]> git.sesse.net Git - mlt/blobdiff - src/modules/avformat/producer_avformat.c
refactor avformat producer_open
[mlt] / src / modules / avformat / producer_avformat.c
index d914dab077e965af934baaa7cd7f22d9aad3890c..c593d210642ab81a8e82c669e7fd92d7132b19e2 100644 (file)
@@ -96,9 +96,9 @@ struct producer_avformat_s
        int current_position;
        int got_picture;
        int top_field_first;
-       int16_t *audio_buffer[ MAX_AUDIO_STREAMS ];
+       uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
        size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
-       int16_t *decode_buffer[ MAX_AUDIO_STREAMS ];
+       uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
        int audio_used[ MAX_AUDIO_STREAMS ];
        int audio_streams;
        int audio_max_stream;
@@ -129,7 +129,8 @@ struct producer_avformat_s
 typedef struct producer_avformat_s *producer_avformat;
 
 // Forward references.
-static int producer_open( producer_avformat self, mlt_profile profile, char *file );
+static int list_components( char* file );
+static int producer_open( producer_avformat self, mlt_profile profile, const char *URL );
 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
 static void producer_avformat_close( producer_avformat );
 static void producer_close( mlt_producer parent );
@@ -147,45 +148,17 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p
 
 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
 {
-       int skip = 0;
+       if ( list_components( file ) )
+               return NULL;
 
-       // Report information about available demuxers and codecs as YAML Tiny
-       if ( file && strstr( file, "f-list" ) )
-       {
-               fprintf( stderr, "---\nformats:\n" );
-               AVInputFormat *format = NULL;
-               while ( ( format = av_iformat_next( format ) ) )
-                       fprintf( stderr, "  - %s\n", format->name );
-               fprintf( stderr, "...\n" );
-               skip = 1;
-       }
-       if ( file && strstr( file, "acodec-list" ) )
-       {
-               fprintf( stderr, "---\naudio_codecs:\n" );
-               AVCodec *codec = NULL;
-               while ( ( codec = av_codec_next( codec ) ) )
-                       if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
-                               fprintf( stderr, "  - %s\n", codec->name );
-               fprintf( stderr, "...\n" );
-               skip = 1;
-       }
-       if ( file && strstr( file, "vcodec-list" ) )
-       {
-               fprintf( stderr, "---\nvideo_codecs:\n" );
-               AVCodec *codec = NULL;
-               while ( ( codec = av_codec_next( codec ) ) )
-                       if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
-                               fprintf( stderr, "  - %s\n", codec->name );
-               fprintf( stderr, "...\n" );
-               skip = 1;
-       }
+       mlt_producer producer = NULL;
 
        // Check that we have a non-NULL argument
-       if ( !skip && file )
+       if ( file )
        {
                // Construct the producer
-               mlt_producer producer = calloc( 1, sizeof( struct mlt_producer_s ) );
                producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
+               producer = calloc( 1, sizeof( struct mlt_producer_s ) );
 
                // Initialise it
                if ( mlt_producer_init( producer, self ) == 0 )
@@ -203,7 +176,7 @@ mlt_producer producer_avformat_init( mlt_profile profile, const char *service, c
 
                        // Register our get_frame implementation
                        producer->get_frame = producer_get_frame;
-                       
+
                        if ( strcmp( service, "avformat-novalidate" ) )
                        {
                                // Open the file
@@ -227,38 +200,76 @@ mlt_producer producer_avformat_init( mlt_profile profile, const char *service, c
                                                av_close_input_file( self->video_format );
                                        self->video_format = NULL;
                                        avformat_unlock();
-       
+
                                        // Default the user-selectable indices from the auto-detected indices
                                        mlt_properties_set_int( properties, "audio_index",  self->audio_index );
                                        mlt_properties_set_int( properties, "video_index",  self->video_index );
-                                       
-#ifdef VDPAU
-                                       mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
-#endif
-                                       mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
                                }
                        }
-                       else
+                       if ( producer )
                        {
 #ifdef VDPAU
                                mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
 #endif
                                mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
                        }
-                       return producer;
                }
        }
-       return NULL;
+       return producer;
+}
+
+int list_components( char* file )
+{
+       int skip = 0;
+
+       // Report information about available demuxers and codecs as YAML Tiny
+       if ( file && strstr( file, "f-list" ) )
+       {
+               fprintf( stderr, "---\nformats:\n" );
+               AVInputFormat *format = NULL;
+               while ( ( format = av_iformat_next( format ) ) )
+                       fprintf( stderr, "  - %s\n", format->name );
+               fprintf( stderr, "...\n" );
+               skip = 1;
+       }
+       if ( file && strstr( file, "acodec-list" ) )
+       {
+               fprintf( stderr, "---\naudio_codecs:\n" );
+               AVCodec *codec = NULL;
+               while ( ( codec = av_codec_next( codec ) ) )
+                       if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
+                               fprintf( stderr, "  - %s\n", codec->name );
+               fprintf( stderr, "...\n" );
+               skip = 1;
+       }
+       if ( file && strstr( file, "vcodec-list" ) )
+       {
+               fprintf( stderr, "---\nvideo_codecs:\n" );
+               AVCodec *codec = NULL;
+               while ( ( codec = av_codec_next( codec ) ) )
+                       if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
+                               fprintf( stderr, "  - %s\n", codec->name );
+               fprintf( stderr, "...\n" );
+               skip = 1;
+       }
+
+       return skip;
 }
 
 /** Find the default streams.
 */
 
-static mlt_properties find_default_streams( mlt_properties meta_media, AVFormatContext *context, int *audio_index, int *video_index )
+static mlt_properties find_default_streams( producer_avformat self )
 {
        int i;
        char key[200];
        AVMetadataTag *tag = NULL;
+       AVFormatContext *context = self->video_format;
+       mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
+
+       // Default to the first audio and video streams found
+       self->audio_index = -1;
+       self->video_index = -1;
 
        mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
 
@@ -279,8 +290,9 @@ static mlt_properties find_default_streams( mlt_properties meta_media, AVFormatC
                switch( codec_context->codec_type )
                {
                        case CODEC_TYPE_VIDEO:
-                               if ( *video_index < 0 )
-                                       *video_index = i;
+                               // Use first video stream
+                               if ( self->video_index < 0 )
+                                       self->video_index = i;
                                mlt_properties_set( meta_media, key, "video" );
                                snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
@@ -325,8 +337,9 @@ static mlt_properties find_default_streams( mlt_properties meta_media, AVFormatC
 #endif
                                break;
                        case CODEC_TYPE_AUDIO:
-                               if ( *audio_index < 0 )
-                                       *audio_index = i;
+                               // Use first audio stream
+                               if ( self->audio_index < 0 )
+                                       self->audio_index = i;
                                mlt_properties_set( meta_media, key, "audio" );
 #if LIBAVCODEC_VERSION_MAJOR > 52
                                snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
@@ -528,249 +541,262 @@ static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVC
        return aspect_ratio;
 }
 
-/** Open the file.
-*/
-
-static int producer_open( producer_avformat self, mlt_profile profile, char *file )
+static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters **parameters )
 {
-       // Return an error code (0 == no error)
-       int error = 0;
+       if ( !URL ) return NULL;
 
-       // Context for avformat
-       AVFormatContext *context = NULL;
-
-       // Get the properties
-       mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
-
-       // We will treat everything with the producer fps
-       double fps = mlt_profile_fps( profile );
-
-       // Lock the service
-       pthread_mutex_init( &self->audio_mutex, NULL );
-       pthread_mutex_init( &self->video_mutex, NULL );
-       pthread_mutex_lock( &self->audio_mutex );
-       pthread_mutex_lock( &self->video_mutex );
-
-       // If "MRL", then create AVInputFormat
-       AVInputFormat *format = NULL;
-       AVFormatParameters *params = NULL;
-       char *standard = NULL;
-       char *mrl = strchr( file, ':' );
-
-       // AV option (0 = both, 1 = video, 2 = audio)
-       int av = 0;
+       char *result = NULL;
+       char *protocol = strdup( URL );
+       char *url = strchr( protocol, ':' );
 
        // Only if there is not a protocol specification that avformat can handle
 #if LIBAVFORMAT_VERSION_MAJOR > 52
-       if ( mrl && !avio_check( file, 0 ) )
+       if ( url && avio_check( URL, 0 ) < 0 )
 #else
-       if ( mrl && !url_exist( file ) )
+       if ( url && !url_exist( URL ) )
 #endif
        {
-               // 'file' becomes format abbreviation
-               mrl[0] = 0;
+               // Truncate protocol string
+               url[0] = 0;
 
                // Lookup the format
-               format = av_find_input_format( file );
+               *format = av_find_input_format( protocol );
 
                // Eat the format designator
-               file = ++mrl;
+               result = ++url;
 
-               if ( format )
+               if ( *format )
                {
                        // Allocate params
-                       params = calloc( sizeof( AVFormatParameters ), 1 );
+                       AVFormatParameters *params = *parameters = calloc( 1, sizeof( AVFormatParameters ) );
 
-                       // These are required by video4linux (defaults)
-                       params->width = 640;
-                       params->height = 480;
-                       params->time_base= (AVRational){1,25};
+                       // These are required by video4linux2 (defaults)
+                       params->width = profile->width;
+                       params->height = profile->height;
+                       params->time_base= (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
                        params->channels = 2;
                        params->sample_rate = 48000;
-               }
 
-               // Parse out params
-               mrl = strchr( file, '?' );
-               while ( mrl )
-               {
-                       mrl[0] = 0;
-                       char *name = strdup( ++mrl );
-                       char *value = strchr( name, ':' );
-                       if ( value )
+                       // Parse out params
+                       url = strchr( url, '?' );
+                       while ( url )
                        {
-                               value[0] = 0;
-                               value++;
-                               char *t = strchr( value, '&' );
-                               if ( t )
-                                       t[0] = 0;
-                               if ( !strcmp( name, "frame_rate" ) )
-                                       params->time_base.den = atoi( value );
-                               else if ( !strcmp( name, "frame_rate_base" ) )
-                                       params->time_base.num = atoi( value );
-                               else if ( !strcmp( name, "sample_rate" ) )
-                                       params->sample_rate = atoi( value );
-                               else if ( !strcmp( name, "channel" ) )
-                                       params->channel = atoi( value );
-                               else if ( !strcmp( name, "channels" ) )
-                                       params->channels = atoi( value );
+                               url[0] = 0;
+                               char *name = strdup( ++url );
+                               char *value = strchr( name, ':' );
+                               if ( value )
+                               {
+                                       value[0] = 0;
+                                       value++;
+                                       char *t = strchr( value, '&' );
+                                       if ( t )
+                                               t[0] = 0;
+                                       if ( !strcmp( name, "frame_rate" ) )
+                                               params->time_base.den = atoi( value );
+                                       else if ( !strcmp( name, "frame_rate_base" ) )
+                                               params->time_base.num = atoi( value );
+                                       else if ( !strcmp( name, "sample_rate" ) )
+                                               params->sample_rate = atoi( value );
+                                       else if ( !strcmp( name, "channel" ) )
+                                               params->channel = atoi( value );
+                                       else if ( !strcmp( name, "channels" ) )
+                                               params->channels = atoi( value );
 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
-                               else if ( !strcmp( name, "pix_fmt" ) )
-                                       params->pix_fmt = av_get_pix_fmt( value );
+                                       else if ( !strcmp( name, "pix_fmt" ) )
+                                               params->pix_fmt = av_get_pix_fmt( value );
 #endif
-                               else if ( !strcmp( name, "width" ) )
-                                       params->width = atoi( value );
-                               else if ( !strcmp( name, "height" ) )
-                                       params->height = atoi( value );
-                               else if ( !strcmp( name, "standard" ) )
-                               {
-                                       standard = strdup( value );
-                                       params->standard = standard;
+                                       else if ( !strcmp( name, "width" ) )
+                                               params->width = atoi( value );
+                                       else if ( !strcmp( name, "height" ) )
+                                               params->height = atoi( value );
+                                       else if ( !strcmp( name, "standard" ) )
+                                               params->standard = strdup( value );
                                }
-                               else if ( !strcmp( name, "av" ) )
-                                       av = atoi( value );
+                               free( name );
+                               url = strchr( url, '&' );
                        }
-                       free( name );
-                       mrl = strchr( mrl, '&' );
                }
        }
+       free( protocol );
+       return result ? strdup( result ) : strdup( URL );
+}
 
-       // Now attempt to open the file
-       error = av_open_input_file( &context, file, format, 0, params ) < 0;
-
-       // Cleanup AVFormatParameters
-       free( standard );
-       free( params );
+static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
+{
+       int error = 0;
 
-       // If successful, then try to get additional info
-       if ( !error )
-       {
-               // Get the stream info
-               error = av_find_stream_info( context ) < 0;
+       // Get the properties
+       mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
 
-               // Continue if no error
-               if ( !error )
-               {
-                       // We will default to the first audio and video streams found
-                       int audio_index = -1;
-                       int video_index = -1;
+       AVFormatContext *format = self->video_format;
 
-                       // Find default audio and video streams
-                       find_default_streams( properties, context, &audio_index, &video_index );
+       // We will treat everything with the producer fps.
+       // TODO: make this more flexible.
+       double fps = mlt_profile_fps( profile );
 
-                       // Now set properties where we can (use default unknowns if required)
-                       if ( context->duration != AV_NOPTS_VALUE )
-                       {
-                               // This isn't going to be accurate for all formats
-                               // Workaround some clips whose estimated duration cause problems:
-                               // http://www.kdenlive.org/mantis/view.php?id=2003
-                               int adjust = -3;
-                               if ( mlt_properties_get( properties, "adjust_length" ) )
-                                       adjust = mlt_properties_get_int( properties, "adjust_length" );
-                               mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps + adjust );
-                               if  ( mlt_properties_get_position( properties, "force_length" ) > 0 )
-                                       frames = mlt_properties_get_position( properties, "force_length" );
-                               mlt_properties_set_position( properties, "out", frames - 1 );
-                               mlt_properties_set_position( properties, "length", frames );
-                       }
+       // Get the duration
+       if ( !mlt_properties_get_int( properties, "_length_computed" ) )
+       {
+               // The _length_computed flag prevents overwriting explicity set length/out/eof properties
+               // when producer_open is called after initial call when restoring or reseting the producer.
+               if ( format->duration != AV_NOPTS_VALUE )
+               {
+                       // This isn't going to be accurate for all formats
+                       mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
+                       mlt_properties_set_position( properties, "out", frames - 1 );
+                       mlt_properties_set_position( properties, "length", frames );
+                       mlt_properties_set_int( properties, "_length_computed", 1 );
+               }
+               else
+               {
+                       // Set live sources to run forever
+                       mlt_properties_set_position( properties, "length", INT_MAX );
+                       mlt_properties_set_position( properties, "out", INT_MAX - 1 );
+                       mlt_properties_set( properties, "eof", "loop" );
+                       mlt_properties_set_int( properties, "_length_computed", 1 );
+               }
+       }
 
-                       if ( context->start_time != AV_NOPTS_VALUE )
-                               self->start_time = context->start_time;
+       if ( format->start_time != AV_NOPTS_VALUE )
+               self->start_time = format->start_time;
 
-                       // Check if we're seekable
-                       // avdevices are typically AVFMT_NOFILE and not seekable
-                       self->seekable = !format || !( format->flags & AVFMT_NOFILE );
-                       if ( context->pb )
-                       {
-                               // protocols can indicate if they support seeking
+       // Check if we're seekable
+       // avdevices are typically AVFMT_NOFILE and not seekable
+       self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
+       if ( format->pb )
+       {
+               // protocols can indicate if they support seeking
 #if LIBAVFORMAT_VERSION_MAJOR > 52
-                               self->seekable = context->pb->seekable;
+               self->seekable = context->pb->seekable;
 #else
-                               URLContext *uc = url_fileno( context->pb );
-                               if ( uc )
-                                       self->seekable = !uc->is_streamed;
+               URLContext *uc = url_fileno( format->pb );
+               if ( uc )
+                       self->seekable = !uc->is_streamed;
 #endif
-                       }
-                       if ( self->seekable )
-                       {
-                               self->seekable = av_seek_frame( context, -1, self->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
-                               mlt_properties_set_int( properties, "seekable", self->seekable );
-                               self->dummy_context = context;
-                               av_open_input_file( &context, file, NULL, 0, NULL );
-                               av_find_stream_info( context );
-                       }
+       }
+       if ( self->seekable )
+       {
+               // Do a more rigourous test of seekable on a disposable context
+               self->seekable = av_seek_frame( format, -1, self->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
+               mlt_properties_set_int( properties, "seekable", self->seekable );
+               self->dummy_context = format;
+               av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
+               format = self->video_format;
+               av_find_stream_info( format );
+       }
 
-                       // Store selected audio and video indexes on properties
-                       self->audio_index = audio_index;
-                       self->video_index = video_index;
-                       self->first_pts = -1;
-                       self->last_position = POSITION_INITIAL;
+       // Fetch the width, height and aspect ratio
+       if ( self->video_index != -1 )
+       {
+               AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
+               mlt_properties_set_int( properties, "width", codec_context->width );
+               mlt_properties_set_int( properties, "height", codec_context->height );
 
-                       // Fetch the width, height and aspect ratio
-                       if ( video_index != -1 )
+               if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
+               {
+                       // Fetch the first frame of DV so we can read it directly
+                       AVPacket pkt;
+                       int ret = 0;
+                       while ( ret >= 0 )
                        {
-                               AVCodecContext *codec_context = context->streams[ video_index ]->codec;
-                               mlt_properties_set_int( properties, "width", codec_context->width );
-                               mlt_properties_set_int( properties, "height", codec_context->height );
-
-                               if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
+                               ret = av_read_frame( format, &pkt );
+                               if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
                                {
-                                       // Fetch the first frame of DV so we can read it directly
-                                       AVPacket pkt;
-                                       int ret = 0;
-                                       while ( ret >= 0 )
-                                       {
-                                               ret = av_read_frame( context, &pkt );
-                                               if ( ret >= 0 && pkt.stream_index == video_index && pkt.size > 0 )
-                                               {
-                                                       get_aspect_ratio( properties, context->streams[ video_index ], codec_context, &pkt );
-                                                       break;
-                                               }
-                                       }
-                               }
-                               else
-                               {
-                                       get_aspect_ratio( properties, context->streams[ video_index ], codec_context, NULL );
+                                       get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, &pkt );
+                                       break;
                                }
+                       }
+               }
+               else
+               {
+                       get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, NULL );
+               }
+
 #ifdef SWSCALE
-                               struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
-                                       codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
-                               if ( context )
-                                       sws_freeContext( context );
-                               else
-                                       error = 1;
+               // Verify that we can convert this to YUV 4:2:2
+               // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
+               struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
+                       codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
+               if ( context )
+                       sws_freeContext( context );
+               else
+                       error = 1;
 #endif
-                       }
+       }
+       return error;
+}
 
-                       // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
-                       if ( av == 0 && audio_index != -1 && video_index != -1 )
-                       {
-                               // We'll use the open one as our video_format
-                               self->video_format = context;
+/** Open the file.
+*/
 
-                               // And open again for our audio context
-                               av_open_input_file( &context, file, NULL, 0, NULL );
-                               av_find_stream_info( context );
+static int producer_open( producer_avformat self, mlt_profile profile, const char *URL )
+{
+       // Return an error code (0 == no error)
+       int error = 0;
 
-                               // Audio context
-                               self->audio_format = context;
-                       }
-                       else if ( av != 2 && video_index != -1 )
+       // Lock the service
+       pthread_mutex_init( &self->audio_mutex, NULL );
+       pthread_mutex_init( &self->video_mutex, NULL );
+       pthread_mutex_lock( &self->audio_mutex );
+       pthread_mutex_lock( &self->video_mutex );
+
+       // Parse URL
+       AVInputFormat *format = NULL;
+       AVFormatParameters *params = NULL;
+       char *filename = parse_url( profile, URL, &format, &params );
+
+       // Now attempt to open the file
+       error = av_open_input_file( &self->video_format, filename, format, 0, params ) < 0;
+
+       // Cleanup AVFormatParameters
+       if ( params )
+       {
+               if ( params->standard )
+                       free( (void*) params->standard );
+               free( params );
+       }
+
+       // If successful, then try to get additional info
+       if ( !error )
+       {
+               // Get the stream info
+               error = av_find_stream_info( self->video_format ) < 0;
+
+               // Continue if no error
+               if ( !error )
+               {
+                       // Find default audio and video streams
+                       find_default_streams( self );
+                       error = get_basic_info( self, profile, filename );
+
+                       // Initialize position info
+                       self->first_pts = -1;
+                       self->last_position = POSITION_INITIAL;
+
+                       // We're going to cheat here - for seekable A/V files, we will have separate contexts
+                       // to support independent seeking of audio from video.
+                       // TODO: Is this really necessary?
+                       if ( self->audio_index != -1 && self->video_index != -1 )
                        {
-                               // We only have a video context
-                               self->video_format = context;
+                               // And open again for our audio context
+                               av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
+                               av_find_stream_info( self->audio_format );
                        }
-                       else if ( audio_index != -1 )
+                       else if ( self->audio_index != -1 )
                        {
                                // We only have an audio context
-                               self->audio_format = context;
+                               self->audio_format = self->video_format;
+                               self->video_format = NULL;
                        }
-                       else
+                       else if ( self->video_index == -1 )
                        {
                                // Something has gone wrong
                                error = -1;
                        }
                }
        }
+       if ( filename )
+               free( filename );
 
        // Unlock the service
        pthread_mutex_unlock( &self->audio_mutex );
@@ -815,6 +841,10 @@ void reopen_video( producer_avformat self, mlt_producer producer, mlt_properties
        }
        mlt_events_unblock( properties, producer );
        apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
+#if LIBAVFORMAT_VERSION_MAJOR > 52
+       if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
+               apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
+#endif
 
        self->audio_index = audio_index;
        if ( self->video_format && video_index > -1 )
@@ -1084,8 +1114,8 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form
        if ( self->image_cache )
        {
                mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position );
-               *buffer = mlt_cache_item_data( item, (int*) format );
-               if ( *buffer )
+               uint8_t *original = mlt_cache_item_data( item, (int*) format );
+               if ( original )
                {
                        // Set the resolution
                        *width = codec_context->width;
@@ -1097,9 +1127,19 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form
 
                        // Cache hit
                        int size = mlt_image_format_size( *format, *width, *height, NULL );
-                       mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
-                       mlt_frame_set_image( frame, *buffer, size, NULL );
-                       // self->top_field_first = mlt_properties_get_int( frame_properties, "top_field_first" );
+                       if ( writable )
+                       {
+                               *buffer = mlt_pool_alloc( size );
+                               mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
+                               memcpy( *buffer, original, size );
+                               mlt_cache_item_close( item );
+                       }
+                       else
+                       {
+                               *buffer = original;
+                               mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
+                               mlt_frame_set_image( frame, *buffer, size, NULL );
+                       }
                        self->got_picture = 1;
 
                        goto exit_get_image;
@@ -1276,7 +1316,6 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form
                int decode_errors = 0;
                int got_picture = 0;
 
-               self->got_picture = 0;
                av_init_packet( &pkt );
 
                // Construct an AVFrame for YUV422 conversion
@@ -1447,13 +1486,13 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form
                                                        else
                                                        {
                                                                mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
-                                                               self->vdpau->is_decoded = 0;
+                                                               image_size = self->vdpau->is_decoded = 0;
                                                        }
                                                }
                                                else
                                                {
                                                        mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
-                                                       got_picture = 0;
+                                                       image_size = got_picture = 0;
                                                }
                                        }
                                        else
@@ -1482,7 +1521,7 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form
                mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release );
        }
        // Try to duplicate last image if there was a decoding failure
-       else if ( !self->got_picture && self->av_frame && self->av_frame->linesize[0] )
+       else if ( !image_size && self->av_frame && self->av_frame->linesize[0] )
        {
                // Duplicate it
                if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
@@ -1624,6 +1663,10 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p
 
                // Process properties as AVOptions
                apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
+#if LIBAVCODEC_VERSION_MAJOR > 52
+               if ( codec->priv_class && codec_context->priv_data )
+                       apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
+#endif
 
                // Reset some image properties
                mlt_properties_set_int( properties, "width", self->video_codec->width );
@@ -1755,7 +1798,13 @@ static void producer_set_up_video( producer_avformat self, mlt_frame frame )
 
                // Process properties as AVOptions
                if ( context )
+               {
                        apply_properties( context, properties, AV_OPT_FLAG_DECODING_PARAM );
+#if LIBAVFORMAT_VERSION_MAJOR > 52
+                       if ( context->iformat && context->iformat->priv_class && context->priv_data )
+                               apply_properties( context->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
+#endif
+               }
        }
 
        // Exception handling for video_index
@@ -1867,6 +1916,15 @@ static int seek_audio( producer_avformat self, mlt_position position, double tim
        return paused;
 }
 
+static int sample_bytes( AVCodecContext *context )
+{
+#if LIBAVCODEC_VERSION_MAJOR > 52
+       return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
+#else
+       return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
+#endif
+}
+
 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
 {
        // Fetch the audio_format
@@ -1882,8 +1940,8 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int
        ReSampleContext *resample = self->audio_resample[ index ];
 
        // Obtain the audio buffers
-       int16_t *audio_buffer = self->audio_buffer[ index ];
-       int16_t *decode_buffer = self->decode_buffer[ index ];
+       uint8_t *audio_buffer = self->audio_buffer[ index ];
+       uint8_t *decode_buffer = self->decode_buffer[ index ];
 
        int audio_used = self->audio_used[ index ];
        uint8_t *ptr = pkt.data;
@@ -1892,11 +1950,12 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int
 
        while ( ptr && ret >= 0 && len > 0 )
        {
-               int data_size = sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE;
+               int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context );
+               int data_size = self->audio_buffer_size[ index ];
 
                // Decode the audio
 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
-               ret = avcodec_decode_audio3( codec_context, decode_buffer, &data_size, &pkt );
+               ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
                ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
 #else
@@ -1915,30 +1974,26 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int
                if ( data_size > 0 )
                {
                        // Figure out how many samples will be needed after resampling
-#if LIBAVCODEC_VERSION_MAJOR > 52
-                       int convert_samples = data_size / codec_context->channels / ( av_get_bits_per_sample_fmt( codec_context->sample_fmt ) / 8 );
-#else
-                       int convert_samples = data_size / codec_context->channels / ( av_get_bits_per_sample_format( codec_context->sample_fmt ) / 8 );
-#endif
-                       int samples_needed = self->resample_factor * convert_samples + 1;
-                       
+                       int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
+                       int samples_needed = self->resample_factor * convert_samples;
+
                        // Resize audio buffer to prevent overflow
-                       if ( audio_used * channels + samples_needed > self->audio_buffer_size[ index ] )
+                       if ( ( audio_used + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
                        {
-                               self->audio_buffer_size[ index ] *= 2;
-                               audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] * sizeof(int16_t) );
+                               self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample;
+                               audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
                        }
                        if ( resample )
                        {
                                // Copy to audio buffer while resampling
-                               int16_t *source = decode_buffer;
-                               int16_t *dest = &audio_buffer[ audio_used * channels ];
-                               audio_used += audio_resample( resample, dest, source, convert_samples );
+                               uint8_t *source = decode_buffer;
+                               uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
+                               audio_used += audio_resample( resample, (short*) dest, (short*) source, convert_samples );
                        }
                        else
                        {
                                // Straight copy to audio buffer
-                               memcpy( &audio_buffer[ audio_used * codec_context->channels ], decode_buffer, data_size );
+                               memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size );
                                audio_used += convert_samples;
                        }
 
@@ -1947,8 +2002,8 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int
                        {
                                *ignore -= 1;
                                audio_used -= samples;
-                               memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) ],
-                                        audio_used * sizeof( int16_t ) );
+                               memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ],
+                                                audio_used * sizeof_sample );
                        }
                }
        }
@@ -2004,6 +2059,8 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
 
        // Fetch the audio_format
        AVFormatContext *context = self->audio_format;
+
+       int sizeof_sample = sizeof( int16_t );
        
        // Determine the tracks to use
        int index = self->audio_index;
@@ -2049,21 +2106,22 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
                        else
                        {
                                codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
+                               sizeof_sample = sample_bytes( codec_context );
                        }
 
                        // Check for audio buffer and create if necessary
-                       self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE;
-                       self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] * sizeof( int16_t ) );
+                       self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof_sample;
+                       self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
 
                        // Check for decoder buffer and create if necessary
-                       self->decode_buffer[ index ] = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
+                       self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
                }
        }
 
        // Get the audio if required
        if ( !paused )
        {
-               int ret = 0;
+               int ret = 0;
                int got_audio = 0;
                AVPacket pkt;
 
@@ -2088,10 +2146,12 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
                        ret = av_read_frame( context, &pkt );
 
                        // We only deal with audio from the selected audio index
-                       if ( ret >= 0 && pkt.data && pkt.size > 0 && ( pkt.stream_index == self->audio_index ||
-                                ( self->audio_index == INT_MAX && context->streams[ pkt.stream_index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
+                       index = pkt.stream_index;
+                       if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
+                                ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
                        {
-                               int channels2 = self->audio_index == INT_MAX ? self->audio_codec[pkt.stream_index]->channels : *channels;
+                               int channels2 = ( self->audio_index == INT_MAX || !self->audio_resample[index] ) ?
+                                       self->audio_codec[index]->channels : *channels;
                                ret = decode_audio( self, &ignore, pkt, channels2, *samples, real_timecode, fps );
                        }
                        av_free_packet( &pkt );
@@ -2109,22 +2169,40 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
                }
 
                // Set some additional return values
+               *format = mlt_audio_s16;
                if ( self->audio_index != INT_MAX && !self->audio_resample[ self->audio_index ] )
                {
-                       *channels  = self->audio_codec[ self->audio_index ]->channels;
-                       *frequency = self->audio_codec[ self->audio_index ]->sample_rate;
+                       index = self->audio_index;
+                       *channels = self->audio_codec[ index ]->channels;
+                       *frequency = self->audio_codec[ index ]->sample_rate;
+                       *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
+                               : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
+                               : mlt_audio_s16;
+                       sizeof_sample = sample_bytes( self->audio_codec[ index ] );
+               }
+               else if ( self->audio_index == INT_MAX )
+               {
+                       // This only works if all audio tracks have the same sample format.
+                       for ( index = 0; index < index_max; index++ )
+                               if ( self->audio_codec[ index ] && !self->audio_resample[ index ] )
+                               {
+                                       *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
+                                               : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
+                                               : mlt_audio_s16;
+                                       sizeof_sample = sample_bytes( self->audio_codec[ index ] );
+                                       break;
+                               }
                }
 
                // Allocate and set the frame's audio buffer
-               int size = *samples * *channels * sizeof(int16_t);
+               int size = mlt_audio_format_size( *format, *samples, *channels );
                *buffer = mlt_pool_alloc( size );
-               *format = mlt_audio_s16;
                mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
 
                // Interleave tracks if audio_index=all
                if ( self->audio_index == INT_MAX )
                {
-                       int16_t *dest = *buffer;
+                       uint8_t *dest = *buffer;
                        int i;
                        for ( i = 0; i < *samples; i++ )
                        {
@@ -2132,18 +2210,18 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
                                if ( self->audio_codec[ index ] )
                                {
                                        int current_channels = self->audio_codec[ index ]->channels;
-                                       int16_t *src = self->audio_buffer[ index ] + i * current_channels;
-                                       memcpy( dest, src, current_channels * sizeof(int16_t) );
-                                       dest += current_channels;
+                                       uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
+                                       memcpy( dest, src, current_channels * sizeof_sample );
+                                       dest += current_channels * sizeof_sample;
                                }
                        }
                        for ( index = 0; index < index_max; index++ )
                        if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
                        {
                                int current_channels = self->audio_codec[ index ]->channels;
-                               int16_t *src = self->audio_buffer[ index ] + *samples * current_channels;
+                               uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
                                self->audio_used[index] -= *samples;
-                               memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof(int16_t) );
+                               memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
                        }
                }
                // Copy a single track to the output buffer
@@ -2154,16 +2232,21 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format
                        // Now handle the audio if we have enough
                        if ( self->audio_used[ index ] > 0 )
                        {
-                               int16_t *src = self->audio_buffer[ index ];
-                               *samples = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
-                               memcpy( *buffer, src, *samples * *channels * sizeof(int16_t) );
-                               self->audio_used[ index ] -= *samples;
-                               memmove( src, &src[ *samples * *channels ], self->audio_used[ index ] * *channels * sizeof(int16_t) );
+                               uint8_t *src = self->audio_buffer[ index ];
+                               // copy samples from audio_buffer
+                               size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
+                               memcpy( *buffer, src, size * *channels * sizeof_sample );
+                               // supply the remaining requested samples as silence
+                               if ( *samples > self->audio_used[ index ] )
+                                       memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
+                               // reposition the samples within audio_buffer
+                               self->audio_used[ index ] -= size;
+                               memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
                        }
                        else
                        {
                                // Otherwise fill with silence
-                               memset( *buffer, 0, *samples * *channels * sizeof(int16_t) );
+                               memset( *buffer, 0, *samples * *channels * sizeof_sample );
                        }
                }
        }
@@ -2214,6 +2297,10 @@ static int audio_codec_init( producer_avformat self, int index, mlt_properties p
 
                // Process properties as AVOptions
                apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
+#if LIBAVCODEC_VERSION_MAJOR > 52
+               if ( codec && codec->priv_class && codec_context->priv_data )
+                       apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
+#endif
        }
        return self->audio_codec[ index ] && self->audio_index > -1;
 }
@@ -2294,8 +2381,8 @@ static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
        // Get the codec(s)
        if ( context && index == INT_MAX )
        {
-               mlt_properties_set_int( frame_properties, "frequency", self->max_frequency );
-               mlt_properties_set_int( frame_properties, "channels", self->total_channels );
+               mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
+               mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
                for ( index = 0; index < context->nb_streams; index++ )
                {
                        if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
@@ -2407,6 +2494,9 @@ static void producer_avformat_close( producer_avformat self )
 
 static void producer_close( mlt_producer parent )
 {
+       // Remove this instance from the cache
+       mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
+
        // Close the parent
        parent->close = NULL;
        mlt_producer_close( parent );