1 /* $Id: decavcodec.c,v 1.6 2005/03/06 04:08:54 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. */
7 /* This module is Handbrake's interface to the ffmpeg decoder library
8 (libavcodec & small parts of libavformat). It contains four Handbrake
11 decavcodec connects HB to an ffmpeg audio decoder
12 decavcodecv connects HB to an ffmpeg video decoder
14 (Two different routines are needed because the ffmpeg library
15 has different decoder calling conventions for audio & video.
16 The audio decoder should have had its name changed to "decavcodeca"
17 but I got lazy.) These work objects are self-contained & follow all
18 of HB's conventions for a decoder module. They can be used like
19 any other HB decoder (deca52, decmpeg2, etc.).
21 decavcodecai "internal" (incestuous?) version of decavcodec
22 decavcodecvi "internal" (incestuous?) version of decavcodecv
24 These routine are functionally equivalent to the routines above but
25 can only be used by the ffmpeg-based stream reader in libhb/stream.c.
26 The reason they exist is because the ffmpeg library leaves some of
27 the information needed by the decoder in the AVStream (the data
28 structure used by the stream reader) and we need to retrieve it
29 to successfully decode frames. But in HB the reader and decoder
30 modules are in completely separate threads and nothing goes between
31 them but hb_buffers containing frames to be decoded. I.e., there's
32 no easy way for the ffmpeg stream reader to pass a pointer to its
33 AVStream over to the ffmpeg video or audio decoder. So the *i work
34 objects use a private back door to the stream reader to get access
35 to the AVStream (routines hb_ffmpeg_avstream and hb_ffmpeg_context)
36 and the codec_param passed to these work objects is the key to this
37 back door (it's basically an index that allows the correct AVStream
40 The normal & *i objects share a lot of code (the basic frame decoding
41 and bitstream info code is factored out into subroutines that can be
42 called by either) but the top level routines of the *i objects
43 (decavcodecviWork, decavcodecviInfo, etc.) are different because:
44 1) they *have* to use the AVCodecContext that's contained in the
45 reader's AVStream rather than just allocating & using their own,
46 2) the Info routines have access to stuff kept in the AVStream in addition
47 to stuff kept in the AVCodecContext. This shouldn't be necessary but
48 crucial information like video frame rate that should be in the
49 AVCodecContext is either missing or wrong in the version of ffmpeg
50 we're currently using.
52 A consequence of the above is that the non-i work objects *can't* use
53 information from the AVStream because there isn't one - they get their
54 data from either the dvd reader or the mpeg reader, not the ffmpeg stream
55 reader. That means that they have to make up for deficiencies in the
56 AVCodecContext info by using stuff kept in the HB "title" struct. It
57 also means that ffmpeg codecs that randomly scatter state needed by
58 the decoder across both the AVCodecContext & the AVStream (e.g., the
59 VC1 decoder) can't easily be used by the HB mpeg stream reader.
64 #include "libavcodec/audioconvert.h"
66 static int decavcodecInit( hb_work_object_t *, hb_job_t * );
67 static int decavcodecWork( hb_work_object_t *, hb_buffer_t **, hb_buffer_t ** );
68 static void decavcodecClose( hb_work_object_t * );
69 static int decavcodecInfo( hb_work_object_t *, hb_work_info_t * );
70 static int decavcodecBSInfo( hb_work_object_t *, const hb_buffer_t *, hb_work_info_t * );
72 hb_work_object_t hb_decavcodec =
75 "MPGA decoder (libavcodec)",
85 // there are nheap items on the heap indexed 1..nheap (i.e., top of
86 // heap is 1). The 0th slot is unused - a marker is put there to check
87 // for overwrite errs.
88 int64_t h[HEAP_SIZE+1];
92 struct hb_work_private_s
95 AVCodecContext *context;
96 AVCodecParserContext *parser;
98 double duration; // frame duration (for video)
99 double pts_next; // next pts we expect to generate
100 int64_t pts; // (video) pts passing from parser to decoder
101 int64_t chap_time; // time of next chap mark (if new_chap != 0)
102 int new_chap; // output chapter mark pending
105 uint32_t decode_errors;
106 int brokenByMicrosoft; // video stream may contain packed b-frames
107 hb_buffer_t* delayq[HEAP_SIZE];
110 struct SwsContext *sws_context; // if we have to rescale or convert color space
113 static int64_t heap_pop( pts_heap_t *heap )
117 if ( heap->nheap <= 0 )
122 // return the top of the heap then put the bottom element on top,
123 // decrease the heap size by one & rebalence the heap.
126 int64_t v = heap->h[heap->nheap--];
128 int child = parent << 1;
129 while ( child <= heap->nheap )
131 // find the smallest of the two children of parent
132 if (child < heap->nheap && heap->h[child] > heap->h[child+1] )
135 if (v <= heap->h[child])
136 // new item is smaller than either child so it's the new parent.
139 // smallest child is smaller than new item so move it up then
140 // check its children.
141 int64_t hp = heap->h[child];
142 heap->h[parent] = hp;
150 static void heap_push( pts_heap_t *heap, int64_t v )
152 if ( heap->nheap < HEAP_SIZE )
157 // stick the new value on the bottom of the heap then bubble it
158 // up to its correct spot.
159 int child = heap->nheap;
161 int parent = child >> 1;
162 if (heap->h[parent] <= v)
165 int64_t hp = heap->h[parent];
173 /***********************************************************************
174 * hb_work_decavcodec_init
175 ***********************************************************************
177 **********************************************************************/
178 static int decavcodecInit( hb_work_object_t * w, hb_job_t * job )
182 hb_work_private_t * pv = calloc( 1, sizeof( hb_work_private_t ) );
183 w->private_data = pv;
187 int codec_id = w->codec_param;
190 codec_id = CODEC_ID_MP2;
192 codec = avcodec_find_decoder( codec_id );
193 pv->parser = av_parser_init( codec_id );
195 pv->context = avcodec_alloc_context();
196 hb_avcodec_open( pv->context, codec );
201 /***********************************************************************
203 ***********************************************************************
205 **********************************************************************/
206 static void decavcodecClose( hb_work_object_t * w )
208 hb_work_private_t * pv = w->private_data;
212 if ( pv->job && pv->context && pv->context->codec )
214 hb_log( "%s-decoder done: %u frames, %u decoder errors, %u drops",
215 pv->context->codec->name, pv->nframes, pv->decode_errors,
218 if ( pv->sws_context )
220 sws_freeContext( pv->sws_context );
224 av_parser_close(pv->parser);
226 if ( pv->context && pv->context->codec )
228 hb_avcodec_close( pv->context );
232 hb_list_close( &pv->list );
240 w->private_data = NULL;
244 /***********************************************************************
246 ***********************************************************************
248 **********************************************************************/
249 static int decavcodecWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
250 hb_buffer_t ** buf_out )
252 hb_work_private_t * pv = w->private_data;
253 hb_buffer_t * in = *buf_in, * buf, * last = NULL;
254 int pos, len, out_size, i, uncompressed_len;
257 unsigned char *parser_output_buffer;
258 int parser_output_buffer_len;
260 if ( (*buf_in)->size <= 0 )
262 /* EOF on input stream - send it downstream & say that we're done */
270 if ( in->start < -1 && pv->pts_next <= 0 )
272 // discard buffers that start before video time 0
276 cur = ( in->start < 0 )? pv->pts_next : in->start;
278 bufaligned = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE );
280 while( pos < in->size )
282 len = av_parser_parse( pv->parser, pv->context,
283 &parser_output_buffer, &parser_output_buffer_len,
284 in->data + pos, in->size - pos, cur, cur );
286 uncompressed_len = 0;
287 if (parser_output_buffer_len)
289 out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE;
290 uncompressed_len = avcodec_decode_audio2( pv->context, bufaligned,
292 parser_output_buffer,
293 parser_output_buffer_len );
300 buf = hb_buffer_init( 2 * out_size );
302 int sample_size_in_bytes = 2; // Default to 2 bytes
303 switch (pv->context->sample_fmt)
306 sample_size_in_bytes = 2;
308 /* We should handle other formats here - but that needs additional format conversion work below */
309 /* For now we'll just report the error and try to carry on */
311 hb_log("decavcodecWork - Unknown Sample Format from avcodec_decode_audio (%d) !", pv->context->sample_fmt);
316 buf->stop = cur + 90000 * ( out_size / (sample_size_in_bytes * pv->context->channels) ) /
317 pv->context->sample_rate;
321 fl32 = (float *) buf->data;
322 for( i = 0; i < out_size / 2; i++ )
329 last = last->next = buf;
333 *buf_out = last = buf;
346 static int decavcodecInfo( hb_work_object_t *w, hb_work_info_t *info )
348 hb_work_private_t *pv = w->private_data;
350 memset( info, 0, sizeof(*info) );
352 if ( pv && pv->context )
354 AVCodecContext *context = pv->context;
355 info->bitrate = context->bit_rate;
356 info->rate = context->time_base.num;
357 info->rate_base = context->time_base.den;
358 info->profile = context->profile;
359 info->level = context->level;
365 static const int chan2layout[] = {
366 HB_INPUT_CH_LAYOUT_MONO, // We should allow no audio really.
367 HB_INPUT_CH_LAYOUT_MONO,
368 HB_INPUT_CH_LAYOUT_STEREO,
369 HB_INPUT_CH_LAYOUT_2F1R,
370 HB_INPUT_CH_LAYOUT_2F2R,
371 HB_INPUT_CH_LAYOUT_3F2R,
372 HB_INPUT_CH_LAYOUT_4F2R,
373 HB_INPUT_CH_LAYOUT_STEREO,
374 HB_INPUT_CH_LAYOUT_STEREO,
377 static int decavcodecBSInfo( hb_work_object_t *w, const hb_buffer_t *buf,
378 hb_work_info_t *info )
380 hb_work_private_t *pv = w->private_data;
383 memset( info, 0, sizeof(*info) );
385 if ( pv && pv->context )
387 return decavcodecInfo( w, info );
390 // We should parse the bitstream to find its parameters but for right
391 // now we just return dummy values if there's a codec that will handle it.
392 AVCodec *codec = avcodec_find_decoder( w->codec_param? w->codec_param :
396 // there's no ffmpeg codec for this audio type - give up
400 static char codec_name[64];
401 info->name = strncpy( codec_name, codec->name, sizeof(codec_name)-1 );
403 AVCodecParserContext *parser = av_parser_init( codec->id );
404 AVCodecContext *context = avcodec_alloc_context();
405 hb_avcodec_open( context, codec );
406 uint8_t *buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE );
407 int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE;
408 unsigned char *pbuffer;
409 int pos = 0, pbuffer_size;
411 while ( pos < buf->size )
413 int len = av_parser_parse( parser, context, &pbuffer, &pbuffer_size,
414 buf->data + pos, buf->size - pos,
415 buf->start, buf->start );
417 if ( pbuffer_size > 0 )
419 len = avcodec_decode_audio2( context, (int16_t*)buffer, &out_size,
420 pbuffer, pbuffer_size );
421 if ( len > 0 && context->sample_rate > 0 )
423 info->bitrate = context->bit_rate;
424 info->rate = context->sample_rate;
426 info->channel_layout = chan2layout[context->channels & 7];
433 av_parser_close( parser );
434 hb_avcodec_close( context );
438 /* -------------------------------------------------------------
439 * General purpose video decoder using libavcodec
442 static uint8_t *copy_plane( uint8_t *dst, uint8_t* src, int dstride, int sstride,
445 if ( dstride == sstride )
447 memcpy( dst, src, dstride * h );
448 return dst + dstride * h;
450 int lbytes = dstride <= sstride? dstride : sstride;
453 memcpy( dst, src, lbytes );
460 // copy one video frame into an HB buf. If the frame isn't in our color space
461 // or at least one of its dimensions is odd, use sws_scale to convert/rescale it.
462 // Otherwise just copy the bits.
463 static hb_buffer_t *copy_frame( hb_work_private_t *pv, AVFrame *frame )
465 AVCodecContext *context = pv->context;
469 // if the dimensions are odd, drop the lsb since h264 requires that
470 // both width and height be even.
471 w = ( context->width >> 1 ) << 1;
472 h = ( context->height >> 1 ) << 1;
476 w = pv->job->title->width;
477 h = pv->job->title->height;
479 hb_buffer_t *buf = hb_video_buffer_init( w, h );
480 uint8_t *dst = buf->data;
482 if ( context->pix_fmt != PIX_FMT_YUV420P || w != context->width ||
483 h != context->height )
485 // have to convert to our internal color space and/or rescale
487 avpicture_fill( &dstpic, dst, PIX_FMT_YUV420P, w, h );
489 if ( ! pv->sws_context )
491 pv->sws_context = sws_getContext( context->width, context->height, context->pix_fmt,
492 w, h, PIX_FMT_YUV420P,
493 SWS_LANCZOS|SWS_ACCURATE_RND,
496 sws_scale( pv->sws_context, frame->data, frame->linesize, 0, h,
497 dstpic.data, dstpic.linesize );
501 dst = copy_plane( dst, frame->data[0], w, frame->linesize[0], h );
502 w = (w + 1) >> 1; h = (h + 1) >> 1;
503 dst = copy_plane( dst, frame->data[1], w, frame->linesize[1], h );
504 dst = copy_plane( dst, frame->data[2], w, frame->linesize[2], h );
509 static int get_frame_buf( AVCodecContext *context, AVFrame *frame )
511 hb_work_private_t *pv = context->opaque;
512 frame->pts = pv->pts;
514 return avcodec_default_get_buffer( context, frame );
517 static void log_chapter( hb_work_private_t *pv, int chap_num, int64_t pts )
519 hb_chapter_t *c = hb_list_item( pv->job->title->list_chapter, chap_num - 1 );
522 hb_log( "%s: \"%s\" (%d) at frame %u time %lld",
523 pv->context->codec->name, c->title, chap_num, pv->nframes, pts );
527 hb_log( "%s: Chapter %d at frame %u time %lld",
528 pv->context->codec->name, chap_num, pv->nframes, pts );
532 static void flushDelayQueue( hb_work_private_t *pv )
535 int slot = pv->nframes & (HEAP_SIZE-1);
537 // flush all the video packets left on our timestamp-reordering delay q
538 while ( ( buf = pv->delayq[slot] ) != NULL )
540 buf->start = heap_pop( &pv->pts_heap );
541 hb_list_add( pv->list, buf );
542 pv->delayq[slot] = NULL;
543 slot = ( slot + 1 ) & (HEAP_SIZE-1);
547 static int decodeFrame( hb_work_private_t *pv, uint8_t *data, int size )
549 int got_picture, oldlevel = 0;
552 if ( global_verbosity_level <= 1 )
554 oldlevel = av_log_get_level();
555 av_log_set_level( AV_LOG_QUIET );
557 if ( avcodec_decode_video( pv->context, &frame, &got_picture, data, size ) < 0 )
561 if ( global_verbosity_level <= 1 )
563 av_log_set_level( oldlevel );
567 // ffmpeg makes it hard to attach a pts to a frame. if the MPEG ES
568 // packet had a pts we handed it to av_parser_parse (if the packet had
569 // no pts we set it to -1 but before the parse we can't distinguish between
570 // the start of a video frame with no pts & an intermediate packet of
571 // some frame which never has a pts). we hope that when parse returns
572 // the frame to us the pts we originally handed it will be in parser->pts.
573 // we put this pts into pv->pts so that when a avcodec_decode_video
574 // finally gets around to allocating an AVFrame to hold the decoded
575 // frame we can stuff that pts into the frame. if all of these relays
576 // worked at this point frame.pts should hold the frame's pts from the
577 // original data stream or -1 if it didn't have one. in the latter case
578 // we generate the next pts in sequence for it.
579 double frame_dur = pv->duration;
580 if ( frame_dur <= 0 )
582 frame_dur = 90000. * (double)pv->context->time_base.num /
583 (double)pv->context->time_base.den;
584 pv->duration = frame_dur;
586 if ( frame.repeat_pict )
588 frame_dur += frame.repeat_pict * frame_dur * 0.5;
590 // XXX Unlike every other video decoder, the Raw decoder doesn't
591 // use the standard buffer allocation routines so we never
592 // get to put a PTS in the frame. Do it now.
593 if ( pv->context->codec_id == CODEC_ID_RAWVIDEO )
598 // If there was no pts for this frame, assume constant frame rate
599 // video & estimate the next frame time from the last & duration.
600 double pts = frame.pts;
605 pv->pts_next = pts + frame_dur;
609 // if we're doing a scan or this content couldn't have been broken
610 // by Microsoft we don't worry about timestamp reordering
611 if ( ! pv->job || ! pv->brokenByMicrosoft )
613 buf = copy_frame( pv, &frame );
615 hb_list_add( pv->list, buf );
620 // XXX This following probably addresses a libavcodec bug but I don't
621 // see an easy fix so we workaround it here.
623 // The M$ 'packed B-frames' atrocity results in decoded frames with
624 // the wrong timestamp. E.g., if there are 2 b-frames the timestamps
625 // we see here will be "2 3 1 5 6 4 ..." instead of "1 2 3 4 5 6".
626 // The frames are actually delivered in the right order but with
627 // the wrong timestamp. To get the correct timestamp attached to
628 // each frame we have a delay queue (longer than the max number of
629 // b-frames) & a sorting heap for the timestamps. As each frame
630 // comes out of the decoder the oldest frame in the queue is removed
631 // and associated with the smallest timestamp. Then the new frame is
632 // added to the queue & its timestamp is pushed on the heap.
633 // This does nothing if the timestamps are correct (i.e., the video
634 // uses a codec that Micro$oft hasn't broken yet) but the frames
635 // get timestamped correctly even when M$ has munged them.
637 // remove the oldest picture from the frame queue (if any) &
638 // give it the smallest timestamp from our heap. The queue size
639 // is a power of two so we get the slot of the oldest by masking
640 // the frame count & this will become the slot of the newest
641 // once we've removed & processed the oldest.
642 int slot = pv->nframes & (HEAP_SIZE-1);
643 if ( ( buf = pv->delayq[slot] ) != NULL )
645 buf->start = heap_pop( &pv->pts_heap );
647 if ( pv->new_chap && buf->start >= pv->chap_time )
649 buf->new_chap = pv->new_chap;
652 log_chapter( pv, buf->new_chap, buf->start );
654 else if ( pv->nframes == 0 )
656 log_chapter( pv, pv->job->chapter_start, buf->start );
658 hb_list_add( pv->list, buf );
661 // add the new frame to the delayq & push its timestamp on the heap
662 pv->delayq[slot] = copy_frame( pv, &frame );
663 heap_push( &pv->pts_heap, pts );
671 static void decodeVideo( hb_work_private_t *pv, uint8_t *data, int size,
672 int64_t pts, int64_t dts )
675 * The following loop is a do..while because we need to handle both
676 * data & the flush at the end (signaled by size=0). At the end there's
677 * generally a frame in the parser & one or more frames in the decoder
678 * (depending on the bframes setting).
684 int len = av_parser_parse( pv->parser, pv->context, &pout, &pout_len,
685 data + pos, size - pos, pts, dts );
690 pv->pts = pv->parser->pts;
691 decodeFrame( pv, pout, pout_len );
693 } while ( pos < size );
695 /* the stuff above flushed the parser, now flush the decoder */
698 while ( decodeFrame( pv, NULL, 0 ) )
701 flushDelayQueue( pv );
705 static hb_buffer_t *link_buf_list( hb_work_private_t *pv )
707 hb_buffer_t *head = hb_list_item( pv->list, 0 );
711 hb_list_rem( pv->list, head );
713 hb_buffer_t *last = head, *buf;
715 while ( ( buf = hb_list_item( pv->list, 0 ) ) != NULL )
717 hb_list_rem( pv->list, buf );
726 static int decavcodecvInit( hb_work_object_t * w, hb_job_t * job )
729 hb_work_private_t *pv = calloc( 1, sizeof( hb_work_private_t ) );
730 w->private_data = pv;
732 pv->list = hb_list_init();
734 int codec_id = w->codec_param;
735 pv->parser = av_parser_init( codec_id );
736 pv->context = avcodec_alloc_context2( CODEC_TYPE_VIDEO );
738 /* we have to wrap ffmpeg's get_buffer to be able to set the pts (?!) */
739 pv->context->opaque = pv;
740 pv->context->get_buffer = get_frame_buf;
745 static int next_hdr( hb_buffer_t *in, int offset )
747 uint8_t *dat = in->data;
748 uint16_t last2 = 0xffff;
749 for ( ; in->size - offset > 1; ++offset )
751 if ( last2 == 0 && dat[offset] == 0x01 )
752 // found an mpeg start code
755 last2 = ( last2 << 8 ) | dat[offset];
761 static int find_hdr( hb_buffer_t *in, int offset, uint8_t hdr_type )
763 if ( in->size - offset < 4 )
764 // not enough room for an mpeg start code
767 for ( ; ( offset = next_hdr( in, offset ) ) >= 0; ++offset )
769 if ( in->data[offset+3] == hdr_type )
776 static int setup_extradata( hb_work_object_t *w, hb_buffer_t *in )
778 hb_work_private_t *pv = w->private_data;
780 // we can't call the avstream funcs but the read_header func in the
781 // AVInputFormat may set up some state in the AVContext. In particular
782 // vc1t_read_header allocates 'extradata' to deal with header issues
783 // related to Microsoft's bizarre engineering notions. We alloc a chunk
784 // of space to make vc1 work then associate the codec with the context.
785 if ( w->codec_param != CODEC_ID_VC1 )
787 // we haven't been inflicted with M$ - allocate a little space as
788 // a marker and return success.
789 pv->context->extradata_size = 16;
790 pv->context->extradata = av_malloc(pv->context->extradata_size);
794 // find the start and and of the sequence header
796 if ( ( shdr = find_hdr( in, 0, 0x0f ) ) < 0 )
798 // didn't find start of seq hdr
801 if ( ( shdr_end = next_hdr( in, shdr + 4 ) ) < 0 )
807 // find the start and and of the entry point header
809 if ( ( ehdr = find_hdr( in, 0, 0x0e ) ) < 0 )
811 // didn't find start of entry point hdr
814 if ( ( ehdr_end = next_hdr( in, ehdr + 4 ) ) < 0 )
820 // found both headers - allocate an extradata big enough to hold both
821 // then copy them into it.
822 pv->context->extradata_size = shdr_end + ehdr_end;
823 pv->context->extradata = av_malloc(pv->context->extradata_size + 8);
824 memcpy( pv->context->extradata, in->data + shdr, shdr_end );
825 memcpy( pv->context->extradata + shdr_end, in->data + ehdr, ehdr_end );
826 memset( pv->context->extradata + shdr_end + ehdr_end, 0, 8);
830 static int decavcodecvWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
831 hb_buffer_t ** buf_out )
833 hb_work_private_t *pv = w->private_data;
834 hb_buffer_t *in = *buf_in;
835 int64_t pts = AV_NOPTS_VALUE;
840 /* if we got an empty buffer signaling end-of-stream send it downstream */
843 decodeVideo( pv, in->data, in->size, pts, dts );
844 hb_list_add( pv->list, in );
845 *buf_out = link_buf_list( pv );
849 // if this is the first frame open the codec (we have to wait for the
850 // first frame because of M$ VC1 braindamage).
851 if ( pv->context->extradata_size == 0 )
853 if ( setup_extradata( w, in ) )
855 // we didn't find the headers needed to set up extradata.
856 // the codec will abort if we open it so just free the buf
857 // and hope we eventually get the info we need.
858 hb_buffer_close( &in );
861 AVCodec *codec = avcodec_find_decoder( w->codec_param );
862 // There's a mis-feature in ffmpeg that causes the context to be
863 // incorrectly initialized the 1st time avcodec_open is called.
864 // If you close it and open a 2nd time, it finishes the job.
865 hb_avcodec_open( pv->context, codec );
866 hb_avcodec_close( pv->context );
867 hb_avcodec_open( pv->context, codec );
873 dts = in->renderOffset;
877 pv->new_chap = in->new_chap;
878 pv->chap_time = pts >= 0? pts : pv->pts_next;
880 decodeVideo( pv, in->data, in->size, pts, dts );
881 hb_buffer_close( &in );
882 *buf_out = link_buf_list( pv );
886 static int decavcodecvInfo( hb_work_object_t *w, hb_work_info_t *info )
888 hb_work_private_t *pv = w->private_data;
890 memset( info, 0, sizeof(*info) );
892 if ( pv && pv->context )
894 AVCodecContext *context = pv->context;
895 info->bitrate = context->bit_rate;
896 info->width = context->width;
897 info->height = context->height;
899 /* ffmpeg gives the frame rate in frames per second while HB wants
900 * it in units of the 27MHz MPEG clock. */
901 info->rate = 27000000;
902 info->rate_base = (int64_t)context->time_base.num * 27000000LL /
903 context->time_base.den;
904 if ( context->ticks_per_frame > 1 )
906 // for ffmpeg 0.5 & later, the H.264 & MPEG-2 time base is
907 // field rate rather than frame rate so convert back to frames.
908 info->rate_base *= context->ticks_per_frame;
911 /* Sometimes there's no pixel aspect set in the source. In that case,
912 assume a 1:1 PAR. Otherwise, preserve the source PAR. */
913 info->pixel_aspect_width = context->sample_aspect_ratio.num ?
914 context->sample_aspect_ratio.num : 1;
915 info->pixel_aspect_height = context->sample_aspect_ratio.den ?
916 context->sample_aspect_ratio.den : 1;
918 /* ffmpeg returns the Pixel Aspect Ratio (PAR). Handbrake wants the
919 * Display Aspect Ratio so we convert by scaling by the Storage
920 * Aspect Ratio (w/h). We do the calc in floating point to get the
922 info->aspect = (double)info->pixel_aspect_width *
923 (double)context->width /
924 (double)info->pixel_aspect_height /
925 (double)context->height;
927 info->profile = context->profile;
928 info->level = context->level;
929 info->name = context->codec->name;
935 static int decavcodecvBSInfo( hb_work_object_t *w, const hb_buffer_t *buf,
936 hb_work_info_t *info )
941 hb_work_object_t hb_decavcodecv =
944 "Video decoder (libavcodec)",
953 // This is a special decoder for ffmpeg streams. The ffmpeg stream reader
954 // includes a parser and passes information from the parser to the decoder
955 // via a codec context kept in the AVStream of the reader's AVFormatContext.
956 // We *have* to use that codec context to decode the stream or we'll get
957 // garbage. ffmpeg_title_scan put a cookie that can be used to get to that
958 // codec context in our codec_param.
960 // this routine gets the appropriate context pointer from the ffmpeg
961 // stream reader. it can't be called until we get the first buffer because
962 // we can't guarantee that reader will be called before the our init
963 // routine and if our init is called first we'll get a pointer to the
964 // old scan stream (which has already been closed).
965 static void init_ffmpeg_context( hb_work_object_t *w )
967 hb_work_private_t *pv = w->private_data;
968 pv->context = hb_ffmpeg_context( w->codec_param );
970 // during scan the decoder gets closed & reopened which will
971 // close the codec so reopen it if it's not there
972 if ( ! pv->context->codec )
974 AVCodec *codec = avcodec_find_decoder( pv->context->codec_id );
975 hb_avcodec_open( pv->context, codec );
977 // set up our best guess at the frame duration.
978 // the frame rate in the codec is usually bogus but it's sometimes
980 AVStream *st = hb_ffmpeg_avstream( w->codec_param );
982 if ( st->nb_frames && st->duration )
984 // compute the average frame duration from the total number
985 // of frames & the total duration.
986 pv->duration = ( (double)st->duration * (double)st->time_base.num ) /
987 ( (double)st->nb_frames * (double)st->time_base.den );
991 // XXX We don't have a frame count or duration so try to use the
992 // far less reliable time base info in the stream.
993 // Because the time bases are so screwed up, we only take values
994 // in the range 8fps - 64fps.
996 if ( st->time_base.num * 64 > st->time_base.den &&
997 st->time_base.den > st->time_base.num * 8 )
1001 else if ( st->r_frame_rate.den * 64 > st->r_frame_rate.num &&
1002 st->r_frame_rate.num > st->r_frame_rate.den * 8 )
1004 tb.num = st->r_frame_rate.den;
1005 tb.den = st->r_frame_rate.num;
1009 tb.num = 1001; /*XXX*/
1010 tb.den = 24000; /*XXX*/
1012 pv->duration = (double)tb.num / (double)tb.den;
1014 pv->duration *= 90000.;
1016 // we have to wrap ffmpeg's get_buffer to be able to set the pts (?!)
1017 pv->context->opaque = pv;
1018 pv->context->get_buffer = get_frame_buf;
1020 // avi, mkv and possibly mp4 containers can contain the M$ VFW packed
1021 // b-frames abortion that messes up frame ordering and timestamps.
1022 // XXX ffmpeg knows which streams are broken but doesn't expose the
1023 // info externally. We should patch ffmpeg to add a flag to the
1024 // codec context for this but until then we mark all ffmpeg streams
1026 pv->brokenByMicrosoft = 1;
1029 static void prepare_ffmpeg_buffer( hb_buffer_t * in )
1031 // ffmpeg requires an extra 8 bytes of zero at the end of the buffer and
1032 // will seg fault in odd, data dependent ways if it's not there. (my guess
1033 // is this is a case of a local performance optimization creating a global
1034 // performance degradation since all the time wasted by extraneous data
1035 // copies & memory zeroing has to be huge compared to the minor reduction
1036 // in inner-loop instructions this affords - modern cpus bottleneck on
1037 // memory bandwidth not instruction bandwidth).
1038 if ( in->size + FF_INPUT_BUFFER_PADDING_SIZE > in->alloc )
1040 // have to realloc to add the padding
1041 hb_buffer_realloc( in, in->size + FF_INPUT_BUFFER_PADDING_SIZE );
1043 memset( in->data + in->size, 0, FF_INPUT_BUFFER_PADDING_SIZE );
1046 static int decavcodecviInit( hb_work_object_t * w, hb_job_t * job )
1049 hb_work_private_t *pv = calloc( 1, sizeof( hb_work_private_t ) );
1050 w->private_data = pv;
1052 pv->list = hb_list_init();
1058 static int decavcodecviWork( hb_work_object_t * w, hb_buffer_t ** buf_in,
1059 hb_buffer_t ** buf_out )
1061 hb_work_private_t *pv = w->private_data;
1062 if ( ! pv->context )
1064 init_ffmpeg_context( w );
1066 hb_buffer_t *in = *buf_in;
1069 /* if we got an empty buffer signaling end-of-stream send it downstream */
1070 if ( in->size == 0 )
1072 /* flush any frames left in the decoder */
1073 while ( decodeFrame( pv, NULL, 0 ) )
1076 flushDelayQueue( pv );
1077 hb_list_add( pv->list, in );
1078 *buf_out = link_buf_list( pv );
1079 return HB_WORK_DONE;
1082 int64_t pts = in->start;
1085 // use the first timestamp as our 'next expected' pts
1086 if ( pv->pts_next < 0 )
1095 pv->new_chap = in->new_chap;
1096 pv->chap_time = pts >= 0? pts : pv->pts_next;
1098 prepare_ffmpeg_buffer( in );
1099 decodeFrame( pv, in->data, in->size );
1100 hb_buffer_close( &in );
1101 *buf_out = link_buf_list( pv );
1105 static int decavcodecviInfo( hb_work_object_t *w, hb_work_info_t *info )
1107 if ( decavcodecvInfo( w, info ) )
1109 hb_work_private_t *pv = w->private_data;
1110 if ( ! pv->context )
1112 init_ffmpeg_context( w );
1114 // we have the frame duration in units of the 90KHz pts clock but
1115 // need it in units of the 27MHz MPEG clock. */
1116 info->rate = 27000000;
1117 info->rate_base = pv->duration * 300.;
1123 static void decodeAudio( hb_work_private_t *pv, uint8_t *data, int size )
1125 AVCodecContext *context = pv->context;
1128 while ( pos < size )
1130 int16_t *buffer = pv->buffer;
1131 if ( buffer == NULL )
1133 pv->buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE );
1134 buffer = pv->buffer;
1136 int out_size = AVCODEC_MAX_AUDIO_FRAME_SIZE;
1137 int len = avcodec_decode_audio2( context, buffer, &out_size,
1138 data + pos, size - pos );
1146 // We require signed 16-bit ints for the output format. If
1147 // we got something different convert it.
1148 if ( context->sample_fmt != SAMPLE_FMT_S16 )
1150 // Note: av_audio_convert seems to be a work-in-progress but
1151 // looks like it will eventually handle general audio
1152 // mixdowns which would allow us much more flexibility
1153 // in handling multichannel audio in HB. If we were doing
1154 // anything more complicated than a one-for-one format
1155 // conversion we'd probably want to cache the converter
1156 // context in the pv.
1157 int isamp = av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1158 AVAudioConvert *ctx = av_audio_convert_alloc( SAMPLE_FMT_S16, 1,
1159 context->sample_fmt, 1,
1161 // get output buffer size (in 2-byte samples) then malloc a buffer
1162 out_size = ( out_size * 2 ) / isamp;
1163 buffer = malloc( out_size );
1165 // we're doing straight sample format conversion which behaves as if
1166 // there were only one channel.
1167 const void * const ibuf[6] = { pv->buffer };
1168 void * const obuf[6] = { buffer };
1169 const int istride[6] = { isamp };
1170 const int ostride[6] = { 2 };
1172 av_audio_convert( ctx, obuf, ostride, ibuf, istride, out_size >> 1 );
1173 av_audio_convert_free( ctx );
1175 hb_buffer_t *buf = hb_buffer_init( 2 * out_size );
1177 // convert from bytes to total samples
1180 double pts = pv->pts_next;
1182 pts += out_size * pv->duration;
1186 float *fl32 = (float *)buf->data;
1188 for( i = 0; i < out_size; ++i )
1190 fl32[i] = buffer[i];
1192 hb_list_add( pv->list, buf );
1194 // if we allocated a buffer for sample format conversion, free it
1195 if ( buffer != pv->buffer )
1203 static int decavcodecaiWork( hb_work_object_t *w, hb_buffer_t **buf_in,
1204 hb_buffer_t **buf_out )
1206 if ( (*buf_in)->size <= 0 )
1208 /* EOF on input stream - send it downstream & say that we're done */
1211 return HB_WORK_DONE;
1214 hb_work_private_t *pv = w->private_data;
1216 if ( (*buf_in)->start < -1 && pv->pts_next <= 0 )
1218 // discard buffers that start before video time 0
1223 if ( ! pv->context )
1225 init_ffmpeg_context( w );
1226 // duration is a scaling factor to go from #bytes in the decoded
1227 // frame to frame time (in 90KHz mpeg ticks). 'channels' converts
1228 // total samples to per-channel samples. 'sample_rate' converts
1229 // per-channel samples to seconds per sample and the 90000
1230 // is mpeg ticks per second.
1231 pv->duration = 90000. /
1232 (double)( pv->context->sample_rate * pv->context->channels );
1234 hb_buffer_t *in = *buf_in;
1236 // if the packet has a timestamp use it if we don't have a timestamp yet
1237 // or if there's been a timing discontinuity of more than 100ms.
1238 if ( in->start >= 0 &&
1239 ( pv->pts_next < 0 || ( in->start - pv->pts_next ) > 90*100 ) )
1241 pv->pts_next = in->start;
1243 prepare_ffmpeg_buffer( in );
1244 decodeAudio( pv, in->data, in->size );
1245 *buf_out = link_buf_list( pv );
1250 hb_work_object_t hb_decavcodecvi =
1253 "Video decoder (ffmpeg streams)",
1261 hb_work_object_t hb_decavcodecai =
1264 "Audio decoder (ffmpeg streams)",