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;
105 int yuv_colorspace, color_primaries;
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;
579 if ( !mlt_properties_get_int( properties, "_length_computed" ) )
581 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
582 // when producer_open is called after initial call when restoring or reseting the producer.
583 if ( format->duration != AV_NOPTS_VALUE )
585 // This isn't going to be accurate for all formats
586 // We will treat everything with the producer fps.
587 mlt_position frames = ( mlt_position )( int )( format->duration *
588 profile->frame_rate_num / profile->frame_rate_den / AV_TIME_BASE);
589 mlt_properties_set_position( properties, "out", frames - 1 );
590 mlt_properties_set_position( properties, "length", frames );
591 mlt_properties_set_int( properties, "_length_computed", 1 );
595 // Set live sources to run forever
596 mlt_properties_set_position( properties, "length", INT_MAX );
597 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
598 mlt_properties_set( properties, "eof", "loop" );
599 mlt_properties_set_int( properties, "_length_computed", 1 );
603 // Check if we're seekable
604 // avdevices are typically AVFMT_NOFILE and not seekable
605 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
608 // protocols can indicate if they support seeking
609 #if LIBAVFORMAT_VERSION_MAJOR >= 53
610 self->seekable = format->pb->seekable;
612 URLContext *uc = url_fileno( format->pb );
614 self->seekable = !uc->is_streamed;
617 if ( self->seekable )
619 // Do a more rigourous test of seekable on a disposable context
620 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
621 mlt_properties_set_int( properties, "seekable", self->seekable );
622 self->dummy_context = format;
623 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
624 self->video_format = NULL;
625 avformat_open_input( &self->video_format, filename, NULL, NULL );
626 avformat_find_stream_info( self->video_format, NULL );
628 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
629 av_find_stream_info( self->video_format );
631 format = self->video_format;
634 // Fetch the width, height and aspect ratio
635 if ( self->video_index != -1 )
637 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
638 mlt_properties_set_int( properties, "width", codec_context->width );
639 mlt_properties_set_int( properties, "height", codec_context->height );
640 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context );
642 // Verify that we can convert this to YUV 4:2:2
643 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
644 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
645 codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
647 sws_freeContext( context );
657 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock )
659 // Return an error code (0 == no error)
661 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
663 if ( !self->is_mutex_init )
665 pthread_mutex_init( &self->audio_mutex, NULL );
666 pthread_mutex_init( &self->video_mutex, NULL );
667 pthread_mutex_init( &self->packets_mutex, NULL );
668 pthread_mutex_init( &self->open_mutex, NULL );
669 self->is_mutex_init = 1;
675 pthread_mutex_lock( &self->audio_mutex );
676 pthread_mutex_lock( &self->video_mutex );
678 mlt_events_block( properties, self->parent );
681 AVInputFormat *format = NULL;
682 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
683 AVDictionary *params = NULL;
685 AVFormatParameters params;
686 memset( ¶ms, 0, sizeof(params) );
688 char *filename = parse_url( profile, URL, &format, ¶ms );
690 // Now attempt to open the file or device with filename
691 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
692 error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0;
694 // If the URL is a network stream URL, then we probably need to open with full URL
695 error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0;
697 error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0;
699 // If the URL is a network stream URL, then we probably need to open with full URL
700 error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0;
703 // Set MLT properties onto video AVFormatContext
704 if ( !error && self->video_format )
706 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
707 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
708 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
711 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
712 av_dict_free( ¶ms );
714 // Cleanup AVFormatParameters
715 if ( params.standard )
716 free( (void*) params.standard );
719 // If successful, then try to get additional info
720 if ( !error && self->video_format )
722 // Get the stream info
723 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
724 error = avformat_find_stream_info( self->video_format, NULL ) < 0;
726 error = av_find_stream_info( self->video_format ) < 0;
729 // Continue if no error
730 if ( !error && self->video_format )
732 // Find default audio and video streams
733 find_default_streams( self );
734 error = get_basic_info( self, profile, filename );
736 // Initialize position info
737 self->first_pts = AV_NOPTS_VALUE;
738 self->last_position = POSITION_INITIAL;
740 if ( !self->audio_format )
742 // We're going to cheat here - for seekable A/V files, we will have separate contexts
743 // to support independent seeking of audio from video.
744 // TODO: Is this really necessary?
745 if ( self->audio_index != -1 && self->video_index != -1 )
747 if ( self->seekable )
749 // And open again for our audio context
750 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
751 avformat_open_input( &self->audio_format, filename, NULL, NULL );
752 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
753 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
754 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
755 avformat_find_stream_info( self->audio_format, NULL );
757 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
758 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
759 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(110<<8)+0)
760 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
761 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
763 av_find_stream_info( self->audio_format );
768 self->audio_format = self->video_format;
771 else if ( self->audio_index != -1 )
773 // We only have an audio context
774 self->audio_format = self->video_format;
775 self->video_format = NULL;
777 else if ( self->video_index == -1 )
779 // Something has gone wrong
782 if ( self->audio_format && !self->audio_streams )
783 get_audio_streams_info( self );
791 self->apackets = mlt_deque_init();
792 self->vpackets = mlt_deque_init();
795 if ( self->dummy_context )
797 pthread_mutex_lock( &self->open_mutex );
798 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
799 avformat_close_input( &self->dummy_context );
801 av_close_input_file( self->dummy_context );
803 self->dummy_context = NULL;
804 pthread_mutex_unlock( &self->open_mutex );
807 // Unlock the service
810 pthread_mutex_unlock( &self->audio_mutex );
811 pthread_mutex_unlock( &self->video_mutex );
813 mlt_events_unblock( properties, self->parent );
818 static void prepare_reopen( producer_avformat self )
820 mlt_service_lock( MLT_PRODUCER_SERVICE( self->parent ) );
821 pthread_mutex_lock( &self->audio_mutex );
822 pthread_mutex_lock( &self->open_mutex );
825 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
827 mlt_pool_release( self->audio_buffer[i] );
828 self->audio_buffer[i] = NULL;
829 av_free( self->decode_buffer[i] );
830 self->decode_buffer[i] = NULL;
831 if ( self->audio_codec[i] )
832 avcodec_close( self->audio_codec[i] );
833 self->audio_codec[i] = NULL;
835 if ( self->video_codec )
836 avcodec_close( self->video_codec );
837 self->video_codec = NULL;
839 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
840 if ( self->seekable && self->audio_format )
841 avformat_close_input( &self->audio_format );
842 if ( self->video_format )
843 avformat_close_input( &self->video_format );
845 if ( self->seekable && self->audio_format )
846 av_close_input_file( self->audio_format );
847 if ( self->video_format )
848 av_close_input_file( self->video_format );
850 self->audio_format = NULL;
851 self->video_format = NULL;
852 pthread_mutex_unlock( &self->open_mutex );
854 // Cleanup the packet queues
856 if ( self->apackets )
858 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
860 av_free_packet( pkt );
863 mlt_deque_close( self->apackets );
864 self->apackets = NULL;
866 if ( self->vpackets )
868 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
870 av_free_packet( pkt );
873 mlt_deque_close( self->vpackets );
874 self->vpackets = NULL;
876 pthread_mutex_unlock( &self->audio_mutex );
877 mlt_service_unlock( MLT_PRODUCER_SERVICE( self->parent ) );
880 static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts )
882 self->invalid_pts_counter += pts == AV_NOPTS_VALUE;
883 self->invalid_dts_counter += dts == AV_NOPTS_VALUE;
884 if ( ( self->invalid_pts_counter <= self->invalid_dts_counter
885 || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE )
891 static void find_first_pts( producer_avformat self, int video_index )
894 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
899 while ( ret >= 0 && toscan-- > 0 )
901 ret = av_read_frame( context, &pkt );
902 if ( ret >= 0 && pkt.stream_index == video_index && ( pkt.flags & PKT_FLAG_KEY ) )
904 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
905 "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n",
906 pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
907 self->first_pts = best_pts( self, pkt.pts, pkt.dts );
908 if ( self->first_pts != AV_NOPTS_VALUE )
911 av_free_packet( &pkt );
913 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
916 static int seek_video( producer_avformat self, mlt_position position,
917 int64_t req_position, int preseek )
919 mlt_producer producer = self->parent;
922 if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
924 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
926 // Fetch the video format context
927 AVFormatContext *context = self->video_format;
929 // Get the video stream
930 AVStream *stream = context->streams[ self->video_index ];
933 AVCodecContext *codec_context = stream->codec;
935 // We may want to use the source fps if available
936 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
937 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
939 if ( self->last_position == POSITION_INITIAL )
940 find_first_pts( self, self->video_index );
942 if ( self->video_frame && position + 1 == self->video_expected )
944 // We're paused - use last image
947 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
949 // Calculate the timestamp for the requested frame
950 int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps );
951 if ( req_position <= 0 )
953 else if ( self->first_pts != AV_NOPTS_VALUE )
954 timestamp += self->first_pts;
955 else if ( context->start_time != AV_NOPTS_VALUE )
956 timestamp += context->start_time;
957 if ( preseek && av_q2d( self->video_time_base ) != 0 )
958 timestamp -= 2 / av_q2d( self->video_time_base );
961 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position " MLT_POSITION_FMT " expected "MLT_POSITION_FMT" last_pos %"PRId64"\n",
962 timestamp, position, self->video_expected, self->last_position );
964 // Seek to the timestamp
965 codec_context->skip_loop_filter = AVDISCARD_NONREF;
966 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
968 // flush any pictures still in decode buffer
969 avcodec_flush_buffers( codec_context );
971 // Remove the cached info relating to the previous position
972 self->current_position = POSITION_INVALID;
973 self->last_position = POSITION_INVALID;
974 av_freep( &self->video_frame );
980 /** Convert a frame position to a time code.
983 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
985 return ( double )position / mlt_producer_get_fps( producer );
988 // Collect information about all audio streams
990 static void get_audio_streams_info( producer_avformat self )
992 // Fetch the audio format context
993 AVFormatContext *context = self->audio_format;
997 i < context->nb_streams;
1000 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
1002 AVCodecContext *codec_context = context->streams[i]->codec;
1003 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1005 // If we don't have a codec and we can't initialise it, we can't do much more...
1006 pthread_mutex_lock( &self->open_mutex );
1007 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1008 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1010 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1013 self->audio_streams++;
1014 self->audio_max_stream = i;
1015 self->total_channels += codec_context->channels;
1016 if ( codec_context->channels > self->max_channel )
1017 self->max_channel = codec_context->channels;
1018 if ( codec_context->sample_rate > self->max_frequency )
1019 self->max_frequency = codec_context->sample_rate;
1020 avcodec_close( codec_context );
1022 pthread_mutex_unlock( &self->open_mutex );
1025 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1026 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1029 static int set_luma_transfer( struct SwsContext *context, int src_colorspace, int dst_colorspace, int full_range )
1031 const int *src_coefficients = sws_getCoefficients( SWS_CS_DEFAULT );
1032 const int *dst_coefficients = sws_getCoefficients( SWS_CS_DEFAULT );
1034 int contrast = 1 << 16;
1035 int saturation = 1 << 16;
1037 switch ( src_colorspace )
1043 src_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1046 src_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1049 src_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1054 switch ( dst_colorspace )
1060 src_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1063 src_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1066 src_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1071 return sws_setColorspaceDetails( context, src_coefficients, full_range, dst_coefficients, full_range,
1072 brightness, contrast, saturation );
1075 static mlt_image_format pick_pix_format( enum PixelFormat pix_fmt )
1083 return mlt_image_rgb24a;
1084 case PIX_FMT_YUV420P:
1085 case PIX_FMT_YUVJ420P:
1086 case PIX_FMT_YUVA420P:
1087 return mlt_image_yuv420p;
1091 case PIX_FMT_MONOWHITE:
1092 case PIX_FMT_MONOBLACK:
1095 return mlt_image_rgb24;
1097 return mlt_image_yuv422;
1101 static mlt_audio_format pick_audio_format( int sample_fmt )
1103 switch ( sample_fmt )
1106 case AV_SAMPLE_FMT_U8:
1107 return mlt_audio_u8;
1108 case AV_SAMPLE_FMT_S16:
1109 return mlt_audio_s16;
1110 case AV_SAMPLE_FMT_S32:
1111 return mlt_audio_s32le;
1112 case AV_SAMPLE_FMT_FLT:
1113 return mlt_audio_f32le;
1114 // planar - this producer converts planar to interleaved
1115 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
1116 case AV_SAMPLE_FMT_U8P:
1117 return mlt_audio_u8;
1118 case AV_SAMPLE_FMT_S16P:
1119 return mlt_audio_s16;
1120 case AV_SAMPLE_FMT_S32P:
1121 return mlt_audio_s32le;
1122 case AV_SAMPLE_FMT_FLTP:
1123 return mlt_audio_f32le;
1126 return mlt_audio_none;
1130 // returns resulting YUV colorspace
1131 static int convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt,
1132 mlt_image_format *format, int width, int height, uint8_t **alpha )
1134 int flags = SWS_BICUBIC | SWS_ACCURATE_RND;
1135 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1136 int result = self->yuv_colorspace;
1139 flags |= SWS_CPU_CAPS_MMX;
1142 flags |= SWS_CPU_CAPS_MMX2;
1145 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "%s @ %dx%d space %d->%d\n",
1146 mlt_image_format_name( *format ),
1147 width, height, self->yuv_colorspace, profile->colorspace );
1149 // extract alpha from planar formats
1150 if ( ( pix_fmt == PIX_FMT_YUVA420P
1151 #if defined(FFUDIV) && LIBAVUTIL_VERSION_INT >= ((51<<16)+(35<<8)+101)
1152 || pix_fmt == PIX_FMT_YUVA444P
1155 *format != mlt_image_rgb24a && *format != mlt_image_opengl &&
1156 frame->data[3] && frame->linesize[3] )
1161 dst = *alpha = mlt_pool_alloc( width * height );
1162 src = frame->data[3];
1164 for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ )
1165 memcpy( dst, src, FFMIN( width, frame->linesize[3] ) );
1168 if ( *format == mlt_image_yuv420p )
1170 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1171 width, height, self->full_luma ? PIX_FMT_YUVJ420P : PIX_FMT_YUV420P,
1172 flags, NULL, NULL, NULL);
1174 output.data[0] = buffer;
1175 output.data[1] = buffer + width * height;
1176 output.data[2] = buffer + ( 5 * width * height ) / 4;
1177 output.linesize[0] = width;
1178 output.linesize[1] = width >> 1;
1179 output.linesize[2] = width >> 1;
1180 if ( !set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma ) )
1181 result = profile->colorspace;
1182 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1183 output.data, output.linesize);
1184 sws_freeContext( context );
1186 else if ( *format == mlt_image_rgb24 )
1188 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1189 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1191 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1192 set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma );
1193 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1194 output.data, output.linesize);
1195 sws_freeContext( context );
1197 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1199 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1200 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1202 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1203 set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma );
1204 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1205 output.data, output.linesize);
1206 sws_freeContext( context );
1210 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1211 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1213 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1214 if ( !set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma ) )
1215 result = profile->colorspace;
1216 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1217 output.data, output.linesize);
1218 sws_freeContext( context );
1223 /** Allocate the image buffer and set it on the frame.
1226 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1230 if ( codec_context->width == 0 || codec_context->height == 0 )
1233 if ( *format == mlt_image_glsl )
1234 *format = pick_pix_format( codec_context->pix_fmt );
1236 *width = codec_context->width;
1237 *height = codec_context->height;
1238 size = mlt_image_format_size( *format, *width, *height, NULL );
1239 *buffer = mlt_pool_alloc( size );
1241 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1248 /** Get an image from a frame.
1251 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1254 producer_avformat self = mlt_frame_pop_service( frame );
1255 mlt_producer producer = self->parent;
1257 // Get the properties from the frame
1258 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1260 // Obtain the frame number of this frame
1261 mlt_position position = mlt_frame_original_position( frame );
1263 // Get the producer properties
1264 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1266 pthread_mutex_lock( &self->video_mutex );
1268 uint8_t *alpha = NULL;
1269 int got_picture = 0;
1272 // Fetch the video format context
1273 AVFormatContext *context = self->video_format;
1275 goto exit_get_image;
1277 // Get the video stream
1278 AVStream *stream = context->streams[ self->video_index ];
1280 // Get codec context
1281 AVCodecContext *codec_context = stream->codec;
1283 // Get the image cache
1284 if ( ! self->image_cache )
1286 // if cache size supplied by environment variable
1287 int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL;
1288 int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0;
1290 // cache size supplied via property
1291 if ( mlt_properties_get( properties, "cache" ) )
1294 cache_size = mlt_properties_get_int( properties, "cache" );
1296 if ( mlt_properties_get_int( properties, "noimagecache" ) )
1298 // create cache if not disabled
1299 if ( !cache_supplied || cache_size > 0 )
1300 self->image_cache = mlt_cache_init();
1301 // set cache size if supplied
1302 if ( self->image_cache && cache_supplied )
1303 mlt_cache_set_size( self->image_cache, cache_size );
1305 if ( self->image_cache )
1307 mlt_frame original = mlt_cache_get_frame( self->image_cache, position );
1310 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
1313 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
1315 mlt_frame_set_alpha( frame, *buffer, size, NULL );
1316 *buffer = mlt_properties_get_data( orig_props, "image", &size );
1317 mlt_frame_set_image( frame, *buffer, size, NULL );
1318 mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL );
1319 *format = mlt_properties_get_int( orig_props, "format" );
1321 // Set the resolution
1322 *width = codec_context->width;
1323 *height = codec_context->height;
1325 // Workaround 1088 encodings missing cropping info.
1326 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1330 goto exit_get_image;
1335 // We may want to use the source fps if available
1336 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1337 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1339 // This is the physical frame position in the source
1340 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1342 // Determines if we have to decode all frames in a sequence
1343 // Temporary hack to improve intra frame only
1344 int must_decode = !( codec_context->codec && codec_context->codec->name ) || (
1345 strcmp( codec_context->codec->name, "dnxhd" ) &&
1346 strcmp( codec_context->codec->name, "dvvideo" ) &&
1347 strcmp( codec_context->codec->name, "huffyuv" ) &&
1348 strcmp( codec_context->codec->name, "mjpeg" ) &&
1349 strcmp( codec_context->codec->name, "rawvideo" ) );
1351 double delay = mlt_properties_get_double( properties, "video_delay" );
1353 // Seek if necessary
1354 const char *interp = mlt_properties_get( frame_properties, "rescale.interp" );
1355 int preseek = must_decode
1356 #if defined(FFUDIV) && LIBAVFORMAT_VERSION_INT >= ((53<<16)+(24<<8)+2)
1357 && ( interp && strcmp( interp, "nearest" ) )
1359 && codec_context->has_b_frames;
1360 int paused = seek_video( self, position, req_position, preseek );
1362 // Seek might have reopened the file
1363 context = self->video_format;
1364 stream = context->streams[ self->video_index ];
1365 codec_context = stream->codec;
1366 if ( *format == mlt_image_none ||
1367 codec_context->pix_fmt == PIX_FMT_ARGB ||
1368 codec_context->pix_fmt == PIX_FMT_RGBA ||
1369 codec_context->pix_fmt == PIX_FMT_ABGR ||
1370 codec_context->pix_fmt == PIX_FMT_BGRA )
1371 *format = pick_pix_format( codec_context->pix_fmt );
1373 // Duplicate the last image if necessary
1374 if ( self->video_frame && self->video_frame->linesize[0]
1375 && ( paused || self->current_position >= req_position ) )
1378 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1380 // Workaround 1088 encodings missing cropping info.
1381 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1384 if ( self->vdpau && self->vdpau->buffer )
1387 picture.data[0] = self->vdpau->buffer;
1388 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1389 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1390 picture.linesize[0] = codec_context->width;
1391 picture.linesize[1] = codec_context->width / 2;
1392 picture.linesize[2] = codec_context->width / 2;
1393 int yuv_colorspace = convert_image( self, (AVFrame*) &picture, *buffer,
1394 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1395 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1399 int yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1400 format, *width, *height, &alpha );
1401 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1408 int64_t int_position = 0;
1409 int decode_errors = 0;
1411 // Construct an AVFrame for YUV422 conversion
1412 if ( !self->video_frame )
1413 self->video_frame = avcodec_alloc_frame( );
1415 while( ret >= 0 && !got_picture )
1418 if ( self->pkt.stream_index == self->video_index )
1419 av_free_packet( &self->pkt );
1420 av_init_packet( &self->pkt );
1421 pthread_mutex_lock( &self->packets_mutex );
1422 if ( mlt_deque_count( self->vpackets ) )
1424 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1430 ret = av_read_frame( context, &self->pkt );
1431 if ( ret >= 0 && !self->seekable && self->pkt.stream_index == self->audio_index )
1433 if ( !av_dup_packet( &self->pkt ) )
1435 AVPacket *tmp = malloc( sizeof(AVPacket) );
1437 mlt_deque_push_back( self->apackets, tmp );
1442 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_image\n", ret );
1443 if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) )
1445 // Try to reconnect to live sources by closing context and codecs,
1446 // and letting next call to get_frame() reopen.
1447 prepare_reopen( self );
1448 pthread_mutex_unlock( &self->packets_mutex );
1449 goto exit_get_image;
1451 if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
1453 mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" );
1454 exit( EXIT_FAILURE );
1458 pthread_mutex_unlock( &self->packets_mutex );
1460 // We only deal with video from the selected video_index
1461 if ( ret >= 0 && self->pkt.stream_index == self->video_index && self->pkt.size > 0 )
1463 int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts );
1464 if ( pts != AV_NOPTS_VALUE )
1466 if ( !self->seekable && self->first_pts == AV_NOPTS_VALUE )
1467 self->first_pts = pts;
1468 if ( self->first_pts != AV_NOPTS_VALUE )
1469 pts -= self->first_pts;
1470 else if ( context->start_time != AV_NOPTS_VALUE )
1471 pts -= context->start_time;
1472 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1473 if ( int_position == self->last_position )
1474 int_position = self->last_position + 1;
1476 mlt_log_debug( MLT_PRODUCER_SERVICE(producer),
1477 "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1478 self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position );
1480 // Make a dumb assumption on streams that contain wild timestamps
1481 if ( abs( req_position - int_position ) > 999 )
1483 int_position = req_position;
1484 mlt_log_warning( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!\n" );
1486 self->last_position = int_position;
1489 if ( must_decode || int_position >= req_position )
1494 if ( self->vdpau->decoder == VDP_INVALID_HANDLE )
1496 vdpau_decoder_init( self );
1498 self->vdpau->is_decoded = 0;
1501 codec_context->reordered_opaque = int_position;
1502 if ( int_position >= req_position )
1503 codec_context->skip_loop_filter = AVDISCARD_NONE;
1504 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1505 ret = avcodec_decode_video2( codec_context, self->video_frame, &got_picture, &self->pkt );
1507 ret = avcodec_decode_video( codec_context, self->video_frame, &got_picture, self->pkt.data, self->pkt.size );
1509 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1512 if ( ++decode_errors <= 10 )
1515 mlt_log_warning( MLT_PRODUCER_SERVICE(producer), "video decoding error %d\n", ret );
1525 // Get position of reordered frame
1526 int_position = self->video_frame->reordered_opaque;
1527 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(106<<8)+0))
1528 pts = best_pts( self, self->video_frame->pkt_pts, self->video_frame->pkt_dts );
1529 if ( pts != AV_NOPTS_VALUE )
1531 if ( self->first_pts != AV_NOPTS_VALUE )
1532 pts -= self->first_pts;
1533 else if ( context->start_time != AV_NOPTS_VALUE )
1534 pts -= context->start_time;
1535 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1539 if ( int_position < req_position )
1541 else if ( int_position >= req_position )
1542 codec_context->skip_loop_filter = AVDISCARD_NONE;
1544 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, self->pkt.flags & PKT_FLAG_KEY );
1547 // Now handle the picture if we have one
1550 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1552 // Workaround 1088 encodings missing cropping info.
1553 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1558 if ( self->vdpau->is_decoded )
1560 struct vdpau_render_state *render = (struct vdpau_render_state*) self->video_frame->data[0];
1562 uint32_t pitches[3];
1563 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1565 if ( !self->vdpau->buffer )
1566 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1567 self->video_frame->data[0] = planes[0] = self->vdpau->buffer;
1568 self->video_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1569 self->video_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1570 self->video_frame->linesize[0] = pitches[0] = codec_context->width;
1571 self->video_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1572 self->video_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1574 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1575 if ( status == VDP_STATUS_OK )
1577 int yuv_colorspace = convert_image( self, self->video_frame, *buffer, PIX_FMT_YUV420P,
1578 format, *width, *height, &alpha );
1579 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1583 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1584 image_size = self->vdpau->is_decoded = 0;
1589 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1590 image_size = got_picture = 0;
1595 int yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1596 format, *width, *height, &alpha );
1597 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1598 self->top_field_first |= self->video_frame->top_field_first;
1599 self->current_position = int_position;
1607 // Free packet data if not video and not live audio packet
1608 if ( self->pkt.stream_index != self->video_index &&
1609 !( !self->seekable && self->pkt.stream_index == self->audio_index ) )
1610 av_free_packet( &self->pkt );
1616 mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release );
1618 if ( image_size > 0 && self->image_cache )
1620 mlt_properties_set_int( frame_properties, "format", *format );
1621 mlt_cache_put_frame( self->image_cache, frame );
1624 // Try to duplicate last image if there was a decoding failure
1625 // TODO: with multithread decoding a partial frame decoding resulting
1626 // in failure also resets av_frame making test below fail.
1627 if ( !image_size && self->video_frame && self->video_frame->linesize[0] )
1630 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1632 // Workaround 1088 encodings missing cropping info.
1633 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1636 if ( self->vdpau && self->vdpau->buffer )
1639 picture.data[0] = self->vdpau->buffer;
1640 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1641 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1642 picture.linesize[0] = codec_context->width;
1643 picture.linesize[1] = codec_context->width / 2;
1644 picture.linesize[2] = codec_context->width / 2;
1645 int yuv_colorspace = convert_image( self, (AVFrame*) &picture, *buffer,
1646 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1647 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1651 int yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1652 format, *width, *height, &alpha );
1653 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1658 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1659 self->video_expected = position + 1;
1663 pthread_mutex_unlock( &self->video_mutex );
1665 // Set the progressive flag
1666 if ( mlt_properties_get( properties, "force_progressive" ) )
1667 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1668 else if ( self->video_frame )
1669 mlt_properties_set_int( frame_properties, "progressive", !self->video_frame->interlaced_frame );
1671 // Set the field order property for this frame
1672 if ( mlt_properties_get( properties, "force_tff" ) )
1673 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1675 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1677 // Set immutable properties of the selected track's (or overridden) source attributes.
1678 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1679 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1680 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1681 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1683 // If we already have RGB, then the full range processing either happened already
1684 // or does not apply (RGB source).
1685 if ( *format == mlt_image_rgb24 || *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1686 mlt_properties_set( frame_properties, "force_full_luma", NULL );
1688 return !got_picture;
1691 /** Process properties as AVOptions and apply to AV context obj
1694 static void apply_properties( void *obj, mlt_properties properties, int flags )
1697 int count = mlt_properties_count( properties );
1698 for ( i = 0; i < count; i++ )
1700 const char *opt_name = mlt_properties_get_name( properties, i );
1701 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(10<<8)+0)
1702 const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, flags );
1704 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1706 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1709 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(12<<8)+0)
1710 av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1711 #elif LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1712 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1713 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1714 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1716 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1722 /** Initialize the video codec context.
1725 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1727 // Initialise the codec if necessary
1728 if ( !self->video_codec )
1730 // Get the video stream
1731 AVStream *stream = self->video_format->streams[ index ];
1733 // Get codec context
1734 AVCodecContext *codec_context = stream->codec;
1737 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1739 if ( codec_context->codec_id == AV_CODEC_ID_H264 )
1741 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1743 if ( vdpau_init( self ) )
1745 self->video_codec = codec_context;
1746 if ( !vdpau_decoder_init( self ) )
1751 codec = avcodec_find_decoder( codec_context->codec_id );
1755 // Initialise multi-threading
1756 int thread_count = mlt_properties_get_int( properties, "threads" );
1757 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1758 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1759 if ( thread_count > 1 )
1760 codec_context->thread_count = thread_count;
1762 // If we don't have a codec and we can't initialise it, we can't do much more...
1763 pthread_mutex_lock( &self->open_mutex );
1764 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1765 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1767 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1770 // Now store the codec with its destructor
1771 self->video_codec = codec_context;
1775 // Remember that we can't use this later
1776 self->video_index = -1;
1777 pthread_mutex_unlock( &self->open_mutex );
1780 pthread_mutex_unlock( &self->open_mutex );
1782 // Process properties as AVOptions
1783 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1784 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(122<<8)+0)
1785 if ( codec->priv_class && codec_context->priv_data )
1786 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1789 // Reset some image properties
1790 if ( self->video_codec )
1792 mlt_properties_set_int( properties, "width", self->video_codec->width );
1793 mlt_properties_set_int( properties, "height", self->video_codec->height );
1794 get_aspect_ratio( properties, stream, self->video_codec );
1797 // Start with the muxer frame rate.
1798 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1799 AVRational frame_rate = stream->avg_frame_rate;
1801 AVRational frame_rate = stream->r_frame_rate;
1803 double fps = av_q2d( frame_rate );
1805 #if LIBAVFORMAT_VERSION_MAJOR < 55
1806 // Verify and sanitize the muxer frame rate.
1807 if ( isnan( fps ) || isinf( fps ) || fps == 0 )
1809 frame_rate = stream->r_frame_rate;
1810 fps = av_q2d( frame_rate );
1813 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) && LIBAVFORMAT_VERSION_MAJOR < 55
1814 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1815 // avg_frame_rate gives some approximate value that does not well match the media.
1816 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1817 // results in some very choppy output, but some value slightly different works
1819 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1821 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1822 fps = av_q2d( frame_rate );
1825 // XXX frame rates less than 1 fps are not considered sane
1826 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1828 // Get the frame rate from the codec.
1829 frame_rate.num = self->video_codec->time_base.den;
1830 frame_rate.den = self->video_codec->time_base.num * self->video_codec->ticks_per_frame;
1831 fps = av_q2d( frame_rate );
1833 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1835 // Use the profile frame rate if all else fails.
1836 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1837 frame_rate.num = profile->frame_rate_num;
1838 frame_rate.den = profile->frame_rate_den;
1841 self->video_time_base = stream->time_base;
1842 if ( mlt_properties_get( properties, "force_fps" ) )
1844 AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 );
1845 self->video_time_base = av_mul_q( stream->time_base, av_div_q( frame_rate, force_fps ) );
1846 frame_rate = force_fps;
1848 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1849 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1851 // Set the YUV colorspace from override or detect
1852 self->yuv_colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1853 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1854 if ( ! self->yuv_colorspace )
1856 switch ( self->video_codec->colorspace )
1858 case AVCOL_SPC_SMPTE240M:
1859 self->yuv_colorspace = 240;
1861 case AVCOL_SPC_BT470BG:
1862 case AVCOL_SPC_SMPTE170M:
1863 self->yuv_colorspace = 601;
1865 case AVCOL_SPC_BT709:
1866 self->yuv_colorspace = 709;
1869 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1870 self->yuv_colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1875 // Let apps get chosen colorspace
1876 mlt_properties_set_int( properties, "meta.media.colorspace", self->yuv_colorspace );
1878 switch ( self->video_codec->color_primaries )
1880 case AVCOL_PRI_BT470BG:
1881 self->color_primaries = 601625;
1883 case AVCOL_PRI_SMPTE170M:
1884 case AVCOL_PRI_SMPTE240M:
1885 self->color_primaries = 601525;
1887 case AVCOL_PRI_BT709:
1888 case AVCOL_PRI_UNSPECIFIED:
1890 self->color_primaries = 709;
1894 self->full_luma = 0;
1895 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(72<<8)+2)
1896 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range );
1897 if ( codec_context->color_range == AVCOL_RANGE_JPEG )
1898 self->full_luma = 1;
1900 if ( mlt_properties_get( properties, "set.force_full_luma" ) )
1901 self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" );
1903 return self->video_codec && self->video_index > -1;
1906 /** Set up video handling.
1909 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1912 mlt_producer producer = self->parent;
1914 // Get the properties
1915 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1917 // Fetch the video format context
1918 AVFormatContext *context = self->video_format;
1920 // Get the video_index
1921 int index = mlt_properties_get_int( properties, "video_index" );
1923 int unlock_needed = 0;
1925 // Reopen the file if necessary
1926 if ( !context && index > -1 )
1929 pthread_mutex_lock( &self->video_mutex );
1930 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1931 mlt_properties_get( properties, "resource" ), 0 );
1932 context = self->video_format;
1935 // Exception handling for video_index
1936 if ( context && index >= (int) context->nb_streams )
1938 // Get the last video stream
1939 for ( index = context->nb_streams - 1;
1940 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1942 mlt_properties_set_int( properties, "video_index", index );
1944 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1946 // Invalidate the video stream
1948 mlt_properties_set_int( properties, "video_index", index );
1951 // Update the video properties if the index changed
1952 if ( context && index > -1 && index != self->video_index )
1954 // Reset the video properties if the index changed
1955 self->video_index = index;
1956 pthread_mutex_lock( &self->open_mutex );
1957 if ( self->video_codec )
1958 avcodec_close( self->video_codec );
1959 self->video_codec = NULL;
1960 pthread_mutex_unlock( &self->open_mutex );
1963 // Get the frame properties
1964 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1967 if ( context && index > -1 && video_codec_init( self, index, properties ) )
1969 // Set the frame properties
1970 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1971 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1972 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1974 // Set the width and height
1975 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
1976 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
1977 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1978 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1979 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1980 mlt_properties_set_int( frame_properties, "colorspace", self->yuv_colorspace );
1981 mlt_properties_set_int( frame_properties, "color_primaries", self->color_primaries );
1982 mlt_properties_set_int( frame_properties, "full_luma", self->full_luma );
1984 // Workaround 1088 encodings missing cropping info.
1985 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1987 mlt_properties_set_int( properties, "meta.media.height", 1080 );
1990 // Add our image operation
1991 mlt_frame_push_service( frame, self );
1992 mlt_frame_push_get_image( frame, producer_get_image );
1996 // If something failed, use test card image
1997 mlt_properties_set_int( frame_properties, "test_image", 1 );
1999 if ( unlock_needed )
2000 pthread_mutex_unlock( &self->video_mutex );
2003 static int seek_audio( producer_avformat self, mlt_position position, double timecode )
2007 // Seek if necessary
2008 if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) )
2010 if ( self->last_position == POSITION_INITIAL )
2012 int video_index = self->video_index;
2013 if ( video_index == -1 )
2014 video_index = first_video_index( self );
2015 if ( video_index >= 0 )
2016 find_first_pts( self, video_index );
2019 if ( position + 1 == self->audio_expected )
2021 // We're paused - silence required
2024 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
2026 AVFormatContext *context = self->audio_format;
2027 int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
2028 if ( context->start_time != AV_NOPTS_VALUE )
2029 timestamp += context->start_time;
2030 if ( timestamp < 0 )
2033 // Set to the real timecode
2034 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
2037 // Clear the usage in the audio buffer
2038 int i = MAX_AUDIO_STREAMS + 1;
2040 self->audio_used[i - 1] = 0;
2046 static int sample_bytes( AVCodecContext *context )
2048 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
2049 return av_get_bytes_per_sample( context->sample_fmt );
2050 #elif LIBAVCODEC_VERSION_MAJOR >= 53
2051 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
2053 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
2057 #if LIBAVCODEC_VERSION_MAJOR >= 55
2058 static void planar_to_interleaved( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample )
2061 for ( s = 0; s < samples; s++ )
2063 for ( c = 0; c < channels; c++ )
2065 memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample );
2066 dest += bytes_per_sample;
2071 static void planar_to_interleaved( uint8_t *dest, uint8_t *src, int samples, int channels, int bytes_per_sample )
2074 for ( s = 0; s < samples; s++ )
2076 for ( c = 0; c < channels; c++ )
2078 memcpy( dest, src + ( c * samples + s ) * bytes_per_sample, bytes_per_sample );
2079 dest += bytes_per_sample;
2085 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
2087 // Fetch the audio_format
2088 AVFormatContext *context = self->audio_format;
2090 // Get the current stream index
2091 int index = pkt.stream_index;
2093 // Get codec context
2094 AVCodecContext *codec_context = self->audio_codec[ index ];
2096 // Obtain the audio buffers
2097 uint8_t *audio_buffer = self->audio_buffer[ index ];
2098 uint8_t *decode_buffer = self->decode_buffer[ index ];
2100 int audio_used = self->audio_used[ index ];
2101 uint8_t *ptr = pkt.data;
2105 while ( ptr && ret >= 0 && len > 0 )
2107 int sizeof_sample = sample_bytes( codec_context );
2108 int data_size = self->audio_buffer_size[ index ];
2111 #if LIBAVCODEC_VERSION_MAJOR >= 55
2112 if ( !self->audio_frame )
2113 self->audio_frame = avcodec_alloc_frame();
2115 avcodec_get_frame_defaults( self->audio_frame );
2116 ret = avcodec_decode_audio4( codec_context, self->audio_frame, &data_size, &pkt );
2118 data_size = av_samples_get_buffer_size( NULL, codec_context->channels,
2119 self->audio_frame->nb_samples, codec_context->sample_fmt, 1 );
2120 decode_buffer = self->audio_frame->data[0];
2123 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2127 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2131 pkt.size = len -= ret;
2132 pkt.data = ptr += ret;
2134 // If decoded successfully
2135 if ( data_size > 0 )
2137 // Figure out how many samples will be needed after resampling
2138 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2140 // Resize audio buffer to prevent overflow
2141 if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2143 self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 2 ) * channels * sizeof_sample;
2144 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2146 uint8_t *dest = &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ];
2147 switch ( codec_context->sample_fmt )
2149 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
2150 case AV_SAMPLE_FMT_U8P:
2151 case AV_SAMPLE_FMT_S16P:
2152 case AV_SAMPLE_FMT_S32P:
2153 case AV_SAMPLE_FMT_FLTP:
2154 #if LIBAVCODEC_VERSION_MAJOR >= 55
2155 planar_to_interleaved( dest, self->audio_frame, convert_samples, codec_context->channels, sizeof_sample );
2157 planar_to_interleaved( dest, decode_buffer, convert_samples, codec_context->channels, sizeof_sample );
2162 // Straight copy to audio buffer
2163 memcpy( dest, decode_buffer, data_size );
2165 audio_used += convert_samples;
2168 while ( *ignore && audio_used )
2171 audio_used -= audio_used > samples ? samples : audio_used;
2172 memmove( audio_buffer, &audio_buffer[ samples * codec_context->channels * sizeof_sample ],
2173 audio_used * sizeof_sample );
2178 // If we're behind, ignore this packet
2179 // Skip this on non-seekable, audio-only inputs.
2180 if ( pkt.pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2182 int64_t pts = pkt.pts;
2183 if ( self->first_pts != AV_NOPTS_VALUE )
2184 pts -= self->first_pts;
2185 else if ( context->start_time != AV_NOPTS_VALUE )
2186 pts -= context->start_time;
2187 double timebase = av_q2d( context->streams[ index ]->time_base );
2188 int64_t int_position = ( int64_t )( timebase * pts * fps + 0.5 );
2189 int64_t req_position = ( int64_t )( timecode * fps + 0.5 );
2191 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
2192 "A pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
2193 pkt.pts, pkt.dts, req_position, self->current_position, int_position );
2195 if ( int_position > 0 )
2197 if ( int_position < req_position )
2198 // We are behind, so skip some
2199 *ignore = req_position - int_position;
2200 else if ( self->audio_index != INT_MAX && int_position > req_position + 2 )
2201 // We are ahead, so seek backwards some more
2202 seek_audio( self, req_position, timecode - 1.0 );
2204 // Cancel the find_first_pts() in seek_audio()
2205 if ( self->video_index == -1 && self->last_position == POSITION_INITIAL )
2206 self->last_position = int_position;
2209 self->audio_used[ index ] = audio_used;
2214 /** Get the audio from a frame.
2216 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2219 producer_avformat self = mlt_frame_pop_audio( frame );
2221 pthread_mutex_lock( &self->audio_mutex );
2223 // Obtain the frame number of this frame
2224 mlt_position position = mlt_frame_original_position( frame );
2226 // Calculate the real time code
2227 double real_timecode = producer_time_of_frame( self->parent, position );
2229 // Get the producer fps
2230 double fps = mlt_producer_get_fps( self->parent );
2232 // Number of frames to ignore (for ffwd)
2233 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2235 // Flag for paused (silence)
2236 int paused = seek_audio( self, position, real_timecode );
2238 // Initialize ignore for all streams from the seek return value
2239 int i = MAX_AUDIO_STREAMS;
2241 ignore[i] = ignore[0];
2243 // Fetch the audio_format
2244 AVFormatContext *context = self->audio_format;
2246 goto exit_get_audio;
2248 int sizeof_sample = sizeof( int16_t );
2250 // Determine the tracks to use
2251 int index = self->audio_index;
2252 int index_max = self->audio_index + 1;
2253 if ( self->audio_index == INT_MAX )
2256 index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
2257 *channels = self->total_channels;
2258 *samples = mlt_sample_calculator( fps, self->max_frequency, position );
2259 *frequency = self->max_frequency;
2262 // Initialize the buffers
2263 for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ )
2265 // Get codec context
2266 AVCodecContext *codec_context = self->audio_codec[ index ];
2268 if ( codec_context && !self->audio_buffer[ index ] )
2270 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2271 sizeof_sample = sample_bytes( codec_context );
2273 // Check for audio buffer and create if necessary
2274 self->audio_buffer_size[ index ] = MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2275 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2277 // Check for decoder buffer and create if necessary
2278 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2282 // Get the audio if required
2283 if ( !paused && *frequency > 0 )
2289 av_init_packet( &pkt );
2291 // Caller requested number samples based on requested sample rate.
2292 if ( self->audio_index != INT_MAX )
2293 *samples = mlt_sample_calculator( fps, self->audio_codec[ self->audio_index ]->sample_rate, position );
2295 while ( ret >= 0 && !got_audio )
2297 // Check if the buffer already contains the samples required
2298 if ( self->audio_index != INT_MAX &&
2299 self->audio_used[ self->audio_index ] >= *samples &&
2300 ignore[ self->audio_index ] == 0 )
2305 else if ( self->audio_index == INT_MAX )
2307 // Check if there is enough audio for all streams
2309 for ( index = 0; got_audio && index < index_max; index++ )
2310 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2317 pthread_mutex_lock( &self->packets_mutex );
2318 if ( mlt_deque_count( self->apackets ) )
2320 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2326 ret = av_read_frame( context, &pkt );
2327 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2329 if ( !av_dup_packet( &pkt ) )
2331 AVPacket *tmp = malloc( sizeof(AVPacket) );
2333 mlt_deque_push_back( self->vpackets, tmp );
2338 mlt_producer producer = self->parent;
2339 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2340 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_audio\n", ret );
2341 if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) )
2343 // Try to reconnect to live sources by closing context and codecs,
2344 // and letting next call to get_frame() reopen.
2345 prepare_reopen( self );
2346 pthread_mutex_unlock( &self->packets_mutex );
2347 goto exit_get_audio;
2349 if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
2351 mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" );
2352 exit( EXIT_FAILURE );
2356 pthread_mutex_unlock( &self->packets_mutex );
2358 // We only deal with audio from the selected audio index
2359 index = pkt.stream_index;
2360 if ( index < MAX_AUDIO_STREAMS && ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2361 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2363 int channels2 = self->audio_codec[index]->channels;
2364 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2367 if ( self->seekable || index != self->video_index )
2368 av_free_packet( &pkt );
2372 // Set some additional return values
2373 *format = mlt_audio_s16;
2374 if ( self->audio_index != INT_MAX )
2376 index = self->audio_index;
2377 *channels = self->audio_codec[ index ]->channels;
2378 *frequency = self->audio_codec[ index ]->sample_rate;
2379 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2380 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2382 else if ( self->audio_index == INT_MAX )
2384 for ( index = 0; index < index_max; index++ )
2385 if ( self->audio_codec[ index ] )
2387 // XXX: This only works if all audio tracks have the same sample format.
2388 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2389 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2394 // Allocate and set the frame's audio buffer
2395 int size = mlt_audio_format_size( *format, *samples, *channels );
2396 *buffer = mlt_pool_alloc( size );
2397 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2399 // Interleave tracks if audio_index=all
2400 if ( self->audio_index == INT_MAX )
2402 uint8_t *dest = *buffer;
2404 for ( i = 0; i < *samples; i++ )
2406 for ( index = 0; index < index_max; index++ )
2407 if ( self->audio_codec[ index ] )
2409 int current_channels = self->audio_codec[ index ]->channels;
2410 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2411 memcpy( dest, src, current_channels * sizeof_sample );
2412 dest += current_channels * sizeof_sample;
2415 for ( index = 0; index < index_max; index++ )
2416 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2418 int current_channels = self->audio_codec[ index ]->channels;
2419 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2420 self->audio_used[index] -= *samples;
2421 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2424 // Copy a single track to the output buffer
2427 index = self->audio_index;
2429 // Now handle the audio if we have enough
2430 if ( self->audio_used[ index ] > 0 )
2432 uint8_t *src = self->audio_buffer[ index ];
2433 // copy samples from audio_buffer
2434 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2435 memcpy( *buffer, src, size * *channels * sizeof_sample );
2436 // supply the remaining requested samples as silence
2437 if ( *samples > self->audio_used[ index ] )
2438 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2439 // reposition the samples within audio_buffer
2440 self->audio_used[ index ] -= size;
2441 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2445 // Otherwise fill with silence
2446 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2453 // Get silence and don't touch the context
2454 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2457 // Regardless of speed (other than paused), we expect to get the next frame
2459 self->audio_expected = position + 1;
2461 pthread_mutex_unlock( &self->audio_mutex );
2466 /** Initialize the audio codec context.
2469 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2471 // Initialise the codec if necessary
2472 if ( !self->audio_codec[ index ] )
2474 // Get codec context
2475 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2478 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2480 // If we don't have a codec and we can't initialise it, we can't do much more...
2481 pthread_mutex_lock( &self->open_mutex );
2482 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
2483 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2485 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2488 // Now store the codec with its destructor
2489 if ( self->audio_codec[ index ] )
2490 avcodec_close( self->audio_codec[ index ] );
2491 self->audio_codec[ index ] = codec_context;
2495 // Remember that we can't use self later
2496 self->audio_index = -1;
2498 pthread_mutex_unlock( &self->open_mutex );
2500 // Process properties as AVOptions
2501 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2502 if ( codec && codec->priv_class && codec_context->priv_data )
2503 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2505 return self->audio_codec[ index ] && self->audio_index > -1;
2508 /** Set up audio handling.
2511 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2514 mlt_producer producer = self->parent;
2516 // Get the properties
2517 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2519 // Fetch the audio format context
2520 AVFormatContext *context = self->audio_format;
2522 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2524 // Get the audio_index
2525 int index = mlt_properties_get_int( properties, "audio_index" );
2527 // Handle all audio tracks
2528 if ( self->audio_index > -1 &&
2529 mlt_properties_get( properties, "audio_index" ) &&
2530 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2533 // Reopen the file if necessary
2534 if ( !context && self->audio_index > -1 && index > -1 )
2536 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2537 mlt_properties_get( properties, "resource" ), 1 );
2538 context = self->audio_format;
2541 // Exception handling for audio_index
2542 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2544 for ( index = context->nb_streams - 1;
2545 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2547 mlt_properties_set_int( properties, "audio_index", index );
2549 if ( context && index > -1 && index < INT_MAX &&
2550 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2552 index = self->audio_index;
2553 mlt_properties_set_int( properties, "audio_index", index );
2555 if ( context && index > -1 && index < INT_MAX &&
2556 pick_audio_format( context->streams[ index ]->codec->sample_fmt ) == mlt_audio_none )
2561 // Update the audio properties if the index changed
2562 if ( context && index > -1 && index != self->audio_index )
2564 pthread_mutex_lock( &self->open_mutex );
2565 if ( self->audio_codec[ self->audio_index ] )
2566 avcodec_close( self->audio_codec[ self->audio_index ] );
2567 self->audio_codec[ self->audio_index ] = NULL;
2568 pthread_mutex_unlock( &self->open_mutex );
2570 if ( self->audio_index != -1 )
2571 self->audio_index = index;
2576 if ( context && index == INT_MAX )
2578 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2579 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2580 for ( index = 0; index < context->nb_streams; index++ )
2582 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2583 audio_codec_init( self, index, properties );
2586 else if ( context && index > -1 && index < MAX_AUDIO_STREAMS &&
2587 audio_codec_init( self, index, properties ) )
2589 mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate );
2590 mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels );
2592 if ( context && index > -1 )
2594 // Add our audio operation
2595 mlt_frame_push_audio( frame, self );
2596 mlt_frame_push_audio( frame, producer_get_audio );
2600 /** Our get frame implementation.
2603 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2605 // Access the private data
2606 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2607 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2608 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2613 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2614 producer->child = self;
2615 self->parent = producer;
2616 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2617 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2620 // Create an empty frame
2621 *frame = mlt_frame_init( service);
2625 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2629 mlt_cache_item_close( cache_item );
2633 // Update timecode on the frame we're creating
2634 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2637 producer_set_up_video( self, *frame );
2640 producer_set_up_audio( self, *frame );
2642 // Set the position of this producer
2643 mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2644 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position );
2646 // Calculate the next timecode
2647 mlt_producer_prepare_next( producer );
2652 static void producer_avformat_close( producer_avformat self )
2654 mlt_log_debug( NULL, "producer_avformat_close\n" );
2656 // Cleanup av contexts
2657 av_free_packet( &self->pkt );
2658 av_free( self->video_frame );
2659 av_free( self->audio_frame );
2660 if ( self->is_mutex_init )
2661 pthread_mutex_lock( &self->open_mutex );
2663 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2665 mlt_pool_release( self->audio_buffer[i] );
2666 av_free( self->decode_buffer[i] );
2667 if ( self->audio_codec[i] )
2668 avcodec_close( self->audio_codec[i] );
2669 self->audio_codec[i] = NULL;
2671 if ( self->video_codec )
2672 avcodec_close( self->video_codec );
2673 self->video_codec = NULL;
2675 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
2676 if ( self->dummy_context )
2677 avformat_close_input( &self->dummy_context );
2678 if ( self->seekable && self->audio_format )
2679 avformat_close_input( &self->audio_format );
2680 if ( self->video_format )
2681 avformat_close_input( &self->video_format );
2683 if ( self->dummy_context )
2684 av_close_input_file( self->dummy_context );
2685 if ( self->seekable && self->audio_format )
2686 av_close_input_file( self->audio_format );
2687 if ( self->video_format )
2688 av_close_input_file( self->video_format );
2690 if ( self->is_mutex_init )
2691 pthread_mutex_unlock( &self->open_mutex );
2693 vdpau_producer_close( self );
2695 if ( self->image_cache )
2696 mlt_cache_close( self->image_cache );
2698 // Cleanup the mutexes
2699 if ( self->is_mutex_init )
2701 pthread_mutex_destroy( &self->audio_mutex );
2702 pthread_mutex_destroy( &self->video_mutex );
2703 pthread_mutex_destroy( &self->packets_mutex );
2704 pthread_mutex_destroy( &self->open_mutex );
2707 // Cleanup the packet queues
2709 if ( self->apackets )
2711 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2713 av_free_packet( pkt );
2716 mlt_deque_close( self->apackets );
2717 self->apackets = NULL;
2719 if ( self->vpackets )
2721 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2723 av_free_packet( pkt );
2726 mlt_deque_close( self->vpackets );
2727 self->vpackets = NULL;
2733 static void producer_close( mlt_producer parent )
2735 // Remove this instance from the cache
2736 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2739 parent->close = NULL;
2740 mlt_producer_close( parent );