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 #define UNCHECKED_BITSTREAM_READER 1
30 #include "libavutil/avassert.h"
31 #include "libavutil/display.h"
32 #include "libavutil/imgutils.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/stereo3d.h"
35 #include "libavutil/timer.h"
38 #include "cabac_functions.h"
39 #include "error_resilience.h"
43 #include "h264chroma.h"
44 #include "h264_mvpred.h"
48 #include "mpegutils.h"
49 #include "rectangle.h"
52 #include "vdpau_compat.h"
54 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
56 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
58 H264Context *h = avctx->priv_data;
59 return h ? h->sps.num_reorder_frames : 0;
62 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
64 int mb_x, int mb_y, int mb_intra, int mb_skipped)
66 H264Context *h = opaque;
67 H264SliceContext *sl = &h->slice_ctx[0];
71 sl->mb_xy = mb_x + mb_y * h->mb_stride;
72 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
74 /* FIXME: It is possible albeit uncommon that slice references
75 * differ between slices. We take the easy approach and ignore
76 * it for now. If this turns out to have any relevance in
77 * practice then correct remapping should be added. */
78 if (ref >= sl->ref_count[0])
80 if (!sl->ref_list[0][ref].data[0]) {
81 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
84 if ((sl->ref_list[0][ref].reference&3) != 3) {
85 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
88 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
90 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
91 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
92 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
94 sl->mb_field_decoding_flag = 0;
95 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
98 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
101 AVCodecContext *avctx = h->avctx;
102 const AVFrame *src = h->cur_pic.f;
103 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
104 int vshift = desc->log2_chroma_h;
105 const int field_pic = h->picture_structure != PICT_FRAME;
111 height = FFMIN(height, avctx->height - y);
113 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
116 if (avctx->draw_horiz_band) {
117 int offset[AV_NUM_DATA_POINTERS];
120 offset[0] = y * src->linesize[0];
122 offset[2] = (y >> vshift) * src->linesize[1];
123 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
128 avctx->draw_horiz_band(avctx, src, offset,
129 y, h->picture_structure, height);
134 * Check if the top & left blocks are available if needed and
135 * change the dc mode so it only uses the available blocks.
137 int ff_h264_check_intra4x4_pred_mode(const H264Context *h, H264SliceContext *sl)
139 static const int8_t top[12] = {
140 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
142 static const int8_t left[12] = {
143 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
147 if (!(sl->top_samples_available & 0x8000)) {
148 for (i = 0; i < 4; i++) {
149 int status = top[sl->intra4x4_pred_mode_cache[scan8[0] + i]];
151 av_log(h->avctx, AV_LOG_ERROR,
152 "top 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] + i] = status;
161 if ((sl->left_samples_available & 0x8888) != 0x8888) {
162 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
163 for (i = 0; i < 4; i++)
164 if (!(sl->left_samples_available & mask[i])) {
165 int status = left[sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
167 av_log(h->avctx, AV_LOG_ERROR,
168 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
169 status, sl->mb_x, sl->mb_y);
170 return AVERROR_INVALIDDATA;
172 sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
178 } // FIXME cleanup like ff_h264_check_intra_pred_mode
181 * Check if the top & left blocks are available if needed and
182 * change the dc mode so it only uses the available blocks.
184 int ff_h264_check_intra_pred_mode(const H264Context *h, H264SliceContext *sl,
185 int mode, int is_chroma)
187 static const int8_t top[4] = { LEFT_DC_PRED8x8, 1, -1, -1 };
188 static const int8_t left[5] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
191 av_log(h->avctx, AV_LOG_ERROR,
192 "out of range intra chroma pred mode at %d %d\n",
194 return AVERROR_INVALIDDATA;
197 if (!(sl->top_samples_available & 0x8000)) {
200 av_log(h->avctx, AV_LOG_ERROR,
201 "top block unavailable for requested intra mode at %d %d\n",
203 return AVERROR_INVALIDDATA;
207 if ((sl->left_samples_available & 0x8080) != 0x8080) {
210 av_log(h->avctx, AV_LOG_ERROR,
211 "left block unavailable for requested intra mode at %d %d\n",
213 return AVERROR_INVALIDDATA;
215 if (is_chroma && (sl->left_samples_available & 0x8080)) {
216 // mad cow disease mode, aka MBAFF + constrained_intra_pred
217 mode = ALZHEIMER_DC_L0T_PRED8x8 +
218 (!(sl->left_samples_available & 0x8000)) +
219 2 * (mode == DC_128_PRED8x8);
226 const uint8_t *ff_h264_decode_nal(H264Context *h, H264SliceContext *sl,
228 int *dst_length, int *consumed, int length)
233 // src[0]&0x80; // forbidden bit
234 h->nal_ref_idc = src[0] >> 5;
235 h->nal_unit_type = src[0] & 0x1F;
240 #define STARTCODE_TEST \
241 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
242 if (src[i + 2] != 3 && src[i + 2] != 0) { \
243 /* startcode, so we must be past the end */ \
249 #if HAVE_FAST_UNALIGNED
250 #define FIND_FIRST_ZERO \
251 if (i > 0 && !src[i]) \
257 for (i = 0; i + 1 < length; i += 9) {
258 if (!((~AV_RN64A(src + i) &
259 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
260 0x8000800080008080ULL))
267 for (i = 0; i + 1 < length; i += 5) {
268 if (!((~AV_RN32A(src + i) &
269 (AV_RN32A(src + i) - 0x01000101U)) &
278 for (i = 0; i + 1 < length; i += 2) {
281 if (i > 0 && src[i - 1] == 0)
287 av_fast_padded_malloc(&sl->rbsp_buffer, &sl->rbsp_buffer_size, length+MAX_MBPAIR_SIZE);
288 dst = sl->rbsp_buffer;
293 if(i>=length-1){ //no escaped 0
295 *consumed= length+1; //+1 for the header
296 if(h->avctx->flags2 & AV_CODEC_FLAG2_FAST){
299 memcpy(dst, src, length);
306 while (si + 2 < length) {
307 // remove escapes (very rare 1:2^22)
308 if (src[si + 2] > 3) {
309 dst[di++] = src[si++];
310 dst[di++] = src[si++];
311 } else if (src[si] == 0 && src[si + 1] == 0 && src[si + 2] != 0) {
312 if (src[si + 2] == 3) { // escape
317 } else // next start code
321 dst[di++] = src[si++];
324 dst[di++] = src[si++];
327 memset(dst + di, 0, AV_INPUT_BUFFER_PADDING_SIZE);
330 *consumed = si + 1; // +1 for the header
331 /* FIXME store exact number of bits in the getbitcontext
332 * (it is needed for decoding) */
337 * Identify the exact end of the bitstream
338 * @return the length of the trailing, or 0 if damaged
340 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
345 ff_tlog(h->avctx, "rbsp trailing %X\n", v);
347 for (r = 1; r < 9; r++) {
355 void ff_h264_free_tables(H264Context *h)
359 av_freep(&h->intra4x4_pred_mode);
360 av_freep(&h->chroma_pred_mode_table);
361 av_freep(&h->cbp_table);
362 av_freep(&h->mvd_table[0]);
363 av_freep(&h->mvd_table[1]);
364 av_freep(&h->direct_table);
365 av_freep(&h->non_zero_count);
366 av_freep(&h->slice_table_base);
367 h->slice_table = NULL;
368 av_freep(&h->list_counts);
370 av_freep(&h->mb2b_xy);
371 av_freep(&h->mb2br_xy);
373 av_buffer_pool_uninit(&h->qscale_table_pool);
374 av_buffer_pool_uninit(&h->mb_type_pool);
375 av_buffer_pool_uninit(&h->motion_val_pool);
376 av_buffer_pool_uninit(&h->ref_index_pool);
378 for (i = 0; i < h->nb_slice_ctx; i++) {
379 H264SliceContext *sl = &h->slice_ctx[i];
381 av_freep(&sl->dc_val_base);
382 av_freep(&sl->er.mb_index2xy);
383 av_freep(&sl->er.error_status_table);
384 av_freep(&sl->er.er_temp_buffer);
386 av_freep(&sl->bipred_scratchpad);
387 av_freep(&sl->edge_emu_buffer);
388 av_freep(&sl->top_borders[0]);
389 av_freep(&sl->top_borders[1]);
391 sl->bipred_scratchpad_allocated = 0;
392 sl->edge_emu_buffer_allocated = 0;
393 sl->top_borders_allocated[0] = 0;
394 sl->top_borders_allocated[1] = 0;
398 int ff_h264_alloc_tables(H264Context *h)
400 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
401 const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1);
404 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
405 row_mb_num, 8 * sizeof(uint8_t), fail)
406 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
408 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
409 big_mb_num * 48 * sizeof(uint8_t), fail)
410 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
411 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
412 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
413 big_mb_num * sizeof(uint16_t), fail)
414 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
415 big_mb_num * sizeof(uint8_t), fail)
416 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[0],
417 row_mb_num, 16 * sizeof(uint8_t), fail);
418 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[1],
419 row_mb_num, 16 * sizeof(uint8_t), fail);
420 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
421 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
423 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
424 4 * big_mb_num * sizeof(uint8_t), fail);
425 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
426 big_mb_num * sizeof(uint8_t), fail)
428 memset(h->slice_table_base, -1,
429 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
430 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
432 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
433 big_mb_num * sizeof(uint32_t), fail);
434 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
435 big_mb_num * sizeof(uint32_t), fail);
436 for (y = 0; y < h->mb_height; y++)
437 for (x = 0; x < h->mb_width; x++) {
438 const int mb_xy = x + y * h->mb_stride;
439 const int b_xy = 4 * x + 4 * y * h->b_stride;
441 h->mb2b_xy[mb_xy] = b_xy;
442 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
445 if (!h->dequant4_coeff[0])
446 ff_h264_init_dequant_tables(h);
451 ff_h264_free_tables(h);
452 return AVERROR(ENOMEM);
457 * Allocate buffers which are not shared amongst multiple threads.
459 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
461 ERContext *er = &sl->er;
462 int mb_array_size = h->mb_height * h->mb_stride;
463 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
464 int c_size = h->mb_stride * (h->mb_height + 1);
465 int yc_size = y_size + 2 * c_size;
468 sl->ref_cache[0][scan8[5] + 1] =
469 sl->ref_cache[0][scan8[7] + 1] =
470 sl->ref_cache[0][scan8[13] + 1] =
471 sl->ref_cache[1][scan8[5] + 1] =
472 sl->ref_cache[1][scan8[7] + 1] =
473 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
475 if (sl != h->slice_ctx) {
476 memset(er, 0, sizeof(*er));
478 if (CONFIG_ERROR_RESILIENCE) {
481 er->avctx = h->avctx;
482 er->decode_mb = h264_er_decode_mb;
484 er->quarter_sample = 1;
486 er->mb_num = h->mb_num;
487 er->mb_width = h->mb_width;
488 er->mb_height = h->mb_height;
489 er->mb_stride = h->mb_stride;
490 er->b8_stride = h->mb_width * 2 + 1;
492 // error resilience code looks cleaner with this
493 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
494 (h->mb_num + 1) * sizeof(int), fail);
496 for (y = 0; y < h->mb_height; y++)
497 for (x = 0; x < h->mb_width; x++)
498 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
500 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
501 h->mb_stride + h->mb_width;
503 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
504 mb_array_size * sizeof(uint8_t), fail);
506 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
507 h->mb_height * h->mb_stride, fail);
509 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
510 yc_size * sizeof(int16_t), fail);
511 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
512 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
513 er->dc_val[2] = er->dc_val[1] + c_size;
514 for (i = 0; i < yc_size; i++)
515 sl->dc_val_base[i] = 1024;
521 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
524 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
525 int parse_extradata);
527 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
529 AVCodecContext *avctx = h->avctx;
532 if (!buf || size <= 0)
537 const unsigned char *p = buf;
542 av_log(avctx, AV_LOG_ERROR,
543 "avcC %d too short\n", size);
544 return AVERROR_INVALIDDATA;
546 /* sps and pps in the avcC always have length coded with 2 bytes,
547 * so put a fake nal_length_size = 2 while parsing them */
548 h->nal_length_size = 2;
549 // Decode sps from avcC
550 cnt = *(p + 5) & 0x1f; // Number of sps
552 for (i = 0; i < cnt; i++) {
553 nalsize = AV_RB16(p) + 2;
554 if(nalsize > size - (p-buf))
555 return AVERROR_INVALIDDATA;
556 ret = decode_nal_units(h, p, nalsize, 1);
558 av_log(avctx, AV_LOG_ERROR,
559 "Decoding sps %d from avcC failed\n", i);
564 // Decode pps from avcC
565 cnt = *(p++); // Number of pps
566 for (i = 0; i < cnt; i++) {
567 nalsize = AV_RB16(p) + 2;
568 if(nalsize > size - (p-buf))
569 return AVERROR_INVALIDDATA;
570 ret = decode_nal_units(h, p, nalsize, 1);
572 av_log(avctx, AV_LOG_ERROR,
573 "Decoding pps %d from avcC failed\n", i);
578 // Store right nal length size that will be used to parse all other nals
579 h->nal_length_size = (buf[4] & 0x03) + 1;
582 ret = decode_nal_units(h, buf, size, 1);
589 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
594 h->backup_width = -1;
595 h->backup_height = -1;
596 h->backup_pix_fmt = AV_PIX_FMT_NONE;
597 h->dequant_coeff_pps = -1;
598 h->current_sps_id = -1;
599 h->cur_chroma_format_idc = -1;
601 h->picture_structure = PICT_FRAME;
602 h->slice_context_count = 1;
603 h->workaround_bugs = avctx->workaround_bugs;
604 h->flags = avctx->flags;
605 h->prev_poc_msb = 1 << 16;
607 h->recovery_frame = -1;
608 h->frame_recovered = 0;
609 h->prev_frame_num = -1;
610 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
612 h->next_outputed_poc = INT_MIN;
613 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
614 h->last_pocs[i] = INT_MIN;
616 ff_h264_reset_sei(h);
618 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
620 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? H264_MAX_THREADS : 1;
621 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
624 return AVERROR(ENOMEM);
627 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
628 h->DPB[i].f = av_frame_alloc();
630 return AVERROR(ENOMEM);
633 h->cur_pic.f = av_frame_alloc();
635 return AVERROR(ENOMEM);
637 h->last_pic_for_ec.f = av_frame_alloc();
638 if (!h->last_pic_for_ec.f)
639 return AVERROR(ENOMEM);
641 for (i = 0; i < h->nb_slice_ctx; i++)
642 h->slice_ctx[i].h264 = h;
647 static AVOnce h264_vlc_init = AV_ONCE_INIT;
649 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
651 H264Context *h = avctx->priv_data;
654 ret = h264_init_context(avctx, h);
659 if (!avctx->has_b_frames)
662 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
664 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
665 return AVERROR_UNKNOWN;
668 if (avctx->codec_id == AV_CODEC_ID_H264) {
669 if (avctx->ticks_per_frame == 1) {
670 if(h->avctx->time_base.den < INT_MAX/2) {
671 h->avctx->time_base.den *= 2;
673 h->avctx->time_base.num /= 2;
675 avctx->ticks_per_frame = 2;
678 if (avctx->extradata_size > 0 && avctx->extradata) {
679 ret = ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
681 ff_h264_free_context(h);
686 if (h->sps.bitstream_restriction_flag &&
687 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
688 h->avctx->has_b_frames = h->sps.num_reorder_frames;
692 avctx->internal->allocate_progress = 1;
694 ff_h264_flush_change(h);
696 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
699 if (h->enable_er && (avctx->active_thread_type & FF_THREAD_SLICE)) {
700 av_log(avctx, AV_LOG_WARNING,
701 "Error resilience with slice threads is enabled. It is unsafe and unsupported and may crash. "
702 "Use it at your own risk\n");
709 static int decode_init_thread_copy(AVCodecContext *avctx)
711 H264Context *h = avctx->priv_data;
714 if (!avctx->internal->is_copy)
717 memset(h, 0, sizeof(*h));
719 ret = h264_init_context(avctx, h);
723 h->context_initialized = 0;
730 * Run setup operations that must be run after slice header decoding.
731 * This includes finding the next displayed frame.
733 * @param h h264 master context
734 * @param setup_finished enough NALs have been read that we can call
735 * ff_thread_finish_setup()
737 static void decode_postinit(H264Context *h, int setup_finished)
739 H264Picture *out = h->cur_pic_ptr;
740 H264Picture *cur = h->cur_pic_ptr;
741 int i, pics, out_of_order, out_idx;
743 h->cur_pic_ptr->f->pict_type = h->pict_type;
745 if (h->next_output_pic)
748 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
749 /* FIXME: if we have two PAFF fields in one packet, we can't start
750 * the next thread here. If we have one field per packet, we can.
751 * The check in decode_nal_units() is not good enough to find this
752 * yet, so we assume the worst for now. */
753 // if (setup_finished)
754 // ff_thread_finish_setup(h->avctx);
755 if (cur->field_poc[0] == INT_MAX && cur->field_poc[1] == INT_MAX)
757 if (h->avctx->hwaccel || h->missing_fields <=1)
761 cur->f->interlaced_frame = 0;
762 cur->f->repeat_pict = 0;
764 /* Signal interlacing information externally. */
765 /* Prioritize picture timing SEI information over used
766 * decoding process if it exists. */
768 if (h->sps.pic_struct_present_flag) {
769 switch (h->sei_pic_struct) {
770 case SEI_PIC_STRUCT_FRAME:
772 case SEI_PIC_STRUCT_TOP_FIELD:
773 case SEI_PIC_STRUCT_BOTTOM_FIELD:
774 cur->f->interlaced_frame = 1;
776 case SEI_PIC_STRUCT_TOP_BOTTOM:
777 case SEI_PIC_STRUCT_BOTTOM_TOP:
778 if (FIELD_OR_MBAFF_PICTURE(h))
779 cur->f->interlaced_frame = 1;
781 // try to flag soft telecine progressive
782 cur->f->interlaced_frame = h->prev_interlaced_frame;
784 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
785 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
786 /* Signal the possibility of telecined film externally
787 * (pic_struct 5,6). From these hints, let the applications
788 * decide if they apply deinterlacing. */
789 cur->f->repeat_pict = 1;
791 case SEI_PIC_STRUCT_FRAME_DOUBLING:
792 cur->f->repeat_pict = 2;
794 case SEI_PIC_STRUCT_FRAME_TRIPLING:
795 cur->f->repeat_pict = 4;
799 if ((h->sei_ct_type & 3) &&
800 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
801 cur->f->interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
803 /* Derive interlacing flag from used decoding process. */
804 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
806 h->prev_interlaced_frame = cur->f->interlaced_frame;
808 if (cur->field_poc[0] != cur->field_poc[1]) {
809 /* Derive top_field_first from field pocs. */
810 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
812 if (h->sps.pic_struct_present_flag) {
813 /* Use picture timing SEI information. Even if it is a
814 * information of a past frame, better than nothing. */
815 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
816 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
817 cur->f->top_field_first = 1;
819 cur->f->top_field_first = 0;
820 } else if (cur->f->interlaced_frame) {
821 /* Default to top field first when pic_struct_present_flag
822 * is not set but interlaced frame detected */
823 cur->f->top_field_first = 1;
825 /* Most likely progressive */
826 cur->f->top_field_first = 0;
830 if (h->sei_frame_packing_present &&
831 h->frame_packing_arrangement_type >= 0 &&
832 h->frame_packing_arrangement_type <= 6 &&
833 h->content_interpretation_type > 0 &&
834 h->content_interpretation_type < 3) {
835 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
837 switch (h->frame_packing_arrangement_type) {
839 stereo->type = AV_STEREO3D_CHECKERBOARD;
842 stereo->type = AV_STEREO3D_COLUMNS;
845 stereo->type = AV_STEREO3D_LINES;
848 if (h->quincunx_subsampling)
849 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
851 stereo->type = AV_STEREO3D_SIDEBYSIDE;
854 stereo->type = AV_STEREO3D_TOPBOTTOM;
857 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
860 stereo->type = AV_STEREO3D_2D;
864 if (h->content_interpretation_type == 2)
865 stereo->flags = AV_STEREO3D_FLAG_INVERT;
869 if (h->sei_display_orientation_present &&
870 (h->sei_anticlockwise_rotation || h->sei_hflip || h->sei_vflip)) {
871 double angle = h->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
872 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
873 AV_FRAME_DATA_DISPLAYMATRIX,
874 sizeof(int32_t) * 9);
876 av_display_rotation_set((int32_t *)rotation->data, angle);
877 av_display_matrix_flip((int32_t *)rotation->data,
878 h->sei_hflip, h->sei_vflip);
882 if (h->sei_reguserdata_afd_present) {
883 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
887 *sd->data = h->active_format_description;
888 h->sei_reguserdata_afd_present = 0;
892 if (h->a53_caption) {
893 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
894 AV_FRAME_DATA_A53_CC,
895 h->a53_caption_size);
897 memcpy(sd->data, h->a53_caption, h->a53_caption_size);
898 av_freep(&h->a53_caption);
899 h->a53_caption_size = 0;
900 h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
903 cur->mmco_reset = h->mmco_reset;
906 // FIXME do something with unavailable reference frames
908 /* Sort B-frames into display order */
909 if (h->sps.bitstream_restriction_flag ||
910 h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
911 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, h->sps.num_reorder_frames);
913 h->low_delay = !h->avctx->has_b_frames;
915 for (i = 0; 1; i++) {
916 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
918 h->last_pocs[i-1] = cur->poc;
921 h->last_pocs[i-1]= h->last_pocs[i];
924 out_of_order = MAX_DELAYED_PIC_COUNT - i;
925 if( cur->f->pict_type == AV_PICTURE_TYPE_B
926 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
927 out_of_order = FFMAX(out_of_order, 1);
928 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
929 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
930 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
931 h->last_pocs[i] = INT_MIN;
932 h->last_pocs[0] = cur->poc;
934 } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
935 av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
936 h->avctx->has_b_frames = out_of_order;
941 while (h->delayed_pic[pics])
944 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
946 h->delayed_pic[pics++] = cur;
947 if (cur->reference == 0)
948 cur->reference = DELAYED_PIC_REF;
950 out = h->delayed_pic[0];
952 for (i = 1; h->delayed_pic[i] &&
953 !h->delayed_pic[i]->f->key_frame &&
954 !h->delayed_pic[i]->mmco_reset;
956 if (h->delayed_pic[i]->poc < out->poc) {
957 out = h->delayed_pic[i];
960 if (h->avctx->has_b_frames == 0 &&
961 (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
962 h->next_outputed_poc = INT_MIN;
963 out_of_order = out->poc < h->next_outputed_poc;
965 if (out_of_order || pics > h->avctx->has_b_frames) {
966 out->reference &= ~DELAYED_PIC_REF;
967 // for frame threading, the owner must be the second field's thread or
968 // else the first thread can release the picture and reuse it unsafely
969 for (i = out_idx; h->delayed_pic[i]; i++)
970 h->delayed_pic[i] = h->delayed_pic[i + 1];
972 if (!out_of_order && pics > h->avctx->has_b_frames) {
973 h->next_output_pic = out;
974 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
975 h->next_outputed_poc = INT_MIN;
977 h->next_outputed_poc = out->poc;
979 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
982 if (h->next_output_pic) {
983 if (h->next_output_pic->recovered) {
984 // We have reached an recovery point and all frames after it in
985 // display order are "recovered".
986 h->frame_recovered |= FRAME_RECOVERED_SEI;
988 h->next_output_pic->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
991 if (setup_finished && !h->avctx->hwaccel) {
992 ff_thread_finish_setup(h->avctx);
994 if (h->avctx->active_thread_type & FF_THREAD_FRAME)
995 h->setup_finished = 1;
999 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl)
1002 int luma_def, chroma_def;
1005 sl->use_weight_chroma = 0;
1006 sl->luma_log2_weight_denom = get_ue_golomb(&sl->gb);
1007 if (h->sps.chroma_format_idc)
1008 sl->chroma_log2_weight_denom = get_ue_golomb(&sl->gb);
1010 if (sl->luma_log2_weight_denom > 7U) {
1011 av_log(h->avctx, AV_LOG_ERROR, "luma_log2_weight_denom %d is out of range\n", sl->luma_log2_weight_denom);
1012 sl->luma_log2_weight_denom = 0;
1014 if (sl->chroma_log2_weight_denom > 7U) {
1015 av_log(h->avctx, AV_LOG_ERROR, "chroma_log2_weight_denom %d is out of range\n", sl->chroma_log2_weight_denom);
1016 sl->chroma_log2_weight_denom = 0;
1019 luma_def = 1 << sl->luma_log2_weight_denom;
1020 chroma_def = 1 << sl->chroma_log2_weight_denom;
1022 for (list = 0; list < 2; list++) {
1023 sl->luma_weight_flag[list] = 0;
1024 sl->chroma_weight_flag[list] = 0;
1025 for (i = 0; i < sl->ref_count[list]; i++) {
1026 int luma_weight_flag, chroma_weight_flag;
1028 luma_weight_flag = get_bits1(&sl->gb);
1029 if (luma_weight_flag) {
1030 sl->luma_weight[i][list][0] = get_se_golomb(&sl->gb);
1031 sl->luma_weight[i][list][1] = get_se_golomb(&sl->gb);
1032 if (sl->luma_weight[i][list][0] != luma_def ||
1033 sl->luma_weight[i][list][1] != 0) {
1035 sl->luma_weight_flag[list] = 1;
1038 sl->luma_weight[i][list][0] = luma_def;
1039 sl->luma_weight[i][list][1] = 0;
1042 if (h->sps.chroma_format_idc) {
1043 chroma_weight_flag = get_bits1(&sl->gb);
1044 if (chroma_weight_flag) {
1046 for (j = 0; j < 2; j++) {
1047 sl->chroma_weight[i][list][j][0] = get_se_golomb(&sl->gb);
1048 sl->chroma_weight[i][list][j][1] = get_se_golomb(&sl->gb);
1049 if (sl->chroma_weight[i][list][j][0] != chroma_def ||
1050 sl->chroma_weight[i][list][j][1] != 0) {
1051 sl->use_weight_chroma = 1;
1052 sl->chroma_weight_flag[list] = 1;
1057 for (j = 0; j < 2; j++) {
1058 sl->chroma_weight[i][list][j][0] = chroma_def;
1059 sl->chroma_weight[i][list][j][1] = 0;
1064 if (sl->slice_type_nos != AV_PICTURE_TYPE_B)
1067 sl->use_weight = sl->use_weight || sl->use_weight_chroma;
1072 * instantaneous decoder refresh.
1074 static void idr(H264Context *h)
1077 ff_h264_remove_all_refs(h);
1079 h->prev_frame_num_offset = 0;
1080 h->prev_poc_msb = 1<<16;
1081 h->prev_poc_lsb = 0;
1082 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1083 h->last_pocs[i] = INT_MIN;
1086 /* forget old pics after a seek */
1087 void ff_h264_flush_change(H264Context *h)
1091 h->next_outputed_poc = INT_MIN;
1092 h->prev_interlaced_frame = 1;
1095 h->prev_frame_num = -1;
1096 if (h->cur_pic_ptr) {
1097 h->cur_pic_ptr->reference = 0;
1098 for (j=i=0; h->delayed_pic[i]; i++)
1099 if (h->delayed_pic[i] != h->cur_pic_ptr)
1100 h->delayed_pic[j++] = h->delayed_pic[i];
1101 h->delayed_pic[j] = NULL;
1103 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1106 ff_h264_reset_sei(h);
1107 h->recovery_frame = -1;
1108 h->frame_recovered = 0;
1109 h->current_slice = 0;
1111 for (i = 0; i < h->nb_slice_ctx; i++)
1112 h->slice_ctx[i].list_count = 0;
1115 /* forget old pics after a seek */
1116 static void flush_dpb(AVCodecContext *avctx)
1118 H264Context *h = avctx->priv_data;
1121 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
1123 ff_h264_flush_change(h);
1125 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
1126 ff_h264_unref_picture(h, &h->DPB[i]);
1127 h->cur_pic_ptr = NULL;
1128 ff_h264_unref_picture(h, &h->cur_pic);
1132 ff_h264_free_tables(h);
1133 h->context_initialized = 0;
1136 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
1138 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
1141 h->frame_num_offset = h->prev_frame_num_offset;
1142 if (h->frame_num < h->prev_frame_num)
1143 h->frame_num_offset += max_frame_num;
1145 if (h->sps.poc_type == 0) {
1146 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
1148 if (h->poc_lsb < h->prev_poc_lsb &&
1149 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
1150 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1151 else if (h->poc_lsb > h->prev_poc_lsb &&
1152 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
1153 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1155 h->poc_msb = h->prev_poc_msb;
1157 field_poc[1] = h->poc_msb + h->poc_lsb;
1158 if (h->picture_structure == PICT_FRAME)
1159 field_poc[1] += h->delta_poc_bottom;
1160 } else if (h->sps.poc_type == 1) {
1161 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1164 if (h->sps.poc_cycle_length != 0)
1165 abs_frame_num = h->frame_num_offset + h->frame_num;
1169 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
1172 expected_delta_per_poc_cycle = 0;
1173 for (i = 0; i < h->sps.poc_cycle_length; i++)
1174 // FIXME integrate during sps parse
1175 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
1177 if (abs_frame_num > 0) {
1178 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1179 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1181 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1182 for (i = 0; i <= frame_num_in_poc_cycle; i++)
1183 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
1187 if (h->nal_ref_idc == 0)
1188 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1190 field_poc[0] = expectedpoc + h->delta_poc[0];
1191 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1193 if (h->picture_structure == PICT_FRAME)
1194 field_poc[1] += h->delta_poc[1];
1196 int poc = 2 * (h->frame_num_offset + h->frame_num);
1198 if (!h->nal_ref_idc)
1205 if (h->picture_structure != PICT_BOTTOM_FIELD)
1206 pic_field_poc[0] = field_poc[0];
1207 if (h->picture_structure != PICT_TOP_FIELD)
1208 pic_field_poc[1] = field_poc[1];
1209 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
1215 * Compute profile from profile_idc and constraint_set?_flags.
1219 * @return profile as defined by FF_PROFILE_H264_*
1221 int ff_h264_get_profile(SPS *sps)
1223 int profile = sps->profile_idc;
1225 switch (sps->profile_idc) {
1226 case FF_PROFILE_H264_BASELINE:
1227 // constraint_set1_flag set to 1
1228 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1230 case FF_PROFILE_H264_HIGH_10:
1231 case FF_PROFILE_H264_HIGH_422:
1232 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1233 // constraint_set3_flag set to 1
1234 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
1241 int ff_set_ref_count(H264Context *h, H264SliceContext *sl)
1243 int ref_count[2], list_count;
1244 int num_ref_idx_active_override_flag;
1246 // set defaults, might be overridden a few lines later
1247 ref_count[0] = h->pps.ref_count[0];
1248 ref_count[1] = h->pps.ref_count[1];
1250 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1252 max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31;
1254 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1255 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1256 num_ref_idx_active_override_flag = get_bits1(&sl->gb);
1258 if (num_ref_idx_active_override_flag) {
1259 ref_count[0] = get_ue_golomb(&sl->gb) + 1;
1260 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1261 ref_count[1] = get_ue_golomb(&sl->gb) + 1;
1263 // full range is spec-ok in this case, even for frames
1267 if (ref_count[0]-1 > max[0] || ref_count[1]-1 > max[1]){
1268 av_log(h->avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", ref_count[0]-1, max[0], ref_count[1]-1, max[1]);
1269 sl->ref_count[0] = sl->ref_count[1] = 0;
1271 return AVERROR_INVALIDDATA;
1274 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1280 ref_count[0] = ref_count[1] = 0;
1283 if (list_count != sl->list_count ||
1284 ref_count[0] != sl->ref_count[0] ||
1285 ref_count[1] != sl->ref_count[1]) {
1286 sl->ref_count[0] = ref_count[0];
1287 sl->ref_count[1] = ref_count[1];
1288 sl->list_count = list_count;
1295 static const uint8_t start_code[] = { 0x00, 0x00, 0x01 };
1297 static int get_bit_length(H264Context *h, const uint8_t *buf,
1298 const uint8_t *ptr, int dst_length,
1299 int i, int next_avc)
1301 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
1302 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
1303 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
1304 h->workaround_bugs |= FF_BUG_TRUNCATED;
1306 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
1307 while (dst_length > 0 && ptr[dst_length - 1] == 0)
1313 return 8 * dst_length - decode_rbsp_trailing(h, ptr + dst_length - 1);
1316 static int get_last_needed_nal(H264Context *h, const uint8_t *buf, int buf_size)
1318 int next_avc = h->is_avc ? 0 : buf_size;
1321 int nals_needed = 0;
1322 int first_slice = 0;
1327 int dst_length, bit_length, consumed;
1330 if (buf_index >= next_avc) {
1331 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1334 next_avc = buf_index + nalsize;
1336 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1337 if (buf_index >= buf_size)
1339 if (buf_index >= next_avc)
1343 ptr = ff_h264_decode_nal(h, &h->slice_ctx[0], buf + buf_index, &dst_length, &consumed,
1344 next_avc - buf_index);
1346 if (!ptr || dst_length < 0)
1347 return AVERROR_INVALIDDATA;
1349 buf_index += consumed;
1351 bit_length = get_bit_length(h, buf, ptr, dst_length,
1352 buf_index, next_avc);
1355 /* packets can sometimes contain multiple PPS/SPS,
1356 * e.g. two PAFF field pictures in one packet, or a demuxer
1357 * which splits NALs strangely if so, when frame threading we
1358 * can't start the next thread until we've read all of them */
1359 switch (h->nal_unit_type) {
1362 nals_needed = nal_index;
1367 init_get_bits(&gb, ptr, bit_length);
1368 if (!get_ue_golomb(&gb) ||
1370 first_slice != h->nal_unit_type)
1371 nals_needed = nal_index;
1373 first_slice = h->nal_unit_type;
1380 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1381 int parse_extradata)
1383 AVCodecContext *const avctx = h->avctx;
1384 H264SliceContext *sl;
1386 unsigned context_count;
1388 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
1393 h->nal_unit_type= 0;
1395 if(!h->slice_context_count)
1396 h->slice_context_count= 1;
1397 h->max_contexts = h->slice_context_count;
1398 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
1399 h->current_slice = 0;
1400 if (!h->first_field)
1401 h->cur_pic_ptr = NULL;
1402 ff_h264_reset_sei(h);
1405 if (h->nal_length_size == 4) {
1406 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
1408 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
1412 if (avctx->active_thread_type & FF_THREAD_FRAME)
1413 nals_needed = get_last_needed_nal(h, buf, buf_size);
1418 next_avc = h->is_avc ? 0 : buf_size;
1428 if (buf_index >= next_avc) {
1429 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1432 next_avc = buf_index + nalsize;
1434 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1435 if (buf_index >= buf_size)
1437 if (buf_index >= next_avc)
1441 sl = &h->slice_ctx[context_count];
1443 ptr = ff_h264_decode_nal(h, sl, buf + buf_index, &dst_length,
1444 &consumed, next_avc - buf_index);
1445 if (!ptr || dst_length < 0) {
1450 bit_length = get_bit_length(h, buf, ptr, dst_length,
1451 buf_index + consumed, next_avc);
1453 if (h->avctx->debug & FF_DEBUG_STARTCODE)
1454 av_log(h->avctx, AV_LOG_DEBUG,
1455 "NAL %d/%d at %d/%d length %d\n",
1456 h->nal_unit_type, h->nal_ref_idc, buf_index, buf_size, dst_length);
1458 if (h->is_avc && (nalsize != consumed) && nalsize)
1459 av_log(h->avctx, AV_LOG_DEBUG,
1460 "AVC: Consumed only %d bytes instead of %d\n",
1463 buf_index += consumed;
1466 if (avctx->skip_frame >= AVDISCARD_NONREF &&
1467 h->nal_ref_idc == 0 &&
1468 h->nal_unit_type != NAL_SEI)
1472 /* Ignore per frame NAL unit type during extradata
1473 * parsing. Decoding slices is not possible in codec init
1475 if (parse_extradata) {
1476 switch (h->nal_unit_type) {
1482 av_log(h->avctx, AV_LOG_WARNING,
1483 "Ignoring NAL %d in global header/extradata\n",
1485 // fall through to next case
1486 case NAL_AUXILIARY_SLICE:
1487 h->nal_unit_type = NAL_FF_IGNORE;
1493 switch (h->nal_unit_type) {
1495 if ((ptr[0] & 0xFC) == 0x98) {
1496 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
1497 h->next_outputed_poc = INT_MIN;
1501 if (h->nal_unit_type != NAL_IDR_SLICE) {
1502 av_log(h->avctx, AV_LOG_ERROR,
1503 "Invalid mix of idr and non-idr slices\n");
1508 if (h->current_slice && (avctx->active_thread_type & FF_THREAD_SLICE)) {
1509 av_log(h, AV_LOG_ERROR, "invalid mixed IDR / non IDR frames cannot be decoded in slice multithreading mode\n");
1510 ret = AVERROR_INVALIDDATA;
1513 idr(h); // FIXME ensure we don't lose some frames if there is reordering
1516 h->has_recovery_point = 1;
1518 init_get_bits(&sl->gb, ptr, bit_length);
1520 if ( nals_needed >= nal_index
1521 || (!(avctx->active_thread_type & FF_THREAD_FRAME) && !context_count))
1524 if ((err = ff_h264_decode_slice_header(h, sl)))
1527 if (h->sei_recovery_frame_cnt >= 0) {
1528 if (h->frame_num != h->sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
1529 h->valid_recovery_point = 1;
1531 if ( h->recovery_frame < 0
1532 || av_mod_uintp2(h->recovery_frame - h->frame_num, h->sps.log2_max_frame_num) > h->sei_recovery_frame_cnt) {
1533 h->recovery_frame = av_mod_uintp2(h->frame_num + h->sei_recovery_frame_cnt, h->sps.log2_max_frame_num);
1535 if (!h->valid_recovery_point)
1536 h->recovery_frame = h->frame_num;
1540 h->cur_pic_ptr->f->key_frame |=
1541 (h->nal_unit_type == NAL_IDR_SLICE);
1543 if (h->nal_unit_type == NAL_IDR_SLICE ||
1544 h->recovery_frame == h->frame_num) {
1545 h->recovery_frame = -1;
1546 h->cur_pic_ptr->recovered = 1;
1548 // If we have an IDR, all frames after it in decoded order are
1550 if (h->nal_unit_type == NAL_IDR_SLICE)
1551 h->frame_recovered |= FRAME_RECOVERED_IDR;
1553 h->cur_pic_ptr->recovered |= h->frame_recovered;
1555 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1558 if (h->current_slice == 1) {
1559 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS))
1560 decode_postinit(h, nal_index >= nals_needed);
1562 if (h->avctx->hwaccel &&
1563 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
1565 #if FF_API_CAP_VDPAU
1566 if (CONFIG_H264_VDPAU_DECODER &&
1567 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
1568 ff_vdpau_h264_picture_start(h);
1572 if (sl->redundant_pic_count == 0) {
1573 if (avctx->hwaccel) {
1574 ret = avctx->hwaccel->decode_slice(avctx,
1575 &buf[buf_index - consumed],
1579 #if FF_API_CAP_VDPAU
1580 } else if (CONFIG_H264_VDPAU_DECODER &&
1581 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU) {
1582 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f->data[0],
1584 sizeof(start_code));
1585 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f->data[0],
1586 &buf[buf_index - consumed],
1596 avpriv_request_sample(avctx, "data partitioning");
1599 init_get_bits(&h->gb, ptr, bit_length);
1600 ret = ff_h264_decode_sei(h);
1601 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1605 init_get_bits(&h->gb, ptr, bit_length);
1606 if (ff_h264_decode_seq_parameter_set(h, 0) >= 0)
1608 if (h->is_avc ? nalsize : 1) {
1609 av_log(h->avctx, AV_LOG_DEBUG,
1610 "SPS decoding failure, trying again with the complete NAL\n");
1612 av_assert0(next_avc - buf_index + consumed == nalsize);
1613 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
1615 init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
1616 8*(next_avc - buf_index + consumed - 1));
1617 if (ff_h264_decode_seq_parameter_set(h, 0) >= 0)
1620 init_get_bits(&h->gb, ptr, bit_length);
1621 ff_h264_decode_seq_parameter_set(h, 1);
1625 init_get_bits(&h->gb, ptr, bit_length);
1626 ret = ff_h264_decode_picture_parameter_set(h, bit_length);
1627 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1631 case NAL_END_SEQUENCE:
1632 case NAL_END_STREAM:
1633 case NAL_FILLER_DATA:
1635 case NAL_AUXILIARY_SLICE:
1640 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
1641 h->nal_unit_type, bit_length);
1644 if (context_count == h->max_contexts) {
1645 ret = ff_h264_execute_decode_slices(h, context_count);
1646 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1651 if (err < 0 || err == SLICE_SKIPED) {
1653 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
1654 sl->ref_count[0] = sl->ref_count[1] = sl->list_count = 0;
1655 } else if (err == SLICE_SINGLETHREAD) {
1656 if (context_count > 1) {
1657 ret = ff_h264_execute_decode_slices(h, context_count - 1);
1658 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1662 /* Slice could not be decoded in parallel mode, restart. Note
1663 * that rbsp_buffer is not transferred, but since we no longer
1664 * run in parallel mode this should not be an issue. */
1665 sl = &h->slice_ctx[0];
1670 if (context_count) {
1671 ret = ff_h264_execute_decode_slices(h, context_count);
1672 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1679 if (h->cur_pic_ptr && !h->droppable) {
1680 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1681 h->picture_structure == PICT_BOTTOM_FIELD);
1684 return (ret < 0) ? ret : buf_index;
1688 * Return the number of bytes consumed for building the current frame.
1690 static int get_consumed_bytes(int pos, int buf_size)
1693 pos = 1; // avoid infinite loops (I doubt that is needed but...)
1694 if (pos + 10 > buf_size)
1695 pos = buf_size; // oops ;)
1700 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
1702 AVFrame *src = srcp->f;
1703 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(src->format);
1705 int ret = av_frame_ref(dst, src);
1709 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(h), 0);
1711 h->backup_width = h->avctx->width;
1712 h->backup_height = h->avctx->height;
1713 h->backup_pix_fmt = h->avctx->pix_fmt;
1715 h->avctx->width = dst->width;
1716 h->avctx->height = dst->height;
1717 h->avctx->pix_fmt = dst->format;
1719 if (srcp->sei_recovery_frame_cnt == 0)
1724 for (i = 0; i < desc->nb_components; i++) {
1725 int hshift = (i > 0) ? desc->log2_chroma_w : 0;
1726 int vshift = (i > 0) ? desc->log2_chroma_h : 0;
1727 int off = ((srcp->crop_left >> hshift) << h->pixel_shift) +
1728 (srcp->crop_top >> vshift) * dst->linesize[i];
1729 dst->data[i] += off;
1734 static int is_extra(const uint8_t *buf, int buf_size)
1736 int cnt= buf[5]&0x1f;
1737 const uint8_t *p= buf+6;
1739 int nalsize= AV_RB16(p) + 2;
1740 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
1748 int nalsize= AV_RB16(p) + 2;
1749 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
1756 static int h264_decode_frame(AVCodecContext *avctx, void *data,
1757 int *got_frame, AVPacket *avpkt)
1759 const uint8_t *buf = avpkt->data;
1760 int buf_size = avpkt->size;
1761 H264Context *h = avctx->priv_data;
1762 AVFrame *pict = data;
1768 h->flags = avctx->flags;
1769 h->setup_finished = 0;
1771 if (h->backup_width != -1) {
1772 avctx->width = h->backup_width;
1773 h->backup_width = -1;
1775 if (h->backup_height != -1) {
1776 avctx->height = h->backup_height;
1777 h->backup_height = -1;
1779 if (h->backup_pix_fmt != AV_PIX_FMT_NONE) {
1780 avctx->pix_fmt = h->backup_pix_fmt;
1781 h->backup_pix_fmt = AV_PIX_FMT_NONE;
1784 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1786 /* end of stream, output what is still in the buffers */
1787 if (buf_size == 0) {
1790 h->cur_pic_ptr = NULL;
1793 // FIXME factorize this with the output code below
1794 out = h->delayed_pic[0];
1797 h->delayed_pic[i] &&
1798 !h->delayed_pic[i]->f->key_frame &&
1799 !h->delayed_pic[i]->mmco_reset;
1801 if (h->delayed_pic[i]->poc < out->poc) {
1802 out = h->delayed_pic[i];
1806 for (i = out_idx; h->delayed_pic[i]; i++)
1807 h->delayed_pic[i] = h->delayed_pic[i + 1];
1810 out->reference &= ~DELAYED_PIC_REF;
1811 ret = output_frame(h, pict, out);
1819 if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
1821 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
1822 if (is_extra(side, side_size))
1823 ff_h264_decode_extradata(h, side, side_size);
1825 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
1826 if (is_extra(buf, buf_size))
1827 return ff_h264_decode_extradata(h, buf, buf_size);
1830 buf_index = decode_nal_units(h, buf, buf_size, 0);
1832 return AVERROR_INVALIDDATA;
1834 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
1835 av_assert0(buf_index <= buf_size);
1839 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
1840 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1841 buf_size >= 4 && !memcmp("Q264", buf, 4))
1843 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1844 return AVERROR_INVALIDDATA;
1847 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1848 (h->mb_y >= h->mb_height && h->mb_height)) {
1849 if (avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)
1850 decode_postinit(h, 1);
1852 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1855 /* Wait for second field. */
1857 if (h->next_output_pic && ((avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
1858 (avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
1859 h->next_output_pic->recovered)) {
1860 if (!h->next_output_pic->recovered)
1861 h->next_output_pic->f->flags |= AV_FRAME_FLAG_CORRUPT;
1863 if (!h->avctx->hwaccel &&
1864 (h->next_output_pic->field_poc[0] == INT_MAX ||
1865 h->next_output_pic->field_poc[1] == INT_MAX)
1868 AVFrame *f = h->next_output_pic->f;
1869 int field = h->next_output_pic->field_poc[0] == INT_MAX;
1870 uint8_t *dst_data[4];
1872 const uint8_t *src_data[4];
1874 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
1876 for (p = 0; p<4; p++) {
1877 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
1878 src_data[p] = f->data[p] + field *f->linesize[p];
1879 linesizes[p] = 2*f->linesize[p];
1882 av_image_copy(dst_data, linesizes, src_data, linesizes,
1883 f->format, f->width, f->height>>1);
1886 ret = output_frame(h, pict, h->next_output_pic);
1890 if (CONFIG_MPEGVIDEO) {
1891 ff_print_debug_info2(h->avctx, pict, NULL,
1892 h->next_output_pic->mb_type,
1893 h->next_output_pic->qscale_table,
1894 h->next_output_pic->motion_val,
1896 h->mb_width, h->mb_height, h->mb_stride, 1);
1901 av_assert0(pict->buf[0] || !*got_frame);
1903 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1905 return get_consumed_bytes(buf_index, buf_size);
1908 av_cold void ff_h264_free_context(H264Context *h)
1912 ff_h264_free_tables(h);
1914 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
1915 ff_h264_unref_picture(h, &h->DPB[i]);
1916 av_frame_free(&h->DPB[i].f);
1918 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
1920 h->cur_pic_ptr = NULL;
1922 for (i = 0; i < h->nb_slice_ctx; i++)
1923 av_freep(&h->slice_ctx[i].rbsp_buffer);
1924 av_freep(&h->slice_ctx);
1925 h->nb_slice_ctx = 0;
1927 for (i = 0; i < MAX_SPS_COUNT; i++)
1928 av_freep(h->sps_buffers + i);
1930 for (i = 0; i < MAX_PPS_COUNT; i++)
1931 av_freep(h->pps_buffers + i);
1934 static av_cold int h264_decode_end(AVCodecContext *avctx)
1936 H264Context *h = avctx->priv_data;
1938 ff_h264_remove_all_refs(h);
1939 ff_h264_free_context(h);
1941 ff_h264_unref_picture(h, &h->cur_pic);
1942 av_frame_free(&h->cur_pic.f);
1943 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1944 av_frame_free(&h->last_pic_for_ec.f);
1949 #define OFFSET(x) offsetof(H264Context, x)
1950 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1951 static const AVOption h264_options[] = {
1952 {"is_avc", "is avc", offsetof(H264Context, is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0},
1953 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
1954 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1958 static const AVClass h264_class = {
1959 .class_name = "H264 Decoder",
1960 .item_name = av_default_item_name,
1961 .option = h264_options,
1962 .version = LIBAVUTIL_VERSION_INT,
1965 static const AVProfile profiles[] = {
1966 { FF_PROFILE_H264_BASELINE, "Baseline" },
1967 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
1968 { FF_PROFILE_H264_MAIN, "Main" },
1969 { FF_PROFILE_H264_EXTENDED, "Extended" },
1970 { FF_PROFILE_H264_HIGH, "High" },
1971 { FF_PROFILE_H264_HIGH_10, "High 10" },
1972 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
1973 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
1974 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
1975 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
1976 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
1977 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
1978 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
1979 { FF_PROFILE_UNKNOWN },
1982 AVCodec ff_h264_decoder = {
1984 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1985 .type = AVMEDIA_TYPE_VIDEO,
1986 .id = AV_CODEC_ID_H264,
1987 .priv_data_size = sizeof(H264Context),
1988 .init = ff_h264_decode_init,
1989 .close = h264_decode_end,
1990 .decode = h264_decode_frame,
1991 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1992 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1993 AV_CODEC_CAP_FRAME_THREADS,
1994 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
1996 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
1997 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1998 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1999 .priv_class = &h264_class,
2002 #if CONFIG_H264_VDPAU_DECODER && FF_API_VDPAU
2003 static const AVClass h264_vdpau_class = {
2004 .class_name = "H264 VDPAU Decoder",
2005 .item_name = av_default_item_name,
2006 .option = h264_options,
2007 .version = LIBAVUTIL_VERSION_INT,
2010 AVCodec ff_h264_vdpau_decoder = {
2011 .name = "h264_vdpau",
2012 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
2013 .type = AVMEDIA_TYPE_VIDEO,
2014 .id = AV_CODEC_ID_H264,
2015 .priv_data_size = sizeof(H264Context),
2016 .init = ff_h264_decode_init,
2017 .close = h264_decode_end,
2018 .decode = h264_decode_frame,
2019 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HWACCEL_VDPAU,
2021 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
2023 .profiles = NULL_IF_CONFIG_SMALL(profiles),
2024 .priv_class = &h264_vdpau_class,