2 * producer_avformat.c -- avformat producer
3 * Copyright (C) 2003-2009 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>
35 # include <libswscale/swscale.h>
37 #if LIBAVCODEC_VERSION_MAJOR > 52
38 #include <libavutil/samplefmt.h>
39 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
40 const char *avcodec_get_sample_fmt_name(int sample_fmt);
43 # include <libavcodec/vdpau.h>
45 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
46 # include <libavutil/pixdesc.h>
49 // System header files
55 #if LIBAVUTIL_VERSION_INT < (50<<16)
56 #define PIX_FMT_RGB32 PIX_FMT_RGBA32
57 #define PIX_FMT_YUYV422 PIX_FMT_YUV422
60 #if LIBAVCODEC_VERSION_MAJOR > 52
61 #include <libavutil/opt.h>
62 #define CODEC_TYPE_VIDEO AVMEDIA_TYPE_VIDEO
63 #define CODEC_TYPE_AUDIO AVMEDIA_TYPE_AUDIO
64 #define PKT_FLAG_KEY AV_PKT_FLAG_KEY
66 #include <libavcodec/opt.h>
69 #define POSITION_INITIAL (-2)
70 #define POSITION_INVALID (-1)
72 #define MAX_AUDIO_STREAMS (10)
73 #define MAX_VDPAU_SURFACES (10)
75 void avformat_lock( );
76 void avformat_unlock( );
78 struct producer_avformat_s
81 AVFormatContext *dummy_context;
82 AVFormatContext *audio_format;
83 AVFormatContext *video_format;
84 AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
85 AVCodecContext *video_codec;
87 ReSampleContext *audio_resample[ MAX_AUDIO_STREAMS ];
88 mlt_position audio_expected;
89 mlt_position video_expected;
97 mlt_position nonseek_position;
100 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
101 size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
102 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
103 int audio_used[ MAX_AUDIO_STREAMS ];
105 int audio_max_stream;
109 unsigned int invalid_pts_counter;
110 double resample_factor;
111 mlt_cache image_cache;
113 pthread_mutex_t video_mutex;
114 pthread_mutex_t audio_mutex;
117 pthread_mutex_t packets_mutex;
122 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 );
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 );
151 /** Constructor for libavformat.
154 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
156 if ( list_components( file ) )
159 mlt_producer producer = NULL;
161 // Check that we have a non-NULL argument
164 // Construct the producer
165 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
166 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
169 if ( mlt_producer_init( producer, self ) == 0 )
171 self->parent = producer;
173 // Get the properties
174 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
176 // Set the resource property (required for all producers)
177 mlt_properties_set( properties, "resource", file );
179 // Register transport implementation with the producer
180 producer->close = (mlt_destructor) producer_close;
182 // Register our get_frame implementation
183 producer->get_frame = producer_get_frame;
185 if ( strcmp( service, "avformat-novalidate" ) )
188 if ( producer_open( self, profile, file ) != 0 )
191 mlt_producer_close( producer );
194 else if ( self->seekable )
196 // Close the file to release resources for large playlists - reopen later as needed
198 if ( self->audio_format )
199 av_close_input_file( self->audio_format );
200 self->audio_format = NULL;
201 if ( self->video_format )
202 av_close_input_file( self->video_format );
203 self->video_format = NULL;
209 // Default the user-selectable indices from the auto-detected indices
210 mlt_properties_set_int( properties, "audio_index", self->audio_index );
211 mlt_properties_set_int( properties, "video_index", self->video_index );
213 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
215 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
222 int list_components( char* file )
226 // Report information about available demuxers and codecs as YAML Tiny
227 if ( file && strstr( file, "f-list" ) )
229 fprintf( stderr, "---\nformats:\n" );
230 AVInputFormat *format = NULL;
231 while ( ( format = av_iformat_next( format ) ) )
232 fprintf( stderr, " - %s\n", format->name );
233 fprintf( stderr, "...\n" );
236 if ( file && strstr( file, "acodec-list" ) )
238 fprintf( stderr, "---\naudio_codecs:\n" );
239 AVCodec *codec = NULL;
240 while ( ( codec = av_codec_next( codec ) ) )
241 if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
242 fprintf( stderr, " - %s\n", codec->name );
243 fprintf( stderr, "...\n" );
246 if ( file && strstr( file, "vcodec-list" ) )
248 fprintf( stderr, "---\nvideo_codecs:\n" );
249 AVCodec *codec = NULL;
250 while ( ( codec = av_codec_next( codec ) ) )
251 if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
252 fprintf( stderr, " - %s\n", codec->name );
253 fprintf( stderr, "...\n" );
260 /** Find the default streams.
263 static mlt_properties find_default_streams( producer_avformat self )
267 AVMetadataTag *tag = NULL;
268 AVFormatContext *context = self->video_format;
269 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
271 // Default to the first audio and video streams found
272 self->audio_index = -1;
273 self->video_index = -1;
275 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
277 // Allow for multiple audio and video streams in the file and select first of each (if available)
278 for( i = 0; i < context->nb_streams; i++ )
280 // Get the codec context
281 AVStream *stream = context->streams[ i ];
282 if ( ! stream ) continue;
283 AVCodecContext *codec_context = stream->codec;
284 if ( ! codec_context ) continue;
285 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
286 if ( ! codec ) continue;
288 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
290 // Determine the type and obtain the first index of each type
291 switch( codec_context->codec_type )
293 case CODEC_TYPE_VIDEO:
294 // Use first video stream
295 if ( self->video_index < 0 )
296 self->video_index = i;
297 mlt_properties_set( meta_media, key, "video" );
298 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
299 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
300 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
301 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
302 ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
303 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
305 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) );
308 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
309 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
310 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
312 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
313 mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den /
314 ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) );
315 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
316 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
317 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
318 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
319 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
320 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
321 switch ( codec_context->colorspace )
323 case AVCOL_SPC_SMPTE240M:
324 mlt_properties_set_int( meta_media, key, 240 );
326 case AVCOL_SPC_BT470BG:
327 case AVCOL_SPC_SMPTE170M:
328 mlt_properties_set_int( meta_media, key, 601 );
330 case AVCOL_SPC_BT709:
331 mlt_properties_set_int( meta_media, key, 709 );
334 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
335 mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
340 case CODEC_TYPE_AUDIO:
341 // Use first audio stream
342 if ( self->audio_index < 0 )
343 self->audio_index = i;
344 mlt_properties_set( meta_media, key, "audio" );
345 #if LIBAVCODEC_VERSION_MAJOR > 52
346 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
347 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
348 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
349 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
350 mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) );
352 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
353 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
354 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
355 mlt_properties_set_int( meta_media, key, codec_context->channels );
360 // snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
361 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
362 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
363 mlt_properties_set( meta_media, key, codec->name );
364 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0))
365 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
366 mlt_properties_set( meta_media, key, codec->long_name );
368 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
369 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
370 // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
371 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
372 // snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
373 // mlt_properties_set_int( meta_media, key, codec_context->profile );
374 // snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
375 // mlt_properties_set_int( meta_media, key, codec_context->level );
378 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
379 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
381 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
383 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
384 mlt_properties_set( meta_media, key, tag->value );
389 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
390 while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
392 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
394 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
395 mlt_properties_set( meta_media, key, tag->value );
399 if ( context->title && strcmp( context->title, "" ) )
400 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
401 if ( context->author && strcmp( context->author, "" ) )
402 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
403 if ( context->copyright && strcmp( context->copyright, "" ) )
404 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
405 if ( context->comment )
406 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
407 if ( context->album )
408 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
410 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
411 if ( context->track )
412 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
418 static inline int dv_is_pal( AVPacket *pkt )
420 return pkt->data[3] & 0x80;
423 static int dv_is_wide( AVPacket *pkt )
425 int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
427 for ( ; i < pkt->size; i += 5 /* packet size */ )
429 if ( pkt->data[ i ] == 0x61 )
431 uint8_t x = pkt->data[ i + 2 ] & 0x7;
432 return ( x == 2 ) || ( x == 7 );
438 static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
440 double aspect_ratio = 1.0;
442 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
446 if ( dv_is_pal( pkt ) )
448 if ( dv_is_wide( pkt ) )
450 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
451 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
455 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
456 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
461 if ( dv_is_wide( pkt ) )
463 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
464 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
468 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
469 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
476 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
477 stream->sample_aspect_ratio;
479 codec_context->sample_aspect_ratio;
481 // Override FFmpeg's notion of DV aspect ratios, which are
482 // based upon a width of 704. Since we do not have a normaliser
483 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
484 // we just coerce the values to facilitate a passive behaviour through
485 // the rescale normaliser when using equivalent producers and consumers.
486 // = display_aspect / (width * height)
487 if ( ar.num == 10 && ar.den == 11 )
490 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
491 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
493 else if ( ar.num == 59 && ar.den == 54 )
496 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
497 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
499 else if ( ar.num == 40 && ar.den == 33 )
502 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
503 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
505 else if ( ar.num == 118 && ar.den == 81 )
508 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
509 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
515 AVRational codec_sar = codec_context->sample_aspect_ratio;
516 AVRational stream_sar =
517 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
518 stream->sample_aspect_ratio;
522 if ( codec_sar.num > 0 )
524 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num );
525 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den );
527 else if ( stream_sar.num > 0 )
529 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num );
530 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den );
534 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 );
535 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 );
538 AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) };
539 aspect_ratio = av_q2d( ar );
540 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
545 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params )
547 if ( !URL ) return NULL;
549 const char *result = URL;
550 char *protocol = strdup( URL );
551 char *url = strchr( protocol, ':' );
553 // Only if there is not a protocol specification that avformat can handle
554 #if LIBAVFORMAT_VERSION_MAJOR > 52
555 if ( url && avio_check( URL, 0 ) < 0 )
557 if ( url && !url_exist( URL ) )
560 // Truncate protocol string
562 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url + 1 );
565 *format = av_find_input_format( protocol );
567 // Eat the format designator
572 // These are required by video4linux2 (defaults)
573 params->width = profile->width;
574 params->height = profile->height;
575 params->time_base= (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
576 params->channels = 2;
577 params->sample_rate = 48000;
580 url = strchr( url, '?' );
584 char *name = strdup( ++url );
585 char *value = strchr( name, ':' );
590 char *t = strchr( value, '&' );
593 if ( !strcmp( name, "frame_rate" ) )
594 params->time_base.den = atoi( value );
595 else if ( !strcmp( name, "frame_rate_base" ) )
596 params->time_base.num = atoi( value );
597 else if ( !strcmp( name, "sample_rate" ) )
598 params->sample_rate = atoi( value );
599 else if ( !strcmp( name, "channel" ) )
600 params->channel = atoi( value );
601 else if ( !strcmp( name, "channels" ) )
602 params->channels = atoi( value );
603 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
604 else if ( !strcmp( name, "pix_fmt" ) )
605 params->pix_fmt = av_get_pix_fmt( value );
607 else if ( !strcmp( name, "width" ) )
608 params->width = atoi( value );
609 else if ( !strcmp( name, "height" ) )
610 params->height = atoi( value );
611 else if ( !strcmp( name, "standard" ) )
612 params->standard = strdup( value );
615 url = strchr( url, '&' );
619 result = strdup( result );
624 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
628 // Get the properties
629 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
631 AVFormatContext *format = self->video_format;
633 // We will treat everything with the producer fps.
634 // TODO: make this more flexible.
635 double fps = mlt_profile_fps( profile );
638 if ( !mlt_properties_get_int( properties, "_length_computed" ) )
640 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
641 // when producer_open is called after initial call when restoring or reseting the producer.
642 if ( format->duration != AV_NOPTS_VALUE )
644 // This isn't going to be accurate for all formats
645 mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
646 mlt_properties_set_position( properties, "out", frames - 1 );
647 mlt_properties_set_position( properties, "length", frames );
648 mlt_properties_set_int( properties, "_length_computed", 1 );
652 // Set live sources to run forever
653 mlt_properties_set_position( properties, "length", INT_MAX );
654 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
655 mlt_properties_set( properties, "eof", "loop" );
656 mlt_properties_set_int( properties, "_length_computed", 1 );
660 if ( format->start_time != AV_NOPTS_VALUE )
661 self->start_time = format->start_time;
663 // Check if we're seekable
664 // avdevices are typically AVFMT_NOFILE and not seekable
665 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
668 // protocols can indicate if they support seeking
669 #if LIBAVFORMAT_VERSION_MAJOR > 52
670 self->seekable = format->pb->seekable;
672 URLContext *uc = url_fileno( format->pb );
674 self->seekable = !uc->is_streamed;
677 if ( self->seekable )
679 // Do a more rigourous test of seekable on a disposable context
680 self->seekable = av_seek_frame( format, -1, self->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
681 mlt_properties_set_int( properties, "seekable", self->seekable );
682 self->dummy_context = format;
683 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
684 format = self->video_format;
685 av_find_stream_info( format );
688 // Fetch the width, height and aspect ratio
689 if ( self->video_index != -1 )
691 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
692 mlt_properties_set_int( properties, "width", codec_context->width );
693 mlt_properties_set_int( properties, "height", codec_context->height );
695 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
697 // Fetch the first frame of DV so we can read it directly
702 ret = av_read_frame( format, &pkt );
703 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
705 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, &pkt );
712 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, NULL );
716 // Verify that we can convert this to YUV 4:2:2
717 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
718 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
719 codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
721 sws_freeContext( context );
732 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL )
734 // Return an error code (0 == no error)
736 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
739 pthread_mutex_init( &self->audio_mutex, NULL );
740 pthread_mutex_init( &self->video_mutex, NULL );
741 pthread_mutex_init( &self->packets_mutex, NULL );
742 mlt_events_block( properties, self->parent );
743 pthread_mutex_lock( &self->audio_mutex );
744 pthread_mutex_lock( &self->video_mutex );
747 AVInputFormat *format = NULL;
748 AVFormatParameters params;
749 memset( ¶ms, 0, sizeof(params) );
750 char *filename = parse_url( profile, URL, &format, ¶ms );
752 // Now attempt to open the file or device with filename
753 error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0;
755 // If the URL is a network stream URL, then we probably need to open with full URL
756 error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0;
758 // Set MLT properties onto video AVFormatContext
759 if ( !error && self->video_format )
761 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
762 #if LIBAVFORMAT_VERSION_MAJOR > 52
763 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
764 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
768 // Cleanup AVFormatParameters
769 if ( params.standard )
770 free( (void*) params.standard );
772 // If successful, then try to get additional info
773 if ( !error && self->video_format )
775 // Get the stream info
776 error = av_find_stream_info( self->video_format ) < 0;
778 // Continue if no error
779 if ( !error && self->video_format )
781 // Find default audio and video streams
782 find_default_streams( self );
783 error = get_basic_info( self, profile, filename );
785 // Initialize position info
786 self->first_pts = -1;
787 self->last_position = POSITION_INITIAL;
789 // We're going to cheat here - for seekable A/V files, we will have separate contexts
790 // to support independent seeking of audio from video.
791 // TODO: Is this really necessary?
792 if ( self->audio_index != -1 && self->video_index != -1 )
794 if ( self->seekable )
796 // And open again for our audio context
797 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
798 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
799 #if LIBAVFORMAT_VERSION_MAJOR > 52
800 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
801 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
803 av_find_stream_info( self->audio_format );
807 self->audio_format = self->video_format;
810 else if ( self->audio_index != -1 )
812 // We only have an audio context
813 self->audio_format = self->video_format;
814 self->video_format = NULL;
816 else if ( self->video_index == -1 )
818 // Something has gone wrong
821 if ( self->audio_format && !self->audio_streams )
822 get_audio_streams_info( self );
829 self->apackets = mlt_deque_init();
830 self->vpackets = mlt_deque_init();
833 if ( self->dummy_context )
835 av_close_input_file( self->dummy_context );
836 self->dummy_context = NULL;
839 // Unlock the service
840 pthread_mutex_unlock( &self->audio_mutex );
841 pthread_mutex_unlock( &self->video_mutex );
842 mlt_events_unblock( properties, self->parent );
847 static void reopen_video( producer_avformat self, mlt_producer producer )
849 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
850 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
851 pthread_mutex_lock( &self->audio_mutex );
853 if ( self->video_codec )
856 avcodec_close( self->video_codec );
859 self->video_codec = NULL;
860 if ( self->dummy_context )
861 av_close_input_file( self->dummy_context );
862 self->dummy_context = NULL;
863 if ( self->video_format )
864 av_close_input_file( self->video_format );
865 self->video_format = NULL;
867 int audio_index = self->audio_index;
868 int video_index = self->video_index;
870 pthread_mutex_unlock( &self->audio_mutex );
871 pthread_mutex_unlock( &self->video_mutex );
872 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
873 mlt_properties_get( properties, "resource" ) );
874 pthread_mutex_lock( &self->video_mutex );
875 pthread_mutex_lock( &self->audio_mutex );
877 self->audio_index = audio_index;
878 if ( self->video_format && video_index > -1 )
880 self->video_index = video_index;
881 video_codec_init( self, video_index, properties );
884 pthread_mutex_unlock( &self->audio_mutex );
885 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
888 static int seek_video( producer_avformat self, mlt_position position,
889 int req_position, int must_decode, int use_new_seek, int *ignore )
891 mlt_producer producer = self->parent;
894 if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
896 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
898 // Fetch the video format context
899 AVFormatContext *context = self->video_format;
901 // Get the video stream
902 AVStream *stream = context->streams[ self->video_index ];
905 AVCodecContext *codec_context = stream->codec;
907 // We may want to use the source fps if available
908 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
909 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
911 if ( self->av_frame && position + 1 == self->video_expected )
913 // We're paused - use last image
916 else if ( !self->seekable && position > self->video_expected && ( position - self->video_expected ) < 250 )
918 // Fast forward - seeking is inefficient for small distances - just ignore following frames
919 *ignore = ( int )( ( position - self->video_expected ) / mlt_producer_get_fps( producer ) * source_fps );
920 codec_context->skip_loop_filter = AVDISCARD_NONREF;
922 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
924 if ( use_new_seek && self->last_position == POSITION_INITIAL )
926 // find first key frame
931 while ( ret >= 0 && toscan-- > 0 )
933 ret = av_read_frame( context, &pkt );
934 if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == self->video_index )
936 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n", pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
937 self->first_pts = pkt.pts;
940 av_free_packet( &pkt );
943 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
946 // Calculate the timestamp for the requested frame
950 timestamp = ( req_position - 0.1 / source_fps ) /
951 ( av_q2d( stream->time_base ) * source_fps );
952 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "pos %d pts %"PRId64" ", req_position, timestamp );
953 if ( self->first_pts > 0 )
954 timestamp += self->first_pts;
955 else if ( context->start_time != AV_NOPTS_VALUE )
956 timestamp += context->start_time;
960 timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
961 if ( context->start_time != AV_NOPTS_VALUE )
962 timestamp += context->start_time;
965 timestamp -= AV_TIME_BASE;
968 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %d\n",
969 timestamp, position, self->video_expected, self->last_position );
971 // Seek to the timestamp
974 codec_context->skip_loop_filter = AVDISCARD_NONREF;
975 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
977 else if ( req_position > 0 || self->last_position <= 0 )
979 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
983 // Re-open video stream when rewinding to beginning from somewhere else.
984 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
985 // not reliably seeking to the first frame across formats.
986 reopen_video( self, producer );
989 // Remove the cached info relating to the previous position
990 self->current_position = POSITION_INVALID;
991 self->last_position = POSITION_INVALID;
992 av_freep( &self->av_frame );
996 // flush any pictures still in decode buffer
997 avcodec_flush_buffers( codec_context );
1004 /** Convert a frame position to a time code.
1007 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1009 return ( double )position / mlt_producer_get_fps( producer );
1012 // Collect information about all audio streams
1014 static void get_audio_streams_info( producer_avformat self )
1016 // Fetch the audio format context
1017 AVFormatContext *context = self->audio_format;
1021 i < context->nb_streams;
1024 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
1026 AVCodecContext *codec_context = context->streams[i]->codec;
1027 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1029 // If we don't have a codec and we can't initialise it, we can't do much more...
1031 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1033 self->audio_streams++;
1034 self->audio_max_stream = i;
1035 self->total_channels += codec_context->channels;
1036 if ( codec_context->channels > self->max_channel )
1037 self->max_channel = codec_context->channels;
1038 if ( codec_context->sample_rate > self->max_frequency )
1039 self->max_frequency = codec_context->sample_rate;
1040 avcodec_close( codec_context );
1045 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1046 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1048 // Other audio-specific initializations
1049 self->resample_factor = 1.0;
1052 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
1054 #if defined(SWSCALE) && (LIBSWSCALE_VERSION_INT >= ((0<<16)+(7<<8)+2))
1056 const int *new_coefficients;
1058 int brightness, contrast, saturation;
1060 if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
1061 &brightness, &contrast, &saturation ) != -1 )
1063 // Don't change these from defaults unless explicitly told to.
1064 if ( use_full_range >= 0 )
1065 full_range = use_full_range;
1066 switch ( colorspace )
1072 new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1075 new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1078 new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1081 new_coefficients = coefficients;
1084 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
1085 brightness, contrast, saturation );
1090 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt,
1091 mlt_image_format *format, int width, int height, int colorspace )
1094 int full_range = -1;
1095 int flags = SWS_BILINEAR | SWS_ACCURATE_RND;
1098 flags |= SWS_CPU_CAPS_MMX;
1101 flags |= SWS_CPU_CAPS_MMX2;
1104 if ( pix_fmt == PIX_FMT_RGB32 )
1106 *format = mlt_image_rgb24a;
1107 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1108 width, height, PIX_FMT_RGBA, flags, NULL, NULL, NULL);
1110 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1111 set_luma_transfer( context, colorspace, full_range );
1112 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1113 output.data, output.linesize);
1114 sws_freeContext( context );
1116 else if ( *format == mlt_image_yuv420p )
1118 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1119 width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1121 output.data[0] = buffer;
1122 output.data[1] = buffer + width * height;
1123 output.data[2] = buffer + ( 5 * width * height ) / 4;
1124 output.linesize[0] = width;
1125 output.linesize[1] = width >> 1;
1126 output.linesize[2] = width >> 1;
1127 set_luma_transfer( context, colorspace, full_range );
1128 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1129 output.data, output.linesize);
1130 sws_freeContext( context );
1132 else if ( *format == mlt_image_rgb24 )
1134 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1135 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1137 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1138 set_luma_transfer( context, colorspace, full_range );
1139 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1140 output.data, output.linesize);
1141 sws_freeContext( context );
1143 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1145 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1146 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1148 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1149 set_luma_transfer( context, colorspace, full_range );
1150 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1151 output.data, output.linesize);
1152 sws_freeContext( context );
1156 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1157 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1159 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1160 set_luma_transfer( context, colorspace, full_range );
1161 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1162 output.data, output.linesize);
1163 sws_freeContext( context );
1166 if ( *format == mlt_image_yuv420p )
1169 pict.data[0] = buffer;
1170 pict.data[1] = buffer + width * height;
1171 pict.data[2] = buffer + ( 5 * width * height ) / 4;
1172 pict.linesize[0] = width;
1173 pict.linesize[1] = width >> 1;
1174 pict.linesize[2] = width >> 1;
1175 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
1177 else if ( *format == mlt_image_rgb24 )
1180 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1181 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
1183 else if ( format == mlt_image_rgb24a || format == mlt_image_opengl )
1186 avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height );
1187 img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height );
1192 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1193 img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height );
1198 /** Allocate the image buffer and set it on the frame.
1201 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1205 if ( codec_context->width == 0 || codec_context->height == 0 )
1208 *width = codec_context->width;
1209 *height = codec_context->height;
1211 if ( codec_context->pix_fmt == PIX_FMT_RGB32 )
1212 size = *width * ( *height + 1 ) * 4;
1214 size = mlt_image_format_size( *format, *width, *height, NULL );
1216 // Construct the output image
1217 *buffer = mlt_pool_alloc( size );
1219 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1226 /** Get an image from a frame.
1229 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1232 producer_avformat self = mlt_frame_pop_service( frame );
1233 mlt_producer producer = self->parent;
1235 // Get the properties from the frame
1236 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1238 // Obtain the frame number of this frame
1239 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
1241 // Get the producer properties
1242 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1244 pthread_mutex_lock( &self->video_mutex );
1246 // Fetch the video format context
1247 AVFormatContext *context = self->video_format;
1249 // Get the video stream
1250 AVStream *stream = context->streams[ self->video_index ];
1252 // Get codec context
1253 AVCodecContext *codec_context = stream->codec;
1255 // Get the image cache
1256 if ( ! self->image_cache && ! mlt_properties_get_int( properties, "noimagecache" ) )
1257 self->image_cache = mlt_cache_init();
1258 if ( self->image_cache )
1260 mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position );
1261 uint8_t *original = mlt_cache_item_data( item, (int*) format );
1264 // Set the resolution
1265 *width = codec_context->width;
1266 *height = codec_context->height;
1268 // Workaround 1088 encodings missing cropping info.
1269 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1273 int size = mlt_image_format_size( *format, *width, *height, NULL );
1276 *buffer = mlt_pool_alloc( size );
1277 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1278 memcpy( *buffer, original, size );
1279 mlt_cache_item_close( item );
1284 mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
1285 mlt_frame_set_image( frame, *buffer, size, NULL );
1287 self->got_picture = 1;
1289 goto exit_get_image;
1298 // Special case ffwd handling
1301 // We may want to use the source fps if available
1302 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1303 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1305 // This is the physical frame position in the source
1306 int req_position = ( int )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1308 // Determines if we have to decode all frames in a sequence
1309 // Temporary hack to improve intra frame only
1310 int must_decode = strcmp( codec_context->codec->name, "dnxhd" ) &&
1311 strcmp( codec_context->codec->name, "dvvideo" ) &&
1312 strcmp( codec_context->codec->name, "huffyuv" ) &&
1313 strcmp( codec_context->codec->name, "mjpeg" ) &&
1314 strcmp( codec_context->codec->name, "rawvideo" );
1316 // Turn on usage of new seek API and PTS for seeking
1317 int use_new_seek = self->seekable &&
1318 codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" );
1319 if ( mlt_properties_get( properties, "new_seek" ) )
1320 use_new_seek = mlt_properties_get_int( properties, "new_seek" );
1322 // Seek if necessary
1323 int paused = seek_video( self, position, req_position, must_decode, use_new_seek, &ignore );
1325 // Seek might have reopened the file
1326 context = self->video_format;
1327 stream = context->streams[ self->video_index ];
1328 codec_context = stream->codec;
1330 // Duplicate the last image if necessary
1331 if ( self->av_frame && self->av_frame->linesize[0] && self->got_picture
1333 || self->current_position == req_position
1334 || ( !use_new_seek && self->current_position > req_position ) ) )
1337 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1339 // Workaround 1088 encodings missing cropping info.
1340 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1343 if ( self->vdpau && self->vdpau->buffer )
1346 picture.data[0] = self->vdpau->buffer;
1347 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1348 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1349 picture.linesize[0] = codec_context->width;
1350 picture.linesize[1] = codec_context->width / 2;
1351 picture.linesize[2] = codec_context->width / 2;
1352 convert_image( (AVFrame*) &picture, *buffer,
1353 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1357 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1358 format, *width, *height, self->colorspace );
1361 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1366 int int_position = 0;
1367 int decode_errors = 0;
1368 int got_picture = 0;
1370 av_init_packet( &pkt );
1372 // Construct an AVFrame for YUV422 conversion
1373 if ( !self->av_frame )
1374 self->av_frame = avcodec_alloc_frame( );
1376 while( ret >= 0 && !got_picture )
1379 pthread_mutex_lock( &self->packets_mutex );
1380 if ( mlt_deque_count( self->vpackets ) )
1382 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1388 ret = av_read_frame( context, &pkt );
1389 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->audio_index )
1391 if ( !av_dup_packet( &pkt ) )
1393 AVPacket *tmp = malloc( sizeof(AVPacket) );
1395 mlt_deque_push_back( self->apackets, tmp );
1399 pthread_mutex_unlock( &self->packets_mutex );
1401 // We only deal with video from the selected video_index
1402 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
1404 // Determine time code of the packet
1407 int64_t pts = pkt.pts;
1408 if ( self->first_pts > 0 )
1409 pts -= self->first_pts;
1410 else if ( context->start_time != AV_NOPTS_VALUE )
1411 pts -= context->start_time;
1412 int_position = ( int64_t )( av_q2d( stream->time_base ) * pts * source_fps + 0.1 ) & 0x7fffffff;
1413 if ( pkt.pts == AV_NOPTS_VALUE )
1415 self->invalid_pts_counter++;
1416 if ( self->invalid_pts_counter > 20 )
1418 mlt_log_panic( MLT_PRODUCER_SERVICE(producer), "\ainvalid PTS; DISABLING NEW_SEEK!\n" );
1419 mlt_properties_set_int( properties, "new_seek", 0 );
1420 int_position = req_position;
1426 self->invalid_pts_counter = 0;
1428 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.pts %"PRId64" req_pos %d cur_pos %d pkt_pos %d\n",
1429 pkt.pts, req_position, self->current_position, int_position );
1433 if ( pkt.dts != AV_NOPTS_VALUE )
1435 double delay = mlt_properties_get_double( properties, "video_delay" );
1436 int_position = ( int64_t )( ( av_q2d( stream->time_base ) * pkt.dts + delay ) * source_fps + 0.5 ) & 0x7fffffff;
1437 if ( context->start_time != AV_NOPTS_VALUE )
1438 int_position -= ( int64_t )( context->start_time * source_fps / AV_TIME_BASE + 0.5 ) & 0x7fffffff;
1439 if ( int_position == self->last_position )
1440 int_position = self->last_position + 1;
1442 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %"PRId64" req_pos %d cur_pos %d pkt_pos %d\n",
1443 pkt.dts, req_position, self->current_position, int_position );
1444 // Make a dumb assumption on streams that contain wild timestamps
1445 if ( abs( req_position - int_position ) > 999 )
1447 int_position = req_position;
1448 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" );
1451 self->last_position = int_position;
1454 if ( must_decode || int_position >= req_position )
1457 if ( g_vdpau && self->vdpau )
1459 if ( g_vdpau->producer != self )
1461 vdpau_decoder_close();
1462 vdpau_decoder_init( self );
1465 self->vdpau->is_decoded = 0;
1468 codec_context->reordered_opaque = pkt.pts;
1469 if ( int_position >= req_position )
1470 codec_context->skip_loop_filter = AVDISCARD_NONE;
1471 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1472 ret = avcodec_decode_video2( codec_context, self->av_frame, &got_picture, &pkt );
1474 ret = avcodec_decode_video( codec_context, self->av_frame, &got_picture, pkt.data, pkt.size );
1476 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1479 if ( ++decode_errors <= 10 )
1492 // Determine time code of the packet
1493 int64_t pts = self->av_frame->reordered_opaque;
1494 if ( self->first_pts > 0 )
1495 pts -= self->first_pts;
1496 else if ( context->start_time != AV_NOPTS_VALUE )
1497 pts -= context->start_time;
1498 int_position = ( int64_t )( av_q2d( stream->time_base) * pts * source_fps + 0.1 ) & 0x7fffffff;
1499 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "got frame %d, key %d\n", int_position, self->av_frame->key_frame );
1502 if ( int_position < req_position )
1507 else if ( int_position >= req_position )
1510 codec_context->skip_loop_filter = AVDISCARD_NONE;
1512 else if ( ignore -- )
1517 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY );
1520 // Now handle the picture if we have one
1523 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1525 // Workaround 1088 encodings missing cropping info.
1526 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1531 if ( self->vdpau->is_decoded )
1533 struct vdpau_render_state *render = (struct vdpau_render_state*) self->av_frame->data[0];
1535 uint32_t pitches[3];
1536 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1538 if ( !self->vdpau->buffer )
1539 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1540 self->av_frame->data[0] = planes[0] = self->vdpau->buffer;
1541 self->av_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1542 self->av_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1543 self->av_frame->linesize[0] = pitches[0] = codec_context->width;
1544 self->av_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1545 self->av_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1547 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1548 if ( status == VDP_STATUS_OK )
1550 convert_image( self->av_frame, *buffer, PIX_FMT_YUV420P,
1551 format, *width, *height, self->colorspace );
1555 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1556 image_size = self->vdpau->is_decoded = 0;
1561 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1562 image_size = got_picture = 0;
1567 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1568 format, *width, *height, self->colorspace );
1569 self->top_field_first |= self->av_frame->top_field_first;
1570 self->current_position = int_position;
1571 self->got_picture = 1;
1578 if ( self->seekable || pkt.stream_index != self->audio_index )
1579 av_free_packet( &pkt );
1583 if ( self->got_picture && image_size > 0 && self->image_cache )
1585 // Copy buffer to image cache
1586 uint8_t *image = mlt_pool_alloc( image_size );
1587 memcpy( image, *buffer, image_size );
1588 mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release );
1590 // Try to duplicate last image if there was a decoding failure
1591 else if ( !image_size && self->av_frame && self->av_frame->linesize[0] )
1594 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1596 // Workaround 1088 encodings missing cropping info.
1597 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1600 if ( self->vdpau && self->vdpau->buffer )
1603 picture.data[0] = self->vdpau->buffer;
1604 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1605 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1606 picture.linesize[0] = codec_context->width;
1607 picture.linesize[1] = codec_context->width / 2;
1608 picture.linesize[2] = codec_context->width / 2;
1609 convert_image( (AVFrame*) &picture, *buffer,
1610 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1614 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1615 format, *width, *height, self->colorspace );
1616 self->got_picture = 1;
1619 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1622 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1623 self->video_expected = position + 1;
1627 pthread_mutex_unlock( &self->video_mutex );
1629 // Set the progressive flag
1630 if ( mlt_properties_get( properties, "force_progressive" ) )
1631 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1632 else if ( self->av_frame )
1633 mlt_properties_set_int( frame_properties, "progressive", !self->av_frame->interlaced_frame );
1635 // Set the field order property for this frame
1636 if ( mlt_properties_get( properties, "force_tff" ) )
1637 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1639 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1641 // Set immutable properties of the selected track's (or overridden) source attributes.
1642 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1643 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1644 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1645 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1647 return !self->got_picture;
1650 /** Process properties as AVOptions and apply to AV context obj
1653 static void apply_properties( void *obj, mlt_properties properties, int flags )
1656 int count = mlt_properties_count( properties );
1657 for ( i = 0; i < count; i++ )
1659 const char *opt_name = mlt_properties_get_name( properties, i );
1660 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1661 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1664 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1665 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1666 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1667 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1669 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1675 /** Initialize the video codec context.
1678 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1680 // Initialise the codec if necessary
1681 if ( !self->video_codec )
1683 // Get the video stream
1684 AVStream *stream = self->video_format->streams[ index ];
1686 // Get codec context
1687 AVCodecContext *codec_context = stream->codec;
1690 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1692 if ( codec_context->codec_id == CODEC_ID_H264 )
1694 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1696 if ( vdpau_init( self ) )
1698 self->video_codec = codec_context;
1699 if ( !vdpau_decoder_init( self ) )
1700 vdpau_decoder_close();
1704 codec = avcodec_find_decoder( codec_context->codec_id );
1708 // Initialise multi-threading
1709 int thread_count = mlt_properties_get_int( properties, "threads" );
1710 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1711 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1712 if ( thread_count > 1 )
1713 codec_context->thread_count = thread_count;
1715 // If we don't have a codec and we can't initialise it, we can't do much more...
1717 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1719 // Now store the codec with its destructor
1720 self->video_codec = codec_context;
1724 // Remember that we can't use this later
1725 self->video_index = -1;
1731 // Process properties as AVOptions
1732 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1733 #if LIBAVCODEC_VERSION_MAJOR > 52
1734 if ( codec->priv_class && codec_context->priv_data )
1735 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1738 // Reset some image properties
1739 mlt_properties_set_int( properties, "width", self->video_codec->width );
1740 mlt_properties_set_int( properties, "height", self->video_codec->height );
1741 // For DV, we'll just use the saved aspect ratio
1742 if ( codec_context->codec_id != CODEC_ID_DVVIDEO )
1743 get_aspect_ratio( properties, stream, self->video_codec, NULL );
1745 // Determine the fps first from the codec
1746 double source_fps = (double) self->video_codec->time_base.den /
1747 ( self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1749 if ( mlt_properties_get( properties, "force_fps" ) )
1751 source_fps = mlt_properties_get_double( properties, "force_fps" );
1752 stream->time_base = av_d2q( source_fps, 1024 );
1753 mlt_properties_set_int( properties, "meta.media.frame_rate_num", stream->time_base.num );
1754 mlt_properties_set_int( properties, "meta.media.frame_rate_den", stream->time_base.den );
1758 // If the muxer reports a frame rate different than the codec
1759 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1760 double muxer_fps = av_q2d( stream->avg_frame_rate );
1761 if ( isnan( muxer_fps ) || muxer_fps == 0 )
1762 muxer_fps = av_q2d( stream->r_frame_rate );
1764 double muxer_fps = av_q2d( stream->r_frame_rate );
1766 // Choose the lesser - the wrong tends to be off by some multiple of 10
1767 source_fps = FFMIN( source_fps, muxer_fps );
1768 if ( source_fps >= 1.0 && ( source_fps < muxer_fps || isnan( muxer_fps ) ) )
1770 mlt_properties_set_int( properties, "meta.media.frame_rate_num", self->video_codec->time_base.den );
1771 mlt_properties_set_int( properties, "meta.media.frame_rate_den", self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1773 else if ( muxer_fps > 0 )
1775 AVRational frame_rate = stream->r_frame_rate;
1776 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1777 // avg_frame_rate gives some approximate value that does not well match the media.
1778 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1779 // results in some very choppy output, but some value slightly different works
1781 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1782 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1783 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1785 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1786 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1790 source_fps = mlt_producer_get_fps( self->parent );
1791 AVRational frame_rate = av_d2q( source_fps, 255 );
1792 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1793 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1797 // source_fps is deprecated in favor of meta.media.frame_rate_num and .frame_rate_den
1798 if ( source_fps > 0 )
1799 mlt_properties_set_double( properties, "source_fps", source_fps );
1801 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( self->parent ) );
1803 // Set the YUV colorspace from override or detect
1804 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1805 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1806 if ( ! self->colorspace )
1808 switch ( self->video_codec->colorspace )
1810 case AVCOL_SPC_SMPTE240M:
1811 self->colorspace = 240;
1813 case AVCOL_SPC_BT470BG:
1814 case AVCOL_SPC_SMPTE170M:
1815 self->colorspace = 601;
1817 case AVCOL_SPC_BT709:
1818 self->colorspace = 709;
1821 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1822 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1827 // Let apps get chosen colorspace
1828 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1830 return self->video_codec && self->video_index > -1;
1833 /** Set up video handling.
1836 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1839 mlt_producer producer = self->parent;
1841 // Get the properties
1842 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1844 // Fetch the video format context
1845 AVFormatContext *context = self->video_format;
1847 // Get the video_index
1848 int index = mlt_properties_get_int( properties, "video_index" );
1850 // Reopen the file if necessary
1851 if ( !context && index > -1 )
1853 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1854 mlt_properties_get( properties, "resource" ) );
1855 context = self->video_format;
1858 // Exception handling for video_index
1859 if ( context && index >= (int) context->nb_streams )
1861 // Get the last video stream
1862 for ( index = context->nb_streams - 1;
1863 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1865 mlt_properties_set_int( properties, "video_index", index );
1867 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1869 // Invalidate the video stream
1871 mlt_properties_set_int( properties, "video_index", index );
1874 // Update the video properties if the index changed
1875 if ( index != self->video_index )
1877 // Reset the video properties if the index changed
1878 self->video_index = index;
1879 if ( self->video_codec )
1882 avcodec_close( self->video_codec );
1885 self->video_codec = NULL;
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 // real_width and real_height are deprecated in favor of meta.media.width and .height
1903 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1904 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1905 mlt_properties_set_int( frame_properties, "real_width", self->video_codec->width );
1906 mlt_properties_set_int( frame_properties, "real_height", self->video_codec->height );
1907 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1908 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
1910 // Workaround 1088 encodings missing cropping info.
1911 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1913 mlt_properties_set_int( properties, "meta.media.height", 1080 );
1914 mlt_properties_set_int( frame_properties, "real_height", 1080 );
1917 // Add our image operation
1918 mlt_frame_push_service( frame, self );
1919 mlt_frame_push_get_image( frame, producer_get_image );
1923 // If something failed, use test card image
1924 mlt_properties_set_int( frame_properties, "test_image", 1 );
1928 static int seek_audio( producer_avformat self, mlt_position position, double timecode, int *ignore )
1932 // Seek if necessary
1933 if ( self->seekable && position != self->audio_expected )
1935 if ( position + 1 == self->audio_expected )
1937 // We're paused - silence required
1940 else if ( !self->seekable && position > self->audio_expected && ( position - self->audio_expected ) < 250 )
1942 // Fast forward - seeking is inefficient for small distances - just ignore following frames
1943 *ignore = position - self->audio_expected;
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 LIBAVCODEC_VERSION_MAJOR > 52
1970 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
1972 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1976 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
1978 // Fetch the audio_format
1979 AVFormatContext *context = self->audio_format;
1981 // Get the current stream index
1982 int index = pkt.stream_index;
1984 // Get codec context
1985 AVCodecContext *codec_context = self->audio_codec[ index ];
1987 // Obtain the resample context if it exists (not always needed)
1988 ReSampleContext *resample = self->audio_resample[ index ];
1990 // Obtain the audio buffers
1991 uint8_t *audio_buffer = self->audio_buffer[ index ];
1992 uint8_t *decode_buffer = self->decode_buffer[ index ];
1994 int audio_used = self->audio_used[ index ];
1995 uint8_t *ptr = pkt.data;
1999 while ( ptr && ret >= 0 && len > 0 )
2001 int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context );
2002 int data_size = self->audio_buffer_size[ index ];
2005 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
2006 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2007 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
2008 ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
2010 ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len );
2014 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2018 pkt.size = len -= ret;
2019 pkt.data = ptr += ret;
2021 // If decoded successfully
2022 if ( data_size > 0 )
2024 // Figure out how many samples will be needed after resampling
2025 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2026 int samples_needed = self->resample_factor * convert_samples;
2028 // Resize audio buffer to prevent overflow
2029 if ( ( audio_used + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2031 self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample;
2032 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2036 // Copy to audio buffer while resampling
2037 uint8_t *source = decode_buffer;
2038 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
2039 audio_used += audio_resample( resample, (short*) dest, (short*) source, convert_samples );
2043 // Straight copy to audio buffer
2044 memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size );
2045 audio_used += convert_samples;
2049 while ( *ignore && audio_used > samples )
2052 audio_used -= samples;
2053 memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ],
2054 audio_used * sizeof_sample );
2059 // If we're behind, ignore this packet
2062 double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt.pts;
2063 int req_position = ( int )( timecode * fps + 0.5 );
2064 int int_position = ( int64_t )( current_pts * fps + 0.5 ) & 0x7fffffff;
2065 if ( context->start_time != AV_NOPTS_VALUE )
2066 int_position -= ( int64_t )( fps * context->start_time / AV_TIME_BASE + 0.5 ) & 0x7fffffff;
2070 if ( int_position < req_position )
2071 // We are behind, so skip some
2074 // We use nb_streams in this test because the tolerance is dependent
2075 // on the interleaving of all streams esp. when there is more than
2076 // one audio stream.
2077 else if ( int_position > req_position + context->nb_streams )
2078 // We are ahead, so seek backwards some more
2079 seek_audio( self, req_position, timecode - 1.0, ignore );
2083 self->audio_used[ index ] = audio_used;
2088 /** Get the audio from a frame.
2090 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2093 producer_avformat self = mlt_frame_pop_audio( frame );
2095 pthread_mutex_lock( &self->audio_mutex );
2097 // Obtain the frame number of this frame
2098 mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" );
2100 // Calculate the real time code
2101 double real_timecode = producer_time_of_frame( self->parent, position );
2103 // Get the producer fps
2104 double fps = mlt_producer_get_fps( self->parent );
2106 // Number of frames to ignore (for ffwd)
2107 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2109 // Flag for paused (silence)
2110 int paused = seek_audio( self, position, real_timecode, &ignore[0] );
2112 // Initialize ignore for all streams from the seek return value
2113 int i = MAX_AUDIO_STREAMS;
2115 ignore[i] = ignore[0];
2117 // Fetch the audio_format
2118 AVFormatContext *context = self->audio_format;
2120 int sizeof_sample = sizeof( int16_t );
2122 // Determine the tracks to use
2123 int index = self->audio_index;
2124 int index_max = self->audio_index + 1;
2125 if ( self->audio_index == INT_MAX )
2128 index_max = context->nb_streams;
2129 *channels = self->total_channels;
2130 *samples = *samples * FFMAX( self->max_frequency, *frequency ) / *frequency;
2131 *frequency = FFMAX( self->max_frequency, *frequency );
2134 // Initialize the resamplers and buffers
2135 for ( ; index < index_max; index++ )
2137 // Get codec context
2138 AVCodecContext *codec_context = self->audio_codec[ index ];
2140 if ( codec_context && !self->audio_buffer[ index ] )
2142 // Check for resample and create if necessary
2143 if ( codec_context->channels <= 2 )
2145 // Determine by how much resampling will increase number of samples
2146 double resample_factor = self->audio_index == INT_MAX ? 1 : (double) *channels / codec_context->channels;
2147 resample_factor *= (double) *frequency / codec_context->sample_rate;
2148 if ( resample_factor > self->resample_factor )
2149 self->resample_factor = resample_factor;
2151 // Create the resampler
2152 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0))
2153 self->audio_resample[ index ] = av_audio_resample_init(
2154 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2155 codec_context->channels, *frequency, codec_context->sample_rate,
2156 SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 );
2158 self->audio_resample[ index ] = audio_resample_init(
2159 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2160 codec_context->channels, *frequency, codec_context->sample_rate );
2165 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2166 sizeof_sample = sample_bytes( codec_context );
2169 // Check for audio buffer and create if necessary
2170 self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2171 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2173 // Check for decoder buffer and create if necessary
2174 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2178 // Get the audio if required
2185 av_init_packet( &pkt );
2187 // If not resampling, give consumer more than requested.
2188 // It requested number samples based on requested frame rate.
2189 // Do not clean this up with a samples *= ...!
2190 if ( self->audio_index != INT_MAX && ! self->audio_resample[ self->audio_index ] )
2191 *samples = *samples * self->audio_codec[ self->audio_index ]->sample_rate / *frequency;
2193 while ( ret >= 0 && !got_audio )
2195 // Check if the buffer already contains the samples required
2196 if ( self->audio_index != INT_MAX &&
2197 self->audio_used[ self->audio_index ] >= *samples &&
2198 ignore[ self->audio_index ] == 0 )
2203 else if ( self->audio_index == INT_MAX )
2205 // Check if there is enough audio for all streams
2207 for ( index = 0; got_audio && index < context->nb_streams; index++ )
2208 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2215 pthread_mutex_lock( &self->packets_mutex );
2216 if ( mlt_deque_count( self->apackets ) )
2218 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2224 ret = av_read_frame( context, &pkt );
2225 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2227 if ( !av_dup_packet( &pkt ) )
2229 AVPacket *tmp = malloc( sizeof(AVPacket) );
2231 mlt_deque_push_back( self->vpackets, tmp );
2235 pthread_mutex_unlock( &self->packets_mutex );
2237 // We only deal with audio from the selected audio index
2238 index = pkt.stream_index;
2239 if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2240 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2242 int channels2 = ( self->audio_index == INT_MAX || !self->audio_resample[index] ) ?
2243 self->audio_codec[index]->channels : *channels;
2244 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2247 if ( self->seekable || index != self->video_index )
2248 av_free_packet( &pkt );
2252 // Set some additional return values
2253 *format = mlt_audio_s16;
2254 if ( self->audio_index != INT_MAX && !self->audio_resample[ self->audio_index ] )
2256 index = self->audio_index;
2257 *channels = self->audio_codec[ index ]->channels;
2258 *frequency = self->audio_codec[ index ]->sample_rate;
2259 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2260 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2262 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2264 else if ( self->audio_index == INT_MAX )
2266 // This only works if all audio tracks have the same sample format.
2267 for ( index = 0; index < index_max; index++ )
2268 if ( self->audio_codec[ index ] && !self->audio_resample[ index ] )
2270 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2271 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2273 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2278 // Allocate and set the frame's audio buffer
2279 int size = mlt_audio_format_size( *format, *samples, *channels );
2280 *buffer = mlt_pool_alloc( size );
2281 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2283 // Interleave tracks if audio_index=all
2284 if ( self->audio_index == INT_MAX )
2286 uint8_t *dest = *buffer;
2288 for ( i = 0; i < *samples; i++ )
2290 for ( index = 0; index < index_max; index++ )
2291 if ( self->audio_codec[ index ] )
2293 int current_channels = self->audio_codec[ index ]->channels;
2294 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2295 memcpy( dest, src, current_channels * sizeof_sample );
2296 dest += current_channels * sizeof_sample;
2299 for ( index = 0; index < index_max; index++ )
2300 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2302 int current_channels = self->audio_codec[ index ]->channels;
2303 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2304 self->audio_used[index] -= *samples;
2305 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2308 // Copy a single track to the output buffer
2311 index = self->audio_index;
2313 // Now handle the audio if we have enough
2314 if ( self->audio_used[ index ] > 0 )
2316 uint8_t *src = self->audio_buffer[ index ];
2317 // copy samples from audio_buffer
2318 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2319 memcpy( *buffer, src, size * *channels * sizeof_sample );
2320 // supply the remaining requested samples as silence
2321 if ( *samples > self->audio_used[ index ] )
2322 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2323 // reposition the samples within audio_buffer
2324 self->audio_used[ index ] -= size;
2325 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2329 // Otherwise fill with silence
2330 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2336 // Get silence and don't touch the context
2337 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2340 // Regardless of speed (other than paused), we expect to get the next frame
2342 self->audio_expected = position + 1;
2344 pthread_mutex_unlock( &self->audio_mutex );
2349 /** Initialize the audio codec context.
2352 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2354 // Initialise the codec if necessary
2355 if ( !self->audio_codec[ index ] )
2357 // Get codec context
2358 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2361 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2363 // If we don't have a codec and we can't initialise it, we can't do much more...
2365 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2367 // Now store the codec with its destructor
2368 if ( self->audio_codec[ index ] )
2369 avcodec_close( self->audio_codec[ index ] );
2370 self->audio_codec[ index ] = codec_context;
2374 // Remember that we can't use self later
2375 self->audio_index = -1;
2379 // Process properties as AVOptions
2380 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2381 #if LIBAVCODEC_VERSION_MAJOR > 52
2382 if ( codec && codec->priv_class && codec_context->priv_data )
2383 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2386 return self->audio_codec[ index ] && self->audio_index > -1;
2389 /** Set up audio handling.
2392 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2395 mlt_producer producer = self->parent;
2397 // Get the properties
2398 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2400 // Fetch the audio format context
2401 AVFormatContext *context = self->audio_format;
2403 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2405 // Get the audio_index
2406 int index = mlt_properties_get_int( properties, "audio_index" );
2408 // Handle all audio tracks
2409 if ( self->audio_index > -1 &&
2410 mlt_properties_get( properties, "audio_index" ) &&
2411 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2414 // Reopen the file if necessary
2415 if ( !context && self->audio_index > -1 && index > -1 )
2417 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2418 mlt_properties_get( properties, "resource" ) );
2419 context = self->audio_format;
2422 // Exception handling for audio_index
2423 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2425 for ( index = context->nb_streams - 1;
2426 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2428 mlt_properties_set_int( properties, "audio_index", index );
2430 if ( context && index > -1 && index < INT_MAX &&
2431 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2433 index = self->audio_index;
2434 mlt_properties_set_int( properties, "audio_index", index );
2437 // Update the audio properties if the index changed
2438 if ( context && index > -1 && index != self->audio_index )
2440 if ( self->audio_codec[ self->audio_index ] )
2443 avcodec_close( self->audio_codec[ self->audio_index ] );
2446 self->audio_codec[ self->audio_index ] = NULL;
2448 if ( self->audio_index != -1 )
2449 self->audio_index = index;
2454 if ( context && index == INT_MAX )
2456 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2457 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2458 for ( index = 0; index < context->nb_streams; index++ )
2460 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2461 audio_codec_init( self, index, properties );
2464 else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2466 // Set the frame properties
2467 if ( index < INT_MAX )
2469 mlt_properties_set_int( frame_properties, "frequency", self->audio_codec[ index ]->sample_rate );
2470 mlt_properties_set_int( frame_properties, "channels", self->audio_codec[ index ]->channels );
2473 if ( context && index > -1 )
2475 // Add our audio operation
2476 mlt_frame_push_audio( frame, self );
2477 mlt_frame_push_audio( frame, producer_get_audio );
2481 /** Our get frame implementation.
2484 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2486 // Access the private data
2487 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2488 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2489 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2494 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2495 producer->child = self;
2496 self->parent = producer;
2497 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2498 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2501 // Create an empty frame
2502 *frame = mlt_frame_init( service);
2506 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2510 mlt_cache_item_close( cache_item );
2514 // Update timecode on the frame we're creating
2515 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2517 // Set the position of this producer
2518 mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2519 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", position );
2522 producer_set_up_video( self, *frame );
2525 producer_set_up_audio( self, *frame );
2527 // Calculate the next timecode
2528 mlt_producer_prepare_next( producer );
2533 static void producer_avformat_close( producer_avformat self )
2535 mlt_log_debug( NULL, "producer_avformat_close\n" );
2537 // Cleanup av contexts
2538 av_free( self->av_frame );
2541 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2543 if ( self->audio_resample[i] )
2544 audio_resample_close( self->audio_resample[i] );
2545 mlt_pool_release( self->audio_buffer[i] );
2546 av_free( self->decode_buffer[i] );
2547 if ( self->audio_codec[i] )
2548 avcodec_close( self->audio_codec[i] );
2550 if ( self->video_codec )
2551 avcodec_close( self->video_codec );
2553 if ( self->dummy_context )
2554 av_close_input_file( self->dummy_context );
2555 if ( self->seekable && self->audio_format )
2556 av_close_input_file( self->audio_format );
2557 if ( self->video_format )
2558 av_close_input_file( self->video_format );
2561 vdpau_producer_close( self );
2563 if ( self->image_cache )
2564 mlt_cache_close( self->image_cache );
2566 // Cleanup the mutexes
2567 pthread_mutex_destroy( &self->audio_mutex );
2568 pthread_mutex_destroy( &self->video_mutex );
2569 pthread_mutex_destroy( &self->packets_mutex );
2571 // Cleanup the packet queues
2573 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2575 av_free_packet( pkt );
2578 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2580 av_free_packet( pkt );
2587 static void producer_close( mlt_producer parent )
2589 // Remove this instance from the cache
2590 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2593 parent->close = NULL;
2594 mlt_producer_close( parent );