Merge commit '19f53840001a7afcdb57b2a797e363ec8081aba3'
[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
36
37 static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
38                                int buf_size)
39 {
40     int i, j;
41     uint32_t state;
42     ParseContext *pc = &h->parse_context;
43     int next_avc= h->is_avc ? 0 : buf_size;
44
45 //    mb_addr= pc->mb_addr - 1;
46     state = pc->state;
47     if (state > 13)
48         state = 7;
49
50     if (h->is_avc && !h->nal_length_size)
51         av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
52
53     for (i = 0; i < buf_size; i++) {
54         if (i >= next_avc) {
55             int nalsize = 0;
56             i = next_avc;
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);
61                 return buf_size;
62             }
63             next_avc = i + nalsize;
64             state    = 5;
65         }
66
67         if (state == 7) {
68             i += h->h264dsp.h264_find_start_code_candidate(buf + i, next_avc - i);
69             if (i < next_avc)
70                 state = 2;
71         } else if (state <= 2) {
72             if (buf[i] == 1)
73                 state ^= 5;            // 2->7, 1->4, 0->5
74             else if (buf[i])
75                 state = 7;
76             else
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) {
82                     i++;
83                     goto found;
84                 }
85             } else if (v == 1 || v == 2 || v == 5) {
86                 state += 8;
87                 continue;
88             }
89             state = 7;
90         } else {
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;
94                 GetBitContext gb;
95
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) {
102                     if (mb <= last_mb)
103                         goto found;
104                 } else
105                     pc->frame_start_found = 1;
106                 state = 7;
107             }
108         }
109     }
110     pc->state = state;
111     if (h->is_avc)
112         return next_avc;
113     return END_NOT_FOUND;
114
115 found:
116     pc->state             = 7;
117     pc->frame_start_found = 0;
118     if (h->is_avc)
119         return next_avc;
120     return i - (state & 5) - 3 * (state > 7);
121 }
122
123 static int scan_mmco_reset(AVCodecParserContext *s)
124 {
125     H264Context *h = s->priv_data;
126
127     h->slice_type_nos = s->pict_type & 3;
128
129     if (h->pps.redundant_pic_cnt_present)
130         get_ue_golomb(&h->gb); // redundant_pic_count
131
132     if (ff_set_ref_count(h) < 0)
133         return AVERROR_INVALIDDATA;
134
135     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
136         int list;
137         for (list = 0; list < h->list_count; list++) {
138             if (get_bits1(&h->gb)) {
139                 int index;
140                 for (index = 0; ; index++) {
141                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
142
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;
150                     } else
151                         break;
152
153                     if (index >= h->ref_count[list]) {
154                         av_log(h->avctx, AV_LOG_ERROR, "reference count overflow\n");
155                         return AVERROR_INVALIDDATA;
156                     }
157                 }
158             }
159         }
160     }
161
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);
165
166     if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
167         int i;
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",
173                        opcode);
174                 return AVERROR_INVALIDDATA;
175             }
176             if (opcode == MMCO_END)
177                return 0;
178             else if (opcode == MMCO_RESET)
179                 return 1;
180
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);
186         }
187     }
188
189     return 0;
190 }
191
192 /**
193  * Parse NAL units of found picture and decode some basic information.
194  *
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.
199  */
200 static inline int parse_nal_units(AVCodecParserContext *s,
201                                   AVCodecContext *avctx,
202                                   const uint8_t *buf, int buf_size)
203 {
204     H264Context *h         = s->priv_data;
205     const uint8_t *buf_end = buf + buf_size;
206     unsigned int pps_id;
207     unsigned int slice_type;
208     int state = -1, got_reset = 0;
209     const uint8_t *ptr;
210     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
211     int field_poc[2];
212
213     /* set some sane default values */
214     s->pict_type         = AV_PICTURE_TYPE_I;
215     s->key_frame         = 0;
216     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
217
218     h->avctx                        = avctx;
219     h->sei_recovery_frame_cnt       = -1;
220     h->sei_dpb_output_delay         = 0;
221     h->sei_cpb_removal_delay        = -1;
222     h->sei_buffering_period_present = 0;
223     h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
224
225     if (!buf_size)
226         return 0;
227
228     for (;;) {
229         int src_length, dst_length, consumed, nalsize = 0;
230         if (h->is_avc) {
231             int i;
232             if (h->nal_length_size >= buf_end - buf) break;
233             nalsize = 0;
234             for (i = 0; i < h->nal_length_size; i++)
235                 nalsize = (nalsize << 8) | *buf++;
236             if (nalsize <= 0 || nalsize > buf_end - buf) {
237                 av_log(h->avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
238                 break;
239             }
240             src_length = nalsize;
241         } else {
242         buf = avpriv_find_start_code(buf, buf_end, &state);
243         if (buf >= buf_end)
244             break;
245         --buf;
246         src_length = buf_end - buf;
247         }
248         switch (state & 0x1f) {
249         case NAL_SLICE:
250         case NAL_IDR_SLICE:
251             // Do not walk the whole buffer just to decode slice header
252             if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
253                 /* IDR or disposable slice
254                  * No need to decode many bytes because MMCOs shall not be present. */
255                 if (src_length > 60)
256                     src_length = 60;
257             } else {
258                 /* To decode up to MMCOs */
259                 if (src_length > 1000)
260                     src_length = 1000;
261             }
262             break;
263         }
264         ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
265         if (ptr == NULL || dst_length < 0)
266             break;
267
268         init_get_bits(&h->gb, ptr, 8 * dst_length);
269         switch (h->nal_unit_type) {
270         case NAL_SPS:
271             ff_h264_decode_seq_parameter_set(h);
272             break;
273         case NAL_PPS:
274             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
275             break;
276         case NAL_SEI:
277             ff_h264_decode_sei(h);
278             break;
279         case NAL_IDR_SLICE:
280             s->key_frame = 1;
281
282             h->prev_frame_num        = 0;
283             h->prev_frame_num_offset = 0;
284             h->prev_poc_msb          =
285             h->prev_poc_lsb          = 0;
286         /* fall through */
287         case NAL_SLICE:
288             get_ue_golomb_long(&h->gb);  // skip first_mb_in_slice
289             slice_type   = get_ue_golomb_31(&h->gb);
290             s->pict_type = golomb_to_pict_type[slice_type % 5];
291             if (h->sei_recovery_frame_cnt >= 0) {
292                 /* key frame, since recovery_frame_cnt is set */
293                 s->key_frame = 1;
294             }
295             pps_id = get_ue_golomb(&h->gb);
296             if (pps_id >= MAX_PPS_COUNT) {
297                 av_log(h->avctx, AV_LOG_ERROR,
298                        "pps_id out of range\n");
299                 return -1;
300             }
301             if (!h->pps_buffers[pps_id]) {
302                 av_log(h->avctx, AV_LOG_ERROR,
303                        "non-existing PPS referenced\n");
304                 return -1;
305             }
306             h->pps = *h->pps_buffers[pps_id];
307             if (!h->sps_buffers[h->pps.sps_id]) {
308                 av_log(h->avctx, AV_LOG_ERROR,
309                        "non-existing SPS referenced\n");
310                 return -1;
311             }
312             h->sps       = *h->sps_buffers[h->pps.sps_id];
313             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
314
315             avctx->profile = ff_h264_get_profile(&h->sps);
316             avctx->level   = h->sps.level_idc;
317
318             if (h->sps.frame_mbs_only_flag) {
319                 h->picture_structure = PICT_FRAME;
320             } else {
321                 if (get_bits1(&h->gb)) { // field_pic_flag
322                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
323                 } else {
324                     h->picture_structure = PICT_FRAME;
325                 }
326             }
327
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);
332
333                 if (h->pps.pic_order_present == 1 &&
334                     h->picture_structure == PICT_FRAME)
335                     h->delta_poc_bottom = get_se_golomb(&h->gb);
336             }
337
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);
341
342                 if (h->pps.pic_order_present == 1 &&
343                     h->picture_structure == PICT_FRAME)
344                     h->delta_poc[1] = get_se_golomb(&h->gb);
345             }
346
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);
351
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);
358                 if (got_reset < 0)
359                     return got_reset;
360             }
361
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) {
366                 if (!got_reset) {
367                     h->prev_poc_msb = h->poc_msb;
368                     h->prev_poc_lsb = h->poc_lsb;
369                 } else {
370                     h->prev_poc_msb = 0;
371                     h->prev_poc_lsb =
372                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
373                 }
374             }
375
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:
380                     s->repeat_pict = 0;
381                     break;
382                 case SEI_PIC_STRUCT_FRAME:
383                 case SEI_PIC_STRUCT_TOP_BOTTOM:
384                 case SEI_PIC_STRUCT_BOTTOM_TOP:
385                     s->repeat_pict = 1;
386                     break;
387                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
388                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
389                     s->repeat_pict = 2;
390                     break;
391                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
392                     s->repeat_pict = 3;
393                     break;
394                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
395                     s->repeat_pict = 5;
396                     break;
397                 default:
398                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
399                     break;
400                 }
401             } else {
402                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
403             }
404
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;
412                         break;
413                     case SEI_PIC_STRUCT_BOTTOM_TOP:
414                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
415                         s->field_order = AV_FIELD_BB;
416                         break;
417                     default:
418                         s->field_order = AV_FIELD_PROGRESSIVE;
419                         break;
420                     }
421                 } else {
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;
426                     else
427                         s->field_order = AV_FIELD_PROGRESSIVE;
428                 }
429             } else {
430                 if (h->picture_structure == PICT_TOP_FIELD)
431                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
432                 else
433                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
434                 s->field_order = AV_FIELD_UNKNOWN;
435             }
436
437             return 0; /* no need to evaluate the rest */
438         }
439         buf += h->is_avc ? nalsize : consumed;
440     }
441     if (q264)
442         return 0;
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);
445     return -1;
446 }
447
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)
452 {
453     H264Context *h   = s->priv_data;
454     ParseContext *pc = &h->parse_context;
455     int next;
456
457     if (!h->got_first) {
458         h->got_first = 1;
459         if (avctx->extradata_size) {
460             h->avctx = avctx;
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)
466                 h->low_delay = 1;
467             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
468         }
469     }
470
471     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
472         next = buf_size;
473     } else {
474         next = h264_find_frame_end(h, buf, buf_size);
475
476         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
477             *poutbuf      = NULL;
478             *poutbuf_size = 0;
479             return buf_size;
480         }
481
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
485         }
486     }
487
488     parse_nal_units(s, avctx, buf, buf_size);
489
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;
494     } else {
495         s->dts_sync_point    = INT_MIN;
496         s->dts_ref_dts_delta = INT_MIN;
497         s->pts_dts_delta     = INT_MIN;
498     }
499
500     if (s->flags & PARSER_FLAG_ONCE) {
501         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
502     }
503
504     *poutbuf      = buf;
505     *poutbuf_size = buf_size;
506     return next;
507 }
508
509 static int h264_split(AVCodecContext *avctx,
510                       const uint8_t *buf, int buf_size)
511 {
512     int i;
513     uint32_t state = -1;
514     int has_sps    = 0;
515
516     for (i = 0; i <= buf_size; i++) {
517         if ((state & 0xFFFFFF1F) == 0x107)
518             has_sps = 1;
519         /*  if ((state&0xFFFFFF1F) == 0x101 ||
520          *     (state&0xFFFFFF1F) == 0x102 ||
521          *     (state&0xFFFFFF1F) == 0x105) {
522          *  }
523          */
524         if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
525             (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
526             if (has_sps) {
527                 while (i > 4 && buf[i - 5] == 0)
528                     i--;
529                 return i - 4;
530             }
531         }
532         if (i < buf_size)
533             state = (state << 8) | buf[i];
534     }
535     return 0;
536 }
537
538 static void close(AVCodecParserContext *s)
539 {
540     H264Context *h   = s->priv_data;
541     ParseContext *pc = &h->parse_context;
542
543     av_free(pc->buffer);
544     ff_h264_free_context(h);
545 }
546
547 static av_cold int init(AVCodecParserContext *s)
548 {
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);
553     return 0;
554 }
555
556 AVCodecParser ff_h264_parser = {
557     .codec_ids      = { AV_CODEC_ID_H264 },
558     .priv_data_size = sizeof(H264Context),
559     .parser_init    = init,
560     .parser_parse   = h264_parse,
561     .parser_close   = close,
562     .split          = h264_split,
563 };