Merge commit '31d2039cb42668ebcf08248bc48bbad44aa05f49'
[ffmpeg.git] / 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
137     h->slice_type_nos = s->pict_type & 3;
138
139     if (h->pps.redundant_pic_cnt_present)
140         get_ue_golomb(&h->gb); // redundant_pic_count
141
142     if (ff_set_ref_count(h) < 0)
143         return AVERROR_INVALIDDATA;
144
145     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
146         int list;
147         for (list = 0; list < h->list_count; list++) {
148             if (get_bits1(&h->gb)) {
149                 int index;
150                 for (index = 0; ; index++) {
151                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
152
153                     if (reordering_of_pic_nums_idc < 3)
154                         get_ue_golomb(&h->gb);
155                     else if (reordering_of_pic_nums_idc > 3) {
156                         av_log(h->avctx, AV_LOG_ERROR,
157                                "illegal reordering_of_pic_nums_idc %d\n",
158                                reordering_of_pic_nums_idc);
159                         return AVERROR_INVALIDDATA;
160                     } else
161                         break;
162
163                     if (index >= h->ref_count[list]) {
164                         av_log(h->avctx, AV_LOG_ERROR,
165                                "reference count %d overflow\n", index);
166                         return AVERROR_INVALIDDATA;
167                     }
168                 }
169             }
170         }
171     }
172
173     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
174         (h->pps.weighted_bipred_idc == 1 && h->slice_type_nos == AV_PICTURE_TYPE_B))
175         ff_pred_weight_table(h);
176
177     if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
178         int i;
179         for (i = 0; i < MAX_MMCO_COUNT; i++) {
180             MMCOOpcode opcode = get_ue_golomb_31(&h->gb);
181             if (opcode > (unsigned) MMCO_LONG) {
182                 av_log(h->avctx, AV_LOG_ERROR,
183                        "illegal memory management control operation %d\n",
184                        opcode);
185                 return AVERROR_INVALIDDATA;
186             }
187             if (opcode == MMCO_END)
188                return 0;
189             else if (opcode == MMCO_RESET)
190                 return 1;
191
192             if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
193                 get_ue_golomb(&h->gb);
194             if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
195                 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
196                 get_ue_golomb_31(&h->gb);
197         }
198     }
199
200     return 0;
201 }
202
203 /**
204  * Parse NAL units of found picture and decode some basic information.
205  *
206  * @param s parser context.
207  * @param avctx codec context.
208  * @param buf buffer with field/frame data.
209  * @param buf_size size of the buffer.
210  */
211 static inline int parse_nal_units(AVCodecParserContext *s,
212                                   AVCodecContext *avctx,
213                                   const uint8_t * const buf, int buf_size)
214 {
215     H264ParseContext *p = s->priv_data;
216     H264Context      *h = &p->h;
217     int buf_index, next_avc;
218     unsigned int pps_id;
219     unsigned int slice_type;
220     int state = -1, got_reset = 0;
221     const uint8_t *ptr;
222     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
223     int field_poc[2];
224
225     /* set some sane default values */
226     s->pict_type         = AV_PICTURE_TYPE_I;
227     s->key_frame         = 0;
228     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
229
230     h->avctx = avctx;
231     ff_h264_reset_sei(h);
232     h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
233
234     if (!buf_size)
235         return 0;
236
237     buf_index     = 0;
238     next_avc      = h->is_avc ? 0 : buf_size;
239     for (;;) {
240         int src_length, dst_length, consumed, nalsize = 0;
241
242         if (buf_index >= next_avc) {
243             nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
244             if (nalsize < 0)
245                 break;
246             next_avc = buf_index + nalsize;
247         } else {
248             buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
249             if (buf_index >= buf_size)
250                 break;
251             if (buf_index >= next_avc)
252                 continue;
253         }
254         src_length = next_avc - buf_index;
255
256         state = buf[buf_index];
257         switch (state & 0x1f) {
258         case NAL_SLICE:
259         case NAL_IDR_SLICE:
260             // Do not walk the whole buffer just to decode slice header
261             if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
262                 /* IDR or disposable slice
263                  * No need to decode many bytes because MMCOs shall not be present. */
264                 if (src_length > 60)
265                     src_length = 60;
266             } else {
267                 /* To decode up to MMCOs */
268                 if (src_length > 1000)
269                     src_length = 1000;
270             }
271             break;
272         }
273         ptr = ff_h264_decode_nal(h, buf + buf_index, &dst_length,
274                                  &consumed, src_length);
275         if (!ptr || dst_length < 0)
276             break;
277
278         buf_index += consumed;
279
280         init_get_bits(&h->gb, ptr, 8 * dst_length);
281         switch (h->nal_unit_type) {
282         case NAL_SPS:
283             ff_h264_decode_seq_parameter_set(h);
284             break;
285         case NAL_PPS:
286             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
287             break;
288         case NAL_SEI:
289             ff_h264_decode_sei(h);
290             break;
291         case NAL_IDR_SLICE:
292             s->key_frame = 1;
293
294             h->prev_frame_num        = 0;
295             h->prev_frame_num_offset = 0;
296             h->prev_poc_msb          =
297             h->prev_poc_lsb          = 0;
298         /* fall through */
299         case NAL_SLICE:
300             get_ue_golomb_long(&h->gb);  // skip first_mb_in_slice
301             slice_type   = get_ue_golomb_31(&h->gb);
302             s->pict_type = golomb_to_pict_type[slice_type % 5];
303             if (h->sei_recovery_frame_cnt >= 0) {
304                 /* key frame, since recovery_frame_cnt is set */
305                 s->key_frame = 1;
306             }
307             pps_id = get_ue_golomb(&h->gb);
308             if (pps_id >= MAX_PPS_COUNT) {
309                 av_log(h->avctx, AV_LOG_ERROR,
310                        "pps_id %u out of range\n", pps_id);
311                 return -1;
312             }
313             if (!h->pps_buffers[pps_id]) {
314                 av_log(h->avctx, AV_LOG_ERROR,
315                        "non-existing PPS %u referenced\n", pps_id);
316                 return -1;
317             }
318             h->pps = *h->pps_buffers[pps_id];
319             if (!h->sps_buffers[h->pps.sps_id]) {
320                 av_log(h->avctx, AV_LOG_ERROR,
321                        "non-existing SPS %u referenced\n", h->pps.sps_id);
322                 return -1;
323             }
324             h->sps       = *h->sps_buffers[h->pps.sps_id];
325             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
326
327             if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
328                 s->key_frame = 1;
329
330             s->coded_width  = 16 * h->sps.mb_width;
331             s->coded_height = 16 * h->sps.mb_height;
332             s->width        = s->coded_width  - (h->sps.crop_right + h->sps.crop_left);
333             s->height       = s->coded_height - (h->sps.crop_top   + h->sps.crop_bottom);
334             if (s->width <= 0 || s->height <= 0) {
335                 s->width  = s->coded_width;
336                 s->height = s->coded_height;
337             }
338
339             switch (h->sps.bit_depth_luma) {
340             case 9:
341                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P9;
342                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
343                 else                   s->format = AV_PIX_FMT_YUV420P9;
344                 break;
345             case 10:
346                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P10;
347                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
348                 else                   s->format = AV_PIX_FMT_YUV420P10;
349                 break;
350             case 8:
351                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P;
352                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
353                 else                   s->format = AV_PIX_FMT_YUV420P;
354                 break;
355             default:
356                 s->format = AV_PIX_FMT_NONE;
357             }
358
359             avctx->profile = ff_h264_get_profile(&h->sps);
360             avctx->level   = h->sps.level_idc;
361
362             if (h->sps.frame_mbs_only_flag) {
363                 h->picture_structure = PICT_FRAME;
364             } else {
365                 if (get_bits1(&h->gb)) { // field_pic_flag
366                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
367                 } else {
368                     h->picture_structure = PICT_FRAME;
369                 }
370             }
371
372             if (h->nal_unit_type == NAL_IDR_SLICE)
373                 get_ue_golomb(&h->gb); /* idr_pic_id */
374             if (h->sps.poc_type == 0) {
375                 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
376
377                 if (h->pps.pic_order_present == 1 &&
378                     h->picture_structure == PICT_FRAME)
379                     h->delta_poc_bottom = get_se_golomb(&h->gb);
380             }
381
382             if (h->sps.poc_type == 1 &&
383                 !h->sps.delta_pic_order_always_zero_flag) {
384                 h->delta_poc[0] = get_se_golomb(&h->gb);
385
386                 if (h->pps.pic_order_present == 1 &&
387                     h->picture_structure == PICT_FRAME)
388                     h->delta_poc[1] = get_se_golomb(&h->gb);
389             }
390
391             /* Decode POC of this picture.
392              * The prev_ values needed for decoding POC of the next picture are not set here. */
393             field_poc[0] = field_poc[1] = INT_MAX;
394             ff_init_poc(h, field_poc, &s->output_picture_number);
395
396             /* Continue parsing to check if MMCO_RESET is present.
397              * FIXME: MMCO_RESET could appear in non-first slice.
398              *        Maybe, we should parse all undisposable non-IDR slice of this
399              *        picture until encountering MMCO_RESET in a slice of it. */
400             if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
401                 got_reset = scan_mmco_reset(s);
402                 if (got_reset < 0)
403                     return got_reset;
404             }
405
406             /* Set up the prev_ values for decoding POC of the next picture. */
407             h->prev_frame_num        = got_reset ? 0 : h->frame_num;
408             h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
409             if (h->nal_ref_idc != 0) {
410                 if (!got_reset) {
411                     h->prev_poc_msb = h->poc_msb;
412                     h->prev_poc_lsb = h->poc_lsb;
413                 } else {
414                     h->prev_poc_msb = 0;
415                     h->prev_poc_lsb =
416                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
417                 }
418             }
419
420             if (h->sps.pic_struct_present_flag) {
421                 switch (h->sei_pic_struct) {
422                 case SEI_PIC_STRUCT_TOP_FIELD:
423                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
424                     s->repeat_pict = 0;
425                     break;
426                 case SEI_PIC_STRUCT_FRAME:
427                 case SEI_PIC_STRUCT_TOP_BOTTOM:
428                 case SEI_PIC_STRUCT_BOTTOM_TOP:
429                     s->repeat_pict = 1;
430                     break;
431                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
432                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
433                     s->repeat_pict = 2;
434                     break;
435                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
436                     s->repeat_pict = 3;
437                     break;
438                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
439                     s->repeat_pict = 5;
440                     break;
441                 default:
442                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
443                     break;
444                 }
445             } else {
446                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
447             }
448
449             if (h->picture_structure == PICT_FRAME) {
450                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
451                 if (h->sps.pic_struct_present_flag) {
452                     switch (h->sei_pic_struct) {
453                     case SEI_PIC_STRUCT_TOP_BOTTOM:
454                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
455                         s->field_order = AV_FIELD_TT;
456                         break;
457                     case SEI_PIC_STRUCT_BOTTOM_TOP:
458                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
459                         s->field_order = AV_FIELD_BB;
460                         break;
461                     default:
462                         s->field_order = AV_FIELD_PROGRESSIVE;
463                         break;
464                     }
465                 } else {
466                     if (field_poc[0] < field_poc[1])
467                         s->field_order = AV_FIELD_TT;
468                     else if (field_poc[0] > field_poc[1])
469                         s->field_order = AV_FIELD_BB;
470                     else
471                         s->field_order = AV_FIELD_PROGRESSIVE;
472                 }
473             } else {
474                 if (h->picture_structure == PICT_TOP_FIELD)
475                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
476                 else
477                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
478                 s->field_order = AV_FIELD_UNKNOWN;
479             }
480
481             return 0; /* no need to evaluate the rest */
482         }
483     }
484     if (q264)
485         return 0;
486     /* didn't find a picture! */
487     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
488     return -1;
489 }
490
491 static int h264_parse(AVCodecParserContext *s,
492                       AVCodecContext *avctx,
493                       const uint8_t **poutbuf, int *poutbuf_size,
494                       const uint8_t *buf, int buf_size)
495 {
496     H264ParseContext *p = s->priv_data;
497     H264Context      *h = &p->h;
498     ParseContext *pc = &p->pc;
499     int next;
500
501     if (!p->got_first) {
502         p->got_first = 1;
503         if (avctx->extradata_size) {
504             h->avctx = avctx;
505             // must be done like in decoder, otherwise opening the parser,
506             // letting it create extradata and then closing and opening again
507             // will cause has_b_frames to be always set.
508             // Note that estimate_timings_from_pts does exactly this.
509             if (!avctx->has_b_frames)
510                 h->low_delay = 1;
511             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
512         }
513     }
514
515     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
516         next = buf_size;
517     } else {
518         next = h264_find_frame_end(p, buf, buf_size);
519
520         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
521             *poutbuf      = NULL;
522             *poutbuf_size = 0;
523             return buf_size;
524         }
525
526         if (next < 0 && next != END_NOT_FOUND) {
527             av_assert1(pc->last_index + next >= 0);
528             h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
529         }
530     }
531
532     parse_nal_units(s, avctx, buf, buf_size);
533
534     if (avctx->framerate.num)
535         avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
536     if (h->sei_cpb_removal_delay >= 0) {
537         s->dts_sync_point    = h->sei_buffering_period_present;
538         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
539         s->pts_dts_delta     = h->sei_dpb_output_delay;
540     } else {
541         s->dts_sync_point    = INT_MIN;
542         s->dts_ref_dts_delta = INT_MIN;
543         s->pts_dts_delta     = INT_MIN;
544     }
545
546     if (s->flags & PARSER_FLAG_ONCE) {
547         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
548     }
549
550     *poutbuf      = buf;
551     *poutbuf_size = buf_size;
552     return next;
553 }
554
555 static int h264_split(AVCodecContext *avctx,
556                       const uint8_t *buf, int buf_size)
557 {
558     uint32_t state = -1;
559     int has_sps    = 0;
560     int has_pps    = 0;
561     const uint8_t *ptr = buf, *end = buf + buf_size;
562     int nalu_type;
563
564     while (ptr < end) {
565         ptr = avpriv_find_start_code(ptr, end, &state);
566         if ((state & 0xFFFFFF00) != 0x100)
567             break;
568         nalu_type = state & 0x1F;
569         if (nalu_type == NAL_SPS) {
570             has_sps = 1;
571         } else if (nalu_type == NAL_PPS)
572             has_pps = 1;
573         /* else if (nalu_type == 0x01 ||
574          *     nalu_type == 0x02 ||
575          *     nalu_type == 0x05) {
576          *  }
577          */
578         else if ((nalu_type != NAL_SEI || has_pps) &&
579                   nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
580                   nalu_type != 0x0f) {
581             if (has_sps) {
582                 while (ptr - 4 > buf && ptr[-5] == 0)
583                     ptr--;
584                 return ptr - 4 - buf;
585             }
586         }
587     }
588
589     return 0;
590 }
591
592 static void h264_close(AVCodecParserContext *s)
593 {
594     H264ParseContext *p = s->priv_data;
595     H264Context      *h = &p->h;
596     ParseContext *pc = &p->pc;
597
598     av_freep(&pc->buffer);
599     ff_h264_free_context(h);
600 }
601
602 static av_cold int init(AVCodecParserContext *s)
603 {
604     H264ParseContext *p = s->priv_data;
605     H264Context      *h = &p->h;
606     h->thread_context[0]   = h;
607     h->slice_context_count = 1;
608     ff_h264dsp_init(&h->h264dsp, 8, 1);
609     return 0;
610 }
611
612 AVCodecParser ff_h264_parser = {
613     .codec_ids      = { AV_CODEC_ID_H264 },
614     .priv_data_size = sizeof(H264ParseContext),
615     .parser_init    = init,
616     .parser_parse   = h264_parse,
617     .parser_close   = h264_close,
618     .split          = h264_split,
619 };