h264: add frame packing as stereo_mode frame metadata
[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 #if HAVE_FAST_UNALIGNED
69             /* we check i < buf_size instead of i + 3 / 7 because it is
70              * simpler and there must be FF_INPUT_BUFFER_PADDING_SIZE
71              * bytes at the end.
72              */
73 #    if HAVE_FAST_64BIT
74             while (i < next_avc &&
75                    !((~*(const uint64_t *)(buf + i) &
76                       (*(const uint64_t *)(buf + i) - 0x0101010101010101ULL)) &
77                       0x8080808080808080ULL))
78                 i += 8;
79 #    else
80             while (i < next_avc &&
81                    !((~*(const uint32_t *)(buf + i) &
82                       (*(const uint32_t *)(buf + i) - 0x01010101U)) &
83                       0x80808080U))
84                 i += 4;
85 #    endif
86 #endif
87             for (; i < next_avc; i++)
88                 if (!buf[i]) {
89                     state = 2;
90                     break;
91                 }
92         } else if (state <= 2) {
93             if (buf[i] == 1)
94                 state ^= 5;            // 2->7, 1->4, 0->5
95             else if (buf[i])
96                 state = 7;
97             else
98                 state >>= 1;           // 2->1, 1->0, 0->0
99         } else if (state <= 5) {
100             int v = buf[i] & 0x1F;
101             if (v == 6 || v == 7 || v == 8 || v == 9) {
102                 if (pc->frame_start_found) {
103                     i++;
104                     goto found;
105                 }
106             } else if (v == 1 || v == 2 || v == 5) {
107                 state += 8;
108                 continue;
109             }
110             state = 7;
111         } else {
112             h->parse_history[h->parse_history_count++]= buf[i];
113             if (h->parse_history_count>3) {
114                 unsigned int mb, last_mb= h->parse_last_mb;
115                 GetBitContext gb;
116
117                 init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
118                 h->parse_history_count=0;
119                 mb= get_ue_golomb_long(&gb);
120                 last_mb= h->parse_last_mb;
121                 h->parse_last_mb= mb;
122                 if (pc->frame_start_found) {
123                     if (mb <= last_mb)
124                         goto found;
125                 } else
126                     pc->frame_start_found = 1;
127                 state = 7;
128             }
129         }
130     }
131     pc->state = state;
132     if (h->is_avc)
133         return next_avc;
134     return END_NOT_FOUND;
135
136 found:
137     pc->state             = 7;
138     pc->frame_start_found = 0;
139     if (h->is_avc)
140         return next_avc;
141     return i - (state & 5) - 3 * (state > 7);
142 }
143
144 /**
145  * Parse NAL units of found picture and decode some basic information.
146  *
147  * @param s parser context.
148  * @param avctx codec context.
149  * @param buf buffer with field/frame data.
150  * @param buf_size size of the buffer.
151  */
152 static inline int parse_nal_units(AVCodecParserContext *s,
153                                   AVCodecContext *avctx,
154                                   const uint8_t *buf, int buf_size)
155 {
156     H264Context *h         = s->priv_data;
157     const uint8_t *buf_end = buf + buf_size;
158     unsigned int pps_id;
159     unsigned int slice_type;
160     int state = -1;
161     const uint8_t *ptr;
162     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
163     int field_poc[2];
164
165     /* set some sane default values */
166     s->pict_type         = AV_PICTURE_TYPE_I;
167     s->key_frame         = 0;
168     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
169
170     h->avctx                        = avctx;
171     h->sei_recovery_frame_cnt       = -1;
172     h->sei_dpb_output_delay         = 0;
173     h->sei_cpb_removal_delay        = -1;
174     h->sei_buffering_period_present = 0;
175     h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
176
177     if (!buf_size)
178         return 0;
179
180     for (;;) {
181         int src_length, dst_length, consumed, nalsize = 0;
182         if (h->is_avc) {
183             int i;
184             if (h->nal_length_size >= buf_end - buf) break;
185             nalsize = 0;
186             for (i = 0; i < h->nal_length_size; i++)
187                 nalsize = (nalsize << 8) | *buf++;
188             if (nalsize <= 0 || nalsize > buf_end - buf) {
189                 av_log(h->avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
190                 break;
191             }
192             src_length = nalsize;
193         } else {
194         buf = avpriv_find_start_code(buf, buf_end, &state);
195         if (buf >= buf_end)
196             break;
197         --buf;
198         src_length = buf_end - buf;
199         }
200         switch (state & 0x1f) {
201         case NAL_SLICE:
202         case NAL_IDR_SLICE:
203             // Do not walk the whole buffer just to decode slice header
204             if (src_length > 20)
205                 src_length = 20;
206             break;
207         }
208         ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
209         if (ptr == NULL || dst_length < 0)
210             break;
211
212         init_get_bits(&h->gb, ptr, 8 * dst_length);
213         switch (h->nal_unit_type) {
214         case NAL_SPS:
215             ff_h264_decode_seq_parameter_set(h);
216             break;
217         case NAL_PPS:
218             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
219             break;
220         case NAL_SEI:
221             ff_h264_decode_sei(h);
222             break;
223         case NAL_IDR_SLICE:
224             s->key_frame = 1;
225
226             h->prev_frame_num        = 0;
227             h->prev_frame_num_offset = 0;
228             h->prev_poc_msb          =
229             h->prev_poc_lsb          = 0;
230         /* fall through */
231         case NAL_SLICE:
232             get_ue_golomb_long(&h->gb);  // skip first_mb_in_slice
233             slice_type   = get_ue_golomb_31(&h->gb);
234             s->pict_type = golomb_to_pict_type[slice_type % 5];
235             if (h->sei_recovery_frame_cnt >= 0) {
236                 /* key frame, since recovery_frame_cnt is set */
237                 s->key_frame = 1;
238             }
239             pps_id = get_ue_golomb(&h->gb);
240             if (pps_id >= MAX_PPS_COUNT) {
241                 av_log(h->avctx, AV_LOG_ERROR,
242                        "pps_id out of range\n");
243                 return -1;
244             }
245             if (!h->pps_buffers[pps_id]) {
246                 av_log(h->avctx, AV_LOG_ERROR,
247                        "non-existing PPS referenced\n");
248                 return -1;
249             }
250             h->pps = *h->pps_buffers[pps_id];
251             if (!h->sps_buffers[h->pps.sps_id]) {
252                 av_log(h->avctx, AV_LOG_ERROR,
253                        "non-existing SPS referenced\n");
254                 return -1;
255             }
256             h->sps       = *h->sps_buffers[h->pps.sps_id];
257             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
258
259             avctx->profile = ff_h264_get_profile(&h->sps);
260             avctx->level   = h->sps.level_idc;
261
262             if (h->sps.frame_mbs_only_flag) {
263                 h->picture_structure = PICT_FRAME;
264             } else {
265                 if (get_bits1(&h->gb)) { // field_pic_flag
266                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
267                 } else {
268                     h->picture_structure = PICT_FRAME;
269                 }
270             }
271
272             if (h->nal_unit_type == NAL_IDR_SLICE)
273                 get_ue_golomb(&h->gb); /* idr_pic_id */
274             if (h->sps.poc_type == 0) {
275                 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
276
277                 if (h->pps.pic_order_present == 1 &&
278                     h->picture_structure == PICT_FRAME)
279                     h->delta_poc_bottom = get_se_golomb(&h->gb);
280             }
281
282             if (h->sps.poc_type == 1 &&
283                 !h->sps.delta_pic_order_always_zero_flag) {
284                 h->delta_poc[0] = get_se_golomb(&h->gb);
285
286                 if (h->pps.pic_order_present == 1 &&
287                     h->picture_structure == PICT_FRAME)
288                     h->delta_poc[1] = get_se_golomb(&h->gb);
289             }
290
291             ff_init_poc(h, field_poc, NULL);
292
293             if (h->sps.pic_struct_present_flag) {
294                 switch (h->sei_pic_struct) {
295                 case SEI_PIC_STRUCT_TOP_FIELD:
296                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
297                     s->repeat_pict = 0;
298                     break;
299                 case SEI_PIC_STRUCT_FRAME:
300                 case SEI_PIC_STRUCT_TOP_BOTTOM:
301                 case SEI_PIC_STRUCT_BOTTOM_TOP:
302                     s->repeat_pict = 1;
303                     break;
304                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
305                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
306                     s->repeat_pict = 2;
307                     break;
308                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
309                     s->repeat_pict = 3;
310                     break;
311                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
312                     s->repeat_pict = 5;
313                     break;
314                 default:
315                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
316                     break;
317                 }
318             } else {
319                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
320             }
321
322             if (h->picture_structure == PICT_FRAME) {
323                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
324                 if (h->sps.pic_struct_present_flag) {
325                     switch (h->sei_pic_struct) {
326                     case SEI_PIC_STRUCT_TOP_BOTTOM:
327                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
328                         s->field_order = AV_FIELD_TT;
329                         break;
330                     case SEI_PIC_STRUCT_BOTTOM_TOP:
331                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
332                         s->field_order = AV_FIELD_BB;
333                         break;
334                     default:
335                         s->field_order = AV_FIELD_PROGRESSIVE;
336                         break;
337                     }
338                 } else {
339                     if (field_poc[0] < field_poc[1])
340                         s->field_order = AV_FIELD_TT;
341                     else if (field_poc[0] > field_poc[1])
342                         s->field_order = AV_FIELD_BB;
343                     else
344                         s->field_order = AV_FIELD_PROGRESSIVE;
345                 }
346             } else {
347                 if (h->picture_structure == PICT_TOP_FIELD)
348                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
349                 else
350                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
351                 s->field_order = AV_FIELD_UNKNOWN;
352             }
353
354             return 0; /* no need to evaluate the rest */
355         }
356         buf += h->is_avc ? nalsize : consumed;
357     }
358     if (q264)
359         return 0;
360     /* didn't find a picture! */
361     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
362     return -1;
363 }
364
365 static int h264_parse(AVCodecParserContext *s,
366                       AVCodecContext *avctx,
367                       const uint8_t **poutbuf, int *poutbuf_size,
368                       const uint8_t *buf, int buf_size)
369 {
370     H264Context *h   = s->priv_data;
371     ParseContext *pc = &h->parse_context;
372     int next;
373
374     if (!h->got_first) {
375         h->got_first = 1;
376         if (avctx->extradata_size) {
377             h->avctx = avctx;
378             // must be done like in decoder, otherwise opening the parser,
379             // letting it create extradata and then closing and opening again
380             // will cause has_b_frames to be always set.
381             // Note that estimate_timings_from_pts does exactly this.
382             if (!avctx->has_b_frames)
383                 h->low_delay = 1;
384             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
385         }
386     }
387
388     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
389         next = buf_size;
390     } else {
391         next = h264_find_frame_end(h, buf, buf_size);
392
393         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
394             *poutbuf      = NULL;
395             *poutbuf_size = 0;
396             return buf_size;
397         }
398
399         if (next < 0 && next != END_NOT_FOUND) {
400             av_assert1(pc->last_index + next >= 0);
401             h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); // update state
402         }
403     }
404
405     parse_nal_units(s, avctx, buf, buf_size);
406
407     if (h->sei_cpb_removal_delay >= 0) {
408         s->dts_sync_point    = h->sei_buffering_period_present;
409         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
410         s->pts_dts_delta     = h->sei_dpb_output_delay;
411     } else {
412         s->dts_sync_point    = INT_MIN;
413         s->dts_ref_dts_delta = INT_MIN;
414         s->pts_dts_delta     = INT_MIN;
415     }
416
417     if (s->flags & PARSER_FLAG_ONCE) {
418         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
419     }
420
421     *poutbuf      = buf;
422     *poutbuf_size = buf_size;
423     return next;
424 }
425
426 static int h264_split(AVCodecContext *avctx,
427                       const uint8_t *buf, int buf_size)
428 {
429     int i;
430     uint32_t state = -1;
431     int has_sps    = 0;
432
433     for (i = 0; i <= buf_size; i++) {
434         if ((state & 0xFFFFFF1F) == 0x107)
435             has_sps = 1;
436         /*  if ((state&0xFFFFFF1F) == 0x101 ||
437          *     (state&0xFFFFFF1F) == 0x102 ||
438          *     (state&0xFFFFFF1F) == 0x105) {
439          *  }
440          */
441         if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
442             (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
443             if (has_sps) {
444                 while (i > 4 && buf[i - 5] == 0)
445                     i--;
446                 return i - 4;
447             }
448         }
449         if (i < buf_size)
450             state = (state << 8) | buf[i];
451     }
452     return 0;
453 }
454
455 static void close(AVCodecParserContext *s)
456 {
457     H264Context *h   = s->priv_data;
458     ParseContext *pc = &h->parse_context;
459
460     av_free(pc->buffer);
461     ff_h264_free_context(h);
462 }
463
464 static av_cold int init(AVCodecParserContext *s)
465 {
466     H264Context *h = s->priv_data;
467     h->thread_context[0]   = h;
468     h->slice_context_count = 1;
469     return 0;
470 }
471
472 AVCodecParser ff_h264_parser = {
473     .codec_ids      = { AV_CODEC_ID_H264 },
474     .priv_data_size = sizeof(H264Context),
475     .parser_init    = init,
476     .parser_parse   = h264_parse,
477     .parser_close   = close,
478     .split          = h264_split,
479 };