michael@0: /* michael@0: * Copyright (c) 2010 The WebM project authors. All Rights Reserved. michael@0: * michael@0: * Use of this source code is governed by a BSD-style license michael@0: * that can be found in the LICENSE file in the root of the source michael@0: * tree. An additional intellectual property rights grant can be found michael@0: * in the file PATENTS. All contributing project authors may michael@0: * be found in the AUTHORS file in the root of the source tree. michael@0: */ michael@0: michael@0: #include michael@0: michael@0: #include "./vp9_rtcd.h" michael@0: #include "vpx_mem/vpx_mem.h" michael@0: #include "vpx_scale/vpx_scale.h" michael@0: michael@0: #include "vp9/common/vp9_alloccommon.h" michael@0: #include "vp9/common/vp9_common.h" michael@0: #include "vp9/common/vp9_entropy.h" michael@0: #include "vp9/common/vp9_entropymode.h" michael@0: #include "vp9/common/vp9_extend.h" michael@0: #include "vp9/common/vp9_idct.h" michael@0: #include "vp9/common/vp9_pred_common.h" michael@0: #include "vp9/common/vp9_quant_common.h" michael@0: #include "vp9/common/vp9_reconintra.h" michael@0: #include "vp9/common/vp9_reconinter.h" michael@0: #include "vp9/common/vp9_seg_common.h" michael@0: #include "vp9/common/vp9_tile_common.h" michael@0: michael@0: #include "vp9/decoder/vp9_dboolhuff.h" michael@0: #include "vp9/decoder/vp9_decodframe.h" michael@0: #include "vp9/decoder/vp9_detokenize.h" michael@0: #include "vp9/decoder/vp9_decodemv.h" michael@0: #include "vp9/decoder/vp9_dsubexp.h" michael@0: #include "vp9/decoder/vp9_onyxd_int.h" michael@0: #include "vp9/decoder/vp9_read_bit_buffer.h" michael@0: #include "vp9/decoder/vp9_thread.h" michael@0: #include "vp9/decoder/vp9_treereader.h" michael@0: michael@0: typedef struct TileWorkerData { michael@0: VP9_COMMON *cm; michael@0: vp9_reader bit_reader; michael@0: DECLARE_ALIGNED(16, MACROBLOCKD, xd); michael@0: DECLARE_ALIGNED(16, unsigned char, token_cache[1024]); michael@0: DECLARE_ALIGNED(16, int16_t, qcoeff[MAX_MB_PLANE][64 * 64]); michael@0: DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]); michael@0: DECLARE_ALIGNED(16, uint16_t, eobs[MAX_MB_PLANE][256]); michael@0: } TileWorkerData; michael@0: michael@0: static int read_be32(const uint8_t *p) { michael@0: return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3]; michael@0: } michael@0: michael@0: static int is_compound_prediction_allowed(const VP9_COMMON *cm) { michael@0: int i; michael@0: for (i = 1; i < ALLOWED_REFS_PER_FRAME; ++i) michael@0: if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1]) michael@0: return 1; michael@0: michael@0: return 0; michael@0: } michael@0: michael@0: static void setup_compound_prediction(VP9_COMMON *cm) { michael@0: if (cm->ref_frame_sign_bias[LAST_FRAME] == michael@0: cm->ref_frame_sign_bias[GOLDEN_FRAME]) { michael@0: cm->comp_fixed_ref = ALTREF_FRAME; michael@0: cm->comp_var_ref[0] = LAST_FRAME; michael@0: cm->comp_var_ref[1] = GOLDEN_FRAME; michael@0: } else if (cm->ref_frame_sign_bias[LAST_FRAME] == michael@0: cm->ref_frame_sign_bias[ALTREF_FRAME]) { michael@0: cm->comp_fixed_ref = GOLDEN_FRAME; michael@0: cm->comp_var_ref[0] = LAST_FRAME; michael@0: cm->comp_var_ref[1] = ALTREF_FRAME; michael@0: } else { michael@0: cm->comp_fixed_ref = LAST_FRAME; michael@0: cm->comp_var_ref[0] = GOLDEN_FRAME; michael@0: cm->comp_var_ref[1] = ALTREF_FRAME; michael@0: } michael@0: } michael@0: michael@0: // len == 0 is not allowed michael@0: static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) { michael@0: return start + len > start && start + len <= end; michael@0: } michael@0: michael@0: static int decode_unsigned_max(struct vp9_read_bit_buffer *rb, int max) { michael@0: const int data = vp9_rb_read_literal(rb, get_unsigned_bits(max)); michael@0: return data > max ? max : data; michael@0: } michael@0: michael@0: static TX_MODE read_tx_mode(vp9_reader *r) { michael@0: TX_MODE tx_mode = vp9_read_literal(r, 2); michael@0: if (tx_mode == ALLOW_32X32) michael@0: tx_mode += vp9_read_bit(r); michael@0: return tx_mode; michael@0: } michael@0: michael@0: static void read_tx_probs(struct tx_probs *tx_probs, vp9_reader *r) { michael@0: int i, j; michael@0: michael@0: for (i = 0; i < TX_SIZE_CONTEXTS; ++i) michael@0: for (j = 0; j < TX_SIZES - 3; ++j) michael@0: vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]); michael@0: michael@0: for (i = 0; i < TX_SIZE_CONTEXTS; ++i) michael@0: for (j = 0; j < TX_SIZES - 2; ++j) michael@0: vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]); michael@0: michael@0: for (i = 0; i < TX_SIZE_CONTEXTS; ++i) michael@0: for (j = 0; j < TX_SIZES - 1; ++j) michael@0: vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]); michael@0: } michael@0: michael@0: static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vp9_reader *r) { michael@0: int i, j; michael@0: for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) michael@0: for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i) michael@0: vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]); michael@0: } michael@0: michael@0: static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp9_reader *r) { michael@0: int i, j; michael@0: for (i = 0; i < INTER_MODE_CONTEXTS; ++i) michael@0: for (j = 0; j < INTER_MODES - 1; ++j) michael@0: vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]); michael@0: } michael@0: michael@0: static INLINE COMPPREDMODE_TYPE read_comp_pred_mode(vp9_reader *r) { michael@0: COMPPREDMODE_TYPE mode = vp9_read_bit(r); michael@0: if (mode) michael@0: mode += vp9_read_bit(r); michael@0: return mode; michael@0: } michael@0: michael@0: static void read_comp_pred(VP9_COMMON *cm, vp9_reader *r) { michael@0: int i; michael@0: michael@0: const int compound_allowed = is_compound_prediction_allowed(cm); michael@0: cm->comp_pred_mode = compound_allowed ? read_comp_pred_mode(r) michael@0: : SINGLE_PREDICTION_ONLY; michael@0: if (compound_allowed) michael@0: setup_compound_prediction(cm); michael@0: michael@0: if (cm->comp_pred_mode == HYBRID_PREDICTION) michael@0: for (i = 0; i < COMP_INTER_CONTEXTS; i++) michael@0: vp9_diff_update_prob(r, &cm->fc.comp_inter_prob[i]); michael@0: michael@0: if (cm->comp_pred_mode != COMP_PREDICTION_ONLY) michael@0: for (i = 0; i < REF_CONTEXTS; i++) { michael@0: vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][0]); michael@0: vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][1]); michael@0: } michael@0: michael@0: if (cm->comp_pred_mode != SINGLE_PREDICTION_ONLY) michael@0: for (i = 0; i < REF_CONTEXTS; i++) michael@0: vp9_diff_update_prob(r, &cm->fc.comp_ref_prob[i]); michael@0: } michael@0: michael@0: static void update_mv_probs(vp9_prob *p, int n, vp9_reader *r) { michael@0: int i; michael@0: for (i = 0; i < n; ++i) michael@0: if (vp9_read(r, NMV_UPDATE_PROB)) michael@0: p[i] = (vp9_read_literal(r, 7) << 1) | 1; michael@0: } michael@0: michael@0: static void read_mv_probs(nmv_context *ctx, int allow_hp, vp9_reader *r) { michael@0: int i, j; michael@0: michael@0: update_mv_probs(ctx->joints, MV_JOINTS - 1, r); michael@0: michael@0: for (i = 0; i < 2; ++i) { michael@0: nmv_component *const comp_ctx = &ctx->comps[i]; michael@0: update_mv_probs(&comp_ctx->sign, 1, r); michael@0: update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r); michael@0: update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r); michael@0: update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r); michael@0: } michael@0: michael@0: for (i = 0; i < 2; ++i) { michael@0: nmv_component *const comp_ctx = &ctx->comps[i]; michael@0: for (j = 0; j < CLASS0_SIZE; ++j) michael@0: update_mv_probs(comp_ctx->class0_fp[j], 3, r); michael@0: update_mv_probs(comp_ctx->fp, 3, r); michael@0: } michael@0: michael@0: if (allow_hp) { michael@0: for (i = 0; i < 2; ++i) { michael@0: nmv_component *const comp_ctx = &ctx->comps[i]; michael@0: update_mv_probs(&comp_ctx->class0_hp, 1, r); michael@0: update_mv_probs(&comp_ctx->hp, 1, r); michael@0: } michael@0: } michael@0: } michael@0: michael@0: static void setup_plane_dequants(VP9_COMMON *cm, MACROBLOCKD *xd, int q_index) { michael@0: int i; michael@0: xd->plane[0].dequant = cm->y_dequant[q_index]; michael@0: michael@0: for (i = 1; i < MAX_MB_PLANE; i++) michael@0: xd->plane[i].dequant = cm->uv_dequant[q_index]; michael@0: } michael@0: michael@0: // Allocate storage for each tile column. michael@0: // TODO(jzern): when max_threads <= 1 the same storage could be used for each michael@0: // tile. michael@0: static void alloc_tile_storage(VP9D_COMP *pbi, int tile_rows, int tile_cols) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols); michael@0: int i, tile_row, tile_col; michael@0: michael@0: CHECK_MEM_ERROR(cm, pbi->mi_streams, michael@0: vpx_realloc(pbi->mi_streams, tile_rows * tile_cols * michael@0: sizeof(*pbi->mi_streams))); michael@0: for (tile_row = 0; tile_row < tile_rows; ++tile_row) { michael@0: for (tile_col = 0; tile_col < tile_cols; ++tile_col) { michael@0: TileInfo tile; michael@0: vp9_tile_init(&tile, cm, tile_row, tile_col); michael@0: pbi->mi_streams[tile_row * tile_cols + tile_col] = michael@0: &cm->mi[tile.mi_row_start * cm->mode_info_stride michael@0: + tile.mi_col_start]; michael@0: } michael@0: } michael@0: michael@0: // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm michael@0: // block where mi unit size is 8x8. michael@0: CHECK_MEM_ERROR(cm, pbi->above_context[0], michael@0: vpx_realloc(pbi->above_context[0], michael@0: sizeof(*pbi->above_context[0]) * MAX_MB_PLANE * michael@0: 2 * aligned_mi_cols)); michael@0: for (i = 1; i < MAX_MB_PLANE; ++i) { michael@0: pbi->above_context[i] = pbi->above_context[0] + michael@0: i * sizeof(*pbi->above_context[0]) * michael@0: 2 * aligned_mi_cols; michael@0: } michael@0: michael@0: // This is sized based on the entire frame. Each tile operates within its michael@0: // column bounds. michael@0: CHECK_MEM_ERROR(cm, pbi->above_seg_context, michael@0: vpx_realloc(pbi->above_seg_context, michael@0: sizeof(*pbi->above_seg_context) * michael@0: aligned_mi_cols)); michael@0: } michael@0: michael@0: static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block, michael@0: TX_SIZE tx_size, int x, int y) { michael@0: struct macroblockd_plane *const pd = &xd->plane[plane]; michael@0: const int eob = pd->eobs[block]; michael@0: if (eob > 0) { michael@0: TX_TYPE tx_type; michael@0: const int plane_type = pd->plane_type; michael@0: const int stride = pd->dst.stride; michael@0: int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block); michael@0: uint8_t *const dst = &pd->dst.buf[4 * y * stride + 4 * x]; michael@0: michael@0: switch (tx_size) { michael@0: case TX_4X4: michael@0: tx_type = get_tx_type_4x4(plane_type, xd, block); michael@0: if (tx_type == DCT_DCT) michael@0: xd->itxm_add(dqcoeff, dst, stride, eob); michael@0: else michael@0: vp9_iht4x4_16_add(dqcoeff, dst, stride, tx_type); michael@0: break; michael@0: case TX_8X8: michael@0: tx_type = get_tx_type_8x8(plane_type, xd); michael@0: vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob); michael@0: break; michael@0: case TX_16X16: michael@0: tx_type = get_tx_type_16x16(plane_type, xd); michael@0: vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob); michael@0: break; michael@0: case TX_32X32: michael@0: tx_type = DCT_DCT; michael@0: vp9_idct32x32_add(dqcoeff, dst, stride, eob); michael@0: break; michael@0: default: michael@0: assert(!"Invalid transform size"); michael@0: } michael@0: michael@0: if (eob == 1) { michael@0: vpx_memset(dqcoeff, 0, 2 * sizeof(dqcoeff[0])); michael@0: } else { michael@0: if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10) michael@0: vpx_memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0])); michael@0: else if (tx_size == TX_32X32 && eob <= 34) michael@0: vpx_memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0])); michael@0: else michael@0: vpx_memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0])); michael@0: } michael@0: } michael@0: } michael@0: michael@0: struct intra_args { michael@0: VP9_COMMON *cm; michael@0: MACROBLOCKD *xd; michael@0: vp9_reader *r; michael@0: uint8_t *token_cache; michael@0: }; michael@0: michael@0: static void predict_and_reconstruct_intra_block(int plane, int block, michael@0: BLOCK_SIZE plane_bsize, michael@0: TX_SIZE tx_size, void *arg) { michael@0: struct intra_args *const args = arg; michael@0: VP9_COMMON *const cm = args->cm; michael@0: MACROBLOCKD *const xd = args->xd; michael@0: struct macroblockd_plane *const pd = &xd->plane[plane]; michael@0: MODE_INFO *const mi = xd->mi_8x8[0]; michael@0: const MB_PREDICTION_MODE mode = (plane == 0) michael@0: ? ((mi->mbmi.sb_type < BLOCK_8X8) ? mi->bmi[block].as_mode michael@0: : mi->mbmi.mode) michael@0: : mi->mbmi.uv_mode; michael@0: int x, y; michael@0: uint8_t *dst; michael@0: txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y); michael@0: dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x]; michael@0: michael@0: if (xd->mb_to_right_edge < 0 || xd->mb_to_bottom_edge < 0) michael@0: extend_for_intra(xd, plane_bsize, plane, block, tx_size); michael@0: michael@0: vp9_predict_intra_block(xd, block >> (tx_size << 1), michael@0: b_width_log2(plane_bsize), tx_size, mode, michael@0: dst, pd->dst.stride, dst, pd->dst.stride); michael@0: michael@0: if (!mi->mbmi.skip_coeff) { michael@0: vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize, x, y, tx_size, michael@0: args->r, args->token_cache); michael@0: inverse_transform_block(xd, plane, block, tx_size, x, y); michael@0: } michael@0: } michael@0: michael@0: struct inter_args { michael@0: VP9_COMMON *cm; michael@0: MACROBLOCKD *xd; michael@0: vp9_reader *r; michael@0: int *eobtotal; michael@0: uint8_t *token_cache; michael@0: }; michael@0: michael@0: static void reconstruct_inter_block(int plane, int block, michael@0: BLOCK_SIZE plane_bsize, michael@0: TX_SIZE tx_size, void *arg) { michael@0: struct inter_args *args = arg; michael@0: VP9_COMMON *const cm = args->cm; michael@0: MACROBLOCKD *const xd = args->xd; michael@0: int x, y; michael@0: txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y); michael@0: michael@0: *args->eobtotal += vp9_decode_block_tokens(cm, xd, plane, block, michael@0: plane_bsize, x, y, tx_size, michael@0: args->r, args->token_cache); michael@0: inverse_transform_block(xd, plane, block, tx_size, x, y); michael@0: } michael@0: michael@0: static void set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd, michael@0: const TileInfo *const tile, michael@0: BLOCK_SIZE bsize, int mi_row, int mi_col) { michael@0: const int bh = num_8x8_blocks_high_lookup[bsize]; michael@0: const int bw = num_8x8_blocks_wide_lookup[bsize]; michael@0: const int offset = mi_row * cm->mode_info_stride + mi_col; michael@0: const int tile_offset = tile->mi_row_start * cm->mode_info_stride + michael@0: tile->mi_col_start; michael@0: michael@0: xd->mi_8x8 = cm->mi_grid_visible + offset; michael@0: xd->prev_mi_8x8 = cm->prev_mi_grid_visible + offset; michael@0: michael@0: // we are using the mode info context stream here michael@0: xd->mi_8x8[0] = xd->mi_stream + offset - tile_offset; michael@0: xd->mi_8x8[0]->mbmi.sb_type = bsize; michael@0: michael@0: // Special case: if prev_mi is NULL, the previous mode info context michael@0: // cannot be used. michael@0: xd->last_mi = cm->prev_mi ? xd->prev_mi_8x8[0] : NULL; michael@0: michael@0: set_skip_context(xd, xd->above_context, xd->left_context, mi_row, mi_col); michael@0: michael@0: // Distance of Mb to the various image edges. These are specified to 8th pel michael@0: // as they are always compared to values that are in 1/8th pel units michael@0: set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols); michael@0: michael@0: setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col); michael@0: } michael@0: michael@0: static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd, michael@0: int idx, int mi_row, int mi_col) { michael@0: MB_MODE_INFO *const mbmi = &xd->mi_8x8[0]->mbmi; michael@0: const int ref = mbmi->ref_frame[idx] - LAST_FRAME; michael@0: const YV12_BUFFER_CONFIG *cfg = get_frame_ref_buffer(cm, ref); michael@0: const struct scale_factors_common *sfc = &cm->active_ref_scale_comm[ref]; michael@0: if (!vp9_is_valid_scale(sfc)) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, michael@0: "Invalid scale factors"); michael@0: michael@0: xd->scale_factor[idx].sfc = sfc; michael@0: setup_pre_planes(xd, idx, cfg, mi_row, mi_col, &xd->scale_factor[idx]); michael@0: xd->corrupted |= cfg->corrupted; michael@0: } michael@0: michael@0: static void decode_modes_b(VP9_COMMON *const cm, MACROBLOCKD *const xd, michael@0: const TileInfo *const tile, michael@0: int mi_row, int mi_col, michael@0: vp9_reader *r, BLOCK_SIZE bsize, michael@0: uint8_t *token_cache) { michael@0: const int less8x8 = bsize < BLOCK_8X8; michael@0: MB_MODE_INFO *mbmi; michael@0: michael@0: set_offsets(cm, xd, tile, bsize, mi_row, mi_col); michael@0: vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r); michael@0: michael@0: if (less8x8) michael@0: bsize = BLOCK_8X8; michael@0: michael@0: // Has to be called after set_offsets michael@0: mbmi = &xd->mi_8x8[0]->mbmi; michael@0: michael@0: if (mbmi->skip_coeff) { michael@0: reset_skip_context(xd, bsize); michael@0: } else { michael@0: if (cm->seg.enabled) michael@0: setup_plane_dequants(cm, xd, vp9_get_qindex(&cm->seg, mbmi->segment_id, michael@0: cm->base_qindex)); michael@0: } michael@0: michael@0: if (!is_inter_block(mbmi)) { michael@0: struct intra_args arg = { michael@0: cm, xd, r, token_cache michael@0: }; michael@0: foreach_transformed_block(xd, bsize, predict_and_reconstruct_intra_block, michael@0: &arg); michael@0: } else { michael@0: // Setup michael@0: set_ref(cm, xd, 0, mi_row, mi_col); michael@0: if (has_second_ref(mbmi)) michael@0: set_ref(cm, xd, 1, mi_row, mi_col); michael@0: michael@0: xd->subpix.filter_x = xd->subpix.filter_y = michael@0: vp9_get_filter_kernel(mbmi->interp_filter); michael@0: michael@0: // Prediction michael@0: vp9_build_inter_predictors_sb(xd, mi_row, mi_col, bsize); michael@0: michael@0: // Reconstruction michael@0: if (!mbmi->skip_coeff) { michael@0: int eobtotal = 0; michael@0: struct inter_args arg = { michael@0: cm, xd, r, &eobtotal, token_cache michael@0: }; michael@0: foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg); michael@0: if (!less8x8 && eobtotal == 0) michael@0: mbmi->skip_coeff = 1; // skip loopfilter michael@0: } michael@0: } michael@0: michael@0: xd->corrupted |= vp9_reader_has_error(r); michael@0: } michael@0: michael@0: static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd, int hbs, michael@0: int mi_row, int mi_col, BLOCK_SIZE bsize, michael@0: vp9_reader *r) { michael@0: const int ctx = partition_plane_context(xd->above_seg_context, michael@0: xd->left_seg_context, michael@0: mi_row, mi_col, bsize); michael@0: const vp9_prob *const probs = get_partition_probs(cm, ctx); michael@0: const int has_rows = (mi_row + hbs) < cm->mi_rows; michael@0: const int has_cols = (mi_col + hbs) < cm->mi_cols; michael@0: PARTITION_TYPE p; michael@0: michael@0: if (has_rows && has_cols) michael@0: p = treed_read(r, vp9_partition_tree, probs); michael@0: else if (!has_rows && has_cols) michael@0: p = vp9_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ; michael@0: else if (has_rows && !has_cols) michael@0: p = vp9_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT; michael@0: else michael@0: p = PARTITION_SPLIT; michael@0: michael@0: if (!cm->frame_parallel_decoding_mode) michael@0: ++cm->counts.partition[ctx][p]; michael@0: michael@0: return p; michael@0: } michael@0: michael@0: static void decode_modes_sb(VP9_COMMON *const cm, MACROBLOCKD *const xd, michael@0: const TileInfo *const tile, michael@0: int mi_row, int mi_col, michael@0: vp9_reader* r, BLOCK_SIZE bsize, michael@0: uint8_t *token_cache) { michael@0: const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2; michael@0: PARTITION_TYPE partition; michael@0: BLOCK_SIZE subsize; michael@0: michael@0: if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) michael@0: return; michael@0: michael@0: partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r); michael@0: subsize = get_subsize(bsize, partition); michael@0: if (subsize < BLOCK_8X8) { michael@0: decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache); michael@0: } else { michael@0: switch (partition) { michael@0: case PARTITION_NONE: michael@0: decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache); michael@0: break; michael@0: case PARTITION_HORZ: michael@0: decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache); michael@0: if (mi_row + hbs < cm->mi_rows) michael@0: decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize, michael@0: token_cache); michael@0: break; michael@0: case PARTITION_VERT: michael@0: decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache); michael@0: if (mi_col + hbs < cm->mi_cols) michael@0: decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize, michael@0: token_cache); michael@0: break; michael@0: case PARTITION_SPLIT: michael@0: decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, subsize, michael@0: token_cache); michael@0: decode_modes_sb(cm, xd, tile, mi_row, mi_col + hbs, r, subsize, michael@0: token_cache); michael@0: decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col, r, subsize, michael@0: token_cache); michael@0: decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize, michael@0: token_cache); michael@0: break; michael@0: default: michael@0: assert(!"Invalid partition type"); michael@0: } michael@0: } michael@0: michael@0: // update partition context michael@0: if (bsize >= BLOCK_8X8 && michael@0: (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT)) michael@0: update_partition_context(xd->above_seg_context, xd->left_seg_context, michael@0: mi_row, mi_col, subsize, bsize); michael@0: } michael@0: michael@0: static void setup_token_decoder(const uint8_t *data, michael@0: const uint8_t *data_end, michael@0: size_t read_size, michael@0: struct vpx_internal_error_info *error_info, michael@0: vp9_reader *r) { michael@0: // Validate the calculated partition length. If the buffer michael@0: // described by the partition can't be fully read, then restrict michael@0: // it to the portion that can be (for EC mode) or throw an error. michael@0: if (!read_is_valid(data, read_size, data_end)) michael@0: vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, michael@0: "Truncated packet or corrupt tile length"); michael@0: michael@0: if (vp9_reader_init(r, data, read_size)) michael@0: vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR, michael@0: "Failed to allocate bool decoder %d", 1); michael@0: } michael@0: michael@0: static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs, michael@0: vp9_reader *r) { michael@0: int i, j, k, l, m; michael@0: michael@0: if (vp9_read_bit(r)) michael@0: for (i = 0; i < BLOCK_TYPES; i++) michael@0: for (j = 0; j < REF_TYPES; j++) michael@0: for (k = 0; k < COEF_BANDS; k++) michael@0: for (l = 0; l < PREV_COEF_CONTEXTS; l++) michael@0: if (k > 0 || l < 3) michael@0: for (m = 0; m < UNCONSTRAINED_NODES; m++) michael@0: vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]); michael@0: } michael@0: michael@0: static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode, michael@0: vp9_reader *r) { michael@0: const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode]; michael@0: TX_SIZE tx_size; michael@0: for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size) michael@0: read_coef_probs_common(fc->coef_probs[tx_size], r); michael@0: } michael@0: michael@0: static void setup_segmentation(struct segmentation *seg, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: int i, j; michael@0: michael@0: seg->update_map = 0; michael@0: seg->update_data = 0; michael@0: michael@0: seg->enabled = vp9_rb_read_bit(rb); michael@0: if (!seg->enabled) michael@0: return; michael@0: michael@0: // Segmentation map update michael@0: seg->update_map = vp9_rb_read_bit(rb); michael@0: if (seg->update_map) { michael@0: for (i = 0; i < SEG_TREE_PROBS; i++) michael@0: seg->tree_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8) michael@0: : MAX_PROB; michael@0: michael@0: seg->temporal_update = vp9_rb_read_bit(rb); michael@0: if (seg->temporal_update) { michael@0: for (i = 0; i < PREDICTION_PROBS; i++) michael@0: seg->pred_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8) michael@0: : MAX_PROB; michael@0: } else { michael@0: for (i = 0; i < PREDICTION_PROBS; i++) michael@0: seg->pred_probs[i] = MAX_PROB; michael@0: } michael@0: } michael@0: michael@0: // Segmentation data update michael@0: seg->update_data = vp9_rb_read_bit(rb); michael@0: if (seg->update_data) { michael@0: seg->abs_delta = vp9_rb_read_bit(rb); michael@0: michael@0: vp9_clearall_segfeatures(seg); michael@0: michael@0: for (i = 0; i < MAX_SEGMENTS; i++) { michael@0: for (j = 0; j < SEG_LVL_MAX; j++) { michael@0: int data = 0; michael@0: const int feature_enabled = vp9_rb_read_bit(rb); michael@0: if (feature_enabled) { michael@0: vp9_enable_segfeature(seg, i, j); michael@0: data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j)); michael@0: if (vp9_is_segfeature_signed(j)) michael@0: data = vp9_rb_read_bit(rb) ? -data : data; michael@0: } michael@0: vp9_set_segdata(seg, i, j, data); michael@0: } michael@0: } michael@0: } michael@0: } michael@0: michael@0: static void setup_loopfilter(struct loopfilter *lf, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: lf->filter_level = vp9_rb_read_literal(rb, 6); michael@0: lf->sharpness_level = vp9_rb_read_literal(rb, 3); michael@0: michael@0: // Read in loop filter deltas applied at the MB level based on mode or ref michael@0: // frame. michael@0: lf->mode_ref_delta_update = 0; michael@0: michael@0: lf->mode_ref_delta_enabled = vp9_rb_read_bit(rb); michael@0: if (lf->mode_ref_delta_enabled) { michael@0: lf->mode_ref_delta_update = vp9_rb_read_bit(rb); michael@0: if (lf->mode_ref_delta_update) { michael@0: int i; michael@0: michael@0: for (i = 0; i < MAX_REF_LF_DELTAS; i++) michael@0: if (vp9_rb_read_bit(rb)) michael@0: lf->ref_deltas[i] = vp9_rb_read_signed_literal(rb, 6); michael@0: michael@0: for (i = 0; i < MAX_MODE_LF_DELTAS; i++) michael@0: if (vp9_rb_read_bit(rb)) michael@0: lf->mode_deltas[i] = vp9_rb_read_signed_literal(rb, 6); michael@0: } michael@0: } michael@0: } michael@0: michael@0: static int read_delta_q(struct vp9_read_bit_buffer *rb, int *delta_q) { michael@0: const int old = *delta_q; michael@0: *delta_q = vp9_rb_read_bit(rb) ? vp9_rb_read_signed_literal(rb, 4) : 0; michael@0: return old != *delta_q; michael@0: } michael@0: michael@0: static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: int update = 0; michael@0: michael@0: cm->base_qindex = vp9_rb_read_literal(rb, QINDEX_BITS); michael@0: update |= read_delta_q(rb, &cm->y_dc_delta_q); michael@0: update |= read_delta_q(rb, &cm->uv_dc_delta_q); michael@0: update |= read_delta_q(rb, &cm->uv_ac_delta_q); michael@0: if (update) michael@0: vp9_init_dequantizer(cm); michael@0: michael@0: xd->lossless = cm->base_qindex == 0 && michael@0: cm->y_dc_delta_q == 0 && michael@0: cm->uv_dc_delta_q == 0 && michael@0: cm->uv_ac_delta_q == 0; michael@0: michael@0: xd->itxm_add = xd->lossless ? vp9_iwht4x4_add : vp9_idct4x4_add; michael@0: } michael@0: michael@0: static INTERPOLATION_TYPE read_interp_filter_type( michael@0: struct vp9_read_bit_buffer *rb) { michael@0: const INTERPOLATION_TYPE literal_to_type[] = { EIGHTTAP_SMOOTH, michael@0: EIGHTTAP, michael@0: EIGHTTAP_SHARP, michael@0: BILINEAR }; michael@0: return vp9_rb_read_bit(rb) ? SWITCHABLE michael@0: : literal_to_type[vp9_rb_read_literal(rb, 2)]; michael@0: } michael@0: michael@0: static void read_frame_size(struct vp9_read_bit_buffer *rb, michael@0: int *width, int *height) { michael@0: const int w = vp9_rb_read_literal(rb, 16) + 1; michael@0: const int h = vp9_rb_read_literal(rb, 16) + 1; michael@0: *width = w; michael@0: *height = h; michael@0: } michael@0: michael@0: static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) { michael@0: cm->display_width = cm->width; michael@0: cm->display_height = cm->height; michael@0: if (vp9_rb_read_bit(rb)) michael@0: read_frame_size(rb, &cm->display_width, &cm->display_height); michael@0: } michael@0: michael@0: static void apply_frame_size(VP9D_COMP *pbi, int width, int height) { michael@0: VP9_COMMON *cm = &pbi->common; michael@0: michael@0: if (cm->width != width || cm->height != height) { michael@0: // Change in frame size. michael@0: if (cm->width == 0 || cm->height == 0) { michael@0: // Assign new frame buffer on first call. michael@0: cm->new_fb_idx = NUM_YV12_BUFFERS - 1; michael@0: cm->fb_idx_ref_cnt[cm->new_fb_idx] = 1; michael@0: } michael@0: michael@0: // TODO(agrange) Don't test width/height, check overall size. michael@0: if (width > cm->width || height > cm->height) { michael@0: // Rescale frame buffers only if they're not big enough already. michael@0: if (vp9_resize_frame_buffers(cm, width, height)) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, michael@0: "Failed to allocate frame buffers"); michael@0: } michael@0: michael@0: cm->width = width; michael@0: cm->height = height; michael@0: michael@0: vp9_update_frame_size(cm); michael@0: } michael@0: michael@0: vp9_realloc_frame_buffer(get_frame_new_buffer(cm), cm->width, cm->height, michael@0: cm->subsampling_x, cm->subsampling_y, michael@0: VP9BORDERINPIXELS); michael@0: } michael@0: michael@0: static void setup_frame_size(VP9D_COMP *pbi, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: int width, height; michael@0: read_frame_size(rb, &width, &height); michael@0: apply_frame_size(pbi, width, height); michael@0: setup_display_size(&pbi->common, rb); michael@0: } michael@0: michael@0: static void setup_frame_size_with_refs(VP9D_COMP *pbi, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: michael@0: int width, height; michael@0: int found = 0, i; michael@0: for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) { michael@0: if (vp9_rb_read_bit(rb)) { michael@0: YV12_BUFFER_CONFIG *const cfg = get_frame_ref_buffer(cm, i); michael@0: width = cfg->y_crop_width; michael@0: height = cfg->y_crop_height; michael@0: found = 1; michael@0: break; michael@0: } michael@0: } michael@0: michael@0: if (!found) michael@0: read_frame_size(rb, &width, &height); michael@0: michael@0: if (!width || !height) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, michael@0: "Referenced frame with invalid size"); michael@0: michael@0: apply_frame_size(pbi, width, height); michael@0: setup_display_size(cm, rb); michael@0: } michael@0: michael@0: static void setup_tile_context(VP9D_COMP *const pbi, MACROBLOCKD *const xd, michael@0: int tile_row, int tile_col) { michael@0: int i; michael@0: const int tile_cols = 1 << pbi->common.log2_tile_cols; michael@0: xd->mi_stream = pbi->mi_streams[tile_row * tile_cols + tile_col]; michael@0: michael@0: for (i = 0; i < MAX_MB_PLANE; ++i) { michael@0: xd->above_context[i] = pbi->above_context[i]; michael@0: } michael@0: // see note in alloc_tile_storage(). michael@0: xd->above_seg_context = pbi->above_seg_context; michael@0: } michael@0: michael@0: static void decode_tile(VP9D_COMP *pbi, const TileInfo *const tile, michael@0: vp9_reader *r) { michael@0: const int num_threads = pbi->oxcf.max_threads; michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: int mi_row, mi_col; michael@0: MACROBLOCKD *xd = &pbi->mb; michael@0: michael@0: if (pbi->do_loopfilter_inline) { michael@0: LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; michael@0: lf_data->frame_buffer = get_frame_new_buffer(cm); michael@0: lf_data->cm = cm; michael@0: lf_data->xd = pbi->mb; michael@0: lf_data->stop = 0; michael@0: lf_data->y_only = 0; michael@0: vp9_loop_filter_frame_init(cm, cm->lf.filter_level); michael@0: } michael@0: michael@0: for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end; michael@0: mi_row += MI_BLOCK_SIZE) { michael@0: // For a SB there are 2 left contexts, each pertaining to a MB row within michael@0: vp9_zero(xd->left_context); michael@0: vp9_zero(xd->left_seg_context); michael@0: for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end; michael@0: mi_col += MI_BLOCK_SIZE) { michael@0: decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64, michael@0: pbi->token_cache); michael@0: } michael@0: michael@0: if (pbi->do_loopfilter_inline) { michael@0: const int lf_start = mi_row - MI_BLOCK_SIZE; michael@0: LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; michael@0: michael@0: // delay the loopfilter by 1 macroblock row. michael@0: if (lf_start < 0) continue; michael@0: michael@0: // decoding has completed: finish up the loop filter in this thread. michael@0: if (mi_row + MI_BLOCK_SIZE >= tile->mi_row_end) continue; michael@0: michael@0: vp9_worker_sync(&pbi->lf_worker); michael@0: lf_data->start = lf_start; michael@0: lf_data->stop = mi_row; michael@0: if (num_threads > 1) { michael@0: vp9_worker_launch(&pbi->lf_worker); michael@0: } else { michael@0: vp9_worker_execute(&pbi->lf_worker); michael@0: } michael@0: } michael@0: } michael@0: michael@0: if (pbi->do_loopfilter_inline) { michael@0: LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; michael@0: michael@0: vp9_worker_sync(&pbi->lf_worker); michael@0: lf_data->start = lf_data->stop; michael@0: lf_data->stop = cm->mi_rows; michael@0: vp9_worker_execute(&pbi->lf_worker); michael@0: } michael@0: } michael@0: michael@0: static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) { michael@0: int min_log2_tile_cols, max_log2_tile_cols, max_ones; michael@0: vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols); michael@0: michael@0: // columns michael@0: max_ones = max_log2_tile_cols - min_log2_tile_cols; michael@0: cm->log2_tile_cols = min_log2_tile_cols; michael@0: while (max_ones-- && vp9_rb_read_bit(rb)) michael@0: cm->log2_tile_cols++; michael@0: michael@0: // rows michael@0: cm->log2_tile_rows = vp9_rb_read_bit(rb); michael@0: if (cm->log2_tile_rows) michael@0: cm->log2_tile_rows += vp9_rb_read_bit(rb); michael@0: } michael@0: michael@0: // Reads the next tile returning its size and adjusting '*data' accordingly michael@0: // based on 'is_last'. michael@0: static size_t get_tile(const uint8_t *const data_end, michael@0: int is_last, michael@0: struct vpx_internal_error_info *error_info, michael@0: const uint8_t **data) { michael@0: size_t size; michael@0: michael@0: if (!is_last) { michael@0: if (!read_is_valid(*data, 4, data_end)) michael@0: vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, michael@0: "Truncated packet or corrupt tile length"); michael@0: michael@0: size = read_be32(*data); michael@0: *data += 4; michael@0: michael@0: if (size > data_end - *data) { michael@0: vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME, michael@0: "Truncated packet or corrupt tile size"); michael@0: } michael@0: } else { michael@0: size = data_end - *data; michael@0: } michael@0: return size; michael@0: } michael@0: michael@0: typedef struct TileBuffer { michael@0: const uint8_t *data; michael@0: size_t size; michael@0: } TileBuffer; michael@0: michael@0: static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: MACROBLOCKD *const xd = &pbi->mb; michael@0: const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols); michael@0: const int tile_cols = 1 << cm->log2_tile_cols; michael@0: const int tile_rows = 1 << cm->log2_tile_rows; michael@0: TileBuffer tile_buffers[4][1 << 6]; michael@0: int tile_row, tile_col; michael@0: const uint8_t *const data_end = pbi->source + pbi->source_sz; michael@0: const uint8_t *end = NULL; michael@0: vp9_reader r; michael@0: michael@0: assert(tile_rows <= 4); michael@0: assert(tile_cols <= (1 << 6)); michael@0: michael@0: // Note: this memset assumes above_context[0], [1] and [2] michael@0: // are allocated as part of the same buffer. michael@0: vpx_memset(pbi->above_context[0], 0, michael@0: sizeof(*pbi->above_context[0]) * MAX_MB_PLANE * 2 * aligned_cols); michael@0: michael@0: vpx_memset(pbi->above_seg_context, 0, michael@0: sizeof(*pbi->above_seg_context) * aligned_cols); michael@0: michael@0: // Load tile data into tile_buffers michael@0: for (tile_row = 0; tile_row < tile_rows; ++tile_row) { michael@0: for (tile_col = 0; tile_col < tile_cols; ++tile_col) { michael@0: const int last_tile = tile_row == tile_rows - 1 && michael@0: tile_col == tile_cols - 1; michael@0: const size_t size = get_tile(data_end, last_tile, &cm->error, &data); michael@0: TileBuffer *const buf = &tile_buffers[tile_row][tile_col]; michael@0: buf->data = data; michael@0: buf->size = size; michael@0: data += size; michael@0: } michael@0: } michael@0: michael@0: // Decode tiles using data from tile_buffers michael@0: for (tile_row = 0; tile_row < tile_rows; ++tile_row) { michael@0: for (tile_col = 0; tile_col < tile_cols; ++tile_col) { michael@0: const int col = pbi->oxcf.inv_tile_order ? tile_cols - tile_col - 1 michael@0: : tile_col; michael@0: const int last_tile = tile_row == tile_rows - 1 && michael@0: col == tile_cols - 1; michael@0: const TileBuffer *const buf = &tile_buffers[tile_row][col]; michael@0: TileInfo tile; michael@0: michael@0: vp9_tile_init(&tile, cm, tile_row, col); michael@0: setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &r); michael@0: setup_tile_context(pbi, xd, tile_row, col); michael@0: decode_tile(pbi, &tile, &r); michael@0: michael@0: if (last_tile) michael@0: end = vp9_reader_find_end(&r); michael@0: } michael@0: } michael@0: michael@0: return end; michael@0: } michael@0: michael@0: static void setup_tile_macroblockd(TileWorkerData *const tile_data) { michael@0: MACROBLOCKD *xd = &tile_data->xd; michael@0: struct macroblockd_plane *const pd = xd->plane; michael@0: int i; michael@0: michael@0: for (i = 0; i < MAX_MB_PLANE; ++i) { michael@0: pd[i].qcoeff = tile_data->qcoeff[i]; michael@0: pd[i].dqcoeff = tile_data->dqcoeff[i]; michael@0: pd[i].eobs = tile_data->eobs[i]; michael@0: vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t)); michael@0: } michael@0: } michael@0: michael@0: static int tile_worker_hook(void *arg1, void *arg2) { michael@0: TileWorkerData *const tile_data = (TileWorkerData*)arg1; michael@0: const TileInfo *const tile = (TileInfo*)arg2; michael@0: int mi_row, mi_col; michael@0: michael@0: for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end; michael@0: mi_row += MI_BLOCK_SIZE) { michael@0: vp9_zero(tile_data->xd.left_context); michael@0: vp9_zero(tile_data->xd.left_seg_context); michael@0: for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end; michael@0: mi_col += MI_BLOCK_SIZE) { michael@0: decode_modes_sb(tile_data->cm, &tile_data->xd, tile, michael@0: mi_row, mi_col, &tile_data->bit_reader, BLOCK_64X64, michael@0: tile_data->token_cache); michael@0: } michael@0: } michael@0: return !tile_data->xd.corrupted; michael@0: } michael@0: michael@0: static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi, const uint8_t *data) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: const uint8_t *const data_end = pbi->source + pbi->source_sz; michael@0: const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols); michael@0: const int tile_cols = 1 << cm->log2_tile_cols; michael@0: const int tile_rows = 1 << cm->log2_tile_rows; michael@0: const int num_workers = MIN(pbi->oxcf.max_threads & ~1, tile_cols); michael@0: int tile_col = 0; michael@0: michael@0: assert(tile_rows == 1); michael@0: (void)tile_rows; michael@0: michael@0: if (num_workers > pbi->num_tile_workers) { michael@0: int i; michael@0: CHECK_MEM_ERROR(cm, pbi->tile_workers, michael@0: vpx_realloc(pbi->tile_workers, michael@0: num_workers * sizeof(*pbi->tile_workers))); michael@0: for (i = pbi->num_tile_workers; i < num_workers; ++i) { michael@0: VP9Worker *const worker = &pbi->tile_workers[i]; michael@0: ++pbi->num_tile_workers; michael@0: michael@0: vp9_worker_init(worker); michael@0: worker->hook = (VP9WorkerHook)tile_worker_hook; michael@0: CHECK_MEM_ERROR(cm, worker->data1, michael@0: vpx_memalign(32, sizeof(TileWorkerData))); michael@0: CHECK_MEM_ERROR(cm, worker->data2, vpx_malloc(sizeof(TileInfo))); michael@0: if (i < num_workers - 1 && !vp9_worker_reset(worker)) { michael@0: vpx_internal_error(&cm->error, VPX_CODEC_ERROR, michael@0: "Tile decoder thread creation failed"); michael@0: } michael@0: } michael@0: } michael@0: michael@0: // Note: this memset assumes above_context[0], [1] and [2] michael@0: // are allocated as part of the same buffer. michael@0: vpx_memset(pbi->above_context[0], 0, michael@0: sizeof(*pbi->above_context[0]) * MAX_MB_PLANE * michael@0: 2 * aligned_mi_cols); michael@0: vpx_memset(pbi->above_seg_context, 0, michael@0: sizeof(*pbi->above_seg_context) * aligned_mi_cols); michael@0: michael@0: while (tile_col < tile_cols) { michael@0: int i; michael@0: for (i = 0; i < num_workers && tile_col < tile_cols; ++i) { michael@0: VP9Worker *const worker = &pbi->tile_workers[i]; michael@0: TileWorkerData *const tile_data = (TileWorkerData*)worker->data1; michael@0: TileInfo *const tile = (TileInfo*)worker->data2; michael@0: const size_t size = michael@0: get_tile(data_end, tile_col == tile_cols - 1, &cm->error, &data); michael@0: michael@0: tile_data->cm = cm; michael@0: tile_data->xd = pbi->mb; michael@0: tile_data->xd.corrupted = 0; michael@0: vp9_tile_init(tile, tile_data->cm, 0, tile_col); michael@0: michael@0: setup_token_decoder(data, data_end, size, &cm->error, michael@0: &tile_data->bit_reader); michael@0: setup_tile_context(pbi, &tile_data->xd, 0, tile_col); michael@0: setup_tile_macroblockd(tile_data); michael@0: michael@0: worker->had_error = 0; michael@0: if (i == num_workers - 1 || tile_col == tile_cols - 1) { michael@0: vp9_worker_execute(worker); michael@0: } else { michael@0: vp9_worker_launch(worker); michael@0: } michael@0: michael@0: data += size; michael@0: ++tile_col; michael@0: } michael@0: michael@0: for (; i > 0; --i) { michael@0: VP9Worker *const worker = &pbi->tile_workers[i - 1]; michael@0: pbi->mb.corrupted |= !vp9_worker_sync(worker); michael@0: } michael@0: } michael@0: michael@0: { michael@0: const int final_worker = (tile_cols + num_workers - 1) % num_workers; michael@0: TileWorkerData *const tile_data = michael@0: (TileWorkerData*)pbi->tile_workers[final_worker].data1; michael@0: return vp9_reader_find_end(&tile_data->bit_reader); michael@0: } michael@0: } michael@0: michael@0: static void check_sync_code(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) { michael@0: if (vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_0 || michael@0: vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_1 || michael@0: vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_2) { michael@0: vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, michael@0: "Invalid frame sync code"); michael@0: } michael@0: } michael@0: michael@0: static void error_handler(void *data, size_t bit_offset) { michael@0: VP9_COMMON *const cm = (VP9_COMMON *)data; michael@0: vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet"); michael@0: } michael@0: michael@0: #define RESERVED \ michael@0: if (vp9_rb_read_bit(rb)) \ michael@0: vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, \ michael@0: "Reserved bit must be unset") michael@0: michael@0: static size_t read_uncompressed_header(VP9D_COMP *pbi, michael@0: struct vp9_read_bit_buffer *rb) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: size_t sz; michael@0: int i; michael@0: michael@0: cm->last_frame_type = cm->frame_type; michael@0: michael@0: if (vp9_rb_read_literal(rb, 2) != VP9_FRAME_MARKER) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, michael@0: "Invalid frame marker"); michael@0: michael@0: cm->version = vp9_rb_read_bit(rb); michael@0: RESERVED; michael@0: michael@0: if (vp9_rb_read_bit(rb)) { michael@0: // show an existing frame directly michael@0: int frame_to_show = cm->ref_frame_map[vp9_rb_read_literal(rb, 3)]; michael@0: ref_cnt_fb(cm->fb_idx_ref_cnt, &cm->new_fb_idx, frame_to_show); michael@0: pbi->refresh_frame_flags = 0; michael@0: cm->lf.filter_level = 0; michael@0: return 0; michael@0: } michael@0: michael@0: cm->frame_type = (FRAME_TYPE) vp9_rb_read_bit(rb); michael@0: cm->show_frame = vp9_rb_read_bit(rb); michael@0: cm->error_resilient_mode = vp9_rb_read_bit(rb); michael@0: michael@0: if (cm->frame_type == KEY_FRAME) { michael@0: check_sync_code(cm, rb); michael@0: michael@0: cm->color_space = vp9_rb_read_literal(rb, 3); // colorspace michael@0: if (cm->color_space != SRGB) { michael@0: vp9_rb_read_bit(rb); // [16,235] (including xvycc) vs [0,255] range michael@0: if (cm->version == 1) { michael@0: cm->subsampling_x = vp9_rb_read_bit(rb); michael@0: cm->subsampling_y = vp9_rb_read_bit(rb); michael@0: vp9_rb_read_bit(rb); // has extra plane michael@0: } else { michael@0: cm->subsampling_y = cm->subsampling_x = 1; michael@0: } michael@0: } else { michael@0: if (cm->version == 1) { michael@0: cm->subsampling_y = cm->subsampling_x = 0; michael@0: vp9_rb_read_bit(rb); // has extra plane michael@0: } else { michael@0: vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, michael@0: "RGB not supported in profile 0"); michael@0: } michael@0: } michael@0: michael@0: pbi->refresh_frame_flags = (1 << NUM_REF_FRAMES) - 1; michael@0: michael@0: for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) michael@0: cm->active_ref_idx[i] = cm->new_fb_idx; michael@0: michael@0: setup_frame_size(pbi, rb); michael@0: } else { michael@0: cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb); michael@0: michael@0: cm->reset_frame_context = cm->error_resilient_mode ? michael@0: 0 : vp9_rb_read_literal(rb, 2); michael@0: michael@0: if (cm->intra_only) { michael@0: check_sync_code(cm, rb); michael@0: michael@0: pbi->refresh_frame_flags = vp9_rb_read_literal(rb, NUM_REF_FRAMES); michael@0: setup_frame_size(pbi, rb); michael@0: } else { michael@0: pbi->refresh_frame_flags = vp9_rb_read_literal(rb, NUM_REF_FRAMES); michael@0: michael@0: for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) { michael@0: const int ref = vp9_rb_read_literal(rb, NUM_REF_FRAMES_LOG2); michael@0: cm->active_ref_idx[i] = cm->ref_frame_map[ref]; michael@0: cm->ref_frame_sign_bias[LAST_FRAME + i] = vp9_rb_read_bit(rb); michael@0: } michael@0: michael@0: setup_frame_size_with_refs(pbi, rb); michael@0: michael@0: cm->allow_high_precision_mv = vp9_rb_read_bit(rb); michael@0: cm->mcomp_filter_type = read_interp_filter_type(rb); michael@0: michael@0: for (i = 0; i < ALLOWED_REFS_PER_FRAME; ++i) michael@0: vp9_setup_scale_factors(cm, i); michael@0: } michael@0: } michael@0: michael@0: if (!cm->error_resilient_mode) { michael@0: cm->refresh_frame_context = vp9_rb_read_bit(rb); michael@0: cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb); michael@0: } else { michael@0: cm->refresh_frame_context = 0; michael@0: cm->frame_parallel_decoding_mode = 1; michael@0: } michael@0: michael@0: // This flag will be overridden by the call to vp9_setup_past_independence michael@0: // below, forcing the use of context 0 for those frame types. michael@0: cm->frame_context_idx = vp9_rb_read_literal(rb, NUM_FRAME_CONTEXTS_LOG2); michael@0: michael@0: if (frame_is_intra_only(cm) || cm->error_resilient_mode) michael@0: vp9_setup_past_independence(cm); michael@0: michael@0: setup_loopfilter(&cm->lf, rb); michael@0: setup_quantization(cm, &pbi->mb, rb); michael@0: setup_segmentation(&cm->seg, rb); michael@0: michael@0: setup_tile_info(cm, rb); michael@0: sz = vp9_rb_read_literal(rb, 16); michael@0: michael@0: if (sz == 0) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, michael@0: "Invalid header size"); michael@0: michael@0: return sz; michael@0: } michael@0: michael@0: static int read_compressed_header(VP9D_COMP *pbi, const uint8_t *data, michael@0: size_t partition_size) { michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: MACROBLOCKD *const xd = &pbi->mb; michael@0: FRAME_CONTEXT *const fc = &cm->fc; michael@0: vp9_reader r; michael@0: int k; michael@0: michael@0: if (vp9_reader_init(&r, data, partition_size)) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, michael@0: "Failed to allocate bool decoder 0"); michael@0: michael@0: cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r); michael@0: if (cm->tx_mode == TX_MODE_SELECT) michael@0: read_tx_probs(&fc->tx_probs, &r); michael@0: read_coef_probs(fc, cm->tx_mode, &r); michael@0: michael@0: for (k = 0; k < MBSKIP_CONTEXTS; ++k) michael@0: vp9_diff_update_prob(&r, &fc->mbskip_probs[k]); michael@0: michael@0: if (!frame_is_intra_only(cm)) { michael@0: nmv_context *const nmvc = &fc->nmvc; michael@0: int i, j; michael@0: michael@0: read_inter_mode_probs(fc, &r); michael@0: michael@0: if (cm->mcomp_filter_type == SWITCHABLE) michael@0: read_switchable_interp_probs(fc, &r); michael@0: michael@0: for (i = 0; i < INTRA_INTER_CONTEXTS; i++) michael@0: vp9_diff_update_prob(&r, &fc->intra_inter_prob[i]); michael@0: michael@0: read_comp_pred(cm, &r); michael@0: michael@0: for (j = 0; j < BLOCK_SIZE_GROUPS; j++) michael@0: for (i = 0; i < INTRA_MODES - 1; ++i) michael@0: vp9_diff_update_prob(&r, &fc->y_mode_prob[j][i]); michael@0: michael@0: for (j = 0; j < PARTITION_CONTEXTS; ++j) michael@0: for (i = 0; i < PARTITION_TYPES - 1; ++i) michael@0: vp9_diff_update_prob(&r, &fc->partition_prob[j][i]); michael@0: michael@0: read_mv_probs(nmvc, cm->allow_high_precision_mv, &r); michael@0: } michael@0: michael@0: return vp9_reader_has_error(&r); michael@0: } michael@0: michael@0: void vp9_init_dequantizer(VP9_COMMON *cm) { michael@0: int q; michael@0: michael@0: for (q = 0; q < QINDEX_RANGE; q++) { michael@0: cm->y_dequant[q][0] = vp9_dc_quant(q, cm->y_dc_delta_q); michael@0: cm->y_dequant[q][1] = vp9_ac_quant(q, 0); michael@0: michael@0: cm->uv_dequant[q][0] = vp9_dc_quant(q, cm->uv_dc_delta_q); michael@0: cm->uv_dequant[q][1] = vp9_ac_quant(q, cm->uv_ac_delta_q); michael@0: } michael@0: } michael@0: michael@0: #ifdef NDEBUG michael@0: #define debug_check_frame_counts(cm) (void)0 michael@0: #else // !NDEBUG michael@0: // Counts should only be incremented when frame_parallel_decoding_mode and michael@0: // error_resilient_mode are disabled. michael@0: static void debug_check_frame_counts(const VP9_COMMON *const cm) { michael@0: FRAME_COUNTS zero_counts; michael@0: vp9_zero(zero_counts); michael@0: assert(cm->frame_parallel_decoding_mode || cm->error_resilient_mode); michael@0: assert(!memcmp(cm->counts.y_mode, zero_counts.y_mode, michael@0: sizeof(cm->counts.y_mode))); michael@0: assert(!memcmp(cm->counts.uv_mode, zero_counts.uv_mode, michael@0: sizeof(cm->counts.uv_mode))); michael@0: assert(!memcmp(cm->counts.partition, zero_counts.partition, michael@0: sizeof(cm->counts.partition))); michael@0: assert(!memcmp(cm->counts.coef, zero_counts.coef, michael@0: sizeof(cm->counts.coef))); michael@0: assert(!memcmp(cm->counts.eob_branch, zero_counts.eob_branch, michael@0: sizeof(cm->counts.eob_branch))); michael@0: assert(!memcmp(cm->counts.switchable_interp, zero_counts.switchable_interp, michael@0: sizeof(cm->counts.switchable_interp))); michael@0: assert(!memcmp(cm->counts.inter_mode, zero_counts.inter_mode, michael@0: sizeof(cm->counts.inter_mode))); michael@0: assert(!memcmp(cm->counts.intra_inter, zero_counts.intra_inter, michael@0: sizeof(cm->counts.intra_inter))); michael@0: assert(!memcmp(cm->counts.comp_inter, zero_counts.comp_inter, michael@0: sizeof(cm->counts.comp_inter))); michael@0: assert(!memcmp(cm->counts.single_ref, zero_counts.single_ref, michael@0: sizeof(cm->counts.single_ref))); michael@0: assert(!memcmp(cm->counts.comp_ref, zero_counts.comp_ref, michael@0: sizeof(cm->counts.comp_ref))); michael@0: assert(!memcmp(&cm->counts.tx, &zero_counts.tx, sizeof(cm->counts.tx))); michael@0: assert(!memcmp(cm->counts.mbskip, zero_counts.mbskip, michael@0: sizeof(cm->counts.mbskip))); michael@0: assert(!memcmp(&cm->counts.mv, &zero_counts.mv, sizeof(cm->counts.mv))); michael@0: } michael@0: #endif // NDEBUG michael@0: michael@0: int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) { michael@0: int i; michael@0: VP9_COMMON *const cm = &pbi->common; michael@0: MACROBLOCKD *const xd = &pbi->mb; michael@0: michael@0: const uint8_t *data = pbi->source; michael@0: const uint8_t *const data_end = pbi->source + pbi->source_sz; michael@0: michael@0: struct vp9_read_bit_buffer rb = { data, data_end, 0, cm, error_handler }; michael@0: const size_t first_partition_size = read_uncompressed_header(pbi, &rb); michael@0: const int keyframe = cm->frame_type == KEY_FRAME; michael@0: const int tile_rows = 1 << cm->log2_tile_rows; michael@0: const int tile_cols = 1 << cm->log2_tile_cols; michael@0: YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm); michael@0: michael@0: if (!first_partition_size) { michael@0: // showing a frame directly michael@0: *p_data_end = data + 1; michael@0: return 0; michael@0: } michael@0: michael@0: if (!pbi->decoded_key_frame && !keyframe) michael@0: return -1; michael@0: michael@0: data += vp9_rb_bytes_read(&rb); michael@0: if (!read_is_valid(data, first_partition_size, data_end)) michael@0: vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, michael@0: "Truncated packet or corrupt header length"); michael@0: michael@0: pbi->do_loopfilter_inline = michael@0: (cm->log2_tile_rows | cm->log2_tile_cols) == 0 && cm->lf.filter_level; michael@0: if (pbi->do_loopfilter_inline && pbi->lf_worker.data1 == NULL) { michael@0: CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, vpx_malloc(sizeof(LFWorkerData))); michael@0: pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker; michael@0: if (pbi->oxcf.max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) { michael@0: vpx_internal_error(&cm->error, VPX_CODEC_ERROR, michael@0: "Loop filter thread creation failed"); michael@0: } michael@0: } michael@0: michael@0: alloc_tile_storage(pbi, tile_rows, tile_cols); michael@0: michael@0: xd->mode_info_stride = cm->mode_info_stride; michael@0: set_prev_mi(cm); michael@0: michael@0: setup_plane_dequants(cm, xd, cm->base_qindex); michael@0: setup_block_dptrs(xd, cm->subsampling_x, cm->subsampling_y); michael@0: michael@0: cm->fc = cm->frame_contexts[cm->frame_context_idx]; michael@0: vp9_zero(cm->counts); michael@0: for (i = 0; i < MAX_MB_PLANE; ++i) michael@0: vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t)); michael@0: michael@0: xd->corrupted = 0; michael@0: new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size); michael@0: michael@0: // TODO(jzern): remove frame_parallel_decoding_mode restriction for michael@0: // single-frame tile decoding. michael@0: if (pbi->oxcf.max_threads > 1 && tile_rows == 1 && tile_cols > 1 && michael@0: cm->frame_parallel_decoding_mode) { michael@0: *p_data_end = decode_tiles_mt(pbi, data + first_partition_size); michael@0: } else { michael@0: *p_data_end = decode_tiles(pbi, data + first_partition_size); michael@0: } michael@0: michael@0: cm->last_width = cm->width; michael@0: cm->last_height = cm->height; michael@0: michael@0: new_fb->corrupted |= xd->corrupted; michael@0: michael@0: if (!pbi->decoded_key_frame) { michael@0: if (keyframe && !new_fb->corrupted) michael@0: pbi->decoded_key_frame = 1; michael@0: else michael@0: vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, michael@0: "A stream must start with a complete key frame"); michael@0: } michael@0: michael@0: if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) { michael@0: vp9_adapt_coef_probs(cm); michael@0: michael@0: if (!frame_is_intra_only(cm)) { michael@0: vp9_adapt_mode_probs(cm); michael@0: vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv); michael@0: } michael@0: } else { michael@0: debug_check_frame_counts(cm); michael@0: } michael@0: michael@0: if (cm->refresh_frame_context) michael@0: cm->frame_contexts[cm->frame_context_idx] = cm->fc; michael@0: michael@0: return 0; michael@0: }