h264: move reading direct_spatial_mv_pred out of ff_set_ref_count()
[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 Libav.
6  *
7  * Libav 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  * Libav 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 Libav; 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 #include <assert.h>
29 #include <stdint.h>
30
31 #include "libavutil/avutil.h"
32 #include "libavutil/error.h"
33 #include "libavutil/log.h"
34 #include "libavutil/mem.h"
35 #include "libavutil/pixfmt.h"
36
37 #include "get_bits.h"
38 #include "golomb.h"
39 #include "h264.h"
40 #include "h264data.h"
41 #include "internal.h"
42 #include "mpegutils.h"
43 #include "parser.h"
44
45 typedef struct H264ParseContext {
46     H264Context h;
47     ParseContext pc;
48     int got_first;
49 } H264ParseContext;
50
51
52 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
53                                int buf_size)
54 {
55     H264Context *h = &p->h;
56     int i;
57     uint32_t state;
58     ParseContext *pc = &p->pc;
59 //    mb_addr= pc->mb_addr - 1;
60     state = pc->state;
61     if (state > 13)
62         state = 7;
63
64     for (i = 0; i < buf_size; i++) {
65         if (state == 7) {
66             i += h->h264dsp.startcode_find_candidate(buf + i, buf_size - i);
67             if (i < buf_size)
68                 state = 2;
69         } else if (state <= 2) {
70             if (buf[i] == 1)
71                 state ^= 5;            // 2->7, 1->4, 0->5
72             else if (buf[i])
73                 state = 7;
74             else
75                 state >>= 1;           // 2->1, 1->0, 0->0
76         } else if (state <= 5) {
77             int nalu_type = buf[i] & 0x1F;
78             if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
79                 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
80                 if (pc->frame_start_found) {
81                     i++;
82                     goto found;
83                 }
84             } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
85                        nalu_type == NAL_IDR_SLICE) {
86                 if (pc->frame_start_found) {
87                     state += 8;
88                     continue;
89                 } else
90                     pc->frame_start_found = 1;
91             }
92             state = 7;
93         } else {
94             // first_mb_in_slice is 0, probably the first nal of a new slice
95             if (buf[i] & 0x80)
96                 goto found;
97             state = 7;
98         }
99     }
100     pc->state = state;
101     return END_NOT_FOUND;
102
103 found:
104     pc->state             = 7;
105     pc->frame_start_found = 0;
106     return i - (state & 5);
107 }
108
109 static int scan_mmco_reset(AVCodecParserContext *s)
110 {
111     H264ParseContext *p = s->priv_data;
112     H264Context      *h = &p->h;
113     H264SliceContext *sl = &h->slice_ctx[0];
114
115     sl->slice_type_nos = s->pict_type & 3;
116
117     if (h->pps.redundant_pic_cnt_present)
118         get_ue_golomb(&sl->gb); // redundant_pic_count
119
120     if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
121         get_bits1(&sl->gb); // direct_spatial_mv_pred
122
123     if (ff_set_ref_count(h, sl) < 0)
124         return AVERROR_INVALIDDATA;
125
126     if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
127         int list;
128         for (list = 0; list < sl->list_count; list++) {
129             if (get_bits1(&sl->gb)) {
130                 int index;
131                 for (index = 0; ; index++) {
132                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&sl->gb);
133
134                     if (reordering_of_pic_nums_idc < 3)
135                         get_ue_golomb(&sl->gb);
136                     else if (reordering_of_pic_nums_idc > 3) {
137                         av_log(h->avctx, AV_LOG_ERROR,
138                                "illegal reordering_of_pic_nums_idc %d\n",
139                                reordering_of_pic_nums_idc);
140                         return AVERROR_INVALIDDATA;
141                     } else
142                         break;
143
144                     if (index >= sl->ref_count[list]) {
145                         av_log(h->avctx, AV_LOG_ERROR,
146                                "reference count %d overflow\n", index);
147                         return AVERROR_INVALIDDATA;
148                     }
149                 }
150             }
151         }
152     }
153
154     if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
155         (h->pps.weighted_bipred_idc == 1 && sl->slice_type_nos == AV_PICTURE_TYPE_B))
156         ff_h264_pred_weight_table(&sl->gb, &h->sps, sl->ref_count, sl->slice_type_nos,
157                                   &sl->pwt);
158
159     if (get_bits1(&sl->gb)) { // adaptive_ref_pic_marking_mode_flag
160         int i;
161         for (i = 0; i < MAX_MMCO_COUNT; i++) {
162             MMCOOpcode opcode = get_ue_golomb_31(&sl->gb);
163             if (opcode > (unsigned) MMCO_LONG) {
164                 av_log(h->avctx, AV_LOG_ERROR,
165                        "illegal memory management control operation %d\n",
166                        opcode);
167                 return AVERROR_INVALIDDATA;
168             }
169             if (opcode == MMCO_END)
170                return 0;
171             else if (opcode == MMCO_RESET)
172                 return 1;
173
174             if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
175                 get_ue_golomb(&sl->gb);
176             if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
177                 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
178                 get_ue_golomb_31(&sl->gb);
179         }
180     }
181
182     return 0;
183 }
184
185 /**
186  * Parse NAL units of found picture and decode some basic information.
187  *
188  * @param s parser context.
189  * @param avctx codec context.
190  * @param buf buffer with field/frame data.
191  * @param buf_size size of the buffer.
192  */
193 static inline int parse_nal_units(AVCodecParserContext *s,
194                                   AVCodecContext *avctx,
195                                   const uint8_t *buf, int buf_size)
196 {
197     H264ParseContext *p = s->priv_data;
198     H264Context      *h = &p->h;
199     H264SliceContext *sl = &h->slice_ctx[0];
200     const uint8_t *buf_end = buf + buf_size;
201
202     H2645NAL nal = { NULL };
203
204     unsigned int pps_id;
205     unsigned int slice_type;
206     int state = -1, got_reset = 0;
207     int field_poc[2];
208     int ret;
209
210     /* set some sane default values */
211     s->pict_type         = AV_PICTURE_TYPE_I;
212     s->key_frame         = 0;
213     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
214
215     h->avctx = avctx;
216     ff_h264_reset_sei(h);
217
218     if (!buf_size)
219         return 0;
220
221     for (;;) {
222         int src_length, consumed;
223         buf = avpriv_find_start_code(buf, buf_end, &state);
224         if (buf >= buf_end)
225             break;
226         --buf;
227         src_length = buf_end - buf;
228         switch (state & 0x1f) {
229         case NAL_SLICE:
230         case NAL_IDR_SLICE:
231             // Do not walk the whole buffer just to decode slice header
232             if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
233                 /* IDR or disposable slice
234                  * No need to decode many bytes because MMCOs shall not be present. */
235                 if (src_length > 60)
236                     src_length = 60;
237             } else {
238                 /* To decode up to MMCOs */
239                 if (src_length > 1000)
240                     src_length = 1000;
241             }
242             break;
243         }
244
245         consumed = ff_h2645_extract_rbsp(buf, src_length, &nal);
246         if (consumed < 0)
247             break;
248
249         ret = init_get_bits(&nal.gb, nal.data, nal.size * 8);
250         if (ret < 0)
251             goto fail;
252         get_bits1(&nal.gb);
253         nal.ref_idc = get_bits(&nal.gb, 2);
254         nal.type    = get_bits(&nal.gb, 5);
255
256         h->gb            = nal.gb;
257         h->nal_ref_idc   = nal.ref_idc;
258         h->nal_unit_type = nal.type;
259
260         switch (h->nal_unit_type) {
261         case NAL_SPS:
262             ff_h264_decode_seq_parameter_set(h);
263             break;
264         case NAL_PPS:
265             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
266             break;
267         case NAL_SEI:
268             ff_h264_decode_sei(h);
269             break;
270         case NAL_IDR_SLICE:
271             s->key_frame = 1;
272
273             h->prev_frame_num        = 0;
274             h->prev_frame_num_offset = 0;
275             h->prev_poc_msb          =
276             h->prev_poc_lsb          = 0;
277         /* fall through */
278         case NAL_SLICE:
279             sl->gb = nal.gb;
280             get_ue_golomb(&sl->gb);  // skip first_mb_in_slice
281             slice_type   = get_ue_golomb_31(&sl->gb);
282             s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
283             if (h->sei_recovery_frame_cnt >= 0) {
284                 /* key frame, since recovery_frame_cnt is set */
285                 s->key_frame = 1;
286             }
287             pps_id = get_ue_golomb(&sl->gb);
288             if (pps_id >= MAX_PPS_COUNT) {
289                 av_log(h->avctx, AV_LOG_ERROR,
290                        "pps_id %u out of range\n", pps_id);
291                 goto fail;
292             }
293             if (!h->pps_buffers[pps_id]) {
294                 av_log(h->avctx, AV_LOG_ERROR,
295                        "non-existing PPS %u referenced\n", pps_id);
296                 goto fail;
297             }
298             h->pps = *h->pps_buffers[pps_id];
299             if (!h->sps_buffers[h->pps.sps_id]) {
300                 av_log(h->avctx, AV_LOG_ERROR,
301                        "non-existing SPS %u referenced\n", h->pps.sps_id);
302                 goto fail;
303             }
304             h->sps       = *h->sps_buffers[h->pps.sps_id];
305             h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
306
307             s->coded_width  = 16 * h->sps.mb_width;
308             s->coded_height = 16 * h->sps.mb_height;
309             s->width        = s->coded_width  - (h->sps.crop_right + h->sps.crop_left);
310             s->height       = s->coded_height - (h->sps.crop_top   + h->sps.crop_bottom);
311             if (s->width <= 0 || s->height <= 0) {
312                 s->width  = s->coded_width;
313                 s->height = s->coded_height;
314             }
315
316             switch (h->sps.bit_depth_luma) {
317             case 9:
318                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P9;
319                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
320                 else                   s->format = AV_PIX_FMT_YUV420P9;
321                 break;
322             case 10:
323                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P10;
324                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
325                 else                   s->format = AV_PIX_FMT_YUV420P10;
326                 break;
327             case 8:
328                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P;
329                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
330                 else                   s->format = AV_PIX_FMT_YUV420P;
331                 break;
332             default:
333                 s->format = AV_PIX_FMT_NONE;
334             }
335
336             avctx->profile = ff_h264_get_profile(&h->sps);
337             avctx->level   = h->sps.level_idc;
338
339             if (h->sps.frame_mbs_only_flag) {
340                 h->picture_structure = PICT_FRAME;
341             } else {
342                 if (get_bits1(&sl->gb)) { // field_pic_flag
343                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&sl->gb); // bottom_field_flag
344                 } else {
345                     h->picture_structure = PICT_FRAME;
346                 }
347             }
348
349             if (h->nal_unit_type == NAL_IDR_SLICE)
350                 get_ue_golomb(&sl->gb); /* idr_pic_id */
351             if (h->sps.poc_type == 0) {
352                 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
353
354                 if (h->pps.pic_order_present == 1 &&
355                     h->picture_structure == PICT_FRAME)
356                     h->delta_poc_bottom = get_se_golomb(&sl->gb);
357             }
358
359             if (h->sps.poc_type == 1 &&
360                 !h->sps.delta_pic_order_always_zero_flag) {
361                 h->delta_poc[0] = get_se_golomb(&sl->gb);
362
363                 if (h->pps.pic_order_present == 1 &&
364                     h->picture_structure == PICT_FRAME)
365                     h->delta_poc[1] = get_se_golomb(&sl->gb);
366             }
367
368             /* Decode POC of this picture.
369              * The prev_ values needed for decoding POC of the next picture are not set here. */
370             field_poc[0] = field_poc[1] = INT_MAX;
371             ff_init_poc(h, field_poc, &s->output_picture_number);
372
373             /* Continue parsing to check if MMCO_RESET is present.
374              * FIXME: MMCO_RESET could appear in non-first slice.
375              *        Maybe, we should parse all undisposable non-IDR slice of this
376              *        picture until encountering MMCO_RESET in a slice of it. */
377             if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
378                 got_reset = scan_mmco_reset(s);
379                 if (got_reset < 0)
380                     goto fail;
381             }
382
383             /* Set up the prev_ values for decoding POC of the next picture. */
384             h->prev_frame_num        = got_reset ? 0 : h->frame_num;
385             h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
386             if (h->nal_ref_idc != 0) {
387                 if (!got_reset) {
388                     h->prev_poc_msb = h->poc_msb;
389                     h->prev_poc_lsb = h->poc_lsb;
390                 } else {
391                     h->prev_poc_msb = 0;
392                     h->prev_poc_lsb =
393                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
394                 }
395             }
396
397             if (h->sps.pic_struct_present_flag) {
398                 switch (h->sei_pic_struct) {
399                 case SEI_PIC_STRUCT_TOP_FIELD:
400                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
401                     s->repeat_pict = 0;
402                     break;
403                 case SEI_PIC_STRUCT_FRAME:
404                 case SEI_PIC_STRUCT_TOP_BOTTOM:
405                 case SEI_PIC_STRUCT_BOTTOM_TOP:
406                     s->repeat_pict = 1;
407                     break;
408                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
409                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
410                     s->repeat_pict = 2;
411                     break;
412                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
413                     s->repeat_pict = 3;
414                     break;
415                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
416                     s->repeat_pict = 5;
417                     break;
418                 default:
419                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
420                     break;
421                 }
422             } else {
423                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
424             }
425
426             if (h->picture_structure == PICT_FRAME) {
427                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
428                 if (h->sps.pic_struct_present_flag) {
429                     switch (h->sei_pic_struct) {
430                     case SEI_PIC_STRUCT_TOP_BOTTOM:
431                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
432                         s->field_order = AV_FIELD_TT;
433                         break;
434                     case SEI_PIC_STRUCT_BOTTOM_TOP:
435                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
436                         s->field_order = AV_FIELD_BB;
437                         break;
438                     default:
439                         s->field_order = AV_FIELD_PROGRESSIVE;
440                         break;
441                     }
442                 } else {
443                     if (field_poc[0] < field_poc[1])
444                         s->field_order = AV_FIELD_TT;
445                     else if (field_poc[0] > field_poc[1])
446                         s->field_order = AV_FIELD_BB;
447                     else
448                         s->field_order = AV_FIELD_PROGRESSIVE;
449                 }
450             } else {
451                 if (h->picture_structure == PICT_TOP_FIELD)
452                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
453                 else
454                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
455                 s->field_order = AV_FIELD_UNKNOWN;
456             }
457
458             av_freep(&nal.rbsp_buffer);
459             return 0; /* no need to evaluate the rest */
460         }
461         buf += consumed;
462     }
463     /* didn't find a picture! */
464     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
465 fail:
466     av_freep(&nal.rbsp_buffer);
467     return -1;
468 }
469
470 static int h264_parse(AVCodecParserContext *s,
471                       AVCodecContext *avctx,
472                       const uint8_t **poutbuf, int *poutbuf_size,
473                       const uint8_t *buf, int buf_size)
474 {
475     H264ParseContext *p = s->priv_data;
476     H264Context      *h = &p->h;
477     ParseContext *pc = &p->pc;
478     int next;
479
480     if (!p->got_first) {
481         p->got_first = 1;
482         if (avctx->extradata_size) {
483             h->avctx = avctx;
484             // must be done like in the decoder.
485             // otherwise opening the parser, creating extradata,
486             // and then closing and opening again
487             // will cause has_b_frames to be always set.
488             // NB: estimate_timings_from_pts behaves exactly like this.
489             if (!avctx->has_b_frames)
490                 h->low_delay = 1;
491             ff_h264_decode_extradata(h);
492         }
493     }
494
495     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
496         next = buf_size;
497     } else {
498         next = h264_find_frame_end(p, buf, buf_size);
499
500         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
501             *poutbuf      = NULL;
502             *poutbuf_size = 0;
503             return buf_size;
504         }
505
506         if (next < 0 && next != END_NOT_FOUND) {
507             assert(pc->last_index + next >= 0);
508             h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
509         }
510     }
511
512     parse_nal_units(s, avctx, buf, buf_size);
513
514     if (h->sei_cpb_removal_delay >= 0) {
515         s->dts_sync_point    = h->sei_buffering_period_present;
516         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
517         s->pts_dts_delta     = h->sei_dpb_output_delay;
518     } else {
519         s->dts_sync_point    = INT_MIN;
520         s->dts_ref_dts_delta = INT_MIN;
521         s->pts_dts_delta     = INT_MIN;
522     }
523
524     if (s->flags & PARSER_FLAG_ONCE) {
525         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
526     }
527
528     *poutbuf      = buf;
529     *poutbuf_size = buf_size;
530     return next;
531 }
532
533 static int h264_split(AVCodecContext *avctx,
534                       const uint8_t *buf, int buf_size)
535 {
536     int i;
537     uint32_t state = -1;
538     int has_sps    = 0;
539
540     for (i = 0; i <= buf_size; i++) {
541         if ((state & 0xFFFFFF1F) == 0x107)
542             has_sps = 1;
543         /*  if((state&0xFFFFFF1F) == 0x101 ||
544          *     (state&0xFFFFFF1F) == 0x102 ||
545          *     (state&0xFFFFFF1F) == 0x105) {
546          *  }
547          */
548         if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x106 &&
549             (state & 0xFFFFFF1F) != 0x107 && (state & 0xFFFFFF1F) != 0x108 &&
550             (state & 0xFFFFFF1F) != 0x109 && (state & 0xFFFFFF1F) != 0x10d &&
551             (state & 0xFFFFFF1F) != 0x10f) {
552             if (has_sps) {
553                 while (i > 4 && buf[i - 5] == 0)
554                     i--;
555                 return i - 4;
556             }
557         }
558         if (i < buf_size)
559             state = (state << 8) | buf[i];
560     }
561     return 0;
562 }
563
564 static void h264_close(AVCodecParserContext *s)
565 {
566     H264ParseContext *p = s->priv_data;
567     H264Context      *h = &p->h;
568     ParseContext *pc = &p->pc;
569
570     av_free(pc->buffer);
571     ff_h264_free_context(h);
572 }
573
574 static av_cold int init(AVCodecParserContext *s)
575 {
576     H264ParseContext *p = s->priv_data;
577     H264Context      *h = &p->h;
578
579     h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
580     if (!h->slice_ctx)
581         return 0;
582     h->nb_slice_ctx = 1;
583
584     h->slice_context_count = 1;
585     ff_h264dsp_init(&h->h264dsp, 8, 1);
586     return 0;
587 }
588
589 AVCodecParser ff_h264_parser = {
590     .codec_ids      = { AV_CODEC_ID_H264 },
591     .priv_data_size = sizeof(H264ParseContext),
592     .parser_init    = init,
593     .parser_parse   = h264_parse,
594     .parser_close   = h264_close,
595     .split          = h264_split,
596 };