2 * producer_avformat.c -- avformat producer
3 * Copyright (C) 2003-2012 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>
34 #include <libswscale/swscale.h>
35 #include <libavutil/samplefmt.h>
36 #include <libavutil/pixdesc.h>
39 # include <libavcodec/vdpau.h>
41 #if (LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0))
42 # include <libavutil/dict.h>
45 // System header files
51 #if LIBAVCODEC_VERSION_MAJOR >= 53
52 #include <libavutil/opt.h>
53 #define CODEC_TYPE_VIDEO AVMEDIA_TYPE_VIDEO
54 #define CODEC_TYPE_AUDIO AVMEDIA_TYPE_AUDIO
55 #define PKT_FLAG_KEY AV_PKT_FLAG_KEY
57 #include <libavcodec/opt.h>
60 #if LIBAVCODEC_VERSION_MAJOR < 55
61 #define AV_CODEC_ID_DVVIDEO CODEC_ID_DVVIDEO
62 #define AV_CODEC_ID_H264 CODEC_ID_H264
65 #define POSITION_INITIAL (-2)
66 #define POSITION_INVALID (-1)
68 #define MAX_AUDIO_STREAMS (32)
69 #define MAX_VDPAU_SURFACES (10)
70 #define MAX_AUDIO_FRAME_SIZE (192000) // 1 second of 48khz 32bit audio
72 struct producer_avformat_s
75 AVFormatContext *dummy_context;
76 AVFormatContext *audio_format;
77 AVFormatContext *video_format;
78 AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
79 AVCodecContext *video_codec;
83 mlt_position audio_expected;
84 mlt_position video_expected;
88 int64_t last_position;
90 int64_t current_position;
91 mlt_position nonseek_position;
93 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
94 size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
95 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
96 int audio_used[ MAX_AUDIO_STREAMS ];
102 unsigned int invalid_pts_counter;
103 unsigned int invalid_dts_counter;
104 mlt_cache image_cache;
107 pthread_mutex_t video_mutex;
108 pthread_mutex_t audio_mutex;
111 pthread_mutex_t packets_mutex;
112 pthread_mutex_t open_mutex;
114 AVRational video_time_base;
119 struct vdpau_render_state render_states[MAX_VDPAU_SURFACES];
133 typedef struct producer_avformat_s *producer_avformat;
135 // Forward references.
136 static int list_components( char* file );
137 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock );
138 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
139 static void producer_avformat_close( producer_avformat );
140 static void producer_close( mlt_producer parent );
141 static void producer_set_up_video( producer_avformat self, mlt_frame frame );
142 static void producer_set_up_audio( producer_avformat self, mlt_frame frame );
143 static void apply_properties( void *obj, mlt_properties properties, int flags );
144 static int video_codec_init( producer_avformat self, int index, mlt_properties properties );
145 static void get_audio_streams_info( producer_avformat self );
146 static mlt_audio_format pick_audio_format( int sample_fmt );
152 /** Constructor for libavformat.
155 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
157 if ( list_components( file ) )
160 mlt_producer producer = NULL;
162 // Check that we have a non-NULL argument
165 // Construct the producer
166 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
167 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
170 if ( mlt_producer_init( producer, self ) == 0 )
172 self->parent = producer;
174 // Get the properties
175 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
177 // Set the resource property (required for all producers)
178 mlt_properties_set( properties, "resource", file );
180 // Register transport implementation with the producer
181 producer->close = (mlt_destructor) producer_close;
183 // Register our get_frame implementation
184 producer->get_frame = producer_get_frame;
186 if ( strcmp( service, "avformat-novalidate" ) )
189 if ( producer_open( self, profile, file, 1 ) != 0 )
192 mlt_producer_close( producer );
194 producer_avformat_close( self );
196 else if ( self->seekable )
198 // Close the file to release resources for large playlists - reopen later as needed
199 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
200 if ( self->audio_format )
201 avformat_close_input( &self->audio_format );
202 if ( self->video_format )
203 avformat_close_input( &self->video_format );
205 if ( self->audio_format )
206 av_close_input_file( self->audio_format );
207 if ( self->video_format )
208 av_close_input_file( self->video_format );
210 self->audio_format = NULL;
211 self->video_format = NULL;
216 // Default the user-selectable indices from the auto-detected indices
217 mlt_properties_set_int( properties, "audio_index", self->audio_index );
218 mlt_properties_set_int( properties, "video_index", self->video_index );
220 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
222 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
229 int list_components( char* file )
233 // Report information about available demuxers and codecs as YAML Tiny
234 if ( file && strstr( file, "f-list" ) )
236 fprintf( stderr, "---\nformats:\n" );
237 AVInputFormat *format = NULL;
238 while ( ( format = av_iformat_next( format ) ) )
239 fprintf( stderr, " - %s\n", format->name );
240 fprintf( stderr, "...\n" );
243 if ( file && strstr( file, "acodec-list" ) )
245 fprintf( stderr, "---\naudio_codecs:\n" );
246 AVCodec *codec = NULL;
247 while ( ( codec = av_codec_next( codec ) ) )
248 if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
249 fprintf( stderr, " - %s\n", codec->name );
250 fprintf( stderr, "...\n" );
253 if ( file && strstr( file, "vcodec-list" ) )
255 fprintf( stderr, "---\nvideo_codecs:\n" );
256 AVCodec *codec = NULL;
257 while ( ( codec = av_codec_next( codec ) ) )
258 if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
259 fprintf( stderr, " - %s\n", codec->name );
260 fprintf( stderr, "...\n" );
267 static int first_video_index( producer_avformat self )
269 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
270 int i = -1; // not found
273 for ( i = 0; i < context->nb_streams; i++ ) {
274 if ( context->streams[i]->codec &&
275 context->streams[i]->codec->codec_type == CODEC_TYPE_VIDEO )
278 if ( i == context->nb_streams )
284 /** Find the default streams.
287 static mlt_properties find_default_streams( producer_avformat self )
291 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
292 AVDictionaryEntry *tag = NULL;
294 AVMetadataTag *tag = NULL;
296 AVFormatContext *context = self->video_format;
297 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
299 // Default to the first audio and video streams found
300 self->audio_index = -1;
301 self->video_index = -1;
303 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
305 // Allow for multiple audio and video streams in the file and select first of each (if available)
306 for( i = 0; i < context->nb_streams; i++ )
308 // Get the codec context
309 AVStream *stream = context->streams[ i ];
310 if ( ! stream ) continue;
311 AVCodecContext *codec_context = stream->codec;
312 if ( ! codec_context ) continue;
313 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
314 if ( ! codec ) continue;
316 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
318 // Determine the type and obtain the first index of each type
319 switch( codec_context->codec_type )
321 case CODEC_TYPE_VIDEO:
322 // Use first video stream
323 if ( self->video_index < 0 )
324 self->video_index = i;
325 mlt_properties_set( meta_media, key, "video" );
326 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
327 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
328 #if LIBAVFORMAT_VERSION_MAJOR < 55
329 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
330 ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
332 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
334 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
335 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
336 snprintf( key, sizeof(key), "meta.media.%d.codec.width", i );
337 mlt_properties_set_int( meta_media, key, codec_context->width );
338 snprintf( key, sizeof(key), "meta.media.%d.codec.height", i );
339 mlt_properties_set_int( meta_media, key, codec_context->height );
340 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
341 AVRational frame_rate = av_inv_q( codec_context->time_base );
342 frame_rate.den *= codec_context->ticks_per_frame;
343 mlt_properties_set_double( meta_media, key, av_q2d( frame_rate ) );
344 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
345 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(3<<8)+0)
346 mlt_properties_set( meta_media, key, av_get_pix_fmt_name( codec_context->pix_fmt ) );
348 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
350 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
351 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
352 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
353 switch ( codec_context->colorspace )
355 case AVCOL_SPC_SMPTE240M:
356 mlt_properties_set_int( meta_media, key, 240 );
358 case AVCOL_SPC_BT470BG:
359 case AVCOL_SPC_SMPTE170M:
360 mlt_properties_set_int( meta_media, key, 601 );
362 case AVCOL_SPC_BT709:
363 mlt_properties_set_int( meta_media, key, 709 );
366 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
367 mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
371 case CODEC_TYPE_AUDIO:
372 if ( !codec_context->channels )
374 // Use first audio stream
375 if ( self->audio_index < 0 && pick_audio_format( codec_context->sample_fmt ) != mlt_audio_none )
376 self->audio_index = i;
378 mlt_properties_set( meta_media, key, "audio" );
379 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
380 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
381 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
382 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
383 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
384 mlt_properties_set_int( meta_media, key, codec_context->channels );
389 // snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
390 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
391 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
392 mlt_properties_set( meta_media, key, codec->name );
393 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
394 mlt_properties_set( meta_media, key, codec->long_name );
395 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
396 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
397 // snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
398 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
399 // snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
400 // mlt_properties_set_int( meta_media, key, codec_context->profile );
401 // snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
402 // mlt_properties_set_int( meta_media, key, codec_context->level );
405 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
406 while ( ( tag = av_dict_get( stream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
408 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
411 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
413 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
414 mlt_properties_set( meta_media, key, tag->value );
418 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
419 while ( ( tag = av_dict_get( context->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
421 while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
424 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
426 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
427 mlt_properties_set( meta_media, key, tag->value );
434 static inline int dv_is_pal( AVPacket *pkt )
436 return pkt->data[3] & 0x80;
439 static int dv_is_wide( AVPacket *pkt )
441 int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
443 for ( ; i < pkt->size; i += 5 /* packet size */ )
445 if ( pkt->data[ i ] == 0x61 )
447 uint8_t x = pkt->data[ i + 2 ] & 0x7;
448 return ( x == 2 ) || ( x == 7 );
454 static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
456 double aspect_ratio = 1.0;
458 if ( codec_context->codec_id == AV_CODEC_ID_DVVIDEO )
462 if ( dv_is_pal( pkt ) )
464 if ( dv_is_wide( pkt ) )
466 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
467 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
471 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
472 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
477 if ( dv_is_wide( pkt ) )
479 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
480 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
484 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
485 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
491 AVRational ar = stream->sample_aspect_ratio;
492 // Override FFmpeg's notion of DV aspect ratios, which are
493 // based upon a width of 704. Since we do not have a normaliser
494 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
495 // we just coerce the values to facilitate a passive behaviour through
496 // the rescale normaliser when using equivalent producers and consumers.
497 // = display_aspect / (width * height)
498 if ( ar.num == 10 && ar.den == 11 )
501 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
502 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
504 else if ( ar.num == 59 && ar.den == 54 )
507 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
508 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
510 else if ( ar.num == 40 && ar.den == 33 )
513 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
514 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
516 else if ( ar.num == 118 && ar.den == 81 )
519 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
520 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
526 AVRational codec_sar = codec_context->sample_aspect_ratio;
527 AVRational stream_sar = stream->sample_aspect_ratio;
528 if ( codec_sar.num > 0 )
530 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num );
531 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den );
533 else if ( stream_sar.num > 0 )
535 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num );
536 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den );
540 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 );
541 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 );
544 AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) };
545 aspect_ratio = av_q2d( ar );
546 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
551 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
552 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVDictionary **params )
554 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params )
557 if ( !URL ) return NULL;
560 char *protocol = strdup( URL );
561 char *url = strchr( protocol, ':' );
563 // Only if there is not a protocol specification that avformat can handle
564 #if LIBAVFORMAT_VERSION_MAJOR >= 53
565 if ( url && avio_check( URL, 0 ) < 0 )
567 if ( url && !url_exist( URL ) )
570 // Truncate protocol string
572 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url + 1 );
575 *format = av_find_input_format( protocol );
577 // Eat the format designator
582 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
583 // support for legacy width and height parameters
587 // These are required by video4linux2 (defaults)
588 params->width = profile->width;
589 params->height = profile->height;
590 if ( !strstr( URL, "&frame_rate" ) )
591 params->time_base = (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
592 params->channels = 2;
593 params->sample_rate = 48000;
597 url = strchr( url, '?' );
601 char *name = strdup( ++url );
602 char *value = strchr( name, '=' );
604 // Also accept : as delimiter for backwards compatibility.
605 value = strchr( name, ':' );
610 char *t = strchr( value, '&' );
613 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
614 // translate old parameters to new av_dict names
615 if ( !strcmp( name, "frame_rate" ) )
616 av_dict_set( params, "framerate", value, 0 );
617 else if ( !strcmp( name, "pix_fmt" ) )
618 av_dict_set( params, "pixel_format", value, 0 );
619 else if ( !strcmp( name, "width" ) )
620 width = strdup( value );
621 else if ( !strcmp( name, "height" ) )
622 height = strdup( value );
624 // generic demux/device option support
625 av_dict_set( params, name, value, 0 );
627 if ( !strcmp( name, "frame_rate" ) )
628 params->time_base.den = atoi( value );
629 else if ( !strcmp( name, "frame_rate_base" ) )
630 params->time_base.num = atoi( value );
631 else if ( !strcmp( name, "sample_rate" ) )
632 params->sample_rate = atoi( value );
633 else if ( !strcmp( name, "channel" ) )
634 params->channel = atoi( value );
635 else if ( !strcmp( name, "channels" ) )
636 params->channels = atoi( value );
637 else if ( !strcmp( name, "pix_fmt" ) )
638 params->pix_fmt = av_get_pix_fmt( value );
639 else if ( !strcmp( name, "width" ) )
640 params->width = atoi( value );
641 else if ( !strcmp( name, "height" ) )
642 params->height = atoi( value );
643 else if ( !strcmp( name, "standard" ) )
644 params->standard = strdup( value );
648 url = strchr( url, '&' );
650 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
651 // continued support for legacy width and height parameters
652 if ( width && height )
654 char *s = malloc( strlen( width ) + strlen( height ) + 2 );
658 av_dict_set( params, "video_size", s, 0 );
661 if ( width ) free( width );
662 if ( height ) free ( height );
665 result = strdup( result );
669 result = strdup( URL );
675 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
679 // Get the properties
680 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
682 AVFormatContext *format = self->video_format;
684 // We will treat everything with the producer fps.
685 // TODO: make this more flexible.
686 double fps = mlt_profile_fps( profile );
689 if ( !mlt_properties_get_int( properties, "_length_computed" ) )
691 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
692 // when producer_open is called after initial call when restoring or reseting the producer.
693 if ( format->duration != AV_NOPTS_VALUE )
695 // This isn't going to be accurate for all formats
696 mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
697 mlt_properties_set_position( properties, "out", frames - 1 );
698 mlt_properties_set_position( properties, "length", frames );
699 mlt_properties_set_int( properties, "_length_computed", 1 );
703 // Set live sources to run forever
704 mlt_properties_set_position( properties, "length", INT_MAX );
705 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
706 mlt_properties_set( properties, "eof", "loop" );
707 mlt_properties_set_int( properties, "_length_computed", 1 );
711 // Check if we're seekable
712 // avdevices are typically AVFMT_NOFILE and not seekable
713 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
716 // protocols can indicate if they support seeking
717 #if LIBAVFORMAT_VERSION_MAJOR >= 53
718 self->seekable = format->pb->seekable;
720 URLContext *uc = url_fileno( format->pb );
722 self->seekable = !uc->is_streamed;
725 if ( self->seekable )
727 // Do a more rigourous test of seekable on a disposable context
728 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
729 mlt_properties_set_int( properties, "seekable", self->seekable );
730 self->dummy_context = format;
731 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
732 self->video_format = NULL;
733 avformat_open_input( &self->video_format, filename, NULL, NULL );
734 avformat_find_stream_info( self->video_format, NULL );
736 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
737 av_find_stream_info( self->video_format );
739 format = self->video_format;
742 // Fetch the width, height and aspect ratio
743 if ( self->video_index != -1 )
745 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
746 mlt_properties_set_int( properties, "width", codec_context->width );
747 mlt_properties_set_int( properties, "height", codec_context->height );
749 if ( codec_context->codec_id == AV_CODEC_ID_DVVIDEO )
751 // Fetch the first frame of DV so we can read it directly
756 ret = av_read_frame( format, &pkt );
757 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
759 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, &pkt );
760 av_free_packet(&pkt);
764 av_free_packet(&pkt);
769 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, NULL );
772 // Verify that we can convert this to YUV 4:2:2
773 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
774 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
775 codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
777 sws_freeContext( context );
787 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock )
789 // Return an error code (0 == no error)
791 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
796 if ( !self->is_mutex_init )
798 pthread_mutex_init( &self->audio_mutex, NULL );
799 pthread_mutex_init( &self->video_mutex, NULL );
800 pthread_mutex_init( &self->packets_mutex, NULL );
801 pthread_mutex_init( &self->open_mutex, NULL );
802 self->is_mutex_init = 1;
804 pthread_mutex_lock( &self->audio_mutex );
805 pthread_mutex_lock( &self->video_mutex );
807 mlt_events_block( properties, self->parent );
810 AVInputFormat *format = NULL;
811 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
812 AVDictionary *params = NULL;
814 AVFormatParameters params;
815 memset( ¶ms, 0, sizeof(params) );
817 char *filename = parse_url( profile, URL, &format, ¶ms );
819 // Now attempt to open the file or device with filename
820 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
821 error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0;
823 // If the URL is a network stream URL, then we probably need to open with full URL
824 error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0;
826 error = av_open_input_file( &self->video_format, filename, format, 0, ¶ms ) < 0;
828 // If the URL is a network stream URL, then we probably need to open with full URL
829 error = av_open_input_file( &self->video_format, URL, format, 0, ¶ms ) < 0;
832 // Set MLT properties onto video AVFormatContext
833 if ( !error && self->video_format )
835 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
836 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
837 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
840 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
841 av_dict_free( ¶ms );
843 // Cleanup AVFormatParameters
844 if ( params.standard )
845 free( (void*) params.standard );
848 // If successful, then try to get additional info
849 if ( !error && self->video_format )
851 // Get the stream info
852 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
853 error = avformat_find_stream_info( self->video_format, NULL ) < 0;
855 error = av_find_stream_info( self->video_format ) < 0;
858 // Continue if no error
859 if ( !error && self->video_format )
861 // Find default audio and video streams
862 find_default_streams( self );
863 error = get_basic_info( self, profile, filename );
865 // Initialize position info
866 self->first_pts = AV_NOPTS_VALUE;
867 self->last_position = POSITION_INITIAL;
869 if ( !self->audio_format )
871 // We're going to cheat here - for seekable A/V files, we will have separate contexts
872 // to support independent seeking of audio from video.
873 // TODO: Is this really necessary?
874 if ( self->audio_index != -1 && self->video_index != -1 )
876 if ( self->seekable )
878 // And open again for our audio context
879 #if LIBAVFORMAT_VERSION_INT > ((53<<16)+(6<<8)+0)
880 avformat_open_input( &self->audio_format, filename, NULL, NULL );
881 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
882 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
883 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
884 avformat_find_stream_info( self->audio_format, NULL );
886 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
887 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
888 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(110<<8)+0)
889 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
890 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
892 av_find_stream_info( self->audio_format );
897 self->audio_format = self->video_format;
900 else if ( self->audio_index != -1 )
902 // We only have an audio context
903 self->audio_format = self->video_format;
904 self->video_format = NULL;
906 else if ( self->video_index == -1 )
908 // Something has gone wrong
911 if ( self->audio_format && !self->audio_streams )
912 get_audio_streams_info( self );
920 self->apackets = mlt_deque_init();
921 self->vpackets = mlt_deque_init();
924 if ( self->dummy_context )
926 pthread_mutex_lock( &self->open_mutex );
927 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
928 avformat_close_input( &self->dummy_context );
930 av_close_input_file( self->dummy_context );
932 self->dummy_context = NULL;
933 pthread_mutex_unlock( &self->open_mutex );
936 // Unlock the service
939 pthread_mutex_unlock( &self->audio_mutex );
940 pthread_mutex_unlock( &self->video_mutex );
942 mlt_events_unblock( properties, self->parent );
947 static void reopen_video( producer_avformat self, mlt_producer producer )
949 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
950 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
951 pthread_mutex_lock( &self->audio_mutex );
952 pthread_mutex_lock( &self->open_mutex );
954 if ( self->video_codec )
955 avcodec_close( self->video_codec );
956 self->video_codec = NULL;
957 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
958 if ( self->dummy_context )
959 avformat_close_input( &self->dummy_context );
960 if ( self->video_format )
961 avformat_close_input( &self->video_format );
963 if ( self->dummy_context )
964 av_close_input_file( self->dummy_context );
965 if ( self->video_format )
966 av_close_input_file( self->video_format );
968 self->dummy_context = NULL;
969 self->video_format = NULL;
970 pthread_mutex_unlock( &self->open_mutex );
972 int audio_index = self->audio_index;
973 int video_index = self->video_index;
975 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
976 mlt_properties_get( properties, "resource" ), 0 );
978 self->audio_index = audio_index;
979 if ( self->video_format && video_index > -1 )
981 self->video_index = video_index;
982 video_codec_init( self, video_index, properties );
985 pthread_mutex_unlock( &self->audio_mutex );
986 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
989 static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts )
991 self->invalid_pts_counter += pts == AV_NOPTS_VALUE;
992 self->invalid_dts_counter += dts == AV_NOPTS_VALUE;
993 if ( ( self->invalid_pts_counter <= self->invalid_dts_counter
994 || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE )
1000 static void find_first_pts( producer_avformat self, int video_index )
1003 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
1008 while ( ret >= 0 && toscan-- > 0 )
1010 ret = av_read_frame( context, &pkt );
1011 if ( ret >= 0 && pkt.stream_index == video_index && ( pkt.flags & PKT_FLAG_KEY ) )
1013 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
1014 "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n",
1015 pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
1016 self->first_pts = best_pts( self, pkt.pts, pkt.dts );
1017 if ( self->first_pts != AV_NOPTS_VALUE )
1020 av_free_packet( &pkt );
1022 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
1025 static int seek_video( producer_avformat self, mlt_position position,
1026 int64_t req_position, int preseek )
1028 mlt_producer producer = self->parent;
1031 if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
1033 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1035 // Fetch the video format context
1036 AVFormatContext *context = self->video_format;
1038 // Get the video stream
1039 AVStream *stream = context->streams[ self->video_index ];
1041 // Get codec context
1042 AVCodecContext *codec_context = stream->codec;
1044 // We may want to use the source fps if available
1045 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1046 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1048 if ( self->last_position == POSITION_INITIAL )
1049 find_first_pts( self, self->video_index );
1051 if ( self->video_frame && position + 1 == self->video_expected )
1053 // We're paused - use last image
1056 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
1058 // Calculate the timestamp for the requested frame
1059 int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps );
1060 if ( req_position <= 0 )
1062 else if ( self->first_pts != AV_NOPTS_VALUE )
1063 timestamp += self->first_pts;
1064 else if ( context->start_time != AV_NOPTS_VALUE )
1065 timestamp += context->start_time;
1066 if ( preseek && av_q2d( self->video_time_base ) != 0 )
1067 timestamp -= 2 / av_q2d( self->video_time_base );
1068 if ( timestamp < 0 )
1070 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %"PRId64"\n",
1071 timestamp, position, self->video_expected, self->last_position );
1073 // Seek to the timestamp
1074 // NOTE: reopen_video is disabled at this time because it is causing trouble with A/V sync.
1075 if ( 1 || req_position > 0 || self->last_position <= 0 )
1077 codec_context->skip_loop_filter = AVDISCARD_NONREF;
1078 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
1080 // flush any pictures still in decode buffer
1081 avcodec_flush_buffers( codec_context );
1085 // Re-open video stream when rewinding to beginning from somewhere else.
1086 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
1087 // not reliably seeking to the first frame across formats.
1088 reopen_video( self, producer );
1091 // Remove the cached info relating to the previous position
1092 self->current_position = POSITION_INVALID;
1093 self->last_position = POSITION_INVALID;
1094 av_freep( &self->video_frame );
1100 /** Convert a frame position to a time code.
1103 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1105 return ( double )position / mlt_producer_get_fps( producer );
1108 // Collect information about all audio streams
1110 static void get_audio_streams_info( producer_avformat self )
1112 // Fetch the audio format context
1113 AVFormatContext *context = self->audio_format;
1117 i < context->nb_streams;
1120 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
1122 AVCodecContext *codec_context = context->streams[i]->codec;
1123 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1125 // If we don't have a codec and we can't initialise it, we can't do much more...
1126 pthread_mutex_lock( &self->open_mutex );
1127 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1128 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1130 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1133 self->audio_streams++;
1134 self->audio_max_stream = i;
1135 self->total_channels += codec_context->channels;
1136 if ( codec_context->channels > self->max_channel )
1137 self->max_channel = codec_context->channels;
1138 if ( codec_context->sample_rate > self->max_frequency )
1139 self->max_frequency = codec_context->sample_rate;
1140 avcodec_close( codec_context );
1142 pthread_mutex_unlock( &self->open_mutex );
1145 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1146 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1149 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
1152 const int *new_coefficients;
1154 int brightness, contrast, saturation;
1156 if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
1157 &brightness, &contrast, &saturation ) != -1 )
1159 // Don't change these from defaults unless explicitly told to.
1160 if ( use_full_range >= 0 )
1161 full_range = use_full_range;
1162 switch ( colorspace )
1168 new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1171 new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1174 new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1177 new_coefficients = coefficients;
1180 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
1181 brightness, contrast, saturation );
1185 static mlt_image_format pick_pix_format( enum PixelFormat pix_fmt )
1193 return mlt_image_rgb24a;
1194 case PIX_FMT_YUV420P:
1195 case PIX_FMT_YUVJ420P:
1196 case PIX_FMT_YUVA420P:
1197 return mlt_image_yuv420p;
1201 case PIX_FMT_MONOWHITE:
1202 case PIX_FMT_MONOBLACK:
1205 return mlt_image_rgb24;
1207 return mlt_image_yuv422;
1211 static mlt_audio_format pick_audio_format( int sample_fmt )
1213 switch ( sample_fmt )
1216 case AV_SAMPLE_FMT_U8:
1217 return mlt_audio_u8;
1218 case AV_SAMPLE_FMT_S16:
1219 return mlt_audio_s16;
1220 case AV_SAMPLE_FMT_S32:
1221 return mlt_audio_s32le;
1222 case AV_SAMPLE_FMT_FLT:
1223 return mlt_audio_f32le;
1224 // planar - this producer converts planar to interleaved
1225 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
1226 case AV_SAMPLE_FMT_U8P:
1227 return mlt_audio_u8;
1228 case AV_SAMPLE_FMT_S16P:
1229 return mlt_audio_s16;
1230 case AV_SAMPLE_FMT_S32P:
1231 return mlt_audio_s32le;
1232 case AV_SAMPLE_FMT_FLTP:
1233 return mlt_audio_f32le;
1236 return mlt_audio_none;
1240 static void convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt,
1241 mlt_image_format *format, int width, int height, uint8_t **alpha )
1243 int flags = SWS_BICUBIC | SWS_ACCURATE_RND;
1246 flags |= SWS_CPU_CAPS_MMX;
1249 flags |= SWS_CPU_CAPS_MMX2;
1252 // extract alpha from planar formats
1253 if ( ( pix_fmt == PIX_FMT_YUVA420P
1254 #if defined(FFUDIV) && LIBAVUTIL_VERSION_INT >= ((51<<16)+(35<<8)+101)
1255 || pix_fmt == PIX_FMT_YUVA444P
1258 *format != mlt_image_rgb24a && *format != mlt_image_opengl &&
1259 frame->data[3] && frame->linesize[3] )
1264 dst = *alpha = mlt_pool_alloc( width * height );
1265 src = frame->data[3];
1267 for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ )
1268 memcpy( dst, src, FFMIN( width, frame->linesize[3] ) );
1271 if ( *format == mlt_image_yuv420p )
1273 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1274 width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1276 output.data[0] = buffer;
1277 output.data[1] = buffer + width * height;
1278 output.data[2] = buffer + ( 5 * width * height ) / 4;
1279 output.linesize[0] = width;
1280 output.linesize[1] = width >> 1;
1281 output.linesize[2] = width >> 1;
1282 set_luma_transfer( context, self->colorspace, -1 );
1283 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1284 output.data, output.linesize);
1285 sws_freeContext( context );
1287 else if ( *format == mlt_image_rgb24 )
1289 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1290 width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1292 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1293 set_luma_transfer( context, self->colorspace, self->full_luma );
1294 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1295 output.data, output.linesize);
1296 sws_freeContext( context );
1298 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1300 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1301 width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1303 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1304 set_luma_transfer( context, self->colorspace, self->full_luma );
1305 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1306 output.data, output.linesize);
1307 sws_freeContext( context );
1311 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1312 width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1314 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1315 set_luma_transfer( context, self->colorspace, -1 );
1316 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1317 output.data, output.linesize);
1318 sws_freeContext( context );
1322 /** Allocate the image buffer and set it on the frame.
1325 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1329 if ( codec_context->width == 0 || codec_context->height == 0 )
1332 if ( *format == mlt_image_glsl )
1333 *format = pick_pix_format( codec_context->pix_fmt );
1335 *width = codec_context->width;
1336 *height = codec_context->height;
1337 size = mlt_image_format_size( *format, *width, *height, NULL );
1338 *buffer = mlt_pool_alloc( size );
1340 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1347 /** Get an image from a frame.
1350 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1353 producer_avformat self = mlt_frame_pop_service( frame );
1354 mlt_producer producer = self->parent;
1356 // Get the properties from the frame
1357 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1359 // Obtain the frame number of this frame
1360 mlt_position position = mlt_frame_original_position( frame );
1362 // Get the producer properties
1363 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1365 pthread_mutex_lock( &self->video_mutex );
1367 // Fetch the video format context
1368 AVFormatContext *context = self->video_format;
1370 // Get the video stream
1371 AVStream *stream = context->streams[ self->video_index ];
1373 // Get codec context
1374 AVCodecContext *codec_context = stream->codec;
1376 uint8_t *alpha = NULL;
1377 int got_picture = 0;
1380 // Get the image cache
1381 if ( ! self->image_cache )
1383 // if cache size supplied by environment variable
1384 int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL;
1385 int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0;
1387 // cache size supplied via property
1388 if ( mlt_properties_get( properties, "cache" ) )
1391 cache_size = mlt_properties_get_int( properties, "cache" );
1393 if ( mlt_properties_get_int( properties, "noimagecache" ) )
1395 // create cache if not disabled
1396 if ( !cache_supplied || cache_size > 0 )
1397 self->image_cache = mlt_cache_init();
1398 // set cache size if supplied
1399 if ( self->image_cache && cache_supplied )
1400 mlt_cache_set_size( self->image_cache, cache_size );
1402 if ( self->image_cache )
1404 mlt_frame original = mlt_cache_get_frame( self->image_cache, position );
1407 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
1410 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
1412 mlt_frame_set_alpha( frame, *buffer, size, NULL );
1413 *buffer = mlt_properties_get_data( orig_props, "image", &size );
1414 mlt_frame_set_image( frame, *buffer, size, NULL );
1415 mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL );
1416 *format = mlt_properties_get_int( orig_props, "format" );
1418 // Set the resolution
1419 *width = codec_context->width;
1420 *height = codec_context->height;
1422 // Workaround 1088 encodings missing cropping info.
1423 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1427 goto exit_get_image;
1432 // We may want to use the source fps if available
1433 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1434 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1436 // This is the physical frame position in the source
1437 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1439 // Determines if we have to decode all frames in a sequence
1440 // Temporary hack to improve intra frame only
1441 int must_decode = !( codec_context->codec && codec_context->codec->name ) || (
1442 strcmp( codec_context->codec->name, "dnxhd" ) &&
1443 strcmp( codec_context->codec->name, "dvvideo" ) &&
1444 strcmp( codec_context->codec->name, "huffyuv" ) &&
1445 strcmp( codec_context->codec->name, "mjpeg" ) &&
1446 strcmp( codec_context->codec->name, "rawvideo" ) );
1448 double delay = mlt_properties_get_double( properties, "video_delay" );
1450 // Seek if necessary
1451 const char *interp = mlt_properties_get( frame_properties, "rescale.interp" );
1452 int preseek = must_decode
1453 #if defined(FFUDIV) && LIBAVFORMAT_VERSION_INT >= ((53<<16)+(24<<8)+2)
1454 && ( interp && strcmp( interp, "nearest" ) )
1456 && codec_context->has_b_frames;
1457 int paused = seek_video( self, position, req_position, preseek );
1459 // Seek might have reopened the file
1460 context = self->video_format;
1461 stream = context->streams[ self->video_index ];
1462 codec_context = stream->codec;
1463 if ( *format == mlt_image_none ||
1464 codec_context->pix_fmt == PIX_FMT_ARGB ||
1465 codec_context->pix_fmt == PIX_FMT_RGBA ||
1466 codec_context->pix_fmt == PIX_FMT_ABGR ||
1467 codec_context->pix_fmt == PIX_FMT_BGRA )
1468 *format = pick_pix_format( codec_context->pix_fmt );
1470 // Duplicate the last image if necessary
1471 if ( self->video_frame && self->video_frame->linesize[0]
1472 && ( paused || self->current_position >= req_position ) )
1475 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1477 // Workaround 1088 encodings missing cropping info.
1478 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1481 if ( self->vdpau && self->vdpau->buffer )
1484 picture.data[0] = self->vdpau->buffer;
1485 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1486 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1487 picture.linesize[0] = codec_context->width;
1488 picture.linesize[1] = codec_context->width / 2;
1489 picture.linesize[2] = codec_context->width / 2;
1490 convert_image( self, (AVFrame*) &picture, *buffer,
1491 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1495 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1496 format, *width, *height, &alpha );
1503 int64_t int_position = 0;
1504 int decode_errors = 0;
1506 // Construct an AVFrame for YUV422 conversion
1507 if ( !self->video_frame )
1508 self->video_frame = avcodec_alloc_frame( );
1510 while( ret >= 0 && !got_picture )
1513 if ( self->pkt.stream_index == self->video_index )
1514 av_free_packet( &self->pkt );
1515 av_init_packet( &self->pkt );
1516 pthread_mutex_lock( &self->packets_mutex );
1517 if ( mlt_deque_count( self->vpackets ) )
1519 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1525 ret = av_read_frame( context, &self->pkt );
1526 if ( ret >= 0 && !self->seekable && self->pkt.stream_index == self->audio_index )
1528 if ( !av_dup_packet( &self->pkt ) )
1530 AVPacket *tmp = malloc( sizeof(AVPacket) );
1532 mlt_deque_push_back( self->apackets, tmp );
1536 pthread_mutex_unlock( &self->packets_mutex );
1538 // We only deal with video from the selected video_index
1539 if ( ret >= 0 && self->pkt.stream_index == self->video_index && self->pkt.size > 0 )
1541 int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts );
1542 if ( pts != AV_NOPTS_VALUE )
1544 if ( !self->seekable && self->first_pts == AV_NOPTS_VALUE )
1545 self->first_pts = pts;
1546 if ( self->first_pts != AV_NOPTS_VALUE )
1547 pts -= self->first_pts;
1548 else if ( context->start_time != AV_NOPTS_VALUE )
1549 pts -= context->start_time;
1550 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1551 if ( int_position == self->last_position )
1552 int_position = self->last_position + 1;
1554 mlt_log_debug( MLT_PRODUCER_SERVICE(producer),
1555 "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1556 self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position );
1558 // Make a dumb assumption on streams that contain wild timestamps
1559 if ( abs( req_position - int_position ) > 999 )
1561 int_position = req_position;
1562 mlt_log_warning( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!\n" );
1564 self->last_position = int_position;
1567 if ( must_decode || int_position >= req_position )
1572 if ( self->vdpau->decoder == VDP_INVALID_HANDLE )
1574 vdpau_decoder_init( self );
1576 self->vdpau->is_decoded = 0;
1579 codec_context->reordered_opaque = int_position;
1580 if ( int_position >= req_position )
1581 codec_context->skip_loop_filter = AVDISCARD_NONE;
1582 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1583 ret = avcodec_decode_video2( codec_context, self->video_frame, &got_picture, &self->pkt );
1585 ret = avcodec_decode_video( codec_context, self->video_frame, &got_picture, self->pkt.data, self->pkt.size );
1587 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1590 if ( ++decode_errors <= 10 )
1601 // Get position of reordered frame
1602 int_position = self->video_frame->reordered_opaque;
1603 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(106<<8)+0))
1604 pts = best_pts( self, self->video_frame->pkt_pts, self->video_frame->pkt_dts );
1605 if ( pts != AV_NOPTS_VALUE )
1607 if ( self->first_pts != AV_NOPTS_VALUE )
1608 pts -= self->first_pts;
1609 else if ( context->start_time != AV_NOPTS_VALUE )
1610 pts -= context->start_time;
1611 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1615 if ( int_position < req_position )
1617 else if ( int_position >= req_position )
1618 codec_context->skip_loop_filter = AVDISCARD_NONE;
1620 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, self->pkt.flags & PKT_FLAG_KEY );
1623 // Now handle the picture if we have one
1626 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1628 // Workaround 1088 encodings missing cropping info.
1629 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1634 if ( self->vdpau->is_decoded )
1636 struct vdpau_render_state *render = (struct vdpau_render_state*) self->video_frame->data[0];
1638 uint32_t pitches[3];
1639 VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1641 if ( !self->vdpau->buffer )
1642 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1643 self->video_frame->data[0] = planes[0] = self->vdpau->buffer;
1644 self->video_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1645 self->video_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1646 self->video_frame->linesize[0] = pitches[0] = codec_context->width;
1647 self->video_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1648 self->video_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1650 VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1651 if ( status == VDP_STATUS_OK )
1653 convert_image( self, self->video_frame, *buffer, PIX_FMT_YUV420P,
1654 format, *width, *height, &alpha );
1658 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1659 image_size = self->vdpau->is_decoded = 0;
1664 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1665 image_size = got_picture = 0;
1670 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1671 format, *width, *height, &alpha );
1672 self->top_field_first |= self->video_frame->top_field_first;
1673 self->current_position = int_position;
1681 // Free packet data if not video and not live audio packet
1682 if ( self->pkt.stream_index != self->video_index &&
1683 !( !self->seekable && self->pkt.stream_index == self->audio_index ) )
1684 av_free_packet( &self->pkt );
1690 mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release );
1692 if ( image_size > 0 && self->image_cache )
1694 mlt_properties_set_int( frame_properties, "format", *format );
1695 mlt_cache_put_frame( self->image_cache, frame );
1698 // Try to duplicate last image if there was a decoding failure
1699 // TODO: with multithread decoding a partial frame decoding resulting
1700 // in failure also resets av_frame making test below fail.
1701 if ( !image_size && self->video_frame && self->video_frame->linesize[0] )
1704 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1706 // Workaround 1088 encodings missing cropping info.
1707 if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1710 if ( self->vdpau && self->vdpau->buffer )
1713 picture.data[0] = self->vdpau->buffer;
1714 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1715 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1716 picture.linesize[0] = codec_context->width;
1717 picture.linesize[1] = codec_context->width / 2;
1718 picture.linesize[2] = codec_context->width / 2;
1719 convert_image( self, (AVFrame*) &picture, *buffer,
1720 PIX_FMT_YUV420P, format, *width, *height, &alpha );
1724 convert_image( self, self->video_frame, *buffer, codec_context->pix_fmt,
1725 format, *width, *height, &alpha );
1730 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1731 self->video_expected = position + 1;
1735 pthread_mutex_unlock( &self->video_mutex );
1737 // Set the progressive flag
1738 if ( mlt_properties_get( properties, "force_progressive" ) )
1739 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1740 else if ( self->video_frame )
1741 mlt_properties_set_int( frame_properties, "progressive", !self->video_frame->interlaced_frame );
1743 // Set the field order property for this frame
1744 if ( mlt_properties_get( properties, "force_tff" ) )
1745 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1747 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1749 // Set immutable properties of the selected track's (or overridden) source attributes.
1750 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1751 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1752 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1753 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1755 // If we already have RGB, then the full range processing either happened already
1756 // or does not apply (RGB source).
1757 if ( *format == mlt_image_rgb24 || *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1758 mlt_properties_set( frame_properties, "force_full_luma", NULL );
1760 return !got_picture;
1763 /** Process properties as AVOptions and apply to AV context obj
1766 static void apply_properties( void *obj, mlt_properties properties, int flags )
1769 int count = mlt_properties_count( properties );
1770 for ( i = 0; i < count; i++ )
1772 const char *opt_name = mlt_properties_get_name( properties, i );
1773 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(10<<8)+0)
1774 const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, flags );
1776 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1778 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1781 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(12<<8)+0)
1782 av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1783 #elif LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1784 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1785 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1786 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1788 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1794 /** Initialize the video codec context.
1797 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1799 // Initialise the codec if necessary
1800 if ( !self->video_codec )
1802 // Get the video stream
1803 AVStream *stream = self->video_format->streams[ index ];
1805 // Get codec context
1806 AVCodecContext *codec_context = stream->codec;
1809 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1811 if ( codec_context->codec_id == AV_CODEC_ID_H264 )
1813 if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1815 if ( vdpau_init( self ) )
1817 self->video_codec = codec_context;
1818 if ( !vdpau_decoder_init( self ) )
1823 codec = avcodec_find_decoder( codec_context->codec_id );
1827 // Initialise multi-threading
1828 int thread_count = mlt_properties_get_int( properties, "threads" );
1829 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1830 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1831 if ( thread_count > 1 )
1832 codec_context->thread_count = thread_count;
1834 // If we don't have a codec and we can't initialise it, we can't do much more...
1835 pthread_mutex_lock( &self->open_mutex );
1836 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
1837 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
1839 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1842 // Now store the codec with its destructor
1843 self->video_codec = codec_context;
1847 // Remember that we can't use this later
1848 self->video_index = -1;
1849 pthread_mutex_unlock( &self->open_mutex );
1852 pthread_mutex_unlock( &self->open_mutex );
1854 // Process properties as AVOptions
1855 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1856 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(122<<8)+0)
1857 if ( codec->priv_class && codec_context->priv_data )
1858 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1861 // Reset some image properties
1862 if ( self->video_codec )
1864 mlt_properties_set_int( properties, "width", self->video_codec->width );
1865 mlt_properties_set_int( properties, "height", self->video_codec->height );
1867 // For DV, we'll just use the saved aspect ratio
1868 if ( codec_context->codec_id != AV_CODEC_ID_DVVIDEO )
1869 get_aspect_ratio( properties, stream, self->video_codec, NULL );
1871 // Start with the muxer frame rate.
1872 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1873 AVRational frame_rate = stream->avg_frame_rate;
1875 AVRational frame_rate = stream->r_frame_rate;
1877 double fps = av_q2d( frame_rate );
1879 #if LIBAVFORMAT_VERSION_MAJOR < 55
1880 // Verify and sanitize the muxer frame rate.
1881 if ( isnan( fps ) || isinf( fps ) || fps == 0 )
1883 frame_rate = stream->r_frame_rate;
1884 fps = av_q2d( frame_rate );
1887 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0) && LIBAVFORMAT_VERSION_MAJOR < 55
1888 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1889 // avg_frame_rate gives some approximate value that does not well match the media.
1890 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1891 // results in some very choppy output, but some value slightly different works
1893 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1895 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1896 fps = av_q2d( frame_rate );
1899 // XXX frame rates less than 1 fps are not considered sane
1900 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1902 // Get the frame rate from the codec.
1903 frame_rate = av_inv_q( self->video_codec->time_base );
1904 frame_rate.den *= self->video_codec->ticks_per_frame;
1905 fps = av_q2d( frame_rate );
1907 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
1909 // Use the profile frame rate if all else fails.
1910 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1911 frame_rate.num = profile->frame_rate_num;
1912 frame_rate.den = profile->frame_rate_den;
1915 self->video_time_base = stream->time_base;
1916 if ( mlt_properties_get( properties, "force_fps" ) )
1918 AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 );
1919 self->video_time_base.num *= frame_rate.num * force_fps.den;
1920 self->video_time_base.den *= frame_rate.den * force_fps.num;
1921 frame_rate = force_fps;
1923 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1924 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1926 // Set the YUV colorspace from override or detect
1927 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1928 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1929 if ( ! self->colorspace )
1931 switch ( self->video_codec->colorspace )
1933 case AVCOL_SPC_SMPTE240M:
1934 self->colorspace = 240;
1936 case AVCOL_SPC_BT470BG:
1937 case AVCOL_SPC_SMPTE170M:
1938 self->colorspace = 601;
1940 case AVCOL_SPC_BT709:
1941 self->colorspace = 709;
1944 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1945 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1950 // Let apps get chosen colorspace
1951 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1953 self->full_luma = -1;
1954 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(72<<8)+2)
1955 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range );
1956 if ( codec_context->color_range == AVCOL_RANGE_JPEG )
1957 self->full_luma = 1;
1959 if ( mlt_properties_get( properties, "set.force_full_luma" ) )
1960 self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" );
1962 return self->video_codec && self->video_index > -1;
1965 /** Set up video handling.
1968 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1971 mlt_producer producer = self->parent;
1973 // Get the properties
1974 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1976 // Fetch the video format context
1977 AVFormatContext *context = self->video_format;
1979 // Get the video_index
1980 int index = mlt_properties_get_int( properties, "video_index" );
1982 // Reopen the file if necessary
1983 if ( !context && index > -1 )
1985 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1986 mlt_properties_get( properties, "resource" ), 1 );
1987 context = self->video_format;
1990 // Exception handling for video_index
1991 if ( context && index >= (int) context->nb_streams )
1993 // Get the last video stream
1994 for ( index = context->nb_streams - 1;
1995 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1997 mlt_properties_set_int( properties, "video_index", index );
1999 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
2001 // Invalidate the video stream
2003 mlt_properties_set_int( properties, "video_index", index );
2006 // Update the video properties if the index changed
2007 if ( index != self->video_index )
2009 // Reset the video properties if the index changed
2010 self->video_index = index;
2011 pthread_mutex_lock( &self->open_mutex );
2012 if ( self->video_codec )
2013 avcodec_close( self->video_codec );
2014 self->video_codec = NULL;
2015 pthread_mutex_unlock( &self->open_mutex );
2018 // Get the frame properties
2019 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2022 if ( context && index > -1 && video_codec_init( self, index, properties ) )
2024 // Set the frame properties
2025 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
2026 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
2027 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
2029 // Set the width and height
2030 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
2031 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
2032 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
2033 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
2034 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
2035 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
2037 // Workaround 1088 encodings missing cropping info.
2038 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
2040 mlt_properties_set_int( properties, "meta.media.height", 1080 );
2043 // Add our image operation
2044 mlt_frame_push_service( frame, self );
2045 mlt_frame_push_get_image( frame, producer_get_image );
2049 // If something failed, use test card image
2050 mlt_properties_set_int( frame_properties, "test_image", 1 );
2054 static int seek_audio( producer_avformat self, mlt_position position, double timecode )
2058 // Seek if necessary
2059 if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) )
2061 if ( self->last_position == POSITION_INITIAL )
2063 int video_index = self->video_index;
2064 if ( video_index == -1 )
2065 video_index = first_video_index( self );
2066 if ( video_index >= 0 )
2067 find_first_pts( self, video_index );
2070 if ( position + 1 == self->audio_expected )
2072 // We're paused - silence required
2075 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
2077 AVFormatContext *context = self->audio_format;
2078 int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
2079 if ( context->start_time != AV_NOPTS_VALUE )
2080 timestamp += context->start_time;
2081 if ( timestamp < 0 )
2084 // Set to the real timecode
2085 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
2088 // Clear the usage in the audio buffer
2089 int i = MAX_AUDIO_STREAMS + 1;
2091 self->audio_used[i - 1] = 0;
2097 static int sample_bytes( AVCodecContext *context )
2099 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(8<<8)+0)
2100 return av_get_bytes_per_sample( context->sample_fmt );
2101 #elif LIBAVCODEC_VERSION_MAJOR >= 53
2102 return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
2104 return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
2108 static void planar_to_interleaved( uint8_t *dest, uint8_t *src, int samples, int channels, int bytes_per_sample )
2111 for ( s = 0; s < samples; s++ )
2113 for ( c = 0; c < channels; c++ )
2115 memcpy( dest, src + ( c * samples + s ) * bytes_per_sample, bytes_per_sample );
2116 dest += bytes_per_sample;
2121 static void planar_to_interleaved2( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample )
2124 for ( s = 0; s < samples; s++ )
2126 for ( c = 0; c < channels; c++ )
2128 memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample );
2129 dest += bytes_per_sample;
2134 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
2136 // Fetch the audio_format
2137 AVFormatContext *context = self->audio_format;
2139 // Get the current stream index
2140 int index = pkt.stream_index;
2142 // Get codec context
2143 AVCodecContext *codec_context = self->audio_codec[ index ];
2145 // Obtain the audio buffers
2146 uint8_t *audio_buffer = self->audio_buffer[ index ];
2147 uint8_t *decode_buffer = self->decode_buffer[ index ];
2149 int audio_used = self->audio_used[ index ];
2150 uint8_t *ptr = pkt.data;
2154 while ( ptr && ret >= 0 && len > 0 )
2156 int sizeof_sample = sample_bytes( codec_context );
2157 int data_size = self->audio_buffer_size[ index ];
2160 #if LIBAVCODEC_VERSION_MAJOR >= 55
2161 if ( !self->audio_frame )
2162 self->audio_frame = avcodec_alloc_frame();
2164 avcodec_get_frame_defaults( self->audio_frame );
2165 ret = avcodec_decode_audio4( codec_context, self->audio_frame, &data_size, &pkt );
2167 data_size = av_samples_get_buffer_size( NULL, codec_context->channels,
2168 self->audio_frame->nb_samples, codec_context->sample_fmt, 1 );
2169 decode_buffer = self->audio_frame->data[0];
2172 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2176 mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2180 pkt.size = len -= ret;
2181 pkt.data = ptr += ret;
2183 // If decoded successfully
2184 if ( data_size > 0 )
2186 // Figure out how many samples will be needed after resampling
2187 int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2189 // Resize audio buffer to prevent overflow
2190 if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2192 self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 2 ) * channels * sizeof_sample;
2193 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2195 uint8_t *dest = &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ];
2196 switch ( codec_context->sample_fmt )
2198 #if LIBAVUTIL_VERSION_INT >= ((51<<16)+(17<<8)+0)
2199 case AV_SAMPLE_FMT_U8P:
2200 case AV_SAMPLE_FMT_S16P:
2201 case AV_SAMPLE_FMT_S32P:
2202 case AV_SAMPLE_FMT_FLTP:
2203 #if LIBAVCODEC_VERSION_MAJOR >= 55
2204 planar_to_interleaved2( dest, self->audio_frame, convert_samples, codec_context->channels, sizeof_sample );
2206 planar_to_interleaved( dest, decode_buffer, convert_samples, codec_context->channels, sizeof_sample );
2211 // Straight copy to audio buffer
2212 memcpy( dest, decode_buffer, data_size );
2214 audio_used += convert_samples;
2217 while ( *ignore && audio_used )
2220 audio_used -= audio_used > samples ? samples : audio_used;
2221 memmove( audio_buffer, &audio_buffer[ samples * codec_context->channels * sizeof_sample ],
2222 audio_used * sizeof_sample );
2227 // If we're behind, ignore this packet
2228 // Skip this on non-seekable, audio-only inputs.
2229 if ( pkt.pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2231 int64_t pts = pkt.pts;
2232 if ( self->first_pts != AV_NOPTS_VALUE )
2233 pts -= self->first_pts;
2234 else if ( context->start_time != AV_NOPTS_VALUE )
2235 pts -= context->start_time;
2236 double timebase = av_q2d( context->streams[ index ]->time_base );
2237 int64_t int_position = ( int64_t )( timebase * pts * fps + 0.5 );
2238 int64_t req_position = ( int64_t )( timecode * fps + 0.5 );
2240 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
2241 "A pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
2242 pkt.pts, pkt.dts, req_position, self->current_position, int_position );
2244 if ( int_position > 0 )
2246 if ( int_position < req_position )
2247 // We are behind, so skip some
2248 *ignore = req_position - int_position;
2249 else if ( self->audio_index != INT_MAX && int_position > req_position + 2 )
2250 // We are ahead, so seek backwards some more
2251 seek_audio( self, req_position, timecode - 1.0 );
2253 // Cancel the find_first_pts() in seek_audio()
2254 if ( self->video_index == -1 && self->last_position == POSITION_INITIAL )
2255 self->last_position = int_position;
2258 self->audio_used[ index ] = audio_used;
2263 /** Get the audio from a frame.
2265 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2268 producer_avformat self = mlt_frame_pop_audio( frame );
2270 pthread_mutex_lock( &self->audio_mutex );
2272 // Obtain the frame number of this frame
2273 mlt_position position = mlt_frame_original_position( frame );
2275 // Calculate the real time code
2276 double real_timecode = producer_time_of_frame( self->parent, position );
2278 // Get the producer fps
2279 double fps = mlt_producer_get_fps( self->parent );
2281 // Number of frames to ignore (for ffwd)
2282 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2284 // Flag for paused (silence)
2285 int paused = seek_audio( self, position, real_timecode );
2287 // Initialize ignore for all streams from the seek return value
2288 int i = MAX_AUDIO_STREAMS;
2290 ignore[i] = ignore[0];
2292 // Fetch the audio_format
2293 AVFormatContext *context = self->audio_format;
2295 int sizeof_sample = sizeof( int16_t );
2297 // Determine the tracks to use
2298 int index = self->audio_index;
2299 int index_max = self->audio_index + 1;
2300 if ( self->audio_index == INT_MAX )
2303 index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
2304 *channels = self->total_channels;
2305 *samples = mlt_sample_calculator( fps, FFMAX( self->max_frequency, *frequency ), position );
2306 *frequency = FFMAX( self->max_frequency, *frequency );
2309 // Initialize the buffers
2310 for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ )
2312 // Get codec context
2313 AVCodecContext *codec_context = self->audio_codec[ index ];
2315 if ( codec_context && !self->audio_buffer[ index ] )
2317 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2318 sizeof_sample = sample_bytes( codec_context );
2320 // Check for audio buffer and create if necessary
2321 self->audio_buffer_size[ index ] = MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2322 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2324 // Check for decoder buffer and create if necessary
2325 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2329 // Get the audio if required
2330 if ( !paused && *frequency > 0 )
2336 av_init_packet( &pkt );
2338 // Caller requested number samples based on requested sample rate.
2339 if ( self->audio_index != INT_MAX )
2340 *samples = mlt_sample_calculator( fps, self->audio_codec[ self->audio_index ]->sample_rate, position );
2342 while ( ret >= 0 && !got_audio )
2344 // Check if the buffer already contains the samples required
2345 if ( self->audio_index != INT_MAX &&
2346 self->audio_used[ self->audio_index ] >= *samples &&
2347 ignore[ self->audio_index ] == 0 )
2352 else if ( self->audio_index == INT_MAX )
2354 // Check if there is enough audio for all streams
2356 for ( index = 0; got_audio && index < index_max; index++ )
2357 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2364 pthread_mutex_lock( &self->packets_mutex );
2365 if ( mlt_deque_count( self->apackets ) )
2367 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2373 ret = av_read_frame( context, &pkt );
2374 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2376 if ( !av_dup_packet( &pkt ) )
2378 AVPacket *tmp = malloc( sizeof(AVPacket) );
2380 mlt_deque_push_back( self->vpackets, tmp );
2384 pthread_mutex_unlock( &self->packets_mutex );
2386 // We only deal with audio from the selected audio index
2387 index = pkt.stream_index;
2388 if ( index < MAX_AUDIO_STREAMS && ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2389 ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2391 int channels2 = self->audio_codec[index]->channels;
2392 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2395 if ( self->seekable || index != self->video_index )
2396 av_free_packet( &pkt );
2400 // Set some additional return values
2401 *format = mlt_audio_s16;
2402 if ( self->audio_index != INT_MAX )
2404 index = self->audio_index;
2405 *channels = self->audio_codec[ index ]->channels;
2406 *frequency = self->audio_codec[ index ]->sample_rate;
2407 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2408 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2410 else if ( self->audio_index == INT_MAX )
2412 for ( index = 0; index < index_max; index++ )
2413 if ( self->audio_codec[ index ] )
2415 // XXX: This only works if all audio tracks have the same sample format.
2416 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2417 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2422 // Allocate and set the frame's audio buffer
2423 int size = mlt_audio_format_size( *format, *samples, *channels );
2424 *buffer = mlt_pool_alloc( size );
2425 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2427 // Interleave tracks if audio_index=all
2428 if ( self->audio_index == INT_MAX )
2430 uint8_t *dest = *buffer;
2432 for ( i = 0; i < *samples; i++ )
2434 for ( index = 0; index < index_max; index++ )
2435 if ( self->audio_codec[ index ] )
2437 int current_channels = self->audio_codec[ index ]->channels;
2438 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2439 memcpy( dest, src, current_channels * sizeof_sample );
2440 dest += current_channels * sizeof_sample;
2443 for ( index = 0; index < index_max; index++ )
2444 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2446 int current_channels = self->audio_codec[ index ]->channels;
2447 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2448 self->audio_used[index] -= *samples;
2449 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2452 // Copy a single track to the output buffer
2455 index = self->audio_index;
2457 // Now handle the audio if we have enough
2458 if ( self->audio_used[ index ] > 0 )
2460 uint8_t *src = self->audio_buffer[ index ];
2461 // copy samples from audio_buffer
2462 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2463 memcpy( *buffer, src, size * *channels * sizeof_sample );
2464 // supply the remaining requested samples as silence
2465 if ( *samples > self->audio_used[ index ] )
2466 memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2467 // reposition the samples within audio_buffer
2468 self->audio_used[ index ] -= size;
2469 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2473 // Otherwise fill with silence
2474 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2480 // Get silence and don't touch the context
2481 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2484 // Regardless of speed (other than paused), we expect to get the next frame
2486 self->audio_expected = position + 1;
2488 pthread_mutex_unlock( &self->audio_mutex );
2493 /** Initialize the audio codec context.
2496 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2498 // Initialise the codec if necessary
2499 if ( !self->audio_codec[ index ] )
2501 // Get codec context
2502 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2505 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2507 // If we don't have a codec and we can't initialise it, we can't do much more...
2508 pthread_mutex_lock( &self->open_mutex );
2509 #if LIBAVCODEC_VERSION_INT >= ((53<<16)+(8<<8)+0)
2510 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2512 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2515 // Now store the codec with its destructor
2516 if ( self->audio_codec[ index ] )
2517 avcodec_close( self->audio_codec[ index ] );
2518 self->audio_codec[ index ] = codec_context;
2522 // Remember that we can't use self later
2523 self->audio_index = -1;
2525 pthread_mutex_unlock( &self->open_mutex );
2527 // Process properties as AVOptions
2528 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2529 if ( codec && codec->priv_class && codec_context->priv_data )
2530 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2532 return self->audio_codec[ index ] && self->audio_index > -1;
2535 /** Set up audio handling.
2538 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2541 mlt_producer producer = self->parent;
2543 // Get the properties
2544 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2546 // Fetch the audio format context
2547 AVFormatContext *context = self->audio_format;
2549 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2551 // Get the audio_index
2552 int index = mlt_properties_get_int( properties, "audio_index" );
2554 // Handle all audio tracks
2555 if ( self->audio_index > -1 &&
2556 mlt_properties_get( properties, "audio_index" ) &&
2557 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2560 // Reopen the file if necessary
2561 if ( !context && self->audio_index > -1 && index > -1 )
2563 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2564 mlt_properties_get( properties, "resource" ), 1 );
2565 context = self->audio_format;
2568 // Exception handling for audio_index
2569 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2571 for ( index = context->nb_streams - 1;
2572 index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2574 mlt_properties_set_int( properties, "audio_index", index );
2576 if ( context && index > -1 && index < INT_MAX &&
2577 context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2579 index = self->audio_index;
2580 mlt_properties_set_int( properties, "audio_index", index );
2582 if ( context && index > -1 && index < INT_MAX &&
2583 pick_audio_format( context->streams[ index ]->codec->sample_fmt ) == mlt_audio_none )
2588 // Update the audio properties if the index changed
2589 if ( context && index > -1 && index != self->audio_index )
2591 pthread_mutex_lock( &self->open_mutex );
2592 if ( self->audio_codec[ self->audio_index ] )
2593 avcodec_close( self->audio_codec[ self->audio_index ] );
2594 self->audio_codec[ self->audio_index ] = NULL;
2595 pthread_mutex_unlock( &self->open_mutex );
2597 if ( self->audio_index != -1 )
2598 self->audio_index = index;
2603 if ( context && index == INT_MAX )
2605 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2606 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2607 for ( index = 0; index < context->nb_streams; index++ )
2609 if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2610 audio_codec_init( self, index, properties );
2613 else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2615 // Set the frame properties
2616 if ( index < MAX_AUDIO_STREAMS )
2618 mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate );
2619 mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels );
2622 if ( context && index > -1 )
2624 // Add our audio operation
2625 mlt_frame_push_audio( frame, self );
2626 mlt_frame_push_audio( frame, producer_get_audio );
2630 /** Our get frame implementation.
2633 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2635 // Access the private data
2636 mlt_service service = MLT_PRODUCER_SERVICE( producer );
2637 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2638 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2643 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2644 producer->child = self;
2645 self->parent = producer;
2646 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2647 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2650 // Create an empty frame
2651 *frame = mlt_frame_init( service);
2655 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2659 mlt_cache_item_close( cache_item );
2663 // Update timecode on the frame we're creating
2664 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2667 producer_set_up_video( self, *frame );
2670 producer_set_up_audio( self, *frame );
2672 // Set the position of this producer
2673 mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2674 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position );
2676 // Calculate the next timecode
2677 mlt_producer_prepare_next( producer );
2682 static void producer_avformat_close( producer_avformat self )
2684 mlt_log_debug( NULL, "producer_avformat_close\n" );
2686 // Cleanup av contexts
2687 av_free_packet( &self->pkt );
2688 av_free( self->video_frame );
2689 av_free( self->audio_frame );
2690 pthread_mutex_lock( &self->open_mutex );
2692 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2694 mlt_pool_release( self->audio_buffer[i] );
2695 av_free( self->decode_buffer[i] );
2696 if ( self->audio_codec[i] )
2697 avcodec_close( self->audio_codec[i] );
2698 self->audio_codec[i] = NULL;
2700 if ( self->video_codec )
2701 avcodec_close( self->video_codec );
2702 self->video_codec = NULL;
2704 #if LIBAVFORMAT_VERSION_INT >= ((53<<16)+(17<<8)+0)
2705 if ( self->dummy_context )
2706 avformat_close_input( &self->dummy_context );
2707 if ( self->seekable && self->audio_format )
2708 avformat_close_input( &self->audio_format );
2709 if ( self->video_format )
2710 avformat_close_input( &self->video_format );
2712 if ( self->dummy_context )
2713 av_close_input_file( self->dummy_context );
2714 if ( self->seekable && self->audio_format )
2715 av_close_input_file( self->audio_format );
2716 if ( self->video_format )
2717 av_close_input_file( self->video_format );
2719 pthread_mutex_unlock( &self->open_mutex );
2721 vdpau_producer_close( self );
2723 if ( self->image_cache )
2724 mlt_cache_close( self->image_cache );
2726 // Cleanup the mutexes
2727 if ( self->is_mutex_init )
2729 pthread_mutex_destroy( &self->audio_mutex );
2730 pthread_mutex_destroy( &self->video_mutex );
2731 pthread_mutex_destroy( &self->packets_mutex );
2732 pthread_mutex_destroy( &self->open_mutex );
2735 // Cleanup the packet queues
2737 if ( self->apackets )
2739 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2741 av_free_packet( pkt );
2744 mlt_deque_close( self->apackets );
2745 self->apackets = NULL;
2747 if ( self->vpackets )
2749 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2751 av_free_packet( pkt );
2754 mlt_deque_close( self->vpackets );
2755 self->vpackets = NULL;
2761 static void producer_close( mlt_producer parent )
2763 // Remove this instance from the cache
2764 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2767 parent->close = NULL;
2768 mlt_producer_close( parent );