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>
28 #include "libavutil/attributes.h"
33 #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;
51 // mb_addr= pc->mb_addr - 1;
56 for (i = 0; i < buf_size; i++) {
58 i += h->h264dsp.startcode_find_candidate(buf + i, buf_size - i);
61 } else if (state <= 2) {
63 state ^= 5; // 2->7, 1->4, 0->5
67 state >>= 1; // 2->1, 1->0, 0->0
68 } else if (state <= 5) {
69 int nalu_type = buf[i] & 0x1F;
70 if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
71 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
72 if (pc->frame_start_found) {
76 } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
77 nalu_type == NAL_IDR_SLICE) {
78 if (pc->frame_start_found) {
82 pc->frame_start_found = 1;
86 // first_mb_in_slice is 0, probably the first nal of a new slice
97 pc->frame_start_found = 0;
98 return i - (state & 5);
101 static int scan_mmco_reset(AVCodecParserContext *s)
103 H264ParseContext *p = s->priv_data;
104 H264Context *h = &p->h;
105 H264SliceContext *sl = &h->slice_ctx[0];
107 sl->slice_type_nos = s->pict_type & 3;
109 if (h->pps.redundant_pic_cnt_present)
110 get_ue_golomb(&sl->gb); // redundant_pic_count
112 if (ff_set_ref_count(h, sl) < 0)
113 return AVERROR_INVALIDDATA;
115 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
117 for (list = 0; list < sl->list_count; list++) {
118 if (get_bits1(&sl->gb)) {
120 for (index = 0; ; index++) {
121 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&sl->gb);
123 if (reordering_of_pic_nums_idc < 3)
124 get_ue_golomb(&sl->gb);
125 else if (reordering_of_pic_nums_idc > 3) {
126 av_log(h->avctx, AV_LOG_ERROR,
127 "illegal reordering_of_pic_nums_idc %d\n",
128 reordering_of_pic_nums_idc);
129 return AVERROR_INVALIDDATA;
133 if (index >= sl->ref_count[list]) {
134 av_log(h->avctx, AV_LOG_ERROR,
135 "reference count %d overflow\n", index);
136 return AVERROR_INVALIDDATA;
143 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
144 (h->pps.weighted_bipred_idc == 1 && sl->slice_type_nos == AV_PICTURE_TYPE_B))
145 ff_pred_weight_table(h, sl);
147 if (get_bits1(&sl->gb)) { // adaptive_ref_pic_marking_mode_flag
149 for (i = 0; i < MAX_MMCO_COUNT; i++) {
150 MMCOOpcode opcode = get_ue_golomb_31(&sl->gb);
151 if (opcode > (unsigned) MMCO_LONG) {
152 av_log(h->avctx, AV_LOG_ERROR,
153 "illegal memory management control operation %d\n",
155 return AVERROR_INVALIDDATA;
157 if (opcode == MMCO_END)
159 else if (opcode == MMCO_RESET)
162 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
163 get_ue_golomb(&sl->gb);
164 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
165 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
166 get_ue_golomb_31(&sl->gb);
174 * Parse NAL units of found picture and decode some basic information.
176 * @param s parser context.
177 * @param avctx codec context.
178 * @param buf buffer with field/frame data.
179 * @param buf_size size of the buffer.
181 static inline int parse_nal_units(AVCodecParserContext *s,
182 AVCodecContext *avctx,
183 const uint8_t *buf, int buf_size)
185 H264ParseContext *p = s->priv_data;
186 H264Context *h = &p->h;
187 H264SliceContext *sl = &h->slice_ctx[0];
188 const uint8_t *buf_end = buf + buf_size;
190 unsigned int slice_type;
191 int state = -1, got_reset = 0;
195 /* set some sane default values */
196 s->pict_type = AV_PICTURE_TYPE_I;
198 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
201 ff_h264_reset_sei(h);
207 int src_length, dst_length, consumed;
208 buf = avpriv_find_start_code(buf, buf_end, &state);
212 src_length = buf_end - buf;
213 switch (state & 0x1f) {
216 // Do not walk the whole buffer just to decode slice header
217 if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
218 /* IDR or disposable slice
219 * No need to decode many bytes because MMCOs shall not be present. */
223 /* To decode up to MMCOs */
224 if (src_length > 1000)
229 ptr = ff_h264_decode_nal(h, sl, buf, &dst_length, &consumed, src_length);
230 if (!ptr || dst_length < 0)
233 init_get_bits(&h->gb, ptr, 8 * dst_length);
234 switch (h->nal_unit_type) {
236 ff_h264_decode_seq_parameter_set(h);
239 ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
242 ff_h264_decode_sei(h);
247 h->prev_frame_num = 0;
248 h->prev_frame_num_offset = 0;
253 init_get_bits(&sl->gb, ptr, 8 * dst_length);
254 get_ue_golomb(&sl->gb); // skip first_mb_in_slice
255 slice_type = get_ue_golomb_31(&sl->gb);
256 s->pict_type = golomb_to_pict_type[slice_type % 5];
257 if (h->sei_recovery_frame_cnt >= 0) {
258 /* key frame, since recovery_frame_cnt is set */
261 pps_id = get_ue_golomb(&sl->gb);
262 if (pps_id >= MAX_PPS_COUNT) {
263 av_log(h->avctx, AV_LOG_ERROR,
264 "pps_id %u out of range\n", pps_id);
267 if (!h->pps_buffers[pps_id]) {
268 av_log(h->avctx, AV_LOG_ERROR,
269 "non-existing PPS %u referenced\n", pps_id);
272 h->pps = *h->pps_buffers[pps_id];
273 if (!h->sps_buffers[h->pps.sps_id]) {
274 av_log(h->avctx, AV_LOG_ERROR,
275 "non-existing SPS %u referenced\n", h->pps.sps_id);
278 h->sps = *h->sps_buffers[h->pps.sps_id];
279 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
281 s->coded_width = 16 * h->sps.mb_width;
282 s->coded_height = 16 * h->sps.mb_height;
283 s->width = s->coded_width - (h->sps.crop_right + h->sps.crop_left);
284 s->height = s->coded_height - (h->sps.crop_top + h->sps.crop_bottom);
285 if (s->width <= 0 || s->height <= 0) {
286 s->width = s->coded_width;
287 s->height = s->coded_height;
290 switch (h->sps.bit_depth_luma) {
292 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P9;
293 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
294 else s->format = AV_PIX_FMT_YUV420P9;
297 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P10;
298 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
299 else s->format = AV_PIX_FMT_YUV420P10;
302 if (CHROMA444(h)) s->format = AV_PIX_FMT_YUV444P;
303 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
304 else s->format = AV_PIX_FMT_YUV420P;
307 s->format = AV_PIX_FMT_NONE;
310 avctx->profile = ff_h264_get_profile(&h->sps);
311 avctx->level = h->sps.level_idc;
313 if (h->sps.frame_mbs_only_flag) {
314 h->picture_structure = PICT_FRAME;
316 if (get_bits1(&sl->gb)) { // field_pic_flag
317 h->picture_structure = PICT_TOP_FIELD + get_bits1(&sl->gb); // bottom_field_flag
319 h->picture_structure = PICT_FRAME;
323 if (h->nal_unit_type == NAL_IDR_SLICE)
324 get_ue_golomb(&sl->gb); /* idr_pic_id */
325 if (h->sps.poc_type == 0) {
326 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
328 if (h->pps.pic_order_present == 1 &&
329 h->picture_structure == PICT_FRAME)
330 h->delta_poc_bottom = get_se_golomb(&sl->gb);
333 if (h->sps.poc_type == 1 &&
334 !h->sps.delta_pic_order_always_zero_flag) {
335 h->delta_poc[0] = get_se_golomb(&sl->gb);
337 if (h->pps.pic_order_present == 1 &&
338 h->picture_structure == PICT_FRAME)
339 h->delta_poc[1] = get_se_golomb(&sl->gb);
342 /* Decode POC of this picture.
343 * The prev_ values needed for decoding POC of the next picture are not set here. */
344 field_poc[0] = field_poc[1] = INT_MAX;
345 ff_init_poc(h, field_poc, &s->output_picture_number);
347 /* Continue parsing to check if MMCO_RESET is present.
348 * FIXME: MMCO_RESET could appear in non-first slice.
349 * Maybe, we should parse all undisposable non-IDR slice of this
350 * picture until encountering MMCO_RESET in a slice of it. */
351 if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
352 got_reset = scan_mmco_reset(s);
357 /* Set up the prev_ values for decoding POC of the next picture. */
358 h->prev_frame_num = got_reset ? 0 : h->frame_num;
359 h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
360 if (h->nal_ref_idc != 0) {
362 h->prev_poc_msb = h->poc_msb;
363 h->prev_poc_lsb = h->poc_lsb;
367 h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
371 if (h->sps.pic_struct_present_flag) {
372 switch (h->sei_pic_struct) {
373 case SEI_PIC_STRUCT_TOP_FIELD:
374 case SEI_PIC_STRUCT_BOTTOM_FIELD:
377 case SEI_PIC_STRUCT_FRAME:
378 case SEI_PIC_STRUCT_TOP_BOTTOM:
379 case SEI_PIC_STRUCT_BOTTOM_TOP:
382 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
383 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
386 case SEI_PIC_STRUCT_FRAME_DOUBLING:
389 case SEI_PIC_STRUCT_FRAME_TRIPLING:
393 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
397 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
400 if (h->picture_structure == PICT_FRAME) {
401 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
402 if (h->sps.pic_struct_present_flag) {
403 switch (h->sei_pic_struct) {
404 case SEI_PIC_STRUCT_TOP_BOTTOM:
405 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
406 s->field_order = AV_FIELD_TT;
408 case SEI_PIC_STRUCT_BOTTOM_TOP:
409 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
410 s->field_order = AV_FIELD_BB;
413 s->field_order = AV_FIELD_PROGRESSIVE;
417 if (field_poc[0] < field_poc[1])
418 s->field_order = AV_FIELD_TT;
419 else if (field_poc[0] > field_poc[1])
420 s->field_order = AV_FIELD_BB;
422 s->field_order = AV_FIELD_PROGRESSIVE;
425 if (h->picture_structure == PICT_TOP_FIELD)
426 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
428 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
429 s->field_order = AV_FIELD_UNKNOWN;
432 return 0; /* no need to evaluate the rest */
436 /* didn't find a picture! */
437 av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
441 static int h264_parse(AVCodecParserContext *s,
442 AVCodecContext *avctx,
443 const uint8_t **poutbuf, int *poutbuf_size,
444 const uint8_t *buf, int buf_size)
446 H264ParseContext *p = s->priv_data;
447 H264Context *h = &p->h;
448 ParseContext *pc = &p->pc;
453 if (avctx->extradata_size) {
455 // must be done like in the decoder.
456 // otherwise opening the parser, creating extradata,
457 // and then closing and opening again
458 // will cause has_b_frames to be always set.
459 // NB: estimate_timings_from_pts behaves exactly like this.
460 if (!avctx->has_b_frames)
462 ff_h264_decode_extradata(h);
466 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
469 next = h264_find_frame_end(p, buf, buf_size);
471 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
477 if (next < 0 && next != END_NOT_FOUND) {
478 assert(pc->last_index + next >= 0);
479 h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
483 parse_nal_units(s, avctx, buf, buf_size);
485 if (h->sei_cpb_removal_delay >= 0) {
486 s->dts_sync_point = h->sei_buffering_period_present;
487 s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
488 s->pts_dts_delta = h->sei_dpb_output_delay;
490 s->dts_sync_point = INT_MIN;
491 s->dts_ref_dts_delta = INT_MIN;
492 s->pts_dts_delta = INT_MIN;
495 if (s->flags & PARSER_FLAG_ONCE) {
496 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
500 *poutbuf_size = buf_size;
504 static int h264_split(AVCodecContext *avctx,
505 const uint8_t *buf, int buf_size)
511 for (i = 0; i <= buf_size; i++) {
512 if ((state & 0xFFFFFF1F) == 0x107)
514 /* if((state&0xFFFFFF1F) == 0x101 ||
515 * (state&0xFFFFFF1F) == 0x102 ||
516 * (state&0xFFFFFF1F) == 0x105) {
519 if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x106 &&
520 (state & 0xFFFFFF1F) != 0x107 && (state & 0xFFFFFF1F) != 0x108 &&
521 (state & 0xFFFFFF1F) != 0x109 && (state & 0xFFFFFF1F) != 0x10d &&
522 (state & 0xFFFFFF1F) != 0x10f) {
524 while (i > 4 && buf[i - 5] == 0)
530 state = (state << 8) | buf[i];
535 static void close(AVCodecParserContext *s)
537 H264ParseContext *p = s->priv_data;
538 H264Context *h = &p->h;
539 ParseContext *pc = &p->pc;
542 ff_h264_free_context(h);
545 static av_cold int init(AVCodecParserContext *s)
547 H264ParseContext *p = s->priv_data;
548 H264Context *h = &p->h;
550 h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
555 h->slice_context_count = 1;
556 ff_h264dsp_init(&h->h264dsp, 8, 1);
560 AVCodecParser ff_h264_parser = {
561 .codec_ids = { AV_CODEC_ID_H264 },
562 .priv_data_size = sizeof(H264ParseContext),
564 .parser_parse = h264_parse,
565 .parser_close = close,