]> git.sesse.net Git - ffmpeg/blob - libavcodec/h264_parser.c
Merge commit '64c81b2cd0dcf1fe66c381a5d2c707dddcf35a7e'
[ffmpeg] / libavcodec / h264_parser.c
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... parser
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
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.
11  *
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.
16  *
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
20  */
21
22 /**
23  * @file
24  * H.264 / AVC / MPEG4 part10 parser.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27
28 #define UNCHECKED_BITSTREAM_READER 1
29
30 #include "libavutil/attributes.h"
31 #include "parser.h"
32 #include "h264data.h"
33 #include "golomb.h"
34 #include "internal.h"
35 #include "mpegutils.h"
36
37 typedef struct H264ParseContext {
38     H264Context h;
39     ParseContext pc;
40     int got_first;
41 } H264ParseContext;
42
43
44 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
45                                int buf_size)
46 {
47     H264Context *h = &p->h;
48     int i, j;
49     uint32_t state;
50     ParseContext *pc = &p->pc;
51
52     int next_avc= h->is_avc ? 0 : buf_size;
53 //    mb_addr= pc->mb_addr - 1;
54     state = pc->state;
55     if (state > 13)
56         state = 7;
57
58     if (h->is_avc && !h->nal_length_size)
59         av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
60
61     for (i = 0; i < buf_size; i++) {
62         if (i >= next_avc) {
63             int nalsize = 0;
64             i = next_avc;
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);
69                 return buf_size;
70             }
71             next_avc = i + nalsize;
72             state    = 5;
73         }
74
75         if (state == 7) {
76             i += h->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
77             if (i < next_avc)
78                 state = 2;
79         } else if (state <= 2) {
80             if (buf[i] == 1)
81                 state ^= 5;            // 2->7, 1->4, 0->5
82             else if (buf[i])
83                 state = 7;
84             else
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) {
91                     i++;
92                     goto found;
93                 }
94             } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
95                        nalu_type == NAL_IDR_SLICE) {
96                 state += 8;
97                 continue;
98             }
99             state = 7;
100         } else {
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;
104                 GetBitContext gb;
105
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) {
111                     if (mb <= last_mb)
112                         goto found;
113                 } else
114                     pc->frame_start_found = 1;
115                 state = 7;
116             }
117         }
118     }
119     pc->state = state;
120     if (h->is_avc)
121         return next_avc;
122     return END_NOT_FOUND;
123
124 found:
125     pc->state             = 7;
126     pc->frame_start_found = 0;
127     if (h->is_avc)
128         return next_avc;
129     return i - (state & 5) - 5 * (state > 7);
130 }
131
132 static int scan_mmco_reset(AVCodecParserContext *s)
133 {
134     H264ParseContext *p = s->priv_data;
135     H264Context      *h = &p->h;
136     H264SliceContext *sl = &h->slice_ctx[0];
137
138     h->slice_type_nos = s->pict_type & 3;
139
140     if (h->pps.redundant_pic_cnt_present)
141         get_ue_golomb(&h->gb); // redundant_pic_count
142
143     if (ff_set_ref_count(h) < 0)
144         return AVERROR_INVALIDDATA;
145
146     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
147         int list;
148         for (list = 0; list < h->list_count; list++) {
149             if (get_bits1(&h->gb)) {
150                 int index;
151                 for (index = 0; ; index++) {
152                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
153
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;
161                     } else
162                         break;
163
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;
168                     }
169                 }
170             }
171         }
172     }
173
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);
177
178     if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
179         int i;
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",
185                        opcode);
186                 return AVERROR_INVALIDDATA;
187             }
188             if (opcode == MMCO_END)
189                return 0;
190             else if (opcode == MMCO_RESET)
191                 return 1;
192
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);
198         }
199     }
200
201     return 0;
202 }
203
204 /**
205  * Parse NAL units of found picture and decode some basic information.
206  *
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.
211  */
212 static inline int parse_nal_units(AVCodecParserContext *s,
213                                   AVCodecContext *avctx,
214                                   const uint8_t * const buf, int buf_size)
215 {
216     H264ParseContext *p = s->priv_data;
217     H264Context      *h = &p->h;
218     int buf_index, next_avc;
219     unsigned int pps_id;
220     unsigned int slice_type;
221     int state = -1, got_reset = 0;
222     const uint8_t *ptr;
223     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
224     int field_poc[2];
225
226     /* set some sane default values */
227     s->pict_type         = AV_PICTURE_TYPE_I;
228     s->key_frame         = 0;
229     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
230
231     h->avctx = avctx;
232     ff_h264_reset_sei(h);
233     h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
234
235     if (!buf_size)
236         return 0;
237
238     buf_index     = 0;
239     next_avc      = h->is_avc ? 0 : buf_size;
240     for (;;) {
241         int src_length, dst_length, consumed, nalsize = 0;
242
243         if (buf_index >= next_avc) {
244             nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
245             if (nalsize < 0)
246                 break;
247             next_avc = buf_index + nalsize;
248         } else {
249             buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
250             if (buf_index >= buf_size)
251                 break;
252             if (buf_index >= next_avc)
253                 continue;
254         }
255         src_length = next_avc - buf_index;
256
257         state = buf[buf_index];
258         switch (state & 0x1f) {
259         case NAL_SLICE:
260         case NAL_IDR_SLICE:
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. */
265                 if (src_length > 60)
266                     src_length = 60;
267             } else {
268                 /* To decode up to MMCOs */
269                 if (src_length > 1000)
270                     src_length = 1000;
271             }
272             break;
273         }
274         ptr = ff_h264_decode_nal(h, buf + buf_index, &dst_length,
275                                  &consumed, src_length);
276         if (!ptr || dst_length < 0)
277             break;
278
279         buf_index += consumed;
280
281         init_get_bits(&h->gb, ptr, 8 * dst_length);
282         switch (h->nal_unit_type) {
283         case NAL_SPS:
284             ff_h264_decode_seq_parameter_set(h);
285             break;
286         case NAL_PPS:
287             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
288             break;
289         case NAL_SEI:
290             ff_h264_decode_sei(h);
291             break;
292         case NAL_IDR_SLICE:
293             s->key_frame = 1;
294
295             h->prev_frame_num        = 0;
296             h->prev_frame_num_offset = 0;
297             h->prev_poc_msb          =
298             h->prev_poc_lsb          = 0;
299         /* fall through */
300         case NAL_SLICE:
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 */
306                 s->key_frame = 1;
307             }
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);
312                 return -1;
313             }
314             if (!h->pps_buffers[pps_id]) {
315                 av_log(h->avctx, AV_LOG_ERROR,
316                        "non-existing PPS %u referenced\n", pps_id);
317                 return -1;
318             }
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);
323                 return -1;
324             }
325             h->sps       = *h->sps_buffers[h->pps.sps_id];
326             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
327
328             if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
329                 s->key_frame = 1;
330
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;
338             }
339
340             switch (h->sps.bit_depth_luma) {
341             case 9:
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;
345                 break;
346             case 10:
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;
350                 break;
351             case 8:
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;
355                 break;
356             default:
357                 s->format = AV_PIX_FMT_NONE;
358             }
359
360             avctx->profile = ff_h264_get_profile(&h->sps);
361             avctx->level   = h->sps.level_idc;
362
363             if (h->sps.frame_mbs_only_flag) {
364                 h->picture_structure = PICT_FRAME;
365             } else {
366                 if (get_bits1(&h->gb)) { // field_pic_flag
367                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
368                 } else {
369                     h->picture_structure = PICT_FRAME;
370                 }
371             }
372
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);
377
378                 if (h->pps.pic_order_present == 1 &&
379                     h->picture_structure == PICT_FRAME)
380                     h->delta_poc_bottom = get_se_golomb(&h->gb);
381             }
382
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);
386
387                 if (h->pps.pic_order_present == 1 &&
388                     h->picture_structure == PICT_FRAME)
389                     h->delta_poc[1] = get_se_golomb(&h->gb);
390             }
391
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);
396
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);
403                 if (got_reset < 0)
404                     return got_reset;
405             }
406
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) {
411                 if (!got_reset) {
412                     h->prev_poc_msb = h->poc_msb;
413                     h->prev_poc_lsb = h->poc_lsb;
414                 } else {
415                     h->prev_poc_msb = 0;
416                     h->prev_poc_lsb =
417                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
418                 }
419             }
420
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:
425                     s->repeat_pict = 0;
426                     break;
427                 case SEI_PIC_STRUCT_FRAME:
428                 case SEI_PIC_STRUCT_TOP_BOTTOM:
429                 case SEI_PIC_STRUCT_BOTTOM_TOP:
430                     s->repeat_pict = 1;
431                     break;
432                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
433                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
434                     s->repeat_pict = 2;
435                     break;
436                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
437                     s->repeat_pict = 3;
438                     break;
439                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
440                     s->repeat_pict = 5;
441                     break;
442                 default:
443                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
444                     break;
445                 }
446             } else {
447                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
448             }
449
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;
457                         break;
458                     case SEI_PIC_STRUCT_BOTTOM_TOP:
459                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
460                         s->field_order = AV_FIELD_BB;
461                         break;
462                     default:
463                         s->field_order = AV_FIELD_PROGRESSIVE;
464                         break;
465                     }
466                 } else {
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;
471                     else
472                         s->field_order = AV_FIELD_PROGRESSIVE;
473                 }
474             } else {
475                 if (h->picture_structure == PICT_TOP_FIELD)
476                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
477                 else
478                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
479                 s->field_order = AV_FIELD_UNKNOWN;
480             }
481
482             return 0; /* no need to evaluate the rest */
483         }
484     }
485     if (q264)
486         return 0;
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);
489     return -1;
490 }
491
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)
496 {
497     H264ParseContext *p = s->priv_data;
498     H264Context      *h = &p->h;
499     ParseContext *pc = &p->pc;
500     int next;
501
502     if (!p->got_first) {
503         p->got_first = 1;
504         if (avctx->extradata_size) {
505             h->avctx = avctx;
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)
511                 h->low_delay = 1;
512             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
513         }
514     }
515
516     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
517         next = buf_size;
518     } else {
519         next = h264_find_frame_end(p, buf, buf_size);
520
521         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
522             *poutbuf      = NULL;
523             *poutbuf_size = 0;
524             return buf_size;
525         }
526
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
530         }
531     }
532
533     parse_nal_units(s, avctx, buf, buf_size);
534
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;
541     } else {
542         s->dts_sync_point    = INT_MIN;
543         s->dts_ref_dts_delta = INT_MIN;
544         s->pts_dts_delta     = INT_MIN;
545     }
546
547     if (s->flags & PARSER_FLAG_ONCE) {
548         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
549     }
550
551     *poutbuf      = buf;
552     *poutbuf_size = buf_size;
553     return next;
554 }
555
556 static int h264_split(AVCodecContext *avctx,
557                       const uint8_t *buf, int buf_size)
558 {
559     uint32_t state = -1;
560     int has_sps    = 0;
561     int has_pps    = 0;
562     const uint8_t *ptr = buf, *end = buf + buf_size;
563     int nalu_type;
564
565     while (ptr < end) {
566         ptr = avpriv_find_start_code(ptr, end, &state);
567         if ((state & 0xFFFFFF00) != 0x100)
568             break;
569         nalu_type = state & 0x1F;
570         if (nalu_type == NAL_SPS) {
571             has_sps = 1;
572         } else if (nalu_type == NAL_PPS)
573             has_pps = 1;
574         /* else if (nalu_type == 0x01 ||
575          *     nalu_type == 0x02 ||
576          *     nalu_type == 0x05) {
577          *  }
578          */
579         else if ((nalu_type != NAL_SEI || has_pps) &&
580                   nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
581                   nalu_type != 0x0f) {
582             if (has_sps) {
583                 while (ptr - 4 > buf && ptr[-5] == 0)
584                     ptr--;
585                 return ptr - 4 - buf;
586             }
587         }
588     }
589
590     return 0;
591 }
592
593 static void h264_close(AVCodecParserContext *s)
594 {
595     H264ParseContext *p = s->priv_data;
596     H264Context      *h = &p->h;
597     ParseContext *pc = &p->pc;
598
599     av_freep(&pc->buffer);
600     ff_h264_free_context(h);
601 }
602
603 static av_cold int init(AVCodecParserContext *s)
604 {
605     H264ParseContext *p = s->priv_data;
606     H264Context      *h = &p->h;
607
608     h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
609     if (!h->slice_ctx)
610         return 0;
611     h->nb_slice_ctx = 1;
612
613     h->thread_context[0]   = h;
614     h->slice_context_count = 1;
615     ff_h264dsp_init(&h->h264dsp, 8, 1);
616     return 0;
617 }
618
619 AVCodecParser ff_h264_parser = {
620     .codec_ids      = { AV_CODEC_ID_H264 },
621     .priv_data_size = sizeof(H264ParseContext),
622     .parser_init    = init,
623     .parser_parse   = h264_parse,
624     .parser_close   = h264_close,
625     .split          = h264_split,
626 };