h264: move the ref lists variables into the per-slice context
[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 "libavutil/attributes.h"
29 #include "parser.h"
30 #include "h264data.h"
31 #include "golomb.h"
32 #include "internal.h"
33 #include "mpegutils.h"
34
35 #include <assert.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;
49     uint32_t state;
50     ParseContext *pc = &p->pc;
51 //    mb_addr= pc->mb_addr - 1;
52     state = pc->state;
53     if (state > 13)
54         state = 7;
55
56     for (i = 0; i < buf_size; i++) {
57         if (state == 7) {
58             i += h->h264dsp.startcode_find_candidate(buf + i, buf_size - i);
59             if (i < buf_size)
60                 state = 2;
61         } else if (state <= 2) {
62             if (buf[i] == 1)
63                 state ^= 5;            // 2->7, 1->4, 0->5
64             else if (buf[i])
65                 state = 7;
66             else
67                 state >>= 1;           // 2->1, 1->0, 0->0
68         } else if (state <= 5) {
69             int nalu_type = buf[i] & 0x1F;
70             if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
71                 nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
72                 if (pc->frame_start_found) {
73                     i++;
74                     goto found;
75                 }
76             } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
77                        nalu_type == NAL_IDR_SLICE) {
78                 if (pc->frame_start_found) {
79                     state += 8;
80                     continue;
81                 } else
82                     pc->frame_start_found = 1;
83             }
84             state = 7;
85         } else {
86             // first_mb_in_slice is 0, probably the first nal of a new slice
87             if (buf[i] & 0x80)
88                 goto found;
89             state = 7;
90         }
91     }
92     pc->state = state;
93     return END_NOT_FOUND;
94
95 found:
96     pc->state             = 7;
97     pc->frame_start_found = 0;
98     return i - (state & 5);
99 }
100
101 static int scan_mmco_reset(AVCodecParserContext *s)
102 {
103     H264ParseContext *p = s->priv_data;
104     H264Context      *h = &p->h;
105     H264SliceContext *sl = &h->slice_ctx[0];
106
107     sl->slice_type_nos = s->pict_type & 3;
108
109     if (h->pps.redundant_pic_cnt_present)
110         get_ue_golomb(&h->gb); // redundant_pic_count
111
112     if (ff_set_ref_count(h, sl) < 0)
113         return AVERROR_INVALIDDATA;
114
115     if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
116         int list;
117         for (list = 0; list < sl->list_count; list++) {
118             if (get_bits1(&h->gb)) {
119                 int index;
120                 for (index = 0; ; index++) {
121                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
122
123                     if (reordering_of_pic_nums_idc < 3)
124                         get_ue_golomb(&h->gb);
125                     else if (reordering_of_pic_nums_idc > 3) {
126                         av_log(h->avctx, AV_LOG_ERROR,
127                                "illegal reordering_of_pic_nums_idc %d\n",
128                                reordering_of_pic_nums_idc);
129                         return AVERROR_INVALIDDATA;
130                     } else
131                         break;
132
133                     if (index >= sl->ref_count[list]) {
134                         av_log(h->avctx, AV_LOG_ERROR,
135                                "reference count %d overflow\n", index);
136                         return AVERROR_INVALIDDATA;
137                     }
138                 }
139             }
140         }
141     }
142
143     if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
144         (h->pps.weighted_bipred_idc == 1 && sl->slice_type_nos == AV_PICTURE_TYPE_B))
145         ff_pred_weight_table(h, sl);
146
147     if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
148         int i;
149         for (i = 0; i < MAX_MMCO_COUNT; i++) {
150             MMCOOpcode opcode = get_ue_golomb_31(&h->gb);
151             if (opcode > (unsigned) MMCO_LONG) {
152                 av_log(h->avctx, AV_LOG_ERROR,
153                        "illegal memory management control operation %d\n",
154                        opcode);
155                 return AVERROR_INVALIDDATA;
156             }
157             if (opcode == MMCO_END)
158                return 0;
159             else if (opcode == MMCO_RESET)
160                 return 1;
161
162             if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
163                 get_ue_golomb(&h->gb);
164             if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
165                 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
166                 get_ue_golomb_31(&h->gb);
167         }
168     }
169
170     return 0;
171 }
172
173 /**
174  * Parse NAL units of found picture and decode some basic information.
175  *
176  * @param s parser context.
177  * @param avctx codec context.
178  * @param buf buffer with field/frame data.
179  * @param buf_size size of the buffer.
180  */
181 static inline int parse_nal_units(AVCodecParserContext *s,
182                                   AVCodecContext *avctx,
183                                   const uint8_t *buf, int buf_size)
184 {
185     H264ParseContext *p = s->priv_data;
186     H264Context      *h = &p->h;
187     const uint8_t *buf_end = buf + buf_size;
188     unsigned int pps_id;
189     unsigned int slice_type;
190     int state = -1, got_reset = 0;
191     const uint8_t *ptr;
192     int field_poc[2];
193
194     /* set some sane default values */
195     s->pict_type         = AV_PICTURE_TYPE_I;
196     s->key_frame         = 0;
197     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
198
199     h->avctx = avctx;
200     ff_h264_reset_sei(h);
201
202     if (!buf_size)
203         return 0;
204
205     for (;;) {
206         int src_length, dst_length, consumed;
207         buf = avpriv_find_start_code(buf, buf_end, &state);
208         if (buf >= buf_end)
209             break;
210         --buf;
211         src_length = buf_end - buf;
212         switch (state & 0x1f) {
213         case NAL_SLICE:
214         case NAL_IDR_SLICE:
215             // Do not walk the whole buffer just to decode slice header
216             if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
217                 /* IDR or disposable slice
218                  * No need to decode many bytes because MMCOs shall not be present. */
219                 if (src_length > 60)
220                     src_length = 60;
221             } else {
222                 /* To decode up to MMCOs */
223                 if (src_length > 1000)
224                     src_length = 1000;
225             }
226             break;
227         }
228         ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
229         if (!ptr || dst_length < 0)
230             break;
231
232         init_get_bits(&h->gb, ptr, 8 * dst_length);
233         switch (h->nal_unit_type) {
234         case NAL_SPS:
235             ff_h264_decode_seq_parameter_set(h);
236             break;
237         case NAL_PPS:
238             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
239             break;
240         case NAL_SEI:
241             ff_h264_decode_sei(h);
242             break;
243         case NAL_IDR_SLICE:
244             s->key_frame = 1;
245
246             h->prev_frame_num        = 0;
247             h->prev_frame_num_offset = 0;
248             h->prev_poc_msb          =
249             h->prev_poc_lsb          = 0;
250         /* fall through */
251         case NAL_SLICE:
252             get_ue_golomb(&h->gb);  // skip first_mb_in_slice
253             slice_type   = get_ue_golomb_31(&h->gb);
254             s->pict_type = golomb_to_pict_type[slice_type % 5];
255             if (h->sei_recovery_frame_cnt >= 0) {
256                 /* key frame, since recovery_frame_cnt is set */
257                 s->key_frame = 1;
258             }
259             pps_id = get_ue_golomb(&h->gb);
260             if (pps_id >= MAX_PPS_COUNT) {
261                 av_log(h->avctx, AV_LOG_ERROR,
262                        "pps_id %u out of range\n", pps_id);
263                 return -1;
264             }
265             if (!h->pps_buffers[pps_id]) {
266                 av_log(h->avctx, AV_LOG_ERROR,
267                        "non-existing PPS %u referenced\n", pps_id);
268                 return -1;
269             }
270             h->pps = *h->pps_buffers[pps_id];
271             if (!h->sps_buffers[h->pps.sps_id]) {
272                 av_log(h->avctx, AV_LOG_ERROR,
273                        "non-existing SPS %u referenced\n", h->pps.sps_id);
274                 return -1;
275             }
276             h->sps       = *h->sps_buffers[h->pps.sps_id];
277             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
278
279             s->coded_width  = 16 * h->sps.mb_width;
280             s->coded_height = 16 * h->sps.mb_height;
281             s->width        = s->coded_width  - (h->sps.crop_right + h->sps.crop_left);
282             s->height       = s->coded_height - (h->sps.crop_top   + h->sps.crop_bottom);
283             if (s->width <= 0 || s->height <= 0) {
284                 s->width  = s->coded_width;
285                 s->height = s->coded_height;
286             }
287
288             switch (h->sps.bit_depth_luma) {
289             case 9:
290                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P9;
291                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
292                 else                   s->format = AV_PIX_FMT_YUV420P9;
293                 break;
294             case 10:
295                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P10;
296                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
297                 else                   s->format = AV_PIX_FMT_YUV420P10;
298                 break;
299             case 8:
300                 if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P;
301                 else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
302                 else                   s->format = AV_PIX_FMT_YUV420P;
303                 break;
304             default:
305                 s->format = AV_PIX_FMT_NONE;
306             }
307
308             avctx->profile = ff_h264_get_profile(&h->sps);
309             avctx->level   = h->sps.level_idc;
310
311             if (h->sps.frame_mbs_only_flag) {
312                 h->picture_structure = PICT_FRAME;
313             } else {
314                 if (get_bits1(&h->gb)) { // field_pic_flag
315                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
316                 } else {
317                     h->picture_structure = PICT_FRAME;
318                 }
319             }
320
321             if (h->nal_unit_type == NAL_IDR_SLICE)
322                 get_ue_golomb(&h->gb); /* idr_pic_id */
323             if (h->sps.poc_type == 0) {
324                 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
325
326                 if (h->pps.pic_order_present == 1 &&
327                     h->picture_structure == PICT_FRAME)
328                     h->delta_poc_bottom = get_se_golomb(&h->gb);
329             }
330
331             if (h->sps.poc_type == 1 &&
332                 !h->sps.delta_pic_order_always_zero_flag) {
333                 h->delta_poc[0] = get_se_golomb(&h->gb);
334
335                 if (h->pps.pic_order_present == 1 &&
336                     h->picture_structure == PICT_FRAME)
337                     h->delta_poc[1] = get_se_golomb(&h->gb);
338             }
339
340             /* Decode POC of this picture.
341              * The prev_ values needed for decoding POC of the next picture are not set here. */
342             field_poc[0] = field_poc[1] = INT_MAX;
343             ff_init_poc(h, field_poc, &s->output_picture_number);
344
345             /* Continue parsing to check if MMCO_RESET is present.
346              * FIXME: MMCO_RESET could appear in non-first slice.
347              *        Maybe, we should parse all undisposable non-IDR slice of this
348              *        picture until encountering MMCO_RESET in a slice of it. */
349             if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
350                 got_reset = scan_mmco_reset(s);
351                 if (got_reset < 0)
352                     return got_reset;
353             }
354
355             /* Set up the prev_ values for decoding POC of the next picture. */
356             h->prev_frame_num        = got_reset ? 0 : h->frame_num;
357             h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
358             if (h->nal_ref_idc != 0) {
359                 if (!got_reset) {
360                     h->prev_poc_msb = h->poc_msb;
361                     h->prev_poc_lsb = h->poc_lsb;
362                 } else {
363                     h->prev_poc_msb = 0;
364                     h->prev_poc_lsb =
365                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
366                 }
367             }
368
369             if (h->sps.pic_struct_present_flag) {
370                 switch (h->sei_pic_struct) {
371                 case SEI_PIC_STRUCT_TOP_FIELD:
372                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
373                     s->repeat_pict = 0;
374                     break;
375                 case SEI_PIC_STRUCT_FRAME:
376                 case SEI_PIC_STRUCT_TOP_BOTTOM:
377                 case SEI_PIC_STRUCT_BOTTOM_TOP:
378                     s->repeat_pict = 1;
379                     break;
380                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
381                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
382                     s->repeat_pict = 2;
383                     break;
384                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
385                     s->repeat_pict = 3;
386                     break;
387                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
388                     s->repeat_pict = 5;
389                     break;
390                 default:
391                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
392                     break;
393                 }
394             } else {
395                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
396             }
397
398             if (h->picture_structure == PICT_FRAME) {
399                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
400                 if (h->sps.pic_struct_present_flag) {
401                     switch (h->sei_pic_struct) {
402                     case SEI_PIC_STRUCT_TOP_BOTTOM:
403                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
404                         s->field_order = AV_FIELD_TT;
405                         break;
406                     case SEI_PIC_STRUCT_BOTTOM_TOP:
407                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
408                         s->field_order = AV_FIELD_BB;
409                         break;
410                     default:
411                         s->field_order = AV_FIELD_PROGRESSIVE;
412                         break;
413                     }
414                 } else {
415                     if (field_poc[0] < field_poc[1])
416                         s->field_order = AV_FIELD_TT;
417                     else if (field_poc[0] > field_poc[1])
418                         s->field_order = AV_FIELD_BB;
419                     else
420                         s->field_order = AV_FIELD_PROGRESSIVE;
421                 }
422             } else {
423                 if (h->picture_structure == PICT_TOP_FIELD)
424                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
425                 else
426                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
427                 s->field_order = AV_FIELD_UNKNOWN;
428             }
429
430             return 0; /* no need to evaluate the rest */
431         }
432         buf += consumed;
433     }
434     /* didn't find a picture! */
435     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
436     return -1;
437 }
438
439 static int h264_parse(AVCodecParserContext *s,
440                       AVCodecContext *avctx,
441                       const uint8_t **poutbuf, int *poutbuf_size,
442                       const uint8_t *buf, int buf_size)
443 {
444     H264ParseContext *p = s->priv_data;
445     H264Context      *h = &p->h;
446     ParseContext *pc = &p->pc;
447     int next;
448
449     if (!p->got_first) {
450         p->got_first = 1;
451         if (avctx->extradata_size) {
452             h->avctx = avctx;
453             // must be done like in the decoder.
454             // otherwise opening the parser, creating extradata,
455             // and then closing and opening again
456             // will cause has_b_frames to be always set.
457             // NB: estimate_timings_from_pts behaves exactly like this.
458             if (!avctx->has_b_frames)
459                 h->low_delay = 1;
460             ff_h264_decode_extradata(h);
461         }
462     }
463
464     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
465         next = buf_size;
466     } else {
467         next = h264_find_frame_end(p, buf, buf_size);
468
469         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
470             *poutbuf      = NULL;
471             *poutbuf_size = 0;
472             return buf_size;
473         }
474
475         if (next < 0 && next != END_NOT_FOUND) {
476             assert(pc->last_index + next >= 0);
477             h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
478         }
479     }
480
481     parse_nal_units(s, avctx, buf, buf_size);
482
483     if (h->sei_cpb_removal_delay >= 0) {
484         s->dts_sync_point    = h->sei_buffering_period_present;
485         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
486         s->pts_dts_delta     = h->sei_dpb_output_delay;
487     } else {
488         s->dts_sync_point    = INT_MIN;
489         s->dts_ref_dts_delta = INT_MIN;
490         s->pts_dts_delta     = INT_MIN;
491     }
492
493     if (s->flags & PARSER_FLAG_ONCE) {
494         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
495     }
496
497     *poutbuf      = buf;
498     *poutbuf_size = buf_size;
499     return next;
500 }
501
502 static int h264_split(AVCodecContext *avctx,
503                       const uint8_t *buf, int buf_size)
504 {
505     int i;
506     uint32_t state = -1;
507     int has_sps    = 0;
508
509     for (i = 0; i <= buf_size; i++) {
510         if ((state & 0xFFFFFF1F) == 0x107)
511             has_sps = 1;
512         /*  if((state&0xFFFFFF1F) == 0x101 ||
513          *     (state&0xFFFFFF1F) == 0x102 ||
514          *     (state&0xFFFFFF1F) == 0x105) {
515          *  }
516          */
517         if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x106 &&
518             (state & 0xFFFFFF1F) != 0x107 && (state & 0xFFFFFF1F) != 0x108 &&
519             (state & 0xFFFFFF1F) != 0x109 && (state & 0xFFFFFF1F) != 0x10d &&
520             (state & 0xFFFFFF1F) != 0x10f) {
521             if (has_sps) {
522                 while (i > 4 && buf[i - 5] == 0)
523                     i--;
524                 return i - 4;
525             }
526         }
527         if (i < buf_size)
528             state = (state << 8) | buf[i];
529     }
530     return 0;
531 }
532
533 static void close(AVCodecParserContext *s)
534 {
535     H264ParseContext *p = s->priv_data;
536     H264Context      *h = &p->h;
537     ParseContext *pc = &p->pc;
538
539     av_free(pc->buffer);
540     ff_h264_free_context(h);
541 }
542
543 static av_cold int init(AVCodecParserContext *s)
544 {
545     H264ParseContext *p = s->priv_data;
546     H264Context      *h = &p->h;
547
548     h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
549     if (!h->slice_ctx)
550         return 0;
551     h->nb_slice_ctx = 1;
552
553     h->thread_context[0]   = h;
554     h->slice_context_count = 1;
555     ff_h264dsp_init(&h->h264dsp, 8, 1);
556     return 0;
557 }
558
559 AVCodecParser ff_h264_parser = {
560     .codec_ids      = { AV_CODEC_ID_H264 },
561     .priv_data_size = sizeof(H264ParseContext),
562     .parser_init    = init,
563     .parser_parse   = h264_parse,
564     .parser_close   = close,
565     .split          = h264_split,
566 };