2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
47 static const uint8_t rem6[QP_MAX_NUM + 1] = {
48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
51 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
63 static const uint8_t field_scan[16+1] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
70 static const uint8_t field_scan8x8[64+1] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
128 static const uint8_t dequant4_coeff_init[6][3] = {
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
151 static void release_unused_pictures(H264Context *h, int remove_current)
155 /* release non reference frames */
156 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
157 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
158 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
159 ff_h264_unref_picture(h, &h->DPB[i]);
164 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
166 const H264Context *h = sl->h264;
167 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
169 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
170 // edge emu needs blocksize + filter length - 1
171 // (= 21x21 for h264)
172 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
174 av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
175 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
176 av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
177 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
179 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
180 !sl->top_borders[0] || !sl->top_borders[1]) {
181 av_freep(&sl->bipred_scratchpad);
182 av_freep(&sl->edge_emu_buffer);
183 av_freep(&sl->top_borders[0]);
184 av_freep(&sl->top_borders[1]);
186 sl->bipred_scratchpad_allocated = 0;
187 sl->edge_emu_buffer_allocated = 0;
188 sl->top_borders_allocated[0] = 0;
189 sl->top_borders_allocated[1] = 0;
190 return AVERROR(ENOMEM);
196 static int init_table_pools(H264Context *h)
198 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
199 const int mb_array_size = h->mb_stride * h->mb_height;
200 const int b4_stride = h->mb_width * 4 + 1;
201 const int b4_array_size = b4_stride * h->mb_height * 4;
203 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
205 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
206 sizeof(uint32_t), av_buffer_allocz);
207 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
208 sizeof(int16_t), av_buffer_allocz);
209 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
211 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
212 !h->ref_index_pool) {
213 av_buffer_pool_uninit(&h->qscale_table_pool);
214 av_buffer_pool_uninit(&h->mb_type_pool);
215 av_buffer_pool_uninit(&h->motion_val_pool);
216 av_buffer_pool_uninit(&h->ref_index_pool);
217 return AVERROR(ENOMEM);
223 static int alloc_picture(H264Context *h, H264Picture *pic)
227 av_assert0(!pic->f->data[0]);
230 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
231 AV_GET_BUFFER_FLAG_REF : 0);
235 pic->crop = h->sps.crop;
236 pic->crop_top = h->sps.crop_top;
237 pic->crop_left= h->sps.crop_left;
239 if (h->avctx->hwaccel) {
240 const AVHWAccel *hwaccel = h->avctx->hwaccel;
241 av_assert0(!pic->hwaccel_picture_private);
242 if (hwaccel->frame_priv_data_size) {
243 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
244 if (!pic->hwaccel_priv_buf)
245 return AVERROR(ENOMEM);
246 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
249 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
250 int h_chroma_shift, v_chroma_shift;
251 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
252 &h_chroma_shift, &v_chroma_shift);
254 for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
255 memset(pic->f->data[1] + pic->f->linesize[1]*i,
256 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
257 memset(pic->f->data[2] + pic->f->linesize[2]*i,
258 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
262 if (!h->qscale_table_pool) {
263 ret = init_table_pools(h);
268 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
269 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
270 if (!pic->qscale_table_buf || !pic->mb_type_buf)
273 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
274 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
276 for (i = 0; i < 2; i++) {
277 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
278 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
279 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
282 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
283 pic->ref_index[i] = pic->ref_index_buf[i]->data;
288 ff_h264_unref_picture(h, pic);
289 return (ret < 0) ? ret : AVERROR(ENOMEM);
292 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
299 static int find_unused_picture(H264Context *h)
303 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
304 if (pic_is_unused(h, &h->DPB[i]))
307 if (i == H264_MAX_PICTURE_COUNT)
308 return AVERROR_INVALIDDATA;
314 static void init_dequant8_coeff_table(H264Context *h)
317 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
319 for (i = 0; i < 6; i++) {
320 h->dequant8_coeff[i] = h->dequant8_buffer[i];
321 for (j = 0; j < i; j++)
322 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
323 64 * sizeof(uint8_t))) {
324 h->dequant8_coeff[i] = h->dequant8_buffer[j];
330 for (q = 0; q < max_qp + 1; q++) {
333 for (x = 0; x < 64; x++)
334 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
335 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
336 h->pps.scaling_matrix8[i][x]) << shift;
341 static void init_dequant4_coeff_table(H264Context *h)
344 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
345 for (i = 0; i < 6; i++) {
346 h->dequant4_coeff[i] = h->dequant4_buffer[i];
347 for (j = 0; j < i; j++)
348 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
349 16 * sizeof(uint8_t))) {
350 h->dequant4_coeff[i] = h->dequant4_buffer[j];
356 for (q = 0; q < max_qp + 1; q++) {
357 int shift = div6[q] + 2;
359 for (x = 0; x < 16; x++)
360 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
361 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
362 h->pps.scaling_matrix4[i][x]) << shift;
367 void ff_h264_init_dequant_tables(H264Context *h)
370 init_dequant4_coeff_table(h);
371 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
373 if (h->pps.transform_8x8_mode)
374 init_dequant8_coeff_table(h);
375 if (h->sps.transform_bypass) {
376 for (i = 0; i < 6; i++)
377 for (x = 0; x < 16; x++)
378 h->dequant4_coeff[i][0][x] = 1 << 6;
379 if (h->pps.transform_8x8_mode)
380 for (i = 0; i < 6; i++)
381 for (x = 0; x < 64; x++)
382 h->dequant8_coeff[i][0][x] = 1 << 6;
386 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
388 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
389 (((pic) && (pic) >= (old_ctx)->DPB && \
390 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
391 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
393 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
394 H264Context *new_base,
395 H264Context *old_base)
399 for (i = 0; i < count; i++) {
400 av_assert1(!from[i] ||
401 IN_RANGE(from[i], old_base, 1) ||
402 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
403 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
407 static int copy_parameter_set(void **to, void **from, int count, int size)
411 for (i = 0; i < count; i++) {
412 if (to[i] && !from[i]) {
414 } else if (from[i] && !to[i]) {
415 to[i] = av_malloc(size);
417 return AVERROR(ENOMEM);
421 memcpy(to[i], from[i], size);
427 #define copy_fields(to, from, start_field, end_field) \
428 memcpy(&(to)->start_field, &(from)->start_field, \
429 (char *)&(to)->end_field - (char *)&(to)->start_field)
431 static int h264_slice_header_init(H264Context *h);
433 int ff_h264_update_thread_context(AVCodecContext *dst,
434 const AVCodecContext *src)
436 H264Context *h = dst->priv_data, *h1 = src->priv_data;
437 int inited = h->context_initialized, err = 0;
445 (h->width != h1->width ||
446 h->height != h1->height ||
447 h->mb_width != h1->mb_width ||
448 h->mb_height != h1->mb_height ||
449 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
450 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
451 h->sps.colorspace != h1->sps.colorspace)) {
456 /* copy block_offset since frame_start may not be called */
457 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
460 if ((ret = copy_parameter_set((void **)h->sps_buffers,
461 (void **)h1->sps_buffers,
462 MAX_SPS_COUNT, sizeof(SPS))) < 0)
465 if ((ret = copy_parameter_set((void **)h->pps_buffers,
466 (void **)h1->pps_buffers,
467 MAX_PPS_COUNT, sizeof(PPS))) < 0)
471 if (need_reinit || !inited) {
472 h->width = h1->width;
473 h->height = h1->height;
474 h->mb_height = h1->mb_height;
475 h->mb_width = h1->mb_width;
476 h->mb_num = h1->mb_num;
477 h->mb_stride = h1->mb_stride;
478 h->b_stride = h1->b_stride;
480 if (h->context_initialized || h1->context_initialized) {
481 if ((err = h264_slice_header_init(h)) < 0) {
482 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
486 /* copy block_offset since frame_start may not be called */
487 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
490 h->avctx->coded_height = h1->avctx->coded_height;
491 h->avctx->coded_width = h1->avctx->coded_width;
492 h->avctx->width = h1->avctx->width;
493 h->avctx->height = h1->avctx->height;
494 h->coded_picture_number = h1->coded_picture_number;
495 h->first_field = h1->first_field;
496 h->picture_structure = h1->picture_structure;
497 h->droppable = h1->droppable;
498 h->low_delay = h1->low_delay;
499 h->backup_width = h1->backup_width;
500 h->backup_height = h1->backup_height;
501 h->backup_pix_fmt = h1->backup_pix_fmt;
503 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
504 ff_h264_unref_picture(h, &h->DPB[i]);
505 if (h1->DPB[i].f->buf[0] &&
506 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
510 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
511 ff_h264_unref_picture(h, &h->cur_pic);
512 if (h1->cur_pic.f->buf[0]) {
513 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
518 h->enable_er = h1->enable_er;
519 h->workaround_bugs = h1->workaround_bugs;
520 h->low_delay = h1->low_delay;
521 h->droppable = h1->droppable;
523 // extradata/NAL handling
524 h->is_avc = h1->is_avc;
525 h->nal_length_size = h1->nal_length_size;
526 h->x264_build = h1->x264_build;
528 // Dequantization matrices
529 // FIXME these are big - can they be only copied when PPS changes?
530 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
532 for (i = 0; i < 6; i++)
533 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
534 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
536 for (i = 0; i < 6; i++)
537 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
538 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
540 h->dequant_coeff_pps = h1->dequant_coeff_pps;
543 copy_fields(h, h1, poc_lsb, current_slice);
545 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
546 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
547 copy_picture_range(h->delayed_pic, h1->delayed_pic,
548 MAX_DELAYED_PIC_COUNT + 2, h, h1);
550 h->frame_recovered = h1->frame_recovered;
556 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
557 h->prev_poc_msb = h->poc_msb;
558 h->prev_poc_lsb = h->poc_lsb;
560 h->prev_frame_num_offset = h->frame_num_offset;
561 h->prev_frame_num = h->frame_num;
563 h->recovery_frame = h1->recovery_frame;
568 static int h264_frame_start(H264Context *h)
572 const int pixel_shift = h->pixel_shift;
574 1<<(h->sps.bit_depth_luma-1),
575 1<<(h->sps.bit_depth_chroma-1),
576 1<<(h->sps.bit_depth_chroma-1),
580 if (!ff_thread_can_start_frame(h->avctx)) {
581 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
585 release_unused_pictures(h, 1);
586 h->cur_pic_ptr = NULL;
588 i = find_unused_picture(h);
590 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
595 pic->reference = h->droppable ? 0 : h->picture_structure;
596 pic->f->coded_picture_number = h->coded_picture_number++;
597 pic->field_picture = h->picture_structure != PICT_FRAME;
598 pic->frame_num = h->frame_num;
601 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
603 * See decode_nal_units().
605 pic->f->key_frame = 0;
608 pic->invalid_gap = 0;
609 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
611 if ((ret = alloc_picture(h, pic)) < 0)
613 if(!h->frame_recovered && !h->avctx->hwaccel
615 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
618 ff_color_frame(pic->f, c);
620 h->cur_pic_ptr = pic;
621 ff_h264_unref_picture(h, &h->cur_pic);
622 if (CONFIG_ERROR_RESILIENCE) {
623 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
626 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
629 for (i = 0; i < h->nb_slice_ctx; i++) {
630 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
631 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
634 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
635 ff_er_frame_start(&h->slice_ctx[0].er);
636 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
637 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
640 for (i = 0; i < 16; i++) {
641 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
642 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
644 for (i = 0; i < 16; i++) {
645 h->block_offset[16 + i] =
646 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
647 h->block_offset[48 + 16 + i] =
648 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
651 /* We mark the current picture as non-reference after allocating it, so
652 * that if we break out due to an error it can be released automatically
653 * in the next ff_mpv_frame_start().
655 h->cur_pic_ptr->reference = 0;
657 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
659 h->next_output_pic = NULL;
661 assert(h->cur_pic_ptr->long_ref == 0);
666 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
668 uint8_t *src_cb, uint8_t *src_cr,
669 int linesize, int uvlinesize,
674 const int pixel_shift = h->pixel_shift;
675 int chroma444 = CHROMA444(h);
676 int chroma422 = CHROMA422(h);
679 src_cb -= uvlinesize;
680 src_cr -= uvlinesize;
682 if (!simple && FRAME_MBAFF(h)) {
685 top_border = sl->top_borders[0][sl->mb_x];
686 AV_COPY128(top_border, src_y + 15 * linesize);
688 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
689 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
692 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
693 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
694 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
695 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
697 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
698 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
700 } else if (chroma422) {
702 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
703 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
705 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
706 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
710 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
711 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
713 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
714 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
719 } else if (MB_MBAFF(sl)) {
725 top_border = sl->top_borders[top_idx][sl->mb_x];
726 /* There are two lines saved, the line above the top macroblock
727 * of a pair, and the line above the bottom macroblock. */
728 AV_COPY128(top_border, src_y + 16 * linesize);
730 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
732 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
735 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
736 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
737 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
738 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
740 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
741 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
743 } else if (chroma422) {
745 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
746 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
748 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
749 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
753 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
754 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
756 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
757 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
764 * Initialize implicit_weight table.
765 * @param field 0/1 initialize the weight for interlaced MBAFF
766 * -1 initializes the rest
768 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
770 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
772 for (i = 0; i < 2; i++) {
773 sl->luma_weight_flag[i] = 0;
774 sl->chroma_weight_flag[i] = 0;
778 if (h->picture_structure == PICT_FRAME) {
779 cur_poc = h->cur_pic_ptr->poc;
781 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
783 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
784 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
786 sl->use_weight_chroma = 0;
790 ref_count0 = sl->ref_count[0];
791 ref_count1 = sl->ref_count[1];
793 cur_poc = h->cur_pic_ptr->field_poc[field];
795 ref_count0 = 16 + 2 * sl->ref_count[0];
796 ref_count1 = 16 + 2 * sl->ref_count[1];
800 sl->use_weight_chroma = 2;
801 sl->luma_log2_weight_denom = 5;
802 sl->chroma_log2_weight_denom = 5;
804 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
805 int64_t poc0 = sl->ref_list[0][ref0].poc;
806 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
808 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
809 int poc1 = sl->ref_list[1][ref1].poc;
810 int td = av_clip_int8(poc1 - poc0);
812 int tb = av_clip_int8(cur_poc - poc0);
813 int tx = (16384 + (FFABS(td) >> 1)) / td;
814 int dist_scale_factor = (tb * tx + 32) >> 8;
815 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
816 w = 64 - dist_scale_factor;
820 sl->implicit_weight[ref0][ref1][0] =
821 sl->implicit_weight[ref0][ref1][1] = w;
823 sl->implicit_weight[ref0][ref1][field] = w;
830 * initialize scan tables
832 static void init_scan_tables(H264Context *h)
835 for (i = 0; i < 16; i++) {
836 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
837 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
838 h->field_scan[i] = TRANSPOSE(field_scan[i]);
841 for (i = 0; i < 64; i++) {
842 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
843 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
844 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
845 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
846 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
849 if (h->sps.transform_bypass) { // FIXME same ugly
850 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
851 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
852 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
853 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
854 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
855 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
857 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
858 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
859 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
860 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
861 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
862 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
866 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
868 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
869 CONFIG_H264_D3D11VA_HWACCEL + \
870 CONFIG_H264_VAAPI_HWACCEL + \
871 (CONFIG_H264_VDA_HWACCEL * 2) + \
872 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
873 CONFIG_H264_VDPAU_HWACCEL)
874 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
875 const enum AVPixelFormat *choices = pix_fmts;
878 switch (h->sps.bit_depth_luma) {
881 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
882 *fmt++ = AV_PIX_FMT_GBRP9;
884 *fmt++ = AV_PIX_FMT_YUV444P9;
885 } else if (CHROMA422(h))
886 *fmt++ = AV_PIX_FMT_YUV422P9;
888 *fmt++ = AV_PIX_FMT_YUV420P9;
892 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
893 *fmt++ = AV_PIX_FMT_GBRP10;
895 *fmt++ = AV_PIX_FMT_YUV444P10;
896 } else if (CHROMA422(h))
897 *fmt++ = AV_PIX_FMT_YUV422P10;
899 *fmt++ = AV_PIX_FMT_YUV420P10;
903 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
904 *fmt++ = AV_PIX_FMT_GBRP12;
906 *fmt++ = AV_PIX_FMT_YUV444P12;
907 } else if (CHROMA422(h))
908 *fmt++ = AV_PIX_FMT_YUV422P12;
910 *fmt++ = AV_PIX_FMT_YUV420P12;
914 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
915 *fmt++ = AV_PIX_FMT_GBRP14;
917 *fmt++ = AV_PIX_FMT_YUV444P14;
918 } else if (CHROMA422(h))
919 *fmt++ = AV_PIX_FMT_YUV422P14;
921 *fmt++ = AV_PIX_FMT_YUV420P14;
924 #if CONFIG_H264_VDPAU_HWACCEL
925 *fmt++ = AV_PIX_FMT_VDPAU;
928 if (h->avctx->colorspace == AVCOL_SPC_RGB)
929 *fmt++ = AV_PIX_FMT_GBRP;
930 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
931 *fmt++ = AV_PIX_FMT_YUVJ444P;
933 *fmt++ = AV_PIX_FMT_YUV444P;
934 } else if (CHROMA422(h)) {
935 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
936 *fmt++ = AV_PIX_FMT_YUVJ422P;
938 *fmt++ = AV_PIX_FMT_YUV422P;
940 #if CONFIG_H264_DXVA2_HWACCEL
941 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
943 #if CONFIG_H264_D3D11VA_HWACCEL
944 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
946 #if CONFIG_H264_VAAPI_HWACCEL
947 *fmt++ = AV_PIX_FMT_VAAPI;
949 #if CONFIG_H264_VDA_HWACCEL
950 *fmt++ = AV_PIX_FMT_VDA_VLD;
951 *fmt++ = AV_PIX_FMT_VDA;
953 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
954 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
956 if (h->avctx->codec->pix_fmts)
957 choices = h->avctx->codec->pix_fmts;
958 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
959 *fmt++ = AV_PIX_FMT_YUVJ420P;
961 *fmt++ = AV_PIX_FMT_YUV420P;
965 av_log(h->avctx, AV_LOG_ERROR,
966 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
967 return AVERROR_INVALIDDATA;
970 *fmt = AV_PIX_FMT_NONE;
972 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
973 if (choices[i] == h->avctx->pix_fmt && !force_callback)
975 return ff_thread_get_format(h->avctx, choices);
978 /* export coded and cropped frame dimensions to AVCodecContext */
979 static int init_dimensions(H264Context *h)
981 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
982 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
983 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
984 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
986 /* handle container cropping */
987 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
988 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
989 h->avctx->width <= width &&
990 h->avctx->height <= height
992 width = h->avctx->width;
993 height = h->avctx->height;
996 if (width <= 0 || height <= 0) {
997 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
999 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1000 return AVERROR_INVALIDDATA;
1002 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1003 h->sps.crop_bottom =
1013 h->avctx->coded_width = h->width;
1014 h->avctx->coded_height = h->height;
1015 h->avctx->width = width;
1016 h->avctx->height = height;
1021 static int h264_slice_header_init(H264Context *h)
1023 int nb_slices = (HAVE_THREADS &&
1024 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1025 h->avctx->thread_count : 1;
1028 ff_set_sar(h->avctx, h->sps.sar);
1029 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1030 &h->chroma_x_shift, &h->chroma_y_shift);
1032 if (h->sps.timing_info_present_flag) {
1033 int64_t den = h->sps.time_scale;
1034 if (h->x264_build < 44U)
1036 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1037 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1040 ff_h264_free_tables(h);
1043 h->prev_interlaced_frame = 1;
1045 init_scan_tables(h);
1046 ret = ff_h264_alloc_tables(h);
1048 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1052 #if FF_API_CAP_VDPAU
1053 if (h->avctx->codec &&
1054 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
1055 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
1056 av_log(h->avctx, AV_LOG_ERROR,
1057 "VDPAU decoding does not support video colorspace.\n");
1058 ret = AVERROR_INVALIDDATA;
1063 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 14 ||
1064 h->sps.bit_depth_luma == 11 || h->sps.bit_depth_luma == 13
1066 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
1067 h->sps.bit_depth_luma);
1068 ret = AVERROR_INVALIDDATA;
1072 h->cur_bit_depth_luma =
1073 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1074 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
1075 h->pixel_shift = h->sps.bit_depth_luma > 8;
1076 h->chroma_format_idc = h->sps.chroma_format_idc;
1077 h->bit_depth_luma = h->sps.bit_depth_luma;
1079 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
1080 h->sps.chroma_format_idc);
1081 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1082 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
1083 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
1084 h->sps.chroma_format_idc);
1085 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
1087 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1090 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1092 max_slices = H264_MAX_THREADS;
1093 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1094 " reducing to %d\n", nb_slices, max_slices);
1095 nb_slices = max_slices;
1097 h->slice_context_count = nb_slices;
1098 h->max_contexts = FFMIN(h->max_contexts, nb_slices);
1100 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1101 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1103 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1107 for (i = 0; i < h->slice_context_count; i++) {
1108 H264SliceContext *sl = &h->slice_ctx[i];
1111 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1112 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1113 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1115 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1116 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1122 h->context_initialized = 1;
1126 ff_h264_free_tables(h);
1127 h->context_initialized = 0;
1131 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1134 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1135 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1136 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1143 * Decode a slice header.
1144 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1146 * @param h h264context
1148 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1150 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1152 unsigned int first_mb_in_slice;
1153 unsigned int pps_id;
1155 unsigned int slice_type, tmp, i, j;
1156 int last_pic_structure, last_pic_droppable;
1158 int needs_reinit = 0;
1159 int field_pic_flag, bottom_field_flag;
1160 int first_slice = sl == h->slice_ctx && !h->current_slice;
1161 int frame_num, droppable, picture_structure;
1162 int mb_aff_frame, last_mb_aff_frame;
1166 av_assert0(!h->setup_finished);
1168 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1169 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1171 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1173 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1174 if (h->current_slice) {
1175 if (h->setup_finished) {
1176 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
1177 return AVERROR_INVALIDDATA;
1179 if (h->max_contexts > 1) {
1180 if (!h->single_decode_warning) {
1181 av_log(h->avctx, AV_LOG_WARNING, "Cannot decode multiple access units as slice threads\n");
1182 h->single_decode_warning = 1;
1184 h->max_contexts = 1;
1185 return SLICE_SINGLETHREAD;
1188 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1189 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1190 h->current_slice = 0;
1193 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1194 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1195 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1196 h->current_slice = 0;
1197 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1198 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1199 h->cur_pic_ptr = NULL;
1203 return AVERROR_INVALIDDATA;
1206 if (!h->first_field) {
1207 if (h->cur_pic_ptr && !h->droppable) {
1208 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1209 h->picture_structure == PICT_BOTTOM_FIELD);
1211 h->cur_pic_ptr = NULL;
1215 if (!h->current_slice)
1216 av_assert0(sl == h->slice_ctx);
1218 slice_type = get_ue_golomb_31(&sl->gb);
1219 if (slice_type > 9) {
1220 av_log(h->avctx, AV_LOG_ERROR,
1221 "slice type %d too large at %d\n",
1222 slice_type, first_mb_in_slice);
1223 return AVERROR_INVALIDDATA;
1225 if (slice_type > 4) {
1227 sl->slice_type_fixed = 1;
1229 sl->slice_type_fixed = 0;
1231 slice_type = golomb_to_pict_type[slice_type];
1232 sl->slice_type = slice_type;
1233 sl->slice_type_nos = slice_type & 3;
1235 if (h->nal_unit_type == NAL_IDR_SLICE &&
1236 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1237 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1238 return AVERROR_INVALIDDATA;
1241 if (h->current_slice == 0 && !h->first_field) {
1243 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1244 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1245 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1246 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE && h->sei_recovery_frame_cnt < 0) ||
1247 h->avctx->skip_frame >= AVDISCARD_ALL) {
1248 return SLICE_SKIPED;
1252 // to make a few old functions happy, it's wrong though
1253 if (!h->setup_finished)
1254 h->pict_type = sl->slice_type;
1256 pps_id = get_ue_golomb(&sl->gb);
1257 if (pps_id >= MAX_PPS_COUNT) {
1258 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1259 return AVERROR_INVALIDDATA;
1261 if (!h->pps_buffers[pps_id]) {
1262 av_log(h->avctx, AV_LOG_ERROR,
1263 "non-existing PPS %u referenced\n",
1265 return AVERROR_INVALIDDATA;
1267 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1268 av_log(h->avctx, AV_LOG_ERROR,
1269 "PPS change from %d to %d forbidden\n",
1270 h->au_pps_id, pps_id);
1271 return AVERROR_INVALIDDATA;
1274 pps = h->pps_buffers[pps_id];
1276 if (!h->sps_buffers[pps->sps_id]) {
1277 av_log(h->avctx, AV_LOG_ERROR,
1278 "non-existing SPS %u referenced\n",
1280 return AVERROR_INVALIDDATA;
1284 h->pps = *h->pps_buffers[pps_id];
1285 } else if (h->setup_finished && h->dequant_coeff_pps != pps_id) {
1286 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1287 return AVERROR_INVALIDDATA;
1290 if (pps->sps_id != h->sps.sps_id ||
1291 pps->sps_id != h->current_sps_id ||
1292 h->sps_buffers[pps->sps_id]->new) {
1295 av_log(h->avctx, AV_LOG_ERROR,
1296 "SPS changed in the middle of the frame\n");
1297 return AVERROR_INVALIDDATA;
1300 h->sps = *h->sps_buffers[h->pps.sps_id];
1302 if (h->mb_width != h->sps.mb_width ||
1303 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1304 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1305 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1309 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1310 h->chroma_format_idc != h->sps.chroma_format_idc)
1313 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1314 (h->sps.bitstream_restriction_flag &&
1315 !h->sps.num_reorder_frames)) {
1316 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1317 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1318 "Reenabling low delay requires a codec flush.\n");
1323 if (h->avctx->has_b_frames < 2)
1324 h->avctx->has_b_frames = !h->low_delay;
1328 must_reinit = (h->context_initialized &&
1329 ( 16*h->sps.mb_width != h->avctx->coded_width
1330 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1331 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1332 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1333 || h->mb_width != h->sps.mb_width
1334 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1336 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1337 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1340 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1343 if (!h->setup_finished) {
1344 h->avctx->profile = ff_h264_get_profile(&h->sps);
1345 h->avctx->level = h->sps.level_idc;
1346 h->avctx->refs = h->sps.ref_frame_count;
1348 h->mb_width = h->sps.mb_width;
1349 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1350 h->mb_num = h->mb_width * h->mb_height;
1351 h->mb_stride = h->mb_width + 1;
1353 h->b_stride = h->mb_width * 4;
1355 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1357 h->width = 16 * h->mb_width;
1358 h->height = 16 * h->mb_height;
1360 ret = init_dimensions(h);
1364 if (h->sps.video_signal_type_present_flag) {
1365 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1367 if (h->sps.colour_description_present_flag) {
1368 if (h->avctx->colorspace != h->sps.colorspace)
1370 h->avctx->color_primaries = h->sps.color_primaries;
1371 h->avctx->color_trc = h->sps.color_trc;
1372 h->avctx->colorspace = h->sps.colorspace;
1377 if (h->context_initialized &&
1378 (must_reinit || needs_reinit)) {
1379 h->context_initialized = 0;
1380 if (sl != h->slice_ctx) {
1381 av_log(h->avctx, AV_LOG_ERROR,
1382 "changing width %d -> %d / height %d -> %d on "
1384 h->width, h->avctx->coded_width,
1385 h->height, h->avctx->coded_height,
1386 h->current_slice + 1);
1387 return AVERROR_INVALIDDATA;
1390 av_assert1(first_slice);
1392 ff_h264_flush_change(h);
1394 if ((ret = get_pixel_format(h, 1)) < 0)
1396 h->avctx->pix_fmt = ret;
1398 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1399 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1401 if ((ret = h264_slice_header_init(h)) < 0) {
1402 av_log(h->avctx, AV_LOG_ERROR,
1403 "h264_slice_header_init() failed\n");
1407 if (!h->context_initialized) {
1408 if (sl != h->slice_ctx) {
1409 av_log(h->avctx, AV_LOG_ERROR,
1410 "Cannot (re-)initialize context during parallel decoding.\n");
1411 return AVERROR_PATCHWELCOME;
1414 if ((ret = get_pixel_format(h, 1)) < 0)
1416 h->avctx->pix_fmt = ret;
1418 if ((ret = h264_slice_header_init(h)) < 0) {
1419 av_log(h->avctx, AV_LOG_ERROR,
1420 "h264_slice_header_init() failed\n");
1425 if (!h->current_slice && h->dequant_coeff_pps != pps_id) {
1426 h->dequant_coeff_pps = pps_id;
1427 ff_h264_init_dequant_tables(h);
1430 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1432 if (h->frame_num != frame_num) {
1433 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1434 h->frame_num, frame_num);
1435 return AVERROR_INVALIDDATA;
1439 if (!h->setup_finished)
1440 h->frame_num = frame_num;
1444 last_mb_aff_frame = h->mb_aff_frame;
1445 last_pic_structure = h->picture_structure;
1446 last_pic_droppable = h->droppable;
1448 droppable = h->nal_ref_idc == 0;
1449 if (h->sps.frame_mbs_only_flag) {
1450 picture_structure = PICT_FRAME;
1452 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1453 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1456 field_pic_flag = get_bits1(&sl->gb);
1458 if (field_pic_flag) {
1459 bottom_field_flag = get_bits1(&sl->gb);
1460 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1462 picture_structure = PICT_FRAME;
1463 mb_aff_frame = h->sps.mb_aff;
1467 if (h->current_slice) {
1468 if (last_pic_structure != picture_structure ||
1469 last_pic_droppable != droppable ||
1470 last_mb_aff_frame != mb_aff_frame) {
1471 av_log(h->avctx, AV_LOG_ERROR,
1472 "Changing field mode (%d -> %d) between slices is not allowed\n",
1473 last_pic_structure, h->picture_structure);
1474 return AVERROR_INVALIDDATA;
1475 } else if (!h->cur_pic_ptr) {
1476 av_log(h->avctx, AV_LOG_ERROR,
1477 "unset cur_pic_ptr on slice %d\n",
1478 h->current_slice + 1);
1479 return AVERROR_INVALIDDATA;
1483 if (!h->setup_finished) {
1484 h->droppable = droppable;
1485 h->picture_structure = picture_structure;
1486 h->mb_aff_frame = mb_aff_frame;
1488 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1490 if (h->current_slice == 0) {
1491 /* Shorten frame num gaps so we don't have to allocate reference
1492 * frames just to throw them away */
1493 if (h->frame_num != h->prev_frame_num) {
1494 int unwrap_prev_frame_num = h->prev_frame_num;
1495 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1497 if (unwrap_prev_frame_num > h->frame_num)
1498 unwrap_prev_frame_num -= max_frame_num;
1500 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1501 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1502 if (unwrap_prev_frame_num < 0)
1503 unwrap_prev_frame_num += max_frame_num;
1505 h->prev_frame_num = unwrap_prev_frame_num;
1509 /* See if we have a decoded first field looking for a pair...
1510 * Here, we're using that to see if we should mark previously
1511 * decode frames as "finished".
1512 * We have to do that before the "dummy" in-between frame allocation,
1513 * since that can modify h->cur_pic_ptr. */
1514 if (h->first_field) {
1515 av_assert0(h->cur_pic_ptr);
1516 av_assert0(h->cur_pic_ptr->f->buf[0]);
1517 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1519 /* Mark old field/frame as completed */
1520 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1521 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1522 last_pic_structure == PICT_BOTTOM_FIELD);
1525 /* figure out if we have a complementary field pair */
1526 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1527 /* Previous field is unmatched. Don't display it, but let it
1528 * remain for reference if marked as such. */
1529 if (last_pic_structure != PICT_FRAME) {
1530 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1531 last_pic_structure == PICT_TOP_FIELD);
1534 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1535 /* This and previous field were reference, but had
1536 * different frame_nums. Consider this field first in
1537 * pair. Throw away previous field except for reference
1539 if (last_pic_structure != PICT_FRAME) {
1540 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1541 last_pic_structure == PICT_TOP_FIELD);
1544 /* Second field in complementary pair */
1545 if (!((last_pic_structure == PICT_TOP_FIELD &&
1546 h->picture_structure == PICT_BOTTOM_FIELD) ||
1547 (last_pic_structure == PICT_BOTTOM_FIELD &&
1548 h->picture_structure == PICT_TOP_FIELD))) {
1549 av_log(h->avctx, AV_LOG_ERROR,
1550 "Invalid field mode combination %d/%d\n",
1551 last_pic_structure, h->picture_structure);
1552 h->picture_structure = last_pic_structure;
1553 h->droppable = last_pic_droppable;
1554 return AVERROR_INVALIDDATA;
1555 } else if (last_pic_droppable != h->droppable) {
1556 avpriv_request_sample(h->avctx,
1557 "Found reference and non-reference fields in the same frame, which");
1558 h->picture_structure = last_pic_structure;
1559 h->droppable = last_pic_droppable;
1560 return AVERROR_PATCHWELCOME;
1566 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1567 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1568 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1569 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1570 h->frame_num, h->prev_frame_num);
1571 if (!h->sps.gaps_in_frame_num_allowed_flag)
1572 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1573 h->last_pocs[i] = INT_MIN;
1574 ret = h264_frame_start(h);
1580 h->prev_frame_num++;
1581 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1582 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1583 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1584 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1585 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1586 ret = ff_generate_sliding_window_mmcos(h, 1);
1587 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1589 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1590 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1592 /* Error concealment: If a ref is missing, copy the previous ref
1594 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1595 * many assumptions about there being no actual duplicates.
1596 * FIXME: This does not copy padding for out-of-frame motion
1597 * vectors. Given we are concealing a lost frame, this probably
1598 * is not noticeable by comparison, but it should be fixed. */
1599 if (h->short_ref_count) {
1601 h->short_ref[0]->f->width == prev->f->width &&
1602 h->short_ref[0]->f->height == prev->f->height &&
1603 h->short_ref[0]->f->format == prev->f->format) {
1604 av_image_copy(h->short_ref[0]->f->data,
1605 h->short_ref[0]->f->linesize,
1606 (const uint8_t **)prev->f->data,
1611 h->short_ref[0]->poc = prev->poc + 2;
1613 h->short_ref[0]->frame_num = h->prev_frame_num;
1617 /* See if we have a decoded first field looking for a pair...
1618 * We're using that to see whether to continue decoding in that
1619 * frame, or to allocate a new one. */
1620 if (h->first_field) {
1621 av_assert0(h->cur_pic_ptr);
1622 av_assert0(h->cur_pic_ptr->f->buf[0]);
1623 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1625 /* figure out if we have a complementary field pair */
1626 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1627 /* Previous field is unmatched. Don't display it, but let it
1628 * remain for reference if marked as such. */
1629 h->missing_fields ++;
1630 h->cur_pic_ptr = NULL;
1631 h->first_field = FIELD_PICTURE(h);
1633 h->missing_fields = 0;
1634 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1635 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1636 h->picture_structure==PICT_BOTTOM_FIELD);
1637 /* This and the previous field had different frame_nums.
1638 * Consider this field first in pair. Throw away previous
1639 * one except for reference purposes. */
1641 h->cur_pic_ptr = NULL;
1643 /* Second field in complementary pair */
1648 /* Frame or first field in a potentially complementary pair */
1649 h->first_field = FIELD_PICTURE(h);
1652 if (!FIELD_PICTURE(h) || h->first_field) {
1653 if (h264_frame_start(h) < 0) {
1655 return AVERROR_INVALIDDATA;
1658 release_unused_pictures(h, 0);
1660 /* Some macroblocks can be accessed before they're available in case
1661 * of lost slices, MBAFF or threading. */
1662 if (FIELD_PICTURE(h)) {
1663 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1664 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1666 memset(h->slice_table, -1,
1667 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1671 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1672 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1673 first_mb_in_slice >= h->mb_num) {
1674 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1675 return AVERROR_INVALIDDATA;
1677 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1678 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1679 FIELD_OR_MBAFF_PICTURE(h);
1680 if (h->picture_structure == PICT_BOTTOM_FIELD)
1681 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1682 av_assert1(sl->mb_y < h->mb_height);
1684 if (h->picture_structure == PICT_FRAME) {
1685 h->curr_pic_num = h->frame_num;
1686 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1688 h->curr_pic_num = 2 * h->frame_num + 1;
1689 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1692 if (h->nal_unit_type == NAL_IDR_SLICE)
1693 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1695 if (h->sps.poc_type == 0) {
1696 int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1698 if (!h->setup_finished)
1699 h->poc_lsb = poc_lsb;
1701 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1702 int delta_poc_bottom = get_se_golomb(&sl->gb);
1703 if (!h->setup_finished)
1704 h->delta_poc_bottom = delta_poc_bottom;
1708 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1709 int delta_poc = get_se_golomb(&sl->gb);
1711 if (!h->setup_finished)
1712 h->delta_poc[0] = delta_poc;
1714 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1715 delta_poc = get_se_golomb(&sl->gb);
1717 if (!h->setup_finished)
1718 h->delta_poc[1] = delta_poc;
1722 if (!h->setup_finished)
1723 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1725 if (h->pps.redundant_pic_cnt_present)
1726 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1728 ret = ff_set_ref_count(h, sl);
1732 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1733 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1735 sl->ref_count[1] = sl->ref_count[0] = 0;
1740 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1741 (h->pps.weighted_bipred_idc == 1 &&
1742 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1743 ff_pred_weight_table(h, sl);
1744 else if (h->pps.weighted_bipred_idc == 2 &&
1745 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1746 implicit_weight_table(h, sl, -1);
1749 for (i = 0; i < 2; i++) {
1750 sl->luma_weight_flag[i] = 0;
1751 sl->chroma_weight_flag[i] = 0;
1755 // If frame-mt is enabled, only update mmco tables for the first slice
1756 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1757 // or h->mmco, which will cause ref list mix-ups and decoding errors
1758 // further down the line. This may break decoding if the first slice is
1759 // corrupt, thus we only do this if frame-mt is enabled.
1760 if (h->nal_ref_idc) {
1761 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1762 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1763 h->current_slice == 0);
1764 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1765 return AVERROR_INVALIDDATA;
1768 if (FRAME_MBAFF(h)) {
1769 ff_h264_fill_mbaff_ref_list(h, sl);
1771 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1772 implicit_weight_table(h, sl, 0);
1773 implicit_weight_table(h, sl, 1);
1777 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1778 ff_h264_direct_dist_scale_factor(h, sl);
1779 ff_h264_direct_ref_list_init(h, sl);
1781 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1782 tmp = get_ue_golomb_31(&sl->gb);
1784 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1785 return AVERROR_INVALIDDATA;
1787 sl->cabac_init_idc = tmp;
1790 sl->last_qscale_diff = 0;
1791 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1792 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1793 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1794 return AVERROR_INVALIDDATA;
1797 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1798 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1799 // FIXME qscale / qp ... stuff
1800 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1801 get_bits1(&sl->gb); /* sp_for_switch_flag */
1802 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1803 sl->slice_type == AV_PICTURE_TYPE_SI)
1804 get_se_golomb(&sl->gb); /* slice_qs_delta */
1806 sl->deblocking_filter = 1;
1807 sl->slice_alpha_c0_offset = 0;
1808 sl->slice_beta_offset = 0;
1809 if (h->pps.deblocking_filter_parameters_present) {
1810 tmp = get_ue_golomb_31(&sl->gb);
1812 av_log(h->avctx, AV_LOG_ERROR,
1813 "deblocking_filter_idc %u out of range\n", tmp);
1814 return AVERROR_INVALIDDATA;
1816 sl->deblocking_filter = tmp;
1817 if (sl->deblocking_filter < 2)
1818 sl->deblocking_filter ^= 1; // 1<->0
1820 if (sl->deblocking_filter) {
1821 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1822 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1823 if (sl->slice_alpha_c0_offset > 12 ||
1824 sl->slice_alpha_c0_offset < -12 ||
1825 sl->slice_beta_offset > 12 ||
1826 sl->slice_beta_offset < -12) {
1827 av_log(h->avctx, AV_LOG_ERROR,
1828 "deblocking filter parameters %d %d out of range\n",
1829 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1830 return AVERROR_INVALIDDATA;
1835 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1836 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1837 h->nal_unit_type != NAL_IDR_SLICE) ||
1838 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1839 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1840 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1841 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1842 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1843 h->nal_ref_idc == 0))
1844 sl->deblocking_filter = 0;
1846 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1847 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1848 /* Cheat slightly for speed:
1849 * Do not bother to deblock across slices. */
1850 sl->deblocking_filter = 2;
1852 h->max_contexts = 1;
1853 if (!h->single_decode_warning) {
1854 av_log(h->avctx, AV_LOG_INFO,
1855 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1856 "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1857 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1858 "but will generate non-standard-compliant output.\n");
1859 h->single_decode_warning = 1;
1861 if (sl != h->slice_ctx) {
1862 av_log(h->avctx, AV_LOG_ERROR,
1863 "Deblocking switched inside frame.\n");
1864 return SLICE_SINGLETHREAD;
1868 sl->qp_thresh = 15 -
1869 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1871 h->pps.chroma_qp_index_offset[0],
1872 h->pps.chroma_qp_index_offset[1]) +
1873 6 * (h->sps.bit_depth_luma - 8);
1875 sl->slice_num = ++h->current_slice;
1878 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1879 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1880 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1881 && sl->slice_num >= MAX_SLICES) {
1882 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1883 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", sl->slice_num, MAX_SLICES);
1886 for (j = 0; j < 2; j++) {
1888 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1889 for (i = 0; i < 16; i++) {
1891 if (j < sl->list_count && i < sl->ref_count[j] &&
1892 sl->ref_list[j][i].parent->f->buf[0]) {
1894 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1895 for (k = 0; k < h->short_ref_count; k++)
1896 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1900 for (k = 0; k < h->long_ref_count; k++)
1901 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1902 id_list[i] = h->short_ref_count + k;
1910 for (i = 0; i < 16; i++)
1911 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1913 ref2frm[18 + 1] = -1;
1914 for (i = 16; i < 48; i++)
1915 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1916 (sl->ref_list[j][i].reference & 3);
1919 h->au_pps_id = pps_id;
1921 h->sps_buffers[h->pps.sps_id]->new = 0;
1922 h->current_sps_id = h->pps.sps_id;
1924 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1925 av_log(h->avctx, AV_LOG_DEBUG,
1926 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
1928 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1930 av_get_picture_type_char(sl->slice_type),
1931 sl->slice_type_fixed ? " fix" : "",
1932 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1933 pps_id, h->frame_num,
1934 h->cur_pic_ptr->field_poc[0],
1935 h->cur_pic_ptr->field_poc[1],
1936 sl->ref_count[0], sl->ref_count[1],
1938 sl->deblocking_filter,
1939 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1941 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1942 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1948 int ff_h264_get_slice_type(const H264SliceContext *sl)
1950 switch (sl->slice_type) {
1951 case AV_PICTURE_TYPE_P:
1953 case AV_PICTURE_TYPE_B:
1955 case AV_PICTURE_TYPE_I:
1957 case AV_PICTURE_TYPE_SP:
1959 case AV_PICTURE_TYPE_SI:
1962 return AVERROR_INVALIDDATA;
1966 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1967 H264SliceContext *sl,
1968 int mb_type, int top_xy,
1969 int left_xy[LEFT_MBS],
1971 int left_type[LEFT_MBS],
1972 int mb_xy, int list)
1974 int b_stride = h->b_stride;
1975 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1976 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1977 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1978 if (USES_LIST(top_type, list)) {
1979 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1980 const int b8_xy = 4 * top_xy + 2;
1981 int *ref2frm = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1982 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1983 ref_cache[0 - 1 * 8] =
1984 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
1985 ref_cache[2 - 1 * 8] =
1986 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
1988 AV_ZERO128(mv_dst - 1 * 8);
1989 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1992 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1993 if (USES_LIST(left_type[LTOP], list)) {
1994 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1995 const int b8_xy = 4 * left_xy[LTOP] + 1;
1996 int *ref2frm = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1997 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1998 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1999 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2000 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2002 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2003 ref_cache[-1 + 16] =
2004 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2006 AV_ZERO32(mv_dst - 1 + 0);
2007 AV_ZERO32(mv_dst - 1 + 8);
2008 AV_ZERO32(mv_dst - 1 + 16);
2009 AV_ZERO32(mv_dst - 1 + 24);
2012 ref_cache[-1 + 16] =
2013 ref_cache[-1 + 24] = LIST_NOT_USED;
2018 if (!USES_LIST(mb_type, list)) {
2019 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2020 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2021 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2022 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2023 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2028 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2029 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
2030 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
2031 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
2032 AV_WN32A(&ref_cache[0 * 8], ref01);
2033 AV_WN32A(&ref_cache[1 * 8], ref01);
2034 AV_WN32A(&ref_cache[2 * 8], ref23);
2035 AV_WN32A(&ref_cache[3 * 8], ref23);
2039 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2040 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2041 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2042 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2043 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2049 * @return non zero if the loop filter can be skipped
2051 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2053 const int mb_xy = sl->mb_xy;
2054 int top_xy, left_xy[LEFT_MBS];
2055 int top_type, left_type[LEFT_MBS];
2059 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2061 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2062 * stuff, I can't imagine that these complex rules are worth it. */
2064 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2065 if (FRAME_MBAFF(h)) {
2066 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2067 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2069 if (left_mb_field_flag != curr_mb_field_flag)
2070 left_xy[LTOP] -= h->mb_stride;
2072 if (curr_mb_field_flag)
2073 top_xy += h->mb_stride &
2074 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2075 if (left_mb_field_flag != curr_mb_field_flag)
2076 left_xy[LBOT] += h->mb_stride;
2080 sl->top_mb_xy = top_xy;
2081 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2082 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2084 /* For sufficiently low qp, filtering wouldn't do anything.
2085 * This is a conservative estimate: could also check beta_offset
2086 * and more accurate chroma_qp. */
2087 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2088 int qp = h->cur_pic.qscale_table[mb_xy];
2089 if (qp <= qp_thresh &&
2090 (left_xy[LTOP] < 0 ||
2091 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2093 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2094 if (!FRAME_MBAFF(h))
2096 if ((left_xy[LTOP] < 0 ||
2097 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2098 (top_xy < h->mb_stride ||
2099 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2104 top_type = h->cur_pic.mb_type[top_xy];
2105 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2106 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2107 if (sl->deblocking_filter == 2) {
2108 if (h->slice_table[top_xy] != sl->slice_num)
2110 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2111 left_type[LTOP] = left_type[LBOT] = 0;
2113 if (h->slice_table[top_xy] == 0xFFFF)
2115 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2116 left_type[LTOP] = left_type[LBOT] = 0;
2118 sl->top_type = top_type;
2119 sl->left_type[LTOP] = left_type[LTOP];
2120 sl->left_type[LBOT] = left_type[LBOT];
2122 if (IS_INTRA(mb_type))
2125 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2126 top_type, left_type, mb_xy, 0);
2127 if (sl->list_count == 2)
2128 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2129 top_type, left_type, mb_xy, 1);
2131 nnz = h->non_zero_count[mb_xy];
2132 nnz_cache = sl->non_zero_count_cache;
2133 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2134 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2135 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2136 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2137 sl->cbp = h->cbp_table[mb_xy];
2140 nnz = h->non_zero_count[top_xy];
2141 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2144 if (left_type[LTOP]) {
2145 nnz = h->non_zero_count[left_xy[LTOP]];
2146 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2147 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2148 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2149 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2152 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2153 * from what the loop filter needs */
2154 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2155 if (IS_8x8DCT(top_type)) {
2156 nnz_cache[4 + 8 * 0] =
2157 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2158 nnz_cache[6 + 8 * 0] =
2159 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2161 if (IS_8x8DCT(left_type[LTOP])) {
2162 nnz_cache[3 + 8 * 1] =
2163 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2165 if (IS_8x8DCT(left_type[LBOT])) {
2166 nnz_cache[3 + 8 * 3] =
2167 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2170 if (IS_8x8DCT(mb_type)) {
2171 nnz_cache[scan8[0]] =
2172 nnz_cache[scan8[1]] =
2173 nnz_cache[scan8[2]] =
2174 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2176 nnz_cache[scan8[0 + 4]] =
2177 nnz_cache[scan8[1 + 4]] =
2178 nnz_cache[scan8[2 + 4]] =
2179 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2181 nnz_cache[scan8[0 + 8]] =
2182 nnz_cache[scan8[1 + 8]] =
2183 nnz_cache[scan8[2 + 8]] =
2184 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2186 nnz_cache[scan8[0 + 12]] =
2187 nnz_cache[scan8[1 + 12]] =
2188 nnz_cache[scan8[2 + 12]] =
2189 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2196 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2198 uint8_t *dest_y, *dest_cb, *dest_cr;
2199 int linesize, uvlinesize, mb_x, mb_y;
2200 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2201 const int old_slice_type = sl->slice_type;
2202 const int pixel_shift = h->pixel_shift;
2203 const int block_h = 16 >> h->chroma_y_shift;
2205 if (sl->deblocking_filter) {
2206 for (mb_x = start_x; mb_x < end_x; mb_x++)
2207 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2209 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2210 sl->slice_num = h->slice_table[mb_xy];
2211 mb_type = h->cur_pic.mb_type[mb_xy];
2212 sl->list_count = h->list_counts[mb_xy];
2216 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2220 dest_y = h->cur_pic.f->data[0] +
2221 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2222 dest_cb = h->cur_pic.f->data[1] +
2223 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2224 mb_y * sl->uvlinesize * block_h;
2225 dest_cr = h->cur_pic.f->data[2] +
2226 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2227 mb_y * sl->uvlinesize * block_h;
2228 // FIXME simplify above
2231 linesize = sl->mb_linesize = sl->linesize * 2;
2232 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2233 if (mb_y & 1) { // FIXME move out of this function?
2234 dest_y -= sl->linesize * 15;
2235 dest_cb -= sl->uvlinesize * (block_h - 1);
2236 dest_cr -= sl->uvlinesize * (block_h - 1);
2239 linesize = sl->mb_linesize = sl->linesize;
2240 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2242 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2244 if (fill_filter_caches(h, sl, mb_type))
2246 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2247 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2249 if (FRAME_MBAFF(h)) {
2250 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2251 linesize, uvlinesize);
2253 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2254 dest_cr, linesize, uvlinesize);
2258 sl->slice_type = old_slice_type;
2260 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2261 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2262 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2265 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2267 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2268 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2269 h->cur_pic.mb_type[mb_xy - 1] :
2270 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2271 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2272 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2276 * Draw edges and report progress for the last MB row.
2278 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2280 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2281 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2282 int height = 16 << FRAME_MBAFF(h);
2283 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2285 if (sl->deblocking_filter) {
2286 if ((top + height) >= pic_height)
2287 height += deblock_border;
2288 top -= deblock_border;
2291 if (top >= pic_height || (top + height) < 0)
2294 height = FFMIN(height, pic_height - top);
2296 height = top + height;
2300 ff_h264_draw_horiz_band(h, sl, top, height);
2302 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2305 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2306 h->picture_structure == PICT_BOTTOM_FIELD);
2309 static void er_add_slice(H264SliceContext *sl,
2310 int startx, int starty,
2311 int endx, int endy, int status)
2313 if (!sl->h264->enable_er)
2316 if (CONFIG_ERROR_RESILIENCE) {
2317 ERContext *er = &sl->h264->slice_ctx[0].er;
2319 ff_er_add_slice(er, startx, starty, endx, endy, status);
2323 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2325 H264SliceContext *sl = arg;
2326 const H264Context *h = sl->h264;
2327 int lf_x_start = sl->mb_x;
2330 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2331 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2333 ret = alloc_scratch_buffers(sl, sl->linesize);
2337 sl->mb_skip_run = -1;
2339 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2341 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2342 avctx->codec_id != AV_CODEC_ID_H264 ||
2343 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2345 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2346 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2348 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2349 prev_status &= ~ VP_START;
2350 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2351 h->slice_ctx[0].er.error_occurred = 1;
2357 align_get_bits(&sl->gb);
2360 ret = ff_init_cabac_decoder(&sl->cabac,
2361 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2362 (get_bits_left(&sl->gb) + 7) / 8);
2366 ff_h264_init_cabac_states(h, sl);
2371 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2372 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2373 sl->next_slice_idx);
2374 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2375 sl->mb_y, ER_MB_ERROR);
2376 return AVERROR_INVALIDDATA;
2379 ret = ff_h264_decode_mb_cabac(h, sl);
2380 // STOP_TIMER("decode_mb_cabac")
2383 ff_h264_hl_decode_mb(h, sl);
2385 // FIXME optimal? or let mb_decode decode 16x32 ?
2386 if (ret >= 0 && FRAME_MBAFF(h)) {
2389 ret = ff_h264_decode_mb_cabac(h, sl);
2392 ff_h264_hl_decode_mb(h, sl);
2395 eos = get_cabac_terminate(&sl->cabac);
2397 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2398 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2399 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2400 sl->mb_y, ER_MB_END);
2401 if (sl->mb_x >= lf_x_start)
2402 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2405 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2406 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2407 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2408 av_log(h->avctx, AV_LOG_ERROR,
2409 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2411 sl->cabac.bytestream_end - sl->cabac.bytestream);
2412 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2413 sl->mb_y, ER_MB_ERROR);
2414 return AVERROR_INVALIDDATA;
2417 if (++sl->mb_x >= h->mb_width) {
2418 loop_filter(h, sl, lf_x_start, sl->mb_x);
2419 sl->mb_x = lf_x_start = 0;
2420 decode_finish_row(h, sl);
2422 if (FIELD_OR_MBAFF_PICTURE(h)) {
2424 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2425 predict_field_decoding_flag(h, sl);
2429 if (eos || sl->mb_y >= h->mb_height) {
2430 ff_tlog(h->avctx, "slice end %d %d\n",
2431 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2432 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2433 sl->mb_y, ER_MB_END);
2434 if (sl->mb_x > lf_x_start)
2435 loop_filter(h, sl, lf_x_start, sl->mb_x);
2443 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2444 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2445 sl->next_slice_idx);
2446 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2447 sl->mb_y, ER_MB_ERROR);
2448 return AVERROR_INVALIDDATA;
2451 ret = ff_h264_decode_mb_cavlc(h, sl);
2454 ff_h264_hl_decode_mb(h, sl);
2456 // FIXME optimal? or let mb_decode decode 16x32 ?
2457 if (ret >= 0 && FRAME_MBAFF(h)) {
2459 ret = ff_h264_decode_mb_cavlc(h, sl);
2462 ff_h264_hl_decode_mb(h, sl);
2467 av_log(h->avctx, AV_LOG_ERROR,
2468 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2469 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2470 sl->mb_y, ER_MB_ERROR);
2474 if (++sl->mb_x >= h->mb_width) {
2475 loop_filter(h, sl, lf_x_start, sl->mb_x);
2476 sl->mb_x = lf_x_start = 0;
2477 decode_finish_row(h, sl);
2479 if (FIELD_OR_MBAFF_PICTURE(h)) {
2481 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2482 predict_field_decoding_flag(h, sl);
2484 if (sl->mb_y >= h->mb_height) {
2485 ff_tlog(h->avctx, "slice end %d %d\n",
2486 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2488 if ( get_bits_left(&sl->gb) == 0
2489 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2490 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2491 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2495 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2496 sl->mb_x, sl->mb_y, ER_MB_END);
2498 return AVERROR_INVALIDDATA;
2503 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2504 ff_tlog(h->avctx, "slice end %d %d\n",
2505 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2507 if (get_bits_left(&sl->gb) == 0) {
2508 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2509 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2510 if (sl->mb_x > lf_x_start)
2511 loop_filter(h, sl, lf_x_start, sl->mb_x);
2515 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2516 sl->mb_y, ER_MB_ERROR);
2518 return AVERROR_INVALIDDATA;
2526 * Call decode_slice() for each context.
2528 * @param h h264 master context
2529 * @param context_count number of contexts to execute
2531 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2533 AVCodecContext *const avctx = h->avctx;
2534 H264SliceContext *sl;
2537 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2539 h->slice_ctx[0].next_slice_idx = INT_MAX;
2541 if (h->avctx->hwaccel
2542 #if FF_API_CAP_VDPAU
2543 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2547 if (context_count == 1) {
2550 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2552 ret = decode_slice(avctx, &h->slice_ctx[0]);
2553 h->mb_y = h->slice_ctx[0].mb_y;
2556 av_assert0(context_count > 0);
2557 for (i = 0; i < context_count; i++) {
2558 int next_slice_idx = h->mb_width * h->mb_height;
2561 sl = &h->slice_ctx[i];
2562 if (CONFIG_ERROR_RESILIENCE) {
2563 sl->er.error_count = 0;
2566 /* make sure none of those slices overlap */
2567 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2568 for (j = 0; j < context_count; j++) {
2569 H264SliceContext *sl2 = &h->slice_ctx[j];
2570 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2572 if (i == j || slice_idx2 < slice_idx)
2574 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2576 sl->next_slice_idx = next_slice_idx;
2579 avctx->execute(avctx, decode_slice, h->slice_ctx,
2580 NULL, context_count, sizeof(h->slice_ctx[0]));
2582 /* pull back stuff from slices to master context */
2583 sl = &h->slice_ctx[context_count - 1];
2585 if (CONFIG_ERROR_RESILIENCE) {
2586 for (i = 1; i < context_count; i++)
2587 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;