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 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 parser.
25 * @author Michael Niedermayer <michaelni@gmx.at>
31 #include "libavutil/avutil.h"
32 #include "libavutil/error.h"
33 #include "libavutil/log.h"
34 #include "libavutil/mem.h"
35 #include "libavutil/pixfmt.h"
42 #include "mpegutils.h"
45 typedef struct H264ParseContext {
52 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
55 H264Context *h = &p->h;
58 ParseContext *pc = &p->pc;
59 // mb_addr= pc->mb_addr - 1;
64 for (i = 0; i < buf_size; i++) {
66 i += h->h264dsp.startcode_find_candidate(buf + i, buf_size - i);
69 } else if (state <= 2) {
71 state ^= 5; // 2->7, 1->4, 0->5
75 state >>= 1; // 2->1, 1->0, 0->0
76 } else if (state <= 5) {
77 int nalu_type = buf[i] & 0x1F;
78 if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
79 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
80 if (pc->frame_start_found) {
84 } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
85 nalu_type == NAL_IDR_SLICE) {
86 if (pc->frame_start_found) {
90 pc->frame_start_found = 1;
94 // first_mb_in_slice is 0, probably the first nal of a new slice
101 return END_NOT_FOUND;
105 pc->frame_start_found = 0;
106 return i - (state & 5);
109 static int scan_mmco_reset(AVCodecParserContext *s)
111 H264ParseContext *p = s->priv_data;
112 H264Context *h = &p->h;
113 H264SliceContext *sl = &h->slice_ctx[0];
114 int list_count, ref_count[2];
116 sl->slice_type_nos = s->pict_type & 3;
118 if (h->pps.redundant_pic_cnt_present)
119 get_ue_golomb(&sl->gb); // redundant_pic_count
121 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
122 get_bits1(&sl->gb); // direct_spatial_mv_pred
124 if (ff_h264_parse_ref_count(&list_count, ref_count, &sl->gb, &h->pps,
125 sl->slice_type_nos, h->picture_structure) < 0)
126 return AVERROR_INVALIDDATA;
128 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
130 for (list = 0; list < list_count; list++) {
131 if (get_bits1(&sl->gb)) {
133 for (index = 0; ; index++) {
134 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&sl->gb);
136 if (reordering_of_pic_nums_idc < 3)
137 get_ue_golomb(&sl->gb);
138 else if (reordering_of_pic_nums_idc > 3) {
139 av_log(h->avctx, AV_LOG_ERROR,
140 "illegal reordering_of_pic_nums_idc %d\n",
141 reordering_of_pic_nums_idc);
142 return AVERROR_INVALIDDATA;
146 if (index >= ref_count[list]) {
147 av_log(h->avctx, AV_LOG_ERROR,
148 "reference count %d overflow\n", index);
149 return AVERROR_INVALIDDATA;
156 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
157 (h->pps.weighted_bipred_idc == 1 && sl->slice_type_nos == AV_PICTURE_TYPE_B))
158 ff_h264_pred_weight_table(&sl->gb, &h->sps, ref_count, sl->slice_type_nos,
161 if (get_bits1(&sl->gb)) { // adaptive_ref_pic_marking_mode_flag
163 for (i = 0; i < MAX_MMCO_COUNT; i++) {
164 MMCOOpcode opcode = get_ue_golomb_31(&sl->gb);
165 if (opcode > (unsigned) MMCO_LONG) {
166 av_log(h->avctx, AV_LOG_ERROR,
167 "illegal memory management control operation %d\n",
169 return AVERROR_INVALIDDATA;
171 if (opcode == MMCO_END)
173 else if (opcode == MMCO_RESET)
176 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
177 get_ue_golomb(&sl->gb);
178 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
179 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
180 get_ue_golomb_31(&sl->gb);
188 * Parse NAL units of found picture and decode some basic information.
190 * @param s parser context.
191 * @param avctx codec context.
192 * @param buf buffer with field/frame data.
193 * @param buf_size size of the buffer.
195 static inline int parse_nal_units(AVCodecParserContext *s,
196 AVCodecContext *avctx,
197 const uint8_t *buf, int buf_size)
199 H264ParseContext *p = s->priv_data;
200 H264Context *h = &p->h;
201 H264SliceContext *sl = &h->slice_ctx[0];
202 const uint8_t *buf_end = buf + buf_size;
204 H2645NAL nal = { NULL };
207 unsigned int slice_type;
208 int state = -1, got_reset = 0;
212 /* set some sane default values */
213 s->pict_type = AV_PICTURE_TYPE_I;
215 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
218 ff_h264_reset_sei(h);
224 int src_length, consumed;
225 buf = avpriv_find_start_code(buf, buf_end, &state);
229 src_length = buf_end - buf;
230 switch (state & 0x1f) {
233 // Do not walk the whole buffer just to decode slice header
234 if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
235 /* IDR or disposable slice
236 * No need to decode many bytes because MMCOs shall not be present. */
240 /* To decode up to MMCOs */
241 if (src_length > 1000)
247 consumed = ff_h2645_extract_rbsp(buf, src_length, &nal);
251 ret = init_get_bits(&nal.gb, nal.data, nal.size * 8);
255 nal.ref_idc = get_bits(&nal.gb, 2);
256 nal.type = get_bits(&nal.gb, 5);
259 h->nal_ref_idc = nal.ref_idc;
260 h->nal_unit_type = nal.type;
262 switch (h->nal_unit_type) {
264 ff_h264_decode_seq_parameter_set(h);
267 ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
270 ff_h264_decode_sei(h);
275 h->prev_frame_num = 0;
276 h->prev_frame_num_offset = 0;
282 get_ue_golomb(&sl->gb); // skip first_mb_in_slice
283 slice_type = get_ue_golomb_31(&sl->gb);
284 s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
285 if (h->sei_recovery_frame_cnt >= 0) {
286 /* key frame, since recovery_frame_cnt is set */
289 pps_id = get_ue_golomb(&sl->gb);
290 if (pps_id >= MAX_PPS_COUNT) {
291 av_log(h->avctx, AV_LOG_ERROR,
292 "pps_id %u out of range\n", pps_id);
295 if (!h->pps_buffers[pps_id]) {
296 av_log(h->avctx, AV_LOG_ERROR,
297 "non-existing PPS %u referenced\n", pps_id);
300 h->pps = *h->pps_buffers[pps_id];
301 if (!h->sps_buffers[h->pps.sps_id]) {
302 av_log(h->avctx, AV_LOG_ERROR,
303 "non-existing SPS %u referenced\n", h->pps.sps_id);
306 h->sps = *h->sps_buffers[h->pps.sps_id];
307 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
309 s->coded_width = 16 * h->sps.mb_width;
310 s->coded_height = 16 * h->sps.mb_height;
311 s->width = s->coded_width - (h->sps.crop_right + h->sps.crop_left);
312 s->height = s->coded_height - (h->sps.crop_top + h->sps.crop_bottom);
313 if (s->width <= 0 || s->height <= 0) {
314 s->width = s->coded_width;
315 s->height = s->coded_height;
318 switch (h->sps.bit_depth_luma) {
320 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P9;
321 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
322 else s->format = AV_PIX_FMT_YUV420P9;
325 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P10;
326 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
327 else s->format = AV_PIX_FMT_YUV420P10;
330 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P;
331 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
332 else s->format = AV_PIX_FMT_YUV420P;
335 s->format = AV_PIX_FMT_NONE;
338 avctx->profile = ff_h264_get_profile(&h->sps);
339 avctx->level = h->sps.level_idc;
341 if (h->sps.frame_mbs_only_flag) {
342 h->picture_structure = PICT_FRAME;
344 if (get_bits1(&sl->gb)) { // field_pic_flag
345 h->picture_structure = PICT_TOP_FIELD + get_bits1(&sl->gb); // bottom_field_flag
347 h->picture_structure = PICT_FRAME;
351 if (h->nal_unit_type == NAL_IDR_SLICE)
352 get_ue_golomb(&sl->gb); /* idr_pic_id */
353 if (h->sps.poc_type == 0) {
354 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
356 if (h->pps.pic_order_present == 1 &&
357 h->picture_structure == PICT_FRAME)
358 h->delta_poc_bottom = get_se_golomb(&sl->gb);
361 if (h->sps.poc_type == 1 &&
362 !h->sps.delta_pic_order_always_zero_flag) {
363 h->delta_poc[0] = get_se_golomb(&sl->gb);
365 if (h->pps.pic_order_present == 1 &&
366 h->picture_structure == PICT_FRAME)
367 h->delta_poc[1] = get_se_golomb(&sl->gb);
370 /* Decode POC of this picture.
371 * The prev_ values needed for decoding POC of the next picture are not set here. */
372 field_poc[0] = field_poc[1] = INT_MAX;
373 ff_init_poc(h, field_poc, &s->output_picture_number);
375 /* Continue parsing to check if MMCO_RESET is present.
376 * FIXME: MMCO_RESET could appear in non-first slice.
377 * Maybe, we should parse all undisposable non-IDR slice of this
378 * picture until encountering MMCO_RESET in a slice of it. */
379 if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
380 got_reset = scan_mmco_reset(s);
385 /* Set up the prev_ values for decoding POC of the next picture. */
386 h->prev_frame_num = got_reset ? 0 : h->frame_num;
387 h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
388 if (h->nal_ref_idc != 0) {
390 h->prev_poc_msb = h->poc_msb;
391 h->prev_poc_lsb = h->poc_lsb;
395 h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
399 if (h->sps.pic_struct_present_flag) {
400 switch (h->sei_pic_struct) {
401 case SEI_PIC_STRUCT_TOP_FIELD:
402 case SEI_PIC_STRUCT_BOTTOM_FIELD:
405 case SEI_PIC_STRUCT_FRAME:
406 case SEI_PIC_STRUCT_TOP_BOTTOM:
407 case SEI_PIC_STRUCT_BOTTOM_TOP:
410 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
411 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
414 case SEI_PIC_STRUCT_FRAME_DOUBLING:
417 case SEI_PIC_STRUCT_FRAME_TRIPLING:
421 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
425 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
428 if (h->picture_structure == PICT_FRAME) {
429 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
430 if (h->sps.pic_struct_present_flag) {
431 switch (h->sei_pic_struct) {
432 case SEI_PIC_STRUCT_TOP_BOTTOM:
433 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
434 s->field_order = AV_FIELD_TT;
436 case SEI_PIC_STRUCT_BOTTOM_TOP:
437 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
438 s->field_order = AV_FIELD_BB;
441 s->field_order = AV_FIELD_PROGRESSIVE;
445 if (field_poc[0] < field_poc[1])
446 s->field_order = AV_FIELD_TT;
447 else if (field_poc[0] > field_poc[1])
448 s->field_order = AV_FIELD_BB;
450 s->field_order = AV_FIELD_PROGRESSIVE;
453 if (h->picture_structure == PICT_TOP_FIELD)
454 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
456 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
457 s->field_order = AV_FIELD_UNKNOWN;
460 av_freep(&nal.rbsp_buffer);
461 return 0; /* no need to evaluate the rest */
465 /* didn't find a picture! */
466 av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
468 av_freep(&nal.rbsp_buffer);
472 static int h264_parse(AVCodecParserContext *s,
473 AVCodecContext *avctx,
474 const uint8_t **poutbuf, int *poutbuf_size,
475 const uint8_t *buf, int buf_size)
477 H264ParseContext *p = s->priv_data;
478 H264Context *h = &p->h;
479 ParseContext *pc = &p->pc;
484 if (avctx->extradata_size) {
486 // must be done like in the decoder.
487 // otherwise opening the parser, creating extradata,
488 // and then closing and opening again
489 // will cause has_b_frames to be always set.
490 // NB: estimate_timings_from_pts behaves exactly like this.
491 if (!avctx->has_b_frames)
493 ff_h264_decode_extradata(h);
497 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
500 next = h264_find_frame_end(p, buf, buf_size);
502 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
508 if (next < 0 && next != END_NOT_FOUND) {
509 assert(pc->last_index + next >= 0);
510 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
514 parse_nal_units(s, avctx, buf, buf_size);
516 if (h->sei_cpb_removal_delay >= 0) {
517 s->dts_sync_point = h->sei_buffering_period_present;
518 s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
519 s->pts_dts_delta = h->sei_dpb_output_delay;
521 s->dts_sync_point = INT_MIN;
522 s->dts_ref_dts_delta = INT_MIN;
523 s->pts_dts_delta = INT_MIN;
526 if (s->flags & PARSER_FLAG_ONCE) {
527 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
531 *poutbuf_size = buf_size;
535 static int h264_split(AVCodecContext *avctx,
536 const uint8_t *buf, int buf_size)
542 for (i = 0; i <= buf_size; i++) {
543 if ((state & 0xFFFFFF1F) == 0x107)
545 /* if((state&0xFFFFFF1F) == 0x101 ||
546 * (state&0xFFFFFF1F) == 0x102 ||
547 * (state&0xFFFFFF1F) == 0x105) {
550 if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x106 &&
551 (state & 0xFFFFFF1F) != 0x107 && (state & 0xFFFFFF1F) != 0x108 &&
552 (state & 0xFFFFFF1F) != 0x109 && (state & 0xFFFFFF1F) != 0x10d &&
553 (state & 0xFFFFFF1F) != 0x10f) {
555 while (i > 4 && buf[i - 5] == 0)
561 state = (state << 8) | buf[i];
566 static void h264_close(AVCodecParserContext *s)
568 H264ParseContext *p = s->priv_data;
569 H264Context *h = &p->h;
570 ParseContext *pc = &p->pc;
573 ff_h264_free_context(h);
576 static av_cold int init(AVCodecParserContext *s)
578 H264ParseContext *p = s->priv_data;
579 H264Context *h = &p->h;
581 h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
586 h->slice_context_count = 1;
587 ff_h264dsp_init(&h->h264dsp, 8, 1);
591 AVCodecParser ff_h264_parser = {
592 .codec_ids = { AV_CODEC_ID_H264 },
593 .priv_data_size = sizeof(H264ParseContext),
595 .parser_parse = h264_parse,
596 .parser_close = h264_close,