OSDN Git Service

Excises xvid from libhb because it's not worthy. Having two different MPEG-4 Part...
[handbrake-jp/handbrake-jp-git.git] / libhb / sync.c
1 /* $Id: sync.c,v 1.38 2005/04/14 21:57:58 titer Exp $
2
3    This file is part of the HandBrake source code.
4    Homepage: <http://handbrake.fr/>.
5    It may be used under the terms of the GNU General Public License. */
6
7 #include "hb.h"
8 #include "hbffmpeg.h"
9 #include <stdio.h>
10 #include "samplerate.h"
11
12 #ifdef INT64_MIN
13 #undef INT64_MIN /* Because it isn't defined correctly in Zeta */
14 #endif
15 #define INT64_MIN (-9223372036854775807LL-1)
16
17 #define AC3_SAMPLES_PER_FRAME 1536
18
19 typedef struct
20 {
21     hb_audio_t * audio;
22
23     int64_t      next_start;    /* start time of next output frame */
24     int64_t      next_pts;      /* start time of next input frame */
25     int64_t      first_drop;    /* PTS of first 'went backwards' frame dropped */
26     int          drop_count;    /* count of 'time went backwards' drops */
27
28     /* Raw */
29     SRC_STATE  * state;
30     SRC_DATA     data;
31
32     /* AC-3 */
33     int          ac3_size;
34     uint8_t    * ac3_buf;
35
36 } hb_sync_audio_t;
37
38 struct hb_work_private_s
39 {
40     hb_job_t * job;
41     int        busy;            // bitmask with one bit for each active input
42                                 // (bit 0 = video; 1 = audio 0, 2 = audio 1, ...
43                                 // appropriate bit is cleared when input gets
44                                 // an eof buf. syncWork returns done when all
45                                 // bits are clear.
46     /* Video */
47     int64_t pts_offset;
48     int64_t next_start;         /* start time of next output frame */
49     int64_t next_pts;           /* start time of next input frame */
50     int64_t first_drop;         /* PTS of first 'went backwards' frame dropped */
51     int drop_count;             /* count of 'time went backwards' drops */
52     int drops;                  /* frames dropped to make a cbr video stream */
53     int dups;                   /* frames duplicated to make a cbr video stream */
54     int video_sequence;
55     int count_frames;
56     int count_frames_max;
57     int chap_mark;              /* to propagate chapter mark across a drop */
58     hb_buffer_t * cur; /* The next picture to process */
59
60     /* Audio */
61     hb_sync_audio_t sync_audio[8];
62     int64_t audio_passthru_slip;
63
64     /* Statistics */
65     uint64_t st_counts[4];
66     uint64_t st_dates[4];
67     uint64_t st_first;
68 };
69
70 /***********************************************************************
71  * Local prototypes
72  **********************************************************************/
73 static void InitAudio( hb_work_object_t * w, int i );
74 static void SyncVideo( hb_work_object_t * w );
75 static void SyncAudio( hb_work_object_t * w, int i );
76 static void InsertSilence( hb_work_object_t * w, int i, int64_t d );
77 static void UpdateState( hb_work_object_t * w );
78
79 /***********************************************************************
80  * hb_work_sync_init
81  ***********************************************************************
82  * Initialize the work object
83  **********************************************************************/
84 int syncInit( hb_work_object_t * w, hb_job_t * job )
85 {
86     hb_title_t       * title = job->title;
87     hb_chapter_t     * chapter;
88     int                i;
89     uint64_t           duration;
90     hb_work_private_t * pv;
91
92     pv = calloc( 1, sizeof( hb_work_private_t ) );
93     w->private_data = pv;
94
95     pv->job            = job;
96     pv->pts_offset     = INT64_MIN;
97
98     /* Calculate how many video frames we are expecting */
99     if (job->pts_to_stop)
100     {
101         duration = job->pts_to_stop + 90000;
102     }
103     else if( job->frame_to_stop )
104     {
105         /* Set the duration to a rough estimate */
106         duration = ( job->frame_to_stop / ( job->vrate / job->vrate_base ) ) * 90000;
107     }
108     else
109     {
110         duration = 0;
111         for( i = job->chapter_start; i <= job->chapter_end; i++ )
112         {
113             chapter   = hb_list_item( title->list_chapter, i - 1 );
114             duration += chapter->duration;
115         }
116         duration += 90000;
117         /* 1 second safety so we're sure we won't miss anything */
118     }
119     pv->count_frames_max = duration * job->vrate / job->vrate_base / 90000;
120
121     hb_log( "sync: expecting %d video frames", pv->count_frames_max );
122     pv->busy |= 1;
123
124     /* Initialize libsamplerate for every audio track we have */
125     if ( ! job->indepth_scan )
126     {
127         for( i = 0; i < hb_list_count( title->list_audio ) && i < 8; i++ )
128         {
129             pv->busy |= ( 1 << (i + 1) );
130             InitAudio( w, i );
131         }
132     }
133
134     return 0;
135 }
136
137 /***********************************************************************
138  * Close
139  ***********************************************************************
140  *
141  **********************************************************************/
142 void syncClose( hb_work_object_t * w )
143 {
144     hb_work_private_t * pv = w->private_data;
145     hb_job_t          * job   = pv->job;
146     hb_title_t        * title = job->title;
147     hb_audio_t        * audio = NULL;
148     int i;
149
150     if( pv->cur )
151     {
152         hb_buffer_close( &pv->cur );
153     }
154
155     hb_log( "sync: got %d frames, %d expected",
156             pv->count_frames, pv->count_frames_max );
157
158     if (pv->drops || pv->dups )
159     {
160         hb_log( "sync: %d frames dropped, %d duplicated", pv->drops, pv->dups );
161     }
162
163     for( i = 0; i < hb_list_count( title->list_audio ); i++ )
164     {
165         audio = hb_list_item( title->list_audio, i );
166         if( audio->config.out.codec == HB_ACODEC_AC3 )
167         {
168             free( pv->sync_audio[i].ac3_buf );
169         }
170         else
171         {
172             src_delete( pv->sync_audio[i].state );
173         }
174     }
175
176     free( pv );
177     w->private_data = NULL;
178 }
179
180 /***********************************************************************
181  * Work
182  ***********************************************************************
183  * The root routine of this work abject
184  *
185  * The way this works is that we are syncing the audio to the PTS of
186  * the last video that we processed. That's why we skip the audio sync
187  * if we haven't got a valid PTS from the video yet.
188  *
189  **********************************************************************/
190 int syncWork( hb_work_object_t * w, hb_buffer_t ** unused1,
191               hb_buffer_t ** unused2 )
192 {
193     hb_work_private_t * pv = w->private_data;
194     int i;
195
196     if ( pv->busy & 1 )
197         SyncVideo( w );
198
199     for( i = 0; i < hb_list_count( pv->job->title->list_audio ); i++ )
200     {
201         if ( pv->busy & ( 1 << (i + 1) ) )
202             SyncAudio( w, i );
203     }
204
205     return ( pv->busy? HB_WORK_OK : HB_WORK_DONE );
206 }
207
208 hb_work_object_t hb_sync =
209 {
210     WORK_SYNC,
211     "Synchronization",
212     syncInit,
213     syncWork,
214     syncClose
215 };
216
217 static void InitAudio( hb_work_object_t * w, int i )
218 {
219     hb_work_private_t * pv = w->private_data;
220     hb_job_t        * job   = pv->job;
221     hb_title_t      * title = job->title;
222     hb_sync_audio_t * sync;
223
224     sync        = &pv->sync_audio[i];
225     sync->audio = hb_list_item( title->list_audio, i );
226
227     if( sync->audio->config.out.codec == HB_ACODEC_AC3 )
228     {
229         /* Have a silent AC-3 frame ready in case we have to fill a
230            gap */
231         AVCodec        * codec;
232         AVCodecContext * c;
233         short          * zeros;
234
235         codec = avcodec_find_encoder( CODEC_ID_AC3 );
236         c     = avcodec_alloc_context();
237
238         c->bit_rate    = sync->audio->config.in.bitrate;
239         c->sample_rate = sync->audio->config.in.samplerate;
240         c->channels    = HB_INPUT_CH_LAYOUT_GET_DISCRETE_COUNT( sync->audio->config.in.channel_layout );
241
242         if( hb_avcodec_open( c, codec ) < 0 )
243         {
244             hb_log( "sync: avcodec_open failed" );
245             return;
246         }
247
248         zeros          = calloc( AC3_SAMPLES_PER_FRAME *
249                                  sizeof( short ) * c->channels, 1 );
250         sync->ac3_size = sync->audio->config.in.bitrate * AC3_SAMPLES_PER_FRAME /
251                              sync->audio->config.in.samplerate / 8;
252         sync->ac3_buf  = malloc( sync->ac3_size );
253
254         if( avcodec_encode_audio( c, sync->ac3_buf, sync->ac3_size,
255                                   zeros ) != sync->ac3_size )
256         {
257             hb_log( "sync: avcodec_encode_audio failed" );
258         }
259
260         free( zeros );
261         hb_avcodec_close( c );
262         av_free( c );
263     }
264     else
265     {
266         /* Initialize libsamplerate */
267         int error;
268         sync->state             = src_new( SRC_SINC_MEDIUM_QUALITY, HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->config.out.mixdown), &error );
269         sync->data.end_of_input = 0;
270     }
271 }
272
273 /***********************************************************************
274  * SyncVideo
275  ***********************************************************************
276  *
277  **********************************************************************/
278 static void SyncVideo( hb_work_object_t * w )
279 {
280     hb_work_private_t * pv = w->private_data;
281     hb_buffer_t * cur, * next, * sub = NULL;
282     hb_job_t * job = pv->job;
283     hb_subtitle_t *subtitle;
284     int i;
285
286     if( !pv->cur && !( pv->cur = hb_fifo_get( job->fifo_raw ) ) )
287     {
288         /* We haven't even got a frame yet */
289         return;
290     }
291     cur = pv->cur;
292     if( cur->size == 0 )
293     {
294         /* we got an end-of-stream. Feed it downstream & signal that we're done. */
295         hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
296
297         /*
298          * Push through any subtitle EOFs in case they were not synced through.
299          */
300         for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
301         {
302             subtitle = hb_list_item( job->list_subtitle, i );
303             if( subtitle->dest == PASSTHRUSUB )
304             {
305                 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
306             }
307         }
308
309         pv->busy &=~ 1;
310         return;
311     }
312
313     /* At this point we have a frame to process. Let's check
314         1) if we will be able to push into the fifo ahead
315         2) if the next frame is there already, since we need it to
316            compute the duration of the current frame*/
317     while( !hb_fifo_is_full( job->fifo_sync ) &&
318            ( next = hb_fifo_see( job->fifo_raw ) ) )
319     {
320         hb_buffer_t * buf_tmp;
321
322         if( next->size == 0 )
323         {
324             /* we got an end-of-stream. Feed it downstream & signal that
325              * we're done. Note that this means we drop the final frame of
326              * video (we don't know its duration). On DVDs the final frame
327              * is often strange and dropping it seems to be a good idea. */
328             hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
329
330             /*
331              * Push through any subtitle EOFs in case they were not synced through.
332              */
333             for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
334             {
335                 subtitle = hb_list_item( job->list_subtitle, i );
336                 if( subtitle->dest == PASSTHRUSUB )
337                 {
338                     hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
339                 }
340             }
341             pv->busy &=~ 1;
342             return;
343         }
344         if( pv->pts_offset == INT64_MIN )
345         {
346             /* This is our first frame */
347             pv->pts_offset = 0;
348             if ( cur->start != 0 )
349             {
350                 /*
351                  * The first pts from a dvd should always be zero but
352                  * can be non-zero with a transport or program stream since
353                  * we're not guaranteed to start on an IDR frame. If we get
354                  * a non-zero initial PTS extend its duration so it behaves
355                  * as if it started at zero so that our audio timing will
356                  * be in sync.
357                  */
358                 hb_log( "sync: first pts is %lld", cur->start );
359                 cur->start = 0;
360             }
361         }
362
363         /*
364          * since the first frame is always 0 and the upstream reader code
365          * is taking care of adjusting for pts discontinuities, we just have
366          * to deal with the next frame's start being in the past. This can
367          * happen when the PTS is adjusted after data loss but video frame
368          * reordering causes some frames with the old clock to appear after
369          * the clock change. This creates frames that overlap in time which
370          * looks to us like time going backward. The downstream muxing code
371          * can deal with overlaps of up to a frame time but anything larger
372          * we handle by dropping frames here.
373          */
374         if ( (int64_t)( next->start - cur->start ) <= 0 ||
375              (int64_t)( (cur->start - pv->audio_passthru_slip ) - pv->next_pts ) < 0 )
376         {
377             if ( pv->first_drop == 0 )
378             {
379                 pv->first_drop = next->start;
380             }
381             ++pv->drop_count;
382             buf_tmp = hb_fifo_get( job->fifo_raw );
383             if ( buf_tmp->new_chap )
384             {
385                 // don't drop a chapter mark when we drop the buffer
386                 pv->chap_mark = buf_tmp->new_chap;
387             }
388             hb_buffer_close( &buf_tmp );
389             continue;
390         }
391         if ( pv->first_drop )
392         {
393             hb_log( "sync: video time didn't advance - dropped %d frames "
394                     "(delta %d ms, current %lld, next %lld, dur %d)",
395                     pv->drop_count, (int)( cur->start - pv->first_drop ) / 90,
396                     cur->start, next->start, (int)( next->start - cur->start ) );
397             pv->first_drop = 0;
398             pv->drop_count = 0;
399         }
400
401         /*
402          * Track the video sequence number localy so that we can sync the audio
403          * to it using the sequence number as well as the PTS.
404          */
405         pv->video_sequence = cur->sequence;
406
407         /*
408          * Look for a subtitle for this frame.
409          *
410          * If found then it will be tagged onto a video buffer of the correct time and 
411          * sent in to the render pipeline. This only needs to be done for VOBSUBs which
412          * get rendered, other types of subtitles can just sit in their raw_queue until
413          * delt with at muxing.
414          */
415         for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
416         {
417             subtitle = hb_list_item( job->list_subtitle, i );
418
419             /*
420              * Rewrite timestamps on subtitles that need it (on raw queue).
421              */
422             if( subtitle->source == CC608SUB ||
423                 subtitle->source == CC708SUB )
424             {
425                 /*
426                  * Rewrite timestamps on subtitles that came from Closed Captions
427                  * since they are using the MPEG2 timestamps.
428                  */
429                 while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
430                 {
431                     /*
432                      * Rewrite the timestamps as and when the video
433                      * (cur->start) reaches the same timestamp as a
434                      * closed caption (sub->start).
435                      *
436                      * What about discontinuity boundaries - not delt
437                      * with here - Van?
438                      *
439                      * Bypass the sync fifo altogether.
440                      */
441                     if( sub->size <= 0 )
442                     {
443                         sub = hb_fifo_get( subtitle->fifo_raw );
444                         hb_fifo_push( subtitle->fifo_out, sub );
445                         sub = NULL;
446                         break;
447                     } else {
448                         /*
449                          * Sync the subtitles to the incoming video, and use
450                          * the matching converted video timestamp.
451                          *
452                          * Note that it doesn't appear that we need to convert 
453                          * timestamps, I guess that they were already correct,
454                          * so just push them through for rendering.
455                          *
456                          */
457                         if( sub->start < cur->start )
458                         {
459                             uint64_t duration;
460                             duration = sub->stop - sub->start;
461                             sub = hb_fifo_get( subtitle->fifo_raw );
462                             hb_fifo_push( subtitle->fifo_out, sub );
463                         } else {
464                             sub = NULL;
465                             break;
466                         }
467                     }
468                 }
469             }
470
471             if( subtitle->source == VOBSUB ) 
472             {
473                 hb_buffer_t * sub2;
474                 while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
475                 {
476                     if( sub->size == 0 )
477                     {
478                         /*
479                          * EOF, pass it through immediately.
480                          */
481                         break;
482                     }
483
484                     /* If two subtitles overlap, make the first one stop
485                        when the second one starts */
486                     sub2 = hb_fifo_see2( subtitle->fifo_raw );
487                     if( sub2 && sub->stop > sub2->start )
488                     {
489                         sub->stop = sub2->start;
490                     }
491                     
492                     // hb_log("0x%x: video seq: %lld  subtitle sequence: %lld",
493                     //       sub, cur->sequence, sub->sequence);
494                     
495                     if( sub->sequence > cur->sequence )
496                     {
497                         /*
498                          * The video is behind where we are, so wait until
499                          * it catches up to the same reader point on the
500                          * DVD. Then our PTS should be in the same region
501                          * as the video.
502                          */
503                         sub = NULL;
504                         break;
505                     }
506                     
507                     if( sub->stop > cur->start ) {
508                         /*
509                          * The stop time is in the future, so fall through
510                          * and we'll deal with it in the next block of
511                          * code.
512                          */
513
514                         /*
515                          * There is a valid subtitle, is it time to display it?
516                          */
517                         if( sub->stop > sub->start)
518                         {
519                             /*
520                              * Normal subtitle which ends after it starts, 
521                              * check to see that the current video is between 
522                              * the start and end.
523                              */
524                             if( cur->start > sub->start &&
525                                 cur->start < sub->stop )
526                             {
527                                 /*
528                                 * We should be playing this, so leave the
529                                 * subtitle in place.
530                                 *
531                                 * fall through to display
532                                 */
533                                 if( ( sub->stop - sub->start ) < ( 2 * 90000 ) )
534                                 {
535                                     /*
536                                      * Subtitle is on for less than three 
537                                      * seconds, extend the time that it is 
538                                      * displayed to make it easier to read. 
539                                      * Make it 3 seconds or until the next
540                                      * subtitle is displayed.
541                                      *
542                                      * This is in response to Indochine which 
543                                      * only displays subs for 1 second - 
544                                      * too fast to read.
545                                      */
546                                     sub->stop = sub->start + ( 2 * 90000 );
547                                 
548                                     sub2 = hb_fifo_see2( subtitle->fifo_raw );
549                                 
550                                     if( sub2 && sub->stop > sub2->start )
551                                     {
552                                         sub->stop = sub2->start;
553                                     }
554                                 }
555                             }
556                             else
557                             {
558                                 /*
559                                  * Defer until the play point is within 
560                                  * the subtitle
561                                  */
562                                 sub = NULL;
563                             }
564                         }
565                         else
566                         {
567                             /*
568                              * The end of the subtitle is less than the start, 
569                              * this is a sign of a PTS discontinuity.
570                              */
571                             if( sub->start > cur->start )
572                             {
573                                 /*
574                                  * we haven't reached the start time yet, or
575                                  * we have jumped backwards after having
576                                  * already started this subtitle.
577                                  */
578                                 if( cur->start < sub->stop )
579                                 {
580                                     /*
581                                      * We have jumped backwards and so should
582                                      * continue displaying this subtitle.
583                                      *
584                                      * fall through to display.
585                                      */
586                                 }
587                                 else
588                                 {
589                                     /*
590                                      * Defer until the play point is 
591                                      * within the subtitle
592                                      */
593                                     sub = NULL;
594                                 }
595                             } else {
596                                 /*
597                                 * Play this subtitle as the start is 
598                                 * greater than our video point.
599                                 *
600                                 * fall through to display/
601                                 */
602                             }
603                         }
604                         break;
605                     }
606                     else
607                     {
608                     
609                         /*
610                          * The subtitle is older than this picture, trash it
611                          */
612                         sub = hb_fifo_get( subtitle->fifo_raw );
613                         hb_buffer_close( &sub );
614                     }
615                 }
616                 
617                 /* If we have a subtitle for this picture, copy it */
618                 /* FIXME: we should avoid this memcpy */
619                 if( sub )
620                 {
621                     if( sub->size > 0 )
622                     {
623                         if( subtitle->dest == RENDERSUB )
624                         {
625                             if ( cur->sub == NULL )
626                             {
627                                 /*
628                                  * Tack onto the video buffer for rendering
629                                  */
630                                 cur->sub         = hb_buffer_init( sub->size );
631                                 cur->sub->x      = sub->x;
632                                 cur->sub->y      = sub->y;
633                                 cur->sub->width  = sub->width;
634                                 cur->sub->height = sub->height;
635                                 memcpy( cur->sub->data, sub->data, sub->size ); 
636                             }
637                         } else {
638                             /*
639                              * Pass-Through, pop it off of the raw queue, 
640                              * rewrite times and make it available to be 
641                              * reencoded.
642                              */
643                             uint64_t sub_duration;
644                             sub = hb_fifo_get( subtitle->fifo_raw );
645                             sub_duration = sub->stop - sub->start;
646                             sub->start = cur->start;
647                             buf_tmp = hb_fifo_see( job->fifo_raw );
648                             int64_t duration = buf_tmp->start - cur->start;
649                             sub->stop = sub->start + duration;
650                             hb_fifo_push( subtitle->fifo_sync, sub );
651                         }
652                     } else {
653                         /*
654                         * EOF - consume for rendered, else pass through
655                         */
656                         if( subtitle->dest == RENDERSUB )
657                         {
658                             sub = hb_fifo_get( subtitle->fifo_raw );
659                             hb_buffer_close( &sub );
660                         } else {
661                             sub = hb_fifo_get( subtitle->fifo_raw );
662                             hb_fifo_push( subtitle->fifo_out, sub );
663                         }
664                     }
665                 }
666             }
667         } // end subtitles
668
669         /*
670          * Adjust the pts of the current frame so that it's contiguous
671          * with the previous frame. The start time of the current frame
672          * has to be the end time of the previous frame and the stop
673          * time has to be the start of the next frame.  We don't
674          * make any adjustments to the source timestamps other than removing
675          * the clock offsets (which also removes pts discontinuities).
676          * This means we automatically encode at the source's frame rate.
677          * MP2 uses an implicit duration (frames end when the next frame
678          * starts) but more advanced containers like MP4 use an explicit
679          * duration. Since we're looking ahead one frame we set the
680          * explicit stop time from the start time of the next frame.
681          */
682         buf_tmp = cur;
683         pv->cur = cur = hb_fifo_get( job->fifo_raw );
684         cur->sub = NULL;
685         pv->next_pts = cur->start;
686         int64_t duration = cur->start - buf_tmp->start;
687         if ( duration <= 0 )
688         {
689             hb_log( "sync: invalid video duration %lld, start %lld, next %lld",
690                     duration, buf_tmp->start, next->start );
691         }
692
693         buf_tmp->start = pv->next_start;
694         pv->next_start += duration;
695         buf_tmp->stop = pv->next_start;
696
697         if ( pv->chap_mark )
698         {
699             // we have a pending chapter mark from a recent drop - put it on this
700             // buffer (this may make it one frame late but we can't do any better).
701             buf_tmp->new_chap = pv->chap_mark;
702             pv->chap_mark = 0;
703         }
704
705         /* Push the frame to the renderer */
706         hb_fifo_push( job->fifo_sync, buf_tmp );
707
708         /* Update UI */
709         UpdateState( w );
710         
711         if( job->frame_to_stop && pv->count_frames > job->frame_to_stop )
712         {
713             // Drop an empty buffer into our output to ensure that things
714             // get flushed all the way out.
715             hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
716             pv->busy &=~ 1;
717             hb_log( "sync: reached %d frames, exiting early (%i busy)",
718                     pv->count_frames, pv->busy );
719             return;
720         }
721
722         /* Make sure we won't get more frames then expected */
723         if( pv->count_frames >= pv->count_frames_max * 2)
724         {
725             hb_log( "sync: got too many frames (%d), exiting early",
726                     pv->count_frames );
727
728             // Drop an empty buffer into our output to ensure that things
729             // get flushed all the way out.
730             hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
731             pv->busy &=~ 1;
732             return;
733         }
734     }
735 }
736
737 static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf,
738                               hb_sync_audio_t *sync, hb_fifo_t *fifo, int i )
739 {
740     int64_t start = sync->next_start;
741     int64_t duration = buf->stop - buf->start;
742
743     sync->next_pts += duration;
744
745     if( audio->config.in.samplerate == audio->config.out.samplerate ||
746         audio->config.out.codec == HB_ACODEC_AC3 ||
747         audio->config.out.codec == HB_ACODEC_DCA )
748     {
749         /*
750          * If we don't have to do sample rate conversion or this audio is 
751          * pass-thru just send the input buffer downstream after adjusting
752          * its timestamps to make the output stream continuous.
753          */
754     }
755     else
756     {
757         /* Not pass-thru - do sample rate conversion */
758         int count_in, count_out;
759         hb_buffer_t * buf_raw = buf;
760         int channel_count = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown) *
761                             sizeof( float );
762
763         count_in  = buf_raw->size / channel_count;
764         /*
765          * When using stupid rates like 44.1 there will always be some
766          * truncation error. E.g., a 1536 sample AC3 frame will turn into a
767          * 1536*44.1/48.0 = 1411.2 sample frame. If we just truncate the .2
768          * the error will build up over time and eventually the audio will
769          * substantially lag the video. libsamplerate will keep track of the
770          * fractional sample & give it to us when appropriate if we give it
771          * an extra sample of space in the output buffer.
772          */
773         count_out = ( duration * audio->config.out.samplerate ) / 90000 + 1;
774
775         sync->data.input_frames = count_in;
776         sync->data.output_frames = count_out;
777         sync->data.src_ratio = (double)audio->config.out.samplerate /
778                                (double)audio->config.in.samplerate;
779
780         buf = hb_buffer_init( count_out * channel_count );
781         sync->data.data_in  = (float *) buf_raw->data;
782         sync->data.data_out = (float *) buf->data;
783         if( src_process( sync->state, &sync->data ) )
784         {
785             /* XXX If this happens, we're screwed */
786             hb_log( "sync: audio %d src_process failed", i );
787         }
788         hb_buffer_close( &buf_raw );
789
790         buf->size = sync->data.output_frames_gen * channel_count;
791         duration = ( sync->data.output_frames_gen * 90000 ) /
792                    audio->config.out.samplerate;
793     }
794     buf->frametype = HB_FRAME_AUDIO;
795     buf->start = start;
796     buf->stop  = start + duration;
797     sync->next_start = start + duration;
798     hb_fifo_push( fifo, buf );
799 }
800
801 /***********************************************************************
802  * SyncAudio
803  ***********************************************************************
804  *
805  **********************************************************************/
806 static void SyncAudio( hb_work_object_t * w, int i )
807 {
808     hb_work_private_t * pv = w->private_data;
809     hb_job_t        * job = pv->job;
810     hb_sync_audio_t * sync = &pv->sync_audio[i];
811     hb_audio_t      * audio = sync->audio;
812     hb_buffer_t     * buf;
813     hb_fifo_t       * fifo;
814     int64_t start;
815
816     if( audio->config.out.codec == HB_ACODEC_AC3 ||
817         audio->config.out.codec == HB_ACODEC_DCA )
818     {
819         fifo = audio->priv.fifo_out;
820     }
821     else
822     {
823         fifo = audio->priv.fifo_sync;
824     }
825
826     while( !hb_fifo_is_full( fifo ) && ( buf = hb_fifo_see( audio->priv.fifo_raw ) ) )
827     {
828         start = buf->start - pv->audio_passthru_slip;
829         /* if the next buffer is an eof send it downstream */
830         if ( buf->size <= 0 )
831         {
832             buf = hb_fifo_get( audio->priv.fifo_raw );
833             hb_fifo_push( fifo, buf );
834             pv->busy &=~ (1 << (i + 1) );
835             return;
836         }
837         if( job->frame_to_stop && pv->count_frames >= job->frame_to_stop )
838         {
839             hb_fifo_push( fifo, hb_buffer_init(0) );
840             pv->busy &=~ (1 << (i + 1) );
841             return;
842         }
843         if ( (int64_t)( start - sync->next_pts ) < 0 )
844         {
845             // audio time went backwards.
846             // If our output clock is more than a half frame ahead of the
847             // input clock drop this frame to move closer to sync.
848             // Otherwise drop frames until the input clock matches the output clock.
849             if ( sync->first_drop || sync->next_start - start > 90*15 )
850             {
851                 // Discard data that's in the past.
852                 if ( sync->first_drop == 0 )
853                 {
854                     sync->first_drop = sync->next_pts;
855                 }
856                 ++sync->drop_count;
857                 buf = hb_fifo_get( audio->priv.fifo_raw );
858                 hb_buffer_close( &buf );
859                 continue;
860             }
861             sync->next_pts = start;
862         }
863         if ( sync->first_drop )
864         {
865             // we were dropping old data but input buf time is now current
866             hb_log( "sync: audio %d time went backwards %d ms, dropped %d frames "
867                     "(next %lld, current %lld)", i,
868                     (int)( sync->next_pts - sync->first_drop ) / 90,
869                     sync->drop_count, sync->first_drop, sync->next_pts );
870             sync->first_drop = 0;
871             sync->drop_count = 0;
872             sync->next_pts = start;
873         }
874         if ( start - sync->next_pts >= (90 * 70) )
875         {
876             if ( start - sync->next_pts > (90000LL * 60) )
877             {
878                 // there's a gap of more than a minute between the last
879                 // frame and this. assume we got a corrupted timestamp
880                 // and just drop the next buf.
881                 hb_log( "sync: %d minute time gap in audio %d - dropping buf"
882                         "  start %lld, next %lld",
883                         (int)((start - sync->next_pts) / (90000*60)),
884                         i, start, sync->next_pts );
885                 buf = hb_fifo_get( audio->priv.fifo_raw );
886                 hb_buffer_close( &buf );
887                 continue;
888             }
889             /*
890              * there's a gap of at least 70ms between the last
891              * frame we processed & the next. Fill it with silence.
892              * Or in the case of DCA, skip some frames from the
893              * other streams.
894              */
895             if( sync->audio->config.out.codec == HB_ACODEC_DCA )
896             {
897                 hb_log( "sync: audio gap %d ms. Skipping frames. Audio %d"
898                         "  start %lld, next %lld",
899                         (int)((start - sync->next_pts) / 90),
900                         i, start, sync->next_pts );
901                 pv->audio_passthru_slip += (start - sync->next_pts);
902                 return;
903             }
904             hb_log( "sync: adding %d ms of silence to audio %d"
905                     "  start %lld, next %lld",
906                     (int)((start - sync->next_pts) / 90),
907                     i, start, sync->next_pts );
908             InsertSilence( w, i, start - sync->next_pts );
909             return;
910         }
911
912         /*
913          * When we get here we've taken care of all the dups and gaps in the
914          * audio stream and are ready to inject the next input frame into
915          * the output stream.
916          */
917         buf = hb_fifo_get( audio->priv.fifo_raw );
918         OutputAudioFrame( job, audio, buf, sync, fifo, i );
919     }
920 }
921
922 static void InsertSilence( hb_work_object_t * w, int i, int64_t duration )
923 {
924     hb_work_private_t * pv = w->private_data;
925     hb_job_t        *job = pv->job;
926     hb_sync_audio_t *sync = &pv->sync_audio[i];
927     hb_buffer_t     *buf;
928     hb_fifo_t       *fifo;
929
930     // to keep pass-thru and regular audio in sync we generate silence in
931     // AC3 frame-sized units. If the silence duration isn't an integer multiple
932     // of the AC3 frame duration we will truncate or round up depending on
933     // which minimizes the timing error.
934     const int frame_dur = ( 90000 * AC3_SAMPLES_PER_FRAME ) /
935                           sync->audio->config.in.samplerate;
936     int frame_count = ( duration + (frame_dur >> 1) ) / frame_dur;
937
938     while ( --frame_count >= 0 )
939     {
940         if( sync->audio->config.out.codec == HB_ACODEC_AC3 )
941         {
942             buf        = hb_buffer_init( sync->ac3_size );
943             buf->start = sync->next_pts;
944             buf->stop  = buf->start + frame_dur;
945             memcpy( buf->data, sync->ac3_buf, buf->size );
946             fifo = sync->audio->priv.fifo_out;
947         }
948         else
949         {
950             buf = hb_buffer_init( AC3_SAMPLES_PER_FRAME * sizeof( float ) *
951                                      HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(
952                                          sync->audio->config.out.mixdown) );
953             buf->start = sync->next_pts;
954             buf->stop  = buf->start + frame_dur;
955             memset( buf->data, 0, buf->size );
956             fifo = sync->audio->priv.fifo_sync;
957         }
958         OutputAudioFrame( job, sync->audio, buf, sync, fifo, i );
959     }
960 }
961
962 static void UpdateState( hb_work_object_t * w )
963 {
964     hb_work_private_t * pv = w->private_data;
965     hb_state_t state;
966
967     if( !pv->count_frames )
968     {
969         pv->st_first = hb_get_date();
970     }
971     pv->count_frames++;
972
973     if( hb_get_date() > pv->st_dates[3] + 1000 )
974     {
975         memmove( &pv->st_dates[0], &pv->st_dates[1],
976                  3 * sizeof( uint64_t ) );
977         memmove( &pv->st_counts[0], &pv->st_counts[1],
978                  3 * sizeof( uint64_t ) );
979         pv->st_dates[3]  = hb_get_date();
980         pv->st_counts[3] = pv->count_frames;
981     }
982
983 #define p state.param.working
984     state.state = HB_STATE_WORKING;
985     p.progress  = (float) pv->count_frames / (float) pv->count_frames_max;
986     if( p.progress > 1.0 )
987     {
988         p.progress = 1.0;
989     }
990     p.rate_cur   = 1000.0 *
991         (float) ( pv->st_counts[3] - pv->st_counts[0] ) /
992         (float) ( pv->st_dates[3] - pv->st_dates[0] );
993     if( hb_get_date() > pv->st_first + 4000 )
994     {
995         int eta;
996         p.rate_avg = 1000.0 * (float) pv->st_counts[3] /
997             (float) ( pv->st_dates[3] - pv->st_first );
998         eta = (float) ( pv->count_frames_max - pv->st_counts[3] ) /
999             p.rate_avg;
1000         p.hours   = eta / 3600;
1001         p.minutes = ( eta % 3600 ) / 60;
1002         p.seconds = eta % 60;
1003     }
1004     else
1005     {
1006         p.rate_avg = 0.0;
1007         p.hours    = -1;
1008         p.minutes  = -1;
1009         p.seconds  = -1;
1010     }
1011 #undef p
1012
1013     hb_set_state( pv->job->h, &state );
1014 }