2 * producer_avformat.c -- avformat producer
3 * Copyright (C) 2003-2009 Ushodaya Enterprises Limited
4 * Author: Charles Yates <charles.yates@pandora.be>
5 * Author: Dan Dennedy <dan@dennedy.org>
6 * Much code borrowed from ffmpeg.c: Copyright (c) 2000-2003 Fabrice Bellard
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 #include <framework/mlt_producer.h>
25 #include <framework/mlt_frame.h>
26 #include <framework/mlt_profile.h>
27 #include <framework/mlt_log.h>
28 #include <framework/mlt_deque.h>
29 #include <framework/mlt_factory.h>
30 #include <framework/mlt_cache.h>
32 // ffmpeg Header files
33 #include <libavformat/avformat.h>
35 # include <libswscale/swscale.h>
37 #if LIBAVCODEC_VERSION_MAJOR >= 53
38 #include <libavutil/samplefmt.h>
39 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
40 const char *avcodec_get_sample_fmt_name(int sample_fmt);
43 # include <libavcodec/vdpau.h>
45 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
46 # include <libavutil/pixdesc.h>
49 // System header files
55 #if LIBAVUTIL_VERSION_INT < (50<<16)
56 #define PIX_FMT_RGB32 PIX_FMT_RGBA32
57 #define PIX_FMT_YUYV422 PIX_FMT_YUV422
60 #if LIBAVCODEC_VERSION_MAJOR >= 53
61 #include <libavutil/opt.h>
62 #define CODEC_TYPE_VIDEO AVMEDIA_TYPE_VIDEO
63 #define CODEC_TYPE_AUDIO AVMEDIA_TYPE_AUDIO
64 #define PKT_FLAG_KEY AV_PKT_FLAG_KEY
66 #include <libavcodec/opt.h>
69 #define POSITION_INITIAL (-2)
70 #define POSITION_INVALID (-1)
72 #define MAX_AUDIO_STREAMS (10)
73 #define MAX_VDPAU_SURFACES (10)
75 void avformat_lock( );
76 void avformat_unlock( );
78 struct producer_avformat_s
81 AVFormatContext *dummy_context;
82 AVFormatContext *audio_format;
83 AVFormatContext *video_format;
84 AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
85 AVCodecContext *video_codec;
87 ReSampleContext *audio_resample[ MAX_AUDIO_STREAMS ];
88 mlt_position audio_expected;
89 mlt_position video_expected;
93 int64_t last_position;
95 int64_t current_position;
96 mlt_position nonseek_position;
99 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
100 size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
101 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
102 int audio_used[ MAX_AUDIO_STREAMS ];
104 int audio_max_stream;
108 unsigned int invalid_pts_counter;
109 double resample_factor;
110 mlt_cache image_cache;
112 pthread_mutex_t video_mutex;
113 pthread_mutex_t audio_mutex;
116 pthread_mutex_t packets_mutex;
121 struct vdpau_render_state render_states[MAX_VDPAU_SURFACES];
135 typedef struct producer_avformat_s *producer_avformat;
137 // Forward references.
138 static int list_components( char* file );
139 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock );
140 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
141 static void producer_avformat_close( producer_avformat );
142 static void producer_close( mlt_producer parent );
143 static void producer_set_up_video( producer_avformat self, mlt_frame frame );
144 static void producer_set_up_audio( producer_avformat self, mlt_frame frame );
145 static void apply_properties( void *obj, mlt_properties properties, int flags );
146 static int video_codec_init( producer_avformat self, int index, mlt_properties properties );
147 static void get_audio_streams_info( producer_avformat self );
153 /** Constructor for libavformat.
156 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
158 if ( list_components( file ) )
161 mlt_producer producer = NULL;
163 // Check that we have a non-NULL argument
166 // Construct the producer
167 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
168 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
171 if ( mlt_producer_init( producer, self ) == 0 )
173 self->parent = producer;
175 // Get the properties
176 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
178 // Set the resource property (required for all producers)
179 mlt_properties_set( properties, "resource", file );
181 // Register transport implementation with the producer
182 producer->close = (mlt_destructor) producer_close;
184 // Register our get_frame implementation
185 producer->get_frame = producer_get_frame;
187 if ( strcmp( service, "avformat-novalidate" ) )
190 if ( producer_open( self, profile, file, 1 ) != 0 )
193 mlt_producer_close( producer );
196 else if ( self->seekable )
198 // Close the file to release resources for large playlists - reopen later as needed
200 if ( self->audio_format )
201 av_close_input_file( self->audio_format );
202 self->audio_format = NULL;
203 if ( self->video_format )
204 av_close_input_file( self->video_format );
205 self->video_format = NULL;
211 // Default the user-selectable indices from the auto-detected indices
212 mlt_properties_set_int( properties, "audio_index", self->audio_index );
213 mlt_properties_set_int( properties, "video_index", self->video_index );
215 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
217 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
224 int list_components( char* file )
228 // Report information about available demuxers and codecs as YAML Tiny
229 if ( file && strstr( file, "f-list" ) )
231 fprintf( stderr, "---\nformats:\n" );
232 AVInputFormat *format = NULL;
233 while ( ( format = av_iformat_next( format ) ) )
234 fprintf( stderr, " - %s\n", format->name );
235 fprintf( stderr, "...\n" );
238 if ( file && strstr( file, "acodec-list" ) )
240 fprintf( stderr, "---\naudio_codecs:\n" );
241 AVCodec *codec = NULL;
242 while ( ( codec = av_codec_next( codec ) ) )
243 if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
244 fprintf( stderr, " - %s\n", codec->name );
245 fprintf( stderr, "...\n" );
248 if ( file && strstr( file, "vcodec-list" ) )
250 fprintf( stderr, "---\nvideo_codecs:\n" );
251 AVCodec *codec = NULL;
252 while ( ( codec = av_codec_next( codec ) ) )
253 if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
254 fprintf( stderr, " - %s\n", codec->name );
255 fprintf( stderr, "...\n" );
262 /** Find the default streams.
265 static mlt_properties find_default_streams( producer_avformat self )
269 AVMetadataTag *tag = NULL;
270 AVFormatContext *context = self->video_format;
271 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
273 // Default to the first audio and video streams found
274 self->audio_index = -1;
275 self->video_index = -1;
277 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
279 // Allow for multiple audio and video streams in the file and select first of each (if available)
280 for( i = 0; i < context->nb_streams; i++ )
282 // Get the codec context
283 AVStream *stream = context->streams[ i ];
284 if ( ! stream ) continue;
285 AVCodecContext *codec_context = stream->codec;
286 if ( ! codec_context ) continue;
287 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
288 if ( ! codec ) continue;
290 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
292 // Determine the type and obtain the first index of each type
293 switch( codec_context->codec_type )
295 case CODEC_TYPE_VIDEO:
296 // Use first video stream
297 if ( self->video_index < 0 )
298 self->video_index = i;
299 mlt_properties_set( meta_media, key, "video" );
300 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
301 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
302 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
303 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
304 ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
305 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
307 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) );
310 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
311 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
312 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
314 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
315 mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den /
316 ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) );
317 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
318 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
319 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
320 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
321 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
322 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
323 switch ( codec_context->colorspace )
325 case AVCOL_SPC_SMPTE240M:
326 mlt_properties_set_int( meta_media, key, 240 );
328 case AVCOL_SPC_BT470BG:
329 case AVCOL_SPC_SMPTE170M:
330 mlt_properties_set_int( meta_media, key, 601 );
332 case AVCOL_SPC_BT709:
333 mlt_properties_set_int( meta_media, key, 709 );
336 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
337 mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
342 case CODEC_TYPE_AUDIO:
343 // Use first audio stream
344 if ( self->audio_index < 0 )
345 self->audio_index = i;
346 mlt_properties_set( meta_media, key, "audio" );
347 #if LIBAVCODEC_VERSION_MAJOR >= 53
348 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
349 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
350 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
351 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
352 mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) );
354 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
355 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
356 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
357 mlt_properties_set_int( meta_media, key, codec_context->channels );
362 // snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
363 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
364 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
365 mlt_properties_set( meta_media, key, codec->name );
366 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0))
367 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
368 mlt_properties_set( meta_media, key, codec->long_name );
370 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
371 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
372 // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
373 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
374 // snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
375 // mlt_properties_set_int( meta_media, key, codec_context->profile );
376 // snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
377 // mlt_properties_set_int( meta_media, key, codec_context->level );
380 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
381 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
383 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
385 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
386 mlt_properties_set( meta_media, key, tag->value );
391 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
392 while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
394 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
396 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
397 mlt_properties_set( meta_media, key, tag->value );
401 if ( context->title && strcmp( context->title, "" ) )
402 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
403 if ( context->author && strcmp( context->author, "" ) )
404 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
405 if ( context->copyright && strcmp( context->copyright, "" ) )
406 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
407 if ( context->comment )
408 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
409 if ( context->album )
410 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
412 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
413 if ( context->track )
414 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
420 static inline int dv_is_pal( AVPacket *pkt )
422 return pkt->data[3] & 0x80;
425 static int dv_is_wide( AVPacket *pkt )
427 int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
429 for ( ; i < pkt->size; i += 5 /* packet size */ )
431 if ( pkt->data[ i ] == 0x61 )
433 uint8_t x = pkt->data[ i + 2 ] & 0x7;
434 return ( x == 2 ) || ( x == 7 );
440 static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
442 double aspect_ratio = 1.0;
444 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
448 if ( dv_is_pal( pkt ) )
450 if ( dv_is_wide( pkt ) )
452 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
453 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
457 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
458 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
463 if ( dv_is_wide( pkt ) )
465 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
466 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
470 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
471 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
478 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
479 stream->sample_aspect_ratio;
481 codec_context->sample_aspect_ratio;
483 // Override FFmpeg's notion of DV aspect ratios, which are
484 // based upon a width of 704. Since we do not have a normaliser
485 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
486 // we just coerce the values to facilitate a passive behaviour through
487 // the rescale normaliser when using equivalent producers and consumers.
488 // = display_aspect / (width * height)
489 if ( ar.num == 10 && ar.den == 11 )
492 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
493 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
495 else if ( ar.num == 59 && ar.den == 54 )
498 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
499 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
501 else if ( ar.num == 40 && ar.den == 33 )
504 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
505 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
507 else if ( ar.num == 118 && ar.den == 81 )
510 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
511 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
517 AVRational codec_sar = codec_context->sample_aspect_ratio;
518 AVRational stream_sar =
519 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
520 stream->sample_aspect_ratio;
524 if ( codec_sar.num > 0 )
526 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num );
527 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den );
529 else if ( stream_sar.num > 0 )
531 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num );
532 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den );
536 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 );
537 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 );
540 AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) };
541 aspect_ratio = av_q2d( ar );
542 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
547 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params )
549 if ( !URL ) return NULL;
551 const char *result = URL;
552 char *protocol = strdup( URL );
553 char *url = strchr( protocol, ':' );
555 // Only if there is not a protocol specification that avformat can handle
556 #if LIBAVFORMAT_VERSION_MAJOR >= 53
557 if ( url && avio_check( URL, 0 ) < 0 )
559 if ( url && !url_exist( URL ) )
562 // Truncate protocol string
564 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url + 1 );
567 *format = av_find_input_format( protocol );
569 // Eat the format designator
574 // These are required by video4linux2 (defaults)
575 params->width = profile->width;
576 params->height = profile->height;
577 if ( !strstr( URL, "&frame_rate" ) )
578 params->time_base = (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
579 params->channels = 2;
580 params->sample_rate = 48000;
583 url = strchr( url, '?' );
587 char *name = strdup( ++url );
588 char *value = strchr( name, '=' );
590 // Also accept : as delimiter for backwards compatibility.
591 value = strchr( name, ':' );
596 char *t = strchr( value, '&' );
599 if ( !strcmp( name, "frame_rate" ) )
600 params->time_base.den = atoi( value );
601 else if ( !strcmp( name, "frame_rate_base" ) )
602 params->time_base.num = atoi( value );
603 else if ( !strcmp( name, "sample_rate" ) )
604 params->sample_rate = atoi( value );
605 else if ( !strcmp( name, "channel" ) )
606 params->channel = atoi( value );
607 else if ( !strcmp( name, "channels" ) )
608 params->channels = atoi( value );
609 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
610 else if ( !strcmp( name, "pix_fmt" ) )
611 params->pix_fmt = av_get_pix_fmt( value );
613 else if ( !strcmp( name, "width" ) )
614 params->width = atoi( value );
615 else if ( !strcmp( name, "height" ) )
616 params->height = atoi( value );
617 else if ( !strcmp( name, "standard" ) )
618 params->standard = strdup( value );
621 url = strchr( url, '&' );
625 result = strdup( result );
630 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
634 // Get the properties
635 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
637 AVFormatContext *format = self->video_format;
639 // We will treat everything with the producer fps.
640 // TODO: make this more flexible.
641 double fps = mlt_profile_fps( profile );
644 if ( !mlt_properties_get_int( properties, "_length_computed" ) )
646 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
647 // when producer_open is called after initial call when restoring or reseting the producer.
648 if ( format->duration != AV_NOPTS_VALUE )
650 // This isn't going to be accurate for all formats
651 mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
652 mlt_properties_set_position( properties, "out", frames - 1 );
653 mlt_properties_set_position( properties, "length", frames );
654 mlt_properties_set_int( properties, "_length_computed", 1 );
658 // Set live sources to run forever
659 mlt_properties_set_position( properties, "length", INT_MAX );
660 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
661 mlt_properties_set( properties, "eof", "loop" );
662 mlt_properties_set_int( properties, "_length_computed", 1 );
666 // Check if we're seekable
667 // avdevices are typically AVFMT_NOFILE and not seekable
668 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
671 // protocols can indicate if they support seeking
672 #if LIBAVFORMAT_VERSION_MAJOR >= 53
673 self->seekable = format->pb->seekable;
675 URLContext *uc = url_fileno( format->pb );
677 self->seekable = !uc->is_streamed;
680 if ( self->seekable )
682 // Do a more rigourous test of seekable on a disposable context
683 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
684 mlt_properties_set_int( properties, "seekable", self->seekable );
685 self->dummy_context = format;
686 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
687 format = self->video_format;
688 av_find_stream_info( format );
691 // Fetch the width, height and aspect ratio
692 if ( self->video_index != -1 )
694 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
695 mlt_properties_set_int( properties, "width", codec_context->width );
696 mlt_properties_set_int( properties, "height", codec_context->height );
698 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
700 // Fetch the first frame of DV so we can read it directly
705 ret = av_read_frame( format, &pkt );
706 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
708 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, &pkt );
715 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, NULL );
719 // Verify that we can convert this to YUV 4:2:2
720 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
721 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
722 codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
724 sws_freeContext( context );
735 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock )
737 // Return an error code (0 == no error)
739 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
744 pthread_mutex_init( &self->audio_mutex, NULL );
745 pthread_mutex_init( &self->video_mutex, NULL );
746 pthread_mutex_init( &self->packets_mutex, NULL );
747 pthread_mutex_lock( &self->audio_mutex );
748 pthread_mutex_lock( &self->video_mutex );
750 mlt_events_block( properties, self->parent );
753 AVInputFormat *format = NULL;
754 AVFormatParameters params;
755 memset( ¶ms, 0, sizeof(params) );
756 char *filename = parse_url( profile, URL, &format, ¶ms );
758 // Now attempt to open the file or device with filename
759 error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0;
761 // If the URL is a network stream URL, then we probably need to open with full URL
762 error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0;
764 // Set MLT properties onto video AVFormatContext
765 if ( !error && self->video_format )
767 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
768 #if LIBAVFORMAT_VERSION_MAJOR >= 53
769 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
770 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
774 // Cleanup AVFormatParameters
775 if ( params.standard )
776 free( (void*) params.standard );
778 // If successful, then try to get additional info
779 if ( !error && self->video_format )
781 // Get the stream info
782 error = av_find_stream_info( self->video_format ) < 0;
784 // Continue if no error
785 if ( !error && self->video_format )
787 // Find default audio and video streams
788 find_default_streams( self );
789 error = get_basic_info( self, profile, filename );
791 // Initialize position info
792 self->first_pts = -1;
793 self->last_position = POSITION_INITIAL;
795 if ( !self->audio_format )
797 // We're going to cheat here - for seekable A/V files, we will have separate contexts
798 // to support independent seeking of audio from video.
799 // TODO: Is this really necessary?
800 if ( self->audio_index != -1 && self->video_index != -1 )
802 if ( self->seekable )
804 // And open again for our audio context
805 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
806 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
807 #if LIBAVFORMAT_VERSION_MAJOR >= 53
808 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
809 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
811 av_find_stream_info( self->audio_format );
815 self->audio_format = self->video_format;
818 else if ( self->audio_index != -1 )
820 // We only have an audio context
821 self->audio_format = self->video_format;
822 self->video_format = NULL;
824 else if ( self->video_index == -1 )
826 // Something has gone wrong
829 if ( self->audio_format && !self->audio_streams )
830 get_audio_streams_info( self );
838 self->apackets = mlt_deque_init();
839 self->vpackets = mlt_deque_init();
842 if ( self->dummy_context )
845 av_close_input_file( self->dummy_context );
847 self->dummy_context = NULL;
850 // Unlock the service
853 pthread_mutex_unlock( &self->audio_mutex );
854 pthread_mutex_unlock( &self->video_mutex );
856 mlt_events_unblock( properties, self->parent );
861 static void reopen_video( producer_avformat self, mlt_producer producer )
863 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
864 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
865 pthread_mutex_lock( &self->audio_mutex );
868 if ( self->video_codec )
869 avcodec_close( self->video_codec );
870 self->video_codec = NULL;
871 if ( self->dummy_context )
872 av_close_input_file( self->dummy_context );
873 self->dummy_context = NULL;
874 if ( self->video_format )
875 av_close_input_file( self->video_format );
876 self->video_format = NULL;
879 int audio_index = self->audio_index;
880 int video_index = self->video_index;
882 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
883 mlt_properties_get( properties, "resource" ), 0 );
885 self->audio_index = audio_index;
886 if ( self->video_format && video_index > -1 )
888 self->video_index = video_index;
889 video_codec_init( self, video_index, properties );
892 pthread_mutex_unlock( &self->audio_mutex );
893 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
896 static int seek_video( producer_avformat self, mlt_position position,
897 int64_t req_position, int must_decode, int use_new_seek, int *ignore )
899 mlt_producer producer = self->parent;
902 if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
904 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
906 // Fetch the video format context
907 AVFormatContext *context = self->video_format;
909 // Get the video stream
910 AVStream *stream = context->streams[ self->video_index ];
913 AVCodecContext *codec_context = stream->codec;
915 // We may want to use the source fps if available
916 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
917 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
919 if ( self->av_frame && position + 1 == self->video_expected )
921 // We're paused - use last image
924 else if ( !self->seekable && position > self->video_expected && ( position - self->video_expected ) < 250 )
926 // Fast forward - seeking is inefficient for small distances - just ignore following frames
927 *ignore = ( int )( ( position - self->video_expected ) / mlt_producer_get_fps( producer ) * source_fps );
928 codec_context->skip_loop_filter = AVDISCARD_NONREF;
930 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
932 if ( use_new_seek && self->last_position == POSITION_INITIAL )
934 // find first key frame
939 while ( ret >= 0 && toscan-- > 0 )
941 ret = av_read_frame( context, &pkt );
942 if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == self->video_index )
944 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n", pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
945 self->first_pts = pkt.pts;
948 av_free_packet( &pkt );
951 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
954 // Calculate the timestamp for the requested frame
958 timestamp = ( req_position - 0.1 / source_fps ) /
959 ( av_q2d( stream->time_base ) * source_fps );
960 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pos %"PRId64" pts %"PRId64"\n", req_position, timestamp );
961 if ( self->first_pts > 0 )
962 timestamp += self->first_pts;
963 else if ( context->start_time != AV_NOPTS_VALUE )
964 timestamp += context->start_time;
968 timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
969 if ( context->start_time != AV_NOPTS_VALUE )
970 timestamp += context->start_time;
973 timestamp -= AV_TIME_BASE;
976 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %"PRId64"\n",
977 timestamp, position, self->video_expected, self->last_position );
979 // Seek to the timestamp
982 codec_context->skip_loop_filter = AVDISCARD_NONREF;
983 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
985 else if ( req_position > 0 || self->last_position <= 0 )
987 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
991 // Re-open video stream when rewinding to beginning from somewhere else.
992 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
993 // not reliably seeking to the first frame across formats.
994 reopen_video( self, producer );
997 // Remove the cached info relating to the previous position
998 self->current_position = POSITION_INVALID;
999 self->last_position = POSITION_INVALID;
1000 av_freep( &self->av_frame );
1004 // flush any pictures still in decode buffer
1005 avcodec_flush_buffers( codec_context );
1012 /** Convert a frame position to a time code.
1015 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1017 return ( double )position / mlt_producer_get_fps( producer );
1020 // Collect information about all audio streams
1022 static void get_audio_streams_info( producer_avformat self )
1024 // Fetch the audio format context
1025 AVFormatContext *context = self->audio_format;
1029 i < context->nb_streams;
1032 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
1034 AVCodecContext *codec_context = context->streams[i]->codec;
1035 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1037 // If we don't have a codec and we can't initialise it, we can't do much more...
1039 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1041 self->audio_streams++;
1042 self->audio_max_stream = i;
1043 self->total_channels += codec_context->channels;
1044 if ( codec_context->channels > self->max_channel )
1045 self->max_channel = codec_context->channels;
1046 if ( codec_context->sample_rate > self->max_frequency )
1047 self->max_frequency = codec_context->sample_rate;
1048 avcodec_close( codec_context );
1053 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1054 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1056 // Other audio-specific initializations
1057 self->resample_factor = 1.0;
1060 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
1062 #if defined(SWSCALE) && (LIBSWSCALE_VERSION_INT >= ((0<<16)+(7<<8)+2))
1064 const int *new_coefficients;
1066 int brightness, contrast, saturation;
1068 if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
1069 &brightness, &contrast, &saturation ) != -1 )
1071 // Don't change these from defaults unless explicitly told to.
1072 if ( use_full_range >= 0 )
1073 full_range = use_full_range;
1074 switch ( colorspace )
1080 new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1083 new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1086 new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1089 new_coefficients = coefficients;
1092 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
1093 brightness, contrast, saturation );
1098 static mlt_image_format pick_format( enum PixelFormat pix_fmt )
1106 return mlt_image_rgb24a;
1107 case PIX_FMT_YUV420P:
1108 case PIX_FMT_YUVJ420P:
1109 case PIX_FMT_YUVA420P:
1110 return mlt_image_yuv420p;
1114 case PIX_FMT_MONOWHITE:
1115 case PIX_FMT_MONOBLACK:
1118 return mlt_image_rgb24;
1120 return mlt_image_yuv422;
1124 static void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt,
1125 mlt_image_format *format, int width, int height, int colorspace )
1128 int full_range = -1;
1129 int flags = SWS_BILINEAR | SWS_ACCURATE_RND;
1132 flags |= SWS_CPU_CAPS_MMX;
1135 flags |= SWS_CPU_CAPS_MMX2;
1138 if ( *format == mlt_image_yuv420p )
1140 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1141 width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1143 output.data[0] = buffer;
1144 output.data[1] = buffer + width * height;
1145 output.data[2] = buffer + ( 5 * width * height ) / 4;
1146 output.linesize[0] = width;
1147 output.linesize[1] = width >> 1;
1148 output.linesize[2] = width >> 1;
1149 set_luma_transfer( context, colorspace, full_range );
1150 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1151 output.data, output.linesize);
1152 sws_freeContext( context );
1154 else if ( *format == mlt_image_rgb24 )
1156 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1157 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1159 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1160 set_luma_transfer( context, colorspace, full_range );
1161 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1162 output.data, output.linesize);
1163 sws_freeContext( context );
1165 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1167 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1168 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1170 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1171 set_luma_transfer( context, colorspace, full_range );
1172 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1173 output.data, output.linesize);
1174 sws_freeContext( context );
1178 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1179 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1181 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1182 set_luma_transfer( context, colorspace, full_range );
1183 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1184 output.data, output.linesize);
1185 sws_freeContext( context );
1188 if ( *format == mlt_image_yuv420p )
1191 pict.data[0] = buffer;
1192 pict.data[1] = buffer + width * height;
1193 pict.data[2] = buffer + ( 5 * width * height ) / 4;
1194 pict.linesize[0] = width;
1195 pict.linesize[1] = width >> 1;
1196 pict.linesize[2] = width >> 1;
1197 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
1199 else if ( *format == mlt_image_rgb24 )
1202 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1203 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
1205 else if ( format == mlt_image_rgb24a || format == mlt_image_opengl )
1208 avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height );
1209 img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height );
1214 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1215 img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height );
1220 /** Allocate the image buffer and set it on the frame.
1223 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1227 if ( codec_context->width == 0 || codec_context->height == 0 )
1229 *width = codec_context->width;
1230 *height = codec_context->height;
1231 size = mlt_image_format_size( *format, *width, *height, NULL );
1232 *buffer = mlt_pool_alloc( size );
1234 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1241 /** Get an image from a frame.
1244 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1247 producer_avformat self = mlt_frame_pop_service( frame );
1248 mlt_producer producer = self->parent;
1250 // Get the properties from the frame
1251 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1253 // Obtain the frame number of this frame
1254 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
1256 // Get the producer properties
1257 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1259 pthread_mutex_lock( &self->video_mutex );
1261 // Fetch the video format context
1262 AVFormatContext *context = self->video_format;
1264 // Get the video stream
1265 AVStream *stream = context->streams[ self->video_index ];
1267 // Get codec context
1268 AVCodecContext *codec_context = stream->codec;
1270 // Get the image cache
1271 if ( ! self->image_cache && ! mlt_properties_get_int( properties, "noimagecache" ) )
1272 self->image_cache = mlt_cache_init();
1273 if ( self->image_cache )
1275 mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position );
1276 uint8_t *original = mlt_cache_item_data( item, (int*) format );
1279 // Set the resolution
1280 *width = codec_context->width;
1281 *height = codec_context->height;
1283 // Workaround 1088 encodings missing cropping info.
1284 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1288 int size = mlt_image_format_size( *format, *width, *height, NULL );
1291 *buffer = mlt_pool_alloc( size );
1292 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1293 memcpy( *buffer, original, size );
1294 mlt_cache_item_close( item );
1299 mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
1300 mlt_frame_set_image( frame, *buffer, size, NULL );
1302 self->got_picture = 1;
1304 goto exit_get_image;
1313 // Special case ffwd handling
1316 // We may want to use the source fps if available
1317 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1318 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1320 // This is the physical frame position in the source
1321 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1323 // Determines if we have to decode all frames in a sequence
1324 // Temporary hack to improve intra frame only
1325 int must_decode = !( codec_context->codec && codec_context->codec->name ) || (
1326 strcmp( codec_context->codec->name, "dnxhd" ) &&
1327 strcmp( codec_context->codec->name, "dvvideo" ) &&
1328 strcmp( codec_context->codec->name, "huffyuv" ) &&
1329 strcmp( codec_context->codec->name, "mjpeg" ) &&
1330 strcmp( codec_context->codec->name, "rawvideo" ) );
1332 // Turn on usage of new seek API and PTS for seeking
1333 int use_new_seek = self->seekable &&
1334 codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" );
1335 if ( mlt_properties_get( properties, "new_seek" ) )
1336 use_new_seek = mlt_properties_get_int( properties, "new_seek" );
1337 double delay = mlt_properties_get_double( properties, "video_delay" );
1339 // Seek if necessary
1340 int paused = seek_video( self, position, req_position, must_decode, use_new_seek, &ignore );
1342 // Seek might have reopened the file
1343 context = self->video_format;
1344 stream = context->streams[ self->video_index ];
1345 codec_context = stream->codec;
1347 if ( *format == mlt_image_none ||
1348 codec_context->pix_fmt == PIX_FMT_ARGB ||
1349 codec_context->pix_fmt == PIX_FMT_RGBA ||
1350 codec_context->pix_fmt == PIX_FMT_ABGR ||
1351 codec_context->pix_fmt == PIX_FMT_BGRA )
1352 *format = pick_format( codec_context->pix_fmt );
1354 // Duplicate the last image if necessary
1355 if ( self->av_frame && self->av_frame->linesize[0] && self->got_picture
1357 || self->current_position == req_position
1358 || ( !use_new_seek && self->current_position > req_position ) ) )
1361 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1363 // Workaround 1088 encodings missing cropping info.
1364 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1367 if ( self->vdpau && self->vdpau->buffer )
1370 picture.data[0] = self->vdpau->buffer;
1371 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1372 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1373 picture.linesize[0] = codec_context->width;
1374 picture.linesize[1] = codec_context->width / 2;
1375 picture.linesize[2] = codec_context->width / 2;
1376 convert_image( (AVFrame*) &picture, *buffer,
1377 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1381 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1382 format, *width, *height, self->colorspace );
1385 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1390 int64_t int_position = 0;
1391 int decode_errors = 0;
1392 int got_picture = 0;
1394 av_init_packet( &pkt );
1396 // Construct an AVFrame for YUV422 conversion
1397 if ( !self->av_frame )
1398 self->av_frame = avcodec_alloc_frame( );
1400 while( ret >= 0 && !got_picture )
1403 pthread_mutex_lock( &self->packets_mutex );
1404 if ( mlt_deque_count( self->vpackets ) )
1406 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1412 ret = av_read_frame( context, &pkt );
1413 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->audio_index )
1415 if ( !av_dup_packet( &pkt ) )
1417 AVPacket *tmp = malloc( sizeof(AVPacket) );
1419 mlt_deque_push_back( self->apackets, tmp );
1423 pthread_mutex_unlock( &self->packets_mutex );
1425 // We only deal with video from the selected video_index
1426 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
1428 // Determine time code of the packet
1431 int64_t pts = pkt.pts;
1432 if ( self->first_pts > 0 )
1433 pts -= self->first_pts;
1434 else if ( context->start_time != AV_NOPTS_VALUE )
1435 pts -= context->start_time;
1436 int_position = ( int64_t )( ( av_q2d( stream->time_base ) * pts + delay ) * source_fps + 0.1 );
1437 if ( pkt.pts == AV_NOPTS_VALUE )
1439 self->invalid_pts_counter++;
1440 if ( self->invalid_pts_counter > 20 )
1442 mlt_log_panic( MLT_PRODUCER_SERVICE(producer), "\ainvalid PTS; DISABLING NEW_SEEK!\n" );
1443 mlt_properties_set_int( properties, "new_seek", 0 );
1444 int_position = req_position;
1450 self->invalid_pts_counter = 0;
1452 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.pts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1453 pkt.pts, req_position, self->current_position, int_position );
1457 if ( pkt.dts != AV_NOPTS_VALUE )
1459 int_position = ( int64_t )( ( av_q2d( stream->time_base ) * pkt.dts + delay ) * source_fps + 0.5 );
1460 if ( context->start_time != AV_NOPTS_VALUE )
1461 int_position -= ( int64_t )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
1462 if ( int_position == self->last_position )
1463 int_position = self->last_position + 1;
1465 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1466 pkt.dts, req_position, self->current_position, int_position );
1467 // Make a dumb assumption on streams that contain wild timestamps
1468 if ( abs( req_position - int_position ) > 999 )
1470 int_position = req_position;
1471 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" );
1474 self->last_position = int_position;
1477 if ( must_decode || int_position >= req_position )
1482 if ( self->vdpau->decoder == VDP_INVALID_HANDLE )
1484 vdpau_decoder_init( self );
1486 self->vdpau->is_decoded = 0;
1489 codec_context->reordered_opaque = pkt.pts;
1490 if ( int_position >= req_position )
1491 codec_context->skip_loop_filter = AVDISCARD_NONE;
1492 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1493 ret = avcodec_decode_video2( codec_context, self->av_frame, &got_picture, &pkt );
1495 ret = avcodec_decode_video( codec_context, self->av_frame, &got_picture, pkt.data, pkt.size );
1497 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1500 if ( ++decode_errors <= 10 )
1513 // Determine time code of the packet
1514 int64_t pts = self->av_frame->reordered_opaque;
1515 if ( self->first_pts > 0 )
1516 pts -= self->first_pts;
1517 else if ( context->start_time != AV_NOPTS_VALUE )
1518 pts -= context->start_time;
1519 int_position = ( int64_t )( av_q2d( stream->time_base) * pts * source_fps + 0.1 );
1520 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "got frame %"PRId64", key %d\n", int_position, self->av_frame->key_frame );
1523 if ( int_position < req_position )
1528 else if ( int_position >= req_position )
1531 codec_context->skip_loop_filter = AVDISCARD_NONE;
1533 else if ( ignore -- )
1538 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY );
1541 // Now handle the picture if we have one
1544 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1546 // Workaround 1088 encodings missing cropping info.
1547 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1552 if ( self->vdpau->is_decoded )
1554 struct vdpau_render_state *render = (struct vdpau_render_state*) self->av_frame->data[0];
1556 uint32_t pitches[3];
1557 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1559 if ( !self->vdpau->buffer )
1560 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1561 self->av_frame->data[0] = planes[0] = self->vdpau->buffer;
1562 self->av_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1563 self->av_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1564 self->av_frame->linesize[0] = pitches[0] = codec_context->width;
1565 self->av_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1566 self->av_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1568 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1569 if ( status == VDP_STATUS_OK )
1571 convert_image( self->av_frame, *buffer, PIX_FMT_YUV420P,
1572 format, *width, *height, self->colorspace );
1576 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1577 image_size = self->vdpau->is_decoded = 0;
1582 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1583 image_size = got_picture = 0;
1588 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1589 format, *width, *height, self->colorspace );
1590 self->top_field_first |= self->av_frame->top_field_first;
1591 self->current_position = int_position;
1592 self->got_picture = 1;
1599 if ( self->seekable || pkt.stream_index != self->audio_index )
1600 av_free_packet( &pkt );
1604 if ( self->got_picture && image_size > 0 && self->image_cache )
1606 // Copy buffer to image cache
1607 uint8_t *image = mlt_pool_alloc( image_size );
1608 memcpy( image, *buffer, image_size );
1609 mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release );
1611 // Try to duplicate last image if there was a decoding failure
1612 else if ( !image_size && self->av_frame && self->av_frame->linesize[0] )
1615 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1617 // Workaround 1088 encodings missing cropping info.
1618 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1621 if ( self->vdpau && self->vdpau->buffer )
1624 picture.data[0] = self->vdpau->buffer;
1625 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1626 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1627 picture.linesize[0] = codec_context->width;
1628 picture.linesize[1] = codec_context->width / 2;
1629 picture.linesize[2] = codec_context->width / 2;
1630 convert_image( (AVFrame*) &picture, *buffer,
1631 PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1635 convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1636 format, *width, *height, self->colorspace );
1637 self->got_picture = 1;
1640 mlt_frame_get_image( frame, buffer, format, width, height, writable );
1643 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1644 self->video_expected = position + 1;
1648 pthread_mutex_unlock( &self->video_mutex );
1650 // Set the progressive flag
1651 if ( mlt_properties_get( properties, "force_progressive" ) )
1652 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1653 else if ( self->av_frame )
1654 mlt_properties_set_int( frame_properties, "progressive", !self->av_frame->interlaced_frame );
1656 // Set the field order property for this frame
1657 if ( mlt_properties_get( properties, "force_tff" ) )
1658 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1660 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1662 // Set immutable properties of the selected track's (or overridden) source attributes.
1663 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1664 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1665 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1666 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1668 return !self->got_picture;
1671 /** Process properties as AVOptions and apply to AV context obj
1674 static void apply_properties( void *obj, mlt_properties properties, int flags )
1677 int count = mlt_properties_count( properties );
1678 for ( i = 0; i < count; i++ )
1680 const char *opt_name = mlt_properties_get_name( properties, i );
1681 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1682 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1685 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1686 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1687 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1688 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1690 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1696 /** Initialize the video codec context.
1699 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1701 // Initialise the codec if necessary
1702 if ( !self->video_codec )
1704 // Get the video stream
1705 AVStream *stream = self->video_format->streams[ index ];
1707 // Get codec context
1708 AVCodecContext *codec_context = stream->codec;
1711 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1713 if ( codec_context->codec_id == CODEC_ID_H264 )
1715 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1717 if ( vdpau_init( self ) )
1719 self->video_codec = codec_context;
1720 if ( !vdpau_decoder_init( self ) )
1725 codec = avcodec_find_decoder( codec_context->codec_id );
1729 // Initialise multi-threading
1730 int thread_count = mlt_properties_get_int( properties, "threads" );
1731 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1732 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1733 if ( thread_count > 1 )
1734 codec_context->thread_count = thread_count;
1736 // If we don't have a codec and we can't initialise it, we can't do much more...
1738 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1740 // Now store the codec with its destructor
1741 self->video_codec = codec_context;
1745 // Remember that we can't use this later
1746 self->video_index = -1;
1752 // Process properties as AVOptions
1753 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1754 #if LIBAVCODEC_VERSION_MAJOR >= 53
1755 if ( codec->priv_class && codec_context->priv_data )
1756 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1759 // Reset some image properties
1760 mlt_properties_set_int( properties, "width", self->video_codec->width );
1761 mlt_properties_set_int( properties, "height", self->video_codec->height );
1762 // For DV, we'll just use the saved aspect ratio
1763 if ( codec_context->codec_id != CODEC_ID_DVVIDEO )
1764 get_aspect_ratio( properties, stream, self->video_codec, NULL );
1766 // Determine the fps first from the codec
1767 double source_fps = (double) self->video_codec->time_base.den /
1768 ( self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1770 if ( mlt_properties_get( properties, "force_fps" ) )
1772 source_fps = mlt_properties_get_double( properties, "force_fps" );
1773 stream->time_base = av_d2q( source_fps, 1024 );
1774 mlt_properties_set_int( properties, "meta.media.frame_rate_num", stream->time_base.num );
1775 mlt_properties_set_int( properties, "meta.media.frame_rate_den", stream->time_base.den );
1779 // If the muxer reports a frame rate different than the codec
1780 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1781 double muxer_fps = av_q2d( stream->avg_frame_rate );
1782 if ( isnan( muxer_fps ) || muxer_fps == 0 )
1783 muxer_fps = av_q2d( stream->r_frame_rate );
1785 double muxer_fps = av_q2d( stream->r_frame_rate );
1787 // Choose the lesser - the wrong tends to be off by some multiple of 10
1788 source_fps = FFMIN( source_fps, muxer_fps );
1789 if ( source_fps >= 1.0 && ( source_fps < muxer_fps || isnan( muxer_fps ) ) )
1791 mlt_properties_set_int( properties, "meta.media.frame_rate_num", self->video_codec->time_base.den );
1792 mlt_properties_set_int( properties, "meta.media.frame_rate_den", self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1794 else if ( muxer_fps > 0 )
1796 AVRational frame_rate = stream->r_frame_rate;
1797 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1798 // avg_frame_rate gives some approximate value that does not well match the media.
1799 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1800 // results in some very choppy output, but some value slightly different works
1802 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1803 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1804 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1806 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1807 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1808 source_fps = av_q2d( frame_rate );
1812 source_fps = mlt_producer_get_fps( self->parent );
1813 AVRational frame_rate = av_d2q( source_fps, 255 );
1814 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1815 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1819 // source_fps is deprecated in favor of meta.media.frame_rate_num and .frame_rate_den
1820 if ( source_fps > 0 )
1821 mlt_properties_set_double( properties, "source_fps", source_fps );
1823 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( self->parent ) );
1825 // Set the YUV colorspace from override or detect
1826 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1827 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1828 if ( ! self->colorspace )
1830 switch ( self->video_codec->colorspace )
1832 case AVCOL_SPC_SMPTE240M:
1833 self->colorspace = 240;
1835 case AVCOL_SPC_BT470BG:
1836 case AVCOL_SPC_SMPTE170M:
1837 self->colorspace = 601;
1839 case AVCOL_SPC_BT709:
1840 self->colorspace = 709;
1843 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1844 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1849 // Let apps get chosen colorspace
1850 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1852 return self->video_codec && self->video_index > -1;
1855 /** Set up video handling.
1858 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1861 mlt_producer producer = self->parent;
1863 // Get the properties
1864 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1866 // Fetch the video format context
1867 AVFormatContext *context = self->video_format;
1869 // Get the video_index
1870 int index = mlt_properties_get_int( properties, "video_index" );
1872 // Reopen the file if necessary
1873 if ( !context && index > -1 )
1875 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1876 mlt_properties_get( properties, "resource" ), 1 );
1877 context = self->video_format;
1880 // Exception handling for video_index
1881 if ( context && index >= (int) context->nb_streams )
1883 // Get the last video stream
1884 for ( index = context->nb_streams - 1;
1885 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1887 mlt_properties_set_int( properties, "video_index", index );
1889 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1891 // Invalidate the video stream
1893 mlt_properties_set_int( properties, "video_index", index );
1896 // Update the video properties if the index changed
1897 if ( index != self->video_index )
1899 // Reset the video properties if the index changed
1900 self->video_index = index;
1902 if ( self->video_codec )
1903 avcodec_close( self->video_codec );
1904 self->video_codec = NULL;
1908 // Get the frame properties
1909 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1912 if ( context && index > -1 && video_codec_init( self, index, properties ) )
1914 // Set the frame properties
1915 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1916 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1917 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1919 // Set the width and height
1920 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
1921 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
1922 // real_width and real_height are deprecated in favor of meta.media.width and .height
1923 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1924 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1925 mlt_properties_set_int( frame_properties, "real_width", self->video_codec->width );
1926 mlt_properties_set_int( frame_properties, "real_height", self->video_codec->height );
1927 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1928 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
1930 // Workaround 1088 encodings missing cropping info.
1931 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1933 mlt_properties_set_int( properties, "meta.media.height", 1080 );
1934 mlt_properties_set_int( frame_properties, "real_height", 1080 );
1937 // Add our image operation
1938 mlt_frame_push_service( frame, self );
1939 mlt_frame_push_get_image( frame, producer_get_image );
1943 // If something failed, use test card image
1944 mlt_properties_set_int( frame_properties, "test_image", 1 );
1948 static int seek_audio( producer_avformat self, mlt_position position, double timecode, int *ignore )
1952 // Seek if necessary
1953 if ( self->seekable && position != self->audio_expected )
1955 if ( position + 1 == self->audio_expected )
1957 // We're paused - silence required
1960 else if ( !self->seekable && position > self->audio_expected && ( position - self->audio_expected ) < 250 )
1962 // Fast forward - seeking is inefficient for small distances - just ignore following frames
1963 *ignore = position - self->audio_expected;
1965 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
1967 AVFormatContext *context = self->audio_format;
1968 int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
1969 if ( context->start_time != AV_NOPTS_VALUE )
1970 timestamp += context->start_time;
1971 if ( timestamp < 0 )
1974 // Set to the real timecode
1975 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
1978 // Clear the usage in the audio buffer
1979 int i = MAX_AUDIO_STREAMS + 1;
1981 self->audio_used[i - 1] = 0;
1987 static int sample_bytes( AVCodecContext *context )
1989 #if LIBAVCODEC_VERSION_MAJOR >= 53
1990 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
1992 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1996 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
1998 // Fetch the audio_format
1999 AVFormatContext *context = self->audio_format;
2001 // Get the current stream index
2002 int index = pkt.stream_index;
2004 // Get codec context
2005 AVCodecContext *codec_context = self->audio_codec[ index ];
2007 // Obtain the resample context if it exists (not always needed)
2008 ReSampleContext *resample = self->audio_resample[ index ];
2010 // Obtain the audio buffers
2011 uint8_t *audio_buffer = self->audio_buffer[ index ];
2012 uint8_t *decode_buffer = self->decode_buffer[ index ];
2014 int audio_used = self->audio_used[ index ];
2015 uint8_t *ptr = pkt.data;
2019 while ( ptr && ret >= 0 && len > 0 )
2021 int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context );
2022 int data_size = self->audio_buffer_size[ index ];
2025 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
2026 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2027 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
2028 ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
2030 ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len );
2034 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2038 pkt.size = len -= ret;
2039 pkt.data = ptr += ret;
2041 // If decoded successfully
2042 if ( data_size > 0 )
2044 // Figure out how many samples will be needed after resampling
2045 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2046 int samples_needed = self->resample_factor * convert_samples;
2048 // Resize audio buffer to prevent overflow
2049 if ( ( audio_used + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2051 self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample;
2052 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2056 // Copy to audio buffer while resampling
2057 uint8_t *source = decode_buffer;
2058 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
2059 audio_used += audio_resample( resample, (short*) dest, (short*) source, convert_samples );
2063 // Straight copy to audio buffer
2064 memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size );
2065 audio_used += convert_samples;
2069 while ( *ignore && audio_used )
2072 audio_used -= audio_used > samples ? samples : audio_used;
2073 memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ],
2074 audio_used * sizeof_sample );
2079 // If we're behind, ignore this packet
2080 // Skip this on non-seekable, audio-only inputs.
2081 if ( pkt.pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2083 double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt.pts;
2084 int64_t req_position = ( int64_t )( timecode * fps + 0.5 );
2085 int64_t int_position = ( int64_t )( current_pts * fps + 0.5 );
2086 if ( context->start_time != AV_NOPTS_VALUE )
2087 int_position -= ( int64_t )( fps * context->start_time / AV_TIME_BASE + 0.5 );
2089 if ( int_position > 0 )
2091 if ( int_position < req_position )
2092 // We are behind, so skip some
2093 *ignore = req_position - int_position;
2094 else if ( self->audio_index != INT_MAX && int_position > req_position + 2 )
2095 // We are ahead, so seek backwards some more
2096 seek_audio( self, req_position, timecode - 1.0, ignore );
2100 self->audio_used[ index ] = audio_used;
2105 /** Get the audio from a frame.
2107 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2110 producer_avformat self = mlt_frame_pop_audio( frame );
2112 pthread_mutex_lock( &self->audio_mutex );
2114 // Obtain the frame number of this frame
2115 mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" );
2117 // Calculate the real time code
2118 double real_timecode = producer_time_of_frame( self->parent, position );
2120 // Get the producer fps
2121 double fps = mlt_producer_get_fps( self->parent );
2123 // Number of frames to ignore (for ffwd)
2124 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2126 // Flag for paused (silence)
2127 int paused = seek_audio( self, position, real_timecode, &ignore[0] );
2129 // Initialize ignore for all streams from the seek return value
2130 int i = MAX_AUDIO_STREAMS;
2132 ignore[i] = ignore[0];
2134 // Fetch the audio_format
2135 AVFormatContext *context = self->audio_format;
2137 int sizeof_sample = sizeof( int16_t );
2139 // Determine the tracks to use
2140 int index = self->audio_index;
2141 int index_max = self->audio_index + 1;
2142 if ( self->audio_index == INT_MAX )
2145 index_max = context->nb_streams;
2146 *channels = self->total_channels;
2147 *samples = *samples * FFMAX( self->max_frequency, *frequency ) / *frequency;
2148 *frequency = FFMAX( self->max_frequency, *frequency );
2151 // Initialize the resamplers and buffers
2152 for ( ; index < index_max; index++ )
2154 // Get codec context
2155 AVCodecContext *codec_context = self->audio_codec[ index ];
2157 if ( codec_context && !self->audio_buffer[ index ] )
2159 // Check for resample and create if necessary
2160 if ( codec_context->channels <= 2 )
2162 // Determine by how much resampling will increase number of samples
2163 double resample_factor = self->audio_index == INT_MAX ? 1 : (double) *channels / codec_context->channels;
2164 resample_factor *= (double) *frequency / codec_context->sample_rate;
2165 if ( resample_factor > self->resample_factor )
2166 self->resample_factor = resample_factor;
2168 // Create the resampler
2169 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0))
2170 self->audio_resample[ index ] = av_audio_resample_init(
2171 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2172 codec_context->channels, *frequency, codec_context->sample_rate,
2173 SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 );
2175 self->audio_resample[ index ] = audio_resample_init(
2176 self->audio_index == INT_MAX ? codec_context->channels : *channels,
2177 codec_context->channels, *frequency, codec_context->sample_rate );
2182 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2183 sizeof_sample = sample_bytes( codec_context );
2186 // Check for audio buffer and create if necessary
2187 self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2188 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2190 // Check for decoder buffer and create if necessary
2191 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2195 // Get the audio if required
2202 av_init_packet( &pkt );
2204 // If not resampling, give consumer more than requested.
2205 // It requested number samples based on requested frame rate.
2206 // Do not clean this up with a samples *= ...!
2207 if ( self->audio_index != INT_MAX && ! self->audio_resample[ self->audio_index ] )
2208 *samples = *samples * self->audio_codec[ self->audio_index ]->sample_rate / *frequency;
2210 while ( ret >= 0 && !got_audio )
2212 // Check if the buffer already contains the samples required
2213 if ( self->audio_index != INT_MAX &&
2214 self->audio_used[ self->audio_index ] >= *samples &&
2215 ignore[ self->audio_index ] == 0 )
2220 else if ( self->audio_index == INT_MAX )
2222 // Check if there is enough audio for all streams
2224 for ( index = 0; got_audio && index < context->nb_streams; index++ )
2225 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2232 pthread_mutex_lock( &self->packets_mutex );
2233 if ( mlt_deque_count( self->apackets ) )
2235 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2241 ret = av_read_frame( context, &pkt );
2242 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2244 if ( !av_dup_packet( &pkt ) )
2246 AVPacket *tmp = malloc( sizeof(AVPacket) );
2248 mlt_deque_push_back( self->vpackets, tmp );
2252 pthread_mutex_unlock( &self->packets_mutex );
2254 // We only deal with audio from the selected audio index
2255 index = pkt.stream_index;
2256 if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2257 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2259 int channels2 = ( self->audio_index == INT_MAX || !self->audio_resample[index] ) ?
2260 self->audio_codec[index]->channels : *channels;
2261 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2264 if ( self->seekable || index != self->video_index )
2265 av_free_packet( &pkt );
2269 // Set some additional return values
2270 *format = mlt_audio_s16;
2271 if ( self->audio_index != INT_MAX && !self->audio_resample[ self->audio_index ] )
2273 index = self->audio_index;
2274 *channels = self->audio_codec[ index ]->channels;
2275 *frequency = self->audio_codec[ index ]->sample_rate;
2276 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2277 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2279 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2281 else if ( self->audio_index == INT_MAX )
2283 // This only works if all audio tracks have the same sample format.
2284 for ( index = 0; index < index_max; index++ )
2285 if ( self->audio_codec[ index ] && !self->audio_resample[ index ] )
2287 *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2288 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2290 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2295 // Allocate and set the frame's audio buffer
2296 int size = mlt_audio_format_size( *format, *samples, *channels );
2297 *buffer = mlt_pool_alloc( size );
2298 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2300 // Interleave tracks if audio_index=all
2301 if ( self->audio_index == INT_MAX )
2303 uint8_t *dest = *buffer;
2305 for ( i = 0; i < *samples; i++ )
2307 for ( index = 0; index < index_max; index++ )
2308 if ( self->audio_codec[ index ] )
2310 int current_channels = self->audio_codec[ index ]->channels;
2311 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2312 memcpy( dest, src, current_channels * sizeof_sample );
2313 dest += current_channels * sizeof_sample;
2316 for ( index = 0; index < index_max; index++ )
2317 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2319 int current_channels = self->audio_codec[ index ]->channels;
2320 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2321 self->audio_used[index] -= *samples;
2322 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2325 // Copy a single track to the output buffer
2328 index = self->audio_index;
2330 // Now handle the audio if we have enough
2331 if ( self->audio_used[ index ] > 0 )
2333 uint8_t *src = self->audio_buffer[ index ];
2334 // copy samples from audio_buffer
2335 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2336 memcpy( *buffer, src, size * *channels * sizeof_sample );
2337 // supply the remaining requested samples as silence
2338 if ( *samples > self->audio_used[ index ] )
2339 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2340 // reposition the samples within audio_buffer
2341 self->audio_used[ index ] -= size;
2342 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2346 // Otherwise fill with silence
2347 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2353 // Get silence and don't touch the context
2354 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2357 // Regardless of speed (other than paused), we expect to get the next frame
2359 self->audio_expected = position + 1;
2361 pthread_mutex_unlock( &self->audio_mutex );
2366 /** Initialize the audio codec context.
2369 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2371 // Initialise the codec if necessary
2372 if ( !self->audio_codec[ index ] )
2374 // Get codec context
2375 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2378 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2380 // If we don't have a codec and we can't initialise it, we can't do much more...
2382 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2384 // Now store the codec with its destructor
2385 if ( self->audio_codec[ index ] )
2386 avcodec_close( self->audio_codec[ index ] );
2387 self->audio_codec[ index ] = codec_context;
2391 // Remember that we can't use self later
2392 self->audio_index = -1;
2396 // Process properties as AVOptions
2397 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2398 #if LIBAVCODEC_VERSION_MAJOR >= 53
2399 if ( codec && codec->priv_class && codec_context->priv_data )
2400 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2403 return self->audio_codec[ index ] && self->audio_index > -1;
2406 /** Set up audio handling.
2409 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2412 mlt_producer producer = self->parent;
2414 // Get the properties
2415 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2417 // Fetch the audio format context
2418 AVFormatContext *context = self->audio_format;
2420 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2422 // Get the audio_index
2423 int index = mlt_properties_get_int( properties, "audio_index" );
2425 // Handle all audio tracks
2426 if ( self->audio_index > -1 &&
2427 mlt_properties_get( properties, "audio_index" ) &&
2428 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2431 // Reopen the file if necessary
2432 if ( !context && self->audio_index > -1 && index > -1 )
2434 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2435 mlt_properties_get( properties, "resource" ), 1 );
2436 context = self->audio_format;
2439 // Exception handling for audio_index
2440 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2442 for ( index = context->nb_streams - 1;
2443 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2445 mlt_properties_set_int( properties, "audio_index", index );
2447 if ( context && index > -1 && index < INT_MAX &&
2448 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2450 index = self->audio_index;
2451 mlt_properties_set_int( properties, "audio_index", index );
2454 // Update the audio properties if the index changed
2455 if ( context && index > -1 && index != self->audio_index )
2458 if ( self->audio_codec[ self->audio_index ] )
2459 avcodec_close( self->audio_codec[ self->audio_index ] );
2460 self->audio_codec[ self->audio_index ] = NULL;
2463 if ( self->audio_index != -1 )
2464 self->audio_index = index;
2469 if ( context && index == INT_MAX )
2471 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2472 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2473 for ( index = 0; index < context->nb_streams; index++ )
2475 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2476 audio_codec_init( self, index, properties );
2479 else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2481 // Set the frame properties
2482 if ( index < INT_MAX )
2484 mlt_properties_set_int( frame_properties, "frequency", self->audio_codec[ index ]->sample_rate );
2485 mlt_properties_set_int( frame_properties, "channels", self->audio_codec[ index ]->channels );
2488 if ( context && index > -1 )
2490 // Add our audio operation
2491 mlt_frame_push_audio( frame, self );
2492 mlt_frame_push_audio( frame, producer_get_audio );
2496 /** Our get frame implementation.
2499 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2501 // Access the private data
2502 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2503 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2504 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2509 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2510 producer->child = self;
2511 self->parent = producer;
2512 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2513 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2516 // Create an empty frame
2517 *frame = mlt_frame_init( service);
2521 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2525 mlt_cache_item_close( cache_item );
2529 // Update timecode on the frame we're creating
2530 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2533 producer_set_up_video( self, *frame );
2536 producer_set_up_audio( self, *frame );
2538 // Set the position of this producer
2539 mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2540 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", position );
2542 // Calculate the next timecode
2543 mlt_producer_prepare_next( producer );
2548 static void producer_avformat_close( producer_avformat self )
2550 mlt_log_debug( NULL, "producer_avformat_close\n" );
2552 // Cleanup av contexts
2553 av_free( self->av_frame );
2556 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2558 if ( self->audio_resample[i] )
2559 audio_resample_close( self->audio_resample[i] );
2560 mlt_pool_release( self->audio_buffer[i] );
2561 av_free( self->decode_buffer[i] );
2562 if ( self->audio_codec[i] )
2563 avcodec_close( self->audio_codec[i] );
2564 self->audio_codec[i] = NULL;
2566 if ( self->video_codec )
2567 avcodec_close( self->video_codec );
2568 self->video_codec = NULL;
2570 if ( self->dummy_context )
2571 av_close_input_file( self->dummy_context );
2572 if ( self->seekable && self->audio_format )
2573 av_close_input_file( self->audio_format );
2574 if ( self->video_format )
2575 av_close_input_file( self->video_format );
2578 vdpau_producer_close( self );
2580 if ( self->image_cache )
2581 mlt_cache_close( self->image_cache );
2583 // Cleanup the mutexes
2584 pthread_mutex_destroy( &self->audio_mutex );
2585 pthread_mutex_destroy( &self->video_mutex );
2586 pthread_mutex_destroy( &self->packets_mutex );
2588 // Cleanup the packet queues
2590 if ( self->apackets )
2592 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2594 av_free_packet( pkt );
2597 mlt_deque_close( self->apackets );
2598 self->apackets = NULL;
2600 if ( self->vpackets )
2602 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2604 av_free_packet( pkt );
2607 mlt_deque_close( self->vpackets );
2608 self->vpackets = NULL;
2614 static void producer_close( mlt_producer parent )
2616 // Remove this instance from the cache
2617 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2620 parent->close = NULL;
2621 mlt_producer_close( parent );