Port folder renaming changes from AOM Manually cherry-picked commits: ceef058 libvpx->libaom part2 3d26d91 libvpx -> libaom cfea7dd vp10/ -> av1/ 3a8eff7 Fix a build issue for a test bf4202e Rename vpx to aom Change-Id: I1b0eb5a40796e3aaf41c58984b4229a439a597dc
diff --git a/av1/decoder/bitreader.h b/av1/decoder/bitreader.h new file mode 100644 index 0000000..75d6aa4 --- /dev/null +++ b/av1/decoder/bitreader.h
@@ -0,0 +1,38 @@ +/* + * Copyright (c) 2016 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +/* The purpose of this header is to provide compile time pluggable bit reader + * implementations with a common interface. */ + +#ifndef VPX10_DECODER_BITREADER_H_ +#define VPX10_DECODER_BITREADER_H_ + +#include "./vpx_config.h" + +#if CONFIG_ANS +#include "av1/common/ans.h" +#include "aom/vp8dx.h" // for vp10_decrypt_cb +#define vp10_reader struct AnsDecoder +#define vp10_reader_has_error ans_reader_has_error +#define vp10_read uabs_read +#define vp10_read_bit uabs_read_bit +#define vp10_read_literal uabs_read_literal +#define vp10_read_tree uabs_read_tree +#else +#include "aom_dsp/bitreader.h" +#define vp10_reader vpx_reader +#define vp10_reader_has_error vpx_reader_has_error +#define vp10_read vpx_read +#define vp10_read_bit vpx_read_bit +#define vp10_read_literal vpx_read_literal +#define vp10_read_tree vpx_read_tree +#endif + +#endif // VPX10_DECODER_BITREADER_H_
diff --git a/av1/decoder/decodeframe.c b/av1/decoder/decodeframe.c new file mode 100644 index 0000000..0f90c20 --- /dev/null +++ b/av1/decoder/decodeframe.c
@@ -0,0 +1,3882 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include <stdlib.h> // qsort() + +#include "./vp10_rtcd.h" +#include "./vpx_config.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" +#include "./vpx_config.h" + +#include "aom_dsp/bitreader_buffer.h" +#include "av1/decoder/bitreader.h" +#include "aom_dsp/vpx_dsp_common.h" +#include "aom_mem/vpx_mem.h" +#include "aom_ports/mem.h" +#include "aom_ports/mem_ops.h" +#include "aom_scale/vpx_scale.h" +#include "aom_util/vpx_thread.h" + +#include "av1/common/alloccommon.h" +#if CONFIG_CLPF +#include "av1/common/clpf.h" +#endif +#include "av1/common/common.h" +#if CONFIG_DERING +#include "av1/common/dering.h" +#endif // CONFIG_DERING +#include "av1/common/entropy.h" +#include "av1/common/entropymode.h" +#include "av1/common/idct.h" +#include "av1/common/thread_common.h" +#include "av1/common/pred_common.h" +#include "av1/common/quant_common.h" +#include "av1/common/reconintra.h" +#include "av1/common/reconinter.h" +#include "av1/common/seg_common.h" +#include "av1/common/tile_common.h" + +#include "av1/decoder/decodeframe.h" +#include "av1/decoder/detokenize.h" +#include "av1/decoder/decodemv.h" +#include "av1/decoder/decoder.h" +#include "av1/decoder/dsubexp.h" + +#define MAX_VPX_HEADER_SIZE 80 + +static int is_compound_reference_allowed(const VP10_COMMON *cm) { + int i; + if (frame_is_intra_only(cm)) return 0; + for (i = 1; i < INTER_REFS_PER_FRAME; ++i) + if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) return 1; + + return 0; +} + +static void setup_compound_reference_mode(VP10_COMMON *cm) { +#if CONFIG_EXT_REFS + 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] = ALTREF_FRAME; +#else + if (cm->ref_frame_sign_bias[LAST_FRAME] == + cm->ref_frame_sign_bias[GOLDEN_FRAME]) { + cm->comp_fixed_ref = ALTREF_FRAME; + cm->comp_var_ref[0] = LAST_FRAME; + cm->comp_var_ref[1] = GOLDEN_FRAME; + } else if (cm->ref_frame_sign_bias[LAST_FRAME] == + cm->ref_frame_sign_bias[ALTREF_FRAME]) { + cm->comp_fixed_ref = GOLDEN_FRAME; + cm->comp_var_ref[0] = LAST_FRAME; + cm->comp_var_ref[1] = ALTREF_FRAME; + } else { + cm->comp_fixed_ref = LAST_FRAME; + cm->comp_var_ref[0] = GOLDEN_FRAME; + cm->comp_var_ref[1] = ALTREF_FRAME; + } +#endif // CONFIG_EXT_REFS +} + +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 int decode_unsigned_max(struct vpx_read_bit_buffer *rb, int max) { + const int data = vpx_rb_read_literal(rb, get_unsigned_bits(max)); + return data > max ? max : data; +} + +static TX_MODE read_tx_mode(struct vpx_read_bit_buffer *rb) { + return vpx_rb_read_bit(rb) ? TX_MODE_SELECT : vpx_rb_read_literal(rb, 2); +} + +static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i, j; + for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) + for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i) + vp10_diff_update_prob(r, &fc->switchable_interp_prob[j][i]); +} + +static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i; +#if CONFIG_REF_MV + for (i = 0; i < NEWMV_MODE_CONTEXTS; ++i) + vp10_diff_update_prob(r, &fc->newmv_prob[i]); + for (i = 0; i < ZEROMV_MODE_CONTEXTS; ++i) + vp10_diff_update_prob(r, &fc->zeromv_prob[i]); + for (i = 0; i < REFMV_MODE_CONTEXTS; ++i) + vp10_diff_update_prob(r, &fc->refmv_prob[i]); + for (i = 0; i < DRL_MODE_CONTEXTS; ++i) + vp10_diff_update_prob(r, &fc->drl_prob[i]); +#if CONFIG_EXT_INTER + vp10_diff_update_prob(r, &fc->new2mv_prob); +#endif // CONFIG_EXT_INTER +#else + int j; + for (i = 0; i < INTER_MODE_CONTEXTS; ++i) + for (j = 0; j < INTER_MODES - 1; ++j) + vp10_diff_update_prob(r, &fc->inter_mode_probs[i][j]); +#endif +} + +#if CONFIG_EXT_INTER +static void read_inter_compound_mode_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i, j; + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (j = 0; j < INTER_MODE_CONTEXTS; ++j) { + for (i = 0; i < INTER_COMPOUND_MODES - 1; ++i) { + vp10_diff_update_prob(r, &fc->inter_compound_mode_probs[j][i]); + } + } + } +} +#endif // CONFIG_EXT_INTER + +static REFERENCE_MODE read_frame_reference_mode( + const VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + if (is_compound_reference_allowed(cm)) { + return vpx_rb_read_bit(rb) + ? REFERENCE_MODE_SELECT + : (vpx_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE); + } else { + return SINGLE_REFERENCE; + } +} + +static void read_frame_reference_mode_probs(VP10_COMMON *cm, vp10_reader *r) { + FRAME_CONTEXT *const fc = cm->fc; + int i, j; + + if (cm->reference_mode == REFERENCE_MODE_SELECT) + for (i = 0; i < COMP_INTER_CONTEXTS; ++i) + vp10_diff_update_prob(r, &fc->comp_inter_prob[i]); + + if (cm->reference_mode != COMPOUND_REFERENCE) { + for (i = 0; i < REF_CONTEXTS; ++i) { + for (j = 0; j < (SINGLE_REFS - 1); ++j) { + vp10_diff_update_prob(r, &fc->single_ref_prob[i][j]); + } + } + } + + if (cm->reference_mode != SINGLE_REFERENCE) { + for (i = 0; i < REF_CONTEXTS; ++i) { +#if CONFIG_EXT_REFS + for (j = 0; j < (FWD_REFS - 1); ++j) + vp10_diff_update_prob(r, &fc->comp_ref_prob[i][j]); + for (j = 0; j < (BWD_REFS - 1); ++j) + vp10_diff_update_prob(r, &fc->comp_bwdref_prob[i][j]); +#else + for (j = 0; j < (COMP_REFS - 1); ++j) + vp10_diff_update_prob(r, &fc->comp_ref_prob[i][j]); +#endif // CONFIG_EXT_REFS + } + } +} + +static void update_mv_probs(vpx_prob *p, int n, vp10_reader *r) { + int i; + for (i = 0; i < n; ++i) vp10_diff_update_prob(r, &p[i]); +} + +static void read_mv_probs(nmv_context *ctx, int allow_hp, vp10_reader *r) { + int i, j; + + update_mv_probs(ctx->joints, MV_JOINTS - 1, r); + +#if CONFIG_REF_MV + vp10_diff_update_prob(r, &ctx->zero_rmv); +#endif + + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + update_mv_probs(&comp_ctx->sign, 1, r); + update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r); + update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r); + update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r); + } + + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + for (j = 0; j < CLASS0_SIZE; ++j) + update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r); + update_mv_probs(comp_ctx->fp, 3, r); + } + + if (allow_hp) { + for (i = 0; i < 2; ++i) { + nmv_component *const comp_ctx = &ctx->comps[i]; + update_mv_probs(&comp_ctx->class0_hp, 1, r); + update_mv_probs(&comp_ctx->hp, 1, r); + } + } +} + +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 eob) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + if (eob > 0) { + tran_low_t *const dqcoeff = pd->dqcoeff; + INV_TXFM_PARAM inv_txfm_param; + inv_txfm_param.tx_type = tx_type; + inv_txfm_param.tx_size = tx_size; + inv_txfm_param.eob = eob; + inv_txfm_param.lossless = xd->lossless[xd->mi[0]->mbmi.segment_id]; + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + inv_txfm_param.bd = xd->bd; + highbd_inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + inv_txfm_add(dqcoeff, dst, stride, &inv_txfm_param); +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + + if (eob == 1) { + dqcoeff[0] = 0; + } else { + if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10) + memset(dqcoeff, 0, 4 * 4 * num_4x4_blocks_wide_txsize_lookup[tx_size] * + sizeof(dqcoeff[0])); +#if CONFIG_EXT_TX + else + memset(dqcoeff, 0, get_tx2d_size(tx_size) * sizeof(dqcoeff[0])); +#else + else if (tx_size == TX_32X32 && eob <= 34) + memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0])); + else + memset(dqcoeff, 0, get_tx2d_size(tx_size) * sizeof(dqcoeff[0])); +#endif + } + } +} + +static void predict_and_reconstruct_intra_block(MACROBLOCKD *const xd, +#if CONFIG_ANS + struct AnsDecoder *const r, +#else + vp10_reader *r, +#endif // CONFIG_ANS + MB_MODE_INFO *const mbmi, + int plane, int row, int col, + TX_SIZE tx_size) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + PREDICTION_MODE mode = (plane == 0) ? mbmi->mode : mbmi->uv_mode; + PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV; + uint8_t *dst; + int block_idx = (row << 1) + col; + dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col]; + + if (mbmi->sb_type < BLOCK_8X8) + if (plane == 0) mode = xd->mi[0]->bmi[(row << 1) + col].as_mode; + + vp10_predict_intra_block(xd, pd->n4_wl, pd->n4_hl, tx_size, mode, dst, + pd->dst.stride, dst, pd->dst.stride, col, row, + plane); + + if (!mbmi->skip) { + TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size); + const scan_order *sc = get_scan(tx_size, tx_type, 0); + const int eob = vp10_decode_block_tokens(xd, plane, sc, col, row, tx_size, + tx_type, r, mbmi->segment_id); + inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride, + eob); + } +} + +#if CONFIG_VAR_TX +static void decode_reconstruct_tx(MACROBLOCKD *const xd, vp10_reader *r, + MB_MODE_INFO *const mbmi, int plane, + BLOCK_SIZE plane_bsize, int block, + int blk_row, int blk_col, TX_SIZE tx_size, + int *eob_total) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const BLOCK_SIZE bsize = txsize_to_bsize[tx_size]; + const int tx_row = blk_row >> (1 - pd->subsampling_y); + const int tx_col = blk_col >> (1 - pd->subsampling_x); + const TX_SIZE plane_tx_size = + plane ? get_uv_tx_size_impl(mbmi->inter_tx_size[tx_row][tx_col], bsize, 0, + 0) + : mbmi->inter_tx_size[tx_row][tx_col]; + int max_blocks_high = num_4x4_blocks_high_lookup[plane_bsize]; + int max_blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize]; + + if (xd->mb_to_bottom_edge < 0) + max_blocks_high += xd->mb_to_bottom_edge >> (5 + pd->subsampling_y); + if (xd->mb_to_right_edge < 0) + max_blocks_wide += xd->mb_to_right_edge >> (5 + pd->subsampling_x); + + if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return; + + if (tx_size == plane_tx_size) { + PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV; + TX_TYPE tx_type = get_tx_type(plane_type, xd, block, plane_tx_size); + const scan_order *sc = get_scan(plane_tx_size, tx_type, 1); + const int eob = + vp10_decode_block_tokens(xd, plane, sc, blk_col, blk_row, plane_tx_size, + tx_type, r, mbmi->segment_id); + inverse_transform_block( + xd, plane, tx_type, plane_tx_size, + &pd->dst.buf[4 * blk_row * pd->dst.stride + 4 * blk_col], + pd->dst.stride, eob); + *eob_total += eob; + } else { + int bsl = b_width_log2_lookup[bsize]; + int i; + + assert(bsl > 0); + --bsl; + + for (i = 0; i < 4; ++i) { + const int offsetr = blk_row + ((i >> 1) << bsl); + const int offsetc = blk_col + ((i & 0x01) << bsl); + int step = num_4x4_blocks_txsize_lookup[tx_size - 1]; + + if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue; + + decode_reconstruct_tx(xd, r, mbmi, plane, plane_bsize, block + i * step, + offsetr, offsetc, tx_size - 1, eob_total); + } + } +} +#endif // CONFIG_VAR_TX + +#if !CONFIG_VAR_TX || CONFIG_SUPERTX || (CONFIG_EXT_TX && CONFIG_RECT_TX) +static int reconstruct_inter_block(MACROBLOCKD *const xd, +#if CONFIG_ANS + struct AnsDecoder *const r, +#else + vp10_reader *r, +#endif + int segment_id, int plane, int row, int col, + TX_SIZE tx_size) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV; + int block_idx = (row << 1) + col; + TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, tx_size); + const scan_order *sc = get_scan(tx_size, tx_type, 1); + const int eob = vp10_decode_block_tokens(xd, plane, sc, col, row, tx_size, + tx_type, r, segment_id); + + inverse_transform_block(xd, plane, tx_type, tx_size, + &pd->dst.buf[4 * row * pd->dst.stride + 4 * col], + pd->dst.stride, eob); + return eob; +} +#endif // !CONFIG_VAR_TX || CONFIG_SUPER_TX + +static INLINE TX_SIZE dec_get_uv_tx_size(const MB_MODE_INFO *mbmi, int n4_wl, + int n4_hl) { + // get minimum log2 num4x4s dimension + const int x = VPXMIN(n4_wl, n4_hl); + return VPXMIN(txsize_sqr_map[mbmi->tx_size], x); +} + +static INLINE void dec_reset_skip_context(MACROBLOCKD *xd) { + int i; + for (i = 0; i < MAX_MB_PLANE; i++) { + struct macroblockd_plane *const pd = &xd->plane[i]; + memset(pd->above_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_w); + memset(pd->left_context, 0, sizeof(ENTROPY_CONTEXT) * pd->n4_h); + } +} + +static void set_plane_n4(MACROBLOCKD *const xd, int bw, int bh, int bwl, + int bhl) { + int i; + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].n4_w = (bw << 1) >> xd->plane[i].subsampling_x; + xd->plane[i].n4_h = (bh << 1) >> xd->plane[i].subsampling_y; + xd->plane[i].n4_wl = bwl - xd->plane[i].subsampling_x; + xd->plane[i].n4_hl = bhl - xd->plane[i].subsampling_y; + } +} + +static MB_MODE_INFO *set_offsets(VP10_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, int bwl, + int bhl) { + const int offset = mi_row * cm->mi_stride + mi_col; + int x, y; + 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]->mbmi.sb_type = bsize; + for (y = 0; y < y_mis; ++y) + for (x = !y; x < x_mis; ++x) { + xd->mi[y * cm->mi_stride + x] = xd->mi[0]; + } + + set_plane_n4(xd, bw, bh, bwl, bhl); + + set_skip_context(xd, mi_row, mi_col); + +#if CONFIG_VAR_TX + xd->max_tx_size = max_txsize_lookup[bsize]; +#endif + + // 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); + + vp10_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col); + return &xd->mi[0]->mbmi; +} + +#if CONFIG_SUPERTX +static MB_MODE_INFO *set_offsets_extend(VP10_COMMON *const cm, + MACROBLOCKD *const xd, + const TileInfo *const tile, + BLOCK_SIZE bsize_pred, int mi_row_pred, + int mi_col_pred, int mi_row_ori, + int mi_col_ori) { + // Used in supertx + // (mi_row_ori, mi_col_ori): location for mv + // (mi_row_pred, mi_col_pred, bsize_pred): region to predict + const int bw = num_8x8_blocks_wide_lookup[bsize_pred]; + const int bh = num_8x8_blocks_high_lookup[bsize_pred]; + const int offset = mi_row_ori * cm->mi_stride + mi_col_ori; + const int bwl = b_width_log2_lookup[bsize_pred]; + const int bhl = b_height_log2_lookup[bsize_pred]; + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = cm->mi + offset; + set_mi_row_col(xd, tile, mi_row_pred, bh, mi_col_pred, bw, cm->mi_rows, + cm->mi_cols); + + xd->up_available = (mi_row_ori > tile->mi_row_start); + xd->left_available = (mi_col_ori > tile->mi_col_start); + + set_plane_n4(xd, bw, bh, bwl, bhl); + + return &xd->mi[0]->mbmi; +} + +static MB_MODE_INFO *set_mb_offsets(VP10_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 offset = mi_row * cm->mi_stride + mi_col; + const TileInfo *const tile = &xd->tile; + int x, y; + + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = cm->mi + offset; + xd->mi[0]->mbmi.sb_type = bsize; + for (y = 0; y < y_mis; ++y) + for (x = !y; x < x_mis; ++x) xd->mi[y * cm->mi_stride + x] = xd->mi[0]; + + set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols); + return &xd->mi[0]->mbmi; +} + +static void set_offsets_topblock(VP10_COMMON *const cm, MACROBLOCKD *const xd, + const TileInfo *const tile, BLOCK_SIZE bsize, + int mi_row, int mi_col) { + const int bw = num_8x8_blocks_wide_lookup[bsize]; + const int bh = num_8x8_blocks_high_lookup[bsize]; + const int offset = mi_row * cm->mi_stride + mi_col; + const int bwl = b_width_log2_lookup[bsize]; + const int bhl = b_height_log2_lookup[bsize]; + + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = cm->mi + offset; + + set_plane_n4(xd, bw, bh, bwl, bhl); + + set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols); + + vp10_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col); +} + +static void set_param_topblock(VP10_COMMON *const cm, MACROBLOCKD *const xd, + BLOCK_SIZE bsize, int mi_row, int mi_col, + int txfm, int skip) { + const int bw = num_8x8_blocks_wide_lookup[bsize]; + const int bh = num_8x8_blocks_high_lookup[bsize]; + const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col); + const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row); + const int offset = mi_row * cm->mi_stride + mi_col; + int x, y; + + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = cm->mi + offset; + + for (y = 0; y < y_mis; ++y) + for (x = 0; x < x_mis; ++x) { + xd->mi[y * cm->mi_stride + x]->mbmi.skip = skip; + xd->mi[y * cm->mi_stride + x]->mbmi.tx_type = txfm; + } +#if CONFIG_VAR_TX + xd->above_txfm_context = cm->above_txfm_context + mi_col; + xd->left_txfm_context = + xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK); + set_txfm_ctxs(xd->mi[0]->mbmi.tx_size, bw, bh, xd); +#endif +} + +static void set_ref(VP10_COMMON *const cm, MACROBLOCKD *const xd, int idx, + int mi_row, int mi_col) { + MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi; + RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME]; + xd->block_refs[idx] = ref_buffer; + if (!vp10_is_valid_scale(&ref_buffer->sf)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid scale factors"); + vp10_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col, + &ref_buffer->sf); + xd->corrupted |= ref_buffer->buf->corrupted; +} + +static void dec_predict_b_extend( + VP10Decoder *const pbi, MACROBLOCKD *const xd, const TileInfo *const tile, + int block, int mi_row_ori, int mi_col_ori, int mi_row_pred, int mi_col_pred, + int mi_row_top, int mi_col_top, uint8_t *dst_buf[3], int dst_stride[3], + BLOCK_SIZE bsize_top, BLOCK_SIZE bsize_pred, int b_sub8x8, int bextend) { + // Used in supertx + // (mi_row_ori, mi_col_ori): location for mv + // (mi_row_pred, mi_col_pred, bsize_pred): region to predict + // (mi_row_top, mi_col_top, bsize_top): region of the top partition size + // block: sub location of sub8x8 blocks + // b_sub8x8: 1: ori is sub8x8; 0: ori is not sub8x8 + // bextend: 1: region to predict is an extension of ori; 0: not + int r = (mi_row_pred - mi_row_top) * MI_SIZE; + int c = (mi_col_pred - mi_col_top) * MI_SIZE; + const int mi_width_top = num_8x8_blocks_wide_lookup[bsize_top]; + const int mi_height_top = num_8x8_blocks_high_lookup[bsize_top]; + MB_MODE_INFO *mbmi; + VP10_COMMON *const cm = &pbi->common; + + if (mi_row_pred < mi_row_top || mi_col_pred < mi_col_top || + mi_row_pred >= mi_row_top + mi_height_top || + mi_col_pred >= mi_col_top + mi_width_top || mi_row_pred >= cm->mi_rows || + mi_col_pred >= cm->mi_cols) + return; + + mbmi = set_offsets_extend(cm, xd, tile, bsize_pred, mi_row_pred, mi_col_pred, + mi_row_ori, mi_col_ori); + set_ref(cm, xd, 0, mi_row_pred, mi_col_pred); + if (has_second_ref(&xd->mi[0]->mbmi)) + set_ref(cm, xd, 1, mi_row_pred, mi_col_pred); + + if (!bextend) { + mbmi->tx_size = b_width_log2_lookup[bsize_top]; + } + + xd->plane[0].dst.stride = dst_stride[0]; + xd->plane[1].dst.stride = dst_stride[1]; + xd->plane[2].dst.stride = dst_stride[2]; + xd->plane[0].dst.buf = dst_buf[0] + + (r >> xd->plane[0].subsampling_y) * dst_stride[0] + + (c >> xd->plane[0].subsampling_x); + xd->plane[1].dst.buf = dst_buf[1] + + (r >> xd->plane[1].subsampling_y) * dst_stride[1] + + (c >> xd->plane[1].subsampling_x); + xd->plane[2].dst.buf = dst_buf[2] + + (r >> xd->plane[2].subsampling_y) * dst_stride[2] + + (c >> xd->plane[2].subsampling_x); + + if (!b_sub8x8) + vp10_build_inter_predictors_sb_extend(xd, +#if CONFIG_EXT_INTER + mi_row_ori, mi_col_ori, +#endif // CONFIG_EXT_INTER + mi_row_pred, mi_col_pred, bsize_pred); + else + vp10_build_inter_predictors_sb_sub8x8_extend(xd, +#if CONFIG_EXT_INTER + mi_row_ori, mi_col_ori, +#endif // CONFIG_EXT_INTER + mi_row_pred, mi_col_pred, + bsize_pred, block); +} + +static void dec_extend_dir(VP10Decoder *const pbi, MACROBLOCKD *const xd, + const TileInfo *const tile, int block, + BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row, + int mi_col, int mi_row_top, int mi_col_top, + uint8_t *dst_buf[3], int dst_stride[3], int dir) { + // dir: 0-lower, 1-upper, 2-left, 3-right + // 4-lowerleft, 5-upperleft, 6-lowerright, 7-upperright + const int mi_width = num_8x8_blocks_wide_lookup[bsize]; + const int mi_height = num_8x8_blocks_high_lookup[bsize]; + int xss = xd->plane[1].subsampling_x; + int yss = xd->plane[1].subsampling_y; + int b_sub8x8 = (bsize < BLOCK_8X8) ? 1 : 0; + BLOCK_SIZE extend_bsize; + int unit, mi_row_pred, mi_col_pred; + + if (dir == 0 || dir == 1) { + extend_bsize = (mi_width == 1 || bsize < BLOCK_8X8 || xss < yss) + ? BLOCK_8X8 + : BLOCK_16X8; + unit = num_8x8_blocks_wide_lookup[extend_bsize]; + mi_row_pred = mi_row + ((dir == 0) ? mi_height : -1); + mi_col_pred = mi_col; + + dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred, + mi_col_pred, mi_row_top, mi_col_top, dst_buf, + dst_stride, top_bsize, extend_bsize, b_sub8x8, 1); + + if (mi_width > unit) { + int i; + assert(!b_sub8x8); + for (i = 0; i < mi_width / unit - 1; i++) { + mi_col_pred += unit; + dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred, + mi_col_pred, mi_row_top, mi_col_top, dst_buf, + dst_stride, top_bsize, extend_bsize, b_sub8x8, 1); + } + } + } else if (dir == 2 || dir == 3) { + extend_bsize = (mi_height == 1 || bsize < BLOCK_8X8 || yss < xss) + ? BLOCK_8X8 + : BLOCK_8X16; + unit = num_8x8_blocks_high_lookup[extend_bsize]; + mi_row_pred = mi_row; + mi_col_pred = mi_col + ((dir == 3) ? mi_width : -1); + + dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred, + mi_col_pred, mi_row_top, mi_col_top, dst_buf, + dst_stride, top_bsize, extend_bsize, b_sub8x8, 1); + + if (mi_height > unit) { + int i; + for (i = 0; i < mi_height / unit - 1; i++) { + mi_row_pred += unit; + dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred, + mi_col_pred, mi_row_top, mi_col_top, dst_buf, + dst_stride, top_bsize, extend_bsize, b_sub8x8, 1); + } + } + } else { + extend_bsize = BLOCK_8X8; + mi_row_pred = mi_row + ((dir == 4 || dir == 6) ? mi_height : -1); + mi_col_pred = mi_col + ((dir == 6 || dir == 7) ? mi_width : -1); + dec_predict_b_extend(pbi, xd, tile, block, mi_row, mi_col, mi_row_pred, + mi_col_pred, mi_row_top, mi_col_top, dst_buf, + dst_stride, top_bsize, extend_bsize, b_sub8x8, 1); + } +} + +static void dec_extend_all(VP10Decoder *const pbi, MACROBLOCKD *const xd, + const TileInfo *const tile, int block, + BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, int mi_row, + int mi_col, int mi_row_top, int mi_col_top, + uint8_t *dst_buf[3], int dst_stride[3]) { + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 0); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 1); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 2); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 3); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 4); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 5); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 6); + dec_extend_dir(pbi, xd, tile, block, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 7); +} + +static void dec_predict_sb_complex(VP10Decoder *const pbi, + MACROBLOCKD *const xd, + const TileInfo *const tile, int mi_row, + int mi_col, int mi_row_top, int mi_col_top, + BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, + uint8_t *dst_buf[3], int dst_stride[3]) { + const VP10_COMMON *const cm = &pbi->common; + const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2; + const PARTITION_TYPE partition = get_partition(cm, mi_row, mi_col, bsize); + const BLOCK_SIZE subsize = get_subsize(bsize, partition); +#if CONFIG_EXT_PARTITION_TYPES + const BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT); +#endif + int i; + const int mi_offset = mi_row * cm->mi_stride + mi_col; + uint8_t *dst_buf1[3], *dst_buf2[3], *dst_buf3[3]; + + DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_TX_SQUARE * 2]); + DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_TX_SQUARE * 2]); + DECLARE_ALIGNED(16, uint8_t, tmp_buf3[MAX_MB_PLANE * MAX_TX_SQUARE * 2]); + int dst_stride1[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE }; + int dst_stride2[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE }; + int dst_stride3[3] = { MAX_TX_SIZE, MAX_TX_SIZE, MAX_TX_SIZE }; + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + int len = sizeof(uint16_t); + dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1); + dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_TX_SQUARE * len); + dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + 2 * MAX_TX_SQUARE * len); + dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2); + dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_TX_SQUARE * len); + dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + 2 * MAX_TX_SQUARE * len); + dst_buf3[0] = CONVERT_TO_BYTEPTR(tmp_buf3); + dst_buf3[1] = CONVERT_TO_BYTEPTR(tmp_buf3 + MAX_TX_SQUARE * len); + dst_buf3[2] = CONVERT_TO_BYTEPTR(tmp_buf3 + 2 * MAX_TX_SQUARE * len); + } else { +#endif + dst_buf1[0] = tmp_buf1; + dst_buf1[1] = tmp_buf1 + MAX_TX_SQUARE; + dst_buf1[2] = tmp_buf1 + 2 * MAX_TX_SQUARE; + dst_buf2[0] = tmp_buf2; + dst_buf2[1] = tmp_buf2 + MAX_TX_SQUARE; + dst_buf2[2] = tmp_buf2 + 2 * MAX_TX_SQUARE; + dst_buf3[0] = tmp_buf3; + dst_buf3[1] = tmp_buf3 + MAX_TX_SQUARE; + dst_buf3[2] = tmp_buf3 + 2 * MAX_TX_SQUARE; +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif + + if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return; + + xd->mi = cm->mi_grid_visible + mi_offset; + xd->mi[0] = cm->mi + mi_offset; + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + } + + switch (partition) { + case PARTITION_NONE: + assert(bsize < top_bsize); + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, bsize, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + break; + case PARTITION_HORZ: + if (bsize == BLOCK_8X8) { + // For sub8x8, predict in 8x8 unit + // First half + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, BLOCK_8X8, 1, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + + // Second half + dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1, + top_bsize, BLOCK_8X8, 1, 1); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + // weighted average to smooth the boundary + xd->plane[0].dst.buf = dst_buf[0]; + xd->plane[0].dst.stride = dst_stride[0]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ, + 0); + } else { + // First half + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 0); + + if (mi_row + hbs < cm->mi_rows) { + // Second half + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, + mi_row + hbs, mi_col, mi_row_top, mi_col_top, + dst_buf1, dst_stride1, top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf1, + dst_stride1); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, 1); + + // weighted average to smooth the boundary + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_HORZ, i); + } + } + } + break; + case PARTITION_VERT: + if (bsize == BLOCK_8X8) { + // First half + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, BLOCK_8X8, 1, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + + // Second half + dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1, + top_bsize, BLOCK_8X8, 1, 1); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + // Smooth + xd->plane[0].dst.buf = dst_buf[0]; + xd->plane[0].dst.stride = dst_stride[0]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[0], dst_stride[0], dst_buf1[0], dst_stride1[0], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT, + 0); + } else { + // First half + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 3); + + // Second half + if (mi_col + hbs < cm->mi_cols) { + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf1, + dst_stride1); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, 2); + + // Smooth + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_VERT, i); + } + } + } + break; + case PARTITION_SPLIT: + if (bsize == BLOCK_8X8) { + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, BLOCK_8X8, 1, 0); + dec_predict_b_extend(pbi, xd, tile, 1, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1, + top_bsize, BLOCK_8X8, 1, 1); + dec_predict_b_extend(pbi, xd, tile, 2, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf2, dst_stride2, + top_bsize, BLOCK_8X8, 1, 1); + dec_predict_b_extend(pbi, xd, tile, 3, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf3, dst_stride3, + top_bsize, BLOCK_8X8, 1, 1); + if (bsize < top_bsize) { + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + dec_extend_all(pbi, xd, tile, 1, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + dec_extend_all(pbi, xd, tile, 2, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf2, dst_stride2); + dec_extend_all(pbi, xd, tile, 3, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf3, dst_stride3); + } + } else { + dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row_top, + mi_col_top, subsize, top_bsize, dst_buf, + dst_stride); + if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) + dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col + hbs, + mi_row_top, mi_col_top, subsize, top_bsize, + dst_buf1, dst_stride1); + if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) + dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col, + mi_row_top, mi_col_top, subsize, top_bsize, + dst_buf2, dst_stride2); + if (mi_row + hbs < cm->mi_rows && mi_col + hbs < cm->mi_cols) + dec_predict_sb_complex(pbi, xd, tile, mi_row + hbs, mi_col + hbs, + mi_row_top, mi_col_top, subsize, top_bsize, + dst_buf3, dst_stride3); + } + for (i = 0; i < MAX_MB_PLANE; i++) { + if (bsize == BLOCK_8X8 && i != 0) + continue; // Skip <4x4 chroma smoothing + if (mi_row < cm->mi_rows && mi_col + hbs < cm->mi_cols) { + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_VERT, i); + if (mi_row + hbs < cm->mi_rows) { + vp10_build_masked_inter_predictor_complex( + xd, dst_buf2[i], dst_stride2[i], dst_buf3[i], dst_stride3[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_VERT, i); + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_HORZ, i); + } + } else if (mi_row + hbs < cm->mi_rows && mi_col < cm->mi_cols) { + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_HORZ, i); + } + } + break; +#if CONFIG_EXT_PARTITION_TYPES + case PARTITION_HORZ_A: + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf2, + dst_stride2, top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf2, dst_stride2, + 1); + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT, + i); + } + for (i = 0; i < MAX_MB_PLANE; i++) { + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ, + i); + } + break; + case PARTITION_VERT_A: + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf2, + dst_stride2, top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf2, + dst_stride2); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf2, + dst_stride2, 2); + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ, + i); + } + for (i = 0; i < MAX_MB_PLANE; i++) { + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf2[i], dst_stride2[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT, + i); + } + break; + case PARTITION_HORZ_B: + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 0); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col, mi_row + hbs, + mi_col, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, mi_col, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs, + mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top, + dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf2, + dst_stride2); + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf1[i]; + xd->plane[i].dst.stride = dst_stride1[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_VERT, i); + } + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_HORZ, + i); + } + break; + case PARTITION_VERT_B: + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, + top_bsize, subsize, 0, 0); + if (bsize < top_bsize) + dec_extend_all(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride); + else + dec_extend_dir(pbi, xd, tile, 0, subsize, top_bsize, mi_row, mi_col, + mi_row_top, mi_col_top, dst_buf, dst_stride, 3); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row, mi_col + hbs, mi_row, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf1, + dst_stride1, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row, mi_col + hbs, + mi_row_top, mi_col_top, dst_buf1, dst_stride1); + + dec_predict_b_extend(pbi, xd, tile, 0, mi_row + hbs, mi_col + hbs, + mi_row + hbs, mi_col + hbs, mi_row_top, mi_col_top, + dst_buf2, dst_stride2, top_bsize, bsize2, 0, 0); + dec_extend_all(pbi, xd, tile, 0, bsize2, top_bsize, mi_row + hbs, + mi_col + hbs, mi_row_top, mi_col_top, dst_buf2, + dst_stride2); + + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf1[i]; + xd->plane[i].dst.stride = dst_stride1[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf1[i], dst_stride1[i], dst_buf2[i], dst_stride2[i], + mi_row, mi_col, mi_row_top, mi_col_top, bsize, top_bsize, + PARTITION_HORZ, i); + } + for (i = 0; i < MAX_MB_PLANE; i++) { + xd->plane[i].dst.buf = dst_buf[i]; + xd->plane[i].dst.stride = dst_stride[i]; + vp10_build_masked_inter_predictor_complex( + xd, dst_buf[i], dst_stride[i], dst_buf1[i], dst_stride1[i], mi_row, + mi_col, mi_row_top, mi_col_top, bsize, top_bsize, PARTITION_VERT, + i); + } + break; +#endif // CONFIG_EXT_PARTITION_TYPES + default: assert(0); + } +} + +static void set_segment_id_supertx(const VP10_COMMON *const cm, + const int mi_row, const int mi_col, + const BLOCK_SIZE bsize) { + const struct segmentation *seg = &cm->seg; + const int miw = + VPXMIN(num_8x8_blocks_wide_lookup[bsize], cm->mi_cols - mi_col); + const int mih = + VPXMIN(num_8x8_blocks_high_lookup[bsize], cm->mi_rows - mi_row); + const int mi_offset = mi_row * cm->mi_stride + mi_col; + MODE_INFO **const mip = cm->mi_grid_visible + mi_offset; + int r, c; + int seg_id_supertx = MAX_SEGMENTS; + + if (!seg->enabled) { + seg_id_supertx = 0; + } else { + // Find the minimum segment_id + for (r = 0; r < mih; r++) + for (c = 0; c < miw; c++) + seg_id_supertx = + VPXMIN(mip[r * cm->mi_stride + c]->mbmi.segment_id, seg_id_supertx); + assert(0 <= seg_id_supertx && seg_id_supertx < MAX_SEGMENTS); + } + + // Assign the the segment_id back to segment_id_supertx + for (r = 0; r < mih; r++) + for (c = 0; c < miw; c++) + mip[r * cm->mi_stride + c]->mbmi.segment_id_supertx = seg_id_supertx; +} +#endif // CONFIG_SUPERTX + +static void decode_block(VP10Decoder *const pbi, MACROBLOCKD *const xd, +#if CONFIG_SUPERTX + int supertx_enabled, +#endif // CONFIG_SUPERTX + int mi_row, int mi_col, vp10_reader *r, +#if CONFIG_EXT_PARTITION_TYPES + PARTITION_TYPE partition, +#endif // CONFIG_EXT_PARTITION_TYPES + BLOCK_SIZE bsize, int bwl, int bhl) { + VP10_COMMON *const cm = &pbi->common; + const int less8x8 = bsize < BLOCK_8X8; + const int bw = 1 << (bwl - 1); + const int bh = 1 << (bhl - 1); + const int x_mis = VPXMIN(bw, cm->mi_cols - mi_col); + const int y_mis = VPXMIN(bh, cm->mi_rows - mi_row); + +#if CONFIG_SUPERTX + MB_MODE_INFO *mbmi; + if (supertx_enabled) { + mbmi = set_mb_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis); + } else { + mbmi = set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis, bwl, + bhl); + } +#if CONFIG_EXT_PARTITION_TYPES + xd->mi[0]->mbmi.partition = partition; +#endif + vp10_read_mode_info(pbi, xd, supertx_enabled, mi_row, mi_col, r, x_mis, + y_mis); +#else + MB_MODE_INFO *mbmi = set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, + y_mis, bwl, bhl); +#if CONFIG_EXT_PARTITION_TYPES + xd->mi[0]->mbmi.partition = partition; +#endif + vp10_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis); +#endif // CONFIG_SUPERTX + + if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) { + const BLOCK_SIZE uv_subsize = + ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y]; + if (uv_subsize == BLOCK_INVALID) + vpx_internal_error(xd->error_info, VPX_CODEC_CORRUPT_FRAME, + "Invalid block size."); + } + +#if CONFIG_SUPERTX + mbmi->segment_id_supertx = MAX_SEGMENTS; + + if (supertx_enabled) { + xd->corrupted |= vp10_reader_has_error(r); + return; + } +#endif // CONFIG_SUPERTX + + if (mbmi->skip) { + dec_reset_skip_context(xd); + } + if (!is_inter_block(mbmi)) { + int plane; + for (plane = 0; plane <= 1; ++plane) { + if (mbmi->palette_mode_info.palette_size[plane]) + vp10_decode_palette_tokens(xd, plane, r); + } + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const TX_SIZE tx_size = + plane ? dec_get_uv_tx_size(mbmi, pd->n4_wl, pd->n4_hl) + : mbmi->tx_size; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + const int stepr = num_4x4_blocks_high_txsize_lookup[tx_size]; + const int stepc = num_4x4_blocks_wide_txsize_lookup[tx_size]; + int row, col; + const int max_blocks_wide = + num_4x4_w + (xd->mb_to_right_edge >= 0 + ? 0 + : xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = + num_4x4_h + (xd->mb_to_bottom_edge >= 0 + ? 0 + : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + + for (row = 0; row < max_blocks_high; row += stepr) + for (col = 0; col < max_blocks_wide; col += stepc) + predict_and_reconstruct_intra_block(xd, r, mbmi, plane, row, col, + tx_size); + } + } else { + // Prediction + vp10_build_inter_predictors_sb(xd, mi_row, mi_col, + VPXMAX(bsize, BLOCK_8X8)); +#if CONFIG_OBMC + if (mbmi->motion_variation == OBMC_CAUSAL) { +#if CONFIG_VP9_HIGHBITDEPTH + DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * MAX_SB_SQUARE]); + DECLARE_ALIGNED(16, uint8_t, tmp_buf2[2 * MAX_MB_PLANE * MAX_SB_SQUARE]); +#else + DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_SB_SQUARE]); + DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_SB_SQUARE]); +#endif // CONFIG_VP9_HIGHBITDEPTH + uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE]; + 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 }; + 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 }; + + assert(mbmi->sb_type >= BLOCK_8X8); +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { + int len = sizeof(uint16_t); + dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1); + dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * len); + dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * 2 * len); + dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2); + dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * len); + dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * 2 * len); + } else { +#endif // CONFIG_VP9_HIGHBITDEPTH + dst_buf1[0] = tmp_buf1; + dst_buf1[1] = tmp_buf1 + MAX_SB_SQUARE; + dst_buf1[2] = tmp_buf1 + MAX_SB_SQUARE * 2; + dst_buf2[0] = tmp_buf2; + dst_buf2[1] = tmp_buf2 + MAX_SB_SQUARE; + dst_buf2[2] = tmp_buf2 + MAX_SB_SQUARE * 2; +#if CONFIG_VP9_HIGHBITDEPTH + } +#endif // CONFIG_VP9_HIGHBITDEPTH + vp10_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1, + dst_width1, dst_height1, + dst_stride1); + vp10_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2, + dst_width2, dst_height2, dst_stride2); + vp10_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, + mi_col); + vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, + dst_stride1, dst_buf2, dst_stride2); + } +#endif // CONFIG_OBMC + + // Reconstruction + if (!mbmi->skip) { + int eobtotal = 0; + int plane; + + for (plane = 0; plane < MAX_MB_PLANE; ++plane) { + const struct macroblockd_plane *const pd = &xd->plane[plane]; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + int row, col; +#if CONFIG_VAR_TX + // TODO(jingning): This can be simplified for decoder performance. + const BLOCK_SIZE plane_bsize = + get_plane_block_size(VPXMAX(bsize, BLOCK_8X8), pd); + const TX_SIZE max_tx_size = max_txsize_lookup[plane_bsize]; + int bw = num_4x4_blocks_wide_txsize_lookup[max_tx_size]; + int bh = num_4x4_blocks_high_txsize_lookup[max_tx_size]; + const int step = num_4x4_blocks_txsize_lookup[max_tx_size]; + int block = 0; +#if CONFIG_EXT_TX && CONFIG_RECT_TX + const TX_SIZE tx_size = + plane ? dec_get_uv_tx_size(mbmi, pd->n4_wl, pd->n4_hl) + : mbmi->tx_size; + + if (tx_size >= TX_SIZES) { // rect txsize is used + const int stepr = num_4x4_blocks_high_txsize_lookup[tx_size]; + const int stepc = num_4x4_blocks_wide_txsize_lookup[tx_size]; + const int max_blocks_wide = + num_4x4_w + + (xd->mb_to_right_edge >= 0 ? 0 : xd->mb_to_right_edge >> + (5 + pd->subsampling_x)); + const int max_blocks_high = + num_4x4_h + + (xd->mb_to_bottom_edge >= 0 ? 0 : xd->mb_to_bottom_edge >> + (5 + pd->subsampling_y)); + + for (row = 0; row < max_blocks_high; row += stepr) + for (col = 0; col < max_blocks_wide; col += stepc) + eobtotal += reconstruct_inter_block(xd, r, mbmi->segment_id, + plane, row, col, tx_size); + } else { +#endif + for (row = 0; row < num_4x4_h; row += bh) { + for (col = 0; col < num_4x4_w; col += bw) { + decode_reconstruct_tx(xd, r, mbmi, plane, plane_bsize, block, row, + col, max_tx_size, &eobtotal); + block += step; + } + } +#if CONFIG_EXT_TX && CONFIG_RECT_TX + } +#endif +#else + const TX_SIZE tx_size = + plane ? dec_get_uv_tx_size(mbmi, pd->n4_wl, pd->n4_hl) + : mbmi->tx_size; + const int stepr = num_4x4_blocks_high_txsize_lookup[tx_size]; + const int stepc = num_4x4_blocks_wide_txsize_lookup[tx_size]; + const int max_blocks_wide = + num_4x4_w + (xd->mb_to_right_edge >= 0 + ? 0 + : xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = + num_4x4_h + + (xd->mb_to_bottom_edge >= 0 ? 0 : xd->mb_to_bottom_edge >> + (5 + pd->subsampling_y)); + + for (row = 0; row < max_blocks_high; row += stepr) + for (col = 0; col < max_blocks_wide; col += stepc) + eobtotal += reconstruct_inter_block(xd, r, mbmi->segment_id, plane, + row, col, tx_size); +#endif + } + + if (!less8x8 && eobtotal == 0) + mbmi->has_no_coeffs = 1; // skip loopfilter + } + } + + xd->corrupted |= vp10_reader_has_error(r); +} + +static INLINE int dec_partition_plane_context(const MACROBLOCKD *xd, int mi_row, + int mi_col, int bsl) { + const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col; + const PARTITION_CONTEXT *left_ctx = + xd->left_seg_context + (mi_row & MAX_MIB_MASK); + int above = (*above_ctx >> bsl) & 1, left = (*left_ctx >> bsl) & 1; + + // assert(bsl >= 0); + + return (left * 2 + above) + bsl * PARTITION_PLOFFSET; +} + +#if !CONFIG_EXT_PARTITION_TYPES +static INLINE void dec_update_partition_context(MACROBLOCKD *xd, int mi_row, + int mi_col, BLOCK_SIZE subsize, + int bw) { + PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col; + PARTITION_CONTEXT *const left_ctx = + xd->left_seg_context + (mi_row & MAX_MIB_MASK); + + // update the partition context at the end notes. set partition bits + // of block sizes larger than the current one to be one, and partition + // bits of smaller block sizes to be zero. + memset(above_ctx, partition_context_lookup[subsize].above, bw); + memset(left_ctx, partition_context_lookup[subsize].left, bw); +} +#endif // !CONFIG_EXT_PARTITION_TYPES + +static PARTITION_TYPE read_partition(VP10_COMMON *cm, MACROBLOCKD *xd, + int mi_row, int mi_col, vp10_reader *r, + int has_rows, int has_cols, +#if CONFIG_EXT_PARTITION_TYPES + BLOCK_SIZE bsize, +#endif + int bsl) { + const int ctx = dec_partition_plane_context(xd, mi_row, mi_col, bsl); + const vpx_prob *const probs = cm->fc->partition_prob[ctx]; + FRAME_COUNTS *counts = xd->counts; + PARTITION_TYPE p; + + if (has_rows && has_cols) +#if CONFIG_EXT_PARTITION_TYPES + if (bsize <= BLOCK_8X8) + p = (PARTITION_TYPE)vp10_read_tree(r, vp10_partition_tree, probs); + else + p = (PARTITION_TYPE)vp10_read_tree(r, vp10_ext_partition_tree, probs); +#else + p = (PARTITION_TYPE)vp10_read_tree(r, vp10_partition_tree, probs); +#endif // CONFIG_EXT_PARTITION_TYPES + else if (!has_rows && has_cols) + p = vp10_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ; + else if (has_rows && !has_cols) + p = vp10_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT; + else + p = PARTITION_SPLIT; + + if (counts) ++counts->partition[ctx][p]; + + return p; +} + +#if CONFIG_SUPERTX +static int read_skip(VP10_COMMON *cm, const MACROBLOCKD *xd, int segment_id, + vp10_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) { + return 1; + } else { + const int ctx = vp10_get_skip_context(xd); + const int skip = vp10_read(r, cm->fc->skip_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->skip[ctx][skip]; + return skip; + } +} +#endif // CONFIG_SUPERTX + +// TODO(slavarnway): eliminate bsize and subsize in future commits +static void decode_partition(VP10Decoder *const pbi, MACROBLOCKD *const xd, +#if CONFIG_SUPERTX + int supertx_enabled, +#endif + int mi_row, int mi_col, vp10_reader *r, + BLOCK_SIZE bsize, int n4x4_l2) { + VP10_COMMON *const cm = &pbi->common; + const int n8x8_l2 = n4x4_l2 - 1; + const int num_8x8_wh = 1 << n8x8_l2; + const int hbs = num_8x8_wh >> 1; + PARTITION_TYPE partition; + BLOCK_SIZE subsize; +#if CONFIG_EXT_PARTITION_TYPES + BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT); +#endif + const int has_rows = (mi_row + hbs) < cm->mi_rows; + const int has_cols = (mi_col + hbs) < cm->mi_cols; +#if CONFIG_SUPERTX + const int read_token = !supertx_enabled; + int skip = 0; + TX_SIZE supertx_size = b_width_log2_lookup[bsize]; + const TileInfo *const tile = &xd->tile; + int txfm = DCT_DCT; +#endif // CONFIG_SUPERTX + + if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return; + + partition = read_partition(cm, xd, mi_row, mi_col, r, has_rows, has_cols, +#if CONFIG_EXT_PARTITION_TYPES + bsize, +#endif + n8x8_l2); + subsize = subsize_lookup[partition][bsize]; // get_subsize(bsize, partition); +#if CONFIG_SUPERTX + if (!frame_is_intra_only(cm) && partition != PARTITION_NONE && + bsize <= MAX_SUPERTX_BLOCK_SIZE && !supertx_enabled && !xd->lossless[0]) { + const int supertx_context = partition_supertx_context_lookup[partition]; + supertx_enabled = + vp10_read(r, cm->fc->supertx_prob[supertx_context][supertx_size]); + if (xd->counts) + xd->counts->supertx[supertx_context][supertx_size][supertx_enabled]++; +#if CONFIG_VAR_TX + if (supertx_enabled) xd->supertx_size = supertx_size; +#endif + } +#endif // CONFIG_SUPERTX + if (!hbs) { + // calculate bmode block dimensions (log 2) + xd->bmode_blocks_wl = 1 >> !!(partition & PARTITION_VERT); + xd->bmode_blocks_hl = 1 >> !!(partition & PARTITION_HORZ); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, 1, 1); + } else { + switch (partition) { + case PARTITION_NONE: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, n4x4_l2, n4x4_l2); + break; + case PARTITION_HORZ: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, n4x4_l2, n8x8_l2); + if (has_rows) + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row + hbs, mi_col, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, n4x4_l2, n8x8_l2); + break; + case PARTITION_VERT: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, n8x8_l2, n4x4_l2); + if (has_cols) + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col + hbs, r, +#if CONFIG_EXT_PARTITION_TYPES + partition, +#endif // CONFIG_EXT_PARTITION_TYPES + subsize, n8x8_l2, n4x4_l2); + break; + case PARTITION_SPLIT: + decode_partition(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r, subsize, n8x8_l2); + decode_partition(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col + hbs, r, subsize, n8x8_l2); + decode_partition(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row + hbs, mi_col, r, subsize, n8x8_l2); + decode_partition(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row + hbs, mi_col + hbs, r, subsize, n8x8_l2); + break; +#if CONFIG_EXT_PARTITION_TYPES + case PARTITION_HORZ_A: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col, r, partition, bsize2, n8x8_l2, n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col + hbs, r, partition, bsize2, n8x8_l2, + n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row + hbs, mi_col, r, partition, subsize, n4x4_l2, + n8x8_l2); + break; + case PARTITION_HORZ_B: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col, r, partition, subsize, n4x4_l2, n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row + hbs, mi_col, r, partition, bsize2, n8x8_l2, + n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row + hbs, mi_col + hbs, r, partition, bsize2, n8x8_l2, + n8x8_l2); + break; + case PARTITION_VERT_A: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col, r, partition, bsize2, n8x8_l2, n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row + hbs, mi_col, r, partition, bsize2, n8x8_l2, + n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col + hbs, r, partition, subsize, n8x8_l2, + n4x4_l2); + break; + case PARTITION_VERT_B: + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col, r, partition, subsize, n8x8_l2, n4x4_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row, mi_col + hbs, r, partition, bsize2, n8x8_l2, + n8x8_l2); + decode_block(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif + mi_row + hbs, mi_col + hbs, r, partition, bsize2, n8x8_l2, + n8x8_l2); + break; +#endif + default: assert(0 && "Invalid partition type"); + } + } + +#if CONFIG_SUPERTX + if (supertx_enabled && read_token) { + uint8_t *dst_buf[3]; + int dst_stride[3], i; + int offset = mi_row * cm->mi_stride + mi_col; + + set_segment_id_supertx(cm, mi_row, mi_col, bsize); + + xd->mi = cm->mi_grid_visible + offset; + xd->mi[0] = cm->mi + offset; + set_mi_row_col(xd, tile, mi_row, num_8x8_blocks_high_lookup[bsize], mi_col, + num_8x8_blocks_wide_lookup[bsize], cm->mi_rows, cm->mi_cols); + set_skip_context(xd, mi_row, mi_col); + skip = read_skip(cm, xd, xd->mi[0]->mbmi.segment_id_supertx, r); + if (skip) { + reset_skip_context(xd, bsize); + } else { +#if CONFIG_EXT_TX + if (get_ext_tx_types(supertx_size, bsize, 1) > 1) { + int eset = get_ext_tx_set(supertx_size, bsize, 1); + if (eset > 0) { + txfm = vp10_read_tree(r, vp10_ext_tx_inter_tree[eset], + cm->fc->inter_ext_tx_prob[eset][supertx_size]); + if (xd->counts) ++xd->counts->inter_ext_tx[eset][supertx_size][txfm]; + } + } +#else + if (supertx_size < TX_32X32) { + txfm = vp10_read_tree(r, vp10_ext_tx_tree, + cm->fc->inter_ext_tx_prob[supertx_size]); + if (xd->counts) ++xd->counts->inter_ext_tx[supertx_size][txfm]; + } +#endif // CONFIG_EXT_TX + } + + vp10_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col); + for (i = 0; i < MAX_MB_PLANE; i++) { + dst_buf[i] = xd->plane[i].dst.buf; + dst_stride[i] = xd->plane[i].dst.stride; + } + dec_predict_sb_complex(pbi, xd, tile, mi_row, mi_col, mi_row, mi_col, bsize, + bsize, dst_buf, dst_stride); + + if (!skip) { + int eobtotal = 0; + MB_MODE_INFO *mbmi; + set_offsets_topblock(cm, xd, tile, bsize, mi_row, mi_col); + mbmi = &xd->mi[0]->mbmi; + mbmi->tx_type = txfm; + assert(mbmi->segment_id_supertx != MAX_SEGMENTS); + for (i = 0; i < MAX_MB_PLANE; ++i) { + const struct macroblockd_plane *const pd = &xd->plane[i]; + const int num_4x4_w = pd->n4_w; + const int num_4x4_h = pd->n4_h; + int row, col; + const TX_SIZE tx_size = + i ? dec_get_uv_tx_size(mbmi, pd->n4_wl, pd->n4_hl) : mbmi->tx_size; + const int stepr = num_4x4_blocks_high_txsize_lookup[tx_size]; + const int stepc = num_4x4_blocks_wide_txsize_lookup[tx_size]; + const int max_blocks_wide = + num_4x4_w + (xd->mb_to_right_edge >= 0 + ? 0 + : xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + const int max_blocks_high = + num_4x4_h + + (xd->mb_to_bottom_edge >= 0 ? 0 : xd->mb_to_bottom_edge >> + (5 + pd->subsampling_y)); + + for (row = 0; row < max_blocks_high; row += stepr) + for (col = 0; col < max_blocks_wide; col += stepc) + eobtotal += reconstruct_inter_block(xd, r, mbmi->segment_id_supertx, + i, row, col, tx_size); + } + if (!(subsize < BLOCK_8X8) && eobtotal == 0) skip = 1; + } + set_param_topblock(cm, xd, bsize, mi_row, mi_col, txfm, skip); + } +#endif // CONFIG_SUPERTX + +#if CONFIG_EXT_PARTITION_TYPES + if (bsize >= BLOCK_8X8) { + switch (partition) { + case PARTITION_SPLIT: + if (bsize > BLOCK_8X8) break; + case PARTITION_NONE: + case PARTITION_HORZ: + case PARTITION_VERT: + update_partition_context(xd, mi_row, mi_col, subsize, bsize); + break; + case PARTITION_HORZ_A: + update_partition_context(xd, mi_row, mi_col, bsize2, subsize); + update_partition_context(xd, mi_row + hbs, mi_col, subsize, subsize); + break; + case PARTITION_HORZ_B: + update_partition_context(xd, mi_row, mi_col, subsize, subsize); + update_partition_context(xd, mi_row + hbs, mi_col, bsize2, subsize); + break; + case PARTITION_VERT_A: + update_partition_context(xd, mi_row, mi_col, bsize2, subsize); + update_partition_context(xd, mi_row, mi_col + hbs, subsize, subsize); + break; + case PARTITION_VERT_B: + update_partition_context(xd, mi_row, mi_col, subsize, subsize); + update_partition_context(xd, mi_row, mi_col + hbs, bsize2, subsize); + break; + default: assert(0 && "Invalid partition type"); + } + } +#else + // update partition context + if (bsize >= BLOCK_8X8 && + (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT)) + dec_update_partition_context(xd, mi_row, mi_col, subsize, num_8x8_wh); +#if DERING_REFINEMENT + if (bsize == BLOCK_64X64) { + if (cm->dering_level != 0 && !sb_all_skip(cm, mi_row, mi_col)) { + cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi.dering_gain = + vpx_read_literal(r, DERING_REFINEMENT_BITS); + } else { + cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi.dering_gain = + 0; + } + } +#endif // DERGING_REFINEMENT +#endif // CONFIG_EXT_PARTITION_TYPES +} + +#if !CONFIG_ANS +static void setup_bool_decoder(const uint8_t *data, const uint8_t *data_end, + const size_t read_size, + struct vpx_internal_error_info *error_info, + vp10_reader *r, vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + // 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)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + if (vpx_reader_init(r, data, read_size, decrypt_cb, decrypt_state)) + vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder %d", 1); +} +#else +static void setup_token_decoder(const uint8_t *data, const uint8_t *data_end, + const size_t read_size, + struct vpx_internal_error_info *error_info, + struct AnsDecoder *const ans, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + (void)decrypt_cb; + (void)decrypt_state; + // 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)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + if (read_size > INT_MAX || ans_read_init(ans, data, (int)read_size)) + vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR, + "Failed to allocate token decoder %d", 1); +} +#endif + +static void read_coef_probs_common(vp10_coeff_probs_model *coef_probs, + vp10_reader *r) { + int i, j, k, l, m; + + if (vp10_read_bit(r)) + for (i = 0; i < PLANE_TYPES; ++i) + for (j = 0; j < REF_TYPES; ++j) + for (k = 0; k < COEF_BANDS; ++k) + for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) + for (m = 0; m < UNCONSTRAINED_NODES; ++m) + vp10_diff_update_prob(r, &coef_probs[i][j][k][l][m]); +} + +static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, + vp10_reader *r) { + const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode]; + TX_SIZE tx_size; + for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size) + read_coef_probs_common(fc->coef_probs[tx_size], r); +#if CONFIG_ANS + vp10_coef_pareto_cdfs(fc); +#endif // CONFIG_ANS +} + +static void setup_segmentation(VP10_COMMON *const cm, + struct vpx_read_bit_buffer *rb) { + struct segmentation *const seg = &cm->seg; + int i, j; + + seg->update_map = 0; + seg->update_data = 0; + + seg->enabled = vpx_rb_read_bit(rb); + if (!seg->enabled) return; + + // Segmentation map update + if (frame_is_intra_only(cm) || cm->error_resilient_mode) { + seg->update_map = 1; + } else { + seg->update_map = vpx_rb_read_bit(rb); + } + if (seg->update_map) { + if (frame_is_intra_only(cm) || cm->error_resilient_mode) { + seg->temporal_update = 0; + } else { + seg->temporal_update = vpx_rb_read_bit(rb); + } + } + + // Segmentation data update + seg->update_data = vpx_rb_read_bit(rb); + if (seg->update_data) { + seg->abs_delta = vpx_rb_read_bit(rb); + + vp10_clearall_segfeatures(seg); + + for (i = 0; i < MAX_SEGMENTS; i++) { + for (j = 0; j < SEG_LVL_MAX; j++) { + int data = 0; + const int feature_enabled = vpx_rb_read_bit(rb); + if (feature_enabled) { + vp10_enable_segfeature(seg, i, j); + data = decode_unsigned_max(rb, vp10_seg_feature_data_max(j)); + if (vp10_is_segfeature_signed(j)) + data = vpx_rb_read_bit(rb) ? -data : data; + } + vp10_set_segdata(seg, i, j, data); + } + } + } +} + +#if CONFIG_LOOP_RESTORATION +static void setup_restoration(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + int i; + RestorationInfo *rsi = &cm->rst_info; + int ntiles; + if (vpx_rb_read_bit(rb)) { + if (vpx_rb_read_bit(rb)) { + rsi->restoration_type = RESTORE_BILATERAL; + ntiles = vp10_get_restoration_ntiles(BILATERAL_TILESIZE, cm->width, + cm->height); + rsi->bilateral_level = (int *)vpx_realloc( + rsi->bilateral_level, sizeof(*rsi->bilateral_level) * ntiles); + assert(rsi->bilateral_level != NULL); + for (i = 0; i < ntiles; ++i) { + if (vpx_rb_read_bit(rb)) { + rsi->bilateral_level[i] = + vpx_rb_read_literal(rb, vp10_bilateral_level_bits(cm)); + } else { + rsi->bilateral_level[i] = -1; + } + } + } else { + rsi->restoration_type = RESTORE_WIENER; + ntiles = + vp10_get_restoration_ntiles(WIENER_TILESIZE, cm->width, cm->height); + rsi->wiener_level = (int *)vpx_realloc( + rsi->wiener_level, sizeof(*rsi->wiener_level) * ntiles); + assert(rsi->wiener_level != NULL); + rsi->vfilter = (int(*)[RESTORATION_HALFWIN])vpx_realloc( + rsi->vfilter, sizeof(*rsi->vfilter) * ntiles); + assert(rsi->vfilter != NULL); + rsi->hfilter = (int(*)[RESTORATION_HALFWIN])vpx_realloc( + rsi->hfilter, sizeof(*rsi->hfilter) * ntiles); + assert(rsi->hfilter != NULL); + for (i = 0; i < ntiles; ++i) { + rsi->wiener_level[i] = vpx_rb_read_bit(rb); + if (rsi->wiener_level[i]) { + rsi->vfilter[i][0] = vpx_rb_read_literal(rb, WIENER_FILT_TAP0_BITS) + + WIENER_FILT_TAP0_MINV; + rsi->vfilter[i][1] = vpx_rb_read_literal(rb, WIENER_FILT_TAP1_BITS) + + WIENER_FILT_TAP1_MINV; + rsi->vfilter[i][2] = vpx_rb_read_literal(rb, WIENER_FILT_TAP2_BITS) + + WIENER_FILT_TAP2_MINV; + rsi->hfilter[i][0] = vpx_rb_read_literal(rb, WIENER_FILT_TAP0_BITS) + + WIENER_FILT_TAP0_MINV; + rsi->hfilter[i][1] = vpx_rb_read_literal(rb, WIENER_FILT_TAP1_BITS) + + WIENER_FILT_TAP1_MINV; + rsi->hfilter[i][2] = vpx_rb_read_literal(rb, WIENER_FILT_TAP2_BITS) + + WIENER_FILT_TAP2_MINV; + } else { + rsi->vfilter[i][0] = rsi->vfilter[i][1] = rsi->vfilter[i][2] = 0; + rsi->hfilter[i][0] = rsi->hfilter[i][1] = rsi->hfilter[i][2] = 0; + } + } + } + } else { + rsi->restoration_type = RESTORE_NONE; + } +} +#endif // CONFIG_LOOP_RESTORATION + +static void setup_loopfilter(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + struct loopfilter *lf = &cm->lf; + lf->filter_level = vpx_rb_read_literal(rb, 6); + lf->sharpness_level = vpx_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 = vpx_rb_read_bit(rb); + if (lf->mode_ref_delta_enabled) { + lf->mode_ref_delta_update = vpx_rb_read_bit(rb); + if (lf->mode_ref_delta_update) { + int i; + + for (i = 0; i < TOTAL_REFS_PER_FRAME; i++) + if (vpx_rb_read_bit(rb)) + lf->ref_deltas[i] = vpx_rb_read_inv_signed_literal(rb, 6); + + for (i = 0; i < MAX_MODE_LF_DELTAS; i++) + if (vpx_rb_read_bit(rb)) + lf->mode_deltas[i] = vpx_rb_read_inv_signed_literal(rb, 6); + } + } +} + +#if CONFIG_CLPF +static void setup_clpf(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + cm->clpf = vpx_rb_read_literal(rb, 1); +} +#endif + +#if CONFIG_DERING +static void setup_dering(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + cm->dering_level = vpx_rb_read_literal(rb, DERING_LEVEL_BITS); +} +#endif // CONFIG_DERING + +static INLINE int read_delta_q(struct vpx_read_bit_buffer *rb) { + return vpx_rb_read_bit(rb) ? vpx_rb_read_inv_signed_literal(rb, 6) : 0; +} + +static void setup_quantization(VP10_COMMON *const cm, + struct vpx_read_bit_buffer *rb) { + cm->base_qindex = vpx_rb_read_literal(rb, QINDEX_BITS); + cm->y_dc_delta_q = read_delta_q(rb); + cm->uv_dc_delta_q = read_delta_q(rb); + cm->uv_ac_delta_q = read_delta_q(rb); + cm->dequant_bit_depth = cm->bit_depth; +#if CONFIG_AOM_QM + cm->using_qmatrix = vpx_rb_read_bit(rb); + if (cm->using_qmatrix) { + cm->min_qmlevel = vpx_rb_read_literal(rb, QM_LEVEL_BITS); + cm->max_qmlevel = vpx_rb_read_literal(rb, QM_LEVEL_BITS); + } else { + cm->min_qmlevel = 0; + cm->max_qmlevel = 0; + } +#endif +} + +static void setup_segmentation_dequant(VP10_COMMON *const cm) { + // Build y/uv dequant values based on segmentation. + int i = 0; +#if CONFIG_AOM_QM + int lossless; + int j = 0; + int qmlevel; + int using_qm = cm->using_qmatrix; + int minqm = cm->min_qmlevel; + int maxqm = cm->max_qmlevel; +#endif +#if CONFIG_NEW_QUANT + int b; + int dq; +#endif // CONFIG_NEW_QUANT + if (cm->seg.enabled) { + for (i = 0; i < MAX_SEGMENTS; ++i) { + const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex); + cm->y_dequant[i][0] = + vp10_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth); + cm->y_dequant[i][1] = vp10_ac_quant(qindex, 0, cm->bit_depth); + cm->uv_dequant[i][0] = + vp10_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth); + cm->uv_dequant[i][1] = + vp10_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth); +#if CONFIG_AOM_QM + lossless = qindex == 0 && cm->y_dc_delta_q == 0 && + cm->uv_dc_delta_q == 0 && cm->uv_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 + qmlevel = (lossless || using_qm == 0) + ? NUM_QM_LEVELS - 1 + : aom_get_qmlevel(cm->base_qindex, minqm, maxqm); + for (j = 0; j < TX_SIZES; ++j) { + cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 0, j, 1); + cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 0, j, 0); + cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 1, j, 1); + cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 1, j, 0); + } +#endif // CONFIG_AOM_QM +#if CONFIG_NEW_QUANT + for (dq = 0; dq < QUANT_PROFILES; dq++) { + for (b = 0; b < COEF_BANDS; ++b) { + vp10_get_dequant_val_nuq(cm->y_dequant[i][b != 0], qindex, b, + cm->y_dequant_nuq[i][dq][b], NULL, dq); + vp10_get_dequant_val_nuq(cm->uv_dequant[i][b != 0], qindex, b, + cm->uv_dequant_nuq[i][dq][b], NULL, dq); + } + } +#endif // CONFIG_NEW_QUANT + } + } else { + const int qindex = cm->base_qindex; + // When segmentation is disabled, only the first value is used. The + // remaining are don't cares. + cm->y_dequant[0][0] = + vp10_dc_quant(qindex, cm->y_dc_delta_q, cm->bit_depth); + cm->y_dequant[0][1] = vp10_ac_quant(qindex, 0, cm->bit_depth); + cm->uv_dequant[0][0] = + vp10_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth); + cm->uv_dequant[0][1] = + vp10_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth); +#if CONFIG_AOM_QM + lossless = qindex == 0 && cm->y_dc_delta_q == 0 && cm->uv_dc_delta_q == 0 && + cm->uv_ac_delta_q == 0; + // No quant weighting when lossless or signalled not using QM + qmlevel = (lossless || using_qm == 0) + ? NUM_QM_LEVELS - 1 + : aom_get_qmlevel(cm->base_qindex, minqm, maxqm); + for (j = 0; j < TX_SIZES; ++j) { + cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 0, j, 1); + cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 0, j, 0); + cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmlevel, 1, j, 1); + cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmlevel, 1, j, 0); + } +#endif +#if CONFIG_NEW_QUANT + for (dq = 0; dq < QUANT_PROFILES; dq++) { + for (b = 0; b < COEF_BANDS; ++b) { + vp10_get_dequant_val_nuq(cm->y_dequant[0][b != 0], qindex, b, + cm->y_dequant_nuq[0][dq][b], NULL, dq); + vp10_get_dequant_val_nuq(cm->uv_dequant[0][b != 0], qindex, b, + cm->uv_dequant_nuq[0][dq][b], NULL, dq); + } + } +#endif // CONFIG_NEW_QUANT + } +} + +static INTERP_FILTER read_interp_filter(struct vpx_read_bit_buffer *rb) { + return vpx_rb_read_bit(rb) ? SWITCHABLE + : vpx_rb_read_literal(rb, 2 + CONFIG_EXT_INTERP); +} + +static void setup_render_size(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + cm->render_width = cm->width; + cm->render_height = cm->height; + if (vpx_rb_read_bit(rb)) + vp10_read_frame_size(rb, &cm->render_width, &cm->render_height); +} + +static void resize_mv_buffer(VP10_COMMON *cm) { + vpx_free(cm->cur_frame->mvs); + cm->cur_frame->mi_rows = cm->mi_rows; + cm->cur_frame->mi_cols = cm->mi_cols; + CHECK_MEM_ERROR(cm, cm->cur_frame->mvs, + (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols, + sizeof(*cm->cur_frame->mvs))); +} + +static void resize_context_buffers(VP10_COMMON *cm, int width, int height) { +#if CONFIG_SIZE_LIMIT + if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) + vpx_internal_error(&cm->error, VPX_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 vp10_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 (vp10_alloc_context_buffers(cm, width, height)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate context buffers"); + } else { + vp10_set_mb_mi(cm, width, height); + } + vp10_init_context_buffers(cm); + cm->width = width; + cm->height = height; + } + if (cm->cur_frame->mvs == NULL || cm->mi_rows > cm->cur_frame->mi_rows || + cm->mi_cols > cm->cur_frame->mi_cols) { + resize_mv_buffer(cm); + } +} + +static void setup_frame_size(VP10_COMMON *cm, struct vpx_read_bit_buffer *rb) { + int width, height; + BufferPool *const pool = cm->buffer_pool; + vp10_read_frame_size(rb, &width, &height); + resize_context_buffers(cm, width, height); + setup_render_size(cm, rb); + + lock_buffer_pool(pool); + if (vpx_realloc_frame_buffer( + get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x, + cm->subsampling_y, +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth, +#endif + VPX_DEC_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); + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x; + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y; + pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth; + pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space; + pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->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 INLINE int valid_ref_frame_img_fmt(vpx_bit_depth_t ref_bit_depth, + int ref_xss, int ref_yss, + vpx_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(VP10_COMMON *cm, + struct vpx_read_bit_buffer *rb) { + int width, height; + int found = 0, i; + int has_valid_ref_frame = 0; + BufferPool *const pool = cm->buffer_pool; + for (i = 0; i < INTER_REFS_PER_FRAME; ++i) { + if (vpx_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; + found = 1; + break; + } + } + + if (!found) { + vp10_read_frame_size(rb, &width, &height); + setup_render_size(cm, rb); + } + + if (width <= 0 || height <= 0) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid frame size"); + + // Check to make sure at least one of frames that this frame references + // has valid dimensions. + for (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) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Referenced frame has invalid size"); + for (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, cm->bit_depth, + cm->subsampling_x, cm->subsampling_y)) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Referenced frame has incompatible color format"); + } + + resize_context_buffers(cm, width, height); + + lock_buffer_pool(pool); + if (vpx_realloc_frame_buffer( + get_frame_new_buffer(cm), cm->width, cm->height, cm->subsampling_x, + cm->subsampling_y, +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth, +#endif + VPX_DEC_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); + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate frame buffer"); + } + unlock_buffer_pool(pool); + + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_x = cm->subsampling_x; + pool->frame_bufs[cm->new_fb_idx].buf.subsampling_y = cm->subsampling_y; + pool->frame_bufs[cm->new_fb_idx].buf.bit_depth = (unsigned int)cm->bit_depth; + pool->frame_bufs[cm->new_fb_idx].buf.color_space = cm->color_space; + pool->frame_bufs[cm->new_fb_idx].buf.color_range = cm->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 read_tile_info(VP10Decoder *const pbi, + struct vpx_read_bit_buffer *const rb) { + VP10_COMMON *const cm = &pbi->common; +#if CONFIG_EXT_TILE +// Read the tile width/height +#if CONFIG_EXT_PARTITION + if (cm->sb_size == BLOCK_128X128) { + cm->tile_width = vpx_rb_read_literal(rb, 5) + 1; + cm->tile_height = vpx_rb_read_literal(rb, 5) + 1; + } else +#endif // CONFIG_EXT_PARTITION + { + cm->tile_width = vpx_rb_read_literal(rb, 6) + 1; + cm->tile_height = vpx_rb_read_literal(rb, 6) + 1; + } + + cm->tile_width <<= cm->mib_size_log2; + cm->tile_height <<= cm->mib_size_log2; + + cm->tile_width = VPXMIN(cm->tile_width, cm->mi_cols); + cm->tile_height = VPXMIN(cm->tile_height, cm->mi_rows); + + // Get the number of tiles + cm->tile_cols = 1; + while (cm->tile_cols * cm->tile_width < cm->mi_cols) ++cm->tile_cols; + + cm->tile_rows = 1; + while (cm->tile_rows * cm->tile_height < cm->mi_rows) ++cm->tile_rows; + + if (cm->tile_cols * cm->tile_rows > 1) { + // Read the number of bytes used to store tile size + pbi->tile_col_size_bytes = vpx_rb_read_literal(rb, 2) + 1; + pbi->tile_size_bytes = vpx_rb_read_literal(rb, 2) + 1; + } +#else + int min_log2_tile_cols, max_log2_tile_cols, max_ones; + vp10_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols); + + // columns + max_ones = max_log2_tile_cols - min_log2_tile_cols; + cm->log2_tile_cols = min_log2_tile_cols; + while (max_ones-- && vpx_rb_read_bit(rb)) cm->log2_tile_cols++; + + if (cm->log2_tile_cols > 6) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid number of tile columns"); + + // rows + cm->log2_tile_rows = vpx_rb_read_bit(rb); + if (cm->log2_tile_rows) cm->log2_tile_rows += vpx_rb_read_bit(rb); + + cm->tile_cols = 1 << cm->log2_tile_cols; + cm->tile_rows = 1 << cm->log2_tile_rows; + + cm->tile_width = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2); + cm->tile_width >>= cm->log2_tile_cols; + cm->tile_height = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2); + cm->tile_height >>= cm->log2_tile_rows; + + // round to integer multiples of superblock size + cm->tile_width = ALIGN_POWER_OF_TWO(cm->tile_width, MAX_MIB_SIZE_LOG2); + cm->tile_height = ALIGN_POWER_OF_TWO(cm->tile_height, MAX_MIB_SIZE_LOG2); + + // tile size magnitude + if (cm->tile_rows > 1 || cm->tile_cols > 1) { + pbi->tile_size_bytes = vpx_rb_read_literal(rb, 2) + 1; + } +#endif // CONFIG_EXT_TILE +} + +static int mem_get_varsize(const uint8_t *src, const 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("Invalid size" && 0); return -1; + } +} + +#if CONFIG_EXT_TILE +// 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, + struct vpx_internal_error_info *error_info, + const uint8_t **data, vpx_decrypt_cb decrypt_cb, + void *decrypt_state, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS], + int tile_size_bytes, int col, int row) { + size_t size; + + size_t copy_size = 0; + const uint8_t *copy_data = NULL; + + if (!read_is_valid(*data, tile_size_bytes, data_end)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + if (decrypt_cb) { + uint8_t be_data[4]; + decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes); + + // Only read number of bytes in cm->tile_size_bytes. + size = mem_get_varsize(be_data, tile_size_bytes); + } else { + size = mem_get_varsize(*data, tile_size_bytes); + } + + // The top bit indicates copy mode + if ((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; + } + + *data += tile_size_bytes; + + if (size > (size_t)(data_end - *data)) + vpx_internal_error(error_info, VPX_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; + + tile_buffers[row][col].raw_data_end = *data; +} + +static void get_tile_buffers( + VP10Decoder *pbi, const uint8_t *data, const uint8_t *data_end, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { + VP10_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; + + if (!have_tiles) { + const uint32_t tile_size = data_end - data; + tile_buffers[0][0].data = data; + tile_buffers[0][0].size = tile_size; + tile_buffers[0][0].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]; + const uint8_t *const data_start = data; + + const int dec_tile_row = VPXMIN(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 = VPXMIN(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; + + size_t tile_col_size; + int r, c; + + // Read tile column sizes for all columns (we need the last tile buffer) + for (c = 0; c < tile_cols; ++c) { + const int is_last = c == tile_cols - 1; + 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 (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 (r = 0; r < (is_last ? tile_rows : tile_rows_end); ++r) { + tile_buffers[r][c].col = c; + + get_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data, + pbi->decrypt_cb, pbi->decrypt_state, tile_buffers, + tile_size_bytes, c, r); + } + } + + // If we have not read the last column, then read it to get the last tile. + if (tile_cols_end != tile_cols) { + c = tile_cols - 1; + + data = tile_col_data_end[c - 1]; + + for (r = 0; r < tile_rows; ++r) { + tile_buffers[r][c].col = c; + + get_tile_buffer(tile_col_data_end[c], &pbi->common.error, &data, + pbi->decrypt_cb, pbi->decrypt_state, tile_buffers, + tile_size_bytes, c, r); + } + } + } +} +#else +// 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 vpx_internal_error_info *error_info, + const uint8_t **data, vpx_decrypt_cb decrypt_cb, + void *decrypt_state, TileBufferDec *const buf) { + size_t size; + + if (!is_last) { + if (!read_is_valid(*data, 4, data_end)) + vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt tile length"); + + if (decrypt_cb) { + uint8_t be_data[4]; + decrypt_cb(decrypt_state, *data, be_data, tile_size_bytes); + size = mem_get_varsize(be_data, tile_size_bytes); + } else { + size = mem_get_varsize(*data, tile_size_bytes); + } + *data += tile_size_bytes; + + if (size > (size_t)(data_end - *data)) + vpx_internal_error(error_info, VPX_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( + VP10Decoder *pbi, const uint8_t *data, const uint8_t *data_end, + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS]) { + VP10_COMMON *const cm = &pbi->common; + int r, c; + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + + for (r = 0; r < tile_rows; ++r) { + for (c = 0; c < tile_cols; ++c) { + const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1); + TileBufferDec *const buf = &tile_buffers[r][c]; + buf->col = c; + get_tile_buffer(data_end, pbi->tile_size_bytes, is_last, &cm->error, + &data, pbi->decrypt_cb, pbi->decrypt_state, buf); + } + } +} +#endif // CONFIG_EXT_TILE + +static const uint8_t *decode_tiles(VP10Decoder *pbi, const uint8_t *data, + const uint8_t *data_end) { + VP10_COMMON *const cm = &pbi->common; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + 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; +#if CONFIG_EXT_TILE + const int dec_tile_row = VPXMIN(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 ? dec_tile_row + 1 : tile_rows; + const int dec_tile_col = VPXMIN(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 inv_col_order = pbi->inv_tile_order && !single_col; + const int inv_row_order = pbi->inv_tile_order && !single_row; +#else + const int tile_rows_start = 0; + const int tile_rows_end = tile_rows; + const int tile_cols_start = 0; + const int tile_cols_end = tile_cols; + const int inv_col_order = pbi->inv_tile_order; + const int inv_row_order = pbi->inv_tile_order; +#endif // CONFIG_EXT_TILE + int tile_row, tile_col; + +#if CONFIG_ENTROPY + cm->do_subframe_update = n_tiles == 1; +#endif // CONFIG_ENTROPY + + if (cm->lf.filter_level && !cm->skip_loop_filter && + pbi->lf_worker.data1 == NULL) { + CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, + vpx_memalign(32, sizeof(LFWorkerData))); + pbi->lf_worker.hook = (VPxWorkerHook)vp10_loop_filter_worker; + if (pbi->max_threads > 1 && !winterface->reset(&pbi->lf_worker)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Loop filter thread creation failed"); + } + } + + if (cm->lf.filter_level && !cm->skip_loop_filter) { + LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1; + // Be sure to sync as we might be resuming after a failed frame decode. + winterface->sync(&pbi->lf_worker); + vp10_loop_filter_data_reset(lf_data, get_frame_new_buffer(cm), cm, + pbi->mb.plane); + } + + assert(tile_rows <= MAX_TILE_ROWS); + assert(tile_cols <= MAX_TILE_COLS); + + get_tile_buffers(pbi, data, data_end, tile_buffers); + + if (pbi->tile_data == NULL || n_tiles != pbi->allocated_tiles) { + vpx_free(pbi->tile_data); + CHECK_MEM_ERROR(cm, pbi->tile_data, + vpx_memalign(32, n_tiles * (sizeof(*pbi->tile_data)))); + pbi->allocated_tiles = n_tiles; + } + + // Load all tile information into tile_data. + for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { + for (tile_col = tile_cols_start; tile_col < tile_cols_end; ++tile_col) { + const TileBufferDec *const buf = &tile_buffers[tile_row][tile_col]; + TileData *const td = pbi->tile_data + tile_cols * tile_row + tile_col; + + td->cm = cm; + td->xd = pbi->mb; + td->xd.corrupted = 0; + td->xd.counts = + cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD + ? &cm->counts + : NULL; + vp10_zero(td->dqcoeff); + vp10_tile_init(&td->xd.tile, td->cm, tile_row, tile_col); +#if !CONFIG_ANS + setup_bool_decoder(buf->data, data_end, buf->size, &cm->error, + &td->bit_reader, pbi->decrypt_cb, pbi->decrypt_state); +#else + setup_token_decoder(buf->data, data_end, buf->size, &cm->error, + &td->bit_reader, pbi->decrypt_cb, pbi->decrypt_state); +#endif + vp10_init_macroblockd(cm, &td->xd, td->dqcoeff); + td->xd.plane[0].color_index_map = td->color_index_map[0]; + td->xd.plane[1].color_index_map = td->color_index_map[1]; + } + } + + 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; + int mi_row = 0; + TileInfo tile_info; + + vp10_tile_set_row(&tile_info, cm, 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; + TileData *const td = pbi->tile_data + tile_cols * row + col; + + vp10_tile_set_col(&tile_info, cm, col); + + vp10_zero_above_context(cm, tile_info.mi_col_start, tile_info.mi_col_end); + + for (mi_row = tile_info.mi_row_start; mi_row < tile_info.mi_row_end; + mi_row += cm->mib_size) { + int mi_col; + + vp10_zero_left_context(&td->xd); + + for (mi_col = tile_info.mi_col_start; mi_col < tile_info.mi_col_end; + mi_col += cm->mib_size) { + decode_partition(pbi, &td->xd, +#if CONFIG_SUPERTX + 0, +#endif // CONFIG_SUPERTX + mi_row, mi_col, &td->bit_reader, cm->sb_size, + b_width_log2_lookup[cm->sb_size]); + } + pbi->mb.corrupted |= td->xd.corrupted; + if (pbi->mb.corrupted) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Failed to decode tile data"); +#if CONFIG_ENTROPY + if (cm->do_subframe_update && + cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { + if ((mi_row + MI_SIZE) % + (MI_SIZE * + VPXMAX(cm->mi_rows / MI_SIZE / COEF_PROBS_BUFS, 1)) == + 0 && + mi_row + MI_SIZE < cm->mi_rows && + cm->coef_probs_update_idx < COEF_PROBS_BUFS - 1) { + vp10_partial_adapt_probs(cm, mi_row, mi_col); + ++cm->coef_probs_update_idx; + } + } +#endif // CONFIG_ENTROPY + } + } + + assert(mi_row > 0); + +#if !CONFIG_VAR_TX + // Loopfilter one tile row. + if (cm->lf.filter_level && !cm->skip_loop_filter) { + LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1; + const int lf_start = VPXMAX(0, tile_info.mi_row_start - cm->mib_size); + const int lf_end = tile_info.mi_row_end - cm->mib_size; + + // Delay the loopfilter if the first tile row is only + // a single superblock high. + if (lf_end <= 0) continue; + + // Decoding has completed. Finish up the loop filter in this thread. + if (tile_info.mi_row_end >= cm->mi_rows) continue; + + winterface->sync(&pbi->lf_worker); + lf_data->start = lf_start; + lf_data->stop = lf_end; + if (pbi->max_threads > 1) { + winterface->launch(&pbi->lf_worker); + } else { + winterface->execute(&pbi->lf_worker); + } + } + + // After loopfiltering, the last 7 row pixels in each superblock row may + // still be changed by the longest loopfilter of the next superblock row. + if (cm->frame_parallel_decode) + vp10_frameworker_broadcast(pbi->cur_buf, mi_row << cm->mib_size_log2); +#endif // !CONFIG_VAR_TX + } + +#if CONFIG_VAR_TX + // Loopfilter the whole frame. + vp10_loop_filter_frame(get_frame_new_buffer(cm), cm, &pbi->mb, + cm->lf.filter_level, 0, 0); +#else + // Loopfilter remaining rows in the frame. + if (cm->lf.filter_level && !cm->skip_loop_filter) { + LFWorkerData *const lf_data = (LFWorkerData *)pbi->lf_worker.data1; + winterface->sync(&pbi->lf_worker); + lf_data->start = lf_data->stop; + lf_data->stop = cm->mi_rows; + winterface->execute(&pbi->lf_worker); + } +#endif // CONFIG_VAR_TX +#if CONFIG_CLPF + if (cm->clpf && !cm->skip_loop_filter) + vp10_clpf_frame(&pbi->cur_buf->buf, cm, &pbi->mb); +#endif +#if CONFIG_DERING + if (cm->dering_level && !cm->skip_loop_filter) { + vp10_dering_frame(&pbi->cur_buf->buf, cm, &pbi->mb, cm->dering_level); + } +#endif // CONFIG_DERING + + if (cm->frame_parallel_decode) + vp10_frameworker_broadcast(pbi->cur_buf, INT_MAX); + +#if CONFIG_EXT_TILE + if (n_tiles == 1) { +#if CONFIG_ANS + return data_end; +#else + // Find the end of the single tile buffer + return vpx_reader_find_end(&pbi->tile_data->bit_reader); +#endif // CONFIG_ANS + } else { + // Return the end of the last tile buffer + return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end; + } +#else +#if CONFIG_ANS + return data_end; +#else + { + // Get last tile data. + TileData *const td = pbi->tile_data + tile_cols * tile_rows - 1; + return vpx_reader_find_end(&td->bit_reader); + } +#endif // CONFIG_ANS +#endif // CONFIG_EXT_TILE +} + +static int tile_worker_hook(TileWorkerData *const tile_data, + const TileInfo *const tile) { + VP10Decoder *const pbi = tile_data->pbi; + const VP10_COMMON *const cm = &pbi->common; + int mi_row, mi_col; + + if (setjmp(tile_data->error_info.jmp)) { + tile_data->error_info.setjmp = 0; + tile_data->xd.corrupted = 1; + return 0; + } + + tile_data->error_info.setjmp = 1; + tile_data->xd.error_info = &tile_data->error_info; + + vp10_zero_above_context(&pbi->common, tile->mi_col_start, tile->mi_col_end); + + for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end; + mi_row += cm->mib_size) { + vp10_zero_left_context(&tile_data->xd); + + for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end; + mi_col += cm->mib_size) { + decode_partition(pbi, &tile_data->xd, +#if CONFIG_SUPERTX + 0, +#endif + mi_row, mi_col, &tile_data->bit_reader, cm->sb_size, + b_width_log2_lookup[cm->sb_size]); + } + } + return !tile_data->xd.corrupted; +} + +// sorts in descending order +static int compare_tile_buffers(const void *a, const void *b) { + const TileBufferDec *const buf1 = (const TileBufferDec *)a; + const TileBufferDec *const buf2 = (const TileBufferDec *)b; + return (int)(buf2->size - buf1->size); +} + +static const uint8_t *decode_tiles_mt(VP10Decoder *pbi, const uint8_t *data, + const uint8_t *data_end) { + VP10_COMMON *const cm = &pbi->common; + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + const int tile_cols = cm->tile_cols; + const int tile_rows = cm->tile_rows; + const int num_workers = VPXMIN(pbi->max_threads & ~1, tile_cols); + TileBufferDec (*const tile_buffers)[MAX_TILE_COLS] = pbi->tile_buffers; +#if CONFIG_EXT_TILE + const int dec_tile_row = VPXMIN(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 ? dec_tile_row + 1 : tile_rows; + const int dec_tile_col = VPXMIN(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; +#else + const int tile_rows_start = 0; + const int tile_rows_end = tile_rows; + const int tile_cols_start = 0; + const int tile_cols_end = tile_cols; +#endif // CONFIG_EXT_TILE + int tile_row, tile_col; + int i; + +#if !(CONFIG_ANS || CONFIG_EXT_TILE) + int final_worker = -1; +#endif // !(CONFIG_ANS || CONFIG_EXT_TILE) + + assert(tile_rows <= MAX_TILE_ROWS); + assert(tile_cols <= MAX_TILE_COLS); + + assert(tile_cols * tile_rows > 1); + +#if CONFIG_ANS + // TODO(any): This might just work now. Needs to be tested. + abort(); // FIXME: Tile parsing broken +#endif // CONFIG_ANS + + // TODO(jzern): See if we can remove the restriction of passing in max + // threads to the decoder. + if (pbi->num_tile_workers == 0) { + const int num_threads = pbi->max_threads & ~1; + CHECK_MEM_ERROR(cm, pbi->tile_workers, + vpx_malloc(num_threads * sizeof(*pbi->tile_workers))); + // Ensure tile data offsets will be properly aligned. This may fail on + // platforms without DECLARE_ALIGNED(). + assert((sizeof(*pbi->tile_worker_data) % 16) == 0); + CHECK_MEM_ERROR( + cm, pbi->tile_worker_data, + vpx_memalign(32, num_threads * sizeof(*pbi->tile_worker_data))); + CHECK_MEM_ERROR(cm, pbi->tile_worker_info, + vpx_malloc(num_threads * sizeof(*pbi->tile_worker_info))); + for (i = 0; i < num_threads; ++i) { + VPxWorker *const worker = &pbi->tile_workers[i]; + ++pbi->num_tile_workers; + + winterface->init(worker); + if (i < num_threads - 1 && !winterface->reset(worker)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Tile decoder thread creation failed"); + } + } + } + + // Reset tile decoding hook + for (i = 0; i < num_workers; ++i) { + VPxWorker *const worker = &pbi->tile_workers[i]; + winterface->sync(worker); + worker->hook = (VPxWorkerHook)tile_worker_hook; + worker->data1 = &pbi->tile_worker_data[i]; + worker->data2 = &pbi->tile_worker_info[i]; + } + + // Initialize thread frame counts. + if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { + for (i = 0; i < num_workers; ++i) { + TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1; + vp10_zero(twd->counts); + } + } + + // Load tile data into tile_buffers + get_tile_buffers(pbi, data, data_end, tile_buffers); + + for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) { + // Sort the buffers in this tile row based on size in descending order. + qsort(&tile_buffers[tile_row][tile_cols_start], + tile_cols_end - tile_cols_start, sizeof(tile_buffers[0][0]), + compare_tile_buffers); + + // Rearrange the tile buffers in this tile row such that per-tile group + // the largest, and presumably the most difficult tile will be decoded in + // the main thread. This should help minimize the number of instances + // where the main thread is waiting for a worker to complete. + { + int group_start; + for (group_start = tile_cols_start; group_start < tile_cols_end; + group_start += num_workers) { + const int group_end = VPXMIN(group_start + num_workers, tile_cols); + const TileBufferDec largest = tile_buffers[tile_row][group_start]; + memmove(&tile_buffers[tile_row][group_start], + &tile_buffers[tile_row][group_start + 1], + (group_end - group_start - 1) * sizeof(tile_buffers[0][0])); + tile_buffers[tile_row][group_end - 1] = largest; + } + } + + for (tile_col = tile_cols_start; tile_col < tile_cols_end;) { + // Launch workers for individual columns + for (i = 0; i < num_workers && tile_col < tile_cols_end; + ++i, ++tile_col) { + TileBufferDec *const buf = &tile_buffers[tile_row][tile_col]; + VPxWorker *const worker = &pbi->tile_workers[i]; + TileWorkerData *const twd = (TileWorkerData *)worker->data1; + TileInfo *const tile_info = (TileInfo *)worker->data2; + + twd->pbi = pbi; + twd->xd = pbi->mb; + twd->xd.corrupted = 0; + twd->xd.counts = + cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD + ? &twd->counts + : NULL; + vp10_zero(twd->dqcoeff); + vp10_tile_init(tile_info, cm, tile_row, buf->col); + vp10_tile_init(&twd->xd.tile, cm, tile_row, buf->col); +#if !CONFIG_ANS + setup_bool_decoder(buf->data, data_end, buf->size, &cm->error, + &twd->bit_reader, pbi->decrypt_cb, + pbi->decrypt_state); +#else + setup_token_decoder(buf->data, data_end, buf->size, &cm->error, + &twd->bit_reader, pbi->decrypt_cb, + pbi->decrypt_state); +#endif // CONFIG_ANS + vp10_init_macroblockd(cm, &twd->xd, twd->dqcoeff); + twd->xd.plane[0].color_index_map = twd->color_index_map[0]; + twd->xd.plane[1].color_index_map = twd->color_index_map[1]; + + worker->had_error = 0; + if (i == num_workers - 1 || tile_col == tile_cols_end - 1) { + winterface->execute(worker); + } else { + winterface->launch(worker); + } + +#if !(CONFIG_ANS || CONFIG_EXT_TILE) + if (tile_row == tile_rows - 1 && buf->col == tile_cols - 1) { + final_worker = i; + } +#endif // !(CONFIG_ANS || CONFIG_EXT_TILE) + } + + // Sync all workers + for (; i > 0; --i) { + VPxWorker *const worker = &pbi->tile_workers[i - 1]; + // TODO(jzern): The tile may have specific error data associated with + // its vpx_internal_error_info which could be propagated to the main + // info in cm. Additionally once the threads have been synced and an + // error is detected, there's no point in continuing to decode tiles. + pbi->mb.corrupted |= !winterface->sync(worker); + } + } + } + + // Accumulate thread frame counts. + if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { + for (i = 0; i < num_workers; ++i) { + TileWorkerData *const twd = (TileWorkerData *)pbi->tile_workers[i].data1; + vp10_accumulate_frame_counts(cm, &twd->counts); + } + } + +#if CONFIG_EXT_TILE + // Return the end of the last tile buffer + return tile_buffers[tile_rows - 1][tile_cols - 1].raw_data_end; +#else +#if CONFIG_ANS + return data_end; +#else + assert(final_worker != -1); + { + TileWorkerData *const twd = + (TileWorkerData *)pbi->tile_workers[final_worker].data1; + return vpx_reader_find_end(&twd->bit_reader); + } +#endif // CONFIG_ANS +#endif // CONFIG_EXT_TILE +} + +static void error_handler(void *data) { + VP10_COMMON *const cm = (VP10_COMMON *)data; + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet"); +} + +static void read_bitdepth_colorspace_sampling(VP10_COMMON *cm, + struct vpx_read_bit_buffer *rb) { + if (cm->profile >= PROFILE_2) { + cm->bit_depth = vpx_rb_read_bit(rb) ? VPX_BITS_12 : VPX_BITS_10; +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth = 1; +#endif + } else { + cm->bit_depth = VPX_BITS_8; +#if CONFIG_VP9_HIGHBITDEPTH + cm->use_highbitdepth = 0; +#endif + } + cm->color_space = vpx_rb_read_literal(rb, 3); + if (cm->color_space != VPX_CS_SRGB) { + // [16,235] (including xvycc) vs [0,255] range + cm->color_range = vpx_rb_read_bit(rb); + if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) { + cm->subsampling_x = vpx_rb_read_bit(rb); + cm->subsampling_y = vpx_rb_read_bit(rb); + if (cm->subsampling_x == 1 && cm->subsampling_y == 1) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "4:2:0 color not supported in profile 1 or 3"); + if (vpx_rb_read_bit(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Reserved bit set"); + } else { + cm->subsampling_y = cm->subsampling_x = 1; + } + } else { + if (cm->profile == PROFILE_1 || cm->profile == PROFILE_3) { + // Note if colorspace is SRGB then 4:4:4 chroma sampling is assumed. + // 4:2:2 or 4:4:0 chroma sampling is not allowed. + cm->subsampling_y = cm->subsampling_x = 0; + if (vpx_rb_read_bit(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Reserved bit set"); + } else { + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "4:4:4 color not supported in profile 0 or 2"); + } + } +} + +static size_t read_uncompressed_header(VP10Decoder *pbi, + struct vpx_read_bit_buffer *rb) { + VP10_COMMON *const cm = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = pool->frame_bufs; + int i, mask, ref_index = 0; + size_t sz; +#if CONFIG_EXT_REFS + cm->last3_frame_type = cm->last2_frame_type; + cm->last2_frame_type = cm->last_frame_type; +#endif // CONFIG_EXT_REFS + cm->last_frame_type = cm->frame_type; + cm->last_intra_only = cm->intra_only; + +#if CONFIG_EXT_REFS + // NOTE: By default all coded frames to be used as a reference + cm->is_reference_frame = 1; +#endif // CONFIG_EXT_REFS + + if (vpx_rb_read_literal(rb, 2) != VPX_FRAME_MARKER) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame marker"); + + cm->profile = vp10_read_profile(rb); +#if CONFIG_VP9_HIGHBITDEPTH + if (cm->profile >= MAX_PROFILES) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Unsupported bitstream profile"); +#else + if (cm->profile >= PROFILE_2) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Unsupported bitstream profile"); +#endif + + cm->show_existing_frame = vpx_rb_read_bit(rb); + + if (cm->show_existing_frame) { + // Show an existing frame directly. + const int frame_to_show = cm->ref_frame_map[vpx_rb_read_literal(rb, 3)]; + + lock_buffer_pool(pool); + if (frame_to_show < 0 || frame_bufs[frame_to_show].ref_count < 1) { + unlock_buffer_pool(pool); + vpx_internal_error(&cm->error, VPX_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); + unlock_buffer_pool(pool); + + cm->lf.filter_level = 0; + cm->show_frame = 1; + pbi->refresh_frame_flags = 0; + + if (cm->frame_parallel_decode) { + for (i = 0; i < REF_FRAMES; ++i) + cm->next_ref_frame_map[i] = cm->ref_frame_map[i]; + } + + return 0; + } + + cm->frame_type = (FRAME_TYPE)vpx_rb_read_bit(rb); + cm->show_frame = vpx_rb_read_bit(rb); + cm->error_resilient_mode = vpx_rb_read_bit(rb); + + if (cm->frame_type == KEY_FRAME) { + if (!vp10_read_sync_code(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame sync code"); + + read_bitdepth_colorspace_sampling(cm, rb); + pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1; + + for (i = 0; i < INTER_REFS_PER_FRAME; ++i) { + cm->frame_refs[i].idx = INVALID_IDX; + cm->frame_refs[i].buf = NULL; + } + + setup_frame_size(cm, rb); + if (pbi->need_resync) { + memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map)); + pbi->need_resync = 0; + } + if (frame_is_intra_only(cm)) + cm->allow_screen_content_tools = vpx_rb_read_bit(rb); + } else { + cm->intra_only = cm->show_frame ? 0 : vpx_rb_read_bit(rb); + + if (cm->error_resilient_mode) { + cm->reset_frame_context = RESET_FRAME_CONTEXT_ALL; + } else { + if (cm->intra_only) { + cm->reset_frame_context = vpx_rb_read_bit(rb) + ? RESET_FRAME_CONTEXT_ALL + : RESET_FRAME_CONTEXT_CURRENT; + } else { + cm->reset_frame_context = vpx_rb_read_bit(rb) + ? RESET_FRAME_CONTEXT_CURRENT + : RESET_FRAME_CONTEXT_NONE; + if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT) + cm->reset_frame_context = vpx_rb_read_bit(rb) + ? RESET_FRAME_CONTEXT_ALL + : RESET_FRAME_CONTEXT_CURRENT; + } + } + + if (cm->intra_only) { + if (!vp10_read_sync_code(rb)) + vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid frame sync code"); + + read_bitdepth_colorspace_sampling(cm, rb); + + pbi->refresh_frame_flags = vpx_rb_read_literal(rb, REF_FRAMES); + setup_frame_size(cm, rb); + 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 = vpx_rb_read_literal(rb, REF_FRAMES); + +#if CONFIG_EXT_REFS + 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; + } +#endif // CONFIG_EXT_REFS + + for (i = 0; i < INTER_REFS_PER_FRAME; ++i) { + const int ref = vpx_rb_read_literal(rb, REF_FRAMES_LOG2); + const int idx = cm->ref_frame_map[ref]; + RefBuffer *const ref_frame = &cm->frame_refs[i]; + ref_frame->idx = idx; + ref_frame->buf = &frame_bufs[idx].buf; + cm->ref_frame_sign_bias[LAST_FRAME + i] = vpx_rb_read_bit(rb); + } + + setup_frame_size_with_refs(cm, rb); + + cm->allow_high_precision_mv = vpx_rb_read_bit(rb); + cm->interp_filter = read_interp_filter(rb); + + for (i = 0; i < INTER_REFS_PER_FRAME; ++i) { + RefBuffer *const ref_buf = &cm->frame_refs[i]; +#if CONFIG_VP9_HIGHBITDEPTH + vp10_setup_scale_factors_for_frame( + &ref_buf->sf, ref_buf->buf->y_crop_width, + ref_buf->buf->y_crop_height, cm->width, cm->height, + cm->use_highbitdepth); +#else + vp10_setup_scale_factors_for_frame( + &ref_buf->sf, ref_buf->buf->y_crop_width, + ref_buf->buf->y_crop_height, cm->width, cm->height); +#endif + } + } + } +#if CONFIG_VP9_HIGHBITDEPTH + get_frame_new_buffer(cm)->bit_depth = cm->bit_depth; +#endif + get_frame_new_buffer(cm)->color_space = cm->color_space; + get_frame_new_buffer(cm)->color_range = cm->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) { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Keyframe / intra-only frame required to reset decoder" + " state"); + } + + if (!cm->error_resilient_mode) { + cm->refresh_frame_context = vpx_rb_read_bit(rb) + ? REFRESH_FRAME_CONTEXT_FORWARD + : REFRESH_FRAME_CONTEXT_BACKWARD; + } else { + cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_FORWARD; + } + + // This flag will be overridden by the call to vp10_setup_past_independence + // below, forcing the use of context 0 for those frame types. + cm->frame_context_idx = vpx_rb_read_literal(rb, FRAME_CONTEXTS_LOG2); + + // Generate next_ref_frame_map. + lock_buffer_pool(pool); + for (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 (frame_is_intra_only(cm) || cm->error_resilient_mode) + vp10_setup_past_independence(cm); + +#if CONFIG_EXT_PARTITION + set_sb_size(cm, vpx_rb_read_bit(rb) ? BLOCK_128X128 : BLOCK_64X64); +#else + set_sb_size(cm, BLOCK_64X64); +#endif // CONFIG_EXT_PARTITION + + setup_loopfilter(cm, rb); +#if CONFIG_CLPF + setup_clpf(cm, rb); +#endif +#if CONFIG_DERING + setup_dering(cm, rb); +#endif +#if CONFIG_LOOP_RESTORATION + setup_restoration(cm, rb); +#endif // CONFIG_LOOP_RESTORATION + setup_quantization(cm, rb); +#if CONFIG_VP9_HIGHBITDEPTH + xd->bd = (int)cm->bit_depth; +#endif + +#if CONFIG_ENTROPY + vp10_default_coef_probs(cm); + if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode || + cm->reset_frame_context == RESET_FRAME_CONTEXT_ALL) { + for (i = 0; i < FRAME_CONTEXTS; ++i) cm->frame_contexts[i] = *cm->fc; + } else if (cm->reset_frame_context == RESET_FRAME_CONTEXT_CURRENT) { + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; + } +#endif // CONFIG_ENTROPY + + setup_segmentation(cm, rb); + + { + int i; + for (i = 0; i < MAX_SEGMENTS; ++i) { + const int qindex = cm->seg.enabled + ? vp10_get_qindex(&cm->seg, i, cm->base_qindex) + : cm->base_qindex; + xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 && + cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0; + } + } + + setup_segmentation_dequant(cm); + cm->tx_mode = + (!cm->seg.enabled && xd->lossless[0]) ? ONLY_4X4 : read_tx_mode(rb); + cm->reference_mode = read_frame_reference_mode(cm, rb); + + read_tile_info(pbi, rb); + sz = vpx_rb_read_literal(rb, 16); + + if (sz == 0) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Invalid header size"); + + return sz; +} + +#if CONFIG_EXT_TX +static void read_ext_tx_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i, j, k; + int s; + for (s = 1; s < EXT_TX_SETS_INTER; ++s) { + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (i = TX_4X4; i < EXT_TX_SIZES; ++i) { + if (!use_inter_ext_tx_for_txsize[s][i]) continue; + for (j = 0; j < num_ext_tx_set_inter[s] - 1; ++j) + vp10_diff_update_prob(r, &fc->inter_ext_tx_prob[s][i][j]); + } + } + } + + for (s = 1; s < EXT_TX_SETS_INTRA; ++s) { + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (i = TX_4X4; i < EXT_TX_SIZES; ++i) { + if (!use_intra_ext_tx_for_txsize[s][i]) continue; + for (j = 0; j < INTRA_MODES; ++j) + for (k = 0; k < num_ext_tx_set_intra[s] - 1; ++k) + vp10_diff_update_prob(r, &fc->intra_ext_tx_prob[s][i][j][k]); + } + } + } +} + +#else + +static void read_ext_tx_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i, j, k; + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (i = TX_4X4; i < EXT_TX_SIZES; ++i) { + for (j = 0; j < TX_TYPES; ++j) + for (k = 0; k < TX_TYPES - 1; ++k) + vp10_diff_update_prob(r, &fc->intra_ext_tx_prob[i][j][k]); + } + } + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (i = TX_4X4; i < EXT_TX_SIZES; ++i) { + for (k = 0; k < TX_TYPES - 1; ++k) + vp10_diff_update_prob(r, &fc->inter_ext_tx_prob[i][k]); + } + } +} +#endif // CONFIG_EXT_TX + +#if CONFIG_SUPERTX +static void read_supertx_probs(FRAME_CONTEXT *fc, vp10_reader *r) { + int i, j; + if (vp10_read(r, GROUP_DIFF_UPDATE_PROB)) { + for (i = 0; i < PARTITION_SUPERTX_CONTEXTS; ++i) { + for (j = 1; j < TX_SIZES; ++j) { + vp10_diff_update_prob(r, &fc->supertx_prob[i][j]); + } + } + } +} +#endif // CONFIG_SUPERTX + +#if CONFIG_GLOBAL_MOTION +static void read_global_motion_params(Global_Motion_Params *params, + vpx_prob *probs, vp10_reader *r) { + GLOBAL_MOTION_TYPE gmtype = + vp10_read_tree(r, vp10_global_motion_types_tree, probs); + params->gmtype = gmtype; + params->motion_params.wmtype = gm_to_trans_type(gmtype); + switch (gmtype) { + case GLOBAL_ZERO: break; + case GLOBAL_AFFINE: + params->motion_params.wmmat[4] = + (vp10_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) * + GM_ALPHA_DECODE_FACTOR); + params->motion_params.wmmat[5] = + vp10_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) * + GM_ALPHA_DECODE_FACTOR + + (1 << WARPEDMODEL_PREC_BITS); + // fallthrough intended + case GLOBAL_ROTZOOM: + params->motion_params.wmmat[2] = + (vp10_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) * + GM_ALPHA_DECODE_FACTOR) + + (1 << WARPEDMODEL_PREC_BITS); + params->motion_params.wmmat[3] = + vp10_read_primitive_symmetric(r, GM_ABS_ALPHA_BITS) * + GM_ALPHA_DECODE_FACTOR; + // fallthrough intended + case GLOBAL_TRANSLATION: + params->motion_params.wmmat[0] = + vp10_read_primitive_symmetric(r, GM_ABS_TRANS_BITS) * + GM_TRANS_DECODE_FACTOR; + params->motion_params.wmmat[1] = + vp10_read_primitive_symmetric(r, GM_ABS_TRANS_BITS) * + GM_TRANS_DECODE_FACTOR; + break; + default: assert(0); + } +} + +static void read_global_motion(VP10_COMMON *cm, vp10_reader *r) { + int frame; + memset(cm->global_motion, 0, sizeof(cm->global_motion)); + for (frame = LAST_FRAME; frame <= ALTREF_FRAME; ++frame) { + read_global_motion_params(&cm->global_motion[frame], + cm->fc->global_motion_types_prob, r); + } +} +#endif // CONFIG_GLOBAL_MOTION + +static int read_compressed_header(VP10Decoder *pbi, const uint8_t *data, + size_t partition_size) { + VP10_COMMON *const cm = &pbi->common; +#if CONFIG_SUPERTX + MACROBLOCKD *const xd = &pbi->mb; +#endif + FRAME_CONTEXT *const fc = cm->fc; + vp10_reader r; + int k, i, j; + +#if !CONFIG_ANS + if (vpx_reader_init(&r, data, partition_size, pbi->decrypt_cb, + pbi->decrypt_state)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate bool decoder 0"); +#else + if (ans_read_init(&r, data, partition_size)) + vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, + "Failed to allocate compressed header ANS decoder"); +#endif // !CONFIG_ANS + + if (cm->tx_mode == TX_MODE_SELECT) { + for (i = 0; i < TX_SIZES - 1; ++i) + for (j = 0; j < TX_SIZE_CONTEXTS; ++j) + for (k = 0; k < i + 1; ++k) + vp10_diff_update_prob(&r, &fc->tx_size_probs[i][j][k]); + } + + read_coef_probs(fc, cm->tx_mode, &r); + +#if CONFIG_VAR_TX + for (k = 0; k < TXFM_PARTITION_CONTEXTS; ++k) + vp10_diff_update_prob(&r, &fc->txfm_partition_prob[k]); +#endif + + for (k = 0; k < SKIP_CONTEXTS; ++k) + vp10_diff_update_prob(&r, &fc->skip_probs[k]); + + if (cm->seg.enabled && cm->seg.update_map) { + if (cm->seg.temporal_update) { + for (k = 0; k < PREDICTION_PROBS; k++) + vp10_diff_update_prob(&r, &cm->fc->seg.pred_probs[k]); + } + for (k = 0; k < MAX_SEGMENTS - 1; k++) + vp10_diff_update_prob(&r, &cm->fc->seg.tree_probs[k]); + } + + for (j = 0; j < INTRA_MODES; j++) + for (i = 0; i < INTRA_MODES - 1; ++i) + vp10_diff_update_prob(&r, &fc->uv_mode_prob[j][i]); + +#if CONFIG_EXT_PARTITION_TYPES + for (i = 0; i < PARTITION_TYPES - 1; ++i) + vp10_diff_update_prob(&r, &fc->partition_prob[0][i]); + for (j = 1; j < PARTITION_CONTEXTS; ++j) + for (i = 0; i < EXT_PARTITION_TYPES - 1; ++i) + vp10_diff_update_prob(&r, &fc->partition_prob[j][i]); +#else + for (j = 0; j < PARTITION_CONTEXTS; ++j) + for (i = 0; i < PARTITION_TYPES - 1; ++i) + vp10_diff_update_prob(&r, &fc->partition_prob[j][i]); +#endif // CONFIG_EXT_PARTITION_TYPES + +#if CONFIG_EXT_INTRA + for (i = 0; i < INTRA_FILTERS + 1; ++i) + for (j = 0; j < INTRA_FILTERS - 1; ++j) + vp10_diff_update_prob(&r, &fc->intra_filter_probs[i][j]); +#endif // CONFIG_EXT_INTRA + + if (frame_is_intra_only(cm)) { + vp10_copy(cm->kf_y_prob, vp10_kf_y_mode_prob); + for (k = 0; k < INTRA_MODES; k++) + for (j = 0; j < INTRA_MODES; j++) + for (i = 0; i < INTRA_MODES - 1; ++i) + vp10_diff_update_prob(&r, &cm->kf_y_prob[k][j][i]); + } else { +#if !CONFIG_REF_MV + nmv_context *const nmvc = &fc->nmvc; +#endif + + read_inter_mode_probs(fc, &r); + +#if CONFIG_EXT_INTER + read_inter_compound_mode_probs(fc, &r); + if (cm->reference_mode != COMPOUND_REFERENCE) { + for (i = 0; i < BLOCK_SIZE_GROUPS; i++) { + if (is_interintra_allowed_bsize_group(i)) { + vp10_diff_update_prob(&r, &fc->interintra_prob[i]); + } + } + for (i = 0; i < BLOCK_SIZE_GROUPS; i++) { + for (j = 0; j < INTERINTRA_MODES - 1; j++) + vp10_diff_update_prob(&r, &fc->interintra_mode_prob[i][j]); + } + for (i = 0; i < BLOCK_SIZES; i++) { + if (is_interintra_allowed_bsize(i) && is_interintra_wedge_used(i)) { + vp10_diff_update_prob(&r, &fc->wedge_interintra_prob[i]); + } + } + } + if (cm->reference_mode != SINGLE_REFERENCE) { + for (i = 0; i < BLOCK_SIZES; i++) { + if (is_interinter_wedge_used(i)) { + vp10_diff_update_prob(&r, &fc->wedge_interinter_prob[i]); + } + } + } +#endif // CONFIG_EXT_INTER + +#if CONFIG_OBMC || CONFIG_WARPED_MOTION + for (i = BLOCK_8X8; i < BLOCK_SIZES; ++i) { + for (j = 0; j < MOTION_VARIATIONS - 1; ++j) + vp10_diff_update_prob(&r, &fc->motvar_prob[i][j]); + } +#endif // CONFIG_OBMC || CONFIG_WARPED_MOTION + + if (cm->interp_filter == SWITCHABLE) read_switchable_interp_probs(fc, &r); + + for (i = 0; i < INTRA_INTER_CONTEXTS; i++) + vp10_diff_update_prob(&r, &fc->intra_inter_prob[i]); + + if (cm->reference_mode != SINGLE_REFERENCE) + setup_compound_reference_mode(cm); + + read_frame_reference_mode_probs(cm, &r); + + for (j = 0; j < BLOCK_SIZE_GROUPS; j++) + for (i = 0; i < INTRA_MODES - 1; ++i) + vp10_diff_update_prob(&r, &fc->y_mode_prob[j][i]); + +#if CONFIG_REF_MV + for (i = 0; i < NMV_CONTEXTS; ++i) + read_mv_probs(&fc->nmvc[i], cm->allow_high_precision_mv, &r); +#else + read_mv_probs(nmvc, cm->allow_high_precision_mv, &r); +#endif + read_ext_tx_probs(fc, &r); +#if CONFIG_SUPERTX + if (!xd->lossless[0]) read_supertx_probs(fc, &r); +#endif +#if CONFIG_GLOBAL_MOTION + read_global_motion(cm, &r); +#endif // CONFIG_GLOBAL_MOTION + } + + return vp10_reader_has_error(&r); +} + +#ifdef NDEBUG +#define debug_check_frame_counts(cm) (void)0 +#else // !NDEBUG +// Counts should only be incremented when frame_parallel_decoding_mode and +// error_resilient_mode are disabled. +static void debug_check_frame_counts(const VP10_COMMON *const cm) { + FRAME_COUNTS zero_counts; + vp10_zero(zero_counts); + assert(cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD || + cm->error_resilient_mode); + assert(!memcmp(cm->counts.y_mode, zero_counts.y_mode, + sizeof(cm->counts.y_mode))); + assert(!memcmp(cm->counts.uv_mode, zero_counts.uv_mode, + sizeof(cm->counts.uv_mode))); + assert(!memcmp(cm->counts.partition, zero_counts.partition, + sizeof(cm->counts.partition))); + assert(!memcmp(cm->counts.coef, zero_counts.coef, sizeof(cm->counts.coef))); + assert(!memcmp(cm->counts.eob_branch, zero_counts.eob_branch, + sizeof(cm->counts.eob_branch))); + assert(!memcmp(cm->counts.switchable_interp, zero_counts.switchable_interp, + sizeof(cm->counts.switchable_interp))); + assert(!memcmp(cm->counts.inter_mode, zero_counts.inter_mode, + sizeof(cm->counts.inter_mode))); +#if CONFIG_EXT_INTER + assert(!memcmp(cm->counts.inter_compound_mode, + zero_counts.inter_compound_mode, + sizeof(cm->counts.inter_compound_mode))); + assert(!memcmp(cm->counts.interintra, zero_counts.interintra, + sizeof(cm->counts.interintra))); + assert(!memcmp(cm->counts.wedge_interintra, zero_counts.wedge_interintra, + sizeof(cm->counts.wedge_interintra))); + assert(!memcmp(cm->counts.wedge_interinter, zero_counts.wedge_interinter, + sizeof(cm->counts.wedge_interinter))); +#endif // CONFIG_EXT_INTER +#if CONFIG_OBMC || CONFIG_WARPED_MOTION + assert(!memcmp(cm->counts.motvar, zero_counts.motvar, + sizeof(cm->counts.motvar))); +#endif // CONFIG_OBMC || CONFIG_WARPED_MOTION + assert(!memcmp(cm->counts.intra_inter, zero_counts.intra_inter, + sizeof(cm->counts.intra_inter))); + assert(!memcmp(cm->counts.comp_inter, zero_counts.comp_inter, + sizeof(cm->counts.comp_inter))); + assert(!memcmp(cm->counts.single_ref, zero_counts.single_ref, + sizeof(cm->counts.single_ref))); + assert(!memcmp(cm->counts.comp_ref, zero_counts.comp_ref, + sizeof(cm->counts.comp_ref))); +#if CONFIG_EXT_REFS + assert(!memcmp(cm->counts.comp_bwdref, zero_counts.comp_bwdref, + sizeof(cm->counts.comp_bwdref))); +#endif // CONFIG_EXT_REFS + assert(!memcmp(&cm->counts.tx_size, &zero_counts.tx_size, + sizeof(cm->counts.tx_size))); + assert(!memcmp(cm->counts.skip, zero_counts.skip, sizeof(cm->counts.skip))); +#if CONFIG_REF_MV + assert( + !memcmp(&cm->counts.mv[0], &zero_counts.mv[0], sizeof(cm->counts.mv[0]))); + assert( + !memcmp(&cm->counts.mv[1], &zero_counts.mv[1], sizeof(cm->counts.mv[0]))); +#else + assert(!memcmp(&cm->counts.mv, &zero_counts.mv, sizeof(cm->counts.mv))); +#endif + assert(!memcmp(cm->counts.inter_ext_tx, zero_counts.inter_ext_tx, + sizeof(cm->counts.inter_ext_tx))); + assert(!memcmp(cm->counts.intra_ext_tx, zero_counts.intra_ext_tx, + sizeof(cm->counts.intra_ext_tx))); +} +#endif // NDEBUG + +static struct vpx_read_bit_buffer *init_read_bit_buffer( + VP10Decoder *pbi, struct vpx_read_bit_buffer *rb, const uint8_t *data, + const uint8_t *data_end, uint8_t clear_data[MAX_VPX_HEADER_SIZE]) { + rb->bit_offset = 0; + rb->error_handler = error_handler; + rb->error_handler_data = &pbi->common; + if (pbi->decrypt_cb) { + const int n = (int)VPXMIN(MAX_VPX_HEADER_SIZE, data_end - data); + pbi->decrypt_cb(pbi->decrypt_state, data, clear_data, n); + rb->bit_buffer = clear_data; + rb->bit_buffer_end = clear_data + n; + } else { + rb->bit_buffer = data; + rb->bit_buffer_end = data_end; + } + return rb; +} + +//------------------------------------------------------------------------------ + +int vp10_read_sync_code(struct vpx_read_bit_buffer *const rb) { + return vpx_rb_read_literal(rb, 8) == VP10_SYNC_CODE_0 && + vpx_rb_read_literal(rb, 8) == VP10_SYNC_CODE_1 && + vpx_rb_read_literal(rb, 8) == VP10_SYNC_CODE_2; +} + +void vp10_read_frame_size(struct vpx_read_bit_buffer *rb, int *width, + int *height) { + *width = vpx_rb_read_literal(rb, 16) + 1; + *height = vpx_rb_read_literal(rb, 16) + 1; +} + +BITSTREAM_PROFILE vp10_read_profile(struct vpx_read_bit_buffer *rb) { + int profile = vpx_rb_read_bit(rb); + profile |= vpx_rb_read_bit(rb) << 1; + if (profile > 2) profile += vpx_rb_read_bit(rb); + return (BITSTREAM_PROFILE)profile; +} + +void vp10_decode_frame(VP10Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, const uint8_t **p_data_end) { + VP10_COMMON *const cm = &pbi->common; + MACROBLOCKD *const xd = &pbi->mb; + struct vpx_read_bit_buffer rb; + int context_updated = 0; + uint8_t clear_data[MAX_VPX_HEADER_SIZE]; + const size_t first_partition_size = read_uncompressed_header( + pbi, init_read_bit_buffer(pbi, &rb, data, data_end, clear_data)); + YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm); + xd->cur_buf = new_fb; +#if CONFIG_GLOBAL_MOTION + xd->global_motion = cm->global_motion; +#endif // CONFIG_GLOBAL_MOTION + + if (!first_partition_size) { +// showing a frame directly +#if CONFIG_EXT_REFS + if (cm->show_existing_frame) + *p_data_end = data + vpx_rb_bytes_read(&rb); + else +#endif // CONFIG_EXT_REFS + *p_data_end = data + (cm->profile <= PROFILE_2 ? 1 : 2); + + return; + } + + data += vpx_rb_bytes_read(&rb); + if (!read_is_valid(data, first_partition_size, data_end)) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Truncated packet or corrupt header length"); + + cm->use_prev_frame_mvs = + !cm->error_resilient_mode && cm->width == cm->last_width && + cm->height == cm->last_height && !cm->last_intra_only && + cm->last_show_frame && (cm->last_frame_type != KEY_FRAME); +#if CONFIG_EXT_REFS + // NOTE(zoeliu): As cm->prev_frame can take neither a frame of + // show_exisiting_frame=1, nor can it take a frame not used as + // a reference, it is probable that by the time it is being + // referred to, the frame buffer it originally points to may + // already get expired and have been reassigned to the current + // newly coded frame. Hence, we need to check whether this is + // the case, and if yes, we have 2 choices: + // (1) Simply disable the use of previous frame mvs; or + // (2) Have cm->prev_frame point to one reference frame buffer, + // e.g. LAST_FRAME. + if (cm->use_prev_frame_mvs && !dec_is_ref_frame_buf(pbi, cm->prev_frame)) { + // Reassign the LAST_FRAME buffer to cm->prev_frame. + RefBuffer *last_fb_ref_buf = &cm->frame_refs[LAST_FRAME - LAST_FRAME]; + cm->prev_frame = &cm->buffer_pool->frame_bufs[last_fb_ref_buf->idx]; + } +#endif // CONFIG_EXT_REFS + + vp10_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y); + + *cm->fc = cm->frame_contexts[cm->frame_context_idx]; + if (!cm->fc->initialized) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Uninitialized entropy context."); + + vp10_zero(cm->counts); + + xd->corrupted = 0; + new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size); + if (new_fb->corrupted) + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data header is corrupted."); + + if (cm->lf.filter_level && !cm->skip_loop_filter) { + vp10_loop_filter_frame_init(cm, cm->lf.filter_level); + } + + // If encoded in frame parallel mode, frame context is ready after decoding + // the frame header. + if (cm->frame_parallel_decode && + cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_BACKWARD) { + VPxWorker *const worker = pbi->frame_worker_owner; + FrameWorkerData *const frame_worker_data = worker->data1; + if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_FORWARD) { + context_updated = 1; + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; + } + vp10_frameworker_lock_stats(worker); + pbi->cur_buf->row = -1; + pbi->cur_buf->col = -1; + frame_worker_data->frame_context_ready = 1; + // Signal the main thread that context is ready. + vp10_frameworker_signal_stats(worker); + vp10_frameworker_unlock_stats(worker); + } + +#if CONFIG_ENTROPY + vp10_copy(cm->starting_coef_probs, cm->fc->coef_probs); + cm->coef_probs_update_idx = 0; +#endif // CONFIG_ENTROPY + + if (pbi->max_threads > 1 +#if CONFIG_EXT_TILE + && pbi->dec_tile_col < 0 // Decoding all columns +#endif // CONFIG_EXT_TILE + && cm->tile_cols > 1) { + // Multi-threaded tile decoder + *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end); + if (!xd->corrupted) { + if (!cm->skip_loop_filter) { + // If multiple threads are used to decode tiles, then we use those + // threads to do parallel loopfiltering. + vp10_loop_filter_frame_mt(new_fb, cm, pbi->mb.plane, + cm->lf.filter_level, 0, 0, pbi->tile_workers, + pbi->num_tile_workers, &pbi->lf_row_sync); + } + } else { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data is corrupted."); + } + } else { + *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end); + } +#if CONFIG_LOOP_RESTORATION + if (cm->rst_info.restoration_type != RESTORE_NONE) { + vp10_loop_restoration_init(&cm->rst_internal, &cm->rst_info, + cm->frame_type == KEY_FRAME, cm->width, + cm->height); + vp10_loop_restoration_rows(new_fb, cm, 0, cm->mi_rows, 0); + } +#endif // CONFIG_LOOP_RESTORATION + + if (!xd->corrupted) { + if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { +#if CONFIG_ENTROPY + cm->partial_prob_update = 0; +#endif // CONFIG_ENTROPY + vp10_adapt_coef_probs(cm); + vp10_adapt_intra_frame_probs(cm); + + if (!frame_is_intra_only(cm)) { + vp10_adapt_inter_frame_probs(cm); + vp10_adapt_mv_probs(cm, cm->allow_high_precision_mv); + } + } else { + debug_check_frame_counts(cm); + } + } else { + vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, + "Decode failed. Frame data is corrupted."); + } + + // Non frame parallel update frame context here. + if (!cm->error_resilient_mode && !context_updated) + cm->frame_contexts[cm->frame_context_idx] = *cm->fc; +}
diff --git a/av1/decoder/decodeframe.h b/av1/decoder/decodeframe.h new file mode 100644 index 0000000..7fdff0b --- /dev/null +++ b/av1/decoder/decodeframe.h
@@ -0,0 +1,33 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DECODEFRAME_H_ +#define VP10_DECODER_DECODEFRAME_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP10Decoder; +struct vpx_read_bit_buffer; + +int vp10_read_sync_code(struct vpx_read_bit_buffer *const rb); +void vp10_read_frame_size(struct vpx_read_bit_buffer *rb, int *width, + int *height); +BITSTREAM_PROFILE vp10_read_profile(struct vpx_read_bit_buffer *rb); + +void vp10_decode_frame(struct VP10Decoder *pbi, const uint8_t *data, + const uint8_t *data_end, const uint8_t **p_data_end); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP10_DECODER_DECODEFRAME_H_
diff --git a/av1/decoder/decodemv.c b/av1/decoder/decodemv.c new file mode 100644 index 0000000..ef776a0 --- /dev/null +++ b/av1/decoder/decodemv.c
@@ -0,0 +1,1782 @@ +/* + Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.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/seg_common.h" + +#include "av1/decoder/decodemv.h" +#include "av1/decoder/decodeframe.h" + +#include "aom_dsp/vpx_dsp_common.h" + +static INLINE int read_uniform(vp10_reader *r, int n) { + int l = get_unsigned_bits(n); + int m = (1 << l) - n; + int v = vp10_read_literal(r, l - 1); + + assert(l != 0); + + if (v < m) + return v; + else + return (v << 1) - m + vp10_read_literal(r, 1); +} + +static PREDICTION_MODE read_intra_mode(vp10_reader *r, const vpx_prob *p) { + return (PREDICTION_MODE)vp10_read_tree(r, vp10_intra_mode_tree, p); +} + +static PREDICTION_MODE read_intra_mode_y(VP10_COMMON *cm, MACROBLOCKD *xd, + vp10_reader *r, int size_group) { + const PREDICTION_MODE y_mode = + read_intra_mode(r, cm->fc->y_mode_prob[size_group]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->y_mode[size_group][y_mode]; + return y_mode; +} + +static PREDICTION_MODE read_intra_mode_uv(VP10_COMMON *cm, MACROBLOCKD *xd, + vp10_reader *r, + PREDICTION_MODE y_mode) { + const PREDICTION_MODE uv_mode = + read_intra_mode(r, cm->fc->uv_mode_prob[y_mode]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->uv_mode[y_mode][uv_mode]; + return uv_mode; +} + +#if CONFIG_EXT_INTER +static INTERINTRA_MODE read_interintra_mode(VP10_COMMON *cm, MACROBLOCKD *xd, + vp10_reader *r, int size_group) { + const INTERINTRA_MODE ii_mode = (INTERINTRA_MODE)vp10_read_tree( + r, vp10_interintra_mode_tree, cm->fc->interintra_mode_prob[size_group]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->interintra_mode[size_group][ii_mode]; + return ii_mode; +} +#endif // CONFIG_EXT_INTER + +static PREDICTION_MODE read_inter_mode(VP10_COMMON *cm, MACROBLOCKD *xd, +#if CONFIG_REF_MV && CONFIG_EXT_INTER + MB_MODE_INFO *mbmi, +#endif + vp10_reader *r, int16_t ctx) { +#if CONFIG_REF_MV + FRAME_COUNTS *counts = xd->counts; + int16_t mode_ctx = ctx & NEWMV_CTX_MASK; + vpx_prob mode_prob = cm->fc->newmv_prob[mode_ctx]; + + if (vp10_read(r, mode_prob) == 0) { + if (counts) ++counts->newmv_mode[mode_ctx][0]; + +#if CONFIG_EXT_INTER + if (has_second_ref(mbmi)) { +#endif // CONFIG_EXT_INTER + return NEWMV; +#if CONFIG_EXT_INTER + } else { + mode_prob = cm->fc->new2mv_prob; + if (vp10_read(r, mode_prob) == 0) { + if (counts) ++counts->new2mv_mode[0]; + return NEWMV; + } else { + if (counts) ++counts->new2mv_mode[1]; + return NEWFROMNEARMV; + } + } +#endif // CONFIG_EXT_INTER + } + if (counts) ++counts->newmv_mode[mode_ctx][1]; + + if (ctx & (1 << ALL_ZERO_FLAG_OFFSET)) return ZEROMV; + + mode_ctx = (ctx >> ZEROMV_OFFSET) & ZEROMV_CTX_MASK; + + mode_prob = cm->fc->zeromv_prob[mode_ctx]; + if (vp10_read(r, mode_prob) == 0) { + if (counts) ++counts->zeromv_mode[mode_ctx][0]; + return ZEROMV; + } + if (counts) ++counts->zeromv_mode[mode_ctx][1]; + + mode_ctx = (ctx >> REFMV_OFFSET) & REFMV_CTX_MASK; + + if (ctx & (1 << SKIP_NEARESTMV_OFFSET)) mode_ctx = 6; + if (ctx & (1 << SKIP_NEARMV_OFFSET)) mode_ctx = 7; + if (ctx & (1 << SKIP_NEARESTMV_SUB8X8_OFFSET)) mode_ctx = 8; + + mode_prob = cm->fc->refmv_prob[mode_ctx]; + + if (vp10_read(r, mode_prob) == 0) { + if (counts) ++counts->refmv_mode[mode_ctx][0]; + + return NEARESTMV; + } else { + if (counts) ++counts->refmv_mode[mode_ctx][1]; + return NEARMV; + } + + // Invalid prediction mode. + assert(0); +#else + const int mode = + vp10_read_tree(r, vp10_inter_mode_tree, cm->fc->inter_mode_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->inter_mode[ctx][mode]; + + return NEARESTMV + mode; +#endif +} + +#if CONFIG_REF_MV +static void read_drl_idx(const VP10_COMMON *cm, MACROBLOCKD *xd, + MB_MODE_INFO *mbmi, vp10_reader *r) { + uint8_t ref_frame_type = vp10_ref_frame_type(mbmi->ref_frame); + mbmi->ref_mv_idx = 0; + + if (mbmi->mode == NEWMV) { + int idx; + for (idx = 0; idx < 2; ++idx) { + if (xd->ref_mv_count[ref_frame_type] > idx + 1) { + uint8_t drl_ctx = vp10_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx); + vpx_prob drl_prob = cm->fc->drl_prob[drl_ctx]; + if (!vp10_read(r, drl_prob)) { + mbmi->ref_mv_idx = idx; + if (xd->counts) ++xd->counts->drl_mode[drl_ctx][0]; + return; + } + mbmi->ref_mv_idx = idx + 1; + if (xd->counts) ++xd->counts->drl_mode[drl_ctx][1]; + } + } + } + + if (mbmi->mode == NEARMV) { + int idx; + // Offset the NEARESTMV mode. + // TODO(jingning): Unify the two syntax decoding loops after the NEARESTMV + // mode is factored in. + for (idx = 1; idx < 3; ++idx) { + if (xd->ref_mv_count[ref_frame_type] > idx + 1) { + uint8_t drl_ctx = vp10_drl_ctx(xd->ref_mv_stack[ref_frame_type], idx); + vpx_prob drl_prob = cm->fc->drl_prob[drl_ctx]; + if (!vp10_read(r, drl_prob)) { + mbmi->ref_mv_idx = idx - 1; + if (xd->counts) ++xd->counts->drl_mode[drl_ctx][0]; + return; + } + mbmi->ref_mv_idx = idx; + if (xd->counts) ++xd->counts->drl_mode[drl_ctx][1]; + } + } + } +} +#endif + +#if CONFIG_EXT_INTER +static PREDICTION_MODE read_inter_compound_mode(VP10_COMMON *cm, + MACROBLOCKD *xd, vp10_reader *r, + int16_t ctx) { + const int mode = vp10_read_tree(r, vp10_inter_compound_mode_tree, + cm->fc->inter_compound_mode_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + + if (counts) ++counts->inter_compound_mode[ctx][mode]; + + assert(is_inter_compound_mode(NEAREST_NEARESTMV + mode)); + return NEAREST_NEARESTMV + mode; +} +#endif // CONFIG_EXT_INTER + +static int read_segment_id(vp10_reader *r, + const struct segmentation_probs *segp) { + return vp10_read_tree(r, vp10_segment_tree, segp->tree_probs); +} + +#if CONFIG_VAR_TX +static void read_tx_size_vartx(VP10_COMMON *cm, MACROBLOCKD *xd, + MB_MODE_INFO *mbmi, FRAME_COUNTS *counts, + TX_SIZE tx_size, int blk_row, int blk_col, + vp10_reader *r) { + int is_split = 0; + const int tx_row = blk_row >> 1; + const int tx_col = blk_col >> 1; + int max_blocks_high = num_4x4_blocks_high_lookup[mbmi->sb_type]; + int max_blocks_wide = num_4x4_blocks_wide_lookup[mbmi->sb_type]; + int ctx = txfm_partition_context(xd->above_txfm_context + tx_col, + xd->left_txfm_context + tx_row, tx_size); + TX_SIZE (*const inter_tx_size) + [MAX_MIB_SIZE] = + (TX_SIZE(*)[MAX_MIB_SIZE]) & mbmi->inter_tx_size[tx_row][tx_col]; + + if (xd->mb_to_bottom_edge < 0) max_blocks_high += xd->mb_to_bottom_edge >> 5; + if (xd->mb_to_right_edge < 0) max_blocks_wide += xd->mb_to_right_edge >> 5; + + if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return; + + is_split = vp10_read(r, cm->fc->txfm_partition_prob[ctx]); + + if (is_split) { + BLOCK_SIZE bsize = txsize_to_bsize[tx_size]; + int bsl = b_width_log2_lookup[bsize]; + int i; + + if (counts) ++counts->txfm_partition[ctx][1]; + + if (tx_size == TX_8X8) { + inter_tx_size[0][0] = TX_4X4; + mbmi->tx_size = TX_4X4; + txfm_partition_update(xd->above_txfm_context + tx_col, + xd->left_txfm_context + tx_row, TX_4X4); + return; + } + + assert(bsl > 0); + --bsl; + for (i = 0; i < 4; ++i) { + int offsetr = blk_row + ((i >> 1) << bsl); + int offsetc = blk_col + ((i & 0x01) << bsl); + read_tx_size_vartx(cm, xd, mbmi, counts, tx_size - 1, offsetr, offsetc, + r); + } + } else { + int idx, idy; + inter_tx_size[0][0] = tx_size; + for (idy = 0; idy < num_4x4_blocks_high_txsize_lookup[tx_size] / 2; ++idy) + for (idx = 0; idx < num_4x4_blocks_wide_txsize_lookup[tx_size] / 2; ++idx) + inter_tx_size[idy][idx] = tx_size; + mbmi->tx_size = tx_size; + if (counts) ++counts->txfm_partition[ctx][0]; + txfm_partition_update(xd->above_txfm_context + tx_col, + xd->left_txfm_context + tx_row, tx_size); + } +} +#endif + +static TX_SIZE read_selected_tx_size(VP10_COMMON *cm, MACROBLOCKD *xd, + int tx_size_cat, vp10_reader *r) { + FRAME_COUNTS *counts = xd->counts; + const int ctx = get_tx_size_context(xd); + int tx_size = vp10_read_tree(r, vp10_tx_size_tree[tx_size_cat], + cm->fc->tx_size_probs[tx_size_cat][ctx]); + if (counts) ++counts->tx_size[tx_size_cat][ctx][tx_size]; + return (TX_SIZE)tx_size; +} + +static TX_SIZE read_tx_size_intra(VP10_COMMON *cm, MACROBLOCKD *xd, + vp10_reader *r) { + TX_MODE tx_mode = cm->tx_mode; + BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type; + if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return TX_4X4; + if (bsize >= BLOCK_8X8) { + if (tx_mode == TX_MODE_SELECT) { + const TX_SIZE tx_size = + read_selected_tx_size(cm, xd, intra_tx_size_cat_lookup[bsize], r); + assert(tx_size <= max_txsize_lookup[bsize]); + return tx_size; + } else { + return tx_size_from_tx_mode(bsize, cm->tx_mode, 0); + } + } else { + return TX_4X4; + } +} + +static TX_SIZE read_tx_size_inter(VP10_COMMON *cm, MACROBLOCKD *xd, + int allow_select, vp10_reader *r) { + TX_MODE tx_mode = cm->tx_mode; + BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type; + if (xd->lossless[xd->mi[0]->mbmi.segment_id]) return TX_4X4; + if (bsize >= BLOCK_8X8) { + if (allow_select && tx_mode == TX_MODE_SELECT) { + const TX_SIZE coded_tx_size = + read_selected_tx_size(cm, xd, inter_tx_size_cat_lookup[bsize], r); +#if !CONFIG_RECT_TX + assert(coded_tx_size <= max_txsize_lookup[bsize]); +#else + if (coded_tx_size > max_txsize_lookup[bsize]) { + assert(coded_tx_size == max_txsize_lookup[bsize] + 1); + return max_txsize_rect_lookup[bsize]; + } +#endif // !CONFIG_RECT_TX + return coded_tx_size; + } else { + return tx_size_from_tx_mode(bsize, cm->tx_mode, 1); + } + } else { +#if CONFIG_EXT_TX && CONFIG_RECT_TX + assert(IMPLIES(tx_mode == ONLY_4X4, bsize == BLOCK_4X4)); + return max_txsize_rect_lookup[bsize]; +#else + return TX_4X4; +#endif + } +} + +static int dec_get_segment_id(const VP10_COMMON *cm, const uint8_t *segment_ids, + int mi_offset, int x_mis, int y_mis) { + int x, y, segment_id = INT_MAX; + + for (y = 0; y < y_mis; y++) + for (x = 0; x < x_mis; x++) + segment_id = + VPXMIN(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(VP10_COMMON *cm, int mi_offset, int x_mis, int y_mis, + int segment_id) { + int x, y; + + assert(segment_id >= 0 && segment_id < MAX_SEGMENTS); + + for (y = 0; y < y_mis; y++) + for (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(VP10_COMMON *const cm, MACROBLOCKD *const xd, + int mi_offset, int x_mis, int y_mis, + vp10_reader *r) { + struct segmentation *const seg = &cm->seg; + FRAME_COUNTS *counts = xd->counts; + struct segmentation_probs *const segp = &cm->fc->seg; + int segment_id; + + if (!seg->enabled) return 0; // Default for disabled segmentation + + assert(seg->update_map && !seg->temporal_update); + + segment_id = read_segment_id(r, segp); + if (counts) ++counts->seg.tree_total[segment_id]; + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static void copy_segment_id(const VP10_COMMON *cm, + const uint8_t *last_segment_ids, + uint8_t *current_segment_ids, int mi_offset, + int x_mis, int y_mis) { + int x, y; + + for (y = 0; y < y_mis; y++) + for (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 read_inter_segment_id(VP10_COMMON *const cm, MACROBLOCKD *const xd, + int mi_row, int mi_col, vp10_reader *r) { + struct segmentation *const seg = &cm->seg; + FRAME_COUNTS *counts = xd->counts; + struct segmentation_probs *const segp = &cm->fc->seg; + MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi; + int predicted_segment_id, segment_id; + const int mi_offset = mi_row * cm->mi_cols + mi_col; + const int bw = num_8x8_blocks_wide_lookup[mbmi->sb_type]; + const int bh = num_8x8_blocks_high_lookup[mbmi->sb_type]; + + // TODO(slavarnway): move x_mis, y_mis into xd ????? + const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw); + const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh); + + if (!seg->enabled) return 0; // Default for disabled segmentation + + predicted_segment_id = cm->last_frame_seg_map + ? dec_get_segment_id(cm, cm->last_frame_seg_map, + mi_offset, x_mis, y_mis) + : 0; + + 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 predicted_segment_id; + } + + if (seg->temporal_update) { + const int ctx = vp10_get_pred_context_seg_id(xd); + const vpx_prob pred_prob = segp->pred_probs[ctx]; + mbmi->seg_id_predicted = vp10_read(r, pred_prob); + if (counts) ++counts->seg.pred[ctx][mbmi->seg_id_predicted]; + if (mbmi->seg_id_predicted) { + segment_id = predicted_segment_id; + } else { + segment_id = read_segment_id(r, segp); + if (counts) ++counts->seg.tree_mispred[segment_id]; + } + } else { + segment_id = read_segment_id(r, segp); + if (counts) ++counts->seg.tree_total[segment_id]; + } + set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id); + return segment_id; +} + +static int read_skip(VP10_COMMON *cm, const MACROBLOCKD *xd, int segment_id, + vp10_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) { + return 1; + } else { + const int ctx = vp10_get_skip_context(xd); + const int skip = vp10_read(r, cm->fc->skip_probs[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->skip[ctx][skip]; + return skip; + } +} + +static void read_palette_mode_info(VP10_COMMON *const cm, MACROBLOCKD *const xd, + vp10_reader *r) { + MODE_INFO *const mi = xd->mi[0]; + MB_MODE_INFO *const mbmi = &mi->mbmi; + const MODE_INFO *const above_mi = xd->above_mi; + const MODE_INFO *const left_mi = xd->left_mi; + const BLOCK_SIZE bsize = mbmi->sb_type; + int i, n, palette_ctx = 0; + PALETTE_MODE_INFO *const pmi = &mbmi->palette_mode_info; + + if (mbmi->mode == DC_PRED) { + if (above_mi) + palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0); + if (left_mi) + palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0); + if (vp10_read( + r, + vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8][palette_ctx])) { + pmi->palette_size[0] = + vp10_read_tree(r, vp10_palette_size_tree, + vp10_default_palette_y_size_prob[bsize - BLOCK_8X8]) + + 2; + n = pmi->palette_size[0]; + for (i = 0; i < n; ++i) + pmi->palette_colors[i] = vp10_read_literal(r, cm->bit_depth); + + xd->plane[0].color_index_map[0] = read_uniform(r, n); + assert(xd->plane[0].color_index_map[0] < n); + } + } + + if (mbmi->uv_mode == DC_PRED) { + if (vp10_read( + r, vp10_default_palette_uv_mode_prob[pmi->palette_size[0] > 0])) { + pmi->palette_size[1] = + vp10_read_tree(r, vp10_palette_size_tree, + vp10_default_palette_uv_size_prob[bsize - BLOCK_8X8]) + + 2; + n = pmi->palette_size[1]; + for (i = 0; i < n; ++i) { + pmi->palette_colors[PALETTE_MAX_SIZE + i] = + vp10_read_literal(r, cm->bit_depth); + pmi->palette_colors[2 * PALETTE_MAX_SIZE + i] = + vp10_read_literal(r, cm->bit_depth); + } + xd->plane[1].color_index_map[0] = read_uniform(r, n); + assert(xd->plane[1].color_index_map[0] < n); + } + } +} + +#if CONFIG_EXT_INTRA +static void read_ext_intra_mode_info(VP10_COMMON *const cm, + MACROBLOCKD *const xd, vp10_reader *r) { + MODE_INFO *const mi = xd->mi[0]; + MB_MODE_INFO *const mbmi = &mi->mbmi; + FRAME_COUNTS *counts = xd->counts; + +#if !ALLOW_FILTER_INTRA_MODES + return; +#endif + if (mbmi->mode == DC_PRED && mbmi->palette_mode_info.palette_size[0] == 0) { + mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = + vp10_read(r, cm->fc->ext_intra_probs[0]); + if (mbmi->ext_intra_mode_info.use_ext_intra_mode[0]) { + mbmi->ext_intra_mode_info.ext_intra_mode[0] = + read_uniform(r, FILTER_INTRA_MODES); + } + if (counts) + ++counts->ext_intra[0][mbmi->ext_intra_mode_info.use_ext_intra_mode[0]]; + } + if (mbmi->uv_mode == DC_PRED && + mbmi->palette_mode_info.palette_size[1] == 0) { + mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = + vp10_read(r, cm->fc->ext_intra_probs[1]); + if (mbmi->ext_intra_mode_info.use_ext_intra_mode[1]) { + mbmi->ext_intra_mode_info.ext_intra_mode[1] = + read_uniform(r, FILTER_INTRA_MODES); + } + if (counts) + ++counts->ext_intra[1][mbmi->ext_intra_mode_info.use_ext_intra_mode[1]]; + } +} + +static void read_intra_angle_info(VP10_COMMON *const cm, MACROBLOCKD *const xd, + vp10_reader *r) { + MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi; + const BLOCK_SIZE bsize = mbmi->sb_type; + const int ctx = vp10_get_pred_context_intra_interp(xd); + int p_angle; + + if (bsize < BLOCK_8X8) return; + + if (mbmi->mode != DC_PRED && mbmi->mode != TM_PRED) { + mbmi->angle_delta[0] = + read_uniform(r, 2 * MAX_ANGLE_DELTAS + 1) - MAX_ANGLE_DELTAS; + p_angle = mode_to_angle_map[mbmi->mode] + mbmi->angle_delta[0] * ANGLE_STEP; + if (vp10_is_intra_filter_switchable(p_angle)) { + FRAME_COUNTS *counts = xd->counts; + mbmi->intra_filter = vp10_read_tree(r, vp10_intra_filter_tree, + cm->fc->intra_filter_probs[ctx]); + if (counts) ++counts->intra_filter[ctx][mbmi->intra_filter]; + } else { + mbmi->intra_filter = INTRA_FILTER_LINEAR; + } + } + + if (mbmi->uv_mode != DC_PRED && mbmi->uv_mode != TM_PRED) { + mbmi->angle_delta[1] = + read_uniform(r, 2 * MAX_ANGLE_DELTAS + 1) - MAX_ANGLE_DELTAS; + } +} +#endif // CONFIG_EXT_INTRA + +static void read_intra_frame_mode_info(VP10_COMMON *const cm, + MACROBLOCKD *const xd, int mi_row, + int mi_col, vp10_reader *r) { + MODE_INFO *const mi = xd->mi[0]; + MB_MODE_INFO *const mbmi = &mi->mbmi; + const MODE_INFO *above_mi = xd->above_mi; + const MODE_INFO *left_mi = xd->left_mi; + const BLOCK_SIZE bsize = mbmi->sb_type; + int i; + const int mi_offset = mi_row * cm->mi_cols + mi_col; + const int bw = xd->plane[0].n4_w >> 1; + const int bh = xd->plane[0].n4_h >> 1; + + // TODO(slavarnway): move x_mis, y_mis into xd ????? + const int x_mis = VPXMIN(cm->mi_cols - mi_col, bw); + const int y_mis = VPXMIN(cm->mi_rows - mi_row, bh); + + mbmi->segment_id = read_intra_segment_id(cm, xd, mi_offset, x_mis, y_mis, r); + mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r); + mbmi->tx_size = read_tx_size_intra(cm, xd, r); + mbmi->ref_frame[0] = INTRA_FRAME; + mbmi->ref_frame[1] = NONE; + + switch (bsize) { + case BLOCK_4X4: + for (i = 0; i < 4; ++i) + mi->bmi[i].as_mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, i)); + mbmi->mode = mi->bmi[3].as_mode; + break; + case BLOCK_4X8: + mi->bmi[0].as_mode = mi->bmi[2].as_mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, 0)); + mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, 1)); + break; + case BLOCK_8X4: + mi->bmi[0].as_mode = mi->bmi[1].as_mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, 0)); + mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, 2)); + break; + default: + mbmi->mode = + read_intra_mode(r, get_y_mode_probs(cm, mi, above_mi, left_mi, 0)); + } + + mbmi->uv_mode = read_intra_mode_uv(cm, xd, r, mbmi->mode); +#if CONFIG_EXT_INTRA + read_intra_angle_info(cm, xd, r); +#endif // CONFIG_EXT_INTRA + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools) + read_palette_mode_info(cm, xd, r); +#if CONFIG_EXT_INTRA + mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0; + mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0; + if (bsize >= BLOCK_8X8) read_ext_intra_mode_info(cm, xd, r); +#endif // CONFIG_EXT_INTRA + + if (!FIXED_TX_TYPE) { +#if CONFIG_EXT_TX + if (get_ext_tx_types(mbmi->tx_size, mbmi->sb_type, 0) > 1 && + cm->base_qindex > 0 && !mbmi->skip && + !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) && + ALLOW_INTRA_EXT_TX) { + FRAME_COUNTS *counts = xd->counts; + int eset = get_ext_tx_set(mbmi->tx_size, mbmi->sb_type, 0); + if (eset > 0) { + mbmi->tx_type = vp10_read_tree( + r, vp10_ext_tx_intra_tree[eset], + cm->fc->intra_ext_tx_prob[eset][mbmi->tx_size][mbmi->mode]); + if (counts) + ++counts + ->intra_ext_tx[eset][mbmi->tx_size][mbmi->mode][mbmi->tx_type]; + } + } else { + mbmi->tx_type = DCT_DCT; + } +#else + if (mbmi->tx_size < TX_32X32 && cm->base_qindex > 0 && !mbmi->skip && + !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) { + FRAME_COUNTS *counts = xd->counts; + TX_TYPE tx_type_nom = intra_mode_to_tx_type_context[mbmi->mode]; + mbmi->tx_type = + vp10_read_tree(r, vp10_ext_tx_tree, + cm->fc->intra_ext_tx_prob[mbmi->tx_size][tx_type_nom]); + if (counts) + ++counts->intra_ext_tx[mbmi->tx_size][tx_type_nom][mbmi->tx_type]; + } else { + mbmi->tx_type = DCT_DCT; + } +#endif // CONFIG_EXT_TX + } +} + +static int read_mv_component(vp10_reader *r, const nmv_component *mvcomp, + int usehp) { + int mag, d, fr, hp; + const int sign = vp10_read(r, mvcomp->sign); + const int mv_class = vp10_read_tree(r, vp10_mv_class_tree, mvcomp->classes); + const int class0 = mv_class == MV_CLASS_0; + + // Integer part + if (class0) { + d = vp10_read_tree(r, vp10_mv_class0_tree, mvcomp->class0); + mag = 0; + } else { + int i; + const int n = mv_class + CLASS0_BITS - 1; // number of bits + + d = 0; + for (i = 0; i < n; ++i) d |= vp10_read(r, mvcomp->bits[i]) << i; + mag = CLASS0_SIZE << (mv_class + 2); + } + + // Fractional part + fr = vp10_read_tree(r, vp10_mv_fp_tree, + class0 ? mvcomp->class0_fp[d] : mvcomp->fp); + + // High precision part (if hp is not used, the default value of the hp is 1) + hp = usehp ? vp10_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp) : 1; + + // Result + mag += ((d << 3) | (fr << 1) | hp) + 1; + return sign ? -mag : mag; +} + +static INLINE void read_mv(vp10_reader *r, MV *mv, const MV *ref, +#if CONFIG_REF_MV + int is_compound, +#endif + const nmv_context *ctx, nmv_context_counts *counts, + int allow_hp) { + MV_JOINT_TYPE joint_type; + const int use_hp = allow_hp && vp10_use_mv_hp(ref); + MV diff = { 0, 0 }; + +#if CONFIG_REF_MV && !CONFIG_EXT_INTER + if (is_compound) { + int is_zero_rmv = vp10_read(r, ctx->zero_rmv); + if (is_zero_rmv) { + joint_type = MV_JOINT_ZERO; + } else { + joint_type = + (MV_JOINT_TYPE)vp10_read_tree(r, vp10_mv_joint_tree, ctx->joints); + } + } else { + joint_type = + (MV_JOINT_TYPE)vp10_read_tree(r, vp10_mv_joint_tree, ctx->joints); + } +#else + joint_type = + (MV_JOINT_TYPE)vp10_read_tree(r, vp10_mv_joint_tree, ctx->joints); +#endif + +#if CONFIG_REF_MV && CONFIG_EXT_INTER + (void)is_compound; +#endif + + if (mv_joint_vertical(joint_type)) + diff.row = read_mv_component(r, &ctx->comps[0], use_hp); + + if (mv_joint_horizontal(joint_type)) + diff.col = read_mv_component(r, &ctx->comps[1], use_hp); + + vp10_inc_mv(&diff, counts, use_hp); + + mv->row = ref->row + diff.row; + mv->col = ref->col + diff.col; +} + +static REFERENCE_MODE read_block_reference_mode(VP10_COMMON *cm, + const MACROBLOCKD *xd, + vp10_reader *r) { + if (cm->reference_mode == REFERENCE_MODE_SELECT) { + const int ctx = vp10_get_reference_mode_context(cm, xd); + const REFERENCE_MODE mode = + (REFERENCE_MODE)vp10_read(r, cm->fc->comp_inter_prob[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->comp_inter[ctx][mode]; + return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE + } else { + return cm->reference_mode; + } +} + +// Read the referncence frame +static void read_ref_frames(VP10_COMMON *const cm, MACROBLOCKD *const xd, + vp10_reader *r, int segment_id, + MV_REFERENCE_FRAME ref_frame[2]) { + FRAME_CONTEXT *const fc = cm->fc; + FRAME_COUNTS *counts = xd->counts; + + 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; + } else { + const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r); + // FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding + if (mode == COMPOUND_REFERENCE) { +#if CONFIG_EXT_REFS + const int idx = cm->ref_frame_sign_bias[cm->comp_bwd_ref[0]]; +#else + const int idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref]; +#endif // CONFIG_EXT_REFS + const int ctx = vp10_get_pred_context_comp_ref_p(cm, xd); + const int bit = vp10_read(r, fc->comp_ref_prob[ctx][0]); + + if (counts) ++counts->comp_ref[ctx][0][bit]; + +#if CONFIG_EXT_REFS + // Decode forward references. + if (!bit) { + const int ctx1 = vp10_get_pred_context_comp_ref_p1(cm, xd); + const int bit1 = vp10_read(r, fc->comp_ref_prob[ctx1][1]); + if (counts) ++counts->comp_ref[ctx1][1][bit1]; + ref_frame[!idx] = cm->comp_fwd_ref[bit1 ? 0 : 1]; + } else { + const int ctx2 = vp10_get_pred_context_comp_ref_p2(cm, xd); + const int bit2 = vp10_read(r, fc->comp_ref_prob[ctx2][2]); + if (counts) ++counts->comp_ref[ctx2][2][bit2]; + ref_frame[!idx] = cm->comp_fwd_ref[bit2 ? 3 : 2]; + } + + // Decode backward references. + { + const int ctx_bwd = vp10_get_pred_context_comp_bwdref_p(cm, xd); + const int bit_bwd = vp10_read(r, fc->comp_bwdref_prob[ctx_bwd][0]); + if (counts) ++counts->comp_bwdref[ctx_bwd][0][bit_bwd]; + ref_frame[idx] = cm->comp_bwd_ref[bit_bwd]; + } +#else + ref_frame[!idx] = cm->comp_var_ref[bit]; + ref_frame[idx] = cm->comp_fixed_ref; +#endif // CONFIG_EXT_REFS + } else if (mode == SINGLE_REFERENCE) { +#if CONFIG_EXT_REFS + const int ctx0 = vp10_get_pred_context_single_ref_p1(xd); + const int bit0 = vp10_read(r, fc->single_ref_prob[ctx0][0]); + if (counts) ++counts->single_ref[ctx0][0][bit0]; + + if (bit0) { + const int ctx1 = vp10_get_pred_context_single_ref_p2(xd); + const int bit1 = vp10_read(r, fc->single_ref_prob[ctx1][1]); + if (counts) ++counts->single_ref[ctx1][1][bit1]; + ref_frame[0] = bit1 ? ALTREF_FRAME : BWDREF_FRAME; + } else { + const int ctx2 = vp10_get_pred_context_single_ref_p3(xd); + const int bit2 = vp10_read(r, fc->single_ref_prob[ctx2][2]); + if (counts) ++counts->single_ref[ctx2][2][bit2]; + if (bit2) { + const int ctx4 = vp10_get_pred_context_single_ref_p5(xd); + const int bit4 = vp10_read(r, fc->single_ref_prob[ctx4][4]); + if (counts) ++counts->single_ref[ctx4][4][bit4]; + ref_frame[0] = bit4 ? GOLDEN_FRAME : LAST3_FRAME; + } else { + const int ctx3 = vp10_get_pred_context_single_ref_p4(xd); + const int bit3 = vp10_read(r, fc->single_ref_prob[ctx3][3]); + if (counts) ++counts->single_ref[ctx3][3][bit3]; + ref_frame[0] = bit3 ? LAST2_FRAME : LAST_FRAME; + } + } +#else + const int ctx0 = vp10_get_pred_context_single_ref_p1(xd); + const int bit0 = vp10_read(r, fc->single_ref_prob[ctx0][0]); + if (counts) ++counts->single_ref[ctx0][0][bit0]; + + if (bit0) { + const int ctx1 = vp10_get_pred_context_single_ref_p2(xd); + const int bit1 = vp10_read(r, fc->single_ref_prob[ctx1][1]); + if (counts) ++counts->single_ref[ctx1][1][bit1]; + ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME; + } else { + ref_frame[0] = LAST_FRAME; + } +#endif // CONFIG_EXT_REFS + + ref_frame[1] = NONE; + } else { + assert(0 && "Invalid prediction mode."); + } + } +} + +#if CONFIG_OBMC || CONFIG_WARPED_MOTION +static MOTION_VARIATION read_motvar_block(VP10_COMMON *const cm, + MACROBLOCKD *const xd, + vp10_reader *r) { + BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type; + FRAME_COUNTS *counts = xd->counts; + MOTION_VARIATION motvar; + + if (is_motvar_allowed(&xd->mi[0]->mbmi)) { + motvar = (MOTION_VARIATION)vp10_read_tree(r, vp10_motvar_tree, + cm->fc->motvar_prob[bsize]); + if (counts) ++counts->motvar[bsize][motvar]; + return motvar; + } else { + return SIMPLE_TRANSLATION; + } +} +#endif // CONFIG_OBMC || CONFIG_WARPED_MOTION + +static INLINE INTERP_FILTER read_interp_filter(VP10_COMMON *const cm, + MACROBLOCKD *const xd, +#if CONFIG_DUAL_FILTER + int dir, +#endif + vp10_reader *r) { +#if CONFIG_EXT_INTERP + if (!vp10_is_interp_needed(xd)) return EIGHTTAP_REGULAR; +#endif + if (cm->interp_filter != SWITCHABLE) { + return cm->interp_filter; + } else { +#if CONFIG_DUAL_FILTER + const int ctx = vp10_get_pred_context_switchable_interp(xd, dir); +#else + const int ctx = vp10_get_pred_context_switchable_interp(xd); +#endif + FRAME_COUNTS *counts = xd->counts; + const INTERP_FILTER type = (INTERP_FILTER)vp10_read_tree( + r, vp10_switchable_interp_tree, cm->fc->switchable_interp_prob[ctx]); + if (counts) ++counts->switchable_interp[ctx][type]; + return type; + } +} + +static void read_intra_block_mode_info(VP10_COMMON *const cm, + MACROBLOCKD *const xd, MODE_INFO *mi, + vp10_reader *r) { + MB_MODE_INFO *const mbmi = &mi->mbmi; + const BLOCK_SIZE bsize = mi->mbmi.sb_type; + int i; + + mbmi->ref_frame[0] = INTRA_FRAME; + mbmi->ref_frame[1] = NONE; + + switch (bsize) { + case BLOCK_4X4: + for (i = 0; i < 4; ++i) + mi->bmi[i].as_mode = read_intra_mode_y(cm, xd, r, 0); + mbmi->mode = mi->bmi[3].as_mode; + break; + case BLOCK_4X8: + mi->bmi[0].as_mode = mi->bmi[2].as_mode = read_intra_mode_y(cm, xd, r, 0); + mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode = + read_intra_mode_y(cm, xd, r, 0); + break; + case BLOCK_8X4: + mi->bmi[0].as_mode = mi->bmi[1].as_mode = read_intra_mode_y(cm, xd, r, 0); + mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode = + read_intra_mode_y(cm, xd, r, 0); + break; + default: + mbmi->mode = read_intra_mode_y(cm, xd, r, size_group_lookup[bsize]); + } + + mbmi->uv_mode = read_intra_mode_uv(cm, xd, r, mbmi->mode); +#if CONFIG_EXT_INTRA + read_intra_angle_info(cm, xd, r); +#endif // CONFIG_EXT_INTRA + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools) + read_palette_mode_info(cm, xd, r); +#if CONFIG_EXT_INTRA + mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0; + mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0; + if (bsize >= BLOCK_8X8) read_ext_intra_mode_info(cm, xd, r); +#endif // CONFIG_EXT_INTRA +} + +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(VP10_COMMON *cm, MACROBLOCKD *xd, + PREDICTION_MODE mode, +#if CONFIG_REF_MV + int block, +#endif + int_mv mv[2], int_mv ref_mv[2], + int_mv nearest_mv[2], int_mv near_mv[2], + int is_compound, int allow_hp, vp10_reader *r) { + int i; + int ret = 1; +#if CONFIG_REF_MV + MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi; + BLOCK_SIZE bsize = mbmi->sb_type; + int_mv *pred_mv = + (bsize >= BLOCK_8X8) ? mbmi->pred_mv : xd->mi[0]->bmi[block].pred_mv_s8; +#endif + + switch (mode) { +#if CONFIG_EXT_INTER + case NEWFROMNEARMV: +#endif // CONFIG_EXT_INTER + case NEWMV: { + FRAME_COUNTS *counts = xd->counts; +#if !CONFIG_REF_MV + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; +#endif + for (i = 0; i < 1 + is_compound; ++i) { +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[i]], + xd->ref_mv_stack[mbmi->ref_frame[i]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, +#if CONFIG_REF_MV + is_compound, +#endif + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + ret = ret && is_mv_valid(&mv[i].as_mv); + +#if CONFIG_REF_MV + pred_mv[i].as_int = ref_mv[i].as_int; +#endif + } + break; + } + case NEARESTMV: { + mv[0].as_int = nearest_mv[0].as_int; + if (is_compound) mv[1].as_int = nearest_mv[1].as_int; + +#if CONFIG_REF_MV + pred_mv[0].as_int = nearest_mv[0].as_int; + if (is_compound) pred_mv[1].as_int = nearest_mv[1].as_int; +#endif + break; + } + case NEARMV: { + mv[0].as_int = near_mv[0].as_int; + if (is_compound) mv[1].as_int = near_mv[1].as_int; + +#if CONFIG_REF_MV + pred_mv[0].as_int = near_mv[0].as_int; + if (is_compound) pred_mv[1].as_int = near_mv[1].as_int; +#endif + break; + } + case ZEROMV: { + mv[0].as_int = 0; + if (is_compound) mv[1].as_int = 0; + +#if CONFIG_REF_MV + pred_mv[0].as_int = 0; + if (is_compound) pred_mv[1].as_int = 0; +#endif + break; + } +#if CONFIG_EXT_INTER + case NEW_NEWMV: { + FRAME_COUNTS *counts = xd->counts; +#if !CONFIG_REF_MV + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; +#endif + assert(is_compound); + for (i = 0; i < 2; ++i) { +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[i]], + xd->ref_mv_stack[mbmi->ref_frame[i]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, is_compound, + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + ret = ret && is_mv_valid(&mv[i].as_mv); + } + 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 NEAREST_NEARMV: { + assert(is_compound); + mv[0].as_int = nearest_mv[0].as_int; + mv[1].as_int = near_mv[1].as_int; + break; + } + case NEAR_NEARESTMV: { + assert(is_compound); + mv[0].as_int = near_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: { + FRAME_COUNTS *counts = xd->counts; +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[0]], + xd->ref_mv_stack[mbmi->ref_frame[0]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, is_compound, + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + assert(is_compound); + ret = ret && is_mv_valid(&mv[0].as_mv); + mv[1].as_int = nearest_mv[1].as_int; + break; + } + case NEAREST_NEWMV: { + FRAME_COUNTS *counts = xd->counts; +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[1]], + xd->ref_mv_stack[mbmi->ref_frame[1]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + mv[0].as_int = nearest_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, is_compound, + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; + mv[0].as_int = nearest_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + assert(is_compound); + ret = ret && is_mv_valid(&mv[1].as_mv); + break; + } + case NEAR_NEWMV: { + FRAME_COUNTS *counts = xd->counts; +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[1]], + xd->ref_mv_stack[mbmi->ref_frame[1]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + mv[0].as_int = near_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, is_compound, + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; + mv[0].as_int = near_mv[0].as_int; + read_mv(r, &mv[1].as_mv, &ref_mv[1].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + assert(is_compound); + + ret = ret && is_mv_valid(&mv[1].as_mv); + break; + } + case NEW_NEARMV: { + FRAME_COUNTS *counts = xd->counts; +#if CONFIG_REF_MV + int nmv_ctx = vp10_nmv_ctx(xd->ref_mv_count[mbmi->ref_frame[0]], + xd->ref_mv_stack[mbmi->ref_frame[0]]); + nmv_context_counts *const mv_counts = + counts ? &counts->mv[nmv_ctx] : NULL; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, is_compound, + &cm->fc->nmvc[nmv_ctx], mv_counts, allow_hp); +#else + nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL; + read_mv(r, &mv[0].as_mv, &ref_mv[0].as_mv, &cm->fc->nmvc, mv_counts, + allow_hp); +#endif + assert(is_compound); + ret = ret && is_mv_valid(&mv[0].as_mv); + mv[1].as_int = near_mv[1].as_int; + break; + } + case ZERO_ZEROMV: { + assert(is_compound); + mv[0].as_int = 0; + mv[1].as_int = 0; + break; + } +#endif // CONFIG_EXT_INTER + default: { return 0; } + } + return ret; +} + +static int read_is_inter_block(VP10_COMMON *const cm, MACROBLOCKD *const xd, + int segment_id, vp10_reader *r) { + if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) { + return get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME; + } else { + const int ctx = vp10_get_intra_inter_context(xd); + const int is_inter = vp10_read(r, cm->fc->intra_inter_prob[ctx]); + FRAME_COUNTS *counts = xd->counts; + if (counts) ++counts->intra_inter[ctx][is_inter]; + return is_inter; + } +} + +static void fpm_sync(void *const data, int mi_row) { + VP10Decoder *const pbi = (VP10Decoder *)data; + vp10_frameworker_wait(pbi->frame_worker_owner, pbi->common.prev_frame, + mi_row << pbi->common.mib_size_log2); +} + +static void read_inter_block_mode_info(VP10Decoder *const pbi, + MACROBLOCKD *const xd, + MODE_INFO *const mi, +#if (CONFIG_OBMC || CONFIG_EXT_INTER) && CONFIG_SUPERTX + int mi_row, int mi_col, vp10_reader *r, + int supertx_enabled) { +#else + int mi_row, int mi_col, vp10_reader *r) { +#endif // CONFIG_OBMC && CONFIG_SUPERTX + VP10_COMMON *const cm = &pbi->common; + MB_MODE_INFO *const mbmi = &mi->mbmi; + 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]; +#if CONFIG_EXT_INTER + int mv_idx; +#endif // CONFIG_EXT_INTER + int ref, is_compound; + int16_t inter_mode_ctx[MODE_CTX_REF_FRAMES]; +#if CONFIG_REF_MV && CONFIG_EXT_INTER + int16_t compound_inter_mode_ctx[MODE_CTX_REF_FRAMES]; +#endif // CONFIG_REF_MV && CONFIG_EXT_INTER + int16_t mode_ctx = 0; + MV_REFERENCE_FRAME ref_frame; + + mbmi->palette_mode_info.palette_size[0] = 0; + mbmi->palette_mode_info.palette_size[1] = 0; + + read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame); + is_compound = has_second_ref(mbmi); + + for (ref = 0; ref < 1 + is_compound; ++ref) { + MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref]; + RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME]; + + xd->block_refs[ref] = ref_buf; + if ((!vp10_is_valid_scale(&ref_buf->sf))) + vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM, + "Reference frame has invalid dimensions"); + vp10_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col, &ref_buf->sf); + } + + for (ref_frame = LAST_FRAME; ref_frame < MODE_CTX_REF_FRAMES; ++ref_frame) { + vp10_find_mv_refs(cm, xd, mi, ref_frame, +#if CONFIG_REF_MV + &xd->ref_mv_count[ref_frame], xd->ref_mv_stack[ref_frame], +#if CONFIG_EXT_INTER + compound_inter_mode_ctx, +#endif // CONFIG_EXT_INTER +#endif + ref_mvs[ref_frame], mi_row, mi_col, fpm_sync, (void *)pbi, + inter_mode_ctx); + } + +#if CONFIG_REF_MV +#if CONFIG_EXT_INTER + if (is_compound) + mode_ctx = compound_inter_mode_ctx[mbmi->ref_frame[0]]; + else +#endif // CONFIG_EXT_INTER + mode_ctx = + vp10_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame, bsize, -1); + mbmi->ref_mv_idx = 0; +#else + mode_ctx = inter_mode_ctx[mbmi->ref_frame[0]]; +#endif + + if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) { + mbmi->mode = ZEROMV; + if (bsize < BLOCK_8X8) { + vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM, + "Invalid usage of segement feature on small blocks"); + return; + } + } else { + if (bsize >= BLOCK_8X8) { +#if CONFIG_EXT_INTER + if (is_compound) + mbmi->mode = read_inter_compound_mode(cm, xd, r, mode_ctx); + else +#endif // CONFIG_EXT_INTER + mbmi->mode = read_inter_mode(cm, xd, +#if CONFIG_REF_MV && CONFIG_EXT_INTER + mbmi, +#endif // CONFIG_REF_MV && CONFIG_EXT_INTER + r, mode_ctx); +#if CONFIG_REF_MV + if (mbmi->mode == NEARMV || mbmi->mode == NEWMV) + read_drl_idx(cm, xd, mbmi, r); +#endif + } + } + +#if CONFIG_EXT_INTER + if (bsize < BLOCK_8X8 || + (mbmi->mode != ZEROMV && mbmi->mode != ZERO_ZEROMV)) { +#else + if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) { +#endif // CONFIG_EXT_INTER + for (ref = 0; ref < 1 + is_compound; ++ref) { + vp10_find_best_ref_mvs(allow_hp, ref_mvs[mbmi->ref_frame[ref]], + &nearestmv[ref], &nearmv[ref]); + } + } + +#if CONFIG_REF_MV + if (mbmi->ref_mv_idx > 0) { + int_mv cur_mv = + xd->ref_mv_stack[mbmi->ref_frame[0]][1 + mbmi->ref_mv_idx].this_mv; + nearmv[0] = cur_mv; + } + +#if CONFIG_EXT_INTER + if (is_compound && bsize >= BLOCK_8X8 && mbmi->mode != ZERO_ZEROMV) { +#else + if (is_compound && bsize >= BLOCK_8X8 && mbmi->mode != NEWMV && + mbmi->mode != ZEROMV) { +#endif // CONFIG_EXT_INTER + uint8_t ref_frame_type = vp10_ref_frame_type(mbmi->ref_frame); + +#if CONFIG_EXT_INTER + if (xd->ref_mv_count[ref_frame_type] > 0) { +#else + if (xd->ref_mv_count[ref_frame_type] == 1 && mbmi->mode == NEARESTMV) { +#endif // CONFIG_EXT_INTER +#if CONFIG_EXT_INTER + if (mbmi->mode == NEAREST_NEARESTMV) { +#endif // CONFIG_EXT_INTER + nearestmv[0] = xd->ref_mv_stack[ref_frame_type][0].this_mv; + nearestmv[1] = xd->ref_mv_stack[ref_frame_type][0].comp_mv; + lower_mv_precision(&nearestmv[0].as_mv, allow_hp); + lower_mv_precision(&nearestmv[1].as_mv, allow_hp); +#if CONFIG_EXT_INTER + } else if (mbmi->mode == NEAREST_NEWMV || mbmi->mode == NEAREST_NEARMV) { + nearestmv[0] = xd->ref_mv_stack[ref_frame_type][0].this_mv; + lower_mv_precision(&nearestmv[0].as_mv, allow_hp); + } else if (mbmi->mode == NEW_NEARESTMV || mbmi->mode == NEAR_NEARESTMV) { + nearestmv[1] = xd->ref_mv_stack[ref_frame_type][0].comp_mv; + lower_mv_precision(&nearestmv[1].as_mv, allow_hp); + } +#endif // CONFIG_EXT_INTER + } + +#if CONFIG_EXT_INTER + if (xd->ref_mv_count[ref_frame_type] > 1) { + if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEAR_NEARESTMV || + mbmi->mode == NEAR_NEARMV) { + nearmv[0] = xd->ref_mv_stack[ref_frame_type][1].this_mv; + lower_mv_precision(&nearmv[0].as_mv, allow_hp); + } + + if (mbmi->mode == NEW_NEARMV || mbmi->mode == NEAREST_NEARMV || + mbmi->mode == NEAR_NEARMV) { + nearmv[1] = xd->ref_mv_stack[ref_frame_type][1].comp_mv; + lower_mv_precision(&nearmv[1].as_mv, allow_hp); + } + } +#else + if (xd->ref_mv_count[ref_frame_type] > 1) { + int ref_mv_idx = 1 + mbmi->ref_mv_idx; + nearestmv[0] = xd->ref_mv_stack[ref_frame_type][0].this_mv; + nearestmv[1] = xd->ref_mv_stack[ref_frame_type][0].comp_mv; + nearmv[0] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].this_mv; + nearmv[1] = xd->ref_mv_stack[ref_frame_type][ref_mv_idx].comp_mv; + } +#endif // CONFIG_EXT_INTER + } +#endif + +#if !CONFIG_EXT_INTERP && !CONFIG_DUAL_FILTER + mbmi->interp_filter = read_interp_filter(cm, xd, r); +#endif // !CONFIG_EXT_INTERP && !CONFIG_DUAL_FILTER + + if (bsize < BLOCK_8X8) { + const int num_4x4_w = 1 << xd->bmode_blocks_wl; + const int num_4x4_h = 1 << xd->bmode_blocks_hl; + int idx, idy; + PREDICTION_MODE b_mode; + int_mv nearest_sub8x8[2], near_sub8x8[2]; +#if CONFIG_EXT_INTER + int_mv ref_mv[2][2]; +#endif // CONFIG_EXT_INTER + for (idy = 0; idy < 2; idy += num_4x4_h) { + for (idx = 0; idx < 2; idx += num_4x4_w) { + int_mv block[2]; + const int j = idy * 2 + idx; + int_mv ref_mv_s8[2]; +#if CONFIG_REF_MV +#if CONFIG_EXT_INTER + if (!is_compound) +#endif // CONFIG_EXT_INTER + mode_ctx = vp10_mode_context_analyzer(inter_mode_ctx, mbmi->ref_frame, + bsize, j); +#endif +#if CONFIG_EXT_INTER + if (is_compound) + b_mode = read_inter_compound_mode(cm, xd, r, mode_ctx); + else +#endif // CONFIG_EXT_INTER + b_mode = read_inter_mode(cm, xd, +#if CONFIG_REF_MV && CONFIG_EXT_INTER + mbmi, +#endif // CONFIG_REF_MV && CONFIG_EXT_INTER + r, mode_ctx); + +#if CONFIG_EXT_INTER + mv_idx = (b_mode == NEWFROMNEARMV) ? 1 : 0; + + if (b_mode != ZEROMV && b_mode != ZERO_ZEROMV) { +#else + if (b_mode != ZEROMV) { +#endif // CONFIG_EXT_INTER +#if CONFIG_REF_MV + CANDIDATE_MV ref_mv_stack[2][MAX_REF_MV_STACK_SIZE]; + uint8_t ref_mv_count[2]; +#endif + for (ref = 0; ref < 1 + is_compound; ++ref) +#if CONFIG_EXT_INTER + { + int_mv mv_ref_list[MAX_MV_REF_CANDIDATES]; + vp10_update_mv_context(xd, mi, mbmi->ref_frame[ref], mv_ref_list, j, + mi_row, mi_col, NULL); +#endif // CONFIG_EXT_INTER + vp10_append_sub8x8_mvs_for_idx( + cm, xd, j, ref, mi_row, mi_col, +#if CONFIG_REF_MV + ref_mv_stack[ref], &ref_mv_count[ref], +#endif +#if CONFIG_EXT_INTER + mv_ref_list, +#endif // CONFIG_EXT_INTER + &nearest_sub8x8[ref], &near_sub8x8[ref]); +#if CONFIG_EXT_INTER + if (have_newmv_in_inter_mode(b_mode)) { + mv_ref_list[0].as_int = nearest_sub8x8[ref].as_int; + mv_ref_list[1].as_int = near_sub8x8[ref].as_int; + vp10_find_best_ref_mvs(allow_hp, mv_ref_list, &ref_mv[0][ref], + &ref_mv[1][ref]); + } + } +#endif // CONFIG_EXT_INTER + } + + for (ref = 0; ref < 1 + is_compound && b_mode != ZEROMV; ++ref) { +#if CONFIG_REF_MV + ref_mv_s8[ref] = nearest_sub8x8[ref]; + lower_mv_precision(&ref_mv_s8[ref].as_mv, allow_hp); +#else + ref_mv_s8[ref] = nearestmv[ref]; +#endif + } +#if CONFIG_EXT_INTER + (void)ref_mv_s8; +#endif + + if (!assign_mv(cm, xd, b_mode, +#if CONFIG_REF_MV + j, +#endif + block, +#if CONFIG_EXT_INTER + ref_mv[mv_idx], +#else + ref_mv_s8, +#endif // CONFIG_EXT_INTER + nearest_sub8x8, near_sub8x8, is_compound, allow_hp, r)) { + xd->corrupted |= 1; + break; + }; + + mi->bmi[j].as_mv[0].as_int = block[0].as_int; + if (is_compound) mi->bmi[j].as_mv[1].as_int = block[1].as_int; + + if (num_4x4_h == 2) mi->bmi[j + 2] = mi->bmi[j]; + if (num_4x4_w == 2) mi->bmi[j + 1] = mi->bmi[j]; + } + } + +#if CONFIG_REF_MV + mbmi->pred_mv[0].as_int = mi->bmi[3].pred_mv_s8[0].as_int; + mbmi->pred_mv[1].as_int = mi->bmi[3].pred_mv_s8[1].as_int; +#endif + mi->mbmi.mode = b_mode; + + mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int; + mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int; + } else { + int ref; + int_mv ref_mv[2]; + ref_mv[0] = nearestmv[0]; + ref_mv[1] = nearestmv[1]; + + for (ref = 0; ref < 1 + is_compound && mbmi->mode == NEWMV; ++ref) { +#if CONFIG_REF_MV + uint8_t ref_frame_type = vp10_ref_frame_type(mbmi->ref_frame); + if (xd->ref_mv_count[ref_frame_type] > 1) { + ref_mv[ref] = + (ref == 0) + ? xd->ref_mv_stack[ref_frame_type][mbmi->ref_mv_idx].this_mv + : xd->ref_mv_stack[ref_frame_type][mbmi->ref_mv_idx].comp_mv; + clamp_mv_ref(&ref_mv[ref].as_mv, xd->n8_w << 3, xd->n8_h << 3, xd); + } +#endif + nearestmv[ref] = ref_mv[ref]; + } + + xd->corrupted |= + !assign_mv(cm, xd, mbmi->mode, +#if CONFIG_REF_MV + 0, +#endif + mbmi->mv, +#if CONFIG_EXT_INTER + mbmi->mode == NEWFROMNEARMV ? nearmv : nearestmv, +#else + ref_mv, +#endif // CONFIG_EXT_INTER + nearestmv, nearmv, is_compound, allow_hp, r); + } + +#if CONFIG_EXT_INTER + mbmi->use_wedge_interintra = 0; + if (cm->reference_mode != COMPOUND_REFERENCE && +#if CONFIG_SUPERTX + !supertx_enabled && +#endif + is_interintra_allowed(mbmi)) { + const int bsize_group = size_group_lookup[bsize]; + const int interintra = vp10_read(r, cm->fc->interintra_prob[bsize_group]); + if (xd->counts) xd->counts->interintra[bsize_group][interintra]++; + assert(mbmi->ref_frame[1] == NONE); + if (interintra) { + const INTERINTRA_MODE interintra_mode = + read_interintra_mode(cm, xd, r, bsize_group); + mbmi->ref_frame[1] = INTRA_FRAME; + mbmi->interintra_mode = interintra_mode; +#if CONFIG_EXT_INTRA + mbmi->ext_intra_mode_info.use_ext_intra_mode[0] = 0; + mbmi->ext_intra_mode_info.use_ext_intra_mode[1] = 0; + mbmi->angle_delta[0] = 0; + mbmi->angle_delta[1] = 0; + mbmi->intra_filter = INTRA_FILTER_LINEAR; +#endif // CONFIG_EXT_INTRA + if (is_interintra_wedge_used(bsize)) { + mbmi->use_wedge_interintra = + vp10_read(r, cm->fc->wedge_interintra_prob[bsize]); + if (xd->counts) + xd->counts->wedge_interintra[bsize][mbmi->use_wedge_interintra]++; + if (mbmi->use_wedge_interintra) { + mbmi->interintra_wedge_index = + vp10_read_literal(r, get_wedge_bits_lookup(bsize)); + mbmi->interintra_wedge_sign = 0; + } + } + } + } +#endif // CONFIG_EXT_INTER + +#if CONFIG_OBMC || CONFIG_WARPED_MOTION + mbmi->motion_variation = SIMPLE_TRANSLATION; +#if CONFIG_SUPERTX + if (!supertx_enabled) +#endif // CONFIG_SUPERTX +#if CONFIG_EXT_INTER + if (mbmi->ref_frame[1] != INTRA_FRAME) +#endif // CONFIG_EXT_INTER + mbmi->motion_variation = read_motvar_block(cm, xd, r); +#endif // CONFIG_OBMC || CONFIG_WARPED_MOTION + +#if CONFIG_EXT_INTER + mbmi->use_wedge_interinter = 0; + if (cm->reference_mode != SINGLE_REFERENCE && + is_inter_compound_mode(mbmi->mode) && +#if CONFIG_OBMC || CONFIG_WARPED_MOTION + !(is_motvar_allowed(mbmi) && + mbmi->motion_variation != SIMPLE_TRANSLATION) && +#endif // CONFIG_OBMC || CONFIG_WARPED_MOTION + is_interinter_wedge_used(bsize)) { + mbmi->use_wedge_interinter = + vp10_read(r, cm->fc->wedge_interinter_prob[bsize]); + if (xd->counts) + xd->counts->wedge_interinter[bsize][mbmi->use_wedge_interinter]++; + if (mbmi->use_wedge_interinter) { + mbmi->interinter_wedge_index = + vp10_read_literal(r, get_wedge_bits_lookup(bsize)); + mbmi->interinter_wedge_sign = vp10_read_bit(r); + } + } +#endif // CONFIG_EXT_INTER + +#if CONFIG_DUAL_FILTER + for (ref = 0; ref < 2; ++ref) { + mbmi->interp_filter[ref] = (cm->interp_filter == SWITCHABLE) + ? EIGHTTAP_REGULAR + : cm->interp_filter; + + if (has_subpel_mv_component(xd->mi[0], xd, ref) || + (mbmi->ref_frame[1] > INTRA_FRAME && + has_subpel_mv_component(xd->mi[0], xd, ref + 2))) + mbmi->interp_filter[ref] = read_interp_filter(cm, xd, ref, r); + } + // The index system worsk as: + // (0, 1) -> (vertical, horizontal) filter types for the first ref frame. + // (2, 3) -> (vertical, horizontal) filter types for the second ref frame. + mbmi->interp_filter[2] = mbmi->interp_filter[0]; + mbmi->interp_filter[3] = mbmi->interp_filter[1]; +#else +#if CONFIG_EXT_INTERP + mbmi->interp_filter = read_interp_filter(cm, xd, r); +#endif // CONFIG_EXT_INTERP +#endif // CONFIG_DUAL_FILTER +} + +static void read_inter_frame_mode_info(VP10Decoder *const pbi, + MACROBLOCKD *const xd, +#if CONFIG_SUPERTX + int supertx_enabled, +#endif // CONFIG_SUPERTX + int mi_row, int mi_col, vp10_reader *r) { + VP10_COMMON *const cm = &pbi->common; + MODE_INFO *const mi = xd->mi[0]; + MB_MODE_INFO *const mbmi = &mi->mbmi; + int inter_block = 1; +#if CONFIG_VAR_TX + BLOCK_SIZE bsize = mbmi->sb_type; +#endif // CONFIG_VAR_TX + + 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, r); +#if CONFIG_SUPERTX + if (!supertx_enabled) { +#endif // CONFIG_SUPERTX + mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r); + inter_block = read_is_inter_block(cm, xd, mbmi->segment_id, r); + +#if CONFIG_VAR_TX + xd->above_txfm_context = cm->above_txfm_context + mi_col; + xd->left_txfm_context = + xd->left_txfm_context_buffer + (mi_row & MAX_MIB_MASK); + if (bsize >= BLOCK_8X8 && cm->tx_mode == TX_MODE_SELECT && !mbmi->skip && + inter_block) { + const TX_SIZE max_tx_size = max_txsize_lookup[bsize]; + const BLOCK_SIZE txb_size = txsize_to_bsize[max_tx_size]; + const int bs = num_4x4_blocks_wide_lookup[txb_size]; + const int width = num_4x4_blocks_wide_lookup[bsize]; + const int height = num_4x4_blocks_high_lookup[bsize]; + int idx, idy; + for (idy = 0; idy < height; idy += bs) + for (idx = 0; idx < width; idx += bs) + read_tx_size_vartx(cm, xd, mbmi, xd->counts, max_tx_size, idy, idx, + r); + if (xd->counts) { + const int ctx = get_tx_size_context(xd); + ++xd->counts->tx_size[max_tx_size - TX_8X8][ctx][mbmi->tx_size]; + } + } else { + if (inter_block) + mbmi->tx_size = read_tx_size_inter(cm, xd, !mbmi->skip, r); + else + mbmi->tx_size = read_tx_size_intra(cm, xd, r); + + if (inter_block) { + const int width = num_4x4_blocks_wide_lookup[bsize]; + const int height = num_4x4_blocks_high_lookup[bsize]; + int idx, idy; + for (idy = 0; idy < height; ++idy) + for (idx = 0; idx < width; ++idx) + mbmi->inter_tx_size[idy >> 1][idx >> 1] = mbmi->tx_size; + } + + set_txfm_ctxs(mbmi->tx_size, xd->n8_w, xd->n8_h, xd); + } +#else + if (inter_block) + mbmi->tx_size = read_tx_size_inter(cm, xd, !mbmi->skip, r); + else + mbmi->tx_size = read_tx_size_intra(cm, xd, r); +#endif // CONFIG_VAR_TX +#if CONFIG_SUPERTX + } +#if CONFIG_VAR_TX + else if (inter_block) { + const int width = num_4x4_blocks_wide_lookup[bsize]; + const int height = num_4x4_blocks_high_lookup[bsize]; + int idx, idy; + xd->mi[0]->mbmi.tx_size = xd->supertx_size; + for (idy = 0; idy < height; ++idy) + for (idx = 0; idx < width; ++idx) + xd->mi[0]->mbmi.inter_tx_size[idy >> 1][idx >> 1] = xd->supertx_size; + } +#endif // CONFIG_VAR_TX +#endif // CONFIG_SUPERTX + + if (inter_block) + read_inter_block_mode_info(pbi, xd, +#if (CONFIG_OBMC || CONFIG_EXT_INTER) && CONFIG_SUPERTX + + mi, mi_row, mi_col, r, supertx_enabled); +#else + mi, mi_row, mi_col, r); +#endif // CONFIG_OBMC && CONFIG_SUPERTX + else + read_intra_block_mode_info(cm, xd, mi, r); + + if (!FIXED_TX_TYPE) { +#if CONFIG_EXT_TX + if (get_ext_tx_types(mbmi->tx_size, mbmi->sb_type, inter_block) > 1 && + cm->base_qindex > 0 && !mbmi->skip && +#if CONFIG_SUPERTX + !supertx_enabled && +#endif // CONFIG_SUPERTX + !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) { + int eset = get_ext_tx_set(mbmi->tx_size, mbmi->sb_type, inter_block); + FRAME_COUNTS *counts = xd->counts; + + if (inter_block) { + if (eset > 0) { + mbmi->tx_type = vp10_read_tree( + r, vp10_ext_tx_inter_tree[eset], + cm->fc->inter_ext_tx_prob[eset][txsize_sqr_map[mbmi->tx_size]]); + if (counts) + ++counts->inter_ext_tx[eset][txsize_sqr_map[mbmi->tx_size]] + [mbmi->tx_type]; + } + } else if (ALLOW_INTRA_EXT_TX) { + if (eset > 0) { + mbmi->tx_type = vp10_read_tree( + r, vp10_ext_tx_intra_tree[eset], + cm->fc->intra_ext_tx_prob[eset][mbmi->tx_size][mbmi->mode]); + if (counts) + ++counts->intra_ext_tx[eset][mbmi->tx_size][mbmi->mode] + [mbmi->tx_type]; + } + } + } else { + mbmi->tx_type = DCT_DCT; + } +#else + if (mbmi->tx_size < TX_32X32 && cm->base_qindex > 0 && !mbmi->skip && +#if CONFIG_SUPERTX + !supertx_enabled && +#endif // CONFIG_SUPERTX + !segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) { + FRAME_COUNTS *counts = xd->counts; + if (inter_block) { + mbmi->tx_type = vp10_read_tree( + r, vp10_ext_tx_tree, cm->fc->inter_ext_tx_prob[mbmi->tx_size]); + if (counts) ++counts->inter_ext_tx[mbmi->tx_size][mbmi->tx_type]; + } else { + const TX_TYPE tx_type_nom = intra_mode_to_tx_type_context[mbmi->mode]; + mbmi->tx_type = vp10_read_tree( + r, vp10_ext_tx_tree, + cm->fc->intra_ext_tx_prob[mbmi->tx_size][tx_type_nom]); + if (counts) + ++counts->intra_ext_tx[mbmi->tx_size][tx_type_nom][mbmi->tx_type]; + } + } else { + mbmi->tx_type = DCT_DCT; + } +#endif // CONFIG_EXT_TX + } +} + +void vp10_read_mode_info(VP10Decoder *const pbi, MACROBLOCKD *xd, +#if CONFIG_SUPERTX + int supertx_enabled, +#endif // CONFIG_SUPERTX + int mi_row, int mi_col, vp10_reader *r, int x_mis, + int y_mis) { + VP10_COMMON *const cm = &pbi->common; + MODE_INFO *const mi = xd->mi[0]; + MV_REF *frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col; + int w, h; + + if (frame_is_intra_only(cm)) { + read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r); +#if CONFIG_REF_MV + for (h = 0; h < y_mis; ++h) { + MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols; + for (w = 0; w < x_mis; ++w) { + MV_REF *const mv = frame_mv + w; + mv->ref_frame[0] = NONE; + mv->ref_frame[1] = NONE; + } + } +#endif + } else { + read_inter_frame_mode_info(pbi, xd, +#if CONFIG_SUPERTX + supertx_enabled, +#endif // CONFIG_SUPERTX + mi_row, mi_col, r); + for (h = 0; h < y_mis; ++h) { + MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols; + for (w = 0; w < x_mis; ++w) { + MV_REF *const mv = frame_mv + w; + mv->ref_frame[0] = mi->mbmi.ref_frame[0]; + mv->ref_frame[1] = mi->mbmi.ref_frame[1]; + mv->mv[0].as_int = mi->mbmi.mv[0].as_int; + mv->mv[1].as_int = mi->mbmi.mv[1].as_int; + } + } + } +}
diff --git a/av1/decoder/decodemv.h b/av1/decoder/decodemv.h new file mode 100644 index 0000000..59fdd70 --- /dev/null +++ b/av1/decoder/decodemv.h
@@ -0,0 +1,34 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DECODEMV_H_ +#define VP10_DECODER_DECODEMV_H_ + +#include "av1/decoder/bitreader.h" + +#include "av1/decoder/decoder.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp10_read_mode_info(VP10Decoder *const pbi, MACROBLOCKD *xd, +#if CONFIG_SUPERTX + int supertx_enabled, +#endif + + int mi_row, int mi_col, vp10_reader *r, int x_mis, + int y_mis); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP10_DECODER_DECODEMV_H_
diff --git a/av1/decoder/decoder.c b/av1/decoder/decoder.c new file mode 100644 index 0000000..4cea36b --- /dev/null +++ b/av1/decoder/decoder.c
@@ -0,0 +1,549 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> +#include <limits.h> +#include <stdio.h> + +#include "./vp10_rtcd.h" +#include "./vpx_dsp_rtcd.h" +#include "./vpx_scale_rtcd.h" + +#include "aom_mem/vpx_mem.h" +#include "aom_ports/system_state.h" +#include "aom_ports/vpx_once.h" +#include "aom_ports/vpx_timer.h" +#include "aom_scale/vpx_scale.h" +#include "aom_util/vpx_thread.h" + +#include "av1/common/alloccommon.h" +#include "av1/common/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" + +static void initialize_dec(void) { + static volatile int init_done = 0; + + if (!init_done) { + vp10_rtcd(); + vpx_dsp_rtcd(); + vpx_scale_rtcd(); + vp10_init_intra_predictors(); +#if CONFIG_EXT_INTER + vp10_init_wedge_masks(); +#endif // CONFIG_EXT_INTER + init_done = 1; + } +} + +static void vp10_dec_setup_mi(VP10_COMMON *cm) { + cm->mi = cm->mip + cm->mi_stride + 1; + cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1; + memset(cm->mi_grid_base, 0, + cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base)); +} + +static int vp10_dec_alloc_mi(VP10_COMMON *cm, int mi_size) { + cm->mip = vpx_calloc(mi_size, sizeof(*cm->mip)); + if (!cm->mip) return 1; + cm->mi_alloc_size = mi_size; + cm->mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO *)); + if (!cm->mi_grid_base) return 1; + return 0; +} + +static void vp10_dec_free_mi(VP10_COMMON *cm) { + vpx_free(cm->mip); + cm->mip = NULL; + vpx_free(cm->mi_grid_base); + cm->mi_grid_base = NULL; +} + +VP10Decoder *vp10_decoder_create(BufferPool *const pool) { + VP10Decoder *volatile const pbi = vpx_memalign(32, sizeof(*pbi)); + VP10_COMMON *volatile const cm = pbi ? &pbi->common : NULL; + + if (!cm) return NULL; + + vp10_zero(*pbi); + + if (setjmp(cm->error.jmp)) { + cm->error.setjmp = 0; + vp10_decoder_remove(pbi); + return NULL; + } + + cm->error.setjmp = 1; + + CHECK_MEM_ERROR(cm, cm->fc, (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc))); + CHECK_MEM_ERROR( + cm, cm->frame_contexts, + (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS, sizeof(*cm->frame_contexts))); + + pbi->need_resync = 1; + 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->ready_for_new_data = 1; + pbi->common.buffer_pool = pool; + + cm->bit_depth = VPX_BITS_8; + cm->dequant_bit_depth = VPX_BITS_8; + + cm->alloc_mi = vp10_dec_alloc_mi; + cm->free_mi = vp10_dec_free_mi; + cm->setup_mi = vp10_dec_setup_mi; + + vp10_loop_filter_init(cm); + +#if CONFIG_AOM_QM + aom_qm_init(cm); +#endif +#if CONFIG_LOOP_RESTORATION + vp10_loop_restoration_precal(); +#endif // CONFIG_LOOP_RESTORATION + + cm->error.setjmp = 0; + + vpx_get_worker_interface()->init(&pbi->lf_worker); + + return pbi; +} + +void vp10_decoder_remove(VP10Decoder *pbi) { + int i; + + if (!pbi) return; + + vpx_get_worker_interface()->end(&pbi->lf_worker); + vpx_free(pbi->lf_worker.data1); + vpx_free(pbi->tile_data); + for (i = 0; i < pbi->num_tile_workers; ++i) { + VPxWorker *const worker = &pbi->tile_workers[i]; + vpx_get_worker_interface()->end(worker); + } + vpx_free(pbi->tile_worker_data); + vpx_free(pbi->tile_worker_info); + vpx_free(pbi->tile_workers); + + if (pbi->num_tile_workers > 0) { + vp10_loop_filter_dealloc(&pbi->lf_row_sync); + } + + vpx_free(pbi); +} + +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; +} + +vpx_codec_err_t vp10_copy_reference_dec(VP10Decoder *pbi, + VPX_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd) { + VP10_COMMON *cm = &pbi->common; + + /* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the + * encoder is using the frame buffers for. This is just a stub to keep the + * vpxenc --test-decode functionality working, and will be replaced in a + * later commit that adds VP9-specific controls for this functionality. + */ + if (ref_frame_flag == VPX_LAST_FLAG) { + const YV12_BUFFER_CONFIG *const cfg = get_ref_frame(cm, 0); + if (cfg == NULL) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "No 'last' reference frame"); + return VPX_CODEC_ERROR; + } + if (!equal_dimensions(cfg, sd)) + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + else + vpx_yv12_copy_frame(cfg, sd); + } else { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, "Invalid reference frame"); + } + + return cm->error.error_code; +} + +vpx_codec_err_t vp10_set_reference_dec(VP10_COMMON *cm, + VPX_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd) { + int idx; + YV12_BUFFER_CONFIG *ref_buf = NULL; + + // TODO(jkoleszar): The decoder doesn't have any real knowledge of what the + // encoder is using the frame buffers for. This is just a stub to keep the + // vpxenc --test-decode functionality working, and will be replaced in a + // later commit that adds VP9-specific controls for this functionality. + + // (Yunqing) The set_reference control depends on the following setting in + // encoder. + // cpi->lst_fb_idx = 0; + // #if CONFIG_EXT_REFS + // cpi->lst2_fb_idx = 1; + // cpi->lst3_fb_idx = 2; + // cpi->gld_fb_idx = 3; + // cpi->bwd_fb_idx = 4; + // cpi->alt_fb_idx = 5; + // #else // CONFIG_EXT_REFS + // cpi->gld_fb_idx = 1; + // cpi->alt_fb_idx = 2; + // #endif // CONFIG_EXT_REFS + + // TODO(zoeliu): To revisit following code and reconsider what assumption we + // may take on the reference frame buffer virtual indexes + if (ref_frame_flag == VPX_LAST_FLAG) { + idx = cm->ref_frame_map[0]; +#if CONFIG_EXT_REFS + } else if (ref_frame_flag == VPX_LAST2_FLAG) { + idx = cm->ref_frame_map[1]; + } else if (ref_frame_flag == VPX_LAST3_FLAG) { + idx = cm->ref_frame_map[2]; + } else if (ref_frame_flag == VPX_GOLD_FLAG) { + idx = cm->ref_frame_map[3]; + } else if (ref_frame_flag == VPX_BWD_FLAG) { + idx = cm->ref_frame_map[4]; + } else if (ref_frame_flag == VPX_ALT_FLAG) { + idx = cm->ref_frame_map[5]; +#else + } else if (ref_frame_flag == VPX_GOLD_FLAG) { + idx = cm->ref_frame_map[1]; + } else if (ref_frame_flag == VPX_ALT_FLAG) { + idx = cm->ref_frame_map[2]; +#endif // CONFIG_EXT_REFS + } else { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, "Invalid reference frame"); + return cm->error.error_code; + } + + if (idx < 0 || idx >= FRAME_BUFFERS) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Invalid reference frame map"); + return cm->error.error_code; + } + + // Get the destination reference buffer. + ref_buf = &cm->buffer_pool->frame_bufs[idx].buf; + + if (!equal_dimensions(ref_buf, sd)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Incorrect buffer dimensions"); + } else { + // Overwrite the reference frame buffer. + vpx_yv12_copy_frame(sd, ref_buf); + } + + return cm->error.error_code; +} + +/* If any buffer updating is signaled it should be done here. */ +static void swap_frame_buffers(VP10Decoder *pbi) { + int ref_index = 0, mask; + VP10_COMMON *const cm = &pbi->common; + BufferPool *const pool = cm->buffer_pool; + RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs; + + lock_buffer_pool(pool); + 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. + for (; ref_index < REF_FRAMES && !cm->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]; + } + + unlock_buffer_pool(pool); + pbi->hold_ref_buf = 0; + cm->frame_to_show = get_frame_new_buffer(cm); + + // TODO(zoeliu): To fix the ref frame buffer update for the scenario of + // cm->frame_parellel_decode == 1 + if (!cm->frame_parallel_decode || !cm->show_frame) { + lock_buffer_pool(pool); + --frame_bufs[cm->new_fb_idx].ref_count; + unlock_buffer_pool(pool); + } + + // 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 vp10_receive_compressed_data(VP10Decoder *pbi, size_t size, + const uint8_t **psource) { + VP10_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; + int retcode = 0; + cm->error.error_code = VPX_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; + } + } + + pbi->ready_for_new_data = 0; + + // Find a free buffer for the new frame, releasing the reference previously + // held. + + // Check if the previous frame was a frame without any references to it. + // Release frame buffer if not decoding in frame parallel mode. + if (!cm->frame_parallel_decode && cm->new_fb_idx >= 0 && + frame_bufs[cm->new_fb_idx].ref_count == 0) + pool->release_fb_cb(pool->cb_priv, + &frame_bufs[cm->new_fb_idx].raw_frame_buffer); + + // 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) return VPX_CODEC_MEM_ERROR; + + // Assign a MV array to the frame buffer. + cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx]; + + pbi->hold_ref_buf = 0; + if (cm->frame_parallel_decode) { + VPxWorker *const worker = pbi->frame_worker_owner; + vp10_frameworker_lock_stats(worker); + frame_bufs[cm->new_fb_idx].frame_worker_owner = worker; + // Reset decoding progress. + pbi->cur_buf = &frame_bufs[cm->new_fb_idx]; + pbi->cur_buf->row = -1; + pbi->cur_buf->col = -1; + vp10_frameworker_unlock_stats(worker); + } else { + pbi->cur_buf = &frame_bufs[cm->new_fb_idx]; + } + + if (setjmp(cm->error.jmp)) { + const VPxWorkerInterface *const winterface = vpx_get_worker_interface(); + int i; + + cm->error.setjmp = 0; + pbi->ready_for_new_data = 1; + + // 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_tile_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. + for (; ref_index < REF_FRAMES && !cm->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); + + vpx_clear_system_state(); + return -1; + } + + cm->error.setjmp = 1; + vp10_decode_frame(pbi, source, source + size, psource); + + swap_frame_buffers(pbi); + +#if CONFIG_EXT_TILE + // For now, we only extend the frame borders when the whole frame is decoded. + // Later, if needed, extend the border for the decoded tile on the frame + // border. + if (pbi->dec_tile_row == -1 && pbi->dec_tile_col == -1) +#endif // CONFIG_EXT_TILE + vpx_extend_frame_inner_borders(cm->frame_to_show); + + vpx_clear_system_state(); + + if (!cm->show_existing_frame) { + cm->last_show_frame = cm->show_frame; + +#if CONFIG_EXT_REFS + // NOTE: It is not supposed to ref to any frame not used as reference + if (cm->is_reference_frame) +#endif // CONFIG_EXT_REFS + cm->prev_frame = cm->cur_frame; + + if (cm->seg.enabled && !cm->frame_parallel_decode) + vp10_swap_current_and_last_seg_map(cm); + } + + // Update progress in frame parallel decode. + if (cm->frame_parallel_decode) { + // Need to lock the mutex here as another thread may + // be accessing this buffer. + VPxWorker *const worker = pbi->frame_worker_owner; + FrameWorkerData *const frame_worker_data = worker->data1; + vp10_frameworker_lock_stats(worker); + + if (cm->show_frame) { + cm->current_video_frame++; + } + frame_worker_data->frame_decoded = 1; + frame_worker_data->frame_context_ready = 1; + vp10_frameworker_signal_stats(worker); + vp10_frameworker_unlock_stats(worker); + } else { + cm->last_width = cm->width; + cm->last_height = cm->height; + if (cm->show_frame) { + cm->current_video_frame++; + } + } + + cm->error.setjmp = 0; + return retcode; +} + +int vp10_get_raw_frame(VP10Decoder *pbi, YV12_BUFFER_CONFIG *sd) { + VP10_COMMON *const cm = &pbi->common; + int ret = -1; + if (pbi->ready_for_new_data == 1) return ret; + + pbi->ready_for_new_data = 1; + + /* no raw frame to show!!! */ + if (!cm->show_frame) return ret; + + pbi->ready_for_new_data = 1; + *sd = *cm->frame_to_show; + ret = 0; + vpx_clear_system_state(); + return ret; +} + +int vp10_get_frame_to_show(VP10Decoder *pbi, YV12_BUFFER_CONFIG *frame) { + VP10_COMMON *const cm = &pbi->common; + + if (!cm->show_frame || !cm->frame_to_show) return -1; + + *frame = *cm->frame_to_show; + return 0; +} + +vpx_codec_err_t vp10_parse_superframe_index(const uint8_t *data, size_t data_sz, + uint32_t sizes[8], int *count, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state) { + // A chunk ending with a byte matching 0xc0 is an invalid chunk unless + // it is a super frame index. If the last byte of real video compression + // data is 0xc0 the encoder must add a 0 byte. If we have the marker but + // not the associated matching marker byte at the front of the index we have + // an invalid bitstream and need to return an error. + + uint8_t marker; + size_t frame_sz_sum = 0; + + assert(data_sz); + marker = read_marker(decrypt_cb, decrypt_state, data + data_sz - 1); + *count = 0; + + if ((marker & 0xe0) == 0xc0) { + const uint32_t frames = (marker & 0x7) + 1; + const uint32_t mag = ((marker >> 3) & 0x3) + 1; + const size_t index_sz = 2 + mag * (frames - 1); + + // This chunk is marked as having a superframe index but doesn't have + // enough data for it, thus it's an invalid superframe index. + if (data_sz < index_sz) return VPX_CODEC_CORRUPT_FRAME; + + { + const uint8_t marker2 = + read_marker(decrypt_cb, decrypt_state, data + data_sz - index_sz); + + // This chunk is marked as having a superframe index but doesn't have + // the matching marker byte at the front of the index therefore it's an + // invalid chunk. + if (marker != marker2) return VPX_CODEC_CORRUPT_FRAME; + } + + { + // Found a valid superframe index. + uint32_t i, j; + const uint8_t *x = &data[data_sz - index_sz + 1]; + + // Frames has a maximum of 8 and mag has a maximum of 4. + uint8_t clear_buffer[32]; + assert(sizeof(clear_buffer) >= frames * mag); + if (decrypt_cb) { + decrypt_cb(decrypt_state, x, clear_buffer, frames * mag); + x = clear_buffer; + } + + for (i = 0; i < frames - 1; ++i) { + uint32_t this_sz = 0; + + for (j = 0; j < mag; ++j) this_sz |= (*x++) << (j * 8); + this_sz += 1; + sizes[i] = this_sz; + frame_sz_sum += this_sz; + } + sizes[i] = (uint32_t)(data_sz - index_sz - frame_sz_sum); + *count = frames; + } + } + return VPX_CODEC_OK; +}
diff --git a/av1/decoder/decoder.h b/av1/decoder/decoder.h new file mode 100644 index 0000000..47a5a7b --- /dev/null +++ b/av1/decoder/decoder.h
@@ -0,0 +1,169 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DECODER_H_ +#define VP10_DECODER_DECODER_H_ + +#include "./vpx_config.h" + +#include "aom/vpx_codec.h" +#include "av1/decoder/bitreader.h" +#include "aom_scale/yv12config.h" +#include "aom_util/vpx_thread.h" + +#include "av1/common/thread_common.h" +#include "av1/common/onyxc_int.h" +#include "av1/decoder/dthread.h" + +#ifdef __cplusplus +extern "C" { +#endif + +// TODO(hkuang): combine this with TileWorkerData. +typedef struct TileData { + VP10_COMMON *cm; + vp10_reader bit_reader; + DECLARE_ALIGNED(16, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(16, tran_low_t, dqcoeff[MAX_TX_SQUARE]); + DECLARE_ALIGNED(16, uint8_t, color_index_map[2][MAX_SB_SQUARE]); +} TileData; + +typedef struct TileWorkerData { + struct VP10Decoder *pbi; + vp10_reader bit_reader; + FRAME_COUNTS counts; + DECLARE_ALIGNED(16, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(16, tran_low_t, dqcoeff[MAX_TX_SQUARE]); + DECLARE_ALIGNED(16, uint8_t, color_index_map[2][MAX_SB_SQUARE]); + struct vpx_internal_error_info error_info; +} TileWorkerData; + +typedef struct TileBufferDec { + const uint8_t *data; + size_t size; + const uint8_t *raw_data_end; // The end of the raw tile buffer in the + // bit stream. + int col; // only used with multi-threaded decoding +} TileBufferDec; + +typedef struct VP10Decoder { + DECLARE_ALIGNED(16, MACROBLOCKD, mb); + + DECLARE_ALIGNED(16, VP10_COMMON, common); + + int ready_for_new_data; + + 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. + + VPxWorker *frame_worker_owner; // frame_worker that owns this pbi. + VPxWorker lf_worker; + VPxWorker *tile_workers; + TileWorkerData *tile_worker_data; + TileInfo *tile_worker_info; + int num_tile_workers; + + TileData *tile_data; + int allocated_tiles; + + TileBufferDec tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS]; + + VP10LfSync lf_row_sync; + + vpx_decrypt_cb decrypt_cb; + void *decrypt_state; + + 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; +#if CONFIG_EXT_TILE + int tile_col_size_bytes; + int dec_tile_row, dec_tile_col; +#endif // CONFIG_EXT_TILE +} VP10Decoder; + +int vp10_receive_compressed_data(struct VP10Decoder *pbi, size_t size, + const uint8_t **dest); + +int vp10_get_raw_frame(struct VP10Decoder *pbi, YV12_BUFFER_CONFIG *sd); + +int vp10_get_frame_to_show(struct VP10Decoder *pbi, YV12_BUFFER_CONFIG *frame); + +vpx_codec_err_t vp10_copy_reference_dec(struct VP10Decoder *pbi, + VPX_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd); + +vpx_codec_err_t vp10_set_reference_dec(VP10_COMMON *cm, + VPX_REFFRAME ref_frame_flag, + YV12_BUFFER_CONFIG *sd); + +static INLINE uint8_t read_marker(vpx_decrypt_cb decrypt_cb, + void *decrypt_state, const uint8_t *data) { + if (decrypt_cb) { + uint8_t marker; + decrypt_cb(decrypt_state, data, &marker, 1); + return marker; + } + return *data; +} + +// This function is exposed for use in tests, as well as the inlined function +// "read_marker". +vpx_codec_err_t vp10_parse_superframe_index(const uint8_t *data, size_t data_sz, + uint32_t sizes[8], int *count, + vpx_decrypt_cb decrypt_cb, + void *decrypt_state); + +struct VP10Decoder *vp10_decoder_create(BufferPool *const pool); + +void vp10_decoder_remove(struct VP10Decoder *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); + } + } +} + +#if CONFIG_EXT_REFS +static INLINE int dec_is_ref_frame_buf(VP10Decoder *const pbi, + RefCntBuffer *frame_buf) { + VP10_COMMON *const cm = &pbi->common; + int i; + for (i = 0; i < INTER_REFS_PER_FRAME; ++i) { + RefBuffer *const ref_frame = &cm->frame_refs[i]; + if (ref_frame->idx == INVALID_IDX) continue; + if (frame_buf == &cm->buffer_pool->frame_bufs[ref_frame->idx]) break; + } + return (i < INTER_REFS_PER_FRAME); +} +#endif // CONFIG_EXT_REFS + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP10_DECODER_DECODER_H_
diff --git a/av1/decoder/detokenize.c b/av1/decoder/detokenize.c new file mode 100644 index 0000000..0fba999 --- /dev/null +++ b/av1/decoder/detokenize.c
@@ -0,0 +1,536 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "aom_mem/vpx_mem.h" +#include "aom_ports/mem.h" + +#include "av1/common/ans.h" +#include "av1/common/blockd.h" +#include "av1/common/common.h" +#include "av1/common/entropy.h" +#include "av1/common/idct.h" + +#include "av1/decoder/detokenize.h" + +#define EOB_CONTEXT_NODE 0 +#define ZERO_CONTEXT_NODE 1 +#define ONE_CONTEXT_NODE 2 +#define LOW_VAL_CONTEXT_NODE 0 +#define TWO_CONTEXT_NODE 1 +#define THREE_CONTEXT_NODE 2 +#define HIGH_LOW_CONTEXT_NODE 3 +#define CAT_ONE_CONTEXT_NODE 4 +#define CAT_THREEFOUR_CONTEXT_NODE 5 +#define CAT_THREE_CONTEXT_NODE 6 +#define CAT_FIVE_CONTEXT_NODE 7 + +#define INCREMENT_COUNT(token) \ + do { \ + if (counts) ++coef_counts[band][ctx][token]; \ + } while (0) + +#if !CONFIG_ANS +static INLINE int read_coeff(const vpx_prob *probs, int n, vp10_reader *r) { + int i, val = 0; + for (i = 0; i < n; ++i) val = (val << 1) | vp10_read(r, probs[i]); + return val; +} + +#if CONFIG_AOM_QM +static int decode_coefs(const MACROBLOCKD *xd, PLANE_TYPE type, + tran_low_t *dqcoeff, TX_SIZE tx_size, TX_TYPE tx_type, + const int16_t *dq, int ctx, const int16_t *scan, + const int16_t *nb, vp10_reader *r, + const qm_val_t *iqm[2][TX_SIZES]) +#else +static int decode_coefs(const MACROBLOCKD *xd, PLANE_TYPE type, + tran_low_t *dqcoeff, TX_SIZE tx_size, TX_TYPE tx_type, + const int16_t *dq, +#if CONFIG_NEW_QUANT + dequant_val_type_nuq *dq_val, +#endif // CONFIG_NEW_QUANT + int ctx, const int16_t *scan, const int16_t *nb, + vp10_reader *r) +#endif +{ + FRAME_COUNTS *counts = xd->counts; + const int max_eob = get_tx2d_size(tx_size); + const FRAME_CONTEXT *const fc = xd->fc; + const int ref = is_inter_block(&xd->mi[0]->mbmi); +#if CONFIG_AOM_QM + const qm_val_t *iqmatrix = iqm[!ref][tx_size]; +#endif + int band, c = 0; + const int tx_size_ctx = txsize_sqr_map[tx_size]; + const vpx_prob(*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] = + fc->coef_probs[tx_size_ctx][type][ref]; + const vpx_prob *prob; + unsigned int(*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1]; + unsigned int(*eob_branch_count)[COEFF_CONTEXTS]; + uint8_t token_cache[MAX_TX_SQUARE]; + const uint8_t *band_translate = get_band_translate(tx_size); + int dq_shift; + int v, token; + int16_t dqv = dq[0]; +#if CONFIG_NEW_QUANT + const tran_low_t *dqv_val = &dq_val[0][0]; +#endif // CONFIG_NEW_QUANT + const uint8_t *cat1_prob; + const uint8_t *cat2_prob; + const uint8_t *cat3_prob; + const uint8_t *cat4_prob; + const uint8_t *cat5_prob; + const uint8_t *cat6_prob; + + if (counts) { + coef_counts = counts->coef[tx_size_ctx][type][ref]; + eob_branch_count = counts->eob_branch[tx_size_ctx][type][ref]; + } + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->bd > VPX_BITS_8) { + if (xd->bd == VPX_BITS_10) { + cat1_prob = vp10_cat1_prob_high10; + cat2_prob = vp10_cat2_prob_high10; + cat3_prob = vp10_cat3_prob_high10; + cat4_prob = vp10_cat4_prob_high10; + cat5_prob = vp10_cat5_prob_high10; + cat6_prob = vp10_cat6_prob_high10; + } else { + cat1_prob = vp10_cat1_prob_high12; + cat2_prob = vp10_cat2_prob_high12; + cat3_prob = vp10_cat3_prob_high12; + cat4_prob = vp10_cat4_prob_high12; + cat5_prob = vp10_cat5_prob_high12; + cat6_prob = vp10_cat6_prob_high12; + } + } else { + cat1_prob = vp10_cat1_prob; + cat2_prob = vp10_cat2_prob; + cat3_prob = vp10_cat3_prob; + cat4_prob = vp10_cat4_prob; + cat5_prob = vp10_cat5_prob; + cat6_prob = vp10_cat6_prob; + } +#else + cat1_prob = vp10_cat1_prob; + cat2_prob = vp10_cat2_prob; + cat3_prob = vp10_cat3_prob; + cat4_prob = vp10_cat4_prob; + cat5_prob = vp10_cat5_prob; + cat6_prob = vp10_cat6_prob; +#endif + + dq_shift = get_tx_scale(xd, tx_type, tx_size); + + while (c < max_eob) { + int val = -1; + band = *band_translate++; + prob = coef_probs[band][ctx]; + if (counts) ++eob_branch_count[band][ctx]; + if (!vp10_read(r, prob[EOB_CONTEXT_NODE])) { + INCREMENT_COUNT(EOB_MODEL_TOKEN); + break; + } + +#if CONFIG_NEW_QUANT + dqv_val = &dq_val[band][0]; +#endif // CONFIG_NEW_QUANT + + while (!vp10_read(r, prob[ZERO_CONTEXT_NODE])) { + INCREMENT_COUNT(ZERO_TOKEN); + dqv = dq[1]; + token_cache[scan[c]] = 0; + ++c; + if (c >= max_eob) return c; // zero tokens at the end (no eob token) + ctx = get_coef_context(nb, token_cache, c); + band = *band_translate++; + prob = coef_probs[band][ctx]; +#if CONFIG_NEW_QUANT + dqv_val = &dq_val[band][0]; +#endif // CONFIG_NEW_QUANT + } + + if (!vp10_read(r, prob[ONE_CONTEXT_NODE])) { + INCREMENT_COUNT(ONE_TOKEN); + token = ONE_TOKEN; + val = 1; + } else { + INCREMENT_COUNT(TWO_TOKEN); + token = vp10_read_tree(r, vp10_coef_con_tree, + vp10_pareto8_full[prob[PIVOT_NODE] - 1]); + switch (token) { + case TWO_TOKEN: + case THREE_TOKEN: + case FOUR_TOKEN: val = token; break; + case CATEGORY1_TOKEN: + val = CAT1_MIN_VAL + read_coeff(cat1_prob, 1, r); + break; + case CATEGORY2_TOKEN: + val = CAT2_MIN_VAL + read_coeff(cat2_prob, 2, r); + break; + case CATEGORY3_TOKEN: + val = CAT3_MIN_VAL + read_coeff(cat3_prob, 3, r); + break; + case CATEGORY4_TOKEN: + val = CAT4_MIN_VAL + read_coeff(cat4_prob, 4, r); + break; + case CATEGORY5_TOKEN: + val = CAT5_MIN_VAL + read_coeff(cat5_prob, 5, r); + break; + case CATEGORY6_TOKEN: { + const int skip_bits = TX_SIZES - 1 - txsize_sqr_up_map[tx_size]; + const uint8_t *cat6p = cat6_prob + skip_bits; +#if CONFIG_VP9_HIGHBITDEPTH + switch (xd->bd) { + case VPX_BITS_8: + val = CAT6_MIN_VAL + read_coeff(cat6p, 14 - skip_bits, r); + break; + case VPX_BITS_10: + val = CAT6_MIN_VAL + read_coeff(cat6p, 16 - skip_bits, r); + break; + case VPX_BITS_12: + val = CAT6_MIN_VAL + read_coeff(cat6p, 18 - skip_bits, r); + break; + default: assert(0); return -1; + } +#else + val = CAT6_MIN_VAL + read_coeff(cat6p, 14 - skip_bits, r); +#endif + break; + } + } + } +#if CONFIG_NEW_QUANT + + v = vp10_dequant_abscoeff_nuq(val, dqv, dqv_val); + v = dq_shift ? ROUND_POWER_OF_TWO(v, dq_shift) : v; +#else +#if CONFIG_AOM_QM + dqv = ((iqmatrix[scan[c]] * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >> + AOM_QM_BITS; +#endif + v = (val * dqv) >> dq_shift; +#endif // CONFIG_NEW_QUANT + +#if CONFIG_COEFFICIENT_RANGE_CHECKING +#if CONFIG_VP9_HIGHBITDEPTH + dqcoeff[scan[c]] = highbd_check_range((vp10_read_bit(r) ? -v : v), xd->bd); +#else + dqcoeff[scan[c]] = check_range(vp10_read_bit(r) ? -v : v); +#endif // CONFIG_VP9_HIGHBITDEPTH +#else + dqcoeff[scan[c]] = vp10_read_bit(r) ? -v : v; +#endif // CONFIG_COEFFICIENT_RANGE_CHECKING + token_cache[scan[c]] = vp10_pt_energy_class[token]; + ++c; + ctx = get_coef_context(nb, token_cache, c); + dqv = dq[1]; + } + + return c; +} +#else // !CONFIG_ANS +static INLINE int read_coeff(const vpx_prob *const probs, int n, + struct AnsDecoder *const ans) { + int i, val = 0; + for (i = 0; i < n; ++i) val = (val << 1) | uabs_read(ans, probs[i]); + return val; +} + +static int decode_coefs_ans(const MACROBLOCKD *const xd, PLANE_TYPE type, + tran_low_t *dqcoeff, TX_SIZE tx_size, + TX_TYPE tx_type, const int16_t *dq, +#if CONFIG_NEW_QUANT + dequant_val_type_nuq *dq_val, +#endif // CONFIG_NEW_QUANT + int ctx, const int16_t *scan, const int16_t *nb, + struct AnsDecoder *const ans) { + FRAME_COUNTS *counts = xd->counts; + const int max_eob = get_tx2d_size(tx_size); + const FRAME_CONTEXT *const fc = xd->fc; + const int ref = is_inter_block(&xd->mi[0]->mbmi); + int band, c = 0; + int skip_eob = 0; + const int tx_size_ctx = txsize_sqr_map[tx_size]; + const vpx_prob(*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] = + fc->coef_probs[tx_size_ctx][type][ref]; + const rans_dec_lut(*coef_cdfs)[COEFF_CONTEXTS] = + fc->coef_cdfs[tx_size_ctx][type][ref]; + const vpx_prob *prob; + const rans_dec_lut *cdf; + unsigned int(*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1]; + unsigned int(*eob_branch_count)[COEFF_CONTEXTS]; + uint8_t token_cache[MAX_TX_SQUARE]; + const uint8_t *band_translate = get_band_translate(tx_size); + int dq_shift; + int v, token; + int16_t dqv = dq[0]; +#if CONFIG_NEW_QUANT + const tran_low_t *dqv_val = &dq_val[0][0]; +#endif // CONFIG_NEW_QUANT + const uint8_t *cat1_prob; + const uint8_t *cat2_prob; + const uint8_t *cat3_prob; + const uint8_t *cat4_prob; + const uint8_t *cat5_prob; + const uint8_t *cat6_prob; + + dq_shift = get_tx_scale(xd, tx_type, tx_size); + + if (counts) { + coef_counts = counts->coef[tx_size_ctx][type][ref]; + eob_branch_count = counts->eob_branch[tx_size_ctx][type][ref]; + } + +#if CONFIG_VP9_HIGHBITDEPTH + if (xd->bd > VPX_BITS_8) { + if (xd->bd == VPX_BITS_10) { + cat1_prob = vp10_cat1_prob_high10; + cat2_prob = vp10_cat2_prob_high10; + cat3_prob = vp10_cat3_prob_high10; + cat4_prob = vp10_cat4_prob_high10; + cat5_prob = vp10_cat5_prob_high10; + cat6_prob = vp10_cat6_prob_high10; + } else { + cat1_prob = vp10_cat1_prob_high12; + cat2_prob = vp10_cat2_prob_high12; + cat3_prob = vp10_cat3_prob_high12; + cat4_prob = vp10_cat4_prob_high12; + cat5_prob = vp10_cat5_prob_high12; + cat6_prob = vp10_cat6_prob_high12; + } + } else { + cat1_prob = vp10_cat1_prob; + cat2_prob = vp10_cat2_prob; + cat3_prob = vp10_cat3_prob; + cat4_prob = vp10_cat4_prob; + cat5_prob = vp10_cat5_prob; + cat6_prob = vp10_cat6_prob; + } +#else + cat1_prob = vp10_cat1_prob; + cat2_prob = vp10_cat2_prob; + cat3_prob = vp10_cat3_prob; + cat4_prob = vp10_cat4_prob; + cat5_prob = vp10_cat5_prob; + cat6_prob = vp10_cat6_prob; +#endif + + while (c < max_eob) { + int val = -1; + band = *band_translate++; + prob = coef_probs[band][ctx]; + if (!skip_eob) { + if (counts) ++eob_branch_count[band][ctx]; + if (!uabs_read(ans, prob[EOB_CONTEXT_NODE])) { + INCREMENT_COUNT(EOB_MODEL_TOKEN); + break; + } + } + +#if CONFIG_NEW_QUANT + dqv_val = &dq_val[band][0]; +#endif // CONFIG_NEW_QUANT + + cdf = &coef_cdfs[band][ctx]; + token = ZERO_TOKEN + rans_read(ans, *cdf); + if (token == ZERO_TOKEN) { + INCREMENT_COUNT(ZERO_TOKEN); + token_cache[scan[c]] = 0; + skip_eob = 1; + } else { + INCREMENT_COUNT(ONE_TOKEN + (token > ONE_TOKEN)); + switch (token) { + case ONE_TOKEN: + case TWO_TOKEN: + case THREE_TOKEN: + case FOUR_TOKEN: val = token; break; + case CATEGORY1_TOKEN: + val = CAT1_MIN_VAL + read_coeff(cat1_prob, 1, ans); + break; + case CATEGORY2_TOKEN: + val = CAT2_MIN_VAL + read_coeff(cat2_prob, 2, ans); + break; + case CATEGORY3_TOKEN: + val = CAT3_MIN_VAL + read_coeff(cat3_prob, 3, ans); + break; + case CATEGORY4_TOKEN: + val = CAT4_MIN_VAL + read_coeff(cat4_prob, 4, ans); + break; + case CATEGORY5_TOKEN: + val = CAT5_MIN_VAL + read_coeff(cat5_prob, 5, ans); + break; + case CATEGORY6_TOKEN: { + const int skip_bits = TX_SIZES - 1 - txsize_sqr_up_map[tx_size]; + const uint8_t *cat6p = cat6_prob + skip_bits; +#if CONFIG_VP9_HIGHBITDEPTH + switch (xd->bd) { + case VPX_BITS_8: + val = CAT6_MIN_VAL + read_coeff(cat6p, 14 - skip_bits, ans); + break; + case VPX_BITS_10: + val = CAT6_MIN_VAL + read_coeff(cat6p, 16 - skip_bits, ans); + break; + case VPX_BITS_12: + val = CAT6_MIN_VAL + read_coeff(cat6p, 18 - skip_bits, ans); + break; + default: assert(0); return -1; + } +#else + val = CAT6_MIN_VAL + read_coeff(cat6p, 14 - skip_bits, ans); +#endif + } break; + } +#if CONFIG_NEW_QUANT + v = vp10_dequant_abscoeff_nuq(val, dqv, dqv_val); + v = dq_shift ? ROUND_POWER_OF_TWO(v, dq_shift) : v; +#else + v = (val * dqv) >> dq_shift; +#endif // CONFIG_NEW_QUANT + +#if CONFIG_COEFFICIENT_RANGE_CHECKING +#if CONFIG_VP9_HIGHBITDEPTH + dqcoeff[scan[c]] = + highbd_check_range((uabs_read_bit(ans) ? -v : v), xd->bd); +#else + dqcoeff[scan[c]] = check_range(uabs_read_bit(ans) ? -v : v); +#endif // CONFIG_VP9_HIGHBITDEPTH +#else + dqcoeff[scan[c]] = uabs_read_bit(ans) ? -v : v; +#endif // CONFIG_COEFFICIENT_RANGE_CHECKING + token_cache[scan[c]] = vp10_pt_energy_class[token]; + skip_eob = 0; + } + ++c; + ctx = get_coef_context(nb, token_cache, c); + dqv = dq[1]; + } + + return c; +} +#endif // !CONFIG_ANS + +// TODO(slavarnway): Decode version of vp10_set_context. Modify +// vp10_set_context +// after testing is complete, then delete this version. +static void dec_set_contexts(const MACROBLOCKD *xd, + struct macroblockd_plane *pd, TX_SIZE tx_size, + int has_eob, int aoff, int loff) { + ENTROPY_CONTEXT *const a = pd->above_context + aoff; + ENTROPY_CONTEXT *const l = pd->left_context + loff; + const int tx_w_in_blocks = num_4x4_blocks_wide_txsize_lookup[tx_size]; + const int tx_h_in_blocks = num_4x4_blocks_high_txsize_lookup[tx_size]; + + // above + if (has_eob && xd->mb_to_right_edge < 0) { + int i; + const int blocks_wide = + pd->n4_w + (xd->mb_to_right_edge >> (5 + pd->subsampling_x)); + int above_contexts = tx_w_in_blocks; + if (above_contexts + aoff > blocks_wide) + above_contexts = blocks_wide - aoff; + + for (i = 0; i < above_contexts; ++i) a[i] = has_eob; + for (i = above_contexts; i < tx_w_in_blocks; ++i) a[i] = 0; + } else { + memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * tx_w_in_blocks); + } + + // left + if (has_eob && xd->mb_to_bottom_edge < 0) { + int i; + const int blocks_high = + pd->n4_h + (xd->mb_to_bottom_edge >> (5 + pd->subsampling_y)); + int left_contexts = tx_h_in_blocks; + if (left_contexts + loff > blocks_high) left_contexts = blocks_high - loff; + + for (i = 0; i < left_contexts; ++i) l[i] = has_eob; + for (i = left_contexts; i < tx_h_in_blocks; ++i) l[i] = 0; + } else { + memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * tx_h_in_blocks); + } +} + +void vp10_decode_palette_tokens(MACROBLOCKD *const xd, int plane, + vp10_reader *r) { + MODE_INFO *const mi = xd->mi[0]; + MB_MODE_INFO *const mbmi = &mi->mbmi; + const BLOCK_SIZE bsize = mbmi->sb_type; + const int rows = (4 * num_4x4_blocks_high_lookup[bsize]) >> + (xd->plane[plane != 0].subsampling_y); + const int cols = (4 * num_4x4_blocks_wide_lookup[bsize]) >> + (xd->plane[plane != 0].subsampling_x); + int color_idx, color_ctx, color_order[PALETTE_MAX_SIZE]; + int n = mbmi->palette_mode_info.palette_size[plane != 0]; + int i, j; + uint8_t *color_map = xd->plane[plane != 0].color_index_map; + const vpx_prob (*const prob)[PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] = + plane ? vp10_default_palette_uv_color_prob + : vp10_default_palette_y_color_prob; + + for (i = 0; i < rows; ++i) { + for (j = (i == 0 ? 1 : 0); j < cols; ++j) { + color_ctx = + vp10_get_palette_color_context(color_map, cols, i, j, n, color_order); + color_idx = vp10_read_tree(r, vp10_palette_color_tree[n - 2], + prob[n - 2][color_ctx]); + assert(color_idx >= 0 && color_idx < n); + color_map[i * cols + j] = color_order[color_idx]; + } + } +} + +int vp10_decode_block_tokens(MACROBLOCKD *const xd, int plane, + const scan_order *sc, int x, int y, + TX_SIZE tx_size, TX_TYPE tx_type, +#if CONFIG_ANS + struct AnsDecoder *const r, +#else + vp10_reader *r, +#endif // CONFIG_ANS + int seg_id) { + struct macroblockd_plane *const pd = &xd->plane[plane]; + const int16_t *const dequant = pd->seg_dequant[seg_id]; + const int ctx = + get_entropy_context(tx_size, pd->above_context + x, pd->left_context + y); +#if CONFIG_NEW_QUANT + int dq = get_dq_profile_from_ctx(ctx); +#endif // CONFIG_NEW_QUANT + +#if !CONFIG_ANS +#if CONFIG_AOM_QM + const int eob = + decode_coefs(xd, pd->plane_type, pd->dqcoeff, tx_size, tx_type, dequant, + ctx, sc->scan, sc->neighbors, r, pd->seg_iqmatrix[seg_id]); +#else + const int eob = + decode_coefs(xd, pd->plane_type, pd->dqcoeff, tx_size, tx_type, dequant, +#if CONFIG_NEW_QUANT + pd->seg_dequant_nuq[seg_id][dq], +#endif // CONFIG_NEW_QUANT + ctx, sc->scan, sc->neighbors, r); +#endif // CONFIG_AOM_QM +#else + const int eob = decode_coefs_ans(xd, pd->plane_type, pd->dqcoeff, tx_size, + tx_type, dequant, +#if CONFIG_NEW_QUANT + pd->seg_dequant_nuq[seg_id][dq], +#endif // CONFIG_NEW_QUANT + ctx, sc->scan, sc->neighbors, r); +#endif // !CONFIG_ANS + dec_set_contexts(xd, pd, tx_size, eob > 0, x, y); + /* + vp10_set_contexts(xd, pd, + get_plane_block_size(xd->mi[0]->mbmi.sb_type, pd), + tx_size, eob > 0, x, y); + */ + return eob; +}
diff --git a/av1/decoder/detokenize.h b/av1/decoder/detokenize.h new file mode 100644 index 0000000..279c193 --- /dev/null +++ b/av1/decoder/detokenize.h
@@ -0,0 +1,38 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DETOKENIZE_H_ +#define VP10_DECODER_DETOKENIZE_H_ + +#include "av1/decoder/decoder.h" +#include "av1/common/ans.h" +#include "av1/common/scan.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp10_decode_palette_tokens(MACROBLOCKD *const xd, int plane, + vp10_reader *r); +int vp10_decode_block_tokens(MACROBLOCKD *const xd, int plane, + const scan_order *sc, int x, int y, + TX_SIZE tx_size, TX_TYPE tx_type, +#if CONFIG_ANS + struct AnsDecoder *const r, +#else + vp10_reader *r, +#endif // CONFIG_ANS + int seg_id); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP10_DECODER_DETOKENIZE_H_
diff --git a/av1/decoder/dsubexp.c b/av1/decoder/dsubexp.c new file mode 100644 index 0000000..146a1de --- /dev/null +++ b/av1/decoder/dsubexp.c
@@ -0,0 +1,82 @@ +/* + Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <assert.h> + +#include "av1/common/entropy.h" + +#include "av1/decoder/dsubexp.h" + +static int inv_recenter_nonneg(int v, int m) { + if (v > 2 * m) return v; + + return (v & 1) ? m - ((v + 1) >> 1) : m + (v >> 1); +} + +static int decode_uniform(vp10_reader *r) { + const int l = 8; + const int m = (1 << l) - 190; + const int v = vp10_read_literal(r, l - 1); + return v < m ? v : (v << 1) - m + vp10_read_bit(r); +} + +static int inv_remap_prob(int v, int m) { + static uint8_t inv_map_table[MAX_PROB - 1] = { + 7, 20, 33, 46, 59, 72, 85, 98, 111, 124, 137, 150, 163, 176, 189, + 202, 215, 228, 241, 254, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, + 12, 13, 14, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, + 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, + 44, 45, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 60, + 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, + 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, + 93, 94, 95, 96, 97, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, + 109, 110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 125, + 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 138, 139, 140, 141, + 142, 143, 144, 145, 146, 147, 148, 149, 151, 152, 153, 154, 155, 156, 157, + 158, 159, 160, 161, 162, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, + 174, 175, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 190, + 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 204, 205, 206, + 207, 208, 209, 210, 211, 212, 213, 214, 216, 217, 218, 219, 220, 221, 222, + 223, 224, 225, 226, 227, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, + 239, 240, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, + }; + assert(v < (int)(sizeof(inv_map_table) / sizeof(inv_map_table[0]))); + v = inv_map_table[v]; + m--; + if ((m << 1) <= MAX_PROB) { + return 1 + inv_recenter_nonneg(v, m); + } else { + return MAX_PROB - inv_recenter_nonneg(v, MAX_PROB - 1 - m); + } +} + +static int decode_term_subexp(vp10_reader *r) { + if (!vp10_read_bit(r)) return vp10_read_literal(r, 4); + if (!vp10_read_bit(r)) return vp10_read_literal(r, 4) + 16; + if (!vp10_read_bit(r)) return vp10_read_literal(r, 5) + 32; + return decode_uniform(r) + 64; +} + +void vp10_diff_update_prob(vp10_reader *r, vpx_prob *p) { + if (vp10_read(r, DIFF_UPDATE_PROB)) { + const int delp = decode_term_subexp(r); + *p = (vpx_prob)inv_remap_prob(delp, *p); + } +} + +int vp10_read_primitive_symmetric(vp10_reader *r, unsigned int mag_bits) { + if (vp10_read_bit(r)) { + int s = vp10_read_bit(r); + int x = vp10_read_literal(r, mag_bits) + 1; + return (s > 0 ? -x : x); + } else { + return 0; + } +}
diff --git a/av1/decoder/dsubexp.h b/av1/decoder/dsubexp.h new file mode 100644 index 0000000..b8980f7 --- /dev/null +++ b/av1/decoder/dsubexp.h
@@ -0,0 +1,31 @@ +/* + * Copyright (c) 2013 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DSUBEXP_H_ +#define VP10_DECODER_DSUBEXP_H_ + +#include "av1/decoder/bitreader.h" + +#ifdef __cplusplus +extern "C" { +#endif + +void vp10_diff_update_prob(vp10_reader *r, vpx_prob *p); + +#ifdef __cplusplus +} // extern "C" +#endif + +// mag_bits is number of bits for magnitude. The alphabet is of size +// 2 * 2^mag_bits + 1, symmetric around 0, where one bit is used to +// indicate 0 or non-zero, mag_bits bits are used to indicate magnitide +// and 1 more bit for the sign if non-zero. +int vp10_read_primitive_symmetric(vp10_reader *r, unsigned int mag_bits); +#endif // VP10_DECODER_DSUBEXP_H_
diff --git a/av1/decoder/dthread.c b/av1/decoder/dthread.c new file mode 100644 index 0000000..d9a2ce1 --- /dev/null +++ b/av1/decoder/dthread.c
@@ -0,0 +1,193 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include "./vpx_config.h" +#include "aom_mem/vpx_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 vp10_frameworker_lock_stats(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_lock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void vp10_frameworker_unlock_stats(VPxWorker *const worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const worker_data = worker->data1; + pthread_mutex_unlock(&worker_data->stats_mutex); +#else + (void)worker; +#endif +} + +void vp10_frameworker_signal_stats(VPxWorker *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 vp10_frameworker_wait(VPxWorker *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. + VPxWorker *const ref_worker = ref_buf->frame_worker_owner; + FrameWorkerData *const ref_worker_data = + (FrameWorkerData *)ref_worker->data1; + const VP10Decoder *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 + + vp10_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; + vp10_frameworker_unlock_stats(ref_worker); + vpx_internal_error(&worker_data->pbi->common.error, + VPX_CODEC_CORRUPT_FRAME, + "Worker %p failed to decode frame", worker); + } + vp10_frameworker_unlock_stats(ref_worker); + } +#else + (void)worker; + (void)ref_buf; + (void)row; + (void)ref_buf; +#endif // CONFIG_MULTITHREAD +} + +void vp10_frameworker_broadcast(RefCntBuffer *const buf, int row) { +#if CONFIG_MULTITHREAD + VPxWorker *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 + + vp10_frameworker_lock_stats(worker); + buf->row = row; + vp10_frameworker_signal_stats(worker); + vp10_frameworker_unlock_stats(worker); +#else + (void)buf; + (void)row; +#endif // CONFIG_MULTITHREAD +} + +void vp10_frameworker_copy_context(VPxWorker *const dst_worker, + VPxWorker *const src_worker) { +#if CONFIG_MULTITHREAD + FrameWorkerData *const src_worker_data = (FrameWorkerData *)src_worker->data1; + FrameWorkerData *const dst_worker_data = (FrameWorkerData *)dst_worker->data1; + VP10_COMMON *const src_cm = &src_worker_data->pbi->common; + VP10_COMMON *const dst_cm = &dst_worker_data->pbi->common; + int i; + + // Wait until source frame's context is ready. + vp10_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; + vp10_frameworker_unlock_stats(src_worker); + + dst_cm->bit_depth = src_cm->bit_depth; +#if CONFIG_VP9_HIGHBITDEPTH + dst_cm->use_highbitdepth = src_cm->use_highbitdepth; +#endif +#if CONFIG_EXT_REFS +// TODO(zoeliu): To handle parallel decoding +#endif // CONFIG_EXT_REFS + 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->subsampling_x = src_cm->subsampling_x; + dst_cm->subsampling_y = src_cm->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.last_sharpness_level = src_cm->lf.sharpness_level; + dst_cm->lf.filter_level = src_cm->lf.filter_level; + memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, TOTAL_REFS_PER_FRAME); + 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/av1/decoder/dthread.h b/av1/decoder/dthread.h new file mode 100644 index 0000000..ef548b6 --- /dev/null +++ b/av1/decoder/dthread.h
@@ -0,0 +1,74 @@ +/* + * Copyright (c) 2014 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef VP10_DECODER_DTHREAD_H_ +#define VP10_DECODER_DTHREAD_H_ + +#include "./vpx_config.h" +#include "aom_util/vpx_thread.h" +#include "aom/internal/vpx_codec_internal.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct VP10Common; +struct VP10Decoder; + +// WorkerData for the FrameWorker thread. It contains all the information of +// the worker and decode structures for decoding a frame. +typedef struct FrameWorkerData { + struct VP10Decoder *pbi; + const uint8_t *data; + const uint8_t *data_end; + size_t data_size; + void *user_priv; + int result; + 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 vp10_frameworker_lock_stats(VPxWorker *const worker); +void vp10_frameworker_unlock_stats(VPxWorker *const worker); +void vp10_frameworker_signal_stats(VPxWorker *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 vp10_frameworker_wait(VPxWorker *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 vp10_frameworker_broadcast(RefCntBuffer *const buf, int row); + +// Copy necessary decoding context from src worker to dst worker. +void vp10_frameworker_copy_context(VPxWorker *const dst_worker, + VPxWorker *const src_worker); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VP10_DECODER_DTHREAD_H_