, tx_size); + av1_set_contexts(xd, pd, plane, plane_bsize, tx_size, cul_level, col, row); + + if (is_inter_block(mbmi)) { + PLANE_TYPE plane_type = get_plane_type(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); + + if (plane == 0) + update_txk_array(mbmi->txk_type, mbmi->sb_type, row, col, tx_size, + tx_type); + } + +#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 +} diff --git a/media/libaom/src/av1/decoder/decodetxb.h b/media/libaom/src/av1/decoder/decodetxb.h new file mode 100644 index 000000000..fe04f6abd --- /dev/null +++ b/media/libaom/src/av1/decoder/decodetxb.h @@ -0,0 +1,32 @@ +/* + * 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. + */ + +#ifndef AOM_AV1_DECODER_DECODETXB_H_ +#define AOM_AV1_DECODER_DECODETXB_H_ + +#include "config/aom_config.h" + +#include "av1/common/blockd.h" +#include "av1/common/onyxc_int.h" +#include "av1/common/txb_common.h" +#include "aom_dsp/bitreader.h" + +uint8_t av1_read_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *const xd, + aom_reader *const r, const int blk_row, + const int blk_col, const int plane, + const TXB_CTX *const txb_ctx, + const TX_SIZE tx_size); + +void av1_read_coeffs_txb_facade(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); +#endif // AOM_AV1_DECODER_DECODETXB_H_ diff --git a/media/libaom/src/av1/decoder/detokenize.c b/media/libaom/src/av1/decoder/detokenize.c new file mode 100644 index 000000000..9d54bd13d --- /dev/null +++ b/media/libaom/src/av1/decoder/detokenize.c @@ -0,0 +1,78 @@ +/* + * 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 "config/aom_config.h" + +#include "aom_mem/aom_mem.h" +#include "aom_ports/mem.h" +#include "av1/common/blockd.h" +#include "av1/decoder/detokenize.h" + +#define ACCT_STR __func__ + +#include "av1/common/common.h" +#include "av1/common/entropy.h" +#include "av1/common/idct.h" + +static void decode_color_map_tokens(Av1ColorMapParam *param, aom_reader *r) { + uint8_t color_order[PALETTE_MAX_SIZE]; + const int n = param->n_colors; + uint8_t *const color_map = param->color_map; + MapCdf color_map_cdf = param->map_cdf; + int plane_block_width = param->plane_width; + int plane_block_height = param->plane_height; + int rows = param->rows; + int cols = param->cols; + + // The first color index. + color_map[0] = av1_read_uniform(r, n); + assert(color_map[0] < n); + + // Run wavefront on the palette map index decoding. + for (int i = 1; i < rows + cols - 1; ++i) { + for (int j = AOMMIN(i, cols - 1); j >= AOMMAX(0, i - rows + 1); --j) { + const int color_ctx = av1_get_palette_color_index_context( + color_map, plane_block_width, (i - j), j, n, color_order, NULL); + const int color_idx = aom_read_symbol( + r, color_map_cdf[n - PALETTE_MIN_SIZE][color_ctx], n, ACCT_STR); + assert(color_idx >= 0 && color_idx < n); + color_map[(i - j) * plane_block_width + j] = color_order[color_idx]; + } + } + // Copy last column to extra columns. + if (cols < plane_block_width) { + for (int i = 0; i < rows; ++i) { + memset(color_map + i * plane_block_width + cols, + color_map[i * plane_block_width + cols - 1], + (plane_block_width - cols)); + } + } + // Copy last row to extra rows. + for (int i = rows; i < plane_block_height; ++i) { + memcpy(color_map + i * plane_block_width, + color_map + (rows - 1) * plane_block_width, plane_block_width); + } +} + +void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane, + aom_reader *r) { + assert(plane == 0 || plane == 1); + Av1ColorMapParam params; + params.color_map = + xd->plane[plane].color_index_map + xd->color_index_map_offset[plane]; + params.map_cdf = plane ? xd->tile_ctx->palette_uv_color_index_cdf + : xd->tile_ctx->palette_y_color_index_cdf; + const MB_MODE_INFO *const mbmi = xd->mi[0]; + params.n_colors = mbmi->palette_mode_info.palette_size[plane]; + av1_get_block_dimensions(mbmi->sb_type, plane, xd, ¶ms.plane_width, + ¶ms.plane_height, ¶ms.rows, ¶ms.cols); + decode_color_map_tokens(¶ms, r); +} diff --git a/media/libaom/src/av1/decoder/detokenize.h b/media/libaom/src/av1/decoder/detokenize.h new file mode 100644 index 000000000..173b437a9 --- /dev/null +++ b/media/libaom/src/av1/decoder/detokenize.h @@ -0,0 +1,29 @@ +/* + * 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 AOM_AV1_DECODER_DETOKENIZE_H_ +#define AOM_AV1_DECODER_DETOKENIZE_H_ + +#include "config/aom_config.h" + +#include "av1/common/scan.h" +#include "av1/decoder/decoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane, aom_reader *r); + +#ifdef __cplusplus +} // extern "C" +#endif +#endif // AOM_AV1_DECODER_DETOKENIZE_H_ diff --git a/media/libaom/src/av1/decoder/dthread.c b/media/libaom/src/av1/decoder/dthread.c new file mode 100644 index 000000000..3946c787a --- /dev/null +++ b/media/libaom/src/av1/decoder/dthread.c @@ -0,0 +1,192 @@ +/* + * 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 "config/aom_config.h" + +#include "aom_mem/aom_mem.h" +#include "av1/common/reconinter.h" +#include "av1/decoder/dthread.h" +#include "av1/decoder/decoder.h" + +// #define DEBUG_THREAD + +// TODO(hkuang): Clean up all the #ifdef in this file. +void av1_frameworker_lock_stats(AVxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_lock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void av1_frameworker_unlock_stats(AVxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_unlock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void av1_frameworker_signal_stats(AVxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + +// TODO(hkuang): Fix the pthread_cond_broadcast in windows wrapper. +#if defined(_WIN32) && !HAVE_PTHREAD_H + pthread_cond_signal(&worker_data->stats_cond); +#else + pthread_cond_broadcast(&worker_data->stats_cond); +#endif + +#else + (void)worker; +#endif +} + +// This macro prevents thread_sanitizer from reporting known concurrent writes. +#if defined(__has_feature) +#if __has_feature(thread_sanitizer) +#define BUILDING_WITH_TSAN +#endif +#endif + +// TODO(hkuang): Remove worker parameter as it is only used in debug code. +void av1_frameworker_wait(AVxWorker *const worker, RefCntBuffer *const ref_buf, + int row) { +#if CONFIG_MULTITHREAD + if (!ref_buf) return; + +#ifndef BUILDING_WITH_TSAN + // The following line of code will get harmless tsan error but it is the key + // to get best performance. + if (ref_buf->row >= row && ref_buf->buf.corrupted != 1) return; +#endif + + { + // Find the worker thread that owns the reference frame. If the reference + // frame has been fully decoded, it may not have owner. + AVxWorker *const ref_worker = ref_buf->frame_worker_owner; + FrameWorkerData *const ref_worker_data = + (FrameWorkerData *)ref_worker->data1; + const AV1Decoder *const pbi = ref_worker_data->pbi; + +#ifdef DEBUG_THREAD + { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + printf("%d %p worker is waiting for %d %p worker (%d) ref %d \r\n", + worker_data->worker_id, worker, ref_worker_data->worker_id, + ref_buf->frame_worker_owner, row, ref_buf->row); + } +#endif + + av1_frameworker_lock_stats(ref_worker); + while (ref_buf->row < row && pbi->cur_buf == ref_buf && + ref_buf->buf.corrupted != 1) { + pthread_cond_wait(&ref_worker_data->stats_cond, + &ref_worker_data->stats_mutex); + } + + if (ref_buf->buf.corrupted == 1) { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + av1_frameworker_unlock_stats(ref_worker); + aom_internal_error(&worker_data->pbi->common.error, + AOM_CODEC_CORRUPT_FRAME, + "Worker %p failed to decode frame", worker); + } + av1_frameworker_unlock_stats(ref_worker); + } +#else + (void)worker; + (void)ref_buf; + (void)row; + (void)ref_buf; +#endif // CONFIG_MULTITHREAD +} + +void av1_frameworker_broadcast(RefCntBuffer *const buf, int row) { +#if CONFIG_MULTITHREAD + AVxWorker *worker = buf->frame_worker_owner; + +#ifdef DEBUG_THREAD + { + FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1; + printf("%d %p worker decode to (%d) \r\n", worker_data->worker_id, + buf->frame_worker_owner, row); + } +#endif + + av1_frameworker_lock_stats(worker); + buf->row = row; + av1_frameworker_signal_stats(worker); + av1_frameworker_unlock_stats(worker); +#else + (void)buf; + (void)row; +#endif // CONFIG_MULTITHREAD +} + +void av1_frameworker_copy_context(AVxWorker *const dst_worker, + AVxWorker *const src_worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const src_worker_data = (FrameWorkerData *)src_worker->data1; + FrameWorkerData *const dst_worker_data = (FrameWorkerData *)dst_worker->data1; + AV1_COMMON *const src_cm = &src_worker_data->pbi->common; + AV1_COMMON *const dst_cm = &dst_worker_data->pbi->common; + int i; + + // Wait until source frame's context is ready. + av1_frameworker_lock_stats(src_worker); + while (!src_worker_data->frame_context_ready) { + pthread_cond_wait(&src_worker_data->stats_cond, + &src_worker_data->stats_mutex); + } + + dst_cm->last_frame_seg_map = src_cm->seg.enabled + ? src_cm->current_frame_seg_map + : src_cm->last_frame_seg_map; + dst_worker_data->pbi->need_resync = src_worker_data->pbi->need_resync; + av1_frameworker_unlock_stats(src_worker); + + dst_cm->seq_params.bit_depth = src_cm->seq_params.bit_depth; + dst_cm->seq_params.use_highbitdepth = src_cm->seq_params.use_highbitdepth; + // TODO(zoeliu): To handle parallel decoding + dst_cm->prev_frame = + src_cm->show_existing_frame ? src_cm->prev_frame : src_cm->cur_frame; + dst_cm->last_width = + !src_cm->show_existing_frame ? src_cm->width : src_cm->last_width; + dst_cm->last_height = + !src_cm->show_existing_frame ? src_cm->height : src_cm->last_height; + dst_cm->seq_params.subsampling_x = src_cm->seq_params.subsampling_x; + dst_cm->seq_params.subsampling_y = src_cm->seq_params.subsampling_y; + dst_cm->frame_type = src_cm->frame_type; + dst_cm->last_show_frame = !src_cm->show_existing_frame + ? src_cm->show_frame + : src_cm->last_show_frame; + for (i = 0; i < REF_FRAMES; ++i) + dst_cm->ref_frame_map[i] = src_cm->next_ref_frame_map[i]; + + memcpy(dst_cm->lf_info.lfthr, src_cm->lf_info.lfthr, + (MAX_LOOP_FILTER + 1) * sizeof(loop_filter_thresh)); + dst_cm->lf.sharpness_level = src_cm->lf.sharpness_level; + dst_cm->lf.filter_level[0] = src_cm->lf.filter_level[0]; + dst_cm->lf.filter_level[1] = src_cm->lf.filter_level[1]; + memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, REF_FRAMES); + memcpy(dst_cm->lf.mode_deltas, src_cm->lf.mode_deltas, MAX_MODE_LF_DELTAS); + dst_cm->seg = src_cm->seg; + memcpy(dst_cm->frame_contexts, src_cm->frame_contexts, + FRAME_CONTEXTS * sizeof(dst_cm->frame_contexts[0])); +#else + (void)dst_worker; + (void)src_worker; +#endif // CONFIG_MULTITHREAD +} diff --git a/media/libaom/src/av1/decoder/dthread.h b/media/libaom/src/av1/decoder/dthread.h new file mode 100644 index 000000000..1d264b07e --- /dev/null +++ b/media/libaom/src/av1/decoder/dthread.h @@ -0,0 +1,82 @@ +/* + * 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 AOM_AV1_DECODER_DTHREAD_H_ +#define AOM_AV1_DECODER_DTHREAD_H_ + +#include "config/aom_config.h" + +#include "aom_util/aom_thread.h" +#include "aom/internal/aom_codec_internal.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct AV1Common; +struct AV1Decoder; +struct ThreadData; + +typedef struct DecWorkerData { + struct ThreadData *td; + const uint8_t *data_end; + struct aom_internal_error_info error_info; +} DecWorkerData; + +// WorkerData for the FrameWorker thread. It contains all the information of +// the worker and decode structures for decoding a frame. +typedef struct FrameWorkerData { + struct AV1Decoder *pbi; + const uint8_t *data; + const uint8_t *data_end; + size_t data_size; + void *user_priv; + int worker_id; + int received_frame; + + // scratch_buffer is used in frame parallel mode only. + // It is used to make a copy of the compressed data. + uint8_t *scratch_buffer; + size_t scratch_buffer_size; + +#if CONFIG_MULTITHREAD + pthread_mutex_t stats_mutex; + pthread_cond_t stats_cond; +#endif + + int frame_context_ready; // Current frame's context is ready to read. + int frame_decoded; // Finished decoding current frame. +} FrameWorkerData; + +void av1_frameworker_lock_stats(AVxWorker *const worker); +void av1_frameworker_unlock_stats(AVxWorker *const worker); +void av1_frameworker_signal_stats(AVxWorker *const worker); + +// Wait until ref_buf has been decoded to row in real pixel unit. +// Note: worker may already finish decoding ref_buf and release it in order to +// start decoding next frame. So need to check whether worker is still decoding +// ref_buf. +void av1_frameworker_wait(AVxWorker *const worker, RefCntBuffer *const ref_buf, + int row); + +// FrameWorker broadcasts its decoding progress so other workers that are +// waiting on it can resume decoding. +void av1_frameworker_broadcast(RefCntBuffer *const buf, int row); + +// Copy necessary decoding context from src worker to dst worker. +void av1_frameworker_copy_context(AVxWorker *const dst_worker, + AVxWorker *const src_worker); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // AOM_AV1_DECODER_DTHREAD_H_ diff --git a/media/libaom/src/av1/decoder/inspection.c b/media/libaom/src/av1/decoder/inspection.c new file mode 100644 index 000000000..e6c89298a --- /dev/null +++ b/media/libaom/src/av1/decoder/inspection.c @@ -0,0 +1,117 @@ +/* + * 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 "av1/decoder/decoder.h" +#include "av1/decoder/inspection.h" +#include "av1/common/enums.h" +#include "av1/common/cdef.h" + +static void ifd_init_mi_rc(insp_frame_data *fd, int mi_cols, int mi_rows) { + fd->mi_cols = mi_cols; + fd->mi_rows = mi_rows; + fd->mi_grid = (insp_mi_data *)aom_malloc(sizeof(insp_mi_data) * fd->mi_rows * + fd->mi_cols); +} + +void ifd_init(insp_frame_data *fd, int frame_width, int frame_height) { + int mi_cols = ALIGN_POWER_OF_TWO(frame_width, 3) >> MI_SIZE_LOG2; + int mi_rows = ALIGN_POWER_OF_TWO(frame_height, 3) >> MI_SIZE_LOG2; + ifd_init_mi_rc(fd, mi_cols, mi_rows); +} + +void ifd_clear(insp_frame_data *fd) { + aom_free(fd->mi_grid); + fd->mi_grid = NULL; +} + +/* TODO(negge) This function may be called by more than one thread when using + a multi-threaded decoder and this may cause a data race. */ +int ifd_inspect(insp_frame_data *fd, void *decoder) { + struct AV1Decoder *pbi = (struct AV1Decoder *)decoder; + AV1_COMMON *const cm = &pbi->common; + if (fd->mi_rows != cm->mi_rows || fd->mi_cols != cm->mi_cols) { + ifd_clear(fd); + ifd_init_mi_rc(fd, cm->mi_rows, cm->mi_cols); + } + fd->show_frame = cm->show_frame; + fd->frame_type = cm->frame_type; + fd->base_qindex = cm->base_qindex; + // Set width and height of the first tile until generic support can be added + TileInfo tile_info; + av1_tile_set_row(&tile_info, cm, 0); + av1_tile_set_col(&tile_info, cm, 0); + fd->tile_mi_cols = tile_info.mi_col_end - tile_info.mi_col_start; + fd->tile_mi_rows = tile_info.mi_row_end - tile_info.mi_row_start; + fd->delta_q_present_flag = cm->delta_q_present_flag; + fd->delta_q_res = cm->delta_q_res; +#if CONFIG_ACCOUNTING + fd->accounting = &pbi->accounting; +#endif + // TODO(negge): copy per frame CDEF data + int i, j; + for (i = 0; i < MAX_SEGMENTS; i++) { + for (j = 0; j < 2; j++) { + fd->y_dequant[i][j] = cm->y_dequant_QTX[i][j]; + fd->u_dequant[i][j] = cm->u_dequant_QTX[i][j]; + fd->v_dequant[i][j] = cm->v_dequant_QTX[i][j]; + } + } + for (j = 0; j < cm->mi_rows; j++) { + for (i = 0; i < cm->mi_cols; i++) { + const MB_MODE_INFO *mbmi = cm->mi_grid_visible[j * cm->mi_stride + i]; + insp_mi_data *mi = &fd->mi_grid[j * cm->mi_cols + i]; + // Segment + mi->segment_id = mbmi->segment_id; + // Motion Vectors + mi->mv[0].row = mbmi->mv[0].as_mv.row; + mi->mv[0].col = mbmi->mv[0].as_mv.col; + mi->mv[1].row = mbmi->mv[1].as_mv.row; + mi->mv[1].col = mbmi->mv[1].as_mv.col; + // Reference Frames + mi->ref_frame[0] = mbmi->ref_frame[0]; + mi->ref_frame[1] = mbmi->ref_frame[1]; + // Prediction Mode + mi->mode = mbmi->mode; + // Prediction Mode for Chromatic planes + if (mi->mode < INTRA_MODES) { + mi->uv_mode = mbmi->uv_mode; + } else { + mi->uv_mode = UV_MODE_INVALID; + } + // Block Size + mi->sb_type = mbmi->sb_type; + // Skip Flag + mi->skip = mbmi->skip; + mi->filter[0] = av1_extract_interp_filter(mbmi->interp_filters, 0); + mi->filter[1] = av1_extract_interp_filter(mbmi->interp_filters, 1); + mi->dual_filter_type = mi->filter[0] * 3 + mi->filter[1]; + // Transform + // TODO(anyone): extract tx type info from mbmi->txk_type[]. + mi->tx_type = DCT_DCT; + mi->tx_size = mbmi->tx_size; + + mi->cdef_level = + cm->cdef_strengths[mbmi->cdef_strength] / CDEF_SEC_STRENGTHS; + mi->cdef_strength = + cm->cdef_strengths[mbmi->cdef_strength] % CDEF_SEC_STRENGTHS; + mi->cdef_strength += mi->cdef_strength == 3; + if (mbmi->uv_mode == UV_CFL_PRED) { + mi->cfl_alpha_idx = mbmi->cfl_alpha_idx; + mi->cfl_alpha_sign = mbmi->cfl_alpha_signs; + } else { + mi->cfl_alpha_idx = 0; + mi->cfl_alpha_sign = 0; + } + // delta_q + mi->current_qindex = mbmi->current_qindex; + } + } + return 1; +} diff --git a/media/libaom/src/av1/decoder/inspection.h b/media/libaom/src/av1/decoder/inspection.h new file mode 100644 index 000000000..7214a9bed --- /dev/null +++ b/media/libaom/src/av1/decoder/inspection.h @@ -0,0 +1,84 @@ +/* + * 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. + */ +#ifndef AOM_AV1_DECODER_INSPECTION_H_ +#define AOM_AV1_DECODER_INSPECTION_H_ + +#ifdef __cplusplus +extern "C" { +#endif // __cplusplus + +#include "av1/common/seg_common.h" +#if CONFIG_ACCOUNTING +#include "av1/decoder/accounting.h" +#endif + +#ifndef AOM_AOMDX_H_ +typedef void (*aom_inspect_cb)(void *decoder, void *data); +#endif + +typedef struct insp_mv insp_mv; + +struct insp_mv { + int16_t row; + int16_t col; +}; + +typedef struct insp_mi_data insp_mi_data; + +struct insp_mi_data { + insp_mv mv[2]; + int16_t ref_frame[2]; + int16_t mode; + int16_t uv_mode; + int16_t sb_type; + int16_t skip; + int16_t segment_id; + int16_t dual_filter_type; + int16_t filter[2]; + int16_t tx_type; + int16_t tx_size; + int16_t cdef_level; + int16_t cdef_strength; + int16_t cfl_alpha_idx; + int16_t cfl_alpha_sign; + int16_t current_qindex; +}; + +typedef struct insp_frame_data insp_frame_data; + +struct insp_frame_data { +#if CONFIG_ACCOUNTING + Accounting *accounting; +#endif + insp_mi_data *mi_grid; + int show_frame; + int frame_type; + int base_qindex; + int mi_rows; + int mi_cols; + int tile_mi_rows; + int tile_mi_cols; + int16_t y_dequant[MAX_SEGMENTS][2]; + int16_t u_dequant[MAX_SEGMENTS][2]; + int16_t v_dequant[MAX_SEGMENTS][2]; + // TODO(negge): add per frame CDEF data + int delta_q_present_flag; + int delta_q_res; +}; + +void ifd_init(insp_frame_data *fd, int frame_width, int frame_height); +void ifd_clear(insp_frame_data *fd); +int ifd_inspect(insp_frame_data *fd, void *decoder); + +#ifdef __cplusplus +} // extern "C" +#endif // __cplusplus +#endif // AOM_AV1_DECODER_INSPECTION_H_ diff --git a/media/libaom/src/av1/decoder/obu.c b/media/libaom/src/av1/decoder/obu.c new file mode 100644 index 000000000..44ecf818e --- /dev/null +++ b/media/libaom/src/av1/decoder/obu.c @@ -0,0 +1,839 @@ +/* + * 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 <assert.h> + +#include "config/aom_config.h" + +#include "aom/aom_codec.h" +#include "aom_dsp/bitreader_buffer.h" +#include "aom_ports/mem_ops.h" + +#include "av1/common/common.h" +#include "av1/common/obu_util.h" +#include "av1/common/timing.h" +#include "av1/decoder/decoder.h" +#include "av1/decoder/decodeframe.h" +#include "av1/decoder/obu.h" + +// Picture prediction structures (0-12 are predefined) in scalability metadata. +typedef enum { + SCALABILITY_L1T2 = 0, + SCALABILITY_L1T3 = 1, + SCALABILITY_L2T1 = 2, + SCALABILITY_L2T2 = 3, + SCALABILITY_L2T3 = 4, + SCALABILITY_S2T1 = 5, + SCALABILITY_S2T2 = 6, + SCALABILITY_S2T3 = 7, + SCALABILITY_L2T1h = 8, + SCALABILITY_L2T2h = 9, + SCALABILITY_L2T3h = 10, + SCALABILITY_S2T1h = 11, + SCALABILITY_S2T2h = 12, + SCALABILITY_S2T3h = 13, + SCALABILITY_SS = 14 +} SCALABILITY_STRUCTURES; + +aom_codec_err_t aom_get_num_layers_from_operating_point_idc( + int operating_point_idc, unsigned int *number_spatial_layers, + unsigned int *number_temporal_layers) { + // derive number of spatial/temporal layers from operating_point_idc + + if (!number_spatial_layers || !number_temporal_layers) + return AOM_CODEC_INVALID_PARAM; + + if (operating_point_idc == 0) { + *number_temporal_layers = 1; + *number_spatial_layers = 1; + } else { + *number_spatial_layers = 0; + *number_temporal_layers = 0; + for (int j = 0; j < MAX_NUM_SPATIAL_LAYERS; j++) { + *number_spatial_layers += + (operating_point_idc >> (j + MAX_NUM_TEMPORAL_LAYERS)) & 0x1; + } + for (int j = 0; j < MAX_NUM_TEMPORAL_LAYERS; j++) { + *number_temporal_layers += (operating_point_idc >> j) & 0x1; + } + } + + return AOM_CODEC_OK; +} + +static int is_obu_in_current_operating_point(AV1Decoder *pbi, + ObuHeader obu_header) { + if (!pbi->current_operating_point) { + return 1; + } + + if ((pbi->current_operating_point >> obu_header.temporal_layer_id) & 0x1 && + (pbi->current_operating_point >> (obu_header.spatial_layer_id + 8)) & + 0x1) { + return 1; + } + return 0; +} + +static int byte_alignment(AV1_COMMON *const cm, + struct aom_read_bit_buffer *const rb) { + while (rb->bit_offset & 7) { + if (aom_rb_read_bit(rb)) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + } + return 0; +} + +static uint32_t read_temporal_delimiter_obu() { return 0; } + +// Returns a boolean that indicates success. +static int read_bitstream_level(BitstreamLevel *bl, + struct aom_read_bit_buffer *rb) { + const uint8_t seq_level_idx = aom_rb_read_literal(rb, LEVEL_BITS); + if (!is_valid_seq_level_idx(seq_level_idx)) return 0; + bl->major = (seq_level_idx >> LEVEL_MINOR_BITS) + LEVEL_MAJOR_MIN; + bl->minor = seq_level_idx & ((1 << LEVEL_MINOR_BITS) - 1); + return 1; +} + +// Returns whether two sequence headers are consistent with each other. +// TODO(huisu,wtc@google.com): make sure the code matches the spec exactly. +static int are_seq_headers_consistent(const SequenceHeader *seq_params_old, + const SequenceHeader *seq_params_new) { + return !memcmp(seq_params_old, seq_params_new, sizeof(SequenceHeader)); +} + +// On success, sets pbi->sequence_header_ready to 1 and returns the number of +// bytes read from 'rb'. +// On failure, sets pbi->common.error.error_code and returns 0. +static uint32_t read_sequence_header_obu(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb) { + AV1_COMMON *const cm = &pbi->common; + const uint32_t saved_bit_offset = rb->bit_offset; + + // Verify rb has been configured to report errors. + assert(rb->error_handler); + + // Use a local variable to store the information as we decode. At the end, + // if no errors have occurred, cm->seq_params is updated. + SequenceHeader sh = cm->seq_params; + SequenceHeader *const seq_params = &sh; + + seq_params->profile = av1_read_profile(rb); + if (seq_params->profile > CONFIG_MAX_DECODE_PROFILE) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return 0; + } + + // Still picture or not + seq_params->still_picture = aom_rb_read_bit(rb); + seq_params->reduced_still_picture_hdr = aom_rb_read_bit(rb); + // Video must have reduced_still_picture_hdr = 0 + if (!seq_params->still_picture && seq_params->reduced_still_picture_hdr) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return 0; + } + + if (seq_params->reduced_still_picture_hdr) { + cm->timing_info_present = 0; + seq_params->decoder_model_info_present_flag = 0; + seq_params->display_model_info_present_flag = 0; + seq_params->operating_points_cnt_minus_1 = 0; + seq_params->operating_point_idc[0] = 0; + if (!read_bitstream_level(&seq_params->level[0], rb)) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return 0; + } + seq_params->tier[0] = 0; + cm->op_params[0].decoder_model_param_present_flag = 0; + cm->op_params[0].display_model_param_present_flag = 0; + } else { + cm->timing_info_present = aom_rb_read_bit(rb); // timing_info_present_flag + if (cm->timing_info_present) { + av1_read_timing_info_header(cm, rb); + + seq_params->decoder_model_info_present_flag = aom_rb_read_bit(rb); + if (seq_params->decoder_model_info_present_flag) + av1_read_decoder_model_info(cm, rb); + } else { + seq_params->decoder_model_info_present_flag = 0; + } + seq_params->display_model_info_present_flag = aom_rb_read_bit(rb); + seq_params->operating_points_cnt_minus_1 = + aom_rb_read_literal(rb, OP_POINTS_CNT_MINUS_1_BITS); + for (int i = 0; i < seq_params->operating_points_cnt_minus_1 + 1; i++) { + seq_params->operating_point_idc[i] = + aom_rb_read_literal(rb, OP_POINTS_IDC_BITS); + if (!read_bitstream_level(&seq_params->level[i], rb)) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return 0; + } + // This is the seq_level_idx[i] > 7 check in the spec. seq_level_idx 7 + // is equivalent to level 3.3. + if (seq_params->level[i].major > 3) + seq_params->tier[i] = aom_rb_read_bit(rb); + else + seq_params->tier[i] = 0; + if (seq_params->decoder_model_info_present_flag) { + cm->op_params[i].decoder_model_param_present_flag = aom_rb_read_bit(rb); + if (cm->op_params[i].decoder_model_param_present_flag) + av1_read_op_parameters_info(cm, rb, i); + } else { + cm->op_params[i].decoder_model_param_present_flag = 0; + } + if (cm->timing_info_present && + (cm->timing_info.equal_picture_interval || + cm->op_params[i].decoder_model_param_present_flag)) { + cm->op_params[i].bitrate = max_level_bitrate( + seq_params->profile, + major_minor_to_seq_level_idx(seq_params->level[i]), + seq_params->tier[i]); + // Level with seq_level_idx = 31 returns a high "dummy" bitrate to pass + // the check + if (cm->op_params[i].bitrate == 0) + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "AV1 does not support this combination of " + "profile, level, and tier."); + // Buffer size in bits/s is bitrate in bits/s * 1 s + cm->op_params[i].buffer_size = cm->op_params[i].bitrate; + } + if (cm->timing_info_present && cm->timing_info.equal_picture_interval && + !cm->op_params[i].decoder_model_param_present_flag) { + // When the decoder_model_parameters are not sent for this op, set + // the default ones that can be used with the resource availability mode + cm->op_params[i].decoder_buffer_delay = 70000; + cm->op_params[i].encoder_buffer_delay = 20000; + cm->op_params[i].low_delay_mode_flag = 0; + } + + if (seq_params->display_model_info_present_flag) { + cm->op_params[i].display_model_param_present_flag = aom_rb_read_bit(rb); + if (cm->op_params[i].display_model_param_present_flag) { + cm->op_params[i].initial_display_delay = + aom_rb_read_literal(rb, 4) + 1; + if (cm->op_params[i].initial_display_delay > 10) + aom_internal_error( + &cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "AV1 does not support more than 10 decoded frames delay"); + } else { + cm->op_params[i].initial_display_delay = 10; + } + } else { + cm->op_params[i].display_model_param_present_flag = 0; + cm->op_params[i].initial_display_delay = 10; + } + } + } + // This decoder supports all levels. Choose operating point provided by + // external means + int operating_point = pbi->operating_point; + if (operating_point < 0 || + operating_point > seq_params->operating_points_cnt_minus_1) + operating_point = 0; + pbi->current_operating_point = + seq_params->operating_point_idc[operating_point]; + if (aom_get_num_layers_from_operating_point_idc( + pbi->current_operating_point, &cm->number_spatial_layers, + &cm->number_temporal_layers) != AOM_CODEC_OK) { + cm->error.error_code = AOM_CODEC_ERROR; + return 0; + } + + av1_read_sequence_header(cm, rb, seq_params); + + av1_read_color_config(rb, pbi->allow_lowbitdepth, seq_params, &cm->error); + if (!(seq_params->subsampling_x == 0 && seq_params->subsampling_y == 0) && + !(seq_params->subsampling_x == 1 && seq_params->subsampling_y == 1) && + !(seq_params->subsampling_x == 1 && seq_params->subsampling_y == 0)) { + aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "Only 4:4:4, 4:2:2 and 4:2:0 are currently supported, " + "%d %d subsampling is not supported.\n", + seq_params->subsampling_x, seq_params->subsampling_y); + } + + seq_params->film_grain_params_present = aom_rb_read_bit(rb); + + if (av1_check_trailing_bits(pbi, rb) != 0) { + // cm->error.error_code is already set. + return 0; + } + + // If a sequence header has been decoded before, we check if the new + // one is consistent with the old one. + if (pbi->sequence_header_ready) { + if (!are_seq_headers_consistent(&cm->seq_params, seq_params)) + pbi->sequence_header_changed = 1; + } + + cm->seq_params = *seq_params; + pbi->sequence_header_ready = 1; + + return ((rb->bit_offset - saved_bit_offset + 7) >> 3); +} + +// On success, returns the frame header size. On failure, calls +// aom_internal_error and does not return. +static uint32_t read_frame_header_obu(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb, + const uint8_t *data, + const uint8_t **p_data_end, + int trailing_bits_present) { + return av1_decode_frame_headers_and_setup(pbi, rb, data, p_data_end, + trailing_bits_present); +} + +static int32_t read_tile_group_header(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb, + int *start_tile, int *end_tile, + int tile_start_implicit) { + AV1_COMMON *const cm = &pbi->common; + uint32_t saved_bit_offset = rb->bit_offset; + int tile_start_and_end_present_flag = 0; + const int num_tiles = pbi->common.tile_rows * pbi->common.tile_cols; + + if (!pbi->common.large_scale_tile && num_tiles > 1) { + tile_start_and_end_present_flag = aom_rb_read_bit(rb); + } + if (pbi->common.large_scale_tile || num_tiles == 1 || + !tile_start_and_end_present_flag) { + *start_tile = 0; + *end_tile = num_tiles - 1; + return ((rb->bit_offset - saved_bit_offset + 7) >> 3); + } + if (tile_start_implicit && tile_start_and_end_present_flag) { + aom_internal_error( + &cm->error, AOM_CODEC_UNSUP_BITSTREAM, + "For OBU_FRAME type obu tile_start_and_end_present_flag must be 0"); + return -1; + } + *start_tile = + aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols); + *end_tile = aom_rb_read_literal(rb, cm->log2_tile_rows + cm->log2_tile_cols); + + return ((rb->bit_offset - saved_bit_offset + 7) >> 3); +} + +static uint32_t read_one_tile_group_obu( + AV1Decoder *pbi, struct aom_read_bit_buffer *rb, int is_first_tg, + const uint8_t *data, const uint8_t *data_end, const uint8_t **p_data_end, + int *is_last_tg, int tile_start_implicit) { + AV1_COMMON *const cm = &pbi->common; + int start_tile, end_tile; + int32_t header_size, tg_payload_size; + + assert((rb->bit_offset & 7) == 0); + assert(rb->bit_buffer + aom_rb_bytes_read(rb) == data); + + header_size = read_tile_group_header(pbi, rb, &start_tile, &end_tile, + tile_start_implicit); + if (header_size == -1 || byte_alignment(cm, rb)) return 0; + if (start_tile > end_tile) return header_size; + data += header_size; + av1_decode_tg_tiles_and_wrapup(pbi, data, data_end, p_data_end, start_tile, + end_tile, is_first_tg); + + tg_payload_size = (uint32_t)(*p_data_end - data); + + // TODO(shan): For now, assume all tile groups received in order + *is_last_tg = end_tile == cm->tile_rows * cm->tile_cols - 1; + return header_size + tg_payload_size; +} + +static void alloc_tile_list_buffer(AV1Decoder *pbi) { + // TODO(yunqing): for now, copy each tile's decoded YUV data directly to the + // output buffer. This needs to be modified according to the application + // requirement. + AV1_COMMON *const cm = &pbi->common; + const int tile_width_in_pixels = cm->tile_width * MI_SIZE; + const int tile_height_in_pixels = cm->tile_height * MI_SIZE; + const int ssy = cm->seq_params.subsampling_y; + const int ssx = cm->seq_params.subsampling_x; + const int num_planes = av1_num_planes(cm); + const size_t yplane_tile_size = tile_height_in_pixels * tile_width_in_pixels; + const size_t uvplane_tile_size = + (num_planes > 1) + ? (tile_height_in_pixels >> ssy) * (tile_width_in_pixels >> ssx) + : 0; + const size_t tile_size = (cm->seq_params.use_highbitdepth ? 2 : 1) * + (yplane_tile_size + 2 * uvplane_tile_size); + pbi->tile_list_size = tile_size * (pbi->tile_count_minus_1 + 1); + + if (pbi->tile_list_size > pbi->buffer_sz) { + if (pbi->tile_list_output != NULL) aom_free(pbi->tile_list_output); + pbi->tile_list_output = NULL; + + pbi->tile_list_output = (uint8_t *)aom_memalign(32, pbi->tile_list_size); + if (pbi->tile_list_output == NULL) + aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, + "Failed to allocate the tile list output buffer"); + pbi->buffer_sz = pbi->tile_list_size; + } +} + +static void copy_decoded_tile_to_tile_list_buffer(AV1Decoder *pbi, + uint8_t **output) { + AV1_COMMON *const cm = &pbi->common; + const int tile_width_in_pixels = cm->tile_width * MI_SIZE; + const int tile_height_in_pixels = cm->tile_height * MI_SIZE; + const int ssy = cm->seq_params.subsampling_y; + const int ssx = cm->seq_params.subsampling_x; + const int num_planes = av1_num_planes(cm); + + // Copy decoded tile to the tile list output buffer. + YV12_BUFFER_CONFIG *cur_frame = get_frame_new_buffer(cm); + const int mi_row = pbi->dec_tile_row * cm->tile_height; + const int mi_col = pbi->dec_tile_col * cm->tile_width; + const int is_hbd = (cur_frame->flags & YV12_FLAG_HIGHBITDEPTH) ? 1 : 0; + uint8_t *bufs[MAX_MB_PLANE] = { NULL, NULL, NULL }; + int strides[MAX_MB_PLANE] = { 0, 0, 0 }; + int plane; + + for (plane = 0; plane < num_planes; ++plane) { + int shift_x = plane > 0 ? ssx : 0; + int shift_y = plane > 0 ? ssy : 0; + + bufs[plane] = cur_frame->buffers[plane]; + strides[plane] = + (plane > 0) ? cur_frame->strides[1] : cur_frame->strides[0]; + + bufs[plane] += mi_row * (MI_SIZE >> shift_y) * strides[plane] + + mi_col * (MI_SIZE >> shift_x); + + if (is_hbd) { + bufs[plane] = (uint8_t *)CONVERT_TO_SHORTPTR(bufs[plane]); + strides[plane] *= 2; + } + + int w, h; + w = (plane > 0 && shift_x > 0) ? ((tile_width_in_pixels + 1) >> shift_x) + : tile_width_in_pixels; + w *= (1 + is_hbd); + h = (plane > 0 && shift_y > 0) ? ((tile_height_in_pixels + 1) >> shift_y) + : tile_height_in_pixels; + int j; + + for (j = 0; j < h; ++j) { + memcpy(*output, bufs[plane], w); + bufs[plane] += strides[plane]; + *output += w; + } + } +} + +// Only called while large_scale_tile = 1. +static uint32_t read_and_decode_one_tile_list(AV1Decoder *pbi, + struct aom_read_bit_buffer *rb, + const uint8_t *data, + const uint8_t *data_end, + const uint8_t **p_data_end, + int *frame_decoding_finished) { + AV1_COMMON *const cm = &pbi->common; + uint32_t tile_list_payload_size = 0; + const int num_tiles = cm->tile_cols * cm->tile_rows; + const int start_tile = 0; + const int end_tile = num_tiles - 1; + int i = 0; + + // Process the tile list info. + pbi->output_frame_width_in_tiles_minus_1 = aom_rb_read_literal(rb, 8); + pbi->output_frame_height_in_tiles_minus_1 = aom_rb_read_literal(rb, 8); + pbi->tile_count_minus_1 = aom_rb_read_literal(rb, 16); + if (pbi->tile_count_minus_1 > MAX_TILES - 1) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return 0; + } + + // Allocate output frame buffer for the tile list. + alloc_tile_list_buffer(pbi); + + uint32_t tile_list_info_bytes = 4; + tile_list_payload_size += tile_list_info_bytes; + data += tile_list_info_bytes; + uint8_t *output = pbi->tile_list_output; + + for (i = 0; i <= pbi->tile_count_minus_1; i++) { + // Process 1 tile. + // Reset the bit reader. + rb->bit_offset = 0; + rb->bit_buffer = data; + + // Read out the tile info. + uint32_t tile_info_bytes = 5; + // Set reference for each tile. + int ref_idx = aom_rb_read_literal(rb, 8); + if (ref_idx >= MAX_EXTERNAL_REFERENCES) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return 0; + } + av1_set_reference_dec(cm, 0, 1, &pbi->ext_refs.refs[ref_idx]); + + pbi->dec_tile_row = aom_rb_read_literal(rb, 8); + pbi->dec_tile_col = aom_rb_read_literal(rb, 8); + if (pbi->dec_tile_row < 0 || pbi->dec_tile_col < 0 || + pbi->dec_tile_row >= cm->tile_rows || + pbi->dec_tile_col >= cm->tile_cols) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return 0; + } + + pbi->coded_tile_data_size = aom_rb_read_literal(rb, 16) + 1; + data += tile_info_bytes; + if ((size_t)(data_end - data) < pbi->coded_tile_data_size) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return 0; + } + + av1_decode_tg_tiles_and_wrapup(pbi, data, data + pbi->coded_tile_data_size, + p_data_end, start_tile, end_tile, 0); + uint32_t tile_payload_size = (uint32_t)(*p_data_end - data); + + tile_list_payload_size += tile_info_bytes + tile_payload_size; + + // Update data ptr for next tile decoding. + data = *p_data_end; + assert(data <= data_end); + + // Copy the decoded tile to the tile list output buffer. + copy_decoded_tile_to_tile_list_buffer(pbi, &output); + } + + *frame_decoding_finished = 1; + return tile_list_payload_size; +} + +static void read_metadata_itut_t35(const uint8_t *data, size_t sz) { + struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL }; + for (size_t i = 0; i < sz; i++) { + aom_rb_read_literal(&rb, 8); + } +} + +static void read_metadata_hdr_cll(const uint8_t *data, size_t sz) { + struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL }; + aom_rb_read_literal(&rb, 16); // max_cll + aom_rb_read_literal(&rb, 16); // max_fall +} + +static void read_metadata_hdr_mdcv(const uint8_t *data, size_t sz) { + struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL }; + for (int i = 0; i < 3; i++) { + aom_rb_read_literal(&rb, 16); // primary_i_chromaticity_x + aom_rb_read_literal(&rb, 16); // primary_i_chromaticity_y + } + + aom_rb_read_literal(&rb, 16); // white_point_chromaticity_x + aom_rb_read_literal(&rb, 16); // white_point_chromaticity_y + + aom_rb_read_unsigned_literal(&rb, 32); // luminance_max + aom_rb_read_unsigned_literal(&rb, 32); // luminance_min +} + +static void scalability_structure(struct aom_read_bit_buffer *rb) { + int spatial_layers_cnt = aom_rb_read_literal(rb, 2); + int spatial_layer_dimensions_present_flag = aom_rb_read_bit(rb); + int spatial_layer_description_present_flag = aom_rb_read_bit(rb); + int temporal_group_description_present_flag = aom_rb_read_bit(rb); + aom_rb_read_literal(rb, 3); // reserved + + if (spatial_layer_dimensions_present_flag) { + int i; + for (i = 0; i < spatial_layers_cnt + 1; i++) { + aom_rb_read_literal(rb, 16); + aom_rb_read_literal(rb, 16); + } + } + if (spatial_layer_description_present_flag) { + int i; + for (i = 0; i < spatial_layers_cnt + 1; i++) { + aom_rb_read_literal(rb, 8); + } + } + if (temporal_group_description_present_flag) { + int i, j, temporal_group_size; + temporal_group_size = aom_rb_read_literal(rb, 8); + for (i = 0; i < temporal_group_size; i++) { + aom_rb_read_literal(rb, 3); + aom_rb_read_bit(rb); + aom_rb_read_bit(rb); + int temporal_group_ref_cnt = aom_rb_read_literal(rb, 3); + for (j = 0; j < temporal_group_ref_cnt; j++) { + aom_rb_read_literal(rb, 8); + } + } + } +} + +static void read_metadata_scalability(const uint8_t *data, size_t sz) { + struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL }; + int scalability_mode_idc = aom_rb_read_literal(&rb, 8); + if (scalability_mode_idc == SCALABILITY_SS) { + scalability_structure(&rb); + } +} + +static void read_metadata_timecode(const uint8_t *data, size_t sz) { + struct aom_read_bit_buffer rb = { data, data + sz, 0, NULL, NULL }; + aom_rb_read_literal(&rb, 5); // counting_type f(5) + int full_timestamp_flag = aom_rb_read_bit(&rb); // full_timestamp_flag f(1) + aom_rb_read_bit(&rb); // discontinuity_flag (f1) + aom_rb_read_bit(&rb); // cnt_dropped_flag f(1) + aom_rb_read_literal(&rb, 9); // n_frames f(9) + if (full_timestamp_flag) { + aom_rb_read_literal(&rb, 6); // seconds_value f(6) + aom_rb_read_literal(&rb, 6); // minutes_value f(6) + aom_rb_read_literal(&rb, 5); // hours_value f(5) + } else { + int seconds_flag = aom_rb_read_bit(&rb); // seconds_flag f(1) + if (seconds_flag) { + aom_rb_read_literal(&rb, 6); // seconds_value f(6) + int minutes_flag = aom_rb_read_bit(&rb); // minutes_flag f(1) + if (minutes_flag) { + aom_rb_read_literal(&rb, 6); // minutes_value f(6) + int hours_flag = aom_rb_read_bit(&rb); // hours_flag f(1) + if (hours_flag) { + aom_rb_read_literal(&rb, 5); // hours_value f(5) + } + } + } + } + // time_offset_length f(5) + int time_offset_length = aom_rb_read_literal(&rb, 5); + if (time_offset_length) { + aom_rb_read_literal(&rb, time_offset_length); // f(time_offset_length) + } +} + +static size_t read_metadata(const uint8_t *data, size_t sz) { + size_t type_length; + uint64_t type_value; + OBU_METADATA_TYPE metadata_type; + if (aom_uleb_decode(data, sz, &type_value, &type_length) < 0) { + return sz; + } + metadata_type = (OBU_METADATA_TYPE)type_value; + if (metadata_type == OBU_METADATA_TYPE_ITUT_T35) { + read_metadata_itut_t35(data + type_length, sz - type_length); + } else if (metadata_type == OBU_METADATA_TYPE_HDR_CLL) { + read_metadata_hdr_cll(data + type_length, sz - type_length); + } else if (metadata_type == OBU_METADATA_TYPE_HDR_MDCV) { + read_metadata_hdr_mdcv(data + type_length, sz - type_length); + } else if (metadata_type == OBU_METADATA_TYPE_SCALABILITY) { + read_metadata_scalability(data + type_length, sz - type_length); + } else if (metadata_type == OBU_METADATA_TYPE_TIMECODE) { + read_metadata_timecode(data + type_length, sz - type_length); + } + + return sz; +} + +// On success, returns a boolean that indicates whether the decoding of the +// current frame is finished. On failure, sets cm->error.error_code and +// returns -1. +int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + const uint8_t **p_data_end) { + AV1_COMMON *const cm = &pbi->common; + int frame_decoding_finished = 0; + int is_first_tg_obu_received = 1; + uint32_t frame_header_size = 0; + ObuHeader obu_header; + memset(&obu_header, 0, sizeof(obu_header)); + pbi->seen_frame_header = 0; + + if (data_end < data) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + + // Reset pbi->camera_frame_header_ready to 0 if cm->large_scale_tile = 0. + if (!cm->large_scale_tile) pbi->camera_frame_header_ready = 0; + + // decode frame as a series of OBUs + while (!frame_decoding_finished && !cm->error.error_code) { + struct aom_read_bit_buffer rb; + size_t payload_size = 0; + size_t decoded_payload_size = 0; + size_t obu_payload_offset = 0; + size_t bytes_read = 0; + const size_t bytes_available = data_end - data; + + if (bytes_available == 0 && !pbi->seen_frame_header) { + *p_data_end = data; + cm->error.error_code = AOM_CODEC_OK; + break; + } + + aom_codec_err_t status = + aom_read_obu_header_and_size(data, bytes_available, cm->is_annexb, + &obu_header, &payload_size, &bytes_read); + + if (status != AOM_CODEC_OK) { + cm->error.error_code = status; + return -1; + } + + // Record obu size header information. + pbi->obu_size_hdr.data = data + obu_header.size; + pbi->obu_size_hdr.size = bytes_read - obu_header.size; + + // Note: aom_read_obu_header_and_size() takes care of checking that this + // doesn't cause 'data' to advance past 'data_end'. + data += bytes_read; + + if ((size_t)(data_end - data) < payload_size) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + + cm->temporal_layer_id = obu_header.temporal_layer_id; + cm->spatial_layer_id = obu_header.spatial_layer_id; + + if (obu_header.type != OBU_TEMPORAL_DELIMITER && + obu_header.type != OBU_SEQUENCE_HEADER && + obu_header.type != OBU_PADDING) { + // don't decode obu if it's not in current operating mode + if (!is_obu_in_current_operating_point(pbi, obu_header)) { + data += payload_size; + continue; + } + } + + av1_init_read_bit_buffer(pbi, &rb, data, data + payload_size); + + switch (obu_header.type) { + case OBU_TEMPORAL_DELIMITER: + decoded_payload_size = read_temporal_delimiter_obu(); + pbi->seen_frame_header = 0; + break; + case OBU_SEQUENCE_HEADER: + decoded_payload_size = read_sequence_header_obu(pbi, &rb); + if (cm->error.error_code != AOM_CODEC_OK) return -1; + break; + case OBU_FRAME_HEADER: + case OBU_REDUNDANT_FRAME_HEADER: + case OBU_FRAME: + // Only decode first frame header received + if (!pbi->seen_frame_header || + (cm->large_scale_tile && !pbi->camera_frame_header_ready)) { + frame_header_size = read_frame_header_obu( + pbi, &rb, data, p_data_end, obu_header.type != OBU_FRAME); + pbi->seen_frame_header = 1; + if (!pbi->ext_tile_debug && cm->large_scale_tile) + pbi->camera_frame_header_ready = 1; + } else { + // TODO(wtc): Verify that the frame_header_obu is identical to the + // original frame_header_obu. For now just skip frame_header_size + // bytes in the bit buffer. + if (frame_header_size > payload_size) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + assert(rb.bit_offset == 0); + rb.bit_offset = 8 * frame_header_size; + } + + decoded_payload_size = frame_header_size; + pbi->frame_header_size = frame_header_size; + + if (cm->show_existing_frame) { + if (obu_header.type == OBU_FRAME) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return -1; + } + frame_decoding_finished = 1; + pbi->seen_frame_header = 0; + break; + } + + // In large scale tile coding, decode the common camera frame header + // before any tile list OBU. + if (!pbi->ext_tile_debug && pbi->camera_frame_header_ready) { + frame_decoding_finished = 1; + // Skip the rest of the frame data. + decoded_payload_size = payload_size; + // Update data_end. + *p_data_end = data_end; + break; + } + + if (obu_header.type != OBU_FRAME) break; + obu_payload_offset = frame_header_size; + // Byte align the reader before reading the tile group. + if (byte_alignment(cm, &rb)) return -1; + AOM_FALLTHROUGH_INTENDED; // fall through to read tile group. + case OBU_TILE_GROUP: + if (!pbi->seen_frame_header) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + if (obu_payload_offset > payload_size) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + decoded_payload_size += read_one_tile_group_obu( + pbi, &rb, is_first_tg_obu_received, data + obu_payload_offset, + data + payload_size, p_data_end, &frame_decoding_finished, + obu_header.type == OBU_FRAME); + is_first_tg_obu_received = 0; + if (frame_decoding_finished) pbi->seen_frame_header = 0; + break; + case OBU_METADATA: + decoded_payload_size = read_metadata(data, payload_size); + break; + case OBU_TILE_LIST: + if (CONFIG_NORMAL_TILE_MODE) { + cm->error.error_code = AOM_CODEC_UNSUP_BITSTREAM; + return -1; + } + + // This OBU type is purely for the large scale tile coding mode. + // The common camera frame header has to be already decoded. + if (!pbi->camera_frame_header_ready) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + + cm->large_scale_tile = 1; + av1_set_single_tile_decoding_mode(cm); + decoded_payload_size = + read_and_decode_one_tile_list(pbi, &rb, data, data + payload_size, + p_data_end, &frame_decoding_finished); + if (cm->error.error_code != AOM_CODEC_OK) return -1; + break; + case OBU_PADDING: + default: + // Skip unrecognized OBUs + decoded_payload_size = payload_size; + break; + } + + // Check that the signalled OBU size matches the actual amount of data read + if (decoded_payload_size > payload_size) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + + // If there are extra padding bytes, they should all be zero + while (decoded_payload_size < payload_size) { + uint8_t padding_byte = data[decoded_payload_size++]; + if (padding_byte != 0) { + cm->error.error_code = AOM_CODEC_CORRUPT_FRAME; + return -1; + } + } + + data += payload_size; + } + + return frame_decoding_finished; +} diff --git a/media/libaom/src/av1/decoder/obu.h b/media/libaom/src/av1/decoder/obu.h new file mode 100644 index 000000000..5ab243fc9 --- /dev/null +++ b/media/libaom/src/av1/decoder/obu.h @@ -0,0 +1,31 @@ +/* + * 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. + */ + +#ifndef AOM_AV1_DECODER_OBU_H_ +#define AOM_AV1_DECODER_OBU_H_ + +#include "aom/aom_codec.h" +#include "av1/decoder/decoder.h" + +// Try to decode one frame from a buffer. +// Returns 1 if we decoded a frame, +// 0 if we didn't decode a frame but that's okay +// (eg, if there was a frame but we skipped it), +// or -1 on error +int aom_decode_frame_from_obus(struct AV1Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, + const uint8_t **p_data_end); + +aom_codec_err_t aom_get_num_layers_from_operating_point_idc( + int operating_point_idc, unsigned int *num_spatial_layers, + unsigned int *num_temporal_layers); + +#endif // AOM_AV1_DECODER_OBU_H_ |