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 Libav.
7 * Libav 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 * Libav 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 Libav; 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/display.h"
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/stereo3d.h"
33 #include "libavutil/timer.h"
36 #include "cabac_functions.h"
37 #include "error_resilience.h"
41 #include "h264chroma.h"
42 #include "h264_mvpred.h"
46 #include "mpegutils.h"
47 #include "rectangle.h"
53 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
55 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
57 int mb_x, int mb_y, int mb_intra, int mb_skipped)
59 H264Context *h = opaque;
60 H264SliceContext *sl = &h->slice_ctx[0];
64 sl->mb_xy = mb_x + mb_y * h->mb_stride;
65 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
67 /* FIXME: It is possible albeit uncommon that slice references
68 * differ between slices. We take the easy approach and ignore
69 * it for now. If this turns out to have any relevance in
70 * practice then correct remapping should be added. */
71 if (ref >= sl->ref_count[0])
73 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
75 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
76 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
77 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
78 assert(!FRAME_MBAFF(h));
79 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
82 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
85 AVCodecContext *avctx = h->avctx;
86 const AVFrame *src = h->cur_pic.f;
87 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
88 int vshift = desc->log2_chroma_h;
89 const int field_pic = h->picture_structure != PICT_FRAME;
95 height = FFMIN(height, avctx->height - y);
97 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
100 if (avctx->draw_horiz_band) {
101 int offset[AV_NUM_DATA_POINTERS];
104 offset[0] = y * src->linesize[0];
106 offset[2] = (y >> vshift) * src->linesize[1];
107 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
112 avctx->draw_horiz_band(avctx, src, offset,
113 y, h->picture_structure, height);
118 * Check if the top & left blocks are available if needed and
119 * change the dc mode so it only uses the available blocks.
121 int ff_h264_check_intra4x4_pred_mode(const H264Context *h, H264SliceContext *sl)
123 static const int8_t top[12] = {
124 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
126 static const int8_t left[12] = {
127 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
131 if (!(sl->top_samples_available & 0x8000)) {
132 for (i = 0; i < 4; i++) {
133 int status = top[sl->intra4x4_pred_mode_cache[scan8[0] + i]];
135 av_log(h->avctx, AV_LOG_ERROR,
136 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
137 status, sl->mb_x, sl->mb_y);
138 return AVERROR_INVALIDDATA;
140 sl->intra4x4_pred_mode_cache[scan8[0] + i] = status;
145 if ((sl->left_samples_available & 0x8888) != 0x8888) {
146 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
147 for (i = 0; i < 4; i++)
148 if (!(sl->left_samples_available & mask[i])) {
149 int status = left[sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
151 av_log(h->avctx, AV_LOG_ERROR,
152 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
153 status, sl->mb_x, sl->mb_y);
154 return AVERROR_INVALIDDATA;
156 sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
162 } // FIXME cleanup like ff_h264_check_intra_pred_mode
165 * Check if the top & left blocks are available if needed and
166 * change the dc mode so it only uses the available blocks.
168 int ff_h264_check_intra_pred_mode(const H264Context *h, H264SliceContext *sl,
169 int mode, int is_chroma)
171 static const int8_t top[4] = { LEFT_DC_PRED8x8, 1, -1, -1 };
172 static const int8_t left[5] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
175 av_log(h->avctx, AV_LOG_ERROR,
176 "out of range intra chroma pred mode at %d %d\n",
178 return AVERROR_INVALIDDATA;
181 if (!(sl->top_samples_available & 0x8000)) {
184 av_log(h->avctx, AV_LOG_ERROR,
185 "top block unavailable for requested intra mode at %d %d\n",
187 return AVERROR_INVALIDDATA;
191 if ((sl->left_samples_available & 0x8080) != 0x8080) {
193 if (is_chroma && (sl->left_samples_available & 0x8080)) {
194 // mad cow disease mode, aka MBAFF + constrained_intra_pred
195 mode = ALZHEIMER_DC_L0T_PRED8x8 +
196 (!(sl->left_samples_available & 0x8000)) +
197 2 * (mode == DC_128_PRED8x8);
200 av_log(h->avctx, AV_LOG_ERROR,
201 "left block unavailable for requested intra mode at %d %d\n",
203 return AVERROR_INVALIDDATA;
210 const uint8_t *ff_h264_decode_nal(H264Context *h, H264SliceContext *sl,
212 int *dst_length, int *consumed, int length)
217 // src[0]&0x80; // forbidden bit
218 h->nal_ref_idc = src[0] >> 5;
219 h->nal_unit_type = src[0] & 0x1F;
224 #define STARTCODE_TEST \
225 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
226 if (src[i + 2] != 3) { \
227 /* startcode, so we must be past the end */ \
233 #if HAVE_FAST_UNALIGNED
234 #define FIND_FIRST_ZERO \
235 if (i > 0 && !src[i]) \
241 for (i = 0; i + 1 < length; i += 9) {
242 if (!((~AV_RN64A(src + i) &
243 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
244 0x8000800080008080ULL))
251 for (i = 0; i + 1 < length; i += 5) {
252 if (!((~AV_RN32A(src + i) &
253 (AV_RN32A(src + i) - 0x01000101U)) &
262 for (i = 0; i + 1 < length; i += 2) {
265 if (i > 0 && src[i - 1] == 0)
271 if (i >= length - 1) { // no escaped 0
272 *dst_length = length;
273 *consumed = length + 1; // +1 for the header
277 av_fast_malloc(&sl->rbsp_buffer, &sl->rbsp_buffer_size,
278 length + AV_INPUT_BUFFER_PADDING_SIZE);
279 dst = sl->rbsp_buffer;
286 while (si + 2 < length) {
287 // remove escapes (very rare 1:2^22)
288 if (src[si + 2] > 3) {
289 dst[di++] = src[si++];
290 dst[di++] = src[si++];
291 } else if (src[si] == 0 && src[si + 1] == 0) {
292 if (src[si + 2] == 3) { // escape
297 } else // next start code
301 dst[di++] = src[si++];
304 dst[di++] = src[si++];
307 memset(dst + di, 0, AV_INPUT_BUFFER_PADDING_SIZE);
310 *consumed = si + 1; // +1 for the header
311 /* FIXME store exact number of bits in the getbitcontext
312 * (it is needed for decoding) */
317 * Identify the exact end of the bitstream
318 * @return the length of the trailing, or 0 if damaged
320 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
325 ff_tlog(h->avctx, "rbsp trailing %X\n", v);
327 for (r = 1; r < 9; r++) {
335 void ff_h264_free_tables(H264Context *h)
339 av_freep(&h->intra4x4_pred_mode);
340 av_freep(&h->chroma_pred_mode_table);
341 av_freep(&h->cbp_table);
342 av_freep(&h->mvd_table[0]);
343 av_freep(&h->mvd_table[1]);
344 av_freep(&h->direct_table);
345 av_freep(&h->non_zero_count);
346 av_freep(&h->slice_table_base);
347 h->slice_table = NULL;
348 av_freep(&h->list_counts);
350 av_freep(&h->mb2b_xy);
351 av_freep(&h->mb2br_xy);
353 av_buffer_pool_uninit(&h->qscale_table_pool);
354 av_buffer_pool_uninit(&h->mb_type_pool);
355 av_buffer_pool_uninit(&h->motion_val_pool);
356 av_buffer_pool_uninit(&h->ref_index_pool);
358 for (i = 0; i < h->nb_slice_ctx; i++) {
359 H264SliceContext *sl = &h->slice_ctx[i];
361 av_freep(&sl->dc_val_base);
362 av_freep(&sl->er.mb_index2xy);
363 av_freep(&sl->er.error_status_table);
364 av_freep(&sl->er.er_temp_buffer);
366 av_freep(&sl->bipred_scratchpad);
367 av_freep(&sl->edge_emu_buffer);
368 av_freep(&sl->top_borders[0]);
369 av_freep(&sl->top_borders[1]);
371 sl->bipred_scratchpad_allocated = 0;
372 sl->edge_emu_buffer_allocated = 0;
373 sl->top_borders_allocated[0] = 0;
374 sl->top_borders_allocated[1] = 0;
378 int ff_h264_alloc_tables(H264Context *h)
380 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
381 const int row_mb_num = h->mb_stride * 2 * h->avctx->thread_count;
384 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
385 row_mb_num * 8 * sizeof(uint8_t), fail)
386 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
388 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
389 big_mb_num * 48 * sizeof(uint8_t), fail)
390 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
391 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
392 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
393 big_mb_num * sizeof(uint16_t), fail)
394 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
395 big_mb_num * sizeof(uint8_t), fail)
396 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
397 16 * row_mb_num * sizeof(uint8_t), fail);
398 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
399 16 * row_mb_num * sizeof(uint8_t), fail);
400 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
401 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
403 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
404 4 * big_mb_num * sizeof(uint8_t), fail);
405 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
406 big_mb_num * sizeof(uint8_t), fail)
408 memset(h->slice_table_base, -1,
409 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
410 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
412 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
413 big_mb_num * sizeof(uint32_t), fail);
414 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
415 big_mb_num * sizeof(uint32_t), fail);
416 for (y = 0; y < h->mb_height; y++)
417 for (x = 0; x < h->mb_width; x++) {
418 const int mb_xy = x + y * h->mb_stride;
419 const int b_xy = 4 * x + 4 * y * h->b_stride;
421 h->mb2b_xy[mb_xy] = b_xy;
422 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
425 if (!h->dequant4_coeff[0])
426 h264_init_dequant_tables(h);
431 ff_h264_free_tables(h);
432 return AVERROR(ENOMEM);
437 * Allocate buffers which are not shared amongst multiple threads.
439 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
441 ERContext *er = &sl->er;
442 int mb_array_size = h->mb_height * h->mb_stride;
443 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
444 int c_size = h->mb_stride * (h->mb_height + 1);
445 int yc_size = y_size + 2 * c_size;
448 sl->ref_cache[0][scan8[5] + 1] =
449 sl->ref_cache[0][scan8[7] + 1] =
450 sl->ref_cache[0][scan8[13] + 1] =
451 sl->ref_cache[1][scan8[5] + 1] =
452 sl->ref_cache[1][scan8[7] + 1] =
453 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
455 if (CONFIG_ERROR_RESILIENCE) {
457 er->avctx = h->avctx;
458 er->decode_mb = h264_er_decode_mb;
460 er->quarter_sample = 1;
462 er->mb_num = h->mb_num;
463 er->mb_width = h->mb_width;
464 er->mb_height = h->mb_height;
465 er->mb_stride = h->mb_stride;
466 er->b8_stride = h->mb_width * 2 + 1;
468 // error resilience code looks cleaner with this
469 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
470 (h->mb_num + 1) * sizeof(int), fail);
472 for (y = 0; y < h->mb_height; y++)
473 for (x = 0; x < h->mb_width; x++)
474 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
476 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
477 h->mb_stride + h->mb_width;
479 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
480 mb_array_size * sizeof(uint8_t), fail);
482 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
483 h->mb_height * h->mb_stride, fail);
485 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
486 yc_size * sizeof(int16_t), fail);
487 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
488 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
489 er->dc_val[2] = er->dc_val[1] + c_size;
490 for (i = 0; i < yc_size; i++)
491 sl->dc_val_base[i] = 1024;
497 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
500 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
501 int parse_extradata);
503 int ff_h264_decode_extradata(H264Context *h)
505 AVCodecContext *avctx = h->avctx;
508 if (avctx->extradata[0] == 1) {
510 unsigned char *p = avctx->extradata;
514 if (avctx->extradata_size < 7) {
515 av_log(avctx, AV_LOG_ERROR,
516 "avcC %d too short\n", avctx->extradata_size);
517 return AVERROR_INVALIDDATA;
519 /* sps and pps in the avcC always have length coded with 2 bytes,
520 * so put a fake nal_length_size = 2 while parsing them */
521 h->nal_length_size = 2;
522 // Decode sps from avcC
523 cnt = *(p + 5) & 0x1f; // Number of sps
525 for (i = 0; i < cnt; i++) {
526 nalsize = AV_RB16(p) + 2;
527 if (p - avctx->extradata + nalsize > avctx->extradata_size)
528 return AVERROR_INVALIDDATA;
529 ret = decode_nal_units(h, p, nalsize, 1);
531 av_log(avctx, AV_LOG_ERROR,
532 "Decoding sps %d from avcC failed\n", i);
537 // Decode pps from avcC
538 cnt = *(p++); // Number of pps
539 for (i = 0; i < cnt; i++) {
540 nalsize = AV_RB16(p) + 2;
541 if (p - avctx->extradata + nalsize > avctx->extradata_size)
542 return AVERROR_INVALIDDATA;
543 ret = decode_nal_units(h, p, nalsize, 1);
545 av_log(avctx, AV_LOG_ERROR,
546 "Decoding pps %d from avcC failed\n", i);
551 // Store right nal length size that will be used to parse all other nals
552 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
555 ret = decode_nal_units(h, avctx->extradata, avctx->extradata_size, 1);
562 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
567 h->dequant_coeff_pps = -1;
569 h->picture_structure = PICT_FRAME;
570 h->slice_context_count = 1;
571 h->workaround_bugs = avctx->workaround_bugs;
572 h->flags = avctx->flags;
573 h->prev_poc_msb = 1 << 16;
575 h->recovery_frame = -1;
576 h->frame_recovered = 0;
578 h->next_outputed_poc = INT_MIN;
579 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
580 h->last_pocs[i] = INT_MIN;
582 ff_h264_reset_sei(h);
584 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
586 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? H264_MAX_THREADS : 1;
587 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
590 return AVERROR(ENOMEM);
593 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
594 h->DPB[i].f = av_frame_alloc();
596 return AVERROR(ENOMEM);
599 h->cur_pic.f = av_frame_alloc();
601 return AVERROR(ENOMEM);
603 for (i = 0; i < h->nb_slice_ctx; i++)
604 h->slice_ctx[i].h264 = h;
609 static AVOnce h264_vlc_init = AV_ONCE_INIT;
611 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
613 H264Context *h = avctx->priv_data;
616 ret = h264_init_context(avctx, h);
621 if (!avctx->has_b_frames)
624 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
626 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
627 return AVERROR_UNKNOWN;
630 if (avctx->codec_id == AV_CODEC_ID_H264) {
631 if (avctx->ticks_per_frame == 1)
632 h->avctx->framerate.num *= 2;
633 avctx->ticks_per_frame = 2;
636 if (avctx->extradata_size > 0 && avctx->extradata) {
637 ret = ff_h264_decode_extradata(h);
639 ff_h264_free_context(h);
644 if (h->sps.bitstream_restriction_flag &&
645 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
646 h->avctx->has_b_frames = h->sps.num_reorder_frames;
650 avctx->internal->allocate_progress = 1;
653 av_log(avctx, AV_LOG_WARNING,
654 "Error resilience is enabled. It is unsafe and unsupported and may crash. "
655 "Use it at your own risk\n");
661 static int decode_init_thread_copy(AVCodecContext *avctx)
663 H264Context *h = avctx->priv_data;
666 if (!avctx->internal->is_copy)
669 memset(h, 0, sizeof(*h));
671 ret = h264_init_context(avctx, h);
675 h->context_initialized = 0;
681 * Run setup operations that must be run after slice header decoding.
682 * This includes finding the next displayed frame.
684 * @param h h264 master context
685 * @param setup_finished enough NALs have been read that we can call
686 * ff_thread_finish_setup()
688 static void decode_postinit(H264Context *h, int setup_finished)
690 H264Picture *out = h->cur_pic_ptr;
691 H264Picture *cur = h->cur_pic_ptr;
692 int i, pics, out_of_order, out_idx;
693 int invalid = 0, cnt = 0;
695 h->cur_pic_ptr->f->pict_type = h->pict_type;
697 if (h->next_output_pic)
700 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
701 /* FIXME: if we have two PAFF fields in one packet, we can't start
702 * the next thread here. If we have one field per packet, we can.
703 * The check in decode_nal_units() is not good enough to find this
704 * yet, so we assume the worst for now. */
705 // if (setup_finished)
706 // ff_thread_finish_setup(h->avctx);
710 cur->f->interlaced_frame = 0;
711 cur->f->repeat_pict = 0;
713 /* Signal interlacing information externally. */
714 /* Prioritize picture timing SEI information over used
715 * decoding process if it exists. */
717 if (h->sps.pic_struct_present_flag) {
718 switch (h->sei_pic_struct) {
719 case SEI_PIC_STRUCT_FRAME:
721 case SEI_PIC_STRUCT_TOP_FIELD:
722 case SEI_PIC_STRUCT_BOTTOM_FIELD:
723 cur->f->interlaced_frame = 1;
725 case SEI_PIC_STRUCT_TOP_BOTTOM:
726 case SEI_PIC_STRUCT_BOTTOM_TOP:
727 if (FIELD_OR_MBAFF_PICTURE(h))
728 cur->f->interlaced_frame = 1;
730 // try to flag soft telecine progressive
731 cur->f->interlaced_frame = h->prev_interlaced_frame;
733 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
734 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
735 /* Signal the possibility of telecined film externally
736 * (pic_struct 5,6). From these hints, let the applications
737 * decide if they apply deinterlacing. */
738 cur->f->repeat_pict = 1;
740 case SEI_PIC_STRUCT_FRAME_DOUBLING:
741 cur->f->repeat_pict = 2;
743 case SEI_PIC_STRUCT_FRAME_TRIPLING:
744 cur->f->repeat_pict = 4;
748 if ((h->sei_ct_type & 3) &&
749 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
750 cur->f->interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
752 /* Derive interlacing flag from used decoding process. */
753 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
755 h->prev_interlaced_frame = cur->f->interlaced_frame;
757 if (cur->field_poc[0] != cur->field_poc[1]) {
758 /* Derive top_field_first from field pocs. */
759 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
761 if (cur->f->interlaced_frame || h->sps.pic_struct_present_flag) {
762 /* Use picture timing SEI information. Even if it is a
763 * information of a past frame, better than nothing. */
764 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
765 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
766 cur->f->top_field_first = 1;
768 cur->f->top_field_first = 0;
770 /* Most likely progressive */
771 cur->f->top_field_first = 0;
775 if (h->sei_frame_packing_present &&
776 h->frame_packing_arrangement_type >= 0 &&
777 h->frame_packing_arrangement_type <= 6 &&
778 h->content_interpretation_type > 0 &&
779 h->content_interpretation_type < 3) {
780 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
784 switch (h->frame_packing_arrangement_type) {
786 stereo->type = AV_STEREO3D_CHECKERBOARD;
789 stereo->type = AV_STEREO3D_COLUMNS;
792 stereo->type = AV_STEREO3D_LINES;
795 if (h->quincunx_subsampling)
796 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
798 stereo->type = AV_STEREO3D_SIDEBYSIDE;
801 stereo->type = AV_STEREO3D_TOPBOTTOM;
804 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
807 stereo->type = AV_STEREO3D_2D;
811 if (h->content_interpretation_type == 2)
812 stereo->flags = AV_STEREO3D_FLAG_INVERT;
815 if (h->sei_display_orientation_present &&
816 (h->sei_anticlockwise_rotation || h->sei_hflip || h->sei_vflip)) {
817 double angle = h->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
818 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
819 AV_FRAME_DATA_DISPLAYMATRIX,
820 sizeof(int32_t) * 9);
824 av_display_rotation_set((int32_t *)rotation->data, angle);
825 av_display_matrix_flip((int32_t *)rotation->data,
826 h->sei_hflip, h->sei_vflip);
829 if (h->sei_reguserdata_afd_present) {
830 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
835 *sd->data = h->active_format_description;
836 h->sei_reguserdata_afd_present = 0;
839 if (h->a53_caption) {
840 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
841 AV_FRAME_DATA_A53_CC,
842 h->a53_caption_size);
846 memcpy(sd->data, h->a53_caption, h->a53_caption_size);
847 av_freep(&h->a53_caption);
848 h->a53_caption_size = 0;
851 // FIXME do something with unavailable reference frames
853 /* Sort B-frames into display order */
854 if (h->sps.bitstream_restriction_flag ||
855 h->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
856 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, h->sps.num_reorder_frames);
858 h->low_delay = !h->avctx->has_b_frames;
861 while (h->delayed_pic[pics])
864 assert(pics <= MAX_DELAYED_PIC_COUNT);
866 h->delayed_pic[pics++] = cur;
867 if (cur->reference == 0)
868 cur->reference = DELAYED_PIC_REF;
870 /* Frame reordering. This code takes pictures from coding order and sorts
871 * them by their incremental POC value into display order. It supports POC
872 * gaps, MMCO reset codes and random resets.
873 * A "display group" can start either with a IDR frame (f.key_frame = 1),
874 * and/or can be closed down with a MMCO reset code. In sequences where
875 * there is no delay, we can't detect that (since the frame was already
876 * output to the user), so we also set h->mmco_reset to detect the MMCO
878 * FIXME: if we detect insufficient delays (as per h->avctx->has_b_frames),
879 * we increase the delay between input and output. All frames affected by
880 * the lag (e.g. those that should have been output before another frame
881 * that we already returned to the user) will be dropped. This is a bug
882 * that we will fix later. */
883 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
884 cnt += out->poc < h->last_pocs[i];
885 invalid += out->poc == INT_MIN;
887 if (!h->mmco_reset && !cur->f->key_frame &&
888 cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
891 h->delayed_pic[pics - 2]->mmco_reset = 2;
893 if (h->mmco_reset || cur->f->key_frame) {
894 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
895 h->last_pocs[i] = INT_MIN;
897 invalid = MAX_DELAYED_PIC_COUNT;
899 out = h->delayed_pic[0];
901 for (i = 1; i < MAX_DELAYED_PIC_COUNT &&
903 !h->delayed_pic[i - 1]->mmco_reset &&
904 !h->delayed_pic[i]->f->key_frame;
906 if (h->delayed_pic[i]->poc < out->poc) {
907 out = h->delayed_pic[i];
910 if (h->avctx->has_b_frames == 0 &&
911 (h->delayed_pic[0]->f->key_frame || h->mmco_reset))
912 h->next_outputed_poc = INT_MIN;
913 out_of_order = !out->f->key_frame && !h->mmco_reset &&
914 (out->poc < h->next_outputed_poc);
916 if (h->sps.bitstream_restriction_flag &&
917 h->avctx->has_b_frames >= h->sps.num_reorder_frames) {
918 } else if (out_of_order && pics - 1 == h->avctx->has_b_frames &&
919 h->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
920 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
921 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, cnt);
924 } else if (h->low_delay &&
925 ((h->next_outputed_poc != INT_MIN &&
926 out->poc > h->next_outputed_poc + 2) ||
927 cur->f->pict_type == AV_PICTURE_TYPE_B)) {
929 h->avctx->has_b_frames++;
932 if (pics > h->avctx->has_b_frames) {
933 out->reference &= ~DELAYED_PIC_REF;
934 // for frame threading, the owner must be the second field's thread or
935 // else the first thread can release the picture and reuse it unsafely
936 for (i = out_idx; h->delayed_pic[i]; i++)
937 h->delayed_pic[i] = h->delayed_pic[i + 1];
939 memmove(h->last_pocs, &h->last_pocs[1],
940 sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
941 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
942 if (!out_of_order && pics > h->avctx->has_b_frames) {
943 h->next_output_pic = out;
944 if (out->mmco_reset) {
946 h->next_outputed_poc = out->poc;
947 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
949 h->next_outputed_poc = INT_MIN;
952 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f->key_frame) {
953 h->next_outputed_poc = INT_MIN;
955 h->next_outputed_poc = out->poc;
960 av_log(h->avctx, AV_LOG_DEBUG, "no picture\n");
963 if (h->next_output_pic) {
964 if (h->next_output_pic->recovered) {
965 // We have reached an recovery point and all frames after it in
966 // display order are "recovered".
967 h->frame_recovered |= FRAME_RECOVERED_SEI;
969 h->next_output_pic->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
972 if (setup_finished && !h->avctx->hwaccel) {
973 ff_thread_finish_setup(h->avctx);
975 if (h->avctx->active_thread_type & FF_THREAD_FRAME)
976 h->setup_finished = 1;
980 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl)
983 int luma_def, chroma_def;
986 sl->use_weight_chroma = 0;
987 sl->luma_log2_weight_denom = get_ue_golomb(&sl->gb);
988 if (h->sps.chroma_format_idc)
989 sl->chroma_log2_weight_denom = get_ue_golomb(&sl->gb);
990 luma_def = 1 << sl->luma_log2_weight_denom;
991 chroma_def = 1 << sl->chroma_log2_weight_denom;
993 for (list = 0; list < 2; list++) {
994 sl->luma_weight_flag[list] = 0;
995 sl->chroma_weight_flag[list] = 0;
996 for (i = 0; i < sl->ref_count[list]; i++) {
997 int luma_weight_flag, chroma_weight_flag;
999 luma_weight_flag = get_bits1(&sl->gb);
1000 if (luma_weight_flag) {
1001 sl->luma_weight[i][list][0] = get_se_golomb(&sl->gb);
1002 sl->luma_weight[i][list][1] = get_se_golomb(&sl->gb);
1003 if (sl->luma_weight[i][list][0] != luma_def ||
1004 sl->luma_weight[i][list][1] != 0) {
1006 sl->luma_weight_flag[list] = 1;
1009 sl->luma_weight[i][list][0] = luma_def;
1010 sl->luma_weight[i][list][1] = 0;
1013 if (h->sps.chroma_format_idc) {
1014 chroma_weight_flag = get_bits1(&sl->gb);
1015 if (chroma_weight_flag) {
1017 for (j = 0; j < 2; j++) {
1018 sl->chroma_weight[i][list][j][0] = get_se_golomb(&sl->gb);
1019 sl->chroma_weight[i][list][j][1] = get_se_golomb(&sl->gb);
1020 if (sl->chroma_weight[i][list][j][0] != chroma_def ||
1021 sl->chroma_weight[i][list][j][1] != 0) {
1022 sl->use_weight_chroma = 1;
1023 sl->chroma_weight_flag[list] = 1;
1028 for (j = 0; j < 2; j++) {
1029 sl->chroma_weight[i][list][j][0] = chroma_def;
1030 sl->chroma_weight[i][list][j][1] = 0;
1035 if (sl->slice_type_nos != AV_PICTURE_TYPE_B)
1038 sl->use_weight = sl->use_weight || sl->use_weight_chroma;
1043 * instantaneous decoder refresh.
1045 static void idr(H264Context *h)
1047 ff_h264_remove_all_refs(h);
1049 h->prev_frame_num_offset =
1051 h->prev_poc_lsb = 0;
1054 /* forget old pics after a seek */
1055 void ff_h264_flush_change(H264Context *h)
1058 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1059 h->last_pocs[i] = INT_MIN;
1060 h->next_outputed_poc = INT_MIN;
1061 h->prev_interlaced_frame = 1;
1064 h->cur_pic_ptr->reference = 0;
1066 ff_h264_reset_sei(h);
1067 h->recovery_frame = -1;
1068 h->frame_recovered = 0;
1071 /* forget old pics after a seek */
1072 static void flush_dpb(AVCodecContext *avctx)
1074 H264Context *h = avctx->priv_data;
1077 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
1079 ff_h264_flush_change(h);
1081 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
1082 ff_h264_unref_picture(h, &h->DPB[i]);
1083 h->cur_pic_ptr = NULL;
1084 ff_h264_unref_picture(h, &h->cur_pic);
1088 ff_h264_free_tables(h);
1089 h->context_initialized = 0;
1092 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
1094 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
1097 h->frame_num_offset = h->prev_frame_num_offset;
1098 if (h->frame_num < h->prev_frame_num)
1099 h->frame_num_offset += max_frame_num;
1101 if (h->sps.poc_type == 0) {
1102 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
1104 if (h->poc_lsb < h->prev_poc_lsb &&
1105 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
1106 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1107 else if (h->poc_lsb > h->prev_poc_lsb &&
1108 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
1109 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1111 h->poc_msb = h->prev_poc_msb;
1113 field_poc[1] = h->poc_msb + h->poc_lsb;
1114 if (h->picture_structure == PICT_FRAME)
1115 field_poc[1] += h->delta_poc_bottom;
1116 } else if (h->sps.poc_type == 1) {
1117 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1120 if (h->sps.poc_cycle_length != 0)
1121 abs_frame_num = h->frame_num_offset + h->frame_num;
1125 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
1128 expected_delta_per_poc_cycle = 0;
1129 for (i = 0; i < h->sps.poc_cycle_length; i++)
1130 // FIXME integrate during sps parse
1131 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
1133 if (abs_frame_num > 0) {
1134 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1135 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1137 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1138 for (i = 0; i <= frame_num_in_poc_cycle; i++)
1139 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
1143 if (h->nal_ref_idc == 0)
1144 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1146 field_poc[0] = expectedpoc + h->delta_poc[0];
1147 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1149 if (h->picture_structure == PICT_FRAME)
1150 field_poc[1] += h->delta_poc[1];
1152 int poc = 2 * (h->frame_num_offset + h->frame_num);
1154 if (!h->nal_ref_idc)
1161 if (h->picture_structure != PICT_BOTTOM_FIELD)
1162 pic_field_poc[0] = field_poc[0];
1163 if (h->picture_structure != PICT_TOP_FIELD)
1164 pic_field_poc[1] = field_poc[1];
1165 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
1171 * Compute profile from profile_idc and constraint_set?_flags.
1175 * @return profile as defined by FF_PROFILE_H264_*
1177 int ff_h264_get_profile(SPS *sps)
1179 int profile = sps->profile_idc;
1181 switch (sps->profile_idc) {
1182 case FF_PROFILE_H264_BASELINE:
1183 // constraint_set1_flag set to 1
1184 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1186 case FF_PROFILE_H264_HIGH_10:
1187 case FF_PROFILE_H264_HIGH_422:
1188 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1189 // constraint_set3_flag set to 1
1190 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
1197 int ff_set_ref_count(H264Context *h, H264SliceContext *sl)
1199 int ref_count[2], list_count;
1200 int num_ref_idx_active_override_flag, max_refs;
1202 // set defaults, might be overridden a few lines later
1203 ref_count[0] = h->pps.ref_count[0];
1204 ref_count[1] = h->pps.ref_count[1];
1206 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1207 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1208 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1209 num_ref_idx_active_override_flag = get_bits1(&sl->gb);
1211 if (num_ref_idx_active_override_flag) {
1212 ref_count[0] = get_ue_golomb(&sl->gb) + 1;
1213 if (ref_count[0] < 1)
1214 return AVERROR_INVALIDDATA;
1215 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1216 ref_count[1] = get_ue_golomb(&sl->gb) + 1;
1217 if (ref_count[1] < 1)
1218 return AVERROR_INVALIDDATA;
1222 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1228 ref_count[0] = ref_count[1] = 0;
1231 max_refs = h->picture_structure == PICT_FRAME ? 16 : 32;
1233 if (ref_count[0] > max_refs || ref_count[1] > max_refs) {
1234 av_log(h->avctx, AV_LOG_ERROR, "reference overflow\n");
1235 sl->ref_count[0] = sl->ref_count[1] = 0;
1236 return AVERROR_INVALIDDATA;
1239 if (list_count != sl->list_count ||
1240 ref_count[0] != sl->ref_count[0] ||
1241 ref_count[1] != sl->ref_count[1]) {
1242 sl->ref_count[0] = ref_count[0];
1243 sl->ref_count[1] = ref_count[1];
1244 sl->list_count = list_count;
1251 static int find_start_code(const uint8_t *buf, int buf_size,
1252 int buf_index, int next_avc)
1254 // start code prefix search
1255 for (; buf_index + 3 < next_avc; buf_index++)
1256 // This should always succeed in the first iteration.
1257 if (buf[buf_index] == 0 &&
1258 buf[buf_index + 1] == 0 &&
1259 buf[buf_index + 2] == 1)
1262 if (buf_index + 3 >= buf_size)
1265 return buf_index + 3;
1268 static int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1269 int buf_size, int *buf_index)
1273 if (*buf_index >= buf_size - h->nal_length_size) {
1274 // the end of the buffer is reached, refill it.
1275 return AVERROR(EAGAIN);
1278 for (i = 0; i < h->nal_length_size; i++)
1279 nalsize = (nalsize << 8) | buf[(*buf_index)++];
1280 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1281 av_log(h->avctx, AV_LOG_ERROR,
1282 "AVC: nal size %d\n", nalsize);
1283 return AVERROR_INVALIDDATA;
1288 static int get_bit_length(H264Context *h, const uint8_t *buf,
1289 const uint8_t *ptr, int dst_length,
1290 int i, int next_avc)
1292 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
1293 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
1294 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
1295 h->workaround_bugs |= FF_BUG_TRUNCATED;
1297 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
1298 while (dst_length > 0 && ptr[dst_length - 1] == 0)
1304 return 8 * dst_length - decode_rbsp_trailing(h, ptr + dst_length - 1);
1307 static int get_last_needed_nal(H264Context *h, const uint8_t *buf, int buf_size)
1309 int next_avc = h->is_avc ? 0 : buf_size;
1312 int nals_needed = 0;
1317 int dst_length, bit_length, consumed;
1320 if (buf_index >= next_avc) {
1321 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1324 next_avc = buf_index + nalsize;
1326 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1327 if (buf_index >= buf_size)
1331 ptr = ff_h264_decode_nal(h, &h->slice_ctx[0], buf + buf_index, &dst_length, &consumed,
1332 next_avc - buf_index);
1334 if (!ptr || dst_length < 0)
1335 return AVERROR_INVALIDDATA;
1337 buf_index += consumed;
1339 bit_length = get_bit_length(h, buf, ptr, dst_length,
1340 buf_index, next_avc);
1343 /* packets can sometimes contain multiple PPS/SPS,
1344 * e.g. two PAFF field pictures in one packet, or a demuxer
1345 * which splits NALs strangely if so, when frame threading we
1346 * can't start the next thread until we've read all of them */
1347 switch (h->nal_unit_type) {
1350 nals_needed = nal_index;
1355 init_get_bits(&gb, ptr, bit_length);
1356 if (!get_ue_golomb(&gb))
1357 nals_needed = nal_index;
1364 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1365 int parse_extradata)
1367 AVCodecContext *const avctx = h->avctx;
1368 H264SliceContext *sl;
1370 unsigned context_count;
1372 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
1376 h->max_contexts = h->slice_context_count;
1377 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
1378 h->current_slice = 0;
1379 if (!h->first_field)
1380 h->cur_pic_ptr = NULL;
1381 ff_h264_reset_sei(h);
1384 if (avctx->active_thread_type & FF_THREAD_FRAME)
1385 nals_needed = get_last_needed_nal(h, buf, buf_size);
1390 next_avc = h->is_avc ? 0 : buf_size;
1400 if (buf_index >= next_avc) {
1401 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1404 next_avc = buf_index + nalsize;
1406 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1407 if (buf_index >= buf_size)
1409 if (buf_index >= next_avc)
1413 sl = &h->slice_ctx[context_count];
1415 ptr = ff_h264_decode_nal(h, sl, buf + buf_index, &dst_length,
1416 &consumed, next_avc - buf_index);
1417 if (!ptr || dst_length < 0) {
1422 bit_length = get_bit_length(h, buf, ptr, dst_length,
1423 buf_index + consumed, next_avc);
1425 if (h->avctx->debug & FF_DEBUG_STARTCODE)
1426 av_log(h->avctx, AV_LOG_DEBUG,
1427 "NAL %d at %d/%d length %d\n",
1428 h->nal_unit_type, buf_index, buf_size, dst_length);
1430 if (h->is_avc && (nalsize != consumed) && nalsize)
1431 av_log(h->avctx, AV_LOG_DEBUG,
1432 "AVC: Consumed only %d bytes instead of %d\n",
1435 buf_index += consumed;
1438 if (avctx->skip_frame >= AVDISCARD_NONREF &&
1439 h->nal_ref_idc == 0 &&
1440 h->nal_unit_type != NAL_SEI)
1444 /* Ignore every NAL unit type except PPS and SPS during extradata
1445 * parsing. Decoding slices is not possible in codec init
1447 if (parse_extradata && HAVE_THREADS &&
1448 (h->avctx->active_thread_type & FF_THREAD_FRAME) &&
1449 (h->nal_unit_type != NAL_PPS &&
1450 h->nal_unit_type != NAL_SPS)) {
1451 if (h->nal_unit_type < NAL_AUD ||
1452 h->nal_unit_type > NAL_AUXILIARY_SLICE)
1453 av_log(avctx, AV_LOG_INFO,
1454 "Ignoring NAL unit %d during extradata parsing\n",
1456 h->nal_unit_type = NAL_FF_IGNORE;
1459 switch (h->nal_unit_type) {
1461 if (h->nal_unit_type != NAL_IDR_SLICE) {
1462 av_log(h->avctx, AV_LOG_ERROR,
1463 "Invalid mix of idr and non-idr slices\n");
1467 idr(h); // FIXME ensure we don't lose some frames if there is reordering
1469 init_get_bits(&sl->gb, ptr, bit_length);
1471 if ((err = ff_h264_decode_slice_header(h, sl)))
1474 if (h->sei_recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
1475 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
1476 ((1 << h->sps.log2_max_frame_num) - 1);
1479 h->cur_pic_ptr->f->key_frame |=
1480 (h->nal_unit_type == NAL_IDR_SLICE) ||
1481 (h->sei_recovery_frame_cnt >= 0);
1483 if (h->nal_unit_type == NAL_IDR_SLICE ||
1484 h->recovery_frame == h->frame_num) {
1485 h->recovery_frame = -1;
1486 h->cur_pic_ptr->recovered = 1;
1488 // If we have an IDR, all frames after it in decoded order are
1490 if (h->nal_unit_type == NAL_IDR_SLICE)
1491 h->frame_recovered |= FRAME_RECOVERED_IDR;
1492 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1494 if (h->current_slice == 1) {
1495 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS))
1496 decode_postinit(h, nal_index >= nals_needed);
1498 if (h->avctx->hwaccel &&
1499 (ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
1503 if (sl->redundant_pic_count == 0 &&
1504 (avctx->skip_frame < AVDISCARD_NONREF ||
1506 (avctx->skip_frame < AVDISCARD_BIDIR ||
1507 sl->slice_type_nos != AV_PICTURE_TYPE_B) &&
1508 (avctx->skip_frame < AVDISCARD_NONKEY ||
1509 h->cur_pic_ptr->f->key_frame) &&
1510 avctx->skip_frame < AVDISCARD_ALL) {
1511 if (avctx->hwaccel) {
1512 ret = avctx->hwaccel->decode_slice(avctx,
1513 &buf[buf_index - consumed],
1524 avpriv_request_sample(avctx, "data partitioning");
1525 ret = AVERROR(ENOSYS);
1529 init_get_bits(&h->gb, ptr, bit_length);
1530 ret = ff_h264_decode_sei(h);
1531 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1535 init_get_bits(&h->gb, ptr, bit_length);
1536 ret = ff_h264_decode_seq_parameter_set(h);
1537 if (ret < 0 && h->is_avc && (nalsize != consumed) && nalsize) {
1538 av_log(h->avctx, AV_LOG_DEBUG,
1539 "SPS decoding failure, trying again with the complete NAL\n");
1540 init_get_bits(&h->gb, buf + buf_index + 1 - consumed,
1542 ff_h264_decode_seq_parameter_set(h);
1547 init_get_bits(&h->gb, ptr, bit_length);
1548 ret = ff_h264_decode_picture_parameter_set(h, bit_length);
1549 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1553 case NAL_END_SEQUENCE:
1554 case NAL_END_STREAM:
1555 case NAL_FILLER_DATA:
1557 case NAL_AUXILIARY_SLICE:
1562 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
1563 h->nal_unit_type, bit_length);
1566 if (context_count == h->max_contexts) {
1567 ret = ff_h264_execute_decode_slices(h, context_count);
1568 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1574 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
1575 sl->ref_count[0] = sl->ref_count[1] = sl->list_count = 0;
1576 } else if (err == 1) {
1577 /* Slice could not be decoded in parallel mode, restart. Note
1578 * that rbsp_buffer is not transferred, but since we no longer
1579 * run in parallel mode this should not be an issue. */
1580 sl = &h->slice_ctx[0];
1585 if (context_count) {
1586 ret = ff_h264_execute_decode_slices(h, context_count);
1587 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1594 if (h->cur_pic_ptr && !h->droppable) {
1595 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1596 h->picture_structure == PICT_BOTTOM_FIELD);
1599 return (ret < 0) ? ret : buf_index;
1603 * Return the number of bytes consumed for building the current frame.
1605 static int get_consumed_bytes(int pos, int buf_size)
1608 pos = 1; // avoid infinite loops (I doubt that is needed but...)
1609 if (pos + 10 > buf_size)
1610 pos = buf_size; // oops ;)
1615 static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src)
1618 int ret = av_frame_ref(dst, src);
1625 for (i = 0; i < 3; i++) {
1626 int hshift = (i > 0) ? h->chroma_x_shift : 0;
1627 int vshift = (i > 0) ? h->chroma_y_shift : 0;
1628 int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) +
1629 (h->sps.crop_top >> vshift) * dst->linesize[i];
1630 dst->data[i] += off;
1635 static int h264_decode_frame(AVCodecContext *avctx, void *data,
1636 int *got_frame, AVPacket *avpkt)
1638 const uint8_t *buf = avpkt->data;
1639 int buf_size = avpkt->size;
1640 H264Context *h = avctx->priv_data;
1641 AVFrame *pict = data;
1645 h->flags = avctx->flags;
1646 h->setup_finished = 0;
1648 /* end of stream, output what is still in the buffers */
1650 if (buf_size == 0) {
1654 h->cur_pic_ptr = NULL;
1656 // FIXME factorize this with the output code below
1657 out = h->delayed_pic[0];
1660 h->delayed_pic[i] &&
1661 !h->delayed_pic[i]->f->key_frame &&
1662 !h->delayed_pic[i]->mmco_reset;
1664 if (h->delayed_pic[i]->poc < out->poc) {
1665 out = h->delayed_pic[i];
1669 for (i = out_idx; h->delayed_pic[i]; i++)
1670 h->delayed_pic[i] = h->delayed_pic[i + 1];
1673 ret = output_frame(h, pict, out->f);
1682 buf_index = decode_nal_units(h, buf, buf_size, 0);
1684 return AVERROR_INVALIDDATA;
1686 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
1691 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
1692 if (avctx->skip_frame >= AVDISCARD_NONREF)
1694 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1695 return AVERROR_INVALIDDATA;
1698 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1699 (h->mb_y >= h->mb_height && h->mb_height)) {
1700 if (avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)
1701 decode_postinit(h, 1);
1703 ff_h264_field_end(h, &h->slice_ctx[0], 0);
1706 if (h->next_output_pic && ((avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
1707 h->next_output_pic->recovered)) {
1708 if (!h->next_output_pic->recovered)
1709 h->next_output_pic->f->flags |= AV_FRAME_FLAG_CORRUPT;
1711 ret = output_frame(h, pict, h->next_output_pic->f);
1718 assert(pict->buf[0] || !*got_frame);
1720 return get_consumed_bytes(buf_index, buf_size);
1723 av_cold void ff_h264_free_context(H264Context *h)
1727 ff_h264_free_tables(h);
1729 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
1730 ff_h264_unref_picture(h, &h->DPB[i]);
1731 av_frame_free(&h->DPB[i].f);
1734 h->cur_pic_ptr = NULL;
1736 for (i = 0; i < h->nb_slice_ctx; i++)
1737 av_freep(&h->slice_ctx[i].rbsp_buffer);
1738 av_freep(&h->slice_ctx);
1739 h->nb_slice_ctx = 0;
1741 for (i = 0; i < MAX_SPS_COUNT; i++)
1742 av_freep(h->sps_buffers + i);
1744 for (i = 0; i < MAX_PPS_COUNT; i++)
1745 av_freep(h->pps_buffers + i);
1748 static av_cold int h264_decode_end(AVCodecContext *avctx)
1750 H264Context *h = avctx->priv_data;
1752 ff_h264_free_context(h);
1754 ff_h264_unref_picture(h, &h->cur_pic);
1755 av_frame_free(&h->cur_pic.f);
1760 #define OFFSET(x) offsetof(H264Context, x)
1761 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1762 static const AVOption h264_options[] = {
1763 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VD },
1767 static const AVClass h264_class = {
1768 .class_name = "h264",
1769 .item_name = av_default_item_name,
1770 .option = h264_options,
1771 .version = LIBAVUTIL_VERSION_INT,
1774 static const AVProfile profiles[] = {
1775 { FF_PROFILE_H264_BASELINE, "Baseline" },
1776 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
1777 { FF_PROFILE_H264_MAIN, "Main" },
1778 { FF_PROFILE_H264_EXTENDED, "Extended" },
1779 { FF_PROFILE_H264_HIGH, "High" },
1780 { FF_PROFILE_H264_HIGH_10, "High 10" },
1781 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
1782 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
1783 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
1784 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
1785 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
1786 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
1787 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
1788 { FF_PROFILE_UNKNOWN },
1791 AVCodec ff_h264_decoder = {
1793 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1794 .type = AVMEDIA_TYPE_VIDEO,
1795 .id = AV_CODEC_ID_H264,
1796 .priv_data_size = sizeof(H264Context),
1797 .init = ff_h264_decode_init,
1798 .close = h264_decode_end,
1799 .decode = h264_decode_frame,
1800 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1801 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1802 AV_CODEC_CAP_FRAME_THREADS,
1803 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
1805 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
1806 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1807 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1808 .priv_class = &h264_class,