]> git.sesse.net Git - vlc/blob - modules/packetizer/mpeg4video.c
Packetizers should trash all blocks with DISCONTINUITY or CORRUPTED flag set.
[vlc] / modules / packetizer / mpeg4video.c
1 /*****************************************************************************
2  * mpeg4video.c: mpeg 4 video packetizer
3  *****************************************************************************
4  * Copyright (C) 2001-2006 the VideoLAN team
5  * $Id$
6  *
7  * Authors: Gildas Bazin <gbazin@videolan.org>
8  *          Laurent Aimar <fenrir@via.ecp.fr>
9  *          Eric Petit <titer@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., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
24  *****************************************************************************/
25
26 /*****************************************************************************
27  * Preamble
28  *****************************************************************************/
29 #include <stdlib.h>                                      /* malloc(), free() */
30
31 #include <vlc/vlc.h>
32 #include <vlc_sout.h>
33 #include <vlc_codec.h>
34 #include <vlc_block.h>
35 #include <vlc_input.h>                  /* hmmm, just for INPUT_RATE_DEFAULT */
36
37 #include "vlc_bits.h"
38 #include "vlc_block_helper.h"
39
40 /*****************************************************************************
41  * Module descriptor
42  *****************************************************************************/
43 static int  Open ( vlc_object_t * );
44 static void Close( vlc_object_t * );
45
46 vlc_module_begin();
47     set_category( CAT_SOUT );
48     set_subcategory( SUBCAT_SOUT_PACKETIZER );
49     set_description( _("MPEG4 video packetizer") );
50     set_capability( "packetizer", 50 );
51     set_callbacks( Open, Close );
52 vlc_module_end();
53
54 /****************************************************************************
55  * Local prototypes
56  ****************************************************************************/
57 static block_t *Packetize( decoder_t *, block_t ** );
58
59 struct decoder_sys_t
60 {
61     /*
62      * Input properties
63      */
64     block_bytestream_t bytestream;
65     int i_state;
66     int i_offset;
67     uint8_t p_startcode[3];
68
69     /*
70      * Common properties
71      */
72     mtime_t i_interpolated_pts;
73     mtime_t i_interpolated_dts;
74     mtime_t i_last_ref_pts;
75     mtime_t i_last_time_ref;
76     mtime_t i_time_ref;
77     mtime_t i_last_time;
78     mtime_t i_last_timeincr;
79
80     unsigned int i_flags;
81
82     int         i_fps_num;
83     int         i_fps_den;
84     int         i_last_incr;
85     int         i_last_incr_diff;
86
87     vlc_bool_t  b_frame;
88
89     /* Current frame being built */
90     block_t    *p_frame;
91     block_t    **pp_last;
92 };
93
94 enum {
95     STATE_NOSYNC,
96     STATE_NEXT_SYNC
97 };
98
99 static block_t *ParseMPEGBlock( decoder_t *, block_t * );
100 static int ParseVOL( decoder_t *, es_format_t *, uint8_t *, int );
101 static int ParseVOP( decoder_t *, block_t * );
102 static int vlc_log2( unsigned int );
103
104 #define VIDEO_OBJECT_MASK                       0x01f
105 #define VIDEO_OBJECT_LAYER_MASK                 0x00f
106
107 #define VIDEO_OBJECT_START_CODE                 0x100
108 #define VIDEO_OBJECT_LAYER_START_CODE           0x120
109 #define VISUAL_OBJECT_SEQUENCE_START_CODE       0x1b0
110 #define VISUAL_OBJECT_SEQUENCE_END_CODE         0x1b1
111 #define USER_DATA_START_CODE                    0x1b2
112 #define GROUP_OF_VOP_START_CODE                 0x1b3
113 #define VIDEO_SESSION_ERROR_CODE                0x1b4
114 #define VISUAL_OBJECT_START_CODE                0x1b5
115 #define VOP_START_CODE                          0x1b6
116 #define FACE_OBJECT_START_CODE                  0x1ba
117 #define FACE_OBJECT_PLANE_START_CODE            0x1bb
118 #define MESH_OBJECT_START_CODE                  0x1bc
119 #define MESH_OBJECT_PLANE_START_CODE            0x1bd
120 #define STILL_TEXTURE_OBJECT_START_CODE         0x1be
121 #define TEXTURE_SPATIAL_LAYER_START_CODE        0x1bf
122 #define TEXTURE_SNR_LAYER_START_CODE            0x1c0
123
124 /*****************************************************************************
125  * Open: probe the packetizer and return score
126  *****************************************************************************/
127 static int Open( vlc_object_t *p_this )
128 {
129     decoder_t     *p_dec = (decoder_t*)p_this;
130     decoder_sys_t *p_sys;
131
132     switch( p_dec->fmt_in.i_codec )
133     {
134         case VLC_FOURCC( 'm', '4', 's', '2'):
135         case VLC_FOURCC( 'M', '4', 'S', '2'):
136         case VLC_FOURCC( 'm', 'p', '4', 's'):
137         case VLC_FOURCC( 'M', 'P', '4', 'S'):
138         case VLC_FOURCC( 'm', 'p', '4', 'v'):
139         case VLC_FOURCC( 'M', 'P', '4', 'V'):
140         case VLC_FOURCC( 'D', 'I', 'V', 'X'):
141         case VLC_FOURCC( 'd', 'i', 'v', 'x'):
142         case VLC_FOURCC( 'X', 'V', 'I', 'D'):
143         case VLC_FOURCC( 'X', 'v', 'i', 'D'):
144         case VLC_FOURCC( 'x', 'v', 'i', 'd'):
145         case VLC_FOURCC( 'D', 'X', '5', '0'):
146         case VLC_FOURCC( 'd', 'x', '5', '0'):
147         case VLC_FOURCC( 0x04, 0,   0,   0):
148         case VLC_FOURCC( '3', 'I', 'V', '2'):
149         case VLC_FOURCC( 'm', '4', 'c', 'c'):
150         case VLC_FOURCC( 'M', '4', 'C', 'C'):
151             break;
152
153         default:
154             return VLC_EGENERIC;
155     }
156
157     /* Allocate the memory needed to store the decoder's structure */
158     if( ( p_dec->p_sys = p_sys = malloc( sizeof(decoder_sys_t) ) ) == NULL )
159     {
160         msg_Err( p_dec, "out of memory" );
161         return VLC_EGENERIC;
162     }
163     memset( p_sys, 0, sizeof(decoder_sys_t) );
164
165     /* Misc init */
166     p_sys->i_state = STATE_NOSYNC;
167     p_sys->bytestream = block_BytestreamInit( p_dec );
168     p_sys->p_startcode[0] = 0;
169     p_sys->p_startcode[1] = 0;
170     p_sys->p_startcode[2] = 1;
171     p_sys->i_offset = 0;
172     p_sys->p_frame = NULL;
173     p_sys->pp_last = &p_sys->p_frame;
174
175     /* Setup properties */
176     es_format_Copy( &p_dec->fmt_out, &p_dec->fmt_in );
177     p_dec->fmt_out.i_codec = VLC_FOURCC( 'm', 'p', '4', 'v' );
178
179     if( p_dec->fmt_in.i_extra )
180     {
181         /* We have a vol */
182         p_dec->fmt_out.i_extra = p_dec->fmt_in.i_extra;
183         p_dec->fmt_out.p_extra = malloc( p_dec->fmt_in.i_extra );
184         memcpy( p_dec->fmt_out.p_extra, p_dec->fmt_in.p_extra,
185                 p_dec->fmt_in.i_extra );
186
187         msg_Dbg( p_dec, "opening with vol size: %d", p_dec->fmt_in.i_extra );
188         ParseVOL( p_dec, &p_dec->fmt_out,
189                   p_dec->fmt_out.p_extra, p_dec->fmt_out.i_extra );
190     }
191     else
192     {
193         /* No vol, we'll have to look for one later on */
194         p_dec->fmt_out.i_extra = 0;
195         p_dec->fmt_out.p_extra = 0;
196     }
197
198     /* Set callback */
199     p_dec->pf_packetize = Packetize;
200
201     return VLC_SUCCESS;
202 }
203
204 /*****************************************************************************
205  * Close: clean up the packetizer
206  *****************************************************************************/
207 static void Close( vlc_object_t *p_this )
208 {
209     decoder_t *p_dec = (decoder_t*)p_this;
210
211     block_BytestreamRelease( &p_dec->p_sys->bytestream );
212     if( p_dec->p_sys->p_frame ) block_ChainRelease( p_dec->p_sys->p_frame );
213     free( p_dec->p_sys );
214 }
215
216 /****************************************************************************
217  * Packetize: the whole thing
218  ****************************************************************************/
219 static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
220 {
221     decoder_sys_t *p_sys = p_dec->p_sys;
222     block_t       *p_pic;
223     mtime_t       i_pts, i_dts;
224
225     if( pp_block == NULL || *pp_block == NULL ) return NULL;
226
227     if( (*pp_block)->i_flags & (BLOCK_FLAG_DISCONTINUITY|BLOCK_FLAG_CORRUPTED) )
228     {
229         p_sys->i_state = STATE_NOSYNC;
230         if( p_sys->p_frame ) block_ChainRelease( p_sys->p_frame );
231         p_sys->p_frame = NULL;
232         p_sys->pp_last = &p_sys->p_frame;
233         block_Release( *pp_block );
234         return NULL;
235     }
236
237     block_BytestreamPush( &p_sys->bytestream, *pp_block );
238
239     while( 1 )
240     {
241         switch( p_sys->i_state )
242         {
243
244         case STATE_NOSYNC:
245             if( block_FindStartcodeFromOffset( &p_sys->bytestream,
246                     &p_sys->i_offset, p_sys->p_startcode, 3 ) == VLC_SUCCESS )
247             {
248                 p_sys->i_state = STATE_NEXT_SYNC;
249             }
250
251             if( p_sys->i_offset )
252             {
253                 block_SkipBytes( &p_sys->bytestream, p_sys->i_offset );
254                 p_sys->i_offset = 0;
255                 block_BytestreamFlush( &p_sys->bytestream );
256             }
257
258             if( p_sys->i_state != STATE_NEXT_SYNC )
259             {
260                 /* Need more data */
261                 return NULL;
262             }
263
264             p_sys->i_offset = 1; /* To find next startcode */
265
266         case STATE_NEXT_SYNC:
267             /* TODO: If p_block == NULL, flush the buffer without checking the
268              * next sync word */
269
270             /* Find the next startcode */
271             if( block_FindStartcodeFromOffset( &p_sys->bytestream,
272                     &p_sys->i_offset, p_sys->p_startcode, 3 ) != VLC_SUCCESS )
273             {
274                 /* Need more data */
275                 return NULL;
276             }
277
278             /* Get the new fragment and set the pts/dts */
279             p_pic = block_New( p_dec, p_sys->i_offset );
280             block_BytestreamFlush( &p_sys->bytestream );
281             p_pic->i_pts = i_pts = p_sys->bytestream.p_block->i_pts;
282             p_pic->i_dts = i_dts = p_sys->bytestream.p_block->i_dts;
283             p_pic->i_rate = p_sys->bytestream.p_block->i_rate;
284
285             block_GetBytes( &p_sys->bytestream, p_pic->p_buffer,
286                             p_pic->i_buffer );
287
288             p_sys->i_offset = 0;
289
290             /* Get picture if any */
291             if( !( p_pic = ParseMPEGBlock( p_dec, p_pic ) ) )
292             {
293                 p_sys->i_state = STATE_NOSYNC;
294                 break;
295             }
296
297             /* don't reuse the same timestamps several times */
298             if( i_pts == p_sys->bytestream.p_block->i_pts &&
299                 i_dts == p_sys->bytestream.p_block->i_dts )
300             {
301                 p_sys->bytestream.p_block->i_pts = 0;
302                 p_sys->bytestream.p_block->i_dts = 0;
303             }
304
305             /* We've just started the stream, wait for the first PTS.
306              * We discard here so we can still get the sequence header. */
307             if( p_sys->i_interpolated_pts <= 0 &&
308                 p_sys->i_interpolated_dts <= 0 )
309             {
310                 msg_Dbg( p_dec, "need a starting pts/dts" );
311                 p_sys->i_state = STATE_NOSYNC;
312                 block_Release( p_pic );
313                 break;
314             }
315
316             /* When starting the stream we can have the first frame with
317              * a null DTS (i_interpolated_pts is initialized to 0) */
318             if( !p_pic->i_dts ) p_pic->i_dts = p_pic->i_pts;
319
320             /* So p_block doesn't get re-added several times */
321             *pp_block = block_BytestreamPop( &p_sys->bytestream );
322
323             p_sys->i_state = STATE_NOSYNC;
324
325             return p_pic;
326         }
327     }
328 }
329
330 /*****************************************************************************
331  * ParseMPEGBlock: Re-assemble fragments into a block containing a picture
332  *****************************************************************************/
333 static block_t *ParseMPEGBlock( decoder_t *p_dec, block_t *p_frag )
334 {
335     decoder_sys_t *p_sys = p_dec->p_sys;
336     block_t *p_pic = NULL;
337
338     if( p_frag->p_buffer[3] == 0xB0 || p_frag->p_buffer[3] == 0xB1 || p_frag->p_buffer[3] == 0xB2 )
339     {   /* VOS and USERDATA */
340 #if 0
341         /* Remove VOS start/end code from the original stream */
342         block_Release( p_frag );
343 #else
344         /* Append the block for now since ts/ps muxers rely on VOL
345          * being present in the stream */
346         block_ChainLastAppend( &p_sys->pp_last, p_frag );
347 #endif
348         return NULL;
349     }
350     if( p_frag->p_buffer[3] >= 0x20 && p_frag->p_buffer[3] <= 0x2f )
351     {
352         /* Copy the complete VOL */
353         if( p_dec->fmt_out.i_extra != p_frag->i_buffer )
354         {
355             p_dec->fmt_out.p_extra =
356                 realloc( p_dec->fmt_out.p_extra, p_frag->i_buffer );
357             p_dec->fmt_out.i_extra = p_frag->i_buffer;
358         }
359         memcpy( p_dec->fmt_out.p_extra, p_frag->p_buffer, p_frag->i_buffer );
360         ParseVOL( p_dec, &p_dec->fmt_out,
361                   p_dec->fmt_out.p_extra, p_dec->fmt_out.i_extra );
362
363 #if 0
364         /* Remove from the original stream */
365         block_Release( p_frag );
366 #else
367         /* Append the block for now since ts/ps muxers rely on VOL
368          * being present in the stream */
369         block_ChainLastAppend( &p_sys->pp_last, p_frag );
370 #endif
371         return NULL;
372     }
373     else
374     {
375         if( !p_dec->fmt_out.i_extra )
376         {
377             msg_Warn( p_dec, "waiting for VOL" );
378             block_Release( p_frag );
379             return NULL;
380         }
381
382         /* Append the block */
383         block_ChainLastAppend( &p_sys->pp_last, p_frag );
384     }
385
386     if( p_frag->p_buffer[3] == 0xb6 &&
387         ParseVOP( p_dec, p_frag ) == VLC_SUCCESS )
388     {
389         /* We are dealing with a VOP */
390         p_pic = block_ChainGather( p_sys->p_frame );
391         p_pic->i_pts = p_sys->i_interpolated_pts;
392         p_pic->i_dts = p_sys->i_interpolated_dts;
393
394         /* Reset context */
395         p_sys->p_frame = NULL;
396         p_sys->pp_last = &p_sys->p_frame;
397     }
398
399     return p_pic;
400 }
401
402 /* ParseVOL:
403  *  TODO:
404  *      - support aspect ratio
405  */
406 static int ParseVOL( decoder_t *p_dec, es_format_t *fmt,
407                      uint8_t *p_vol, int i_vol )
408 {
409     decoder_sys_t *p_sys = p_dec->p_sys;
410     int i_vo_type, i_vo_ver_id, i_ar, i_shape;
411     bs_t s;
412
413     for( ;; )
414     {
415         if( p_vol[0] == 0x00 && p_vol[1] == 0x00 && p_vol[2] == 0x01 &&
416             p_vol[3] >= 0x20 && p_vol[3] <= 0x2f ) break;
417
418         p_vol++; i_vol--;
419         if( i_vol <= 4 ) return VLC_EGENERIC;
420     }
421
422     bs_init( &s, &p_vol[4], i_vol - 4 );
423
424     bs_skip( &s, 1 );   /* random access */
425     i_vo_type = bs_read( &s, 8 );
426     if( bs_read1( &s ) )
427     {
428         i_vo_ver_id = bs_read( &s, 4 );
429         bs_skip( &s, 3 );
430     }
431     else
432     {
433         i_vo_ver_id = 1;
434     }
435     i_ar = bs_read( &s, 4 );
436     if( i_ar == 0xf )
437     {
438         int i_ar_width, i_ar_height;
439
440         i_ar_width = bs_read( &s, 8 );
441         i_ar_height= bs_read( &s, 8 );
442     }
443     if( bs_read1( &s ) )
444     {
445         int i_chroma_format;
446         int i_low_delay;
447
448         /* vol control parameter */
449         i_chroma_format = bs_read( &s, 2 );
450         i_low_delay = bs_read1( &s );
451
452         if( bs_read1( &s ) )
453         {
454             bs_skip( &s, 16 );
455             bs_skip( &s, 16 );
456             bs_skip( &s, 16 );
457             bs_skip( &s, 3 );
458             bs_skip( &s, 11 );
459             bs_skip( &s, 1 );
460             bs_skip( &s, 16 );
461         }
462     }
463     /* shape 0->RECT, 1->BIN, 2->BIN_ONLY, 3->GRAY */
464     i_shape = bs_read( &s, 2 );
465     if( i_shape == 3 && i_vo_ver_id != 1 )
466     {
467         bs_skip( &s, 4 );
468     }
469
470     if( !bs_read1( &s ) ) return VLC_EGENERIC; /* Marker */
471
472     p_sys->i_fps_num = bs_read( &s, 16 ); /* Time increment resolution*/
473     if( !p_sys->i_fps_num ) p_sys->i_fps_num = 1;
474
475     if( !bs_read1( &s ) ) return VLC_EGENERIC; /* Marker */
476
477     if( bs_read1( &s ) )
478     {
479         int i_time_increment_bits = vlc_log2( p_sys->i_fps_num - 1 ) + 1;
480
481         if( i_time_increment_bits < 1 ) i_time_increment_bits = 1;
482
483         p_sys->i_fps_den = bs_read( &s, i_time_increment_bits );
484     }
485     if( i_shape == 0 )
486     {
487         bs_skip( &s, 1 );
488         fmt->video.i_width = bs_read( &s, 13 );
489         bs_skip( &s, 1 );
490         fmt->video.i_height= bs_read( &s, 13 );
491         bs_skip( &s, 1 );
492     }
493
494     return VLC_SUCCESS;
495 }
496
497 static int ParseVOP( decoder_t *p_dec, block_t *p_vop )
498 {
499     decoder_sys_t *p_sys = p_dec->p_sys;
500     int64_t i_time_increment, i_time_ref;
501     int i_modulo_time_base = 0, i_time_increment_bits;
502     bs_t s;
503
504     bs_init( &s, &p_vop->p_buffer[4], p_vop->i_buffer - 4 );
505
506     switch( bs_read( &s, 2 ) )
507     {
508     case 0:
509         p_sys->i_flags = BLOCK_FLAG_TYPE_I;
510         break;
511     case 1:
512         p_sys->i_flags = BLOCK_FLAG_TYPE_P;
513         break;
514     case 2:
515         p_sys->i_flags = BLOCK_FLAG_TYPE_B;
516         p_sys->b_frame = VLC_TRUE;
517         break;
518     case 3: /* gni ? */
519         p_sys->i_flags = BLOCK_FLAG_TYPE_PB;
520         break;
521     }
522
523     while( bs_read( &s, 1 ) ) i_modulo_time_base++;
524     if( !bs_read1( &s ) ) return VLC_EGENERIC; /* Marker */
525
526     /* VOP time increment */
527     i_time_increment_bits = vlc_log2(p_dec->p_sys->i_fps_num - 1) + 1;
528     if( i_time_increment_bits < 1 ) i_time_increment_bits = 1;
529     i_time_increment = bs_read( &s, i_time_increment_bits );
530
531     /* Interpolate PTS/DTS */
532     if( !(p_sys->i_flags & BLOCK_FLAG_TYPE_B) )
533     {
534         p_sys->i_last_time_ref = p_sys->i_time_ref;
535         p_sys->i_time_ref +=
536             (i_modulo_time_base * p_dec->p_sys->i_fps_num);
537         i_time_ref = p_sys->i_time_ref;
538     }
539     else
540     {
541         i_time_ref = p_sys->i_last_time_ref +
542             (i_modulo_time_base * p_dec->p_sys->i_fps_num);
543     }
544
545 #if 0
546     msg_Err( p_dec, "interp pts/dts (%lli,%lli), pts/dts (%lli,%lli)",
547              p_sys->i_interpolated_pts, p_sys->i_interpolated_dts,
548              p_vop->i_pts, p_vop->i_dts );
549 #endif
550
551     if( p_dec->p_sys->i_fps_num < 5 && /* Work-around buggy streams */
552         p_dec->fmt_in.video.i_frame_rate > 0 &&
553         p_dec->fmt_in.video.i_frame_rate_base > 0 )
554     {
555         p_sys->i_interpolated_pts += I64C(1000000) *
556         p_dec->fmt_in.video.i_frame_rate_base *
557         p_vop->i_rate / INPUT_RATE_DEFAULT /
558         p_dec->fmt_in.video.i_frame_rate;
559     }
560     else if( p_dec->p_sys->i_fps_num )
561         p_sys->i_interpolated_pts +=
562             ( I64C(1000000) * (i_time_ref + i_time_increment -
563               p_sys->i_last_time - p_sys->i_last_timeincr) *
564               p_vop->i_rate / INPUT_RATE_DEFAULT /
565               p_dec->p_sys->i_fps_num );
566
567     p_sys->i_last_time = i_time_ref;
568     p_sys->i_last_timeincr = i_time_increment;
569
570     /* Correct interpolated dts when we receive a new pts/dts */
571     if( p_vop->i_pts > 0 )
572         p_sys->i_interpolated_pts = p_vop->i_pts;
573     if( p_vop->i_dts > 0 )
574         p_sys->i_interpolated_dts = p_vop->i_dts;
575
576     if( (p_sys->i_flags & BLOCK_FLAG_TYPE_B) || !p_sys->b_frame )
577     {
578         /* Trivial case (DTS == PTS) */
579
580         p_sys->i_interpolated_dts = p_sys->i_interpolated_pts;
581
582         if( p_vop->i_pts > 0 )
583             p_sys->i_interpolated_dts = p_vop->i_pts;
584         if( p_vop->i_dts > 0 )
585             p_sys->i_interpolated_dts = p_vop->i_dts;
586
587         p_sys->i_interpolated_pts = p_sys->i_interpolated_dts;
588     }
589     else
590     {
591         if( p_sys->i_last_ref_pts > 0 )
592             p_sys->i_interpolated_dts = p_sys->i_last_ref_pts;
593
594         p_sys->i_last_ref_pts = p_sys->i_interpolated_pts;
595     }
596
597     return VLC_SUCCESS;
598 }
599
600 /* look at ffmpeg av_log2 ;) */
601 static int vlc_log2( unsigned int v )
602 {
603     int n = 0;
604     static const int vlc_log2_table[16] =
605     {
606         0,0,1,1,2,2,2,2, 3,3,3,3,3,3,3,3
607     };
608
609     if( v&0xffff0000 )
610     {
611         v >>= 16;
612         n += 16;
613     }
614     if( v&0xff00 )
615     {
616         v >>= 8;
617         n += 8;
618     }
619     if( v&0xf0 )
620     {
621         v >>= 4;
622         n += 4;
623     }
624     n += vlc_log2_table[v];
625
626     return n;
627 }