h264_parser: AVC support.
[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 #include "parser.h"
29 #include "h264data.h"
30 #include "golomb.h"
31
32 #include <assert.h>
33
34
35 static int ff_h264_find_frame_end(H264Context *h, const uint8_t *buf, int buf_size)
36 {
37     int i, j;
38     uint32_t state;
39     ParseContext *pc = &(h->s.parse_context);
40     int next_avc= h->is_avc ? 0 : buf_size;
41
42 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
43 //    mb_addr= pc->mb_addr - 1;
44     state= pc->state;
45     if(state>13)
46         state= 7;
47
48     if(h->is_avc && !h->nal_length_size)
49         av_log(h->s.avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
50
51     for(i=0; i<buf_size; i++){
52         if(i >= next_avc) {
53             int nalsize = 0;
54             i = next_avc;
55             for(j = 0; j < h->nal_length_size; j++)
56                 nalsize = (nalsize << 8) | buf[i++];
57             if(nalsize <= 0 || nalsize > buf_size - i){
58                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
59                 return buf_size;
60             }
61             next_avc= i + nalsize;
62             state= 5;
63         }
64
65         if(state==7){
66 #if HAVE_FAST_UNALIGNED
67         /* we check i<buf_size instead of i+3/7 because its simpler
68          * and there should be FF_INPUT_BUFFER_PADDING_SIZE bytes at the end
69          */
70 #    if HAVE_FAST_64BIT
71             while(i<next_avc && !((~*(const uint64_t*)(buf+i) & (*(const uint64_t*)(buf+i) - 0x0101010101010101ULL)) & 0x8080808080808080ULL))
72                 i+=8;
73 #    else
74             while(i<next_avc && !((~*(const uint32_t*)(buf+i) & (*(const uint32_t*)(buf+i) - 0x01010101U)) & 0x80808080U))
75                 i+=4;
76 #    endif
77 #endif
78             for(; i<next_avc; i++){
79                 if(!buf[i]){
80                     state=2;
81                     break;
82                 }
83             }
84         }else if(state<=2){
85             if(buf[i]==1)   state^= 5; //2->7, 1->4, 0->5
86             else if(buf[i]) state = 7;
87             else            state>>=1; //2->1, 1->0, 0->0
88         }else if(state<=5){
89             int v= buf[i] & 0x1F;
90             if(v==6 || v==7 || v==8 || v==9){
91                 if(pc->frame_start_found){
92                     i++;
93                     goto found;
94                 }
95             }else if(v==1 || v==2 || v==5){
96                 if(pc->frame_start_found){
97                     state+=8;
98                     continue;
99                 }else
100                     pc->frame_start_found = 1;
101             }
102             state= 7;
103         }else{
104             if(buf[i] & 0x80)
105                 goto found;
106             state= 7;
107         }
108     }
109     pc->state= state;
110     if(h->is_avc)
111         return next_avc;
112     return END_NOT_FOUND;
113
114 found:
115     pc->state=7;
116     pc->frame_start_found= 0;
117     if(h->is_avc)
118         return next_avc;
119     return i-(state&5);
120 }
121
122 /**
123  * Parse NAL units of found picture and decode some basic information.
124  *
125  * @param s parser context.
126  * @param avctx codec context.
127  * @param buf buffer with field/frame data.
128  * @param buf_size size of the buffer.
129  */
130 static inline int parse_nal_units(AVCodecParserContext *s,
131                                   AVCodecContext *avctx,
132                                   const uint8_t *buf, int buf_size)
133 {
134     H264Context *h = s->priv_data;
135     const uint8_t *buf_end = buf + buf_size;
136     unsigned int pps_id;
137     unsigned int slice_type;
138     int state = -1;
139     const uint8_t *ptr;
140
141     /* set some sane default values */
142     s->pict_type = AV_PICTURE_TYPE_I;
143     s->key_frame = 0;
144
145     h->s.avctx= avctx;
146     h->sei_recovery_frame_cnt = -1;
147     h->sei_dpb_output_delay         =  0;
148     h->sei_cpb_removal_delay        = -1;
149     h->sei_buffering_period_present =  0;
150
151     if (!buf_size)
152         return 0;
153
154     for(;;) {
155         int src_length, dst_length, consumed;
156         buf = ff_find_start_code(buf, buf_end, &state);
157         if(buf >= buf_end)
158             break;
159         --buf;
160         src_length = buf_end - buf;
161         switch (state & 0x1f) {
162         case NAL_SLICE:
163         case NAL_IDR_SLICE:
164             // Do not walk the whole buffer just to decode slice header
165             if (src_length > 20)
166                 src_length = 20;
167             break;
168         }
169         ptr= ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
170         if (ptr==NULL || dst_length < 0)
171             break;
172
173         init_get_bits(&h->s.gb, ptr, 8*dst_length);
174         switch(h->nal_unit_type) {
175         case NAL_SPS:
176             ff_h264_decode_seq_parameter_set(h);
177             break;
178         case NAL_PPS:
179             ff_h264_decode_picture_parameter_set(h, h->s.gb.size_in_bits);
180             break;
181         case NAL_SEI:
182             ff_h264_decode_sei(h);
183             break;
184         case NAL_IDR_SLICE:
185             s->key_frame = 1;
186             /* fall through */
187         case NAL_SLICE:
188             get_ue_golomb(&h->s.gb);  // skip first_mb_in_slice
189             slice_type = get_ue_golomb_31(&h->s.gb);
190             s->pict_type = golomb_to_pict_type[slice_type % 5];
191             if (h->sei_recovery_frame_cnt >= 0) {
192                 /* key frame, since recovery_frame_cnt is set */
193                 s->key_frame = 1;
194             }
195             pps_id= get_ue_golomb(&h->s.gb);
196             if(pps_id>=MAX_PPS_COUNT) {
197                 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
198                 return -1;
199             }
200             if(!h->pps_buffers[pps_id]) {
201                 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS referenced\n");
202                 return -1;
203             }
204             h->pps= *h->pps_buffers[pps_id];
205             if(!h->sps_buffers[h->pps.sps_id]) {
206                 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS referenced\n");
207                 return -1;
208             }
209             h->sps = *h->sps_buffers[h->pps.sps_id];
210             h->frame_num = get_bits(&h->s.gb, h->sps.log2_max_frame_num);
211
212             avctx->profile = ff_h264_get_profile(&h->sps);
213             avctx->level   = h->sps.level_idc;
214
215             if(h->sps.frame_mbs_only_flag){
216                 h->s.picture_structure= PICT_FRAME;
217             }else{
218                 if(get_bits1(&h->s.gb)) { //field_pic_flag
219                     h->s.picture_structure= PICT_TOP_FIELD + get_bits1(&h->s.gb); //bottom_field_flag
220                 } else {
221                     h->s.picture_structure= PICT_FRAME;
222                 }
223             }
224
225             if(h->sps.pic_struct_present_flag) {
226                 switch (h->sei_pic_struct) {
227                     case SEI_PIC_STRUCT_TOP_FIELD:
228                     case SEI_PIC_STRUCT_BOTTOM_FIELD:
229                         s->repeat_pict = 0;
230                         break;
231                     case SEI_PIC_STRUCT_FRAME:
232                     case SEI_PIC_STRUCT_TOP_BOTTOM:
233                     case SEI_PIC_STRUCT_BOTTOM_TOP:
234                         s->repeat_pict = 1;
235                         break;
236                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
237                     case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
238                         s->repeat_pict = 2;
239                         break;
240                     case SEI_PIC_STRUCT_FRAME_DOUBLING:
241                         s->repeat_pict = 3;
242                         break;
243                     case SEI_PIC_STRUCT_FRAME_TRIPLING:
244                         s->repeat_pict = 5;
245                         break;
246                     default:
247                         s->repeat_pict = h->s.picture_structure == PICT_FRAME ? 1 : 0;
248                         break;
249                 }
250             } else {
251                 s->repeat_pict = h->s.picture_structure == PICT_FRAME ? 1 : 0;
252             }
253
254             return 0; /* no need to evaluate the rest */
255         }
256         buf += consumed;
257     }
258     /* didn't find a picture! */
259     av_log(h->s.avctx, AV_LOG_ERROR, "missing picture in access unit\n");
260     return -1;
261 }
262
263 static int h264_parse(AVCodecParserContext *s,
264                       AVCodecContext *avctx,
265                       const uint8_t **poutbuf, int *poutbuf_size,
266                       const uint8_t *buf, int buf_size)
267 {
268     H264Context *h = s->priv_data;
269     ParseContext *pc = &h->s.parse_context;
270     int next;
271
272     if (!h->got_first) {
273         h->got_first = 1;
274         if (avctx->extradata_size) {
275             h->s.avctx = avctx;
276             // must be done like in decoder, otherwise opening the parser,
277             // letting it create extradata and then closing and opening again
278             // will cause has_b_frames to be always set.
279             // Note that estimate_timings_from_pts does exactly this.
280             if (!avctx->has_b_frames)
281                 h->s.low_delay = 1;
282             ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
283         }
284     }
285
286     if(s->flags & PARSER_FLAG_COMPLETE_FRAMES){
287         next= buf_size;
288     }else{
289         next= ff_h264_find_frame_end(h, buf, buf_size);
290
291         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
292             *poutbuf = NULL;
293             *poutbuf_size = 0;
294             return buf_size;
295         }
296
297         if(next<0 && next != END_NOT_FOUND){
298             assert(pc->last_index + next >= 0 );
299             ff_h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); //update state
300         }
301     }
302
303     if(!h->is_avc){
304     parse_nal_units(s, avctx, buf, buf_size);
305
306     if (h->sei_cpb_removal_delay >= 0) {
307         s->dts_sync_point    = h->sei_buffering_period_present;
308         s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
309         s->pts_dts_delta     = h->sei_dpb_output_delay;
310     } else {
311         s->dts_sync_point    = INT_MIN;
312         s->dts_ref_dts_delta = INT_MIN;
313         s->pts_dts_delta     = INT_MIN;
314     }
315
316     if (s->flags & PARSER_FLAG_ONCE) {
317         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
318     }
319     }
320
321     *poutbuf = buf;
322     *poutbuf_size = buf_size;
323     return next;
324 }
325
326 static int h264_split(AVCodecContext *avctx,
327                       const uint8_t *buf, int buf_size)
328 {
329     int i;
330     uint32_t state = -1;
331     int has_sps= 0;
332
333     for(i=0; i<=buf_size; i++){
334         if((state&0xFFFFFF1F) == 0x107)
335             has_sps=1;
336 /*        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
337         }*/
338         if((state&0xFFFFFF00) == 0x100 && (state&0xFFFFFF1F) != 0x107 && (state&0xFFFFFF1F) != 0x108 && (state&0xFFFFFF1F) != 0x109){
339             if(has_sps){
340                 while(i>4 && buf[i-5]==0) i--;
341                 return i-4;
342             }
343         }
344         if (i<buf_size)
345             state= (state<<8) | buf[i];
346     }
347     return 0;
348 }
349
350 static void close(AVCodecParserContext *s)
351 {
352     H264Context *h = s->priv_data;
353     ParseContext *pc = &h->s.parse_context;
354
355     av_free(pc->buffer);
356     ff_h264_free_context(h);
357 }
358
359 static int init(AVCodecParserContext *s)
360 {
361     H264Context *h = s->priv_data;
362     h->thread_context[0] = h;
363     return 0;
364 }
365
366 AVCodecParser ff_h264_parser = {
367     { CODEC_ID_H264 },
368     sizeof(H264Context),
369     init,
370     h264_parse,
371     close,
372     h264_split,
373 };