git.sesse.net Git - mlt/blob - src/modules/avformat/producer_avformat.c

   1 /*
   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
   7  *
   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.
  12  *
  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.
  17  *
  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
  21  */
  22
  23 // MLT Header files
  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>
  31
  32 // ffmpeg Header files
  33 #include <libavformat/avformat.h>
  34 #ifdef SWSCALE
  35 #  include <libswscale/swscale.h>
  36 #endif
  37 #if LIBAVCODEC_VERSION_MAJOR > 52
  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);
  41 #endif
  42 #ifdef VDPAU
  43 #  include <libavcodec/vdpau.h>
  44 #endif
  45 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
  46 #  include <libavutil/pixdesc.h>
  47 #endif
  48
  49 // System header files
  50 #include <stdlib.h>
  51 #include <string.h>
  52 #include <pthread.h>
  53 #include <limits.h>
  54
  55 #if LIBAVUTIL_VERSION_INT < (50<<16)
  56 #define PIX_FMT_RGB32 PIX_FMT_RGBA32
  57 #define PIX_FMT_YUYV422 PIX_FMT_YUV422
  58 #endif
  59
  60 #if LIBAVCODEC_VERSION_MAJOR > 52
  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
  65 #else
  66 #include <libavcodec/opt.h>
  67 #endif
  68
  69 #define POSITION_INITIAL (-2)
  70 #define POSITION_INVALID (-1)
  71
  72 #define MAX_AUDIO_STREAMS (10)
  73 #define MAX_VDPAU_SURFACES (10)
  74
  75 void avformat_lock( );
  76 void avformat_unlock( );
  77
  78 struct producer_avformat_s
  79 {
  80         mlt_producer parent;
  81         AVFormatContext *dummy_context;
  82         AVFormatContext *audio_format;
  83         AVFormatContext *video_format;
  84         AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
  85         AVCodecContext *video_codec;
  86         AVFrame *av_frame;
  87         ReSampleContext *audio_resample[ MAX_AUDIO_STREAMS ];
  88         mlt_position audio_expected;
  89         mlt_position video_expected;
  90         int audio_index;
  91         int video_index;
  92         double start_time;
  93         int first_pts;
  94         int64_t last_position;
  95         int seekable;
  96         int64_t current_position;
  97         mlt_position nonseek_position;
  98         int got_picture;
  99         int top_field_first;
 100         uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
 101         size_t audio_buffer_size[ MAX_AUDIO_STREAMS ];
 102         uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
 103         int audio_used[ MAX_AUDIO_STREAMS ];
 104         int audio_streams;
 105         int audio_max_stream;
 106         int total_channels;
 107         int max_channel;
 108         int max_frequency;
 109         unsigned int invalid_pts_counter;
 110         double resample_factor;
 111         mlt_cache image_cache;
 112         int colorspace;
 113         pthread_mutex_t video_mutex;
 114         pthread_mutex_t audio_mutex;
 115         mlt_deque apackets;
 116         mlt_deque vpackets;
 117         pthread_mutex_t packets_mutex;
 118 #ifdef VDPAU
 119         struct
 120         {
 121                 // from FFmpeg
 122                 struct vdpau_render_state render_states[MAX_VDPAU_SURFACES];
 123
 124                 // internal
 125                 mlt_deque deque;
 126                 int b_age;
 127                 int ip_age[2];
 128                 int is_decoded;
 129                 uint8_t *buffer;
 130         } *vdpau;
 131 #endif
 132 };
 133 typedef struct producer_avformat_s *producer_avformat;
 134
 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
 147 #ifdef VDPAU
 148 #include "vdpau.c"
 149 #endif
 150
 151 /** Constructor for libavformat.
 152 */
 153
 154 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
 155 {
 156         if ( list_components( file ) )
 157                 return NULL;
 158
 159         mlt_producer producer = NULL;
 160
 161         // Check that we have a non-NULL argument
 162         if ( file )
 163         {
 164                 // Construct the producer
 165                 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
 166                 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
 167
 168                 // Initialise it
 169                 if ( mlt_producer_init( producer, self ) == 0 )
 170                 {
 171                         self->parent = producer;
 172
 173                         // Get the properties
 174                         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
 175
 176                         // Set the resource property (required for all producers)
 177                         mlt_properties_set( properties, "resource", file );
 178
 179                         // Register transport implementation with the producer
 180                         producer->close = (mlt_destructor) producer_close;
 181
 182                         // Register our get_frame implementation
 183                         producer->get_frame = producer_get_frame;
 184
 185                         if ( strcmp( service, "avformat-novalidate" ) )
 186                         {
 187                                 // Open the file
 188                                 if ( producer_open( self, profile, file, 1 ) != 0 )
 189                                 {
 190                                         // Clean up
 191                                         mlt_producer_close( producer );
 192                                         producer = NULL;
 193                                 }
 194                                 else if ( self->seekable )
 195                                 {
 196                                         // Close the file to release resources for large playlists - reopen later as needed
 197                                         avformat_lock();
 198                                         if ( self->audio_format )
 199                                                 av_close_input_file( self->audio_format );
 200                                         self->audio_format = NULL;
 201                                         if ( self->video_format )
 202                                                 av_close_input_file( self->video_format );
 203                                         self->video_format = NULL;
 204                                         avformat_unlock();
 205                                 }
 206                         }
 207                         if ( producer )
 208                         {
 209                                 // Default the user-selectable indices from the auto-detected indices
 210                                 mlt_properties_set_int( properties, "audio_index",  self->audio_index );
 211                                 mlt_properties_set_int( properties, "video_index",  self->video_index );
 212 #ifdef VDPAU
 213                                 mlt_service_cache_set_size( MLT_PRODUCER_SERVICE(producer), "producer_avformat", 5 );
 214 #endif
 215                                 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
 216                         }
 217                 }
 218         }
 219         return producer;
 220 }
 221
 222 int list_components( char* file )
 223 {
 224         int skip = 0;
 225
 226         // Report information about available demuxers and codecs as YAML Tiny
 227         if ( file && strstr( file, "f-list" ) )
 228         {
 229                 fprintf( stderr, "---\nformats:\n" );
 230                 AVInputFormat *format = NULL;
 231                 while ( ( format = av_iformat_next( format ) ) )
 232                         fprintf( stderr, "  - %s\n", format->name );
 233                 fprintf( stderr, "...\n" );
 234                 skip = 1;
 235         }
 236         if ( file && strstr( file, "acodec-list" ) )
 237         {
 238                 fprintf( stderr, "---\naudio_codecs:\n" );
 239                 AVCodec *codec = NULL;
 240                 while ( ( codec = av_codec_next( codec ) ) )
 241                         if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
 242                                 fprintf( stderr, "  - %s\n", codec->name );
 243                 fprintf( stderr, "...\n" );
 244                 skip = 1;
 245         }
 246         if ( file && strstr( file, "vcodec-list" ) )
 247         {
 248                 fprintf( stderr, "---\nvideo_codecs:\n" );
 249                 AVCodec *codec = NULL;
 250                 while ( ( codec = av_codec_next( codec ) ) )
 251                         if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
 252                                 fprintf( stderr, "  - %s\n", codec->name );
 253                 fprintf( stderr, "...\n" );
 254                 skip = 1;
 255         }
 256
 257         return skip;
 258 }
 259
 260 /** Find the default streams.
 261 */
 262
 263 static mlt_properties find_default_streams( producer_avformat self )
 264 {
 265         int i;
 266         char key[200];
 267         AVMetadataTag *tag = NULL;
 268         AVFormatContext *context = self->video_format;
 269         mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
 270
 271         // Default to the first audio and video streams found
 272         self->audio_index = -1;
 273         self->video_index = -1;
 274
 275         mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
 276
 277         // Allow for multiple audio and video streams in the file and select first of each (if available)
 278         for( i = 0; i < context->nb_streams; i++ )
 279         {
 280                 // Get the codec context
 281                 AVStream *stream = context->streams[ i ];
 282                 if ( ! stream ) continue;
 283                 AVCodecContext *codec_context = stream->codec;
 284                 if ( ! codec_context ) continue;
 285                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
 286                 if ( ! codec ) continue;
 287
 288                 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
 289
 290                 // Determine the type and obtain the first index of each type
 291                 switch( codec_context->codec_type )
 292                 {
 293                         case CODEC_TYPE_VIDEO:
 294                                 // Use first video stream
 295                                 if ( self->video_index < 0 )
 296                                         self->video_index = i;
 297                                 mlt_properties_set( meta_media, key, "video" );
 298                                 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
 299 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
 300                                 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
 301                                 if ( isnan( ffmpeg_fps ) || ffmpeg_fps == 0 )
 302                                         ffmpeg_fps = av_q2d( context->streams[ i ]->r_frame_rate );
 303                                 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
 304 #else
 305                                 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) );
 306 #endif
 307
 308 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 309                                 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
 310                                 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
 311 #endif
 312                                 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
 313                                 mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den /
 314                                                                                    ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) );
 315                                 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
 316                                 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
 317                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
 318                                 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
 319 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
 320                                 snprintf( key, sizeof(key), "meta.media.%d.codec.colorspace", i );
 321                                 switch ( codec_context->colorspace )
 322                                 {
 323                                 case AVCOL_SPC_SMPTE240M:
 324                                         mlt_properties_set_int( meta_media, key, 240 );
 325                                         break;
 326                                 case AVCOL_SPC_BT470BG:
 327                                 case AVCOL_SPC_SMPTE170M:
 328                                         mlt_properties_set_int( meta_media, key, 601 );
 329                                         break;
 330                                 case AVCOL_SPC_BT709:
 331                                         mlt_properties_set_int( meta_media, key, 709 );
 332                                         break;
 333                                 default:
 334                                         // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
 335                                         mlt_properties_set_int( meta_media, key, codec_context->width * codec_context->height > 750000 ? 709 : 601 );
 336                                         break;
 337                                 }
 338 #endif
 339                                 break;
 340                         case CODEC_TYPE_AUDIO:
 341                                 // Use first audio stream
 342                                 if ( self->audio_index < 0 )
 343                                         self->audio_index = i;
 344                                 mlt_properties_set( meta_media, key, "audio" );
 345 #if LIBAVCODEC_VERSION_MAJOR > 52
 346                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
 347                                 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_context->sample_fmt ) );
 348 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
 349                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
 350                                 mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) );
 351 #endif
 352                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
 353                                 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
 354                                 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
 355                                 mlt_properties_set_int( meta_media, key, codec_context->channels );
 356                                 break;
 357                         default:
 358                                 break;
 359                 }
 360 //              snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
 361 //              mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
 362                 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
 363                 mlt_properties_set( meta_media, key, codec->name );
 364 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0))
 365                 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
 366                 mlt_properties_set( meta_media, key, codec->long_name );
 367 #endif
 368                 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
 369                 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
 370 //              snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
 371 //              mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
 372 //              snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
 373 //              mlt_properties_set_int( meta_media, key, codec_context->profile );
 374 //              snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
 375 //              mlt_properties_set_int( meta_media, key, codec_context->level );
 376
 377                 // Read Metadata
 378 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
 379                 while ( ( tag = av_metadata_get( stream->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
 380                 {
 381                         if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
 382                         {
 383                                 snprintf( key, sizeof(key), "meta.attr.%d.stream.%s.markup", i, tag->key );
 384                                 mlt_properties_set( meta_media, key, tag->value );
 385                         }
 386                 }
 387 #endif
 388         }
 389 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(31<<8)+0)
 390         while ( ( tag = av_metadata_get( context->metadata, "", tag, AV_METADATA_IGNORE_SUFFIX ) ) )
 391         {
 392                 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
 393                 {
 394                         snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
 395                         mlt_properties_set( meta_media, key, tag->value );
 396                 }
 397         }
 398 #else
 399         if ( context->title && strcmp( context->title, "" ) )
 400                 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
 401         if ( context->author && strcmp( context->author, "" ) )
 402                 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
 403         if ( context->copyright && strcmp( context->copyright, "" ) )
 404                 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
 405         if ( context->comment )
 406                 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
 407         if ( context->album )
 408                 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
 409         if ( context->year )
 410                 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
 411         if ( context->track )
 412                 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
 413 #endif
 414
 415         return meta_media;
 416 }
 417
 418 static inline int dv_is_pal( AVPacket *pkt )
 419 {
 420         return pkt->data[3] & 0x80;
 421 }
 422
 423 static int dv_is_wide( AVPacket *pkt )
 424 {
 425         int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
 426
 427         for ( ; i < pkt->size; i += 5 /* packet size */ )
 428         {
 429                 if ( pkt->data[ i ] == 0x61 )
 430                 {
 431                         uint8_t x = pkt->data[ i + 2 ] & 0x7;
 432                         return ( x == 2 ) || ( x == 7 );
 433                 }
 434         }
 435         return 0;
 436 }
 437
 438 static double get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
 439 {
 440         double aspect_ratio = 1.0;
 441
 442         if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
 443         {
 444                 if ( pkt )
 445                 {
 446                         if ( dv_is_pal( pkt ) )
 447                         {
 448                                 if ( dv_is_wide( pkt ) )
 449                                 {
 450                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
 451                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
 452                                 }
 453                                 else
 454                                 {
 455                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
 456                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
 457                                 }
 458                         }
 459                         else
 460                         {
 461                                 if ( dv_is_wide( pkt ) )
 462                                 {
 463                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
 464                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
 465                                 }
 466                                 else
 467                                 {
 468                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
 469                                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
 470                                 }
 471                         }
 472                 }
 473                 else
 474                 {
 475                         AVRational ar =
 476 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 477                                 stream->sample_aspect_ratio;
 478 #else
 479                                 codec_context->sample_aspect_ratio;
 480 #endif
 481                         // Override FFmpeg's notion of DV aspect ratios, which are
 482                         // based upon a width of 704. Since we do not have a normaliser
 483                         // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
 484                         // we just coerce the values to facilitate a passive behaviour through
 485                         // the rescale normaliser when using equivalent producers and consumers.
 486                         // = display_aspect / (width * height)
 487                         if ( ar.num == 10 && ar.den == 11 )
 488                         {
 489                                 // 4:3 NTSC
 490                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 8 );
 491                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 9 );
 492                         }
 493                         else if ( ar.num == 59 && ar.den == 54 )
 494                         {
 495                                 // 4:3 PAL
 496                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 16 );
 497                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 15 );
 498                         }
 499                         else if ( ar.num == 40 && ar.den == 33 )
 500                         {
 501                                 // 16:9 NTSC
 502                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 32 );
 503                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 27 );
 504                         }
 505                         else if ( ar.num == 118 && ar.den == 81 )
 506                         {
 507                                 // 16:9 PAL
 508                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 64 );
 509                                 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 45 );
 510                         }
 511                 }
 512         }
 513         else
 514         {
 515                 AVRational codec_sar = codec_context->sample_aspect_ratio;
 516                 AVRational stream_sar =
 517 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 518                         stream->sample_aspect_ratio;
 519 #else
 520                         { 0, 1 };
 521 #endif
 522                 if ( codec_sar.num > 0 )
 523                 {
 524                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", codec_sar.num );
 525                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", codec_sar.den );
 526                 }
 527                 else if ( stream_sar.num > 0 )
 528                 {
 529                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", stream_sar.num );
 530                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", stream_sar.den );
 531                 }
 532                 else
 533                 {
 534                         mlt_properties_set_int( properties, "meta.media.sample_aspect_num", 1 );
 535                         mlt_properties_set_int( properties, "meta.media.sample_aspect_den", 1 );
 536                 }
 537         }
 538         AVRational ar = { mlt_properties_get_double( properties, "meta.media.sample_aspect_num" ), mlt_properties_get_double( properties, "meta.media.sample_aspect_den" ) };
 539         aspect_ratio = av_q2d( ar );
 540         mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
 541
 542         return aspect_ratio;
 543 }
 544
 545 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVFormatParameters *params )
 546 {
 547         if ( !URL ) return NULL;
 548
 549         const char *result = URL;
 550         char *protocol = strdup( URL );
 551         char *url = strchr( protocol, ':' );
 552
 553         // Only if there is not a protocol specification that avformat can handle
 554 #if LIBAVFORMAT_VERSION_MAJOR > 52
 555         if ( url && avio_check( URL, 0 ) < 0 )
 556 #else
 557         if ( url && !url_exist( URL ) )
 558 #endif
 559         {
 560                 // Truncate protocol string
 561                 url[0] = 0;
 562                 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url + 1 );
 563
 564                 // Lookup the format
 565                 *format = av_find_input_format( protocol );
 566
 567                 // Eat the format designator
 568                 result = ++url;
 569
 570                 if ( *format )
 571                 {
 572                         // These are required by video4linux2 (defaults)
 573                         params->width = profile->width;
 574                         params->height = profile->height;
 575                         if ( !strstr( URL, "&frame_rate" ) )
 576                                 params->time_base = (AVRational){ profile->frame_rate_den, profile->frame_rate_num };
 577                         params->channels = 2;
 578                         params->sample_rate = 48000;
 579
 580                         // Parse out params
 581                         url = strchr( url, '?' );
 582                         while ( url )
 583                         {
 584                                 url[0] = 0;
 585                                 char *name = strdup( ++url );
 586                                 char *value = strchr( name, ':' );
 587                                 if ( value )
 588                                 {
 589                                         value[0] = 0;
 590                                         value++;
 591                                         char *t = strchr( value, '&' );
 592                                         if ( t )
 593                                                 t[0] = 0;
 594                                         if ( !strcmp( name, "frame_rate" ) )
 595                                                 params->time_base.den = atoi( value );
 596                                         else if ( !strcmp( name, "frame_rate_base" ) )
 597                                                 params->time_base.num = atoi( value );
 598                                         else if ( !strcmp( name, "sample_rate" ) )
 599                                                 params->sample_rate = atoi( value );
 600                                         else if ( !strcmp( name, "channel" ) )
 601                                                 params->channel = atoi( value );
 602                                         else if ( !strcmp( name, "channels" ) )
 603                                                 params->channels = atoi( value );
 604 #if (LIBAVUTIL_VERSION_INT > ((50<<16)+(7<<8)+0))
 605                                         else if ( !strcmp( name, "pix_fmt" ) )
 606                                                 params->pix_fmt = av_get_pix_fmt( value );
 607 #endif
 608                                         else if ( !strcmp( name, "width" ) )
 609                                                 params->width = atoi( value );
 610                                         else if ( !strcmp( name, "height" ) )
 611                                                 params->height = atoi( value );
 612                                         else if ( !strcmp( name, "standard" ) )
 613                                                 params->standard = strdup( value );
 614                                 }
 615                                 free( name );
 616                                 url = strchr( url, '&' );
 617                         }
 618                 }
 619         }
 620         result = strdup( result );
 621         free( protocol );
 622         return result;
 623 }
 624
 625 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
 626 {
 627         int error = 0;
 628
 629         // Get the properties
 630         mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
 631
 632         AVFormatContext *format = self->video_format;
 633
 634         // We will treat everything with the producer fps.
 635         // TODO: make this more flexible.
 636         double fps = mlt_profile_fps( profile );
 637
 638         // Get the duration
 639         if ( !mlt_properties_get_int( properties, "_length_computed" ) )
 640         {
 641                 // The _length_computed flag prevents overwriting explicity set length/out/eof properties
 642                 // when producer_open is called after initial call when restoring or reseting the producer.
 643                 if ( format->duration != AV_NOPTS_VALUE )
 644                 {
 645                         // This isn't going to be accurate for all formats
 646                         mlt_position frames = ( mlt_position )( ( ( double )format->duration / ( double )AV_TIME_BASE ) * fps );
 647                         mlt_properties_set_position( properties, "out", frames - 1 );
 648                         mlt_properties_set_position( properties, "length", frames );
 649                         mlt_properties_set_int( properties, "_length_computed", 1 );
 650                 }
 651                 else
 652                 {
 653                         // Set live sources to run forever
 654                         mlt_properties_set_position( properties, "length", INT_MAX );
 655                         mlt_properties_set_position( properties, "out", INT_MAX - 1 );
 656                         mlt_properties_set( properties, "eof", "loop" );
 657                         mlt_properties_set_int( properties, "_length_computed", 1 );
 658                 }
 659         }
 660
 661         if ( format->start_time != AV_NOPTS_VALUE )
 662                 self->start_time = format->start_time;
 663
 664         // Check if we're seekable
 665         // avdevices are typically AVFMT_NOFILE and not seekable
 666         self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
 667         if ( format->pb )
 668         {
 669                 // protocols can indicate if they support seeking
 670 #if LIBAVFORMAT_VERSION_MAJOR > 52
 671                 self->seekable = format->pb->seekable;
 672 #else
 673                 URLContext *uc = url_fileno( format->pb );
 674                 if ( uc )
 675                         self->seekable = !uc->is_streamed;
 676 #endif
 677         }
 678         if ( self->seekable )
 679         {
 680                 // Do a more rigourous test of seekable on a disposable context
 681                 self->seekable = av_seek_frame( format, -1, self->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
 682                 mlt_properties_set_int( properties, "seekable", self->seekable );
 683                 self->dummy_context = format;
 684                 av_open_input_file( &self->video_format, filename, NULL, 0, NULL );
 685                 format = self->video_format;
 686                 av_find_stream_info( format );
 687         }
 688
 689         // Fetch the width, height and aspect ratio
 690         if ( self->video_index != -1 )
 691         {
 692                 AVCodecContext *codec_context = format->streams[ self->video_index ]->codec;
 693                 mlt_properties_set_int( properties, "width", codec_context->width );
 694                 mlt_properties_set_int( properties, "height", codec_context->height );
 695
 696                 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
 697                 {
 698                         // Fetch the first frame of DV so we can read it directly
 699                         AVPacket pkt;
 700                         int ret = 0;
 701                         while ( ret >= 0 )
 702                         {
 703                                 ret = av_read_frame( format, &pkt );
 704                                 if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
 705                                 {
 706                                         get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, &pkt );
 707                                         break;
 708                                 }
 709                         }
 710                 }
 711                 else
 712                 {
 713                         get_aspect_ratio( properties, format->streams[ self->video_index ], codec_context, NULL );
 714                 }
 715
 716 #ifdef SWSCALE
 717                 // Verify that we can convert this to YUV 4:2:2
 718                 // TODO: we can now also return RGB and RGBA and quite possibly more in the future.
 719                 struct SwsContext *context = sws_getContext( codec_context->width, codec_context->height, codec_context->pix_fmt,
 720                         codec_context->width, codec_context->height, PIX_FMT_YUYV422, SWS_BILINEAR, NULL, NULL, NULL);
 721                 if ( context )
 722                         sws_freeContext( context );
 723                 else
 724                         error = 1;
 725 #endif
 726         }
 727         return error;
 728 }
 729
 730 /** Open the file.
 731 */
 732
 733 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock )
 734 {
 735         // Return an error code (0 == no error)
 736         int error = 0;
 737         mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
 738
 739         // Lock the service
 740         if ( take_lock )
 741         {
 742                 pthread_mutex_init( &self->audio_mutex, NULL );
 743                 pthread_mutex_init( &self->video_mutex, NULL );
 744                 pthread_mutex_init( &self->packets_mutex, NULL );
 745                 pthread_mutex_lock( &self->audio_mutex );
 746                 pthread_mutex_lock( &self->video_mutex );
 747         }
 748         mlt_events_block( properties, self->parent );
 749
 750         // Parse URL
 751         AVInputFormat *format = NULL;
 752         AVFormatParameters params;
 753         memset( &params, 0, sizeof(params) );
 754         char *filename = parse_url( profile, URL, &format, &params );
 755
 756         // Now attempt to open the file or device with filename
 757         error = av_open_input_file( &self->video_format, filename, format, 0, &params ) < 0;
 758         if ( error )
 759                 // If the URL is a network stream URL, then we probably need to open with full URL
 760                 error = av_open_input_file( &self->video_format, URL, format, 0, &params ) < 0;
 761
 762         // Set MLT properties onto video AVFormatContext
 763         if ( !error && self->video_format )
 764         {
 765                 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
 766 #if LIBAVFORMAT_VERSION_MAJOR > 52
 767                 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
 768                         apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
 769 #endif
 770         }
 771
 772         // Cleanup AVFormatParameters
 773         if ( params.standard )
 774                 free( (void*) params.standard );
 775
 776         // If successful, then try to get additional info
 777         if ( !error && self->video_format )
 778         {
 779                 // Get the stream info
 780                 error = av_find_stream_info( self->video_format ) < 0;
 781
 782                 // Continue if no error
 783                 if ( !error && self->video_format )
 784                 {
 785                         // Find default audio and video streams
 786                         find_default_streams( self );
 787                         error = get_basic_info( self, profile, filename );
 788
 789                         // Initialize position info
 790                         self->first_pts = -1;
 791                         self->last_position = POSITION_INITIAL;
 792
 793                         if ( !self->audio_format )
 794                         {
 795                                 // We're going to cheat here - for seekable A/V files, we will have separate contexts
 796                                 // to support independent seeking of audio from video.
 797                                 // TODO: Is this really necessary?
 798                                 if ( self->audio_index != -1 && self->video_index != -1 )
 799                                 {
 800                                         if ( self->seekable )
 801                                         {
 802                                                 // And open again for our audio context
 803                                                 av_open_input_file( &self->audio_format, filename, NULL, 0, NULL );
 804                                                 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
 805 #if LIBAVFORMAT_VERSION_MAJOR > 52
 806                                                 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
 807                                                         apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
 808 #endif
 809                                                 av_find_stream_info( self->audio_format );
 810                                         }
 811                                         else
 812                                         {
 813                                                 self->audio_format = self->video_format;
 814                                         }
 815                                 }
 816                                 else if ( self->audio_index != -1 )
 817                                 {
 818                                         // We only have an audio context
 819                                         self->audio_format = self->video_format;
 820                                         self->video_format = NULL;
 821                                 }
 822                                 else if ( self->video_index == -1 )
 823                                 {
 824                                         // Something has gone wrong
 825                                         error = -1;
 826                                 }
 827                                 if ( self->audio_format && !self->audio_streams )
 828                                         get_audio_streams_info( self );
 829                         }
 830                 }
 831         }
 832         if ( filename )
 833                 free( filename );
 834         if ( !error )
 835         {
 836                 self->apackets = mlt_deque_init();
 837                 self->vpackets = mlt_deque_init();
 838         }
 839
 840         if ( self->dummy_context )
 841         {
 842                 av_close_input_file( self->dummy_context );
 843                 self->dummy_context = NULL;
 844         }
 845
 846         // Unlock the service
 847         if ( take_lock )
 848         {
 849                 pthread_mutex_unlock( &self->audio_mutex );
 850                 pthread_mutex_unlock( &self->video_mutex );
 851         }
 852         mlt_events_unblock( properties, self->parent );
 853
 854         return error;
 855 }
 856
 857 static void reopen_video( producer_avformat self, mlt_producer producer )
 858 {
 859         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
 860         mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
 861         pthread_mutex_lock( &self->audio_mutex );
 862
 863         if ( self->video_codec )
 864         {
 865                 avformat_lock();
 866                 avcodec_close( self->video_codec );
 867                 avformat_unlock();
 868         }
 869         self->video_codec = NULL;
 870         if ( self->dummy_context )
 871                 av_close_input_file( self->dummy_context );
 872         self->dummy_context = NULL;
 873         if ( self->video_format )
 874                 av_close_input_file( self->video_format );
 875         self->video_format = NULL;
 876
 877         int audio_index = self->audio_index;
 878         int video_index = self->video_index;
 879
 880         producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
 881                 mlt_properties_get( properties, "resource" ), 0 );
 882
 883         self->audio_index = audio_index;
 884         if ( self->video_format && video_index > -1 )
 885         {
 886                 self->video_index = video_index;
 887                 video_codec_init( self, video_index, properties );
 888         }
 889
 890         pthread_mutex_unlock( &self->audio_mutex );
 891         mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
 892 }
 893
 894 static int seek_video( producer_avformat self, mlt_position position,
 895         int64_t req_position, int must_decode, int use_new_seek, int *ignore )
 896 {
 897         mlt_producer producer = self->parent;
 898         int paused = 0;
 899
 900         if ( self->seekable && ( position != self->video_expected || self->last_position < 0 ) )
 901         {
 902                 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
 903
 904                 // Fetch the video format context
 905                 AVFormatContext *context = self->video_format;
 906
 907                 // Get the video stream
 908                 AVStream *stream = context->streams[ self->video_index ];
 909
 910                 // Get codec context
 911                 AVCodecContext *codec_context = stream->codec;
 912
 913                 // We may want to use the source fps if available
 914                 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
 915                         mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
 916
 917                 if ( self->av_frame && position + 1 == self->video_expected )
 918                 {
 919                         // We're paused - use last image
 920                         paused = 1;
 921                 }
 922                 else if ( !self->seekable && position > self->video_expected && ( position - self->video_expected ) < 250 )
 923                 {
 924                         // Fast forward - seeking is inefficient for small distances - just ignore following frames
 925                         *ignore = ( int )( ( position - self->video_expected ) / mlt_producer_get_fps( producer ) * source_fps );
 926                         codec_context->skip_loop_filter = AVDISCARD_NONREF;
 927                 }
 928                 else if ( self->seekable && ( position < self->video_expected || position - self->video_expected >= 12 || self->last_position < 0 ) )
 929                 {
 930                         if ( use_new_seek && self->last_position == POSITION_INITIAL )
 931                         {
 932                                 // find first key frame
 933                                 int ret = 0;
 934                                 int toscan = 100;
 935                                 AVPacket pkt;
 936
 937                                 while ( ret >= 0 && toscan-- > 0 )
 938                                 {
 939                                         ret = av_read_frame( context, &pkt );
 940                                         if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == self->video_index )
 941                                         {
 942                                                 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) );
 943                                                 self->first_pts = pkt.pts;
 944                                                 toscan = 0;
 945                                         }
 946                                         av_free_packet( &pkt );
 947                                 }
 948                                 // Rewind
 949                                 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
 950                         }
 951
 952                         // Calculate the timestamp for the requested frame
 953                         int64_t timestamp;
 954                         if ( use_new_seek )
 955                         {
 956                                 timestamp = ( req_position - 0.1 / source_fps ) /
 957                                         ( av_q2d( stream->time_base ) * source_fps );
 958                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pos %"PRId64" pts %"PRId64"\n", req_position, timestamp );
 959                                 if ( self->first_pts > 0 )
 960                                         timestamp += self->first_pts;
 961                                 else if ( context->start_time != AV_NOPTS_VALUE )
 962                                         timestamp += context->start_time;
 963                         }
 964                         else
 965                         {
 966                                 timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
 967                                 if ( context->start_time != AV_NOPTS_VALUE )
 968                                         timestamp += context->start_time;
 969                         }
 970                         if ( must_decode )
 971                                 timestamp -= AV_TIME_BASE;
 972                         if ( timestamp < 0 )
 973                                 timestamp = 0;
 974                         mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position %d expected %d last_pos %"PRId64"\n",
 975                                 timestamp, position, self->video_expected, self->last_position );
 976
 977                         // Seek to the timestamp
 978                         if ( use_new_seek )
 979                         {
 980                                 codec_context->skip_loop_filter = AVDISCARD_NONREF;
 981                                 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
 982                         }
 983                         else if ( req_position > 0 || self->last_position <= 0 )
 984                         {
 985                                 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
 986                         }
 987                         else
 988                         {
 989                                 // Re-open video stream when rewinding to beginning from somewhere else.
 990                                 // This is rather ugly, and I prefer not to do it this way, but ffmpeg is
 991                                 // not reliably seeking to the first frame across formats.
 992                                 reopen_video( self, producer );
 993                         }
 994
 995                         // Remove the cached info relating to the previous position
 996                         self->current_position = POSITION_INVALID;
 997                         self->last_position = POSITION_INVALID;
 998                         av_freep( &self->av_frame );
 999
1000                         if ( use_new_seek )
1001                         {
1002                                 // flush any pictures still in decode buffer
1003                                 avcodec_flush_buffers( codec_context );
1004                         }
1005                 }
1006         }
1007         return paused;
1008 }
1009
1010 /** Convert a frame position to a time code.
1011 */
1012
1013 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1014 {
1015         return ( double )position / mlt_producer_get_fps( producer );
1016 }
1017
1018 // Collect information about all audio streams
1019
1020 static void get_audio_streams_info( producer_avformat self )
1021 {
1022         // Fetch the audio format context
1023         AVFormatContext *context = self->audio_format;
1024         int i;
1025
1026         for ( i = 0;
1027                   i < context->nb_streams;
1028                   i++ )
1029         {
1030                 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
1031                 {
1032                         AVCodecContext *codec_context = context->streams[i]->codec;
1033                         AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1034
1035                         // If we don't have a codec and we can't initialise it, we can't do much more...
1036                         avformat_lock( );
1037                         if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1038                         {
1039                                 self->audio_streams++;
1040                                 self->audio_max_stream = i;
1041                                 self->total_channels += codec_context->channels;
1042                                 if ( codec_context->channels > self->max_channel )
1043                                         self->max_channel = codec_context->channels;
1044                                 if ( codec_context->sample_rate > self->max_frequency )
1045                                         self->max_frequency = codec_context->sample_rate;
1046                                 avcodec_close( codec_context );
1047                         }
1048                         avformat_unlock( );
1049                 }
1050         }
1051         mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1052                 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1053
1054         // Other audio-specific initializations
1055         self->resample_factor = 1.0;
1056 }
1057
1058 static void set_luma_transfer( struct SwsContext *context, int colorspace, int use_full_range )
1059 {
1060 #if defined(SWSCALE) && (LIBSWSCALE_VERSION_INT >= ((0<<16)+(7<<8)+2))
1061         int *coefficients;
1062         const int *new_coefficients;
1063         int full_range;
1064         int brightness, contrast, saturation;
1065
1066         if ( sws_getColorspaceDetails( context, &coefficients, &full_range, &coefficients, &full_range,
1067                         &brightness, &contrast, &saturation ) != -1 )
1068         {
1069                 // Don't change these from defaults unless explicitly told to.
1070                 if ( use_full_range >= 0 )
1071                         full_range = use_full_range;
1072                 switch ( colorspace )
1073                 {
1074                 case 170:
1075                 case 470:
1076                 case 601:
1077                 case 624:
1078                         new_coefficients = sws_getCoefficients( SWS_CS_ITU601 );
1079                         break;
1080                 case 240:
1081                         new_coefficients = sws_getCoefficients( SWS_CS_SMPTE240M );
1082                         break;
1083                 case 709:
1084                         new_coefficients = sws_getCoefficients( SWS_CS_ITU709 );
1085                         break;
1086                 default:
1087                         new_coefficients = coefficients;
1088                         break;
1089                 }
1090                 sws_setColorspaceDetails( context, new_coefficients, full_range, new_coefficients, full_range,
1091                         brightness, contrast, saturation );
1092         }
1093 #endif
1094 }
1095
1096 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt,
1097         mlt_image_format *format, int width, int height, int colorspace )
1098 {
1099 #ifdef SWSCALE
1100         int full_range = -1;
1101         int flags = SWS_BILINEAR | SWS_ACCURATE_RND;
1102
1103 #ifdef USE_MMX
1104         flags |= SWS_CPU_CAPS_MMX;
1105 #endif
1106 #ifdef USE_SSE
1107         flags |= SWS_CPU_CAPS_MMX2;
1108 #endif
1109
1110         if ( pix_fmt == PIX_FMT_RGB32 )
1111         {
1112                 *format = mlt_image_rgb24a;
1113                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1114                         width, height, PIX_FMT_RGBA, flags, NULL, NULL, NULL);
1115                 AVPicture output;
1116                 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1117                 set_luma_transfer( context, colorspace, full_range );
1118                 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1119                         output.data, output.linesize);
1120                 sws_freeContext( context );
1121         }
1122         else if ( *format == mlt_image_yuv420p )
1123         {
1124                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1125                         width, height, PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1126                 AVPicture output;
1127                 output.data[0] = buffer;
1128                 output.data[1] = buffer + width * height;
1129                 output.data[2] = buffer + ( 5 * width * height ) / 4;
1130                 output.linesize[0] = width;
1131                 output.linesize[1] = width >> 1;
1132                 output.linesize[2] = width >> 1;
1133                 set_luma_transfer( context, colorspace, full_range );
1134                 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1135                         output.data, output.linesize);
1136                 sws_freeContext( context );
1137         }
1138         else if ( *format == mlt_image_rgb24 )
1139         {
1140                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1141                         width, height, PIX_FMT_RGB24, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1142                 AVPicture output;
1143                 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1144                 set_luma_transfer( context, colorspace, full_range );
1145                 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1146                         output.data, output.linesize);
1147                 sws_freeContext( context );
1148         }
1149         else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1150         {
1151                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1152                         width, height, PIX_FMT_RGBA, flags | SWS_FULL_CHR_H_INT, NULL, NULL, NULL);
1153                 AVPicture output;
1154                 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
1155                 set_luma_transfer( context, colorspace, full_range );
1156                 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1157                         output.data, output.linesize);
1158                 sws_freeContext( context );
1159         }
1160         else
1161         {
1162                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1163                         width, height, PIX_FMT_YUYV422, flags | SWS_FULL_CHR_H_INP, NULL, NULL, NULL);
1164                 AVPicture output;
1165                 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1166                 set_luma_transfer( context, colorspace, full_range );
1167                 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1168                         output.data, output.linesize);
1169                 sws_freeContext( context );
1170         }
1171 #else
1172         if ( *format == mlt_image_yuv420p )
1173         {
1174                 AVPicture pict;
1175                 pict.data[0] = buffer;
1176                 pict.data[1] = buffer + width * height;
1177                 pict.data[2] = buffer + ( 5 * width * height ) / 4;
1178                 pict.linesize[0] = width;
1179                 pict.linesize[1] = width >> 1;
1180                 pict.linesize[2] = width >> 1;
1181                 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
1182         }
1183         else if ( *format == mlt_image_rgb24 )
1184         {
1185                 AVPicture output;
1186                 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
1187                 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
1188         }
1189         else if ( format == mlt_image_rgb24a || format == mlt_image_opengl )
1190         {
1191                 AVPicture output;
1192                 avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height );
1193                 img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height );
1194         }
1195         else
1196         {
1197                 AVPicture output;
1198                 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
1199                 img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height );
1200         }
1201 #endif
1202 }
1203
1204 /** Allocate the image buffer and set it on the frame.
1205 */
1206
1207 static int allocate_buffer( mlt_frame frame, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
1208 {
1209         int size = 0;
1210
1211         if ( codec_context->width == 0 || codec_context->height == 0 )
1212                 return size;
1213
1214         *width = codec_context->width;
1215         *height = codec_context->height;
1216
1217         if ( codec_context->pix_fmt == PIX_FMT_RGB32 )
1218                 size = *width * ( *height + 1 ) * 4;
1219         else
1220                 size = mlt_image_format_size( *format, *width, *height, NULL );
1221
1222         // Construct the output image
1223         *buffer = mlt_pool_alloc( size );
1224         if ( *buffer )
1225                 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1226         else
1227                 size = 0;
1228
1229         return size;
1230 }
1231
1232 /** Get an image from a frame.
1233 */
1234
1235 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1236 {
1237         // Get the producer
1238         producer_avformat self = mlt_frame_pop_service( frame );
1239         mlt_producer producer = self->parent;
1240
1241         // Get the properties from the frame
1242         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1243
1244         // Obtain the frame number of this frame
1245         mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
1246
1247         // Get the producer properties
1248         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1249
1250         pthread_mutex_lock( &self->video_mutex );
1251
1252         // Fetch the video format context
1253         AVFormatContext *context = self->video_format;
1254
1255         // Get the video stream
1256         AVStream *stream = context->streams[ self->video_index ];
1257
1258         // Get codec context
1259         AVCodecContext *codec_context = stream->codec;
1260
1261         // Get the image cache
1262         if ( ! self->image_cache && ! mlt_properties_get_int( properties, "noimagecache" ) )
1263                 self->image_cache = mlt_cache_init();
1264         if ( self->image_cache )
1265         {
1266                 mlt_cache_item item = mlt_cache_get( self->image_cache, (void*) position );
1267                 uint8_t *original = mlt_cache_item_data( item, (int*) format );
1268                 if ( original )
1269                 {
1270                         // Set the resolution
1271                         *width = codec_context->width;
1272                         *height = codec_context->height;
1273
1274                         // Workaround 1088 encodings missing cropping info.
1275                         if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1276                                 *height = 1080;
1277
1278                         // Cache hit
1279                         int size = mlt_image_format_size( *format, *width, *height, NULL );
1280                         if ( writable )
1281                         {
1282                                 *buffer = mlt_pool_alloc( size );
1283                                 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1284                                 memcpy( *buffer, original, size );
1285                                 mlt_cache_item_close( item );
1286                         }
1287                         else
1288                         {
1289                                 *buffer = original;
1290                                 mlt_properties_set_data( frame_properties, "avformat.image_cache", item, 0, ( mlt_destructor )mlt_cache_item_close, NULL );
1291                                 mlt_frame_set_image( frame, *buffer, size, NULL );
1292                         }
1293                         self->got_picture = 1;
1294
1295                         goto exit_get_image;
1296                 }
1297         }
1298         // Cache miss
1299         int image_size = 0;
1300
1301         // Packet
1302         AVPacket pkt;
1303
1304         // Special case ffwd handling
1305         int ignore = 0;
1306
1307         // We may want to use the source fps if available
1308         double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1309                 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1310
1311         // This is the physical frame position in the source
1312         int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1313
1314         // Determines if we have to decode all frames in a sequence
1315         // Temporary hack to improve intra frame only
1316         int must_decode = strcmp( codec_context->codec->name, "dnxhd" ) &&
1317                                   strcmp( codec_context->codec->name, "dvvideo" ) &&
1318                                   strcmp( codec_context->codec->name, "huffyuv" ) &&
1319                                   strcmp( codec_context->codec->name, "mjpeg" ) &&
1320                                   strcmp( codec_context->codec->name, "rawvideo" );
1321
1322         // Turn on usage of new seek API and PTS for seeking
1323         int use_new_seek = self->seekable &&
1324                 codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" );
1325         if ( mlt_properties_get( properties, "new_seek" ) )
1326                 use_new_seek = mlt_properties_get_int( properties, "new_seek" );
1327
1328         // Seek if necessary
1329         int paused = seek_video( self, position, req_position, must_decode, use_new_seek, &ignore );
1330
1331         // Seek might have reopened the file
1332         context = self->video_format;
1333         stream = context->streams[ self->video_index ];
1334         codec_context = stream->codec;
1335
1336         // Duplicate the last image if necessary
1337         if ( self->av_frame && self->av_frame->linesize[0] && self->got_picture
1338                  && ( paused
1339                           || self->current_position == req_position
1340                           || ( !use_new_seek && self->current_position > req_position ) ) )
1341         {
1342                 // Duplicate it
1343                 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1344                 {
1345                         // Workaround 1088 encodings missing cropping info.
1346                         if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1347                                 *height = 1080;
1348 #ifdef VDPAU
1349                         if ( self->vdpau && self->vdpau->buffer )
1350                         {
1351                                 AVPicture picture;
1352                                 picture.data[0] = self->vdpau->buffer;
1353                                 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1354                                 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1355                                 picture.linesize[0] = codec_context->width;
1356                                 picture.linesize[1] = codec_context->width / 2;
1357                                 picture.linesize[2] = codec_context->width / 2;
1358                                 convert_image( (AVFrame*) &picture, *buffer,
1359                                         PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1360                         }
1361                         else
1362 #endif
1363                         convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1364                                 format, *width, *height, self->colorspace );
1365                 }
1366                 else
1367                         mlt_frame_get_image( frame, buffer, format, width, height, writable );
1368         }
1369         else
1370         {
1371                 int ret = 0;
1372                 int64_t int_position = 0;
1373                 int decode_errors = 0;
1374                 int got_picture = 0;
1375
1376                 av_init_packet( &pkt );
1377
1378                 // Construct an AVFrame for YUV422 conversion
1379                 if ( !self->av_frame )
1380                         self->av_frame = avcodec_alloc_frame( );
1381
1382                 while( ret >= 0 && !got_picture )
1383                 {
1384                         // Read a packet
1385                         pthread_mutex_lock( &self->packets_mutex );
1386                         if ( mlt_deque_count( self->vpackets ) )
1387                         {
1388                                 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1389                                 pkt = *tmp;
1390                                 free( tmp );
1391                         }
1392                         else
1393                         {
1394                                 ret = av_read_frame( context, &pkt );
1395                                 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->audio_index )
1396                                 {
1397                                         if ( !av_dup_packet( &pkt ) )
1398                                         {
1399                                                 AVPacket *tmp = malloc( sizeof(AVPacket) );
1400                                                 *tmp = pkt;
1401                                                 mlt_deque_push_back( self->apackets, tmp );
1402                                         }
1403                                 }
1404                         }
1405                         pthread_mutex_unlock( &self->packets_mutex );
1406
1407                         // We only deal with video from the selected video_index
1408                         if ( ret >= 0 && pkt.stream_index == self->video_index && pkt.size > 0 )
1409                         {
1410                                 // Determine time code of the packet
1411                                 if ( use_new_seek )
1412                                 {
1413                                         int64_t pts = pkt.pts;
1414                                         if ( self->first_pts > 0 )
1415                                                 pts -= self->first_pts;
1416                                         else if ( context->start_time != AV_NOPTS_VALUE )
1417                                                 pts -= context->start_time;
1418                                         int_position = ( int64_t )( av_q2d( stream->time_base ) * pts * source_fps + 0.1 );
1419                                         if ( pkt.pts == AV_NOPTS_VALUE )
1420                                         {
1421                                                 self->invalid_pts_counter++;
1422                                                 if ( self->invalid_pts_counter > 20 )
1423                                                 {
1424                                                         mlt_log_panic( MLT_PRODUCER_SERVICE(producer), "\ainvalid PTS; DISABLING NEW_SEEK!\n" );
1425                                                         mlt_properties_set_int( properties, "new_seek", 0 );
1426                                                         int_position = req_position;
1427                                                         use_new_seek = 0;
1428                                                 }
1429                                         }
1430                                         else
1431                                         {
1432                                                 self->invalid_pts_counter = 0;
1433                                         }
1434                                         mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.pts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1435                                                 pkt.pts, req_position, self->current_position, int_position );
1436                                 }
1437                                 else
1438                                 {
1439                                         if ( pkt.dts != AV_NOPTS_VALUE )
1440                                         {
1441                                                 double delay = mlt_properties_get_double( properties, "video_delay" );
1442                                                 int_position = ( int64_t )( ( av_q2d( stream->time_base ) * pkt.dts + delay ) * source_fps + 0.5 );
1443                                                 if ( context->start_time != AV_NOPTS_VALUE )
1444                                                         int_position -= ( int64_t )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
1445                                                 if ( int_position == self->last_position )
1446                                                         int_position = self->last_position + 1;
1447                                         }
1448                                         mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1449                                                 pkt.dts, req_position, self->current_position, int_position );
1450                                         // Make a dumb assumption on streams that contain wild timestamps
1451                                         if ( abs( req_position - int_position ) > 999 )
1452                                         {
1453                                                 int_position = req_position;
1454                                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" );
1455                                         }
1456                                 }
1457                                 self->last_position = int_position;
1458
1459                                 // Decode the image
1460                                 if ( must_decode || int_position >= req_position )
1461                                 {
1462 #ifdef VDPAU
1463                                         if ( g_vdpau && self->vdpau )
1464                                         {
1465                                                 if ( g_vdpau->producer != self )
1466                                                 {
1467                                                         vdpau_decoder_close();
1468                                                         vdpau_decoder_init( self );
1469                                                 }
1470                                                 if ( self->vdpau )
1471                                                         self->vdpau->is_decoded = 0;
1472                                         }
1473 #endif
1474                                         codec_context->reordered_opaque = pkt.pts;
1475                                         if ( int_position >= req_position )
1476                                                 codec_context->skip_loop_filter = AVDISCARD_NONE;
1477 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
1478                                         ret = avcodec_decode_video2( codec_context, self->av_frame, &got_picture, &pkt );
1479 #else
1480                                         ret = avcodec_decode_video( codec_context, self->av_frame, &got_picture, pkt.data, pkt.size );
1481 #endif
1482                                         // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1483                                         if ( ret < 0 )
1484                                         {
1485                                                 if ( ++decode_errors <= 10 )
1486                                                         ret = 0;
1487                                         }
1488                                         else
1489                                         {
1490                                                 decode_errors = 0;
1491                                         }
1492                                 }
1493
1494                                 if ( got_picture )
1495                                 {
1496                                         if ( use_new_seek )
1497                                         {
1498                                                 // Determine time code of the packet
1499                                                 int64_t pts = self->av_frame->reordered_opaque;
1500                                                 if ( self->first_pts > 0 )
1501                                                         pts -= self->first_pts;
1502                                                 else if ( context->start_time != AV_NOPTS_VALUE )
1503                                                         pts -= context->start_time;
1504                                                 int_position = ( int64_t )( av_q2d( stream->time_base) * pts * source_fps + 0.1 );
1505                                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "got frame %"PRId64", key %d\n", int_position, self->av_frame->key_frame );
1506                                         }
1507                                         // Handle ignore
1508                                         if ( int_position < req_position )
1509                                         {
1510                                                 ignore = 0;
1511                                                 got_picture = 0;
1512                                         }
1513                                         else if ( int_position >= req_position )
1514                                         {
1515                                                 ignore = 0;
1516                                                 codec_context->skip_loop_filter = AVDISCARD_NONE;
1517                                         }
1518                                         else if ( ignore -- )
1519                                         {
1520                                                 got_picture = 0;
1521                                         }
1522                                 }
1523                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY );
1524                         }
1525
1526                         // Now handle the picture if we have one
1527                         if ( got_picture )
1528                         {
1529                                 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1530                                 {
1531                                         // Workaround 1088 encodings missing cropping info.
1532                                         if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1533                                                 *height = 1080;
1534 #ifdef VDPAU
1535                                         if ( self->vdpau )
1536                                         {
1537                                                 if ( self->vdpau->is_decoded )
1538                                                 {
1539                                                         struct vdpau_render_state *render = (struct vdpau_render_state*) self->av_frame->data[0];
1540                                                         void *planes[3];
1541                                                         uint32_t pitches[3];
1542                                                         VdpYCbCrFormat dest_format = VDP_YCBCR_FORMAT_YV12;
1543
1544                                                         if ( !self->vdpau->buffer )
1545                                                                 self->vdpau->buffer = mlt_pool_alloc( codec_context->width * codec_context->height * 3 / 2 );
1546                                                         self->av_frame->data[0] = planes[0] = self->vdpau->buffer;
1547                                                         self->av_frame->data[2] = planes[1] = self->vdpau->buffer + codec_context->width * codec_context->height;
1548                                                         self->av_frame->data[1] = planes[2] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1549                                                         self->av_frame->linesize[0] = pitches[0] = codec_context->width;
1550                                                         self->av_frame->linesize[1] = pitches[1] = codec_context->width / 2;
1551                                                         self->av_frame->linesize[2] = pitches[2] = codec_context->width / 2;
1552
1553                                                         VdpStatus status = vdp_surface_get_bits( render->surface, dest_format, planes, pitches );
1554                                                         if ( status == VDP_STATUS_OK )
1555                                                         {
1556                                                                 convert_image( self->av_frame, *buffer, PIX_FMT_YUV420P,
1557                                                                         format, *width, *height, self->colorspace );
1558                                                         }
1559                                                         else
1560                                                         {
1561                                                                 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU Error: %s\n", vdp_get_error_string( status ) );
1562                                                                 image_size = self->vdpau->is_decoded = 0;
1563                                                         }
1564                                                 }
1565                                                 else
1566                                                 {
1567                                                         mlt_log_error( MLT_PRODUCER_SERVICE(producer), "VDPAU error in VdpDecoderRender\n" );
1568                                                         image_size = got_picture = 0;
1569                                                 }
1570                                         }
1571                                         else
1572 #endif
1573                                         convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1574                                                 format, *width, *height, self->colorspace );
1575                                         self->top_field_first |= self->av_frame->top_field_first;
1576                                         self->current_position = int_position;
1577                                         self->got_picture = 1;
1578                                 }
1579                                 else
1580                                 {
1581                                         got_picture = 0;
1582                                 }
1583                         }
1584                         if ( self->seekable || pkt.stream_index != self->audio_index )
1585                                 av_free_packet( &pkt );
1586                 }
1587         }
1588
1589         if ( self->got_picture && image_size > 0 && self->image_cache )
1590         {
1591                 // Copy buffer to image cache
1592                 uint8_t *image = mlt_pool_alloc( image_size );
1593                 memcpy( image, *buffer, image_size );
1594                 mlt_cache_put( self->image_cache, (void*) position, image, *format, mlt_pool_release );
1595         }
1596         // Try to duplicate last image if there was a decoding failure
1597         else if ( !image_size && self->av_frame && self->av_frame->linesize[0] )
1598         {
1599                 // Duplicate it
1600                 if ( ( image_size = allocate_buffer( frame, codec_context, buffer, format, width, height ) ) )
1601                 {
1602                         // Workaround 1088 encodings missing cropping info.
1603                         if ( *height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1604                                 *height = 1080;
1605 #ifdef VDPAU
1606                         if ( self->vdpau && self->vdpau->buffer )
1607                         {
1608                                 AVPicture picture;
1609                                 picture.data[0] = self->vdpau->buffer;
1610                                 picture.data[2] = self->vdpau->buffer + codec_context->width * codec_context->height;
1611                                 picture.data[1] = self->vdpau->buffer + codec_context->width * codec_context->height * 5 / 4;
1612                                 picture.linesize[0] = codec_context->width;
1613                                 picture.linesize[1] = codec_context->width / 2;
1614                                 picture.linesize[2] = codec_context->width / 2;
1615                                 convert_image( (AVFrame*) &picture, *buffer,
1616                                         PIX_FMT_YUV420P, format, *width, *height, self->colorspace );
1617                         }
1618                         else
1619 #endif
1620                         convert_image( self->av_frame, *buffer, codec_context->pix_fmt,
1621                                 format, *width, *height, self->colorspace );
1622                         self->got_picture = 1;
1623                 }
1624                 else
1625                         mlt_frame_get_image( frame, buffer, format, width, height, writable );
1626         }
1627
1628         // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1629         self->video_expected = position + 1;
1630
1631 exit_get_image:
1632
1633         pthread_mutex_unlock( &self->video_mutex );
1634
1635         // Set the progressive flag
1636         if ( mlt_properties_get( properties, "force_progressive" ) )
1637                 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
1638         else if ( self->av_frame )
1639                 mlt_properties_set_int( frame_properties, "progressive", !self->av_frame->interlaced_frame );
1640
1641         // Set the field order property for this frame
1642         if ( mlt_properties_get( properties, "force_tff" ) )
1643                 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
1644         else
1645                 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
1646
1647         // Set immutable properties of the selected track's (or overridden) source attributes.
1648         mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1649         mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
1650         mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
1651         mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1652
1653         return !self->got_picture;
1654 }
1655
1656 /** Process properties as AVOptions and apply to AV context obj
1657 */
1658
1659 static void apply_properties( void *obj, mlt_properties properties, int flags )
1660 {
1661         int i;
1662         int count = mlt_properties_count( properties );
1663         for ( i = 0; i < count; i++ )
1664         {
1665                 const char *opt_name = mlt_properties_get_name( properties, i );
1666                 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1667                 if ( opt_name && mlt_properties_get( properties, opt_name ) )
1668                 {
1669                         if ( opt )
1670 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1671                                 av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1672 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1673                                 av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1674 #else
1675                                 av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1676 #endif
1677                 }
1678         }
1679 }
1680
1681 /** Initialize the video codec context.
1682  */
1683
1684 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
1685 {
1686         // Initialise the codec if necessary
1687         if ( !self->video_codec )
1688         {
1689                 // Get the video stream
1690                 AVStream *stream = self->video_format->streams[ index ];
1691
1692                 // Get codec context
1693                 AVCodecContext *codec_context = stream->codec;
1694
1695                 // Find the codec
1696                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1697 #ifdef VDPAU
1698                 if ( codec_context->codec_id == CODEC_ID_H264 )
1699                 {
1700                         if ( ( codec = avcodec_find_decoder_by_name( "h264_vdpau" ) ) )
1701                         {
1702                                 if ( vdpau_init( self ) )
1703                                 {
1704                                         self->video_codec = codec_context;
1705                                         if ( !vdpau_decoder_init( self ) )
1706                                                 vdpau_decoder_close();
1707                                 }
1708                         }
1709                         if ( !self->vdpau )
1710                                 codec = avcodec_find_decoder( codec_context->codec_id );
1711                 }
1712 #endif
1713
1714                 // Initialise multi-threading
1715                 int thread_count = mlt_properties_get_int( properties, "threads" );
1716                 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1717                         thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1718                 if ( thread_count > 1 )
1719                         codec_context->thread_count = thread_count;
1720
1721                 // If we don't have a codec and we can't initialise it, we can't do much more...
1722                 avformat_lock( );
1723                 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1724                 {
1725                         // Now store the codec with its destructor
1726                         self->video_codec = codec_context;
1727                 }
1728                 else
1729                 {
1730                         // Remember that we can't use this later
1731                         self->video_index = -1;
1732                         avformat_unlock( );
1733                         return 0;
1734                 }
1735                 avformat_unlock( );
1736
1737                 // Process properties as AVOptions
1738                 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1739 #if LIBAVCODEC_VERSION_MAJOR > 52
1740                 if ( codec->priv_class && codec_context->priv_data )
1741                         apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1742 #endif
1743
1744                 // Reset some image properties
1745                 mlt_properties_set_int( properties, "width", self->video_codec->width );
1746                 mlt_properties_set_int( properties, "height", self->video_codec->height );
1747                 // For DV, we'll just use the saved aspect ratio
1748                 if ( codec_context->codec_id != CODEC_ID_DVVIDEO )
1749                         get_aspect_ratio( properties, stream, self->video_codec, NULL );
1750
1751                 // Determine the fps first from the codec
1752                 double source_fps = (double) self->video_codec->time_base.den /
1753                                                                    ( self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1754
1755                 if ( mlt_properties_get( properties, "force_fps" ) )
1756                 {
1757                         source_fps = mlt_properties_get_double( properties, "force_fps" );
1758                         stream->time_base = av_d2q( source_fps, 1024 );
1759                         mlt_properties_set_int( properties, "meta.media.frame_rate_num", stream->time_base.num );
1760                         mlt_properties_set_int( properties, "meta.media.frame_rate_den", stream->time_base.den );
1761                 }
1762                 else
1763                 {
1764                         // If the muxer reports a frame rate different than the codec
1765 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1766                         double muxer_fps = av_q2d( stream->avg_frame_rate );
1767                         if ( isnan( muxer_fps ) || muxer_fps == 0 )
1768                                 muxer_fps = av_q2d( stream->r_frame_rate );
1769 #else
1770                         double muxer_fps = av_q2d( stream->r_frame_rate );
1771 #endif
1772                         // Choose the lesser - the wrong tends to be off by some multiple of 10
1773                         source_fps = FFMIN( source_fps, muxer_fps );
1774                         if ( source_fps >= 1.0 && ( source_fps < muxer_fps || isnan( muxer_fps ) ) )
1775                         {
1776                                 mlt_properties_set_int( properties, "meta.media.frame_rate_num", self->video_codec->time_base.den );
1777                                 mlt_properties_set_int( properties, "meta.media.frame_rate_den", self->video_codec->time_base.num == 0 ? 1 : self->video_codec->time_base.num );
1778                         }
1779                         else if ( muxer_fps > 0 )
1780                         {
1781                                 AVRational frame_rate = stream->r_frame_rate;
1782                                 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
1783                                 // avg_frame_rate gives some approximate value that does not well match the media.
1784                                 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
1785                                 // results in some very choppy output, but some value slightly different works
1786                                 // great.
1787 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(42<<8)+0)
1788                                 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
1789                                         frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
1790 #endif
1791                                 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1792                                 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1793                         }
1794                         else
1795                         {
1796                                 source_fps = mlt_producer_get_fps( self->parent );
1797                                 AVRational frame_rate = av_d2q( source_fps, 255 );
1798                                 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
1799                                 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
1800                         }
1801                 }
1802
1803                 // source_fps is deprecated in favor of meta.media.frame_rate_num and .frame_rate_den
1804                 if ( source_fps > 0 )
1805                         mlt_properties_set_double( properties, "source_fps", source_fps );
1806                 else
1807                         mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( self->parent ) );
1808
1809                 // Set the YUV colorspace from override or detect
1810                 self->colorspace = mlt_properties_get_int( properties, "force_colorspace" );
1811 #if LIBAVCODEC_VERSION_INT > ((52<<16)+(28<<8)+0)
1812                 if ( ! self->colorspace )
1813                 {
1814                         switch ( self->video_codec->colorspace )
1815                         {
1816                         case AVCOL_SPC_SMPTE240M:
1817                                 self->colorspace = 240;
1818                                 break;
1819                         case AVCOL_SPC_BT470BG:
1820                         case AVCOL_SPC_SMPTE170M:
1821                                 self->colorspace = 601;
1822                                 break;
1823                         case AVCOL_SPC_BT709:
1824                                 self->colorspace = 709;
1825                                 break;
1826                         default:
1827                                 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
1828                                 self->colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
1829                                 break;
1830                         }
1831                 }
1832 #endif
1833                 // Let apps get chosen colorspace
1834                 mlt_properties_set_int( properties, "meta.media.colorspace", self->colorspace );
1835         }
1836         return self->video_codec && self->video_index > -1;
1837 }
1838
1839 /** Set up video handling.
1840 */
1841
1842 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
1843 {
1844         // Get the producer
1845         mlt_producer producer = self->parent;
1846
1847         // Get the properties
1848         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1849
1850         // Fetch the video format context
1851         AVFormatContext *context = self->video_format;
1852
1853         // Get the video_index
1854         int index = mlt_properties_get_int( properties, "video_index" );
1855
1856         // Reopen the file if necessary
1857         if ( !context && index > -1 )
1858         {
1859                 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1860                         mlt_properties_get( properties, "resource" ), 1 );
1861                 context = self->video_format;
1862         }
1863
1864         // Exception handling for video_index
1865         if ( context && index >= (int) context->nb_streams )
1866         {
1867                 // Get the last video stream
1868                 for ( index = context->nb_streams - 1;
1869                           index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1870                           index-- );
1871                 mlt_properties_set_int( properties, "video_index", index );
1872         }
1873         if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1874         {
1875                 // Invalidate the video stream
1876                 index = -1;
1877                 mlt_properties_set_int( properties, "video_index", index );
1878         }
1879
1880         // Update the video properties if the index changed
1881         if ( index != self->video_index )
1882         {
1883                 // Reset the video properties if the index changed
1884                 self->video_index = index;
1885                 if ( self->video_codec )
1886                 {
1887                         avformat_lock();
1888                         avcodec_close( self->video_codec );
1889                         avformat_unlock();
1890                 }
1891                 self->video_codec = NULL;
1892         }
1893
1894         // Get the frame properties
1895         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1896
1897         // Get the codec
1898         if ( context && index > -1 && video_codec_init( self, index, properties ) )
1899         {
1900                 // Set the frame properties
1901                 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1902                 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1903                         force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1904
1905                 // Set the width and height
1906                 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
1907                 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
1908                 // real_width and real_height are deprecated in favor of meta.media.width and .height
1909                 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
1910                 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
1911                 mlt_properties_set_int( frame_properties, "real_width", self->video_codec->width );
1912                 mlt_properties_set_int( frame_properties, "real_height", self->video_codec->height );
1913                 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1914                 mlt_properties_set_int( frame_properties, "colorspace", self->colorspace );
1915
1916                 // Workaround 1088 encodings missing cropping info.
1917                 if ( self->video_codec->height == 1088 && mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) ) == 16.0/9.0 )
1918                 {
1919                         mlt_properties_set_int( properties, "meta.media.height", 1080 );
1920                         mlt_properties_set_int( frame_properties, "real_height", 1080 );
1921                 }
1922
1923                 // Add our image operation
1924                 mlt_frame_push_service( frame, self );
1925                 mlt_frame_push_get_image( frame, producer_get_image );
1926         }
1927         else
1928         {
1929                 // If something failed, use test card image
1930                 mlt_properties_set_int( frame_properties, "test_image", 1 );
1931         }
1932 }
1933
1934 static int seek_audio( producer_avformat self, mlt_position position, double timecode, int *ignore )
1935 {
1936         int paused = 0;
1937
1938         // Seek if necessary
1939         if ( self->seekable && position != self->audio_expected )
1940         {
1941                 if ( position + 1 == self->audio_expected )
1942                 {
1943                         // We're paused - silence required
1944                         paused = 1;
1945                 }
1946                 else if ( !self->seekable && position > self->audio_expected && ( position - self->audio_expected ) < 250 )
1947                 {
1948                         // Fast forward - seeking is inefficient for small distances - just ignore following frames
1949                         *ignore = position - self->audio_expected;
1950                 }
1951                 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
1952                 {
1953                         AVFormatContext *context = self->audio_format;
1954                         int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
1955                         if ( context->start_time != AV_NOPTS_VALUE )
1956                                 timestamp += context->start_time;
1957                         if ( timestamp < 0 )
1958                                 timestamp = 0;
1959
1960                         // Set to the real timecode
1961                         if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
1962                                 paused = 1;
1963
1964                         // Clear the usage in the audio buffer
1965                         int i = MAX_AUDIO_STREAMS + 1;
1966                         while ( --i )
1967                                 self->audio_used[i - 1] = 0;
1968                 }
1969         }
1970         return paused;
1971 }
1972
1973 static int sample_bytes( AVCodecContext *context )
1974 {
1975 #if LIBAVCODEC_VERSION_MAJOR > 52
1976         return av_get_bits_per_sample_fmt( context->sample_fmt ) / 8;
1977 #else
1978         return av_get_bits_per_sample_format( context->sample_fmt ) / 8;
1979 #endif
1980 }
1981
1982 static int decode_audio( producer_avformat self, int *ignore, AVPacket pkt, int channels, int samples, double timecode, double fps )
1983 {
1984         // Fetch the audio_format
1985         AVFormatContext *context = self->audio_format;
1986
1987         // Get the current stream index
1988         int index = pkt.stream_index;
1989
1990         // Get codec context
1991         AVCodecContext *codec_context = self->audio_codec[ index ];
1992
1993         // Obtain the resample context if it exists (not always needed)
1994         ReSampleContext *resample = self->audio_resample[ index ];
1995
1996         // Obtain the audio buffers
1997         uint8_t *audio_buffer = self->audio_buffer[ index ];
1998         uint8_t *decode_buffer = self->decode_buffer[ index ];
1999
2000         int audio_used = self->audio_used[ index ];
2001         uint8_t *ptr = pkt.data;
2002         int len = pkt.size;
2003         int ret = 0;
2004
2005         while ( ptr && ret >= 0 && len > 0 )
2006         {
2007                 int sizeof_sample = resample? sizeof( int16_t ) : sample_bytes( codec_context );
2008                 int data_size = self->audio_buffer_size[ index ];
2009
2010                 // Decode the audio
2011 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(26<<8)+0))
2012                 ret = avcodec_decode_audio3( codec_context, (int16_t*) decode_buffer, &data_size, &pkt );
2013 #elif (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
2014                 ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
2015 #else
2016                 ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len );
2017 #endif
2018                 if ( ret < 0 )
2019                 {
2020                         mlt_log_warning( MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", ret );
2021                         break;
2022                 }
2023
2024                 pkt.size = len -= ret;
2025                 pkt.data = ptr += ret;
2026
2027                 // If decoded successfully
2028                 if ( data_size > 0 )
2029                 {
2030                         // Figure out how many samples will be needed after resampling
2031                         int convert_samples = data_size / codec_context->channels / sample_bytes( codec_context );
2032                         int samples_needed = self->resample_factor * convert_samples;
2033
2034                         // Resize audio buffer to prevent overflow
2035                         if ( ( audio_used + samples_needed ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2036                         {
2037                                 self->audio_buffer_size[ index ] = ( audio_used + samples_needed * 2 ) * channels * sizeof_sample;
2038                                 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2039                         }
2040                         if ( resample )
2041                         {
2042                                 // Copy to audio buffer while resampling
2043                                 uint8_t *source = decode_buffer;
2044                                 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
2045                                 audio_used += audio_resample( resample, (short*) dest, (short*) source, convert_samples );
2046                         }
2047                         else
2048                         {
2049                                 // Straight copy to audio buffer
2050                                 memcpy( &audio_buffer[ audio_used * codec_context->channels * sizeof_sample ], decode_buffer, data_size );
2051                                 audio_used += convert_samples;
2052                         }
2053
2054                         // Handle ignore
2055                         while ( *ignore && audio_used > samples )
2056                         {
2057                                 *ignore -= 1;
2058                                 audio_used -= samples;
2059                                 memmove( audio_buffer, &audio_buffer[ samples * (resample? channels : codec_context->channels) * sizeof_sample ],
2060                                                  audio_used * sizeof_sample );
2061                         }
2062                 }
2063         }
2064
2065         // If we're behind, ignore this packet
2066         if ( pkt.pts >= 0 )
2067         {
2068                 double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt.pts;
2069                 int64_t req_position = ( int64_t )( timecode * fps + 0.5 );
2070                 int64_t int_position = ( int64_t )( current_pts * fps + 0.5 );
2071                 if ( context->start_time != AV_NOPTS_VALUE )
2072                         int_position -= ( int64_t )( fps * context->start_time / AV_TIME_BASE + 0.5 );
2073
2074                 if ( *ignore == 0 )
2075                 {
2076                         if ( int_position < req_position )
2077                                 // We are behind, so skip some
2078                                 *ignore = 1;
2079
2080                         // We use nb_streams in this test because the tolerance is dependent
2081                         // on the interleaving of all streams esp. when there is more than
2082                         // one audio stream.
2083                         else if ( int_position > req_position + context->nb_streams )
2084                                 // We are ahead, so seek backwards some more
2085                                 seek_audio( self, req_position, timecode - 1.0, ignore );
2086                 }
2087         }
2088
2089         self->audio_used[ index ] = audio_used;
2090
2091         return ret;
2092 }
2093
2094 /** Get the audio from a frame.
2095 */
2096 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2097 {
2098         // Get the producer
2099         producer_avformat self = mlt_frame_pop_audio( frame );
2100
2101         pthread_mutex_lock( &self->audio_mutex );
2102
2103         // Obtain the frame number of this frame
2104         mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" );
2105
2106         // Calculate the real time code
2107         double real_timecode = producer_time_of_frame( self->parent, position );
2108
2109         // Get the producer fps
2110         double fps = mlt_producer_get_fps( self->parent );
2111
2112         // Number of frames to ignore (for ffwd)
2113         int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2114
2115         // Flag for paused (silence)
2116         int paused = seek_audio( self, position, real_timecode, &ignore[0] );
2117
2118         // Initialize ignore for all streams from the seek return value
2119         int i = MAX_AUDIO_STREAMS;
2120         while ( i-- )
2121                 ignore[i] = ignore[0];
2122
2123         // Fetch the audio_format
2124         AVFormatContext *context = self->audio_format;
2125
2126         int sizeof_sample = sizeof( int16_t );
2127
2128         // Determine the tracks to use
2129         int index = self->audio_index;
2130         int index_max = self->audio_index + 1;
2131         if ( self->audio_index == INT_MAX )
2132         {
2133                 index = 0;
2134                 index_max = context->nb_streams;
2135                 *channels = self->total_channels;
2136                 *samples = *samples * FFMAX( self->max_frequency, *frequency ) / *frequency;
2137                 *frequency = FFMAX( self->max_frequency, *frequency );
2138         }
2139
2140         // Initialize the resamplers and buffers
2141         for ( ; index < index_max; index++ )
2142         {
2143                 // Get codec context
2144                 AVCodecContext *codec_context = self->audio_codec[ index ];
2145
2146                 if ( codec_context && !self->audio_buffer[ index ] )
2147                 {
2148                         // Check for resample and create if necessary
2149                         if ( codec_context->channels <= 2 )
2150                         {
2151                                 // Determine by how much resampling will increase number of samples
2152                                 double resample_factor = self->audio_index == INT_MAX ? 1 : (double) *channels / codec_context->channels;
2153                                 resample_factor *= (double) *frequency / codec_context->sample_rate;
2154                                 if ( resample_factor > self->resample_factor )
2155                                         self->resample_factor = resample_factor;
2156
2157                                 // Create the resampler
2158 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0))
2159                                 self->audio_resample[ index ] = av_audio_resample_init(
2160                                         self->audio_index == INT_MAX ? codec_context->channels : *channels,
2161                                         codec_context->channels, *frequency, codec_context->sample_rate,
2162                                         SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 );
2163 #else
2164                                 self->audio_resample[ index ] = audio_resample_init(
2165                                         self->audio_index == INT_MAX ? codec_context->channels : *channels,
2166                                         codec_context->channels, *frequency, codec_context->sample_rate );
2167 #endif
2168                         }
2169                         else
2170                         {
2171                                 codec_context->request_channels = self->audio_index == INT_MAX ? codec_context->channels : *channels;
2172                                 sizeof_sample = sample_bytes( codec_context );
2173                         }
2174
2175                         // Check for audio buffer and create if necessary
2176                         self->audio_buffer_size[ index ] = AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2177                         self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2178
2179                         // Check for decoder buffer and create if necessary
2180                         self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2181                 }
2182         }
2183
2184         // Get the audio if required
2185         if ( !paused )
2186         {
2187                 int ret = 0;
2188                 int got_audio = 0;
2189                 AVPacket pkt;
2190
2191                 av_init_packet( &pkt );
2192
2193                 // If not resampling, give consumer more than requested.
2194                 // It requested number samples based on requested frame rate.
2195                 // Do not clean this up with a samples *= ...!
2196                 if ( self->audio_index != INT_MAX && ! self->audio_resample[ self->audio_index ] )
2197                         *samples = *samples * self->audio_codec[ self->audio_index ]->sample_rate / *frequency;
2198
2199                 while ( ret >= 0 && !got_audio )
2200                 {
2201                         // Check if the buffer already contains the samples required
2202                         if ( self->audio_index != INT_MAX &&
2203                                  self->audio_used[ self->audio_index ] >= *samples &&
2204                                  ignore[ self->audio_index ] == 0 )
2205                         {
2206                                 got_audio = 1;
2207                                 break;
2208                         }
2209                         else if ( self->audio_index == INT_MAX )
2210                         {
2211                                 // Check if there is enough audio for all streams
2212                                 got_audio = 1;
2213                                 for ( index = 0; got_audio && index < context->nb_streams; index++ )
2214                                         if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2215                                                 got_audio = 0;
2216                                 if ( got_audio )
2217                                         break;
2218                         }
2219
2220                         // Read a packet
2221                         pthread_mutex_lock( &self->packets_mutex );
2222                         if ( mlt_deque_count( self->apackets ) )
2223                         {
2224                                 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2225                                 pkt = *tmp;
2226                                 free( tmp );
2227                         }
2228                         else
2229                         {
2230                                 ret = av_read_frame( context, &pkt );
2231                                 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2232                                 {
2233                                         if ( !av_dup_packet( &pkt ) )
2234                                         {
2235                                                 AVPacket *tmp = malloc( sizeof(AVPacket) );
2236                                                 *tmp = pkt;
2237                                                 mlt_deque_push_back( self->vpackets, tmp );
2238                                         }
2239                                 }
2240                         }
2241                         pthread_mutex_unlock( &self->packets_mutex );
2242
2243                         // We only deal with audio from the selected audio index
2244                         index = pkt.stream_index;
2245                         if ( ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2246                                  ( self->audio_index == INT_MAX && context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
2247                         {
2248                                 int channels2 = ( self->audio_index == INT_MAX || !self->audio_resample[index] ) ?
2249                                         self->audio_codec[index]->channels : *channels;
2250                                 ret = decode_audio( self, &ignore[index], pkt, channels2, *samples, real_timecode, fps );
2251                         }
2252
2253                         if ( self->seekable || index != self->video_index )
2254                                 av_free_packet( &pkt );
2255
2256                 }
2257
2258                 // Set some additional return values
2259                 *format = mlt_audio_s16;
2260                 if ( self->audio_index != INT_MAX && !self->audio_resample[ self->audio_index ] )
2261                 {
2262                         index = self->audio_index;
2263                         *channels = self->audio_codec[ index ]->channels;
2264                         *frequency = self->audio_codec[ index ]->sample_rate;
2265                         *format = self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_S32 ? mlt_audio_s32le
2266                                 : self->audio_codec[ index ]->sample_fmt == SAMPLE_FMT_FLT ? mlt_audio_f32le
2267                                 : mlt_audio_s16;
2268                         sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2269                 }
2270                 else if ( self->audio_index == INT_MAX )
2271                 {
2272                         // This only works if all audio tracks have the same sample format.
2273                         for ( index = 0; index < index_max; index++ )
2274                                 if ( self->audio_codec[ index ] && !self->audio_resample[ index ] )
2275                                 {
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
2278                                                 : mlt_audio_s16;
2279                                         sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2280                                         break;
2281                                 }
2282                 }
2283
2284                 // Allocate and set the frame's audio buffer
2285                 int size = mlt_audio_format_size( *format, *samples, *channels );
2286                 *buffer = mlt_pool_alloc( size );
2287                 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2288
2289                 // Interleave tracks if audio_index=all
2290                 if ( self->audio_index == INT_MAX )
2291                 {
2292                         uint8_t *dest = *buffer;
2293                         int i;
2294                         for ( i = 0; i < *samples; i++ )
2295                         {
2296                                 for ( index = 0; index < index_max; index++ )
2297                                 if ( self->audio_codec[ index ] )
2298                                 {
2299                                         int current_channels = self->audio_codec[ index ]->channels;
2300                                         uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2301                                         memcpy( dest, src, current_channels * sizeof_sample );
2302                                         dest += current_channels * sizeof_sample;
2303                                 }
2304                         }
2305                         for ( index = 0; index < index_max; index++ )
2306                         if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2307                         {
2308                                 int current_channels = self->audio_codec[ index ]->channels;
2309                                 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2310                                 self->audio_used[index] -= *samples;
2311                                 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2312                         }
2313                 }
2314                 // Copy a single track to the output buffer
2315                 else
2316                 {
2317                         index = self->audio_index;
2318
2319                         // Now handle the audio if we have enough
2320                         if ( self->audio_used[ index ] > 0 )
2321                         {
2322                                 uint8_t *src = self->audio_buffer[ index ];
2323                                 // copy samples from audio_buffer
2324                                 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2325                                 memcpy( *buffer, src, size * *channels * sizeof_sample );
2326                                 // supply the remaining requested samples as silence
2327                                 if ( *samples > self->audio_used[ index ] )
2328                                         memset( *buffer + size * *channels * sizeof_sample, 0, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2329                                 // reposition the samples within audio_buffer
2330                                 self->audio_used[ index ] -= size;
2331                                 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2332                         }
2333                         else
2334                         {
2335                                 // Otherwise fill with silence
2336                                 memset( *buffer, 0, *samples * *channels * sizeof_sample );
2337                         }
2338                 }
2339         }
2340         else
2341         {
2342                 // Get silence and don't touch the context
2343                 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2344         }
2345
2346         // Regardless of speed (other than paused), we expect to get the next frame
2347         if ( !paused )
2348                 self->audio_expected = position + 1;
2349
2350         pthread_mutex_unlock( &self->audio_mutex );
2351
2352         return 0;
2353 }
2354
2355 /** Initialize the audio codec context.
2356 */
2357
2358 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2359 {
2360         // Initialise the codec if necessary
2361         if ( !self->audio_codec[ index ] )
2362         {
2363                 // Get codec context
2364                 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2365
2366                 // Find the codec
2367                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2368
2369                 // If we don't have a codec and we can't initialise it, we can't do much more...
2370                 avformat_lock( );
2371                 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
2372                 {
2373                         // Now store the codec with its destructor
2374                         if ( self->audio_codec[ index ] )
2375                                 avcodec_close( self->audio_codec[ index ] );
2376                         self->audio_codec[ index ] = codec_context;
2377                 }
2378                 else
2379                 {
2380                         // Remember that we can't use self later
2381                         self->audio_index = -1;
2382                 }
2383                 avformat_unlock( );
2384
2385                 // Process properties as AVOptions
2386                 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2387 #if LIBAVCODEC_VERSION_MAJOR > 52
2388                 if ( codec && codec->priv_class && codec_context->priv_data )
2389                         apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2390 #endif
2391         }
2392         return self->audio_codec[ index ] && self->audio_index > -1;
2393 }
2394
2395 /** Set up audio handling.
2396 */
2397
2398 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2399 {
2400         // Get the producer
2401         mlt_producer producer = self->parent;
2402
2403         // Get the properties
2404         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2405
2406         // Fetch the audio format context
2407         AVFormatContext *context = self->audio_format;
2408
2409         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2410
2411         // Get the audio_index
2412         int index = mlt_properties_get_int( properties, "audio_index" );
2413
2414         // Handle all audio tracks
2415         if ( self->audio_index > -1 &&
2416              mlt_properties_get( properties, "audio_index" ) &&
2417              !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2418                 index = INT_MAX;
2419
2420         // Reopen the file if necessary
2421         if ( !context && self->audio_index > -1 && index > -1 )
2422         {
2423                 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2424                         mlt_properties_get( properties, "resource" ), 1 );
2425                 context = self->audio_format;
2426         }
2427
2428         // Exception handling for audio_index
2429         if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2430         {
2431                 for ( index = context->nb_streams - 1;
2432                           index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
2433                           index-- );
2434                 mlt_properties_set_int( properties, "audio_index", index );
2435         }
2436         if ( context && index > -1 && index < INT_MAX &&
2437                  context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
2438         {
2439                 index = self->audio_index;
2440                 mlt_properties_set_int( properties, "audio_index", index );
2441         }
2442
2443         // Update the audio properties if the index changed
2444         if ( context && index > -1 && index != self->audio_index )
2445         {
2446                 if ( self->audio_codec[ self->audio_index ] )
2447                 {
2448                         avformat_lock();
2449                         avcodec_close( self->audio_codec[ self->audio_index ] );
2450                         avformat_unlock();
2451                 }
2452                 self->audio_codec[ self->audio_index ] = NULL;
2453         }
2454         if ( self->audio_index != -1 )
2455                 self->audio_index = index;
2456         else
2457                 index = -1;
2458
2459         // Get the codec(s)
2460         if ( context && index == INT_MAX )
2461         {
2462                 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
2463                 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
2464                 for ( index = 0; index < context->nb_streams; index++ )
2465                 {
2466                         if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
2467                                 audio_codec_init( self, index, properties );
2468                 }
2469         }
2470         else if ( context && index > -1 && audio_codec_init( self, index, properties ) )
2471         {
2472                 // Set the frame properties
2473                 if ( index < INT_MAX )
2474                 {
2475                         mlt_properties_set_int( frame_properties, "frequency", self->audio_codec[ index ]->sample_rate );
2476                         mlt_properties_set_int( frame_properties, "channels", self->audio_codec[ index ]->channels );
2477                 }
2478         }
2479         if ( context && index > -1 )
2480         {
2481                 // Add our audio operation
2482                 mlt_frame_push_audio( frame, self );
2483                 mlt_frame_push_audio( frame, producer_get_audio );
2484         }
2485 }
2486
2487 /** Our get frame implementation.
2488 */
2489
2490 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
2491 {
2492         // Access the private data
2493         mlt_service service = MLT_PRODUCER_SERVICE( producer );
2494         mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
2495         producer_avformat self = mlt_cache_item_data( cache_item, NULL );
2496
2497         // If cache miss
2498         if ( !self )
2499         {
2500                 self = calloc( 1, sizeof( struct producer_avformat_s ) );
2501                 producer->child = self;
2502                 self->parent = producer;
2503                 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
2504                 cache_item = mlt_service_cache_get( service, "producer_avformat" );
2505         }
2506
2507         // Create an empty frame
2508         *frame = mlt_frame_init( service);
2509
2510         if ( *frame )
2511         {
2512                 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
2513         }
2514         else
2515         {
2516                 mlt_cache_item_close( cache_item );
2517                 return 1;
2518         }
2519
2520         // Update timecode on the frame we're creating
2521         mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
2522
2523         // Set the position of this producer
2524         mlt_position position = self->seekable ? mlt_producer_frame( producer ) : self->nonseek_position++;
2525         mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", position );
2526
2527         // Set up the video
2528         producer_set_up_video( self, *frame );
2529
2530         // Set up the audio
2531         producer_set_up_audio( self, *frame );
2532
2533         // Calculate the next timecode
2534         mlt_producer_prepare_next( producer );
2535
2536         return 0;
2537 }
2538
2539 static void producer_avformat_close( producer_avformat self )
2540 {
2541         mlt_log_debug( NULL, "producer_avformat_close\n" );
2542
2543         // Cleanup av contexts
2544         av_free( self->av_frame );
2545         avformat_lock();
2546         int i;
2547         for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
2548         {
2549                 if ( self->audio_resample[i] )
2550                         audio_resample_close( self->audio_resample[i] );
2551                 mlt_pool_release( self->audio_buffer[i] );
2552                 av_free( self->decode_buffer[i] );
2553                 if ( self->audio_codec[i] )
2554                         avcodec_close( self->audio_codec[i] );
2555         }
2556         if ( self->video_codec )
2557                 avcodec_close( self->video_codec );
2558         // Close the file
2559         if ( self->dummy_context )
2560                 av_close_input_file( self->dummy_context );
2561         if ( self->seekable && self->audio_format )
2562                 av_close_input_file( self->audio_format );
2563         if ( self->video_format )
2564                 av_close_input_file( self->video_format );
2565         avformat_unlock();
2566 #ifdef VDPAU
2567         vdpau_producer_close( self );
2568 #endif
2569         if ( self->image_cache )
2570                 mlt_cache_close( self->image_cache );
2571
2572         // Cleanup the mutexes
2573         pthread_mutex_destroy( &self->audio_mutex );
2574         pthread_mutex_destroy( &self->video_mutex );
2575         pthread_mutex_destroy( &self->packets_mutex );
2576
2577         // Cleanup the packet queues
2578         AVPacket *pkt;
2579         while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
2580         {
2581                 av_free_packet( pkt );
2582                 free( pkt );
2583         }
2584         while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
2585         {
2586                 av_free_packet( pkt );
2587                 free( pkt );
2588         }
2589
2590         free( self );
2591 }
2592
2593 static void producer_close( mlt_producer parent )
2594 {
2595         // Remove this instance from the cache
2596         mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
2597
2598         // Close the parent
2599         parent->close = NULL;
2600         mlt_producer_close( parent );
2601
2602         // Free the memory
2603         free( parent );
2604 }