producer_avformat.c: bugfix detection of aspect ratio for DV AVI (applies to raw...
[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 else
78 {
79 // Close the file to release resources for large playlists - reopen later as needed
80 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
81 mlt_properties_set_data( properties, "audio_context", NULL, 0, NULL, NULL );
82 mlt_properties_set_data( properties, "video_context", NULL, 0, NULL, NULL );
83
84 // Default the user-selectable indices from the auto-detected indices
85 mlt_properties_set_int( properties, "audio_index", mlt_properties_get_int( properties, "_audio_index" ) );
86 mlt_properties_set_int( properties, "video_index", mlt_properties_get_int( properties, "_video_index" ) );
87 }
88 }
89 }
90
91 return this;
92 }
93
94 /** Find the default streams.
95 */
96
97 static void find_default_streams( AVFormatContext *context, int *audio_index, int *video_index )
98 {
99 int i;
100
101 // Allow for multiple audio and video streams in the file and select first of each (if available)
102 for( i = 0; i < context->nb_streams; i++ )
103 {
104 // Get the codec context
105 AVCodecContext *codec_context = context->streams[ i ]->codec;
106
107 if ( avcodec_find_decoder( codec_context->codec_id ) == NULL )
108 continue;
109
110 // Determine the type and obtain the first index of each type
111 switch( codec_context->codec_type )
112 {
113 case CODEC_TYPE_VIDEO:
114 if ( *video_index < 0 )
115 *video_index = i;
116 break;
117 case CODEC_TYPE_AUDIO:
118 if ( *audio_index < 0 )
119 *audio_index = i;
120 break;
121 default:
122 break;
123 }
124 }
125 }
126
127 /** Producer file destructor.
128 */
129
130 static void producer_file_close( void *context )
131 {
132 if ( context != NULL )
133 {
134 // Lock the mutex now
135 avformat_lock( );
136
137 // Close the file
138 av_close_input_file( context );
139
140 // Unlock the mutex now
141 avformat_unlock( );
142 }
143 }
144
145 /** Producer file destructor.
146 */
147
148 static void producer_codec_close( void *codec )
149 {
150 if ( codec != NULL )
151 {
152 // Lock the mutex now
153 avformat_lock( );
154
155 // Close the file
156 avcodec_close( codec );
157
158 // Unlock the mutex now
159 avformat_unlock( );
160 }
161 }
162
163 static inline int dv_is_pal( AVPacket *pkt )
164 {
165 return pkt->data[3] & 0x80;
166 }
167
168 static int dv_is_wide( AVPacket *pkt )
169 {
170 int i = 80 /* block size */ *3 /* VAUX starts at block 3 */ +3 /* skip block header */;
171
172 for ( ; i < pkt->size; i += 5 /* packet size */ )
173 {
174 if ( pkt->data[ i ] == 0x61 )
175 {
176 uint8_t x = pkt->data[ i + 2 ] & 0x7;
177 return ( x == 2 ) || ( x == 7 );
178 }
179 }
180 return 0;
181 }
182
183 static double get_aspect_ratio( AVStream *stream, AVCodecContext *codec_context, AVPacket *pkt )
184 {
185 double aspect_ratio = 1.0;
186
187 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
188 {
189 if ( pkt )
190 {
191 if ( dv_is_pal( pkt ) )
192 {
193 aspect_ratio = dv_is_wide( pkt )
194 ? 64.0/45.0 // 16:9 PAL
195 : 16.0/15.0; // 4:3 PAL
196 }
197 else
198 {
199 aspect_ratio = dv_is_wide( pkt )
200 ? 32.0/27.0 // 16:9 NTSC
201 : 8.0/9.0; // 4:3 NTSC
202 }
203 }
204 else
205 {
206 AVRational ar =
207 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
208 stream->sample_aspect_ratio;
209 #else
210 codec_context->sample_aspect_ratio;
211 #endif
212 // Override FFmpeg's notion of DV aspect ratios, which are
213 // based upon a width of 704. Since we do not have a normaliser
214 // that crops (nor is cropping 720 wide ITU-R 601 video always desirable)
215 // we just coerce the values to facilitate a passive behaviour through
216 // the rescale normaliser when using equivalent producers and consumers.
217 // = display_aspect / (width * height)
218 if ( ar.num == 10 && ar.den == 11 )
219 aspect_ratio = 8.0/9.0; // 4:3 NTSC
220 else if ( ar.num == 59 && ar.den == 54 )
221 aspect_ratio = 16.0/15.0; // 4:3 PAL
222 else if ( ar.num == 40 && ar.den == 33 )
223 aspect_ratio = 32.0/27.0; // 16:9 NTSC
224 else if ( ar.num == 118 && ar.den == 81 )
225 aspect_ratio = 64.0/45.0; // 16:9 PAL
226 }
227 }
228 else
229 {
230 AVRational codec_sar = codec_context->sample_aspect_ratio;
231 AVRational stream_sar =
232 #if LIBAVFORMAT_VERSION_INT >= ((52<<16)+(21<<8)+0)
233 stream->sample_aspect_ratio;
234 #else
235 { 0, 1 };
236 #endif
237 if ( codec_sar.num > 0 )
238 aspect_ratio = av_q2d( codec_sar );
239 else if ( stream_sar.num > 0 )
240 aspect_ratio = av_q2d( stream_sar );
241 }
242 return aspect_ratio;
243 }
244
245 /** Open the file.
246 */
247
248 static int producer_open( mlt_producer this, mlt_profile profile, char *file )
249 {
250 // Return an error code (0 == no error)
251 int error = 0;
252
253 // Context for avformat
254 AVFormatContext *context = NULL;
255
256 // Get the properties
257 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
258
259 // We will treat everything with the producer fps
260 double fps = mlt_profile_fps( profile );
261
262 // Lock the mutex now
263 avformat_lock( );
264
265 // If "MRL", then create AVInputFormat
266 AVInputFormat *format = NULL;
267 AVFormatParameters *params = NULL;
268 char *standard = NULL;
269 char *mrl = strchr( file, ':' );
270
271 // AV option (0 = both, 1 = video, 2 = audio)
272 int av = 0;
273
274 // Setting lowest log level
275 av_log_set_level( -1 );
276
277 // Only if there is not a protocol specification that avformat can handle
278 if ( mrl && !url_exist( file ) )
279 {
280 // 'file' becomes format abbreviation
281 mrl[0] = 0;
282
283 // Lookup the format
284 format = av_find_input_format( file );
285
286 // Eat the format designator
287 file = ++mrl;
288
289 if ( format )
290 {
291 // Allocate params
292 params = calloc( sizeof( AVFormatParameters ), 1 );
293
294 // These are required by video4linux (defaults)
295 params->width = 640;
296 params->height = 480;
297 params->time_base= (AVRational){1,25};
298 // params->device = file;
299 params->channels = 2;
300 params->sample_rate = 48000;
301 }
302
303 // XXX: this does not work anymore since avdevice
304 // TODO: make producer_avddevice?
305 // Parse out params
306 mrl = strchr( file, '?' );
307 while ( mrl )
308 {
309 mrl[0] = 0;
310 char *name = strdup( ++mrl );
311 char *value = strchr( name, ':' );
312 if ( value )
313 {
314 value[0] = 0;
315 value++;
316 char *t = strchr( value, '&' );
317 if ( t )
318 t[0] = 0;
319 if ( !strcmp( name, "frame_rate" ) )
320 params->time_base.den = atoi( value );
321 else if ( !strcmp( name, "frame_rate_base" ) )
322 params->time_base.num = atoi( value );
323 else if ( !strcmp( name, "sample_rate" ) )
324 params->sample_rate = atoi( value );
325 else if ( !strcmp( name, "channels" ) )
326 params->channels = atoi( value );
327 else if ( !strcmp( name, "width" ) )
328 params->width = atoi( value );
329 else if ( !strcmp( name, "height" ) )
330 params->height = atoi( value );
331 else if ( !strcmp( name, "standard" ) )
332 {
333 standard = strdup( value );
334 params->standard = standard;
335 }
336 else if ( !strcmp( name, "av" ) )
337 av = atoi( value );
338 }
339 free( name );
340 mrl = strchr( mrl, '&' );
341 }
342 }
343
344 // Now attempt to open the file
345 error = av_open_input_file( &context, file, format, 0, params ) < 0;
346
347 // Cleanup AVFormatParameters
348 free( standard );
349 free( params );
350
351 // If successful, then try to get additional info
352 if ( error == 0 )
353 {
354 // Get the stream info
355 error = av_find_stream_info( context ) < 0;
356
357 // Continue if no error
358 if ( error == 0 )
359 {
360 // We will default to the first audio and video streams found
361 int audio_index = -1;
362 int video_index = -1;
363 int av_bypass = 0;
364
365 // Now set properties where we can (use default unknowns if required)
366 if ( context->duration != AV_NOPTS_VALUE )
367 {
368 // This isn't going to be accurate for all formats
369 mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps + 0.5 );
370 mlt_properties_set_position( properties, "out", frames - 1 );
371 mlt_properties_set_position( properties, "length", frames );
372 }
373
374 // Find default audio and video streams
375 find_default_streams( context, &audio_index, &video_index );
376
377 if ( context->start_time != AV_NOPTS_VALUE )
378 mlt_properties_set_double( properties, "_start_time", context->start_time );
379
380 // Check if we're seekable (something funny about mpeg here :-/)
381 if ( strcmp( file, "pipe:" ) && strncmp( file, "http://", 6 ) && strncmp( file, "udp:", 4 ) && strncmp( file, "tcp:", 4 ) && strncmp( file, "rtsp:", 5 ) && strncmp( file, "rtp:", 4 ) )
382 {
383 mlt_properties_set_int( properties, "seekable", av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ), AVSEEK_FLAG_BACKWARD ) >= 0 );
384 mlt_properties_set_data( properties, "dummy_context", context, 0, producer_file_close, NULL );
385 av_open_input_file( &context, file, NULL, 0, NULL );
386 av_find_stream_info( context );
387 }
388 else
389 av_bypass = 1;
390
391 // Store selected audio and video indexes on properties
392 mlt_properties_set_int( properties, "_audio_index", audio_index );
393 mlt_properties_set_int( properties, "_video_index", video_index );
394 mlt_properties_set_int( properties, "_last_position", -1 );
395
396 // Fetch the width, height and aspect ratio
397 if ( video_index != -1 )
398 {
399 AVCodecContext *codec_context = context->streams[ video_index ]->codec;
400 mlt_properties_set_int( properties, "width", codec_context->width );
401 mlt_properties_set_int( properties, "height", codec_context->height );
402
403 if ( codec_context->codec_id == CODEC_ID_DVVIDEO )
404 {
405 // Fetch the first frame of DV so we can read it directly
406 AVPacket pkt;
407 int ret = 0;
408 while ( ret >= 0 )
409 {
410 ret = av_read_frame( context, &pkt );
411 if ( ret >= 0 && pkt.stream_index == video_index && pkt.size > 0 )
412 {
413 mlt_properties_set_double( properties, "aspect_ratio",
414 get_aspect_ratio( context->streams[ video_index ], codec_context, &pkt ) );
415 break;
416 }
417 }
418 }
419 else
420 {
421 mlt_properties_set_double( properties, "aspect_ratio",
422 get_aspect_ratio( context->streams[ video_index ], codec_context, NULL ) );
423 }
424 }
425
426 // Read Metadata
427 if (context->title != NULL)
428 mlt_properties_set(properties, "meta.attr.title.markup", context->title );
429 if (context->author != NULL)
430 mlt_properties_set(properties, "meta.attr.author.markup", context->author );
431 if (context->copyright != NULL)
432 mlt_properties_set(properties, "meta.attr.copyright.markup", context->copyright );
433 if (context->comment != NULL)
434 mlt_properties_set(properties, "meta.attr.comment.markup", context->comment );
435 if (context->album != NULL)
436 mlt_properties_set(properties, "meta.attr.album.markup", context->album );
437 if (context->year != 0)
438 mlt_properties_set_int(properties, "meta.attr.year.markup", context->year );
439 if (context->track != 0)
440 mlt_properties_set_int(properties, "meta.attr.track.markup", context->track );
441
442 // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
443 if ( av == 0 && audio_index != -1 && video_index != -1 )
444 {
445 // We'll use the open one as our video_context
446 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
447
448 // And open again for our audio context
449 av_open_input_file( &context, file, NULL, 0, NULL );
450 av_find_stream_info( context );
451
452 // Audio context
453 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
454 }
455 else if ( av != 2 && video_index != -1 )
456 {
457 // We only have a video context
458 mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
459 }
460 else if ( audio_index != -1 )
461 {
462 // We only have an audio context
463 mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
464 }
465 else
466 {
467 // Something has gone wrong
468 error = -1;
469 }
470
471 mlt_properties_set_int( properties, "av_bypass", av_bypass );
472 }
473 }
474
475 // Unlock the mutex now
476 avformat_unlock( );
477
478 return error;
479 }
480
481 /** Convert a frame position to a time code.
482 */
483
484 static double producer_time_of_frame( mlt_producer this, mlt_position position )
485 {
486 return ( double )position / mlt_producer_get_fps( this );
487 }
488
489 static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, mlt_image_format format, int width, int height )
490 {
491 #ifdef SWSCALE
492 if ( format == mlt_image_yuv420p )
493 {
494 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
495 width, height, PIX_FMT_YUV420P, SWS_FAST_BILINEAR, NULL, NULL, NULL);
496 AVPicture output;
497 output.data[0] = buffer;
498 output.data[1] = buffer + width * height;
499 output.data[2] = buffer + ( 3 * width * height ) / 2;
500 output.linesize[0] = width;
501 output.linesize[1] = width >> 1;
502 output.linesize[2] = width >> 1;
503 sws_scale( context, frame->data, frame->linesize, 0, height,
504 output.data, output.linesize);
505 sws_freeContext( context );
506 }
507 else if ( format == mlt_image_rgb24 )
508 {
509 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
510 width, height, PIX_FMT_RGB24, SWS_FAST_BILINEAR, NULL, NULL, NULL);
511 AVPicture output;
512 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
513 sws_scale( context, frame->data, frame->linesize, 0, height,
514 output.data, output.linesize);
515 sws_freeContext( context );
516 }
517 else
518 {
519 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
520 width, height, PIX_FMT_YUYV422, SWS_FAST_BILINEAR, NULL, NULL, NULL);
521 AVPicture output;
522 avpicture_fill( &output, buffer, PIX_FMT_YUYV422, width, height );
523 sws_scale( context, frame->data, frame->linesize, 0, height,
524 output.data, output.linesize);
525 sws_freeContext( context );
526 }
527 #else
528 if ( format == mlt_image_yuv420p )
529 {
530 AVPicture pict;
531 pict.data[0] = buffer;
532 pict.data[1] = buffer + width * height;
533 pict.data[2] = buffer + ( 3 * width * height ) / 2;
534 pict.linesize[0] = width;
535 pict.linesize[1] = width >> 1;
536 pict.linesize[2] = width >> 1;
537 img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
538 }
539 else if ( format == mlt_image_rgb24 )
540 {
541 AVPicture output;
542 avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
543 img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
544 }
545 else
546 {
547 AVPicture output;
548 avpicture_fill( &output, buffer, PIX_FMT_YUV422, width, height );
549 img_convert( &output, PIX_FMT_YUV422, (AVPicture *)frame, pix_fmt, width, height );
550 }
551 #endif
552 }
553
554 /** Get an image from a frame.
555 */
556
557 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
558 {
559 // Get the properties from the frame
560 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
561
562 // Obtain the frame number of this frame
563 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
564
565 // Get the producer
566 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
567
568 // Get the producer properties
569 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
570
571 // Fetch the video_context
572 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
573
574 // Get the video_index
575 int index = mlt_properties_get_int( properties, "video_index" );
576
577 // Obtain the expected frame numer
578 mlt_position expected = mlt_properties_get_position( properties, "_video_expected" );
579
580 // Get the video stream
581 AVStream *stream = context->streams[ index ];
582
583 // Get codec context
584 AVCodecContext *codec_context = stream->codec;
585
586 // Packet
587 AVPacket pkt;
588
589 // Get the conversion frame
590 AVFrame *av_frame = mlt_properties_get_data( properties, "av_frame", NULL );
591
592 // Special case pause handling flag
593 int paused = 0;
594
595 // Special case ffwd handling
596 int ignore = 0;
597
598 // We may want to use the source fps if available
599 double source_fps = mlt_properties_get_double( properties, "source_fps" );
600 double fps = mlt_producer_get_fps( this );
601
602 // This is the physical frame position in the source
603 int req_position = ( int )( position / fps * source_fps + 0.5 );
604
605 // Get the seekable status
606 int seekable = mlt_properties_get_int( properties, "seekable" );
607
608 // Generate the size in bytes
609 int size = 0;
610
611 // Hopefully provide better support for streams...
612 int av_bypass = mlt_properties_get_int( properties, "av_bypass" );
613
614 // Determines if we have to decode all frames in a sequence
615 int must_decode = 1;
616
617 // Set the result arguments that we know here (only *buffer is now required)
618 *width = codec_context->width;
619 *height = codec_context->height;
620
621 switch ( *format )
622 {
623 case mlt_image_yuv420p:
624 size = *width * 3 * ( *height + 1 ) / 2;
625 break;
626 case mlt_image_rgb24:
627 size = *width * ( *height + 1 ) * 3;
628 break;
629 default:
630 *format = mlt_image_yuv422;
631 size = *width * ( *height + 1 ) * 2;
632 break;
633 }
634
635 // Set this on the frame properties
636 mlt_properties_set_int( frame_properties, "width", *width );
637 mlt_properties_set_int( frame_properties, "height", *height );
638
639 // Construct the output image
640 *buffer = mlt_pool_alloc( size );
641
642 // Temporary hack to improve intra frame only
643 must_decode = strcmp( codec_context->codec->name, "mjpeg" ) &&
644 strcmp( codec_context->codec->name, "rawvideo" ) &&
645 strcmp( codec_context->codec->name, "dvvideo" );
646
647 // Seek if necessary
648 if ( position != expected )
649 {
650 if ( av_frame != NULL && position + 1 == expected )
651 {
652 // We're paused - use last image
653 paused = 1;
654 }
655 else if ( !seekable && position > expected && ( position - expected ) < 250 )
656 {
657 // Fast forward - seeking is inefficient for small distances - just ignore following frames
658 ignore = ( int )( ( position - expected ) / fps * source_fps );
659 }
660 else if ( seekable && ( position < expected || position - expected >= 12 ) )
661 {
662 // Calculate the timestamp for the requested frame
663 int64_t timestamp = ( int64_t )( ( double )req_position / source_fps * AV_TIME_BASE + 0.5 );
664 if ( ( uint64_t )context->start_time != AV_NOPTS_VALUE )
665 timestamp += context->start_time;
666 if ( must_decode )
667 timestamp -= AV_TIME_BASE;
668 if ( timestamp < 0 )
669 timestamp = 0;
670
671 // Set to the timestamp
672 av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD );
673
674 // Remove the cached info relating to the previous position
675 mlt_properties_set_int( properties, "_current_position", -1 );
676 mlt_properties_set_int( properties, "_last_position", -1 );
677 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
678 av_frame = NULL;
679 }
680 }
681
682 // Duplicate the last image if necessary (see comment on rawvideo below)
683 int current_position = mlt_properties_get_int( properties, "_current_position" );
684 int got_picture = mlt_properties_get_int( properties, "_got_picture" );
685 if ( av_frame != NULL && got_picture && ( paused || current_position >= req_position ) && av_bypass == 0 )
686 {
687 // Duplicate it
688 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
689
690 // Set this on the frame properties
691 mlt_properties_set_data( frame_properties, "image", *buffer, size, ( mlt_destructor )mlt_pool_release, NULL );
692 }
693 else
694 {
695 int ret = 0;
696 int int_position = 0;
697 got_picture = 0;
698
699 av_init_packet( &pkt );
700
701 // Construct an AVFrame for YUV422 conversion
702 if ( av_frame == NULL )
703 {
704 av_frame = avcodec_alloc_frame( );
705 mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL );
706 }
707
708 while( ret >= 0 && !got_picture )
709 {
710 // Read a packet
711 ret = av_read_frame( context, &pkt );
712
713 // We only deal with video from the selected video_index
714 if ( ret >= 0 && pkt.stream_index == index && pkt.size > 0 )
715 {
716 // Determine time code of the packet
717 int_position = ( int )( av_q2d( stream->time_base ) * pkt.dts * source_fps + 0.5 );
718 if ( context->start_time != AV_NOPTS_VALUE )
719 int_position -= ( int )( context->start_time * source_fps / AV_TIME_BASE + 0.5 );
720 int last_position = mlt_properties_get_int( properties, "_last_position" );
721 if ( int_position == last_position )
722 int_position = last_position + 1;
723 mlt_properties_set_int( properties, "_last_position", int_position );
724
725 // Decode the image
726 if ( must_decode || int_position >= req_position )
727 ret = avcodec_decode_video( codec_context, av_frame, &got_picture, pkt.data, pkt.size );
728
729 if ( got_picture )
730 {
731 // Handle ignore
732 if ( int_position < req_position )
733 {
734 ignore = 0;
735 got_picture = 0;
736 }
737 else if ( int_position >= req_position )
738 {
739 ignore = 0;
740 }
741 else if ( ignore -- )
742 {
743 got_picture = 0;
744 }
745 }
746 av_free_packet( &pkt );
747 }
748 else if ( ret >= 0 )
749 {
750 av_free_packet( &pkt );
751 }
752
753 // Now handle the picture if we have one
754 if ( got_picture )
755 {
756 mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame );
757 mlt_properties_set_int( properties, "top_field_first", av_frame->top_field_first );
758 convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );
759 mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL );
760 mlt_properties_set_int( properties, "_current_position", int_position );
761 mlt_properties_set_int( properties, "_got_picture", 1 );
762 }
763 }
764 if ( !got_picture )
765 mlt_frame_get_image( frame, buffer, format, width, height, writable );
766 }
767
768 // Very untidy - for rawvideo, the packet contains the frame, hence the free packet
769 // above will break the pause behaviour - so we wipe the frame now
770 if ( !strcmp( codec_context->codec->name, "rawvideo" ) )
771 mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
772
773 // Set the field order property for this frame
774 mlt_properties_set_int( frame_properties, "top_field_first", mlt_properties_get_int( properties, "top_field_first" ) );
775
776 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
777 mlt_properties_set_position( properties, "_video_expected", position + 1 );
778
779 return 0;
780 }
781
782 /** Set up video handling.
783 */
784
785 static void producer_set_up_video( mlt_producer this, mlt_frame frame )
786 {
787 // Get the properties
788 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
789
790 // Fetch the video_context
791 AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );
792
793 // Get the video_index
794 int index = mlt_properties_get_int( properties, "video_index" );
795
796 // Reopen the file if necessary
797 if ( !context && index > -1 )
798 {
799 mlt_events_block( properties, this );
800 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(this) ),
801 mlt_properties_get( properties, "resource" ) );
802 context = mlt_properties_get_data( properties, "video_context", NULL );
803 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
804 mlt_events_unblock( properties, this );
805 }
806
807 // Exception handling for video_index
808 if ( context && index >= (int) context->nb_streams )
809 {
810 // Get the last video stream
811 for ( index = context->nb_streams - 1; index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO; --index );
812 mlt_properties_set_int( properties, "video_index", index );
813 }
814 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_VIDEO )
815 {
816 // Invalidate the video stream
817 index = -1;
818 mlt_properties_set_int( properties, "video_index", index );
819 }
820
821 // Get the frame properties
822 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
823
824 if ( context && index > -1 )
825 {
826 // Get the video stream
827 AVStream *stream = context->streams[ index ];
828
829 // Get codec context
830 AVCodecContext *codec_context = stream->codec;
831
832 // Get the codec
833 AVCodec *codec = mlt_properties_get_data( properties, "video_codec", NULL );
834
835 // Update the video properties if the index changed
836 if ( index != mlt_properties_get_int( properties, "_video_index" ) )
837 {
838 // Reset the video properties if the index changed
839 mlt_properties_set_int( properties, "_video_index", index );
840 mlt_properties_set_data( properties, "video_codec", NULL, 0, NULL, NULL );
841 mlt_properties_set_int( properties, "width", codec_context->width );
842 mlt_properties_set_int( properties, "height", codec_context->height );
843 // TODO: get the first usable AVPacket and reset the stream position
844 mlt_properties_set_double( properties, "aspect_ratio",
845 get_aspect_ratio( context->streams[ index ], codec_context, NULL ) );
846 codec = NULL;
847 }
848
849 // Initialise the codec if necessary
850 if ( codec == NULL )
851 {
852 // Initialise multi-threading
853 int thread_count = mlt_properties_get_int( properties, "threads" );
854 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
855 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
856 if ( thread_count > 1 )
857 {
858 avcodec_thread_init( codec_context, thread_count );
859 codec_context->thread_count = thread_count;
860 }
861
862 // Find the codec
863 codec = avcodec_find_decoder( codec_context->codec_id );
864
865 // If we don't have a codec and we can't initialise it, we can't do much more...
866 avformat_lock( );
867 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
868 {
869 // Now store the codec with its destructor
870 mlt_properties_set_data( properties, "video_codec", codec_context, 0, producer_codec_close, NULL );
871 }
872 else
873 {
874 // Remember that we can't use this later
875 mlt_properties_set_int( properties, "video_index", -1 );
876 }
877 avformat_unlock( );
878 }
879
880 // No codec, no show...
881 if ( codec != NULL )
882 {
883 double source_fps = 0;
884 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
885 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
886 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
887
888 // Determine the fps
889 source_fps = ( double )codec_context->time_base.den / ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num );
890
891 // We'll use fps if it's available
892 if ( source_fps > 0 )
893 mlt_properties_set_double( properties, "source_fps", source_fps );
894 else
895 mlt_properties_set_double( properties, "source_fps", mlt_producer_get_fps( this ) );
896 mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
897
898 // Set the width and height
899 mlt_properties_set_int( frame_properties, "width", codec_context->width );
900 mlt_properties_set_int( frame_properties, "height", codec_context->height );
901 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
902
903 mlt_frame_push_get_image( frame, producer_get_image );
904 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
905 }
906 else
907 {
908 mlt_properties_set_int( frame_properties, "test_image", 1 );
909 }
910 }
911 else
912 {
913 mlt_properties_set_int( frame_properties, "test_image", 1 );
914 }
915 }
916
917 /** Get the audio from a frame.
918 */
919
920 static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
921 {
922 // Get the properties from the frame
923 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
924
925 // Obtain the frame number of this frame
926 mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );
927
928 // Get the producer
929 mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );
930
931 // Get the producer properties
932 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
933
934 // Fetch the audio_context
935 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
936
937 // Get the audio_index
938 int index = mlt_properties_get_int( properties, "audio_index" );
939
940 // Get the seekable status
941 int seekable = mlt_properties_get_int( properties, "seekable" );
942
943 // Obtain the expected frame numer
944 mlt_position expected = mlt_properties_get_position( properties, "_audio_expected" );
945
946 // Obtain the resample context if it exists (not always needed)
947 ReSampleContext *resample = mlt_properties_get_data( properties, "audio_resample", NULL );
948
949 // Obtain the audio buffer
950 int16_t *audio_buffer = mlt_properties_get_data( properties, "audio_buffer", NULL );
951
952 // Get amount of audio used
953 int audio_used = mlt_properties_get_int( properties, "_audio_used" );
954
955 // Calculate the real time code
956 double real_timecode = producer_time_of_frame( this, position );
957
958 // Get the audio stream
959 AVStream *stream = context->streams[ index ];
960
961 // Get codec context
962 AVCodecContext *codec_context = stream->codec;
963
964 // Packet
965 AVPacket pkt;
966
967 // Number of frames to ignore (for ffwd)
968 int ignore = 0;
969
970 // Flag for paused (silence)
971 int paused = 0;
972
973 // Check for resample and create if necessary
974 if ( resample == NULL && codec_context->channels <= 2 )
975 {
976 // Create the resampler
977 resample = audio_resample_init( *channels, codec_context->channels, *frequency, codec_context->sample_rate );
978
979 // And store it on properties
980 mlt_properties_set_data( properties, "audio_resample", resample, 0, ( mlt_destructor )audio_resample_close, NULL );
981 }
982 else if ( resample == NULL )
983 {
984 *channels = codec_context->channels;
985 *frequency = codec_context->sample_rate;
986 }
987
988 // Check for audio buffer and create if necessary
989 if ( audio_buffer == NULL )
990 {
991 // Allocate the audio buffer
992 audio_buffer = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );
993
994 // And store it on properties for reuse
995 mlt_properties_set_data( properties, "audio_buffer", audio_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
996 }
997
998 // Seek if necessary
999 if ( position != expected )
1000 {
1001 if ( position + 1 == expected )
1002 {
1003 // We're paused - silence required
1004 paused = 1;
1005 }
1006 else if ( !seekable && position > expected && ( position - expected ) < 250 )
1007 {
1008 // Fast forward - seeking is inefficient for small distances - just ignore following frames
1009 ignore = position - expected;
1010 }
1011 else if ( position < expected || position - expected >= 12 )
1012 {
1013 // Set to the real timecode
1014 if ( av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ) + real_timecode * 1000000.0, AVSEEK_FLAG_BACKWARD ) != 0 )
1015 paused = 1;
1016
1017 // Clear the usage in the audio buffer
1018 audio_used = 0;
1019 }
1020 }
1021
1022 // Get the audio if required
1023 if ( !paused )
1024 {
1025 int ret = 0;
1026 int got_audio = 0;
1027 int16_t *temp = av_malloc( sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE );
1028
1029 av_init_packet( &pkt );
1030
1031 while( ret >= 0 && !got_audio )
1032 {
1033 // Check if the buffer already contains the samples required
1034 if ( audio_used >= *samples && ignore == 0 )
1035 {
1036 got_audio = 1;
1037 break;
1038 }
1039
1040 // Read a packet
1041 ret = av_read_frame( context, &pkt );
1042
1043 int len = pkt.size;
1044 uint8_t *ptr = pkt.data;
1045
1046 // We only deal with audio from the selected audio_index
1047 while ( ptr != NULL && ret >= 0 && pkt.stream_index == index && len > 0 )
1048 {
1049 int data_size = sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE;
1050
1051 // Decode the audio
1052 #if (LIBAVCODEC_VERSION_INT >= ((51<<16)+(29<<8)+0))
1053 ret = avcodec_decode_audio2( codec_context, temp, &data_size, ptr, len );
1054 #else
1055 ret = avcodec_decode_audio( codec_context, temp, &data_size, ptr, len );
1056 #endif
1057 if ( ret < 0 )
1058 {
1059 ret = 0;
1060 break;
1061 }
1062
1063 len -= ret;
1064 ptr += ret;
1065
1066 if ( data_size > 0 )
1067 {
1068 if ( resample != NULL )
1069 {
1070 audio_used += audio_resample( resample, &audio_buffer[ audio_used * *channels ], temp, data_size / ( codec_context->channels * sizeof( int16_t ) ) );
1071 }
1072 else
1073 {
1074 memcpy( &audio_buffer[ audio_used * *channels ], temp, data_size );
1075 audio_used += data_size / ( codec_context->channels * sizeof( int16_t ) );
1076 }
1077
1078 // Handle ignore
1079 while ( ignore && audio_used > *samples )
1080 {
1081 ignore --;
1082 audio_used -= *samples;
1083 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * sizeof( int16_t ) );
1084 }
1085 }
1086
1087 // If we're behind, ignore this packet
1088 if ( pkt.pts >= 0 )
1089 {
1090 float current_pts = av_q2d( stream->time_base ) * pkt.pts;
1091 if ( seekable && ( !ignore && current_pts <= ( real_timecode - 0.02 ) ) )
1092 ignore = 1;
1093 }
1094 }
1095
1096 // We're finished with this packet regardless
1097 av_free_packet( &pkt );
1098 }
1099
1100 *buffer = mlt_pool_alloc( *samples * *channels * sizeof( int16_t ) );
1101 mlt_properties_set_data( frame_properties, "audio", *buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
1102
1103 // Now handle the audio if we have enough
1104 if ( audio_used >= *samples )
1105 {
1106 memcpy( *buffer, audio_buffer, *samples * *channels * sizeof( int16_t ) );
1107 audio_used -= *samples;
1108 memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * *channels * sizeof( int16_t ) );
1109 }
1110 else
1111 {
1112 memset( *buffer, 0, *samples * *channels * sizeof( int16_t ) );
1113 }
1114
1115 // Store the number of audio samples still available
1116 mlt_properties_set_int( properties, "_audio_used", audio_used );
1117
1118 // Release the temporary audio
1119 av_free( temp );
1120 }
1121 else
1122 {
1123 // Get silence and don't touch the context
1124 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
1125 }
1126
1127 // Regardless of speed (other than paused), we expect to get the next frame
1128 if ( !paused )
1129 mlt_properties_set_position( properties, "_audio_expected", position + 1 );
1130
1131 return 0;
1132 }
1133
1134 /** Set up audio handling.
1135 */
1136
1137 static void producer_set_up_audio( mlt_producer this, mlt_frame frame )
1138 {
1139 // Get the properties
1140 mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );
1141
1142 // Fetch the audio_context
1143 AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );
1144
1145 // Get the audio_index
1146 int index = mlt_properties_get_int( properties, "audio_index" );
1147
1148 // Reopen the file if necessary
1149 if ( !context && index > -1 )
1150 {
1151 mlt_events_block( properties, this );
1152 producer_open( this, mlt_service_profile( MLT_PRODUCER_SERVICE(this) ),
1153 mlt_properties_get( properties, "resource" ) );
1154 context = mlt_properties_get_data( properties, "audio_context", NULL );
1155 mlt_properties_set_data( properties, "dummy_context", NULL, 0, NULL, NULL );
1156 mlt_events_unblock( properties, this );
1157 }
1158
1159 // Exception handling for audio_index
1160 if ( context && index >= (int) context->nb_streams )
1161 {
1162 for ( index = context->nb_streams - 1; index >= 0 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO; --index );
1163 mlt_properties_set_int( properties, "audio_index", index );
1164 }
1165 if ( context && index > -1 && context->streams[ index ]->codec->codec_type != CODEC_TYPE_AUDIO )
1166 {
1167 index = -1;
1168 mlt_properties_set_int( properties, "audio_index", index );
1169 }
1170
1171 // Update the audio properties if the index changed
1172 if ( index > -1 && index != mlt_properties_get_int( properties, "_audio_index" ) ) {
1173 mlt_properties_set_int( properties, "_audio_index", index );
1174 mlt_properties_set_data( properties, "audio_codec", NULL, 0, NULL, NULL );
1175 }
1176
1177 // Deal with audio context
1178 if ( context != NULL && index > -1 )
1179 {
1180 // Get the frame properties
1181 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1182
1183 // Get the audio stream
1184 AVStream *stream = context->streams[ index ];
1185
1186 // Get codec context
1187 AVCodecContext *codec_context = stream->codec;
1188
1189 // Get the codec
1190 AVCodec *codec = mlt_properties_get_data( properties, "audio_codec", NULL );
1191
1192 // Initialise the codec if necessary
1193 if ( codec == NULL )
1194 {
1195 // Find the codec
1196 codec = avcodec_find_decoder( codec_context->codec_id );
1197
1198 // If we don't have a codec and we can't initialise it, we can't do much more...
1199 avformat_lock( );
1200 if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
1201 {
1202 // Now store the codec with its destructor
1203 mlt_properties_set_data( properties, "audio_codec", codec_context, 0, producer_codec_close, NULL );
1204
1205 }
1206 else
1207 {
1208 // Remember that we can't use this later
1209 mlt_properties_set_int( properties, "audio_index", -1 );
1210 }
1211 avformat_unlock( );
1212 }
1213
1214 // No codec, no show...
1215 if ( codec != NULL )
1216 {
1217 mlt_frame_push_audio( frame, producer_get_audio );
1218 mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
1219 mlt_properties_set_int( frame_properties, "frequency", codec_context->sample_rate );
1220 mlt_properties_set_int( frame_properties, "channels", codec_context->channels );
1221 }
1222 }
1223 }
1224
1225 /** Our get frame implementation.
1226 */
1227
1228 static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index )
1229 {
1230 // Create an empty frame
1231 *frame = mlt_frame_init( MLT_PRODUCER_SERVICE( this ) );
1232
1233 // Update timecode on the frame we're creating
1234 mlt_frame_set_position( *frame, mlt_producer_position( this ) );
1235
1236 // Set the position of this producer
1237 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( this ) );
1238
1239 // Set up the video
1240 producer_set_up_video( this, *frame );
1241
1242 // Set up the audio
1243 producer_set_up_audio( this, *frame );
1244
1245 // Set the aspect_ratio
1246 mlt_properties_set_double( MLT_FRAME_PROPERTIES( *frame ), "aspect_ratio", mlt_properties_get_double( MLT_PRODUCER_PROPERTIES( this ), "aspect_ratio" ) );
1247
1248 // Calculate the next timecode
1249 mlt_producer_prepare_next( this );
1250
1251 return 0;
1252 }