]> git.sesse.net Git - mlt/commitdiff
Add multitrack audio encoding.
authorDan Dennedy <dan@dennedy.org>
Thu, 13 May 2010 04:07:09 +0000 (21:07 -0700)
committerDan Dennedy <dan@dennedy.org>
Thu, 13 May 2010 04:07:09 +0000 (21:07 -0700)
This is a check point for the first working version. Changes are
forthcoming.

src/modules/avformat/consumer_avformat.c

index 9dacb50a52991f1a6ce3ca1af66b9fbc76c533d8..834f5b4262ae7d11c75afb5a8afb7d8c8e93af76 100644 (file)
@@ -46,7 +46,8 @@
 #define PIX_FMT_YUYV422 PIX_FMT_YUV422
 #endif
 
-#define AUDIO_ENCODE_BUFFER_SIZE (48000 * 2)
+#define MAX_AUDIO_STREAMS (8)
+#define AUDIO_ENCODE_BUFFER_SIZE (48000 * 2 * MAX_AUDIO_STREAMS)
 
 //
 // This structure should be extended and made globally available in mlt
@@ -142,7 +143,7 @@ static int consumer_is_stopped( mlt_consumer this );
 static void *consumer_thread( void *arg );
 static void consumer_close( mlt_consumer this );
 
-/** Initialise the dv consumer.
+/** Initialise the consumer.
 */
 
 mlt_consumer consumer_avformat_init( mlt_profile profile, char *arg )
@@ -372,13 +373,13 @@ static void apply_properties( void *obj, mlt_properties properties, int flags, i
 /** Add an audio output stream
 */
 
-static AVStream *add_audio_stream( mlt_consumer this, AVFormatContext *oc, int codec_id )
+static AVStream *add_audio_stream( mlt_consumer this, AVFormatContext *oc, int codec_id, int channels )
 {
        // Get the properties
        mlt_properties properties = MLT_CONSUMER_PROPERTIES( this );
 
        // Create a new stream
-       AVStream *st = av_new_stream( oc, 1 );
+       AVStream *st = av_new_stream( oc, oc->nb_streams );
 
        // If created, then initialise from properties
        if ( st != NULL ) 
@@ -435,7 +436,7 @@ static AVStream *add_audio_stream( mlt_consumer this, AVFormatContext *oc, int c
                // Set parameters controlled by MLT
                c->sample_rate = mlt_properties_get_int( properties, "frequency" );
                c->time_base = ( AVRational ){ 1, c->sample_rate };
-               c->channels = mlt_properties_get_int( properties, "channels" );
+               c->channels = channels;
 
                if ( mlt_properties_get( properties, "alang" ) != NULL )
                        strncpy( st->language, mlt_properties_get( properties, "alang" ), sizeof( st->language ) );
@@ -513,7 +514,7 @@ static AVStream *add_video_stream( mlt_consumer this, AVFormatContext *oc, int c
        mlt_properties properties = MLT_CONSUMER_PROPERTIES( this );
 
        // Create a new stream
-       AVStream *st = av_new_stream( oc, 0 );
+       AVStream *st = av_new_stream( oc, oc->nb_streams );
 
        if ( st != NULL ) 
        {
@@ -844,7 +845,8 @@ static void *consumer_thread( void *arg )
        mlt_image_format img_fmt = mlt_image_yuv422;
 
        // For receiving audio samples back from the fifo
-       int16_t *buffer = av_malloc( AUDIO_ENCODE_BUFFER_SIZE );
+       int16_t *audio_buf_1 = av_malloc( AUDIO_ENCODE_BUFFER_SIZE );
+       int16_t *audio_buf_2 = NULL;
        int count = 0;
 
        // Allocate the context
@@ -855,17 +857,15 @@ static void *consumer_thread( void *arg )
 #endif
 
        // Streams
-       AVStream *audio_st = NULL;
        AVStream *video_st = NULL;
+       AVStream *audio_st[ MAX_AUDIO_STREAMS ];
+       int is_audio_initialized = 0;
 
        // Time stamps
        double audio_pts = 0;
        double video_pts = 0;
 
-       // Loop variable
-       int i;
-
-       // Frames despatched
+       // Frames dispatched
        long int frames = 0;
        long int total_time = 0;
 
@@ -880,6 +880,15 @@ static void *consumer_thread( void *arg )
        int audio_codec_id;
        int video_codec_id;
 
+       // Misc
+       char key[27];
+       mlt_properties frame_meta_properties = mlt_properties_new();
+
+       // Initialize audio_st
+       int i = MAX_AUDIO_STREAMS;
+       while ( i-- )
+               audio_st[i] = NULL;
+
        // Check for user selected format first
        if ( format != NULL )
 #if LIBAVFORMAT_VERSION_INT < ((52<<16)+(45<<8)+0)
@@ -964,11 +973,19 @@ static void *consumer_thread( void *arg )
        oc->oformat = fmt;
        snprintf( oc->filename, sizeof(oc->filename), "%s", filename );
 
-       // Add audio and video streams 
+       // Get the first frame
+       frame = mlt_consumer_rt_frame( this );
+
+       // Add audio and video streams
        if ( video_codec_id != CODEC_ID_NONE )
                video_st = add_video_stream( this, oc, video_codec_id );
        if ( audio_codec_id != CODEC_ID_NONE )
-               audio_st = add_audio_stream( this, oc, audio_codec_id );
+       {
+               i = 0;
+               if ( frame )
+                       i = mlt_properties_get_int( MLT_FRAME_PROPERTIES(frame), "meta.map.audio.0.channels" );
+               audio_st[0] = add_audio_stream( this, oc, audio_codec_id, i ? i : channels );
+       }
 
        // Set the parameters (even though we have none...)
        if ( av_set_parameters(oc, NULL) >= 0 ) 
@@ -988,11 +1005,11 @@ static void *consumer_thread( void *arg )
 
                if ( video_st && !open_video( oc, video_st ) )
                        video_st = NULL;
-               if ( audio_st )
+               if ( audio_st[0] ) // Add audio streams as needed later
                {
-                       audio_input_frame_size = open_audio( oc, audio_st, audio_outbuf_size );
+                       audio_input_frame_size = open_audio( oc, audio_st[0], audio_outbuf_size );
                        if ( !audio_input_frame_size )
-                               audio_st = NULL;
+                               audio_st[0] = NULL;
                }
 
                // Open the output file, if needed
@@ -1005,8 +1022,9 @@ static void *consumer_thread( void *arg )
                        }
                }
        
-               // Write the stream header, if any
-               if ( mlt_properties_get_int( properties, "running" ) )
+               // Write the stream header now if there are no audio streams.
+               // Otherwise, we wait until we possibly add additional audio streams below.
+               if ( ! audio_st[0] && mlt_properties_get_int( properties, "running" ) )
                        av_write_header( oc );
        }
        else
@@ -1020,7 +1038,7 @@ static void *consumer_thread( void *arg )
                output = alloc_picture( video_st->codec->pix_fmt, width, height );
 
        // Last check - need at least one stream
-       if ( audio_st == NULL && video_st == NULL )
+       if ( !audio_st[0] && !video_st )
                mlt_properties_set_int( properties, "running", 0 );
 
        // Get the starting time (can ignore the times above)
@@ -1030,13 +1048,10 @@ static void *consumer_thread( void *arg )
        while( mlt_properties_get_int( properties, "running" ) &&
               ( !terminated || ( video_st && mlt_deque_count( queue ) ) ) )
        {
-               // Get the frame
-               frame = mlt_consumer_rt_frame( this );
-
                // Check that we have a frame to work with
                if ( frame != NULL )
                {
-                       // Increment frames despatched
+                       // Increment frames dispatched
                        frames ++;
 
                        // Default audio args
@@ -1046,11 +1061,36 @@ static void *consumer_thread( void *arg )
                        terminated = terminate_on_pause && mlt_properties_get_double( frame_properties, "_speed" ) == 0.0;
 
                        // Get audio and append to the fifo
-                       if ( !terminated && audio_st )
+                       if ( !terminated && audio_st[0] )
                        {
                                samples = mlt_sample_calculator( fps, frequency, count ++ );
                                mlt_frame_get_audio( frame, (void**) &pcm, &aud_fmt, &frequency, &channels, &samples );
 
+                               // Add audio streams based on channel mapping
+                               if ( ! is_audio_initialized )
+                               {
+                                       int j = 0;
+                                       for ( i = 0; i < MAX_AUDIO_STREAMS && j < channels; i++ )
+                                       {
+                                               sprintf( key, "meta.map.audio.%d.channels", i );
+                                               int map_channels = mlt_properties_get_int( frame_properties, key );
+                                               if ( !map_channels && i > 0 )
+                                                       map_channels = channels - j;
+                                               if ( map_channels )
+                                               {
+                                                       if ( i && ( audio_st[i] = add_audio_stream( this, oc, audio_codec_id, map_channels ) )
+                                                                  && ( ! open_audio( oc, audio_st[i], audio_outbuf_size ) ) )
+                                                               audio_st[i] = NULL;
+                                               }
+                                               j += i ? map_channels : audio_st[0]->codec->channels;
+                                       }
+                                       mlt_properties_pass( frame_meta_properties, frame_properties, "meta.map.audio." );
+
+                                       // Write the stream header now that we have all streams
+                                       av_write_header( oc );
+                                       is_audio_initialized = 1;
+                               }
+
                                // Create the fifo if we don't have one
                                if ( fifo == NULL )
                                {
@@ -1077,41 +1117,101 @@ static void *consumer_thread( void *arg )
                while ( 1 )
                {
                        // Write interleaved audio and video frames
-                       if ( !video_st || ( video_st && audio_st && audio_pts < video_pts ) )
+                       if ( !video_st || ( video_st && audio_st[0] && audio_pts < video_pts ) )
                        {
                                if ( ( video_st && terminated ) || ( channels * audio_input_frame_size ) < sample_fifo_used( fifo ) )
                                {
-                                       AVCodecContext *c = audio_st->codec;
-                                       AVPacket pkt;
                                        int n = FFMIN( FFMIN( channels * audio_input_frame_size, sample_fifo_used( fifo ) ), AUDIO_ENCODE_BUFFER_SIZE );
-
                                        if ( n > 0 )
-                                               sample_fifo_fetch( fifo, buffer, n );
+                                               sample_fifo_fetch( fifo, audio_buf_1, n );
                                        else
-                                               memset( buffer, 0, AUDIO_ENCODE_BUFFER_SIZE );
-                                       
-                                       av_init_packet( &pkt );
-                                       pkt.size = avcodec_encode_audio( c, audio_outbuf, audio_outbuf_size, buffer );
-                                       
-                                       // Write the compressed frame in the media file
-                                       if ( c->coded_frame && c->coded_frame->pts != AV_NOPTS_VALUE )
+                                               memset( audio_buf_1, 0, AUDIO_ENCODE_BUFFER_SIZE );
+                                       samples = n / channels;
+
+                                       // Extract the audio channels according to channel mapping
+                                       int j = 0; // channel offset into interleaved source buffer
+                                       int map_start = 0, map_channels = 0;
+                                       for ( i = 0; i < MAX_AUDIO_STREAMS && audio_st[i] && j < channels; i++ )
                                        {
-                                               pkt.pts = av_rescale_q( c->coded_frame->pts, c->time_base, audio_st->time_base );
-                                               mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "audio pkt pts %lld frame pts %lld", pkt.pts, c->coded_frame->pts );
-                                       }
-                                       pkt.flags |= PKT_FLAG_KEY;
-                                       pkt.stream_index= audio_st->index;
-                                       pkt.data= audio_outbuf;
+                                               AVStream *stream = audio_st[i];
+                                               AVCodecContext *codec = stream->codec;
+                                               AVPacket pkt;
 
-                                       if ( pkt.size > 0 )
-                                               if ( av_interleaved_write_frame( oc, &pkt ) != 0 )
-                                                       mlt_log_error( MLT_CONSUMER_SERVICE( this ), "error writing audio frame\n" );
+                                               av_init_packet( &pkt );
 
-                                       mlt_log_debug( MLT_CONSUMER_SERVICE( this ), " frame_size %d\n", c->frame_size );
-                                       if ( audio_codec_id == CODEC_ID_VORBIS )
-                                               audio_pts = (double)c->coded_frame->pts * av_q2d( audio_st->time_base );
-                                       else
-                                               audio_pts = (double)audio_st->pts.val * av_q2d( audio_st->time_base );
+                                               // Get the audio channel mapping
+                                               sprintf( key, "%d.channels", i );
+                                               map_channels = mlt_properties_get_int( frame_meta_properties, key );
+                                               sprintf( key, "%d.start", i );
+                                               if ( mlt_properties_get( frame_meta_properties, key ) )
+                                                       map_start = mlt_properties_get_int( frame_meta_properties, key );
+                                               else
+                                                       map_start = -1;
+
+                                               // Optimized for no channel remapping.
+                                               if ( !map_channels && map_start == -1 )
+                                               {
+                                                       // Encode the audio.
+                                                       pkt.size = avcodec_encode_audio( codec, audio_outbuf, audio_outbuf_size, audio_buf_1 );
+                                               }
+                                               else
+                                               {
+                                                       int ch; // channel offset into interleaved dest buffer
+
+                                                       // If last map..channels not specific, use the remaining channels.
+                                                       if ( !map_channels )
+                                                               map_channels = channels - j;
+
+                                                       // Clear or allocate the audio buffer.
+                                                       if ( !audio_buf_2 )
+                                                               audio_buf_2 = av_mallocz( AUDIO_ENCODE_BUFFER_SIZE );
+                                                       else
+                                                               memset( audio_buf_2, 0, AUDIO_ENCODE_BUFFER_SIZE );
+
+                                                       // Interleave the audio buffer with the #channels for this stream.
+                                                       for ( ch = 0; ch < map_channels && j < channels; ch++, j++ )
+                                                       {
+                                                               int16_t *src = audio_buf_1 + ( map_start > -1 ? ( map_start + ch ) : j );
+                                                               int16_t *dest = audio_buf_2 + ch;
+                                                               int s = samples + 1;
+
+                                                               while ( --s ) {
+                                                                       *dest = *src;
+                                                                       dest += map_channels;
+                                                                       src += channels;
+                                                               }
+                                                       }
+
+                                                       // Encode the audio.
+                                                       pkt.size = avcodec_encode_audio( codec, audio_outbuf, audio_outbuf_size, audio_buf_2 );
+                                               }
+
+                                               // Write the compressed frame in the media file
+                                               if ( codec->coded_frame && codec->coded_frame->pts != AV_NOPTS_VALUE )
+                                               {
+                                                       pkt.pts = av_rescale_q( codec->coded_frame->pts, codec->time_base, stream->time_base );
+                                                       mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "audio stream %d pkt pts %lld frame pts %lld",
+                                                               stream->index, pkt.pts, codec->coded_frame->pts );
+                                               }
+                                               pkt.flags |= PKT_FLAG_KEY;
+                                               pkt.stream_index = stream->index;
+                                               pkt.data = audio_outbuf;
+
+                                               if ( pkt.size > 0 && stream->pts.den )
+                                               {
+                                                       if ( av_interleaved_write_frame( oc, &pkt ) )
+                                                               mlt_log_error( MLT_CONSUMER_SERVICE( this ), "error writing audio frame %d\n", frames - 1 );
+                                               }
+
+                                               mlt_log_debug( MLT_CONSUMER_SERVICE( this ), " frame_size %d\n", codec->frame_size );
+                                               if ( i == 0 )
+                                               {
+                                                       if ( audio_codec_id == CODEC_ID_VORBIS )
+                                                               audio_pts = (double)codec->coded_frame->pts * av_q2d( stream->time_base );
+                                                       else
+                                                               audio_pts = (double)stream->pts.val * av_q2d( stream->time_base );
+                                               }
+                                       }
                                }
                                else
                                {
@@ -1258,8 +1358,8 @@ static void *consumer_thread( void *arg )
                                        break;
                                }
                        }
-                       if ( audio_st )
-                               mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "audio pts %lld (%f) ", audio_st->pts.val, audio_pts );
+                       if ( audio_st[0] )
+                               mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "audio pts %lld (%f) ", audio_st[0]->pts.val, audio_pts );
                        if ( video_st )
                                mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "video pts %lld (%f) ", video_st->pts.val, video_pts );
                        mlt_log_debug( MLT_CONSUMER_SERVICE( this ), "\n" );
@@ -1280,6 +1380,9 @@ static void *consumer_thread( void *arg )
                                nanosleep( &t, NULL );
                        }
                }
+
+               // Get the next frame
+               frame = mlt_consumer_rt_frame( this );
        }
 
 #ifdef FLUSH
@@ -1346,6 +1449,9 @@ static void *consumer_thread( void *arg )
        }
 #endif
 
+       // Write the trailer, if any
+       av_write_trailer( oc );
+
        // XXX ugly hack to prevent x264 from crashing on multi-threaded encoding
        int pass = mlt_properties_get_int( properties, "pass" );
        int thread_count = mlt_properties_get_int( properties, "threads" );
@@ -1356,22 +1462,19 @@ static void *consumer_thread( void *arg )
        // close each codec
        if ( video_st && !multithreaded_x264 )
                close_video(oc, video_st);
-       if (audio_st)
-               close_audio(oc, audio_st);
-
-       // Write the trailer, if any
-       av_write_trailer(oc);
+       for ( i = 0; i < MAX_AUDIO_STREAMS && audio_st[i]; i++ )
+               close_audio( oc, audio_st[i] );
 
        // Free the streams
-       for(i = 0; i < oc->nb_streams; i++)
-               av_freep(&oc->streams[i]);
+       for ( i = 0; i < oc->nb_streams; i++ )
+               av_freep( &oc->streams[i] );
 
        // Close the output file
-       if (!(fmt->flags & AVFMT_NOFILE))
+       if ( !( fmt->flags & AVFMT_NOFILE ) )
 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(0<<8)+0)
-               url_fclose(oc->pb);
+               url_fclose( oc->pb );
 #else
-               url_fclose(&oc->pb);
+               url_fclose( &oc->pb );
 #endif
 
        // Clean up input and output frames
@@ -1381,15 +1484,17 @@ static void *consumer_thread( void *arg )
        av_free( input->data[0] );
        av_free( input );
        av_free( video_outbuf );
-       av_free( buffer );
+       av_free( audio_buf_1 );
+       av_free( audio_buf_2 );
 
        // Free the stream
-       av_free(oc);
+       av_free( oc );
 
        // Just in case we terminated on pause
        mlt_properties_set_int( properties, "running", 0 );
 
        mlt_consumer_stopped( this );
+       mlt_properties_close( frame_meta_properties );
 
        if ( pass == 2 )
        {