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"
37 static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
42 ParseContext *pc = &h->parse_context;
43 int next_avc= h->is_avc ? 0 : buf_size;
45 // mb_addr= pc->mb_addr - 1;
50 if (h->is_avc && !h->nal_length_size)
51 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
53 for (i = 0; i < buf_size; i++) {
57 for (j = 0; j < h->nal_length_size; j++)
58 nalsize = (nalsize << 8) | buf[i++];
59 if (nalsize <= 0 || nalsize > buf_size - i) {
60 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
63 next_avc = i + nalsize;
68 i += h->h264dsp.h264_find_start_code_candidate(buf + i, next_avc - i);
71 } else if (state <= 2) {
73 state ^= 5; // 2->7, 1->4, 0->5
77 state >>= 1; // 2->1, 1->0, 0->0
78 } else if (state <= 5) {
79 int v = buf[i] & 0x1F;
80 if (v == 6 || v == 7 || v == 8 || v == 9) {
81 if (pc->frame_start_found) {
85 } else if (v == 1 || v == 2 || v == 5) {
91 h->parse_history[h->parse_history_count++]= buf[i];
92 if (h->parse_history_count>3) {
93 unsigned int mb, last_mb= h->parse_last_mb;
96 init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
97 h->parse_history_count=0;
98 mb= get_ue_golomb_long(&gb);
99 last_mb= h->parse_last_mb;
100 h->parse_last_mb= mb;
101 if (pc->frame_start_found) {
105 pc->frame_start_found = 1;
113 return END_NOT_FOUND;
117 pc->frame_start_found = 0;
120 return i - (state & 5) - 3 * (state > 7);
123 static int scan_mmco_reset(AVCodecParserContext *s)
125 H264Context *h = s->priv_data;
127 h->slice_type_nos = s->pict_type & 3;
129 if (h->pps.redundant_pic_cnt_present)
130 get_ue_golomb(&h->gb); // redundant_pic_count
132 if (ff_set_ref_count(h) < 0)
133 return AVERROR_INVALIDDATA;
135 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
137 for (list = 0; list < h->list_count; list++) {
138 if (get_bits1(&h->gb)) {
140 for (index = 0; ; index++) {
141 unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
143 if (reordering_of_pic_nums_idc < 3)
144 get_ue_golomb(&h->gb);
145 else if (reordering_of_pic_nums_idc > 3) {
146 av_log(h->avctx, AV_LOG_ERROR,
147 "illegal reordering_of_pic_nums_idc %d\n",
148 reordering_of_pic_nums_idc);
149 return AVERROR_INVALIDDATA;
153 if (index >= h->ref_count[list]) {
154 av_log(h->avctx, AV_LOG_ERROR, "reference count overflow\n");
155 return AVERROR_INVALIDDATA;
162 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
163 (h->pps.weighted_bipred_idc == 1 && h->slice_type_nos == AV_PICTURE_TYPE_B))
164 ff_pred_weight_table(h);
166 if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
168 for (i = 0; i < MAX_MMCO_COUNT; i++) {
169 MMCOOpcode opcode = get_ue_golomb_31(&h->gb);
170 if (opcode > (unsigned) MMCO_LONG) {
171 av_log(h->avctx, AV_LOG_ERROR,
172 "illegal memory management control operation %d\n",
174 return AVERROR_INVALIDDATA;
176 if (opcode == MMCO_END)
178 else if (opcode == MMCO_RESET)
181 if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
182 get_ue_golomb(&h->gb);
183 if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
184 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
185 get_ue_golomb_31(&h->gb);
193 * Parse NAL units of found picture and decode some basic information.
195 * @param s parser context.
196 * @param avctx codec context.
197 * @param buf buffer with field/frame data.
198 * @param buf_size size of the buffer.
200 static inline int parse_nal_units(AVCodecParserContext *s,
201 AVCodecContext *avctx,
202 const uint8_t *buf, int buf_size)
204 H264Context *h = s->priv_data;
205 const uint8_t *buf_end = buf + buf_size;
207 unsigned int slice_type;
208 int state = -1, got_reset = 0;
210 int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
213 /* set some sane default values */
214 s->pict_type = AV_PICTURE_TYPE_I;
216 s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
219 ff_h264_reset_sei(h);
220 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
226 int src_length, dst_length, consumed, nalsize = 0;
229 if (h->nal_length_size >= buf_end - buf) break;
231 for (i = 0; i < h->nal_length_size; i++)
232 nalsize = (nalsize << 8) | *buf++;
233 if (nalsize <= 0 || nalsize > buf_end - buf) {
234 av_log(h->avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
237 src_length = nalsize;
239 buf = avpriv_find_start_code(buf, buf_end, &state);
243 src_length = buf_end - buf;
245 switch (state & 0x1f) {
248 // Do not walk the whole buffer just to decode slice header
249 if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
250 /* IDR or disposable slice
251 * No need to decode many bytes because MMCOs shall not be present. */
255 /* To decode up to MMCOs */
256 if (src_length > 1000)
261 ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
262 if (ptr == NULL || dst_length < 0)
265 init_get_bits(&h->gb, ptr, 8 * dst_length);
266 switch (h->nal_unit_type) {
268 ff_h264_decode_seq_parameter_set(h);
271 ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
274 ff_h264_decode_sei(h);
279 h->prev_frame_num = 0;
280 h->prev_frame_num_offset = 0;
285 get_ue_golomb_long(&h->gb); // skip first_mb_in_slice
286 slice_type = get_ue_golomb_31(&h->gb);
287 s->pict_type = golomb_to_pict_type[slice_type % 5];
288 if (h->sei_recovery_frame_cnt >= 0) {
289 /* key frame, since recovery_frame_cnt is set */
292 pps_id = get_ue_golomb(&h->gb);
293 if (pps_id >= MAX_PPS_COUNT) {
294 av_log(h->avctx, AV_LOG_ERROR,
295 "pps_id out of range\n");
298 if (!h->pps_buffers[pps_id]) {
299 av_log(h->avctx, AV_LOG_ERROR,
300 "non-existing PPS referenced\n");
303 h->pps = *h->pps_buffers[pps_id];
304 if (!h->sps_buffers[h->pps.sps_id]) {
305 av_log(h->avctx, AV_LOG_ERROR,
306 "non-existing SPS referenced\n");
309 h->sps = *h->sps_buffers[h->pps.sps_id];
310 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
312 if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
315 avctx->profile = ff_h264_get_profile(&h->sps);
316 avctx->level = h->sps.level_idc;
318 if (h->sps.frame_mbs_only_flag) {
319 h->picture_structure = PICT_FRAME;
321 if (get_bits1(&h->gb)) { // field_pic_flag
322 h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
324 h->picture_structure = PICT_FRAME;
328 if (h->nal_unit_type == NAL_IDR_SLICE)
329 get_ue_golomb(&h->gb); /* idr_pic_id */
330 if (h->sps.poc_type == 0) {
331 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
333 if (h->pps.pic_order_present == 1 &&
334 h->picture_structure == PICT_FRAME)
335 h->delta_poc_bottom = get_se_golomb(&h->gb);
338 if (h->sps.poc_type == 1 &&
339 !h->sps.delta_pic_order_always_zero_flag) {
340 h->delta_poc[0] = get_se_golomb(&h->gb);
342 if (h->pps.pic_order_present == 1 &&
343 h->picture_structure == PICT_FRAME)
344 h->delta_poc[1] = get_se_golomb(&h->gb);
347 /* Decode POC of this picture.
348 * The prev_ values needed for decoding POC of the next picture are not set here. */
349 field_poc[0] = field_poc[1] = INT_MAX;
350 ff_init_poc(h, field_poc, &s->output_picture_number);
352 /* Continue parsing to check if MMCO_RESET is present.
353 * FIXME: MMCO_RESET could appear in non-first slice.
354 * Maybe, we should parse all undisposable non-IDR slice of this
355 * picture until encountering MMCO_RESET in a slice of it. */
356 if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
357 got_reset = scan_mmco_reset(s);
362 /* Set up the prev_ values for decoding POC of the next picture. */
363 h->prev_frame_num = got_reset ? 0 : h->frame_num;
364 h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
365 if (h->nal_ref_idc != 0) {
367 h->prev_poc_msb = h->poc_msb;
368 h->prev_poc_lsb = h->poc_lsb;
372 h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
376 if (h->sps.pic_struct_present_flag) {
377 switch (h->sei_pic_struct) {
378 case SEI_PIC_STRUCT_TOP_FIELD:
379 case SEI_PIC_STRUCT_BOTTOM_FIELD:
382 case SEI_PIC_STRUCT_FRAME:
383 case SEI_PIC_STRUCT_TOP_BOTTOM:
384 case SEI_PIC_STRUCT_BOTTOM_TOP:
387 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
388 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
391 case SEI_PIC_STRUCT_FRAME_DOUBLING:
394 case SEI_PIC_STRUCT_FRAME_TRIPLING:
398 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
402 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
405 if (h->picture_structure == PICT_FRAME) {
406 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
407 if (h->sps.pic_struct_present_flag) {
408 switch (h->sei_pic_struct) {
409 case SEI_PIC_STRUCT_TOP_BOTTOM:
410 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
411 s->field_order = AV_FIELD_TT;
413 case SEI_PIC_STRUCT_BOTTOM_TOP:
414 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
415 s->field_order = AV_FIELD_BB;
418 s->field_order = AV_FIELD_PROGRESSIVE;
422 if (field_poc[0] < field_poc[1])
423 s->field_order = AV_FIELD_TT;
424 else if (field_poc[0] > field_poc[1])
425 s->field_order = AV_FIELD_BB;
427 s->field_order = AV_FIELD_PROGRESSIVE;
430 if (h->picture_structure == PICT_TOP_FIELD)
431 s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
433 s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
434 s->field_order = AV_FIELD_UNKNOWN;
437 return 0; /* no need to evaluate the rest */
439 buf += h->is_avc ? nalsize : consumed;
443 /* didn't find a picture! */
444 av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
448 static int h264_parse(AVCodecParserContext *s,
449 AVCodecContext *avctx,
450 const uint8_t **poutbuf, int *poutbuf_size,
451 const uint8_t *buf, int buf_size)
453 H264Context *h = s->priv_data;
454 ParseContext *pc = &h->parse_context;
459 if (avctx->extradata_size) {
461 // must be done like in decoder, otherwise opening the parser,
462 // letting it create extradata and then closing and opening again
463 // will cause has_b_frames to be always set.
464 // Note that estimate_timings_from_pts does exactly this.
465 if (!avctx->has_b_frames)
467 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
471 if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
474 next = h264_find_frame_end(h, buf, buf_size);
476 if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
482 if (next < 0 && next != END_NOT_FOUND) {
483 av_assert1(pc->last_index + next >= 0);
484 h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); // update state
488 parse_nal_units(s, avctx, buf, buf_size);
490 if (h->sei_cpb_removal_delay >= 0) {
491 s->dts_sync_point = h->sei_buffering_period_present;
492 s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
493 s->pts_dts_delta = h->sei_dpb_output_delay;
495 s->dts_sync_point = INT_MIN;
496 s->dts_ref_dts_delta = INT_MIN;
497 s->pts_dts_delta = INT_MIN;
500 if (s->flags & PARSER_FLAG_ONCE) {
501 s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
505 *poutbuf_size = buf_size;
509 static int h264_split(AVCodecContext *avctx,
510 const uint8_t *buf, int buf_size)
516 for (i = 0; i <= buf_size; i++) {
517 if ((state & 0xFFFFFF1F) == 0x107)
519 /* if ((state&0xFFFFFF1F) == 0x101 ||
520 * (state&0xFFFFFF1F) == 0x102 ||
521 * (state&0xFFFFFF1F) == 0x105) {
524 if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
525 (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
527 while (i > 4 && buf[i - 5] == 0)
533 state = (state << 8) | buf[i];
538 static void close(AVCodecParserContext *s)
540 H264Context *h = s->priv_data;
541 ParseContext *pc = &h->parse_context;
544 ff_h264_free_context(h);
547 static av_cold int init(AVCodecParserContext *s)
549 H264Context *h = s->priv_data;
550 h->thread_context[0] = h;
551 h->slice_context_count = 1;
552 ff_h264dsp_init(&h->h264dsp, 8, 1);
556 AVCodecParser ff_h264_parser = {
557 .codec_ids = { AV_CODEC_ID_H264 },
558 .priv_data_size = sizeof(H264Context),
560 .parser_parse = h264_parse,
561 .parser_close = close,