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;
66 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
67 int buf_size, void *logctx)
71 ParseContext *pc = &p->pc;
73 int next_avc = p->is_avc ? 0 : buf_size;
74 // mb_addr= pc->mb_addr - 1;
79 if (p->is_avc && !p->nal_length_size)
80 av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
82 for (i = 0; i < buf_size; i++) {
86 for (j = 0; j < p->nal_length_size; j++)
87 nalsize = (nalsize << 8) | buf[i++];
88 if (nalsize <= 0 || nalsize > buf_size - i) {
89 av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
92 next_avc = i + nalsize;
97 i += p->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
100 } else if (state <= 2) {
102 state ^= 5; // 2->7, 1->4, 0->5
106 state >>= 1; // 2->1, 1->0, 0->0
107 } else if (state <= 5) {
108 int nalu_type = buf[i] & 0x1F;
109 if (nalu_type == H264_NAL_SEI || nalu_type == H264_NAL_SPS ||
110 nalu_type == H264_NAL_PPS || nalu_type == H264_NAL_AUD) {
111 if (pc->frame_start_found) {
115 } else if (nalu_type == H264_NAL_SLICE || nalu_type == H264_NAL_DPA ||
116 nalu_type == H264_NAL_IDR_SLICE) {
122 p->parse_history[p->parse_history_count++] = buf[i];
123 if (p->parse_history_count > 5) {
124 unsigned int mb, last_mb = p->parse_last_mb;
127 init_get_bits(&gb, p->parse_history, 8*p->parse_history_count);
128 p->parse_history_count = 0;
129 mb= get_ue_golomb_long(&gb);
130 p->parse_last_mb = mb;
131 if (pc->frame_start_found) {
135 pc->frame_start_found = 1;
143 return END_NOT_FOUND;
147 pc->frame_start_found = 0;
150 return i - (state & 5) - 5 * (state > 7);
153 static int scan_mmco_reset(AVCodecParserContext *s, GetBitContext *gb,
156 H264PredWeightTable pwt;
157 int slice_type_nos = s->pict_type & 3;
158 H264ParseContext *p = s->priv_data;
159 int list_count, ref_count[2];
162 if (p->ps.pps->redundant_pic_cnt_present)
163 get_ue_golomb(gb); // redundant_pic_count
165 if (slice_type_nos == AV_PICTURE_TYPE_B)
166 get_bits1(gb); // direct_spatial_mv_pred
168 if (ff_h264_parse_ref_count(&list_count, ref_count, gb, p->ps.pps,
169 slice_type_nos, p->picture_structure, logctx) < 0)
170 return AVERROR_INVALIDDATA;
172 if (slice_type_nos != AV_PICTURE_TYPE_I) {
174 for (list = 0; list < list_count; list++) {
177 for (index = 0; ; index++) {
178 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(gb);
180 if (reordering_of_pic_nums_idc < 3)
181 get_ue_golomb_long(gb);
182 else if (reordering_of_pic_nums_idc > 3) {
183 av_log(logctx, AV_LOG_ERROR,
184 "illegal reordering_of_pic_nums_idc %d\n",
185 reordering_of_pic_nums_idc);
186 return AVERROR_INVALIDDATA;
190 if (index >= ref_count[list]) {
191 av_log(logctx, AV_LOG_ERROR,
192 "reference count %d overflow\n", index);
193 return AVERROR_INVALIDDATA;
200 if ((p->ps.pps->weighted_pred && slice_type_nos == AV_PICTURE_TYPE_P) ||
201 (p->ps.pps->weighted_bipred_idc == 1 && slice_type_nos == AV_PICTURE_TYPE_B))
202 ff_h264_pred_weight_table(gb, p->ps.sps, ref_count, slice_type_nos,
205 if (get_bits1(gb)) { // adaptive_ref_pic_marking_mode_flag
207 for (i = 0; i < MAX_MMCO_COUNT; i++) {
208 MMCOOpcode opcode = get_ue_golomb_31(gb);
209 if (opcode > (unsigned) MMCO_LONG) {
210 av_log(logctx, AV_LOG_ERROR,
211 "illegal memory management control operation %d\n",
213 return AVERROR_INVALIDDATA;
215 if (opcode == MMCO_END)
217 else if (opcode == MMCO_RESET)
220 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
221 get_ue_golomb_long(gb); // difference_of_pic_nums_minus1
222 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
223 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
224 get_ue_golomb_31(gb);
231 static inline int get_avc_nalsize(H264ParseContext *p, const uint8_t *buf,
232 int buf_size, int *buf_index, void *logctx)
236 if (*buf_index >= buf_size - p->nal_length_size) {
237 // the end of the buffer is reached, refill it
238 return AVERROR(EAGAIN);
241 for (i = 0; i < p->nal_length_size; i++)
242 nalsize = ((unsigned)nalsize << 8) | buf[(*buf_index)++];
243 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
244 av_log(logctx, AV_LOG_ERROR,
245 "AVC: nal size %d\n", nalsize);
246 return AVERROR_INVALIDDATA;
252 * Parse NAL units of found picture and decode some basic information.
254 * @param s parser context.
255 * @param avctx codec context.
256 * @param buf buffer with field/frame data.
257 * @param buf_size size of the buffer.
259 static inline int parse_nal_units(AVCodecParserContext *s,
260 AVCodecContext *avctx,
261 const uint8_t * const buf, int buf_size)
263 H264ParseContext *p = s->priv_data;
264 H2645NAL nal = { NULL };
265 int buf_index, next_avc;
267 unsigned int slice_type;
268 int state = -1, got_reset = 0;
269 int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
273 /* set some sane default values */
274 s->pict_type = AV_PICTURE_TYPE_I;
276 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
278 ff_h264_sei_uninit(&p->sei);
279 p->sei.frame_packing.frame_packing_arrangement_cancel_flag = -1;
285 next_avc = p->is_avc ? 0 : buf_size;
288 int src_length, consumed, nalsize = 0;
290 if (buf_index >= next_avc) {
291 nalsize = get_avc_nalsize(p, buf, buf_size, &buf_index, avctx);
294 next_avc = buf_index + nalsize;
296 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
297 if (buf_index >= buf_size)
299 if (buf_index >= next_avc)
302 src_length = next_avc - buf_index;
304 state = buf[buf_index];
305 switch (state & 0x1f) {
307 case H264_NAL_IDR_SLICE:
308 // Do not walk the whole buffer just to decode slice header
309 if ((state & 0x1f) == H264_NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
310 /* IDR or disposable slice
311 * No need to decode many bytes because MMCOs shall not be present. */
315 /* To decode up to MMCOs */
316 if (src_length > 1000)
321 consumed = ff_h2645_extract_rbsp(buf + buf_index, src_length, &nal);
325 buf_index += consumed;
327 ret = init_get_bits8(&nal.gb, nal.data, nal.size);
331 nal.ref_idc = get_bits(&nal.gb, 2);
332 nal.type = get_bits(&nal.gb, 5);
336 ff_h264_decode_seq_parameter_set(&nal.gb, avctx, &p->ps, 0);
339 ff_h264_decode_picture_parameter_set(&nal.gb, avctx, &p->ps,
343 ff_h264_sei_decode(&p->sei, &nal.gb, &p->ps, avctx);
345 case H264_NAL_IDR_SLICE:
348 p->poc.prev_frame_num = 0;
349 p->poc.prev_frame_num_offset = 0;
350 p->poc.prev_poc_msb =
351 p->poc.prev_poc_lsb = 0;
354 get_ue_golomb_long(&nal.gb); // skip first_mb_in_slice
355 slice_type = get_ue_golomb_31(&nal.gb);
356 s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
357 if (p->sei.recovery_point.recovery_frame_cnt >= 0) {
358 /* key frame, since recovery_frame_cnt is set */
361 pps_id = get_ue_golomb(&nal.gb);
362 if (pps_id >= MAX_PPS_COUNT) {
363 av_log(avctx, AV_LOG_ERROR,
364 "pps_id %u out of range\n", pps_id);
367 if (!p->ps.pps_list[pps_id]) {
368 av_log(avctx, AV_LOG_ERROR,
369 "non-existing PPS %u referenced\n", pps_id);
373 av_buffer_unref(&p->ps.pps_ref);
374 av_buffer_unref(&p->ps.sps_ref);
377 p->ps.pps_ref = av_buffer_ref(p->ps.pps_list[pps_id]);
380 p->ps.pps = (const PPS*)p->ps.pps_ref->data;
382 if (!p->ps.sps_list[p->ps.pps->sps_id]) {
383 av_log(avctx, AV_LOG_ERROR,
384 "non-existing SPS %u referenced\n", p->ps.pps->sps_id);
388 p->ps.sps_ref = av_buffer_ref(p->ps.sps_list[p->ps.pps->sps_id]);
391 p->ps.sps = (const SPS*)p->ps.sps_ref->data;
395 // heuristic to detect non marked keyframes
396 if (p->ps.sps->ref_frame_count <= 1 && p->ps.pps->ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
399 p->poc.frame_num = get_bits(&nal.gb, sps->log2_max_frame_num);
401 s->coded_width = 16 * sps->mb_width;
402 s->coded_height = 16 * sps->mb_height;
403 s->width = s->coded_width - (sps->crop_right + sps->crop_left);
404 s->height = s->coded_height - (sps->crop_top + sps->crop_bottom);
405 if (s->width <= 0 || s->height <= 0) {
406 s->width = s->coded_width;
407 s->height = s->coded_height;
410 switch (sps->bit_depth_luma) {
412 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P9;
413 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P9;
414 else s->format = AV_PIX_FMT_YUV420P9;
417 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P10;
418 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P10;
419 else s->format = AV_PIX_FMT_YUV420P10;
422 if (sps->chroma_format_idc == 3) s->format = AV_PIX_FMT_YUV444P;
423 else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P;
424 else s->format = AV_PIX_FMT_YUV420P;
427 s->format = AV_PIX_FMT_NONE;
430 avctx->profile = ff_h264_get_profile(sps);
431 avctx->level = sps->level_idc;
433 if (sps->frame_mbs_only_flag) {
434 p->picture_structure = PICT_FRAME;
436 if (get_bits1(&nal.gb)) { // field_pic_flag
437 p->picture_structure = PICT_TOP_FIELD + get_bits1(&nal.gb); // bottom_field_flag
439 p->picture_structure = PICT_FRAME;
443 if (nal.type == H264_NAL_IDR_SLICE)
444 get_ue_golomb_long(&nal.gb); /* idr_pic_id */
445 if (sps->poc_type == 0) {
446 p->poc.poc_lsb = get_bits(&nal.gb, sps->log2_max_poc_lsb);
448 if (p->ps.pps->pic_order_present == 1 &&
449 p->picture_structure == PICT_FRAME)
450 p->poc.delta_poc_bottom = get_se_golomb(&nal.gb);
453 if (sps->poc_type == 1 &&
454 !sps->delta_pic_order_always_zero_flag) {
455 p->poc.delta_poc[0] = get_se_golomb(&nal.gb);
457 if (p->ps.pps->pic_order_present == 1 &&
458 p->picture_structure == PICT_FRAME)
459 p->poc.delta_poc[1] = get_se_golomb(&nal.gb);
462 /* Decode POC of this picture.
463 * The prev_ values needed for decoding POC of the next picture are not set here. */
464 field_poc[0] = field_poc[1] = INT_MAX;
465 ff_h264_init_poc(field_poc, &s->output_picture_number, sps,
466 &p->poc, p->picture_structure, nal.ref_idc);
468 /* Continue parsing to check if MMCO_RESET is present.
469 * FIXME: MMCO_RESET could appear in non-first slice.
470 * Maybe, we should parse all undisposable non-IDR slice of this
471 * picture until encountering MMCO_RESET in a slice of it. */
472 if (nal.ref_idc && nal.type != H264_NAL_IDR_SLICE) {
473 got_reset = scan_mmco_reset(s, &nal.gb, avctx);
478 /* Set up the prev_ values for decoding POC of the next picture. */
479 p->poc.prev_frame_num = got_reset ? 0 : p->poc.frame_num;
480 p->poc.prev_frame_num_offset = got_reset ? 0 : p->poc.frame_num_offset;
481 if (nal.ref_idc != 0) {
483 p->poc.prev_poc_msb = p->poc.poc_msb;
484 p->poc.prev_poc_lsb = p->poc.poc_lsb;
486 p->poc.prev_poc_msb = 0;
487 p->poc.prev_poc_lsb =
488 p->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
492 if (sps->pic_struct_present_flag) {
493 switch (p->sei.picture_timing.pic_struct) {
494 case SEI_PIC_STRUCT_TOP_FIELD:
495 case SEI_PIC_STRUCT_BOTTOM_FIELD:
498 case SEI_PIC_STRUCT_FRAME:
499 case SEI_PIC_STRUCT_TOP_BOTTOM:
500 case SEI_PIC_STRUCT_BOTTOM_TOP:
503 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
504 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
507 case SEI_PIC_STRUCT_FRAME_DOUBLING:
510 case SEI_PIC_STRUCT_FRAME_TRIPLING:
514 s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
518 s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
521 if (p->picture_structure == PICT_FRAME) {
522 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
523 if (sps->pic_struct_present_flag) {
524 switch (p->sei.picture_timing.pic_struct) {
525 case SEI_PIC_STRUCT_TOP_BOTTOM:
526 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
527 s->field_order = AV_FIELD_TT;
529 case SEI_PIC_STRUCT_BOTTOM_TOP:
530 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
531 s->field_order = AV_FIELD_BB;
534 s->field_order = AV_FIELD_PROGRESSIVE;
538 if (field_poc[0] < field_poc[1])
539 s->field_order = AV_FIELD_TT;
540 else if (field_poc[0] > field_poc[1])
541 s->field_order = AV_FIELD_BB;
543 s->field_order = AV_FIELD_PROGRESSIVE;
546 if (p->picture_structure == PICT_TOP_FIELD)
547 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
549 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
550 s->field_order = AV_FIELD_UNKNOWN;
553 av_freep(&nal.rbsp_buffer);
554 return 0; /* no need to evaluate the rest */
558 av_freep(&nal.rbsp_buffer);
561 /* didn't find a picture! */
562 av_log(avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
564 av_freep(&nal.rbsp_buffer);
568 static int h264_parse(AVCodecParserContext *s,
569 AVCodecContext *avctx,
570 const uint8_t **poutbuf, int *poutbuf_size,
571 const uint8_t *buf, int buf_size)
573 H264ParseContext *p = s->priv_data;
574 ParseContext *pc = &p->pc;
579 if (avctx->extradata_size) {
580 ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
581 &p->ps, &p->is_avc, &p->nal_length_size,
582 avctx->err_recognition, avctx);
586 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
589 next = h264_find_frame_end(p, buf, buf_size, avctx);
591 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
597 if (next < 0 && next != END_NOT_FOUND) {
598 av_assert1(pc->last_index + next >= 0);
599 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next, avctx); // update state
603 parse_nal_units(s, avctx, buf, buf_size);
605 if (avctx->framerate.num)
606 avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
607 if (p->sei.picture_timing.cpb_removal_delay >= 0) {
608 s->dts_sync_point = p->sei.buffering_period.present;
609 s->dts_ref_dts_delta = p->sei.picture_timing.cpb_removal_delay;
610 s->pts_dts_delta = p->sei.picture_timing.dpb_output_delay;
612 s->dts_sync_point = INT_MIN;
613 s->dts_ref_dts_delta = INT_MIN;
614 s->pts_dts_delta = INT_MIN;
617 if (s->flags & PARSER_FLAG_ONCE) {
618 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
622 *poutbuf_size = buf_size;
626 static int h264_split(AVCodecContext *avctx,
627 const uint8_t *buf, int buf_size)
632 const uint8_t *ptr = buf, *end = buf + buf_size;
636 ptr = avpriv_find_start_code(ptr, end, &state);
637 if ((state & 0xFFFFFF00) != 0x100)
639 nalu_type = state & 0x1F;
640 if (nalu_type == H264_NAL_SPS) {
642 } else if (nalu_type == H264_NAL_PPS)
644 /* else if (nalu_type == 0x01 ||
645 * nalu_type == 0x02 ||
646 * nalu_type == 0x05) {
649 else if ((nalu_type != H264_NAL_SEI || has_pps) &&
650 nalu_type != H264_NAL_AUD && nalu_type != H264_NAL_SPS_EXT &&
653 while (ptr - 4 > buf && ptr[-5] == 0)
655 return ptr - 4 - buf;
663 static void h264_close(AVCodecParserContext *s)
665 H264ParseContext *p = s->priv_data;
666 ParseContext *pc = &p->pc;
668 av_freep(&pc->buffer);
670 ff_h264_sei_uninit(&p->sei);
671 ff_h264_ps_uninit(&p->ps);
674 static av_cold int init(AVCodecParserContext *s)
676 H264ParseContext *p = s->priv_data;
678 ff_h264dsp_init(&p->h264dsp, 8, 1);
682 AVCodecParser ff_h264_parser = {
683 .codec_ids = { AV_CODEC_ID_H264 },
684 .priv_data_size = sizeof(H264ParseContext),
686 .parser_parse = h264_parse,
687 .parser_close = h264_close,