X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=src%2Fmodules%2Favformat%2Fproducer_avformat.c;h=a228ec733b56952ba2fd11262700c8c4d73224ec;hb=424496d4cbce82ca24cc85536bd9d3fdd6593067;hp=9387a6632616c0f13e40c348d0d8ea71801b9aea;hpb=f9023d71359af736f2f899c6f118721d0ce301d3;p=mlt diff --git a/src/modules/avformat/producer_avformat.c b/src/modules/avformat/producer_avformat.c index 9387a663..f2530308 100644 --- a/src/modules/avformat/producer_avformat.c +++ b/src/modules/avformat/producer_avformat.c @@ -1,6 +1,6 @@ /* * producer_avformat.c -- avformat producer - * Copyright (C) 2003-2009 Ushodaya Enterprises Limited + * Copyright (C) 2003-2012 Ushodaya Enterprises Limited * Author: Charles Yates * Author: Dan Dennedy * Much code borrowed from ffmpeg.c: Copyright (c) 2000-2003 Fabrice Bellard @@ -31,19 +31,15 @@ // ffmpeg Header files #include -#ifdef SWSCALE -# include -#endif -#if LIBAVCODEC_VERSION_MAJOR > 52 +#include #include -#elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) -const char *avcodec_get_sample_fmt_name(int sample_fmt); -#endif +#include + #ifdef VDPAU # include #endif -#if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0)) -# include +#if (LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)) +# include #endif // System header files @@ -52,12 +48,7 @@ const char *avcodec_get_sample_fmt_name(int sample_fmt); #include #include -#if LIBAVUTIL_VERSION_INT < (50<<16) -#define PIX_FMT_RGB32 PIX_FMT_RGBA32 -#define PIX_FMT_YUYV422 PIX_FMT_YUV422 -#endif - -#if LIBAVCODEC_VERSION_MAJOR > 52 +#if LIBAVCODEC_VERSION_MAJOR >= 53 #include #define CODEC_TYPE_VIDEO AVMEDIA_TYPE_VIDEO #define CODEC_TYPE_AUDIO AVMEDIA_TYPE_AUDIO @@ -66,14 +57,17 @@ const char *avcodec_get_sample_fmt_name(int sample_fmt); #include #endif +#if LIBAVCODEC_VERSION_MAJOR < 55 +#define AV_CODEC_ID_DVVIDEO CODEC_ID_DVVIDEO +#define AV_CODEC_ID_H264 CODEC_ID_H264 +#endif + #define POSITION_INITIAL (-2) #define POSITION_INVALID (-1) -#define MAX_AUDIO_STREAMS (10) +#define MAX_AUDIO_STREAMS (32) #define MAX_VDPAU_SURFACES (10) - -void avformat_lock( ); -void avformat_unlock( ); +#define MAX_AUDIO_FRAME_SIZE (192000) // 1 second of 48khz 32bit audio struct producer_avformat_s { @@ -83,19 +77,18 @@ struct producer_avformat_s AVFormatContext *video_format; AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ]; AVCodecContext *video_codec; - AVFrame *av_frame; - ReSampleContext *audio_resample[ MAX_AUDIO_STREAMS ]; + AVFrame *video_frame; + AVFrame *audio_frame; + AVPacket pkt; mlt_position audio_expected; mlt_position video_expected; int audio_index; int video_index; - double start_time; - int first_pts; + int64_t first_pts; int64_t last_position; int seekable; int64_t current_position; mlt_position nonseek_position; - int got_picture; int top_field_first; uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ]; size_t audio_buffer_size[ MAX_AUDIO_STREAMS ]; @@ -107,14 +100,18 @@ struct producer_avformat_s int max_channel; int max_frequency; unsigned int invalid_pts_counter; - double resample_factor; + unsigned int invalid_dts_counter; mlt_cache image_cache; - int colorspace; + int yuv_colorspace, color_primaries; + int full_luma; pthread_mutex_t video_mutex; pthread_mutex_t audio_mutex; mlt_deque apackets; mlt_deque vpackets; pthread_mutex_t packets_mutex; + pthread_mutex_t open_mutex; + int is_mutex_init; + AVRational video_time_base; #ifdef VDPAU struct { @@ -127,6 +124,9 @@ struct producer_avformat_s int ip_age[2]; int is_decoded; uint8_t *buffer; + + VdpDevice device; + VdpDecoder decoder; } *vdpau; #endif }; @@ -143,6 +143,7 @@ static void producer_set_up_audio( producer_avformat self, mlt_frame frame ); static void apply_properties( void *obj, mlt_properties properties, int flags ); static int video_codec_init( producer_avformat self, int index, mlt_properties properties ); static void get_audio_streams_info( producer_avformat self ); +static mlt_audio_format pick_audio_format( int sample_fmt ); #ifdef VDPAU #include "vdpau.c" @@ -190,18 +191,24 @@ mlt_producer producer_avformat_init( mlt_profile profile, const char *service, c // Clean up mlt_producer_close( producer ); producer = NULL; + producer_avformat_close( self ); } else if ( self->seekable ) { // Close the file to release resources for large playlists - reopen later as needed - avformat_lock(); +#if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0) + if ( self->audio_format ) + avformat_close_input( &self->audio_format ); + if ( self->video_format ) + avformat_close_input( &self->video_format ); +#else if ( self->audio_format ) av_close_input_file( self->audio_format ); - self->audio_format = NULL; if ( self->video_format ) av_close_input_file( self->video_format ); +#endif + self->audio_format = NULL; self->video_format = NULL; - avformat_unlock(); } } if ( producer ) @@ -257,6 +264,23 @@ int list_components( char* file ) return skip; } +static int first_video_index( producer_avformat self ) +{ + AVFormatContext *context = self->video_format? self->video_format : self->audio_format; + int i = -1; // not found + + if ( context ) { + for ( i = 0; i < context->nb_streams; i++ ) { + if ( context->streams[i]->codec && + context->streams[i]->codec->codec_type == CODEC_TYPE_VIDEO ) + break; + } + if ( i == context->nb_streams ) + i = -1; + } + return i; +} + /** Find the default streams. */ @@ -264,7 +288,11 @@ static mlt_properties find_default_streams( producer_avformat self ) { int i; char key[200]; +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0) + AVDictionaryEntry *tag = NULL; +#else AVMetadataTag *tag = NULL; +#endif AVFormatContext *context = self->video_format; mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent ); @@ -296,27 +324,30 @@ static mlt_properties find_default_streams( producer_avformat self ) 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) double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate ); +#if LIBAVFORMAT_VERSION_MAJOR < 55 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 ) ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate ); - mlt_properties_set_double( meta_media, key, ffmpeg_fps ); -#else - mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) ); #endif + mlt_properties_set_double( meta_media, key, ffmpeg_fps ); -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0) snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i ); mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) ); -#endif + snprintf( key, sizeof(key), "meta.media.%d.codec.width", i ); + mlt_properties_set_int( meta_media, key, codec_context->width ); + snprintf( key, sizeof(key), "meta.media.%d.codec.height", i ); + mlt_properties_set_int( meta_media, key, codec_context->height ); snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i ); - mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den / - ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) ); + AVRational frame_rate = { codec_context->time_base.den, codec_context->time_base.num * codec_context->ticks_per_frame }; + mlt_properties_set_double( meta_media, key, av_q2d( frame_rate ) ); snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i ); +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(3<<8)+0) + mlt_properties_set( meta_media, key, av_get_pix_fmt_name( codec_context->pix_fmt ) ); +#else mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) ); +#endif snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i ); mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) ); -#if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0) snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i ); switch ( codec_context->colorspace ) { @@ -335,20 +366,17 @@ static mlt_properties find_default_streams( producer_avformat self ) mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 ); break; } -#endif break; case CODEC_TYPE_AUDIO: + if ( !codec_context->channels ) + break; // Use first audio stream - if ( self->audio_index < 0 ) + if ( self->audio_index < 0 && pick_audio_format( codec_context->sample_fmt ) != mlt_audio_none ) 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 ); mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) ); -#elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0)) - snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i ); - mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) ); -#endif snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i ); mlt_properties_set_int( meta_media, key, codec_context->sample_rate ); snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i ); @@ -361,10 +389,8 @@ static mlt_properties find_default_streams( producer_avformat self ) // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) ); snprintf( key, sizeof(key), "meta.media.%d.codec.name", i ); mlt_properties_set( meta_media, key, codec->name ); -#if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0)) snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i ); mlt_properties_set( meta_media, key, codec->long_name ); -#endif snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i ); mlt_properties_set_int( meta_media, key, codec_context->bit_rate ); // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i ); @@ -375,8 +401,11 @@ static mlt_properties find_default_streams( producer_avformat self ) // mlt_properties_set_int( meta_media, key, codec_context->level ); // Read Metadata -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0) +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0) + while ( ( tag = av_dict_get( stream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) ) +#else while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) ) +#endif { if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) ) { @@ -384,10 +413,12 @@ static mlt_properties find_default_streams( producer_avformat self ) mlt_properties_set( meta_media, key, tag->value ); } } -#endif } -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0) +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0) + while ( ( tag = av_dict_get( context->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) ) +#else while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) ) +#endif { if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) ) { @@ -395,163 +426,36 @@ static mlt_properties find_default_streams( producer_avformat self ) mlt_properties_set( meta_media, key, tag->value ); } } -#else - if ( context->title && strcmp( context->title, "" ) ) - mlt_properties_set(properties, "meta.attr.title.markup", context->title ); - if ( context->author && strcmp( context->author, "" ) ) - mlt_properties_set(properties, "meta.attr.author.markup", context->author ); - if ( context->copyright && strcmp( context->copyright, "" ) ) - mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright ); - if ( context->comment ) - mlt_properties_set(properties, "meta.attr.comment.markup", context->comment ); - if ( context->album ) - mlt_properties_set(properties, "meta.attr.album.markup", context->album ); - if ( context->year ) - mlt_properties_set_int(properties, "meta.attr.year.markup", context->year ); - if ( context->track ) - mlt_properties_set_int(properties, "meta.attr.track.markup", context->track ); -#endif return meta_media; } -static inline int dv_is_pal( AVPacket *pkt ) +static void get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context ) { - return pkt->data[3] & 0x80; + AVRational sar = stream->sample_aspect_ratio; + if ( sar.num <= 0 || sar.den <= 0 ) + sar = codec_context->sample_aspect_ratio; + if ( sar.num <= 0 || sar.den <= 0 ) + sar.num = sar.den = 1; + mlt_properties_set_int( properties, "meta.media.sample_aspect_num", sar.num ); + mlt_properties_set_int( properties, "meta.media.sample_aspect_den", sar.den ); + mlt_properties_set_double( properties, "aspect_ratio", av_q2d( sar ) ); } -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( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt ) -{ - double aspect_ratio = 1.0; - - if ( codec_context->codec_id == CODEC_ID_DVVIDEO ) - { - if ( pkt ) - { - if ( dv_is_pal( pkt ) ) - { - if ( dv_is_wide( pkt ) ) - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 ); - } - else - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 ); - } - } - else - { - if ( dv_is_wide( pkt ) ) - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 ); - } - else - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 ); - } - } - } - else - { - AVRational ar = -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0) - stream->sample_aspect_ratio; +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) +static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVDictionary **params ) #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 ) - { - // 4:3 NTSC - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 ); - } - else if ( ar.num == 59 && ar.den == 54 ) - { - // 4:3 PAL - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 ); - } - else if ( ar.num == 40 && ar.den == 33 ) - { - // 16:9 NTSC - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 ); - } - else if ( ar.num == 118 && ar.den == 81 ) - { - // 16:9 PAL - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 ); - } - } - } - else - { - AVRational codec_sar = codec_context->sample_aspect_ratio; - AVRational stream_sar = -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0) - stream->sample_aspect_ratio; -#else - { 0, 1 }; -#endif - if ( codec_sar.num > 0 ) - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den ); - } - else if ( stream_sar.num > 0 ) - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den ); - } - else - { - mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 ); - mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 ); - } - } - AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) }; - aspect_ratio = av_q2d( ar ); - mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio ); - - return aspect_ratio; -} - static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params ) +#endif { if ( !URL ) return NULL; - const char *result = URL; + 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 LIBAVFORMAT_VERSION_MAJOR >= 53 if ( url && avio_check( URL, 0 ) < 0 ) #else if ( url && !url_exist( URL ) ) @@ -569,6 +473,11 @@ static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **fo if ( *format ) { +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + // support for legacy width and height parameters + char *width = NULL; + char *height = NULL; +#else // These are required by video4linux2 (defaults) params->width = profile->width; params->height = profile->height; @@ -576,6 +485,7 @@ static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **fo params->time_base = (AVRational){ profile->frame_rate_den, profile->frame_rate_num }; params->channels = 2; params->sample_rate = 48000; +#endif // Parse out params url = strchr( url, '?' ); @@ -583,7 +493,10 @@ static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **fo { url[0] = 0; char *name = strdup( ++url ); - char *value = strchr( name, ':' ); + char *value = strchr( name, '=' ); + if ( !value ) + // Also accept : as delimiter for backwards compatibility. + value = strchr( name, ':' ); if ( value ) { value[0] = 0; @@ -591,6 +504,20 @@ static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **fo char *t = strchr( value, '&' ); if ( t ) t[0] = 0; +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + // translate old parameters to new av_dict names + if ( !strcmp( name, "frame_rate" ) ) + av_dict_set( params, "framerate", value, 0 ); + else if ( !strcmp( name, "pix_fmt" ) ) + av_dict_set( params, "pixel_format", value, 0 ); + else if ( !strcmp( name, "width" ) ) + width = strdup( value ); + else if ( !strcmp( name, "height" ) ) + height = strdup( value ); + else + // generic demux/device option support + av_dict_set( params, name, value, 0 ); +#else if ( !strcmp( name, "frame_rate" ) ) params->time_base.den = atoi( value ); else if ( !strcmp( name, "frame_rate_base" ) ) @@ -601,23 +528,40 @@ static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **fo 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 ); -#endif 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 ); +#endif } free( name ); url = strchr( url, '&' ); } +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + // continued support for legacy width and height parameters + if ( width && height ) + { + char *s = malloc( strlen( width ) + strlen( height ) + 2 ); + strcpy( s, width ); + strcat( s, "x"); + strcat( s, height ); + av_dict_set( params, "video_size", s, 0 ); + free( s ); + } + if ( width ) free( width ); + if ( height ) free ( height ); +#endif } + result = strdup( result ); + } + else + { + result = strdup( URL ); } - result = strdup( result ); free( protocol ); return result; } @@ -631,10 +575,6 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch AVFormatContext *format = self->video_format; - // We will treat everything with the producer fps. - // TODO: make this more flexible. - double fps = mlt_profile_fps( profile ); - // Get the duration if ( !mlt_properties_get_int( properties, "_length_computed" ) ) { @@ -643,7 +583,9 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch 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 ); + // We will treat everything with the producer fps. + mlt_position frames = ( mlt_position )( int )( format->duration * + profile->frame_rate_num / profile->frame_rate_den / AV_TIME_BASE); mlt_properties_set_position( properties, "out", frames - 1 ); mlt_properties_set_position( properties, "length", frames ); mlt_properties_set_int( properties, "_length_computed", 1 ); @@ -658,16 +600,13 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch } } - 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->iformat || !( format->iformat->flags & AVFMT_NOFILE ); if ( format->pb ) { // protocols can indicate if they support seeking -#if LIBAVFORMAT_VERSION_MAJOR > 52 +#if LIBAVFORMAT_VERSION_MAJOR >= 53 self->seekable = format->pb->seekable; #else URLContext *uc = url_fileno( format->pb ); @@ -678,12 +617,18 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch 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; + self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0; mlt_properties_set_int( properties, "seekable", self->seekable ); self->dummy_context = format; +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + self->video_format = NULL; + avformat_open_input( &self->video_format, filename, NULL, NULL ); + avformat_find_stream_info( self->video_format, NULL ); +#else av_open_input_file( &self->video_format, filename, NULL, 0, NULL ); + av_find_stream_info( self->video_format ); +#endif format = self->video_format; - av_find_stream_info( format ); } // Fetch the width, height and aspect ratio @@ -692,28 +637,8 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch 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 ); + get_aspect_ratio( properties, format->streams[ self->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( format, &pkt ); - if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 ) - { - 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 // 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, @@ -722,7 +647,6 @@ static int get_basic_info( producer_avformat self, mlt_profile profile, const ch sws_freeContext( context ); else error = 1; -#endif } return error; } @@ -736,12 +660,18 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha int error = 0; mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent ); - // Lock the service - if ( take_lock ) + if ( !self->is_mutex_init ) { pthread_mutex_init( &self->audio_mutex, NULL ); pthread_mutex_init( &self->video_mutex, NULL ); pthread_mutex_init( &self->packets_mutex, NULL ); + pthread_mutex_init( &self->open_mutex, NULL ); + self->is_mutex_init = 1; + } + + // Lock the service + if ( take_lock ) + { pthread_mutex_lock( &self->audio_mutex ); pthread_mutex_lock( &self->video_mutex ); } @@ -749,35 +679,52 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha // Parse URL AVInputFormat *format = NULL; +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + AVDictionary *params = NULL; +#else AVFormatParameters params; memset( ¶ms, 0, sizeof(params) ); +#endif char *filename = parse_url( profile, URL, &format, ¶ms ); // Now attempt to open the file or device with filename +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0; + if ( error ) + // If the URL is a network stream URL, then we probably need to open with full URL + error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0; +#else error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0; if ( error ) // If the URL is a network stream URL, then we probably need to open with full URL error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0; +#endif // Set MLT properties onto video AVFormatContext if ( !error && self->video_format ) { 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 } +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + av_dict_free( ¶ms ); +#else // Cleanup AVFormatParameters if ( params.standard ) free( (void*) params.standard ); +#endif // If successful, then try to get additional info if ( !error && self->video_format ) { // Get the stream info +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + error = avformat_find_stream_info( self->video_format, NULL ) < 0; +#else error = av_find_stream_info( self->video_format ) < 0; +#endif // Continue if no error if ( !error && self->video_format ) @@ -787,7 +734,7 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha error = get_basic_info( self, profile, filename ); // Initialize position info - self->first_pts = -1; + self->first_pts = AV_NOPTS_VALUE; self->last_position = POSITION_INITIAL; if ( !self->audio_format ) @@ -800,13 +747,21 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha if ( self->seekable ) { // And open again for our audio context - av_open_input_file( &self->audio_format, filename, NULL, 0, NULL ); +#if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0) + avformat_open_input( &self->audio_format, filename, NULL, NULL ); apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM ); -#if LIBAVFORMAT_VERSION_MAJOR > 52 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data ) apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM ); + avformat_find_stream_info( self->audio_format, NULL ); +#else + av_open_input_file( &self->audio_format, filename, NULL, 0, NULL ); + apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM ); +#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(110<<8)+0) + if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data ) + apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM ); #endif av_find_stream_info( self->audio_format ); +#endif } else { @@ -839,8 +794,14 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha if ( self->dummy_context ) { + pthread_mutex_lock( &self->open_mutex ); +#if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0) + avformat_close_input( &self->dummy_context ); +#else av_close_input_file( self->dummy_context ); +#endif self->dummy_context = NULL; + pthread_mutex_unlock( &self->open_mutex ); } // Unlock the service @@ -854,45 +815,106 @@ static int producer_open( producer_avformat self, mlt_profile profile, const cha return error; } -static void reopen_video( producer_avformat self, mlt_producer producer ) +static void prepare_reopen( producer_avformat self ) { - mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer ); - mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) ); + mlt_service_lock( MLT_PRODUCER_SERVICE( self->parent ) ); pthread_mutex_lock( &self->audio_mutex ); + pthread_mutex_lock( &self->open_mutex ); - if ( self->video_codec ) + int i; + for ( i = 0; i < MAX_AUDIO_STREAMS; i++ ) { - avformat_lock(); - avcodec_close( self->video_codec ); - avformat_unlock(); + mlt_pool_release( self->audio_buffer[i] ); + self->audio_buffer[i] = NULL; + av_free( self->decode_buffer[i] ); + self->decode_buffer[i] = NULL; + if ( self->audio_codec[i] ) + avcodec_close( self->audio_codec[i] ); + self->audio_codec[i] = NULL; } + if ( self->video_codec ) + avcodec_close( self->video_codec ); self->video_codec = NULL; - if ( self->dummy_context ) - av_close_input_file( self->dummy_context ); - self->dummy_context = NULL; + +#if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0) + if ( self->seekable && self->audio_format ) + avformat_close_input( &self->audio_format ); + if ( self->video_format ) + avformat_close_input( &self->video_format ); +#else + if ( self->seekable && self->audio_format ) + av_close_input_file( self->audio_format ); if ( self->video_format ) av_close_input_file( self->video_format ); +#endif + self->audio_format = NULL; self->video_format = NULL; + pthread_mutex_unlock( &self->open_mutex ); + + // Cleanup the packet queues + AVPacket *pkt; + if ( self->apackets ) + { + while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) ) + { + av_free_packet( pkt ); + free( pkt ); + } + mlt_deque_close( self->apackets ); + self->apackets = NULL; + } + if ( self->vpackets ) + { + while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) ) + { + av_free_packet( pkt ); + free( pkt ); + } + mlt_deque_close( self->vpackets ); + self->vpackets = NULL; + } + pthread_mutex_unlock( &self->audio_mutex ); + mlt_service_unlock( MLT_PRODUCER_SERVICE( self->parent ) ); +} - int audio_index = self->audio_index; - int video_index = self->video_index; +static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts ) +{ + self->invalid_pts_counter += pts == AV_NOPTS_VALUE; + self->invalid_dts_counter += dts == AV_NOPTS_VALUE; + if ( ( self->invalid_pts_counter <= self->invalid_dts_counter + || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE ) + return pts; + else + return dts; +} - producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ), - mlt_properties_get( properties, "resource" ), 0 ); +static void find_first_pts( producer_avformat self, int video_index ) +{ + // find initial PTS + AVFormatContext *context = self->video_format? self->video_format : self->audio_format; + int ret = 0; + int toscan = 500; + AVPacket pkt; - self->audio_index = audio_index; - if ( self->video_format && video_index > -1 ) + while ( ret >= 0 && toscan-- > 0 ) { - self->video_index = video_index; - video_codec_init( self, video_index, properties ); + ret = av_read_frame( context, &pkt ); + if ( ret >= 0 && pkt.stream_index == video_index && ( pkt.flags & PKT_FLAG_KEY ) ) + { + mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), + "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n", + pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) ); + self->first_pts = best_pts( self, pkt.pts, pkt.dts ); + if ( self->first_pts != AV_NOPTS_VALUE ) + toscan = 0; + } + av_free_packet( &pkt ); } - - pthread_mutex_unlock( &self->audio_mutex ); - mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) ); + av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD ); } static int seek_video( producer_avformat self, mlt_position position, - int64_t req_position, int must_decode, int use_new_seek, int *ignore ) + int64_t req_position, int preseek ) { mlt_producer producer = self->parent; int paused = 0; @@ -914,94 +936,42 @@ static int seek_video( producer_avformat self, mlt_position position, double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) / mlt_properties_get_double( properties, "meta.media.frame_rate_den" ); - if ( self->av_frame && position + 1 == self->video_expected ) + if ( self->last_position == POSITION_INITIAL ) + find_first_pts( self, self->video_index ); + + if ( self->video_frame && position + 1 == self->video_expected ) { // We're paused - use last image paused = 1; } - else if ( !self->seekable && position > self->video_expected && ( position - self->video_expected ) < 250 ) - { - // Fast forward - seeking is inefficient for small distances - just ignore following frames - *ignore = ( int )( ( position - self->video_expected ) / mlt_producer_get_fps( producer ) * source_fps ); - codec_context->skip_loop_filter = AVDISCARD_NONREF; - } else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) ) { - if ( use_new_seek && self->last_position == POSITION_INITIAL ) - { - // find first key frame - int ret = 0; - int toscan = 100; - AVPacket pkt; - - while ( ret >= 0 && toscan-- > 0 ) - { - ret = av_read_frame( context, &pkt ); - if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == self->video_index ) - { - mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n", pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) ); - self->first_pts = pkt.pts; - toscan = 0; - } - av_free_packet( &pkt ); - } - // Rewind - av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD ); - } - // Calculate the timestamp for the requested frame - int64_t timestamp; - if ( use_new_seek ) - { - timestamp = ( req_position - 0.1 / source_fps ) / - ( av_q2d( stream->time_base ) * source_fps ); - mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "pos %"PRId64" pts %"PRId64" ", req_position, timestamp ); - if ( self->first_pts > 0 ) - timestamp += self->first_pts; - else if ( context->start_time != AV_NOPTS_VALUE ) - timestamp += context->start_time; - } - else - { - timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 ); - if ( context->start_time != AV_NOPTS_VALUE ) - timestamp += context->start_time; - } - if ( must_decode ) - timestamp -= AV_TIME_BASE; + int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps ); + if ( req_position <= 0 ) + timestamp = 0; + else if ( self->first_pts != AV_NOPTS_VALUE ) + timestamp += self->first_pts; + else if ( context->start_time != AV_NOPTS_VALUE ) + timestamp += context->start_time; + if ( preseek && av_q2d( self->video_time_base ) != 0 ) + timestamp -= 2 / av_q2d( self->video_time_base ); if ( timestamp < 0 ) timestamp = 0; - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %"PRId64"\n", + mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position " MLT_POSITION_FMT " expected "MLT_POSITION_FMT" last_pos %"PRId64"\n", timestamp, position, self->video_expected, self->last_position ); // Seek to the timestamp - if ( use_new_seek ) - { - codec_context->skip_loop_filter = AVDISCARD_NONREF; - av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD ); - } - else if ( req_position > 0 || self->last_position <= 0 ) - { - av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ); - } - else - { - // Re-open video stream when rewinding to beginning from somewhere else. - // This is rather ugly, and I prefer not to do it this way, but ffmpeg is - // not reliably seeking to the first frame across formats. - reopen_video( self, producer ); - } + codec_context->skip_loop_filter = AVDISCARD_NONREF; + av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD ); + + // flush any pictures still in decode buffer + avcodec_flush_buffers( codec_context ); // Remove the cached info relating to the previous position self->current_position = POSITION_INVALID; self->last_position = POSITION_INVALID; - av_freep( &self->av_frame ); - - if ( use_new_seek ) - { - // flush any pictures still in decode buffer - avcodec_flush_buffers( codec_context ); - } + av_freep( &self->video_frame ); } } return paused; @@ -1033,8 +1003,12 @@ static void get_audio_streams_info( producer_avformat self ) AVCodec *codec = avcodec_find_decoder( codec_context->codec_id ); // If we don't have a codec and we can't initialise it, we can't do much more... - avformat_lock( ); + pthread_mutex_lock( &self->open_mutex ); +#if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0) + if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 ) +#else if ( codec && avcodec_open( codec_context, codec ) >= 0 ) +#endif { self->audio_streams++; self->audio_max_stream = i; @@ -1045,19 +1019,15 @@ static void get_audio_streams_info( producer_avformat self ) self->max_frequency = codec_context->sample_rate; avcodec_close( codec_context ); } - avformat_unlock( ); + pthread_mutex_unlock( &self->open_mutex ); } } mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n", self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel ); - - // Other audio-specific initializations - self->resample_factor = 1.0; } -static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range ) +static void set_luma_transfer( struct SwsContext *context, int yuv_colorspace, int use_full_range ) { -#if defined(SWSCALE) && (LIBSWSCALE_VERSION_INT >= ((0<<16)+(7<<8)+2)) int *coefficients; const int *new_coefficients; int full_range; @@ -1069,7 +1039,7 @@ static void set_luma_transfer( struct SwsContext *context, int colorspace, int u // Don't change these from defaults unless explicitly told to. if ( use_full_range >= 0 ) full_range = use_full_range; - switch ( colorspace ) + switch ( yuv_colorspace ) { case 170: case 470: @@ -1090,15 +1060,67 @@ static void set_luma_transfer( struct SwsContext *context, int colorspace, int u sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range, brightness, contrast, saturation ); } +} + +static mlt_image_format pick_pix_format( enum PixelFormat pix_fmt ) +{ + switch ( pix_fmt ) + { + case PIX_FMT_ARGB: + case PIX_FMT_RGBA: + case PIX_FMT_ABGR: + case PIX_FMT_BGRA: + return mlt_image_rgb24a; + case PIX_FMT_YUV420P: + case PIX_FMT_YUVJ420P: + case PIX_FMT_YUVA420P: + return mlt_image_yuv420p; + case PIX_FMT_RGB24: + case PIX_FMT_BGR24: + case PIX_FMT_GRAY8: + case PIX_FMT_MONOWHITE: + case PIX_FMT_MONOBLACK: + case PIX_FMT_RGB8: + case PIX_FMT_BGR8: + return mlt_image_rgb24; + default: + return mlt_image_yuv422; + } +} + +static mlt_audio_format pick_audio_format( int sample_fmt ) +{ + switch ( sample_fmt ) + { + // interleaved + case AV_SAMPLE_FMT_U8: + return mlt_audio_u8; + case AV_SAMPLE_FMT_S16: + return mlt_audio_s16; + case AV_SAMPLE_FMT_S32: + return mlt_audio_s32le; + case AV_SAMPLE_FMT_FLT: + return mlt_audio_f32le; + // planar - this producer converts planar to interleaved +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0) + case AV_SAMPLE_FMT_U8P: + return mlt_audio_u8; + case AV_SAMPLE_FMT_S16P: + return mlt_audio_s16; + case AV_SAMPLE_FMT_S32P: + return mlt_audio_s32le; + case AV_SAMPLE_FMT_FLTP: + return mlt_audio_f32le; #endif + default: + return mlt_audio_none; + } } -static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, - mlt_image_format *format, int width, int height, int colorspace ) +static void convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt, + mlt_image_format *format, int width, int height, uint8_t **alpha ) { -#ifdef SWSCALE - int full_range = -1; - int flags = SWS_BILINEAR | SWS_ACCURATE_RND; + int flags = SWS_BICUBIC | SWS_ACCURATE_RND; #ifdef USE_MMX flags |= SWS_CPU_CAPS_MMX; @@ -1107,19 +1129,26 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, flags |= SWS_CPU_CAPS_MMX2; #endif - if ( pix_fmt == PIX_FMT_RGB32 ) + // extract alpha from planar formats + if ( ( pix_fmt == PIX_FMT_YUVA420P +#if defined(FFUDIV) && LIBAVUTIL_VERSION_INT >= ((51<<16)+(35<<8)+101) + || pix_fmt == PIX_FMT_YUVA444P +#endif + ) && + *format != mlt_image_rgb24a && *format != mlt_image_opengl && + frame->data[3] && frame->linesize[3] ) { - *format = mlt_image_rgb24a; - struct SwsContext *context = sws_getContext( width, height, pix_fmt, - width, height, PIX_FMT_RGBA, flags, NULL, NULL, NULL); - AVPicture output; - avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height ); - set_luma_transfer( context, colorspace, full_range ); - sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height, - output.data, output.linesize); - sws_freeContext( context ); + int i; + uint8_t *src, *dst; + + dst = *alpha = mlt_pool_alloc( width * height ); + src = frame->data[3]; + + for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ ) + memcpy( dst, src, FFMIN( width, frame->linesize[3] ) ); } - else if ( *format == mlt_image_yuv420p ) + + if ( *format == mlt_image_yuv420p ) { struct SwsContext *context = sws_getContext( width, height, pix_fmt, width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL); @@ -1130,7 +1159,7 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, output.linesize[0] = width; output.linesize[1] = width >> 1; output.linesize[2] = width >> 1; - set_luma_transfer( context, colorspace, full_range ); + set_luma_transfer( context, self->yuv_colorspace, -1 ); sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height, output.data, output.linesize); sws_freeContext( context ); @@ -1141,7 +1170,7 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL); AVPicture output; avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height ); - set_luma_transfer( context, colorspace, full_range ); + set_luma_transfer( context, self->yuv_colorspace, self->full_luma ); sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height, output.data, output.linesize); sws_freeContext( context ); @@ -1152,7 +1181,7 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL); AVPicture output; avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height ); - set_luma_transfer( context, colorspace, full_range ); + set_luma_transfer( context, self->yuv_colorspace, self->full_luma ); sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height, output.data, output.linesize); sws_freeContext( context ); @@ -1163,42 +1192,11 @@ static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL); AVPicture output; avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height ); - set_luma_transfer( context, colorspace, full_range ); + set_luma_transfer( context, self->yuv_colorspace, -1 ); sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height, output.data, output.linesize); sws_freeContext( context ); } -#else - if ( *format == mlt_image_yuv420p ) - { - AVPicture pict; - pict.data[0] = buffer; - pict.data[1] = buffer + width * height; - pict.data[2] = buffer + ( 5 * width * height ) / 4; - pict.linesize[0] = width; - pict.linesize[1] = width >> 1; - pict.linesize[2] = width >> 1; - img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height ); - } - else if ( *format == mlt_image_rgb24 ) - { - AVPicture output; - avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height ); - img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height ); - } - else if ( format == mlt_image_rgb24a || format == mlt_image_opengl ) - { - AVPicture output; - avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height ); - img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height ); - } - else - { - AVPicture output; - avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height ); - img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height ); - } -#endif } /** Allocate the image buffer and set it on the frame. @@ -1211,15 +1209,12 @@ static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint if ( codec_context->width == 0 || codec_context->height == 0 ) return size; + if ( *format == mlt_image_glsl ) + *format = pick_pix_format( codec_context->pix_fmt ); + *width = codec_context->width; *height = codec_context->height; - - if ( codec_context->pix_fmt == PIX_FMT_RGB32 ) - size = *width * ( *height + 1 ) * 4; - else - size = mlt_image_format_size( *format, *width, *height, NULL ); - - // Construct the output image + size = mlt_image_format_size( *format, *width, *height, NULL ); *buffer = mlt_pool_alloc( size ); if ( *buffer ) mlt_frame_set_image( frame, *buffer, size, mlt_pool_release ); @@ -1242,15 +1237,21 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame ); // Obtain the frame number of this frame - mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" ); + mlt_position position = mlt_frame_original_position( frame ); // Get the producer properties mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer ); pthread_mutex_lock( &self->video_mutex ); + uint8_t *alpha = NULL; + int got_picture = 0; + int image_size = 0; + // Fetch the video format context AVFormatContext *context = self->video_format; + if ( !context ) + goto exit_get_image; // Get the video stream AVStream *stream = context->streams[ self->video_index ]; @@ -1259,14 +1260,43 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form AVCodecContext *codec_context = stream->codec; // Get the image cache - if ( ! self->image_cache && ! mlt_properties_get_int( properties, "noimagecache" ) ) - self->image_cache = mlt_cache_init(); + if ( ! self->image_cache ) + { + // if cache size supplied by environment variable + int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL; + int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0; + + // cache size supplied via property + if ( mlt_properties_get( properties, "cache" ) ) + { + cache_supplied = 1; + cache_size = mlt_properties_get_int( properties, "cache" ); + } + if ( mlt_properties_get_int( properties, "noimagecache" ) ) + cache_size = 0; + // create cache if not disabled + if ( !cache_supplied || cache_size > 0 ) + self->image_cache = mlt_cache_init(); + // set cache size if supplied + if ( self->image_cache && cache_supplied ) + mlt_cache_set_size( self->image_cache, cache_size ); + } if ( self->image_cache ) { - mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position ); - uint8_t *original = mlt_cache_item_data( item, (int*) format ); + mlt_frame original = mlt_cache_get_frame( self->image_cache, position ); if ( original ) { + mlt_properties orig_props = MLT_FRAME_PROPERTIES( original ); + int size = 0; + + *buffer = mlt_properties_get_data( orig_props, "alpha", &size ); + if (*buffer) + mlt_frame_set_alpha( frame, *buffer, size, NULL ); + *buffer = mlt_properties_get_data( orig_props, "image", &size ); + mlt_frame_set_image( frame, *buffer, size, NULL ); + mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL ); + *format = mlt_properties_get_int( orig_props, "format" ); + // Set the resolution *width = codec_context->width; *height = codec_context->height; @@ -1275,34 +1305,11 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 ) *height = 1080; - // Cache hit - int size = mlt_image_format_size( *format, *width, *height, NULL ); - 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; - + got_picture = 1; goto exit_get_image; } } // Cache miss - int image_size = 0; - - // Packet - AVPacket pkt; - - // Special case ffwd handling - int ignore = 0; // We may want to use the source fps if available double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) / @@ -1313,31 +1320,38 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form // Determines if we have to decode all frames in a sequence // Temporary hack to improve intra frame only - int must_decode = strcmp( codec_context->codec->name, "dnxhd" ) && + int must_decode = !( codec_context->codec && codec_context->codec->name ) || ( + strcmp( codec_context->codec->name, "dnxhd" ) && strcmp( codec_context->codec->name, "dvvideo" ) && strcmp( codec_context->codec->name, "huffyuv" ) && strcmp( codec_context->codec->name, "mjpeg" ) && - strcmp( codec_context->codec->name, "rawvideo" ); + strcmp( codec_context->codec->name, "rawvideo" ) ); - // Turn on usage of new seek API and PTS for seeking - int use_new_seek = self->seekable && - codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" ); - if ( mlt_properties_get( properties, "new_seek" ) ) - use_new_seek = mlt_properties_get_int( properties, "new_seek" ); + double delay = mlt_properties_get_double( properties, "video_delay" ); // Seek if necessary - int paused = seek_video( self, position, req_position, must_decode, use_new_seek, &ignore ); + const char *interp = mlt_properties_get( frame_properties, "rescale.interp" ); + int preseek = must_decode +#if defined(FFUDIV) && LIBAVFORMAT_VERSION_INT >= ((53<<16)+(24<<8)+2) + && ( interp && strcmp( interp, "nearest" ) ) +#endif + && codec_context->has_b_frames; + int paused = seek_video( self, position, req_position, preseek ); // Seek might have reopened the file context = self->video_format; stream = context->streams[ self->video_index ]; codec_context = stream->codec; + if ( *format == mlt_image_none || + codec_context->pix_fmt == PIX_FMT_ARGB || + codec_context->pix_fmt == PIX_FMT_RGBA || + codec_context->pix_fmt == PIX_FMT_ABGR || + codec_context->pix_fmt == PIX_FMT_BGRA ) + *format = pick_pix_format( codec_context->pix_fmt ); // Duplicate the last image if necessary - if ( self->av_frame && self->av_frame->linesize[0] && self->got_picture - && ( paused - || self->current_position == req_position - || ( !use_new_seek && self->current_position > req_position ) ) ) + if ( self->video_frame && self->video_frame->linesize[0] + && ( paused || self->current_position >= req_position ) ) { // Duplicate it if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) ) @@ -1355,104 +1369,96 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form picture.linesize[0] = codec_context->width; picture.linesize[1] = codec_context->width / 2; picture.linesize[2] = codec_context->width / 2; - convert_image( (AVFrame*) &picture, *buffer, - PIX_FMT_YUV420P, format, *width, *height, self->colorspace ); + convert_image( self, (AVFrame*) &picture, *buffer, + PIX_FMT_YUV420P, format, *width, *height, &alpha ); } else #endif - convert_image( self->av_frame, *buffer, codec_context->pix_fmt, - format, *width, *height, self->colorspace ); + convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt, + format, *width, *height, &alpha ); + got_picture = 1; } - else - mlt_frame_get_image( frame, buffer, format, width, height, writable ); } else { int ret = 0; int64_t int_position = 0; int decode_errors = 0; - int got_picture = 0; - - av_init_packet( &pkt ); // Construct an AVFrame for YUV422 conversion - if ( !self->av_frame ) - self->av_frame = avcodec_alloc_frame( ); + if ( !self->video_frame ) + self->video_frame = avcodec_alloc_frame( ); while( ret >= 0 && !got_picture ) { // Read a packet + if ( self->pkt.stream_index == self->video_index ) + av_free_packet( &self->pkt ); + av_init_packet( &self->pkt ); pthread_mutex_lock( &self->packets_mutex ); if ( mlt_deque_count( self->vpackets ) ) { AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets ); - pkt = *tmp; + self->pkt = *tmp; free( tmp ); } else { - ret = av_read_frame( context, &pkt ); - if ( ret >= 0 && !self->seekable && pkt.stream_index == self->audio_index ) + ret = av_read_frame( context, &self->pkt ); + if ( ret >= 0 && !self->seekable && self->pkt.stream_index == self->audio_index ) { - if ( !av_dup_packet( &pkt ) ) + if ( !av_dup_packet( &self->pkt ) ) { AVPacket *tmp = malloc( sizeof(AVPacket) ); - *tmp = pkt; + *tmp = self->pkt; mlt_deque_push_back( self->apackets, tmp ); } } + else if ( ret < 0 ) + { + mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_image\n", ret ); + if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) ) + { + // Try to reconnect to live sources by closing context and codecs, + // and letting next call to get_frame() reopen. + prepare_reopen( self ); + pthread_mutex_unlock( &self->packets_mutex ); + goto exit_get_image; + } + if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) ) + { + mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" ); + exit( EXIT_FAILURE ); + } + } } pthread_mutex_unlock( &self->packets_mutex ); // We only deal with video from the selected video_index - if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 ) + if ( ret >= 0 && self->pkt.stream_index == self->video_index && self->pkt.size > 0 ) { - // Determine time code of the packet - if ( use_new_seek ) + int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts ); + if ( pts != AV_NOPTS_VALUE ) { - int64_t pts = pkt.pts; - if ( self->first_pts > 0 ) + if ( !self->seekable && self->first_pts == AV_NOPTS_VALUE ) + self->first_pts = pts; + if ( self->first_pts != AV_NOPTS_VALUE ) pts -= self->first_pts; else if ( context->start_time != AV_NOPTS_VALUE ) pts -= context->start_time; - int_position = ( int64_t )( av_q2d( stream->time_base ) * pts * source_fps + 0.1 ); - if ( pkt.pts == AV_NOPTS_VALUE ) - { - self->invalid_pts_counter++; - if ( self->invalid_pts_counter > 20 ) - { - mlt_log_panic( MLT_PRODUCER_SERVICE(producer), "\ainvalid PTS; DISABLING NEW_SEEK!\n" ); - mlt_properties_set_int( properties, "new_seek", 0 ); - int_position = req_position; - use_new_seek = 0; - } - } - else - { - self->invalid_pts_counter = 0; - } - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.pts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n", - pkt.pts, req_position, self->current_position, int_position ); + int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 ); + if ( int_position == self->last_position ) + int_position = self->last_position + 1; } - else + mlt_log_debug( MLT_PRODUCER_SERVICE(producer), + "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n", + self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position ); + + // Make a dumb assumption on streams that contain wild timestamps + if ( abs( req_position - int_position ) > 999 ) { - if ( pkt.dts != AV_NOPTS_VALUE ) - { - double delay = mlt_properties_get_double( properties, "video_delay" ); - int_position = ( int64_t )( ( av_q2d( stream->time_base ) * pkt.dts + delay ) * source_fps + 0.5 ); - if ( context->start_time != AV_NOPTS_VALUE ) - int_position -= ( int64_t )( context->start_time * source_fps / AV_TIME_BASE + 0.5 ); - if ( int_position == self->last_position ) - int_position = self->last_position + 1; - } - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n", - pkt.dts, req_position, self->current_position, int_position ); - // Make a dumb assumption on streams that contain wild timestamps - if ( abs( req_position - int_position ) > 999 ) - { - int_position = req_position; - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" ); - } + int_position = req_position; + mlt_log_warning( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!\n" ); } self->last_position = int_position; @@ -1460,30 +1466,30 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form if ( must_decode || int_position >= req_position ) { #ifdef VDPAU - if ( g_vdpau && self->vdpau ) + if ( self->vdpau ) { - if ( g_vdpau->producer != self ) + if ( self->vdpau->decoder == VDP_INVALID_HANDLE ) { - vdpau_decoder_close(); vdpau_decoder_init( self ); } - if ( self->vdpau ) - self->vdpau->is_decoded = 0; + self->vdpau->is_decoded = 0; } #endif - codec_context->reordered_opaque = pkt.pts; + codec_context->reordered_opaque = int_position; if ( int_position >= req_position ) codec_context->skip_loop_filter = AVDISCARD_NONE; #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0)) - ret = avcodec_decode_video2( codec_context, self->av_frame, &got_picture, &pkt ); + ret = avcodec_decode_video2( codec_context, self->video_frame, &got_picture, &self->pkt ); #else - ret = avcodec_decode_video( codec_context, self->av_frame, &got_picture, pkt.data, pkt.size ); + ret = avcodec_decode_video( codec_context, self->video_frame, &got_picture, self->pkt.data, self->pkt.size ); #endif // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors. if ( ret < 0 ) { if ( ++decode_errors <= 10 ) ret = 0; + else + mlt_log_warning( MLT_PRODUCER_SERVICE(producer), "video decoding error %d\n", ret ); } else { @@ -1493,34 +1499,26 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form if ( got_picture ) { - if ( use_new_seek ) + // Get position of reordered frame + int_position = self->video_frame->reordered_opaque; +#if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(106<<8)+0)) + pts = best_pts( self, self->video_frame->pkt_pts, self->video_frame->pkt_dts ); + if ( pts != AV_NOPTS_VALUE ) { - // Determine time code of the packet - int64_t pts = self->av_frame->reordered_opaque; - if ( self->first_pts > 0 ) + if ( self->first_pts != AV_NOPTS_VALUE ) pts -= self->first_pts; else if ( context->start_time != AV_NOPTS_VALUE ) pts -= context->start_time; - int_position = ( int64_t )( av_q2d( stream->time_base) * pts * source_fps + 0.1 ); - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "got frame %"PRId64", key %d\n", int_position, self->av_frame->key_frame ); + int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 ); } - // Handle ignore +#endif + if ( int_position < req_position ) - { - ignore = 0; got_picture = 0; - } else if ( int_position >= req_position ) - { - ignore = 0; codec_context->skip_loop_filter = AVDISCARD_NONE; - } - else if ( ignore -- ) - { - got_picture = 0; - } } - mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY ); + mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, self->pkt.flags & PKT_FLAG_KEY ); } // Now handle the picture if we have one @@ -1536,25 +1534,25 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form { if ( self->vdpau->is_decoded ) { - struct vdpau_render_state *render = (struct vdpau_render_state*) self->av_frame->data[0]; + struct vdpau_render_state *render = (struct vdpau_render_state*) self->video_frame->data[0]; void *planes[3]; uint32_t pitches[3]; VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12; if ( !self->vdpau->buffer ) self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 ); - self->av_frame->data[0] = planes[0] = self->vdpau->buffer; - self->av_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height; - self->av_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4; - self->av_frame->linesize[0] = pitches[0] = codec_context->width; - self->av_frame->linesize[1] = pitches[1] = codec_context->width / 2; - self->av_frame->linesize[2] = pitches[2] = codec_context->width / 2; + self->video_frame->data[0] = planes[0] = self->vdpau->buffer; + self->video_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height; + self->video_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4; + self->video_frame->linesize[0] = pitches[0] = codec_context->width; + self->video_frame->linesize[1] = pitches[1] = codec_context->width / 2; + self->video_frame->linesize[2] = pitches[2] = codec_context->width / 2; VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches ); if ( status == VDP_STATUS_OK ) { - convert_image( self->av_frame, *buffer, PIX_FMT_YUV420P, - format, *width, *height, self->colorspace ); + convert_image( self, self->video_frame, *buffer, PIX_FMT_YUV420P, + format, *width, *height, &alpha ); } else { @@ -1570,31 +1568,38 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form } else #endif - convert_image( self->av_frame, *buffer, codec_context->pix_fmt, - format, *width, *height, self->colorspace ); - self->top_field_first |= self->av_frame->top_field_first; + convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt, + format, *width, *height, &alpha ); + self->top_field_first |= self->video_frame->top_field_first; self->current_position = int_position; - self->got_picture = 1; } else { got_picture = 0; } } - if ( self->seekable || pkt.stream_index != self->audio_index ) - av_free_packet( &pkt ); + + // Free packet data if not video and not live audio packet + if ( self->pkt.stream_index != self->video_index && + !( !self->seekable && self->pkt.stream_index == self->audio_index ) ) + av_free_packet( &self->pkt ); } } - if ( self->got_picture && image_size > 0 && self->image_cache ) + // set alpha + if ( alpha ) + mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release ); + + if ( image_size > 0 && self->image_cache ) { - // Copy buffer to image cache - uint8_t *image = mlt_pool_alloc( image_size ); - memcpy( image, *buffer, image_size ); - mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release ); + mlt_properties_set_int( frame_properties, "format", *format ); + mlt_cache_put_frame( self->image_cache, frame ); } + // Try to duplicate last image if there was a decoding failure - else if ( !image_size && self->av_frame && self->av_frame->linesize[0] ) + // TODO: with multithread decoding a partial frame decoding resulting + // in failure also resets av_frame making test below fail. + if ( !image_size && self->video_frame && self->video_frame->linesize[0] ) { // Duplicate it if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) ) @@ -1612,17 +1617,15 @@ static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_form picture.linesize[0] = codec_context->width; picture.linesize[1] = codec_context->width / 2; picture.linesize[2] = codec_context->width / 2; - convert_image( (AVFrame*) &picture, *buffer, - PIX_FMT_YUV420P, format, *width, *height, self->colorspace ); + convert_image( self, (AVFrame*) &picture, *buffer, + PIX_FMT_YUV420P, format, *width, *height, &alpha ); } else #endif - convert_image( self->av_frame, *buffer, codec_context->pix_fmt, - format, *width, *height, self->colorspace ); - self->got_picture = 1; + convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt, + format, *width, *height, &alpha ); + got_picture = 1; } - else - mlt_frame_get_image( frame, buffer, format, width, height, writable ); } // Regardless of speed, we expect to get the next frame (cos we ain't too bright) @@ -1635,8 +1638,8 @@ exit_get_image: // Set the progressive flag if ( mlt_properties_get( properties, "force_progressive" ) ) mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) ); - else if ( self->av_frame ) - mlt_properties_set_int( frame_properties, "progressive", !self->av_frame->interlaced_frame ); + else if ( self->video_frame ) + mlt_properties_set_int( frame_properties, "progressive", !self->video_frame->interlaced_frame ); // Set the field order property for this frame if ( mlt_properties_get( properties, "force_tff" ) ) @@ -1650,7 +1653,12 @@ exit_get_image: mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) ); mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) ); - return !self->got_picture; + // If we already have RGB, then the full range processing either happened already + // or does not apply (RGB source). + if ( *format == mlt_image_rgb24 || *format == mlt_image_rgb24a || *format == mlt_image_opengl ) + mlt_properties_set( frame_properties, "force_full_luma", NULL ); + + return !got_picture; } /** Process properties as AVOptions and apply to AV context obj @@ -1663,11 +1671,17 @@ static void apply_properties( void *obj, mlt_properties properties, int flags ) for ( i = 0; i < count; i++ ) { const char *opt_name = mlt_properties_get_name( properties, i ); +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(10<<8)+0) + const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, flags ); +#else const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags ); +#endif if ( opt_name && mlt_properties_get( properties, opt_name ) ) { if ( opt ) -#if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0) +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(12<<8)+0) + av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), 0 ); +#elif LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0) av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL ); #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0) av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 ); @@ -1695,7 +1709,7 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p // Find the codec AVCodec *codec = avcodec_find_decoder( codec_context->codec_id ); #ifdef VDPAU - if ( codec_context->codec_id == CODEC_ID_H264 ) + if ( codec_context->codec_id == AV_CODEC_ID_H264 ) { if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) ) { @@ -1703,7 +1717,7 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p { self->video_codec = codec_context; if ( !vdpau_decoder_init( self ) ) - vdpau_decoder_close(); + vdpau_fini( self ); } } if ( !self->vdpau ) @@ -1719,8 +1733,12 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p codec_context->thread_count = thread_count; // If we don't have a codec and we can't initialise it, we can't do much more... - avformat_lock( ); + pthread_mutex_lock( &self->open_mutex ); +#if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0) + if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 ) +#else if ( codec && avcodec_open( codec_context, codec ) >= 0 ) +#endif { // Now store the codec with its destructor self->video_codec = codec_context; @@ -1729,109 +1747,131 @@ static int video_codec_init( producer_avformat self, int index, mlt_properties p { // Remember that we can't use this later self->video_index = -1; - avformat_unlock( ); + pthread_mutex_unlock( &self->open_mutex ); return 0; } - avformat_unlock( ); + pthread_mutex_unlock( &self->open_mutex ); // 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 LIBAVCODEC_VERSION_INT >= ((52<<16)+(122<<8)+0) 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 ); - mlt_properties_set_int( properties, "height", self->video_codec->height ); - // For DV, we'll just use the saved aspect ratio - if ( codec_context->codec_id != CODEC_ID_DVVIDEO ) - get_aspect_ratio( properties, stream, self->video_codec, NULL ); - - // Determine the fps first from the codec - double source_fps = (double) self->video_codec->time_base.den / - ( self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num ); - - if ( mlt_properties_get( properties, "force_fps" ) ) + if ( self->video_codec ) { - source_fps = mlt_properties_get_double( properties, "force_fps" ); - stream->time_base = av_d2q( source_fps, 1024 ); - mlt_properties_set_int( properties, "meta.media.frame_rate_num", stream->time_base.num ); - mlt_properties_set_int( properties, "meta.media.frame_rate_den", stream->time_base.den ); + mlt_properties_set_int( properties, "width", self->video_codec->width ); + mlt_properties_set_int( properties, "height", self->video_codec->height ); + get_aspect_ratio( properties, stream, self->video_codec ); } - else - { - // If the muxer reports a frame rate different than the codec + + // Start with the muxer frame rate. #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) - double muxer_fps = av_q2d( stream->avg_frame_rate ); - if ( isnan( muxer_fps ) || muxer_fps == 0 ) - muxer_fps = av_q2d( stream->r_frame_rate ); + AVRational frame_rate = stream->avg_frame_rate; #else - double muxer_fps = av_q2d( stream->r_frame_rate ); + AVRational frame_rate = stream->r_frame_rate; #endif - // Choose the lesser - the wrong tends to be off by some multiple of 10 - source_fps = FFMIN( source_fps, muxer_fps ); - if ( source_fps >= 1.0 && ( source_fps < muxer_fps || isnan( muxer_fps ) ) ) - { - mlt_properties_set_int( properties, "meta.media.frame_rate_num", self->video_codec->time_base.den ); - mlt_properties_set_int( properties, "meta.media.frame_rate_den", self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num ); - } - else if ( muxer_fps > 0 ) - { - AVRational frame_rate = stream->r_frame_rate; - // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid, - // avg_frame_rate gives some approximate value that does not well match the media. - // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly - // results in some very choppy output, but some value slightly different works - // great. -#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) - if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 ) - frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 ); + double fps = av_q2d( frame_rate ); + +#if LIBAVFORMAT_VERSION_MAJOR < 55 + // Verify and sanitize the muxer frame rate. + if ( isnan( fps ) || isinf( fps ) || fps == 0 ) + { + frame_rate = stream->r_frame_rate; + fps = av_q2d( frame_rate ); + } #endif - mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num ); - mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den ); - } - else - { - source_fps = mlt_producer_get_fps( self->parent ); - AVRational frame_rate = av_d2q( source_fps, 255 ); - mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num ); - mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den ); - } +#if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) && LIBAVFORMAT_VERSION_MAJOR < 55 + // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid, + // avg_frame_rate gives some approximate value that does not well match the media. + // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly + // results in some very choppy output, but some value slightly different works + // great. + if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 ) + { + frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 ); + fps = av_q2d( frame_rate ); + } +#endif + // XXX frame rates less than 1 fps are not considered sane + if ( isnan( fps ) || isinf( fps ) || fps < 1.0 ) + { + // Get the frame rate from the codec. + frame_rate.num = self->video_codec->time_base.den; + frame_rate.den = self->video_codec->time_base.num * self->video_codec->ticks_per_frame; + fps = av_q2d( frame_rate ); + } + if ( isnan( fps ) || isinf( fps ) || fps < 1.0 ) + { + // Use the profile frame rate if all else fails. + mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) ); + frame_rate.num = profile->frame_rate_num; + frame_rate.den = profile->frame_rate_den; } - // source_fps is deprecated in favor of meta.media.frame_rate_num and .frame_rate_den - if ( source_fps > 0 ) - mlt_properties_set_double( properties, "source_fps", source_fps ); - else - mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( self->parent ) ); + self->video_time_base = stream->time_base; + if ( mlt_properties_get( properties, "force_fps" ) ) + { + AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 ); + self->video_time_base = av_mul_q( stream->time_base, av_div_q( frame_rate, force_fps ) ); + frame_rate = force_fps; + } + mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num ); + mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den ); // Set the YUV colorspace from override or detect - self->colorspace = mlt_properties_get_int( properties, "force_colorspace" ); -#if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0) - if ( ! self->colorspace ) + self->yuv_colorspace = mlt_properties_get_int( properties, "force_colorspace" ); +#if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0) + if ( ! self->yuv_colorspace ) { switch ( self->video_codec->colorspace ) { case AVCOL_SPC_SMPTE240M: - self->colorspace = 240; + self->yuv_colorspace = 240; break; case AVCOL_SPC_BT470BG: case AVCOL_SPC_SMPTE170M: - self->colorspace = 601; + self->yuv_colorspace = 601; break; case AVCOL_SPC_BT709: - self->colorspace = 709; + self->yuv_colorspace = 709; break; default: // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV" - self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601; + self->yuv_colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601; break; } } #endif // Let apps get chosen colorspace - mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace ); + mlt_properties_set_int( properties, "meta.media.colorspace", self->yuv_colorspace ); + + switch ( self->video_codec->color_primaries ) + { + case AVCOL_PRI_BT470BG: + self->color_primaries = 601625; + break; + case AVCOL_PRI_SMPTE170M: + case AVCOL_PRI_SMPTE240M: + self->color_primaries = 601525; + break; + case AVCOL_PRI_BT709: + case AVCOL_PRI_UNSPECIFIED: + default: + self->color_primaries = 709; + break; + } + + self->full_luma = -1; +#if LIBAVCODEC_VERSION_INT >= ((52<<16)+(72<<8)+2) + mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range ); + if ( codec_context->color_range == AVCOL_RANGE_JPEG ) + self->full_luma = 1; +#endif + if ( mlt_properties_get( properties, "set.force_full_luma" ) ) + self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" ); } return self->video_codec && self->video_index > -1; } @@ -1853,11 +1893,15 @@ static void producer_set_up_video( producer_avformat self, mlt_frame frame ) // Get the video_index int index = mlt_properties_get_int( properties, "video_index" ); + int unlock_needed = 0; + // Reopen the file if necessary if ( !context && index > -1 ) { + unlock_needed = 1; + pthread_mutex_lock( &self->video_mutex ); producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ), - mlt_properties_get( properties, "resource" ), 1 ); + mlt_properties_get( properties, "resource" ), 0 ); context = self->video_format; } @@ -1882,13 +1926,11 @@ static void producer_set_up_video( producer_avformat self, mlt_frame frame ) { // Reset the video properties if the index changed self->video_index = index; + pthread_mutex_lock( &self->open_mutex ); if ( self->video_codec ) - { - avformat_lock(); avcodec_close( self->video_codec ); - avformat_unlock(); - } self->video_codec = NULL; + pthread_mutex_unlock( &self->open_mutex ); } // Get the frame properties @@ -1905,19 +1947,17 @@ static void producer_set_up_video( producer_avformat self, mlt_frame frame ) // Set the width and height mlt_properties_set_int( frame_properties, "width", self->video_codec->width ); mlt_properties_set_int( frame_properties, "height", self->video_codec->height ); - // real_width and real_height are deprecated in favor of meta.media.width and .height mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width ); mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height ); - mlt_properties_set_int( frame_properties, "real_width", self->video_codec->width ); - mlt_properties_set_int( frame_properties, "real_height", self->video_codec->height ); mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio ); - mlt_properties_set_int( frame_properties, "colorspace", self->colorspace ); + mlt_properties_set_int( frame_properties, "colorspace", self->yuv_colorspace ); + mlt_properties_set_int( frame_properties, "color_primaries", self->color_primaries ); + mlt_properties_set_int( frame_properties, "full_luma", self->full_luma ); // Workaround 1088 encodings missing cropping info. if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 ) { mlt_properties_set_int( properties, "meta.media.height", 1080 ); - mlt_properties_set_int( frame_properties, "real_height", 1080 ); } // Add our image operation @@ -1929,25 +1969,31 @@ static void producer_set_up_video( producer_avformat self, mlt_frame frame ) // If something failed, use test card image mlt_properties_set_int( frame_properties, "test_image", 1 ); } + if ( unlock_needed ) + pthread_mutex_unlock( &self->video_mutex ); } -static int seek_audio( producer_avformat self, mlt_position position, double timecode, int *ignore ) +static int seek_audio( producer_avformat self, mlt_position position, double timecode ) { int paused = 0; // Seek if necessary - if ( self->seekable && position != self->audio_expected ) + if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) ) { + if ( self->last_position == POSITION_INITIAL ) + { + int video_index = self->video_index; + if ( video_index == -1 ) + video_index = first_video_index( self ); + if ( video_index >= 0 ) + find_first_pts( self, video_index ); + } + if ( position + 1 == self->audio_expected ) { // We're paused - silence required paused = 1; } - else if ( !self->seekable && position > self->audio_expected && ( position - self->audio_expected ) < 250 ) - { - // Fast forward - seeking is inefficient for small distances - just ignore following frames - *ignore = position - self->audio_expected; - } else if ( position < self->audio_expected || position - self->audio_expected >= 12 ) { AVFormatContext *context = self->audio_format; @@ -1972,13 +2018,43 @@ static int seek_audio( producer_avformat self, mlt_position position, double tim static int sample_bytes( AVCodecContext *context ) { -#if LIBAVCODEC_VERSION_MAJOR > 52 +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0) + return av_get_bytes_per_sample( context->sample_fmt ); +#elif LIBAVCODEC_VERSION_MAJOR >= 53 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8; #else return av_get_bits_per_sample_format( context->sample_fmt ) / 8; #endif } +#if LIBAVCODEC_VERSION_MAJOR >= 55 +static void planar_to_interleaved( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample ) +{ + int s, c; + for ( s = 0; s < samples; s++ ) + { + for ( c = 0; c < channels; c++ ) + { + memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample ); + dest += bytes_per_sample; + } + } +} +#else +static void planar_to_interleaved( uint8_t *dest, uint8_t *src, int samples, int channels, int bytes_per_sample ) +{ + int s, c; + for ( s = 0; s < samples; s++ ) + { + for ( c = 0; c < channels; c++ ) + { + memcpy( dest, src + ( c * samples + s ) * bytes_per_sample, bytes_per_sample ); + dest += bytes_per_sample; + } + } +} +#endif + static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps ) { // Fetch the audio_format @@ -1990,9 +2066,6 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int // Get codec context AVCodecContext *codec_context = self->audio_codec[ index ]; - // Obtain the resample context if it exists (not always needed) - ReSampleContext *resample = self->audio_resample[ index ]; - // Obtain the audio buffers uint8_t *audio_buffer = self->audio_buffer[ index ]; uint8_t *decode_buffer = self->decode_buffer[ index ]; @@ -2004,16 +2077,23 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int while ( ptr && ret >= 0 && len > 0 ) { - int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context ); + int sizeof_sample = 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, (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 ); +#if LIBAVCODEC_VERSION_MAJOR >= 55 + if ( !self->audio_frame ) + self->audio_frame = avcodec_alloc_frame(); + else + avcodec_get_frame_defaults( self->audio_frame ); + ret = avcodec_decode_audio4( codec_context, self->audio_frame, &data_size, &pkt ); + if ( data_size ) { + data_size = av_samples_get_buffer_size( NULL, codec_context->channels, + self->audio_frame->nb_samples, codec_context->sample_fmt, 1 ); + decode_buffer = self->audio_frame->data[0]; + } #else - ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len ); + ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt ); #endif if ( ret < 0 ) { @@ -2029,61 +2109,74 @@ static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int { // Figure out how many samples will be needed after resampling 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 + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] ) + if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] ) { - self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample; + self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 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 - 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 + uint8_t *dest = &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ]; + switch ( codec_context->sample_fmt ) { +#if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0) + case AV_SAMPLE_FMT_U8P: + case AV_SAMPLE_FMT_S16P: + case AV_SAMPLE_FMT_S32P: + case AV_SAMPLE_FMT_FLTP: +#if LIBAVCODEC_VERSION_MAJOR >= 55 + planar_to_interleaved( dest, self->audio_frame, convert_samples, codec_context->channels, sizeof_sample ); +#else + planar_to_interleaved( dest, decode_buffer, convert_samples, codec_context->channels, sizeof_sample ); +#endif + break; +#endif + default: // Straight copy to audio buffer - memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size ); - audio_used += convert_samples; + memcpy( dest, decode_buffer, data_size ); } + audio_used += convert_samples; // Handle ignore - while ( *ignore && audio_used > samples ) + while ( *ignore && audio_used ) { *ignore -= 1; - audio_used -= samples; - memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ], + audio_used -= audio_used > samples ? samples : audio_used; + memmove( audio_buffer, &audio_buffer[ samples * codec_context->channels * sizeof_sample ], audio_used * sizeof_sample ); } } } // If we're behind, ignore this packet - if ( pkt.pts >= 0 ) - { - double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt.pts; + // Skip this on non-seekable, audio-only inputs. + if ( pkt.pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 ) + { + int64_t pts = pkt.pts; + if ( self->first_pts != AV_NOPTS_VALUE ) + pts -= self->first_pts; + else if ( context->start_time != AV_NOPTS_VALUE ) + pts -= context->start_time; + double timebase = av_q2d( context->streams[ index ]->time_base ); + int64_t int_position = ( int64_t )( timebase * pts * fps + 0.5 ); int64_t req_position = ( int64_t )( timecode * fps + 0.5 ); - int64_t int_position = ( int64_t )( current_pts * fps + 0.5 ); - if ( context->start_time != AV_NOPTS_VALUE ) - int_position -= ( int64_t )( fps * context->start_time / AV_TIME_BASE + 0.5 ); - if ( *ignore == 0 ) + mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), + "A pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n", + pkt.pts, pkt.dts, req_position, self->current_position, int_position ); + + if ( int_position > 0 ) { if ( int_position < req_position ) // We are behind, so skip some - *ignore = 1; - - // We use nb_streams in this test because the tolerance is dependent - // on the interleaving of all streams esp. when there is more than - // one audio stream. - else if ( int_position > req_position + context->nb_streams ) + *ignore = req_position - int_position; + else if ( self->audio_index != INT_MAX && int_position > req_position + 2 ) // We are ahead, so seek backwards some more - seek_audio( self, req_position, timecode - 1.0, ignore ); + seek_audio( self, req_position, timecode - 1.0 ); } + // Cancel the find_first_pts() in seek_audio() + if ( self->video_index == -1 && self->last_position == POSITION_INITIAL ) + self->last_position = int_position; } self->audio_used[ index ] = audio_used; @@ -2101,7 +2194,7 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format pthread_mutex_lock( &self->audio_mutex ); // Obtain the frame number of this frame - mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" ); + mlt_position position = mlt_frame_original_position( frame ); // Calculate the real time code double real_timecode = producer_time_of_frame( self->parent, position ); @@ -2113,7 +2206,7 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format int ignore[ MAX_AUDIO_STREAMS ] = { 0 }; // Flag for paused (silence) - int paused = seek_audio( self, position, real_timecode, &ignore[0] ); + int paused = seek_audio( self, position, real_timecode ); // Initialize ignore for all streams from the seek return value int i = MAX_AUDIO_STREAMS; @@ -2122,6 +2215,8 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format // Fetch the audio_format AVFormatContext *context = self->audio_format; + if ( !context ) + goto exit_get_audio; int sizeof_sample = sizeof( int16_t ); @@ -2131,49 +2226,25 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format if ( self->audio_index == INT_MAX ) { index = 0; - index_max = context->nb_streams; + index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams ); *channels = self->total_channels; - *samples = *samples * FFMAX( self->max_frequency, *frequency ) / *frequency; - *frequency = FFMAX( self->max_frequency, *frequency ); + *samples = mlt_sample_calculator( fps, self->max_frequency, position ); + *frequency = self->max_frequency; } - // Initialize the resamplers and buffers - for ( ; index < index_max; index++ ) + // Initialize the buffers + for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ ) { // Get codec context AVCodecContext *codec_context = self->audio_codec[ index ]; if ( codec_context && !self->audio_buffer[ index ] ) { - // Check for resample and create if necessary - if ( codec_context->channels <= 2 ) - { - // Determine by how much resampling will increase number of samples - double resample_factor = self->audio_index == INT_MAX ? 1 : (double) *channels / codec_context->channels; - resample_factor *= (double) *frequency / codec_context->sample_rate; - if ( resample_factor > self->resample_factor ) - self->resample_factor = resample_factor; - - // Create the resampler -#if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0)) - self->audio_resample[ index ] = av_audio_resample_init( - self->audio_index == INT_MAX ? codec_context->channels : *channels, - codec_context->channels, *frequency, codec_context->sample_rate, - SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 ); -#else - self->audio_resample[ index ] = audio_resample_init( - self->audio_index == INT_MAX ? codec_context->channels : *channels, - codec_context->channels, *frequency, codec_context->sample_rate ); -#endif - } - else - { - codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels; - sizeof_sample = sample_bytes( codec_context ); - } + 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 * sizeof_sample; + self->audio_buffer_size[ index ] = 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 @@ -2182,7 +2253,7 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format } // Get the audio if required - if ( !paused ) + if ( !paused && *frequency > 0 ) { int ret = 0; int got_audio = 0; @@ -2190,11 +2261,9 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format av_init_packet( &pkt ); - // If not resampling, give consumer more than requested. - // It requested number samples based on requested frame rate. - // Do not clean this up with a samples *= ...! - if ( self->audio_index != INT_MAX && ! self->audio_resample[ self->audio_index ] ) - *samples = *samples * self->audio_codec[ self->audio_index ]->sample_rate / *frequency; + // Caller requested number samples based on requested sample rate. + if ( self->audio_index != INT_MAX ) + *samples = mlt_sample_calculator( fps, self->audio_codec[ self->audio_index ]->sample_rate, position ); while ( ret >= 0 && !got_audio ) { @@ -2210,7 +2279,7 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format { // Check if there is enough audio for all streams got_audio = 1; - for ( index = 0; got_audio && index < context->nb_streams; index++ ) + for ( index = 0; got_audio && index < index_max; index++ ) if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] ) got_audio = 0; if ( got_audio ) @@ -2237,16 +2306,34 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format mlt_deque_push_back( self->vpackets, tmp ); } } + else if ( ret < 0 ) + { + mlt_producer producer = self->parent; + mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer ); + mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_audio\n", ret ); + if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) ) + { + // Try to reconnect to live sources by closing context and codecs, + // and letting next call to get_frame() reopen. + prepare_reopen( self ); + pthread_mutex_unlock( &self->packets_mutex ); + goto exit_get_audio; + } + if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) ) + { + mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" ); + exit( EXIT_FAILURE ); + } + } } pthread_mutex_unlock( &self->packets_mutex ); // We only deal with audio from the selected audio index index = pkt.stream_index; - if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index || + if ( index < MAX_AUDIO_STREAMS && 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_resample[index] ) ? - self->audio_codec[index]->channels : *channels; + int channels2 = self->audio_codec[index]->channels; ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps ); } @@ -2257,25 +2344,21 @@ 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 ] ) + if ( self->audio_index != INT_MAX ) { 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; + *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt ); 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 ] ) + if ( self->audio_codec[ 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; + // XXX: This only works if all audio tracks have the same sample format. + *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt ); sizeof_sample = sample_bytes( self->audio_codec[ index ] ); break; } @@ -2339,6 +2422,7 @@ static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format } else { +exit_get_audio: // Get silence and don't touch the context mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples ); } @@ -2367,8 +2451,12 @@ static int audio_codec_init( producer_avformat self, int index, mlt_properties p AVCodec *codec = avcodec_find_decoder( codec_context->codec_id ); // If we don't have a codec and we can't initialise it, we can't do much more... - avformat_lock( ); + pthread_mutex_lock( &self->open_mutex ); +#if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0) + if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 ) +#else if ( codec && avcodec_open( codec_context, codec ) >= 0 ) +#endif { // Now store the codec with its destructor if ( self->audio_codec[ index ] ) @@ -2380,14 +2468,12 @@ static int audio_codec_init( producer_avformat self, int index, mlt_properties p // Remember that we can't use self later self->audio_index = -1; } - avformat_unlock( ); + pthread_mutex_unlock( &self->open_mutex ); // 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; } @@ -2439,17 +2525,20 @@ static void producer_set_up_audio( producer_avformat self, mlt_frame frame ) index = self->audio_index; mlt_properties_set_int( properties, "audio_index", index ); } + if ( context && index > -1 && index < INT_MAX && + pick_audio_format( context->streams[ index ]->codec->sample_fmt ) == mlt_audio_none ) + { + index = -1; + } // Update the audio properties if the index changed if ( context && index > -1 && index != self->audio_index ) { + pthread_mutex_lock( &self->open_mutex ); if ( self->audio_codec[ self->audio_index ] ) - { - avformat_lock(); avcodec_close( self->audio_codec[ self->audio_index ] ); - avformat_unlock(); - } self->audio_codec[ self->audio_index ] = NULL; + pthread_mutex_unlock( &self->open_mutex ); } if ( self->audio_index != -1 ) self->audio_index = index; @@ -2467,14 +2556,11 @@ static void producer_set_up_audio( producer_avformat self, mlt_frame frame ) audio_codec_init( self, index, properties ); } } - else if ( context && index > -1 && audio_codec_init( self, index, properties ) ) + else if ( context && index > -1 && index < MAX_AUDIO_STREAMS && + audio_codec_init( self, index, properties ) ) { - // Set the frame properties - if ( index < INT_MAX ) - { - mlt_properties_set_int( frame_properties, "frequency", self->audio_codec[ index ]->sample_rate ); - mlt_properties_set_int( frame_properties, "channels", self->audio_codec[ index ]->channels ); - } + mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate ); + mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels ); } if ( context && index > -1 ) { @@ -2520,16 +2606,16 @@ static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int i // Update timecode on the frame we're creating mlt_frame_set_position( *frame, mlt_producer_position( producer ) ); - // Set the position of this producer - mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++; - mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", position ); - // Set up the video producer_set_up_video( self, *frame ); // Set up the audio producer_set_up_audio( self, *frame ); + // Set the position of this producer + mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++; + mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position ); + // Calculate the next timecode mlt_producer_prepare_next( producer ); @@ -2541,28 +2627,41 @@ static void producer_avformat_close( producer_avformat self ) mlt_log_debug( NULL, "producer_avformat_close\n" ); // Cleanup av contexts - av_free( self->av_frame ); - avformat_lock(); + av_free_packet( &self->pkt ); + av_free( self->video_frame ); + av_free( self->audio_frame ); + if ( self->is_mutex_init ) + pthread_mutex_lock( &self->open_mutex ); int i; for ( i = 0; i < MAX_AUDIO_STREAMS; i++ ) { - if ( self->audio_resample[i] ) - audio_resample_close( self->audio_resample[i] ); mlt_pool_release( self->audio_buffer[i] ); av_free( self->decode_buffer[i] ); if ( self->audio_codec[i] ) avcodec_close( self->audio_codec[i] ); + self->audio_codec[i] = NULL; } if ( self->video_codec ) avcodec_close( self->video_codec ); + self->video_codec = NULL; // Close the file +#if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0) + if ( self->dummy_context ) + avformat_close_input( &self->dummy_context ); + if ( self->seekable && self->audio_format ) + avformat_close_input( &self->audio_format ); + if ( self->video_format ) + avformat_close_input( &self->video_format ); +#else if ( self->dummy_context ) av_close_input_file( self->dummy_context ); if ( self->seekable && self->audio_format ) av_close_input_file( self->audio_format ); if ( self->video_format ) av_close_input_file( self->video_format ); - avformat_unlock(); +#endif + if ( self->is_mutex_init ) + pthread_mutex_unlock( &self->open_mutex ); #ifdef VDPAU vdpau_producer_close( self ); #endif @@ -2570,21 +2669,35 @@ static void producer_avformat_close( producer_avformat self ) mlt_cache_close( self->image_cache ); // Cleanup the mutexes - pthread_mutex_destroy( &self->audio_mutex ); - pthread_mutex_destroy( &self->video_mutex ); - pthread_mutex_destroy( &self->packets_mutex ); + if ( self->is_mutex_init ) + { + pthread_mutex_destroy( &self->audio_mutex ); + pthread_mutex_destroy( &self->video_mutex ); + pthread_mutex_destroy( &self->packets_mutex ); + pthread_mutex_destroy( &self->open_mutex ); + } // Cleanup the packet queues AVPacket *pkt; - while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) ) + if ( self->apackets ) { - av_free_packet( pkt ); - free( pkt ); + while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) ) + { + av_free_packet( pkt ); + free( pkt ); + } + mlt_deque_close( self->apackets ); + self->apackets = NULL; } - while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) ) + if ( self->vpackets ) { - av_free_packet( pkt ); - free( pkt ); + while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) ) + { + av_free_packet( pkt ); + free( pkt ); + } + mlt_deque_close( self->vpackets ); + self->vpackets = NULL; } free( self );