6e3e65b28a1b063d756bb01caf24f0b69c21ce8a
[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, int buf_size)
38 {
39     int i, j;
40     uint32_t state;
41     ParseContext *pc = &h->parse_context;
42     int next_avc= h->is_avc ? 0 : buf_size;
43
44 //    mb_addr= pc->mb_addr - 1;
45     state= pc->state;
46     if(state>13)
47         state= 7;
48
49     if(h->is_avc && !h->nal_length_size)
50         av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
51
52     for(i=0; i<buf_size; i++){
53         if(i >= next_avc) {
54             int nalsize = 0;
55             i = next_avc;
56             for(j = 0; j < h->nal_length_size; j++)
57                 nalsize = (nalsize << 8) | buf[i++];
58             if(nalsize <= 0 || nalsize > buf_size - i){
59                 av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
60                 return buf_size;
61             }
62             next_avc= i + nalsize;
63             state= 5;
64         }
65
66         if(state==7){
67 #if HAVE_FAST_UNALIGNED
68         /* we check i<buf_size instead of i+3/7 because its simpler
69          * and there should be FF_INPUT_BUFFER_PADDING_SIZE bytes at the end
70          */
71 #    if HAVE_FAST_64BIT
72             while(i<next_avc && !((~*(const uint64_t*)(buf+i) & (*(const uint64_t*)(buf+i) - 0x0101010101010101ULL)) & 0x8080808080808080ULL))
73                 i+=8;
74 #    else
75             while(i<next_avc && !((~*(const uint32_t*)(buf+i) & (*(const uint32_t*)(buf+i) - 0x01010101U)) & 0x80808080U))
76                 i+=4;
77 #    endif
78 #endif
79             for(; i<next_avc; i++){
80                 if(!buf[i]){
81                     state=2;
82                     break;
83                 }
84             }
85         }else if(state<=2){
86             if(buf[i]==1)   state^= 5; //2->7, 1->4, 0->5
87             else if(buf[i]) state = 7;
88             else            state>>=1; //2->1, 1->0, 0->0
89         }else if(state<=5){
90             int v= buf[i] & 0x1F;
91             if(v==6 || v==7 || v==8 || v==9){
92                 if(pc->frame_start_found){
93                     i++;
94                     goto found;
95                 }
96             }else if(v==1 || v==2 || v==5){
97                 state+=8;
98                 continue;
99             }
100             state= 7;
101         }else{
102             h->parse_history[h->parse_history_count++]= buf[i];
103             if(h->parse_history_count>3){
104                 unsigned int mb, last_mb= h->parse_last_mb;
105                 GetBitContext gb;
106
107                 init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
108                 h->parse_history_count=0;
109                 mb= get_ue_golomb_long(&gb);
110                 last_mb= h->parse_last_mb;
111                 h->parse_last_mb= mb;
112                 if(pc->frame_start_found){
113                     if(mb <= last_mb)
114                         goto found;
115                 }else
116                     pc->frame_start_found = 1;
117                 state= 7;
118             }
119         }
120     }
121     pc->state= state;
122     if(h->is_avc)
123         return next_avc;
124     return END_NOT_FOUND;
125
126 found:
127     pc->state=7;
128     pc->frame_start_found= 0;
129     if(h->is_avc)
130         return next_avc;
131     return i-(state&5) - 3*(state>7);
132 }
133
134 /**
135  * Parse NAL units of found picture and decode some basic information.
136  *
137  * @param s parser context.
138  * @param avctx codec context.
139  * @param buf buffer with field/frame data.
140  * @param buf_size size of the buffer.
141  */
142 static inline int parse_nal_units(AVCodecParserContext *s,
143                                   AVCodecContext *avctx,
144                                   const uint8_t *buf, int buf_size)
145 {
146     H264Context *h = s->priv_data;
147     const uint8_t *buf_end = buf + buf_size;
148     unsigned int pps_id;
149     unsigned int slice_type;
150     int state = -1;
151     const uint8_t *ptr;
152     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
153     int field_poc[2];
154
155     /* set some sane default values */
156     s->pict_type = AV_PICTURE_TYPE_I;
157     s->key_frame = 0;
158     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
159
160     h->avctx= avctx;
161     h->sei_recovery_frame_cnt = -1;
162     h->sei_dpb_output_delay         =  0;
163     h->sei_cpb_removal_delay        = -1;
164     h->sei_buffering_period_present =  0;
165
166     if (!buf_size)
167         return 0;
168
169     for(;;) {
170         int src_length, dst_length, consumed, nalsize = 0;
171         if (h->is_avc) {
172             int i;
173             if (h->nal_length_size >= buf_end - buf) break;
174             nalsize = 0;
175             for (i = 0; i < h->nal_length_size; i++)
176                 nalsize = (nalsize << 8) | *buf++;
177             if (nalsize <= 0 || nalsize > buf_end - buf) {
178                 av_log(h->avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
179                 break;
180             }
181             src_length = nalsize;
182         } else {
183         buf = avpriv_find_start_code(buf, buf_end, &state);
184         if(buf >= buf_end)
185             break;
186         --buf;
187         src_length = buf_end - buf;
188         }
189         switch (state & 0x1f) {
190         case NAL_SLICE:
191         case NAL_IDR_SLICE:
192             // Do not walk the whole buffer just to decode slice header
193             if (src_length > 20)
194                 src_length = 20;
195             break;
196         }
197         ptr= ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
198         if (ptr==NULL || dst_length < 0)
199             break;
200
201         init_get_bits(&h->gb, ptr, 8*dst_length);
202         switch(h->nal_unit_type) {
203         case NAL_SPS:
204             ff_h264_decode_seq_parameter_set(h);
205             break;
206         case NAL_PPS:
207             ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
208             break;
209         case NAL_SEI:
210             ff_h264_decode_sei(h);
211             break;
212         case NAL_IDR_SLICE:
213             s->key_frame = 1;
214
215             h->prev_frame_num        = 0;
216             h->prev_frame_num_offset = 0;
217             h->prev_poc_msb          =
218             h->prev_poc_lsb          = 0;
219             /* fall through */
220         case NAL_SLICE:
221             get_ue_golomb_long(&h->gb);  // skip first_mb_in_slice
222             slice_type = get_ue_golomb_31(&h->gb);
223             s->pict_type = golomb_to_pict_type[slice_type % 5];
224             if (h->sei_recovery_frame_cnt >= 0) {
225                 /* key frame, since recovery_frame_cnt is set */
226                 s->key_frame = 1;
227             }
228             pps_id= get_ue_golomb(&h->gb);
229             if(pps_id>=MAX_PPS_COUNT) {
230                 av_log(h->avctx, AV_LOG_ERROR, "pps_id out of range\n");
231                 return -1;
232             }
233             if(!h->pps_buffers[pps_id]) {
234                 av_log(h->avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
235                 return -1;
236             }
237             h->pps= *h->pps_buffers[pps_id];
238             if(!h->sps_buffers[h->pps.sps_id]) {
239                 av_log(h->avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
240                 return -1;
241             }
242             h->sps = *h->sps_buffers[h->pps.sps_id];
243             h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
244
245             avctx->profile = ff_h264_get_profile(&h->sps);
246             avctx->level   = h->sps.level_idc;
247
248             if(h->sps.frame_mbs_only_flag){
249                 h->picture_structure= PICT_FRAME;
250             }else{
251                 if(get_bits1(&h->gb)) { //field_pic_flag
252                     h->picture_structure= PICT_TOP_FIELD + get_bits1(&h->gb); //bottom_field_flag
253                 } else {
254                     h->picture_structure= PICT_FRAME;
255                 }
256             }
257
258             if (h->nal_unit_type == NAL_IDR_SLICE)
259                 get_ue_golomb(&h->gb); /* idr_pic_id */
260             if (h->sps.poc_type == 0) {
261                 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
262
263                 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
264                     h->delta_poc_bottom = get_se_golomb(&h->gb);
265             }
266
267             if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
268                 h->delta_poc[0] = get_se_golomb(&h->gb);
269
270                 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
271                     h->delta_poc[1] = get_se_golomb(&h->gb);
272             }
273
274             ff_init_poc(h, field_poc, NULL);
275
276             if(h->sps.pic_struct_present_flag) {
277                 switch (h->sei_pic_struct) {
278                     case SEI_PIC_STRUCT_TOP_FIELD:
279                     case SEI_PIC_STRUCT_BOTTOM_FIELD:
280                         s->repeat_pict = 0;
281                         break;
282                     case SEI_PIC_STRUCT_FRAME:
283                     case SEI_PIC_STRUCT_TOP_BOTTOM:
284                     case SEI_PIC_STRUCT_BOTTOM_TOP:
285                         s->repeat_pict = 1;
286                         break;
287                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
288                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
289                         s->repeat_pict = 2;
290                         break;
291                     case SEI_PIC_STRUCT_FRAME_DOUBLING:
292                         s->repeat_pict = 3;
293                         break;
294                     case SEI_PIC_STRUCT_FRAME_TRIPLING:
295                         s->repeat_pict = 5;
296                         break;
297                     default:
298                         s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
299                         break;
300                 }
301             } else {
302                 s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
303             }
304
305             if (h->picture_structure == PICT_FRAME) {
306                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
307                 if (h->sps.pic_struct_present_flag) {
308                     switch (h->sei_pic_struct) {
309                         case SEI_PIC_STRUCT_TOP_BOTTOM:
310                         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
311                             s->field_order = AV_FIELD_TT;
312                             break;
313                         case SEI_PIC_STRUCT_BOTTOM_TOP:
314                         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
315                             s->field_order = AV_FIELD_BB;
316                             break;
317                         default:
318                             s->field_order = AV_FIELD_PROGRESSIVE;
319                             break;
320                     }
321                 } else {
322                     if (field_poc[0] < field_poc[1])
323                         s->field_order = AV_FIELD_TT;
324                     else if (field_poc[0] > field_poc[1])
325                         s->field_order = AV_FIELD_BB;
326                     else
327                         s->field_order = AV_FIELD_PROGRESSIVE;
328                 }
329             } else {
330                 if (h->picture_structure == PICT_TOP_FIELD)
331                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
332                 else
333                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
334                 s->field_order = AV_FIELD_UNKNOWN;
335             }
336
337             return 0; /* no need to evaluate the rest */
338         }
339         buf += h->is_avc ? nalsize : consumed;
340     }
341     if (q264)
342         return 0;
343     /* didn't find a picture! */
344     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
345     return -1;
346 }
347
348 static int h264_parse(AVCodecParserContext *s,
349                       AVCodecContext *avctx,
350                       const uint8_t **poutbuf, int *poutbuf_size,
351                       const uint8_t *buf, int buf_size)
352 {
353     H264Context *h = s->priv_data;
354     ParseContext *pc = &h->parse_context;
355     int next;
356
357     if (!h->got_first) {
358         h->got_first = 1;
359         if (avctx->extradata_size) {
360             h->avctx = avctx;
361             // must be done like in decoder, otherwise opening the parser,
362             // letting it create extradata and then closing and opening again
363             // will cause has_b_frames to be always set.
364             // Note that estimate_timings_from_pts does exactly this.
365             if (!avctx->has_b_frames)
366                 h->low_delay = 1;
367             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
368         }
369     }
370
371     if(s->flags & PARSER_FLAG_COMPLETE_FRAMES){
372         next= buf_size;
373     }else{
374         next = h264_find_frame_end(h, buf, buf_size);
375
376         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
377             *poutbuf = NULL;
378             *poutbuf_size = 0;
379             return buf_size;
380         }
381
382         if(next<0 && next != END_NOT_FOUND){
383             av_assert1(pc->last_index + next >= 0 );
384             h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); // update state
385         }
386     }
387
388     parse_nal_units(s, avctx, buf, buf_size);
389
390     if (h->sei_cpb_removal_delay >= 0) {
391         s->dts_sync_point    = h->sei_buffering_period_present;
392         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
393         s->pts_dts_delta     = h->sei_dpb_output_delay;
394     } else {
395         s->dts_sync_point    = INT_MIN;
396         s->dts_ref_dts_delta = INT_MIN;
397         s->pts_dts_delta     = INT_MIN;
398     }
399
400     if (s->flags & PARSER_FLAG_ONCE) {
401         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
402     }
403
404     *poutbuf = buf;
405     *poutbuf_size = buf_size;
406     return next;
407 }
408
409 static int h264_split(AVCodecContext *avctx,
410                       const uint8_t *buf, int buf_size)
411 {
412     int i;
413     uint32_t state = -1;
414     int has_sps= 0;
415
416     for(i=0; i<=buf_size; i++){
417         if((state&0xFFFFFF1F) == 0x107)
418             has_sps=1;
419 /*        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
420         }*/
421         if((state&0xFFFFFF00) == 0x100 && (state&0xFFFFFF1F) != 0x107 && (state&0xFFFFFF1F) != 0x108 && (state&0xFFFFFF1F) != 0x109){
422             if(has_sps){
423                 while(i>4 && buf[i-5]==0) i--;
424                 return i-4;
425             }
426         }
427         if (i<buf_size)
428             state= (state<<8) | buf[i];
429     }
430     return 0;
431 }
432
433 static void close(AVCodecParserContext *s)
434 {
435     H264Context *h = s->priv_data;
436     ParseContext *pc = &h->parse_context;
437
438     av_free(pc->buffer);
439     ff_h264_free_context(h);
440 }
441
442 static av_cold int init(AVCodecParserContext *s)
443 {
444     H264Context *h = s->priv_data;
445     h->thread_context[0] = h;
446     h->slice_context_count = 1;
447     return 0;
448 }
449
450 AVCodecParser ff_h264_parser = {
451     .codec_ids      = { AV_CODEC_ID_H264 },
452     .priv_data_size = sizeof(H264Context),
453     .parser_init    = init,
454     .parser_parse   = h264_parse,
455     .parser_close   = close,
456     .split          = h264_split,
457 };