2 * H.26L/H.264/AVC/JVT/14496-10/... parser
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 / MPEG-4 part10 parser.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #define UNCHECKED_BITSTREAM_READER 1
33 #include "libavutil/avutil.h"
34 #include "libavutil/error.h"
35 #include "libavutil/log.h"
36 #include "libavutil/mem.h"
37 #include "libavutil/pixfmt.h"
47 #include "mpegutils.h"
50 typedef struct H264ParseContext {
53 H264DSPContext h264dsp;
59 int picture_structure;
60 uint8_t parse_history[6];
61 int parse_history_count;
63 int64_t reference_dts;
64 int last_frame_num, last_picture_structure;
68 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
69 int buf_size, void *logctx)
73 ParseContext *pc = &p->pc;
75 int next_avc = p->is_avc ? 0 : buf_size;
76 // mb_addr= pc->mb_addr - 1;
81 if (p->is_avc && !p->nal_length_size)
82 av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
84 for (i = 0; i < buf_size; i++) {
88 for (j = 0; j < p->nal_length_size; j++)
89 nalsize = (nalsize << 8) | buf[i++];
90 if (nalsize <= 0 || nalsize > buf_size - i) {
91 av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
94 next_avc = i + nalsize;
99 i += p->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
102 } else if (state <= 2) {
104 state ^= 5; // 2->7, 1->4, 0->5
108 state >>= 1; // 2->1, 1->0, 0->0
109 } else if (state <= 5) {
110 int nalu_type = buf[i] & 0x1F;
111 if (nalu_type == H264_NAL_SEI || nalu_type == H264_NAL_SPS ||
112 nalu_type == H264_NAL_PPS || nalu_type == H264_NAL_AUD) {
113 if (pc->frame_start_found) {
117 } else if (nalu_type == H264_NAL_SLICE || nalu_type == H264_NAL_DPA ||
118 nalu_type == H264_NAL_IDR_SLICE) {
124 p->parse_history[p->parse_history_count++] = buf[i];
125 if (p->parse_history_count > 5) {
126 unsigned int mb, last_mb = p->parse_last_mb;
129 init_get_bits(&gb, p->parse_history, 8*p->parse_history_count);
130 p->parse_history_count = 0;
131 mb= get_ue_golomb_long(&gb);
132 p->parse_last_mb = mb;
133 if (pc->frame_start_found) {
137 pc->frame_start_found = 1;
145 return END_NOT_FOUND;
149 pc->frame_start_found = 0;
152 return i - (state & 5) - 5 * (state > 7);
155 static int scan_mmco_reset(AVCodecParserContext *s, GetBitContext *gb,
158 H264PredWeightTable pwt;
159 int slice_type_nos = s->pict_type & 3;
160 H264ParseContext *p = s->priv_data;
161 int list_count, ref_count[2];
164 if (p->ps.pps->redundant_pic_cnt_present)
165 get_ue_golomb(gb); // redundant_pic_count
167 if (slice_type_nos == AV_PICTURE_TYPE_B)
168 get_bits1(gb); // direct_spatial_mv_pred
170 if (ff_h264_parse_ref_count(&list_count, ref_count, gb, p->ps.pps,
171 slice_type_nos, p->picture_structure, logctx) < 0)
172 return AVERROR_INVALIDDATA;
174 if (slice_type_nos != AV_PICTURE_TYPE_I) {
176 for (list = 0; list < list_count; list++) {
179 for (index = 0; ; index++) {
180 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(gb);
182 if (reordering_of_pic_nums_idc < 3)
183 get_ue_golomb_long(gb);
184 else if (reordering_of_pic_nums_idc > 3) {
185 av_log(logctx, AV_LOG_ERROR,
186 "illegal reordering_of_pic_nums_idc %d\n",
187 reordering_of_pic_nums_idc);
188 return AVERROR_INVALIDDATA;
192 if (index >= ref_count[list]) {
193 av_log(logctx, AV_LOG_ERROR,
194 "reference count %d overflow\n", index);
195 return AVERROR_INVALIDDATA;
202 if ((p->ps.pps->weighted_pred && slice_type_nos == AV_PICTURE_TYPE_P) ||
203 (p->ps.pps->weighted_bipred_idc == 1 && slice_type_nos == AV_PICTURE_TYPE_B))
204 ff_h264_pred_weight_table(gb, p->ps.sps, ref_count, slice_type_nos,
205 &pwt, p->picture_structure, logctx);
207 if (get_bits1(gb)) { // adaptive_ref_pic_marking_mode_flag
209 for (i = 0; i < MAX_MMCO_COUNT; i++) {
210 MMCOOpcode opcode = get_ue_golomb_31(gb);
211 if (opcode > (unsigned) MMCO_LONG) {
212 av_log(logctx, AV_LOG_ERROR,
213 "illegal memory management control operation %d\n",
215 return AVERROR_INVALIDDATA;
217 if (opcode == MMCO_END)
219 else if (opcode == MMCO_RESET)
222 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
223 get_ue_golomb_long(gb); // difference_of_pic_nums_minus1
224 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
225 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
226 get_ue_golomb_31(gb);
234 * Parse NAL units of found picture and decode some basic information.
236 * @param s parser context.
237 * @param avctx codec context.
238 * @param buf buffer with field/frame data.
239 * @param buf_size size of the buffer.
241 static inline int parse_nal_units(AVCodecParserContext *s,
242 AVCodecContext *avctx,
243 const uint8_t * const buf, int buf_size)
245 H264ParseContext *p = s->priv_data;
246 H2645RBSP rbsp = { NULL };
247 H2645NAL nal = { NULL };
248 int buf_index, next_avc;
250 unsigned int slice_type;
251 int state = -1, got_reset = 0;
252 int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
256 /* set some sane default values */
257 s->pict_type = AV_PICTURE_TYPE_I;
259 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
261 ff_h264_sei_uninit(&p->sei);
262 p->sei.frame_packing.frame_packing_arrangement_cancel_flag = -1;
267 av_fast_padded_malloc(&rbsp.rbsp_buffer, &rbsp.rbsp_buffer_alloc_size, buf_size);
268 if (!rbsp.rbsp_buffer)
269 return AVERROR(ENOMEM);
272 next_avc = p->is_avc ? 0 : buf_size;
275 int src_length, consumed, nalsize = 0;
277 if (buf_index >= next_avc) {
278 nalsize = get_nalsize(p->nal_length_size, buf, buf_size, &buf_index, avctx);
281 next_avc = buf_index + nalsize;
283 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
284 if (buf_index >= buf_size)
286 if (buf_index >= next_avc)
289 src_length = next_avc - buf_index;
291 state = buf[buf_index];
292 switch (state & 0x1f) {
294 case H264_NAL_IDR_SLICE:
295 // Do not walk the whole buffer just to decode slice header
296 if ((state & 0x1f) == H264_NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
297 /* IDR or disposable slice
298 * No need to decode many bytes because MMCOs shall not be present. */
302 /* To decode up to MMCOs */
303 if (src_length > 1000)
308 consumed = ff_h2645_extract_rbsp(buf + buf_index, src_length, &rbsp, &nal, 1);
312 buf_index += consumed;
314 ret = init_get_bits8(&nal.gb, nal.data, nal.size);
318 nal.ref_idc = get_bits(&nal.gb, 2);
319 nal.type = get_bits(&nal.gb, 5);
323 ff_h264_decode_seq_parameter_set(&nal.gb, avctx, &p->ps, 0);
326 ff_h264_decode_picture_parameter_set(&nal.gb, avctx, &p->ps,
330 ff_h264_sei_decode(&p->sei, &nal.gb, &p->ps, avctx);
332 case H264_NAL_IDR_SLICE:
335 p->poc.prev_frame_num = 0;
336 p->poc.prev_frame_num_offset = 0;
337 p->poc.prev_poc_msb =
338 p->poc.prev_poc_lsb = 0;
341 get_ue_golomb_long(&nal.gb); // skip first_mb_in_slice
342 slice_type = get_ue_golomb_31(&nal.gb);
343 s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
344 if (p->sei.recovery_point.recovery_frame_cnt >= 0) {
345 /* key frame, since recovery_frame_cnt is set */
348 pps_id = get_ue_golomb(&nal.gb);
349 if (pps_id >= MAX_PPS_COUNT) {
350 av_log(avctx, AV_LOG_ERROR,
351 "pps_id %u out of range\n", pps_id);
354 if (!p->ps.pps_list[pps_id]) {
355 av_log(avctx, AV_LOG_ERROR,
356 "non-existing PPS %u referenced\n", pps_id);
360 av_buffer_unref(&p->ps.pps_ref);
361 av_buffer_unref(&p->ps.sps_ref);
364 p->ps.pps_ref = av_buffer_ref(p->ps.pps_list[pps_id]);
367 p->ps.pps = (const PPS*)p->ps.pps_ref->data;
369 if (!p->ps.sps_list[p->ps.pps->sps_id]) {
370 av_log(avctx, AV_LOG_ERROR,
371 "non-existing SPS %u referenced\n", p->ps.pps->sps_id);
375 p->ps.sps_ref = av_buffer_ref(p->ps.sps_list[p->ps.pps->sps_id]);
378 p->ps.sps = (const SPS*)p->ps.sps_ref->data;
382 // heuristic to detect non marked keyframes
383 if (p->ps.sps->ref_frame_count <= 1 && p->ps.pps->ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
386 p->poc.frame_num = get_bits(&nal.gb, sps->log2_max_frame_num);
388 s->coded_width = 16 * sps->mb_width;
389 s->coded_height = 16 * sps->mb_height;
390 s->width = s->coded_width - (sps->crop_right + sps->crop_left);
391 s->height = s->coded_height - (sps->crop_top + sps->crop_bottom);
392 if (s->width <= 0 || s->height <= 0) {
393 s->width = s->coded_width;
394 s->height = s->coded_height;
397 switch (sps->bit_depth_luma) {
399 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P9;
400 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P9;
401 else s->format = AV_PIX_FMT_YUV420P9;
404 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P10;
405 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P10;
406 else s->format = AV_PIX_FMT_YUV420P10;
409 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P;
410 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P;
411 else s->format = AV_PIX_FMT_YUV420P;
414 s->format = AV_PIX_FMT_NONE;
417 avctx->profile = ff_h264_get_profile(sps);
418 avctx->level = sps->level_idc;
420 if (sps->frame_mbs_only_flag) {
421 p->picture_structure = PICT_FRAME;
423 if (get_bits1(&nal.gb)) { // field_pic_flag
424 p->picture_structure = PICT_TOP_FIELD + get_bits1(&nal.gb); // bottom_field_flag
426 p->picture_structure = PICT_FRAME;
430 if (nal.type == H264_NAL_IDR_SLICE)
431 get_ue_golomb_long(&nal.gb); /* idr_pic_id */
432 if (sps->poc_type == 0) {
433 p->poc.poc_lsb = get_bits(&nal.gb, sps->log2_max_poc_lsb);
435 if (p->ps.pps->pic_order_present == 1 &&
436 p->picture_structure == PICT_FRAME)
437 p->poc.delta_poc_bottom = get_se_golomb(&nal.gb);
440 if (sps->poc_type == 1 &&
441 !sps->delta_pic_order_always_zero_flag) {
442 p->poc.delta_poc[0] = get_se_golomb(&nal.gb);
444 if (p->ps.pps->pic_order_present == 1 &&
445 p->picture_structure == PICT_FRAME)
446 p->poc.delta_poc[1] = get_se_golomb(&nal.gb);
449 /* Decode POC of this picture.
450 * The prev_ values needed for decoding POC of the next picture are not set here. */
451 field_poc[0] = field_poc[1] = INT_MAX;
452 ff_h264_init_poc(field_poc, &s->output_picture_number, sps,
453 &p->poc, p->picture_structure, nal.ref_idc);
455 /* Continue parsing to check if MMCO_RESET is present.
456 * FIXME: MMCO_RESET could appear in non-first slice.
457 * Maybe, we should parse all undisposable non-IDR slice of this
458 * picture until encountering MMCO_RESET in a slice of it. */
459 if (nal.ref_idc && nal.type != H264_NAL_IDR_SLICE) {
460 got_reset = scan_mmco_reset(s, &nal.gb, avctx);
465 /* Set up the prev_ values for decoding POC of the next picture. */
466 p->poc.prev_frame_num = got_reset ? 0 : p->poc.frame_num;
467 p->poc.prev_frame_num_offset = got_reset ? 0 : p->poc.frame_num_offset;
468 if (nal.ref_idc != 0) {
470 p->poc.prev_poc_msb = p->poc.poc_msb;
471 p->poc.prev_poc_lsb = p->poc.poc_lsb;
473 p->poc.prev_poc_msb = 0;
474 p->poc.prev_poc_lsb =
475 p->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
479 if (sps->pic_struct_present_flag && p->sei.picture_timing.present) {
480 switch (p->sei.picture_timing.pic_struct) {
481 case H264_SEI_PIC_STRUCT_TOP_FIELD:
482 case H264_SEI_PIC_STRUCT_BOTTOM_FIELD:
485 case H264_SEI_PIC_STRUCT_FRAME:
486 case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
487 case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
490 case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
491 case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
494 case H264_SEI_PIC_STRUCT_FRAME_DOUBLING:
497 case H264_SEI_PIC_STRUCT_FRAME_TRIPLING:
501 s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
505 s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
508 if (p->picture_structure == PICT_FRAME) {
509 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
510 if (sps->pic_struct_present_flag && p->sei.picture_timing.present) {
511 switch (p->sei.picture_timing.pic_struct) {
512 case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
513 case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
514 s->field_order = AV_FIELD_TT;
516 case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
517 case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
518 s->field_order = AV_FIELD_BB;
521 s->field_order = AV_FIELD_PROGRESSIVE;
525 if (field_poc[0] < field_poc[1])
526 s->field_order = AV_FIELD_TT;
527 else if (field_poc[0] > field_poc[1])
528 s->field_order = AV_FIELD_BB;
530 s->field_order = AV_FIELD_PROGRESSIVE;
533 if (p->picture_structure == PICT_TOP_FIELD)
534 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
536 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
537 if (p->poc.frame_num == p->last_frame_num &&
538 p->last_picture_structure != AV_PICTURE_STRUCTURE_UNKNOWN &&
539 p->last_picture_structure != AV_PICTURE_STRUCTURE_FRAME &&
540 p->last_picture_structure != s->picture_structure) {
541 if (p->last_picture_structure == AV_PICTURE_STRUCTURE_TOP_FIELD)
542 s->field_order = AV_FIELD_TT;
544 s->field_order = AV_FIELD_BB;
546 s->field_order = AV_FIELD_UNKNOWN;
548 p->last_picture_structure = s->picture_structure;
549 p->last_frame_num = p->poc.frame_num;
552 av_freep(&rbsp.rbsp_buffer);
553 return 0; /* no need to evaluate the rest */
557 av_freep(&rbsp.rbsp_buffer);
560 /* didn't find a picture! */
561 av_log(avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
563 av_freep(&rbsp.rbsp_buffer);
567 static int h264_parse(AVCodecParserContext *s,
568 AVCodecContext *avctx,
569 const uint8_t **poutbuf, int *poutbuf_size,
570 const uint8_t *buf, int buf_size)
572 H264ParseContext *p = s->priv_data;
573 ParseContext *pc = &p->pc;
578 if (avctx->extradata_size) {
579 ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
580 &p->ps, &p->is_avc, &p->nal_length_size,
581 avctx->err_recognition, avctx);
585 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
588 next = h264_find_frame_end(p, buf, buf_size, avctx);
590 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
596 if (next < 0 && next != END_NOT_FOUND) {
597 av_assert1(pc->last_index + next >= 0);
598 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next, avctx); // update state
602 parse_nal_units(s, avctx, buf, buf_size);
604 if (avctx->framerate.num)
605 avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
606 if (p->sei.picture_timing.cpb_removal_delay >= 0) {
607 s->dts_sync_point = p->sei.buffering_period.present;
608 s->dts_ref_dts_delta = p->sei.picture_timing.cpb_removal_delay;
609 s->pts_dts_delta = p->sei.picture_timing.dpb_output_delay;
611 s->dts_sync_point = INT_MIN;
612 s->dts_ref_dts_delta = INT_MIN;
613 s->pts_dts_delta = INT_MIN;
616 if (s->flags & PARSER_FLAG_ONCE) {
617 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
620 if (s->dts_sync_point >= 0) {
621 int64_t den = avctx->time_base.den * (int64_t)avctx->pkt_timebase.num;
623 int64_t num = avctx->time_base.num * (int64_t)avctx->pkt_timebase.den;
624 if (s->dts != AV_NOPTS_VALUE) {
625 // got DTS from the stream, update reference timestamp
626 p->reference_dts = s->dts - av_rescale(s->dts_ref_dts_delta, num, den);
627 } else if (p->reference_dts != AV_NOPTS_VALUE) {
628 // compute DTS based on reference timestamp
629 s->dts = p->reference_dts + av_rescale(s->dts_ref_dts_delta, num, den);
632 if (p->reference_dts != AV_NOPTS_VALUE && s->pts == AV_NOPTS_VALUE)
633 s->pts = s->dts + av_rescale(s->pts_dts_delta, num, den);
635 if (s->dts_sync_point > 0)
636 p->reference_dts = s->dts; // new reference
641 *poutbuf_size = buf_size;
645 static int h264_split(AVCodecContext *avctx,
646 const uint8_t *buf, int buf_size)
651 const uint8_t *ptr = buf, *end = buf + buf_size;
655 ptr = avpriv_find_start_code(ptr, end, &state);
656 if ((state & 0xFFFFFF00) != 0x100)
658 nalu_type = state & 0x1F;
659 if (nalu_type == H264_NAL_SPS) {
661 } else if (nalu_type == H264_NAL_PPS)
663 /* else if (nalu_type == 0x01 ||
664 * nalu_type == 0x02 ||
665 * nalu_type == 0x05) {
668 else if ((nalu_type != H264_NAL_SEI || has_pps) &&
669 nalu_type != H264_NAL_AUD && nalu_type != H264_NAL_SPS_EXT &&
672 while (ptr - 4 > buf && ptr[-5] == 0)
674 return ptr - 4 - buf;
682 static void h264_close(AVCodecParserContext *s)
684 H264ParseContext *p = s->priv_data;
685 ParseContext *pc = &p->pc;
687 av_freep(&pc->buffer);
689 ff_h264_sei_uninit(&p->sei);
690 ff_h264_ps_uninit(&p->ps);
693 static av_cold int init(AVCodecParserContext *s)
695 H264ParseContext *p = s->priv_data;
697 p->reference_dts = AV_NOPTS_VALUE;
698 p->last_frame_num = INT_MAX;
699 ff_h264dsp_init(&p->h264dsp, 8, 1);
703 AVCodecParser ff_h264_parser = {
704 .codec_ids = { AV_CODEC_ID_H264 },
705 .priv_data_size = sizeof(H264ParseContext),
707 .parser_parse = h264_parse,
708 .parser_close = h264_close,