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 / MPEG4 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"
44 #include "mpegutils.h"
47 typedef struct H264ParseContext {
54 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
57 H264Context *h = &p->h;
60 ParseContext *pc = &p->pc;
62 int next_avc= h->is_avc ? 0 : buf_size;
63 // mb_addr= pc->mb_addr - 1;
68 if (h->is_avc && !h->nal_length_size)
69 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
71 for (i = 0; i < buf_size; i++) {
75 for (j = 0; j < h->nal_length_size; j++)
76 nalsize = (nalsize << 8) | buf[i++];
77 if (nalsize <= 0 || nalsize > buf_size - i) {
78 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
81 next_avc = i + nalsize;
86 i += h->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
89 } else if (state <= 2) {
91 state ^= 5; // 2->7, 1->4, 0->5
95 state >>= 1; // 2->1, 1->0, 0->0
96 } else if (state <= 5) {
97 int nalu_type = buf[i] & 0x1F;
98 if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
99 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
100 if (pc->frame_start_found) {
104 } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
105 nalu_type == NAL_IDR_SLICE) {
111 h->parse_history[h->parse_history_count++]= buf[i];
112 if (h->parse_history_count>5) {
113 unsigned int mb, last_mb= h->parse_last_mb;
116 init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
117 h->parse_history_count=0;
118 mb= get_ue_golomb_long(&gb);
119 h->parse_last_mb= mb;
120 if (pc->frame_start_found) {
124 pc->frame_start_found = 1;
132 return END_NOT_FOUND;
136 pc->frame_start_found = 0;
139 return i - (state & 5) - 5 * (state > 7);
142 static int scan_mmco_reset(AVCodecParserContext *s)
144 H264ParseContext *p = s->priv_data;
145 H264Context *h = &p->h;
146 H264SliceContext *sl = &h->slice_ctx[0];
148 sl->slice_type_nos = s->pict_type & 3;
150 if (h->pps.redundant_pic_cnt_present)
151 get_ue_golomb(&sl->gb); // redundant_pic_count
153 if (ff_set_ref_count(h, sl) < 0)
154 return AVERROR_INVALIDDATA;
156 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
158 for (list = 0; list < sl->list_count; list++) {
159 if (get_bits1(&sl->gb)) {
161 for (index = 0; ; index++) {
162 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&sl->gb);
164 if (reordering_of_pic_nums_idc < 3)
165 get_ue_golomb_long(&sl->gb);
166 else if (reordering_of_pic_nums_idc > 3) {
167 av_log(h->avctx, AV_LOG_ERROR,
168 "illegal reordering_of_pic_nums_idc %d\n",
169 reordering_of_pic_nums_idc);
170 return AVERROR_INVALIDDATA;
174 if (index >= sl->ref_count[list]) {
175 av_log(h->avctx, AV_LOG_ERROR,
176 "reference count %d overflow\n", index);
177 return AVERROR_INVALIDDATA;
184 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
185 (h->pps.weighted_bipred_idc == 1 && sl->slice_type_nos == AV_PICTURE_TYPE_B))
186 ff_pred_weight_table(h, sl);
188 if (get_bits1(&sl->gb)) { // adaptive_ref_pic_marking_mode_flag
190 for (i = 0; i < MAX_MMCO_COUNT; i++) {
191 MMCOOpcode opcode = get_ue_golomb_31(&sl->gb);
192 if (opcode > (unsigned) MMCO_LONG) {
193 av_log(h->avctx, AV_LOG_ERROR,
194 "illegal memory management control operation %d\n",
196 return AVERROR_INVALIDDATA;
198 if (opcode == MMCO_END)
200 else if (opcode == MMCO_RESET)
203 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
204 get_ue_golomb_long(&sl->gb); // difference_of_pic_nums_minus1
205 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
206 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
207 get_ue_golomb_31(&sl->gb);
215 * Parse NAL units of found picture and decode some basic information.
217 * @param s parser context.
218 * @param avctx codec context.
219 * @param buf buffer with field/frame data.
220 * @param buf_size size of the buffer.
222 static inline int parse_nal_units(AVCodecParserContext *s,
223 AVCodecContext *avctx,
224 const uint8_t * const buf, int buf_size)
226 H264ParseContext *p = s->priv_data;
227 H264Context *h = &p->h;
228 H264SliceContext *sl = &h->slice_ctx[0];
229 int buf_index, next_avc;
231 unsigned int slice_type;
232 int state = -1, got_reset = 0;
234 int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
237 /* set some sane default values */
238 s->pict_type = AV_PICTURE_TYPE_I;
240 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
243 ff_h264_reset_sei(h);
244 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
250 next_avc = h->is_avc ? 0 : buf_size;
252 int src_length, dst_length, consumed, nalsize = 0;
254 if (buf_index >= next_avc) {
255 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
258 next_avc = buf_index + nalsize;
260 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
261 if (buf_index >= buf_size)
263 if (buf_index >= next_avc)
266 src_length = next_avc - buf_index;
268 state = buf[buf_index];
269 switch (state & 0x1f) {
272 // Do not walk the whole buffer just to decode slice header
273 if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
274 /* IDR or disposable slice
275 * No need to decode many bytes because MMCOs shall not be present. */
279 /* To decode up to MMCOs */
280 if (src_length > 1000)
285 ptr = ff_h264_decode_nal(h, sl, buf + buf_index, &dst_length,
286 &consumed, src_length);
287 if (!ptr || dst_length < 0)
290 buf_index += consumed;
292 init_get_bits(&h->gb, ptr, 8 * dst_length);
293 switch (h->nal_unit_type) {
295 ff_h264_decode_seq_parameter_set(h, 0);
298 ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
301 ff_h264_decode_sei(h);
306 h->prev_frame_num = 0;
307 h->prev_frame_num_offset = 0;
312 init_get_bits(&sl->gb, ptr, 8 * dst_length);
313 get_ue_golomb_long(&sl->gb); // skip first_mb_in_slice
314 slice_type = get_ue_golomb_31(&sl->gb);
315 s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
316 if (h->sei_recovery_frame_cnt >= 0) {
317 /* key frame, since recovery_frame_cnt is set */
320 pps_id = get_ue_golomb(&sl->gb);
321 if (pps_id >= MAX_PPS_COUNT) {
322 av_log(h->avctx, AV_LOG_ERROR,
323 "pps_id %u out of range\n", pps_id);
326 if (!h->pps_buffers[pps_id]) {
327 av_log(h->avctx, AV_LOG_ERROR,
328 "non-existing PPS %u referenced\n", pps_id);
331 h->pps = *h->pps_buffers[pps_id];
332 if (!h->sps_buffers[h->pps.sps_id]) {
333 av_log(h->avctx, AV_LOG_ERROR,
334 "non-existing SPS %u referenced\n", h->pps.sps_id);
337 h->sps = *h->sps_buffers[h->pps.sps_id];
338 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
340 if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
343 s->coded_width = 16 * h->sps.mb_width;
344 s->coded_height = 16 * h->sps.mb_height;
345 s->width = s->coded_width - (h->sps.crop_right + h->sps.crop_left);
346 s->height = s->coded_height - (h->sps.crop_top + h->sps.crop_bottom);
347 if (s->width <= 0 || s->height <= 0) {
348 s->width = s->coded_width;
349 s->height = s->coded_height;
352 switch (h->sps.bit_depth_luma) {
354 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P9;
355 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
356 else s->format = AV_PIX_FMT_YUV420P9;
359 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P10;
360 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
361 else s->format = AV_PIX_FMT_YUV420P10;
364 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P;
365 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
366 else s->format = AV_PIX_FMT_YUV420P;
369 s->format = AV_PIX_FMT_NONE;
372 avctx->profile = ff_h264_get_profile(&h->sps);
373 avctx->level = h->sps.level_idc;
375 if (h->sps.frame_mbs_only_flag) {
376 h->picture_structure = PICT_FRAME;
378 if (get_bits1(&sl->gb)) { // field_pic_flag
379 h->picture_structure = PICT_TOP_FIELD + get_bits1(&sl->gb); // bottom_field_flag
381 h->picture_structure = PICT_FRAME;
385 if (h->nal_unit_type == NAL_IDR_SLICE)
386 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
387 if (h->sps.poc_type == 0) {
388 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
390 if (h->pps.pic_order_present == 1 &&
391 h->picture_structure == PICT_FRAME)
392 h->delta_poc_bottom = get_se_golomb(&sl->gb);
395 if (h->sps.poc_type == 1 &&
396 !h->sps.delta_pic_order_always_zero_flag) {
397 h->delta_poc[0] = get_se_golomb(&sl->gb);
399 if (h->pps.pic_order_present == 1 &&
400 h->picture_structure == PICT_FRAME)
401 h->delta_poc[1] = get_se_golomb(&sl->gb);
404 /* Decode POC of this picture.
405 * The prev_ values needed for decoding POC of the next picture are not set here. */
406 field_poc[0] = field_poc[1] = INT_MAX;
407 ff_init_poc(h, field_poc, &s->output_picture_number);
409 /* Continue parsing to check if MMCO_RESET is present.
410 * FIXME: MMCO_RESET could appear in non-first slice.
411 * Maybe, we should parse all undisposable non-IDR slice of this
412 * picture until encountering MMCO_RESET in a slice of it. */
413 if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
414 got_reset = scan_mmco_reset(s);
419 /* Set up the prev_ values for decoding POC of the next picture. */
420 h->prev_frame_num = got_reset ? 0 : h->frame_num;
421 h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
422 if (h->nal_ref_idc != 0) {
424 h->prev_poc_msb = h->poc_msb;
425 h->prev_poc_lsb = h->poc_lsb;
429 h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
433 if (h->sps.pic_struct_present_flag) {
434 switch (h->sei_pic_struct) {
435 case SEI_PIC_STRUCT_TOP_FIELD:
436 case SEI_PIC_STRUCT_BOTTOM_FIELD:
439 case SEI_PIC_STRUCT_FRAME:
440 case SEI_PIC_STRUCT_TOP_BOTTOM:
441 case SEI_PIC_STRUCT_BOTTOM_TOP:
444 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
445 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
448 case SEI_PIC_STRUCT_FRAME_DOUBLING:
451 case SEI_PIC_STRUCT_FRAME_TRIPLING:
455 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
459 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
462 if (h->picture_structure == PICT_FRAME) {
463 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
464 if (h->sps.pic_struct_present_flag) {
465 switch (h->sei_pic_struct) {
466 case SEI_PIC_STRUCT_TOP_BOTTOM:
467 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
468 s->field_order = AV_FIELD_TT;
470 case SEI_PIC_STRUCT_BOTTOM_TOP:
471 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
472 s->field_order = AV_FIELD_BB;
475 s->field_order = AV_FIELD_PROGRESSIVE;
479 if (field_poc[0] < field_poc[1])
480 s->field_order = AV_FIELD_TT;
481 else if (field_poc[0] > field_poc[1])
482 s->field_order = AV_FIELD_BB;
484 s->field_order = AV_FIELD_PROGRESSIVE;
487 if (h->picture_structure == PICT_TOP_FIELD)
488 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
490 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
491 s->field_order = AV_FIELD_UNKNOWN;
494 return 0; /* no need to evaluate the rest */
499 /* didn't find a picture! */
500 av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
504 static int h264_parse(AVCodecParserContext *s,
505 AVCodecContext *avctx,
506 const uint8_t **poutbuf, int *poutbuf_size,
507 const uint8_t *buf, int buf_size)
509 H264ParseContext *p = s->priv_data;
510 H264Context *h = &p->h;
511 ParseContext *pc = &p->pc;
516 if (avctx->extradata_size) {
518 // must be done like in decoder, otherwise opening the parser,
519 // letting it create extradata and then closing and opening again
520 // will cause has_b_frames to be always set.
521 // Note that estimate_timings_from_pts does exactly this.
522 if (!avctx->has_b_frames)
524 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
528 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
531 next = h264_find_frame_end(p, buf, buf_size);
533 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
539 if (next < 0 && next != END_NOT_FOUND) {
540 av_assert1(pc->last_index + next >= 0);
541 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
545 parse_nal_units(s, avctx, buf, buf_size);
547 if (avctx->framerate.num)
548 avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
549 if (h->sei_cpb_removal_delay >= 0) {
550 s->dts_sync_point = h->sei_buffering_period_present;
551 s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
552 s->pts_dts_delta = h->sei_dpb_output_delay;
554 s->dts_sync_point = INT_MIN;
555 s->dts_ref_dts_delta = INT_MIN;
556 s->pts_dts_delta = INT_MIN;
559 if (s->flags & PARSER_FLAG_ONCE) {
560 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
564 *poutbuf_size = buf_size;
568 static int h264_split(AVCodecContext *avctx,
569 const uint8_t *buf, int buf_size)
574 const uint8_t *ptr = buf, *end = buf + buf_size;
578 ptr = avpriv_find_start_code(ptr, end, &state);
579 if ((state & 0xFFFFFF00) != 0x100)
581 nalu_type = state & 0x1F;
582 if (nalu_type == NAL_SPS) {
584 } else if (nalu_type == NAL_PPS)
586 /* else if (nalu_type == 0x01 ||
587 * nalu_type == 0x02 ||
588 * nalu_type == 0x05) {
591 else if ((nalu_type != NAL_SEI || has_pps) &&
592 nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
595 while (ptr - 4 > buf && ptr[-5] == 0)
597 return ptr - 4 - buf;
605 static void h264_close(AVCodecParserContext *s)
607 H264ParseContext *p = s->priv_data;
608 H264Context *h = &p->h;
609 ParseContext *pc = &p->pc;
611 av_freep(&pc->buffer);
612 ff_h264_free_context(h);
615 static av_cold int init(AVCodecParserContext *s)
617 H264ParseContext *p = s->priv_data;
618 H264Context *h = &p->h;
620 h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
625 h->slice_context_count = 1;
626 ff_h264dsp_init(&h->h264dsp, 8, 1);
630 AVCodecParser ff_h264_parser = {
631 .codec_ids = { AV_CODEC_ID_H264 },
632 .priv_data_size = sizeof(H264ParseContext),
634 .parser_parse = h264_parse,
635 .parser_close = h264_close,