X-Git-Url: http://research.m1stereo.tv/gitweb?a=blobdiff_plain;f=src%2Fmodules%2Favformat%2Fproducer_avformat.c;h=3763bdb0ff7c08aa6f568fd0ed42bfbb7953a21d;hb=bf13b4d4753f0a2c8d73664e970e0d7b1b4a5a31;hp=7f294bbb9d065c5c789e856223bd1d36021426ba;hpb=027dc39ab6e64c5b50901996c4977dc45cadb1c3;p=melted diff --git a/src/modules/avformat/producer_avformat.c b/src/modules/avformat/producer_avformat.c index 7f294bb..3763bdb 100644 --- a/src/modules/avformat/producer_avformat.c +++ b/src/modules/avformat/producer_avformat.c @@ -25,8 +25,12 @@ // ffmpeg Header files #include +#include #ifdef SWSCALE -#include +# include +#endif +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) +# include "audioconvert.h" #endif // System header files @@ -160,32 +164,70 @@ static void producer_codec_close( void *codec ) } } -static double get_aspect_ratio( AVStream *stream, AVCodecContext *codec_context ) +static inline int dv_is_pal( AVPacket *pkt ) +{ + return pkt->data[3] & 0x80; +} + +static int dv_is_wide( AVPacket *pkt ) +{ + int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */; + + for ( ; i < pkt->size; i += 5 /* packet size */ ) + { + if ( pkt->data[ i ] == 0x61 ) + { + uint8_t x = pkt->data[ i + 2 ] & 0x7; + return ( x == 2 ) || ( x == 7 ); + } + } + return 0; +} + +static double get_aspect_ratio( AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt ) { double aspect_ratio = 1.0; if ( codec_context->codec_id == CODEC_ID_DVVIDEO ) { - AVRational ar = -// #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0) -// stream->sample_aspect_ratio; -// #else - codec_context->sample_aspect_ratio; -// #endif - // Override FFmpeg's notion of DV aspect ratios, which are - // based upon a width of 704. Since we do not have a normaliser - // that crops (nor is cropping 720 wide ITU-R 601 video always desirable) - // we just coerce the values to facilitate a passive behaviour through - // the rescale normaliser when using equivalent producers and consumers. - // = display_aspect / (width * height) - if ( ar.num == 10 && ar.den == 11 ) - aspect_ratio = 8.0/9.0; // 4:3 NTSC - else if ( ar.num == 59 && ar.den == 54 ) - aspect_ratio = 16.0/15.0; // 4:3 PAL - else if ( ar.num == 40 && ar.den == 33 ) - aspect_ratio = 32.0/27.0; // 16:9 NTSC - else if ( ar.num == 118 && ar.den == 81 ) - aspect_ratio = 64.0/45.0; // 16:9 PAL + if ( pkt ) + { + if ( dv_is_pal( pkt ) ) + { + aspect_ratio = dv_is_wide( pkt ) + ? 64.0/45.0 // 16:9 PAL + : 16.0/15.0; // 4:3 PAL + } + else + { + aspect_ratio = dv_is_wide( pkt ) + ? 32.0/27.0 // 16:9 NTSC + : 8.0/9.0; // 4:3 NTSC + } + } + else + { + AVRational ar = +#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0) + stream->sample_aspect_ratio; +#else + codec_context->sample_aspect_ratio; +#endif + // Override FFmpeg's notion of DV aspect ratios, which are + // based upon a width of 704. Since we do not have a normaliser + // that crops (nor is cropping 720 wide ITU-R 601 video always desirable) + // we just coerce the values to facilitate a passive behaviour through + // the rescale normaliser when using equivalent producers and consumers. + // = display_aspect / (width * height) + if ( ar.num == 10 && ar.den == 11 ) + aspect_ratio = 8.0/9.0; // 4:3 NTSC + else if ( ar.num == 59 && ar.den == 54 ) + aspect_ratio = 16.0/15.0; // 4:3 PAL + else if ( ar.num == 40 && ar.den == 33 ) + aspect_ratio = 32.0/27.0; // 16:9 NTSC + else if ( ar.num == 118 && ar.den == 81 ) + aspect_ratio = 64.0/45.0; // 16:9 PAL + } } else { @@ -361,8 +403,28 @@ static int producer_open( mlt_producer this, mlt_profile profile, char *file ) 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 ); - mlt_properties_set_double( properties, "aspect_ratio", - get_aspect_ratio( context->streams[ video_index ], codec_context ) ); + + 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 ) + { + ret = av_read_frame( context, &pkt ); + if ( ret >= 0 && pkt.stream_index == video_index && pkt.size > 0 ) + { + mlt_properties_set_double( properties, "aspect_ratio", + get_aspect_ratio( context->streams[ video_index ], codec_context, &pkt ) ); + break; + } + } + } + else + { + mlt_properties_set_double( properties, "aspect_ratio", + get_aspect_ratio( context->streams[ video_index ], codec_context, NULL ) ); + } } // Read Metadata @@ -493,6 +555,45 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, #endif } +/** Allocate the image buffer and set it on the frame. +*/ + +static int allocate_buffer( mlt_properties frame_properties, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height ) +{ + int size = 0; + + if ( codec_context->width == 0 || codec_context->height == 0 ) + return size; + + *width = codec_context->width; + *height = codec_context->height; + mlt_properties_set_int( frame_properties, "width", *width ); + mlt_properties_set_int( frame_properties, "height", *height ); + + switch ( *format ) + { + case mlt_image_yuv420p: + size = *width * 3 * ( *height + 1 ) / 2; + break; + case mlt_image_rgb24: + size = *width * ( *height + 1 ) * 3; + break; + default: + *format = mlt_image_yuv422; + size = *width * ( *height + 1 ) * 2; + break; + } + + // Construct the output image + *buffer = mlt_pool_alloc( size ); + if ( *buffer ) + mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL ); + else + size = 0; + + return size; +} + /** Get an image from a frame. */ @@ -547,40 +648,12 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form // Get the seekable status int seekable = mlt_properties_get_int( properties, "seekable" ); - // Generate the size in bytes - int size = 0; - // Hopefully provide better support for streams... int av_bypass = mlt_properties_get_int( properties, "av_bypass" ); // Determines if we have to decode all frames in a sequence int must_decode = 1; - // Set the result arguments that we know here (only *buffer is now required) - *width = codec_context->width; - *height = codec_context->height; - - switch ( *format ) - { - case mlt_image_yuv420p: - size = *width * 3 * ( *height + 1 ) / 2; - break; - case mlt_image_rgb24: - size = *width * ( *height + 1 ) * 3; - break; - default: - *format = mlt_image_yuv422; - size = *width * ( *height + 1 ) * 2; - break; - } - - // Set this on the frame properties - mlt_properties_set_int( frame_properties, "width", *width ); - mlt_properties_set_int( frame_properties, "height", *height ); - - // Construct the output image - *buffer = mlt_pool_alloc( size ); - // Temporary hack to improve intra frame only must_decode = strcmp( codec_context->codec->name, "mjpeg" ) && strcmp( codec_context->codec->name, "rawvideo" ) && @@ -627,10 +700,10 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form if ( av_frame != NULL && got_picture && ( paused || current_position >= req_position ) && av_bypass == 0 ) { // Duplicate it - convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height ); - - // Set this on the frame properties - mlt_properties_set_data( frame_properties, "image", *buffer, size, ( mlt_destructor )mlt_pool_release, NULL ); + if ( allocate_buffer( frame_properties, codec_context, buffer, format, width, height ) ) + convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height ); + else + mlt_frame_get_image( frame, buffer, format, width, height, writable ); } else { @@ -642,10 +715,7 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form // Construct an AVFrame for YUV422 conversion if ( av_frame == NULL ) - { av_frame = avcodec_alloc_frame( ); - mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL ); - } while( ret >= 0 && !got_picture ) { @@ -695,12 +765,19 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form // Now handle the picture if we have one if ( got_picture ) { - mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame ); - mlt_properties_set_int( properties, "top_field_first", av_frame->top_field_first ); - convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height ); - mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL ); - mlt_properties_set_int( properties, "_current_position", int_position ); - mlt_properties_set_int( properties, "_got_picture", 1 ); + if ( allocate_buffer( frame_properties, codec_context, buffer, format, width, height ) ) + { + convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height ); + mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame ); + mlt_properties_set_int( properties, "top_field_first", av_frame->top_field_first ); + mlt_properties_set_int( properties, "_current_position", int_position ); + mlt_properties_set_int( properties, "_got_picture", 1 ); + mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL ); + } + else + { + got_picture = 0; + } } } if ( !got_picture ) @@ -721,6 +798,26 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form return 0; } +/** Process properties as AVOptions and apply to AV context obj +*/ + +static void apply_properties( void *obj, mlt_properties properties, int flags ) +{ + int i; + int count = mlt_properties_count( properties ); + for ( i = 0; i < count; i++ ) + { + const char *opt_name = mlt_properties_get_name( properties, i ); + const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags ); + if ( opt != NULL ) +#if LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0) + av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 ); +#else + av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) ); +#endif + } +} + /** Set up video handling. */ @@ -744,6 +841,9 @@ static void producer_set_up_video( mlt_producer this, mlt_frame frame ) context = mlt_properties_get_data( properties, "video_context", NULL ); mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL ); mlt_events_unblock( properties, this ); + + // Process properties as AVOptions + apply_properties( context, properties, AV_OPT_FLAG_DECODING_PARAM ); } // Exception handling for video_index @@ -782,8 +882,9 @@ static void producer_set_up_video( mlt_producer this, mlt_frame frame ) mlt_properties_set_data( properties, "video_codec", NULL, 0, NULL, NULL ); mlt_properties_set_int( properties, "width", codec_context->width ); mlt_properties_set_int( properties, "height", codec_context->height ); + // TODO: get the first usable AVPacket and reset the stream position mlt_properties_set_double( properties, "aspect_ratio", - get_aspect_ratio( context->streams[ index ], codec_context ) ); + get_aspect_ratio( context->streams[ index ], codec_context, NULL ) ); codec = NULL; } @@ -814,17 +915,21 @@ static void producer_set_up_video( mlt_producer this, mlt_frame frame ) { // Remember that we can't use this later mlt_properties_set_int( properties, "video_index", -1 ); + index = -1; } avformat_unlock( ); + + // Process properties as AVOptions + apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM ); } // No codec, no show... - if ( codec != NULL ) + if ( codec && index > -1 ) { double source_fps = 0; double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" ); double aspect_ratio = ( force_aspect_ratio > 0.0 ) ? - force_aspect_ratio : get_aspect_ratio( stream, codec_context ); + force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" ); // Determine the fps source_fps = ( double )codec_context->time_base.den / ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ); @@ -887,8 +992,15 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form // Obtain the resample context if it exists (not always needed) ReSampleContext *resample = mlt_properties_get_data( properties, "audio_resample", NULL ); - // Obtain the audio buffer +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) + // Get the format converter context if it exists + AVAudioConvert *convert = mlt_properties_get_data( properties, "audio_convert", NULL ); +#endif + + // Obtain the audio buffers int16_t *audio_buffer = mlt_properties_get_data( properties, "audio_buffer", NULL ); + int16_t *decode_buffer = mlt_properties_get_data( properties, "decode_buffer", NULL ); + int16_t *convert_buffer = mlt_properties_get_data( properties, "convert_buffer", NULL ); // Get amount of audio used int audio_used = mlt_properties_get_int( properties, "_audio_used" ); @@ -926,6 +1038,17 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form *frequency = codec_context->sample_rate; } +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) + // Check for audio format converter and create if necessary + // TODO: support higher resolutions than 16-bit. + if ( convert == NULL && codec_context->sample_fmt != SAMPLE_FMT_S16 ) + { + // Create single channel converter for interleaved with no mixing matrix + convert = av_audio_convert_alloc( SAMPLE_FMT_S16, 1, codec_context->sample_fmt, 1, NULL, 0 ); + mlt_properties_set_data( properties, "audio_convert", convert, 0, ( mlt_destructor )av_audio_convert_free, NULL ); + } +#endif + // Check for audio buffer and create if necessary if ( audio_buffer == NULL ) { @@ -936,6 +1059,28 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form mlt_properties_set_data( properties, "audio_buffer", audio_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL ); } + // Check for decoder buffer and create if necessary + if ( decode_buffer == NULL ) + { + // Allocate the audio buffer + decode_buffer = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) ); + + // And store it on properties for reuse + mlt_properties_set_data( properties, "decode_buffer", decode_buffer, 0, ( mlt_destructor )av_free, NULL ); + } + +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) + // Check for format converter buffer and create if necessary + if ( resample && convert && convert_buffer == NULL ) + { + // Allocate the audio buffer + convert_buffer = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) ); + + // And store it on properties for reuse + mlt_properties_set_data( properties, "convert_buffer", convert_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL ); + } +#endif + // Seek if necessary if ( position != expected ) { @@ -965,7 +1110,6 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form { int ret = 0; int got_audio = 0; - int16_t *temp = av_malloc( sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE ); av_init_packet( &pkt ); @@ -991,9 +1135,9 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form // Decode the audio #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0)) - ret = avcodec_decode_audio2( codec_context, temp, &data_size, ptr, len ); + ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len ); #else - ret = avcodec_decode_audio( codec_context, temp, &data_size, ptr, len ); + ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len ); #endif if ( ret < 0 ) { @@ -1006,14 +1150,41 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form if ( data_size > 0 ) { - if ( resample != NULL ) + int src_stride[6]= { av_get_bits_per_sample_format( codec_context->sample_fmt ) / 8 }; + int dst_stride[6]= { av_get_bits_per_sample_format( SAMPLE_FMT_S16 ) / 8 }; + + if ( resample ) { - audio_used += audio_resample( resample, &audio_buffer[ audio_used * *channels ], temp, data_size / ( codec_context->channels * sizeof( int16_t ) ) ); + int16_t *source = decode_buffer; + int16_t *dest = &audio_buffer[ audio_used * *channels ]; + int convert_samples = data_size / src_stride[0]; + +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) + if ( convert ) + { + const void *src_buf[6] = { decode_buffer }; + void *dst_buf[6] = { convert_buffer }; + av_audio_convert( convert, dst_buf, dst_stride, src_buf, src_stride, convert_samples ); + source = convert_buffer; + } +#endif + audio_used += audio_resample( resample, dest, source, convert_samples / codec_context->channels ); } else { - memcpy( &audio_buffer[ audio_used * *channels ], temp, data_size ); - audio_used += data_size / ( codec_context->channels * sizeof( int16_t ) ); +#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) + if ( convert ) + { + const void *src_buf[6] = { decode_buffer }; + void *dst_buf[6] = { &audio_buffer[ audio_used * *channels ] }; + av_audio_convert( convert, dst_buf, dst_stride, src_buf, src_stride, data_size / src_stride[0] ); + } + else +#endif + { + memcpy( &audio_buffer[ audio_used * *channels ], decode_buffer, data_size ); + } + audio_used += data_size / *channels / src_stride[0]; } // Handle ignore @@ -1055,9 +1226,6 @@ static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_form // Store the number of audio samples still available mlt_properties_set_int( properties, "_audio_used", audio_used ); - - // Release the temporary audio - av_free( temp ); } else { @@ -1110,7 +1278,8 @@ static void producer_set_up_audio( mlt_producer this, mlt_frame frame ) } // Update the audio properties if the index changed - if ( index > -1 && index != mlt_properties_get_int( properties, "_audio_index" ) ) { + if ( index > -1 && index != mlt_properties_get_int( properties, "_audio_index" ) ) + { mlt_properties_set_int( properties, "_audio_index", index ); mlt_properties_set_data( properties, "audio_codec", NULL, 0, NULL, NULL ); } @@ -1148,12 +1317,16 @@ static void producer_set_up_audio( mlt_producer this, mlt_frame frame ) { // Remember that we can't use this later mlt_properties_set_int( properties, "audio_index", -1 ); + index = -1; } avformat_unlock( ); + + // Process properties as AVOptions + apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM ); } // No codec, no show... - if ( codec != NULL ) + if ( codec && index > -1 ) { mlt_frame_push_audio( frame, producer_get_audio ); mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );