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;
99 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
100 size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
101 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
102 int audio_used[ MAX_AUDIO_STREAMS ];
104 int audio_max_stream;
108 unsigned int invalid_pts_counter;
109 double resample_factor;
110 mlt_cache image_cache;
112 pthread_mutex_t video_mutex;
113 pthread_mutex_t audio_mutex;
118 struct vdpau_render_state render_states[MAX_VDPAU_SURFACES];
129 typedef struct producer_avformat_s *producer_avformat;
131 // Forward references.
132 static int list_components( char* file );
133 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL );
134 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
135 static void producer_avformat_close( producer_avformat );
136 static void producer_close( mlt_producer parent );
137 static void producer_set_up_video( producer_avformat self, mlt_frame frame );
138 static void producer_set_up_audio( producer_avformat self, mlt_frame frame );
139 static void apply_properties( void *obj, mlt_properties properties, int flags );
140 static int video_codec_init( producer_avformat self, int index, mlt_properties properties );
146 /** Constructor for libavformat.
149 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
151 if ( list_components( file ) )
154 mlt_producer producer = NULL;
156 // Check that we have a non-NULL argument
159 // Construct the producer
160 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
161 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
164 if ( mlt_producer_init( producer, self ) == 0 )
166 self->parent = producer;
168 // Get the properties
169 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
171 // Set the resource property (required for all producers)
172 mlt_properties_set( properties, "resource", file );
174 // Register transport implementation with the producer
175 producer->close = (mlt_destructor) producer_close;
177 // Register our get_frame implementation
178 producer->get_frame = producer_get_frame;
180 if ( strcmp( service, "avformat-novalidate" ) )
183 if ( producer_open( self, profile, file ) != 0 )
186 mlt_producer_close( producer );
191 // Close the file to release resources for large playlists - reopen later as needed
193 if ( self->dummy_context )
194 av_close_input_file( self->dummy_context );
195 self->dummy_context = NULL;
196 if ( self->audio_format )
197 av_close_input_file( self->audio_format );
198 self->audio_format = NULL;
199 if ( self->video_format )
200 av_close_input_file( self->video_format );
201 self->video_format = NULL;
204 // Default the user-selectable indices from the auto-detected indices
205 mlt_properties_set_int( properties, "audio_index", self->audio_index );
206 mlt_properties_set_int( properties, "video_index", self->video_index );
212 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
214 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
221 int list_components( char* file )
225 // Report information about available demuxers and codecs as YAML Tiny
226 if ( file && strstr( file, "f-list" ) )
228 fprintf( stderr, "---\nformats:\n" );
229 AVInputFormat *format = NULL;
230 while ( ( format = av_iformat_next( format ) ) )
231 fprintf( stderr, " - %s\n", format->name );
232 fprintf( stderr, "...\n" );
235 if ( file && strstr( file, "acodec-list" ) )
237 fprintf( stderr, "---\naudio_codecs:\n" );
238 AVCodec *codec = NULL;
239 while ( ( codec = av_codec_next( codec ) ) )
240 if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
241 fprintf( stderr, " - %s\n", codec->name );
242 fprintf( stderr, "...\n" );
245 if ( file && strstr( file, "vcodec-list" ) )
247 fprintf( stderr, "---\nvideo_codecs:\n" );
248 AVCodec *codec = NULL;
249 while ( ( codec = av_codec_next( codec ) ) )
250 if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
251 fprintf( stderr, " - %s\n", codec->name );
252 fprintf( stderr, "...\n" );
259 /** Find the default streams.
262 static mlt_properties find_default_streams( producer_avformat self )
266 AVMetadataTag *tag = NULL;
267 AVFormatContext *context = self->video_format;
268 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
270 // Default to the first audio and video streams found
271 self->audio_index = -1;
272 self->video_index = -1;
274 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
276 // Allow for multiple audio and video streams in the file and select first of each (if available)
277 for( i = 0; i < context->nb_streams; i++ )
279 // Get the codec context
280 AVStream *stream = context->streams[ i ];
281 if ( ! stream ) continue;
282 AVCodecContext *codec_context = stream->codec;
283 if ( ! codec_context ) continue;
284 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
285 if ( ! codec ) continue;
287 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
289 // Determine the type and obtain the first index of each type
290 switch( codec_context->codec_type )
292 case CODEC_TYPE_VIDEO:
293 // Use first video stream
294 if ( self->video_index < 0 )
295 self->video_index = i;
296 mlt_properties_set( meta_media, key, "video" );
297 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
298 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
299 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
300 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
301 ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
302 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
304 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) );
307 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
308 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
309 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
311 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
312 mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den /
313 ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) );
314 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
315 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
316 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
317 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
318 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
319 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
320 switch ( codec_context->colorspace )
322 case AVCOL_SPC_SMPTE240M:
323 mlt_properties_set_int( meta_media, key, 240 );
325 case AVCOL_SPC_BT470BG:
326 case AVCOL_SPC_SMPTE170M:
327 mlt_properties_set_int( meta_media, key, 601 );
329 case AVCOL_SPC_BT709:
330 mlt_properties_set_int( meta_media, key, 709 );
333 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
334 mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
339 case CODEC_TYPE_AUDIO:
340 // Use first audio stream
341 if ( self->audio_index < 0 )
342 self->audio_index = i;
343 mlt_properties_set( meta_media, key, "audio" );
344 #if LIBAVCODEC_VERSION_MAJOR > 52
345 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
346 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
347 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
348 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
349 mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) );
351 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
352 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
353 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
354 mlt_properties_set_int( meta_media, key, codec_context->channels );
359 // snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
360 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
361 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
362 mlt_properties_set( meta_media, key, codec->name );
363 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0))
364 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
365 mlt_properties_set( meta_media, key, codec->long_name );
367 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
368 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
369 // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
370 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
371 // snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
372 // mlt_properties_set_int( meta_media, key, codec_context->profile );
373 // snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
374 // mlt_properties_set_int( meta_media, key, codec_context->level );
377 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
378 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
380 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
382 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
383 mlt_properties_set( meta_media, key, tag->value );
388 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
389 while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
391 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
393 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
394 mlt_properties_set( meta_media, key, tag->value );
398 if ( context->title && strcmp( context->title, "" ) )
399 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
400 if ( context->author && strcmp( context->author, "" ) )
401 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
402 if ( context->copyright && strcmp( context->copyright, "" ) )
403 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
404 if ( context->comment )
405 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
406 if ( context->album )
407 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
409 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
410 if ( context->track )
411 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
417 static inline int dv_is_pal( AVPacket *pkt )
419 return pkt->data[3] & 0x80;
422 static int dv_is_wide( AVPacket *pkt )
424 int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
426 for ( ; i < pkt->size; i += 5 /* packet size */ )
428 if ( pkt->data[ i ] == 0x61 )
430 uint8_t x = pkt->data[ i + 2 ] & 0x7;
431 return ( x == 2 ) || ( x == 7 );
437 static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
439 double aspect_ratio = 1.0;
441 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
445 if ( dv_is_pal( pkt ) )
447 if ( dv_is_wide( pkt ) )
449 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
450 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
454 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
455 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
460 if ( dv_is_wide( pkt ) )
462 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
463 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
467 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
468 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
475 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
476 stream->sample_aspect_ratio;
478 codec_context->sample_aspect_ratio;
480 // Override FFmpeg's notion of DV aspect ratios, which are
481 // based upon a width of 704. Since we do not have a normaliser
482 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
483 // we just coerce the values to facilitate a passive behaviour through
484 // the rescale normaliser when using equivalent producers and consumers.
485 // = display_aspect / (width * height)
486 if ( ar.num == 10 && ar.den == 11 )
489 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
490 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
492 else if ( ar.num == 59 && ar.den == 54 )
495 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
496 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
498 else if ( ar.num == 40 && ar.den == 33 )
501 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
502 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
504 else if ( ar.num == 118 && ar.den == 81 )
507 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
508 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
514 AVRational codec_sar = codec_context->sample_aspect_ratio;
515 AVRational stream_sar =
516 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
517 stream->sample_aspect_ratio;
521 if ( codec_sar.num > 0 )
523 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num );
524 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den );
526 else if ( stream_sar.num > 0 )
528 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num );
529 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den );
533 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 );
534 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 );
537 AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) };
538 aspect_ratio = av_q2d( ar );
539 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
544 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters **parameters )
546 if ( !URL ) return NULL;
549 char *protocol = strdup( URL );
550 char *url = strchr( protocol, ':' );
552 // Only if there is not a protocol specification that avformat can handle
553 #if LIBAVFORMAT_VERSION_MAJOR > 52
554 if ( url && avio_check( URL, 0 ) < 0 )
556 if ( url && !url_exist( URL ) )
559 // Truncate protocol string
563 *format = av_find_input_format( protocol );
565 // Eat the format designator
571 AVFormatParameters *params = *parameters = calloc( 1, sizeof( AVFormatParameters ) );
573 // These are required by video4linux2 (defaults)
574 params->width = profile->width;
575 params->height = profile->height;
576 params->time_base= (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
577 params->channels = 2;
578 params->sample_rate = 48000;
581 url = strchr( url, '?' );
585 char *name = strdup( ++url );
586 char *value = strchr( name, ':' );
591 char *t = strchr( value, '&' );
594 if ( !strcmp( name, "frame_rate" ) )
595 params->time_base.den = atoi( value );
596 else if ( !strcmp( name, "frame_rate_base" ) )
597 params->time_base.num = atoi( value );
598 else if ( !strcmp( name, "sample_rate" ) )
599 params->sample_rate = atoi( value );
600 else if ( !strcmp( name, "channel" ) )
601 params->channel = atoi( value );
602 else if ( !strcmp( name, "channels" ) )
603 params->channels = atoi( value );
604 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
605 else if ( !strcmp( name, "pix_fmt" ) )
606 params->pix_fmt = av_get_pix_fmt( value );
608 else if ( !strcmp( name, "width" ) )
609 params->width = atoi( value );
610 else if ( !strcmp( name, "height" ) )
611 params->height = atoi( value );
612 else if ( !strcmp( name, "standard" ) )
613 params->standard = strdup( value );
616 url = strchr( url, '&' );
621 return result ? strdup( result ) : strdup( URL );
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 = context->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)
738 pthread_mutex_init( &self->audio_mutex, NULL );
739 pthread_mutex_init( &self->video_mutex, NULL );
740 pthread_mutex_lock( &self->audio_mutex );
741 pthread_mutex_lock( &self->video_mutex );
744 AVInputFormat *format = NULL;
745 AVFormatParameters *params = NULL;
746 char *filename = parse_url( profile, URL, &format, ¶ms );
748 // Now attempt to open the file
749 error = av_open_input_file( &self->video_format, filename, format, 0, params ) < 0;
751 // Cleanup AVFormatParameters
754 if ( params->standard )
755 free( (void*) params->standard );
759 // If successful, then try to get additional info
762 // Get the stream info
763 error = av_find_stream_info( self->video_format ) < 0;
765 // Continue if no error
768 // Find default audio and video streams
769 find_default_streams( self );
770 error = get_basic_info( self, profile, filename );
772 // Initialize position info
773 self->first_pts = -1;
774 self->last_position = POSITION_INITIAL;
776 // We're going to cheat here - for seekable A/V files, we will have separate contexts
777 // to support independent seeking of audio from video.
778 // TODO: Is this really necessary?
779 if ( self->audio_index != -1 && self->video_index != -1 )
781 // And open again for our audio context
782 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
783 av_find_stream_info( self->audio_format );
785 else if ( self->audio_index != -1 )
787 // We only have an audio context
788 self->audio_format = self->video_format;
789 self->video_format = NULL;
791 else if ( self->video_index == -1 )
793 // Something has gone wrong
801 // Unlock the service
802 pthread_mutex_unlock( &self->audio_mutex );
803 pthread_mutex_unlock( &self->video_mutex );
808 void reopen_video( producer_avformat self, mlt_producer producer, mlt_properties properties )
810 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
811 pthread_mutex_lock( &self->audio_mutex );
813 if ( self->video_codec )
816 avcodec_close( self->video_codec );
819 self->video_codec = NULL;
820 if ( self->dummy_context )
821 av_close_input_file( self->dummy_context );
822 self->dummy_context = NULL;
823 if ( self->video_format )
824 av_close_input_file( self->video_format );
825 self->video_format = NULL;
827 int audio_index = self->audio_index;
828 int video_index = self->video_index;
830 mlt_events_block( properties, producer );
831 pthread_mutex_unlock( &self->audio_mutex );
832 pthread_mutex_unlock( &self->video_mutex );
833 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
834 mlt_properties_get( properties, "resource" ) );
835 pthread_mutex_lock( &self->video_mutex );
836 pthread_mutex_lock( &self->audio_mutex );
837 if ( self->dummy_context )
839 av_close_input_file( self->dummy_context );
840 self->dummy_context = NULL;
842 mlt_events_unblock( properties, producer );
843 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
844 #if LIBAVFORMAT_VERSION_MAJOR > 52
845 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
846 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
849 self->audio_index = audio_index;
850 if ( self->video_format && video_index > -1 )
852 self->video_index = video_index;
853 video_codec_init( self, video_index, properties );
856 pthread_mutex_unlock( &self->audio_mutex );
857 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
860 /** Convert a frame position to a time code.
863 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
865 return ( double )position / mlt_producer_get_fps( producer );
868 // Collect information about all audio streams
870 static void get_audio_streams_info( producer_avformat self )
872 // Fetch the audio format context
873 AVFormatContext *context = self->audio_format;
877 i < context->nb_streams;
880 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
882 AVCodecContext *codec_context = context->streams[i]->codec;
883 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
885 // If we don't have a codec and we can't initialise it, we can't do much more...
887 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
889 self->audio_streams++;
890 self->audio_max_stream = i;
891 self->total_channels += codec_context->channels;
892 if ( codec_context->channels > self->max_channel )
893 self->max_channel = codec_context->channels;
894 if ( codec_context->sample_rate > self->max_frequency )
895 self->max_frequency = codec_context->sample_rate;
896 avcodec_close( codec_context );
901 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
902 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
904 // Other audio-specific initializations
905 self->resample_factor = 1.0;
908 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
910 #if defined(SWSCALE) && (LIBSWSCALE_VERSION_INT >= ((0<<16)+(7<<8)+2))
912 const int *new_coefficients;
914 int brightness, contrast, saturation;
916 if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
917 &brightness, &contrast, &saturation ) != -1 )
919 // Don't change these from defaults unless explicitly told to.
920 if ( use_full_range >= 0 )
921 full_range = use_full_range;
922 switch ( colorspace )
928 new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
931 new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
934 new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
937 new_coefficients = coefficients;
940 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
941 brightness, contrast, saturation );
946 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt,
947 mlt_image_format *format, int width, int height, int colorspace )
951 int flags = SWS_BILINEAR | SWS_ACCURATE_RND;
954 flags |= SWS_CPU_CAPS_MMX;
957 flags |= SWS_CPU_CAPS_MMX2;
960 if ( pix_fmt == PIX_FMT_RGB32 )
962 *format = mlt_image_rgb24a;
963 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
964 width, height, PIX_FMT_RGBA, flags, NULL, NULL, NULL);
966 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
967 set_luma_transfer( context, colorspace, full_range );
968 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
969 output.data, output.linesize);
970 sws_freeContext( context );
972 else if ( *format == mlt_image_yuv420p )
974 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
975 width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
977 output.data[0] = buffer;
978 output.data[1] = buffer + width * height;
979 output.data[2] = buffer + ( 5 * width * height ) / 4;
980 output.linesize[0] = width;
981 output.linesize[1] = width >> 1;
982 output.linesize[2] = width >> 1;
983 set_luma_transfer( context, colorspace, full_range );
984 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
985 output.data, output.linesize);
986 sws_freeContext( context );
988 else if ( *format == mlt_image_rgb24 )
990 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
991 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
993 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
994 set_luma_transfer( context, colorspace, full_range );
995 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
996 output.data, output.linesize);
997 sws_freeContext( context );
999 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1001 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1002 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1004 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1005 set_luma_transfer( context, colorspace, full_range );
1006 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1007 output.data, output.linesize);
1008 sws_freeContext( context );
1012 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1013 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1015 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1016 set_luma_transfer( context, colorspace, full_range );
1017 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1018 output.data, output.linesize);
1019 sws_freeContext( context );
1022 if ( *format == mlt_image_yuv420p )
1025 pict.data[0] = buffer;
1026 pict.data[1] = buffer + width * height;
1027 pict.data[2] = buffer + ( 5 * width * height ) / 4;
1028 pict.linesize[0] = width;
1029 pict.linesize[1] = width >> 1;
1030 pict.linesize[2] = width >> 1;
1031 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
1033 else if ( *format == mlt_image_rgb24 )
1036 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1037 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
1039 else if ( format == mlt_image_rgb24a || format == mlt_image_opengl )
1042 avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height );
1043 img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height );
1048 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1049 img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height );
1054 /** Allocate the image buffer and set it on the frame.
1057 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1061 if ( codec_context->width == 0 || codec_context->height == 0 )
1064 *width = codec_context->width;
1065 *height = codec_context->height;
1067 if ( codec_context->pix_fmt == PIX_FMT_RGB32 )
1068 size = *width * ( *height + 1 ) * 4;
1070 size = mlt_image_format_size( *format, *width, *height, NULL );
1072 // Construct the output image
1073 *buffer = mlt_pool_alloc( size );
1075 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1082 /** Get an image from a frame.
1085 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1088 producer_avformat self = mlt_frame_pop_service( frame );
1089 mlt_producer producer = self->parent;
1091 // Get the properties from the frame
1092 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1094 // Obtain the frame number of this frame
1095 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
1097 // Get the producer properties
1098 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1100 pthread_mutex_lock( &self->video_mutex );
1102 // Fetch the video format context
1103 AVFormatContext *context = self->video_format;
1105 // Get the video stream
1106 AVStream *stream = context->streams[ self->video_index ];
1108 // Get codec context
1109 AVCodecContext *codec_context = stream->codec;
1111 // Get the image cache
1112 if ( ! self->image_cache && ! mlt_properties_get_int( properties, "noimagecache" ) )
1113 self->image_cache = mlt_cache_init();
1114 if ( self->image_cache )
1116 mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position );
1117 uint8_t *original = mlt_cache_item_data( item, (int*) format );
1120 // Set the resolution
1121 *width = codec_context->width;
1122 *height = codec_context->height;
1124 // Workaround 1088 encodings missing cropping info.
1125 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1129 int size = mlt_image_format_size( *format, *width, *height, NULL );
1132 *buffer = mlt_pool_alloc( size );
1133 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1134 memcpy( *buffer, original, size );
1135 mlt_cache_item_close( item );
1140 mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
1141 mlt_frame_set_image( frame, *buffer, size, NULL );
1143 self->got_picture = 1;
1145 goto exit_get_image;
1154 // Special case pause handling flag
1157 // Special case ffwd handling
1160 // We may want to use the source fps if available
1161 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1162 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1163 double fps = mlt_producer_get_fps( producer );
1165 // This is the physical frame position in the source
1166 int req_position = ( int )( position / fps * source_fps + 0.5 );
1168 // Determines if we have to decode all frames in a sequence
1169 // Temporary hack to improve intra frame only
1170 int must_decode = strcmp( codec_context->codec->name, "dnxhd" ) &&
1171 strcmp( codec_context->codec->name, "dvvideo" ) &&
1172 strcmp( codec_context->codec->name, "huffyuv" ) &&
1173 strcmp( codec_context->codec->name, "mjpeg" ) &&
1174 strcmp( codec_context->codec->name, "rawvideo" );
1176 int last_position = self->last_position;
1178 // Turn on usage of new seek API and PTS for seeking
1179 int use_new_seek = codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" );
1180 if ( mlt_properties_get( properties, "new_seek" ) )
1181 use_new_seek = mlt_properties_get_int( properties, "new_seek" );
1183 // Seek if necessary
1184 if ( position != self->video_expected || last_position < 0 )
1186 if ( self->av_frame && position + 1 == self->video_expected )
1188 // We're paused - use last image
1191 else if ( !self->seekable && position > self->video_expected && ( position - self->video_expected ) < 250 )
1193 // Fast forward - seeking is inefficient for small distances - just ignore following frames
1194 ignore = ( int )( ( position - self->video_expected ) / fps * source_fps );
1195 codec_context->skip_loop_filter = AVDISCARD_NONREF;
1197 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || last_position < 0 ) )
1199 if ( use_new_seek && last_position == POSITION_INITIAL )
1201 // find first key frame
1205 while ( ret >= 0 && toscan-- > 0 )
1207 ret = av_read_frame( context, &pkt );
1208 if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == self->video_index )
1210 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) );
1211 self->first_pts = pkt.pts;
1214 av_free_packet( &pkt );
1217 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
1220 // Calculate the timestamp for the requested frame
1224 timestamp = ( req_position - 0.1 / source_fps ) /
1225 ( av_q2d( stream->time_base ) * source_fps );
1226 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "pos %d pts %"PRId64" ", req_position, timestamp );
1227 if ( self->first_pts > 0 )
1228 timestamp += self->first_pts;
1229 else if ( context->start_time != AV_NOPTS_VALUE )
1230 timestamp += context->start_time;
1234 timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
1235 if ( context->start_time != AV_NOPTS_VALUE )
1236 timestamp += context->start_time;
1239 timestamp -= AV_TIME_BASE;
1240 if ( timestamp < 0 )
1242 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %d\n",
1243 timestamp, position, self->video_expected, last_position );
1245 // Seek to the timestamp
1248 codec_context->skip_loop_filter = AVDISCARD_NONREF;
1249 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
1251 else if ( req_position > 0 || last_position <= 0 )
1253 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
1257 // Re-open video stream when rewinding to beginning from somewhere else.
1258 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
1259 // not reliably seeking to the first frame across formats.
1260 reopen_video( self, producer, properties );
1261 context = self->video_format;
1262 stream = context->streams[ self->video_index ];
1263 codec_context = stream->codec;
1266 // Remove the cached info relating to the previous position
1267 self->current_position = POSITION_INVALID;
1268 self->last_position = POSITION_INVALID;
1269 av_freep( &self->av_frame );
1273 // flush any pictures still in decode buffer
1274 avcodec_flush_buffers( codec_context );
1279 // Duplicate the last image if necessary
1280 if ( self->av_frame && self->av_frame->linesize[0] && self->got_picture && self->seekable
1282 || self->current_position == req_position
1283 || ( !use_new_seek && self->current_position > req_position ) ) )
1286 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1288 // Workaround 1088 encodings missing cropping info.
1289 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1292 if ( self->vdpau && self->vdpau->buffer )
1295 picture.data[0] = self->vdpau->buffer;
1296 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1297 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1298 picture.linesize[0] = codec_context->width;
1299 picture.linesize[1] = codec_context->width / 2;
1300 picture.linesize[2] = codec_context->width / 2;
1301 convert_image( (AVFrame*) &picture, *buffer,
1302 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1306 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1307 format, *width, *height, self->colorspace );
1310 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1315 int int_position = 0;
1316 int decode_errors = 0;
1317 int got_picture = 0;
1319 av_init_packet( &pkt );
1321 // Construct an AVFrame for YUV422 conversion
1322 if ( !self->av_frame )
1323 self->av_frame = avcodec_alloc_frame( );
1325 while( ret >= 0 && !got_picture )
1328 ret = av_read_frame( context, &pkt );
1330 // We only deal with video from the selected video_index
1331 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
1333 // Determine time code of the packet
1336 int64_t pts = pkt.pts;
1337 if ( self->first_pts > 0 )
1338 pts -= self->first_pts;
1339 else if ( context->start_time != AV_NOPTS_VALUE )
1340 pts -= context->start_time;
1341 int_position = ( int )( av_q2d( stream->time_base ) * pts * source_fps + 0.1 );
1342 if ( pkt.pts == AV_NOPTS_VALUE )
1344 self->invalid_pts_counter++;
1345 if ( self->invalid_pts_counter > 20 )
1347 mlt_log_panic( MLT_PRODUCER_SERVICE(producer), "\ainvalid PTS; DISABLING NEW_SEEK!\n" );
1348 mlt_properties_set_int( properties, "new_seek", 0 );
1349 int_position = req_position;
1355 self->invalid_pts_counter = 0;
1357 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.pts %"PRId64" req_pos %d cur_pos %d pkt_pos %d\n",
1358 pkt.pts, req_position, self->current_position, int_position );
1362 if ( self->seekable && pkt.dts != AV_NOPTS_VALUE )
1364 int_position = ( int )( av_q2d( stream->time_base ) * pkt.dts * source_fps + 0.5 );
1365 if ( context->start_time != AV_NOPTS_VALUE )
1366 int_position -= ( int )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
1367 last_position = self->last_position;
1368 if ( int_position == last_position )
1369 int_position = last_position + 1;
1373 int_position = req_position;
1375 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %"PRId64" req_pos %d cur_pos %d pkt_pos %d\n",
1376 pkt.dts, req_position, self->current_position, int_position );
1377 // Make a dumb assumption on streams that contain wild timestamps
1378 if ( abs( req_position - int_position ) > 999 )
1380 int_position = req_position;
1381 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" );
1384 self->last_position = int_position;
1387 if ( must_decode || int_position >= req_position )
1390 if ( g_vdpau && self->vdpau )
1392 if ( g_vdpau->producer != self )
1394 vdpau_decoder_close();
1395 vdpau_decoder_init( self );
1398 self->vdpau->is_decoded = 0;
1401 codec_context->reordered_opaque = pkt.pts;
1402 if ( int_position >= req_position )
1403 codec_context->skip_loop_filter = AVDISCARD_NONE;
1404 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1405 ret = avcodec_decode_video2( codec_context, self->av_frame, &got_picture, &pkt );
1407 ret = avcodec_decode_video( codec_context, self->av_frame, &got_picture, pkt.data, pkt.size );
1409 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1412 if ( ++decode_errors <= 10 )
1425 // Determine time code of the packet
1426 int64_t pts = self->av_frame->reordered_opaque;
1427 if ( self->first_pts > 0 )
1428 pts -= self->first_pts;
1429 else if ( context->start_time != AV_NOPTS_VALUE )
1430 pts -= context->start_time;
1431 int_position = ( int )( av_q2d( stream->time_base) * pts * source_fps + 0.1 );
1432 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "got frame %d, key %d\n", int_position, self->av_frame->key_frame );
1435 if ( int_position < req_position )
1440 else if ( int_position >= req_position )
1443 codec_context->skip_loop_filter = AVDISCARD_NONE;
1445 else if ( ignore -- )
1450 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY );
1453 // Now handle the picture if we have one
1456 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1458 // Workaround 1088 encodings missing cropping info.
1459 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1464 if ( self->vdpau->is_decoded )
1466 struct vdpau_render_state *render = (struct vdpau_render_state*) self->av_frame->data[0];
1468 uint32_t pitches[3];
1469 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1471 if ( !self->vdpau->buffer )
1472 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1473 self->av_frame->data[0] = planes[0] = self->vdpau->buffer;
1474 self->av_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1475 self->av_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1476 self->av_frame->linesize[0] = pitches[0] = codec_context->width;
1477 self->av_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1478 self->av_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1480 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1481 if ( status == VDP_STATUS_OK )
1483 convert_image( self->av_frame, *buffer, PIX_FMT_YUV420P,
1484 format, *width, *height, self->colorspace );
1488 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1489 image_size = self->vdpau->is_decoded = 0;
1494 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1495 image_size = got_picture = 0;
1500 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1501 format, *width, *height, self->colorspace );
1502 self->top_field_first |= self->av_frame->top_field_first;
1503 self->current_position = int_position;
1504 self->got_picture = 1;
1512 av_free_packet( &pkt );
1516 if ( self->got_picture && image_size > 0 && self->image_cache )
1518 // Copy buffer to image cache
1519 uint8_t *image = mlt_pool_alloc( image_size );
1520 memcpy( image, *buffer, image_size );
1521 mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release );
1523 // Try to duplicate last image if there was a decoding failure
1524 else if ( !image_size && self->av_frame && self->av_frame->linesize[0] )
1527 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1529 // Workaround 1088 encodings missing cropping info.
1530 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1533 if ( self->vdpau && self->vdpau->buffer )
1536 picture.data[0] = self->vdpau->buffer;
1537 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1538 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1539 picture.linesize[0] = codec_context->width;
1540 picture.linesize[1] = codec_context->width / 2;
1541 picture.linesize[2] = codec_context->width / 2;
1542 convert_image( (AVFrame*) &picture, *buffer,
1543 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1547 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1548 format, *width, *height, self->colorspace );
1549 self->got_picture = 1;
1552 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1555 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1556 self->video_expected = position + 1;
1560 pthread_mutex_unlock( &self->video_mutex );
1562 // Set the progressive flag
1563 if ( mlt_properties_get( properties, "force_progressive" ) )
1564 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1565 else if ( self->av_frame )
1566 mlt_properties_set_int( frame_properties, "progressive", !self->av_frame->interlaced_frame );
1568 // Set the field order property for this frame
1569 if ( mlt_properties_get( properties, "force_tff" ) )
1570 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1572 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1574 // Set immutable properties of the selected track's (or overridden) source attributes.
1575 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1576 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1577 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1578 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1580 return !self->got_picture;
1583 /** Process properties as AVOptions and apply to AV context obj
1586 static void apply_properties( void *obj, mlt_properties properties, int flags )
1589 int count = mlt_properties_count( properties );
1590 for ( i = 0; i < count; i++ )
1592 const char *opt_name = mlt_properties_get_name( properties, i );
1593 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1594 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1597 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1598 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1599 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1600 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1602 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1608 /** Initialize the video codec context.
1611 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1613 // Initialise the codec if necessary
1614 if ( !self->video_codec )
1616 // Get the video stream
1617 AVStream *stream = self->video_format->streams[ index ];
1619 // Get codec context
1620 AVCodecContext *codec_context = stream->codec;
1623 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1625 if ( codec_context->codec_id == CODEC_ID_H264 )
1627 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1629 if ( vdpau_init( self ) )
1631 self->video_codec = codec_context;
1632 if ( !vdpau_decoder_init( self ) )
1633 vdpau_decoder_close();
1637 codec = avcodec_find_decoder( codec_context->codec_id );
1641 // Initialise multi-threading
1642 int thread_count = mlt_properties_get_int( properties, "threads" );
1643 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1644 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1645 if ( thread_count > 1 )
1646 codec_context->thread_count = thread_count;
1648 // If we don't have a codec and we can't initialise it, we can't do much more...
1650 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1652 // Now store the codec with its destructor
1653 self->video_codec = codec_context;
1657 // Remember that we can't use this later
1658 self->video_index = -1;
1664 // Process properties as AVOptions
1665 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1666 #if LIBAVCODEC_VERSION_MAJOR > 52
1667 if ( codec->priv_class && codec_context->priv_data )
1668 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1671 // Reset some image properties
1672 mlt_properties_set_int( properties, "width", self->video_codec->width );
1673 mlt_properties_set_int( properties, "height", self->video_codec->height );
1674 // For DV, we'll just use the saved aspect ratio
1675 if ( codec_context->codec_id != CODEC_ID_DVVIDEO )
1676 get_aspect_ratio( properties, stream, self->video_codec, NULL );
1678 // Determine the fps first from the codec
1679 double source_fps = (double) self->video_codec->time_base.den /
1680 ( self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1682 if ( mlt_properties_get( properties, "force_fps" ) )
1684 source_fps = mlt_properties_get_double( properties, "force_fps" );
1685 stream->time_base = av_d2q( source_fps, 1024 );
1686 mlt_properties_set_int( properties, "meta.media.frame_rate_num", stream->time_base.num );
1687 mlt_properties_set_int( properties, "meta.media.frame_rate_den", stream->time_base.den );
1691 // If the muxer reports a frame rate different than the codec
1692 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1693 double muxer_fps = av_q2d( stream->avg_frame_rate );
1694 if ( isnan( muxer_fps ) || muxer_fps == 0 )
1695 muxer_fps = av_q2d( stream->r_frame_rate );
1697 double muxer_fps = av_q2d( stream->r_frame_rate );
1699 // Choose the lesser - the wrong tends to be off by some multiple of 10
1700 source_fps = FFMIN( source_fps, muxer_fps );
1701 if ( source_fps >= 1.0 && ( source_fps < muxer_fps || isnan( muxer_fps ) ) )
1703 mlt_properties_set_int( properties, "meta.media.frame_rate_num", self->video_codec->time_base.den );
1704 mlt_properties_set_int( properties, "meta.media.frame_rate_den", self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1706 else if ( muxer_fps > 0 )
1708 AVRational frame_rate = stream->r_frame_rate;
1709 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1710 // avg_frame_rate gives some approximate value that does not well match the media.
1711 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1712 // results in some very choppy output, but some value slightly different works
1714 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1715 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1716 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1718 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1719 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1723 source_fps = mlt_producer_get_fps( self->parent );
1724 AVRational frame_rate = av_d2q( source_fps, 255 );
1725 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1726 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1730 // source_fps is deprecated in favor of meta.media.frame_rate_num and .frame_rate_den
1731 if ( source_fps > 0 )
1732 mlt_properties_set_double( properties, "source_fps", source_fps );
1734 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( self->parent ) );
1736 // Set the YUV colorspace from override or detect
1737 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1738 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1739 if ( ! self->colorspace )
1741 switch ( self->video_codec->colorspace )
1743 case AVCOL_SPC_SMPTE240M:
1744 self->colorspace = 240;
1746 case AVCOL_SPC_BT470BG:
1747 case AVCOL_SPC_SMPTE170M:
1748 self->colorspace = 601;
1750 case AVCOL_SPC_BT709:
1751 self->colorspace = 709;
1754 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1755 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1760 // Let apps get chosen colorspace
1761 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1763 return self->video_codec && self->video_index > -1;
1766 /** Set up video handling.
1769 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1772 mlt_producer producer = self->parent;
1774 // Get the properties
1775 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1777 // Fetch the video format context
1778 AVFormatContext *context = self->video_format;
1780 // Get the video_index
1781 int index = mlt_properties_get_int( properties, "video_index" );
1783 // Reopen the file if necessary
1784 if ( !context && index > -1 )
1786 mlt_events_block( properties, producer );
1787 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1788 mlt_properties_get( properties, "resource" ) );
1789 context = self->video_format;
1790 if ( self->dummy_context )
1792 av_close_input_file( self->dummy_context );
1793 self->dummy_context = NULL;
1795 mlt_events_unblock( properties, producer );
1796 if ( self->audio_format && !self->audio_streams )
1797 get_audio_streams_info( self );
1799 // Process properties as AVOptions
1802 apply_properties( context, properties, AV_OPT_FLAG_DECODING_PARAM );
1803 #if LIBAVFORMAT_VERSION_MAJOR > 52
1804 if ( context->iformat && context->iformat->priv_class && context->priv_data )
1805 apply_properties( context->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
1810 // Exception handling for video_index
1811 if ( context && index >= (int) context->nb_streams )
1813 // Get the last video stream
1814 for ( index = context->nb_streams - 1;
1815 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1817 mlt_properties_set_int( properties, "video_index", index );
1819 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1821 // Invalidate the video stream
1823 mlt_properties_set_int( properties, "video_index", index );
1826 // Update the video properties if the index changed
1827 if ( index != self->video_index )
1829 // Reset the video properties if the index changed
1830 self->video_index = index;
1831 if ( self->video_codec )
1834 avcodec_close( self->video_codec );
1837 self->video_codec = NULL;
1840 // Get the frame properties
1841 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1844 if ( context && index > -1 && video_codec_init( self, index, properties ) )
1846 // Set the frame properties
1847 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1848 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1849 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1851 // Set the width and height
1852 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
1853 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
1854 // real_width and real_height are deprecated in favor of meta.media.width and .height
1855 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1856 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1857 mlt_properties_set_int( frame_properties, "real_width", self->video_codec->width );
1858 mlt_properties_set_int( frame_properties, "real_height", self->video_codec->height );
1859 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1860 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
1862 // Workaround 1088 encodings missing cropping info.
1863 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1865 mlt_properties_set_int( properties, "meta.media.height", 1080 );
1866 mlt_properties_set_int( frame_properties, "real_height", 1080 );
1869 // Add our image operation
1870 mlt_frame_push_service( frame, self );
1871 mlt_frame_push_get_image( frame, producer_get_image );
1875 // If something failed, use test card image
1876 mlt_properties_set_int( frame_properties, "test_image", 1 );
1880 static int seek_audio( producer_avformat self, mlt_position position, double timecode, int *ignore )
1884 // Seek if necessary
1885 if ( position != self->audio_expected )
1887 if ( position + 1 == self->audio_expected )
1889 // We're paused - silence required
1892 else if ( !self->seekable && position > self->audio_expected && ( position - self->audio_expected ) < 250 )
1894 // Fast forward - seeking is inefficient for small distances - just ignore following frames
1895 *ignore = position - self->audio_expected;
1897 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
1899 AVFormatContext *context = self->audio_format;
1900 int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
1901 if ( context->start_time != AV_NOPTS_VALUE )
1902 timestamp += context->start_time;
1903 if ( timestamp < 0 )
1906 // Set to the real timecode
1907 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
1910 // Clear the usage in the audio buffer
1911 int i = MAX_AUDIO_STREAMS + 1;
1913 self->audio_used[i - 1] = 0;
1919 static int sample_bytes( AVCodecContext *context )
1921 #if LIBAVCODEC_VERSION_MAJOR > 52
1922 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
1924 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1928 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
1930 // Fetch the audio_format
1931 AVFormatContext *context = self->audio_format;
1933 // Get the current stream index
1934 int index = pkt.stream_index;
1936 // Get codec context
1937 AVCodecContext *codec_context = self->audio_codec[ index ];
1939 // Obtain the resample context if it exists (not always needed)
1940 ReSampleContext *resample = self->audio_resample[ index ];
1942 // Obtain the audio buffers
1943 uint8_t *audio_buffer = self->audio_buffer[ index ];
1944 uint8_t *decode_buffer = self->decode_buffer[ index ];
1946 int audio_used = self->audio_used[ index ];
1947 uint8_t *ptr = pkt.data;
1951 while ( ptr && ret >= 0 && len > 0 )
1953 int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context );
1954 int data_size = self->audio_buffer_size[ index ];
1957 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1958 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
1959 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
1960 ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
1962 ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len );
1966 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
1970 pkt.size = len -= ret;
1971 pkt.data = ptr += ret;
1973 // If decoded successfully
1974 if ( data_size > 0 )
1976 // Figure out how many samples will be needed after resampling
1977 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
1978 int samples_needed = self->resample_factor * convert_samples;
1980 // Resize audio buffer to prevent overflow
1981 if ( ( audio_used + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
1983 self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample;
1984 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
1988 // Copy to audio buffer while resampling
1989 uint8_t *source = decode_buffer;
1990 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
1991 audio_used += audio_resample( resample, (short*) dest, (short*) source, convert_samples );
1995 // Straight copy to audio buffer
1996 memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size );
1997 audio_used += convert_samples;
2001 while ( *ignore && audio_used > samples )
2004 audio_used -= samples;
2005 memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ],
2006 audio_used * sizeof_sample );
2011 // If we're behind, ignore this packet
2014 double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt.pts;
2015 int req_position = ( int )( timecode * fps + 0.5 );
2016 int int_position = ( int )( current_pts * fps + 0.5 );
2017 if ( context->start_time != AV_NOPTS_VALUE )
2018 int_position -= ( int )( fps * context->start_time / AV_TIME_BASE + 0.5 );
2020 if ( self->seekable && *ignore == 0 )
2022 if ( int_position < req_position )
2023 // We are behind, so skip some
2025 else if ( int_position > req_position + 2 )
2026 // We are ahead, so seek backwards some more
2027 seek_audio( self, req_position, timecode - 1.0, ignore );
2031 self->audio_used[ index ] = audio_used;
2036 /** Get the audio from a frame.
2038 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2041 producer_avformat self = mlt_frame_pop_audio( frame );
2043 pthread_mutex_lock( &self->audio_mutex );
2045 // Obtain the frame number of this frame
2046 mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" );
2048 // Calculate the real time code
2049 double real_timecode = producer_time_of_frame( self->parent, position );
2051 // Get the producer fps
2052 double fps = mlt_producer_get_fps( self->parent );
2054 // Number of frames to ignore (for ffwd)
2057 // Flag for paused (silence)
2058 int paused = seek_audio( self, position, real_timecode, &ignore );
2060 // Fetch the audio_format
2061 AVFormatContext *context = self->audio_format;
2063 int sizeof_sample = sizeof( int16_t );
2065 // Determine the tracks to use
2066 int index = self->audio_index;
2067 int index_max = self->audio_index + 1;
2068 if ( self->audio_index == INT_MAX )
2071 index_max = context->nb_streams;
2072 *channels = self->total_channels;
2073 *samples = *samples * FFMAX( self->max_frequency, *frequency ) / *frequency;
2074 *frequency = FFMAX( self->max_frequency, *frequency );
2077 // Initialize the resamplers and buffers
2078 for ( ; index < index_max; index++ )
2080 // Get codec context
2081 AVCodecContext *codec_context = self->audio_codec[ index ];
2083 if ( codec_context && !self->audio_buffer[ index ] )
2085 // Check for resample and create if necessary
2086 if ( codec_context->channels <= 2 )
2088 // Determine by how much resampling will increase number of samples
2089 double resample_factor = self->audio_index == INT_MAX ? 1 : (double) *channels / codec_context->channels;
2090 resample_factor *= (double) *frequency / codec_context->sample_rate;
2091 if ( resample_factor > self->resample_factor )
2092 self->resample_factor = resample_factor;
2094 // Create the resampler
2095 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0))
2096 self->audio_resample[ index ] = av_audio_resample_init(
2097 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2098 codec_context->channels, *frequency, codec_context->sample_rate,
2099 SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 );
2101 self->audio_resample[ index ] = audio_resample_init(
2102 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2103 codec_context->channels, *frequency, codec_context->sample_rate );
2108 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2109 sizeof_sample = sample_bytes( codec_context );
2112 // Check for audio buffer and create if necessary
2113 self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2114 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2116 // Check for decoder buffer and create if necessary
2117 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2121 // Get the audio if required
2128 av_init_packet( &pkt );
2130 // If not resampling, give consumer more than requested.
2131 // It requested number samples based on requested frame rate.
2132 // Do not clean this up with a samples *= ...!
2133 if ( self->audio_index != INT_MAX && ! self->audio_resample[ self->audio_index ] )
2134 *samples = *samples * self->audio_codec[ self->audio_index ]->sample_rate / *frequency;
2136 while ( ret >= 0 && !got_audio )
2138 // Check if the buffer already contains the samples required
2139 if ( self->audio_index != INT_MAX && self->audio_used[ self->audio_index ] >= *samples && ignore == 0 )
2146 ret = av_read_frame( context, &pkt );
2148 // We only deal with audio from the selected audio index
2149 index = pkt.stream_index;
2150 if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2151 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2153 int channels2 = ( self->audio_index == INT_MAX || !self->audio_resample[index] ) ?
2154 self->audio_codec[index]->channels : *channels;
2155 ret = decode_audio( self, &ignore, pkt, channels2, *samples, real_timecode, fps );
2157 av_free_packet( &pkt );
2159 if ( self->audio_index == INT_MAX && ret >= 0 )
2161 // Determine if there is enough audio for all streams
2163 for ( index = 0; index < context->nb_streams; index++ )
2165 if ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples )
2171 // Set some additional return values
2172 *format = mlt_audio_s16;
2173 if ( self->audio_index != INT_MAX && !self->audio_resample[ self->audio_index ] )
2175 index = self->audio_index;
2176 *channels = self->audio_codec[ index ]->channels;
2177 *frequency = self->audio_codec[ index ]->sample_rate;
2178 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2179 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2181 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2183 else if ( self->audio_index == INT_MAX )
2185 // This only works if all audio tracks have the same sample format.
2186 for ( index = 0; index < index_max; index++ )
2187 if ( self->audio_codec[ index ] && !self->audio_resample[ index ] )
2189 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2190 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2192 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2197 // Allocate and set the frame's audio buffer
2198 int size = mlt_audio_format_size( *format, *samples, *channels );
2199 *buffer = mlt_pool_alloc( size );
2200 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2202 // Interleave tracks if audio_index=all
2203 if ( self->audio_index == INT_MAX )
2205 uint8_t *dest = *buffer;
2207 for ( i = 0; i < *samples; i++ )
2209 for ( index = 0; index < index_max; index++ )
2210 if ( self->audio_codec[ index ] )
2212 int current_channels = self->audio_codec[ index ]->channels;
2213 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2214 memcpy( dest, src, current_channels * sizeof_sample );
2215 dest += current_channels * sizeof_sample;
2218 for ( index = 0; index < index_max; index++ )
2219 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2221 int current_channels = self->audio_codec[ index ]->channels;
2222 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2223 self->audio_used[index] -= *samples;
2224 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2227 // Copy a single track to the output buffer
2230 index = self->audio_index;
2232 // Now handle the audio if we have enough
2233 if ( self->audio_used[ index ] > 0 )
2235 uint8_t *src = self->audio_buffer[ index ];
2236 // copy samples from audio_buffer
2237 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2238 memcpy( *buffer, src, size * *channels * sizeof_sample );
2239 // supply the remaining requested samples as silence
2240 if ( *samples > self->audio_used[ index ] )
2241 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2242 // reposition the samples within audio_buffer
2243 self->audio_used[ index ] -= size;
2244 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2248 // Otherwise fill with silence
2249 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2255 // Get silence and don't touch the context
2256 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2259 // Regardless of speed (other than paused), we expect to get the next frame
2261 self->audio_expected = position + 1;
2263 pthread_mutex_unlock( &self->audio_mutex );
2268 /** Initialize the audio codec context.
2271 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2273 // Initialise the codec if necessary
2274 if ( !self->audio_codec[ index ] )
2276 // Get codec context
2277 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2280 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2282 // If we don't have a codec and we can't initialise it, we can't do much more...
2284 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2286 // Now store the codec with its destructor
2287 if ( self->audio_codec[ index ] )
2288 avcodec_close( self->audio_codec[ index ] );
2289 self->audio_codec[ index ] = codec_context;
2293 // Remember that we can't use self later
2294 self->audio_index = -1;
2298 // Process properties as AVOptions
2299 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2300 #if LIBAVCODEC_VERSION_MAJOR > 52
2301 if ( codec && codec->priv_class && codec_context->priv_data )
2302 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2305 return self->audio_codec[ index ] && self->audio_index > -1;
2308 /** Set up audio handling.
2311 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2314 mlt_producer producer = self->parent;
2316 // Get the properties
2317 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2319 // Fetch the audio format context
2320 AVFormatContext *context = self->audio_format;
2322 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2324 // Get the audio_index
2325 int index = mlt_properties_get_int( properties, "audio_index" );
2327 // Handle all audio tracks
2328 if ( self->audio_index > -1 &&
2329 mlt_properties_get( properties, "audio_index" ) &&
2330 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2333 // Reopen the file if necessary
2334 if ( !context && self->audio_index > -1 && index > -1 )
2336 mlt_events_block( properties, producer );
2337 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2338 mlt_properties_get( properties, "resource" ) );
2339 context = self->audio_format;
2340 if ( self->dummy_context )
2342 av_close_input_file( self->dummy_context );
2343 self->dummy_context = NULL;
2345 mlt_events_unblock( properties, producer );
2346 if ( self->audio_format )
2347 get_audio_streams_info( self );
2350 // Exception handling for audio_index
2351 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2353 for ( index = context->nb_streams - 1;
2354 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2356 mlt_properties_set_int( properties, "audio_index", index );
2358 if ( context && index > -1 && index < INT_MAX &&
2359 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2361 index = self->audio_index;
2362 mlt_properties_set_int( properties, "audio_index", index );
2365 // Update the audio properties if the index changed
2366 if ( context && index > -1 && index != self->audio_index )
2368 if ( self->audio_codec[ self->audio_index ] )
2371 avcodec_close( self->audio_codec[ self->audio_index ] );
2374 self->audio_codec[ self->audio_index ] = NULL;
2376 if ( self->audio_index != -1 )
2377 self->audio_index = index;
2382 if ( context && index == INT_MAX )
2384 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2385 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2386 for ( index = 0; index < context->nb_streams; index++ )
2388 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2389 audio_codec_init( self, index, properties );
2392 else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2394 // Set the frame properties
2395 if ( index < INT_MAX )
2397 mlt_properties_set_int( frame_properties, "frequency", self->audio_codec[ index ]->sample_rate );
2398 mlt_properties_set_int( frame_properties, "channels", self->audio_codec[ index ]->channels );
2401 if ( context && index > -1 )
2403 // Add our audio operation
2404 mlt_frame_push_audio( frame, self );
2405 mlt_frame_push_audio( frame, producer_get_audio );
2409 /** Our get frame implementation.
2412 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2414 // Access the private data
2415 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2416 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2417 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2422 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2423 producer->child = self;
2424 self->parent = producer;
2425 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2426 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2429 // Create an empty frame
2430 *frame = mlt_frame_init( service);
2434 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2438 mlt_cache_item_close( cache_item );
2442 // Update timecode on the frame we're creating
2443 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2445 // Set the position of this producer
2446 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( producer ) );
2449 producer_set_up_video( self, *frame );
2452 producer_set_up_audio( self, *frame );
2454 // Calculate the next timecode
2455 mlt_producer_prepare_next( producer );
2460 static void producer_avformat_close( producer_avformat self )
2462 mlt_log_debug( NULL, "producer_avformat_close\n" );
2464 av_free( self->av_frame );
2467 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2469 if ( self->audio_resample[i] )
2470 audio_resample_close( self->audio_resample[i] );
2471 mlt_pool_release( self->audio_buffer[i] );
2472 av_free( self->decode_buffer[i] );
2473 if ( self->audio_codec[i] )
2474 avcodec_close( self->audio_codec[i] );
2476 if ( self->video_codec )
2477 avcodec_close( self->video_codec );
2478 if ( self->dummy_context )
2479 av_close_input_file( self->dummy_context );
2480 if ( self->audio_format )
2481 av_close_input_file( self->audio_format );
2482 if ( self->video_format )
2483 av_close_input_file( self->video_format );
2486 vdpau_producer_close( self );
2488 if ( self->image_cache )
2489 mlt_cache_close( self->image_cache );
2490 pthread_mutex_destroy( &self->audio_mutex );
2491 pthread_mutex_destroy( &self->video_mutex );
2495 static void producer_close( mlt_producer parent )
2497 // Remove this instance from the cache
2498 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2501 parent->close = NULL;
2502 mlt_producer_close( parent );