1 /* $Id: sync.c,v 1.38 2005/04/14 21:57:58 titer Exp $
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. */
10 #include "samplerate.h"
13 #undef INT64_MIN /* Because it isn't defined correctly in Zeta */
15 #define INT64_MIN (-9223372036854775807LL-1)
17 #define AC3_SAMPLES_PER_FRAME 1536
22 int ref; /* Reference count to tell us when it's unused */
24 int64_t audio_passthru_slip;
25 int64_t video_pts_slip;
28 /* Frame based point-to-point support */
29 int64_t audio_pts_thresh;
31 hb_cond_t * next_frame;
39 int64_t next_start; /* start time of next output frame */
40 int64_t next_pts; /* start time of next input frame */
41 int64_t first_drop; /* PTS of first 'went backwards' frame dropped */
42 int drop_count; /* count of 'time went backwards' drops */
58 int64_t next_start; /* start time of next output frame */
59 int64_t next_pts; /* start time of next input frame */
60 int64_t first_drop; /* PTS of first 'went backwards' frame dropped */
61 int drop_count; /* count of 'time went backwards' drops */
62 int drops; /* frames dropped to make a cbr video stream */
63 int dups; /* frames duplicated to make a cbr video stream */
66 int chap_mark; /* to propagate chapter mark across a drop */
67 hb_buffer_t * cur; /* The next picture to process */
70 uint64_t st_counts[4];
75 struct hb_work_private_s
78 hb_sync_common_t * common;
81 hb_sync_video_t video;
82 hb_sync_audio_t audio;
86 /***********************************************************************
88 **********************************************************************/
89 static void getPtsOffset( hb_work_object_t * w );
90 static int checkPtsOffset( hb_work_object_t * w );
91 static void InitAudio( hb_job_t * job, hb_sync_common_t * common, int i );
92 static void InsertSilence( hb_work_object_t * w, int64_t d );
93 static void UpdateState( hb_work_object_t * w );
94 static void UpdateSearchState( hb_work_object_t * w, int64_t start );
95 static hb_buffer_t * OutputAudioFrame( hb_audio_t *audio, hb_buffer_t *buf,
96 hb_sync_audio_t *sync );
98 /***********************************************************************
100 ***********************************************************************
101 * Initialize the work object
102 **********************************************************************/
103 hb_work_object_t * hb_sync_init( hb_job_t * job )
105 hb_title_t * title = job->title;
106 hb_chapter_t * chapter;
109 hb_work_private_t * pv;
110 hb_sync_video_t * sync;
111 hb_work_object_t * w;
112 hb_work_object_t * ret = NULL;
114 pv = calloc( 1, sizeof( hb_work_private_t ) );
115 sync = &pv->type.video;
116 pv->common = calloc( 1, sizeof( hb_sync_common_t ) );
118 pv->common->mutex = hb_lock_init();
119 pv->common->audio_pts_thresh = 0;
120 pv->common->next_frame = hb_cond_init();
121 pv->common->pts_count = 1;
122 if ( job->frame_to_start || job->pts_to_start )
124 pv->common->start_found = 0;
128 pv->common->start_found = 1;
131 ret = w = hb_get_work( WORK_SYNC_VIDEO );
132 w->private_data = pv;
133 w->fifo_in = job->fifo_raw;
134 w->fifo_out = job->fifo_sync;
137 pv->common->pts_offset = INT64_MIN;
138 sync->first_frame = 1;
142 /* We already have an accurate frame count from pass 1 */
143 hb_interjob_t * interjob = hb_interjob_get( job->h );
144 sync->count_frames_max = interjob->frame_count;
148 /* Calculate how many video frames we are expecting */
149 if ( job->pts_to_stop )
151 duration = job->pts_to_stop + 90000;
153 else if( job->frame_to_stop )
155 /* Set the duration to a rough estimate */
156 duration = ( job->frame_to_stop / ( title->rate / title->rate_base ) ) * 90000;
161 for( i = job->chapter_start; i <= job->chapter_end; i++ )
163 chapter = hb_list_item( title->list_chapter, i - 1 );
164 duration += chapter->duration;
167 /* 1 second safety so we're sure we won't miss anything */
169 sync->count_frames_max = duration * title->rate / title->rate_base / 90000;
172 hb_log( "sync: expecting %d video frames", sync->count_frames_max );
174 /* Initialize libsamplerate for every audio track we have */
175 if ( ! job->indepth_scan )
177 for( i = 0; i < hb_list_count( title->list_audio ) && i < 8; i++ )
179 InitAudio( job, pv->common, i );
182 pv->common->first_pts = malloc( sizeof(int64_t) * pv->common->pts_count );
183 for ( i = 0; i < pv->common->pts_count; i++ )
184 pv->common->first_pts[i] = INT64_MAX;
189 /***********************************************************************
191 ***********************************************************************
193 **********************************************************************/
194 void syncVideoClose( hb_work_object_t * w )
196 hb_work_private_t * pv = w->private_data;
197 hb_job_t * job = pv->job;
198 hb_sync_video_t * sync = &pv->type.video;
200 // Wake up audio sync if it's still waiting on condition.
201 pv->common->pts_offset = 0;
202 pv->common->start_found = 1;
203 hb_cond_broadcast( pv->common->next_frame );
207 hb_buffer_close( &sync->cur );
210 hb_log( "sync: got %d frames, %d expected",
211 pv->common->count_frames, sync->count_frames_max );
213 /* save data for second pass */
216 /* Preserve frame count for better accuracy in pass 2 */
217 hb_interjob_t * interjob = hb_interjob_get( job->h );
218 interjob->frame_count = pv->common->count_frames;
219 interjob->last_job = job->sequence_id;
220 interjob->total_time = sync->next_start;
223 if (sync->drops || sync->dups )
225 hb_log( "sync: %d frames dropped, %d duplicated",
226 sync->drops, sync->dups );
229 hb_lock( pv->common->mutex );
230 if ( --pv->common->ref == 0 )
232 hb_unlock( pv->common->mutex );
233 hb_lock_close( &pv->common->mutex );
238 hb_unlock( pv->common->mutex );
242 w->private_data = NULL;
245 /***********************************************************************
247 ***********************************************************************
249 **********************************************************************/
250 int syncVideoWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
251 hb_buffer_t ** buf_out )
253 hb_buffer_t * cur, * next, * sub = NULL;
254 hb_work_private_t * pv = w->private_data;
255 hb_job_t * job = pv->job;
256 hb_subtitle_t * subtitle;
257 hb_sync_video_t * sync = &pv->type.video;
264 /* Wait for start of point-to-point encoding */
265 if( !pv->common->start_found )
267 hb_sync_video_t * sync = &pv->type.video;
269 if( next->size == 0 )
272 pv->common->start_found = 1;
273 hb_cond_broadcast( pv->common->next_frame );
276 * Push through any subtitle EOFs in case they
277 * were not synced through.
279 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
281 subtitle = hb_list_item( job->list_subtitle, i );
282 if( subtitle->config.dest == PASSTHRUSUB )
284 if( subtitle->source == VOBSUB )
285 hb_fifo_push( subtitle->fifo_sync, hb_buffer_init( 0 ) );
287 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
292 if ( pv->common->count_frames < job->frame_to_start ||
293 next->start < job->pts_to_start )
295 // Flush any subtitles that have pts prior to the
297 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
299 subtitle = hb_list_item( job->list_subtitle, i );
300 while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
302 if ( sub->start > next->start )
304 sub = hb_fifo_get( subtitle->fifo_raw );
305 hb_buffer_close( &sub );
308 hb_lock( pv->common->mutex );
309 // Tell the audio threads what must be dropped
310 pv->common->audio_pts_thresh = next->start;
311 hb_cond_broadcast( pv->common->next_frame );
312 hb_unlock( pv->common->mutex );
314 UpdateSearchState( w, next->start );
315 hb_buffer_close( &next );
319 hb_lock( pv->common->mutex );
320 pv->common->start_found = 1;
321 pv->common->count_frames = 0;
322 hb_cond_broadcast( pv->common->next_frame );
323 hb_unlock( pv->common->mutex );
327 /* Wait till we can determine the initial pts of all streams */
328 if( pv->common->pts_offset == INT64_MIN )
330 pv->common->first_pts[0] = next->start;
331 hb_lock( pv->common->mutex );
332 while( pv->common->pts_offset == INT64_MIN )
334 // Full fifos will make us wait forever, so get the
335 // pts offset from the available streams if full
336 if ( hb_fifo_is_full( job->fifo_raw ) )
339 hb_cond_broadcast( pv->common->next_frame );
341 else if ( checkPtsOffset( w ) )
342 hb_cond_broadcast( pv->common->next_frame );
344 hb_cond_timedwait( pv->common->next_frame, pv->common->mutex, 200 );
346 hb_unlock( pv->common->mutex );
352 if( sync->cur->size == 0 )
354 /* we got an end-of-stream as our first video packet?
355 * Feed it downstream & signal that we're done.
357 *buf_out = hb_buffer_init( 0 );
359 pv->common->start_found = 1;
360 hb_cond_broadcast( pv->common->next_frame );
363 * Push through any subtitle EOFs in case they
364 * were not synced through.
366 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
368 subtitle = hb_list_item( job->list_subtitle, i );
369 if( subtitle->config.dest == PASSTHRUSUB )
371 if( subtitle->source == VOBSUB )
372 hb_fifo_push( subtitle->fifo_sync, hb_buffer_init( 0 ) );
374 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
382 /* At this point we have a frame to process. Let's check
383 1) if we will be able to push into the fifo ahead
384 2) if the next frame is there already, since we need it to
385 compute the duration of the current frame*/
386 if( next->size == 0 )
388 hb_buffer_close( &next );
390 cur->start = sync->next_start;
391 cur->stop = cur->start + 90000. / ((double)job->vrate / (double)job->vrate_base);
393 /* Make sure last frame is reflected in frame count */
394 pv->common->count_frames++;
396 /* Push the frame to the renderer */
397 hb_fifo_push( job->fifo_sync, cur );
400 /* we got an end-of-stream. Feed it downstream & signal that
401 * we're done. Note that this means we drop the final frame of
402 * video (we don't know its duration). On DVDs the final frame
403 * is often strange and dropping it seems to be a good idea. */
404 *buf_out = hb_buffer_init( 0 );
407 * Push through any subtitle EOFs in case they were not synced through.
409 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
411 subtitle = hb_list_item( job->list_subtitle, i );
412 if( subtitle->config.dest == PASSTHRUSUB )
414 if( subtitle->source == VOBSUB )
415 hb_fifo_push( subtitle->fifo_sync, hb_buffer_init( 0 ) );
417 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
420 pv->common->start_found = 1;
421 hb_cond_broadcast( pv->common->next_frame );
425 /* Check for end of point-to-point frame encoding */
426 if( job->frame_to_stop && pv->common->count_frames > job->frame_to_stop )
428 // Drop an empty buffer into our output to ensure that things
429 // get flushed all the way out.
430 hb_buffer_close( &sync->cur );
431 hb_buffer_close( &next );
432 *buf_out = hb_buffer_init( 0 );
433 hb_log( "sync: reached %d frames, exiting early",
434 pv->common->count_frames );
437 * Push through any subtitle EOFs in case they were not synced through.
439 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
441 subtitle = hb_list_item( job->list_subtitle, i );
442 if( subtitle->config.dest == PASSTHRUSUB )
444 if( subtitle->source == VOBSUB )
445 hb_fifo_push( subtitle->fifo_sync, hb_buffer_init( 0 ) );
447 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
453 /* Check for end of point-to-point pts encoding */
454 if( job->pts_to_stop && sync->next_start >= job->pts_to_stop )
456 // Drop an empty buffer into our output to ensure that things
457 // get flushed all the way out.
458 hb_log( "sync: reached pts %"PRId64", exiting early",
460 hb_buffer_close( &sync->cur );
461 hb_buffer_close( &next );
462 *buf_out = hb_buffer_init( 0 );
465 * Push through any subtitle EOFs in case they were not synced through.
467 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
469 subtitle = hb_list_item( job->list_subtitle, i );
470 if( subtitle->config.dest == PASSTHRUSUB )
472 if( subtitle->source == VOBSUB )
473 hb_fifo_push( subtitle->fifo_sync, hb_buffer_init( 0 ) );
475 hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
481 if( sync->first_frame )
483 /* This is our first frame */
484 if ( cur->start > pv->common->pts_offset )
487 * The first pts from a dvd should always be zero but
488 * can be non-zero with a transport or program stream since
489 * we're not guaranteed to start on an IDR frame. If we get
490 * a non-zero initial PTS extend its duration so it behaves
491 * as if it started at zero so that our audio timing will
494 hb_log( "sync: first pts is %"PRId64, cur->start );
495 cur->start = pv->common->pts_offset;
497 sync->first_frame = 0;
501 * since the first frame is always 0 and the upstream reader code
502 * is taking care of adjusting for pts discontinuities, we just have
503 * to deal with the next frame's start being in the past. This can
504 * happen when the PTS is adjusted after data loss but video frame
505 * reordering causes some frames with the old clock to appear after
506 * the clock change. This creates frames that overlap in time which
507 * looks to us like time going backward. The downstream muxing code
508 * can deal with overlaps of up to a frame time but anything larger
509 * we handle by dropping frames here.
511 hb_lock( pv->common->mutex );
512 if ( (int64_t)( next->start - pv->common->video_pts_slip - cur->start ) <= 0 )
514 if ( sync->first_drop == 0 )
516 sync->first_drop = next->start;
519 if (next->start - cur->start > 0)
521 sync->pts_skip += next->start - cur->start;
522 pv->common->video_pts_slip -= next->start - cur->start;
524 hb_unlock( pv->common->mutex );
525 if ( next->new_chap )
527 // don't drop a chapter mark when we drop the buffer
528 sync->chap_mark = next->new_chap;
530 hb_buffer_close( &next );
533 hb_unlock( pv->common->mutex );
534 if ( sync->first_drop )
536 hb_log( "sync: video time didn't advance - dropped %d frames "
537 "(delta %d ms, current %"PRId64", next %"PRId64", dur %d)",
538 sync->drop_count, (int)( cur->start - sync->first_drop ) / 90,
539 cur->start, next->start, (int)( next->start - cur->start ) );
540 sync->first_drop = 0;
541 sync->drop_count = 0;
545 * Track the video sequence number localy so that we can sync the audio
546 * to it using the sequence number as well as the PTS.
548 sync->video_sequence = cur->sequence;
551 * Look for a subtitle for this frame.
553 * If found then it will be tagged onto a video buffer of the correct time and
554 * sent in to the render pipeline. This only needs to be done for VOBSUBs which
555 * get rendered, other types of subtitles can just sit in their raw_queue until
556 * delt with at muxing.
558 for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
560 subtitle = hb_list_item( job->list_subtitle, i );
563 * Rewrite timestamps on subtitles that need it (on raw queue).
565 if( subtitle->source == CC608SUB ||
566 subtitle->source == CC708SUB ||
567 subtitle->source == SRTSUB ||
568 subtitle->source == UTF8SUB ||
569 subtitle->source == TX3GSUB)
572 * Rewrite timestamps on subtitles that came from Closed Captions
573 * since they are using the MPEG2 timestamps.
575 while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
578 * Rewrite the timestamps as and when the video
579 * (cur->start) reaches the same timestamp as a
580 * closed caption (sub->start).
582 * What about discontinuity boundaries - not delt
585 * Bypass the sync fifo altogether.
589 sub = hb_fifo_get( subtitle->fifo_raw );
590 hb_fifo_push( subtitle->fifo_out, sub );
595 * Sync the subtitles to the incoming video, and use
596 * the matching converted video timestamp.
598 * Note that it doesn't appear that we need to convert
599 * timestamps, I guess that they were already correct,
600 * so just push them through for rendering.
603 if( sub->start < cur->start )
605 sub = hb_fifo_get( subtitle->fifo_raw );
606 hb_fifo_push( subtitle->fifo_out, sub );
615 if( subtitle->source == VOBSUB )
618 while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
623 * EOF, pass it through immediately.
628 /* If two subtitles overlap, make the first one stop
629 when the second one starts */
630 sub2 = hb_fifo_see2( subtitle->fifo_raw );
631 if( sub2 && sub->stop > sub2->start )
633 sub->stop = sub2->start;
636 // hb_log("0x%x: video seq: %"PRId64" subtitle sequence: %"PRId64,
637 // sub, cur->sequence, sub->sequence);
639 if( sub->sequence > cur->sequence )
642 * The video is behind where we are, so wait until
643 * it catches up to the same reader point on the
644 * DVD. Then our PTS should be in the same region
651 if( sub->stop > cur->start ) {
653 * The stop time is in the future, so fall through
654 * and we'll deal with it in the next block of
659 * There is a valid subtitle, is it time to display it?
661 if( sub->stop > sub->start)
664 * Normal subtitle which ends after it starts,
665 * check to see that the current video is between
668 if( cur->start > sub->start &&
669 cur->start < sub->stop )
672 * We should be playing this, so leave the
675 * fall through to display
677 if( ( sub->stop - sub->start ) < ( 2 * 90000 ) )
680 * Subtitle is on for less than three
681 * seconds, extend the time that it is
682 * displayed to make it easier to read.
683 * Make it 3 seconds or until the next
684 * subtitle is displayed.
686 * This is in response to Indochine which
687 * only displays subs for 1 second -
690 sub->stop = sub->start + ( 2 * 90000 );
692 sub2 = hb_fifo_see2( subtitle->fifo_raw );
694 if( sub2 && sub->stop > sub2->start )
696 sub->stop = sub2->start;
703 * Defer until the play point is within
712 * The end of the subtitle is less than the start,
713 * this is a sign of a PTS discontinuity.
715 if( sub->start > cur->start )
718 * we haven't reached the start time yet, or
719 * we have jumped backwards after having
720 * already started this subtitle.
722 if( cur->start < sub->stop )
725 * We have jumped backwards and so should
726 * continue displaying this subtitle.
728 * fall through to display.
734 * Defer until the play point is
735 * within the subtitle
741 * Play this subtitle as the start is
742 * greater than our video point.
744 * fall through to display/
754 * The subtitle is older than this picture, trash it
756 sub = hb_fifo_get( subtitle->fifo_raw );
757 hb_buffer_close( &sub );
761 /* If we have a subtitle for this picture, copy it */
762 /* FIXME: we should avoid this memcpy */
767 if( subtitle->config.dest == RENDERSUB )
769 if ( cur->sub == NULL )
772 * Tack onto the video buffer for rendering
774 cur->sub = hb_buffer_init( sub->size );
775 cur->sub->x = sub->x;
776 cur->sub->y = sub->y;
777 cur->sub->width = sub->width;
778 cur->sub->height = sub->height;
779 memcpy( cur->sub->data, sub->data, sub->size );
783 * Pass-Through, pop it off of the raw queue,
785 sub = hb_fifo_get( subtitle->fifo_raw );
786 hb_fifo_push( subtitle->fifo_sync, sub );
790 * EOF - consume for rendered, else pass through
792 if( subtitle->config.dest == RENDERSUB )
794 sub = hb_fifo_get( subtitle->fifo_raw );
795 hb_buffer_close( &sub );
797 sub = hb_fifo_get( subtitle->fifo_raw );
798 hb_fifo_push( subtitle->fifo_sync, sub );
806 * Adjust the pts of the current frame so that it's contiguous
807 * with the previous frame. The start time of the current frame
808 * has to be the end time of the previous frame and the stop
809 * time has to be the start of the next frame. We don't
810 * make any adjustments to the source timestamps other than removing
811 * the clock offsets (which also removes pts discontinuities).
812 * This means we automatically encode at the source's frame rate.
813 * MP2 uses an implicit duration (frames end when the next frame
814 * starts) but more advanced containers like MP4 use an explicit
815 * duration. Since we're looking ahead one frame we set the
816 * explicit stop time from the start time of the next frame.
819 sync->cur = cur = next;
821 sync->next_pts = cur->start;
822 int64_t duration = cur->start - sync->pts_skip - (*buf_out)->start;
826 hb_log( "sync: invalid video duration %"PRId64", start %"PRId64", next %"PRId64"",
827 duration, (*buf_out)->start, next->start );
830 (*buf_out)->start = sync->next_start;
831 sync->next_start += duration;
832 (*buf_out)->stop = sync->next_start;
834 if ( sync->chap_mark )
836 // we have a pending chapter mark from a recent drop - put it on this
837 // buffer (this may make it one frame late but we can't do any better).
838 (*buf_out)->new_chap = sync->chap_mark;
848 // sync*Init does nothing because sync has a special initializer
849 // that takes care of initializing video and all audio tracks
850 int syncVideoInit( hb_work_object_t * w, hb_job_t * job)
855 hb_work_object_t hb_sync_video =
858 "Video Synchronization",
864 /***********************************************************************
866 ***********************************************************************
868 **********************************************************************/
869 void syncAudioClose( hb_work_object_t * w )
871 hb_work_private_t * pv = w->private_data;
872 hb_sync_audio_t * sync = &pv->type.audio;
874 if( w->audio->config.out.codec == HB_ACODEC_AC3 )
876 free( sync->ac3_buf );
880 src_delete( sync->state );
883 hb_lock( pv->common->mutex );
884 if ( --pv->common->ref == 0 )
886 hb_unlock( pv->common->mutex );
887 hb_lock_close( &pv->common->mutex );
892 hb_unlock( pv->common->mutex );
896 w->private_data = NULL;
899 int syncAudioInit( hb_work_object_t * w, hb_job_t * job)
904 /***********************************************************************
906 ***********************************************************************
908 **********************************************************************/
909 static int syncAudioWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
910 hb_buffer_t ** buf_out )
912 hb_work_private_t * pv = w->private_data;
913 hb_job_t * job = pv->job;
914 hb_sync_audio_t * sync = &pv->type.audio;
921 /* if the next buffer is an eof send it downstream */
922 if ( buf->size <= 0 )
924 hb_buffer_close( &buf );
925 *buf_out = hb_buffer_init( 0 );
929 /* Wait for start frame if doing point-to-point */
930 hb_lock( pv->common->mutex );
931 while ( !pv->common->start_found )
933 if ( buf->start < pv->common->audio_pts_thresh )
935 hb_buffer_close( &buf );
936 hb_unlock( pv->common->mutex );
939 while ( !pv->common->start_found &&
940 buf->start >= pv->common->audio_pts_thresh )
942 hb_cond_timedwait( pv->common->next_frame, pv->common->mutex, 200 );
945 if ( buf->start < pv->common->audio_pts_thresh )
947 hb_buffer_close( &buf );
948 hb_unlock( pv->common->mutex );
951 hb_unlock( pv->common->mutex );
953 /* Wait till we can determine the initial pts of all streams */
954 if( pv->common->pts_offset == INT64_MIN )
956 pv->common->first_pts[sync->index+1] = buf->start;
957 hb_lock( pv->common->mutex );
958 while( pv->common->pts_offset == INT64_MIN )
960 // Full fifos will make us wait forever, so get the
961 // pts offset from the available streams if full
962 if (hb_fifo_is_full(w->fifo_in))
965 hb_cond_broadcast( pv->common->next_frame );
967 else if ( checkPtsOffset( w ) )
968 hb_cond_broadcast( pv->common->next_frame );
970 hb_cond_timedwait( pv->common->next_frame, pv->common->mutex, 200 );
972 hb_unlock( pv->common->mutex );
975 if( job->frame_to_stop && pv->common->count_frames >= job->frame_to_stop )
977 hb_buffer_close( &buf );
978 *buf_out = hb_buffer_init( 0 );
982 if( job->pts_to_stop && sync->next_start >= job->pts_to_stop )
984 hb_buffer_close( &buf );
985 *buf_out = hb_buffer_init( 0 );
989 hb_lock( pv->common->mutex );
990 start = buf->start - pv->common->audio_passthru_slip;
991 hb_unlock( pv->common->mutex );
992 if ( (int64_t)( start - sync->next_pts ) < 0 )
994 // audio time went backwards.
995 // If our output clock is more than a half frame ahead of the
996 // input clock drop this frame to move closer to sync.
997 // Otherwise drop frames until the input clock matches the output clock.
998 if ( sync->first_drop || sync->next_start - start > 90*15 )
1000 // Discard data that's in the past.
1001 if ( sync->first_drop == 0 )
1003 sync->first_drop = sync->next_pts;
1006 hb_buffer_close( &buf );
1009 sync->next_pts = start;
1011 if ( sync->first_drop )
1013 // we were dropping old data but input buf time is now current
1014 hb_log( "sync: audio %d time went backwards %d ms, dropped %d frames "
1015 "(next %"PRId64", current %"PRId64")", w->audio->id,
1016 (int)( sync->next_pts - sync->first_drop ) / 90,
1017 sync->drop_count, sync->first_drop, sync->next_pts );
1018 sync->first_drop = 0;
1019 sync->drop_count = 0;
1020 sync->next_pts = start;
1022 if ( start - sync->next_pts >= (90 * 70) )
1024 if ( start - sync->next_pts > (90000LL * 60) )
1026 // there's a gap of more than a minute between the last
1027 // frame and this. assume we got a corrupted timestamp
1028 // and just drop the next buf.
1029 hb_log( "sync: %d minute time gap in audio %d - dropping buf"
1030 " start %"PRId64", next %"PRId64,
1031 (int)((start - sync->next_pts) / (90000*60)),
1032 w->audio->id, start, sync->next_pts );
1033 hb_buffer_close( &buf );
1037 * there's a gap of at least 70ms between the last
1038 * frame we processed & the next. Fill it with silence.
1039 * Or in the case of DCA, skip some frames from the
1042 if( w->audio->config.out.codec == HB_ACODEC_DCA )
1044 hb_log( "sync: audio gap %d ms. Skipping frames. Audio %d"
1045 " start %"PRId64", next %"PRId64,
1046 (int)((start - sync->next_pts) / 90),
1047 w->audio->id, start, sync->next_pts );
1048 hb_lock( pv->common->mutex );
1049 pv->common->audio_passthru_slip += (start - sync->next_pts);
1050 pv->common->video_pts_slip += (start - sync->next_pts);
1051 hb_unlock( pv->common->mutex );
1055 hb_log( "sync: adding %d ms of silence to audio %d"
1056 " start %"PRId64", next %"PRId64,
1057 (int)((start - sync->next_pts) / 90),
1058 w->audio->id, start, sync->next_pts );
1059 InsertSilence( w, start - sync->next_pts );
1063 * When we get here we've taken care of all the dups and gaps in the
1064 * audio stream and are ready to inject the next input frame into
1065 * the output stream.
1067 *buf_out = OutputAudioFrame( w->audio, buf, sync );
1071 hb_work_object_t hb_sync_audio =
1074 "AudioSynchronization",
1080 static void InitAudio( hb_job_t * job, hb_sync_common_t * common, int i )
1082 hb_work_object_t * w;
1083 hb_work_private_t * pv;
1084 hb_title_t * title = job->title;
1085 hb_sync_audio_t * sync;
1087 pv = calloc( 1, sizeof( hb_work_private_t ) );
1088 sync = &pv->type.audio;
1091 pv->common = common;
1093 pv->common->pts_count++;
1095 w = hb_get_work( WORK_SYNC_AUDIO );
1096 w->private_data = pv;
1097 w->audio = hb_list_item( title->list_audio, i );
1098 w->fifo_in = w->audio->priv.fifo_raw;
1100 if( w->audio->config.out.codec == HB_ACODEC_AC3 ||
1101 w->audio->config.out.codec == HB_ACODEC_DCA )
1103 w->fifo_out = w->audio->priv.fifo_out;
1107 w->fifo_out = w->audio->priv.fifo_sync;
1110 if( w->audio->config.out.codec == HB_ACODEC_AC3 )
1112 /* Have a silent AC-3 frame ready in case we have to fill a
1118 codec = avcodec_find_encoder( CODEC_ID_AC3 );
1119 c = avcodec_alloc_context();
1121 c->bit_rate = w->audio->config.in.bitrate;
1122 c->sample_rate = w->audio->config.in.samplerate;
1123 c->channels = HB_INPUT_CH_LAYOUT_GET_DISCRETE_COUNT( w->audio->config.in.channel_layout );
1125 if( hb_avcodec_open( c, codec ) < 0 )
1127 hb_log( "sync: avcodec_open failed" );
1131 zeros = calloc( AC3_SAMPLES_PER_FRAME *
1132 sizeof( short ) * c->channels, 1 );
1133 sync->ac3_size = w->audio->config.in.bitrate * AC3_SAMPLES_PER_FRAME /
1134 w->audio->config.in.samplerate / 8;
1135 sync->ac3_buf = malloc( sync->ac3_size );
1137 if( avcodec_encode_audio( c, sync->ac3_buf, sync->ac3_size,
1138 zeros ) != sync->ac3_size )
1140 hb_log( "sync: avcodec_encode_audio failed" );
1144 hb_avcodec_close( c );
1149 /* Initialize libsamplerate */
1151 sync->state = src_new( SRC_SINC_MEDIUM_QUALITY,
1152 HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(
1153 w->audio->config.out.mixdown), &error );
1154 sync->data.end_of_input = 0;
1156 hb_list_add( job->list_work, w );
1159 static hb_buffer_t * OutputAudioFrame( hb_audio_t *audio, hb_buffer_t *buf,
1160 hb_sync_audio_t *sync )
1162 int64_t start = sync->next_start;
1163 int64_t duration = buf->stop - buf->start;
1165 sync->next_pts += duration;
1167 if( audio->config.in.samplerate == audio->config.out.samplerate ||
1168 audio->config.out.codec == HB_ACODEC_AC3 ||
1169 audio->config.out.codec == HB_ACODEC_DCA )
1172 * If we don't have to do sample rate conversion or this audio is
1173 * pass-thru just send the input buffer downstream after adjusting
1174 * its timestamps to make the output stream continuous.
1179 /* Not pass-thru - do sample rate conversion */
1180 int count_in, count_out;
1181 hb_buffer_t * buf_raw = buf;
1182 int channel_count = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown) *
1185 count_in = buf_raw->size / channel_count;
1187 * When using stupid rates like 44.1 there will always be some
1188 * truncation error. E.g., a 1536 sample AC3 frame will turn into a
1189 * 1536*44.1/48.0 = 1411.2 sample frame. If we just truncate the .2
1190 * the error will build up over time and eventually the audio will
1191 * substantially lag the video. libsamplerate will keep track of the
1192 * fractional sample & give it to us when appropriate if we give it
1193 * an extra sample of space in the output buffer.
1195 count_out = ( duration * audio->config.out.samplerate ) / 90000 + 1;
1197 sync->data.input_frames = count_in;
1198 sync->data.output_frames = count_out;
1199 sync->data.src_ratio = (double)audio->config.out.samplerate /
1200 (double)audio->config.in.samplerate;
1202 buf = hb_buffer_init( count_out * channel_count );
1203 sync->data.data_in = (float *) buf_raw->data;
1204 sync->data.data_out = (float *) buf->data;
1205 if( src_process( sync->state, &sync->data ) )
1207 /* XXX If this happens, we're screwed */
1208 hb_log( "sync: audio %d src_process failed", audio->id );
1210 hb_buffer_close( &buf_raw );
1212 buf->size = sync->data.output_frames_gen * channel_count;
1213 duration = ( sync->data.output_frames_gen * 90000 ) /
1214 audio->config.out.samplerate;
1216 buf->frametype = HB_FRAME_AUDIO;
1218 buf->stop = start + duration;
1219 sync->next_start = start + duration;
1223 static void InsertSilence( hb_work_object_t * w, int64_t duration )
1225 hb_work_private_t * pv = w->private_data;
1226 hb_sync_audio_t *sync = &pv->type.audio;
1230 // to keep pass-thru and regular audio in sync we generate silence in
1231 // AC3 frame-sized units. If the silence duration isn't an integer multiple
1232 // of the AC3 frame duration we will truncate or round up depending on
1233 // which minimizes the timing error.
1234 const int frame_dur = ( 90000 * AC3_SAMPLES_PER_FRAME ) /
1235 w->audio->config.in.samplerate;
1236 int frame_count = ( duration + (frame_dur >> 1) ) / frame_dur;
1238 while ( --frame_count >= 0 )
1240 if( w->audio->config.out.codec == HB_ACODEC_AC3 )
1242 buf = hb_buffer_init( sync->ac3_size );
1243 buf->start = sync->next_pts;
1244 buf->stop = buf->start + frame_dur;
1245 memcpy( buf->data, sync->ac3_buf, buf->size );
1246 fifo = w->audio->priv.fifo_out;
1250 buf = hb_buffer_init( AC3_SAMPLES_PER_FRAME * sizeof( float ) *
1251 HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(
1252 w->audio->config.out.mixdown) );
1253 buf->start = sync->next_pts;
1254 buf->stop = buf->start + frame_dur;
1255 memset( buf->data, 0, buf->size );
1256 fifo = w->audio->priv.fifo_sync;
1258 buf = OutputAudioFrame( w->audio, buf, sync );
1259 hb_fifo_push( fifo, buf );
1263 static void UpdateState( hb_work_object_t * w )
1265 hb_work_private_t * pv = w->private_data;
1266 hb_sync_video_t * sync = &pv->type.video;
1269 if( !pv->common->count_frames )
1271 sync->st_first = hb_get_date();
1272 pv->job->st_pause_date = -1;
1273 pv->job->st_paused = 0;
1275 pv->common->count_frames++;
1277 if( hb_get_date() > sync->st_dates[3] + 1000 )
1279 memmove( &sync->st_dates[0], &sync->st_dates[1],
1280 3 * sizeof( uint64_t ) );
1281 memmove( &sync->st_counts[0], &sync->st_counts[1],
1282 3 * sizeof( uint64_t ) );
1283 sync->st_dates[3] = hb_get_date();
1284 sync->st_counts[3] = pv->common->count_frames;
1287 #define p state.param.working
1288 state.state = HB_STATE_WORKING;
1289 p.progress = (float) pv->common->count_frames / (float) sync->count_frames_max;
1290 if( p.progress > 1.0 )
1294 p.rate_cur = 1000.0 *
1295 (float) ( sync->st_counts[3] - sync->st_counts[0] ) /
1296 (float) ( sync->st_dates[3] - sync->st_dates[0] );
1297 if( hb_get_date() > sync->st_first + 4000 )
1300 p.rate_avg = 1000.0 * (float) sync->st_counts[3] /
1301 (float) ( sync->st_dates[3] - sync->st_first - pv->job->st_paused);
1302 eta = (float) ( sync->count_frames_max - sync->st_counts[3] ) /
1304 p.hours = eta / 3600;
1305 p.minutes = ( eta % 3600 ) / 60;
1306 p.seconds = eta % 60;
1317 hb_set_state( pv->job->h, &state );
1320 static void UpdateSearchState( hb_work_object_t * w, int64_t start )
1322 hb_work_private_t * pv = w->private_data;
1323 hb_sync_video_t * sync = &pv->type.video;
1328 now = hb_get_date();
1329 if( !pv->common->count_frames )
1331 sync->st_first = now;
1332 pv->job->st_pause_date = -1;
1333 pv->job->st_paused = 0;
1335 pv->common->count_frames++;
1337 #define p state.param.working
1338 state.state = HB_STATE_SEARCHING;
1339 if ( pv->job->frame_to_start )
1340 p.progress = (float) pv->common->count_frames /
1341 (float) pv->job->frame_to_start;
1342 else if ( pv->job->pts_to_start )
1343 p.progress = (float) start / (float) pv->job->pts_to_start;
1346 if( p.progress > 1.0 )
1350 if (now > sync->st_first)
1354 if ( pv->job->frame_to_start )
1356 avg = 1000.0 * (double)pv->common->count_frames / (now - sync->st_first);
1357 eta = ( pv->job->frame_to_start - pv->common->count_frames ) / avg;
1359 else if ( pv->job->pts_to_start )
1361 avg = 1000.0 * (double)start / (now - sync->st_first);
1362 eta = ( pv->job->pts_to_start - start ) / avg;
1364 p.hours = eta / 3600;
1365 p.minutes = ( eta % 3600 ) / 60;
1366 p.seconds = eta % 60;
1377 hb_set_state( pv->job->h, &state );
1380 static void getPtsOffset( hb_work_object_t * w )
1382 hb_work_private_t * pv = w->private_data;
1384 int64_t first_pts = INT64_MAX;
1386 for( i = 0; i < pv->common->pts_count; i++ )
1388 if ( pv->common->first_pts[i] < first_pts )
1389 first_pts = pv->common->first_pts[i];
1391 pv->common->audio_passthru_slip = pv->common->pts_offset = first_pts;
1395 static int checkPtsOffset( hb_work_object_t * w )
1397 hb_work_private_t * pv = w->private_data;
1400 for( i = 0; i < pv->common->pts_count; i++ )
1402 if ( pv->common->first_pts[i] == INT64_MAX )