h264_parser: Fix order of operations
[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
34 #include <assert.h>
35
36
37 static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
38                                int buf_size)
39 {
40     int i;
41     uint32_t state;
42     ParseContext *pc = &h->parse_context;
43 //    mb_addr= pc->mb_addr - 1;
44     state = pc->state;
45     if (state > 13)
46         state = 7;
47
48     for (i = 0; i < buf_size; i++) {
49         if (state == 7) {
50             i += h->h264dsp.h264_find_start_code_candidate(buf + i, buf_size - i);
51             if (i < buf_size)
52                 state = 2;
53         } else if (state <= 2) {
54             if (buf[i] == 1)
55                 state ^= 5;            // 2->7, 1->4, 0->5
56             else if (buf[i])
57                 state = 7;
58             else
59                 state >>= 1;           // 2->1, 1->0, 0->0
60         } else if (state <= 5) {
61             int v = buf[i] & 0x1F;
62             if (v == 6 || v == 7 || v == 8 || v == 9) {
63                 if (pc->frame_start_found) {
64                     i++;
65                     goto found;
66                 }
67             } else if (v == 1 || v == 2 || v == 5) {
68                 if (pc->frame_start_found) {
69                     state += 8;
70                     continue;
71                 } else
72                     pc->frame_start_found = 1;
73             }
74             state = 7;
75         } else {
76             if (buf[i] & 0x80)
77                 goto found;
78             state = 7;
79         }
80     }
81     pc->state = state;
82     return END_NOT_FOUND;
83
84 found:
85     pc->state             = 7;
86     pc->frame_start_found = 0;
87     return i - (state & 5);
88 }
89
90 static int scan_mmco_reset(AVCodecParserContext *s)
91 {
92     H264Context *h = s->priv_data;
93
94     h->slice_type_nos = s->pict_type & 3;
95
96     if (h->pps.redundant_pic_cnt_present)
97         get_ue_golomb(&h->gb); // redundant_pic_count
98
99     if (ff_set_ref_count(h) < 0)
100         return AVERROR_INVALIDDATA;
101
102     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
103         int list;
104         for (list = 0; list < h->list_count; list++) {
105             if (get_bits1(&h->gb)) {
106                 int index;
107                 for (index = 0; ; index++) {
108                     unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(&h->gb);
109
110                     if (reordering_of_pic_nums_idc < 3)
111                         get_ue_golomb(&h->gb);
112                     else if (reordering_of_pic_nums_idc > 3) {
113                         av_log(h->avctx, AV_LOG_ERROR,
114                                "illegal reordering_of_pic_nums_idc %d\n",
115                                reordering_of_pic_nums_idc);
116                         return AVERROR_INVALIDDATA;
117                     } else
118                         break;
119
120                     if (index >= h->ref_count[list]) {
121                         av_log(h->avctx, AV_LOG_ERROR, "reference count overflow\n");
122                         return AVERROR_INVALIDDATA;
123                     }
124                 }
125             }
126         }
127     }
128
129     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
130         (h->pps.weighted_bipred_idc == 1 && h->slice_type_nos == AV_PICTURE_TYPE_B))
131         ff_pred_weight_table(h);
132
133     if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
134         int i;
135         for (i = 0; i < MAX_MMCO_COUNT; i++) {
136             MMCOOpcode opcode = get_ue_golomb_31(&h->gb);
137             if (opcode > (unsigned) MMCO_LONG) {
138                 av_log(h->avctx, AV_LOG_ERROR,
139                        "illegal memory management control operation %d\n",
140                        opcode);
141                 return AVERROR_INVALIDDATA;
142             }
143             if (opcode == MMCO_END)
144                return 0;
145             else if (opcode == MMCO_RESET)
146                 return 1;
147
148             if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
149                 get_ue_golomb(&h->gb);
150             if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
151                 opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
152                 get_ue_golomb_31(&h->gb);
153         }
154     }
155
156     return 0;
157 }
158
159 /**
160  * Parse NAL units of found picture and decode some basic information.
161  *
162  * @param s parser context.
163  * @param avctx codec context.
164  * @param buf buffer with field/frame data.
165  * @param buf_size size of the buffer.
166  */
167 static inline int parse_nal_units(AVCodecParserContext *s,
168                                   AVCodecContext *avctx,
169                                   const uint8_t *buf, int buf_size)
170 {
171     H264Context *h         = s->priv_data;
172     const uint8_t *buf_end = buf + buf_size;
173     unsigned int pps_id;
174     unsigned int slice_type;
175     int state = -1, got_reset = 0;
176     const uint8_t *ptr;
177     int field_poc[2];
178
179     /* set some sane default values */
180     s->pict_type         = AV_PICTURE_TYPE_I;
181     s->key_frame         = 0;
182     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
183
184     h->avctx                        = avctx;
185     h->sei_recovery_frame_cnt       = -1;
186     h->sei_dpb_output_delay         = 0;
187     h->sei_cpb_removal_delay        = -1;
188     h->sei_buffering_period_present = 0;
189
190     if (!buf_size)
191         return 0;
192
193     for (;;) {
194         int src_length, dst_length, consumed;
195         buf = avpriv_find_start_code(buf, buf_end, &state);
196         if (buf >= buf_end)
197             break;
198         --buf;
199         src_length = buf_end - buf;
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 ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
205                 /* IDR or disposable slice
206                  * No need to decode many bytes because MMCOs shall not be present. */
207                 if (src_length > 60)
208                     src_length = 60;
209             } else {
210                 /* To decode up to MMCOs */
211                 if (src_length > 1000)
212                     src_length = 1000;
213             }
214             break;
215         }
216         ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
217         if (ptr == NULL || dst_length < 0)
218             break;
219
220         init_get_bits(&h->gb, ptr, 8 * dst_length);
221         switch (h->nal_unit_type) {
222         case NAL_SPS:
223             ff_h264_decode_seq_parameter_set(h);
224             break;
225         case NAL_PPS:
226             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
227             break;
228         case NAL_SEI:
229             ff_h264_decode_sei(h);
230             break;
231         case NAL_IDR_SLICE:
232             s->key_frame = 1;
233
234             h->prev_frame_num        = 0;
235             h->prev_frame_num_offset = 0;
236             h->prev_poc_msb          =
237             h->prev_poc_lsb          = 0;
238         /* fall through */
239         case NAL_SLICE:
240             get_ue_golomb(&h->gb);  // skip first_mb_in_slice
241             slice_type   = get_ue_golomb_31(&h->gb);
242             s->pict_type = golomb_to_pict_type[slice_type % 5];
243             if (h->sei_recovery_frame_cnt >= 0) {
244                 /* key frame, since recovery_frame_cnt is set */
245                 s->key_frame = 1;
246             }
247             pps_id = get_ue_golomb(&h->gb);
248             if (pps_id >= MAX_PPS_COUNT) {
249                 av_log(h->avctx, AV_LOG_ERROR,
250                        "pps_id out of range\n");
251                 return -1;
252             }
253             if (!h->pps_buffers[pps_id]) {
254                 av_log(h->avctx, AV_LOG_ERROR,
255                        "non-existing PPS referenced\n");
256                 return -1;
257             }
258             h->pps = *h->pps_buffers[pps_id];
259             if (!h->sps_buffers[h->pps.sps_id]) {
260                 av_log(h->avctx, AV_LOG_ERROR,
261                        "non-existing SPS referenced\n");
262                 return -1;
263             }
264             h->sps       = *h->sps_buffers[h->pps.sps_id];
265             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
266
267             avctx->profile = ff_h264_get_profile(&h->sps);
268             avctx->level   = h->sps.level_idc;
269
270             if (h->sps.frame_mbs_only_flag) {
271                 h->picture_structure = PICT_FRAME;
272             } else {
273                 if (get_bits1(&h->gb)) { // field_pic_flag
274                     h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
275                 } else {
276                     h->picture_structure = PICT_FRAME;
277                 }
278             }
279
280             if (h->nal_unit_type == NAL_IDR_SLICE)
281                 get_ue_golomb(&h->gb); /* idr_pic_id */
282             if (h->sps.poc_type == 0) {
283                 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
284
285                 if (h->pps.pic_order_present == 1 &&
286                     h->picture_structure == PICT_FRAME)
287                     h->delta_poc_bottom = get_se_golomb(&h->gb);
288             }
289
290             if (h->sps.poc_type == 1 &&
291                 !h->sps.delta_pic_order_always_zero_flag) {
292                 h->delta_poc[0] = get_se_golomb(&h->gb);
293
294                 if (h->pps.pic_order_present == 1 &&
295                     h->picture_structure == PICT_FRAME)
296                     h->delta_poc[1] = get_se_golomb(&h->gb);
297             }
298
299             /* Decode POC of this picture.
300              * The prev_ values needed for decoding POC of the next picture are not set here. */
301             field_poc[0] = field_poc[1] = INT_MAX;
302             ff_init_poc(h, field_poc, &s->output_picture_number);
303
304             /* Continue parsing to check if MMCO_RESET is present.
305              * FIXME: MMCO_RESET could appear in non-first slice.
306              *        Maybe, we should parse all undisposable non-IDR slice of this
307              *        picture until encountering MMCO_RESET in a slice of it. */
308             if (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
309                 got_reset = scan_mmco_reset(s);
310                 if (got_reset < 0)
311                     return got_reset;
312             }
313
314             /* Set up the prev_ values for decoding POC of the next picture. */
315             h->prev_frame_num        = got_reset ? 0 : h->frame_num;
316             h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
317             if (h->nal_ref_idc != 0) {
318                 if (!got_reset) {
319                     h->prev_poc_msb = h->poc_msb;
320                     h->prev_poc_lsb = h->poc_lsb;
321                 } else {
322                     h->prev_poc_msb = 0;
323                     h->prev_poc_lsb =
324                         h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
325                 }
326             }
327
328             if (h->sps.pic_struct_present_flag) {
329                 switch (h->sei_pic_struct) {
330                 case SEI_PIC_STRUCT_TOP_FIELD:
331                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
332                     s->repeat_pict = 0;
333                     break;
334                 case SEI_PIC_STRUCT_FRAME:
335                 case SEI_PIC_STRUCT_TOP_BOTTOM:
336                 case SEI_PIC_STRUCT_BOTTOM_TOP:
337                     s->repeat_pict = 1;
338                     break;
339                 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
340                 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
341                     s->repeat_pict = 2;
342                     break;
343                 case SEI_PIC_STRUCT_FRAME_DOUBLING:
344                     s->repeat_pict = 3;
345                     break;
346                 case SEI_PIC_STRUCT_FRAME_TRIPLING:
347                     s->repeat_pict = 5;
348                     break;
349                 default:
350                     s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
351                     break;
352                 }
353             } else {
354                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
355             }
356
357             if (h->picture_structure == PICT_FRAME) {
358                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
359                 if (h->sps.pic_struct_present_flag) {
360                     switch (h->sei_pic_struct) {
361                     case SEI_PIC_STRUCT_TOP_BOTTOM:
362                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
363                         s->field_order = AV_FIELD_TT;
364                         break;
365                     case SEI_PIC_STRUCT_BOTTOM_TOP:
366                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
367                         s->field_order = AV_FIELD_BB;
368                         break;
369                     default:
370                         s->field_order = AV_FIELD_PROGRESSIVE;
371                         break;
372                     }
373                 } else {
374                     if (field_poc[0] < field_poc[1])
375                         s->field_order = AV_FIELD_TT;
376                     else if (field_poc[0] > field_poc[1])
377                         s->field_order = AV_FIELD_BB;
378                     else
379                         s->field_order = AV_FIELD_PROGRESSIVE;
380                 }
381             } else {
382                 if (h->picture_structure == PICT_TOP_FIELD)
383                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
384                 else
385                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
386                 s->field_order = AV_FIELD_UNKNOWN;
387             }
388
389             return 0; /* no need to evaluate the rest */
390         }
391         buf += consumed;
392     }
393     /* didn't find a picture! */
394     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
395     return -1;
396 }
397
398 static int h264_parse(AVCodecParserContext *s,
399                       AVCodecContext *avctx,
400                       const uint8_t **poutbuf, int *poutbuf_size,
401                       const uint8_t *buf, int buf_size)
402 {
403     H264Context *h   = s->priv_data;
404     ParseContext *pc = &h->parse_context;
405     int next;
406
407     if (!h->got_first) {
408         h->got_first = 1;
409         if (avctx->extradata_size) {
410             h->avctx = avctx;
411             // must be done like in the decoder.
412             // otherwise opening the parser, creating extradata,
413             // and then closing and opening again
414             // will cause has_b_frames to be always set.
415             // NB: estimate_timings_from_pts behaves exactly like this.
416             if (!avctx->has_b_frames)
417                 h->low_delay = 1;
418             ff_h264_decode_extradata(h);
419         }
420     }
421
422     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
423         next = buf_size;
424     } else {
425         next = h264_find_frame_end(h, buf, buf_size);
426
427         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
428             *poutbuf      = NULL;
429             *poutbuf_size = 0;
430             return buf_size;
431         }
432
433         if (next < 0 && next != END_NOT_FOUND) {
434             assert(pc->last_index + next >= 0);
435             h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); // update state
436         }
437     }
438
439     parse_nal_units(s, avctx, buf, buf_size);
440
441     if (h->sei_cpb_removal_delay >= 0) {
442         s->dts_sync_point    = h->sei_buffering_period_present;
443         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
444         s->pts_dts_delta     = h->sei_dpb_output_delay;
445     } else {
446         s->dts_sync_point    = INT_MIN;
447         s->dts_ref_dts_delta = INT_MIN;
448         s->pts_dts_delta     = INT_MIN;
449     }
450
451     if (s->flags & PARSER_FLAG_ONCE) {
452         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
453     }
454
455     *poutbuf      = buf;
456     *poutbuf_size = buf_size;
457     return next;
458 }
459
460 static int h264_split(AVCodecContext *avctx,
461                       const uint8_t *buf, int buf_size)
462 {
463     int i;
464     uint32_t state = -1;
465     int has_sps    = 0;
466
467     for (i = 0; i <= buf_size; i++) {
468         if ((state & 0xFFFFFF1F) == 0x107)
469             has_sps = 1;
470         /*  if((state&0xFFFFFF1F) == 0x101 ||
471          *     (state&0xFFFFFF1F) == 0x102 ||
472          *     (state&0xFFFFFF1F) == 0x105) {
473          *  }
474          */
475         if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
476             (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
477             if (has_sps) {
478                 while (i > 4 && buf[i - 5] == 0)
479                     i--;
480                 return i - 4;
481             }
482         }
483         if (i < buf_size)
484             state = (state << 8) | buf[i];
485     }
486     return 0;
487 }
488
489 static void close(AVCodecParserContext *s)
490 {
491     H264Context *h   = s->priv_data;
492     ParseContext *pc = &h->parse_context;
493
494     av_free(pc->buffer);
495     ff_h264_free_context(h);
496 }
497
498 static av_cold int init(AVCodecParserContext *s)
499 {
500     H264Context *h = s->priv_data;
501     h->thread_context[0]   = h;
502     h->slice_context_count = 1;
503     ff_h264dsp_init(&h->h264dsp, 8, 1);
504     return 0;
505 }
506
507 AVCodecParser ff_h264_parser = {
508     .codec_ids      = { AV_CODEC_ID_H264 },
509     .priv_data_size = sizeof(H264Context),
510     .parser_init    = init,
511     .parser_parse   = h264_parse,
512     .parser_close   = close,
513     .split          = h264_split,
514 };