]> git.sesse.net Git - vlc/blob - modules/packetizer/h264.c
* h264: bug fix (give valid date + split at the right place).
[vlc] / modules / packetizer / h264.c
1 /*****************************************************************************
2  * h264.c: h264/avc video packetizer
3  *****************************************************************************
4  * Copyright (C) 2001, 2002 VideoLAN
5  * $Id$
6  *
7  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
8  *          Eric Petit <titer@videolan.org>
9  *          Gildas Bazin <gbazin@videolan.org>
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License as published by
13  * the Free Software Foundation; either version 2 of the License, or
14  * (at your option) any later version.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  * GNU General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public License
22  * along with this program; if not, write to the Free Software
23  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
24  *****************************************************************************/
25
26 /*****************************************************************************
27  * Preamble
28  *****************************************************************************/
29 #include <stdlib.h>                                      /* malloc(), free() */
30
31 #include <vlc/vlc.h>
32 #include <vlc/decoder.h>
33 #include <vlc/sout.h>
34
35 #include "vlc_block_helper.h"
36 #include "vlc_bits.h"
37
38 /*****************************************************************************
39  * Module descriptor
40  *****************************************************************************/
41 static int  Open ( vlc_object_t * );
42 static void Close( vlc_object_t * );
43
44 vlc_module_begin();
45     set_category( CAT_SOUT );
46     set_subcategory( SUBCAT_SOUT_PACKETIZER );
47     set_description( _("H264 video packetizer") );
48     set_capability( "packetizer", 50 );
49     set_callbacks( Open, Close );
50 vlc_module_end();
51
52
53 /****************************************************************************
54  * Local prototypes
55  ****************************************************************************/
56 static block_t *Packetize( decoder_t *, block_t ** );
57 static block_t *PacketizeAVC1( decoder_t *, block_t ** );
58
59 struct decoder_sys_t
60 {
61     block_bytestream_t bytestream;
62
63     int     i_state;
64     int     i_offset;
65     uint8_t startcode[4];
66
67     vlc_bool_t b_slice;
68     block_t    *p_frame;
69
70     vlc_bool_t   b_sps;
71
72     /* avcC data */
73     int i_avcC_length_size;
74
75     /* Useful values of the Sequence Parameter Set */
76     int i_log2_max_frame_num;
77     int b_frame_mbs_only;
78
79     /* Useful values of the Slice Header */
80     int i_nal_type;
81     int i_nal_ref_idc;
82     int i_idr_pic_id;
83     int i_frame_num;
84 };
85
86 enum
87 {
88     STATE_NOSYNC,
89     STATE_NEXT_SYNC,
90 };
91
92 enum nal_unit_type_e
93 {
94     NAL_UNKNOWN = 0,
95     NAL_SLICE   = 1,
96     NAL_SLICE_DPA   = 2,
97     NAL_SLICE_DPB   = 3,
98     NAL_SLICE_DPC   = 4,
99     NAL_SLICE_IDR   = 5,    /* ref_idc != 0 */
100     NAL_SEI         = 6,    /* ref_idc == 0 */
101     NAL_SPS         = 7,
102     NAL_PPS         = 8,
103     NAL_AU_DELIMITER= 9
104     /* ref_idc == 0 for 6,9,10,11,12 */
105 };
106
107 enum nal_priority_e
108 {
109     NAL_PRIORITY_DISPOSABLE = 0,
110     NAL_PRIORITY_LOW        = 1,
111     NAL_PRIORITY_HIGH       = 2,
112     NAL_PRIORITY_HIGHEST    = 3,
113 };
114
115 static block_t *ParseNALBlock( decoder_t *, block_t * );
116
117 static block_t *nal_get_annexeb( decoder_t *, uint8_t *p, int );
118
119 /*****************************************************************************
120  * Open: probe the packetizer and return score
121  *****************************************************************************/
122 static int Open( vlc_object_t *p_this )
123 {
124     decoder_t     *p_dec = (decoder_t*)p_this;
125     decoder_sys_t *p_sys;
126
127     if( p_dec->fmt_in.i_codec != VLC_FOURCC( 'h', '2', '6', '4') &&
128         p_dec->fmt_in.i_codec != VLC_FOURCC( 'H', '2', '6', '4') &&
129         p_dec->fmt_in.i_codec != VLC_FOURCC( 'V', 'S', 'S', 'H') &&
130         p_dec->fmt_in.i_codec != VLC_FOURCC( 'v', 's', 's', 'h') &&
131         ( p_dec->fmt_in.i_codec != VLC_FOURCC( 'a', 'v', 'c', '1') ||
132           p_dec->fmt_in.i_extra < 7 ) )
133     {
134         return VLC_EGENERIC;
135     }
136
137     /* Allocate the memory needed to store the decoder's structure */
138     if( ( p_dec->p_sys = p_sys = malloc( sizeof(decoder_sys_t) ) ) == NULL )
139     {
140         msg_Err( p_dec, "out of memory" );
141         return VLC_EGENERIC;
142     }
143     p_sys->i_state = STATE_NOSYNC;
144     p_sys->i_offset = 0;
145     p_sys->startcode[0] = 0;
146     p_sys->startcode[1] = 0;
147     p_sys->startcode[2] = 0;
148     p_sys->startcode[3] = 1;
149     p_sys->bytestream = block_BytestreamInit( p_dec );
150     p_sys->b_slice = VLC_FALSE;
151     p_sys->p_frame = NULL;
152     p_sys->b_sps   = VLC_FALSE;
153
154     p_sys->i_nal_type = -1;
155     p_sys->i_nal_ref_idc = -1;
156     p_sys->i_idr_pic_id = -1;
157     p_sys->i_frame_num = -1;
158
159     /* Setup properties */
160     es_format_Copy( &p_dec->fmt_out, &p_dec->fmt_in );
161     p_dec->fmt_out.i_codec = VLC_FOURCC( 'h', '2', '6', '4' );
162     /* FIXME: FFMPEG isn't happy at all if you leave this */
163     if( p_dec->fmt_out.i_extra ) free( p_dec->fmt_out.p_extra );
164     p_dec->fmt_out.i_extra = 0; p_dec->fmt_out.p_extra = 0;
165
166     if( p_dec->fmt_in.i_codec == VLC_FOURCC( 'a', 'v', 'c', '1' ) )
167     {
168         uint8_t *p = &((uint8_t*)p_dec->fmt_in.p_extra)[4];
169         int i_sps, i_pps;
170         int i;
171
172         /* Parse avcC */
173         p_sys->i_avcC_length_size = 1 + ((*p++)&0x03);
174
175         /* Read SPS */
176         i_sps = (*p++)&0x1f;
177
178         for( i = 0; i < i_sps; i++ )
179         {
180             int i_length = GetWBE( p );
181             block_t *p_sps = nal_get_annexeb( p_dec, p + 2, i_length );
182
183             ParseNALBlock( p_dec, p_sps );
184             p += 2 + i_length;
185         }
186         /* Read PPS */
187         i_pps = *p++;
188         for( i = 0; i < i_pps; i++ )
189         {
190             int i_length = GetWBE( p );
191             block_t *p_pps = nal_get_annexeb( p_dec, p + 2, i_length );
192
193             ParseNALBlock( p_dec, p_pps );
194             p += 2 + i_length;
195         }
196         msg_Dbg( p_dec, "avcC length size=%d sps=%d pps=%d",
197                  p_sys->i_avcC_length_size, i_sps, i_pps );
198
199         /* Set callback */
200         p_dec->pf_packetize = PacketizeAVC1;
201     }
202     else
203     {
204         /* Set callback */
205         p_dec->pf_packetize = Packetize;
206     }
207
208     return VLC_SUCCESS;
209 }
210
211 /*****************************************************************************
212  * Close: clean up the packetizer
213  *****************************************************************************/
214 static void Close( vlc_object_t *p_this )
215 {
216     decoder_t *p_dec = (decoder_t*)p_this;
217     decoder_sys_t *p_sys = p_dec->p_sys;
218
219     block_BytestreamRelease( &p_sys->bytestream );
220     free( p_sys );
221 }
222
223 /****************************************************************************
224  * Packetize: the whole thing
225  ****************************************************************************/
226 static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
227 {
228     decoder_sys_t *p_sys = p_dec->p_sys;
229     block_t       *p_pic;
230
231     if( !pp_block || !*pp_block ) return NULL;
232
233     block_BytestreamPush( &p_sys->bytestream, *pp_block );
234
235     for( ;; )
236     {
237         switch( p_sys->i_state )
238         {
239             case STATE_NOSYNC:
240                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
241                       &p_sys->i_offset, p_sys->startcode+1, 3 ) == VLC_SUCCESS)
242                 {
243                     p_sys->i_state = STATE_NEXT_SYNC;
244                 }
245
246                 if( p_sys->i_offset )
247                 {
248                     block_SkipBytes( &p_sys->bytestream, p_sys->i_offset );
249                     p_sys->i_offset = 0;
250                     block_BytestreamFlush( &p_sys->bytestream );
251                 }
252
253                 if( p_sys->i_state != STATE_NEXT_SYNC )
254                 {
255                     /* Need more data */
256                     return NULL;
257                 }
258
259                 p_sys->i_offset = 1; /* To find next startcode */
260
261             case STATE_NEXT_SYNC:
262                 /* Find the next startcode */
263                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
264                       &p_sys->i_offset, p_sys->startcode, 3 ) != VLC_SUCCESS)
265                 {
266                     if( block_FindStartcodeFromOffset( &p_sys->bytestream,
267                           &p_sys->i_offset, p_sys->startcode+1, 3 ) !=
268                         VLC_SUCCESS )
269                     {
270                         /* Need more data */
271                         return NULL;
272                     }
273                 }
274
275                 /* Get the new fragment and set the pts/dts */
276                 p_pic = block_New( p_dec, p_sys->i_offset );
277                 p_pic->i_pts = p_sys->bytestream.p_block->i_pts;
278                 p_pic->i_dts = p_sys->bytestream.p_block->i_dts;
279
280                 block_GetBytes( &p_sys->bytestream, p_pic->p_buffer,
281                                 p_pic->i_buffer );
282
283                 p_sys->i_offset = 0;
284
285                 /* Parse the NAL */
286                 if( !( p_pic = ParseNALBlock( p_dec, p_pic ) ) )
287                 {
288                     p_sys->i_state = STATE_NOSYNC;
289                     break;
290                 }
291 #if 0
292                 msg_Dbg( p_dec, "pts="I64Fd" dts="I64Fd,
293                          p_pic->i_pts, p_pic->i_dts );
294 #endif
295
296                 /* So p_block doesn't get re-added several times */
297                 *pp_block = block_BytestreamPop( &p_sys->bytestream );
298
299                 p_sys->i_state = STATE_NOSYNC;
300
301                 return p_pic;
302         }
303     }
304 }
305
306 /****************************************************************************
307  * PacketizeAVC1: the whole thing
308  ****************************************************************************/
309 static block_t *PacketizeAVC1( decoder_t *p_dec, block_t **pp_block )
310 {
311     decoder_sys_t *p_sys = p_dec->p_sys;
312     block_t       *p_block;
313     block_t       *p_ret = NULL;
314     uint8_t       *p;
315
316     if( !pp_block || !*pp_block ) return NULL;
317
318     p_block = *pp_block;
319     *pp_block = NULL;
320
321     for( p = p_block->p_buffer; p < &p_block->p_buffer[p_block->i_buffer]; )
322     {
323         block_t *p_pic;
324         int i_size = 0;
325         int i;
326
327         for( i = 0; i < p_sys->i_avcC_length_size; i++ )
328         {
329             i_size = (i_size << 8) | (*p++);
330         }
331
332         if( i_size > 0 )
333         {
334             block_t *p_part = nal_get_annexeb( p_dec, p, i_size );
335
336             p_part->i_dts = p_block->i_dts;
337             p_part->i_pts = p_block->i_pts;
338
339             /* Parse the NAL */
340             if( ( p_pic = ParseNALBlock( p_dec, p_part ) ) )
341             {
342                 block_ChainAppend( &p_ret, p_pic );
343             }
344         }
345         p += i_size;
346     }
347
348     return p_ret;
349 }
350
351 static block_t *nal_get_annexeb( decoder_t *p_dec, uint8_t *p, int i_size )
352 {
353     block_t *p_nal;
354
355     p_nal = block_New( p_dec, 3 + i_size );
356
357     /* Add start code */
358     p_nal->p_buffer[0] = 0x00;
359     p_nal->p_buffer[1] = 0x00;
360     p_nal->p_buffer[2] = 0x01;
361
362     /* Copy nalu */
363     memcpy( &p_nal->p_buffer[3], p, i_size );
364
365     return p_nal;
366 }
367
368 static void nal_get_decoded( uint8_t **pp_ret, int *pi_ret,
369                              uint8_t *src, int i_src )
370 {
371     uint8_t *end = &src[i_src];
372     uint8_t *dst = malloc( i_src );
373
374     *pp_ret = dst;
375
376     while( src < end )
377     {
378         if( src < end - 3 && src[0] == 0x00 && src[1] == 0x00 &&
379             src[2] == 0x03 )
380         {
381             *dst++ = 0x00;
382             *dst++ = 0x00;
383
384             src += 3;
385             continue;
386         }
387         *dst++ = *src++;
388     }
389
390     *pi_ret = dst - *pp_ret;
391 }
392
393 static inline int bs_read_ue( bs_t *s )
394 {
395     int i = 0;
396
397     while( bs_read1( s ) == 0 && s->p < s->p_end && i < 32 )
398     {
399         i++;
400     }
401     return( ( 1 << i) - 1 + bs_read( s, i ) );
402 }
403
404 static inline int bs_read_se( bs_t *s )
405 {
406     int val = bs_read_ue( s );
407
408     return val&0x01 ? (val+1)/2 : -(val/2);
409 }
410
411
412 static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
413 {
414     decoder_sys_t *p_sys = p_dec->p_sys;
415     block_t *p_pic = NULL;
416
417     const int i_nal_ref_idc = (p_frag->p_buffer[3] >> 5)&0x03;
418     const int i_nal_type = p_frag->p_buffer[3]&0x1f;
419
420 #define OUTPUT \
421     do {                                                \
422         p_pic = block_ChainGather( p_sys->p_frame );    \
423         p_pic->i_length = 0;    /* FIXME */             \
424                                                         \
425         p_sys->p_frame = NULL;                          \
426         p_sys->b_slice = VLC_FALSE;                     \
427     } while(0)
428
429
430     if( p_sys->b_slice && !p_sys->b_sps )
431     {
432         block_ChainRelease( p_sys->p_frame );
433         msg_Warn( p_dec, "waiting for SPS" );
434
435         /* Reset context */
436         p_sys->p_frame = NULL;
437         p_sys->b_slice = VLC_FALSE;
438     }
439
440     if( !p_sys->b_sps &&
441         i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
442     {
443         p_sys->b_slice = VLC_TRUE;
444         /* Fragment will be discarded later on */
445     }
446     else if( i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
447     {
448         uint8_t *dec;
449         int i_dec, i_first_mb, i_slice_type, i_frame_num, i_pic_flags = 0;
450         vlc_bool_t b_pic = VLC_FALSE;
451         bs_t s;
452
453         /* do not convert the whole frame */
454         nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
455                          __MIN( p_frag->i_buffer - 4, 60 ) );
456         bs_init( &s, dec, i_dec );
457
458         /* first_mb_in_slice */
459         i_first_mb = bs_read_ue( &s );
460
461         /* slice_type */
462         switch( (i_slice_type = bs_read_ue( &s )) )
463         {
464             case 0: case 5:
465                 i_pic_flags = BLOCK_FLAG_TYPE_P;
466                 break;
467             case 1: case 6:
468                 i_pic_flags = BLOCK_FLAG_TYPE_B;
469                 break;
470             case 2: case 7:
471                 i_pic_flags = BLOCK_FLAG_TYPE_I;
472                 break;
473             case 3: case 8: /* SP */
474                 i_pic_flags = BLOCK_FLAG_TYPE_P;
475                 break;
476             case 4: case 9:
477                 i_pic_flags = BLOCK_FLAG_TYPE_I;
478                 break;
479         }
480
481         /* pic_parameter_set_id */
482         bs_read_ue( &s );
483         /* frame_num */
484         i_frame_num = bs_read( &s, p_sys->i_log2_max_frame_num + 4 );
485
486         /* Detection of the first VCL NAL unit of a primary coded picture
487          * (cf. 7.4.1.2.4) */
488         if( i_frame_num != p_sys->i_frame_num ||
489             ( (i_nal_ref_idc != p_sys->i_nal_ref_idc) &&
490               (!i_nal_ref_idc || !p_sys->i_nal_ref_idc) ) )
491         {
492             b_pic = VLC_TRUE;
493         }
494         p_sys->i_frame_num = i_frame_num;
495         p_sys->i_nal_ref_idc = i_nal_ref_idc;
496
497         if( !p_sys->b_frame_mbs_only )
498         {
499             /* field_pic_flag */
500             if( bs_read( &s, 1 ) )
501             {
502                 /* bottom_field_flag */
503                 bs_read( &s, 1 );
504             }
505         }
506
507         if( i_nal_type == NAL_SLICE_IDR )
508         {
509             /* id_pic_id */
510             int i_idr_pic_id = bs_read_ue( &s );
511             if( p_sys->i_nal_type != i_nal_type ) b_pic = VLC_TRUE;
512             if( p_sys->i_idr_pic_id != i_idr_pic_id ) b_pic = VLC_TRUE;
513             p_sys->i_idr_pic_id = i_idr_pic_id;
514         }
515         p_sys->i_nal_type = i_nal_type;
516
517         if( b_pic && p_sys->b_slice )
518             OUTPUT;
519
520         p_sys->b_slice = VLC_TRUE;
521
522         free( dec );
523     }
524     else if( i_nal_type == NAL_SPS )
525     {
526         uint8_t *dec;
527         int     i_dec;
528         bs_t s;
529         int i_tmp;
530
531         msg_Dbg( p_dec, "found NAL_SPS" );
532
533         p_sys->b_sps = VLC_TRUE;
534
535         nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
536                          p_frag->i_buffer - 4 );
537
538         bs_init( &s, dec, i_dec );
539         /* Skip profile(8), constraint_set012, reserver(5), level(8) */
540         bs_skip( &s, 8 + 1+1+1 + 5 + 8 );
541         /* sps id */
542         bs_read_ue( &s );
543         /* Skip i_log2_max_frame_num */
544         p_sys->i_log2_max_frame_num = bs_read_ue( &s );
545         /* Read poc_type */
546         i_tmp = bs_read_ue( &s );
547         if( i_tmp == 0 )
548         {
549             /* skip i_log2_max_poc_lsb */
550             bs_read_ue( &s );
551         }
552         else if( i_tmp == 1 )
553         {
554             int i_cycle;
555             /* skip b_delta_pic_order_always_zero */
556             bs_skip( &s, 1 );
557             /* skip i_offset_for_non_ref_pic */
558             bs_read_se( &s );
559             /* skip i_offset_for_top_to_bottom_field */
560             bs_read_se( &s );
561             /* read i_num_ref_frames_in_poc_cycle */
562             i_cycle = bs_read_ue( &s );
563             if( i_cycle > 256 ) i_cycle = 256;
564             while( i_cycle > 0 )
565             {
566                 /* skip i_offset_for_ref_frame */
567                 bs_read_se(&s );
568             }
569         }
570         /* i_num_ref_frames */
571         bs_read_ue( &s );
572         /* b_gaps_in_frame_num_value_allowed */
573         bs_skip( &s, 1 );
574
575         /* Read size */
576         p_dec->fmt_out.video.i_width  = 16 * ( bs_read_ue( &s ) + 1 );
577         p_dec->fmt_out.video.i_height = 16 * ( bs_read_ue( &s ) + 1 );
578
579         /* b_frame_mbs_only */
580         p_sys->b_frame_mbs_only = bs_read( &s, 1 );
581         if( p_sys->b_frame_mbs_only == 0 )
582         {
583             bs_skip( &s, 1 );
584         }
585         /* b_direct8x8_inference */
586         bs_skip( &s, 1 );
587
588         /* crop */
589         i_tmp = bs_read( &s, 1 );
590         if( i_tmp )
591         {
592             /* left */
593             bs_read_ue( &s );
594             /* right */
595             bs_read_ue( &s );
596             /* top */
597             bs_read_ue( &s );
598             /* bottom */
599             bs_read_ue( &s );
600         }
601
602         /* vui */
603         i_tmp = bs_read( &s, 1 );
604         if( i_tmp )
605         {
606             /* read the aspect ratio part if any FIXME check it */
607             i_tmp = bs_read( &s, 1 );
608             if( i_tmp )
609             {
610                 static const struct { int w, h; } sar[14] =
611                 {
612                     { 0,   0 }, { 1,   1 }, { 12, 11 }, { 10, 11 },
613                     { 16, 11 }, { 40, 33 }, { 24, 11 }, { 20, 11 },
614                     { 32, 11 }, { 80, 33 }, { 18, 11 }, { 15, 11 },
615                     { 64, 33 }, { 160,99 },
616                 };
617                 int i_sar = bs_read( &s, 8 );
618                 int w, h;
619
620                 if( i_sar < 14 )
621                 {
622                     w = sar[i_sar].w;
623                     h = sar[i_sar].h;
624                 }
625                 else
626                 {
627                     w = bs_read( &s, 16 );
628                     h = bs_read( &s, 16 );
629                 }
630                 p_dec->fmt_out.video.i_aspect =
631                     VOUT_ASPECT_FACTOR * w / h * p_dec->fmt_out.video.i_width /
632                     p_dec->fmt_out.video.i_height;
633             }
634         }
635
636         free( dec );
637
638
639         if( p_sys->b_slice )
640             OUTPUT;
641     }
642     else if( i_nal_type == NAL_PPS )
643     {
644         bs_t s;
645         bs_init( &s, &p_frag->p_buffer[4], p_frag->i_buffer - 4 );
646
647         /* TODO */
648         msg_Dbg( p_dec, "found NAL_PPS" );
649
650         if( p_sys->b_slice )
651             OUTPUT;
652     }
653     else if( i_nal_type == NAL_AU_DELIMITER ||
654              i_nal_type == NAL_SEI ||
655              ( i_nal_type >= 13 && i_nal_type <= 18 ) )
656     {
657         if( p_sys->b_slice )
658             OUTPUT;
659     }
660
661 #undef OUTPUT
662
663     /* Append the block */
664     block_ChainAppend( &p_sys->p_frame, p_frag );
665
666     return p_pic;
667 }