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
30 #include "libavutil/attributes.h"
35 #include "mpegutils.h"
37 typedef struct H264ParseContext {
44 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
47 H264Context *h = &p->h;
50 ParseContext *pc = &p->pc;
52 int next_avc= h->is_avc ? 0 : buf_size;
53 // mb_addr= pc->mb_addr - 1;
58 if (h->is_avc && !h->nal_length_size)
59 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
61 for (i = 0; i < buf_size; i++) {
65 for (j = 0; j < h->nal_length_size; j++)
66 nalsize = (nalsize << 8) | buf[i++];
67 if (nalsize <= 0 || nalsize > buf_size - i) {
68 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
71 next_avc = i + nalsize;
76 i += h->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
79 } else if (state <= 2) {
81 state ^= 5; // 2->7, 1->4, 0->5
85 state >>= 1; // 2->1, 1->0, 0->0
86 } else if (state <= 5) {
87 int nalu_type = buf[i] & 0x1F;
88 if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
89 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
90 if (pc->frame_start_found) {
94 } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
95 nalu_type == NAL_IDR_SLICE) {
101 h->parse_history[h->parse_history_count++]= buf[i];
102 if (h->parse_history_count>5) {
103 unsigned int mb, last_mb= h->parse_last_mb;
106 init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
107 h->parse_history_count=0;
108 mb= get_ue_golomb_long(&gb);
109 h->parse_last_mb= mb;
110 if (pc->frame_start_found) {
114 pc->frame_start_found = 1;
122 return END_NOT_FOUND;
126 pc->frame_start_found = 0;
129 return i - (state & 5) - 5 * (state > 7);
132 static int scan_mmco_reset(AVCodecParserContext *s)
134 H264ParseContext *p = s->priv_data;
135 H264Context *h = &p->h;
136 H264SliceContext *sl = &h->slice_ctx[0];
138 h->slice_type_nos = s->pict_type & 3;
140 if (h->pps.redundant_pic_cnt_present)
141 get_ue_golomb(&h->gb); // redundant_pic_count
143 if (ff_set_ref_count(h) < 0)
144 return AVERROR_INVALIDDATA;
146 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
148 for (list = 0; list < h->list_count; list++) {
149 if (get_bits1(&h->gb)) {
151 for (index = 0; ; index++) {
152 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
154 if (reordering_of_pic_nums_idc < 3)
155 get_ue_golomb(&h->gb);
156 else if (reordering_of_pic_nums_idc > 3) {
157 av_log(h->avctx, AV_LOG_ERROR,
158 "illegal reordering_of_pic_nums_idc %d\n",
159 reordering_of_pic_nums_idc);
160 return AVERROR_INVALIDDATA;
164 if (index >= h->ref_count[list]) {
165 av_log(h->avctx, AV_LOG_ERROR,
166 "reference count %d overflow\n", index);
167 return AVERROR_INVALIDDATA;
174 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
175 (h->pps.weighted_bipred_idc == 1 && h->slice_type_nos == AV_PICTURE_TYPE_B))
176 ff_pred_weight_table(h, sl);
178 if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
180 for (i = 0; i < MAX_MMCO_COUNT; i++) {
181 MMCOOpcode opcode = get_ue_golomb_31(&h->gb);
182 if (opcode > (unsigned) MMCO_LONG) {
183 av_log(h->avctx, AV_LOG_ERROR,
184 "illegal memory management control operation %d\n",
186 return AVERROR_INVALIDDATA;
188 if (opcode == MMCO_END)
190 else if (opcode == MMCO_RESET)
193 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
194 get_ue_golomb(&h->gb);
195 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
196 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
197 get_ue_golomb_31(&h->gb);
205 * Parse NAL units of found picture and decode some basic information.
207 * @param s parser context.
208 * @param avctx codec context.
209 * @param buf buffer with field/frame data.
210 * @param buf_size size of the buffer.
212 static inline int parse_nal_units(AVCodecParserContext *s,
213 AVCodecContext *avctx,
214 const uint8_t * const buf, int buf_size)
216 H264ParseContext *p = s->priv_data;
217 H264Context *h = &p->h;
218 int buf_index, next_avc;
220 unsigned int slice_type;
221 int state = -1, got_reset = 0;
223 int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
226 /* set some sane default values */
227 s->pict_type = AV_PICTURE_TYPE_I;
229 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
232 ff_h264_reset_sei(h);
233 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
239 next_avc = h->is_avc ? 0 : buf_size;
241 int src_length, dst_length, consumed, nalsize = 0;
243 if (buf_index >= next_avc) {
244 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
247 next_avc = buf_index + nalsize;
249 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
250 if (buf_index >= buf_size)
252 if (buf_index >= next_avc)
255 src_length = next_avc - buf_index;
257 state = buf[buf_index];
258 switch (state & 0x1f) {
261 // Do not walk the whole buffer just to decode slice header
262 if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
263 /* IDR or disposable slice
264 * No need to decode many bytes because MMCOs shall not be present. */
268 /* To decode up to MMCOs */
269 if (src_length > 1000)
274 ptr = ff_h264_decode_nal(h, buf + buf_index, &dst_length,
275 &consumed, src_length);
276 if (!ptr || dst_length < 0)
279 buf_index += consumed;
281 init_get_bits(&h->gb, ptr, 8 * dst_length);
282 switch (h->nal_unit_type) {
284 ff_h264_decode_seq_parameter_set(h);
287 ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
290 ff_h264_decode_sei(h);
295 h->prev_frame_num = 0;
296 h->prev_frame_num_offset = 0;
301 get_ue_golomb_long(&h->gb); // skip first_mb_in_slice
302 slice_type = get_ue_golomb_31(&h->gb);
303 s->pict_type = golomb_to_pict_type[slice_type % 5];
304 if (h->sei_recovery_frame_cnt >= 0) {
305 /* key frame, since recovery_frame_cnt is set */
308 pps_id = get_ue_golomb(&h->gb);
309 if (pps_id >= MAX_PPS_COUNT) {
310 av_log(h->avctx, AV_LOG_ERROR,
311 "pps_id %u out of range\n", pps_id);
314 if (!h->pps_buffers[pps_id]) {
315 av_log(h->avctx, AV_LOG_ERROR,
316 "non-existing PPS %u referenced\n", pps_id);
319 h->pps = *h->pps_buffers[pps_id];
320 if (!h->sps_buffers[h->pps.sps_id]) {
321 av_log(h->avctx, AV_LOG_ERROR,
322 "non-existing SPS %u referenced\n", h->pps.sps_id);
325 h->sps = *h->sps_buffers[h->pps.sps_id];
326 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
328 if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
331 s->coded_width = 16 * h->sps.mb_width;
332 s->coded_height = 16 * h->sps.mb_height;
333 s->width = s->coded_width - (h->sps.crop_right + h->sps.crop_left);
334 s->height = s->coded_height - (h->sps.crop_top + h->sps.crop_bottom);
335 if (s->width <= 0 || s->height <= 0) {
336 s->width = s->coded_width;
337 s->height = s->coded_height;
340 switch (h->sps.bit_depth_luma) {
342 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P9;
343 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
344 else s->format = AV_PIX_FMT_YUV420P9;
347 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P10;
348 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
349 else s->format = AV_PIX_FMT_YUV420P10;
352 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P;
353 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
354 else s->format = AV_PIX_FMT_YUV420P;
357 s->format = AV_PIX_FMT_NONE;
360 avctx->profile = ff_h264_get_profile(&h->sps);
361 avctx->level = h->sps.level_idc;
363 if (h->sps.frame_mbs_only_flag) {
364 h->picture_structure = PICT_FRAME;
366 if (get_bits1(&h->gb)) { // field_pic_flag
367 h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
369 h->picture_structure = PICT_FRAME;
373 if (h->nal_unit_type == NAL_IDR_SLICE)
374 get_ue_golomb(&h->gb); /* idr_pic_id */
375 if (h->sps.poc_type == 0) {
376 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
378 if (h->pps.pic_order_present == 1 &&
379 h->picture_structure == PICT_FRAME)
380 h->delta_poc_bottom = get_se_golomb(&h->gb);
383 if (h->sps.poc_type == 1 &&
384 !h->sps.delta_pic_order_always_zero_flag) {
385 h->delta_poc[0] = get_se_golomb(&h->gb);
387 if (h->pps.pic_order_present == 1 &&
388 h->picture_structure == PICT_FRAME)
389 h->delta_poc[1] = get_se_golomb(&h->gb);
392 /* Decode POC of this picture.
393 * The prev_ values needed for decoding POC of the next picture are not set here. */
394 field_poc[0] = field_poc[1] = INT_MAX;
395 ff_init_poc(h, field_poc, &s->output_picture_number);
397 /* Continue parsing to check if MMCO_RESET is present.
398 * FIXME: MMCO_RESET could appear in non-first slice.
399 * Maybe, we should parse all undisposable non-IDR slice of this
400 * picture until encountering MMCO_RESET in a slice of it. */
401 if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
402 got_reset = scan_mmco_reset(s);
407 /* Set up the prev_ values for decoding POC of the next picture. */
408 h->prev_frame_num = got_reset ? 0 : h->frame_num;
409 h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
410 if (h->nal_ref_idc != 0) {
412 h->prev_poc_msb = h->poc_msb;
413 h->prev_poc_lsb = h->poc_lsb;
417 h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
421 if (h->sps.pic_struct_present_flag) {
422 switch (h->sei_pic_struct) {
423 case SEI_PIC_STRUCT_TOP_FIELD:
424 case SEI_PIC_STRUCT_BOTTOM_FIELD:
427 case SEI_PIC_STRUCT_FRAME:
428 case SEI_PIC_STRUCT_TOP_BOTTOM:
429 case SEI_PIC_STRUCT_BOTTOM_TOP:
432 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
433 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
436 case SEI_PIC_STRUCT_FRAME_DOUBLING:
439 case SEI_PIC_STRUCT_FRAME_TRIPLING:
443 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
447 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
450 if (h->picture_structure == PICT_FRAME) {
451 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
452 if (h->sps.pic_struct_present_flag) {
453 switch (h->sei_pic_struct) {
454 case SEI_PIC_STRUCT_TOP_BOTTOM:
455 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
456 s->field_order = AV_FIELD_TT;
458 case SEI_PIC_STRUCT_BOTTOM_TOP:
459 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
460 s->field_order = AV_FIELD_BB;
463 s->field_order = AV_FIELD_PROGRESSIVE;
467 if (field_poc[0] < field_poc[1])
468 s->field_order = AV_FIELD_TT;
469 else if (field_poc[0] > field_poc[1])
470 s->field_order = AV_FIELD_BB;
472 s->field_order = AV_FIELD_PROGRESSIVE;
475 if (h->picture_structure == PICT_TOP_FIELD)
476 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
478 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
479 s->field_order = AV_FIELD_UNKNOWN;
482 return 0; /* no need to evaluate the rest */
487 /* didn't find a picture! */
488 av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
492 static int h264_parse(AVCodecParserContext *s,
493 AVCodecContext *avctx,
494 const uint8_t **poutbuf, int *poutbuf_size,
495 const uint8_t *buf, int buf_size)
497 H264ParseContext *p = s->priv_data;
498 H264Context *h = &p->h;
499 ParseContext *pc = &p->pc;
504 if (avctx->extradata_size) {
506 // must be done like in decoder, otherwise opening the parser,
507 // letting it create extradata and then closing and opening again
508 // will cause has_b_frames to be always set.
509 // Note that estimate_timings_from_pts does exactly this.
510 if (!avctx->has_b_frames)
512 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
516 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
519 next = h264_find_frame_end(p, buf, buf_size);
521 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
527 if (next < 0 && next != END_NOT_FOUND) {
528 av_assert1(pc->last_index + next >= 0);
529 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
533 parse_nal_units(s, avctx, buf, buf_size);
535 if (avctx->framerate.num)
536 avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
537 if (h->sei_cpb_removal_delay >= 0) {
538 s->dts_sync_point = h->sei_buffering_period_present;
539 s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
540 s->pts_dts_delta = h->sei_dpb_output_delay;
542 s->dts_sync_point = INT_MIN;
543 s->dts_ref_dts_delta = INT_MIN;
544 s->pts_dts_delta = INT_MIN;
547 if (s->flags & PARSER_FLAG_ONCE) {
548 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
552 *poutbuf_size = buf_size;
556 static int h264_split(AVCodecContext *avctx,
557 const uint8_t *buf, int buf_size)
562 const uint8_t *ptr = buf, *end = buf + buf_size;
566 ptr = avpriv_find_start_code(ptr, end, &state);
567 if ((state & 0xFFFFFF00) != 0x100)
569 nalu_type = state & 0x1F;
570 if (nalu_type == NAL_SPS) {
572 } else if (nalu_type == NAL_PPS)
574 /* else if (nalu_type == 0x01 ||
575 * nalu_type == 0x02 ||
576 * nalu_type == 0x05) {
579 else if ((nalu_type != NAL_SEI || has_pps) &&
580 nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
583 while (ptr - 4 > buf && ptr[-5] == 0)
585 return ptr - 4 - buf;
593 static void h264_close(AVCodecParserContext *s)
595 H264ParseContext *p = s->priv_data;
596 H264Context *h = &p->h;
597 ParseContext *pc = &p->pc;
599 av_freep(&pc->buffer);
600 ff_h264_free_context(h);
603 static av_cold int init(AVCodecParserContext *s)
605 H264ParseContext *p = s->priv_data;
606 H264Context *h = &p->h;
608 h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
613 h->thread_context[0] = h;
614 h->slice_context_count = 1;
615 ff_h264dsp_init(&h->h264dsp, 8, 1);
619 AVCodecParser ff_h264_parser = {
620 .codec_ids = { AV_CODEC_ID_H264 },
621 .priv_data_size = sizeof(H264ParseContext),
623 .parser_parse = h264_parse,
624 .parser_close = h264_close,