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

   1 /*
   2  * producer_avformat.c -- avformat producer
   3  * Copyright (C) 2003-2004 Ushodaya Enterprises Limited
   4  * Author: Charles Yates <charles.yates@pandora.be>
   5  * Much code borrowed from ffmpeg.c: Copyright (c) 2000-2003 Fabrice Bellard
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  */
  21
  22 // MLT Header files
  23 #include <framework/mlt_producer.h>
  24 #include <framework/mlt_frame.h>
  25 #include <framework/mlt_profile.h>
  26 #include <framework/mlt_log.h>
  27
  28 // ffmpeg Header files
  29 #include <avformat.h>
  30 #include <opt.h>
  31 #ifdef SWSCALE
  32 #  include <swscale.h>
  33 #endif
  34 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
  35 #  include "audioconvert.h"
  36 #endif
  37
  38 // System header files
  39 #include <stdlib.h>
  40 #include <string.h>
  41 #include <pthread.h>
  42 #include <limits.h>
  43
  44 #if LIBAVUTIL_VERSION_INT < (50<<16)
  45 #define PIX_FMT_RGB32 PIX_FMT_RGBA32
  46 #define PIX_FMT_YUYV422 PIX_FMT_YUV422
  47 #endif
  48
  49 #define POSITION_INITIAL (-2)
  50 #define POSITION_INVALID (-1)
  51
  52 #define MAX_AUDIO_STREAMS (8)
  53
  54 void avformat_lock( );
  55 void avformat_unlock( );
  56
  57 struct producer_avformat_s
  58 {
  59         struct mlt_producer_s parent;
  60         AVFormatContext *dummy_context;
  61         AVFormatContext *audio_format;
  62         AVFormatContext *video_format;
  63         AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
  64         AVCodecContext *video_codec;
  65         AVFrame *av_frame;
  66         ReSampleContext *audio_resample[ MAX_AUDIO_STREAMS ];
  67         mlt_position audio_expected;
  68         mlt_position video_expected;
  69         int audio_index;
  70         int video_index;
  71         double start_time;
  72         int first_pts;
  73         int last_position;
  74         int seekable;
  75         int current_position;
  76         int got_picture;
  77         int top_field_first;
  78         int16_t *audio_buffer[ MAX_AUDIO_STREAMS ];
  79         int16_t *decode_buffer[ MAX_AUDIO_STREAMS ];
  80         int audio_used[ MAX_AUDIO_STREAMS ];
  81         int audio_streams;
  82         int audio_max_stream;
  83         int total_channels;
  84         int max_channel;
  85         int max_frequency;
  86 };
  87 typedef struct producer_avformat_s *producer_avformat;
  88
  89 // Forward references.
  90 static int producer_open( producer_avformat this, mlt_profile profile, char *file );
  91 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index );
  92 static void producer_format_close( void *context );
  93 static void producer_close( mlt_producer parent );
  94
  95 /** Constructor for libavformat.
  96 */
  97
  98 mlt_producer producer_avformat_init( mlt_profile profile, char *file )
  99 {
 100         int skip = 0;
 101
 102         // Report information about available demuxers and codecs as YAML Tiny
 103         if ( file && strstr( file, "f-list" ) )
 104         {
 105                 fprintf( stderr, "---\nformats:\n" );
 106                 AVInputFormat *format = NULL;
 107                 while ( ( format = av_iformat_next( format ) ) )
 108                         fprintf( stderr, "  - %s\n", format->name );
 109                 fprintf( stderr, "...\n" );
 110                 skip = 1;
 111         }
 112         if ( file && strstr( file, "acodec-list" ) )
 113         {
 114                 fprintf( stderr, "---\naudio_codecs:\n" );
 115                 AVCodec *codec = NULL;
 116                 while ( ( codec = av_codec_next( codec ) ) )
 117                         if ( codec->decode && codec->type == CODEC_TYPE_AUDIO )
 118                                 fprintf( stderr, "  - %s\n", codec->name );
 119                 fprintf( stderr, "...\n" );
 120                 skip = 1;
 121         }
 122         if ( file && strstr( file, "vcodec-list" ) )
 123         {
 124                 fprintf( stderr, "---\nvideo_codecs:\n" );
 125                 AVCodec *codec = NULL;
 126                 while ( ( codec = av_codec_next( codec ) ) )
 127                         if ( codec->decode && codec->type == CODEC_TYPE_VIDEO )
 128                                 fprintf( stderr, "  - %s\n", codec->name );
 129                 fprintf( stderr, "...\n" );
 130                 skip = 1;
 131         }
 132
 133         // Check that we have a non-NULL argument
 134         if ( !skip && file )
 135         {
 136                 // Construct the producer
 137                 producer_avformat this = calloc( 1, sizeof( struct producer_avformat_s ) );
 138
 139                 // Initialise it
 140                 if ( mlt_producer_init( &this->parent, this ) == 0 )
 141                 {
 142                         mlt_producer producer = &this->parent;
 143
 144                         // Get the properties
 145                         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
 146
 147                         // Set the resource property (required for all producers)
 148                         mlt_properties_set( properties, "resource", file );
 149
 150                         // Register transport implementation with the producer
 151                         producer->close = (mlt_destructor) producer_close;
 152
 153                         // Register our get_frame implementation
 154                         producer->get_frame = producer_get_frame;
 155
 156                         // Open the file
 157                         if ( producer_open( this, profile, file ) != 0 )
 158                         {
 159                                 // Clean up
 160                                 mlt_producer_close( producer );
 161                                 this = NULL;
 162                         }
 163                         else
 164                         {
 165                                 // Close the file to release resources for large playlists - reopen later as needed
 166                                 producer_format_close( this->dummy_context );
 167                                 this->dummy_context = NULL;
 168                                 producer_format_close( this->audio_format );
 169                                 this->audio_format = NULL;
 170                                 producer_format_close( this->video_format );
 171                                 this->video_format = NULL;
 172
 173                                 // Default the user-selectable indices from the auto-detected indices
 174                                 mlt_properties_set_int( properties, "audio_index",  this->audio_index );
 175                                 mlt_properties_set_int( properties, "video_index",  this->video_index );
 176                         }
 177                         return producer;
 178                 }
 179         }
 180         return NULL;
 181 }
 182
 183 /** Find the default streams.
 184 */
 185
 186 static mlt_properties find_default_streams( mlt_properties meta_media, AVFormatContext *context, int *audio_index, int *video_index )
 187 {
 188         int i;
 189         char key[200];
 190
 191         mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
 192
 193         // Allow for multiple audio and video streams in the file and select first of each (if available)
 194         for( i = 0; i < context->nb_streams; i++ )
 195         {
 196                 // Get the codec context
 197                 AVStream *stream = context->streams[ i ];
 198                 if ( ! stream ) continue;
 199                 AVCodecContext *codec_context = stream->codec;
 200                 if ( ! codec_context ) continue;
 201                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
 202                 if ( ! codec ) continue;
 203
 204                 snprintf( key, sizeof(key), "meta.media.%d.stream.type", i );
 205
 206                 // Determine the type and obtain the first index of each type
 207                 switch( codec_context->codec_type )
 208                 {
 209                         case CODEC_TYPE_VIDEO:
 210                                 if ( *video_index < 0 )
 211                                         *video_index = i;
 212                                 mlt_properties_set( meta_media, key, "video" );
 213                                 snprintf( key, sizeof(key), "meta.media.%d.stream.frame_rate", i );
 214                                 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->r_frame_rate ) );
 215 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 216                                 snprintf( key, sizeof(key), "meta.media.%d.stream.sample_aspect_ratio", i );
 217                                 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
 218 #endif
 219                                 snprintf( key, sizeof(key), "meta.media.%d.codec.frame_rate", i );
 220                                 mlt_properties_set_double( meta_media, key, (double) codec_context->time_base.den /
 221                                                                                    ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num ) );
 222                                 snprintf( key, sizeof(key), "meta.media.%d.codec.pix_fmt", i );
 223                                 mlt_properties_set( meta_media, key, avcodec_get_pix_fmt_name( codec_context->pix_fmt ) );
 224                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_aspect_ratio", i );
 225                                 mlt_properties_set_double( meta_media, key, av_q2d( codec_context->sample_aspect_ratio ) );
 226                                 break;
 227                         case CODEC_TYPE_AUDIO:
 228                                 if ( *audio_index < 0 )
 229                                         *audio_index = i;
 230                                 mlt_properties_set( meta_media, key, "audio" );
 231 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(71<<8)+0))
 232                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_fmt", i );
 233                                 mlt_properties_set( meta_media, key, avcodec_get_sample_fmt_name( codec_context->sample_fmt ) );
 234 #endif
 235                                 snprintf( key, sizeof(key), "meta.media.%d.codec.sample_rate", i );
 236                                 mlt_properties_set_int( meta_media, key, codec_context->sample_rate );
 237                                 snprintf( key, sizeof(key), "meta.media.%d.codec.channels", i );
 238                                 mlt_properties_set_int( meta_media, key, codec_context->channels );
 239                                 break;
 240                         default:
 241                                 break;
 242                 }
 243 //              snprintf( key, sizeof(key), "meta.media.%d.stream.time_base", i );
 244 //              mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
 245                 snprintf( key, sizeof(key), "meta.media.%d.codec.name", i );
 246                 mlt_properties_set( meta_media, key, codec->name );
 247 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(55<<8)+0))
 248                 snprintf( key, sizeof(key), "meta.media.%d.codec.long_name", i );
 249                 mlt_properties_set( meta_media, key, codec->long_name );
 250 #endif
 251                 snprintf( key, sizeof(key), "meta.media.%d.codec.bit_rate", i );
 252                 mlt_properties_set_int( meta_media, key, codec_context->bit_rate );
 253 //              snprintf( key, sizeof(key), "meta.media.%d.codec.time_base", i );
 254 //              mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
 255 //              snprintf( key, sizeof(key), "meta.media.%d.codec.profile", i );
 256 //              mlt_properties_set_int( meta_media, key, codec_context->profile );
 257 //              snprintf( key, sizeof(key), "meta.media.%d.codec.level", i );
 258 //              mlt_properties_set_int( meta_media, key, codec_context->level );
 259         }
 260
 261         return meta_media;
 262 }
 263
 264 /** Producer file destructor.
 265 */
 266
 267 static void producer_format_close( void *context )
 268 {
 269         if ( context )
 270         {
 271                 // Lock the mutex now
 272                 avformat_lock( );
 273
 274                 // Close the file
 275                 av_close_input_file( context );
 276
 277                 // Unlock the mutex now
 278                 avformat_unlock( );
 279         }
 280 }
 281
 282 /** Producer file destructor.
 283 */
 284
 285 static void producer_codec_close( void *codec )
 286 {
 287         if ( codec )
 288         {
 289                 // Lock the mutex now
 290                 avformat_lock( );
 291
 292                 // Close the file
 293                 avcodec_close( codec );
 294
 295                 // Unlock the mutex now
 296                 avformat_unlock( );
 297         }
 298 }
 299
 300 static inline int dv_is_pal( AVPacket *pkt )
 301 {
 302         return pkt->data[3] & 0x80;
 303 }
 304
 305 static int dv_is_wide( AVPacket *pkt )
 306 {
 307         int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
 308
 309         for ( ; i < pkt->size; i += 5 /* packet size */ )
 310         {
 311                 if ( pkt->data[ i ] == 0x61 )
 312                 {
 313                         uint8_t x = pkt->data[ i + 2 ] & 0x7;
 314                         return ( x == 2 ) || ( x == 7 );
 315                 }
 316         }
 317         return 0;
 318 }
 319
 320 static double get_aspect_ratio( AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
 321 {
 322         double aspect_ratio = 1.0;
 323
 324         if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
 325         {
 326                 if ( pkt )
 327                 {
 328                         if ( dv_is_pal( pkt ) )
 329                         {
 330                                 aspect_ratio = dv_is_wide( pkt )
 331                                         ? 64.0/45.0 // 16:9 PAL
 332                                         : 16.0/15.0; // 4:3 PAL
 333                         }
 334                         else
 335                         {
 336                                 aspect_ratio = dv_is_wide( pkt )
 337                                         ? 32.0/27.0 // 16:9 NTSC
 338                                         : 8.0/9.0; // 4:3 NTSC
 339                         }
 340                 }
 341                 else
 342                 {
 343                         AVRational ar =
 344 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 345                                 stream->sample_aspect_ratio;
 346 #else
 347                                 codec_context->sample_aspect_ratio;
 348 #endif
 349                         // Override FFmpeg's notion of DV aspect ratios, which are
 350                         // based upon a width of 704. Since we do not have a normaliser
 351                         // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
 352                         // we just coerce the values to facilitate a passive behaviour through
 353                         // the rescale normaliser when using equivalent producers and consumers.
 354                         // = display_aspect / (width * height)
 355                         if ( ar.num == 10 && ar.den == 11 )
 356                                 aspect_ratio = 8.0/9.0; // 4:3 NTSC
 357                         else if ( ar.num == 59 && ar.den == 54 )
 358                                 aspect_ratio = 16.0/15.0; // 4:3 PAL
 359                         else if ( ar.num == 40 && ar.den == 33 )
 360                                 aspect_ratio = 32.0/27.0; // 16:9 NTSC
 361                         else if ( ar.num == 118 && ar.den == 81 )
 362                                 aspect_ratio = 64.0/45.0; // 16:9 PAL
 363                 }
 364         }
 365         else
 366         {
 367                 AVRational codec_sar = codec_context->sample_aspect_ratio;
 368                 AVRational stream_sar =
 369 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
 370                         stream->sample_aspect_ratio;
 371 #else
 372                         { 0, 1 };
 373 #endif
 374                 if ( codec_sar.num > 0 )
 375                         aspect_ratio = av_q2d( codec_sar );
 376                 else if ( stream_sar.num > 0 )
 377                         aspect_ratio = av_q2d( stream_sar );
 378         }
 379         return aspect_ratio;
 380 }
 381
 382 /** Open the file.
 383 */
 384
 385 static int producer_open( producer_avformat this, mlt_profile profile, char *file )
 386 {
 387         // Return an error code (0 == no error)
 388         int error = 0;
 389
 390         // Context for avformat
 391         AVFormatContext *context = NULL;
 392
 393         // Get the properties
 394         mlt_properties properties = MLT_PRODUCER_PROPERTIES( &this->parent );
 395
 396         // We will treat everything with the producer fps
 397         double fps = mlt_profile_fps( profile );
 398
 399         // Lock the mutex now
 400         avformat_lock( );
 401
 402         // If "MRL", then create AVInputFormat
 403         AVInputFormat *format = NULL;
 404         AVFormatParameters *params = NULL;
 405         char *standard = NULL;
 406         char *mrl = strchr( file, ':' );
 407
 408         // AV option (0 = both, 1 = video, 2 = audio)
 409         int av = 0;
 410
 411         // Only if there is not a protocol specification that avformat can handle
 412         if ( mrl && !url_exist( file ) )
 413         {
 414                 // 'file' becomes format abbreviation
 415                 mrl[0] = 0;
 416
 417                 // Lookup the format
 418                 format = av_find_input_format( file );
 419
 420                 // Eat the format designator
 421                 file = ++mrl;
 422
 423                 if ( format )
 424                 {
 425                         // Allocate params
 426                         params = calloc( sizeof( AVFormatParameters ), 1 );
 427
 428                         // These are required by video4linux (defaults)
 429                         params->width = 640;
 430                         params->height = 480;
 431                         params->time_base= (AVRational){1,25};
 432                         // params->device = file;
 433                         params->channels = 2;
 434                         params->sample_rate = 48000;
 435                 }
 436
 437                 // XXX: this does not work anymore since avdevice
 438                 // TODO: make producer_avddevice?
 439                 // Parse out params
 440                 mrl = strchr( file, '?' );
 441                 while ( mrl )
 442                 {
 443                         mrl[0] = 0;
 444                         char *name = strdup( ++mrl );
 445                         char *value = strchr( name, ':' );
 446                         if ( value )
 447                         {
 448                                 value[0] = 0;
 449                                 value++;
 450                                 char *t = strchr( value, '&' );
 451                                 if ( t )
 452                                         t[0] = 0;
 453                                 if ( !strcmp( name, "frame_rate" ) )
 454                                         params->time_base.den = atoi( value );
 455                                 else if ( !strcmp( name, "frame_rate_base" ) )
 456                                         params->time_base.num = atoi( value );
 457                                 else if ( !strcmp( name, "sample_rate" ) )
 458                                         params->sample_rate = atoi( value );
 459                                 else if ( !strcmp( name, "channels" ) )
 460                                         params->channels = atoi( value );
 461                                 else if ( !strcmp( name, "width" ) )
 462                                         params->width = atoi( value );
 463                                 else if ( !strcmp( name, "height" ) )
 464                                         params->height = atoi( value );
 465                                 else if ( !strcmp( name, "standard" ) )
 466                                 {
 467                                         standard = strdup( value );
 468                                         params->standard = standard;
 469                                 }
 470                                 else if ( !strcmp( name, "av" ) )
 471                                         av = atoi( value );
 472                         }
 473                         free( name );
 474                         mrl = strchr( mrl, '&' );
 475                 }
 476         }
 477
 478         // Now attempt to open the file
 479         error = av_open_input_file( &context, file, format, 0, params ) < 0;
 480
 481         // Cleanup AVFormatParameters
 482         free( standard );
 483         free( params );
 484
 485         // If successful, then try to get additional info
 486         if ( !error )
 487         {
 488                 // Get the stream info
 489                 error = av_find_stream_info( context ) < 0;
 490
 491                 // Continue if no error
 492                 if ( !error )
 493                 {
 494                         // We will default to the first audio and video streams found
 495                         int audio_index = -1;
 496                         int video_index = -1;
 497
 498                         // Now set properties where we can (use default unknowns if required)
 499                         if ( context->duration != AV_NOPTS_VALUE )
 500                         {
 501                                 // This isn't going to be accurate for all formats
 502                                 mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps + 0.5 );
 503                                 mlt_properties_set_position( properties, "out", frames - 1 );
 504                                 mlt_properties_set_position( properties, "length", frames );
 505                         }
 506
 507                         // Find default audio and video streams
 508                         find_default_streams( properties, context, &audio_index, &video_index );
 509
 510                         if ( context->start_time != AV_NOPTS_VALUE )
 511                                 this->start_time = context->start_time;
 512
 513                         // Check if we're seekable (something funny about mpeg here :-/)
 514                         if ( strncmp( file, "pipe:", 5 ) &&
 515                                  strncmp( file, "http:", 5 ) &&
 516                                  strncmp( file, "udp:", 4 )  &&
 517                                  strncmp( file, "tcp:", 4 )  &&
 518                                  strncmp( file, "rtsp:", 5 ) &&
 519                                  strncmp( file, "rtp:", 4 ) )
 520                         {
 521                                 this->seekable = av_seek_frame( context, -1, this->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
 522                                 mlt_properties_set_int( properties, "seekable", this->seekable );
 523                                 producer_format_close( this->dummy_context );
 524                                 this->dummy_context = context;
 525                                 av_open_input_file( &context, file, NULL, 0, NULL );
 526                                 av_find_stream_info( context );
 527                         }
 528
 529                         // Store selected audio and video indexes on properties
 530                         this->audio_index = audio_index;
 531                         this->video_index = video_index;
 532                         this->first_pts = -1;
 533                         this->last_position = POSITION_INITIAL;
 534
 535                         // Fetch the width, height and aspect ratio
 536                         if ( video_index != -1 )
 537                         {
 538                                 AVCodecContext *codec_context = context->streams[ video_index ]->codec;
 539                                 mlt_properties_set_int( properties, "width", codec_context->width );
 540                                 mlt_properties_set_int( properties, "height", codec_context->height );
 541
 542                                 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
 543                                 {
 544                                         // Fetch the first frame of DV so we can read it directly
 545                                         AVPacket pkt;
 546                                         int ret = 0;
 547                                         while ( ret >= 0 )
 548                                         {
 549                                                 ret = av_read_frame( context, &pkt );
 550                                                 if ( ret >= 0 && pkt.stream_index == video_index && pkt.size > 0 )
 551                                                 {
 552                                                         mlt_properties_set_double( properties, "aspect_ratio",
 553                                                                 get_aspect_ratio( context->streams[ video_index ], codec_context, &pkt ) );
 554                                                         break;
 555                                                 }
 556                                         }
 557                                 }
 558                                 else
 559                                 {
 560                                         mlt_properties_set_double( properties, "aspect_ratio",
 561                                                 get_aspect_ratio( context->streams[ video_index ], codec_context, NULL ) );
 562                                 }
 563                         }
 564
 565                         // Read Metadata
 566                         if ( context->title )
 567                                 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
 568                         if ( context->author )
 569                                 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
 570                         if ( context->copyright )
 571                                 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
 572                         if ( context->comment )
 573                                 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
 574                         if ( context->album )
 575                                 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
 576                         if ( context->year )
 577                                 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
 578                         if ( context->track )
 579                                 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
 580
 581                         // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
 582                         if ( av == 0 && audio_index != -1 && video_index != -1 )
 583                         {
 584                                 // We'll use the open one as our video_format
 585                                 avformat_unlock();
 586                                 producer_format_close( this->video_format );
 587                                 avformat_lock();
 588                                 this->video_format = context;
 589
 590                                 // And open again for our audio context
 591                                 av_open_input_file( &context, file, NULL, 0, NULL );
 592                                 av_find_stream_info( context );
 593
 594                                 // Audio context
 595                                 avformat_unlock();
 596                                 producer_format_close( this->audio_format );
 597                                 avformat_lock();
 598                                 this->audio_format = context;
 599                         }
 600                         else if ( av != 2 && video_index != -1 )
 601                         {
 602                                 // We only have a video context
 603                                 avformat_unlock();
 604                                 producer_format_close( this->video_format );
 605                                 avformat_lock();
 606                                 this->video_format = context;
 607                         }
 608                         else if ( audio_index != -1 )
 609                         {
 610                                 // We only have an audio context
 611                                 avformat_unlock();
 612                                 producer_format_close( this->audio_format );
 613                                 avformat_lock();
 614                                 this->audio_format = context;
 615                         }
 616                         else
 617                         {
 618                                 // Something has gone wrong
 619                                 error = -1;
 620                         }
 621                 }
 622         }
 623
 624         // Unlock the mutex now
 625         avformat_unlock( );
 626
 627         return error;
 628 }
 629
 630 /** Convert a frame position to a time code.
 631 */
 632
 633 static double producer_time_of_frame( mlt_producer this, mlt_position position )
 634 {
 635         return ( double )position / mlt_producer_get_fps( this );
 636 }
 637
 638                 // Collect information about all audio streams
 639
 640 static void get_audio_streams_info( producer_avformat this )
 641 {
 642         // Fetch the audio format context
 643         AVFormatContext *context = this->audio_format;
 644         int i;
 645
 646         for ( i = 0;
 647                   i < context->nb_streams;
 648                   i++ )
 649         {
 650                 if ( context->streams[i]->codec->codec_type == CODEC_TYPE_AUDIO )
 651                 {
 652                         AVCodecContext *codec_context = context->streams[i]->codec;
 653                         AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
 654
 655                         // If we don't have a codec and we can't initialise it, we can't do much more...
 656                         avformat_lock( );
 657                         if ( codec && avcodec_open( codec_context, codec ) >= 0 )
 658                         {
 659                                 this->audio_streams++;
 660                                 this->audio_max_stream = i;
 661                                 this->total_channels += codec_context->channels;
 662                                 if ( codec_context->channels > this->max_channel )
 663                                         this->max_channel = codec_context->channels;
 664                                 if ( codec_context->sample_rate > this->max_frequency )
 665                                         this->max_frequency = codec_context->sample_rate;
 666                                 avcodec_close( codec_context );
 667                         }
 668                         avformat_unlock( );
 669                 }
 670         }
 671         mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
 672                 this->audio_streams, this->audio_max_stream, this->total_channels, this->max_channel );
 673 }
 674
 675 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, mlt_image_format *format, int width, int height )
 676 {
 677 #ifdef SWSCALE
 678         if ( pix_fmt == PIX_FMT_RGB32 )
 679         {
 680                 *format = mlt_image_rgb24a;
 681                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
 682                         width, height, PIX_FMT_RGBA, SWS_FAST_BILINEAR, NULL, NULL, NULL);
 683                 AVPicture output;
 684                 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
 685                 sws_scale( context, frame->data, frame->linesize, 0, height,
 686                         output.data, output.linesize);
 687                 sws_freeContext( context );
 688         }
 689         else if ( *format == mlt_image_yuv420p )
 690         {
 691                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
 692                         width, height, PIX_FMT_YUV420P, SWS_FAST_BILINEAR, NULL, NULL, NULL);
 693                 AVPicture output;
 694                 output.data[0] = buffer;
 695                 output.data[1] = buffer + width * height;
 696                 output.data[2] = buffer + ( 3 * width * height ) / 2;
 697                 output.linesize[0] = width;
 698                 output.linesize[1] = width >> 1;
 699                 output.linesize[2] = width >> 1;
 700                 sws_scale( context, frame->data, frame->linesize, 0, height,
 701                         output.data, output.linesize);
 702                 sws_freeContext( context );
 703         }
 704         else if ( *format == mlt_image_rgb24 )
 705         {
 706                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
 707                         width, height, PIX_FMT_RGB24, SWS_FAST_BILINEAR, NULL, NULL, NULL);
 708                 AVPicture output;
 709                 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
 710                 sws_scale( context, frame->data, frame->linesize, 0, height,
 711                         output.data, output.linesize);
 712                 sws_freeContext( context );
 713         }
 714         else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
 715         {
 716                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
 717                         width, height, PIX_FMT_RGBA, SWS_FAST_BILINEAR, NULL, NULL, NULL);
 718                 AVPicture output;
 719                 avpicture_fill( &output, buffer, PIX_FMT_RGBA, width, height );
 720                 sws_scale( context, frame->data, frame->linesize, 0, height,
 721                         output.data, output.linesize);
 722                 sws_freeContext( context );
 723         }
 724         else
 725         {
 726                 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
 727                         width, height, PIX_FMT_YUYV422, SWS_FAST_BILINEAR, NULL, NULL, NULL);
 728                 AVPicture output;
 729                 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
 730                 sws_scale( context, frame->data, frame->linesize, 0, height,
 731                         output.data, output.linesize);
 732                 sws_freeContext( context );
 733         }
 734 #else
 735         if ( *format == mlt_image_yuv420p )
 736         {
 737                 AVPicture pict;
 738                 pict.data[0] = buffer;
 739                 pict.data[1] = buffer + width * height;
 740                 pict.data[2] = buffer + ( 3 * width * height ) / 2;
 741                 pict.linesize[0] = width;
 742                 pict.linesize[1] = width >> 1;
 743                 pict.linesize[2] = width >> 1;
 744                 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
 745         }
 746         else if ( *format == mlt_image_rgb24 )
 747         {
 748                 AVPicture output;
 749                 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
 750                 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
 751         }
 752         else if ( format == mlt_image_rgb24a || format == mlt_image_opengl )
 753         {
 754                 AVPicture output;
 755                 avpicture_fill( &output, buffer, PIX_FMT_RGB32, width, height );
 756                 img_convert( &output, PIX_FMT_RGB32, (AVPicture *)frame, pix_fmt, width, height );
 757         }
 758         else
 759         {
 760                 AVPicture output;
 761                 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
 762                 img_convert( &output, PIX_FMT_YUYV422, (AVPicture *)frame, pix_fmt, width, height );
 763         }
 764 #endif
 765 }
 766
 767 /** Allocate the image buffer and set it on the frame.
 768 */
 769
 770 static int allocate_buffer( mlt_properties frame_properties, AVCodecContext *codec_context, uint8_t **buffer, mlt_image_format *format, int *width, int *height )
 771 {
 772         int size = 0;
 773
 774         if ( codec_context->width == 0 || codec_context->height == 0 )
 775                 return size;
 776
 777         *width = codec_context->width;
 778         *height = codec_context->height;
 779         mlt_properties_set_int( frame_properties, "width", *width );
 780         mlt_properties_set_int( frame_properties, "height", *height );
 781
 782         if ( codec_context->pix_fmt == PIX_FMT_RGB32 )
 783                 size = *width * ( *height + 1 ) * 4;
 784         else switch ( *format )
 785         {
 786                 case mlt_image_yuv420p:
 787                         size = *width * 3 * ( *height + 1 ) / 2;
 788                         break;
 789                 case mlt_image_rgb24:
 790                         size = *width * ( *height + 1 ) * 3;
 791                         break;
 792                 case mlt_image_rgb24a:
 793                 case mlt_image_opengl:
 794                         size = *width * ( *height + 1 ) * 4;
 795                         break;
 796                 default:
 797                         *format = mlt_image_yuv422;
 798                         size = *width * ( *height + 1 ) * 2;
 799                         break;
 800         }
 801
 802         // Construct the output image
 803         *buffer = mlt_pool_alloc( size );
 804         if ( *buffer )
 805                 mlt_properties_set_data( frame_properties, "image", *buffer, size, mlt_pool_release, NULL );
 806         else
 807                 size = 0;
 808
 809         return size;
 810 }
 811
 812 /** Get an image from a frame.
 813 */
 814
 815 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
 816 {
 817         // Get the producer
 818         producer_avformat this = mlt_frame_pop_service( frame );
 819         mlt_producer producer = &this->parent;
 820
 821         // Get the properties from the frame
 822         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
 823
 824         // Obtain the frame number of this frame
 825         mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
 826
 827         // Get the producer properties
 828         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
 829
 830         avformat_lock();
 831
 832         // Fetch the video format context
 833         AVFormatContext *context = this->video_format;
 834
 835         // Get the video stream
 836         AVStream *stream = context->streams[ this->video_index ];
 837
 838         // Get codec context
 839         AVCodecContext *codec_context = stream->codec;
 840
 841         // Packet
 842         AVPacket pkt;
 843
 844         // Special case pause handling flag
 845         int paused = 0;
 846
 847         // Special case ffwd handling
 848         int ignore = 0;
 849
 850         // We may want to use the source fps if available
 851         double source_fps = mlt_properties_get_double( properties, "source_fps" );
 852         double fps = mlt_producer_get_fps( producer );
 853
 854         // This is the physical frame position in the source
 855         int req_position = ( int )( position / fps * source_fps + 0.5 );
 856
 857         // Determines if we have to decode all frames in a sequence
 858         // Temporary hack to improve intra frame only
 859         int must_decode = strcmp( codec_context->codec->name, "dnxhd" ) &&
 860                                   strcmp( codec_context->codec->name, "dvvideo" ) &&
 861                                   strcmp( codec_context->codec->name, "huffyuv" ) &&
 862                                   strcmp( codec_context->codec->name, "mjpeg" ) &&
 863                                   strcmp( codec_context->codec->name, "rawvideo" );
 864
 865         int last_position = this->last_position;
 866
 867         // Turn on usage of new seek API and PTS for seeking
 868         int use_new_seek = codec_context->codec_id == CODEC_ID_H264 && !strcmp( context->iformat->name, "mpegts" );
 869         if ( mlt_properties_get( properties, "new_seek" ) )
 870                 use_new_seek = mlt_properties_get_int( properties, "new_seek" );
 871
 872         // Seek if necessary
 873         if ( position != this->video_expected || last_position < 0 )
 874         {
 875                 if ( this->av_frame && position + 1 == this->video_expected )
 876                 {
 877                         // We're paused - use last image
 878                         paused = 1;
 879                 }
 880                 else if ( !this->seekable && position > this->video_expected && ( position - this->video_expected ) < 250 )
 881                 {
 882                         // Fast forward - seeking is inefficient for small distances - just ignore following frames
 883                         ignore = ( int )( ( position - this->video_expected ) / fps * source_fps );
 884                         codec_context->skip_loop_filter = AVDISCARD_NONREF;
 885                 }
 886                 else if ( this->seekable && ( position < this->video_expected || position - this->video_expected >= 12 || last_position < 0 ) )
 887                 {
 888                         if ( use_new_seek && last_position == POSITION_INITIAL )
 889                         {
 890                                 // find first key frame
 891                                 int ret = 0;
 892                                 int toscan = 100;
 893
 894                                 while ( ret >= 0 && toscan-- > 0 )
 895                                 {
 896                                         ret = av_read_frame( context, &pkt );
 897                                         if ( ret >= 0 && ( pkt.flags & PKT_FLAG_KEY ) && pkt.stream_index == this->video_index )
 898                                         {
 899                                                 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "first_pts %lld dts %lld pts_dts_delta %d\n", pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
 900                                                 this->first_pts = pkt.pts;
 901                                                 toscan = 0;
 902                                         }
 903                                         av_free_packet( &pkt );
 904                                 }
 905                                 // Rewind
 906                                 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
 907                         }
 908
 909                         // Calculate the timestamp for the requested frame
 910                         int64_t timestamp;
 911                         if ( use_new_seek )
 912                         {
 913                                 timestamp = ( req_position - 0.1 / source_fps ) /
 914                                         ( av_q2d( stream->time_base ) * source_fps );
 915                                 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "pos %d pts %lld ", req_position, timestamp );
 916                                 if ( this->first_pts > 0 )
 917                                         timestamp += this->first_pts;
 918                                 else if ( context->start_time != AV_NOPTS_VALUE )
 919                                         timestamp += context->start_time;
 920                         }
 921                         else
 922                         {
 923                                 timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
 924                                 if ( context->start_time != AV_NOPTS_VALUE )
 925                                         timestamp += context->start_time;
 926                         }
 927                         if ( must_decode )
 928                                 timestamp -= AV_TIME_BASE;
 929                         if ( timestamp < 0 )
 930                                 timestamp = 0;
 931                         mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %lld position %d expected %d last_pos %d\n",
 932                                 timestamp, position, this->video_expected, last_position );
 933
 934                         // Seek to the timestamp
 935                         if ( use_new_seek )
 936                         {
 937                                 codec_context->skip_loop_filter = AVDISCARD_NONREF;
 938                                 av_seek_frame( context, this->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
 939                         }
 940                         else
 941                         {
 942                                 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
 943                         }
 944
 945                         // Remove the cached info relating to the previous position
 946                         this->current_position = POSITION_INVALID;
 947                         this->last_position = POSITION_INVALID;
 948                         av_freep( &this->av_frame );
 949
 950                         if ( use_new_seek )
 951                         {
 952                                 // flush any pictures still in decode buffer
 953                                 avcodec_flush_buffers( codec_context );
 954                         }
 955                 }
 956         }
 957
 958         // Duplicate the last image if necessary (see comment on rawvideo below)
 959         if ( this->av_frame && this->got_picture && this->seekable
 960                  && ( paused
 961                           || this->current_position == req_position
 962                           || ( !use_new_seek && this->current_position > req_position ) ) )
 963         {
 964                 // Duplicate it
 965                 if ( allocate_buffer( frame_properties, codec_context, buffer, format, width, height ) )
 966                         convert_image( this->av_frame, *buffer, codec_context->pix_fmt, format, *width, *height );
 967                 else
 968                         mlt_frame_get_image( frame, buffer, format, width, height, writable );
 969         }
 970         else
 971         {
 972                 int ret = 0;
 973                 int int_position = 0;
 974                 int decode_errors = 0;
 975                 int got_picture = 0;
 976
 977                 av_init_packet( &pkt );
 978
 979                 // Construct an AVFrame for YUV422 conversion
 980                 if ( !this->av_frame )
 981                         this->av_frame = avcodec_alloc_frame( );
 982
 983                 while( ret >= 0 && !got_picture )
 984                 {
 985                         // Read a packet
 986                         ret = av_read_frame( context, &pkt );
 987
 988                         // We only deal with video from the selected video_index
 989                         if ( ret >= 0 && pkt.stream_index == this->video_index && pkt.size > 0 )
 990                         {
 991                                 // Determine time code of the packet
 992                                 if ( use_new_seek )
 993                                 {
 994                                         int64_t pts = pkt.pts;
 995                                         if ( this->first_pts > 0 )
 996                                                 pts -= this->first_pts;
 997                                         else if ( context->start_time != AV_NOPTS_VALUE )
 998                                                 pts -= context->start_time;
 999                                         int_position = ( int )( av_q2d( stream->time_base ) * pts * source_fps + 0.1 );
1000                                 }
1001                                 else
1002                                 {
1003                                         if ( pkt.dts != AV_NOPTS_VALUE )
1004                                         {
1005                                                 int_position = ( int )( av_q2d( stream->time_base ) * pkt.dts * source_fps + 0.5 );
1006                                                 if ( context->start_time != AV_NOPTS_VALUE )
1007                                                         int_position -= ( int )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
1008                                                 last_position = this->last_position;
1009                                                 if ( int_position == last_position )
1010                                                         int_position = last_position + 1;
1011                                         }
1012                                         else
1013                                         {
1014                                                 int_position = req_position;
1015                                         }
1016                                         mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "pkt.dts %llu req_pos %d cur_pos %d pkt_pos %d",
1017                                                 pkt.dts, req_position, this->current_position, int_position );
1018                                         // Make a dumb assumption on streams that contain wild timestamps
1019                                         if ( abs( req_position - int_position ) > 999 )
1020                                         {
1021                                                 int_position = req_position;
1022                                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP!" );
1023                                         }
1024                                 }
1025                                 this->last_position = int_position;
1026
1027                                 // Decode the image
1028                                 if ( must_decode || int_position >= req_position )
1029                                 {
1030                                         codec_context->reordered_opaque = pkt.pts;
1031                                         if ( int_position >= req_position )
1032                                                 codec_context->skip_loop_filter = AVDISCARD_NONE;
1033                                         ret = avcodec_decode_video( codec_context, this->av_frame, &got_picture, pkt.data, pkt.size );
1034                                         // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1035                                         if ( ret < 0 )
1036                                         {
1037                                                 if ( ++decode_errors <= 10 )
1038                                                         ret = 0;
1039                                         }
1040                                         else
1041                                         {
1042                                                 decode_errors = 0;
1043                                         }
1044                                 }
1045
1046                                 if ( got_picture )
1047                                 {
1048                                         if ( use_new_seek )
1049                                         {
1050                                                 // Determine time code of the packet
1051                                                 int64_t pts = this->av_frame->reordered_opaque;
1052                                                 if ( this->first_pts > 0 )
1053                                                         pts -= this->first_pts;
1054                                                 else if ( context->start_time != AV_NOPTS_VALUE )
1055                                                         pts -= context->start_time;
1056                                                 int_position = ( int )( av_q2d( stream->time_base) * pts * source_fps + 0.1 );
1057                                                 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "got frame %d, key %d\n", int_position, this->av_frame->key_frame );
1058                                         }
1059                                         // Handle ignore
1060                                         if ( int_position < req_position )
1061                                         {
1062                                                 ignore = 0;
1063                                                 got_picture = 0;
1064                                         }
1065                                         else if ( int_position >= req_position )
1066                                         {
1067                                                 ignore = 0;
1068                                                 codec_context->skip_loop_filter = AVDISCARD_NONE;
1069                                         }
1070                                         else if ( ignore -- )
1071                                         {
1072                                                 got_picture = 0;
1073                                         }
1074                                 }
1075                                 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), " got_pic %d key %d\n", got_picture, pkt.flags & PKT_FLAG_KEY );
1076                                 av_free_packet( &pkt );
1077                         }
1078                         else if ( ret >= 0 )
1079                         {
1080                                 av_free_packet( &pkt );
1081                         }
1082
1083                         // Now handle the picture if we have one
1084                         if ( got_picture )
1085                         {
1086                                 if ( allocate_buffer( frame_properties, codec_context, buffer, format, width, height ) )
1087                                 {
1088                                         convert_image( this->av_frame, *buffer, codec_context->pix_fmt, format, *width, *height );
1089                                         if ( !mlt_properties_get( properties, "force_progressive" ) )
1090                                                 mlt_properties_set_int( frame_properties, "progressive", !this->av_frame->interlaced_frame );
1091                                         this->top_field_first |= this->av_frame->top_field_first;
1092                                         this->current_position = int_position;
1093                                         this->got_picture = 1;
1094                                 }
1095                                 else
1096                                 {
1097                                         got_picture = 0;
1098                                 }
1099                         }
1100                 }
1101                 if ( !got_picture )
1102                         mlt_frame_get_image( frame, buffer, format, width, height, writable );
1103         }
1104
1105         // Very untidy - for rawvideo, the packet contains the frame, hence the free packet
1106         // above will break the pause behaviour - so we wipe the frame now
1107         if ( !strcmp( codec_context->codec->name, "rawvideo" ) )
1108                 av_freep( &this->av_frame );
1109
1110         avformat_unlock();
1111
1112         // Set the field order property for this frame
1113         mlt_properties_set_int( frame_properties, "top_field_first", this->top_field_first );
1114
1115         // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
1116         this->video_expected = position + 1;
1117
1118         return 0;
1119 }
1120
1121 /** Process properties as AVOptions and apply to AV context obj
1122 */
1123
1124 static void apply_properties( void *obj, mlt_properties properties, int flags )
1125 {
1126         int i;
1127         int count = mlt_properties_count( properties );
1128         for ( i = 0; i < count; i++ )
1129         {
1130                 const char *opt_name = mlt_properties_get_name( properties, i );
1131                 const AVOption *opt = av_find_opt( obj, opt_name, NULL, flags, flags );
1132                 if ( opt )
1133 #if LIBAVCODEC_VERSION_INT >= ((52<<16)+(7<<8)+0)
1134                         av_set_string3( obj, opt_name, mlt_properties_get( properties, opt_name), 0, NULL );
1135 #elif LIBAVCODEC_VERSION_INT >= ((51<<16)+(59<<8)+0)
1136                         av_set_string2( obj, opt_name, mlt_properties_get( properties, opt_name), 0 );
1137 #else
1138                         av_set_string( obj, opt_name, mlt_properties_get( properties, opt_name) );
1139 #endif
1140         }
1141 }
1142
1143 /** Initialize the video codec context.
1144  */
1145
1146 static int video_codec_init( producer_avformat this, int index, mlt_properties properties )
1147 {
1148         // Initialise the codec if necessary
1149         if ( !this->video_codec )
1150         {
1151                 // Get the video stream
1152                 AVStream *stream = this->video_format->streams[ index ];
1153
1154                 // Get codec context
1155                 AVCodecContext *codec_context = stream->codec;
1156
1157                 // Find the codec
1158                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1159
1160                 // Initialise multi-threading
1161                 int thread_count = mlt_properties_get_int( properties, "threads" );
1162                 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
1163                         thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
1164                 if ( thread_count > 1 )
1165                 {
1166                         avcodec_thread_init( codec_context, thread_count );
1167                         codec_context->thread_count = thread_count;
1168                 }
1169
1170                 // If we don't have a codec and we can't initialise it, we can't do much more...
1171                 avformat_lock( );
1172                 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1173                 {
1174                         // Now store the codec with its destructor
1175                         producer_codec_close( this->video_codec );
1176                         this->video_codec = codec_context;
1177                 }
1178                 else
1179                 {
1180                         // Remember that we can't use this later
1181                         this->video_index = -1;
1182                 }
1183                 avformat_unlock( );
1184
1185                 // Process properties as AVOptions
1186                 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1187
1188                 // Reset some image properties
1189                 mlt_properties_set_int( properties, "width", this->video_codec->width );
1190                 mlt_properties_set_int( properties, "height", this->video_codec->height );
1191                 mlt_properties_set_double( properties, "aspect_ratio", get_aspect_ratio( stream, this->video_codec, NULL ) );
1192
1193                 // Determine the fps first from the codec
1194                 double source_fps = (double) this->video_codec->time_base.den /
1195                                                                    ( this->video_codec->time_base.num == 0 ? 1 : this->video_codec->time_base.num );
1196
1197                 // If the muxer reports a frame rate different than the codec
1198                 double muxer_fps = av_q2d( stream->r_frame_rate );
1199                 if ( source_fps != muxer_fps )
1200                         // Choose the lesser - the wrong tends to be off by some multiple of 10
1201                         source_fps = muxer_fps < source_fps ? muxer_fps : source_fps;
1202
1203                 // We'll use fps if it's available
1204                 if ( source_fps > 0 )
1205                         mlt_properties_set_double( properties, "source_fps", source_fps );
1206                 else
1207                         mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( &this->parent ) );
1208         }
1209         return this->video_codec && this->video_index > -1;
1210 }
1211
1212 /** Set up video handling.
1213 */
1214
1215 static void producer_set_up_video( producer_avformat this, mlt_frame frame )
1216 {
1217         // Get the producer
1218         mlt_producer producer = &this->parent;
1219
1220         // Get the properties
1221         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1222
1223         // Fetch the video format context
1224         AVFormatContext *context = this->video_format;
1225
1226         // Get the video_index
1227         int index = mlt_properties_get_int( properties, "video_index" );
1228
1229         // Reopen the file if necessary
1230         if ( !context && index > -1 )
1231         {
1232                 mlt_events_block( properties, producer );
1233                 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1234                         mlt_properties_get( properties, "resource" ) );
1235                 context = this->video_format;
1236                 producer_format_close( this->dummy_context );
1237                 this->dummy_context = NULL;
1238                 mlt_events_unblock( properties, producer );
1239                 get_audio_streams_info( this );
1240
1241                 // Process properties as AVOptions
1242                 apply_properties( context, properties, AV_OPT_FLAG_DECODING_PARAM );
1243         }
1244
1245         // Exception handling for video_index
1246         if ( context && index >= (int) context->nb_streams )
1247         {
1248                 // Get the last video stream
1249                 for ( index = context->nb_streams - 1;
1250                           index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO;
1251                           index-- );
1252                 mlt_properties_set_int( properties, "video_index", index );
1253         }
1254         if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
1255         {
1256                 // Invalidate the video stream
1257                 index = -1;
1258                 mlt_properties_set_int( properties, "video_index", index );
1259         }
1260
1261         // Update the video properties if the index changed
1262         if ( index != this->video_index )
1263         {
1264                 // Reset the video properties if the index changed
1265                 this->video_index = index;
1266                 producer_codec_close( this->video_codec );
1267                 this->video_codec = NULL;
1268         }
1269
1270         // Get the frame properties
1271         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1272
1273         // Get the codec
1274         if ( context && index > -1 && video_codec_init( this, index, properties ) )
1275         {
1276                 // Set the frame properties
1277                 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
1278                 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
1279                         force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
1280
1281                 // Set the width and height
1282                 mlt_properties_set_int( frame_properties, "width", this->video_codec->width );
1283                 mlt_properties_set_int( frame_properties, "height", this->video_codec->height );
1284                 mlt_properties_set_int( frame_properties, "real_width", this->video_codec->width );
1285                 mlt_properties_set_int( frame_properties, "real_height", this->video_codec->height );
1286                 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
1287                 if ( mlt_properties_get( properties, "force_progressive" ) )
1288                         mlt_properties_set_int( frame_properties, "progressive", mlt_properties_get_int( properties, "force_progressive" ) );
1289
1290                 // Add our image operation
1291                 mlt_frame_push_service( frame, this );
1292                 mlt_frame_push_get_image( frame, producer_get_image );
1293         }
1294         else
1295         {
1296                 // If something failed, use test card image
1297                 mlt_properties_set_int( frame_properties, "test_image", 1 );
1298         }
1299 }
1300
1301 static int seek_audio( producer_avformat this, mlt_position position, double timecode, int *ignore )
1302 {
1303         int paused = 0;
1304
1305         // Fetch the audio_format
1306         AVFormatContext *context = this->audio_format;
1307
1308         // Seek if necessary
1309         if ( position != this->audio_expected )
1310         {
1311                 if ( position + 1 == this->audio_expected )
1312                 {
1313                         // We're paused - silence required
1314                         paused = 1;
1315                 }
1316                 else if ( !this->seekable && position > this->audio_expected && ( position - this->audio_expected ) < 250 )
1317                 {
1318                         // Fast forward - seeking is inefficient for small distances - just ignore following frames
1319                         *ignore = position - this->audio_expected;
1320                 }
1321                 else if ( position < this->audio_expected || position - this->audio_expected >= 12 )
1322                 {
1323                         int64_t timestamp = ( int64_t )( timecode * AV_TIME_BASE + 0.5 );
1324                         if ( context->start_time != AV_NOPTS_VALUE )
1325                                 timestamp += context->start_time;
1326                         if ( timestamp < 0 )
1327                                 timestamp = 0;
1328
1329                         // Set to the real timecode
1330                         if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
1331                                 paused = 1;
1332
1333                         // Clear the usage in the audio buffer
1334                         int i = MAX_AUDIO_STREAMS + 1;
1335                         while ( --i )
1336                                 this->audio_used[i - 1] = 0;
1337                 }
1338         }
1339         return paused;
1340 }
1341
1342 static int decode_audio( producer_avformat this, int *ignore, AVPacket *pkt, int samples, double timecode, double source_fps )
1343 {
1344         // Fetch the audio_format
1345         AVFormatContext *context = this->audio_format;
1346
1347         // Get the current stream index
1348         int index = pkt->stream_index;
1349
1350         // Get codec context
1351         AVCodecContext *codec_context = this->audio_codec[ index ];
1352
1353         // Obtain the resample context if it exists (not always needed)
1354         ReSampleContext *resample = this->audio_resample[ index ];
1355
1356         // Obtain the audio buffers
1357         int16_t *audio_buffer = this->audio_buffer[ index ];
1358         int16_t *decode_buffer = this->decode_buffer[ index ];
1359
1360         int audio_used = this->audio_used[ index ];
1361         int channels = codec_context->channels;
1362         int data_size = sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE;
1363         uint8_t *ptr = pkt->data;
1364         int len = pkt->size;
1365         int ret = 0;
1366
1367         while ( ptr && ret >= 0 && len > 0 )
1368         {
1369                 // Decode the audio
1370 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
1371                 ret = avcodec_decode_audio2( codec_context, decode_buffer, &data_size, ptr, len );
1372 #else
1373                 ret = avcodec_decode_audio( codec_context, decode_buffer, &data_size, ptr, len );
1374 #endif
1375                 if ( ret < 0 )
1376                 {
1377                         ret = 0;
1378                         break;
1379                 }
1380
1381                 len -= ret;
1382                 ptr += ret;
1383
1384                 // If decoded successfully and will not overflow
1385                 if ( data_size > 0 && ( audio_used * channels + data_size  / sizeof(int16_t) < AVCODEC_MAX_AUDIO_FRAME_SIZE ) )
1386                 {
1387                         if ( resample )
1388                         {
1389                                 int16_t *source = decode_buffer;
1390                                 int16_t *dest = &audio_buffer[ audio_used * channels ];
1391                                 int convert_samples = data_size / channels / av_get_bits_per_sample_format( codec_context->sample_fmt ) * 8;
1392                                 audio_used += audio_resample( resample, dest, source, convert_samples );
1393                         }
1394                         else
1395                         {
1396                                 memcpy( &audio_buffer[ audio_used * channels ], decode_buffer, data_size );
1397                                 audio_used += data_size / channels / av_get_bits_per_sample_format( codec_context->sample_fmt ) * 8;
1398                         }
1399
1400                         // Handle ignore
1401                         while ( *ignore && audio_used > samples )
1402                         {
1403                                 *ignore -= 1;
1404                                 audio_used -= samples;
1405                                 memmove( audio_buffer, &audio_buffer[ samples * channels ], audio_used * sizeof( int16_t ) );
1406                         }
1407                 }
1408                 else
1409                 {
1410                         mlt_log_error( MLT_PRODUCER_SERVICE( &this->parent ), "audio_buffer overflow: audio_used %d data_size %d\n", audio_used, data_size );
1411                 }
1412         }
1413
1414         // If we're behind, ignore this packet
1415         if ( pkt->pts >= 0 )
1416         {
1417                 double current_pts = av_q2d( context->streams[ index ]->time_base ) * pkt->pts;
1418                 int req_position = ( int )( timecode * source_fps + 0.5 );
1419                 int int_position = ( int )( current_pts * source_fps + 0.5 );
1420
1421                 if ( context->start_time != AV_NOPTS_VALUE )
1422                         int_position -= ( int )( source_fps * context->start_time / AV_TIME_BASE + 0.5 );
1423                 if ( this->seekable && *ignore == 0 && int_position < req_position )
1424                         *ignore = 1;
1425         }
1426
1427         this->audio_used[ index ] = audio_used;
1428
1429         return ret;
1430 }
1431
1432 /** Get the audio from a frame.
1433 */
1434
1435 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
1436 {
1437         // Get the producer
1438         producer_avformat this = mlt_frame_pop_audio( frame );
1439
1440         // Obtain the frame number of this frame
1441         mlt_position position = mlt_properties_get_position( MLT_FRAME_PROPERTIES( frame ), "avformat_position" );
1442
1443         // Calculate the real time code
1444         double real_timecode = producer_time_of_frame( &this->parent, position );
1445
1446         // Get the source fps
1447         double source_fps = mlt_properties_get_double( MLT_PRODUCER_PROPERTIES( &this->parent ), "source_fps" );
1448
1449         // Number of frames to ignore (for ffwd)
1450         int ignore = 0;
1451
1452         // Flag for paused (silence)
1453         int paused = seek_audio( this, position, real_timecode, &ignore );
1454
1455         // Fetch the audio_format
1456         AVFormatContext *context = this->audio_format;
1457
1458         // Determine the tracks to use
1459         int index = this->audio_index;
1460         int index_max = this->audio_index + 1;
1461         if ( this->audio_index == INT_MAX )
1462         {
1463                 index = 0;
1464                 index_max = context->nb_streams;
1465                 *channels = this->total_channels;
1466                 *frequency = this->max_frequency;
1467         }
1468
1469         // Initialize the resamplers and buffers
1470         for ( ; index < index_max; index++ )
1471         {
1472                 // Get codec context
1473                 AVCodecContext *codec_context = this->audio_codec[ index ];
1474
1475                 if ( codec_context && !this->audio_buffer[ index ] )
1476                 {
1477                         // Check for resample and create if necessary
1478                         if ( codec_context->channels <= 2 )
1479                         {
1480                                 // Create the resampler
1481 #if (LIBAVCODEC_VERSION_INT >= ((52<<16)+(15<<8)+0))
1482                                 this->audio_resample[ index ] = av_audio_resample_init(
1483                                         this->audio_index == INT_MAX ? codec_context->channels : *channels,
1484                                         codec_context->channels, *frequency, codec_context->sample_rate,
1485                                         SAMPLE_FMT_S16, codec_context->sample_fmt, 16, 10, 0, 0.8 );
1486 #else
1487                                 this->audio_resample[ index ] = audio_resample_init(
1488                                         this->audio_index == INT_MAX ? codec_context->channels : *channels,
1489                                         codec_context->channels, *frequency, codec_context->sample_rate );
1490 #endif
1491                         }
1492                         else
1493                         {
1494                                 codec_context->request_channels = this->audio_index == INT_MAX ? codec_context->channels : *channels;
1495                         }
1496
1497                         // Check for audio buffer and create if necessary
1498                         this->audio_buffer[ index ] = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
1499
1500                         // Check for decoder buffer and create if necessary
1501                         this->decode_buffer[ index ] = av_malloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
1502                 }
1503         }
1504
1505         // Get the audio if required
1506         if ( !paused )
1507         {
1508                 int ret = 0;
1509                 int got_audio = 0;
1510                 AVPacket pkt;
1511
1512                 av_init_packet( &pkt );
1513
1514                 while ( ret >= 0 && !got_audio )
1515                 {
1516                         // Check if the buffer already contains the samples required
1517                         if ( this->audio_index != INT_MAX && this->audio_used[ this->audio_index ] >= *samples && ignore == 0 )
1518                         {
1519                                 got_audio = 1;
1520                                 break;
1521                         }
1522
1523                         // Read a packet
1524                         ret = av_read_frame( context, &pkt );
1525
1526                         // We only deal with audio from the selected audio index
1527                         if ( ret >= 0 && pkt.data && pkt.size > 0 && ( pkt.stream_index == this->audio_index ||
1528                                  ( this->audio_index == INT_MAX && context->streams[ pkt.stream_index ]->codec->codec_type == CODEC_TYPE_AUDIO ) ) )
1529                                 ret = decode_audio( this, &ignore, &pkt, *samples, real_timecode, source_fps );
1530                         av_free_packet( &pkt );
1531
1532                         if ( this->audio_index == INT_MAX && ret >= 0 )
1533                         {
1534                                 // Determine if there is enough audio for all streams
1535                                 got_audio = 1;
1536                                 for ( index = 0; index < context->nb_streams; index++ )
1537                                 {
1538                                         if ( this->audio_codec[ index ] && this->audio_used[ index ] < *samples )
1539                                                 got_audio = 0;
1540                                 }
1541                         }
1542                 }
1543
1544                 // Allocate and set the frame's audio buffer
1545                 int size = *samples * *channels * sizeof(int16_t);
1546                 *buffer = mlt_pool_alloc( size );
1547                 *format = mlt_audio_s16;
1548                 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
1549
1550                 // Interleave tracks if audio_index=all
1551                 if ( this->audio_index == INT_MAX )
1552                 {
1553                         int16_t *dest = *buffer;
1554                         int i;
1555                         for ( i = 0; i < *samples; i++ )
1556                         {
1557                                 for ( index = 0; index < index_max; index++ )
1558                                 if ( this->audio_codec[ index ] )
1559                                 {
1560                                         int current_channels = this->audio_codec[ index ]->channels;
1561                                         int16_t *src = this->audio_buffer[ index ] + i * current_channels;
1562                                         memcpy( dest, src, current_channels * sizeof(int16_t) );
1563                                         dest += current_channels;
1564                                 }
1565                         }
1566                         for ( index = 0; index < index_max; index++ )
1567                         if ( this->audio_codec[ index ] )
1568                         {
1569                                 int current_channels = this->audio_codec[ index ]->channels;
1570                                 int16_t *src = this->audio_buffer[ index ] + *samples * current_channels;
1571                                 this->audio_used[index] -= *samples;
1572                                 memmove( this->audio_buffer[ index ], src, this->audio_used[ index ] * current_channels * sizeof(int16_t) );
1573                         }
1574                 }
1575                 // Copy a single track to the output buffer
1576                 else
1577                 {
1578                         index = this->audio_index;
1579
1580                         // Now handle the audio if we have enough
1581                         if ( this->audio_used[ index ] >= *samples )
1582                         {
1583                                 int16_t *src = this->audio_buffer[ index ];
1584                                 memcpy( *buffer, src, *samples * *channels * sizeof(int16_t) );
1585                                 this->audio_used[ index ] -= *samples;
1586                                 memmove( src, &src[ *samples * *channels ], this->audio_used[ index ] * *channels * sizeof(int16_t) );
1587                         }
1588                         else
1589                         {
1590                                 // Otherwise fill with silence
1591                                 memset( *buffer, 0, *samples * *channels * sizeof(int16_t) );
1592                         }
1593                         if ( !this->audio_resample[ index ] )
1594                         {
1595                                 // TODO: uncomment and remove following line when full multi-channel support is ready
1596                                 // *channels = codec_context->channels;
1597                                 *frequency = this->audio_codec[ index ]->sample_rate;
1598                         }
1599                 }
1600         }
1601         else
1602         {
1603                 // Get silence and don't touch the context
1604                 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
1605         }
1606
1607         // Regardless of speed (other than paused), we expect to get the next frame
1608         if ( !paused )
1609                 this->audio_expected = position + 1;
1610
1611         return 0;
1612 }
1613
1614 /** Initialize the audio codec context.
1615 */
1616
1617 static int audio_codec_init( producer_avformat this, int index, mlt_properties properties )
1618 {
1619         // Initialise the codec if necessary
1620         if ( !this->audio_codec[ index ] )
1621         {
1622                 // Get codec context
1623                 AVCodecContext *codec_context = this->audio_format->streams[index]->codec;
1624
1625                 // Find the codec
1626                 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
1627
1628                 // If we don't have a codec and we can't initialise it, we can't do much more...
1629                 avformat_lock( );
1630                 if ( codec && avcodec_open( codec_context, codec ) >= 0 )
1631                 {
1632                         // Now store the codec with its destructor
1633                         avformat_unlock();
1634                         producer_codec_close( this->audio_codec[index] );
1635                         this->audio_codec[ index ] = codec_context;
1636                 }
1637                 else
1638                 {
1639                         // Remember that we can't use this later
1640                         this->audio_index = -1;
1641                         avformat_unlock( );
1642                 }
1643
1644                 // Process properties as AVOptions
1645                 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
1646         }
1647         return this->audio_codec[ index ] && this->audio_index > -1;
1648 }
1649
1650 /** Set up audio handling.
1651 */
1652
1653 static void producer_set_up_audio( producer_avformat this, mlt_frame frame )
1654 {
1655         // Get the producer
1656         mlt_producer producer = &this->parent;
1657
1658         // Get the properties
1659         mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1660
1661         // Fetch the audio format context
1662         AVFormatContext *context = this->audio_format;
1663
1664         mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1665
1666         // Get the audio_index
1667         int index = mlt_properties_get_int( properties, "audio_index" );
1668
1669         // Handle all audio tracks
1670         if ( mlt_properties_get( properties, "audio_index" ) &&
1671                  !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
1672                 index = INT_MAX;
1673
1674         // Reopen the file if necessary
1675         if ( !context && index > -1 )
1676         {
1677                 mlt_events_block( properties, producer );
1678                 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
1679                         mlt_properties_get( properties, "resource" ) );
1680                 context = this->audio_format;
1681                 producer_format_close( this->dummy_context );
1682                 this->dummy_context = NULL;
1683                 mlt_events_unblock( properties, producer );
1684                 get_audio_streams_info( this );
1685         }
1686
1687         // Exception handling for audio_index
1688         if ( context && index >= (int) context->nb_streams && index < INT_MAX )
1689         {
1690                 for ( index = context->nb_streams - 1;
1691                           index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO;
1692                           index-- );
1693                 mlt_properties_set_int( properties, "audio_index", index );
1694         }
1695         if ( context && index > -1 && index < INT_MAX &&
1696                  context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
1697         {
1698                 index = -1;
1699                 mlt_properties_set_int( properties, "audio_index", index );
1700         }
1701
1702         // Update the audio properties if the index changed
1703         if ( index > -1 && index != this->audio_index )
1704         {
1705                 producer_codec_close( this->audio_codec[ this->audio_index ] );
1706                 this->audio_codec[ this->audio_index ] = NULL;
1707         }
1708         this->audio_index = index;
1709
1710         // Get the codec(s)
1711         if ( context && index == INT_MAX )
1712         {
1713                 mlt_properties_set_int( frame_properties, "frequency", this->max_frequency );
1714                 mlt_properties_set_int( frame_properties, "channels", this->total_channels );
1715                 for ( index = 0; index < context->nb_streams; index++ )
1716                 {
1717                         if ( context->streams[ index ]->codec->codec_type == CODEC_TYPE_AUDIO )
1718                                 audio_codec_init( this, index, properties );
1719                 }
1720         }
1721         else if ( context && index > -1 && audio_codec_init( this, index, properties ) )
1722         {
1723                 // Set the frame properties
1724                 if ( index < INT_MAX )
1725                 {
1726                         mlt_properties_set_int( frame_properties, "frequency", this->audio_codec[ index ]->sample_rate );
1727                         mlt_properties_set_int( frame_properties, "channels", this->audio_codec[ index ]->channels );
1728                 }
1729         }
1730         if ( context && index > -1 )
1731         {
1732                 // Add our audio operation
1733                 mlt_frame_push_audio( frame, this );
1734                 mlt_frame_push_audio( frame, producer_get_audio );
1735         }
1736 }
1737
1738 /** Our get frame implementation.
1739 */
1740
1741 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
1742 {
1743         // Access the private data
1744         producer_avformat this = producer->child;
1745
1746         // Create an empty frame
1747         *frame = mlt_frame_init( MLT_PRODUCER_SERVICE( producer ) );
1748
1749         // Update timecode on the frame we're creating
1750         mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
1751
1752         // Set the position of this producer
1753         mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( producer ) );
1754
1755         // Set up the video
1756         producer_set_up_video( this, *frame );
1757
1758         // Set up the audio
1759         producer_set_up_audio( this, *frame );
1760
1761         // Calculate the next timecode
1762         mlt_producer_prepare_next( producer );
1763
1764         return 0;
1765 }
1766
1767 static void producer_close( mlt_producer parent )
1768 {
1769         // Obtain this
1770         producer_avformat this = parent->child;
1771
1772         // Close the file
1773         av_free( this->av_frame );
1774         int i;
1775         for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
1776         {
1777                 if ( this->audio_resample[i] )
1778                         audio_resample_close( this->audio_resample[i] );
1779                 mlt_pool_release( this->audio_buffer[i] );
1780                 av_free( this->decode_buffer[i] );
1781                 producer_codec_close( this->audio_codec[i] );
1782         }
1783         producer_codec_close( this->video_codec );
1784         producer_format_close( this->dummy_context );
1785         producer_format_close( this->audio_format );
1786         producer_format_close( this->video_format );
1787
1788         // Close the parent
1789         parent->close = NULL;
1790         mlt_producer_close( parent );
1791
1792         // Free the memory
1793         free( this );
1794 }