2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
12 #include <stdlib.h> // qsort()
14 #include "./vp9_rtcd.h"
15 #include "./vpx_scale_rtcd.h"
17 #include "vpx_mem/vpx_mem.h"
18 #include "vpx_scale/vpx_scale.h"
20 #include "vp9/common/vp9_alloccommon.h"
21 #include "vp9/common/vp9_common.h"
22 #include "vp9/common/vp9_entropy.h"
23 #include "vp9/common/vp9_entropymode.h"
24 #include "vp9/common/vp9_idct.h"
25 #include "vp9/common/vp9_pred_common.h"
26 #include "vp9/common/vp9_quant_common.h"
27 #include "vp9/common/vp9_reconintra.h"
28 #include "vp9/common/vp9_reconinter.h"
29 #include "vp9/common/vp9_seg_common.h"
30 #include "vp9/common/vp9_tile_common.h"
32 #include "vp9/decoder/vp9_decodeframe.h"
33 #include "vp9/decoder/vp9_detokenize.h"
34 #include "vp9/decoder/vp9_decodemv.h"
35 #include "vp9/decoder/vp9_dsubexp.h"
36 #include "vp9/decoder/vp9_onyxd_int.h"
37 #include "vp9/decoder/vp9_read_bit_buffer.h"
38 #include "vp9/decoder/vp9_reader.h"
39 #include "vp9/decoder/vp9_thread.h"
41 typedef struct TileWorkerData {
43 vp9_reader bit_reader;
44 DECLARE_ALIGNED(16, MACROBLOCKD, xd);
45 DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]);
48 static int read_be32(const uint8_t *p) {
49 return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
52 static int is_compound_reference_allowed(const VP9_COMMON *cm) {
54 for (i = 1; i < REFS_PER_FRAME; ++i)
55 if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1])
61 static void setup_compound_reference(VP9_COMMON *cm) {
62 if (cm->ref_frame_sign_bias[LAST_FRAME] ==
63 cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
64 cm->comp_fixed_ref = ALTREF_FRAME;
65 cm->comp_var_ref[0] = LAST_FRAME;
66 cm->comp_var_ref[1] = GOLDEN_FRAME;
67 } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
68 cm->ref_frame_sign_bias[ALTREF_FRAME]) {
69 cm->comp_fixed_ref = GOLDEN_FRAME;
70 cm->comp_var_ref[0] = LAST_FRAME;
71 cm->comp_var_ref[1] = ALTREF_FRAME;
73 cm->comp_fixed_ref = LAST_FRAME;
74 cm->comp_var_ref[0] = GOLDEN_FRAME;
75 cm->comp_var_ref[1] = ALTREF_FRAME;
79 static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
80 return len != 0 && len <= (size_t)(end - start);
83 static int decode_unsigned_max(struct vp9_read_bit_buffer *rb, int max) {
84 const int data = vp9_rb_read_literal(rb, get_unsigned_bits(max));
85 return data > max ? max : data;
88 static TX_MODE read_tx_mode(vp9_reader *r) {
89 TX_MODE tx_mode = vp9_read_literal(r, 2);
90 if (tx_mode == ALLOW_32X32)
91 tx_mode += vp9_read_bit(r);
95 static void read_tx_mode_probs(struct tx_probs *tx_probs, vp9_reader *r) {
98 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
99 for (j = 0; j < TX_SIZES - 3; ++j)
100 vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);
102 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
103 for (j = 0; j < TX_SIZES - 2; ++j)
104 vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);
106 for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
107 for (j = 0; j < TX_SIZES - 1; ++j)
108 vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
111 static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
113 for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
114 for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
115 vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
118 static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
120 for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
121 for (j = 0; j < INTER_MODES - 1; ++j)
122 vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
125 static REFERENCE_MODE read_reference_mode(VP9_COMMON *cm, vp9_reader *r) {
126 if (is_compound_reference_allowed(cm)) {
127 REFERENCE_MODE mode = vp9_read_bit(r);
129 mode += vp9_read_bit(r);
130 setup_compound_reference(cm);
133 return SINGLE_REFERENCE;
137 static void read_reference_mode_probs(VP9_COMMON *cm, vp9_reader *r) {
139 if (cm->reference_mode == REFERENCE_MODE_SELECT)
140 for (i = 0; i < COMP_INTER_CONTEXTS; i++)
141 vp9_diff_update_prob(r, &cm->fc.comp_inter_prob[i]);
143 if (cm->reference_mode != COMPOUND_REFERENCE)
144 for (i = 0; i < REF_CONTEXTS; i++) {
145 vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][0]);
146 vp9_diff_update_prob(r, &cm->fc.single_ref_prob[i][1]);
149 if (cm->reference_mode != SINGLE_REFERENCE)
150 for (i = 0; i < REF_CONTEXTS; i++)
151 vp9_diff_update_prob(r, &cm->fc.comp_ref_prob[i]);
154 static void update_mv_probs(vp9_prob *p, int n, vp9_reader *r) {
156 for (i = 0; i < n; ++i)
157 if (vp9_read(r, NMV_UPDATE_PROB))
158 p[i] = (vp9_read_literal(r, 7) << 1) | 1;
161 static void read_mv_probs(nmv_context *ctx, int allow_hp, vp9_reader *r) {
164 update_mv_probs(ctx->joints, MV_JOINTS - 1, r);
166 for (i = 0; i < 2; ++i) {
167 nmv_component *const comp_ctx = &ctx->comps[i];
168 update_mv_probs(&comp_ctx->sign, 1, r);
169 update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r);
170 update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r);
171 update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r);
174 for (i = 0; i < 2; ++i) {
175 nmv_component *const comp_ctx = &ctx->comps[i];
176 for (j = 0; j < CLASS0_SIZE; ++j)
177 update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
178 update_mv_probs(comp_ctx->fp, 3, r);
182 for (i = 0; i < 2; ++i) {
183 nmv_component *const comp_ctx = &ctx->comps[i];
184 update_mv_probs(&comp_ctx->class0_hp, 1, r);
185 update_mv_probs(&comp_ctx->hp, 1, r);
190 static void setup_plane_dequants(VP9_COMMON *cm, MACROBLOCKD *xd, int q_index) {
192 xd->plane[0].dequant = cm->y_dequant[q_index];
194 for (i = 1; i < MAX_MB_PLANE; i++)
195 xd->plane[i].dequant = cm->uv_dequant[q_index];
198 // Allocate storage for each tile column.
199 // TODO(jzern): when max_threads <= 1 the same storage could be used for each
201 static void alloc_tile_storage(VP9D_COMP *pbi, int tile_rows, int tile_cols) {
202 VP9_COMMON *const cm = &pbi->common;
203 const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
204 int i, tile_row, tile_col;
206 CHECK_MEM_ERROR(cm, pbi->mi_streams,
207 vpx_realloc(pbi->mi_streams, tile_rows * tile_cols *
208 sizeof(*pbi->mi_streams)));
209 for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
210 for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
212 vp9_tile_init(&tile, cm, tile_row, tile_col);
213 pbi->mi_streams[tile_row * tile_cols + tile_col] =
214 &cm->mi[tile.mi_row_start * cm->mode_info_stride
215 + tile.mi_col_start];
219 // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
220 // block where mi unit size is 8x8.
221 CHECK_MEM_ERROR(cm, pbi->above_context[0],
222 vpx_realloc(pbi->above_context[0],
223 sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
224 2 * aligned_mi_cols));
225 for (i = 1; i < MAX_MB_PLANE; ++i) {
226 pbi->above_context[i] = pbi->above_context[0] +
227 i * sizeof(*pbi->above_context[0]) *
231 // This is sized based on the entire frame. Each tile operates within its
233 CHECK_MEM_ERROR(cm, pbi->above_seg_context,
234 vpx_realloc(pbi->above_seg_context,
235 sizeof(*pbi->above_seg_context) *
239 static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
240 TX_SIZE tx_size, uint8_t *dst, int stride,
242 struct macroblockd_plane *const pd = &xd->plane[plane];
245 const int plane_type = pd->plane_type;
246 int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
249 tx_type = get_tx_type_4x4(plane_type, xd, block);
250 if (tx_type == DCT_DCT)
251 xd->itxm_add(dqcoeff, dst, stride, eob);
253 vp9_iht4x4_16_add(dqcoeff, dst, stride, tx_type);
256 tx_type = get_tx_type_8x8(plane_type, xd);
257 vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
260 tx_type = get_tx_type_16x16(plane_type, xd);
261 vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
265 vp9_idct32x32_add(dqcoeff, dst, stride, eob);
268 assert(0 && "Invalid transform size");
272 vpx_memset(dqcoeff, 0, 2 * sizeof(dqcoeff[0]));
274 if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
275 vpx_memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
276 else if (tx_size == TX_32X32 && eob <= 34)
277 vpx_memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
279 vpx_memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
290 static void predict_and_reconstruct_intra_block(int plane, int block,
291 BLOCK_SIZE plane_bsize,
292 TX_SIZE tx_size, void *arg) {
293 struct intra_args *const args = arg;
294 VP9_COMMON *const cm = args->cm;
295 MACROBLOCKD *const xd = args->xd;
296 struct macroblockd_plane *const pd = &xd->plane[plane];
297 MODE_INFO *const mi = xd->mi_8x8[0];
298 const MB_PREDICTION_MODE mode = (plane == 0)
299 ? ((mi->mbmi.sb_type < BLOCK_8X8) ? mi->bmi[block].as_mode
304 txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
305 dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x];
307 vp9_predict_intra_block(xd, block >> (tx_size << 1),
308 b_width_log2(plane_bsize), tx_size, mode,
309 dst, pd->dst.stride, dst, pd->dst.stride,
312 if (!mi->mbmi.skip_coeff) {
313 const int eob = vp9_decode_block_tokens(cm, xd, plane, block,
314 plane_bsize, x, y, tx_size,
316 inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride,
328 static void reconstruct_inter_block(int plane, int block,
329 BLOCK_SIZE plane_bsize,
330 TX_SIZE tx_size, void *arg) {
331 struct inter_args *args = arg;
332 VP9_COMMON *const cm = args->cm;
333 MACROBLOCKD *const xd = args->xd;
334 struct macroblockd_plane *const pd = &xd->plane[plane];
336 txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
337 eob = vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize, x, y,
339 inverse_transform_block(xd, plane, block, tx_size,
340 &pd->dst.buf[4 * y * pd->dst.stride + 4 * x],
341 pd->dst.stride, eob);
342 *args->eobtotal += eob;
345 static void set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
346 const TileInfo *const tile,
347 BLOCK_SIZE bsize, int mi_row, int mi_col) {
348 const int bw = num_8x8_blocks_wide_lookup[bsize];
349 const int bh = num_8x8_blocks_high_lookup[bsize];
350 const int x_mis = MIN(bw, cm->mi_cols - mi_col);
351 const int y_mis = MIN(bh, cm->mi_rows - mi_row);
352 const int offset = mi_row * cm->mode_info_stride + mi_col;
353 const int tile_offset = tile->mi_row_start * cm->mode_info_stride +
357 xd->mi_8x8 = cm->mi_grid_visible + offset;
358 xd->prev_mi_8x8 = cm->prev_mi_grid_visible + offset;
359 // Special case: if prev_mi is NULL, the previous mode info context
361 xd->last_mi = cm->prev_mi ? xd->prev_mi_8x8[0] : NULL;
363 xd->mi_8x8[0] = xd->mi_stream + offset - tile_offset;
364 xd->mi_8x8[0]->mbmi.sb_type = bsize;
365 for (y = 0; y < y_mis; ++y)
366 for (x = !y; x < x_mis; ++x)
367 xd->mi_8x8[y * cm->mode_info_stride + x] = xd->mi_8x8[0];
369 set_skip_context(xd, xd->above_context, xd->left_context, mi_row, mi_col);
371 // Distance of Mb to the various image edges. These are specified to 8th pel
372 // as they are always compared to values that are in 1/8th pel units
373 set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
375 setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
378 static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
379 int idx, int mi_row, int mi_col) {
380 MB_MODE_INFO *const mbmi = &xd->mi_8x8[0]->mbmi;
381 RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
382 xd->block_refs[idx] = ref_buffer;
383 if (!vp9_is_valid_scale(&ref_buffer->sf))
384 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
385 "Invalid scale factors");
386 setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col, &ref_buffer->sf);
387 xd->corrupted |= ref_buffer->buf->corrupted;
390 static void decode_modes_b(VP9_COMMON *const cm, MACROBLOCKD *const xd,
391 const TileInfo *const tile,
392 int mi_row, int mi_col,
393 vp9_reader *r, BLOCK_SIZE bsize) {
394 const int less8x8 = bsize < BLOCK_8X8;
397 set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
398 vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r);
403 // Has to be called after set_offsets
404 mbmi = &xd->mi_8x8[0]->mbmi;
406 if (mbmi->skip_coeff) {
407 reset_skip_context(xd, bsize);
410 setup_plane_dequants(cm, xd, vp9_get_qindex(&cm->seg, mbmi->segment_id,
414 if (!is_inter_block(mbmi)) {
415 struct intra_args arg = { cm, xd, r };
416 foreach_transformed_block(xd, bsize, predict_and_reconstruct_intra_block,
420 set_ref(cm, xd, 0, mi_row, mi_col);
421 if (has_second_ref(mbmi))
422 set_ref(cm, xd, 1, mi_row, mi_col);
424 xd->subpix.filter_x = xd->subpix.filter_y =
425 vp9_get_filter_kernel(mbmi->interp_filter);
428 vp9_dec_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
431 if (!mbmi->skip_coeff) {
433 struct inter_args arg = { cm, xd, r, &eobtotal };
434 foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
435 if (!less8x8 && eobtotal == 0)
436 mbmi->skip_coeff = 1; // skip loopfilter
440 xd->corrupted |= vp9_reader_has_error(r);
443 static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd, int hbs,
444 int mi_row, int mi_col, BLOCK_SIZE bsize,
446 const int ctx = partition_plane_context(xd->above_seg_context,
447 xd->left_seg_context,
448 mi_row, mi_col, bsize);
449 const vp9_prob *const probs = get_partition_probs(cm, ctx);
450 const int has_rows = (mi_row + hbs) < cm->mi_rows;
451 const int has_cols = (mi_col + hbs) < cm->mi_cols;
454 if (has_rows && has_cols)
455 p = vp9_read_tree(r, vp9_partition_tree, probs);
456 else if (!has_rows && has_cols)
457 p = vp9_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ;
458 else if (has_rows && !has_cols)
459 p = vp9_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT;
463 if (!cm->frame_parallel_decoding_mode)
464 ++cm->counts.partition[ctx][p];
469 static void decode_modes_sb(VP9_COMMON *const cm, MACROBLOCKD *const xd,
470 const TileInfo *const tile,
471 int mi_row, int mi_col,
472 vp9_reader* r, BLOCK_SIZE bsize) {
473 const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
474 PARTITION_TYPE partition;
477 if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
480 partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
481 subsize = get_subsize(bsize, partition);
482 if (subsize < BLOCK_8X8) {
483 decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
487 decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
490 decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
491 if (mi_row + hbs < cm->mi_rows)
492 decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
495 decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
496 if (mi_col + hbs < cm->mi_cols)
497 decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
499 case PARTITION_SPLIT:
500 decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, subsize);
501 decode_modes_sb(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
502 decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
503 decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize);
506 assert(0 && "Invalid partition type");
510 // update partition context
511 if (bsize >= BLOCK_8X8 &&
512 (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
513 update_partition_context(xd->above_seg_context, xd->left_seg_context,
514 mi_row, mi_col, subsize, bsize);
517 static void setup_token_decoder(const uint8_t *data,
518 const uint8_t *data_end,
520 struct vpx_internal_error_info *error_info,
522 // Validate the calculated partition length. If the buffer
523 // described by the partition can't be fully read, then restrict
524 // it to the portion that can be (for EC mode) or throw an error.
525 if (!read_is_valid(data, read_size, data_end))
526 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
527 "Truncated packet or corrupt tile length");
529 if (vp9_reader_init(r, data, read_size))
530 vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR,
531 "Failed to allocate bool decoder %d", 1);
534 static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs,
539 for (i = 0; i < PLANE_TYPES; ++i)
540 for (j = 0; j < REF_TYPES; ++j)
541 for (k = 0; k < COEF_BANDS; ++k)
542 for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
543 for (m = 0; m < UNCONSTRAINED_NODES; ++m)
544 vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
547 static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode,
549 const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
551 for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size)
552 read_coef_probs_common(fc->coef_probs[tx_size], r);
555 static void setup_segmentation(struct segmentation *seg,
556 struct vp9_read_bit_buffer *rb) {
560 seg->update_data = 0;
562 seg->enabled = vp9_rb_read_bit(rb);
566 // Segmentation map update
567 seg->update_map = vp9_rb_read_bit(rb);
568 if (seg->update_map) {
569 for (i = 0; i < SEG_TREE_PROBS; i++)
570 seg->tree_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
573 seg->temporal_update = vp9_rb_read_bit(rb);
574 if (seg->temporal_update) {
575 for (i = 0; i < PREDICTION_PROBS; i++)
576 seg->pred_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
579 for (i = 0; i < PREDICTION_PROBS; i++)
580 seg->pred_probs[i] = MAX_PROB;
584 // Segmentation data update
585 seg->update_data = vp9_rb_read_bit(rb);
586 if (seg->update_data) {
587 seg->abs_delta = vp9_rb_read_bit(rb);
589 vp9_clearall_segfeatures(seg);
591 for (i = 0; i < MAX_SEGMENTS; i++) {
592 for (j = 0; j < SEG_LVL_MAX; j++) {
594 const int feature_enabled = vp9_rb_read_bit(rb);
595 if (feature_enabled) {
596 vp9_enable_segfeature(seg, i, j);
597 data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j));
598 if (vp9_is_segfeature_signed(j))
599 data = vp9_rb_read_bit(rb) ? -data : data;
601 vp9_set_segdata(seg, i, j, data);
607 static void setup_loopfilter(struct loopfilter *lf,
608 struct vp9_read_bit_buffer *rb) {
609 lf->filter_level = vp9_rb_read_literal(rb, 6);
610 lf->sharpness_level = vp9_rb_read_literal(rb, 3);
612 // Read in loop filter deltas applied at the MB level based on mode or ref
614 lf->mode_ref_delta_update = 0;
616 lf->mode_ref_delta_enabled = vp9_rb_read_bit(rb);
617 if (lf->mode_ref_delta_enabled) {
618 lf->mode_ref_delta_update = vp9_rb_read_bit(rb);
619 if (lf->mode_ref_delta_update) {
622 for (i = 0; i < MAX_REF_LF_DELTAS; i++)
623 if (vp9_rb_read_bit(rb))
624 lf->ref_deltas[i] = vp9_rb_read_signed_literal(rb, 6);
626 for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
627 if (vp9_rb_read_bit(rb))
628 lf->mode_deltas[i] = vp9_rb_read_signed_literal(rb, 6);
633 static int read_delta_q(struct vp9_read_bit_buffer *rb, int *delta_q) {
634 const int old = *delta_q;
635 *delta_q = vp9_rb_read_bit(rb) ? vp9_rb_read_signed_literal(rb, 4) : 0;
636 return old != *delta_q;
639 static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd,
640 struct vp9_read_bit_buffer *rb) {
643 cm->base_qindex = vp9_rb_read_literal(rb, QINDEX_BITS);
644 update |= read_delta_q(rb, &cm->y_dc_delta_q);
645 update |= read_delta_q(rb, &cm->uv_dc_delta_q);
646 update |= read_delta_q(rb, &cm->uv_ac_delta_q);
648 vp9_init_dequantizer(cm);
650 xd->lossless = cm->base_qindex == 0 &&
651 cm->y_dc_delta_q == 0 &&
652 cm->uv_dc_delta_q == 0 &&
653 cm->uv_ac_delta_q == 0;
655 xd->itxm_add = xd->lossless ? vp9_iwht4x4_add : vp9_idct4x4_add;
658 static INTERPOLATION_TYPE read_interp_filter_type(
659 struct vp9_read_bit_buffer *rb) {
660 const INTERPOLATION_TYPE literal_to_type[] = { EIGHTTAP_SMOOTH,
664 return vp9_rb_read_bit(rb) ? SWITCHABLE
665 : literal_to_type[vp9_rb_read_literal(rb, 2)];
668 static void read_frame_size(struct vp9_read_bit_buffer *rb,
669 int *width, int *height) {
670 const int w = vp9_rb_read_literal(rb, 16) + 1;
671 const int h = vp9_rb_read_literal(rb, 16) + 1;
676 static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
677 cm->display_width = cm->width;
678 cm->display_height = cm->height;
679 if (vp9_rb_read_bit(rb))
680 read_frame_size(rb, &cm->display_width, &cm->display_height);
683 static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
684 VP9_COMMON *cm = &pbi->common;
686 if (cm->width != width || cm->height != height) {
687 // Change in frame size.
688 // TODO(agrange) Don't test width/height, check overall size.
689 if (width > cm->width || height > cm->height) {
690 // Rescale frame buffers only if they're not big enough already.
691 if (vp9_resize_frame_buffers(cm, width, height))
692 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
693 "Failed to allocate frame buffers");
699 vp9_update_frame_size(cm);
702 if (cm->fb_list != NULL) {
703 vpx_codec_frame_buffer_t *const ext_fb = &cm->fb_list[cm->new_fb_idx];
704 if (vp9_realloc_frame_buffer(get_frame_new_buffer(cm),
705 cm->width, cm->height,
706 cm->subsampling_x, cm->subsampling_y,
707 VP9_DEC_BORDER_IN_PIXELS, ext_fb,
708 cm->realloc_fb_cb, cm->user_priv)) {
709 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
710 "Failed to allocate external frame buffer");
713 vp9_realloc_frame_buffer(get_frame_new_buffer(cm), cm->width, cm->height,
714 cm->subsampling_x, cm->subsampling_y,
715 VP9_DEC_BORDER_IN_PIXELS, NULL, NULL, NULL);
719 static void setup_frame_size(VP9D_COMP *pbi,
720 struct vp9_read_bit_buffer *rb) {
722 read_frame_size(rb, &width, &height);
723 apply_frame_size(pbi, width, height);
724 setup_display_size(&pbi->common, rb);
727 static void setup_frame_size_with_refs(VP9D_COMP *pbi,
728 struct vp9_read_bit_buffer *rb) {
729 VP9_COMMON *const cm = &pbi->common;
733 for (i = 0; i < REFS_PER_FRAME; ++i) {
734 if (vp9_rb_read_bit(rb)) {
735 YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
736 width = buf->y_crop_width;
737 height = buf->y_crop_height;
744 read_frame_size(rb, &width, &height);
746 if (!width || !height)
747 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
748 "Referenced frame with invalid size");
750 apply_frame_size(pbi, width, height);
751 setup_display_size(cm, rb);
754 static void setup_tile_context(VP9D_COMP *const pbi, MACROBLOCKD *const xd,
755 int tile_row, int tile_col) {
757 const int tile_cols = 1 << pbi->common.log2_tile_cols;
758 xd->mi_stream = pbi->mi_streams[tile_row * tile_cols + tile_col];
760 for (i = 0; i < MAX_MB_PLANE; ++i) {
761 xd->above_context[i] = pbi->above_context[i];
763 // see note in alloc_tile_storage().
764 xd->above_seg_context = pbi->above_seg_context;
767 static void decode_tile(VP9D_COMP *pbi, const TileInfo *const tile,
769 const int num_threads = pbi->oxcf.max_threads;
770 VP9_COMMON *const cm = &pbi->common;
772 MACROBLOCKD *xd = &pbi->mb;
774 if (pbi->do_loopfilter_inline) {
775 LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
776 lf_data->frame_buffer = get_frame_new_buffer(cm);
778 lf_data->xd = pbi->mb;
781 vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
784 for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
785 mi_row += MI_BLOCK_SIZE) {
786 // For a SB there are 2 left contexts, each pertaining to a MB row within
787 vp9_zero(xd->left_context);
788 vp9_zero(xd->left_seg_context);
789 for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
790 mi_col += MI_BLOCK_SIZE) {
791 decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64);
794 if (pbi->do_loopfilter_inline) {
795 const int lf_start = mi_row - MI_BLOCK_SIZE;
796 LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
798 // delay the loopfilter by 1 macroblock row.
799 if (lf_start < 0) continue;
801 // decoding has completed: finish up the loop filter in this thread.
802 if (mi_row + MI_BLOCK_SIZE >= tile->mi_row_end) continue;
804 vp9_worker_sync(&pbi->lf_worker);
805 lf_data->start = lf_start;
806 lf_data->stop = mi_row;
807 if (num_threads > 1) {
808 vp9_worker_launch(&pbi->lf_worker);
810 vp9_worker_execute(&pbi->lf_worker);
815 if (pbi->do_loopfilter_inline) {
816 LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
818 vp9_worker_sync(&pbi->lf_worker);
819 lf_data->start = lf_data->stop;
820 lf_data->stop = cm->mi_rows;
821 vp9_worker_execute(&pbi->lf_worker);
825 static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
826 int min_log2_tile_cols, max_log2_tile_cols, max_ones;
827 vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
830 max_ones = max_log2_tile_cols - min_log2_tile_cols;
831 cm->log2_tile_cols = min_log2_tile_cols;
832 while (max_ones-- && vp9_rb_read_bit(rb))
833 cm->log2_tile_cols++;
836 cm->log2_tile_rows = vp9_rb_read_bit(rb);
837 if (cm->log2_tile_rows)
838 cm->log2_tile_rows += vp9_rb_read_bit(rb);
841 // Reads the next tile returning its size and adjusting '*data' accordingly
842 // based on 'is_last'.
843 static size_t get_tile(const uint8_t *const data_end,
845 struct vpx_internal_error_info *error_info,
846 const uint8_t **data) {
850 if (!read_is_valid(*data, 4, data_end))
851 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
852 "Truncated packet or corrupt tile length");
854 size = read_be32(*data);
857 if (size > (size_t)(data_end - *data))
858 vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
859 "Truncated packet or corrupt tile size");
861 size = data_end - *data;
866 typedef struct TileBuffer {
869 int col; // only used with multi-threaded decoding
872 static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
873 VP9_COMMON *const cm = &pbi->common;
874 MACROBLOCKD *const xd = &pbi->mb;
875 const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols);
876 const int tile_cols = 1 << cm->log2_tile_cols;
877 const int tile_rows = 1 << cm->log2_tile_rows;
878 TileBuffer tile_buffers[4][1 << 6];
879 int tile_row, tile_col;
880 const uint8_t *const data_end = pbi->source + pbi->source_sz;
881 const uint8_t *end = NULL;
884 assert(tile_rows <= 4);
885 assert(tile_cols <= (1 << 6));
887 // Note: this memset assumes above_context[0], [1] and [2]
888 // are allocated as part of the same buffer.
889 vpx_memset(pbi->above_context[0], 0,
890 sizeof(*pbi->above_context[0]) * MAX_MB_PLANE * 2 * aligned_cols);
892 vpx_memset(pbi->above_seg_context, 0,
893 sizeof(*pbi->above_seg_context) * aligned_cols);
895 // Load tile data into tile_buffers
896 for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
897 for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
898 const int last_tile = tile_row == tile_rows - 1 &&
899 tile_col == tile_cols - 1;
900 const size_t size = get_tile(data_end, last_tile, &cm->error, &data);
901 TileBuffer *const buf = &tile_buffers[tile_row][tile_col];
908 // Decode tiles using data from tile_buffers
909 for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
910 for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
911 const int col = pbi->oxcf.inv_tile_order ? tile_cols - tile_col - 1
913 const int last_tile = tile_row == tile_rows - 1 &&
914 col == tile_cols - 1;
915 const TileBuffer *const buf = &tile_buffers[tile_row][col];
918 vp9_tile_init(&tile, cm, tile_row, col);
919 setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &r);
920 setup_tile_context(pbi, xd, tile_row, col);
921 decode_tile(pbi, &tile, &r);
924 end = vp9_reader_find_end(&r);
931 static void setup_tile_macroblockd(TileWorkerData *const tile_data) {
932 MACROBLOCKD *xd = &tile_data->xd;
933 struct macroblockd_plane *const pd = xd->plane;
936 for (i = 0; i < MAX_MB_PLANE; ++i) {
937 pd[i].dqcoeff = tile_data->dqcoeff[i];
938 vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t));
942 static int tile_worker_hook(void *arg1, void *arg2) {
943 TileWorkerData *const tile_data = (TileWorkerData*)arg1;
944 const TileInfo *const tile = (TileInfo*)arg2;
947 for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
948 mi_row += MI_BLOCK_SIZE) {
949 vp9_zero(tile_data->xd.left_context);
950 vp9_zero(tile_data->xd.left_seg_context);
951 for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
952 mi_col += MI_BLOCK_SIZE) {
953 decode_modes_sb(tile_data->cm, &tile_data->xd, tile,
954 mi_row, mi_col, &tile_data->bit_reader, BLOCK_64X64);
957 return !tile_data->xd.corrupted;
960 // sorts in descending order
961 static int compare_tile_buffers(const void *a, const void *b) {
962 const TileBuffer *const buf1 = (const TileBuffer*)a;
963 const TileBuffer *const buf2 = (const TileBuffer*)b;
964 if (buf1->size < buf2->size) {
966 } else if (buf1->size == buf2->size) {
973 static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi, const uint8_t *data) {
974 VP9_COMMON *const cm = &pbi->common;
975 const uint8_t *bit_reader_end = NULL;
976 const uint8_t *const data_end = pbi->source + pbi->source_sz;
977 const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
978 const int tile_cols = 1 << cm->log2_tile_cols;
979 const int tile_rows = 1 << cm->log2_tile_rows;
980 const int num_workers = MIN(pbi->oxcf.max_threads & ~1, tile_cols);
981 TileBuffer tile_buffers[1 << 6];
983 int final_worker = -1;
985 assert(tile_cols <= (1 << 6));
986 assert(tile_rows == 1);
989 if (num_workers > pbi->num_tile_workers) {
991 CHECK_MEM_ERROR(cm, pbi->tile_workers,
992 vpx_realloc(pbi->tile_workers,
993 num_workers * sizeof(*pbi->tile_workers)));
994 for (i = pbi->num_tile_workers; i < num_workers; ++i) {
995 VP9Worker *const worker = &pbi->tile_workers[i];
996 ++pbi->num_tile_workers;
998 vp9_worker_init(worker);
999 worker->hook = (VP9WorkerHook)tile_worker_hook;
1000 CHECK_MEM_ERROR(cm, worker->data1,
1001 vpx_memalign(32, sizeof(TileWorkerData)));
1002 CHECK_MEM_ERROR(cm, worker->data2, vpx_malloc(sizeof(TileInfo)));
1003 if (i < num_workers - 1 && !vp9_worker_reset(worker)) {
1004 vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
1005 "Tile decoder thread creation failed");
1010 // Note: this memset assumes above_context[0], [1] and [2]
1011 // are allocated as part of the same buffer.
1012 vpx_memset(pbi->above_context[0], 0,
1013 sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
1014 2 * aligned_mi_cols);
1015 vpx_memset(pbi->above_seg_context, 0,
1016 sizeof(*pbi->above_seg_context) * aligned_mi_cols);
1018 // Load tile data into tile_buffers
1019 for (n = 0; n < tile_cols; ++n) {
1021 get_tile(data_end, n == tile_cols - 1, &cm->error, &data);
1022 TileBuffer *const buf = &tile_buffers[n];
1029 // Sort the buffers based on size in descending order.
1030 qsort(tile_buffers, tile_cols, sizeof(tile_buffers[0]), compare_tile_buffers);
1032 // Rearrange the tile buffers such that per-tile group the largest, and
1033 // presumably the most difficult, tile will be decoded in the main thread.
1034 // This should help minimize the number of instances where the main thread is
1035 // waiting for a worker to complete.
1037 int group_start = 0;
1038 while (group_start < tile_cols) {
1039 const TileBuffer largest = tile_buffers[group_start];
1040 const int group_end = MIN(group_start + num_workers, tile_cols) - 1;
1041 memmove(tile_buffers + group_start, tile_buffers + group_start + 1,
1042 (group_end - group_start) * sizeof(tile_buffers[0]));
1043 tile_buffers[group_end] = largest;
1044 group_start = group_end + 1;
1049 while (n < tile_cols) {
1051 for (i = 0; i < num_workers && n < tile_cols; ++i) {
1052 VP9Worker *const worker = &pbi->tile_workers[i];
1053 TileWorkerData *const tile_data = (TileWorkerData*)worker->data1;
1054 TileInfo *const tile = (TileInfo*)worker->data2;
1055 TileBuffer *const buf = &tile_buffers[n];
1058 tile_data->xd = pbi->mb;
1059 tile_data->xd.corrupted = 0;
1060 vp9_tile_init(tile, tile_data->cm, 0, buf->col);
1062 setup_token_decoder(buf->data, data_end, buf->size, &cm->error,
1063 &tile_data->bit_reader);
1064 setup_tile_context(pbi, &tile_data->xd, 0, buf->col);
1065 setup_tile_macroblockd(tile_data);
1067 worker->had_error = 0;
1068 if (i == num_workers - 1 || n == tile_cols - 1) {
1069 vp9_worker_execute(worker);
1071 vp9_worker_launch(worker);
1074 if (buf->col == tile_cols - 1) {
1081 for (; i > 0; --i) {
1082 VP9Worker *const worker = &pbi->tile_workers[i - 1];
1083 pbi->mb.corrupted |= !vp9_worker_sync(worker);
1085 if (final_worker > -1) {
1086 TileWorkerData *const tile_data =
1087 (TileWorkerData*)pbi->tile_workers[final_worker].data1;
1088 bit_reader_end = vp9_reader_find_end(&tile_data->bit_reader);
1093 return bit_reader_end;
1096 static void check_sync_code(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
1097 if (vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_0 ||
1098 vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_1 ||
1099 vp9_rb_read_literal(rb, 8) != VP9_SYNC_CODE_2) {
1100 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1101 "Invalid frame sync code");
1105 static void error_handler(void *data, size_t bit_offset) {
1106 VP9_COMMON *const cm = (VP9_COMMON *)data;
1107 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet");
1111 if (vp9_rb_read_bit(rb)) \
1112 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM, \
1113 "Reserved bit must be unset")
1115 static size_t read_uncompressed_header(VP9D_COMP *pbi,
1116 struct vp9_read_bit_buffer *rb) {
1117 VP9_COMMON *const cm = &pbi->common;
1121 cm->last_frame_type = cm->frame_type;
1123 if (vp9_rb_read_literal(rb, 2) != VP9_FRAME_MARKER)
1124 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1125 "Invalid frame marker");
1127 cm->version = vp9_rb_read_bit(rb);
1130 cm->show_existing_frame = vp9_rb_read_bit(rb);
1131 if (cm->show_existing_frame) {
1132 // show an existing frame directly
1133 int frame_to_show = cm->ref_frame_map[vp9_rb_read_literal(rb, 3)];
1134 ref_cnt_fb(cm->fb_idx_ref_cnt, &cm->new_fb_idx, frame_to_show);
1135 pbi->refresh_frame_flags = 0;
1136 cm->lf.filter_level = 0;
1140 cm->frame_type = (FRAME_TYPE) vp9_rb_read_bit(rb);
1141 cm->show_frame = vp9_rb_read_bit(rb);
1142 cm->error_resilient_mode = vp9_rb_read_bit(rb);
1144 if (cm->frame_type == KEY_FRAME) {
1145 check_sync_code(cm, rb);
1147 cm->color_space = vp9_rb_read_literal(rb, 3); // colorspace
1148 if (cm->color_space != SRGB) {
1149 vp9_rb_read_bit(rb); // [16,235] (including xvycc) vs [0,255] range
1150 if (cm->version == 1) {
1151 cm->subsampling_x = vp9_rb_read_bit(rb);
1152 cm->subsampling_y = vp9_rb_read_bit(rb);
1153 vp9_rb_read_bit(rb); // has extra plane
1155 cm->subsampling_y = cm->subsampling_x = 1;
1158 if (cm->version == 1) {
1159 cm->subsampling_y = cm->subsampling_x = 0;
1160 vp9_rb_read_bit(rb); // has extra plane
1162 vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
1163 "RGB not supported in profile 0");
1167 pbi->refresh_frame_flags = (1 << REF_FRAMES) - 1;
1169 for (i = 0; i < REFS_PER_FRAME; ++i) {
1170 cm->frame_refs[i].idx = cm->new_fb_idx;
1171 cm->frame_refs[i].buf = get_frame_new_buffer(cm);
1174 setup_frame_size(pbi, rb);
1176 cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb);
1178 cm->reset_frame_context = cm->error_resilient_mode ?
1179 0 : vp9_rb_read_literal(rb, 2);
1181 if (cm->intra_only) {
1182 check_sync_code(cm, rb);
1184 pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
1185 setup_frame_size(pbi, rb);
1187 pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
1189 for (i = 0; i < REFS_PER_FRAME; ++i) {
1190 const int ref = vp9_rb_read_literal(rb, REF_FRAMES_LOG2);
1191 const int idx = cm->ref_frame_map[ref];
1192 cm->frame_refs[i].idx = idx;
1193 cm->frame_refs[i].buf = &cm->yv12_fb[idx];
1194 cm->ref_frame_sign_bias[LAST_FRAME + i] = vp9_rb_read_bit(rb);
1197 setup_frame_size_with_refs(pbi, rb);
1199 cm->allow_high_precision_mv = vp9_rb_read_bit(rb);
1200 cm->mcomp_filter_type = read_interp_filter_type(rb);
1202 for (i = 0; i < REFS_PER_FRAME; ++i) {
1203 RefBuffer *const ref_buf = &cm->frame_refs[i];
1204 vp9_setup_scale_factors_for_frame(&ref_buf->sf,
1205 ref_buf->buf->y_crop_width,
1206 ref_buf->buf->y_crop_height,
1207 cm->width, cm->height);
1208 if (vp9_is_scaled(&ref_buf->sf))
1209 vp9_extend_frame_borders(ref_buf->buf,
1210 cm->subsampling_x, cm->subsampling_y);
1215 if (!cm->error_resilient_mode) {
1216 cm->refresh_frame_context = vp9_rb_read_bit(rb);
1217 cm->frame_parallel_decoding_mode = vp9_rb_read_bit(rb);
1219 cm->refresh_frame_context = 0;
1220 cm->frame_parallel_decoding_mode = 1;
1223 // This flag will be overridden by the call to vp9_setup_past_independence
1224 // below, forcing the use of context 0 for those frame types.
1225 cm->frame_context_idx = vp9_rb_read_literal(rb, FRAME_CONTEXTS_LOG2);
1227 if (frame_is_intra_only(cm) || cm->error_resilient_mode)
1228 vp9_setup_past_independence(cm);
1230 setup_loopfilter(&cm->lf, rb);
1231 setup_quantization(cm, &pbi->mb, rb);
1232 setup_segmentation(&cm->seg, rb);
1234 setup_tile_info(cm, rb);
1235 sz = vp9_rb_read_literal(rb, 16);
1238 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1239 "Invalid header size");
1244 static int read_compressed_header(VP9D_COMP *pbi, const uint8_t *data,
1245 size_t partition_size) {
1246 VP9_COMMON *const cm = &pbi->common;
1247 MACROBLOCKD *const xd = &pbi->mb;
1248 FRAME_CONTEXT *const fc = &cm->fc;
1252 if (vp9_reader_init(&r, data, partition_size))
1253 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
1254 "Failed to allocate bool decoder 0");
1256 cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
1257 if (cm->tx_mode == TX_MODE_SELECT)
1258 read_tx_mode_probs(&fc->tx_probs, &r);
1259 read_coef_probs(fc, cm->tx_mode, &r);
1261 for (k = 0; k < MBSKIP_CONTEXTS; ++k)
1262 vp9_diff_update_prob(&r, &fc->mbskip_probs[k]);
1264 if (!frame_is_intra_only(cm)) {
1265 nmv_context *const nmvc = &fc->nmvc;
1268 read_inter_mode_probs(fc, &r);
1270 if (cm->mcomp_filter_type == SWITCHABLE)
1271 read_switchable_interp_probs(fc, &r);
1273 for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
1274 vp9_diff_update_prob(&r, &fc->intra_inter_prob[i]);
1276 cm->reference_mode = read_reference_mode(cm, &r);
1277 read_reference_mode_probs(cm, &r);
1279 for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
1280 for (i = 0; i < INTRA_MODES - 1; ++i)
1281 vp9_diff_update_prob(&r, &fc->y_mode_prob[j][i]);
1283 for (j = 0; j < PARTITION_CONTEXTS; ++j)
1284 for (i = 0; i < PARTITION_TYPES - 1; ++i)
1285 vp9_diff_update_prob(&r, &fc->partition_prob[j][i]);
1287 read_mv_probs(nmvc, cm->allow_high_precision_mv, &r);
1290 return vp9_reader_has_error(&r);
1293 void vp9_init_dequantizer(VP9_COMMON *cm) {
1296 for (q = 0; q < QINDEX_RANGE; q++) {
1297 cm->y_dequant[q][0] = vp9_dc_quant(q, cm->y_dc_delta_q);
1298 cm->y_dequant[q][1] = vp9_ac_quant(q, 0);
1300 cm->uv_dequant[q][0] = vp9_dc_quant(q, cm->uv_dc_delta_q);
1301 cm->uv_dequant[q][1] = vp9_ac_quant(q, cm->uv_ac_delta_q);
1306 #define debug_check_frame_counts(cm) (void)0
1308 // Counts should only be incremented when frame_parallel_decoding_mode and
1309 // error_resilient_mode are disabled.
1310 static void debug_check_frame_counts(const VP9_COMMON *const cm) {
1311 FRAME_COUNTS zero_counts;
1312 vp9_zero(zero_counts);
1313 assert(cm->frame_parallel_decoding_mode || cm->error_resilient_mode);
1314 assert(!memcmp(cm->counts.y_mode, zero_counts.y_mode,
1315 sizeof(cm->counts.y_mode)));
1316 assert(!memcmp(cm->counts.uv_mode, zero_counts.uv_mode,
1317 sizeof(cm->counts.uv_mode)));
1318 assert(!memcmp(cm->counts.partition, zero_counts.partition,
1319 sizeof(cm->counts.partition)));
1320 assert(!memcmp(cm->counts.coef, zero_counts.coef,
1321 sizeof(cm->counts.coef)));
1322 assert(!memcmp(cm->counts.eob_branch, zero_counts.eob_branch,
1323 sizeof(cm->counts.eob_branch)));
1324 assert(!memcmp(cm->counts.switchable_interp, zero_counts.switchable_interp,
1325 sizeof(cm->counts.switchable_interp)));
1326 assert(!memcmp(cm->counts.inter_mode, zero_counts.inter_mode,
1327 sizeof(cm->counts.inter_mode)));
1328 assert(!memcmp(cm->counts.intra_inter, zero_counts.intra_inter,
1329 sizeof(cm->counts.intra_inter)));
1330 assert(!memcmp(cm->counts.comp_inter, zero_counts.comp_inter,
1331 sizeof(cm->counts.comp_inter)));
1332 assert(!memcmp(cm->counts.single_ref, zero_counts.single_ref,
1333 sizeof(cm->counts.single_ref)));
1334 assert(!memcmp(cm->counts.comp_ref, zero_counts.comp_ref,
1335 sizeof(cm->counts.comp_ref)));
1336 assert(!memcmp(&cm->counts.tx, &zero_counts.tx, sizeof(cm->counts.tx)));
1337 assert(!memcmp(cm->counts.mbskip, zero_counts.mbskip,
1338 sizeof(cm->counts.mbskip)));
1339 assert(!memcmp(&cm->counts.mv, &zero_counts.mv, sizeof(cm->counts.mv)));
1343 int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
1345 VP9_COMMON *const cm = &pbi->common;
1346 MACROBLOCKD *const xd = &pbi->mb;
1348 const uint8_t *data = pbi->source;
1349 const uint8_t *const data_end = pbi->source + pbi->source_sz;
1351 struct vp9_read_bit_buffer rb = { data, data_end, 0, cm, error_handler };
1352 const size_t first_partition_size = read_uncompressed_header(pbi, &rb);
1353 const int keyframe = cm->frame_type == KEY_FRAME;
1354 const int tile_rows = 1 << cm->log2_tile_rows;
1355 const int tile_cols = 1 << cm->log2_tile_cols;
1356 YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm);
1357 xd->cur_buf = new_fb;
1359 if (!first_partition_size) {
1360 // showing a frame directly
1361 *p_data_end = data + 1;
1365 if (!pbi->decoded_key_frame && !keyframe)
1368 data += vp9_rb_bytes_read(&rb);
1369 if (!read_is_valid(data, first_partition_size, data_end))
1370 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1371 "Truncated packet or corrupt header length");
1373 pbi->do_loopfilter_inline =
1374 (cm->log2_tile_rows | cm->log2_tile_cols) == 0 && cm->lf.filter_level;
1375 if (pbi->do_loopfilter_inline && pbi->lf_worker.data1 == NULL) {
1376 CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, vpx_malloc(sizeof(LFWorkerData)));
1377 pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker;
1378 if (pbi->oxcf.max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) {
1379 vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
1380 "Loop filter thread creation failed");
1384 alloc_tile_storage(pbi, tile_rows, tile_cols);
1386 xd->mode_info_stride = cm->mode_info_stride;
1389 setup_plane_dequants(cm, xd, cm->base_qindex);
1390 setup_block_dptrs(xd, cm->subsampling_x, cm->subsampling_y);
1392 cm->fc = cm->frame_contexts[cm->frame_context_idx];
1393 vp9_zero(cm->counts);
1394 for (i = 0; i < MAX_MB_PLANE; ++i)
1395 vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t));
1398 new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
1400 // TODO(jzern): remove frame_parallel_decoding_mode restriction for
1401 // single-frame tile decoding.
1402 if (pbi->oxcf.max_threads > 1 && tile_rows == 1 && tile_cols > 1 &&
1403 cm->frame_parallel_decoding_mode) {
1404 *p_data_end = decode_tiles_mt(pbi, data + first_partition_size);
1406 *p_data_end = decode_tiles(pbi, data + first_partition_size);
1409 cm->last_width = cm->width;
1410 cm->last_height = cm->height;
1412 new_fb->corrupted |= xd->corrupted;
1414 if (!pbi->decoded_key_frame) {
1415 if (keyframe && !new_fb->corrupted)
1416 pbi->decoded_key_frame = 1;
1418 vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
1419 "A stream must start with a complete key frame");
1422 if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) {
1423 vp9_adapt_coef_probs(cm);
1425 if (!frame_is_intra_only(cm)) {
1426 vp9_adapt_mode_probs(cm);
1427 vp9_adapt_mv_probs(cm, cm->allow_high_precision_mv);
1430 debug_check_frame_counts(cm);
1433 if (cm->refresh_frame_context)
1434 cm->frame_contexts[cm->frame_context_idx] = cm->fc;