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 + FF_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, FF_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 tprintf(h->avctx, "rbsp trailing %X\n", v);
327 for (r = 1; r < 9; r++) {
335 void ff_h264_free_tables(H264Context *h, int free_rbsp)
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 if (free_rbsp && h->DPB) {
359 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
360 ff_h264_unref_picture(h, &h->DPB[i]);
363 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
364 h->DPB[i].needs_realloc = 1;
367 h->cur_pic_ptr = NULL;
369 for (i = 0; i < h->nb_slice_ctx; i++) {
370 H264SliceContext *sl = &h->slice_ctx[i];
372 av_freep(&sl->dc_val_base);
373 av_freep(&sl->er.mb_index2xy);
374 av_freep(&sl->er.error_status_table);
375 av_freep(&sl->er.er_temp_buffer);
377 av_freep(&sl->bipred_scratchpad);
378 av_freep(&sl->edge_emu_buffer);
379 av_freep(&sl->top_borders[0]);
380 av_freep(&sl->top_borders[1]);
382 sl->bipred_scratchpad_allocated = 0;
383 sl->edge_emu_buffer_allocated = 0;
384 sl->top_borders_allocated[0] = 0;
385 sl->top_borders_allocated[1] = 0;
388 av_freep(&sl->rbsp_buffer);
389 sl->rbsp_buffer_size = 0;
394 int ff_h264_alloc_tables(H264Context *h)
396 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
397 const int row_mb_num = h->mb_stride * 2 * h->avctx->thread_count;
400 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
401 row_mb_num * 8 * sizeof(uint8_t), fail)
402 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
404 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
405 big_mb_num * 48 * sizeof(uint8_t), fail)
406 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
407 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
408 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
409 big_mb_num * sizeof(uint16_t), fail)
410 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
411 big_mb_num * sizeof(uint8_t), fail)
412 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
413 16 * row_mb_num * sizeof(uint8_t), fail);
414 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
415 16 * row_mb_num * sizeof(uint8_t), fail);
416 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
417 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
419 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
420 4 * big_mb_num * sizeof(uint8_t), fail);
421 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
422 big_mb_num * sizeof(uint8_t), fail)
424 memset(h->slice_table_base, -1,
425 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
426 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
428 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
429 big_mb_num * sizeof(uint32_t), fail);
430 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
431 big_mb_num * sizeof(uint32_t), fail);
432 for (y = 0; y < h->mb_height; y++)
433 for (x = 0; x < h->mb_width; x++) {
434 const int mb_xy = x + y * h->mb_stride;
435 const int b_xy = 4 * x + 4 * y * h->b_stride;
437 h->mb2b_xy[mb_xy] = b_xy;
438 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
441 if (!h->dequant4_coeff[0])
442 h264_init_dequant_tables(h);
445 h->DPB = av_mallocz_array(H264_MAX_PICTURE_COUNT, sizeof(*h->DPB));
448 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
449 av_frame_unref(&h->DPB[i].f);
450 av_frame_unref(&h->cur_pic.f);
456 ff_h264_free_tables(h, 1);
457 return AVERROR(ENOMEM);
462 * Allocate buffers which are not shared amongst multiple threads.
464 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
466 ERContext *er = &sl->er;
467 int mb_array_size = h->mb_height * h->mb_stride;
468 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
469 int c_size = h->mb_stride * (h->mb_height + 1);
470 int yc_size = y_size + 2 * c_size;
473 sl->ref_cache[0][scan8[5] + 1] =
474 sl->ref_cache[0][scan8[7] + 1] =
475 sl->ref_cache[0][scan8[13] + 1] =
476 sl->ref_cache[1][scan8[5] + 1] =
477 sl->ref_cache[1][scan8[7] + 1] =
478 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
480 if (CONFIG_ERROR_RESILIENCE) {
482 er->avctx = h->avctx;
483 er->decode_mb = h264_er_decode_mb;
485 er->quarter_sample = 1;
487 er->mb_num = h->mb_num;
488 er->mb_width = h->mb_width;
489 er->mb_height = h->mb_height;
490 er->mb_stride = h->mb_stride;
491 er->b8_stride = h->mb_width * 2 + 1;
493 // error resilience code looks cleaner with this
494 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
495 (h->mb_num + 1) * sizeof(int), fail);
497 for (y = 0; y < h->mb_height; y++)
498 for (x = 0; x < h->mb_width; x++)
499 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
501 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
502 h->mb_stride + h->mb_width;
504 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
505 mb_array_size * sizeof(uint8_t), fail);
507 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
508 h->mb_height * h->mb_stride, fail);
510 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
511 yc_size * sizeof(int16_t), fail);
512 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
513 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
514 er->dc_val[2] = er->dc_val[1] + c_size;
515 for (i = 0; i < yc_size; i++)
516 sl->dc_val_base[i] = 1024;
522 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
525 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
526 int parse_extradata);
528 int ff_h264_decode_extradata(H264Context *h)
530 AVCodecContext *avctx = h->avctx;
533 if (avctx->extradata[0] == 1) {
535 unsigned char *p = avctx->extradata;
539 if (avctx->extradata_size < 7) {
540 av_log(avctx, AV_LOG_ERROR,
541 "avcC %d too short\n", avctx->extradata_size);
542 return AVERROR_INVALIDDATA;
544 /* sps and pps in the avcC always have length coded with 2 bytes,
545 * so put a fake nal_length_size = 2 while parsing them */
546 h->nal_length_size = 2;
547 // Decode sps from avcC
548 cnt = *(p + 5) & 0x1f; // Number of sps
550 for (i = 0; i < cnt; i++) {
551 nalsize = AV_RB16(p) + 2;
552 if (p - avctx->extradata + nalsize > avctx->extradata_size)
553 return AVERROR_INVALIDDATA;
554 ret = decode_nal_units(h, p, nalsize, 1);
556 av_log(avctx, AV_LOG_ERROR,
557 "Decoding sps %d from avcC failed\n", i);
562 // Decode pps from avcC
563 cnt = *(p++); // Number of pps
564 for (i = 0; i < cnt; i++) {
565 nalsize = AV_RB16(p) + 2;
566 if (p - avctx->extradata + nalsize > avctx->extradata_size)
567 return AVERROR_INVALIDDATA;
568 ret = decode_nal_units(h, p, nalsize, 1);
570 av_log(avctx, AV_LOG_ERROR,
571 "Decoding pps %d from avcC failed\n", i);
576 // Store right nal length size that will be used to parse all other nals
577 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
580 ret = decode_nal_units(h, avctx->extradata, avctx->extradata_size, 1);
587 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
589 H264Context *h = avctx->priv_data;
595 h->dequant_coeff_pps = -1;
596 h->cur_chroma_format_idc = -1;
598 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
599 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
601 h->picture_structure = PICT_FRAME;
602 h->slice_context_count = 1;
603 h->workaround_bugs = avctx->workaround_bugs;
604 h->flags = avctx->flags;
607 // s->decode_mb = ff_h263_decode_mb;
608 if (!avctx->has_b_frames)
611 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
613 ff_h264_decode_init_vlc();
615 ff_init_cabac_states();
617 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? H264_MAX_THREADS : 1;
618 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
621 return AVERROR(ENOMEM);
624 for (i = 0; i < h->nb_slice_ctx; i++)
625 h->slice_ctx[i].h264 = h;
627 h->outputed_poc = h->next_outputed_poc = INT_MIN;
628 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
629 h->last_pocs[i] = INT_MIN;
630 h->prev_poc_msb = 1 << 16;
632 ff_h264_reset_sei(h);
633 h->recovery_frame = -1;
634 h->frame_recovered = 0;
635 if (avctx->codec_id == AV_CODEC_ID_H264) {
636 if (avctx->ticks_per_frame == 1)
637 h->avctx->framerate.num *= 2;
638 avctx->ticks_per_frame = 2;
641 if (avctx->extradata_size > 0 && avctx->extradata) {
642 ret = ff_h264_decode_extradata(h);
644 ff_h264_free_context(h);
649 if (h->sps.bitstream_restriction_flag &&
650 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
651 h->avctx->has_b_frames = h->sps.num_reorder_frames;
655 avctx->internal->allocate_progress = 1;
658 av_log(avctx, AV_LOG_WARNING,
659 "Error resilience is enabled. It is unsafe and unsupported and may crash. "
660 "Use it at your own risk\n");
666 static int decode_init_thread_copy(AVCodecContext *avctx)
668 H264Context *h = avctx->priv_data;
671 if (!avctx->internal->is_copy)
673 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
674 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
676 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? H264_MAX_THREADS : 1;
677 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
680 return AVERROR(ENOMEM);
683 for (i = 0; i < h->nb_slice_ctx; i++)
684 h->slice_ctx[i].h264 = h;
687 h->context_initialized = 0;
693 * Run setup operations that must be run after slice header decoding.
694 * This includes finding the next displayed frame.
696 * @param h h264 master context
697 * @param setup_finished enough NALs have been read that we can call
698 * ff_thread_finish_setup()
700 static void decode_postinit(H264Context *h, int setup_finished)
702 H264Picture *out = h->cur_pic_ptr;
703 H264Picture *cur = h->cur_pic_ptr;
704 int i, pics, out_of_order, out_idx;
705 int invalid = 0, cnt = 0;
707 h->cur_pic_ptr->f.pict_type = h->pict_type;
709 if (h->next_output_pic)
712 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
713 /* FIXME: if we have two PAFF fields in one packet, we can't start
714 * the next thread here. If we have one field per packet, we can.
715 * The check in decode_nal_units() is not good enough to find this
716 * yet, so we assume the worst for now. */
717 // if (setup_finished)
718 // ff_thread_finish_setup(h->avctx);
722 cur->f.interlaced_frame = 0;
723 cur->f.repeat_pict = 0;
725 /* Signal interlacing information externally. */
726 /* Prioritize picture timing SEI information over used
727 * decoding process if it exists. */
729 if (h->sps.pic_struct_present_flag) {
730 switch (h->sei_pic_struct) {
731 case SEI_PIC_STRUCT_FRAME:
733 case SEI_PIC_STRUCT_TOP_FIELD:
734 case SEI_PIC_STRUCT_BOTTOM_FIELD:
735 cur->f.interlaced_frame = 1;
737 case SEI_PIC_STRUCT_TOP_BOTTOM:
738 case SEI_PIC_STRUCT_BOTTOM_TOP:
739 if (FIELD_OR_MBAFF_PICTURE(h))
740 cur->f.interlaced_frame = 1;
742 // try to flag soft telecine progressive
743 cur->f.interlaced_frame = h->prev_interlaced_frame;
745 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
746 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
747 /* Signal the possibility of telecined film externally
748 * (pic_struct 5,6). From these hints, let the applications
749 * decide if they apply deinterlacing. */
750 cur->f.repeat_pict = 1;
752 case SEI_PIC_STRUCT_FRAME_DOUBLING:
753 cur->f.repeat_pict = 2;
755 case SEI_PIC_STRUCT_FRAME_TRIPLING:
756 cur->f.repeat_pict = 4;
760 if ((h->sei_ct_type & 3) &&
761 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
762 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
764 /* Derive interlacing flag from used decoding process. */
765 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
767 h->prev_interlaced_frame = cur->f.interlaced_frame;
769 if (cur->field_poc[0] != cur->field_poc[1]) {
770 /* Derive top_field_first from field pocs. */
771 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
773 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
774 /* Use picture timing SEI information. Even if it is a
775 * information of a past frame, better than nothing. */
776 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
777 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
778 cur->f.top_field_first = 1;
780 cur->f.top_field_first = 0;
782 /* Most likely progressive */
783 cur->f.top_field_first = 0;
787 if (h->sei_frame_packing_present &&
788 h->frame_packing_arrangement_type >= 0 &&
789 h->frame_packing_arrangement_type <= 6 &&
790 h->content_interpretation_type > 0 &&
791 h->content_interpretation_type < 3) {
792 AVStereo3D *stereo = av_stereo3d_create_side_data(&cur->f);
796 switch (h->frame_packing_arrangement_type) {
798 stereo->type = AV_STEREO3D_CHECKERBOARD;
801 stereo->type = AV_STEREO3D_COLUMNS;
804 stereo->type = AV_STEREO3D_LINES;
807 if (h->quincunx_subsampling)
808 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
810 stereo->type = AV_STEREO3D_SIDEBYSIDE;
813 stereo->type = AV_STEREO3D_TOPBOTTOM;
816 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
819 stereo->type = AV_STEREO3D_2D;
823 if (h->content_interpretation_type == 2)
824 stereo->flags = AV_STEREO3D_FLAG_INVERT;
827 if (h->sei_display_orientation_present &&
828 (h->sei_anticlockwise_rotation || h->sei_hflip || h->sei_vflip)) {
829 double angle = h->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
830 AVFrameSideData *rotation = av_frame_new_side_data(&cur->f,
831 AV_FRAME_DATA_DISPLAYMATRIX,
832 sizeof(int32_t) * 9);
836 av_display_rotation_set((int32_t *)rotation->data, angle);
837 av_display_matrix_flip((int32_t *)rotation->data,
838 h->sei_hflip, h->sei_vflip);
841 // FIXME do something with unavailable reference frames
843 /* Sort B-frames into display order */
845 if (h->sps.bitstream_restriction_flag &&
846 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
847 h->avctx->has_b_frames = h->sps.num_reorder_frames;
851 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
852 !h->sps.bitstream_restriction_flag) {
853 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
858 while (h->delayed_pic[pics])
861 assert(pics <= MAX_DELAYED_PIC_COUNT);
863 h->delayed_pic[pics++] = cur;
864 if (cur->reference == 0)
865 cur->reference = DELAYED_PIC_REF;
867 /* Frame reordering. This code takes pictures from coding order and sorts
868 * them by their incremental POC value into display order. It supports POC
869 * gaps, MMCO reset codes and random resets.
870 * A "display group" can start either with a IDR frame (f.key_frame = 1),
871 * and/or can be closed down with a MMCO reset code. In sequences where
872 * there is no delay, we can't detect that (since the frame was already
873 * output to the user), so we also set h->mmco_reset to detect the MMCO
875 * FIXME: if we detect insufficient delays (as per h->avctx->has_b_frames),
876 * we increase the delay between input and output. All frames affected by
877 * the lag (e.g. those that should have been output before another frame
878 * that we already returned to the user) will be dropped. This is a bug
879 * that we will fix later. */
880 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
881 cnt += out->poc < h->last_pocs[i];
882 invalid += out->poc == INT_MIN;
884 if (!h->mmco_reset && !cur->f.key_frame &&
885 cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
888 h->delayed_pic[pics - 2]->mmco_reset = 2;
890 if (h->mmco_reset || cur->f.key_frame) {
891 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
892 h->last_pocs[i] = INT_MIN;
894 invalid = MAX_DELAYED_PIC_COUNT;
896 out = h->delayed_pic[0];
898 for (i = 1; i < MAX_DELAYED_PIC_COUNT &&
900 !h->delayed_pic[i - 1]->mmco_reset &&
901 !h->delayed_pic[i]->f.key_frame;
903 if (h->delayed_pic[i]->poc < out->poc) {
904 out = h->delayed_pic[i];
907 if (h->avctx->has_b_frames == 0 &&
908 (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
909 h->next_outputed_poc = INT_MIN;
910 out_of_order = !out->f.key_frame && !h->mmco_reset &&
911 (out->poc < h->next_outputed_poc);
913 if (h->sps.bitstream_restriction_flag &&
914 h->avctx->has_b_frames >= h->sps.num_reorder_frames) {
915 } else if (out_of_order && pics - 1 == h->avctx->has_b_frames &&
916 h->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
917 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
918 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, cnt);
921 } else if (h->low_delay &&
922 ((h->next_outputed_poc != INT_MIN &&
923 out->poc > h->next_outputed_poc + 2) ||
924 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
926 h->avctx->has_b_frames++;
929 if (pics > h->avctx->has_b_frames) {
930 out->reference &= ~DELAYED_PIC_REF;
931 // for frame threading, the owner must be the second field's thread or
932 // else the first thread can release the picture and reuse it unsafely
933 for (i = out_idx; h->delayed_pic[i]; i++)
934 h->delayed_pic[i] = h->delayed_pic[i + 1];
936 memmove(h->last_pocs, &h->last_pocs[1],
937 sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
938 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
939 if (!out_of_order && pics > h->avctx->has_b_frames) {
940 h->next_output_pic = out;
941 if (out->mmco_reset) {
943 h->next_outputed_poc = out->poc;
944 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
946 h->next_outputed_poc = INT_MIN;
949 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
950 h->next_outputed_poc = INT_MIN;
952 h->next_outputed_poc = out->poc;
957 av_log(h->avctx, AV_LOG_DEBUG, "no picture\n");
960 if (h->next_output_pic) {
961 if (h->next_output_pic->recovered) {
962 // We have reached an recovery point and all frames after it in
963 // display order are "recovered".
964 h->frame_recovered |= FRAME_RECOVERED_SEI;
966 h->next_output_pic->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
969 if (setup_finished && !h->avctx->hwaccel)
970 ff_thread_finish_setup(h->avctx);
973 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl)
976 int luma_def, chroma_def;
979 sl->use_weight_chroma = 0;
980 sl->luma_log2_weight_denom = get_ue_golomb(&sl->gb);
981 if (h->sps.chroma_format_idc)
982 sl->chroma_log2_weight_denom = get_ue_golomb(&sl->gb);
983 luma_def = 1 << sl->luma_log2_weight_denom;
984 chroma_def = 1 << sl->chroma_log2_weight_denom;
986 for (list = 0; list < 2; list++) {
987 sl->luma_weight_flag[list] = 0;
988 sl->chroma_weight_flag[list] = 0;
989 for (i = 0; i < sl->ref_count[list]; i++) {
990 int luma_weight_flag, chroma_weight_flag;
992 luma_weight_flag = get_bits1(&sl->gb);
993 if (luma_weight_flag) {
994 sl->luma_weight[i][list][0] = get_se_golomb(&sl->gb);
995 sl->luma_weight[i][list][1] = get_se_golomb(&sl->gb);
996 if (sl->luma_weight[i][list][0] != luma_def ||
997 sl->luma_weight[i][list][1] != 0) {
999 sl->luma_weight_flag[list] = 1;
1002 sl->luma_weight[i][list][0] = luma_def;
1003 sl->luma_weight[i][list][1] = 0;
1006 if (h->sps.chroma_format_idc) {
1007 chroma_weight_flag = get_bits1(&sl->gb);
1008 if (chroma_weight_flag) {
1010 for (j = 0; j < 2; j++) {
1011 sl->chroma_weight[i][list][j][0] = get_se_golomb(&sl->gb);
1012 sl->chroma_weight[i][list][j][1] = get_se_golomb(&sl->gb);
1013 if (sl->chroma_weight[i][list][j][0] != chroma_def ||
1014 sl->chroma_weight[i][list][j][1] != 0) {
1015 sl->use_weight_chroma = 1;
1016 sl->chroma_weight_flag[list] = 1;
1021 for (j = 0; j < 2; j++) {
1022 sl->chroma_weight[i][list][j][0] = chroma_def;
1023 sl->chroma_weight[i][list][j][1] = 0;
1028 if (sl->slice_type_nos != AV_PICTURE_TYPE_B)
1031 sl->use_weight = sl->use_weight || sl->use_weight_chroma;
1036 * instantaneous decoder refresh.
1038 static void idr(H264Context *h)
1040 ff_h264_remove_all_refs(h);
1042 h->prev_frame_num_offset =
1044 h->prev_poc_lsb = 0;
1047 /* forget old pics after a seek */
1048 void ff_h264_flush_change(H264Context *h)
1051 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1052 h->last_pocs[i] = INT_MIN;
1053 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1054 h->prev_interlaced_frame = 1;
1057 h->cur_pic_ptr->reference = 0;
1059 ff_h264_reset_sei(h);
1060 h->recovery_frame = -1;
1061 h->frame_recovered = 0;
1064 /* forget old pics after a seek */
1065 static void flush_dpb(AVCodecContext *avctx)
1067 H264Context *h = avctx->priv_data;
1070 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
1072 ff_h264_flush_change(h);
1075 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
1076 ff_h264_unref_picture(h, &h->DPB[i]);
1077 h->cur_pic_ptr = NULL;
1078 ff_h264_unref_picture(h, &h->cur_pic);
1082 ff_h264_free_tables(h, 1);
1083 h->context_initialized = 0;
1086 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
1088 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
1091 h->frame_num_offset = h->prev_frame_num_offset;
1092 if (h->frame_num < h->prev_frame_num)
1093 h->frame_num_offset += max_frame_num;
1095 if (h->sps.poc_type == 0) {
1096 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
1098 if (h->poc_lsb < h->prev_poc_lsb &&
1099 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
1100 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1101 else if (h->poc_lsb > h->prev_poc_lsb &&
1102 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
1103 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1105 h->poc_msb = h->prev_poc_msb;
1107 field_poc[1] = h->poc_msb + h->poc_lsb;
1108 if (h->picture_structure == PICT_FRAME)
1109 field_poc[1] += h->delta_poc_bottom;
1110 } else if (h->sps.poc_type == 1) {
1111 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1114 if (h->sps.poc_cycle_length != 0)
1115 abs_frame_num = h->frame_num_offset + h->frame_num;
1119 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
1122 expected_delta_per_poc_cycle = 0;
1123 for (i = 0; i < h->sps.poc_cycle_length; i++)
1124 // FIXME integrate during sps parse
1125 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
1127 if (abs_frame_num > 0) {
1128 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1129 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1131 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1132 for (i = 0; i <= frame_num_in_poc_cycle; i++)
1133 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
1137 if (h->nal_ref_idc == 0)
1138 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1140 field_poc[0] = expectedpoc + h->delta_poc[0];
1141 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1143 if (h->picture_structure == PICT_FRAME)
1144 field_poc[1] += h->delta_poc[1];
1146 int poc = 2 * (h->frame_num_offset + h->frame_num);
1148 if (!h->nal_ref_idc)
1155 if (h->picture_structure != PICT_BOTTOM_FIELD)
1156 pic_field_poc[0] = field_poc[0];
1157 if (h->picture_structure != PICT_TOP_FIELD)
1158 pic_field_poc[1] = field_poc[1];
1159 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
1165 * Compute profile from profile_idc and constraint_set?_flags.
1169 * @return profile as defined by FF_PROFILE_H264_*
1171 int ff_h264_get_profile(SPS *sps)
1173 int profile = sps->profile_idc;
1175 switch (sps->profile_idc) {
1176 case FF_PROFILE_H264_BASELINE:
1177 // constraint_set1_flag set to 1
1178 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1180 case FF_PROFILE_H264_HIGH_10:
1181 case FF_PROFILE_H264_HIGH_422:
1182 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1183 // constraint_set3_flag set to 1
1184 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
1191 int ff_h264_set_parameter_from_sps(H264Context *h)
1193 if (h->flags & CODEC_FLAG_LOW_DELAY ||
1194 (h->sps.bitstream_restriction_flag &&
1195 !h->sps.num_reorder_frames)) {
1196 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1197 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1198 "Reenabling low delay requires a codec flush.\n");
1203 if (h->avctx->has_b_frames < 2)
1204 h->avctx->has_b_frames = !h->low_delay;
1206 if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1207 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
1208 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
1209 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1210 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
1211 h->pixel_shift = h->sps.bit_depth_luma > 8;
1213 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
1214 h->sps.chroma_format_idc);
1215 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1216 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
1217 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
1218 h->sps.chroma_format_idc);
1219 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
1221 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
1222 h->sps.bit_depth_luma);
1223 return AVERROR_INVALIDDATA;
1229 int ff_set_ref_count(H264Context *h, H264SliceContext *sl)
1231 int ref_count[2], list_count;
1232 int num_ref_idx_active_override_flag, max_refs;
1234 // set defaults, might be overridden a few lines later
1235 ref_count[0] = h->pps.ref_count[0];
1236 ref_count[1] = h->pps.ref_count[1];
1238 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1239 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1240 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1241 num_ref_idx_active_override_flag = get_bits1(&sl->gb);
1243 if (num_ref_idx_active_override_flag) {
1244 ref_count[0] = get_ue_golomb(&sl->gb) + 1;
1245 if (ref_count[0] < 1)
1246 return AVERROR_INVALIDDATA;
1247 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1248 ref_count[1] = get_ue_golomb(&sl->gb) + 1;
1249 if (ref_count[1] < 1)
1250 return AVERROR_INVALIDDATA;
1254 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1260 ref_count[0] = ref_count[1] = 0;
1263 max_refs = h->picture_structure == PICT_FRAME ? 16 : 32;
1265 if (ref_count[0] > max_refs || ref_count[1] > max_refs) {
1266 av_log(h->avctx, AV_LOG_ERROR, "reference overflow\n");
1267 sl->ref_count[0] = sl->ref_count[1] = 0;
1268 return AVERROR_INVALIDDATA;
1271 if (list_count != sl->list_count ||
1272 ref_count[0] != sl->ref_count[0] ||
1273 ref_count[1] != sl->ref_count[1]) {
1274 sl->ref_count[0] = ref_count[0];
1275 sl->ref_count[1] = ref_count[1];
1276 sl->list_count = list_count;
1283 static int find_start_code(const uint8_t *buf, int buf_size,
1284 int buf_index, int next_avc)
1286 // start code prefix search
1287 for (; buf_index + 3 < next_avc; buf_index++)
1288 // This should always succeed in the first iteration.
1289 if (buf[buf_index] == 0 &&
1290 buf[buf_index + 1] == 0 &&
1291 buf[buf_index + 2] == 1)
1294 if (buf_index + 3 >= buf_size)
1297 return buf_index + 3;
1300 static int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1301 int buf_size, int *buf_index)
1305 if (*buf_index >= buf_size - h->nal_length_size)
1308 for (i = 0; i < h->nal_length_size; i++)
1309 nalsize = (nalsize << 8) | buf[(*buf_index)++];
1310 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1311 av_log(h->avctx, AV_LOG_ERROR,
1312 "AVC: nal size %d\n", nalsize);
1318 static int get_bit_length(H264Context *h, const uint8_t *buf,
1319 const uint8_t *ptr, int dst_length,
1320 int i, int next_avc)
1322 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
1323 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
1324 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
1325 h->workaround_bugs |= FF_BUG_TRUNCATED;
1327 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
1328 while (dst_length > 0 && ptr[dst_length - 1] == 0)
1334 return 8 * dst_length - decode_rbsp_trailing(h, ptr + dst_length - 1);
1337 static int get_last_needed_nal(H264Context *h, const uint8_t *buf, int buf_size)
1339 int next_avc = h->is_avc ? 0 : buf_size;
1342 int nals_needed = 0;
1347 int dst_length, bit_length, consumed;
1350 if (buf_index >= next_avc) {
1351 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1354 next_avc = buf_index + nalsize;
1356 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1357 if (buf_index >= buf_size)
1361 ptr = ff_h264_decode_nal(h, &h->slice_ctx[0], buf + buf_index, &dst_length, &consumed,
1362 next_avc - buf_index);
1364 if (!ptr || dst_length < 0)
1365 return AVERROR_INVALIDDATA;
1367 buf_index += consumed;
1369 bit_length = get_bit_length(h, buf, ptr, dst_length,
1370 buf_index, next_avc);
1373 /* packets can sometimes contain multiple PPS/SPS,
1374 * e.g. two PAFF field pictures in one packet, or a demuxer
1375 * which splits NALs strangely if so, when frame threading we
1376 * can't start the next thread until we've read all of them */
1377 switch (h->nal_unit_type) {
1380 nals_needed = nal_index;
1385 init_get_bits(&gb, ptr, bit_length);
1386 if (!get_ue_golomb(&gb))
1387 nals_needed = nal_index;
1394 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1395 int parse_extradata)
1397 AVCodecContext *const avctx = h->avctx;
1398 H264SliceContext *sl;
1400 unsigned context_count;
1402 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
1406 h->max_contexts = h->slice_context_count;
1407 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) {
1408 h->current_slice = 0;
1409 if (!h->first_field)
1410 h->cur_pic_ptr = NULL;
1411 ff_h264_reset_sei(h);
1414 if (avctx->active_thread_type & FF_THREAD_FRAME)
1415 nals_needed = get_last_needed_nal(h, buf, buf_size);
1420 next_avc = h->is_avc ? 0 : buf_size;
1430 if (buf_index >= next_avc) {
1431 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1434 next_avc = buf_index + nalsize;
1436 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1437 if (buf_index >= buf_size)
1439 if (buf_index >= next_avc)
1443 sl = &h->slice_ctx[context_count];
1445 ptr = ff_h264_decode_nal(h, sl, buf + buf_index, &dst_length,
1446 &consumed, next_avc - buf_index);
1447 if (!ptr || dst_length < 0) {
1452 bit_length = get_bit_length(h, buf, ptr, dst_length,
1453 buf_index + consumed, next_avc);
1455 if (h->avctx->debug & FF_DEBUG_STARTCODE)
1456 av_log(h->avctx, AV_LOG_DEBUG,
1457 "NAL %d at %d/%d length %d\n",
1458 h->nal_unit_type, buf_index, buf_size, dst_length);
1460 if (h->is_avc && (nalsize != consumed) && nalsize)
1461 av_log(h->avctx, AV_LOG_DEBUG,
1462 "AVC: Consumed only %d bytes instead of %d\n",
1465 buf_index += consumed;
1468 if (avctx->skip_frame >= AVDISCARD_NONREF &&
1469 h->nal_ref_idc == 0 &&
1470 h->nal_unit_type != NAL_SEI)
1474 /* Ignore every NAL unit type except PPS and SPS during extradata
1475 * parsing. Decoding slices is not possible in codec init
1477 if (parse_extradata && HAVE_THREADS &&
1478 (h->avctx->active_thread_type & FF_THREAD_FRAME) &&
1479 (h->nal_unit_type != NAL_PPS &&
1480 h->nal_unit_type != NAL_SPS)) {
1481 if (h->nal_unit_type < NAL_AUD ||
1482 h->nal_unit_type > NAL_AUXILIARY_SLICE)
1483 av_log(avctx, AV_LOG_INFO,
1484 "Ignoring NAL unit %d during extradata parsing\n",
1486 h->nal_unit_type = NAL_FF_IGNORE;
1489 switch (h->nal_unit_type) {
1491 if (h->nal_unit_type != NAL_IDR_SLICE) {
1492 av_log(h->avctx, AV_LOG_ERROR,
1493 "Invalid mix of idr and non-idr slices\n");
1497 idr(h); // FIXME ensure we don't lose some frames if there is reordering
1499 init_get_bits(&sl->gb, ptr, bit_length);
1501 if ((err = ff_h264_decode_slice_header(h, sl)))
1504 if (h->sei_recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
1505 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
1506 ((1 << h->sps.log2_max_frame_num) - 1);
1509 h->cur_pic_ptr->f.key_frame |=
1510 (h->nal_unit_type == NAL_IDR_SLICE) ||
1511 (h->sei_recovery_frame_cnt >= 0);
1513 if (h->nal_unit_type == NAL_IDR_SLICE ||
1514 h->recovery_frame == h->frame_num) {
1515 h->recovery_frame = -1;
1516 h->cur_pic_ptr->recovered = 1;
1518 // If we have an IDR, all frames after it in decoded order are
1520 if (h->nal_unit_type == NAL_IDR_SLICE)
1521 h->frame_recovered |= FRAME_RECOVERED_IDR;
1522 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1524 if (h->current_slice == 1) {
1525 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
1526 decode_postinit(h, nal_index >= nals_needed);
1528 if (h->avctx->hwaccel &&
1529 (ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
1533 if (sl->redundant_pic_count == 0 &&
1534 (avctx->skip_frame < AVDISCARD_NONREF ||
1536 (avctx->skip_frame < AVDISCARD_BIDIR ||
1537 sl->slice_type_nos != AV_PICTURE_TYPE_B) &&
1538 (avctx->skip_frame < AVDISCARD_NONKEY ||
1539 sl->slice_type_nos == AV_PICTURE_TYPE_I) &&
1540 avctx->skip_frame < AVDISCARD_ALL) {
1541 if (avctx->hwaccel) {
1542 ret = avctx->hwaccel->decode_slice(avctx,
1543 &buf[buf_index - consumed],
1554 avpriv_request_sample(avctx, "data partitioning");
1555 ret = AVERROR(ENOSYS);
1559 init_get_bits(&h->gb, ptr, bit_length);
1560 ret = ff_h264_decode_sei(h);
1561 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1565 init_get_bits(&h->gb, ptr, bit_length);
1566 ret = ff_h264_decode_seq_parameter_set(h);
1567 if (ret < 0 && h->is_avc && (nalsize != consumed) && nalsize) {
1568 av_log(h->avctx, AV_LOG_DEBUG,
1569 "SPS decoding failure, trying again with the complete NAL\n");
1570 init_get_bits(&h->gb, buf + buf_index + 1 - consumed,
1572 ff_h264_decode_seq_parameter_set(h);
1575 ret = ff_h264_set_parameter_from_sps(h);
1581 init_get_bits(&h->gb, ptr, bit_length);
1582 ret = ff_h264_decode_picture_parameter_set(h, bit_length);
1583 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1587 case NAL_END_SEQUENCE:
1588 case NAL_END_STREAM:
1589 case NAL_FILLER_DATA:
1591 case NAL_AUXILIARY_SLICE:
1596 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
1597 h->nal_unit_type, bit_length);
1600 if (context_count == h->max_contexts) {
1601 ret = ff_h264_execute_decode_slices(h, context_count);
1602 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1608 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
1609 sl->ref_count[0] = sl->ref_count[1] = sl->list_count = 0;
1610 } else if (err == 1) {
1611 /* Slice could not be decoded in parallel mode, restart. Note
1612 * that rbsp_buffer is not transferred, but since we no longer
1613 * run in parallel mode this should not be an issue. */
1614 sl = &h->slice_ctx[0];
1619 if (context_count) {
1620 ret = ff_h264_execute_decode_slices(h, context_count);
1621 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1628 if (h->cur_pic_ptr && !h->droppable) {
1629 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1630 h->picture_structure == PICT_BOTTOM_FIELD);
1633 return (ret < 0) ? ret : buf_index;
1637 * Return the number of bytes consumed for building the current frame.
1639 static int get_consumed_bytes(int pos, int buf_size)
1642 pos = 1; // avoid infinite loops (I doubt that is needed but...)
1643 if (pos + 10 > buf_size)
1644 pos = buf_size; // oops ;)
1649 static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src)
1652 int ret = av_frame_ref(dst, src);
1659 for (i = 0; i < 3; i++) {
1660 int hshift = (i > 0) ? h->chroma_x_shift : 0;
1661 int vshift = (i > 0) ? h->chroma_y_shift : 0;
1662 int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) +
1663 (h->sps.crop_top >> vshift) * dst->linesize[i];
1664 dst->data[i] += off;
1669 static int h264_decode_frame(AVCodecContext *avctx, void *data,
1670 int *got_frame, AVPacket *avpkt)
1672 const uint8_t *buf = avpkt->data;
1673 int buf_size = avpkt->size;
1674 H264Context *h = avctx->priv_data;
1675 AVFrame *pict = data;
1679 h->flags = avctx->flags;
1681 /* end of stream, output what is still in the buffers */
1683 if (buf_size == 0) {
1687 h->cur_pic_ptr = NULL;
1689 // FIXME factorize this with the output code below
1690 out = h->delayed_pic[0];
1693 h->delayed_pic[i] &&
1694 !h->delayed_pic[i]->f.key_frame &&
1695 !h->delayed_pic[i]->mmco_reset;
1697 if (h->delayed_pic[i]->poc < out->poc) {
1698 out = h->delayed_pic[i];
1702 for (i = out_idx; h->delayed_pic[i]; i++)
1703 h->delayed_pic[i] = h->delayed_pic[i + 1];
1706 ret = output_frame(h, pict, &out->f);
1715 buf_index = decode_nal_units(h, buf, buf_size, 0);
1717 return AVERROR_INVALIDDATA;
1719 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
1724 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
1725 if (avctx->skip_frame >= AVDISCARD_NONREF)
1727 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1728 return AVERROR_INVALIDDATA;
1731 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
1732 (h->mb_y >= h->mb_height && h->mb_height)) {
1733 if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
1734 decode_postinit(h, 1);
1736 ff_h264_field_end(h, &h->slice_ctx[0], 0);
1739 if (h->next_output_pic && ((avctx->flags & CODEC_FLAG_OUTPUT_CORRUPT) ||
1740 h->next_output_pic->recovered)) {
1741 if (!h->next_output_pic->recovered)
1742 h->next_output_pic->f.flags |= AV_FRAME_FLAG_CORRUPT;
1744 ret = output_frame(h, pict, &h->next_output_pic->f);
1751 assert(pict->buf[0] || !*got_frame);
1753 return get_consumed_bytes(buf_index, buf_size);
1756 av_cold void ff_h264_free_context(H264Context *h)
1760 ff_h264_free_tables(h, 1); // FIXME cleanup init stuff perhaps
1762 av_freep(&h->slice_ctx);
1763 h->nb_slice_ctx = 0;
1765 for (i = 0; i < MAX_SPS_COUNT; i++)
1766 av_freep(h->sps_buffers + i);
1768 for (i = 0; i < MAX_PPS_COUNT; i++)
1769 av_freep(h->pps_buffers + i);
1772 static av_cold int h264_decode_end(AVCodecContext *avctx)
1774 H264Context *h = avctx->priv_data;
1776 ff_h264_free_context(h);
1778 ff_h264_unref_picture(h, &h->cur_pic);
1783 #define OFFSET(x) offsetof(H264Context, x)
1784 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1785 static const AVOption h264_options[] = {
1786 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VD },
1790 static const AVClass h264_class = {
1791 .class_name = "h264",
1792 .item_name = av_default_item_name,
1793 .option = h264_options,
1794 .version = LIBAVUTIL_VERSION_INT,
1797 static const AVProfile profiles[] = {
1798 { FF_PROFILE_H264_BASELINE, "Baseline" },
1799 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
1800 { FF_PROFILE_H264_MAIN, "Main" },
1801 { FF_PROFILE_H264_EXTENDED, "Extended" },
1802 { FF_PROFILE_H264_HIGH, "High" },
1803 { FF_PROFILE_H264_HIGH_10, "High 10" },
1804 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
1805 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
1806 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
1807 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
1808 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
1809 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
1810 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
1811 { FF_PROFILE_UNKNOWN },
1814 AVCodec ff_h264_decoder = {
1816 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1817 .type = AVMEDIA_TYPE_VIDEO,
1818 .id = AV_CODEC_ID_H264,
1819 .priv_data_size = sizeof(H264Context),
1820 .init = ff_h264_decode_init,
1821 .close = h264_decode_end,
1822 .decode = h264_decode_frame,
1823 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
1824 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
1825 CODEC_CAP_FRAME_THREADS,
1827 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
1828 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1829 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1830 .priv_class = &h264_class,