producer_avformat.c: do not free AVPacket if av_read_frame fails.
[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 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
20
21 // Local header files
22 #include "producer_avformat.h"
23
24 // MLT Header files
25 #include <framework/mlt_frame.h>
26
27 // ffmpeg Header files
28 #include <avformat.h>
29 #ifdef SWSCALE
30 #include <swscale.h>
31 #endif
32
33 // System header files
34 #include <stdlib.h>
35 #include <string.h>
36 #include <pthread.h>
37 #include <math.h>
38
39 void avformat_lock( );
40 void avformat_unlock( );
41
42 // Forward references.
43 static int producer_open( mlt_producer this, char *file );
44 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index );
45
46 /** Constructor for libavformat.
47 */
48
49 mlt_producer producer_avformat_init( char *file )
50 {
51 mlt_producer this = NULL;
52
53 // Check that we have a non-NULL argument
54 if ( file != NULL )
55 {
56 // Construct the producer
57 this = calloc( 1, sizeof( struct mlt_producer_s ) );
58
59 // Initialise it
60 if ( mlt_producer_init( this, NULL ) == 0 )
61 {
62 // Get the properties
63 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
64
65 // Set the resource property (required for all producers)
66 mlt_properties_set( properties, "resource", file );
67
68 // Register our get_frame implementation
69 this->get_frame = producer_get_frame;
70
71 // Open the file
72 if ( producer_open( this, file ) != 0 )
73 {
74 // Clean up
75 mlt_producer_close( this );
76 this = NULL;
77 }
78 }
79 }
80
81 return this;
82 }
83
84 /** Find the default streams.
85 */
86
87 static void find_default_streams( AVFormatContext *context, int *audio_index, int *video_index )
88 {
89 int i;
90
91 // Allow for multiple audio and video streams in the file and select first of each (if available)
92 for( i = 0; i < context->nb_streams; i++ )
93 {
94 // Get the codec context
95 AVCodecContext *codec_context = context->streams[ i ]->codec;
96
97 if ( avcodec_find_decoder( codec_context->codec_id ) == NULL )
98 continue;
99
100 // Determine the type and obtain the first index of each type
101 switch( codec_context->codec_type )
102 {
103 case CODEC_TYPE_VIDEO:
104 if ( *video_index < 0 )
105 *video_index = i;
106 break;
107 case CODEC_TYPE_AUDIO:
108 if ( *audio_index < 0 )
109 *audio_index = i;
110 break;
111 default:
112 break;
113 }
114 }
115 }
116
117 /** Producer file destructor.
118 */
119
120 static void producer_file_close( void *context )
121 {
122 if ( context != NULL )
123 {
124 // Lock the mutex now
125 avformat_lock( );
126
127 // Close the file
128 av_close_input_file( context );
129
130 // Unlock the mutex now
131 avformat_unlock( );
132 }
133 }
134
135 /** Producer file destructor.
136 */
137
138 static void producer_codec_close( void *codec )
139 {
140 if ( codec != NULL )
141 {
142 // Lock the mutex now
143 avformat_lock( );
144
145 // Close the file
146 avcodec_close( codec );
147
148 // Unlock the mutex now
149 avformat_unlock( );
150 }
151 }
152
153 /** Open the file.
154 */
155
156 static int producer_open( mlt_producer this, char *file )
157 {
158 // Return an error code (0 == no error)
159 int error = 0;
160
161 // Context for avformat
162 AVFormatContext *context = NULL;
163
164 // Get the properties
165 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
166
167 // We will treat everything with the producer fps
168 double fps = mlt_producer_get_fps( this );
169
170 // Lock the mutex now
171 avformat_lock( );
172
173 // If "MRL", then create AVInputFormat
174 AVInputFormat *format = NULL;
175 AVFormatParameters *params = NULL;
176 char *standard = NULL;
177 char *mrl = strchr( file, ':' );
178
179 // AV option (0 = both, 1 = video, 2 = audio)
180 int av = 0;
181
182 // Setting lowest log level
183 av_log_set_level( -1 );
184
185 // Only if there is not a protocol specification that avformat can handle
186 if ( mrl && !url_exist( file ) )
187 {
188 // 'file' becomes format abbreviation
189 mrl[0] = 0;
190
191 // Lookup the format
192 format = av_find_input_format( file );
193
194 // Eat the format designator
195 file = ++mrl;
196
197 if ( format )
198 {
199 // Allocate params
200 params = calloc( sizeof( AVFormatParameters ), 1 );
201
202 // These are required by video4linux (defaults)
203 params->width = 640;
204 params->height = 480;
205 params->time_base= (AVRational){1,25};
206 // params->device = file;
207 params->channels = 2;
208 params->sample_rate = 48000;
209 }
210
211 // XXX: this does not work anymore since avdevice
212 // TODO: make producer_avddevice?
213 // Parse out params
214 mrl = strchr( file, '?' );
215 while ( mrl )
216 {
217 mrl[0] = 0;
218 char *name = strdup( ++mrl );
219 char *value = strchr( name, ':' );
220 if ( value )
221 {
222 value[0] = 0;
223 value++;
224 char *t = strchr( value, '&' );
225 if ( t )
226 t[0] = 0;
227 if ( !strcmp( name, "frame_rate" ) )
228 params->time_base.den = atoi( value );
229 else if ( !strcmp( name, "frame_rate_base" ) )
230 params->time_base.num = atoi( value );
231 else if ( !strcmp( name, "sample_rate" ) )
232 params->sample_rate = atoi( value );
233 else if ( !strcmp( name, "channels" ) )
234 params->channels = atoi( value );
235 else if ( !strcmp( name, "width" ) )
236 params->width = atoi( value );
237 else if ( !strcmp( name, "height" ) )
238 params->height = atoi( value );
239 else if ( !strcmp( name, "standard" ) )
240 {
241 standard = strdup( value );
242 params->standard = standard;
243 }
244 else if ( !strcmp( name, "av" ) )
245 av = atoi( value );
246 }
247 free( name );
248 mrl = strchr( mrl, '&' );
249 }
250 }
251
252 // Now attempt to open the file
253 error = av_open_input_file( &context, file, format, 0, params ) < 0;
254
255 // Cleanup AVFormatParameters
256 free( standard );
257 free( params );
258
259 // If successful, then try to get additional info
260 if ( error == 0 )
261 {
262 // Get the stream info
263 error = av_find_stream_info( context ) < 0;
264
265 // Continue if no error
266 if ( error == 0 )
267 {
268 // We will default to the first audio and video streams found
269 int audio_index = -1;
270 int video_index = -1;
271 int av_bypass = 0;
272
273 // Now set properties where we can (use default unknowns if required)
274 if ( context->duration != AV_NOPTS_VALUE )
275 {
276 // This isn't going to be accurate for all formats
277 mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps + 0.5 );
278 mlt_properties_set_position( properties, "out", frames - 1 );
279 mlt_properties_set_position( properties, "length", frames );
280 }
281
282 // Find default audio and video streams
283 find_default_streams( context, &audio_index, &video_index );
284
285 if ( context->start_time != AV_NOPTS_VALUE )
286 mlt_properties_set_double( properties, "_start_time", context->start_time );
287
288 // Check if we're seekable (something funny about mpeg here :-/)
289 if ( strcmp( file, "pipe:" ) && strncmp( file, "http://", 6 ) )
290 {
291 mlt_properties_set_int( properties, "seekable", av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ), AVSEEK_FLAG_BACKWARD ) >= 0 );
292 mlt_properties_set_data( properties, "dummy_context", context, 0, producer_file_close, NULL );
293 av_open_input_file( &context, file, NULL, 0, NULL );
294 av_find_stream_info( context );
295 }
296 else
297 av_bypass = 1;
298
299 // Store selected audio and video indexes on properties
300 mlt_properties_set_int( properties, "audio_index", audio_index );
301 mlt_properties_set_int( properties, "video_index", video_index );
302 mlt_properties_set_int( properties, "_last_position", -1 );
303
304 // Fetch the width, height and aspect ratio
305 if ( video_index != -1 )
306 {
307 AVCodecContext *codec_context = context->streams[ video_index ]->codec;
308 mlt_properties_set_int( properties, "width", codec_context->width );
309 mlt_properties_set_int( properties, "height", codec_context->height );
310 mlt_properties_set_double( properties, "aspect_ratio", av_q2d( codec_context->sample_aspect_ratio ) );
311 }
312
313 // Read Metadata
314 if (context->title != NULL)
315 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
316 if (context->author != NULL)
317 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
318 if (context->copyright != NULL)
319 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
320 if (context->comment != NULL)
321 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
322 if (context->album != NULL)
323 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
324 if (context->year != 0)
325 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
326 if (context->track != 0)
327 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
328
329 // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
330 if ( av == 0 && !av_bypass && audio_index != -1 && video_index != -1 )
331 {
332 // We'll use the open one as our video_context
333 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
334
335 // And open again for our audio context
336 av_open_input_file( &context, file, NULL, 0, NULL );
337 av_find_stream_info( context );
338
339 // Audio context
340 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
341 }
342 else if ( av != 2 && video_index != -1 )
343 {
344 // We only have a video context
345 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
346 }
347 else if ( audio_index != -1 )
348 {
349 // We only have an audio context
350 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
351 }
352 else
353 {
354 // Something has gone wrong
355 error = -1;
356 }
357
358 mlt_properties_set_int( properties, "av_bypass", av_bypass );
359 }
360 }
361
362 // Unlock the mutex now
363 avformat_unlock( );
364
365 return error;
366 }
367
368 /** Convert a frame position to a time code.
369 */
370
371 static double producer_time_of_frame( mlt_producer this, mlt_position position )
372 {
373 return ( double )position / mlt_producer_get_fps( this );
374 }
375
376 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, mlt_image_format format, int width, int height )
377 {
378 #ifdef SWSCALE
379 if ( format == mlt_image_yuv420p )
380 {
381 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
382 width, height, PIX_FMT_YUV420P, SWS_FAST_BILINEAR, NULL, NULL, NULL);
383 AVPicture output;
384 output.data[0] = buffer;
385 output.data[1] = buffer + width * height;
386 output.data[2] = buffer + ( 3 * width * height ) / 2;
387 output.linesize[0] = width;
388 output.linesize[1] = width >> 1;
389 output.linesize[2] = width >> 1;
390 sws_scale( context, frame->data, frame->linesize, 0, height,
391 output.data, output.linesize);
392 sws_freeContext( context );
393 }
394 else if ( format == mlt_image_rgb24 )
395 {
396 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
397 width, height, PIX_FMT_RGB24, SWS_FAST_BILINEAR, NULL, NULL, NULL);
398 AVPicture output;
399 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
400 sws_scale( context, frame->data, frame->linesize, 0, height,
401 output.data, output.linesize);
402 sws_freeContext( context );
403 }
404 else
405 {
406 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
407 width, height, PIX_FMT_YUYV422, SWS_FAST_BILINEAR, NULL, NULL, NULL);
408 AVPicture output;
409 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
410 sws_scale( context, frame->data, frame->linesize, 0, height,
411 output.data, output.linesize);
412 sws_freeContext( context );
413 }
414 #else
415 if ( format == mlt_image_yuv420p )
416 {
417 AVPicture pict;
418 pict.data[0] = buffer;
419 pict.data[1] = buffer + width * height;
420 pict.data[2] = buffer + ( 3 * width * height ) / 2;
421 pict.linesize[0] = width;
422 pict.linesize[1] = width >> 1;
423 pict.linesize[2] = width >> 1;
424 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
425 }
426 else if ( format == mlt_image_rgb24 )
427 {
428 AVPicture output;
429 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
430 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
431 }
432 else
433 {
434 AVPicture output;
435 avpicture_fill( &output, buffer, PIX_FMT_YUV422, width, height );
436 img_convert( &output, PIX_FMT_YUV422, (AVPicture *)frame, pix_fmt, width, height );
437 }
438 #endif
439 }
440
441 /** Get an image from a frame.
442 */
443
444 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
445 {
446 // Get the properties from the frame
447 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
448
449 // Obtain the frame number of this frame
450 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
451
452 // Get the producer
453 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
454
455 // Get the producer properties
456 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
457
458 // Fetch the video_context
459 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
460
461 // Get the video_index
462 int index = mlt_properties_get_int( properties, "video_index" );
463
464 // Obtain the expected frame numer
465 mlt_position expected = mlt_properties_get_position( properties, "_video_expected" );
466
467 // Get the video stream
468 AVStream *stream = context->streams[ index ];
469
470 // Get codec context
471 AVCodecContext *codec_context = stream->codec;
472
473 // Packet
474 AVPacket pkt;
475
476 // Get the conversion frame
477 AVFrame *av_frame = mlt_properties_get_data( properties, "av_frame", NULL );
478
479 // Special case pause handling flag
480 int paused = 0;
481
482 // Special case ffwd handling
483 int ignore = 0;
484
485 // We may want to use the source fps if available
486 double source_fps = mlt_properties_get_double( properties, "source_fps" );
487 double fps = mlt_producer_get_fps( this );
488
489 // This is the physical frame position in the source
490 int req_position = ( int )( position / fps * source_fps + 0.5 );
491
492 // Get the seekable status
493 int seekable = mlt_properties_get_int( properties, "seekable" );
494
495 // Generate the size in bytes
496 int size = 0;
497
498 // Hopefully provide better support for streams...
499 int av_bypass = mlt_properties_get_int( properties, "av_bypass" );
500
501 // Determines if we have to decode all frames in a sequence
502 int must_decode = 1;
503
504 // Set the result arguments that we know here (only *buffer is now required)
505 *width = codec_context->width;
506 *height = codec_context->height;
507
508 switch ( *format )
509 {
510 case mlt_image_yuv420p:
511 size = *width * 3 * ( *height + 1 ) / 2;
512 break;
513 case mlt_image_rgb24:
514 size = *width * ( *height + 1 ) * 3;
515 break;
516 default:
517 *format = mlt_image_yuv422;
518 size = *width * ( *height + 1 ) * 2;
519 break;
520 }
521
522 // Set this on the frame properties
523 mlt_properties_set_int( frame_properties, "width", *width );
524 mlt_properties_set_int( frame_properties, "height", *height );
525
526 // Construct the output image
527 *buffer = mlt_pool_alloc( size );
528
529 // Temporary hack to improve intra frame only
530 must_decode = strcmp( codec_context->codec->name, "mjpeg" ) &&
531 strcmp( codec_context->codec->name, "rawvideo" ) &&
532 strcmp( codec_context->codec->name, "dvvideo" );
533
534 // Seek if necessary
535 if ( position != expected )
536 {
537 if ( av_frame != NULL && position + 1 == expected )
538 {
539 // We're paused - use last image
540 paused = 1;
541 }
542 else if ( !seekable && position > expected && ( position - expected ) < 250 )
543 {
544 // Fast forward - seeking is inefficient for small distances - just ignore following frames
545 ignore = ( int )( ( position - expected ) / fps * source_fps );
546 }
547 else if ( seekable && ( position < expected || position - expected >= 12 ) )
548 {
549 // Calculate the timestamp for the requested frame
550 int64_t timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
551 if ( ( uint64_t )context->start_time != AV_NOPTS_VALUE )
552 timestamp += context->start_time;
553 if ( must_decode )
554 timestamp -= AV_TIME_BASE;
555 if ( timestamp < 0 )
556 timestamp = 0;
557
558 // Set to the timestamp
559 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
560
561 // Remove the cached info relating to the previous position
562 mlt_properties_set_int( properties, "_current_position", -1 );
563 mlt_properties_set_int( properties, "_last_position", -1 );
564 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
565 av_frame = NULL;
566 }
567 }
568
569 // Duplicate the last image if necessary (see comment on rawvideo below)
570 int current_position = mlt_properties_get_int( properties, "_current_position" );
571 int got_picture = mlt_properties_get_int( properties, "_got_picture" );
572 if ( av_frame != NULL && got_picture && ( paused || current_position >= req_position ) && av_bypass == 0 )
573 {
574 // Duplicate it
575 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
576
577 // Set this on the frame properties
578 mlt_properties_set_data( frame_properties, "image", *buffer, size, ( mlt_destructor )mlt_pool_release, NULL );
579 }
580 else
581 {
582 int ret = 0;
583 int int_position = 0;
584 got_picture = 0;
585
586 av_init_packet( &pkt );
587
588 // Construct an AVFrame for YUV422 conversion
589 if ( av_frame == NULL )
590 {
591 av_frame = avcodec_alloc_frame( );
592 mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL );
593 }
594
595 while( ret >= 0 && !got_picture )
596 {
597 // Read a packet
598 ret = av_read_frame( context, &pkt );
599
600 // We only deal with video from the selected video_index
601 if ( ret >= 0 && pkt.stream_index == index && pkt.size > 0 )
602 {
603 // Determine time code of the packet
604 int_position = ( int )( av_q2d( stream->time_base ) * pkt.dts * source_fps + 0.5 );
605 if ( context->start_time != AV_NOPTS_VALUE )
606 int_position -= ( int )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
607 int last_position = mlt_properties_get_int( properties, "_last_position" );
608 if ( int_position == last_position )
609 int_position = last_position + 1;
610 mlt_properties_set_int( properties, "_last_position", int_position );
611
612 // Decode the image
613 if ( must_decode || int_position >= req_position )
614 ret = avcodec_decode_video( codec_context, av_frame, &got_picture, pkt.data, pkt.size );
615
616 if ( got_picture )
617 {
618 // Handle ignore
619 if ( int_position < req_position )
620 {
621 ignore = 0;
622 got_picture = 0;
623 }
624 else if ( int_position >= req_position )
625 {
626 ignore = 0;
627 }
628 else if ( ignore -- )
629 {
630 got_picture = 0;
631 }
632 }
633 av_free_packet( &pkt );
634 }
635 else if ( ret >= 0 )
636 {
637 av_free_packet( &pkt );
638 }
639
640 // Now handle the picture if we have one
641 if ( got_picture )
642 {
643 mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame );
644 mlt_properties_set_int( frame_properties, "top_field_first", av_frame->top_field_first );
645 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
646 mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL );
647 mlt_properties_set_int( properties, "_current_position", int_position );
648 mlt_properties_set_int( properties, "_got_picture", 1 );
649 }
650 }
651 }
652
653 // Very untidy - for rawvideo, the packet contains the frame, hence the free packet
654 // above will break the pause behaviour - so we wipe the frame now
655 if ( !strcmp( codec_context->codec->name, "rawvideo" ) )
656 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
657
658 // Set the field order property for this frame
659 mlt_properties_set_int( frame_properties, "top_field_first", mlt_properties_get_int( properties, "top_field_first" ) );
660
661 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
662 mlt_properties_set_position( properties, "_video_expected", position + 1 );
663
664 return 0;
665 }
666
667 /** Set up video handling.
668 */
669
670 static void producer_set_up_video( mlt_producer this, mlt_frame frame )
671 {
672 // Get the properties
673 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
674
675 // Fetch the video_context
676 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
677
678 // Get the video_index
679 int index = mlt_properties_get_int( properties, "video_index" );
680
681 // Get the frame properties
682 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
683
684 if ( context != NULL && index != -1 )
685 {
686 // Get the video stream
687 AVStream *stream = context->streams[ index ];
688
689 // Get codec context
690 AVCodecContext *codec_context = stream->codec;
691
692 // Get the codec
693 AVCodec *codec = mlt_properties_get_data( properties, "video_codec", NULL );
694
695 // Initialise the codec if necessary
696 if ( codec == NULL )
697 {
698 // Find the codec
699 codec = avcodec_find_decoder( codec_context->codec_id );
700
701 // If we don't have a codec and we can't initialise it, we can't do much more...
702 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
703 {
704 // Now store the codec with its destructor
705 mlt_properties_set_data( properties, "video_codec", codec_context, 0, producer_codec_close, NULL );
706 }
707 else
708 {
709 // Remember that we can't use this later
710 mlt_properties_set_int( properties, "video_index", -1 );
711 }
712 }
713
714 // No codec, no show...
715 if ( codec != NULL )
716 {
717 double source_fps = 0;
718 int norm_aspect_ratio = mlt_properties_get_int( properties, "norm_aspect_ratio" );
719 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
720 double aspect_ratio;
721
722 // XXX: We won't know the real aspect ratio until an image is decoded
723 // but we do need it now (to satisfy filter_resize) - take a guess based
724 // on pal/ntsc
725 if ( force_aspect_ratio > 0.0 )
726 {
727 aspect_ratio = force_aspect_ratio;
728 }
729 else if ( !norm_aspect_ratio && codec_context->sample_aspect_ratio.num > 0 )
730 {
731 aspect_ratio = av_q2d( codec_context->sample_aspect_ratio );
732 }
733 else
734 {
735 int is_pal = mlt_producer_get_fps( this ) == 25.0;
736 aspect_ratio = is_pal ? 59.0/54.0 : 10.0/11.0;
737 }
738
739 // Determine the fps
740 source_fps = ( double )codec_context->time_base.den / ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num );
741
742 // We'll use fps if it's available
743 if ( source_fps > 0 )
744 mlt_properties_set_double( properties, "source_fps", source_fps );
745 else
746 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( this ) );
747 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
748
749 // Set the width and height
750 mlt_properties_set_int( frame_properties, "width", codec_context->width );
751 mlt_properties_set_int( frame_properties, "height", codec_context->height );
752 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
753
754 mlt_frame_push_get_image( frame, producer_get_image );
755 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
756 }
757 else
758 {
759 mlt_properties_set_int( frame_properties, "test_image", 1 );
760 }
761 }
762 else
763 {
764 mlt_properties_set_int( frame_properties, "test_image", 1 );
765 }
766 }
767
768 /** Get the audio from a frame.
769 */
770
771 static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
772 {
773 // Get the properties from the frame
774 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
775
776 // Obtain the frame number of this frame
777 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
778
779 // Get the producer
780 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
781
782 // Get the producer properties
783 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
784
785 // Fetch the audio_context
786 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
787
788 // Get the audio_index
789 int index = mlt_properties_get_int( properties, "audio_index" );
790
791 // Get the seekable status
792 int seekable = mlt_properties_get_int( properties, "seekable" );
793
794 // Obtain the expected frame numer
795 mlt_position expected = mlt_properties_get_position( properties, "_audio_expected" );
796
797 // Obtain the resample context if it exists (not always needed)
798 ReSampleContext *resample = mlt_properties_get_data( properties, "audio_resample", NULL );
799
800 // Obtain the audio buffer
801 int16_t *audio_buffer = mlt_properties_get_data( properties, "audio_buffer", NULL );
802
803 // Get amount of audio used
804 int audio_used = mlt_properties_get_int( properties, "_audio_used" );
805
806 // Calculate the real time code
807 double real_timecode = producer_time_of_frame( this, position );
808
809 // Get the audio stream
810 AVStream *stream = context->streams[ index ];
811
812 // Get codec context
813 AVCodecContext *codec_context = stream->codec;
814
815 // Packet
816 AVPacket pkt;
817
818 // Number of frames to ignore (for ffwd)
819 int ignore = 0;
820
821 // Flag for paused (silence)
822 int paused = 0;
823
824 // Check for resample and create if necessary
825 if ( resample == NULL && codec_context->channels <= 2 )
826 {
827 // Create the resampler
828 resample = audio_resample_init( *channels, codec_context->channels, *frequency, codec_context->sample_rate );
829
830 // And store it on properties
831 mlt_properties_set_data( properties, "audio_resample", resample, 0, ( mlt_destructor )audio_resample_close, NULL );
832 }
833 else if ( resample == NULL )
834 {
835 *channels = codec_context->channels;
836 *frequency = codec_context->sample_rate;
837 }
838
839 // Check for audio buffer and create if necessary
840 if ( audio_buffer == NULL )
841 {
842 // Allocate the audio buffer
843 audio_buffer = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
844
845 // And store it on properties for reuse
846 mlt_properties_set_data( properties, "audio_buffer", audio_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
847 }
848
849 // Seek if necessary
850 if ( position != expected )
851 {
852 if ( position + 1 == expected )
853 {
854 // We're paused - silence required
855 paused = 1;
856 }
857 else if ( !seekable && position > expected && ( position - expected ) < 250 )
858 {
859 // Fast forward - seeking is inefficient for small distances - just ignore following frames
860 ignore = position - expected;
861 }
862 else if ( position < expected || position - expected >= 12 )
863 {
864 // Set to the real timecode
865 if ( av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ) + real_timecode * 1000000.0, AVSEEK_FLAG_BACKWARD ) != 0 )
866 paused = 1;
867
868 // Clear the usage in the audio buffer
869 audio_used = 0;
870 }
871 }
872
873 // Get the audio if required
874 if ( !paused )
875 {
876 int ret = 0;
877 int got_audio = 0;
878 int16_t *temp = av_malloc( sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE );
879
880 av_init_packet( &pkt );
881
882 while( ret >= 0 && !got_audio )
883 {
884 // Check if the buffer already contains the samples required
885 if ( audio_used >= *samples && ignore == 0 )
886 {
887 got_audio = 1;
888 break;
889 }
890
891 // Read a packet
892 ret = av_read_frame( context, &pkt );
893
894 int len = pkt.size;
895 uint8_t *ptr = pkt.data;
896
897 // We only deal with audio from the selected audio_index
898 while ( ptr != NULL && ret >= 0 && pkt.stream_index == index && len > 0 )
899 {
900 int data_size = sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE;
901
902 // Decode the audio
903 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
904 ret = avcodec_decode_audio2( codec_context, temp, &data_size, ptr, len );
905 #else
906 ret = avcodec_decode_audio( codec_context, temp, &data_size, ptr, len );
907 #endif
908 if ( ret < 0 )
909 {
910 ret = 0;
911 break;
912 }
913
914 len -= ret;
915 ptr += ret;
916
917 if ( data_size > 0 )
918 {
919 if ( resample != NULL )
920 {
921 audio_used += audio_resample( resample, &audio_buffer[ audio_used * *channels ], temp, data_size / ( codec_context->channels * sizeof( int16_t ) ) );
922 }
923 else
924 {
925 memcpy( &audio_buffer[ audio_used * *channels ], temp, data_size );
926 audio_used += data_size / ( codec_context->channels * sizeof( int16_t ) );
927 }
928
929 // Handle ignore
930 while ( ignore && audio_used > *samples )
931 {
932 ignore --;
933 audio_used -= *samples;
934 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * sizeof( int16_t ) );
935 }
936 }
937
938 // If we're behind, ignore this packet
939 float current_pts = av_q2d( stream->time_base ) * pkt.pts;
940 if ( seekable && ( !ignore && current_pts <= ( real_timecode - 0.02 ) ) )
941 ignore = 1;
942 }
943
944 // We're finished with this packet regardless
945 av_free_packet( &pkt );
946 }
947
948 *buffer = mlt_pool_alloc( *samples * *channels * sizeof( int16_t ) );
949 mlt_properties_set_data( frame_properties, "audio", *buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
950
951 // Now handle the audio if we have enough
952 if ( audio_used >= *samples )
953 {
954 memcpy( *buffer, audio_buffer, *samples * *channels * sizeof( int16_t ) );
955 audio_used -= *samples;
956 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * *channels * sizeof( int16_t ) );
957 }
958 else
959 {
960 memset( *buffer, 0, *samples * *channels * sizeof( int16_t ) );
961 }
962
963 // Store the number of audio samples still available
964 mlt_properties_set_int( properties, "_audio_used", audio_used );
965
966 // Release the temporary audio
967 av_free( temp );
968 }
969 else
970 {
971 // Get silence and don't touch the context
972 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
973 }
974
975 // Regardless of speed (other than paused), we expect to get the next frame
976 if ( !paused )
977 mlt_properties_set_position( properties, "_audio_expected", position + 1 );
978
979 return 0;
980 }
981
982 /** Set up audio handling.
983 */
984
985 static void producer_set_up_audio( mlt_producer this, mlt_frame frame )
986 {
987 // Get the properties
988 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
989
990 // Fetch the audio_context
991 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
992
993 // Get the audio_index
994 int index = mlt_properties_get_int( properties, "audio_index" );
995
996 // Deal with audio context
997 if ( context != NULL && index != -1 )
998 {
999 // Get the frame properties
1000 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1001
1002 // Get the audio stream
1003 AVStream *stream = context->streams[ index ];
1004
1005 // Get codec context
1006 AVCodecContext *codec_context = stream->codec;
1007
1008 // Get the codec
1009 AVCodec *codec = mlt_properties_get_data( properties, "audio_codec", NULL );
1010
1011 // Initialise the codec if necessary
1012 if ( codec == NULL )
1013 {
1014 // Find the codec
1015 codec = avcodec_find_decoder( codec_context->codec_id );
1016
1017 // If we don't have a codec and we can't initialise it, we can't do much more...
1018 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
1019 {
1020 // Now store the codec with its destructor
1021 mlt_properties_set_data( properties, "audio_codec", codec_context, 0, producer_codec_close, NULL );
1022
1023 }
1024 else
1025 {
1026 // Remember that we can't use this later
1027 mlt_properties_set_int( properties, "audio_index", -1 );
1028 }
1029 }
1030
1031 // No codec, no show...
1032 if ( codec != NULL )
1033 {
1034 mlt_frame_push_audio( frame, producer_get_audio );
1035 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
1036 mlt_properties_set_int( frame_properties, "frequency", codec_context->sample_rate );
1037 mlt_properties_set_int( frame_properties, "channels", codec_context->channels );
1038 }
1039 }
1040 }
1041
1042 /** Our get frame implementation.
1043 */
1044
1045 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index )
1046 {
1047 // Create an empty frame
1048 *frame = mlt_frame_init( );
1049
1050 // Update timecode on the frame we're creating
1051 mlt_frame_set_position( *frame, mlt_producer_position( this ) );
1052
1053 // Set the position of this producer
1054 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( this ) );
1055
1056 // Set up the video
1057 producer_set_up_video( this, *frame );
1058
1059 // Set up the audio
1060 producer_set_up_audio( this, *frame );
1061
1062 // Set the aspect_ratio
1063 mlt_properties_set_double( MLT_FRAME_PROPERTIES( *frame ), "aspect_ratio", mlt_properties_get_double( MLT_PRODUCER_PROPERTIES( this ), "aspect_ratio" ) );
1064
1065 // Calculate the next timecode
1066 mlt_producer_prepare_next( this );
1067
1068 return 0;
1069 }