]> git.sesse.net Git - vlc/blob - modules/packetizer/h264.c
* h264: less debug.
[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     vlc_bool_t   b_pps;
72
73     /* avcC data */
74     int i_avcC_length_size;
75
76     /* Useful values of the Sequence Parameter Set */
77     int i_log2_max_frame_num;
78     int b_frame_mbs_only;
79
80     /* Useful values of the Slice Header */
81     int i_nal_type;
82     int i_nal_ref_idc;
83     int i_idr_pic_id;
84     int i_frame_num;
85 };
86
87 enum
88 {
89     STATE_NOSYNC,
90     STATE_NEXT_SYNC,
91 };
92
93 enum nal_unit_type_e
94 {
95     NAL_UNKNOWN = 0,
96     NAL_SLICE   = 1,
97     NAL_SLICE_DPA   = 2,
98     NAL_SLICE_DPB   = 3,
99     NAL_SLICE_DPC   = 4,
100     NAL_SLICE_IDR   = 5,    /* ref_idc != 0 */
101     NAL_SEI         = 6,    /* ref_idc == 0 */
102     NAL_SPS         = 7,
103     NAL_PPS         = 8,
104     NAL_AU_DELIMITER= 9
105     /* ref_idc == 0 for 6,9,10,11,12 */
106 };
107
108 enum nal_priority_e
109 {
110     NAL_PRIORITY_DISPOSABLE = 0,
111     NAL_PRIORITY_LOW        = 1,
112     NAL_PRIORITY_HIGH       = 2,
113     NAL_PRIORITY_HIGHEST    = 3,
114 };
115
116 static block_t *ParseNALBlock( decoder_t *, block_t * );
117
118 static block_t *nal_get_annexeb( decoder_t *, uint8_t *p, int );
119
120 /*****************************************************************************
121  * Open: probe the packetizer and return score
122  *****************************************************************************/
123 static int Open( vlc_object_t *p_this )
124 {
125     decoder_t     *p_dec = (decoder_t*)p_this;
126     decoder_sys_t *p_sys;
127
128     if( p_dec->fmt_in.i_codec != VLC_FOURCC( 'h', '2', '6', '4') &&
129         p_dec->fmt_in.i_codec != VLC_FOURCC( 'H', '2', '6', '4') &&
130         p_dec->fmt_in.i_codec != VLC_FOURCC( 'V', 'S', 'S', 'H') &&
131         p_dec->fmt_in.i_codec != VLC_FOURCC( 'v', 's', 's', 'h') &&
132         ( p_dec->fmt_in.i_codec != VLC_FOURCC( 'a', 'v', 'c', '1') ||
133           p_dec->fmt_in.i_extra < 7 ) )
134     {
135         return VLC_EGENERIC;
136     }
137
138     /* Allocate the memory needed to store the decoder's structure */
139     if( ( p_dec->p_sys = p_sys = malloc( sizeof(decoder_sys_t) ) ) == NULL )
140     {
141         msg_Err( p_dec, "out of memory" );
142         return VLC_EGENERIC;
143     }
144     p_sys->i_state = STATE_NOSYNC;
145     p_sys->i_offset = 0;
146     p_sys->startcode[0] = 0;
147     p_sys->startcode[1] = 0;
148     p_sys->startcode[2] = 0;
149     p_sys->startcode[3] = 1;
150     p_sys->bytestream = block_BytestreamInit( p_dec );
151     p_sys->b_slice = VLC_FALSE;
152     p_sys->p_frame = NULL;
153     p_sys->b_sps   = VLC_FALSE;
154     p_sys->b_pps   = VLC_FALSE;
155
156     p_sys->i_nal_type = -1;
157     p_sys->i_nal_ref_idc = -1;
158     p_sys->i_idr_pic_id = -1;
159     p_sys->i_frame_num = -1;
160
161     /* Setup properties */
162     es_format_Copy( &p_dec->fmt_out, &p_dec->fmt_in );
163     p_dec->fmt_out.i_codec = VLC_FOURCC( 'h', '2', '6', '4' );
164     /* FIXME: FFMPEG isn't happy at all if you leave this */
165     if( p_dec->fmt_out.i_extra ) free( p_dec->fmt_out.p_extra );
166     p_dec->fmt_out.i_extra = 0; p_dec->fmt_out.p_extra = 0;
167
168     if( p_dec->fmt_in.i_codec == VLC_FOURCC( 'a', 'v', 'c', '1' ) )
169     {
170         uint8_t *p = &((uint8_t*)p_dec->fmt_in.p_extra)[4];
171         int i_sps, i_pps;
172         int i;
173
174         /* Parse avcC */
175         p_sys->i_avcC_length_size = 1 + ((*p++)&0x03);
176
177         /* Read SPS */
178         i_sps = (*p++)&0x1f;
179
180         for( i = 0; i < i_sps; i++ )
181         {
182             int i_length = GetWBE( p );
183             block_t *p_sps = nal_get_annexeb( p_dec, p + 2, i_length );
184
185             ParseNALBlock( p_dec, p_sps );
186             p += 2 + i_length;
187         }
188         /* Read PPS */
189         i_pps = *p++;
190         for( i = 0; i < i_pps; i++ )
191         {
192             int i_length = GetWBE( p );
193             block_t *p_pps = nal_get_annexeb( p_dec, p + 2, i_length );
194
195             ParseNALBlock( p_dec, p_pps );
196             p += 2 + i_length;
197         }
198         msg_Dbg( p_dec, "avcC length size=%d sps=%d pps=%d",
199                  p_sys->i_avcC_length_size, i_sps, i_pps );
200
201         /* Set callback */
202         p_dec->pf_packetize = PacketizeAVC1;
203     }
204     else
205     {
206         /* Set callback */
207         p_dec->pf_packetize = Packetize;
208     }
209
210     return VLC_SUCCESS;
211 }
212
213 /*****************************************************************************
214  * Close: clean up the packetizer
215  *****************************************************************************/
216 static void Close( vlc_object_t *p_this )
217 {
218     decoder_t *p_dec = (decoder_t*)p_this;
219     decoder_sys_t *p_sys = p_dec->p_sys;
220
221     block_BytestreamRelease( &p_sys->bytestream );
222     free( p_sys );
223 }
224
225 /****************************************************************************
226  * Packetize: the whole thing
227  ****************************************************************************/
228 static block_t *Packetize( decoder_t *p_dec, block_t **pp_block )
229 {
230     decoder_sys_t *p_sys = p_dec->p_sys;
231     block_t       *p_pic;
232
233     if( !pp_block || !*pp_block ) return NULL;
234
235     block_BytestreamPush( &p_sys->bytestream, *pp_block );
236
237     for( ;; )
238     {
239         switch( p_sys->i_state )
240         {
241             case STATE_NOSYNC:
242                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
243                       &p_sys->i_offset, p_sys->startcode+1, 3 ) == VLC_SUCCESS)
244                 {
245                     p_sys->i_state = STATE_NEXT_SYNC;
246                 }
247
248                 if( p_sys->i_offset )
249                 {
250                     block_SkipBytes( &p_sys->bytestream, p_sys->i_offset );
251                     p_sys->i_offset = 0;
252                     block_BytestreamFlush( &p_sys->bytestream );
253                 }
254
255                 if( p_sys->i_state != STATE_NEXT_SYNC )
256                 {
257                     /* Need more data */
258                     return NULL;
259                 }
260
261                 p_sys->i_offset = 1; /* To find next startcode */
262
263             case STATE_NEXT_SYNC:
264                 /* Find the next startcode */
265                 if( block_FindStartcodeFromOffset( &p_sys->bytestream,
266                       &p_sys->i_offset, p_sys->startcode, 3 ) != VLC_SUCCESS)
267                 {
268                     if( block_FindStartcodeFromOffset( &p_sys->bytestream,
269                           &p_sys->i_offset, p_sys->startcode+1, 3 ) !=
270                         VLC_SUCCESS )
271                     {
272                         /* Need more data */
273                         return NULL;
274                     }
275                 }
276
277                 /* Get the new fragment and set the pts/dts */
278                 p_pic = block_New( p_dec, p_sys->i_offset );
279                 p_pic->i_pts = p_sys->bytestream.p_block->i_pts;
280                 p_pic->i_dts = p_sys->bytestream.p_block->i_dts;
281
282                 block_GetBytes( &p_sys->bytestream, p_pic->p_buffer,
283                                 p_pic->i_buffer );
284
285                 p_sys->i_offset = 0;
286
287                 /* Parse the NAL */
288                 if( !( p_pic = ParseNALBlock( p_dec, p_pic ) ) )
289                 {
290                     p_sys->i_state = STATE_NOSYNC;
291                     break;
292                 }
293 #if 0
294                 msg_Dbg( p_dec, "pts="I64Fd" dts="I64Fd,
295                          p_pic->i_pts, p_pic->i_dts );
296 #endif
297
298                 /* So p_block doesn't get re-added several times */
299                 *pp_block = block_BytestreamPop( &p_sys->bytestream );
300
301                 p_sys->i_state = STATE_NOSYNC;
302
303                 return p_pic;
304         }
305     }
306 }
307
308 /****************************************************************************
309  * PacketizeAVC1: the whole thing
310  ****************************************************************************/
311 static block_t *PacketizeAVC1( decoder_t *p_dec, block_t **pp_block )
312 {
313     decoder_sys_t *p_sys = p_dec->p_sys;
314     block_t       *p_block;
315     block_t       *p_ret = NULL;
316     uint8_t       *p;
317
318     if( !pp_block || !*pp_block ) return NULL;
319
320     p_block = *pp_block;
321     *pp_block = NULL;
322
323     for( p = p_block->p_buffer; p < &p_block->p_buffer[p_block->i_buffer]; )
324     {
325         block_t *p_pic;
326         int i_size = 0;
327         int i;
328
329         for( i = 0; i < p_sys->i_avcC_length_size; i++ )
330         {
331             i_size = (i_size << 8) | (*p++);
332         }
333
334         if( i_size > 0 )
335         {
336             block_t *p_part = nal_get_annexeb( p_dec, p, i_size );
337
338             p_part->i_dts = p_block->i_dts;
339             p_part->i_pts = p_block->i_pts;
340
341             /* Parse the NAL */
342             if( ( p_pic = ParseNALBlock( p_dec, p_part ) ) )
343             {
344                 block_ChainAppend( &p_ret, p_pic );
345             }
346         }
347         p += i_size;
348     }
349
350     return p_ret;
351 }
352
353 static block_t *nal_get_annexeb( decoder_t *p_dec, uint8_t *p, int i_size )
354 {
355     block_t *p_nal;
356
357     p_nal = block_New( p_dec, 3 + i_size );
358
359     /* Add start code */
360     p_nal->p_buffer[0] = 0x00;
361     p_nal->p_buffer[1] = 0x00;
362     p_nal->p_buffer[2] = 0x01;
363
364     /* Copy nalu */
365     memcpy( &p_nal->p_buffer[3], p, i_size );
366
367     return p_nal;
368 }
369
370 static void nal_get_decoded( uint8_t **pp_ret, int *pi_ret,
371                              uint8_t *src, int i_src )
372 {
373     uint8_t *end = &src[i_src];
374     uint8_t *dst = malloc( i_src );
375
376     *pp_ret = dst;
377
378     while( src < end )
379     {
380         if( src < end - 3 && src[0] == 0x00 && src[1] == 0x00 &&
381             src[2] == 0x03 )
382         {
383             *dst++ = 0x00;
384             *dst++ = 0x00;
385
386             src += 3;
387             continue;
388         }
389         *dst++ = *src++;
390     }
391
392     *pi_ret = dst - *pp_ret;
393 }
394
395 static inline int bs_read_ue( bs_t *s )
396 {
397     int i = 0;
398
399     while( bs_read1( s ) == 0 && s->p < s->p_end && i < 32 )
400     {
401         i++;
402     }
403     return( ( 1 << i) - 1 + bs_read( s, i ) );
404 }
405
406 static inline int bs_read_se( bs_t *s )
407 {
408     int val = bs_read_ue( s );
409
410     return val&0x01 ? (val+1)/2 : -(val/2);
411 }
412
413
414 static block_t *ParseNALBlock( decoder_t *p_dec, block_t *p_frag )
415 {
416     decoder_sys_t *p_sys = p_dec->p_sys;
417     block_t *p_pic = NULL;
418
419     const int i_nal_ref_idc = (p_frag->p_buffer[3] >> 5)&0x03;
420     const int i_nal_type = p_frag->p_buffer[3]&0x1f;
421
422 #define OUTPUT \
423     do {                                                \
424         p_pic = block_ChainGather( p_sys->p_frame );    \
425         p_pic->i_length = 0;    /* FIXME */             \
426                                                         \
427         p_sys->p_frame = NULL;                          \
428         p_sys->b_slice = VLC_FALSE;                     \
429     } while(0)
430
431
432     if( p_sys->b_slice && !p_sys->b_sps )
433     {
434         block_ChainRelease( p_sys->p_frame );
435         msg_Warn( p_dec, "waiting for SPS" );
436
437         /* Reset context */
438         p_sys->p_frame = NULL;
439         p_sys->b_slice = VLC_FALSE;
440     }
441
442     if( !p_sys->b_sps &&
443         i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
444     {
445         p_sys->b_slice = VLC_TRUE;
446         /* Fragment will be discarded later on */
447     }
448     else if( i_nal_type >= NAL_SLICE && i_nal_type <= NAL_SLICE_IDR )
449     {
450         uint8_t *dec;
451         int i_dec, i_first_mb, i_slice_type, i_frame_num, i_pic_flags = 0;
452         vlc_bool_t b_pic = VLC_FALSE;
453         bs_t s;
454
455         /* do not convert the whole frame */
456         nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
457                          __MIN( p_frag->i_buffer - 4, 60 ) );
458         bs_init( &s, dec, i_dec );
459
460         /* first_mb_in_slice */
461         i_first_mb = bs_read_ue( &s );
462
463         /* slice_type */
464         switch( (i_slice_type = bs_read_ue( &s )) )
465         {
466             case 0: case 5:
467                 i_pic_flags = BLOCK_FLAG_TYPE_P;
468                 break;
469             case 1: case 6:
470                 i_pic_flags = BLOCK_FLAG_TYPE_B;
471                 break;
472             case 2: case 7:
473                 i_pic_flags = BLOCK_FLAG_TYPE_I;
474                 break;
475             case 3: case 8: /* SP */
476                 i_pic_flags = BLOCK_FLAG_TYPE_P;
477                 break;
478             case 4: case 9:
479                 i_pic_flags = BLOCK_FLAG_TYPE_I;
480                 break;
481         }
482
483         /* pic_parameter_set_id */
484         bs_read_ue( &s );
485         /* frame_num */
486         i_frame_num = bs_read( &s, p_sys->i_log2_max_frame_num + 4 );
487
488         /* Detection of the first VCL NAL unit of a primary coded picture
489          * (cf. 7.4.1.2.4) */
490         if( i_frame_num != p_sys->i_frame_num ||
491             ( (i_nal_ref_idc != p_sys->i_nal_ref_idc) &&
492               (!i_nal_ref_idc || !p_sys->i_nal_ref_idc) ) )
493         {
494             b_pic = VLC_TRUE;
495         }
496         p_sys->i_frame_num = i_frame_num;
497         p_sys->i_nal_ref_idc = i_nal_ref_idc;
498
499         if( !p_sys->b_frame_mbs_only )
500         {
501             /* field_pic_flag */
502             if( bs_read( &s, 1 ) )
503             {
504                 /* bottom_field_flag */
505                 bs_read( &s, 1 );
506             }
507         }
508
509         if( i_nal_type == NAL_SLICE_IDR )
510         {
511             /* id_pic_id */
512             int i_idr_pic_id = bs_read_ue( &s );
513             if( p_sys->i_nal_type != i_nal_type ) b_pic = VLC_TRUE;
514             if( p_sys->i_idr_pic_id != i_idr_pic_id ) b_pic = VLC_TRUE;
515             p_sys->i_idr_pic_id = i_idr_pic_id;
516         }
517         p_sys->i_nal_type = i_nal_type;
518
519         if( b_pic && p_sys->b_slice )
520             OUTPUT;
521
522         p_sys->b_slice = VLC_TRUE;
523
524         free( dec );
525     }
526     else if( i_nal_type == NAL_SPS )
527     {
528         uint8_t *dec;
529         int     i_dec;
530         bs_t s;
531         int i_tmp;
532
533         if( !p_sys->b_sps )
534             msg_Dbg( p_dec, "found NAL_SPS" );
535
536         p_sys->b_sps = VLC_TRUE;
537
538         nal_get_decoded( &dec, &i_dec, &p_frag->p_buffer[4],
539                          p_frag->i_buffer - 4 );
540
541         bs_init( &s, dec, i_dec );
542         /* Skip profile(8), constraint_set012, reserver(5), level(8) */
543         bs_skip( &s, 8 + 1+1+1 + 5 + 8 );
544         /* sps id */
545         bs_read_ue( &s );
546         /* Skip i_log2_max_frame_num */
547         p_sys->i_log2_max_frame_num = bs_read_ue( &s );
548         /* Read poc_type */
549         i_tmp = bs_read_ue( &s );
550         if( i_tmp == 0 )
551         {
552             /* skip i_log2_max_poc_lsb */
553             bs_read_ue( &s );
554         }
555         else if( i_tmp == 1 )
556         {
557             int i_cycle;
558             /* skip b_delta_pic_order_always_zero */
559             bs_skip( &s, 1 );
560             /* skip i_offset_for_non_ref_pic */
561             bs_read_se( &s );
562             /* skip i_offset_for_top_to_bottom_field */
563             bs_read_se( &s );
564             /* read i_num_ref_frames_in_poc_cycle */
565             i_cycle = bs_read_ue( &s );
566             if( i_cycle > 256 ) i_cycle = 256;
567             while( i_cycle > 0 )
568             {
569                 /* skip i_offset_for_ref_frame */
570                 bs_read_se(&s );
571             }
572         }
573         /* i_num_ref_frames */
574         bs_read_ue( &s );
575         /* b_gaps_in_frame_num_value_allowed */
576         bs_skip( &s, 1 );
577
578         /* Read size */
579         p_dec->fmt_out.video.i_width  = 16 * ( bs_read_ue( &s ) + 1 );
580         p_dec->fmt_out.video.i_height = 16 * ( bs_read_ue( &s ) + 1 );
581
582         /* b_frame_mbs_only */
583         p_sys->b_frame_mbs_only = bs_read( &s, 1 );
584         if( p_sys->b_frame_mbs_only == 0 )
585         {
586             bs_skip( &s, 1 );
587         }
588         /* b_direct8x8_inference */
589         bs_skip( &s, 1 );
590
591         /* crop */
592         i_tmp = bs_read( &s, 1 );
593         if( i_tmp )
594         {
595             /* left */
596             bs_read_ue( &s );
597             /* right */
598             bs_read_ue( &s );
599             /* top */
600             bs_read_ue( &s );
601             /* bottom */
602             bs_read_ue( &s );
603         }
604
605         /* vui */
606         i_tmp = bs_read( &s, 1 );
607         if( i_tmp )
608         {
609             /* read the aspect ratio part if any FIXME check it */
610             i_tmp = bs_read( &s, 1 );
611             if( i_tmp )
612             {
613                 static const struct { int w, h; } sar[14] =
614                 {
615                     { 0,   0 }, { 1,   1 }, { 12, 11 }, { 10, 11 },
616                     { 16, 11 }, { 40, 33 }, { 24, 11 }, { 20, 11 },
617                     { 32, 11 }, { 80, 33 }, { 18, 11 }, { 15, 11 },
618                     { 64, 33 }, { 160,99 },
619                 };
620                 int i_sar = bs_read( &s, 8 );
621                 int w, h;
622
623                 if( i_sar < 14 )
624                 {
625                     w = sar[i_sar].w;
626                     h = sar[i_sar].h;
627                 }
628                 else
629                 {
630                     w = bs_read( &s, 16 );
631                     h = bs_read( &s, 16 );
632                 }
633                 p_dec->fmt_out.video.i_aspect =
634                     VOUT_ASPECT_FACTOR * w / h * p_dec->fmt_out.video.i_width /
635                     p_dec->fmt_out.video.i_height;
636             }
637         }
638
639         free( dec );
640
641
642         if( p_sys->b_slice )
643             OUTPUT;
644     }
645     else if( i_nal_type == NAL_PPS )
646     {
647         bs_t s;
648         bs_init( &s, &p_frag->p_buffer[4], p_frag->i_buffer - 4 );
649
650         if( !p_sys->b_pps )
651             msg_Dbg( p_dec, "found NAL_PPS" );
652         p_sys->b_pps = VLC_TRUE;
653
654         /* TODO */
655
656         if( p_sys->b_slice )
657             OUTPUT;
658     }
659     else if( i_nal_type == NAL_AU_DELIMITER ||
660              i_nal_type == NAL_SEI ||
661              ( i_nal_type >= 13 && i_nal_type <= 18 ) )
662     {
663         if( p_sys->b_slice )
664             OUTPUT;
665     }
666
667 #undef OUTPUT
668
669     /* Append the block */
670     block_ChainAppend( &p_sys->p_frame, p_frag );
671
672     return p_pic;
673 }