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"
46 static const uint8_t field_scan[16+1] = {
47 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
48 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
49 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
50 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
53 static const uint8_t field_scan8x8[64+1] = {
54 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
55 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
56 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
57 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
58 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
59 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
60 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
61 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
62 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
63 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
64 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
65 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
66 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
67 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
68 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
69 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
72 static const uint8_t field_scan8x8_cavlc[64+1] = {
73 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
74 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
75 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
76 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
77 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
78 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
79 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
80 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
81 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
82 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
83 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
84 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
85 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
86 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
87 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
88 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
91 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
92 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
93 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
94 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
95 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
96 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
97 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
98 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
99 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
100 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
101 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
102 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
103 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
104 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
105 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
106 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
107 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
108 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
111 static void release_unused_pictures(H264Context *h, int remove_current)
115 /* release non reference frames */
116 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
117 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
118 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
119 ff_h264_unref_picture(h, &h->DPB[i]);
124 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
126 const H264Context *h = sl->h264;
127 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
129 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
130 // edge emu needs blocksize + filter length - 1
131 // (= 21x21 for h264)
132 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
134 av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
135 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
136 av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
137 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
139 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
140 !sl->top_borders[0] || !sl->top_borders[1]) {
141 av_freep(&sl->bipred_scratchpad);
142 av_freep(&sl->edge_emu_buffer);
143 av_freep(&sl->top_borders[0]);
144 av_freep(&sl->top_borders[1]);
146 sl->bipred_scratchpad_allocated = 0;
147 sl->edge_emu_buffer_allocated = 0;
148 sl->top_borders_allocated[0] = 0;
149 sl->top_borders_allocated[1] = 0;
150 return AVERROR(ENOMEM);
156 static int init_table_pools(H264Context *h)
158 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
159 const int mb_array_size = h->mb_stride * h->mb_height;
160 const int b4_stride = h->mb_width * 4 + 1;
161 const int b4_array_size = b4_stride * h->mb_height * 4;
163 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
165 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
166 sizeof(uint32_t), av_buffer_allocz);
167 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
168 sizeof(int16_t), av_buffer_allocz);
169 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
171 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
172 !h->ref_index_pool) {
173 av_buffer_pool_uninit(&h->qscale_table_pool);
174 av_buffer_pool_uninit(&h->mb_type_pool);
175 av_buffer_pool_uninit(&h->motion_val_pool);
176 av_buffer_pool_uninit(&h->ref_index_pool);
177 return AVERROR(ENOMEM);
183 static int alloc_picture(H264Context *h, H264Picture *pic)
187 av_assert0(!pic->f->data[0]);
190 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
191 AV_GET_BUFFER_FLAG_REF : 0);
195 pic->crop = h->sps.crop;
196 pic->crop_top = h->sps.crop_top;
197 pic->crop_left= h->sps.crop_left;
199 if (h->avctx->hwaccel) {
200 const AVHWAccel *hwaccel = h->avctx->hwaccel;
201 av_assert0(!pic->hwaccel_picture_private);
202 if (hwaccel->frame_priv_data_size) {
203 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
204 if (!pic->hwaccel_priv_buf)
205 return AVERROR(ENOMEM);
206 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
209 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
210 int h_chroma_shift, v_chroma_shift;
211 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
212 &h_chroma_shift, &v_chroma_shift);
214 for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
215 memset(pic->f->data[1] + pic->f->linesize[1]*i,
216 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
217 memset(pic->f->data[2] + pic->f->linesize[2]*i,
218 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
222 if (!h->qscale_table_pool) {
223 ret = init_table_pools(h);
228 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
229 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
230 if (!pic->qscale_table_buf || !pic->mb_type_buf)
233 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
234 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
236 for (i = 0; i < 2; i++) {
237 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
238 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
239 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
242 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
243 pic->ref_index[i] = pic->ref_index_buf[i]->data;
248 ff_h264_unref_picture(h, pic);
249 return (ret < 0) ? ret : AVERROR(ENOMEM);
252 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
259 static int find_unused_picture(H264Context *h)
263 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
264 if (pic_is_unused(h, &h->DPB[i]))
267 if (i == H264_MAX_PICTURE_COUNT)
268 return AVERROR_INVALIDDATA;
274 static void init_dequant8_coeff_table(H264Context *h)
277 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
279 for (i = 0; i < 6; i++) {
280 h->dequant8_coeff[i] = h->dequant8_buffer[i];
281 for (j = 0; j < i; j++)
282 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
283 64 * sizeof(uint8_t))) {
284 h->dequant8_coeff[i] = h->dequant8_buffer[j];
290 for (q = 0; q < max_qp + 1; q++) {
291 int shift = ff_h264_quant_div6[q];
292 int idx = ff_h264_quant_rem6[q];
293 for (x = 0; x < 64; x++)
294 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
295 ((uint32_t)ff_h264_dequant8_coeff_init[idx][ff_h264_dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
296 h->pps.scaling_matrix8[i][x]) << shift;
301 static void init_dequant4_coeff_table(H264Context *h)
304 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
305 for (i = 0; i < 6; i++) {
306 h->dequant4_coeff[i] = h->dequant4_buffer[i];
307 for (j = 0; j < i; j++)
308 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
309 16 * sizeof(uint8_t))) {
310 h->dequant4_coeff[i] = h->dequant4_buffer[j];
316 for (q = 0; q < max_qp + 1; q++) {
317 int shift = ff_h264_quant_div6[q] + 2;
318 int idx = ff_h264_quant_rem6[q];
319 for (x = 0; x < 16; x++)
320 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
321 ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
322 h->pps.scaling_matrix4[i][x]) << shift;
327 void ff_h264_init_dequant_tables(H264Context *h)
330 init_dequant4_coeff_table(h);
331 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
333 if (h->pps.transform_8x8_mode)
334 init_dequant8_coeff_table(h);
335 if (h->sps.transform_bypass) {
336 for (i = 0; i < 6; i++)
337 for (x = 0; x < 16; x++)
338 h->dequant4_coeff[i][0][x] = 1 << 6;
339 if (h->pps.transform_8x8_mode)
340 for (i = 0; i < 6; i++)
341 for (x = 0; x < 64; x++)
342 h->dequant8_coeff[i][0][x] = 1 << 6;
346 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
348 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
349 (((pic) && (pic) >= (old_ctx)->DPB && \
350 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
351 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
353 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
354 H264Context *new_base,
355 H264Context *old_base)
359 for (i = 0; i < count; i++) {
360 av_assert1(!from[i] ||
361 IN_RANGE(from[i], old_base, 1) ||
362 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
363 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
367 static int copy_parameter_set(void **to, void **from, int count, int size)
371 for (i = 0; i < count; i++) {
372 if (to[i] && !from[i]) {
374 } else if (from[i] && !to[i]) {
375 to[i] = av_malloc(size);
377 return AVERROR(ENOMEM);
381 memcpy(to[i], from[i], size);
387 #define copy_fields(to, from, start_field, end_field) \
388 memcpy(&(to)->start_field, &(from)->start_field, \
389 (char *)&(to)->end_field - (char *)&(to)->start_field)
391 static int h264_slice_header_init(H264Context *h);
393 int ff_h264_update_thread_context(AVCodecContext *dst,
394 const AVCodecContext *src)
396 H264Context *h = dst->priv_data, *h1 = src->priv_data;
397 int inited = h->context_initialized, err = 0;
405 (h->width != h1->width ||
406 h->height != h1->height ||
407 h->mb_width != h1->mb_width ||
408 h->mb_height != h1->mb_height ||
409 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
410 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
411 h->sps.colorspace != h1->sps.colorspace)) {
416 /* copy block_offset since frame_start may not be called */
417 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
420 if ((ret = copy_parameter_set((void **)h->sps_buffers,
421 (void **)h1->sps_buffers,
422 MAX_SPS_COUNT, sizeof(SPS))) < 0)
425 if ((ret = copy_parameter_set((void **)h->pps_buffers,
426 (void **)h1->pps_buffers,
427 MAX_PPS_COUNT, sizeof(PPS))) < 0)
431 if (need_reinit || !inited) {
432 h->width = h1->width;
433 h->height = h1->height;
434 h->mb_height = h1->mb_height;
435 h->mb_width = h1->mb_width;
436 h->mb_num = h1->mb_num;
437 h->mb_stride = h1->mb_stride;
438 h->b_stride = h1->b_stride;
440 if (h->context_initialized || h1->context_initialized) {
441 if ((err = h264_slice_header_init(h)) < 0) {
442 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
446 /* copy block_offset since frame_start may not be called */
447 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
450 h->avctx->coded_height = h1->avctx->coded_height;
451 h->avctx->coded_width = h1->avctx->coded_width;
452 h->avctx->width = h1->avctx->width;
453 h->avctx->height = h1->avctx->height;
454 h->coded_picture_number = h1->coded_picture_number;
455 h->first_field = h1->first_field;
456 h->picture_structure = h1->picture_structure;
457 h->droppable = h1->droppable;
458 h->low_delay = h1->low_delay;
459 h->backup_width = h1->backup_width;
460 h->backup_height = h1->backup_height;
461 h->backup_pix_fmt = h1->backup_pix_fmt;
463 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
464 ff_h264_unref_picture(h, &h->DPB[i]);
465 if (h1->DPB[i].f->buf[0] &&
466 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
470 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
471 ff_h264_unref_picture(h, &h->cur_pic);
472 if (h1->cur_pic.f->buf[0]) {
473 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
478 h->enable_er = h1->enable_er;
479 h->workaround_bugs = h1->workaround_bugs;
480 h->low_delay = h1->low_delay;
481 h->droppable = h1->droppable;
483 // extradata/NAL handling
484 h->is_avc = h1->is_avc;
485 h->nal_length_size = h1->nal_length_size;
486 h->x264_build = h1->x264_build;
488 // Dequantization matrices
489 // FIXME these are big - can they be only copied when PPS changes?
490 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
492 for (i = 0; i < 6; i++)
493 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
494 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
496 for (i = 0; i < 6; i++)
497 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
498 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
500 h->dequant_coeff_pps = h1->dequant_coeff_pps;
503 copy_fields(h, h1, poc_lsb, current_slice);
505 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
506 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
507 copy_picture_range(h->delayed_pic, h1->delayed_pic,
508 MAX_DELAYED_PIC_COUNT + 2, h, h1);
510 h->frame_recovered = h1->frame_recovered;
516 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
517 h->prev_poc_msb = h->poc_msb;
518 h->prev_poc_lsb = h->poc_lsb;
520 h->prev_frame_num_offset = h->frame_num_offset;
521 h->prev_frame_num = h->frame_num;
523 h->recovery_frame = h1->recovery_frame;
528 static int h264_frame_start(H264Context *h)
532 const int pixel_shift = h->pixel_shift;
534 1<<(h->sps.bit_depth_luma-1),
535 1<<(h->sps.bit_depth_chroma-1),
536 1<<(h->sps.bit_depth_chroma-1),
540 if (!ff_thread_can_start_frame(h->avctx)) {
541 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
545 release_unused_pictures(h, 1);
546 h->cur_pic_ptr = NULL;
548 i = find_unused_picture(h);
550 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
555 pic->reference = h->droppable ? 0 : h->picture_structure;
556 pic->f->coded_picture_number = h->coded_picture_number++;
557 pic->field_picture = h->picture_structure != PICT_FRAME;
558 pic->frame_num = h->frame_num;
561 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
563 * See decode_nal_units().
565 pic->f->key_frame = 0;
568 pic->invalid_gap = 0;
569 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
571 if ((ret = alloc_picture(h, pic)) < 0)
573 if(!h->frame_recovered && !h->avctx->hwaccel
575 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
578 ff_color_frame(pic->f, c);
580 h->cur_pic_ptr = pic;
581 ff_h264_unref_picture(h, &h->cur_pic);
582 if (CONFIG_ERROR_RESILIENCE) {
583 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
586 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
589 for (i = 0; i < h->nb_slice_ctx; i++) {
590 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
591 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
594 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
595 ff_er_frame_start(&h->slice_ctx[0].er);
596 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
597 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
600 for (i = 0; i < 16; i++) {
601 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
602 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
604 for (i = 0; i < 16; i++) {
605 h->block_offset[16 + i] =
606 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
607 h->block_offset[48 + 16 + i] =
608 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
611 /* We mark the current picture as non-reference after allocating it, so
612 * that if we break out due to an error it can be released automatically
613 * in the next ff_mpv_frame_start().
615 h->cur_pic_ptr->reference = 0;
617 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
619 h->next_output_pic = NULL;
621 assert(h->cur_pic_ptr->long_ref == 0);
626 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
628 uint8_t *src_cb, uint8_t *src_cr,
629 int linesize, int uvlinesize,
634 const int pixel_shift = h->pixel_shift;
635 int chroma444 = CHROMA444(h);
636 int chroma422 = CHROMA422(h);
639 src_cb -= uvlinesize;
640 src_cr -= uvlinesize;
642 if (!simple && FRAME_MBAFF(h)) {
645 top_border = sl->top_borders[0][sl->mb_x];
646 AV_COPY128(top_border, src_y + 15 * linesize);
648 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
649 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
652 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
653 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
654 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
655 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
657 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
658 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
660 } else if (chroma422) {
662 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
663 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
665 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
666 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
670 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
671 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
673 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
674 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
679 } else if (MB_MBAFF(sl)) {
685 top_border = sl->top_borders[top_idx][sl->mb_x];
686 /* There are two lines saved, the line above the top macroblock
687 * of a pair, and the line above the bottom macroblock. */
688 AV_COPY128(top_border, src_y + 16 * linesize);
690 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
692 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
695 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
696 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
697 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
698 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
700 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
701 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
703 } else if (chroma422) {
705 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
706 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
708 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
709 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
713 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
714 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
716 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
717 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
724 * Initialize implicit_weight table.
725 * @param field 0/1 initialize the weight for interlaced MBAFF
726 * -1 initializes the rest
728 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
730 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
732 for (i = 0; i < 2; i++) {
733 sl->pwt.luma_weight_flag[i] = 0;
734 sl->pwt.chroma_weight_flag[i] = 0;
738 if (h->picture_structure == PICT_FRAME) {
739 cur_poc = h->cur_pic_ptr->poc;
741 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
743 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
744 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
745 sl->pwt.use_weight = 0;
746 sl->pwt.use_weight_chroma = 0;
750 ref_count0 = sl->ref_count[0];
751 ref_count1 = sl->ref_count[1];
753 cur_poc = h->cur_pic_ptr->field_poc[field];
755 ref_count0 = 16 + 2 * sl->ref_count[0];
756 ref_count1 = 16 + 2 * sl->ref_count[1];
759 sl->pwt.use_weight = 2;
760 sl->pwt.use_weight_chroma = 2;
761 sl->pwt.luma_log2_weight_denom = 5;
762 sl->pwt.chroma_log2_weight_denom = 5;
764 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
765 int64_t poc0 = sl->ref_list[0][ref0].poc;
766 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
768 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
769 int poc1 = sl->ref_list[1][ref1].poc;
770 int td = av_clip_int8(poc1 - poc0);
772 int tb = av_clip_int8(cur_poc - poc0);
773 int tx = (16384 + (FFABS(td) >> 1)) / td;
774 int dist_scale_factor = (tb * tx + 32) >> 8;
775 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
776 w = 64 - dist_scale_factor;
780 sl->pwt.implicit_weight[ref0][ref1][0] =
781 sl->pwt.implicit_weight[ref0][ref1][1] = w;
783 sl->pwt.implicit_weight[ref0][ref1][field] = w;
790 * initialize scan tables
792 static void init_scan_tables(H264Context *h)
795 for (i = 0; i < 16; i++) {
796 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
797 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
798 h->field_scan[i] = TRANSPOSE(field_scan[i]);
801 for (i = 0; i < 64; i++) {
802 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
803 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
804 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
805 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
806 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
809 if (h->sps.transform_bypass) { // FIXME same ugly
810 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
811 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
812 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
813 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
814 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
815 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
817 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
818 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
819 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
820 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
821 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
822 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
826 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
828 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
829 CONFIG_H264_D3D11VA_HWACCEL + \
830 CONFIG_H264_VAAPI_HWACCEL + \
831 (CONFIG_H264_VDA_HWACCEL * 2) + \
832 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
833 CONFIG_H264_VDPAU_HWACCEL)
834 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
835 const enum AVPixelFormat *choices = pix_fmts;
838 switch (h->sps.bit_depth_luma) {
841 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
842 *fmt++ = AV_PIX_FMT_GBRP9;
844 *fmt++ = AV_PIX_FMT_YUV444P9;
845 } else if (CHROMA422(h))
846 *fmt++ = AV_PIX_FMT_YUV422P9;
848 *fmt++ = AV_PIX_FMT_YUV420P9;
852 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
853 *fmt++ = AV_PIX_FMT_GBRP10;
855 *fmt++ = AV_PIX_FMT_YUV444P10;
856 } else if (CHROMA422(h))
857 *fmt++ = AV_PIX_FMT_YUV422P10;
859 *fmt++ = AV_PIX_FMT_YUV420P10;
863 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
864 *fmt++ = AV_PIX_FMT_GBRP12;
866 *fmt++ = AV_PIX_FMT_YUV444P12;
867 } else if (CHROMA422(h))
868 *fmt++ = AV_PIX_FMT_YUV422P12;
870 *fmt++ = AV_PIX_FMT_YUV420P12;
874 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
875 *fmt++ = AV_PIX_FMT_GBRP14;
877 *fmt++ = AV_PIX_FMT_YUV444P14;
878 } else if (CHROMA422(h))
879 *fmt++ = AV_PIX_FMT_YUV422P14;
881 *fmt++ = AV_PIX_FMT_YUV420P14;
884 #if CONFIG_H264_VDPAU_HWACCEL
885 *fmt++ = AV_PIX_FMT_VDPAU;
888 if (h->avctx->colorspace == AVCOL_SPC_RGB)
889 *fmt++ = AV_PIX_FMT_GBRP;
890 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
891 *fmt++ = AV_PIX_FMT_YUVJ444P;
893 *fmt++ = AV_PIX_FMT_YUV444P;
894 } else if (CHROMA422(h)) {
895 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
896 *fmt++ = AV_PIX_FMT_YUVJ422P;
898 *fmt++ = AV_PIX_FMT_YUV422P;
900 #if CONFIG_H264_DXVA2_HWACCEL
901 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
903 #if CONFIG_H264_D3D11VA_HWACCEL
904 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
906 #if CONFIG_H264_VAAPI_HWACCEL
907 *fmt++ = AV_PIX_FMT_VAAPI;
909 #if CONFIG_H264_VDA_HWACCEL
910 *fmt++ = AV_PIX_FMT_VDA_VLD;
911 *fmt++ = AV_PIX_FMT_VDA;
913 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
914 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
916 if (h->avctx->codec->pix_fmts)
917 choices = h->avctx->codec->pix_fmts;
918 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
919 *fmt++ = AV_PIX_FMT_YUVJ420P;
921 *fmt++ = AV_PIX_FMT_YUV420P;
925 av_log(h->avctx, AV_LOG_ERROR,
926 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
927 return AVERROR_INVALIDDATA;
930 *fmt = AV_PIX_FMT_NONE;
932 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
933 if (choices[i] == h->avctx->pix_fmt && !force_callback)
935 return ff_thread_get_format(h->avctx, choices);
938 /* export coded and cropped frame dimensions to AVCodecContext */
939 static int init_dimensions(H264Context *h)
941 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
942 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
943 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
944 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
946 /* handle container cropping */
947 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
948 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
949 h->avctx->width <= width &&
950 h->avctx->height <= height
952 width = h->avctx->width;
953 height = h->avctx->height;
956 if (width <= 0 || height <= 0) {
957 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
959 if (h->avctx->err_recognition & AV_EF_EXPLODE)
960 return AVERROR_INVALIDDATA;
962 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
973 h->avctx->coded_width = h->width;
974 h->avctx->coded_height = h->height;
975 h->avctx->width = width;
976 h->avctx->height = height;
981 static int h264_slice_header_init(H264Context *h)
983 int nb_slices = (HAVE_THREADS &&
984 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
985 h->avctx->thread_count : 1;
988 ff_set_sar(h->avctx, h->sps.sar);
989 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
990 &h->chroma_x_shift, &h->chroma_y_shift);
992 if (h->sps.timing_info_present_flag) {
993 int64_t den = h->sps.time_scale;
994 if (h->x264_build < 44U)
996 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
997 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1000 ff_h264_free_tables(h);
1003 h->prev_interlaced_frame = 1;
1005 init_scan_tables(h);
1006 ret = ff_h264_alloc_tables(h);
1008 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1012 #if FF_API_CAP_VDPAU
1013 if (h->avctx->codec &&
1014 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
1015 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
1016 av_log(h->avctx, AV_LOG_ERROR,
1017 "VDPAU decoding does not support video colorspace.\n");
1018 ret = AVERROR_INVALIDDATA;
1023 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 14 ||
1024 h->sps.bit_depth_luma == 11 || h->sps.bit_depth_luma == 13
1026 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
1027 h->sps.bit_depth_luma);
1028 ret = AVERROR_INVALIDDATA;
1032 h->cur_bit_depth_luma =
1033 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1034 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
1035 h->pixel_shift = h->sps.bit_depth_luma > 8;
1036 h->chroma_format_idc = h->sps.chroma_format_idc;
1037 h->bit_depth_luma = h->sps.bit_depth_luma;
1039 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
1040 h->sps.chroma_format_idc);
1041 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1042 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
1043 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
1044 h->sps.chroma_format_idc);
1045 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
1047 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1050 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1052 max_slices = H264_MAX_THREADS;
1053 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1054 " reducing to %d\n", nb_slices, max_slices);
1055 nb_slices = max_slices;
1057 h->slice_context_count = nb_slices;
1058 h->max_contexts = FFMIN(h->max_contexts, nb_slices);
1060 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1061 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1063 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1067 for (i = 0; i < h->slice_context_count; i++) {
1068 H264SliceContext *sl = &h->slice_ctx[i];
1071 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1072 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1073 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1075 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1076 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1082 h->context_initialized = 1;
1086 ff_h264_free_tables(h);
1087 h->context_initialized = 0;
1091 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1094 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1095 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1096 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1103 * Decode a slice header.
1104 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1106 * @param h h264context
1108 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1110 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1112 unsigned int first_mb_in_slice;
1113 unsigned int pps_id;
1115 unsigned int slice_type, tmp, i, j;
1116 int last_pic_structure, last_pic_droppable;
1118 int needs_reinit = 0;
1119 int field_pic_flag, bottom_field_flag;
1120 int first_slice = sl == h->slice_ctx && !h->current_slice;
1121 int frame_num, droppable, picture_structure;
1122 int mb_aff_frame, last_mb_aff_frame;
1126 av_assert0(!h->setup_finished);
1128 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1129 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1131 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1133 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1134 if (h->current_slice) {
1135 if (h->setup_finished) {
1136 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
1137 return AVERROR_INVALIDDATA;
1139 if (h->max_contexts > 1) {
1140 if (!h->single_decode_warning) {
1141 av_log(h->avctx, AV_LOG_WARNING, "Cannot decode multiple access units as slice threads\n");
1142 h->single_decode_warning = 1;
1144 h->max_contexts = 1;
1145 return SLICE_SINGLETHREAD;
1148 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1149 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1150 h->current_slice = 0;
1153 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1154 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1155 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1156 h->current_slice = 0;
1157 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1158 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1159 h->cur_pic_ptr = NULL;
1163 return AVERROR_INVALIDDATA;
1166 if (!h->first_field) {
1167 if (h->cur_pic_ptr && !h->droppable) {
1168 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1169 h->picture_structure == PICT_BOTTOM_FIELD);
1171 h->cur_pic_ptr = NULL;
1175 if (!h->current_slice)
1176 av_assert0(sl == h->slice_ctx);
1178 slice_type = get_ue_golomb_31(&sl->gb);
1179 if (slice_type > 9) {
1180 av_log(h->avctx, AV_LOG_ERROR,
1181 "slice type %d too large at %d\n",
1182 slice_type, first_mb_in_slice);
1183 return AVERROR_INVALIDDATA;
1185 if (slice_type > 4) {
1187 sl->slice_type_fixed = 1;
1189 sl->slice_type_fixed = 0;
1191 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1192 sl->slice_type = slice_type;
1193 sl->slice_type_nos = slice_type & 3;
1195 if (h->nal_unit_type == NAL_IDR_SLICE &&
1196 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1197 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1198 return AVERROR_INVALIDDATA;
1201 if (h->current_slice == 0 && !h->first_field) {
1203 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1204 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1205 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1206 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE && h->sei_recovery_frame_cnt < 0) ||
1207 h->avctx->skip_frame >= AVDISCARD_ALL) {
1208 return SLICE_SKIPED;
1212 // to make a few old functions happy, it's wrong though
1213 if (!h->setup_finished)
1214 h->pict_type = sl->slice_type;
1216 pps_id = get_ue_golomb(&sl->gb);
1217 if (pps_id >= MAX_PPS_COUNT) {
1218 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1219 return AVERROR_INVALIDDATA;
1221 if (!h->pps_buffers[pps_id]) {
1222 av_log(h->avctx, AV_LOG_ERROR,
1223 "non-existing PPS %u referenced\n",
1225 return AVERROR_INVALIDDATA;
1227 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1228 av_log(h->avctx, AV_LOG_ERROR,
1229 "PPS change from %d to %d forbidden\n",
1230 h->au_pps_id, pps_id);
1231 return AVERROR_INVALIDDATA;
1234 pps = h->pps_buffers[pps_id];
1236 if (!h->sps_buffers[pps->sps_id]) {
1237 av_log(h->avctx, AV_LOG_ERROR,
1238 "non-existing SPS %u referenced\n",
1240 return AVERROR_INVALIDDATA;
1244 h->pps = *h->pps_buffers[pps_id];
1246 if (h->pps.sps_id != pps->sps_id ||
1247 h->pps.transform_8x8_mode != pps->transform_8x8_mode ||
1248 (h->setup_finished && h->dequant_coeff_pps != pps_id)) {
1249 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1250 return AVERROR_INVALIDDATA;
1254 if (pps->sps_id != h->sps.sps_id ||
1255 pps->sps_id != h->current_sps_id ||
1256 h->sps_buffers[pps->sps_id]->new) {
1259 av_log(h->avctx, AV_LOG_ERROR,
1260 "SPS changed in the middle of the frame\n");
1261 return AVERROR_INVALIDDATA;
1264 h->sps = *h->sps_buffers[h->pps.sps_id];
1266 if (h->mb_width != h->sps.mb_width ||
1267 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1268 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1269 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1273 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1274 h->chroma_format_idc != h->sps.chroma_format_idc)
1277 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1278 (h->sps.bitstream_restriction_flag &&
1279 !h->sps.num_reorder_frames)) {
1280 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1281 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1282 "Reenabling low delay requires a codec flush.\n");
1287 if (h->avctx->has_b_frames < 2)
1288 h->avctx->has_b_frames = !h->low_delay;
1292 must_reinit = (h->context_initialized &&
1293 ( 16*h->sps.mb_width != h->avctx->coded_width
1294 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1295 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1296 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1297 || h->mb_width != h->sps.mb_width
1298 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1300 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1301 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1304 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1307 if (!h->setup_finished) {
1308 h->avctx->profile = ff_h264_get_profile(&h->sps);
1309 h->avctx->level = h->sps.level_idc;
1310 h->avctx->refs = h->sps.ref_frame_count;
1312 h->mb_width = h->sps.mb_width;
1313 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1314 h->mb_num = h->mb_width * h->mb_height;
1315 h->mb_stride = h->mb_width + 1;
1317 h->b_stride = h->mb_width * 4;
1319 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1321 h->width = 16 * h->mb_width;
1322 h->height = 16 * h->mb_height;
1324 ret = init_dimensions(h);
1328 if (h->sps.video_signal_type_present_flag) {
1329 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1331 if (h->sps.colour_description_present_flag) {
1332 if (h->avctx->colorspace != h->sps.colorspace)
1334 h->avctx->color_primaries = h->sps.color_primaries;
1335 h->avctx->color_trc = h->sps.color_trc;
1336 h->avctx->colorspace = h->sps.colorspace;
1341 if (h->context_initialized &&
1342 (must_reinit || needs_reinit)) {
1343 h->context_initialized = 0;
1344 if (sl != h->slice_ctx) {
1345 av_log(h->avctx, AV_LOG_ERROR,
1346 "changing width %d -> %d / height %d -> %d on "
1348 h->width, h->avctx->coded_width,
1349 h->height, h->avctx->coded_height,
1350 h->current_slice + 1);
1351 return AVERROR_INVALIDDATA;
1354 av_assert1(first_slice);
1356 ff_h264_flush_change(h);
1358 if ((ret = get_pixel_format(h, 1)) < 0)
1360 h->avctx->pix_fmt = ret;
1362 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1363 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1365 if ((ret = h264_slice_header_init(h)) < 0) {
1366 av_log(h->avctx, AV_LOG_ERROR,
1367 "h264_slice_header_init() failed\n");
1371 if (!h->context_initialized) {
1372 if (sl != h->slice_ctx) {
1373 av_log(h->avctx, AV_LOG_ERROR,
1374 "Cannot (re-)initialize context during parallel decoding.\n");
1375 return AVERROR_PATCHWELCOME;
1378 if ((ret = get_pixel_format(h, 1)) < 0)
1380 h->avctx->pix_fmt = ret;
1382 if ((ret = h264_slice_header_init(h)) < 0) {
1383 av_log(h->avctx, AV_LOG_ERROR,
1384 "h264_slice_header_init() failed\n");
1389 if (!h->current_slice && h->dequant_coeff_pps != pps_id) {
1390 h->dequant_coeff_pps = pps_id;
1391 ff_h264_init_dequant_tables(h);
1394 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1396 if (h->frame_num != frame_num) {
1397 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1398 h->frame_num, frame_num);
1399 return AVERROR_INVALIDDATA;
1403 if (!h->setup_finished)
1404 h->frame_num = frame_num;
1408 last_mb_aff_frame = h->mb_aff_frame;
1409 last_pic_structure = h->picture_structure;
1410 last_pic_droppable = h->droppable;
1412 droppable = h->nal_ref_idc == 0;
1413 if (h->sps.frame_mbs_only_flag) {
1414 picture_structure = PICT_FRAME;
1416 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1417 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1420 field_pic_flag = get_bits1(&sl->gb);
1422 if (field_pic_flag) {
1423 bottom_field_flag = get_bits1(&sl->gb);
1424 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1426 picture_structure = PICT_FRAME;
1427 mb_aff_frame = h->sps.mb_aff;
1431 if (h->current_slice) {
1432 if (last_pic_structure != picture_structure ||
1433 last_pic_droppable != droppable ||
1434 last_mb_aff_frame != mb_aff_frame) {
1435 av_log(h->avctx, AV_LOG_ERROR,
1436 "Changing field mode (%d -> %d) between slices is not allowed\n",
1437 last_pic_structure, h->picture_structure);
1438 return AVERROR_INVALIDDATA;
1439 } else if (!h->cur_pic_ptr) {
1440 av_log(h->avctx, AV_LOG_ERROR,
1441 "unset cur_pic_ptr on slice %d\n",
1442 h->current_slice + 1);
1443 return AVERROR_INVALIDDATA;
1447 if (!h->setup_finished) {
1448 h->droppable = droppable;
1449 h->picture_structure = picture_structure;
1450 h->mb_aff_frame = mb_aff_frame;
1452 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1454 if (h->current_slice == 0) {
1455 /* Shorten frame num gaps so we don't have to allocate reference
1456 * frames just to throw them away */
1457 if (h->frame_num != h->prev_frame_num) {
1458 int unwrap_prev_frame_num = h->prev_frame_num;
1459 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1461 if (unwrap_prev_frame_num > h->frame_num)
1462 unwrap_prev_frame_num -= max_frame_num;
1464 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1465 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1466 if (unwrap_prev_frame_num < 0)
1467 unwrap_prev_frame_num += max_frame_num;
1469 h->prev_frame_num = unwrap_prev_frame_num;
1473 /* See if we have a decoded first field looking for a pair...
1474 * Here, we're using that to see if we should mark previously
1475 * decode frames as "finished".
1476 * We have to do that before the "dummy" in-between frame allocation,
1477 * since that can modify h->cur_pic_ptr. */
1478 if (h->first_field) {
1479 av_assert0(h->cur_pic_ptr);
1480 av_assert0(h->cur_pic_ptr->f->buf[0]);
1481 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1483 /* Mark old field/frame as completed */
1484 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1485 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1486 last_pic_structure == PICT_BOTTOM_FIELD);
1489 /* figure out if we have a complementary field pair */
1490 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1491 /* Previous field is unmatched. Don't display it, but let it
1492 * remain for reference if marked as such. */
1493 if (last_pic_structure != PICT_FRAME) {
1494 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1495 last_pic_structure == PICT_TOP_FIELD);
1498 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1499 /* This and previous field were reference, but had
1500 * different frame_nums. Consider this field first in
1501 * pair. Throw away previous field except for reference
1503 if (last_pic_structure != PICT_FRAME) {
1504 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1505 last_pic_structure == PICT_TOP_FIELD);
1508 /* Second field in complementary pair */
1509 if (!((last_pic_structure == PICT_TOP_FIELD &&
1510 h->picture_structure == PICT_BOTTOM_FIELD) ||
1511 (last_pic_structure == PICT_BOTTOM_FIELD &&
1512 h->picture_structure == PICT_TOP_FIELD))) {
1513 av_log(h->avctx, AV_LOG_ERROR,
1514 "Invalid field mode combination %d/%d\n",
1515 last_pic_structure, h->picture_structure);
1516 h->picture_structure = last_pic_structure;
1517 h->droppable = last_pic_droppable;
1518 return AVERROR_INVALIDDATA;
1519 } else if (last_pic_droppable != h->droppable) {
1520 avpriv_request_sample(h->avctx,
1521 "Found reference and non-reference fields in the same frame, which");
1522 h->picture_structure = last_pic_structure;
1523 h->droppable = last_pic_droppable;
1524 return AVERROR_PATCHWELCOME;
1530 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1531 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1532 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1533 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1534 h->frame_num, h->prev_frame_num);
1535 if (!h->sps.gaps_in_frame_num_allowed_flag)
1536 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1537 h->last_pocs[i] = INT_MIN;
1538 ret = h264_frame_start(h);
1544 h->prev_frame_num++;
1545 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1546 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1547 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1548 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1549 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1550 ret = ff_generate_sliding_window_mmcos(h, 1);
1551 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1553 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1554 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1556 /* Error concealment: If a ref is missing, copy the previous ref
1558 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1559 * many assumptions about there being no actual duplicates.
1560 * FIXME: This does not copy padding for out-of-frame motion
1561 * vectors. Given we are concealing a lost frame, this probably
1562 * is not noticeable by comparison, but it should be fixed. */
1563 if (h->short_ref_count) {
1565 h->short_ref[0]->f->width == prev->f->width &&
1566 h->short_ref[0]->f->height == prev->f->height &&
1567 h->short_ref[0]->f->format == prev->f->format) {
1568 av_image_copy(h->short_ref[0]->f->data,
1569 h->short_ref[0]->f->linesize,
1570 (const uint8_t **)prev->f->data,
1575 h->short_ref[0]->poc = prev->poc + 2;
1577 h->short_ref[0]->frame_num = h->prev_frame_num;
1581 /* See if we have a decoded first field looking for a pair...
1582 * We're using that to see whether to continue decoding in that
1583 * frame, or to allocate a new one. */
1584 if (h->first_field) {
1585 av_assert0(h->cur_pic_ptr);
1586 av_assert0(h->cur_pic_ptr->f->buf[0]);
1587 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1589 /* figure out if we have a complementary field pair */
1590 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1591 /* Previous field is unmatched. Don't display it, but let it
1592 * remain for reference if marked as such. */
1593 h->missing_fields ++;
1594 h->cur_pic_ptr = NULL;
1595 h->first_field = FIELD_PICTURE(h);
1597 h->missing_fields = 0;
1598 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1599 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1600 h->picture_structure==PICT_BOTTOM_FIELD);
1601 /* This and the previous field had different frame_nums.
1602 * Consider this field first in pair. Throw away previous
1603 * one except for reference purposes. */
1605 h->cur_pic_ptr = NULL;
1607 /* Second field in complementary pair */
1612 /* Frame or first field in a potentially complementary pair */
1613 h->first_field = FIELD_PICTURE(h);
1616 if (!FIELD_PICTURE(h) || h->first_field) {
1617 if (h264_frame_start(h) < 0) {
1619 return AVERROR_INVALIDDATA;
1622 release_unused_pictures(h, 0);
1624 /* Some macroblocks can be accessed before they're available in case
1625 * of lost slices, MBAFF or threading. */
1626 if (FIELD_PICTURE(h)) {
1627 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1628 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1630 memset(h->slice_table, -1,
1631 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1635 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1636 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1637 first_mb_in_slice >= h->mb_num) {
1638 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1639 return AVERROR_INVALIDDATA;
1641 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1642 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1643 FIELD_OR_MBAFF_PICTURE(h);
1644 if (h->picture_structure == PICT_BOTTOM_FIELD)
1645 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1646 av_assert1(sl->mb_y < h->mb_height);
1648 if (h->picture_structure == PICT_FRAME) {
1649 h->curr_pic_num = h->frame_num;
1650 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1652 h->curr_pic_num = 2 * h->frame_num + 1;
1653 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1656 if (h->nal_unit_type == NAL_IDR_SLICE)
1657 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1659 if (h->sps.poc_type == 0) {
1660 int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1662 if (!h->setup_finished)
1663 h->poc_lsb = poc_lsb;
1665 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1666 int delta_poc_bottom = get_se_golomb(&sl->gb);
1667 if (!h->setup_finished)
1668 h->delta_poc_bottom = delta_poc_bottom;
1672 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1673 int delta_poc = get_se_golomb(&sl->gb);
1675 if (!h->setup_finished)
1676 h->delta_poc[0] = delta_poc;
1678 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1679 delta_poc = get_se_golomb(&sl->gb);
1681 if (!h->setup_finished)
1682 h->delta_poc[1] = delta_poc;
1686 if (!h->setup_finished)
1687 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1689 if (h->pps.redundant_pic_cnt_present)
1690 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1692 ret = ff_set_ref_count(h, sl);
1696 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1697 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1699 sl->ref_count[1] = sl->ref_count[0] = 0;
1704 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1705 (h->pps.weighted_bipred_idc == 1 &&
1706 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1707 ff_h264_pred_weight_table(&sl->gb, &h->sps, sl->ref_count,
1708 sl->slice_type_nos, &sl->pwt);
1709 else if (h->pps.weighted_bipred_idc == 2 &&
1710 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1711 implicit_weight_table(h, sl, -1);
1713 sl->pwt.use_weight = 0;
1714 for (i = 0; i < 2; i++) {
1715 sl->pwt.luma_weight_flag[i] = 0;
1716 sl->pwt.chroma_weight_flag[i] = 0;
1720 // If frame-mt is enabled, only update mmco tables for the first slice
1721 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1722 // or h->mmco, which will cause ref list mix-ups and decoding errors
1723 // further down the line. This may break decoding if the first slice is
1724 // corrupt, thus we only do this if frame-mt is enabled.
1725 if (h->nal_ref_idc) {
1726 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1727 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1728 h->current_slice == 0);
1729 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1730 return AVERROR_INVALIDDATA;
1733 if (FRAME_MBAFF(h)) {
1734 ff_h264_fill_mbaff_ref_list(h, sl);
1736 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1737 implicit_weight_table(h, sl, 0);
1738 implicit_weight_table(h, sl, 1);
1742 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1743 ff_h264_direct_dist_scale_factor(h, sl);
1744 ff_h264_direct_ref_list_init(h, sl);
1746 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1747 tmp = get_ue_golomb_31(&sl->gb);
1749 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1750 return AVERROR_INVALIDDATA;
1752 sl->cabac_init_idc = tmp;
1755 sl->last_qscale_diff = 0;
1756 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1757 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1758 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1759 return AVERROR_INVALIDDATA;
1762 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1763 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1764 // FIXME qscale / qp ... stuff
1765 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1766 get_bits1(&sl->gb); /* sp_for_switch_flag */
1767 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1768 sl->slice_type == AV_PICTURE_TYPE_SI)
1769 get_se_golomb(&sl->gb); /* slice_qs_delta */
1771 sl->deblocking_filter = 1;
1772 sl->slice_alpha_c0_offset = 0;
1773 sl->slice_beta_offset = 0;
1774 if (h->pps.deblocking_filter_parameters_present) {
1775 tmp = get_ue_golomb_31(&sl->gb);
1777 av_log(h->avctx, AV_LOG_ERROR,
1778 "deblocking_filter_idc %u out of range\n", tmp);
1779 return AVERROR_INVALIDDATA;
1781 sl->deblocking_filter = tmp;
1782 if (sl->deblocking_filter < 2)
1783 sl->deblocking_filter ^= 1; // 1<->0
1785 if (sl->deblocking_filter) {
1786 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1787 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1788 if (sl->slice_alpha_c0_offset > 12 ||
1789 sl->slice_alpha_c0_offset < -12 ||
1790 sl->slice_beta_offset > 12 ||
1791 sl->slice_beta_offset < -12) {
1792 av_log(h->avctx, AV_LOG_ERROR,
1793 "deblocking filter parameters %d %d out of range\n",
1794 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1795 return AVERROR_INVALIDDATA;
1800 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1801 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1802 h->nal_unit_type != NAL_IDR_SLICE) ||
1803 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1804 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1805 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1806 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1807 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1808 h->nal_ref_idc == 0))
1809 sl->deblocking_filter = 0;
1811 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1812 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1813 /* Cheat slightly for speed:
1814 * Do not bother to deblock across slices. */
1815 sl->deblocking_filter = 2;
1817 h->max_contexts = 1;
1818 if (!h->single_decode_warning) {
1819 av_log(h->avctx, AV_LOG_INFO,
1820 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1821 "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"
1822 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1823 "but will generate non-standard-compliant output.\n");
1824 h->single_decode_warning = 1;
1826 if (sl != h->slice_ctx) {
1827 av_log(h->avctx, AV_LOG_ERROR,
1828 "Deblocking switched inside frame.\n");
1829 return SLICE_SINGLETHREAD;
1833 sl->qp_thresh = 15 -
1834 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1836 h->pps.chroma_qp_index_offset[0],
1837 h->pps.chroma_qp_index_offset[1]) +
1838 6 * (h->sps.bit_depth_luma - 8);
1840 sl->slice_num = ++h->current_slice;
1843 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1844 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1845 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1846 && sl->slice_num >= MAX_SLICES) {
1847 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1848 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);
1851 for (j = 0; j < 2; j++) {
1853 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1854 for (i = 0; i < 16; i++) {
1856 if (j < sl->list_count && i < sl->ref_count[j] &&
1857 sl->ref_list[j][i].parent->f->buf[0]) {
1859 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1860 for (k = 0; k < h->short_ref_count; k++)
1861 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1865 for (k = 0; k < h->long_ref_count; k++)
1866 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1867 id_list[i] = h->short_ref_count + k;
1875 for (i = 0; i < 16; i++)
1876 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1878 ref2frm[18 + 1] = -1;
1879 for (i = 16; i < 48; i++)
1880 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1881 (sl->ref_list[j][i].reference & 3);
1884 h->au_pps_id = pps_id;
1886 h->sps_buffers[h->pps.sps_id]->new = 0;
1887 h->current_sps_id = h->pps.sps_id;
1889 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1890 av_log(h->avctx, AV_LOG_DEBUG,
1891 "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",
1893 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1895 av_get_picture_type_char(sl->slice_type),
1896 sl->slice_type_fixed ? " fix" : "",
1897 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1898 pps_id, h->frame_num,
1899 h->cur_pic_ptr->field_poc[0],
1900 h->cur_pic_ptr->field_poc[1],
1901 sl->ref_count[0], sl->ref_count[1],
1903 sl->deblocking_filter,
1904 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1906 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1907 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1913 int ff_h264_get_slice_type(const H264SliceContext *sl)
1915 switch (sl->slice_type) {
1916 case AV_PICTURE_TYPE_P:
1918 case AV_PICTURE_TYPE_B:
1920 case AV_PICTURE_TYPE_I:
1922 case AV_PICTURE_TYPE_SP:
1924 case AV_PICTURE_TYPE_SI:
1927 return AVERROR_INVALIDDATA;
1931 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1932 H264SliceContext *sl,
1933 int mb_type, int top_xy,
1934 int left_xy[LEFT_MBS],
1936 int left_type[LEFT_MBS],
1937 int mb_xy, int list)
1939 int b_stride = h->b_stride;
1940 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1941 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1942 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1943 if (USES_LIST(top_type, list)) {
1944 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1945 const int b8_xy = 4 * top_xy + 2;
1946 int *ref2frm = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1947 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1948 ref_cache[0 - 1 * 8] =
1949 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
1950 ref_cache[2 - 1 * 8] =
1951 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
1953 AV_ZERO128(mv_dst - 1 * 8);
1954 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1957 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1958 if (USES_LIST(left_type[LTOP], list)) {
1959 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1960 const int b8_xy = 4 * left_xy[LTOP] + 1;
1961 int *ref2frm = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1962 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1963 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1964 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1965 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1967 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1968 ref_cache[-1 + 16] =
1969 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1971 AV_ZERO32(mv_dst - 1 + 0);
1972 AV_ZERO32(mv_dst - 1 + 8);
1973 AV_ZERO32(mv_dst - 1 + 16);
1974 AV_ZERO32(mv_dst - 1 + 24);
1977 ref_cache[-1 + 16] =
1978 ref_cache[-1 + 24] = LIST_NOT_USED;
1983 if (!USES_LIST(mb_type, list)) {
1984 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1985 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1986 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1987 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1988 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1993 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1994 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1995 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
1996 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
1997 AV_WN32A(&ref_cache[0 * 8], ref01);
1998 AV_WN32A(&ref_cache[1 * 8], ref01);
1999 AV_WN32A(&ref_cache[2 * 8], ref23);
2000 AV_WN32A(&ref_cache[3 * 8], ref23);
2004 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2005 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2006 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2007 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2008 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2013 * @return non zero if the loop filter can be skipped
2015 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2017 const int mb_xy = sl->mb_xy;
2018 int top_xy, left_xy[LEFT_MBS];
2019 int top_type, left_type[LEFT_MBS];
2023 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2025 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2026 * stuff, I can't imagine that these complex rules are worth it. */
2028 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2029 if (FRAME_MBAFF(h)) {
2030 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2031 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2033 if (left_mb_field_flag != curr_mb_field_flag)
2034 left_xy[LTOP] -= h->mb_stride;
2036 if (curr_mb_field_flag)
2037 top_xy += h->mb_stride &
2038 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2039 if (left_mb_field_flag != curr_mb_field_flag)
2040 left_xy[LBOT] += h->mb_stride;
2044 sl->top_mb_xy = top_xy;
2045 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2046 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2048 /* For sufficiently low qp, filtering wouldn't do anything.
2049 * This is a conservative estimate: could also check beta_offset
2050 * and more accurate chroma_qp. */
2051 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2052 int qp = h->cur_pic.qscale_table[mb_xy];
2053 if (qp <= qp_thresh &&
2054 (left_xy[LTOP] < 0 ||
2055 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2057 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2058 if (!FRAME_MBAFF(h))
2060 if ((left_xy[LTOP] < 0 ||
2061 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2062 (top_xy < h->mb_stride ||
2063 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2068 top_type = h->cur_pic.mb_type[top_xy];
2069 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2070 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2071 if (sl->deblocking_filter == 2) {
2072 if (h->slice_table[top_xy] != sl->slice_num)
2074 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2075 left_type[LTOP] = left_type[LBOT] = 0;
2077 if (h->slice_table[top_xy] == 0xFFFF)
2079 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2080 left_type[LTOP] = left_type[LBOT] = 0;
2082 sl->top_type = top_type;
2083 sl->left_type[LTOP] = left_type[LTOP];
2084 sl->left_type[LBOT] = left_type[LBOT];
2086 if (IS_INTRA(mb_type))
2089 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2090 top_type, left_type, mb_xy, 0);
2091 if (sl->list_count == 2)
2092 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2093 top_type, left_type, mb_xy, 1);
2095 nnz = h->non_zero_count[mb_xy];
2096 nnz_cache = sl->non_zero_count_cache;
2097 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2098 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2099 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2100 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2101 sl->cbp = h->cbp_table[mb_xy];
2104 nnz = h->non_zero_count[top_xy];
2105 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2108 if (left_type[LTOP]) {
2109 nnz = h->non_zero_count[left_xy[LTOP]];
2110 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2111 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2112 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2113 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2116 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2117 * from what the loop filter needs */
2118 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2119 if (IS_8x8DCT(top_type)) {
2120 nnz_cache[4 + 8 * 0] =
2121 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2122 nnz_cache[6 + 8 * 0] =
2123 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2125 if (IS_8x8DCT(left_type[LTOP])) {
2126 nnz_cache[3 + 8 * 1] =
2127 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2129 if (IS_8x8DCT(left_type[LBOT])) {
2130 nnz_cache[3 + 8 * 3] =
2131 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2134 if (IS_8x8DCT(mb_type)) {
2135 nnz_cache[scan8[0]] =
2136 nnz_cache[scan8[1]] =
2137 nnz_cache[scan8[2]] =
2138 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2140 nnz_cache[scan8[0 + 4]] =
2141 nnz_cache[scan8[1 + 4]] =
2142 nnz_cache[scan8[2 + 4]] =
2143 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2145 nnz_cache[scan8[0 + 8]] =
2146 nnz_cache[scan8[1 + 8]] =
2147 nnz_cache[scan8[2 + 8]] =
2148 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2150 nnz_cache[scan8[0 + 12]] =
2151 nnz_cache[scan8[1 + 12]] =
2152 nnz_cache[scan8[2 + 12]] =
2153 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2160 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2162 uint8_t *dest_y, *dest_cb, *dest_cr;
2163 int linesize, uvlinesize, mb_x, mb_y;
2164 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2165 const int old_slice_type = sl->slice_type;
2166 const int pixel_shift = h->pixel_shift;
2167 const int block_h = 16 >> h->chroma_y_shift;
2169 if (sl->deblocking_filter) {
2170 for (mb_x = start_x; mb_x < end_x; mb_x++)
2171 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2173 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2174 sl->slice_num = h->slice_table[mb_xy];
2175 mb_type = h->cur_pic.mb_type[mb_xy];
2176 sl->list_count = h->list_counts[mb_xy];
2180 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2184 dest_y = h->cur_pic.f->data[0] +
2185 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2186 dest_cb = h->cur_pic.f->data[1] +
2187 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2188 mb_y * sl->uvlinesize * block_h;
2189 dest_cr = h->cur_pic.f->data[2] +
2190 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2191 mb_y * sl->uvlinesize * block_h;
2192 // FIXME simplify above
2195 linesize = sl->mb_linesize = sl->linesize * 2;
2196 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2197 if (mb_y & 1) { // FIXME move out of this function?
2198 dest_y -= sl->linesize * 15;
2199 dest_cb -= sl->uvlinesize * (block_h - 1);
2200 dest_cr -= sl->uvlinesize * (block_h - 1);
2203 linesize = sl->mb_linesize = sl->linesize;
2204 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2206 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2208 if (fill_filter_caches(h, sl, mb_type))
2210 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2211 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2213 if (FRAME_MBAFF(h)) {
2214 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2215 linesize, uvlinesize);
2217 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2218 dest_cr, linesize, uvlinesize);
2222 sl->slice_type = old_slice_type;
2224 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2225 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2226 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2229 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2231 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2232 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2233 h->cur_pic.mb_type[mb_xy - 1] :
2234 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2235 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2236 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2240 * Draw edges and report progress for the last MB row.
2242 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2244 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2245 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2246 int height = 16 << FRAME_MBAFF(h);
2247 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2249 if (sl->deblocking_filter) {
2250 if ((top + height) >= pic_height)
2251 height += deblock_border;
2252 top -= deblock_border;
2255 if (top >= pic_height || (top + height) < 0)
2258 height = FFMIN(height, pic_height - top);
2260 height = top + height;
2264 ff_h264_draw_horiz_band(h, sl, top, height);
2266 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2269 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2270 h->picture_structure == PICT_BOTTOM_FIELD);
2273 static void er_add_slice(H264SliceContext *sl,
2274 int startx, int starty,
2275 int endx, int endy, int status)
2277 if (!sl->h264->enable_er)
2280 if (CONFIG_ERROR_RESILIENCE) {
2281 ERContext *er = &sl->h264->slice_ctx[0].er;
2283 ff_er_add_slice(er, startx, starty, endx, endy, status);
2287 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2289 H264SliceContext *sl = arg;
2290 const H264Context *h = sl->h264;
2291 int lf_x_start = sl->mb_x;
2294 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2295 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2297 ret = alloc_scratch_buffers(sl, sl->linesize);
2301 sl->mb_skip_run = -1;
2303 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2305 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2306 avctx->codec_id != AV_CODEC_ID_H264 ||
2307 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2309 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2310 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2312 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2313 prev_status &= ~ VP_START;
2314 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2315 h->slice_ctx[0].er.error_occurred = 1;
2321 align_get_bits(&sl->gb);
2324 ret = ff_init_cabac_decoder(&sl->cabac,
2325 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2326 (get_bits_left(&sl->gb) + 7) / 8);
2330 ff_h264_init_cabac_states(h, sl);
2335 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2336 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2337 sl->next_slice_idx);
2338 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2339 sl->mb_y, ER_MB_ERROR);
2340 return AVERROR_INVALIDDATA;
2343 ret = ff_h264_decode_mb_cabac(h, sl);
2344 // STOP_TIMER("decode_mb_cabac")
2347 ff_h264_hl_decode_mb(h, sl);
2349 // FIXME optimal? or let mb_decode decode 16x32 ?
2350 if (ret >= 0 && FRAME_MBAFF(h)) {
2353 ret = ff_h264_decode_mb_cabac(h, sl);
2356 ff_h264_hl_decode_mb(h, sl);
2359 eos = get_cabac_terminate(&sl->cabac);
2361 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2362 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2363 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2364 sl->mb_y, ER_MB_END);
2365 if (sl->mb_x >= lf_x_start)
2366 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2369 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2370 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2371 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2372 av_log(h->avctx, AV_LOG_ERROR,
2373 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2375 sl->cabac.bytestream_end - sl->cabac.bytestream);
2376 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2377 sl->mb_y, ER_MB_ERROR);
2378 return AVERROR_INVALIDDATA;
2381 if (++sl->mb_x >= h->mb_width) {
2382 loop_filter(h, sl, lf_x_start, sl->mb_x);
2383 sl->mb_x = lf_x_start = 0;
2384 decode_finish_row(h, sl);
2386 if (FIELD_OR_MBAFF_PICTURE(h)) {
2388 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2389 predict_field_decoding_flag(h, sl);
2393 if (eos || sl->mb_y >= h->mb_height) {
2394 ff_tlog(h->avctx, "slice end %d %d\n",
2395 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2396 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2397 sl->mb_y, ER_MB_END);
2398 if (sl->mb_x > lf_x_start)
2399 loop_filter(h, sl, lf_x_start, sl->mb_x);
2407 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2408 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2409 sl->next_slice_idx);
2410 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2411 sl->mb_y, ER_MB_ERROR);
2412 return AVERROR_INVALIDDATA;
2415 ret = ff_h264_decode_mb_cavlc(h, sl);
2418 ff_h264_hl_decode_mb(h, sl);
2420 // FIXME optimal? or let mb_decode decode 16x32 ?
2421 if (ret >= 0 && FRAME_MBAFF(h)) {
2423 ret = ff_h264_decode_mb_cavlc(h, sl);
2426 ff_h264_hl_decode_mb(h, sl);
2431 av_log(h->avctx, AV_LOG_ERROR,
2432 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2433 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2434 sl->mb_y, ER_MB_ERROR);
2438 if (++sl->mb_x >= h->mb_width) {
2439 loop_filter(h, sl, lf_x_start, sl->mb_x);
2440 sl->mb_x = lf_x_start = 0;
2441 decode_finish_row(h, sl);
2443 if (FIELD_OR_MBAFF_PICTURE(h)) {
2445 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2446 predict_field_decoding_flag(h, sl);
2448 if (sl->mb_y >= h->mb_height) {
2449 ff_tlog(h->avctx, "slice end %d %d\n",
2450 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2452 if ( get_bits_left(&sl->gb) == 0
2453 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2454 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2455 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2459 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2460 sl->mb_x, sl->mb_y, ER_MB_END);
2462 return AVERROR_INVALIDDATA;
2467 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2468 ff_tlog(h->avctx, "slice end %d %d\n",
2469 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2471 if (get_bits_left(&sl->gb) == 0) {
2472 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2473 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2474 if (sl->mb_x > lf_x_start)
2475 loop_filter(h, sl, lf_x_start, sl->mb_x);
2479 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2480 sl->mb_y, ER_MB_ERROR);
2482 return AVERROR_INVALIDDATA;
2490 * Call decode_slice() for each context.
2492 * @param h h264 master context
2493 * @param context_count number of contexts to execute
2495 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2497 AVCodecContext *const avctx = h->avctx;
2498 H264SliceContext *sl;
2501 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2503 h->slice_ctx[0].next_slice_idx = INT_MAX;
2505 if (h->avctx->hwaccel
2506 #if FF_API_CAP_VDPAU
2507 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2511 if (context_count == 1) {
2514 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2516 ret = decode_slice(avctx, &h->slice_ctx[0]);
2517 h->mb_y = h->slice_ctx[0].mb_y;
2520 av_assert0(context_count > 0);
2521 for (i = 0; i < context_count; i++) {
2522 int next_slice_idx = h->mb_width * h->mb_height;
2525 sl = &h->slice_ctx[i];
2526 if (CONFIG_ERROR_RESILIENCE) {
2527 sl->er.error_count = 0;
2530 /* make sure none of those slices overlap */
2531 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2532 for (j = 0; j < context_count; j++) {
2533 H264SliceContext *sl2 = &h->slice_ctx[j];
2534 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2536 if (i == j || slice_idx2 < slice_idx)
2538 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2540 sl->next_slice_idx = next_slice_idx;
2543 avctx->execute(avctx, decode_slice, h->slice_ctx,
2544 NULL, context_count, sizeof(h->slice_ctx[0]));
2546 /* pull back stuff from slices to master context */
2547 sl = &h->slice_ctx[context_count - 1];
2549 if (CONFIG_ERROR_RESILIENCE) {
2550 for (i = 1; i < context_count; i++)
2551 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;