2 * producer_avformat.c -- avformat producer
3 * Copyright (C) 2003-2012 Ushodaya Enterprises Limited
4 * Author: Charles Yates <charles.yates@pandora.be>
5 * Author: Dan Dennedy <dan@dennedy.org>
6 * Much code borrowed from ffmpeg.c: Copyright (c) 2000-2003 Fabrice Bellard
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 #include <framework/mlt_producer.h>
25 #include <framework/mlt_frame.h>
26 #include <framework/mlt_profile.h>
27 #include <framework/mlt_log.h>
28 #include <framework/mlt_deque.h>
29 #include <framework/mlt_factory.h>
30 #include <framework/mlt_cache.h>
32 // ffmpeg Header files
33 #include <libavformat/avformat.h>
34 #include <libswscale/swscale.h>
35 #include <libavutil/samplefmt.h>
36 #include <libavutil/pixdesc.h>
39 # include <libavcodec/vdpau.h>
41 #if (LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0))
42 # include <libavutil/dict.h>
45 // System header files
51 #if LIBAVCODEC_VERSION_MAJOR >= 53
52 #include <libavutil/opt.h>
53 #define CODEC_TYPE_VIDEO AVMEDIA_TYPE_VIDEO
54 #define CODEC_TYPE_AUDIO AVMEDIA_TYPE_AUDIO
55 #define PKT_FLAG_KEY AV_PKT_FLAG_KEY
57 #include <libavcodec/opt.h>
60 #if LIBAVCODEC_VERSION_MAJOR < 55
61 #define AV_CODEC_ID_DVVIDEO CODEC_ID_DVVIDEO
62 #define AV_CODEC_ID_H264 CODEC_ID_H264
65 #define POSITION_INITIAL (-2)
66 #define POSITION_INVALID (-1)
68 #define MAX_AUDIO_STREAMS (32)
69 #define MAX_VDPAU_SURFACES (10)
70 #define MAX_AUDIO_FRAME_SIZE (192000) // 1 second of 48khz 32bit audio
72 struct producer_avformat_s
75 AVFormatContext *dummy_context;
76 AVFormatContext *audio_format;
77 AVFormatContext *video_format;
78 AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
79 AVCodecContext *video_codec;
83 mlt_position audio_expected;
84 mlt_position video_expected;
88 int64_t last_position;
90 int64_t current_position;
91 mlt_position nonseek_position;
93 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
94 size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
95 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
96 int audio_used[ MAX_AUDIO_STREAMS ];
102 unsigned int invalid_pts_counter;
103 unsigned int invalid_dts_counter;
104 mlt_cache image_cache;
107 pthread_mutex_t video_mutex;
108 pthread_mutex_t audio_mutex;
111 pthread_mutex_t packets_mutex;
112 pthread_mutex_t open_mutex;
114 AVRational video_time_base;
119 struct vdpau_render_state render_states[MAX_VDPAU_SURFACES];
133 typedef struct producer_avformat_s *producer_avformat;
135 // Forward references.
136 static int list_components( char* file );
137 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock );
138 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
139 static void producer_avformat_close( producer_avformat );
140 static void producer_close( mlt_producer parent );
141 static void producer_set_up_video( producer_avformat self, mlt_frame frame );
142 static void producer_set_up_audio( producer_avformat self, mlt_frame frame );
143 static void apply_properties( void *obj, mlt_properties properties, int flags );
144 static int video_codec_init( producer_avformat self, int index, mlt_properties properties );
145 static void get_audio_streams_info( producer_avformat self );
146 static mlt_audio_format pick_audio_format( int sample_fmt );
152 /** Constructor for libavformat.
155 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
157 if ( list_components( file ) )
160 mlt_producer producer = NULL;
162 // Check that we have a non-NULL argument
165 // Construct the producer
166 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
167 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
170 if ( mlt_producer_init( producer, self ) == 0 )
172 self->parent = producer;
174 // Get the properties
175 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
177 // Set the resource property (required for all producers)
178 mlt_properties_set( properties, "resource", file );
180 // Register transport implementation with the producer
181 producer->close = (mlt_destructor) producer_close;
183 // Register our get_frame implementation
184 producer->get_frame = producer_get_frame;
186 if ( strcmp( service, "avformat-novalidate" ) )
189 if ( producer_open( self, profile, file, 1 ) != 0 )
192 mlt_producer_close( producer );
194 producer_avformat_close( self );
196 else if ( self->seekable )
198 // Close the file to release resources for large playlists - reopen later as needed
199 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
200 if ( self->audio_format )
201 avformat_close_input( &self->audio_format );
202 if ( self->video_format )
203 avformat_close_input( &self->video_format );
205 if ( self->audio_format )
206 av_close_input_file( self->audio_format );
207 if ( self->video_format )
208 av_close_input_file( self->video_format );
210 self->audio_format = NULL;
211 self->video_format = NULL;
216 // Default the user-selectable indices from the auto-detected indices
217 mlt_properties_set_int( properties, "audio_index", self->audio_index );
218 mlt_properties_set_int( properties, "video_index", self->video_index );
220 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
222 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
229 int list_components( char* file )
233 // Report information about available demuxers and codecs as YAML Tiny
234 if ( file && strstr( file, "f-list" ) )
236 fprintf( stderr, "---\nformats:\n" );
237 AVInputFormat *format = NULL;
238 while ( ( format = av_iformat_next( format ) ) )
239 fprintf( stderr, " - %s\n", format->name );
240 fprintf( stderr, "...\n" );
243 if ( file && strstr( file, "acodec-list" ) )
245 fprintf( stderr, "---\naudio_codecs:\n" );
246 AVCodec *codec = NULL;
247 while ( ( codec = av_codec_next( codec ) ) )
248 if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
249 fprintf( stderr, " - %s\n", codec->name );
250 fprintf( stderr, "...\n" );
253 if ( file && strstr( file, "vcodec-list" ) )
255 fprintf( stderr, "---\nvideo_codecs:\n" );
256 AVCodec *codec = NULL;
257 while ( ( codec = av_codec_next( codec ) ) )
258 if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
259 fprintf( stderr, " - %s\n", codec->name );
260 fprintf( stderr, "...\n" );
267 static int first_video_index( producer_avformat self )
269 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
270 int i = -1; // not found
273 for ( i = 0; i < context->nb_streams; i++ ) {
274 if ( context->streams[i]->codec &&
275 context->streams[i]->codec->codec_type == CODEC_TYPE_VIDEO )
278 if ( i == context->nb_streams )
284 /** Find the default streams.
287 static mlt_properties find_default_streams( producer_avformat self )
291 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
292 AVDictionaryEntry *tag = NULL;
294 AVMetadataTag *tag = NULL;
296 AVFormatContext *context = self->video_format;
297 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
299 // Default to the first audio and video streams found
300 self->audio_index = -1;
301 self->video_index = -1;
303 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
305 // Allow for multiple audio and video streams in the file and select first of each (if available)
306 for( i = 0; i < context->nb_streams; i++ )
308 // Get the codec context
309 AVStream *stream = context->streams[ i ];
310 if ( ! stream ) continue;
311 AVCodecContext *codec_context = stream->codec;
312 if ( ! codec_context ) continue;
313 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
314 if ( ! codec ) continue;
316 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
318 // Determine the type and obtain the first index of each type
319 switch( codec_context->codec_type )
321 case CODEC_TYPE_VIDEO:
322 // Use first video stream
323 if ( self->video_index < 0 )
324 self->video_index = i;
325 mlt_properties_set( meta_media, key, "video" );
326 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
327 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
328 #if LIBAVFORMAT_VERSION_MAJOR < 55
329 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
330 ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
332 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
334 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
335 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
336 snprintf( key, sizeof(key), "meta.media.%d.codec.width", i );
337 mlt_properties_set_int( meta_media, key, codec_context->width );
338 snprintf( key, sizeof(key), "meta.media.%d.codec.height", i );
339 mlt_properties_set_int( meta_media, key, codec_context->height );
340 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
341 AVRational frame_rate = { codec_context->time_base.den, codec_context->time_base.num * codec_context->ticks_per_frame };
342 mlt_properties_set_double( meta_media, key, av_q2d( frame_rate ) );
343 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
344 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(3<<8)+0)
345 mlt_properties_set( meta_media, key, av_get_pix_fmt_name( codec_context->pix_fmt ) );
347 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
349 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
350 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
351 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
352 switch ( codec_context->colorspace )
354 case AVCOL_SPC_SMPTE240M:
355 mlt_properties_set_int( meta_media, key, 240 );
357 case AVCOL_SPC_BT470BG:
358 case AVCOL_SPC_SMPTE170M:
359 mlt_properties_set_int( meta_media, key, 601 );
361 case AVCOL_SPC_BT709:
362 mlt_properties_set_int( meta_media, key, 709 );
365 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
366 mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
370 case CODEC_TYPE_AUDIO:
371 if ( !codec_context->channels )
373 // Use first audio stream
374 if ( self->audio_index < 0 && pick_audio_format( codec_context->sample_fmt ) != mlt_audio_none )
375 self->audio_index = i;
377 mlt_properties_set( meta_media, key, "audio" );
378 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
379 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
380 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
381 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
382 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
383 mlt_properties_set_int( meta_media, key, codec_context->channels );
388 // snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
389 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
390 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
391 mlt_properties_set( meta_media, key, codec->name );
392 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
393 mlt_properties_set( meta_media, key, codec->long_name );
394 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
395 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
396 // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
397 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
398 // snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
399 // mlt_properties_set_int( meta_media, key, codec_context->profile );
400 // snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
401 // mlt_properties_set_int( meta_media, key, codec_context->level );
404 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
405 while ( ( tag = av_dict_get( stream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
407 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
410 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
412 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
413 mlt_properties_set( meta_media, key, tag->value );
417 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
418 while ( ( tag = av_dict_get( context->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
420 while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
423 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
425 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
426 mlt_properties_set( meta_media, key, tag->value );
433 static void get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context )
435 AVRational sar = stream->sample_aspect_ratio;
436 if ( sar.num <= 0 || sar.den <= 0 )
437 sar = codec_context->sample_aspect_ratio;
438 if ( sar.num <= 0 || sar.den <= 0 )
439 sar.num = sar.den = 1;
440 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", sar.num );
441 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", sar.den );
442 mlt_properties_set_double( properties, "aspect_ratio", av_q2d( sar ) );
445 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
446 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVDictionary **params )
448 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params )
451 if ( !URL ) return NULL;
454 char *protocol = strdup( URL );
455 char *url = strchr( protocol, ':' );
457 // Only if there is not a protocol specification that avformat can handle
458 #if LIBAVFORMAT_VERSION_MAJOR >= 53
459 if ( url && avio_check( URL, 0 ) < 0 )
461 if ( url && !url_exist( URL ) )
464 // Truncate protocol string
466 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url + 1 );
469 *format = av_find_input_format( protocol );
471 // Eat the format designator
476 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
477 // support for legacy width and height parameters
481 // These are required by video4linux2 (defaults)
482 params->width = profile->width;
483 params->height = profile->height;
484 if ( !strstr( URL, "&frame_rate" ) )
485 params->time_base = (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
486 params->channels = 2;
487 params->sample_rate = 48000;
491 url = strchr( url, '?' );
495 char *name = strdup( ++url );
496 char *value = strchr( name, '=' );
498 // Also accept : as delimiter for backwards compatibility.
499 value = strchr( name, ':' );
504 char *t = strchr( value, '&' );
507 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
508 // translate old parameters to new av_dict names
509 if ( !strcmp( name, "frame_rate" ) )
510 av_dict_set( params, "framerate", value, 0 );
511 else if ( !strcmp( name, "pix_fmt" ) )
512 av_dict_set( params, "pixel_format", value, 0 );
513 else if ( !strcmp( name, "width" ) )
514 width = strdup( value );
515 else if ( !strcmp( name, "height" ) )
516 height = strdup( value );
518 // generic demux/device option support
519 av_dict_set( params, name, value, 0 );
521 if ( !strcmp( name, "frame_rate" ) )
522 params->time_base.den = atoi( value );
523 else if ( !strcmp( name, "frame_rate_base" ) )
524 params->time_base.num = atoi( value );
525 else if ( !strcmp( name, "sample_rate" ) )
526 params->sample_rate = atoi( value );
527 else if ( !strcmp( name, "channel" ) )
528 params->channel = atoi( value );
529 else if ( !strcmp( name, "channels" ) )
530 params->channels = atoi( value );
531 else if ( !strcmp( name, "pix_fmt" ) )
532 params->pix_fmt = av_get_pix_fmt( value );
533 else if ( !strcmp( name, "width" ) )
534 params->width = atoi( value );
535 else if ( !strcmp( name, "height" ) )
536 params->height = atoi( value );
537 else if ( !strcmp( name, "standard" ) )
538 params->standard = strdup( value );
542 url = strchr( url, '&' );
544 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
545 // continued support for legacy width and height parameters
546 if ( width && height )
548 char *s = malloc( strlen( width ) + strlen( height ) + 2 );
552 av_dict_set( params, "video_size", s, 0 );
555 if ( width ) free( width );
556 if ( height ) free ( height );
559 result = strdup( result );
563 result = strdup( URL );
569 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
573 // Get the properties
574 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
576 AVFormatContext *format = self->video_format;
578 // We will treat everything with the producer fps.
579 // TODO: make this more flexible.
580 double fps = mlt_profile_fps( profile );
583 if ( !mlt_properties_get_int( properties, "_length_computed" ) )
585 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
586 // when producer_open is called after initial call when restoring or reseting the producer.
587 if ( format->duration != AV_NOPTS_VALUE )
589 // This isn't going to be accurate for all formats
590 mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
591 mlt_properties_set_position( properties, "out", frames - 1 );
592 mlt_properties_set_position( properties, "length", frames );
593 mlt_properties_set_int( properties, "_length_computed", 1 );
597 // Set live sources to run forever
598 mlt_properties_set_position( properties, "length", INT_MAX );
599 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
600 mlt_properties_set( properties, "eof", "loop" );
601 mlt_properties_set_int( properties, "_length_computed", 1 );
605 // Check if we're seekable
606 // avdevices are typically AVFMT_NOFILE and not seekable
607 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
610 // protocols can indicate if they support seeking
611 #if LIBAVFORMAT_VERSION_MAJOR >= 53
612 self->seekable = format->pb->seekable;
614 URLContext *uc = url_fileno( format->pb );
616 self->seekable = !uc->is_streamed;
619 if ( self->seekable )
621 // Do a more rigourous test of seekable on a disposable context
622 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
623 mlt_properties_set_int( properties, "seekable", self->seekable );
624 self->dummy_context = format;
625 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
626 self->video_format = NULL;
627 avformat_open_input( &self->video_format, filename, NULL, NULL );
628 avformat_find_stream_info( self->video_format, NULL );
630 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
631 av_find_stream_info( self->video_format );
633 format = self->video_format;
636 // Fetch the width, height and aspect ratio
637 if ( self->video_index != -1 )
639 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
640 mlt_properties_set_int( properties, "width", codec_context->width );
641 mlt_properties_set_int( properties, "height", codec_context->height );
642 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context );
644 // Verify that we can convert this to YUV 4:2:2
645 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
646 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
647 codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
649 sws_freeContext( context );
659 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock )
661 // Return an error code (0 == no error)
663 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
668 if ( !self->is_mutex_init )
670 pthread_mutex_init( &self->audio_mutex, NULL );
671 pthread_mutex_init( &self->video_mutex, NULL );
672 pthread_mutex_init( &self->packets_mutex, NULL );
673 pthread_mutex_init( &self->open_mutex, NULL );
674 self->is_mutex_init = 1;
676 pthread_mutex_lock( &self->audio_mutex );
677 pthread_mutex_lock( &self->video_mutex );
679 mlt_events_block( properties, self->parent );
682 AVInputFormat *format = NULL;
683 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
684 AVDictionary *params = NULL;
686 AVFormatParameters params;
687 memset( ¶ms, 0, sizeof(params) );
689 char *filename = parse_url( profile, URL, &format, ¶ms );
691 // Now attempt to open the file or device with filename
692 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
693 error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0;
695 // If the URL is a network stream URL, then we probably need to open with full URL
696 error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0;
698 error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0;
700 // If the URL is a network stream URL, then we probably need to open with full URL
701 error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0;
704 // Set MLT properties onto video AVFormatContext
705 if ( !error && self->video_format )
707 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
708 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
709 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
712 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
713 av_dict_free( ¶ms );
715 // Cleanup AVFormatParameters
716 if ( params.standard )
717 free( (void*) params.standard );
720 // If successful, then try to get additional info
721 if ( !error && self->video_format )
723 // Get the stream info
724 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
725 error = avformat_find_stream_info( self->video_format, NULL ) < 0;
727 error = av_find_stream_info( self->video_format ) < 0;
730 // Continue if no error
731 if ( !error && self->video_format )
733 // Find default audio and video streams
734 find_default_streams( self );
735 error = get_basic_info( self, profile, filename );
737 // Initialize position info
738 self->first_pts = AV_NOPTS_VALUE;
739 self->last_position = POSITION_INITIAL;
741 if ( !self->audio_format )
743 // We're going to cheat here - for seekable A/V files, we will have separate contexts
744 // to support independent seeking of audio from video.
745 // TODO: Is this really necessary?
746 if ( self->audio_index != -1 && self->video_index != -1 )
748 if ( self->seekable )
750 // And open again for our audio context
751 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
752 avformat_open_input( &self->audio_format, filename, NULL, NULL );
753 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
754 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
755 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
756 avformat_find_stream_info( self->audio_format, NULL );
758 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
759 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
760 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(110<<8)+0)
761 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
762 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
764 av_find_stream_info( self->audio_format );
769 self->audio_format = self->video_format;
772 else if ( self->audio_index != -1 )
774 // We only have an audio context
775 self->audio_format = self->video_format;
776 self->video_format = NULL;
778 else if ( self->video_index == -1 )
780 // Something has gone wrong
783 if ( self->audio_format && !self->audio_streams )
784 get_audio_streams_info( self );
792 self->apackets = mlt_deque_init();
793 self->vpackets = mlt_deque_init();
796 if ( self->dummy_context )
798 pthread_mutex_lock( &self->open_mutex );
799 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
800 avformat_close_input( &self->dummy_context );
802 av_close_input_file( self->dummy_context );
804 self->dummy_context = NULL;
805 pthread_mutex_unlock( &self->open_mutex );
808 // Unlock the service
811 pthread_mutex_unlock( &self->audio_mutex );
812 pthread_mutex_unlock( &self->video_mutex );
814 mlt_events_unblock( properties, self->parent );
819 static void reopen_video( producer_avformat self, mlt_producer producer )
821 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
822 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
823 pthread_mutex_lock( &self->audio_mutex );
824 pthread_mutex_lock( &self->open_mutex );
826 if ( self->video_codec )
827 avcodec_close( self->video_codec );
828 self->video_codec = NULL;
829 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
830 if ( self->dummy_context )
831 avformat_close_input( &self->dummy_context );
832 if ( self->video_format )
833 avformat_close_input( &self->video_format );
835 if ( self->dummy_context )
836 av_close_input_file( self->dummy_context );
837 if ( self->video_format )
838 av_close_input_file( self->video_format );
840 self->dummy_context = NULL;
841 self->video_format = NULL;
842 pthread_mutex_unlock( &self->open_mutex );
844 int audio_index = self->audio_index;
845 int video_index = self->video_index;
847 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
848 mlt_properties_get( properties, "resource" ), 0 );
850 self->audio_index = audio_index;
851 if ( self->video_format && video_index > -1 )
853 self->video_index = video_index;
854 video_codec_init( self, video_index, properties );
857 pthread_mutex_unlock( &self->audio_mutex );
858 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
861 static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts )
863 self->invalid_pts_counter += pts == AV_NOPTS_VALUE;
864 self->invalid_dts_counter += dts == AV_NOPTS_VALUE;
865 if ( ( self->invalid_pts_counter <= self->invalid_dts_counter
866 || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE )
872 static void find_first_pts( producer_avformat self, int video_index )
875 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
880 while ( ret >= 0 && toscan-- > 0 )
882 ret = av_read_frame( context, &pkt );
883 if ( ret >= 0 && pkt.stream_index == video_index && ( pkt.flags & PKT_FLAG_KEY ) )
885 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
886 "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n",
887 pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
888 self->first_pts = best_pts( self, pkt.pts, pkt.dts );
889 if ( self->first_pts != AV_NOPTS_VALUE )
892 av_free_packet( &pkt );
894 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
897 static int seek_video( producer_avformat self, mlt_position position,
898 int64_t req_position, int preseek )
900 mlt_producer producer = self->parent;
903 if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
905 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
907 // Fetch the video format context
908 AVFormatContext *context = self->video_format;
910 // Get the video stream
911 AVStream *stream = context->streams[ self->video_index ];
914 AVCodecContext *codec_context = stream->codec;
916 // We may want to use the source fps if available
917 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
918 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
920 if ( self->last_position == POSITION_INITIAL )
921 find_first_pts( self, self->video_index );
923 if ( self->video_frame && position + 1 == self->video_expected )
925 // We're paused - use last image
928 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
930 // Calculate the timestamp for the requested frame
931 int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps );
932 if ( req_position <= 0 )
934 else if ( self->first_pts != AV_NOPTS_VALUE )
935 timestamp += self->first_pts;
936 else if ( context->start_time != AV_NOPTS_VALUE )
937 timestamp += context->start_time;
938 if ( preseek && av_q2d( self->video_time_base ) != 0 )
939 timestamp -= 2 / av_q2d( self->video_time_base );
942 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %"PRId64"\n",
943 timestamp, position, self->video_expected, self->last_position );
945 // Seek to the timestamp
946 // NOTE: reopen_video is disabled at this time because it is causing trouble with A/V sync.
947 if ( 1 || req_position > 0 || self->last_position <= 0 )
949 codec_context->skip_loop_filter = AVDISCARD_NONREF;
950 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
952 // flush any pictures still in decode buffer
953 avcodec_flush_buffers( codec_context );
957 // Re-open video stream when rewinding to beginning from somewhere else.
958 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
959 // not reliably seeking to the first frame across formats.
960 reopen_video( self, producer );
963 // Remove the cached info relating to the previous position
964 self->current_position = POSITION_INVALID;
965 self->last_position = POSITION_INVALID;
966 av_freep( &self->video_frame );
972 /** Convert a frame position to a time code.
975 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
977 return ( double )position / mlt_producer_get_fps( producer );
980 // Collect information about all audio streams
982 static void get_audio_streams_info( producer_avformat self )
984 // Fetch the audio format context
985 AVFormatContext *context = self->audio_format;
989 i < context->nb_streams;
992 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
994 AVCodecContext *codec_context = context->streams[i]->codec;
995 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
997 // If we don't have a codec and we can't initialise it, we can't do much more...
998 pthread_mutex_lock( &self->open_mutex );
999 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1000 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1002 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1005 self->audio_streams++;
1006 self->audio_max_stream = i;
1007 self->total_channels += codec_context->channels;
1008 if ( codec_context->channels > self->max_channel )
1009 self->max_channel = codec_context->channels;
1010 if ( codec_context->sample_rate > self->max_frequency )
1011 self->max_frequency = codec_context->sample_rate;
1012 avcodec_close( codec_context );
1014 pthread_mutex_unlock( &self->open_mutex );
1017 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1018 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1021 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
1024 const int *new_coefficients;
1026 int brightness, contrast, saturation;
1028 if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
1029 &brightness, &contrast, &saturation ) != -1 )
1031 // Don't change these from defaults unless explicitly told to.
1032 if ( use_full_range >= 0 )
1033 full_range = use_full_range;
1034 switch ( colorspace )
1040 new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1043 new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1046 new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1049 new_coefficients = coefficients;
1052 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
1053 brightness, contrast, saturation );
1057 static mlt_image_format pick_pix_format( enum PixelFormat pix_fmt )
1065 return mlt_image_rgb24a;
1066 case PIX_FMT_YUV420P:
1067 case PIX_FMT_YUVJ420P:
1068 case PIX_FMT_YUVA420P:
1069 return mlt_image_yuv420p;
1073 case PIX_FMT_MONOWHITE:
1074 case PIX_FMT_MONOBLACK:
1077 return mlt_image_rgb24;
1079 return mlt_image_yuv422;
1083 static mlt_audio_format pick_audio_format( int sample_fmt )
1085 switch ( sample_fmt )
1088 case AV_SAMPLE_FMT_U8:
1089 return mlt_audio_u8;
1090 case AV_SAMPLE_FMT_S16:
1091 return mlt_audio_s16;
1092 case AV_SAMPLE_FMT_S32:
1093 return mlt_audio_s32le;
1094 case AV_SAMPLE_FMT_FLT:
1095 return mlt_audio_f32le;
1096 // planar - this producer converts planar to interleaved
1097 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
1098 case AV_SAMPLE_FMT_U8P:
1099 return mlt_audio_u8;
1100 case AV_SAMPLE_FMT_S16P:
1101 return mlt_audio_s16;
1102 case AV_SAMPLE_FMT_S32P:
1103 return mlt_audio_s32le;
1104 case AV_SAMPLE_FMT_FLTP:
1105 return mlt_audio_f32le;
1108 return mlt_audio_none;
1112 static void convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt,
1113 mlt_image_format *format, int width, int height, uint8_t **alpha )
1115 int flags = SWS_BICUBIC | SWS_ACCURATE_RND;
1118 flags |= SWS_CPU_CAPS_MMX;
1121 flags |= SWS_CPU_CAPS_MMX2;
1124 // extract alpha from planar formats
1125 if ( ( pix_fmt == PIX_FMT_YUVA420P
1126 #if defined(FFUDIV) && LIBAVUTIL_VERSION_INT >= ((51<<16)+(35<<8)+101)
1127 || pix_fmt == PIX_FMT_YUVA444P
1130 *format != mlt_image_rgb24a && *format != mlt_image_opengl &&
1131 frame->data[3] && frame->linesize[3] )
1136 dst = *alpha = mlt_pool_alloc( width * height );
1137 src = frame->data[3];
1139 for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ )
1140 memcpy( dst, src, FFMIN( width, frame->linesize[3] ) );
1143 if ( *format == mlt_image_yuv420p )
1145 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1146 width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1148 output.data[0] = buffer;
1149 output.data[1] = buffer + width * height;
1150 output.data[2] = buffer + ( 5 * width * height ) / 4;
1151 output.linesize[0] = width;
1152 output.linesize[1] = width >> 1;
1153 output.linesize[2] = width >> 1;
1154 set_luma_transfer( context, self->colorspace, -1 );
1155 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1156 output.data, output.linesize);
1157 sws_freeContext( context );
1159 else if ( *format == mlt_image_rgb24 )
1161 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1162 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1164 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1165 set_luma_transfer( context, self->colorspace, self->full_luma );
1166 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1167 output.data, output.linesize);
1168 sws_freeContext( context );
1170 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1172 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1173 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1175 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1176 set_luma_transfer( context, self->colorspace, self->full_luma );
1177 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1178 output.data, output.linesize);
1179 sws_freeContext( context );
1183 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1184 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1186 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1187 set_luma_transfer( context, self->colorspace, -1 );
1188 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1189 output.data, output.linesize);
1190 sws_freeContext( context );
1194 /** Allocate the image buffer and set it on the frame.
1197 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1201 if ( codec_context->width == 0 || codec_context->height == 0 )
1204 if ( *format == mlt_image_glsl )
1205 *format = pick_pix_format( codec_context->pix_fmt );
1207 *width = codec_context->width;
1208 *height = codec_context->height;
1209 size = mlt_image_format_size( *format, *width, *height, NULL );
1210 *buffer = mlt_pool_alloc( size );
1212 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1219 /** Get an image from a frame.
1222 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1225 producer_avformat self = mlt_frame_pop_service( frame );
1226 mlt_producer producer = self->parent;
1228 // Get the properties from the frame
1229 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1231 // Obtain the frame number of this frame
1232 mlt_position position = mlt_frame_original_position( frame );
1234 // Get the producer properties
1235 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1237 pthread_mutex_lock( &self->video_mutex );
1239 // Fetch the video format context
1240 AVFormatContext *context = self->video_format;
1242 // Get the video stream
1243 AVStream *stream = context->streams[ self->video_index ];
1245 // Get codec context
1246 AVCodecContext *codec_context = stream->codec;
1248 uint8_t *alpha = NULL;
1249 int got_picture = 0;
1252 // Get the image cache
1253 if ( ! self->image_cache )
1255 // if cache size supplied by environment variable
1256 int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL;
1257 int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0;
1259 // cache size supplied via property
1260 if ( mlt_properties_get( properties, "cache" ) )
1263 cache_size = mlt_properties_get_int( properties, "cache" );
1265 if ( mlt_properties_get_int( properties, "noimagecache" ) )
1267 // create cache if not disabled
1268 if ( !cache_supplied || cache_size > 0 )
1269 self->image_cache = mlt_cache_init();
1270 // set cache size if supplied
1271 if ( self->image_cache && cache_supplied )
1272 mlt_cache_set_size( self->image_cache, cache_size );
1274 if ( self->image_cache )
1276 mlt_frame original = mlt_cache_get_frame( self->image_cache, position );
1279 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
1282 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
1284 mlt_frame_set_alpha( frame, *buffer, size, NULL );
1285 *buffer = mlt_properties_get_data( orig_props, "image", &size );
1286 mlt_frame_set_image( frame, *buffer, size, NULL );
1287 mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL );
1288 *format = mlt_properties_get_int( orig_props, "format" );
1290 // Set the resolution
1291 *width = codec_context->width;
1292 *height = codec_context->height;
1294 // Workaround 1088 encodings missing cropping info.
1295 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1299 goto exit_get_image;
1304 // We may want to use the source fps if available
1305 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1306 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1308 // This is the physical frame position in the source
1309 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1311 // Determines if we have to decode all frames in a sequence
1312 // Temporary hack to improve intra frame only
1313 int must_decode = !( codec_context->codec && codec_context->codec->name ) || (
1314 strcmp( codec_context->codec->name, "dnxhd" ) &&
1315 strcmp( codec_context->codec->name, "dvvideo" ) &&
1316 strcmp( codec_context->codec->name, "huffyuv" ) &&
1317 strcmp( codec_context->codec->name, "mjpeg" ) &&
1318 strcmp( codec_context->codec->name, "rawvideo" ) );
1320 double delay = mlt_properties_get_double( properties, "video_delay" );
1322 // Seek if necessary
1323 const char *interp = mlt_properties_get( frame_properties, "rescale.interp" );
1324 int preseek = must_decode
1325 #if defined(FFUDIV) && LIBAVFORMAT_VERSION_INT >= ((53<<16)+(24<<8)+2)
1326 && ( interp && strcmp( interp, "nearest" ) )
1328 && codec_context->has_b_frames;
1329 int paused = seek_video( self, position, req_position, preseek );
1331 // Seek might have reopened the file
1332 context = self->video_format;
1333 stream = context->streams[ self->video_index ];
1334 codec_context = stream->codec;
1335 if ( *format == mlt_image_none ||
1336 codec_context->pix_fmt == PIX_FMT_ARGB ||
1337 codec_context->pix_fmt == PIX_FMT_RGBA ||
1338 codec_context->pix_fmt == PIX_FMT_ABGR ||
1339 codec_context->pix_fmt == PIX_FMT_BGRA )
1340 *format = pick_pix_format( codec_context->pix_fmt );
1342 // Duplicate the last image if necessary
1343 if ( self->video_frame && self->video_frame->linesize[0]
1344 && ( paused || self->current_position >= req_position ) )
1347 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1349 // Workaround 1088 encodings missing cropping info.
1350 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1353 if ( self->vdpau && self->vdpau->buffer )
1356 picture.data[0] = self->vdpau->buffer;
1357 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1358 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1359 picture.linesize[0] = codec_context->width;
1360 picture.linesize[1] = codec_context->width / 2;
1361 picture.linesize[2] = codec_context->width / 2;
1362 convert_image( self, (AVFrame*) &picture, *buffer,
1363 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1367 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1368 format, *width, *height, &alpha );
1375 int64_t int_position = 0;
1376 int decode_errors = 0;
1378 // Construct an AVFrame for YUV422 conversion
1379 if ( !self->video_frame )
1380 self->video_frame = avcodec_alloc_frame( );
1382 while( ret >= 0 && !got_picture )
1385 if ( self->pkt.stream_index == self->video_index )
1386 av_free_packet( &self->pkt );
1387 av_init_packet( &self->pkt );
1388 pthread_mutex_lock( &self->packets_mutex );
1389 if ( mlt_deque_count( self->vpackets ) )
1391 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1397 ret = av_read_frame( context, &self->pkt );
1398 if ( ret >= 0 && !self->seekable && self->pkt.stream_index == self->audio_index )
1400 if ( !av_dup_packet( &self->pkt ) )
1402 AVPacket *tmp = malloc( sizeof(AVPacket) );
1404 mlt_deque_push_back( self->apackets, tmp );
1408 pthread_mutex_unlock( &self->packets_mutex );
1410 // We only deal with video from the selected video_index
1411 if ( ret >= 0 && self->pkt.stream_index == self->video_index && self->pkt.size > 0 )
1413 int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts );
1414 if ( pts != AV_NOPTS_VALUE )
1416 if ( !self->seekable && self->first_pts == AV_NOPTS_VALUE )
1417 self->first_pts = pts;
1418 if ( self->first_pts != AV_NOPTS_VALUE )
1419 pts -= self->first_pts;
1420 else if ( context->start_time != AV_NOPTS_VALUE )
1421 pts -= context->start_time;
1422 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1423 if ( int_position == self->last_position )
1424 int_position = self->last_position + 1;
1426 mlt_log_debug( MLT_PRODUCER_SERVICE(producer),
1427 "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1428 self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position );
1430 // Make a dumb assumption on streams that contain wild timestamps
1431 if ( abs( req_position - int_position ) > 999 )
1433 int_position = req_position;
1434 mlt_log_warning( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!\n" );
1436 self->last_position = int_position;
1439 if ( must_decode || int_position >= req_position )
1444 if ( self->vdpau->decoder == VDP_INVALID_HANDLE )
1446 vdpau_decoder_init( self );
1448 self->vdpau->is_decoded = 0;
1451 codec_context->reordered_opaque = int_position;
1452 if ( int_position >= req_position )
1453 codec_context->skip_loop_filter = AVDISCARD_NONE;
1454 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1455 ret = avcodec_decode_video2( codec_context, self->video_frame, &got_picture, &self->pkt );
1457 ret = avcodec_decode_video( codec_context, self->video_frame, &got_picture, self->pkt.data, self->pkt.size );
1459 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1462 if ( ++decode_errors <= 10 )
1473 // Get position of reordered frame
1474 int_position = self->video_frame->reordered_opaque;
1475 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(106<<8)+0))
1476 pts = best_pts( self, self->video_frame->pkt_pts, self->video_frame->pkt_dts );
1477 if ( pts != AV_NOPTS_VALUE )
1479 if ( self->first_pts != AV_NOPTS_VALUE )
1480 pts -= self->first_pts;
1481 else if ( context->start_time != AV_NOPTS_VALUE )
1482 pts -= context->start_time;
1483 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1487 if ( int_position < req_position )
1489 else if ( int_position >= req_position )
1490 codec_context->skip_loop_filter = AVDISCARD_NONE;
1492 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, self->pkt.flags & PKT_FLAG_KEY );
1495 // Now handle the picture if we have one
1498 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1500 // Workaround 1088 encodings missing cropping info.
1501 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1506 if ( self->vdpau->is_decoded )
1508 struct vdpau_render_state *render = (struct vdpau_render_state*) self->video_frame->data[0];
1510 uint32_t pitches[3];
1511 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1513 if ( !self->vdpau->buffer )
1514 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1515 self->video_frame->data[0] = planes[0] = self->vdpau->buffer;
1516 self->video_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1517 self->video_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1518 self->video_frame->linesize[0] = pitches[0] = codec_context->width;
1519 self->video_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1520 self->video_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1522 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1523 if ( status == VDP_STATUS_OK )
1525 convert_image( self, self->video_frame, *buffer, PIX_FMT_YUV420P,
1526 format, *width, *height, &alpha );
1530 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1531 image_size = self->vdpau->is_decoded = 0;
1536 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1537 image_size = got_picture = 0;
1542 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1543 format, *width, *height, &alpha );
1544 self->top_field_first |= self->video_frame->top_field_first;
1545 self->current_position = int_position;
1553 // Free packet data if not video and not live audio packet
1554 if ( self->pkt.stream_index != self->video_index &&
1555 !( !self->seekable && self->pkt.stream_index == self->audio_index ) )
1556 av_free_packet( &self->pkt );
1562 mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release );
1564 if ( image_size > 0 && self->image_cache )
1566 mlt_properties_set_int( frame_properties, "format", *format );
1567 mlt_cache_put_frame( self->image_cache, frame );
1570 // Try to duplicate last image if there was a decoding failure
1571 // TODO: with multithread decoding a partial frame decoding resulting
1572 // in failure also resets av_frame making test below fail.
1573 if ( !image_size && self->video_frame && self->video_frame->linesize[0] )
1576 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1578 // Workaround 1088 encodings missing cropping info.
1579 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1582 if ( self->vdpau && self->vdpau->buffer )
1585 picture.data[0] = self->vdpau->buffer;
1586 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1587 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1588 picture.linesize[0] = codec_context->width;
1589 picture.linesize[1] = codec_context->width / 2;
1590 picture.linesize[2] = codec_context->width / 2;
1591 convert_image( self, (AVFrame*) &picture, *buffer,
1592 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1596 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1597 format, *width, *height, &alpha );
1602 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1603 self->video_expected = position + 1;
1607 pthread_mutex_unlock( &self->video_mutex );
1609 // Set the progressive flag
1610 if ( mlt_properties_get( properties, "force_progressive" ) )
1611 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1612 else if ( self->video_frame )
1613 mlt_properties_set_int( frame_properties, "progressive", !self->video_frame->interlaced_frame );
1615 // Set the field order property for this frame
1616 if ( mlt_properties_get( properties, "force_tff" ) )
1617 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1619 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1621 // Set immutable properties of the selected track's (or overridden) source attributes.
1622 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1623 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1624 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1625 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1627 // If we already have RGB, then the full range processing either happened already
1628 // or does not apply (RGB source).
1629 if ( *format == mlt_image_rgb24 || *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1630 mlt_properties_set( frame_properties, "force_full_luma", NULL );
1632 return !got_picture;
1635 /** Process properties as AVOptions and apply to AV context obj
1638 static void apply_properties( void *obj, mlt_properties properties, int flags )
1641 int count = mlt_properties_count( properties );
1642 for ( i = 0; i < count; i++ )
1644 const char *opt_name = mlt_properties_get_name( properties, i );
1645 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(10<<8)+0)
1646 const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, flags );
1648 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1650 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1653 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(12<<8)+0)
1654 av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1655 #elif LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1656 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1657 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1658 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1660 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1666 /** Initialize the video codec context.
1669 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1671 // Initialise the codec if necessary
1672 if ( !self->video_codec )
1674 // Get the video stream
1675 AVStream *stream = self->video_format->streams[ index ];
1677 // Get codec context
1678 AVCodecContext *codec_context = stream->codec;
1681 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1683 if ( codec_context->codec_id == AV_CODEC_ID_H264 )
1685 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1687 if ( vdpau_init( self ) )
1689 self->video_codec = codec_context;
1690 if ( !vdpau_decoder_init( self ) )
1695 codec = avcodec_find_decoder( codec_context->codec_id );
1699 // Initialise multi-threading
1700 int thread_count = mlt_properties_get_int( properties, "threads" );
1701 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1702 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1703 if ( thread_count > 1 )
1704 codec_context->thread_count = thread_count;
1706 // If we don't have a codec and we can't initialise it, we can't do much more...
1707 pthread_mutex_lock( &self->open_mutex );
1708 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1709 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1711 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1714 // Now store the codec with its destructor
1715 self->video_codec = codec_context;
1719 // Remember that we can't use this later
1720 self->video_index = -1;
1721 pthread_mutex_unlock( &self->open_mutex );
1724 pthread_mutex_unlock( &self->open_mutex );
1726 // Process properties as AVOptions
1727 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1728 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(122<<8)+0)
1729 if ( codec->priv_class && codec_context->priv_data )
1730 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1733 // Reset some image properties
1734 if ( self->video_codec )
1736 mlt_properties_set_int( properties, "width", self->video_codec->width );
1737 mlt_properties_set_int( properties, "height", self->video_codec->height );
1738 get_aspect_ratio( properties, stream, self->video_codec );
1741 // Start with the muxer frame rate.
1742 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1743 AVRational frame_rate = stream->avg_frame_rate;
1745 AVRational frame_rate = stream->r_frame_rate;
1747 double fps = av_q2d( frame_rate );
1749 #if LIBAVFORMAT_VERSION_MAJOR < 55
1750 // Verify and sanitize the muxer frame rate.
1751 if ( isnan( fps ) || isinf( fps ) || fps == 0 )
1753 frame_rate = stream->r_frame_rate;
1754 fps = av_q2d( frame_rate );
1757 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) && LIBAVFORMAT_VERSION_MAJOR < 55
1758 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1759 // avg_frame_rate gives some approximate value that does not well match the media.
1760 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1761 // results in some very choppy output, but some value slightly different works
1763 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1765 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1766 fps = av_q2d( frame_rate );
1769 // XXX frame rates less than 1 fps are not considered sane
1770 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1772 // Get the frame rate from the codec.
1773 frame_rate.num = self->video_codec->time_base.den;
1774 frame_rate.den = self->video_codec->time_base.num * self->video_codec->ticks_per_frame;
1775 fps = av_q2d( frame_rate );
1777 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1779 // Use the profile frame rate if all else fails.
1780 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1781 frame_rate.num = profile->frame_rate_num;
1782 frame_rate.den = profile->frame_rate_den;
1785 self->video_time_base = stream->time_base;
1786 if ( mlt_properties_get( properties, "force_fps" ) )
1788 AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 );
1789 self->video_time_base.num *= frame_rate.num * force_fps.den;
1790 self->video_time_base.den *= frame_rate.den * force_fps.num;
1791 frame_rate = force_fps;
1793 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1794 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1796 // Set the YUV colorspace from override or detect
1797 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1798 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1799 if ( ! self->colorspace )
1801 switch ( self->video_codec->colorspace )
1803 case AVCOL_SPC_SMPTE240M:
1804 self->colorspace = 240;
1806 case AVCOL_SPC_BT470BG:
1807 case AVCOL_SPC_SMPTE170M:
1808 self->colorspace = 601;
1810 case AVCOL_SPC_BT709:
1811 self->colorspace = 709;
1814 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1815 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1820 // Let apps get chosen colorspace
1821 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1823 self->full_luma = -1;
1824 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(72<<8)+2)
1825 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range );
1826 if ( codec_context->color_range == AVCOL_RANGE_JPEG )
1827 self->full_luma = 1;
1829 if ( mlt_properties_get( properties, "set.force_full_luma" ) )
1830 self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" );
1832 return self->video_codec && self->video_index > -1;
1835 /** Set up video handling.
1838 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1841 mlt_producer producer = self->parent;
1843 // Get the properties
1844 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1846 // Fetch the video format context
1847 AVFormatContext *context = self->video_format;
1849 // Get the video_index
1850 int index = mlt_properties_get_int( properties, "video_index" );
1852 // Reopen the file if necessary
1853 if ( !context && index > -1 )
1855 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1856 mlt_properties_get( properties, "resource" ), 1 );
1857 context = self->video_format;
1860 // Exception handling for video_index
1861 if ( context && index >= (int) context->nb_streams )
1863 // Get the last video stream
1864 for ( index = context->nb_streams - 1;
1865 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1867 mlt_properties_set_int( properties, "video_index", index );
1869 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1871 // Invalidate the video stream
1873 mlt_properties_set_int( properties, "video_index", index );
1876 // Update the video properties if the index changed
1877 if ( index != self->video_index )
1879 // Reset the video properties if the index changed
1880 self->video_index = index;
1881 pthread_mutex_lock( &self->open_mutex );
1882 if ( self->video_codec )
1883 avcodec_close( self->video_codec );
1884 self->video_codec = NULL;
1885 pthread_mutex_unlock( &self->open_mutex );
1888 // Get the frame properties
1889 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1892 if ( context && index > -1 && video_codec_init( self, index, properties ) )
1894 // Set the frame properties
1895 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1896 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1897 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1899 // Set the width and height
1900 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
1901 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
1902 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1903 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1904 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1905 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
1907 // Workaround 1088 encodings missing cropping info.
1908 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1910 mlt_properties_set_int( properties, "meta.media.height", 1080 );
1913 // Add our image operation
1914 mlt_frame_push_service( frame, self );
1915 mlt_frame_push_get_image( frame, producer_get_image );
1919 // If something failed, use test card image
1920 mlt_properties_set_int( frame_properties, "test_image", 1 );
1924 static int seek_audio( producer_avformat self, mlt_position position, double timecode )
1928 // Seek if necessary
1929 if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) )
1931 if ( self->last_position == POSITION_INITIAL )
1933 int video_index = self->video_index;
1934 if ( video_index == -1 )
1935 video_index = first_video_index( self );
1936 if ( video_index >= 0 )
1937 find_first_pts( self, video_index );
1940 if ( position + 1 == self->audio_expected )
1942 // We're paused - silence required
1945 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
1947 AVFormatContext *context = self->audio_format;
1948 int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
1949 if ( context->start_time != AV_NOPTS_VALUE )
1950 timestamp += context->start_time;
1951 if ( timestamp < 0 )
1954 // Set to the real timecode
1955 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
1958 // Clear the usage in the audio buffer
1959 int i = MAX_AUDIO_STREAMS + 1;
1961 self->audio_used[i - 1] = 0;
1967 static int sample_bytes( AVCodecContext *context )
1969 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
1970 return av_get_bytes_per_sample( context->sample_fmt );
1971 #elif LIBAVCODEC_VERSION_MAJOR >= 53
1972 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
1974 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1978 static void planar_to_interleaved( uint8_t *dest, uint8_t *src, int samples, int channels, int bytes_per_sample )
1981 for ( s = 0; s < samples; s++ )
1983 for ( c = 0; c < channels; c++ )
1985 memcpy( dest, src + ( c * samples + s ) * bytes_per_sample, bytes_per_sample );
1986 dest += bytes_per_sample;
1991 static void planar_to_interleaved2( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample )
1994 for ( s = 0; s < samples; s++ )
1996 for ( c = 0; c < channels; c++ )
1998 memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample );
1999 dest += bytes_per_sample;
2004 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
2006 // Fetch the audio_format
2007 AVFormatContext *context = self->audio_format;
2009 // Get the current stream index
2010 int index = pkt.stream_index;
2012 // Get codec context
2013 AVCodecContext *codec_context = self->audio_codec[ index ];
2015 // Obtain the audio buffers
2016 uint8_t *audio_buffer = self->audio_buffer[ index ];
2017 uint8_t *decode_buffer = self->decode_buffer[ index ];
2019 int audio_used = self->audio_used[ index ];
2020 uint8_t *ptr = pkt.data;
2024 while ( ptr && ret >= 0 && len > 0 )
2026 int sizeof_sample = sample_bytes( codec_context );
2027 int data_size = self->audio_buffer_size[ index ];
2030 #if LIBAVCODEC_VERSION_MAJOR >= 55
2031 if ( !self->audio_frame )
2032 self->audio_frame = avcodec_alloc_frame();
2034 avcodec_get_frame_defaults( self->audio_frame );
2035 ret = avcodec_decode_audio4( codec_context, self->audio_frame, &data_size, &pkt );
2037 data_size = av_samples_get_buffer_size( NULL, codec_context->channels,
2038 self->audio_frame->nb_samples, codec_context->sample_fmt, 1 );
2039 decode_buffer = self->audio_frame->data[0];
2042 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2046 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2050 pkt.size = len -= ret;
2051 pkt.data = ptr += ret;
2053 // If decoded successfully
2054 if ( data_size > 0 )
2056 // Figure out how many samples will be needed after resampling
2057 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2059 // Resize audio buffer to prevent overflow
2060 if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2062 self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 2 ) * channels * sizeof_sample;
2063 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2065 uint8_t *dest = &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ];
2066 switch ( codec_context->sample_fmt )
2068 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
2069 case AV_SAMPLE_FMT_U8P:
2070 case AV_SAMPLE_FMT_S16P:
2071 case AV_SAMPLE_FMT_S32P:
2072 case AV_SAMPLE_FMT_FLTP:
2073 #if LIBAVCODEC_VERSION_MAJOR >= 55
2074 planar_to_interleaved2( dest, self->audio_frame, convert_samples, codec_context->channels, sizeof_sample );
2076 planar_to_interleaved( dest, decode_buffer, convert_samples, codec_context->channels, sizeof_sample );
2081 // Straight copy to audio buffer
2082 memcpy( dest, decode_buffer, data_size );
2084 audio_used += convert_samples;
2087 while ( *ignore && audio_used )
2090 audio_used -= audio_used > samples ? samples : audio_used;
2091 memmove( audio_buffer, &audio_buffer[ samples * codec_context->channels * sizeof_sample ],
2092 audio_used * sizeof_sample );
2097 // If we're behind, ignore this packet
2098 // Skip this on non-seekable, audio-only inputs.
2099 if ( pkt.pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2101 int64_t pts = pkt.pts;
2102 if ( self->first_pts != AV_NOPTS_VALUE )
2103 pts -= self->first_pts;
2104 else if ( context->start_time != AV_NOPTS_VALUE )
2105 pts -= context->start_time;
2106 double timebase = av_q2d( context->streams[ index ]->time_base );
2107 int64_t int_position = ( int64_t )( timebase * pts * fps + 0.5 );
2108 int64_t req_position = ( int64_t )( timecode * fps + 0.5 );
2110 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
2111 "A pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
2112 pkt.pts, pkt.dts, req_position, self->current_position, int_position );
2114 if ( int_position > 0 )
2116 if ( int_position < req_position )
2117 // We are behind, so skip some
2118 *ignore = req_position - int_position;
2119 else if ( self->audio_index != INT_MAX && int_position > req_position + 2 )
2120 // We are ahead, so seek backwards some more
2121 seek_audio( self, req_position, timecode - 1.0 );
2123 // Cancel the find_first_pts() in seek_audio()
2124 if ( self->video_index == -1 && self->last_position == POSITION_INITIAL )
2125 self->last_position = int_position;
2128 self->audio_used[ index ] = audio_used;
2133 /** Get the audio from a frame.
2135 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2138 producer_avformat self = mlt_frame_pop_audio( frame );
2140 pthread_mutex_lock( &self->audio_mutex );
2142 // Obtain the frame number of this frame
2143 mlt_position position = mlt_frame_original_position( frame );
2145 // Calculate the real time code
2146 double real_timecode = producer_time_of_frame( self->parent, position );
2148 // Get the producer fps
2149 double fps = mlt_producer_get_fps( self->parent );
2151 // Number of frames to ignore (for ffwd)
2152 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2154 // Flag for paused (silence)
2155 int paused = seek_audio( self, position, real_timecode );
2157 // Initialize ignore for all streams from the seek return value
2158 int i = MAX_AUDIO_STREAMS;
2160 ignore[i] = ignore[0];
2162 // Fetch the audio_format
2163 AVFormatContext *context = self->audio_format;
2165 int sizeof_sample = sizeof( int16_t );
2167 // Determine the tracks to use
2168 int index = self->audio_index;
2169 int index_max = self->audio_index + 1;
2170 if ( self->audio_index == INT_MAX )
2173 index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
2174 *channels = self->total_channels;
2175 *samples = mlt_sample_calculator( fps, FFMAX( self->max_frequency, *frequency ), position );
2176 *frequency = FFMAX( self->max_frequency, *frequency );
2179 // Initialize the buffers
2180 for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ )
2182 // Get codec context
2183 AVCodecContext *codec_context = self->audio_codec[ index ];
2185 if ( codec_context && !self->audio_buffer[ index ] )
2187 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2188 sizeof_sample = sample_bytes( codec_context );
2190 // Check for audio buffer and create if necessary
2191 self->audio_buffer_size[ index ] = MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2192 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2194 // Check for decoder buffer and create if necessary
2195 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2199 // Get the audio if required
2200 if ( !paused && *frequency > 0 )
2206 av_init_packet( &pkt );
2208 // Caller requested number samples based on requested sample rate.
2209 if ( self->audio_index != INT_MAX )
2210 *samples = mlt_sample_calculator( fps, self->audio_codec[ self->audio_index ]->sample_rate, position );
2212 while ( ret >= 0 && !got_audio )
2214 // Check if the buffer already contains the samples required
2215 if ( self->audio_index != INT_MAX &&
2216 self->audio_used[ self->audio_index ] >= *samples &&
2217 ignore[ self->audio_index ] == 0 )
2222 else if ( self->audio_index == INT_MAX )
2224 // Check if there is enough audio for all streams
2226 for ( index = 0; got_audio && index < index_max; index++ )
2227 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2234 pthread_mutex_lock( &self->packets_mutex );
2235 if ( mlt_deque_count( self->apackets ) )
2237 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2243 ret = av_read_frame( context, &pkt );
2244 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2246 if ( !av_dup_packet( &pkt ) )
2248 AVPacket *tmp = malloc( sizeof(AVPacket) );
2250 mlt_deque_push_back( self->vpackets, tmp );
2254 pthread_mutex_unlock( &self->packets_mutex );
2256 // We only deal with audio from the selected audio index
2257 index = pkt.stream_index;
2258 if ( index < MAX_AUDIO_STREAMS && ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2259 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2261 int channels2 = self->audio_codec[index]->channels;
2262 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2265 if ( self->seekable || index != self->video_index )
2266 av_free_packet( &pkt );
2270 // Set some additional return values
2271 *format = mlt_audio_s16;
2272 if ( self->audio_index != INT_MAX )
2274 index = self->audio_index;
2275 *channels = self->audio_codec[ index ]->channels;
2276 *frequency = self->audio_codec[ index ]->sample_rate;
2277 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2278 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2280 else if ( self->audio_index == INT_MAX )
2282 for ( index = 0; index < index_max; index++ )
2283 if ( self->audio_codec[ index ] )
2285 // XXX: This only works if all audio tracks have the same sample format.
2286 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2287 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2292 // Allocate and set the frame's audio buffer
2293 int size = mlt_audio_format_size( *format, *samples, *channels );
2294 *buffer = mlt_pool_alloc( size );
2295 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2297 // Interleave tracks if audio_index=all
2298 if ( self->audio_index == INT_MAX )
2300 uint8_t *dest = *buffer;
2302 for ( i = 0; i < *samples; i++ )
2304 for ( index = 0; index < index_max; index++ )
2305 if ( self->audio_codec[ index ] )
2307 int current_channels = self->audio_codec[ index ]->channels;
2308 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2309 memcpy( dest, src, current_channels * sizeof_sample );
2310 dest += current_channels * sizeof_sample;
2313 for ( index = 0; index < index_max; index++ )
2314 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2316 int current_channels = self->audio_codec[ index ]->channels;
2317 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2318 self->audio_used[index] -= *samples;
2319 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2322 // Copy a single track to the output buffer
2325 index = self->audio_index;
2327 // Now handle the audio if we have enough
2328 if ( self->audio_used[ index ] > 0 )
2330 uint8_t *src = self->audio_buffer[ index ];
2331 // copy samples from audio_buffer
2332 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2333 memcpy( *buffer, src, size * *channels * sizeof_sample );
2334 // supply the remaining requested samples as silence
2335 if ( *samples > self->audio_used[ index ] )
2336 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2337 // reposition the samples within audio_buffer
2338 self->audio_used[ index ] -= size;
2339 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2343 // Otherwise fill with silence
2344 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2350 // Get silence and don't touch the context
2351 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2354 // Regardless of speed (other than paused), we expect to get the next frame
2356 self->audio_expected = position + 1;
2358 pthread_mutex_unlock( &self->audio_mutex );
2363 /** Initialize the audio codec context.
2366 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2368 // Initialise the codec if necessary
2369 if ( !self->audio_codec[ index ] )
2371 // Get codec context
2372 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2375 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2377 // If we don't have a codec and we can't initialise it, we can't do much more...
2378 pthread_mutex_lock( &self->open_mutex );
2379 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
2380 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2382 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2385 // Now store the codec with its destructor
2386 if ( self->audio_codec[ index ] )
2387 avcodec_close( self->audio_codec[ index ] );
2388 self->audio_codec[ index ] = codec_context;
2392 // Remember that we can't use self later
2393 self->audio_index = -1;
2395 pthread_mutex_unlock( &self->open_mutex );
2397 // Process properties as AVOptions
2398 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2399 if ( codec && codec->priv_class && codec_context->priv_data )
2400 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2402 return self->audio_codec[ index ] && self->audio_index > -1;
2405 /** Set up audio handling.
2408 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2411 mlt_producer producer = self->parent;
2413 // Get the properties
2414 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2416 // Fetch the audio format context
2417 AVFormatContext *context = self->audio_format;
2419 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2421 // Get the audio_index
2422 int index = mlt_properties_get_int( properties, "audio_index" );
2424 // Handle all audio tracks
2425 if ( self->audio_index > -1 &&
2426 mlt_properties_get( properties, "audio_index" ) &&
2427 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2430 // Reopen the file if necessary
2431 if ( !context && self->audio_index > -1 && index > -1 )
2433 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2434 mlt_properties_get( properties, "resource" ), 1 );
2435 context = self->audio_format;
2438 // Exception handling for audio_index
2439 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2441 for ( index = context->nb_streams - 1;
2442 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2444 mlt_properties_set_int( properties, "audio_index", index );
2446 if ( context && index > -1 && index < INT_MAX &&
2447 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2449 index = self->audio_index;
2450 mlt_properties_set_int( properties, "audio_index", index );
2452 if ( context && index > -1 && index < INT_MAX &&
2453 pick_audio_format( context->streams[ index ]->codec->sample_fmt ) == mlt_audio_none )
2458 // Update the audio properties if the index changed
2459 if ( context && index > -1 && index != self->audio_index )
2461 pthread_mutex_lock( &self->open_mutex );
2462 if ( self->audio_codec[ self->audio_index ] )
2463 avcodec_close( self->audio_codec[ self->audio_index ] );
2464 self->audio_codec[ self->audio_index ] = NULL;
2465 pthread_mutex_unlock( &self->open_mutex );
2467 if ( self->audio_index != -1 )
2468 self->audio_index = index;
2473 if ( context && index == INT_MAX )
2475 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2476 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2477 for ( index = 0; index < context->nb_streams; index++ )
2479 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2480 audio_codec_init( self, index, properties );
2483 else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2485 // Set the frame properties
2486 if ( index < MAX_AUDIO_STREAMS )
2488 mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate );
2489 mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels );
2492 if ( context && index > -1 )
2494 // Add our audio operation
2495 mlt_frame_push_audio( frame, self );
2496 mlt_frame_push_audio( frame, producer_get_audio );
2500 /** Our get frame implementation.
2503 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2505 // Access the private data
2506 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2507 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2508 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2513 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2514 producer->child = self;
2515 self->parent = producer;
2516 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2517 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2520 // Create an empty frame
2521 *frame = mlt_frame_init( service);
2525 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2529 mlt_cache_item_close( cache_item );
2533 // Update timecode on the frame we're creating
2534 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2537 producer_set_up_video( self, *frame );
2540 producer_set_up_audio( self, *frame );
2542 // Set the position of this producer
2543 mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2544 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position );
2546 // Calculate the next timecode
2547 mlt_producer_prepare_next( producer );
2552 static void producer_avformat_close( producer_avformat self )
2554 mlt_log_debug( NULL, "producer_avformat_close\n" );
2556 // Cleanup av contexts
2557 av_free_packet( &self->pkt );
2558 av_free( self->video_frame );
2559 av_free( self->audio_frame );
2560 pthread_mutex_lock( &self->open_mutex );
2562 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2564 mlt_pool_release( self->audio_buffer[i] );
2565 av_free( self->decode_buffer[i] );
2566 if ( self->audio_codec[i] )
2567 avcodec_close( self->audio_codec[i] );
2568 self->audio_codec[i] = NULL;
2570 if ( self->video_codec )
2571 avcodec_close( self->video_codec );
2572 self->video_codec = NULL;
2574 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
2575 if ( self->dummy_context )
2576 avformat_close_input( &self->dummy_context );
2577 if ( self->seekable && self->audio_format )
2578 avformat_close_input( &self->audio_format );
2579 if ( self->video_format )
2580 avformat_close_input( &self->video_format );
2582 if ( self->dummy_context )
2583 av_close_input_file( self->dummy_context );
2584 if ( self->seekable && self->audio_format )
2585 av_close_input_file( self->audio_format );
2586 if ( self->video_format )
2587 av_close_input_file( self->video_format );
2589 pthread_mutex_unlock( &self->open_mutex );
2591 vdpau_producer_close( self );
2593 if ( self->image_cache )
2594 mlt_cache_close( self->image_cache );
2596 // Cleanup the mutexes
2597 if ( self->is_mutex_init )
2599 pthread_mutex_destroy( &self->audio_mutex );
2600 pthread_mutex_destroy( &self->video_mutex );
2601 pthread_mutex_destroy( &self->packets_mutex );
2602 pthread_mutex_destroy( &self->open_mutex );
2605 // Cleanup the packet queues
2607 if ( self->apackets )
2609 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2611 av_free_packet( pkt );
2614 mlt_deque_close( self->apackets );
2615 self->apackets = NULL;
2617 if ( self->vpackets )
2619 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2621 av_free_packet( pkt );
2624 mlt_deque_close( self->vpackets );
2625 self->vpackets = NULL;
2631 static void producer_close( mlt_producer parent )
2633 // Remove this instance from the cache
2634 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2637 parent->close = NULL;
2638 mlt_producer_close( parent );