producer_avformat.c: bugfix regression playing audio-only files.
[melted] / 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
26 // ffmpeg Header files
27 #include <avformat.h>
28 #ifdef SWSCALE
29 #include <swscale.h>
30 #endif
31
32 // System header files
33 #include <stdlib.h>
34 #include <string.h>
35 #include <pthread.h>
36 #include <math.h>
37
38 void avformat_lock( );
39 void avformat_unlock( );
40
41 // Forward references.
42 static int producer_open( mlt_producer this, mlt_profile profile, char *file );
43 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index );
44
45 /** Constructor for libavformat.
46 */
47
48 mlt_producer producer_avformat_init( mlt_profile profile, char *file )
49 {
50 mlt_producer this = NULL;
51
52 // Check that we have a non-NULL argument
53 if ( file != NULL )
54 {
55 // Construct the producer
56 this = calloc( 1, sizeof( struct mlt_producer_s ) );
57
58 // Initialise it
59 if ( mlt_producer_init( this, NULL ) == 0 )
60 {
61 // Get the properties
62 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
63
64 // Set the resource property (required for all producers)
65 mlt_properties_set( properties, "resource", file );
66
67 // Register our get_frame implementation
68 this->get_frame = producer_get_frame;
69
70 // Open the file
71 if ( producer_open( this, profile, file ) != 0 )
72 {
73 // Clean up
74 mlt_producer_close( this );
75 this = NULL;
76 }
77
78 // Close the file to release resources for large playlists - reopen later as needed
79 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
80 mlt_properties_set_data( properties, "audio_context", NULL, 0, NULL, NULL );
81 mlt_properties_set_data( properties, "video_context", NULL, 0, NULL, NULL );
82 }
83 }
84
85 return this;
86 }
87
88 /** Find the default streams.
89 */
90
91 static void find_default_streams( AVFormatContext *context, int *audio_index, int *video_index )
92 {
93 int i;
94
95 // Allow for multiple audio and video streams in the file and select first of each (if available)
96 for( i = 0; i < context->nb_streams; i++ )
97 {
98 // Get the codec context
99 AVCodecContext *codec_context = context->streams[ i ]->codec;
100
101 if ( avcodec_find_decoder( codec_context->codec_id ) == NULL )
102 continue;
103
104 // Determine the type and obtain the first index of each type
105 switch( codec_context->codec_type )
106 {
107 case CODEC_TYPE_VIDEO:
108 if ( *video_index < 0 )
109 *video_index = i;
110 break;
111 case CODEC_TYPE_AUDIO:
112 if ( *audio_index < 0 )
113 *audio_index = i;
114 break;
115 default:
116 break;
117 }
118 }
119 }
120
121 /** Producer file destructor.
122 */
123
124 static void producer_file_close( void *context )
125 {
126 if ( context != NULL )
127 {
128 // Lock the mutex now
129 avformat_lock( );
130
131 // Close the file
132 av_close_input_file( context );
133
134 // Unlock the mutex now
135 avformat_unlock( );
136 }
137 }
138
139 /** Producer file destructor.
140 */
141
142 static void producer_codec_close( void *codec )
143 {
144 if ( codec != NULL )
145 {
146 // Lock the mutex now
147 avformat_lock( );
148
149 // Close the file
150 avcodec_close( codec );
151
152 // Unlock the mutex now
153 avformat_unlock( );
154 }
155 }
156
157 /** Open the file.
158 */
159
160 static int producer_open( mlt_producer this, mlt_profile profile, char *file )
161 {
162 // Return an error code (0 == no error)
163 int error = 0;
164
165 // Context for avformat
166 AVFormatContext *context = NULL;
167
168 // Get the properties
169 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
170
171 // We will treat everything with the producer fps
172 double fps = mlt_profile_fps( profile );
173
174 // Lock the mutex now
175 avformat_lock( );
176
177 // If "MRL", then create AVInputFormat
178 AVInputFormat *format = NULL;
179 AVFormatParameters *params = NULL;
180 char *standard = NULL;
181 char *mrl = strchr( file, ':' );
182
183 // AV option (0 = both, 1 = video, 2 = audio)
184 int av = 0;
185
186 // Setting lowest log level
187 av_log_set_level( -1 );
188
189 // Only if there is not a protocol specification that avformat can handle
190 if ( mrl && !url_exist( file ) )
191 {
192 // 'file' becomes format abbreviation
193 mrl[0] = 0;
194
195 // Lookup the format
196 format = av_find_input_format( file );
197
198 // Eat the format designator
199 file = ++mrl;
200
201 if ( format )
202 {
203 // Allocate params
204 params = calloc( sizeof( AVFormatParameters ), 1 );
205
206 // These are required by video4linux (defaults)
207 params->width = 640;
208 params->height = 480;
209 params->time_base= (AVRational){1,25};
210 // params->device = file;
211 params->channels = 2;
212 params->sample_rate = 48000;
213 }
214
215 // XXX: this does not work anymore since avdevice
216 // TODO: make producer_avddevice?
217 // Parse out params
218 mrl = strchr( file, '?' );
219 while ( mrl )
220 {
221 mrl[0] = 0;
222 char *name = strdup( ++mrl );
223 char *value = strchr( name, ':' );
224 if ( value )
225 {
226 value[0] = 0;
227 value++;
228 char *t = strchr( value, '&' );
229 if ( t )
230 t[0] = 0;
231 if ( !strcmp( name, "frame_rate" ) )
232 params->time_base.den = atoi( value );
233 else if ( !strcmp( name, "frame_rate_base" ) )
234 params->time_base.num = atoi( value );
235 else if ( !strcmp( name, "sample_rate" ) )
236 params->sample_rate = atoi( value );
237 else if ( !strcmp( name, "channels" ) )
238 params->channels = atoi( value );
239 else if ( !strcmp( name, "width" ) )
240 params->width = atoi( value );
241 else if ( !strcmp( name, "height" ) )
242 params->height = atoi( value );
243 else if ( !strcmp( name, "standard" ) )
244 {
245 standard = strdup( value );
246 params->standard = standard;
247 }
248 else if ( !strcmp( name, "av" ) )
249 av = atoi( value );
250 }
251 free( name );
252 mrl = strchr( mrl, '&' );
253 }
254 }
255
256 // Now attempt to open the file
257 error = av_open_input_file( &context, file, format, 0, params ) < 0;
258
259 // Cleanup AVFormatParameters
260 free( standard );
261 free( params );
262
263 // If successful, then try to get additional info
264 if ( error == 0 )
265 {
266 // Get the stream info
267 error = av_find_stream_info( context ) < 0;
268
269 // Continue if no error
270 if ( error == 0 )
271 {
272 // We will default to the first audio and video streams found
273 int audio_index = -1;
274 int video_index = -1;
275 int av_bypass = 0;
276
277 // Now set properties where we can (use default unknowns if required)
278 if ( context->duration != AV_NOPTS_VALUE )
279 {
280 // This isn't going to be accurate for all formats
281 mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps + 0.5 );
282 mlt_properties_set_position( properties, "out", frames - 1 );
283 mlt_properties_set_position( properties, "length", frames );
284 }
285
286 // Find default audio and video streams
287 find_default_streams( context, &audio_index, &video_index );
288
289 if ( context->start_time != AV_NOPTS_VALUE )
290 mlt_properties_set_double( properties, "_start_time", context->start_time );
291
292 // Check if we're seekable (something funny about mpeg here :-/)
293 if ( strcmp( file, "pipe:" ) && strncmp( file, "http://", 6 ) )
294 {
295 mlt_properties_set_int( properties, "seekable", av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ), AVSEEK_FLAG_BACKWARD ) >= 0 );
296 mlt_properties_set_data( properties, "dummy_context", context, 0, producer_file_close, NULL );
297 av_open_input_file( &context, file, NULL, 0, NULL );
298 av_find_stream_info( context );
299 }
300 else
301 av_bypass = 1;
302
303 // Store selected audio and video indexes on properties
304 mlt_properties_set_int( properties, "audio_index", audio_index );
305 mlt_properties_set_int( properties, "video_index", video_index );
306 mlt_properties_set_int( properties, "_last_position", -1 );
307
308 // Fetch the width, height and aspect ratio
309 if ( video_index != -1 )
310 {
311 AVCodecContext *codec_context = context->streams[ video_index ]->codec;
312 mlt_properties_set_int( properties, "width", codec_context->width );
313 mlt_properties_set_int( properties, "height", codec_context->height );
314 mlt_properties_set_double( properties, "aspect_ratio", av_q2d( codec_context->sample_aspect_ratio ) );
315 }
316
317 // Read Metadata
318 if (context->title != NULL)
319 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
320 if (context->author != NULL)
321 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
322 if (context->copyright != NULL)
323 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
324 if (context->comment != NULL)
325 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
326 if (context->album != NULL)
327 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
328 if (context->year != 0)
329 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
330 if (context->track != 0)
331 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
332
333 // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
334 if ( av == 0 && !av_bypass && audio_index != -1 && video_index != -1 )
335 {
336 // We'll use the open one as our video_context
337 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
338
339 // And open again for our audio context
340 av_open_input_file( &context, file, NULL, 0, NULL );
341 av_find_stream_info( context );
342
343 // Audio context
344 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
345 }
346 else if ( av != 2 && video_index != -1 )
347 {
348 // We only have a video context
349 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
350 }
351 else if ( audio_index != -1 )
352 {
353 // We only have an audio context
354 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
355 }
356 else
357 {
358 // Something has gone wrong
359 error = -1;
360 }
361
362 mlt_properties_set_int( properties, "av_bypass", av_bypass );
363 }
364 }
365
366 // Unlock the mutex now
367 avformat_unlock( );
368
369 return error;
370 }
371
372 /** Convert a frame position to a time code.
373 */
374
375 static double producer_time_of_frame( mlt_producer this, mlt_position position )
376 {
377 return ( double )position / mlt_producer_get_fps( this );
378 }
379
380 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, mlt_image_format format, int width, int height )
381 {
382 #ifdef SWSCALE
383 if ( format == mlt_image_yuv420p )
384 {
385 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
386 width, height, PIX_FMT_YUV420P, SWS_FAST_BILINEAR, NULL, NULL, NULL);
387 AVPicture output;
388 output.data[0] = buffer;
389 output.data[1] = buffer + width * height;
390 output.data[2] = buffer + ( 3 * width * height ) / 2;
391 output.linesize[0] = width;
392 output.linesize[1] = width >> 1;
393 output.linesize[2] = width >> 1;
394 sws_scale( context, frame->data, frame->linesize, 0, height,
395 output.data, output.linesize);
396 sws_freeContext( context );
397 }
398 else if ( format == mlt_image_rgb24 )
399 {
400 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
401 width, height, PIX_FMT_RGB24, SWS_FAST_BILINEAR, NULL, NULL, NULL);
402 AVPicture output;
403 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
404 sws_scale( context, frame->data, frame->linesize, 0, height,
405 output.data, output.linesize);
406 sws_freeContext( context );
407 }
408 else
409 {
410 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
411 width, height, PIX_FMT_YUYV422, SWS_FAST_BILINEAR, NULL, NULL, NULL);
412 AVPicture output;
413 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
414 sws_scale( context, frame->data, frame->linesize, 0, height,
415 output.data, output.linesize);
416 sws_freeContext( context );
417 }
418 #else
419 if ( format == mlt_image_yuv420p )
420 {
421 AVPicture pict;
422 pict.data[0] = buffer;
423 pict.data[1] = buffer + width * height;
424 pict.data[2] = buffer + ( 3 * width * height ) / 2;
425 pict.linesize[0] = width;
426 pict.linesize[1] = width >> 1;
427 pict.linesize[2] = width >> 1;
428 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
429 }
430 else if ( format == mlt_image_rgb24 )
431 {
432 AVPicture output;
433 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
434 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
435 }
436 else
437 {
438 AVPicture output;
439 avpicture_fill( &output, buffer, PIX_FMT_YUV422, width, height );
440 img_convert( &output, PIX_FMT_YUV422, (AVPicture *)frame, pix_fmt, width, height );
441 }
442 #endif
443 }
444
445 /** Get an image from a frame.
446 */
447
448 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
449 {
450 // Get the properties from the frame
451 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
452
453 // Obtain the frame number of this frame
454 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
455
456 // Get the producer
457 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
458
459 // Get the producer properties
460 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
461
462 // Fetch the video_context
463 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
464
465 // Get the video_index
466 int index = mlt_properties_get_int( properties, "video_index" );
467
468 // Obtain the expected frame numer
469 mlt_position expected = mlt_properties_get_position( properties, "_video_expected" );
470
471 // Get the video stream
472 AVStream *stream = context->streams[ index ];
473
474 // Get codec context
475 AVCodecContext *codec_context = stream->codec;
476
477 // Packet
478 AVPacket pkt;
479
480 // Get the conversion frame
481 AVFrame *av_frame = mlt_properties_get_data( properties, "av_frame", NULL );
482
483 // Special case pause handling flag
484 int paused = 0;
485
486 // Special case ffwd handling
487 int ignore = 0;
488
489 // We may want to use the source fps if available
490 double source_fps = mlt_properties_get_double( properties, "source_fps" );
491 double fps = mlt_producer_get_fps( this );
492
493 // This is the physical frame position in the source
494 int req_position = ( int )( position / fps * source_fps + 0.5 );
495
496 // Get the seekable status
497 int seekable = mlt_properties_get_int( properties, "seekable" );
498
499 // Generate the size in bytes
500 int size = 0;
501
502 // Hopefully provide better support for streams...
503 int av_bypass = mlt_properties_get_int( properties, "av_bypass" );
504
505 // Determines if we have to decode all frames in a sequence
506 int must_decode = 1;
507
508 // Set the result arguments that we know here (only *buffer is now required)
509 *width = codec_context->width;
510 *height = codec_context->height;
511
512 switch ( *format )
513 {
514 case mlt_image_yuv420p:
515 size = *width * 3 * ( *height + 1 ) / 2;
516 break;
517 case mlt_image_rgb24:
518 size = *width * ( *height + 1 ) * 3;
519 break;
520 default:
521 *format = mlt_image_yuv422;
522 size = *width * ( *height + 1 ) * 2;
523 break;
524 }
525
526 // Set this on the frame properties
527 mlt_properties_set_int( frame_properties, "width", *width );
528 mlt_properties_set_int( frame_properties, "height", *height );
529
530 // Construct the output image
531 *buffer = mlt_pool_alloc( size );
532
533 // Temporary hack to improve intra frame only
534 must_decode = strcmp( codec_context->codec->name, "mjpeg" ) &&
535 strcmp( codec_context->codec->name, "rawvideo" ) &&
536 strcmp( codec_context->codec->name, "dvvideo" );
537
538 // Seek if necessary
539 if ( position != expected )
540 {
541 if ( av_frame != NULL && position + 1 == expected )
542 {
543 // We're paused - use last image
544 paused = 1;
545 }
546 else if ( !seekable && position > expected && ( position - expected ) < 250 )
547 {
548 // Fast forward - seeking is inefficient for small distances - just ignore following frames
549 ignore = ( int )( ( position - expected ) / fps * source_fps );
550 }
551 else if ( seekable && ( position < expected || position - expected >= 12 ) )
552 {
553 // Calculate the timestamp for the requested frame
554 int64_t timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
555 if ( ( uint64_t )context->start_time != AV_NOPTS_VALUE )
556 timestamp += context->start_time;
557 if ( must_decode )
558 timestamp -= AV_TIME_BASE;
559 if ( timestamp < 0 )
560 timestamp = 0;
561
562 // Set to the timestamp
563 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
564
565 // Remove the cached info relating to the previous position
566 mlt_properties_set_int( properties, "_current_position", -1 );
567 mlt_properties_set_int( properties, "_last_position", -1 );
568 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
569 av_frame = NULL;
570 }
571 }
572
573 // Duplicate the last image if necessary (see comment on rawvideo below)
574 int current_position = mlt_properties_get_int( properties, "_current_position" );
575 int got_picture = mlt_properties_get_int( properties, "_got_picture" );
576 if ( av_frame != NULL && got_picture && ( paused || current_position >= req_position ) && av_bypass == 0 )
577 {
578 // Duplicate it
579 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
580
581 // Set this on the frame properties
582 mlt_properties_set_data( frame_properties, "image", *buffer, size, ( mlt_destructor )mlt_pool_release, NULL );
583 }
584 else
585 {
586 int ret = 0;
587 int int_position = 0;
588 got_picture = 0;
589
590 av_init_packet( &pkt );
591
592 // Construct an AVFrame for YUV422 conversion
593 if ( av_frame == NULL )
594 {
595 av_frame = avcodec_alloc_frame( );
596 mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL );
597 }
598
599 while( ret >= 0 && !got_picture )
600 {
601 // Read a packet
602 ret = av_read_frame( context, &pkt );
603
604 // We only deal with video from the selected video_index
605 if ( ret >= 0 && pkt.stream_index == index && pkt.size > 0 )
606 {
607 // Determine time code of the packet
608 int_position = ( int )( av_q2d( stream->time_base ) * pkt.dts * source_fps + 0.5 );
609 if ( context->start_time != AV_NOPTS_VALUE )
610 int_position -= ( int )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
611 int last_position = mlt_properties_get_int( properties, "_last_position" );
612 if ( int_position == last_position )
613 int_position = last_position + 1;
614 mlt_properties_set_int( properties, "_last_position", int_position );
615
616 // Decode the image
617 if ( must_decode || int_position >= req_position )
618 ret = avcodec_decode_video( codec_context, av_frame, &got_picture, pkt.data, pkt.size );
619
620 if ( got_picture )
621 {
622 // Handle ignore
623 if ( int_position < req_position )
624 {
625 ignore = 0;
626 got_picture = 0;
627 }
628 else if ( int_position >= req_position )
629 {
630 ignore = 0;
631 }
632 else if ( ignore -- )
633 {
634 got_picture = 0;
635 }
636 }
637 av_free_packet( &pkt );
638 }
639 else if ( ret >= 0 )
640 {
641 av_free_packet( &pkt );
642 }
643
644 // Now handle the picture if we have one
645 if ( got_picture )
646 {
647 mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame );
648 mlt_properties_set_int( properties, "top_field_first", av_frame->top_field_first );
649 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
650 mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL );
651 mlt_properties_set_int( properties, "_current_position", int_position );
652 mlt_properties_set_int( properties, "_got_picture", 1 );
653 }
654 }
655 if ( !got_picture )
656 mlt_frame_get_image( frame, buffer, format, width, height, writable );
657 }
658
659 // Very untidy - for rawvideo, the packet contains the frame, hence the free packet
660 // above will break the pause behaviour - so we wipe the frame now
661 if ( !strcmp( codec_context->codec->name, "rawvideo" ) )
662 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
663
664 // Set the field order property for this frame
665 mlt_properties_set_int( frame_properties, "top_field_first", mlt_properties_get_int( properties, "top_field_first" ) );
666
667 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
668 mlt_properties_set_position( properties, "_video_expected", position + 1 );
669
670 return 0;
671 }
672
673 /** Set up video handling.
674 */
675
676 static void producer_set_up_video( mlt_producer this, mlt_frame frame )
677 {
678 // Get the properties
679 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
680
681 // Fetch the video_context
682 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
683
684 // Get the video_index
685 int index = mlt_properties_get_int( properties, "video_index" );
686
687 // Reopen the file if necessary
688 if ( !context && index != -1 )
689 {
690 mlt_events_block( properties, this );
691 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(this) ),
692 mlt_properties_get( properties, "resource" ) );
693 context = mlt_properties_get_data( properties, "video_context", NULL );
694 index = mlt_properties_get_int( properties, "video_index" );
695 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
696 mlt_events_unblock( properties, this );
697 }
698
699 // Get the frame properties
700 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
701
702 if ( context != NULL && index != -1 )
703 {
704 // Get the video stream
705 AVStream *stream = context->streams[ index ];
706
707 // Get codec context
708 AVCodecContext *codec_context = stream->codec;
709
710 // Get the codec
711 AVCodec *codec = mlt_properties_get_data( properties, "video_codec", NULL );
712
713 // Initialise the codec if necessary
714 if ( codec == NULL )
715 {
716 // Initialise multi-threading
717 int thread_count = mlt_properties_get_int( properties, "threads" );
718 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
719 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
720 if ( thread_count > 1 )
721 {
722 avcodec_thread_init( codec_context, thread_count );
723 codec_context->thread_count = thread_count;
724 }
725
726 // Find the codec
727 codec = avcodec_find_decoder( codec_context->codec_id );
728
729 // If we don't have a codec and we can't initialise it, we can't do much more...
730 avformat_lock( );
731 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
732 {
733 // Now store the codec with its destructor
734 mlt_properties_set_data( properties, "video_codec", codec_context, 0, producer_codec_close, NULL );
735 }
736 else
737 {
738 // Remember that we can't use this later
739 mlt_properties_set_int( properties, "video_index", -1 );
740 }
741 avformat_unlock( );
742 }
743
744 // No codec, no show...
745 if ( codec != NULL )
746 {
747 double source_fps = 0;
748 int norm_aspect_ratio = mlt_properties_get_int( properties, "norm_aspect_ratio" );
749 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
750 double aspect_ratio;
751
752 if ( strcmp( codec_context->codec->name, "dvvideo" ) == 0 )
753 {
754 // Override FFmpeg's notion of DV aspect ratios, which are
755 // based upon a width of 704. Since we do not have a normaliser
756 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
757 // we just coerce the values to facilitate a passive behaviour through
758 // the rescale normaliser when using equivalent producers and consumers.
759 // = display_aspect / (width * height)
760 if ( codec_context->sample_aspect_ratio.num == 10 &&
761 codec_context->sample_aspect_ratio.den == 11 )
762 force_aspect_ratio = 8.0/9.0; // 4:3 NTSC
763 else if ( codec_context->sample_aspect_ratio.num == 59 &&
764 codec_context->sample_aspect_ratio.den == 54 )
765 force_aspect_ratio = 16.0/15.0; // 4:3 PAL
766 else if ( codec_context->sample_aspect_ratio.num == 40 &&
767 codec_context->sample_aspect_ratio.den == 33 )
768 force_aspect_ratio = 32.0/27.0; // 16:9 NTSC
769 else if ( codec_context->sample_aspect_ratio.num == 118 &&
770 codec_context->sample_aspect_ratio.den == 81 )
771 force_aspect_ratio = 64.0/45.0; // 16:9 PAL
772 }
773
774 // XXX: We won't know the real aspect ratio until an image is decoded
775 // but we do need it now (to satisfy filter_resize) - take a guess based
776 // on pal/ntsc
777 if ( force_aspect_ratio > 0.0 )
778 {
779 aspect_ratio = force_aspect_ratio;
780 }
781 else if ( !norm_aspect_ratio && codec_context->sample_aspect_ratio.num > 0 )
782 {
783 aspect_ratio = av_q2d( codec_context->sample_aspect_ratio );
784 }
785 else
786 {
787 aspect_ratio = 1.0;
788 }
789
790 // Determine the fps
791 source_fps = ( double )codec_context->time_base.den / ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num );
792
793 // We'll use fps if it's available
794 if ( source_fps > 0 )
795 mlt_properties_set_double( properties, "source_fps", source_fps );
796 else
797 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( this ) );
798 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
799
800 // Set the width and height
801 mlt_properties_set_int( frame_properties, "width", codec_context->width );
802 mlt_properties_set_int( frame_properties, "height", codec_context->height );
803 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
804
805 mlt_frame_push_get_image( frame, producer_get_image );
806 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
807 }
808 else
809 {
810 mlt_properties_set_int( frame_properties, "test_image", 1 );
811 }
812 }
813 else
814 {
815 mlt_properties_set_int( frame_properties, "test_image", 1 );
816 }
817 }
818
819 /** Get the audio from a frame.
820 */
821
822 static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
823 {
824 // Get the properties from the frame
825 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
826
827 // Obtain the frame number of this frame
828 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
829
830 // Get the producer
831 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
832
833 // Get the producer properties
834 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
835
836 // Fetch the audio_context
837 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
838
839 // Get the audio_index
840 int index = mlt_properties_get_int( properties, "audio_index" );
841
842 // Get the seekable status
843 int seekable = mlt_properties_get_int( properties, "seekable" );
844
845 // Obtain the expected frame numer
846 mlt_position expected = mlt_properties_get_position( properties, "_audio_expected" );
847
848 // Obtain the resample context if it exists (not always needed)
849 ReSampleContext *resample = mlt_properties_get_data( properties, "audio_resample", NULL );
850
851 // Obtain the audio buffer
852 int16_t *audio_buffer = mlt_properties_get_data( properties, "audio_buffer", NULL );
853
854 // Get amount of audio used
855 int audio_used = mlt_properties_get_int( properties, "_audio_used" );
856
857 // Calculate the real time code
858 double real_timecode = producer_time_of_frame( this, position );
859
860 // Get the audio stream
861 AVStream *stream = context->streams[ index ];
862
863 // Get codec context
864 AVCodecContext *codec_context = stream->codec;
865
866 // Packet
867 AVPacket pkt;
868
869 // Number of frames to ignore (for ffwd)
870 int ignore = 0;
871
872 // Flag for paused (silence)
873 int paused = 0;
874
875 // Check for resample and create if necessary
876 if ( resample == NULL && codec_context->channels <= 2 )
877 {
878 // Create the resampler
879 resample = audio_resample_init( *channels, codec_context->channels, *frequency, codec_context->sample_rate );
880
881 // And store it on properties
882 mlt_properties_set_data( properties, "audio_resample", resample, 0, ( mlt_destructor )audio_resample_close, NULL );
883 }
884 else if ( resample == NULL )
885 {
886 *channels = codec_context->channels;
887 *frequency = codec_context->sample_rate;
888 }
889
890 // Check for audio buffer and create if necessary
891 if ( audio_buffer == NULL )
892 {
893 // Allocate the audio buffer
894 audio_buffer = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
895
896 // And store it on properties for reuse
897 mlt_properties_set_data( properties, "audio_buffer", audio_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
898 }
899
900 // Seek if necessary
901 if ( position != expected )
902 {
903 if ( position + 1 == expected )
904 {
905 // We're paused - silence required
906 paused = 1;
907 }
908 else if ( !seekable && position > expected && ( position - expected ) < 250 )
909 {
910 // Fast forward - seeking is inefficient for small distances - just ignore following frames
911 ignore = position - expected;
912 }
913 else if ( position < expected || position - expected >= 12 )
914 {
915 // Set to the real timecode
916 if ( av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ) + real_timecode * 1000000.0, AVSEEK_FLAG_BACKWARD ) != 0 )
917 paused = 1;
918
919 // Clear the usage in the audio buffer
920 audio_used = 0;
921 }
922 }
923
924 // Get the audio if required
925 if ( !paused )
926 {
927 int ret = 0;
928 int got_audio = 0;
929 int16_t *temp = av_malloc( sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE );
930
931 av_init_packet( &pkt );
932
933 while( ret >= 0 && !got_audio )
934 {
935 // Check if the buffer already contains the samples required
936 if ( audio_used >= *samples && ignore == 0 )
937 {
938 got_audio = 1;
939 break;
940 }
941
942 // Read a packet
943 ret = av_read_frame( context, &pkt );
944
945 int len = pkt.size;
946 uint8_t *ptr = pkt.data;
947
948 // We only deal with audio from the selected audio_index
949 while ( ptr != NULL && ret >= 0 && pkt.stream_index == index && len > 0 )
950 {
951 int data_size = sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE;
952
953 // Decode the audio
954 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
955 ret = avcodec_decode_audio2( codec_context, temp, &data_size, ptr, len );
956 #else
957 ret = avcodec_decode_audio( codec_context, temp, &data_size, ptr, len );
958 #endif
959 if ( ret < 0 )
960 {
961 ret = 0;
962 break;
963 }
964
965 len -= ret;
966 ptr += ret;
967
968 if ( data_size > 0 )
969 {
970 if ( resample != NULL )
971 {
972 audio_used += audio_resample( resample, &audio_buffer[ audio_used * *channels ], temp, data_size / ( codec_context->channels * sizeof( int16_t ) ) );
973 }
974 else
975 {
976 memcpy( &audio_buffer[ audio_used * *channels ], temp, data_size );
977 audio_used += data_size / ( codec_context->channels * sizeof( int16_t ) );
978 }
979
980 // Handle ignore
981 while ( ignore && audio_used > *samples )
982 {
983 ignore --;
984 audio_used -= *samples;
985 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * sizeof( int16_t ) );
986 }
987 }
988
989 // If we're behind, ignore this packet
990 float current_pts = av_q2d( stream->time_base ) * pkt.pts;
991 if ( seekable && ( !ignore && current_pts <= ( real_timecode - 0.02 ) ) )
992 ignore = 1;
993 }
994
995 // We're finished with this packet regardless
996 av_free_packet( &pkt );
997 }
998
999 *buffer = mlt_pool_alloc( *samples * *channels * sizeof( int16_t ) );
1000 mlt_properties_set_data( frame_properties, "audio", *buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
1001
1002 // Now handle the audio if we have enough
1003 if ( audio_used >= *samples )
1004 {
1005 memcpy( *buffer, audio_buffer, *samples * *channels * sizeof( int16_t ) );
1006 audio_used -= *samples;
1007 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * *channels * sizeof( int16_t ) );
1008 }
1009 else
1010 {
1011 memset( *buffer, 0, *samples * *channels * sizeof( int16_t ) );
1012 }
1013
1014 // Store the number of audio samples still available
1015 mlt_properties_set_int( properties, "_audio_used", audio_used );
1016
1017 // Release the temporary audio
1018 av_free( temp );
1019 }
1020 else
1021 {
1022 // Get silence and don't touch the context
1023 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
1024 }
1025
1026 // Regardless of speed (other than paused), we expect to get the next frame
1027 if ( !paused )
1028 mlt_properties_set_position( properties, "_audio_expected", position + 1 );
1029
1030 return 0;
1031 }
1032
1033 /** Set up audio handling.
1034 */
1035
1036 static void producer_set_up_audio( mlt_producer this, mlt_frame frame )
1037 {
1038 // Get the properties
1039 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
1040
1041 // Fetch the audio_context
1042 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
1043
1044 // Get the audio_index
1045 int index = mlt_properties_get_int( properties, "audio_index" );
1046
1047 // Reopen the file if necessary
1048 if ( !context && index != -1 )
1049 {
1050 mlt_events_block( properties, this );
1051 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(this) ),
1052 mlt_properties_get( properties, "resource" ) );
1053 context = mlt_properties_get_data( properties, "audio_context", NULL );
1054 index = mlt_properties_get_int( properties, "audio_index" );
1055 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
1056 mlt_events_unblock( properties, this );
1057 }
1058
1059 // Deal with audio context
1060 if ( context != NULL && index != -1 )
1061 {
1062 // Get the frame properties
1063 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1064
1065 // Get the audio stream
1066 AVStream *stream = context->streams[ index ];
1067
1068 // Get codec context
1069 AVCodecContext *codec_context = stream->codec;
1070
1071 // Get the codec
1072 AVCodec *codec = mlt_properties_get_data( properties, "audio_codec", NULL );
1073
1074 // Initialise the codec if necessary
1075 if ( codec == NULL )
1076 {
1077 // Find the codec
1078 codec = avcodec_find_decoder( codec_context->codec_id );
1079
1080 // If we don't have a codec and we can't initialise it, we can't do much more...
1081 avformat_lock( );
1082 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
1083 {
1084 // Now store the codec with its destructor
1085 mlt_properties_set_data( properties, "audio_codec", codec_context, 0, producer_codec_close, NULL );
1086
1087 }
1088 else
1089 {
1090 // Remember that we can't use this later
1091 mlt_properties_set_int( properties, "audio_index", -1 );
1092 }
1093 avformat_unlock( );
1094 }
1095
1096 // No codec, no show...
1097 if ( codec != NULL )
1098 {
1099 mlt_frame_push_audio( frame, producer_get_audio );
1100 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
1101 mlt_properties_set_int( frame_properties, "frequency", codec_context->sample_rate );
1102 mlt_properties_set_int( frame_properties, "channels", codec_context->channels );
1103 }
1104 }
1105 }
1106
1107 /** Our get frame implementation.
1108 */
1109
1110 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index )
1111 {
1112 // Create an empty frame
1113 *frame = mlt_frame_init( MLT_PRODUCER_SERVICE( this ) );
1114
1115 // Update timecode on the frame we're creating
1116 mlt_frame_set_position( *frame, mlt_producer_position( this ) );
1117
1118 // Set the position of this producer
1119 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( this ) );
1120
1121 // Set up the video
1122 producer_set_up_video( this, *frame );
1123
1124 // Set up the audio
1125 producer_set_up_audio( this, *frame );
1126
1127 // Set the aspect_ratio
1128 mlt_properties_set_double( MLT_FRAME_PROPERTIES( *frame ), "aspect_ratio", mlt_properties_get_double( MLT_PRODUCER_PROPERTIES( this ), "aspect_ratio" ) );
1129
1130 // Calculate the next timecode
1131 mlt_producer_prepare_next( this );
1132
1133 return 0;
1134 }